Datasets:

mlcore

/

arxiv-classifier

like 1

1801.00487

3

1801

Full orientation control of epitaxial MoS2 on hBN assisted by substrate defects

Inversion asymmetry in two-dimensional materials grants them fascinating properties such as spin-coupled valley degrees of freedom and piezoelectricity, but at the cost of inversion domain boundaries if the epitaxy of the grown 2D layer -- on a polar substrate -- cannot adequately distinguish what are often near-degenerate 0{\deg} and 180{\deg} orientations. We employ first-principles calculations to identify a method to lift this near-degeneracy: the energetic distinction between eclipsed and staggered configurations during nucleation at a point defect in the substrate. For monolayer MoS2 grown on hexagonal boron nitride, the predicted defect complex can be more stable than common MoS2 point defects because it is both a donor-acceptor pair and a Frenkel pair shared between adjacent layers of a 2D heterostack. Orientation control is verified in experiments that achieve ~90% consistency in the orientation of as-grown triangular MoS2 flakes on hBN, as confirmed by aberration-corrected scanning/transmission electron microscopy. This defect-enhanced orientational epitaxy could provide a general mechanism to break the near-degeneracy of 0/180{\deg} orientations of polar 2D materials on polar substrates, overcoming a long-standing impediment to scalable synthesis of single-crystal 2D semiconductors.

cond-mat.mes-hall

cond-mat

Full orientation control of epitaxial MoS2 on hBN assisted by substrate defects Fu Zhang1,2*, Yuanxi Wang2,3†*, Chad Erb1,4, Ke Wang4, Parivash Moradifar1,4, Vincent H. Crespi1,2,3,5‡, Nasim Alem1,2§ 1 Department of Materials Science and Engineering, Pennsylvania State University, University Park, PA 16802, USA. 2 Center for Two Dimensional and Layered Materials, Pennsylvania State University, University Park, PA 16802, USA. 3 2-Dimensional Crystal Consortium, Pennsylvania State University, University Park, PA 16802, USA. 4 Materials Research Institute, Pennsylvania State University, University Park, PA 16802, USA. 5 Department of Physics, Department of Chemistry, The Pennsylvania State University, University Park, PA 16802, USA. Inversion asymmetry in two-dimensional materials grants them fascinating properties such as spin-coupled valley degrees of freedom and piezoelectricity, but at the cost of inversion domain boundaries if the epitaxy of the grown 2D layer -- on a polar substrate -- cannot adequately distinguish what are often near-degenerate 0° and 180° orientations. We employ first-principles calculations to identify a method to lift this near-degeneracy: the energetic distinction between eclipsed and staggered configurations during nucleation at a point defect in the substrate. For monolayer MoS2 grown on hexagonal boron nitride, the predicted defect complex can be more stable than common MoS2 point defects because it is both a donor-acceptor pair and a Frenkel pair shared between adjacent layers of a 2D heterostack. Orientation control is verified in experiments that achieve ~90% consistency in the orientation of as-grown triangular MoS2 flakes on hBN, as confirmed by aberration-corrected scanning/transmission electron microscopy. This defect-enhanced orientational epitaxy could provide a general mechanism to break the near-degeneracy of 0/180° orientations of polar 2D materials on polar substrates, overcoming a long-standing impediment to scalable synthesis of single-crystal 2D semiconductors. †[email protected] ‡[email protected] §[email protected] * These authors contributed equally to this work. I. INTRODUCTION these between form at near-degeneracy typically strong enough energetically The breaking of in-plane inversion symmetry in polar two- dimensional (2D) crystals such as monolayer MoS2 introduces novel physics such as coupled spin-valley degrees of freedom [1,2] and in-plane piezoelectricity [3,4]. Yet such blessings come with a curse: While the interactions of polar 2D layers with near-commensurate polar substrates to disfavor arbitrary are orientations and favor two discrete orientations 180° apart, they are too weak to break the remaining two orientations [5,6]. The inversion domain boundaries that then interfaces of merging crystallites [7,8] can degrade device performance [9] and may induce undesirable multilayer growth [10]. Such inversion domain boundaries also complicate the growth of insulators such as Bi2Se3 [11], high-Tc topological superconductors [12], and 3D binary semiconductors [12] (even on carefully chosen lattice-matched substrates). Growth of high-quality single crystals is often associated with the discovery of new physics [13 -- 16]; such growth outcomes have been impeded in polar 2D materials by the ubiquitous presence of inversion grain boundaries. Prior efforts to suppress inversion domain formation include guiding lateral growth at step edges [11,12] (at the lateral the than grew prior work risk of inducing undesirable multilayer growth), or limiting nucleation density [10] (at the cost of slower growth rate). Interesting transition metal dichalcogenides (TMD) directly on hexagonal boron nitride (hBN) by powder vapor transport (PVT), chemical vapor deposition [5,6,17 -- 19], or thermal decomposition [20] to achieve scalability better that of mechanically transferred heterostructures [21 -- 27], but never achieved full orientational epitaxy (i.e. distinguishing inverted domains). The minimum requirement of distinguishing inversion domains in the grown TMD layer is the breaking of in-plane inversion symmetry in the substrate, limiting potential choices to layered compounds such as hBN and semiconductor surfaces such as the (0001) facets of GaN and sapphire. Here we focus on an hBN substrate due to its lack of surface inhomogeneity and dangling bonds [6]. We employ first-principles calculations to identify common intrinsic defects in the hBN substrate that can amplify the distinction between the 0° and 180° stacking geometries and enable full epitaxial growth: a paradoxical defect-enhanced orientational epitaxy in which structural defects (in the substrate) improve material quality in the layer grown above. Similar orientation control then observed experimentally by growing MoS2 on exfoliated hBN substrates using PVT, with excellent (~90%) orientational epitaxy. The geometry of the resulting population of is triangular flakes is compatible with a near-seamless monolayer containing very inversion domain boundaries. Aberration-corrected scanning/transmission electron microscopy (AC-S/TEM) confirms the atomic structure and orientation of the MoS2/hBN system. few II. STACKING DEGENERACY OF INVERSION DOMAINS their ~28% We begin by revisiting the difficulty in lifting the 0/180º near-degeneracy for TMDs stacked on commensurate or near-commensurate substrates. The local minimum energy states for MoS2 stacked onto itself occurs at 0/180° interlayer orientations corresponding to the 2H and 3R polytypes with only 5 meV difference per MoS2 unit [28]. The stacking orientation preference of hBN with itself is likewise weak [29]. The orientational preference of a MoS2 overlayer on a hBN substrate is expected to be even weaker, given lattice mismatch. Indeed, density functional theory (DFT) calculations performed with three different implementations of vdW corrections (DFT- D3 [30], DFT-TS [31], and vdW-DF2 [32]) in a periodic- approximate supercell that contains a 4×4 (5×5) supercell of MoS2 (hBN) yield a 0/180° orientational preference of at most 0.5 meV per MoS2 unit (see Appendix A for details), where the stacking with reversed bond polarities (defined by elemental electronegativities, see Fig. 1a) is only slightly preferred. This near-degeneracy is not surprising, since each atom in one layer systematically samples a variety of local FIG. 1. (a) Top view of pristine MoS2 on hBN, where S atoms sample a variety of local environments, eclipsing B(oron), N(itrogen), or H(ollow) sites. (b) Stable defect pairings in a 2D heterostack are likely Frenkel pairs: an adatom in one layer (red filled) binding strongly to a vacancy in the other layer (blue empty). (c) The Moad+VB complex has the strongest defect pair binding energy (notation described in main text). (d) Formation energies of MoS2 defects isolated in a monolayer (solid lines), paired with VB (dashed), and paired with VN (dotted), as function of sulfur chemical potential and in a nitrogen-rich setting. environments in the other layer across their interface (Fig. 1a). While this energy difference can be made significant given sufficient area, the energy barrier across intermediate orientations between 0° and 60° (60° is symmetry-equivalent to 180°) also scales with area, and at a faster rate of 2 meV/MoS2 (see Appendix A), effectively trapping the growing layer at 0° or 60°. The orientation is thus likely set when the MoS2 flake is too small for the stacking energetics of its interior to overcome thermal fluctuations. Can the spatial averaging across the supercell be broken by making some specific location(s) in the flake special? Along these lines, we first consider finite-size effects -- i.e. edge effects and incomplete spatial averaging -- by examining the orientational energetics of finite sulfur- passivated MoS2 clusters, including those with areas smaller than the smallest possible coincident supercell and the smallest known MoxSy cluster Mo3S13 (Fig. S1). Even in these cases, a marginal preference of at most 2 meV per Mo was found. An intriguing orientation preference found in a recent work differs in that it used Mo6S6 clusters with unpassivated metal-terminated edges [33]. III. DISTINGUISHING INVERSION DOMAINS BY A DEFECT COMPLEX We next consider whether the spatial averaging (and the associated near-degeneracy) can be interrupted by a localized structural defect in the hBN substrate. Such defects may also act as natural nucleation sites. To find defects that can strengthen interlayer orientational coupling (i.e. correlating the polarities of hBN and MoS2 more strongly), we systematically examine three types of pairwise interactions: between a MoS2 point defect and pristine hBN, between an hBN point defect and pristine MoS2, and between point defects in both MoS2 and hBN, as tabulated in Fig. 1c. Darker colors indicate stronger pairwise binding Ebinding =!"#$%&'( -- !)*+, &'( -- !./0&'( -- Eadhesion, where Eadhesion is the pristine van der Waals interlayer adhesion, so that Ebinding = 0 for pristine MoS2 stacked on pristine hBN (top left of table). Moad, Sad, VS, MoS are respectively an Mo adatom, S adatom, S vacancy, and Mo substituting S, chosen from common MoS2 defects with formation energies below 3 eV within the experimentally accessible range of sulfur chemical potentials [34]. The(cid:1) and(cid:3)symbols indicate MoS2 defects on the sulfur plane away from or adjacent to the hBN layer. VB, VN, BN, NB, Bad, Nad, are B or N vacancies, antisite B or N (i.e. substituting N or B), and B or N adatoms respectively. We do not consider defects with higher degrees of complexity since they have higher formation energies (see Appendix B) and degrade epitaxy, as discuss later. We find the most strongly bound defects to be proximate adatom-vacancy pairs, with the 9.1 eV VB+Moad binding being by far the strongest. Such combinations are interlayer Frenkel pairs: adatom-vacancy complexes that were originally studied for H Mo S B N Mo on MoS2 N B Mo interstitial 2.90Å 2.15Å 5.05Å Mo on B vacancy in hBN FIG. 2. A Mo interstitial atom (red) between MoS2 and a VB in hBN in a 4×4/(5×5) supercell equilibrates to a 5.05 Å interlayer spacing, which is very close to the 4.96 Å spacing of pristine MoS2 on pristine hBN. The individual separations of Mo from each of these sheets in isolation also sum to essentially the same value. Thus Mo+VB on hBN can nucleate the growth of a MoS2 overlayer with surprisingly little deformation of the ideal bilayer spacing. their compliance with charge neutrality and constant stoichiometry (i.e. without electron and elemental reservoirs [35]). Frenkel pairs typically appear as low- energy defect complexes in materials with large differences in cation and anion radii (to accommodate the interstitial), where they leave no detectable remnant if they recombine. By contrast, the "interstitial" in an interlayer Frenkel pair is actually an adatom that is accommodated by the van der Waals gap, and 'recombination' of the adatom on one sheet with a vacancy in a chemically distinct sheet leaves a distinguishable defect complex, as schematically shown in Fig. 1b. Since the VB+Moad pair binds the strongest (Fig. 1c), we focus on it here and then show that its orientational control function generalizes to other defect pairs such as VN+Moad. This choice is further justified by the calculated formation energies of defect pairs [34,35] !"#$%&'( -- Epristine-MoS2/hBN -- niµi, where ni and µi are the number of i atoms added or removed from the pristine heterostack and their chemical potentials, with the usual constraint from achieving thermodynamic equilibrium with pristine sheets µMo+2µS=EMoS2 and µB+µN=EhBN. Defect pair formation energies are shown in Fig. 1d as functions of µS (referenced from the per-atom energy of solid α-S) and for µN set to the per-atom energy of N2 (the nitrogen-rich limit [36]): Among the various defects in MoS2, Moad (solid red) is the only defect that is stabilized when paired with VB (dashed red). (If X is an isolated MoS2 defect, the X+VB binding energy needs to be stronger than VB formation energy to stabilize X+VB against X [35]). We therefore exclude other defect combinations involving e.g. Sad or MoS for the present study. Defect functional calculations are also shown in Fig. S2. Even though the Moad+VB formation energy of at least 2 eV would still yield a negligible defect concentration, hBN defects should be preexisting so that the VB contribution to the formation energy need not be accounted for. The native VB in hBN before MoS2 growth are expected to be out-of-equilibrium and passivated by hydrogen, since hBN samples are synthesized from hydrides and since H-passivated VB is formation energies from hybrid ~7.7 eV more stable than VB [36], with a large migration barrier rendering them immobile below their annealing temperature of at least ~1000 -- 2000 K [36]. Thus, taking the fully passivated VB+3H complex as immobile (out of equilibrium), taking Moad as mobile (in equilibrium) with formation energy EMo, and taking their binding energy as Ebinding = EVB+3H + EMo -- 3µH -- EVB+Mo (positive for Mo replacing 3H), then following the mass action law [37], one is tempted to conclude that the percentage of VB that is exp[(Ebinding -- EMo)/kBT]. Thus combine with Moad Moad+VB pairing will approach completion as Ebinding overpowers EMo. However, this requirement on Ebinding can be alleviated. Just like defects can be immobilized by high migration barriers and become out of equilibrium [37], so can defect pairs be locked by high binding energies and become out of equilibrium. Removing each H and Mo from VB requires 2.3 -- 2.7 eV and 9.1 eV respectively, so if unbinding occurs at 1000 K, our MoS2 growth temperature, it would occur at rates of 2 -- 200 s -- 1 and 10 -- 32 s -- 1. Therefore as along as Ebinding>0, Moad will irreversibly replace H due to the much longer timescale of its unbinding. Indeed, Ebinding = 9.1 -- 7.7=1.4 eV for Moad. The earliest event in the formation of VB+Moad is presumably the binding of a Mo atom to a VB (VB are common in hBN [38]) by 9.6 eV, consistent with the reported strong binding between VB and transition metal atoms in general [39] and the strong binding of transition metal atoms to pyridinic-nitrogen defects in graphene in particular [40] (structurally similar to VB). The under- coordinated Mo atoms available in partially decomposed Mo on Boron vacancy Interstitial, eclipsed Interstitial, staggered ) V e ( y g r e n E Interstitial Mo N bonded to interstitial Mo FIG. 3. Spin-polarized DFT band structures of a Mo atom bound to a VB, and a Moad+VB complex with an eclipsed and staggered configuration; Fermi levels are set to zero. The two band structures in each panel represent the majority and minority spin channel. States localized on the interstitial Mo and its three nearest-neighbor N atoms are colored red and blue respectively. Nitrogen levels in the valence band rise in energy when eclipsed, reflecting the repulsion between the N and S atoms bonded to the interstitial Mo. MOCVD precursors such as Mo(CO)x or CVD precursors such as MoOx or MoSxOy should also bind strongly to VB (see Ref. [41] for the case of MoO3). The full growth kinetics for the nucleation of MoS2 at a VB is beyond the scope of the present study, but the most plausible such route begins with the VB-bound Mo adatom first coordinating to ambient S. Strikingly, these sulfur atoms can then form the VB+Moad interlayer Frenkel pair by incorporating directly into a MoS2 overlayer that sits above the hBN layer. Fig. 2 shows this configuration with a structurally relaxed 4×4 MoS2 on 5×5 hBN supercell: the 5.05 Å interlayer separation is very close to the 4.96 Å van der Waals separation of pristine MoS2 on hBN. Thus the Mo interstitial above VB essentially "takes up no space" in the interlayer gallery. In a further interesting coincidence, the adatom heights of two "constituent" systems -- Mo above VB+hBN (2.15 Å) and Mo above pristine MoS2 (2.90 Å) -- sum to nearly the same value. The energetic comparisons between the 0/180º stacking described earlier are now reexamined -- now including a VB+Moad complex -- with very different results. The orientation where the three sulfur atoms and three nitrogen atoms nearest to the Mo interstitial are staggered is strongly favored, by 0.88 eV per Mo interstitial, over the opposite orientation where they are eclipsed (Fig. 3). A similar 0.5 0.4 0.3 0.2 0.1 0.0 DEHeVL Ê Ê Ê Ï ‡ Ú 0 ‡ Ï Ú Ú ‡ Ï 10 Triangle centered on: Ê Ï Ú ‡ Ê Ï Ú ‡ Ê Ï Ú ‡ Ê Ú ‡ Ï Ê Ú Ï ‡ 20 30 Angle q Ê Boron vacancy + Mo ‡ Boron Ï Nitrogen Ú Hollow Ê Ú Ï ‡ 40 Ê ‡ Ï Ú ‡ Ï Ê Ú 50 Ï Ú ‡ Ê Ï Ú ‡ Ê 60 Angle θ FIG. 4. Total energies (relative to the ground state for each case) of finite MoS2 flakes on monolayer hBN with a boron vacancy and Mo interstitial (black, connected plot). The weaker stacking energy variations without VB+Moad are shown in the scattered plots in color, where the center of the flake lies above a B (red squares), N (blue diamonds), or hollow site (green triangles) during rotation. preference is well-known in the conformational isomers of molecules such as ethane [42]. This defect-mediated orientational preference appears to be generic, as we also found substantial (~0.5 eV) orientational preferences for VN+Moad and other defect-pair structures (see Fig. S3). Finally, to demonstrate the absence of local minima at other intermediate orientations and the robustness of this orientation preference against edge effects, we examined finite MoS2 triangles on hBN with interstitial VB+Moad at the centers and again found a substantial preference of ~0.5 eV, as shown by the connected black dots in Fig. 4 (details are discussed in the SM). The much weaker variation in stacking energy of the same flake on hBN without VB+Moad is shown in the scattered plots, where the center of the flake lies above a B (red squares), N (blue diamonds), or hollow site (green triangles) as the flake is rotated. Electronic structure calculations reveal the origin of the strong binding of VB+Moad and its mechanism of orientation control: the interlayer Frenkel pair is also a donor-acceptor pair. A VB accepts three electrons from a transition metal (e.g. Mo) upon adsorption [39], leaving three degenerate occupied Mo d orbitals within the band gap, as shown by the occupied red bands in Fig. 3 (the two columns in each panel are for the majority and minority spin channel). When a MoS2 layer is added, these mid-gap states split differently for the two stacking orientations, but with similar summed band energies. In contrast, the eigenvalues for the orbitals of the nitrogen atoms bonded to the Mo interstitial lie much higher for the eclipsed geometry, due to the repulsion (with possible electrostatic and steric contributions [42,43]) from the sulfur above (blue bands in Fig. 3). This effect has been verified with hybrid functional calculations (Fig. S4), which generally provide more accurate formation energies [34,35]. The orientation preference does not extend to bilayer MoS2 with a Mo interstitial, which does not charge transfer to either sheet. IV. GROWTH EXPERIMENTS ON PRISTINE AND positions and defect level PLASMA-TREATED HEXAGONAL BN Taken in total, these results demonstrate how VB+Moad and similar defects could induce epitaxial growth of MoS2 with full orientation control. Is this mechanism borne out by experiment? To this end, MoS2 was grown on freestanding hBN (on a TEM grid) as well as on Si/SiO2-supported hBN using a PVT growth protocol that prioritizes the initial heterogeneous nucleation of metal species at the boron vacancy sites (see Ref. [41] and Fig. S5 for details). Raman and photoluminescence spectroscopy of this MoS2 grown on hBN are similar to those of free-standing MoS2, verifying the quality of the hBN substrate (Fig. S6, in contrast to MoS2 on Si/SiO2). Within the triangular MoS2 flakes revealed by scanning electron microscopy in Fig. 5a (with more images in Fig. S8), ~90% have a single, consistent orientation in the upper region of the hBN Ref. [38] for a demonstration of the drastic decrease in vacancy visibility when layer number increases from one to four). While interstitial metal atoms may be more reliably imaged (as reported elsewhere for the WSe2/hBN system [44]), the defect-mediated orientational control mechanism described here can be tested to a certain degree by establishing that only isolated point defects support full orientation control of MoS2, i.e. more geometrically complex defects in hBN such as multivacancy voids or step edges should not facilitate orientational epitaxy. To test this hypothesis, a population of vacancies was introduced through a pre-growth reactive ion etching of suspended hBN films for 0, 10 or 30 seconds [41]. MoS2 flakes were then grown on these plasma-treated hBN substrates with identical precursors, growth temperatures, and growth times. ADF-STEM imaging (Fig. 6a-c) along with SAED (Fig. 6d- f) reveal that plasma treatment increases the total number of MoS2 flakes (likely due to a higher density of nucleation sites) while losing epitaxy, as quantified by the histograms of MoS2 misorientation angles with respect to hBN in Fig. 6g-i (see also Fig. S11). High-resolution electron microscopy images (Figs. 6j-l) confirm that ion-irradiated substrate. The 0°/180° stacking degeneracy is nearly fully lifted. Such flakes can merge into a monolayer film nearly free of inversion domain boundaries, as suggested by ADF- STEM images in Fig. S9. A correlation between triangle orientation and the hBN surface polarity is also the most parsimonious explanation for the observed reversal of triangle orientation across a step edge in the h-BN substrate (dashed line in Fig. 5a), noting that the layer polarity of AA′-stacked hBN reverses across an odd-layer number step edge. Although a direct measurement of step height is not available due to its coverage by multilayer MoS2 and measurement uncertainty in estimating bulk hBN thicknesses, any other explanation for this reversal would require that an alternative property not related to lattice polarity both change across the step edge and also control the lattice polarity of the MoS2 flakes. The possibility that the observed orientation inversion reflects an inversion of the thermodynamic or kinetic Wulff shape is also unlikely since step edges do not interrupt Wulff shapes (except for possibly truncating corners) and also since it would imply abrupt spatial changes in the growth conditions, which vary continuously on millimeter length scales. In contrast to the clear orientation preference on hBN, second-layer MoS2 flakes stacked on the first-layer MoS2 film (lower right of Fig. 5a) lack preferred alignment. The bright-field TEM image and corresponding selected-area electron diffraction (SAED, Fig. 5b) confirm a precise alignment of parallel zigzag edges between hBN and MoS2 (see Fig. S10 for additional characterization). Unlike in Fig. 5a, both 0/180º orientations are seen in Fig. 5b because growth occurred on both sides of free-standing hBN. Direct imaging of single isolated boron vacancies in multilayer hBN substrates that are covered by MoS2 is not feasible because each imaged hBN lattice site is actually a full atomic column due to the bulk hBN AA′ stacking (see a 90% 83% b 500nm 5 μm FIG. 5. (a) SEM image of triangular MoS2 flakes on hBN. An hBN step edge separates two regions, each with 83% or 90% of the flakes at the same orientation. Inset shows the same image color-coded by orientation. (b) TEM image of triangular MoS2 flakes grown on freestanding hBN where its crystallinity and alignment with the hBN substrate are verified by the selected area electron diffraction from the circled area. FIG. 6. Effect of different reactive ion etching time. (a-c) ADF- STEM images of as-grown MoS2/hBN heterostructures after 0, 10 or 30 seconds of etching, whose degree of epitaxy is examined by (d-f) selected area diffraction, yielding (g-i) histograms of MoS2 misorientation angles with respect to hBN. Corresponding (j-l) HRTEM images of hBN substrates for different etching times show more complex defect structures in the etched films. including many hBN contains a much higher defect population with higher complexity, larger-scale voids and associated step edges, consistent with the loss of the stagger/eclipse mechanism these more geometrically complex defects. Fresh hBN step edges created by etching should significantly promote the growth of MoS2 flakes with random orientations, as suggested by the observed random orientations of MoS2 flakes grown at pre-existing step edges (from the hBN sample without plasma treatment, Fig. S12). around V. CONCLUSION interlayer The present work demonstrates that, although vacancies in a crystal are an obvious degradation of translational order, their spatially "sharp" physical nature and well-controlled angular structure can paradoxically enhance the sensitivity of a system to orientational order, especially during the critical stage of nucleation, by accentuating orientation- dependent interactions. Defect-assisted orientational epitaxy exploits the identical structure and orientation of a given type of point defect (e.g. VB) across a polar crystalline substrate. Even given full orientation uniformity and coalescence, translational mismatch is still a concern upon the merging of two grains. However, no such boundaries have been reported for TMDs thus far, presumably due to being outcompeted energetically by perfect stitches (see Fig. S9 and Ref. [45]). If there are no strong substrate registry effects (e.g. TMDs on hBN), the strain energy distributed deep into the flake interior across a lateral distance D from the boundary scales as D(1/D)2 =1/D, so stitching is more favorable than grain boundary formation for large D (i.e. large-enough flakes). To our knowledge, the only report of zero-tilt boundaries in a 2D material so far is for graphene on high-registry Ni [46]. Even misoriented grains almost always stitch together tightly (a chain of rhombi [9,45]), underlining film coalescence in these systems. One can thus envision defect- enhanced epitaxy (also possibly seed molecules [17]) as providing a general means to promote well-oriented layer- by-layer growth of 2D heterostructures. These insights into the atomistic mechanisms of orientation control can help guide further improvements to film crystallinity, as has been recently achieved in the growth of WSe2 on hBN using MOCVD with a strong suppression of inversion domains [44]. For example, introducing transition metal precursors of the same kind as the parent film can minimize the trapping of competing precursors that may otherwise 'poison' substrate vacancies. Coalescence techniques [47] can then be combined with orientational control to achieve monocrystallinity. into dense mirror boundaries the propensity for ACKNOWLEDGEMENTS F. Zhang and N. Alem acknowledge support from NSF under EFRI 2-DARE Grant 1433378. Y. Wang and V. H. Crespi acknowledge the National Science Foundation Materials Innovation Platform Two-Dimensional Crystal Consortium under DMR-1539916 and XSEDE (TG- DMR170050) for the allocation of computational resources on the LSU superMIC cluster. We also gratefully acknowledge the Center for 2-Dimensional and Layered Materials (2DLM) at the Pennsylvania State University. P. Moradifar and N. Alem acknowledge support from MRSEC under NSF DMR-1420620. The authors are grateful to Professor Joan Redwing for use of the APCVD system. APPENDIX A: ORIENTATION PREFERENCE OF PRISTINE PERIODIC STRUCTURES In 5×5 hBN + 4×4 MoS2 supercells, the relative energies of different stacking orientations and translations are calculated with three implementations of vdW corrections and are shown in Fig. 7. The three implementations agree that energies are not sensitive to translation (as shown by the clustering of the dots at 0° and 60° respectively), while the orientation preference increases from 0.1 meV (per MoS2) for vdW-DF2 to 0.3 meV for DFT-D3, and to 0.5 meV for DFT-TS. Alternatively, a √21×√21 h-BN supercell (5a+1b) and a √13×√13 MoS2 supercell (4A+1B) can be used to construct a heterostructure with strain less than 1% [48], where a and b are the lattice vectors for h-BN and A and B are for MoS2. Since both supercell lattice vectors (5a+1b) and (4A+1B) lie about 15° degrees away from the zigzag direction, the same heterostructure supercell can fit stacking geometries close to 0°, 30°, and 60° (more accurately, 3°, 25°, 35° and 57°). Thus the two near-ground- state stackings (3° and 57°) can be fairly compared with the two intermeditate twist angles (25° and 35°), i.e. with the remaining 1% artifical strain cancelled out when comparing relative energies. The energy difference between 3° and 57° is 0.4 meV per MoS2 unit, consistent with the estimate using 2.5 2.0 1.5 1.0 0.5 0.0 DEHmeVL 0º Mo S B N Ê ÊÊ Ê ÊÊ ‡‡‡‡‡‡ ÏÏÏ ÏÏÏ 60 Angle θ 60º Ê ÏÏÏÏÏÏ ÊÊÊ ‡‡‡‡‡‡ ÊÊ 0 20 40 Angle q 3º 25º 57º 35º and translations calculated with FIG. 7. Relative energies (per MoS2 unit) of different stacking orientations three implementations of vdW corrections. Point markers are computed with 5×5+4×4 supercells and bar markers are computed with √21×√21+√13×√13 supercells. The energy difference between the two stable stacking orientations of 0° and 60° is small compared with the barrier separating them (at intermediate twist angles). site nucleation for MoS2, with and sulfur are set to E(H2) and E(α-Sulfur). As shown in Fig. 8b, both hydrogen passivated VB and VN are favored against VBN over a wide Fermi energy range under N-rich and B-rich conditions. For unpassivated VB and VN, at least one is favored against VBN over the same range. Thus the hBN substrate likely hosts a predominate population of the most favorable monovacancy point defect, each serving as a consistent orientations. These results also reflect a strong binding between sulfur and VN into SN (similar to the highly stable ON impurity in Ref. [36]), since its +1 charged state is isoelectronic to pristine hBN. The strong S-VN binding and Mo-VB binding (see discussion in main text) are consistent with the STEM image in Ref. [44] revealing transition metal and chalcogen atoms always trapped at different sublattices of hBN. [1] H. Zeng, J. Dai, W. Yao, D. Xiao, and X. Cui, Nat. Nanotechnol. 7, 490 (2012). K. F. Mak, K. He, J. Shan, and T. F. Heinz, Nat. Nanotechnol. 7, 494 (2012). [3] W. Wu, L. Wang, Y. Li, F. Zhang, L. Lin, S. Niu, D. Chenet, X. Zhang, Y. Hao, T. F. Heinz, J. Hone, and Z. L. Wang, Nature 514, 470 (2014). H. Zhu, Y. Wang, J. Xiao, M. Liu, S. Xiong, Z. J. Wong, Z. Ye, Y. Ye, X. Yin, and X. Zhang, Nat. Nanotechnol. 10, 151 (2014). [5] M. Okada, T. Sawazaki, K. Watanabe, T. Taniguch, H. Hibino, H. Shinohara, and R. Kitaura, ACS Nano 8, 8273 (2014). A. Yan, J. Velasco, S. Kahn, K. Watanabe, T. Taniguchi, F. Wang, M. F. Crommie, and A. Zettl, Nano Lett. 15, 6324 (2015). [7] W. Auwärter, M. Muntwiler, J. Osterwalder, and T. [2] [4] [6] Greber, Surf. Sci. 545, L735 (2003). F. Orlando, P. Lacovig, L. Omiciuolo, N. G. Apostol, R. Larciprete, A. Baraldi, and S. Lizzit, ACS Nano 8, 12063 (2014). [9] W. Zhou, X. Zou, S. Najmaei, Z. Liu, Y. Shi, J. Kong, J. Lou, P. M. Ajayan, B. I. Yakobson, and J.- C. Idrobo, Nano Lett. 13, 2615 (2013). [10] K. Kang, S. Xie, L. Huang, Y. Han, P. Y. Huang, K. F. Mak, C.-J. Kim, D. Muller, and J. Park, Nature 520, 656 (2015). [11] N. V. Tarakina, S. Schreyeck, M. Luysberg, S. Grauer, C. Schumacher, G. Karczewski, K. Brunner, C. Gould, H. Buhmann, R. E. Dunin-Borkowski, and L. W. Molenkamp, Adv. Mater. Interfaces 1, 1400134 (2014). [12] D. G. Schlom and J. S. Harris, in Mol. Beam Ep. (Elsevier, 1995), pp. 505 -- 622. [13] K. I. Bolotin, F. Ghahari, M. D. Shulman, H. L. Stormer, and P. Kim, Nature 462, 196 (2009). [14] C. R. Dean, A. F. Young, P. Cadden-Zimansky, L. Wang, H. Ren, K. Watanabe, T. Taniguchi, P. Kim, N-rich L L H H B-rich [8] L 1 4 5 1 2 3 V e Fermi energy from hBN VBMHeVL Fermi energy from hBN VBMHeVL VN VN+H VN+2H L H V e 2 3 4 5 H a Electrostatic correction 0 -1 ÊÊ ‡ ‡ Ï Ï -- Ecorr (eV) -EcorrHeVL -2 -3 -4 Ê ‡ Ï Ê ‡ Ï q=±1 Ê q=±2 ‡ q=±3 -5 0.00 0.05 0.10 0.15 0.20 Ï 1 / Nsuper 1êNsuper b V e y g r e n e n o i t a m r o F y g r e n e n o i t a m r o F 12 10 8 6 4 2 0 -2 0 12 10 8 6 4 2 0 -2 0 VBN VB VB+H VB+2H VB+3H VN+3H SN V e y g r e n e n o i t a m r o F y g r e n e n o i t a m r o F 12 10 8 6 4 2 0 -2 0 12 10 8 6 4 2 0 -2 0 the 5×5+4×4 heterostructure supercell, while near-30° stackings are about 2 meV per MoS2 unit above the 0° ground state. This barrier should span over a wide range between 0° and 60° [49], implying that, if edge effects are ignored, a flake would translate and rotate on the substrate with negligible corrugation until it is trapped by a 0° or 60° stacking. APPENDIX B: VACANCY TYPES IN HEXAGONAL BN To determine whether intrinsic hBN defects with complexities higher than monovacancies need to be considered, we calculated defect formation energies of VB, VN, divacancy VBN, their various hydrogen passivated complexes, and sulfur substitution of nitrogen SN, as functions of the nitrogen chemical potential µN and the Fermi level (for charged defects) within density functional theory. Calculation methods closely follow prior studies with similar results [36,50 -- 52], where potential alignment for the correction of spurious electrostatic interactions in supercell calculations is performed following the Freysoldt- Neugebauer-Van de Walle scheme [53] as implemented in Ref. [54]; parameters for the model dielectric profile of hBN follow those of Ref. [55] where the in-plane and out- of-plane dielectric constant of the hBN slab is properly defined. The correction energies for various supercell sizes and charged states are shown in Fig. 8a where each extrapolation towards Nsuper(cid:2)(cid:4) (using the functional form of Ref. [54]) is set to zero. The final correction energies for the Nsuper×Nsuper×1=5×5×1 supercell geometry we used are +0.55, +2.18, and +4.90 eV for q=±1, ±2, and ±3. The experimentally accessible µN is limited within µN = E(N2) and µN=E(hBN) -- E(α-Boron), corresponding to N-rich and B-rich conditions. The chemical potentials for hydrogen 5 4 3 2 1 4 3 2 1 hBN VBM (eV) hBN VBM (eV) Fermi energy from 5 Fermi energy from Fermi energy from hBN VBMHeVL Fermi energy from hBN VBMHeVL FIG. 8. (a) Corrections to spurious electrostatic interactions in supercell geometries of Nsuper×Nsuper×1, for charged states of q=±1, ±2, and ±3. (b) Formation energies of VB, VN, and VBN, as a function of the Fermi level at two nitrogen chemical potentials. Both VB and VN are stabilized by H-passivation and are favored against VBN over a wide Fermi energy range. J. Hone, and K. L. Shepard, Nat. Phys. 7, 693 (2011). [15] X. Du, I. Skachko, F. Duerr, A. Luican, and E. Y. Andrei, Nature 462, 192 (2009). [16] Y. Zhang, Y.-W. Tan, H. L. Stormer, and P. Kim, Nature 438, 201 (2005). [21] [18] [19] [17] X. Ling, Y.-H. Lee, Y. Lin, W. Fang, L. Yu, M. S. Dresselhaus, and J. Kong, Nano Lett. 14, 464 (2014). S. Wang, X. Wang, and J. H. Warner, ACS Nano 9, 5246 (2015). L. Fu, Y. Sun, N. Wu, R. G. Mendes, L. Chen, Z. Xu, T. Zhang, M. H. Rümmeli, B. Rellinghaus, D. Pohl, L. Zhuang, and L. Fu, ACS Nano 10, 2063 (2016). [20] Y. Shi, W. Zhou, A.-Y. Lu, W. Fang, Y.-H. Lee, A. L. Hsu, S. M. Kim, K. K. Kim, H. Y. Yang, L.-J. Li, J.-C. Idrobo, and J. Kong, Nano Lett. 12, 2784 (2012). T. Georgiou, R. Jalil, B. D. Belle, L. Britnell, R. V. Gorbachev, S. V. Morozov, Y.-J. Kim, A. Gholinia, S. J. Haigh, O. Makarovsky, L. Eaves, L. A. Ponomarenko, A. K. Geim, K. S. Novoselov, and A. Mishchenko, Nat. Nanotechnol. 8, 100 (2012). [22] B. Hunt, J. D. Sanchez-Yamagishi, A. F. Young, M. Yankowitz, B. J. LeRoy, K. Watanabe, T. Taniguchi, P. Moon, M. Koshino, P. Jarillo-Herrero, and R. C. Ashoori, Science 340, 1427 (2013). [23] C.-H. Lee, G.-H. Lee, A. M. van der Zande, W. Chen, Y. Li, M. Han, X. Cui, G. Arefe, C. Nuckolls, T. F. Heinz, J. Guo, J. Hone, and P. Kim, Nat. Nanotechnol. 9, 676 (2014). S. Larentis, J. R. Tolsma, B. Fallahazad, D. C. Dillen, K. Kim, A. H. MacDonald, and E. Tutuc, Nano Lett. 14, 2039 (2014). [25] O. Lopez-Sanchez, E. Alarcon Llado, V. Koman, A. Fontcuberta i Morral, A. Radenovic, and A. Kis, ACS Nano 8, 3042 (2014). [26] X. Cui, G.-H. Lee, Y. D. Kim, G. Arefe, P. Y. Huang, C.-H. Lee, D. A. Chenet, X. Zhang, L. Wang, F. Ye, F. Pizzocchero, B. S. Jessen, K. Watanabe, T. Taniguchi, D. A. Muller, T. Low, P. Kim, and J. Hone, Nat. Nanotechnol. 10, 534 (2015). [27] C. R. R. Dean, A. F. F. Young, I. Meric, C. Lee, L. Wang, S. Sorgenfrei, K. Watanabe, T. Taniguchi, P. Kim, K. L. L. Shepard, and J. Hone, Nat. Nanotechnol. 5, 722 (2010). J. He, K. Hummer, and C. Franchini, Phys. Rev. B 89, 075409 (2014). [29] G. Constantinescu, A. Kuc, and T. Heine, Phys. Rev. [24] [28] Lett. 111, 036104 (2013). S. Grimme, J. Antony, S. Ehrlich, and H. Krieg, J. Chem. Phys. 132, 154104 (2010). [30] [31] A. Tkatchenko and M. Scheffler, Phys. Rev. Lett. [53] C. Freysoldt, J. Neugebauer, and C. G. Van de Walle, 102, 073005 (2009). Phys. Rev. Lett. 102, 016402 (2009). [32] K. Lee, É. D. Murray, L. Kong, B. I. Lundqvist, and D. C. Langreth, Phys. Rev. B 82, 081101 (2010). [33] D. Fu, X. Zhao, Y. Zhang, L. Li, H. Xu, A. Jang, S. I. Yoon, P. Song, S. M. Poh, T. Ren, Z. Ding, W. Fu, T. J. Shin, H. S. Shin, S. T. Pantelides, W. Zhou, and K. P. Loh, J. Am. Chem. Soc. 139, 9392 (2017). [34] H.-P. Komsa and A. V. Krasheninnikov, Phys. Rev. B 91, 125304 (2015). [35] C. Freysoldt, B. Grabowski, T. Hickel, J. Neugebauer, G. Kresse, A. Janotti, and C. G. Van De Walle, Rev. Mod. Phys. 86, 253 (2014). L. Weston, D. Wickramaratne, M. Mackoit, A. Alkauskas, and C. G. Van de Walle, Phys. Rev. B 97, 214104 (2018). [37] C. G. Van de Walle and J. Neugebauer, J. Appl. [36] Phys. 95, 3851 (2004). [38] N. Alem, R. Erni, C. Kisielowski, M. D. Rossell, W. Gannett, and A. Zettl, Phys. Rev. B 80, 155425 (2009). [39] B. Huang, H. Xiang, J. Yu, and S. H. Wei, Phys. Rev. Lett. 108, 206802 (2012). [40] Y.-C. Lin, P.-Y. Teng, C.-H. Yeh, M. Koshino, P.-W. Chiu, and K. Suenaga, Nano Lett. 15, 7408 (2015). See Supplemental Materials. [41] [42] Y. Mo, Wiley Interdiscip. Rev. Comput. Mol. Sci. 1, 164 (2011). S. Liu, J. Chem. Phys. 126, 244103 (2007). [43] [44] X. Zhang, F. Zhang, Y. Wang, D. S. Schulman, T. Zhang, A. Bansal, N. Alem, S. Das, V. H. Crespi, M. Terrones, and J. M. Redwing, ACS Nano acsnano.8b09230 (2019). [46] [45] H. Yu, Z. Yang, L. Du, J. Zhang, J. Shi, W. Chen, P. Chen, M. Liao, J. Zhao, J. Meng, G. Wang, J. Zhu, R. Yang, D. Shi, L. Gu, and G. Zhang, Small 13, 1603005 (2017). J. Lahiri, Y. Lin, P. Bozkurt, I. I. Oleynik, and M. Batzill, Nat. Nanotechnol. 5, 326 (2010). [47] X. Zhang, T. H. Choudhury, M. Chubarov, Y. Xiang, B. Jariwala, F. Zhang, N. Alem, G.-C. Wang, J. A. Robinson, and J. M. Redwing, Nano Lett. 18, 1049 (2018). [48] H. P. Komsa and A. V. Krasheninnikov, Phys. Rev. B 88, 085318 (2013). [49] D. Dumcenco, D. Ovchinnikov, K. Marinov, P. Lazić, M. Gibertini, N. Marzari, O. L. Sanchez, Y.- C. Kung, D. Krasnozhon, M.-W. Chen, S. Bertolazzi, P. Gillet, A. Fontcuberta i Morral, A. Radenovic, and A. Kis, ACS Nano 9, 4611 (2015). S. Okada, Phys. Rev. B 80, 161404 (2009). [50] [51] W. Orellana and H. Chacham, Phys. Rev. B 63, [52] B. Huang and H. Lee, Phys. Rev. B 86, 245406 125205 (2001). (2012). [79] W. Yang, G. Chen, Z. Shi, C.-C. Liu, L. Zhang, G. Xie, M. Cheng, D. Wang, R. Yang, D. Shi, K. Watanabe, T. Taniguchi, Y. Yao, Y. Zhang, and G. Zhang, Nat. Mater. 12, 792 (2013). [80] A. Summerfield, A. Davies, T. S. Cheng, V. V. Korolkov, Y. Cho, C. J. Mellor, C. T. Foxon, A. N. Khlobystov, K. Watanabe, T. Taniguchi, L. Eaves, S. V. Novikov, and P. H. Beton, Sci. Rep. 6, 22440 (2016). J. Wu, B. Wang, Y. Wei, R. Yang, and M. Dresselhaus, Mater. Res. Lett. 1, 200 (2013). J. A. Oliveira, W. B. De Almeida, and H. A. Duarte, Chem. Phys. Lett. 372, 650 (2003). [82] [81] [54] M. H. Naik and M. Jain, Comput. Phys. Commun. 13261 (2013). 226, 114 (2018). [56] [55] H. P. Komsa, N. Berseneva, A. V. Krasheninnikov, and R. M. Nieminen, Phys. Rev. X 4, 031044 (2014). J. P. Perdew, K. Burke, and M. Ernzerhof, Phys. Rev. Lett. 77, 3865 (1996). J. P. Perdew, K. Burke, and M. Ernzerhof, Phys. Rev. Lett. 78, 1396 (1997). [58] D. Joubert, Phys. Rev. B 59, 1758 (1999). [59] [60] G. Kresse and J. Furthmüller, Phys. Rev. B 54, P. E. Blöchl, Phys. Rev. B 50, 17953 (1994). [57] 11169 (1996). T. Björkman, A. Gulans, A. V. Krasheninnikov, and R. M. Nieminen, J. Phys. Condens. Matter 24, 424218 (2012). T. Björkman, A. Gulans, A. V. Krasheninnikov, and R. M. Nieminen, Phys. Rev. Lett. 108, 235502 (2012). T. Bučko, S. Lebègue, J. Hafner, and J. G. Ángyán, Phys. Rev. B 87, 064110 (2013). J. Heyd, G. E. Scuseria, and M. Ernzerhof, J. Chem. Phys. 118, 8207 (2003). J. Heyd, G. E. Scuseria, and M. Ernzerhof, J. Chem. Phys. 124, 219906 (2006). J. Kibsgaard, T. F. Jaramillo, and F. Besenbacher, Nat. Chem. 6, 248 (2014). [61] [62] [63] [64] [65] [66] [69] [70] [67] C. Attaccalite, M. Bockstedte, A. Marini, A. Rubio, and L. Wirtz, Phys. Rev. B 83, 144115 (2011). [68] F. Zhang, C. Erb, L. Runkle, X. Zhang, and N. Alem, Nanotechnology 29, 025602 (2018). F. Zhang, M. A. AlSaud, M. Hainey, K. Wang, J. M. Redwing, and N. Alem, Microsc. Microanal. 22, 1640 (2016). F. Zhang, K. Momeni, M. A. AlSaud, A. Azizi, M. F. Hainey, J. M. Redwing, L. Q. Chen, and N. Alem, 2D Mater. 4, 025029 (2017). [71] M. Bosi, RSC Adv. 5, 75500 (2015). [72] C. Lee, H. Yan, L. E. Brus, T. F. Heinz, J. Hone, and S. Ryu, ACS Nano 4, 2695 (2010). [73] A. Splendiani, L. Sun, Y. Zhang, T. Li, J. Kim, C. Y. Chim, G. Galli, and F. Wang, Nano Lett 10, 1271 (2010). [74] K. F. Mak, C. Lee, J. Hone, J. Shan, and T. F. Heinz, Phys. Rev. Lett. 105, 136805 (2010). [75] N. Alem, Q. M. Ramasse, C. R. Seabourne, O. V. Yazyev, K. Erickson, M. C. Sarahan, C. Kisielowski, A. J. Scott, S. G. Louie, and A. Zettl, Phys. Rev. Lett. 109, 205502 (2012). J. C. Meyer, A. Chuvilin, G. Algara-Siller, J. Biskupek, and U. Kaiser, Nano Lett. 9, 2683 (2009). [77] A. Zobelli, A. Gloter, C. P. Ewels, G. Seifert, and C. [76] Colliex, Phys. Rev. B 75, 245402 (2007). [78] M. Neek-Amal, J. Beheshtian, A. Sadeghi, K. H. Michel, and F. M. Peeters, J. Phys. Chem. C 117, Supplemental Materials for "Full orientation control of epitaxial MoS2 on hBN assisted by substrate defects" Fu Zhang1,2,4*, Yuanxi Wang2,3*, Chad Erb1,4, Ke Wang4, Parivash Moradifar1,4, Vincent Crespi1,2,3,5, Nasim Alem1,2 1 Department of Materials Science and Engineering, Pennsylvania State University, University 2 Center for Two Dimensional and Layered Materials, Pennsylvania State University, Park, PA 16802, USA. University Park, PA 16802, USA. 3 2-Dimensional Crystal Consortium, Pennsylvania State University, University Park, PA 4 Materials Research Institute, Pennsylvania State University, University Park, PA 16802, 16802, USA. USA. 5 Department of Physics, Department of Chemistry, The Pennsylvania State University, University Park, PA 16802, USA. * These authors contributed equally to this work. First-principles calculation Density functional theory calculations were performed using the Perdew-Burke- Ernzerhof parametrization of the generalized gradient approximation (GGA-PBE) exchange-correlation functional [1,2] and pseudopotentials constructed from the projector augmented wave (PAW) method [3,4], as implemented in the Vienna Ab initio Simulation Package (VASP) [5]. Van der Waals corrections were included using the DFT-D3 [6], DFT-TS [7], and vdW-DF2 [8] methods. Both DFT-D3 and DFT-TS show excellent agreement with random phase approximation treatments of the van der Waals interaction in the interlayer binding energy of bulk MoS2 (<10% error) [9 -- 11], but overbind hBN layers by 80 -- 100% [10 -- 12]. vdW-DF yields a similar binding energy for MoS2 and better binding energy for hBN [10,11]. All corrections yield excellent results for corrugation, i.e. the energy variation upon sliding adjacent layers relative to each other. Ionic relaxations were all performed at the PBE level with vdW corrections using the DFT-D3 method (unless otherwise noted, e.g. for calculations using DFT-TS and vdW-DF2) until forces were smaller than 0.01 eV/Å. Hybrid functional eigenvalues and total energies were calculated using the range-separated form of Heyd, Scuseria, and Ernzerhof (HSE06) [13,14] and using structures relaxed at the HSE06 level until forces were smaller than 0.02 eV/Å. Orientation preference: finite flakes For finite triangles on pristine hBN, not only are 0 and 60° close in energy (< 2 meV), they are in fact both slightly disfavored due to edge effects, as seen from the complete orientation map of stacking energy in Fig. 4 in the main text (colored and scattered markers). This indicates that orientation control, if determined purely by van der Waals interactions between pristine sheets, can only be effective at a later stage when the triangle is sufficiently large that the substrate adhesion scaling as L2 dominates over any edge effects, which scale as L. The near degeneracy extends down to the smallest known Mo/S cluster with a structure similar to the hexagonal motif in MoS2 [15]: for this Mo3S13 cluster (Fig. S1), the orientational preference is calculated to be only 2 meV 1 per Mo. As with the previous series of calculations with a MoS2 flake floating on h-BN, we again calculate the relative energies as the stacking orientation varies between 0 and 60°, but now in the presence of boron vacancy + Mo interstitial; these results are discussed in the main text and shown by the black connected dots in Fig. 4. Since placing Moad under a Mo site at the center of a finite triangular flake requires triangles with a side length of l=3 (too small to exclude corner effects) or l=6 (computationally too expensive, including the underlying hBN), we chose to place Moad under the hollow site with l=4. The corresponding periodic structure is in Fig. S3. This result is also the most direct test of orientation control for small transient clusters within the current scope of computational feasibility, since this is the earliest emergence of MoS2 crystallinity that persists, i.e. the minimal prerequisite for the manifestation of a well- defined "orientation selectivity". Even if a transient oxide or oxysulfide cluster with high symmetry (e.g. C3v) was computationally found to have a preferred orientation, the eventual MoS2 flake does not necessarily inherit this orientation, since bonding characters and polarities might change. Figure S1. Mo3S13 cluster on monolayer h-BN. 2 PBE N-rich PBE B-rich Defect formation energy (eV) Defect formation energy (eV) 6 5 4 3 2 1 0 6 5 4 3 2 1 0 6 5 4 3 2 1 0 6 5 4 3 2 1 0 -1.5 -1.0 -0.5 0.0 0.5 μS relative to bulk S(eV) HSE N-rich -1.5 -1.0 -0.5 0.0 μS relative to bulk S(eV) 0.5 -1.5 -1.0 -0.5 0.0 0.5 μS relative to bulk S(eV) HSE B-rich -1.5 -1.0 -0.5 0.0 0.5 μS relative to bulk S(eV) Figure S2. Defect and defect complex formation energies calculated using PBE (upper panels) and HSE06 (lower panels) in the nitrogen-rich (left) and boron-rich (right) limit. In every case the DFT-D3 method is used for van der Waals corrections. The Mo adatom is the only MoS2 defect that becomes stabilized when paired with an hBN defect (a boron vacancy). Alternative vacancy types and Mo interstitial lateral positions In contrast to a boron vacancy favoring the staggered configuration over the eclipsed one, a nitrogen vacancy (Mo+VN) favors the eclipsed configuration over the staggered one, by 0.49 eV, as shown in Fig. S3. This is not surprising since the boron atoms (with partial positive charge) closest to the Mo interstitial (a VN hybridizes with a Mo interstial rather than accept charge from it [16]) prefer proximity to the negatively charged sulfur atoms above. This preference, although seemingly opposite to that of a boron vacancy, would in fact not affect the orientation selectivity of MoS2 flakes on hBN: a VN favoring the eclipsed configuration and a VB favoring the staggered configuration on the same hBN layer orient the MoS2 overlayer the same. 3 A Mo interstial above a VB at the lateral "metal" site (i.e. under a metal atom in the MoS2 layer) was investigated in the main article; for the case where the Mo interstitial is at the lateral "hollow" site, as shown in Fig. S3, the staggered configuration is still favored, by 0.53 eV. Mo interstitials at "metal" sites are presumably more likely in reality since DFT calculations show that they are always more energetically favorable than "hollow" site Mo interstitials. Figure S3. A nitrogen vacancy favors the eclipsed configuration by 0.49 eV. A Mo interstitial at a "hollow" site favors the staggered configuration by 0.53 eV. Hybrid functional calculations Defect levels for test cases (e.g. boron and nitrogen vacancies, not shown here) are taken from Γ-point eigenvalues using the HSE06 hybrid functional [17,18] and are verified to be consistent with previously reported values [19,20]. Defect levels for the staggered and eclipsed stacking geometries discussed in the main text (Mo interstitial at the MoS2 "metal" site) are then calculated at the HSE06 level as well and compared with those calculated from PBE in Fig. S4, where the occupied defect levels are populated with spin symbols. The overall effect is an increase in band gap and defect levels shifting deeper into the mid-gap region, with no significant change in the ordering of occupied levels. The total energy of the staggered geometry calculated at the HSE06 level is lower than the eclipsed case by 0.62 eV. 4  Figure S4. Defect levels for the "eclipsed" and "staggered" stacking geometries calculated from PBE and HSE06. Occupied levels are populated with spin symbols, where blue and red indicate two spin polarizations and paired spins indicate degenerate spin states. Figure S5. (a) Schematic of PVT system. (b) Annular dark field (ADF)-STEM image of a Mo- terminated MoS2 edge. Atmosphere Pressure PVT synthesis of MoS2/hBN heterostructure The hBN flakes were mechanically exfoliated from powder (grade PT 110, Momentive Performance Materials) and placed on a Si/SiO2 substrate, before they were transferred to a Au quantifoil TEM grid using PMMA-assisted transfer [21]. The combination of mechanical exfoliation and transfer produces freestanding hBN films with high quality surfaces. Mechanically exfoliated hBN (ME-hBN) both in freestanding form and supported by a Si/SiO2 substrate (300nm SiO2 thickness) were used as templates to fabricate MoS2/hBN heterostructures. The method of synthesizing monolayer MoS2 was reported elsewhere [22]. The TEM grid was placed on a Si/SiO2 substrate which follows downstream of the hot-zone crucible. MoS2 was grown on ME-hBN by powder vapor transport at atmospheric pressure in a 15mm diameter horizontal tube furnace [23], as shown in Fig. S5. High-purity nitrogen 5 (200 sccm) was introduced into the furnace throughout the process. Approximately 1 mg of MoO3 (99.99%, Sigma Aldrich) is placed in a crucible located in the hot zone of the furnace (~700°C). Sulfur powder (300mg, 99.5% Alfa Aesar) was placed upsteam of the hot zone, at a temperature of ~230°C, while the hot zone follows a ramp to 700°C over thirty minutes, a ten-minute hold at 700°C, and cool down without feedback. Growth protocol We briefly highlight two main differences between our growth protocol and that of Yu. et al. [24], which is operationally closest to ours among existing studies but reported a mixture of MoS2 at 0 and 180º on hBN. First, while our MoO3 precursor and sulfur powder temperatures are ramped up simultaneously, Yu et al. introduced MoO3 into the cavity after S vapor filled the cavity. Exposure to S prior to Mo precursor arrival likely passivates hBN vacancies, which would disable the proposed mechanism of Mo-based orientation control. Second, Yu et. al. adopted a three-zone setup with typical temperatures of 115ºC (Sulfur), 450 -- 580ºC (MoO3 with delayed entry), and 750ºC (substrates), while we adopted a two-zone setup with higher MoO3 temperature (i.e. higher Mo precursor partial pressure). A higher Mo precursor vapor pressure would favor direct heterogeneous nucleation of MoS2 seeds on the substrate rather than the deposition of nucleated MoS2 clusters from the gas phase to the substrate [25], facilitating the proposed Mo-based seeding on hBN. Characterization by Raman spectroscopy, photoluminescence, EDS, and EELS Raman spectra (Fig. S6a) reveals a frequency difference between the in-plane E12g and out-of-plane A1g modes that varies from 19 to 24 cm -- 1 across a MoS2 flake grown on BN, reflecting a variation in number of layers that is confirmed by contrast variations in the ADF-STEM image (Fig. S7) and is consistent with previous reports for PVT- synthesized MoS2 crystals [26]. The monolayer regions show a photoluminescence peak at 1.89 eV (Fig. S6c), close to the optical band gap of freestanding exfoliated MoS2 monolayers (1.90 eV) [27], with a similar ~50 meV full width at half-maximum [28]. The sharp, intense photoluminescence similar to that of free-standing flakes verifies hBN as an excellent substrate to preserve the intrinsic properties of the 2D sheet it supports, in contrast to MoS2 grown directly on Si/SiO2 (Fig. S6b). Figure S6. Raman and PL comparison between MoS2 grown on hBN and directly on Si/SiO2. 6 Figure S7. ADF-STEM image of triangular-shaped MoS2 grown on a freestanding ME-hBN flake. Layer thickness can be identified by different contrast. The growth on the free-standing hBN can occur on both sides of the surface making both 0 and 180º degree orientations possible. Figure S8. SEM image of more areas of triangular MoS2 flakes epitaxially grown on ME-hBN on a Si/SiO2 substrate (scale bar = 2 µm). Red triangles indicate the dominant orientation. 7  Figure S9. ADF-STEM image of MoS2 flakes merging into a single-crystal monolayer film free of inversion domain boundaries. Electron energy loss spectroscopy (EELS) performed on the hBN surface reveals two characteristic peaks at the boron and nitrogen edges (Fig. S10a) which correlate with σ* sp2 bonds and π* bonds [29], establishing the hexagonal honeycomb structure of hBN. Energy-dispersive X-ray spectroscopy mapping (EDS, lower panel of Fig. S10) also confirms the chemical fingerprint of the as-grown heterostructure. EDS elemental maps for nitrogen, sulfur and molybdenum are shown, indicating a uniform distribution of N in the hBN substrate, while Mo and S are locally observed within the triangular domain. All the STEM EDS maps were collected using the superX EDS quad-detectors on Titan3. The EDS boron elemental map is inconclusive because of low signal levels due to weak X-ray generation from this low atomic number element. Figure S10. (a) Point EEL spectrum of hBN surface clearly shows the boron and nitrogen edges (point is marked by red cross in panel b). (lower panels) STEM-EDS elemental maps for nitrogen, sulfur, and molybdenum. 8 Reactive Ion Etching The hBN flakes were placed on Si/SiO2 substrate for reactive ion etching. 50 sccm of oxygen gas was introduced into a Tepla M4L Plasma Etch system operating at a pressure of 200 mTorr, with a radio frequency (13.56 MHz) power of 50W to generate plasma. Different etching times were applied to suspended hBN surfaces while fixing other plasma generation parameters. After plasma treatment, we use the same PMMA transfer method to obtain freestanding hBN substrates for the growth of MoS2/hBN heterostructures. Quantifying the degree of epitaxy upon etching With increasing etching time, the degree of epitaxy tends to decrease and the size of as- synthesized MoS2 flakes decreases (Fig. S11), since more surface defects and dangling bonds are generated that can act as active nucleation sites for growth of MoS2 flakes. This observation suggests that the relation of defect concentration and degree of orientation control is an interplay of both defect density and defect type: while the number of nucleation sites is increased upon etching, more complex defect structures (e.g. step edges at the edge of triangular voids) do not facilitate epitaxy. Moreover, whereas computational studies suggest that a single defect under a grown region can determine the stacking orientation at early stages of growth (i.e. while the flakes are still nanometer-sized), a higher density of the induced defects could impair this mechanism by interacting strongly with flake edges. Figure S11. Statistical analysis of (a) the degree of epitaxy, (b) domain size and (c) substrate coverage of the as-grown MoS2 on hBN. 9 Figure S12. ADF-STEM image of MoS2 domains at step edges (of the pristine hBN without any plasma treatment) shows random orientations due to more complex interactions with the h- BN step. Material Characterization Scanning electron microscopy was carried out on a Leo 1530 FESEM. Raman/PL characterization was conducted on a 532nm Witec confocal Raman system at an operation power of ~1 mW. AC-S/TEM imaging and spectroscopy were carried out on a FEI Titan3 60-300 microscope operating at 80kV with a monochromated gun and spherical aberration corrected lenses, providing sub-angstrom resolution. A high angle annular dark field (HAADF) detector was used for the ADF-STEM imaging. The HAADF detector (Fischione) had a collection angle of 51 -- 300 mrad, a beam current of 45pA, and beam convergence angle of 30 mrad (C2 aperture of 70um) for STEM image acquisition. Imaging of surface defects was carried out at the electron dose of ~5000 e -- /Å2·s to minimize structural damage [30]. The HREM imaging condition for hBN surface defects was tuned to a negative Cs to provide white atom contrast at a slight over focus. 10 Residual strain in the hBN substrate Strain can amplify hBN's sublattice asymmetry by e.g. charge redistribution between B and N [31] and uniaxial strain in particular can further lower lattice symmetry. To test the possible role of strain in orientational epitaxy, in Fig. S13 we extended the calculations discussed in Fig. 1 in the main text and calculated the variation of stacking energies for uniformly strained and uniaxially strained (along zigzag or armchair) hBN substrates at a strain of 2% using √21×√21+√13×√13 supercells and DFT-D3 for dispersion forces. This is larger than the typical 0 -- 1% residual strain in bulk hBN reported in literature [32,33] and below the nonlinear elastic threshold [34] of 8%. Fig. S13 shows that the 0/180º near-degeneracy and the rotational barrier persist under strain. Figure S13. Relative energies (per MoS2 unit) of different stacking orientations for uniformly strained and uniaxially strained hBN substrates at a magnitude of 2%, calculated with DFT-D3 and √21×√21+√13×√13 supercells. Strong binding between VB and precursors other than an isolated Mo atom To demonstrate VB can bind to more realistic metal precursor clusters, we examine the example of MoO3 among the commonly studied MoOx clusters [35] since the suboxides have oxidation states in between the Mo case (examined in the main text) and the trioxide case (examined here). Spin-polarized DFT calculations show that a MoO3 cluster binds to hBN by 1.5 eV and to VB by 4.5 eV (Fig. S14), demonstrating the strong likelihood of VB trapping metal precursors due to its dangling bond character. The detailed structures of MoO3 are not discussed here for reasons mentioned in the previous section on finite flakes. 11  Figure S14. Relaxed structures of MoO3 on pristine hBN and on a boron vacancy. Reference [1] [2] [3] D. Joubert, Phys. Rev. B 59, 1758 (1999). [4] [5] G. Kresse and J. Furthmüller, Phys. Rev. B 54, 11169 (1996). [6] P. E. Blöchl, Phys. Rev. B 50, 17953 (1994). J. P. Perdew, K. Burke, and M. Ernzerhof, Phys. Rev. Lett. 77, 3865 (1996). J. P. Perdew, K. Burke, and M. Ernzerhof, Phys. Rev. Lett. 78, 1396 (1997). S. Grimme, J. Antony, S. Ehrlich, and H. Krieg, J. Chem. Phys. 132, 154104 (2010). [7] A. Tkatchenko and M. Scheffler, Phys. Rev. Lett. 102, 073005 (2009). [8] K. Lee, É. D. Murray, L. Kong, B. I. Lundqvist, and D. C. Langreth, Phys. Rev. B 82, 081101 (2010). J. He, K. Hummer, and C. Franchini, Phys. Rev. B 89, 075409 (2014). [9] [10] T. Björkman, A. Gulans, A. V. Krasheninnikov, and R. M. Nieminen, J. Phys. Condens. Matter 24, 424218 (2012). [11] T. Björkman, A. Gulans, A. V. Krasheninnikov, and R. M. Nieminen, Phys. Rev. [12] T. Bučko, S. Lebègue, J. Hafner, and J. G. Ángyán, Phys. Rev. B 87, 064110 Lett. 108, 235502 (2012). (2013). [13] J. Heyd, G. E. Scuseria, and M. Ernzerhof, J. Chem. Phys. 118, 8207 (2003). [14] J. Heyd, G. E. Scuseria, and M. Ernzerhof, J. Chem. Phys. 124, 219906 (2006). [15] J. Kibsgaard, T. F. Jaramillo, and F. Besenbacher, Nat. Chem. 6, 248 (2014). [16] B. Huang, H. Xiang, J. Yu, and S. H. Wei, Phys. Rev. Lett. 108, 206802 (2012). [17] H.-P. Komsa and A. V. Krasheninnikov, Phys. Rev. B 91, 125304 (2015). [18] C. Freysoldt, B. Grabowski, T. Hickel, J. Neugebauer, G. Kresse, A. Janotti, and C. G. Van De Walle, Rev. Mod. Phys. 86, 253 (2014). [19] C. Attaccalite, M. Bockstedte, A. Marini, A. Rubio, and L. Wirtz, Phys. Rev. B 83, 144115 (2011). 12 [20] W. Orellana and H. Chacham, Phys. Rev. B 63, 125205 (2001). [21] F. Zhang, C. Erb, L. Runkle, X. Zhang, and N. Alem, Nanotechnology 29, 025602 (2018). [22] F. Zhang, M. A. AlSaud, M. Hainey, K. Wang, J. M. Redwing, and N. Alem, Microsc. Microanal. 22, 1640 (2016). [23] F. Zhang, K. Momeni, M. A. AlSaud, A. Azizi, M. F. Hainey, J. M. Redwing, L. Q. Chen, and N. Alem, 2D Mater. 4, 025029 (2017). [24] H. Yu, Z. Yang, L. Du, J. Zhang, J. Shi, W. Chen, P. Chen, M. Liao, J. Zhao, J. Meng, G. Wang, J. Zhu, R. Yang, D. Shi, L. Gu, and G. Zhang, Small 13, 1603005 (2017). [25] M. Bosi, RSC Adv. 5, 75500 (2015). [26] C. Lee, H. Yan, L. E. Brus, T. F. Heinz, J. Hone, and S. Ryu, ACS Nano 4, 2695 [27] A. Splendiani, L. Sun, Y. Zhang, T. Li, J. Kim, C. Y. Chim, G. Galli, and F. Wang, Nano Lett 10, 1271 (2010). [28] K. F. Mak, C. Lee, J. Hone, J. Shan, and T. F. Heinz, Phys. Rev. Lett. 105, 136805 (2010). (2010). [29] N. Alem, Q. M. Ramasse, C. R. Seabourne, O. V. Yazyev, K. Erickson, M. C. Sarahan, C. Kisielowski, A. J. Scott, S. G. Louie, and A. Zettl, Phys. Rev. Lett. 109, 205502 (2012). [30] J. C. Meyer, A. Chuvilin, G. Algara-Siller, J. Biskupek, and U. Kaiser, Nano Lett. 9, 2683 (2009). [31] M. Neek-Amal, J. Beheshtian, A. Sadeghi, K. H. Michel, and F. M. Peeters, J. Phys. Chem. C 117, 13261 (2013). [32] W. Yang, G. Chen, Z. Shi, C.-C. Liu, L. Zhang, G. Xie, M. Cheng, D. Wang, R. Yang, D. Shi, K. Watanabe, T. Taniguchi, Y. Yao, Y. Zhang, and G. Zhang, Nat. Mater. 12, 792 (2013). [33] A. Summerfield, A. Davies, T. S. Cheng, V. V. Korolkov, Y. Cho, C. J. Mellor, C. T. Foxon, A. N. Khlobystov, K. Watanabe, T. Taniguchi, L. Eaves, S. V. Novikov, and P. H. Beton, Sci. Rep. 6, 22440 (2016). [34] J. Wu, B. Wang, Y. Wei, R. Yang, and M. Dresselhaus, Mater. Res. Lett. 1, 200 [35] J. A. Oliveira, W. B. De Almeida, and H. A. Duarte, Chem. Phys. Lett. 372, 650 (2013). (2003). 13 

1604.01330

2

1604

All-dielectric phase-change reconfigurable metasurface

We harness non-volatile, amorphous-crystalline transitions in the chalcogenide phase-change medium germanium antimony telluride (GST) to realize optically-switchable, all-dielectric metamaterials. Nanostructured, subwavelength-thickness films of GST present high-quality resonances that are spectrally shifted by laser-induced structural transitions, providing reflectivity and transmission switching contrast ratios of up to 5:1 (7 dB) at visible/near-infrared wavelengths selected by design.

cond-mat.mes-hall

cond-mat

All-dielectric phase-change reconfigurable metasurface Artemios Karvounis1, Behrad Gholipour1, Kevin F. MacDonald1, and Nikolay I. Zheludev1, 2 1 Optoelectronics Research Centre & Centre for Photonic Metamaterials, University of Southampton, Southampton, SO17 1BJ, UK 2 Centre for Disruptive Photonic Technologies, School of Physical and Mathematical Sciences & The Photonics Institute, Nanyang Technological University, Singapore 637371 We harness non-volatile, amorphous-crystalline transitions in the chalcogenide phase-change medium germanium antimony telluride (GST) to realize optically-switchable, all-dielectric metamaterials. Nanostructured, subwavelength-thickness films of GST present high-quality resonances that are spectrally shifted by laser-induced structural transitions, providing reflectivity and transmission switching contrast ratios of up to 5:1 (7 dB) at visible/near-infrared wavelengths selected by design. of plasmonic (noble including From their emergence as a paradigm for engineering new passive electromagnetic properties such as negative refractive index or perfect absorption, metamaterial concepts have extended rapidly to include a wealth of dynamic - switchable, tunable, reconfigurable, and nonlinear optical functionalities, typically through the hybridization metal) metamaterials/surfaces with active media.1 Phase-change materials, vanadium dioxide,6-8 gallium,9 and liquid crystals10-12 have featured prominently in this evolution. We now show that the chalcogenides offer a uniquely flexible platform for the realization of non-volatile, optically-switchable all- Subwavelength-thickness dielectric metamaterials. germanium antimony telluride (GST) nano-grating metasurfaces (Fig. 1) provide high-quality (Q ≥ 20) near- infrared resonances that can be spectrally shifted by optically-induced crystallization to deliver reflection and transmission switching contrast ratios up to 5:1 (7 dB). chalcogenides,2-5 To mitigate the substantial Ohmic losses encountered in plasmonic metamaterials at optical frequencies, which compromise many applications, while also improving manufacturing process practicality and compatibility with established (opto)electronic technologies, considerable effort has been devoted of late to the realization of 'all- dielectric' metamaterials, presenting resonances based upon the excitation of Mie as opposed to plasmonic (displacement as opposed to conduction current) modes in high-index, low-loss dielectric as opposed to noble metal nanostructures. A wide range of passive all-dielectric metasurface planar optical elements for steering, splitting, filtering, focusing and variously manipulating beams have been demonstrated, very typically using silicon for visible to near-IR wavelengths.13-17 Active functionalities have been demonstrated on the basis of hybridization of a silicon metasurface with a liquid crystal,18 two photon absorption on silicon metasurfaces19, 20 and nonlinear optomechanical reconfiguration in a free-standing silicon membrane metasurface.21 By virtue of their compositionally-controlled high- index, low-loss characteristics, which extend over a broad Figure 1. All-chalcogenide nano-grating metasurface. (a) Oblique incidence and (b) cross-sectional scanning electron microscope images of a 750 nm period grating fabricated by focused ion beam milling in a 300 nm thick amorphous GST film on silica. (c) Numerically simulated distribution of the y-component of electric field in the xz plane, overlaid with arrows denoting the direction and magnitude of magnetic field, for a unit cell of such a grating at resonance [at wavelength λ = 1235 nm for P = 750 nm; slot width s = 130 nm]. spectral range from the visible to long-wave infrared, and can moreover be reversibly switched (electrically or optically) in a non-volatile fashion, the chalcogenides (binary and ternary sulphides, selenides and tellurides) provide an exceptionally adaptable material base for the realization of optically reconfigurable meta-devices. Their phase-change properties – reversible transitions between amorphous and crystalline states with markedly different optical and electronic properties – have been utilized for decades in optical data storage and more in electronic phase-change RAM.22 The recently crystalline-to-amorphous transition is a melt-quenching process initiated by a short (few ns or less), intense excitation that momentarily raises the local temperature above the melting point Tm; the amorphous-to-crystalline transition is an annealing process requiring a longer (sub- µs), lower intensity excitation to hold the material above 1 transmission and its glass transition temperature Tg (but below Tm) for a short time. The latter can also be achieved through an accumulation of sub-threshold (including fs laser pulse) excitations, facilitating reproducible 'greyscale' and neuromorphic switching modes of interest for all-optical data and image processing, and harnessed recently for direct, reversible laser writing of planar optical elements and short/mid-wave IR metamaterials in a chalcogenide thin film.23-26 Here, we demonstrate structurally engineered high- reflection quality near-infrared resonances in planar (300 nm thick) dielectric nano- grating metasurfaces of amorphous germanium antimony telluride (Ge2Sb2Te5 or GST - a widely used composition in data storage applications), and the non-volatile switching of laser-induced crystallization of the chalcogenide. We employ nano- grating array metasurface patterns of subwavelength periodicity (Fig. 1), similar to those used, for example, in demonstrations of active nanophotonic photodetectors and tunable filters.27-29 A thin (subwavelength) film of a transparent medium at normal incidence has properties of reflection and transmission dependent on its thickness and complex refractive index. Periodically structuring such a film on the subwavelength scale has the effect of introducing narrow reflection/transmission resonances via the interaction between thin film interference and grating mode.30, 31 Such structures are non-diffractive and thus behave in the far field as homogenous layers.32 In the case of anisotropic structuring, the resultant optical properties are dependent on the polarization of incident light. resonances via these GST films with a thickness of 300 nm were deposited on optically flat quartz substrates by RF sputtering (Kurt J. Lesker Nano 38). A base pressure of 5×10-5 mbar is achieved prior to deposition and high-purity argon is used as the sputtering gas (70 ccpm to strike, 37 ccpm to maintain the plasma). The substrate is held within 10K of room temperature on a rotating platen 150 mm from the target to produce low-stress amorphous films. Nano- grating metasurface patterns, with a fixed slot width s ~130 nm and periods P from 750 to 950 nm, each covering an area of approximately 20 µm × 20 µm, were etched through the GST layer by focused ion beam (FIB) milling (Fig. 1). The normal-incidence transmission and reflection characteristics of these GST nano-grating metasurfaces were subsequently quantified, for incident polarizations parallel and perpendicular to the grating lines (along the y and x directions defined in Fig. 1, or TE and TM orientations of the grating, respectively), using a microspectrophotometer (CRAIC QDI2010) with a sampling domain size of 15 µm × 15 µm. Unstructured, amorphous GST is broadly transparent in the near-IR range, with measured transmission at a thickness of 300 nm >70% between 1300 and 1800 nm (Fig. 2); absorption being <20% in this spectral range (<1% above 1500 nm). Nano-grating metasurface structures introduce pronounced resonances, with quality factors Q more than 20 (Q = λr/Δλ where λr is the resonance frequency and Δλ is the half-maximum Figure 2. (a, b) Microspectrophotometrically measured Reflection R, transmission T and absorption A [=1-{R+T}] spectra for 300 nm thick amorphous GST nano-grating metamaterials with a selection of periods P [as labelled; slot width s = 130 nm], under TE-polarized illumination, alongside spectra for the unstructured amorphous GST film. (c) Numerically simulated transmission spectra calculated using a non-dispersive GST refractive index value for each grating period [as labelled] selected to reproduce experimental resonance positions and widths. linewidth), for TE polarized light, as shown in Figs. 2a and 2b, at spectral positions directly proportional to the nano-grating period P. This behavior is replicated in finite element (Comsol MultiPhysics) numerical simulations. These assume in all cases a lossless non-dispersive refractive index of 1.46 for the silica substrate, normally incident narrowband plane wave illumination and, by virtue of periodic boundary conditions, a grating pattern of infinite extent in the xy plane. Using in the first instance ellipsometrically obtained values for the complex refractive index of the unstructured amorphous GST film (see Supplemental Material33), which has a weakly dispersive real part ~2.5 2 (decreasing slowly with increasing wavelength from 2.6 to 2.4 between 1000 and 1700 nm) and a loss coefficient <0.045 across most of the near-IR spectral range (rising below 1130 nm to reach 0.09 at 1000 nm), a good qualitative match is obtained, albeit with values of Q >30 (see Supplemental Fig. S2 33). Using instead the real and imaginary parts of GST refractive index as free parameters in the model, an improved match to the experimentally observed spectral positions and widths of nano-grating resonances is achieved with non-dispersive refractive index values for each grating period (as shown in Fig. 2c) that are marginally higher in both the real and imaginary part than the ellipsometric values at the corresponding resonant wavelength. The discrepancies between ellipsometrically derived and experimental/fitted spectra are related to manufacturing imperfections, i.e. deviations from the ideal model geometry such as slight over-milling of grating lines into the substrate, and to contamination / stoichiometric change in the GST during FIB milling, which effects some change in refractive index. to the incident angle of The simulations show that the TE resonance is associated with the excitation of anti-phased displacement currents (in the ±y direction) along the core and sides of each GST 'nanowire', and a circulating pattern of magnetic field centered within the wire as illustrated in Fig. 1c. For the orthogonal TM polarization, the spectral dispersion of reflectivity and transmission (as measured by the microspectrophotometer, which employs an objective of numerical aperture 0.28) is more complex: Resonances, again at wavelengths proportional to P, are split as a result of the structures' sensitivity in this orientation (see Supplemental Material33). By engaging the phase-change properties of GST,22 the resonances of all-chalcogenide metasurfaces can be optically switched in a non-volatile fashion. In the present case, GST nano-gratings are converted from the as- deposited amorphous phase to a crystalline state by laser excitation at a wavelength of 532 nm (selected for its strong absorption in GST). This annealing is achieved by raster-scanning the beam, with a spot diameter of ~5 μm and continuous wave intensity of ~3 mW/μm2, over the sample to bring the GST momentarily to a temperature above its glass-transition point Tg but below its melting point Tm (around 110 and 630°C respectively,34, 35 though exact values will vary with factors including film thickness, composition and density). light The resultant change in GST's complex refractive index produces a change in the spectral dispersion of the nano- grating resonances, bringing about substantial changes in the metasurface transmission and reflection, especially at wavelengths close to the resonance – absolute levels are seen to increase/decrease by as much as a factor of five (Fig. 3). An increase in the real part of the GST refractive index red-shifts metasurface resonances by approximately 150 nm, while the concomitant increase in the imaginary part of the index is primarily responsible for increasing the resonance linewidth and broadband (non-resonant) absorption, particularly at shorter wavelengths. transmission spectra for Figure 3. (a) Microspectrophotometrically measured TE-mode reflectivity and the as-deposited amorphous and laser-annealed [partially] crystalline phases of a 300 nm thick GST nano-grating metamaterial with a period P = 850 nm [slot width s = 130 nm]. (b, c) Spectral dispersion of TE- mode reflection (b) and transmission (c) switching contrast, evaluated as 10𝑙𝑜𝑔 𝐴 𝐶 where A and C are respectively the amorphous and crystalline levels, for a selection of GST nano- grating periods [as labelled]. in laser-annealed GST Repeating the above process of matching numerically modelled spectra to experimentally observed nano- grating resonance positions and widths yields (for all three grating periods, again under the non-dispersive approximation) a refractive index value of 2.85 + 0.09i for the the nano-gratings (see Supplemental Fig. S3 33). The induced index change of ~10% is substantial but yet smaller and less wavelength- dependent than may be expected on the basis of ellipsometric data for unstructured crystalline GST33. This indicates strongly is stoichiometrically modified and/or only partially crystallized,5, 36, 37 which is to be anticipated primarily as a consequence of the FIB milling process (reduction of refractive index due to irradiation,38 creation of defects and gallium implantation) and because nanostructuring unavoidably modifies the thermal properties of the film (i.e. the energy absorbed from the laser beam at a given point; the temperature achieved; and the rates of temperature increase/decrease). the nanostructured GST that Nonetheless, the refractive index change (resonance spectral shift) achieved experimentally may be considered close to optimal in that it maximizes transmission and reflectivity contrast. For example, the 1470 nm reflectivity maximum of the amorphous 850 nm period 3 grating becomes a reflectivity minimum in the (partially) crystalline ratio (Ramorphous:Rcrystalline) of 5:1 (7 dB). contrast giving state, a The reverse crystalline-to-amorphous transition is not demonstrated as part of the present study: This melt- quench process would require transient heating of the GST to a temperature above Tm, which would lead to geometric deformation and chemical degradation of the samples. Robust metasurfaces supporting reversible switching over many cycles may be realized by encapsulating the GST nanostructure in the manner of the functional chalcogenide layers within rewritable optical discs, which are located between protective layers of ZnS:SiO2. In this regard, all-chalcogenide metasurfaces hold a notable advantage over hybrid plasmonic- metal/chalcogenide metamaterials,2-5 which ultimately require similar passivation layers (and thereby inevitably sacrifice switching contrast due the separation necessarily introduced between the active chalcogenide component and the surface, i.e. optical near-field, of the plasmonic metal). All-chalcogenide metasurfaces can also offer lower insertion losses and greater ease of fabrication (via a single lithographic step in a single layer of CMOS- compatible material) than plasmonic hybrid structures. to switching of In summary, we have realized all-dielectric photonic metasurfaces using a chalcogenide phase-change material platform, and demonstrated high-contrast, non-volatile, optically-induced their near-infrared resonant reflectivity and transmission characteristics. Subwavelength (300 nm; <λ/5) films of germanium antimony telluride (GST) structured with non-diffractive sub-wavelength grating patterns, present high-quality resonances that are spectrally shifted by as much as 10% as a consequence of a laser-induced (amorphous- crystalline) structural transitions in the GST, providing switching contrast ratios of up to 5:1 (7 dB) in reflection and 1:3 (-5 dB) in transmission. Within the transparency range of the (unstructured) host medium, high switching contrast wavebands can be engineered by design, i.e. appropriate selection of metasurface pattern geometry and dimensions. GST – a material with an established industrial footprint in optical and electronic data storage, can readily be structured for telecommunications applications at 1550 nm, while other members of the extensive (sulphide, selenide and telluride) chalcogenide family may also provide similar active, all-dielectric metasurface functionality in the visible range and at infrared wavelengths out to 20 µm. platforms for offer incremental chalcogenides Among material all-dielectric metamaterials, unparalleled compositional variety (i.e. range and variability of material parameters) and non-volatile (including binary22 as well as switching functionality. A wealth of reconfigurable and self- 'flat-optic' adaptive applications including switchable/tunable bandpass filter, lens, beam deflection and optical limiting components. It is interesting to note that the GST metasurfaces' resonant reflectivity and transmission changes occur in the opposite direction to subwavelength-thickness 'greyscale'23) envisaged, may be in the unstructured chalcogenide those (e.g. crystallization increases the reflectivity of an unstructured film but decreases that of a nano-grating at resonance). The ability of a single medium to provide both high and low reflectivity/transmission (signal on and off) levels in the same phase state, such that they can be simultaneously inverted via a homogenous, sample-wide structural transition, may be of interest in image processing as well as the above metasurface optics applications. Sciences Research Council This work was supported by the Engineering and Physical [grants EP/G060363/1 and EP/M009122/1], the Samsung Advanced Institute of Technology [collaboration project IO140325-01462-01], The Royal Society, the Singapore Ministry of Education [grant MOE2011-T3-1-005], and the Singapore Agency for Science, Technology and Research (A*STAR) [SERC project 1223600007]. Following a period of embargo, the data from this paper will be available from the University of Southampton repository at http://dx.doi.org/10.5258/SOTON/392918. 1. N. I. Zheludev and Y. S. Kivshar, Nat. Mater. 11, 917-924 (2012). 2. Z. L. Sámson, K. F. MacDonald, F. De Angelis, B. Gholipour, K. J. Knight, C. C. Huang, E. Di Fabrizio, D. W. Hewak and N. I. Zheludev, Appl. Phys. Lett. 96, 143105 (2010). 3. B. Gholipour, J. Zhang, K. F. MacDonald, D. W. Hewak and N. I. Zheludev, Adv. Mater. 25, 3050-3054 (2013). 4. A. Tittl, A. K. U. Michel, M. Schäferling, X. Yin, B. Gholipour, L. Cui, M. Wuttig, T. Taubner, F. Neubrech and H. Giessen, Adv. Mater. 27, 4597–4603 (2015). 5. A. K. U. Michel, D. N. Chigrin, T. W. W. Mass, K. Schönauer, M. Salinga, M. Wuttig and T. Taubner, Nano Lett. 13, 3470−3475 (2013). 6. T. Driscoll, H.-T. Kim, B.-G. Chae, B.-J. Kim, Y.-W. Lee, N. M. Jokerst, S. Palit, D. R. Smith, M. Di Ventra and D. N. Basov, Science 325, 1518-1521 (2009). 7. M. J. Dicken, K. Aydin, I. M. Pryce, L. A. Sweatlock, E. M. Boyd, S. Walavalkar, J. Ma and H. A. Atwater, Opt. Express 17 (20), 18330-18339 (2009). 8. D. Y. Lei, K. Appavoo, F. Ligmajer, Y. Sonnefraud, R. F. Haglund Jr. and S. A. Maier, ACS Photon. 2, 1306−1313 (2015). 9. R. F. Waters, P. A. Hobson, K. F. MacDonald and N. I. Zheludev, Appl. Phys. Lett. 107, 081102 (2015). 10. S. Xiao, V. P. Drachev, A. V. Kildishev, X. Ni, U. K. Chettiar, H. K. Yuan and V. M. Shalaev, Nature 466, 735-738 (2010). 11. B. Kang, J. H. Woo, E. Choi, H. H. Lee, E. S. Kim, J. Kim, T. J. Hwang, Y. S. Park, D. H. Kim and J. W. Wu, Opt. Express 18 (16), 16492-16498 (2010). 12. O. Buchnev, J. Y. Ou, M. Kaczmarek, N. I. Zheludev and V. A. Fedotov, Opt. Express 21 (2), 1633-1638 (2013). 13. J. Zhang, K. F. MacDonald and N. I. Zheludev, Opt. Express 21 (22), 26721–26728 (2013). 14. P. Moitra, Y. Yang, Z. Anderson, I. I. Kravchenko, D. P. Briggs and J. Valentine, Nat. Photon. 7, 791-795 (2013). 15. K. E. Chong, B. Hopkins, I. Staude, A. E. Miroshnichenko, J. Dominguez, M. Decker, D. N. Neshev, I. Brener and Y. Kivshar, Small 10, 1985-1990 (2014). 16. D. Lin, P. Fan, E. Hasman and M. L. Brongersma, Science 345, 298-302 (2014). 17. A. Arbabi, Y. Horie, M. Bagheri and A. Faraon, Nat. Nanotech. 10, 937-944 (2015). 18. J. Sautter, I. Staude, M. Decker, E. Rusak, D. N. Neshev, I. Brener and Y. Kivshar, ACS Nano 9, 4308–4315 (2015). 19. M. R. Shcherbakov, P. P. Vabishchevich, A. S. Shorokhov, K. E. 4 Chong, D. Y. Choi, I. Staude, A. E. Miroshnichenko, D. N. Neshev, A. A. Fedyanin and Y. Kivshar, Nano Lett. 15, 6985−6990 (2015). 20. Y. Yang, W. Wang, A. Boulesbaa, I. I. Kravchenko, D. P. Briggs, A. Puretzky, D. Geohegan and J. Valentine, Nano Lett. 15, 7388−7393 (2015). Commun. 6, 7591 (2015). 30. S. Fan, W. Suh and J. D. Joannopoulos, J. Opt. Soc. Am. A 20, 569-572 (2003). 31. G. D'Aguanno, D. de Ceglia, N. Mattiucci and M. J. Bloemer, Opt. Lett. 36, 1984-1986 (2011). 32. P. Lalanne and J. P. Hugonin, J. Opt. Soc. Am. A 15, 1843-1851 21. A. Karvounis, J. Ou, W. Wu, K. F. MacDonald and N. I. (1998). Zheludev, Appl. Phys. Lett. 107, 191110 (2015). 22. M. Wuttig and N. Yamada, Nat. Mater. 6, 824-832 (2007). 23. Q. Wang, J. Maddock, E. T. F. Rogers, T. Roy, C. Craig, K. F. MacDonald, D. W. Hewak and N. I. Zheludev, Appl. Phys. Lett. 104, 121105 (2014). 24. Q. Wang, E. T. F. Rogers, B. Gholipour, C. M. Wang, Y. Guanghui, J. Teng and N. I. Zheludev, Nat. Photon. 10, 60-65 (2016). 33. See supplemental material at [URL will be inserted by AIP] for ellipsometric refractive index data for unstructured GST and associated simulations of amorphous GST nano-grating transmission; Experimental transmission spectra and simulated fittings for laser-annealed nano-gratings; Sample data for and computational modelling of TM-polarized illumination. 34. J. Orava, A. L. Greer, B. Gholipour, D. W. Hewak and C. E. Smith, Nat. Mater. 11 (4), 279-283 (2012). 25. C. D. Wright, Y. Liu, K. I. Kohary, M. M. Aziz and R. J. Hicken, 35. B. Gholipour, C. C. Huang and D. W. Hewak, J. Mater. Sci.: Adv. Mater. 23, 3408-3413 (2011). Mater. Electron. 26, 4763–4769 (2015). 26. D. Kuzum, R. G. D. Jeyasingh, B. Lee and H. S. P. Wong, Nano 36. B. Sa, N. Miao, J. Zhou, Z. Sun and R. Ahuja, Physical Lett. 12, 2179-2186 (2012). 27. A. Christ, S. G. Tikhodeev, N. A. Gippius, J. Kuhl and H. Giessen, Phys. Rev. Lett. 91, 183901 (2003). 28. A. Sobhani, M. W. Knight, Y. Wang, B. Zheng, N. S. King, L. V. Brown, Z. Fang, P. Nordlander and N. J. Halas, Nat. Commun. 4, 1643 (2013). 29. S. J. Kim, P. Fan, J. H. Kang and M. L. Brongersma, Nat. Chemistry Chemical Physics 12, 1585–1588 (2010). 37. T. Siegrist, P. Jost, H. Volker, M. Woda, P. Merkelbach, C. Schlockermann and M. Wuttig, Nat. Mater. 10, 202-208 (2011). 38. D. P. San-Román-Alerigi, D. H. Anjum, Y. Zhang, X. Yang, A. Benslimane, T. K. Ng, M. N. Hedhili, M. Alsunaidi and B. S. Ooi, J. Appl. Phys. 113, 044116 (2013). 5 All-dielectric phase-change reconfigurable metasurface: Supplemental Material Artemios Karvounis1, Behrad Gholipour1, Kevin F. MacDonald1, and Nikolay I. Zheludev1, 2 1 Optoelectronics Research Centre & Centre for Photonic Metamaterials, University of Southampton, Southampton, SO17 1BJ, UK 2 Centre for Disruptive Photonic Technologies, School of Physical and Mathematical Sciences & The Photonics Institute, Nanyang Technological University, Singapore 637371 Figure S1. Near-IR dispersion of complex refractive indices, from spectroscopic ellipsometry, of the as-deposited amorphous and laser-annealed crystalline phases of a 300 nm thick unstructured Ge:Sb:Te film on silica. Figure S2. Numerically simulated transmission spectra for 300 nm thick amorphous GST nano-grating metamaterials with a selection of periods [as labelled; slot width s = 130 nm], under TE-polarized illumination, calculated using ellipsometircally measured values for the [weakly dispersive] complex refractive index of unstructured amorphous GST. 1 00.20.40.60.82461000120014001600Extinction coefficient kRefractive Index nWavelength, nmnamorphkamorphncrystkcrystFigure S3. (a) Microspectrophotometrically measured transmission spectra for 300 nm thick laser-annealed [partially] crystalline GST nano-grating metamaterials with a selection of periods P [as labelled; slot width s = 130 nm], under TE-polarized illumination. (b) Corresponding numerically simulated transmission spectra calculated using a non-dispersive GST refractive index value of 2.85 + 0.09i that reproduces the experimental resonance positions and widths for all grating periods. ----------------------- TM-polarized illumination Under TM-polarized illumination nano-grating resonances are split (Supplementary Fig. S4) as a result of the structures' strong sensitivity in this orientation to the incident angle of light: The microspectrophotometer employs an objective with a numerical aperture of 0.28, thereby illuminating samples at incident angles θ Figure S4. Microspectrophotometrically measured reflection and transmission spectra for a 300 nm thick amorphous GST nano-grating metamaterial with a period P = 750 nm [slot width s = 130 nm], under TM-polarized illumination. 2 ranging from zero and ~16°. This is of little consequence to the TE mode (experimental data are reproduced well by numerical simulations assuming ideally normal incidence), but for the TM polarization spatial symmetry is broken by the slightest deviation from normal incidence, leading to the observed resonance splitting.1, 2 The TM resonances are characterized by displacement currents circulating in the xz plane (forming magnetic dipoles oriented along y) – a single symmetric loop centered within each nanowire at singularly normal incidence, θ = 0°; more complex asymmetric double-loop distributions at off-normal angles (Supplementary Fig. S5). Figure S5. Numerically simulated TM-mode transmission spectra for a 300 nm thick GST nano-grating metamaterial, with a period P = 750 nm and slot width at the lower and upper surfaces of the GST layer of 130 and 450 nm respectively, for incident angles between 0º and 8º [as labelled; vertically offset for clarity]. Field maps above show the distribution of the x- component of electric field in the xz plane for a unit cell of the metasurface at the singular normal incidence resonance [λ = 1235 nm] and the two minima [= 1145 and 1285 nm] of the split resonance for an incident angle of 8º. 1. A. Christ, S. G. Tikhodeev, N. A. Gippius, J. Kuhl and H. Giessen, Phys. Rev. Lett. 91, 183901 (2003). 2. D. L. Brundrett, E. N. Glytsis and T. K. Gaylord, Opt. Lett. 23, 700-702 (1998). 3 

1809.01381

1

1809

Charge Transport Behavior of 1D Gold Chiral Nanojunctions

Understanding the process of electron tunneling in chirality-induced single-molecule junctions is imperative for the development of nanoscale switching and artificial nanomotors. Based on the combined non-equilibrium Green functions formalism and the ground-state density functional theory, we present here the charge transport behavior of chiral gold (7,3) nanowires (NWs) in comparison with various other chiral and achiral 1D gold nanostructures as the principal leads to form stable single-molecule junctions. For sigma-saturated alkane chains, we find that the contact potential barriers vary widely with the achiral leads but not with the chiral ones, although a close resemblance exists in the tunneling constants. Lower energy gaps for single-molecule junctions with Au(7,3)NWs ensure better electronic conductance even after allowing for the low thermal loss, due mainly to the close-packed arrangements of atoms with minimum wire tension. Our first-principles quantum transport analysis further suggests that chiral Au(7,3)NWs render higher electronic conductance than chiral gold (5,3) nanotubes (NTs), once bridged by either sigma-saturated or pi-conjugated molecular moieties. It, however, turns out that asymmetricity in the characteristics of channel formation at the lead-molecule contact remains often associated with chiral Au(7,3)NWs only.

cond-mat.mes-hall

cond-mat

Charge Transport Behavior of 1D Gold Chiral Nanojunctions Talem Rebeda Roya,b, Arijit Sena,b∗ aSRM Research Institute, SRM Institute of Science and Technology, Chennai 603203, India bDepartment of Physics and Nanotechnology, SRM Institute of Science and Technology, Chennai 603203, India Abstract Understanding the process of electron tunneling in chirality-induced single- molecule junctions is imperative for the development of nanoscale switching and artificial nanomotors. Based on the combined non-equilibrium Greens functions formalism and the ground-state density functional theory, we present here the charge transport behavior of chiral gold (7,3) nanowires (NWs) in comparison with various other chiral and achiral 1D gold nanostructures as the principal leads to form stable single-molecule junctions. For σ-saturated alkane chains, we find that the contact potential barriers vary widely with the achiral leads but not with the chiral ones, although a close resemblance exists in the tunneling constants. Lower energy gaps for single-molecule junctions with Au(7,3)NWs ensure better electronic conductance even after allowing for the low thermal loss, due mainly to the close-packed arrangements of atoms with minimum wire ten- sion. Our f irst − principles quantum transport analysis further suggests that chiral Au(7,3)NWs render higher electronic conductance than chiral gold (5,3) nanotubes (NTs), once bridged by either σ-saturated or π-conjugated molecu- lar moieties. It, however, turns out that asymmetricity in the characteristics of channel formation at the lead-molecule contact remains often associated with chiral Au(7,3)NWs only. ∗Corresponding author Email address: [email protected] (Talem Rebeda Roya,b, Arijit Sena,b∗) Preprint submitted to Journal of LATEX Templates September 6, 2018 Keywords: Charge transport, Single-molecule junctions, Chiral gold nanowires, Achiral gold nanowires, Chiral gold nanotubes 1. Introduction Recent years have seen a sizeable amount of study on one-dimensional (1D) nanostructures in the form of nanotubes (NTs) and nanowires (NWs), primarily because of their potential applications in nanodevice technology [1, 2, 3, 4, 5, 6, 7, 8, 9, 10]. Among these, chiral nanostructures continue to garner inter- est due especially to their characteristic helicity that often displays the ability to control the physical properties in somewhat more flexible way. From con- formational point of view, achiral nanostructures possess mirror planes while chiral nanostructures display no mirror symmetry but have glide planes. Mag- netoelectronic properties are often found to be quite sensitive to chirality [1], apart from the band-related differences like band symmetry and band spacing between chiral and achiral carbon nanotubes (CNTs). Further, the impurity concentration and curvature effect turn out to be much stronger in chiral CNTs than in achiral ones[2]. Although a major focus has remained for quite some time on developing CNT based field effect transistors (CNT-FETs) in a robust way, metallic nanotubes and nanowires have also received much attention[4, 5, 6, 7, 8, 9] in the process. The underlying lattice for a gold nanotube is a 2D triangular network with one atom per unit cell, in contrast to a CNT that has a hexagonal framework. Another important difference is that gold nanotubes, unlike CNTs, always dis- play metalicity due to its unique electronic nature. Using a tight-binding spiral model, Yevtushenko et al [9] have demonstrated that electronic transport may get affected considerably by the interplay between chirality and nonlinearity in a chiral CNT. However, the works by Manrique et al [10] reveal that like in CNTs, chiral currents are also oscillatory functions of energy in Au(5,3)NTs, but the magnitude of chiral currents are much larger in the latter. The first experimental observation of chiral Au(5,3)NT based nanobridges was reported 2 by Oshima et al [4]. Later, Senger et al [6] predicted that such conformation would possess a magic size[5, 6] in respect of its minimum wire tension. We have recently shown that chiral Au(5,3)NTs can form stable single-molecule junctions having similar conductance as achiral Au(100)NWs[11]. In this work, we intend to address how the metallic chirality, as shown in Fig. 1, may af- fect the electronic transport in respect of achiral counterparts while forming a single-molecule junction. 2. Computational methods Several two probe junctions formed by leads of chiral as well as achiral symmetry were constructed being abridged by either a σ-saturated alkane chain of varying length or a π-conjugated benzene ring. Leads were also chosen as either gold nanotubes (AuNTs) or gold nanowires (AuNWs) having chiral or achiral character. All these leads were pre-optimized using the densiy functional theory (DFT) before the formation of respective two probe junctions. The lattice parameters along the x and y directions were maintained at a minimum of 24 A to allow for sufficient isolation during the 1D relaxation. The optimized lattice constant along the z direction in the unit cell turns out to be 4.08, 7.06, 20.73 and 28.55 A respectively for Au(100)NW, Au(111)NW, Au(5,3)NT and Au(7,3)NW based leads. The number of atoms in the respective unit cell varies from as minimum as 9 for achiral Au(100)NW to as maximum as 85 for chiral AU(7,3)NW. The molecular moieties in all these heterojunctions were optimized within a force tolerance limit of 0.05eV /A, with a basis of linear combination of atomic orbitals (LCAO), as implemented in the SIESTA package[12]. For the transport calculations, we employed the non-equilibrium Greens functions (NEGF)[13, 14, 15] formalism by utilizing non-orthogonal localized double−ζ basis sets, while core electrons were treated by norm-conserving pseu- dopotentials. In our NEGF-DFT calculations[16, 17, 18, 19], the density mesh cutoff was set to 200 Ry and the k-point sampling grid was chosen as 1×1×400. The exchange and correlation were handled within the generalized gradient ap- 3 proximation (GGA) based on the revised Perdew, Burke, and Ernzerhof (rPBE) functional[20]. The transmission probability, T (E, Vb), at a given energy (E) and bias (Vb), can be obtained from the generalized Landauer theory[21] as T (E, Vb) = T r[ΓL(E)Gr(E)ΓR(E)Ga(E)] (1) Where Gr(a) denote the retarded (advanced) Greens functions of the scattering (cid:80) region and ΓL(R) = (cid:61)(cid:80)r the coupling of the molecule,(cid:80)r L(R)(E) −(cid:80)r† L(R)(E) signify the line widths due to L(R)(E), with the left (L) and right (R) leads, as determined from the unperturbed lead Greens functions. The conductances were obtained from the transmission at the Fermi level such that G = G0 i Ti, where G0 is the conductance quantum[21] while Ti refers to the eigenvalue of the i-th eigenchannel. 3. Results and discussion From the bandstructure analyses, we know that Au(5,3)NT and Au(7,3)NW possess respectively five and six available channels[6] for conduction leading to the electronic conductance values of 5G0 and 6G0. The fact that the num- ber of helical strands in Au(7,3)NW exceeds that of Au(5,3)NT leads to better conductance in the former [see Table 1]. However, when these 1D nanostruc- tures, acting as leads, form single-molecule junctions with the same active el- ement, appreciable diversity crops up in the behavior of electron transmission. Even as the length of the alkanedithiol [ADT, HS − (CH2)N − SH] molecular chain increases, the resonant transmission peaks undergo noticeable reduction in amplitude in case of Au(5,3)NTs, working as leads (see Fig. 2). Further, Au(100)NW provides the high conductance (HC) while Au(111)NW, the low conductance (LC) values even for the increasing molecular length, which is in tune with our previous work [22]. In contrast, 1D gold chiral nanostructures, irrespective of being NTs or NWs, exhibit only HC in alkanedithiol SMJs. As Table 1 suggests, our calculated HC and LC values auger well with the available experimental data. 4 l o i h t i d e n a k l a d e t a r u t a s - σ e h t f o h t g n e l e h t s a s n o i t c n u j e l u c e l o m - e l g n i s l a r i h c d n a l a r i h c a s u o i r a v r o f e c n a t c u d n o c e h t f o y d u t s e v i t a r a p m o c A : 1 e l b a T h t i w d e t a i c o s s a , d e t s i l o s l a e r a a t a d l a t n e m i r e p x e e l b a l i a v a e h T . s p u o r g e n e l y h t e m e h t f o ) N ( r e b m u n e h t f o s m r e t n i d e s a e r c n i s i n i a h c r a l u c e l o m . s d a e l g n i e b s r e t s u l c u A l a r i h c l a r i h c A ) . p x E ( . l a C ) 0 G ( e c n a t c u d n o C W N ) 3 , 7 ( u A 2 − 3 − 0 1 × 5 . 1 0 1 × 3 . 3 T N ) 3 , 5 ( u A 2 − 0 1 × 0 . 1 W N ) 1 1 1 ( u A 3 − 0 1 × 3 . 1 ] 3 2 [ ) 3 − 0 1 × 4 . 1 ( 3 − 0 1 × 6 . 1 ] 3 2 [ ) 4 − 0 1 × 8 . 1 ( 4 − 0 1 × 0 . 2 ] 4 2 [ ) 4 − 0 1 × 0 . 3 ( 4 − 0 1 × 6 . 3 4 − 0 1 × 9 . 1 5 − 0 1 × 7 . 2 5 − 0 1 × 5 . 4 ] 3 2 [ ) 5 − 0 1 × 6 . 2 ( 5 − 0 1 × 8 . 2 ] 5 2 , 3 2 [ ) 6 − 0 1 × 7 . 5 , 6 − 0 1 × 0 . 4 ( 6 − 0 1 × 0 . 4 W N ) 0 0 1 ( u A 2 − 0 1 × 0 . 1 ] 3 2 [ ) 3 − 0 1 × 0 . 1 ( ] 3 2 [ ) 4 − 0 1 × 2 . 2 ( ] 4 2 [ ) 5 − 0 1 × 0 . 2 ( 3 − 4 − 5 − 0 1 × 0 . 1 0 1 × 7 . 1 0 1 × 5 . 2 N 4 6 8 0 1 5 From Fig. 2(a-d), we find that the transmission peak for achiral Au(100)NW based SMJs at -0.22 eV remains strong even as the length of the alkanedithiol molecular wire increases. It arises from the band edge states of Au(100)NW leading to Fano resonance through quantum interference, mediated by the through- bond as well as the through-space tunneling events[22]. However, the other peak near the Fermi level at about -0.1 eV appears in the form of Breit-Wigner res- onance due to effective coupling between the lead and the respective alkane chain. Such coupling indeed varies as the lead topology and also, the molecular conformation tend to differ. To understand this, we calculate the renormalized molecular levels (RMLs) from the diagonalization of the self-consistent Hamil- tonian projected on the molecular moiety[13]. A similar trend is observed for achiral Au(111)NW based SMJs as well at, however, different electronic en- ergies. A transmission peak would arise whenever there is an available state of similar orbital character in the leads, which can interact with the RMLs in order to yield the resonance. As exhibited in Fig. 2(e-h), the two reso- nant transmission peaks near the Fermi level for chiral Au(5,3)NT based SMJs at respectively -0.1 and -0.2 eV disappear as the number (N) of the methy- lene groups increases beyond N=6. This is because the lead-molecule coupling gets weakened abruptly with the increase in the molecular length. Contrary to it, the resonant transmission peak at about -0.06 eV, associated with chiral Au(7,3)NW based SMJs, remains prominent even up to N=10, though gets re- duced as usual in amplitude. However, the primary lead-molecule interaction in all these heterojunctions has π character, stemming from the px orbital of S linkers and the dxz orbital of Au adatoms. Contour plots in the inset of Fig. 3 for the local density of states (LDOS) at the Fermi level imply quantum tun- neling as the principal mechanism of charge transport in the chiral Au(7,3)NW system as well. The through-space tunneling strength[22, 26] begins to weaken as the tunneling length expands because of the exponential behavior in the elec- tronic conductance (G) with the increasing molecular length (L), as described by GG0 = AGexp(−βLL), where βLrefers to the tunneling decay parameter and lnAG, the intercept. From the plot of natural logarithm of electronic conduc- 6 tance versus molecular length for a set of alkanedithiol SMJs associated with 1D gold chiral and achiral nanostructures, as shown in Fig. 3, it is apparent that chiral nanoleads yield the high conductance (HC), while achiral ones out of Au(111)NW are only responsible for the low conductance (LC). As we see in Table 2, the tunneling decay parameter does not depend much on chirality though the intercepts as well as the contact resistance vary a lot. To further analyze the contact properties, we calculated the contact decay constant (βC) and also, the contact potential barrier height (φC), according to Ref. 27. It turns out that the available experimental values corroborate often to the chiral systems only, as far as the contact properties are concerned[27]. Table 2: A comparative study of the decay constant (βL), the intercept (lnAG), the contact conductance (GC ), the contact decay constant (βC ) and the contact potential barrier height (φC ) for various achiral as well as chiral single-molecule junctions, as obtained from our first-principles analysis. The available experimental data,associated with Au clusters being the leads, are also listed. Contact properties Achiral Chiral Au cluster (Exp.) Au(100)NW[22] Au(111)NW[22] Au(5,3)NT Au(7,3)NW βL(A−1) InAG GC (G0) βC (A−1) φC (eV ) 1.1 -0.10 1.11 0.023 0.001 0.9 -2.78 0.062 0.63 0.39 1.0 -0.47 0.62 0.1 0.01 0.9 -0.46 0.63 0.1 0.01 0.84±0.04[28] -0.43[28] 0.65[28] 0.115±0.035[27] 0.01[27] We further examine the behavior of charge transport in single-molecule junc- tions with respect to the benzenedithiol (BDT) molecule. As Fig. 4(a-d) shows, the electronic conductance increases manifold, in sharp contrast to alkanedithiol SMJs, irrespective of the lead chirality. Such enhancement occurs because of the channel enhancement with the availability of π-conjugated molecular orbitals. It is further evident from the respective transmission eigenstates having the iso- value of 0.4, associated with the principal resonant peaks near the Fermi level. The large width in the transmission peaks of achiral Au(100)NW based SMJs signify the onset of strong lead-molecule coupling while such coupling appears to 7 be weaker with chiral systems. From Fig. 4(c), the single resonant peak at the lowest unoccupied molecular orbital (LUMO) of the chiral Au(5,3)NT based SMJs implies the LUMO-mediated electron transmission. On the other hand, at the highest occupied molecular orbital (HOMO) of the chiral Au(7,3)NW based SMJs signifies the HOMO-mediated transmission, as indicated in Fig. 4(d). However, in all these heterojunctions, σ channel appears to be better coupled through dz2 orbital of the Au adatoms. 4. Conclusions We have demonstrated from f irst−principles the charge transport behavior in various stable single-molecule junctions, made up of chiral as well as achiral leads at the nanoscale. Apparently, there lies a considerable difference in the transmission pattern when a single molecule is trapped between leads of diverse topology in forming a break junction. While single-molecule junctions with chiral leads like Au(5,3)NTs and Au(7,3)NWs give a close resemblance to those with achiral leads like Au(100)NWs in the length dependence of electronic con- ductance, the contact potentials differ appreciably from chiral to achiral hetero- junctions and sometimes, even within achiral systems as well. For σ-saturated alkane chains, Au(7,3)NWs give rise to higher electronic conductance with asym- metric lead-molecule contact than Au(5,3)NTs that display rather symmetric- ity in the π-channel formation. On the other hand, single-molecule junctions having π-conjugated molecular moieties bridging Au(5,3)NTs render lower elec- tronic conductance out of the LUMO-mediated transmission than those linking Au(7,3)NWs as leads following the HOMO-mediated one, though with strong σ-channel coupling. Understanding the charge transport mechanism of stable single-molecule junctions having chirality features at the nanoscale leads can potentially help develop molecular switching and artificial nanomotors, much of which exist still at the laboratory level only. 8 5. Acknowledgement This work was supported by DST NanoMission, Govt. of India, via Project No. SR/NM/NS1062/2012. We are thankful to the National PARAM Super- computing Facility (NPSF), Centre for Development of Advanced Computing (C-DAC), along with SRM-HPCC, for facilitating the high-performance com- puting. References References [1] F. L. Shyu, C. C. Tsai, C. H. Lee, M. F. Lin, Magnetoelectronic properties of chiral carbon nanotubes and tori, Journal of Physics: Condensed Matter 18 (35) (2006) 8313. URL http://stacks.iop.org/0953-8984/18/i=35/a=016 [2] R. Geetha, V. Gayathri, Strong effect of chirality on doped single walled carbon nanotubes, Physica Scripta 80 (2) (2009) 025701. URL http://stacks.iop.org/1402-4896/80/i=2/a=025701 [3] P. Gao, H. Li, Q. Zhang, N. Peng, D. He, Carbon nanotube field-effect transistors functionalized with self-assembly gold nanocrystals, Nanotech- nology 21 (9) (2010) 095202. URL http://stacks.iop.org/0957-4484/21/i=9/a=095202 [4] Y. Oshima, A. Onga, K. Takayanagi, Helical gold nanotube synthesized at 150 k, Phys. Rev. Lett. 91 (2003) 205503. doi:10.1103/PhysRevLett.91. 205503. URL https://link.aps.org/doi/10.1103/PhysRevLett.91.205503 [5] E. Tosatti, S. Prestipino, S. Kostlmeier, A. D. Corso, F. D. Di Tolla, String tension and stability of magic tip-suspended nanowires, Science 291 (5502) (2001) 288 -- 290. arXiv:http://science.sciencemag.org/content/291/ 9 5502/288.full.pdf, doi:10.1126/science.291.5502.288. URL http://science.sciencemag.org/content/291/5502/288 [6] R. T. Senger, S. Dag, S. Ciraci, Chiral single-wall gold nanotubes, Phys. Rev. Lett. 93 (2004) 196807. doi:10.1103/PhysRevLett.93.196807. URL https://link.aps.org/doi/10.1103/PhysRevLett.93.196807 [7] C. J. Lambert, S. W. D. Bailey, J. Cserti, Oscillating chiral currents in nanotubes: A route to nanoscale magnetic test tubes, Phys. Rev. B 78 (2008) 233405. doi:10.1103/PhysRevB.78.233405. URL https://link.aps.org/doi/10.1103/PhysRevB.78.233405 [8] M. del Valle, C. Tejedor, G. Cuniberti, Scaling of the conductance in gold nanotubes, Phys. Rev. B 74 (2006) 045408. doi:10.1103/PhysRevB.74. 045408. URL https://link.aps.org/doi/10.1103/PhysRevB.74.045408 [9] O. M. Yevtushenko, G. Y. Slepyan, S. A. Maksimenko, A. Lakhtakia, D. A. Romanov, Nonlinear electron transport effects in a chiral carbon nanotube, Phys. Rev. Lett. 79 (1997) 1102 -- 1105. doi:10.1103/PhysRevLett.79. 1102. URL https://link.aps.org/doi/10.1103/PhysRevLett.79.1102 [10] D. Z. Manrique, J. Cserti, C. J. Lambert, Chiral currents in gold nanotubes, Phys. Rev. B 81 (2010) 073103. doi:10.1103/PhysRevB.81.073103. URL https://link.aps.org/doi/10.1103/PhysRevB.81.073103 [11] A. Sen, C.-J. Lin, C.-C. Kaun, Single-molecule conductance through chi- ral gold nanotubes, The Journal of Physical Chemistry C 117 (26) (2013) 13676 -- 13680. arXiv:http://dx.doi.org/10.1021/jp402531p, doi:10. 1021/jp402531p. URL http://dx.doi.org/10.1021/jp402531p [12] J. M. Soler, E. Artacho, J. D. Gale, A. Garca, J. Junquera, P. Ordejn, D. Snchez-Portal, The siesta method for ab initio order- n materials simu- 10 lation, Journal of Physics: Condensed Matter 14 (11) (2002) 2745. URL http://stacks.iop.org/0953-8984/14/i=11/a=302 [13] J. Taylor, H. Guo, J. Wang, Ab initio, Phys. Rev. B 63 (2001) 245407. doi:10.1103/PhysRevB.63.245407. URL https://link.aps.org/doi/10.1103/PhysRevB.63.245407 [14] M. Brandbyge, J.-L. Mozos, P. Ordej´on, J. Taylor, K. Stokbro, Density- functional method for nonequilibrium electron transport, Phys. Rev. B 65 (2002) 165401. doi:10.1103/PhysRevB.65.165401. URL https://link.aps.org/doi/10.1103/PhysRevB.65.165401 [15] N. Liu, L. Zhang, X. Chen, X. Kong, X. Zheng, H. Guo, Negative differen- tial resistance in gesi core-shell transport junctions: the role of local sp2 hy- bridization, Nanoscale 8 (2016) 16026 -- 16033. doi:10.1039/C6NR05087E. URL http://dx.doi.org/10.1039/C6NR05087E [16] X. Chen, C. K. Wong, C. A. Yuan, G. Zhang, Impact of the func- tional group on the working range of polyaniline as carbon diox- ide sensors, Sensors and Actuators B: Chemical 175 (Supplement C) (2012) 15 -- 21, selected Papers presented at Eurosensors XXV. doi:https://doi.org/10.1016/j.snb.2011.11.054. URL http://www.sciencedirect.com/science/article/pii/ S0925400511010598 [17] X. Chen, C. A. Yuan, C. K. Wong, H. Ye, S. Y. Leung, G. Zhang, Molec- ular modeling of protonic acid doping of emeraldine base polyaniline for chemical sensors, Sensors and Actuators B: Chemical 174 (Supplement C) (2012) 210 -- 216. doi:https://doi.org/10.1016/j.snb.2012.08.042. URL http://www.sciencedirect.com/science/article/pii/ S0925400512008568 [18] Y. Zhang, C. Tan, Q. Yang, H. Ye, X.-P. Chen, Arsenic phosphorus mono- layer: A promising candidate for h 2 s sensor and no degradation with 11 high sensitivity and selectivity, IEEE Electron Device Letters 38 (9) (2017) 1321 -- 1324. [19] F.-F. Hu, H.-Y. Tang, C.-J. Tan, H.-Y. Ye, X.-P. Chen, G.-Q. Zhang, Ni- trogen dioxide gas sensor based on monolayer sns: A first-principles study, IEEE Electron Device Letters 38 (7) (2017) 983 -- 986. [20] Y. Zhang, W. Yang, Comment on "generalized gradient approxima- tion made simple", Phys. Rev. Lett. 80 (1998) 890 -- 890. doi:10.1103/ PhysRevLett.80.890. URL https://link.aps.org/doi/10.1103/PhysRevLett.80.890 [21] R. Landauer, Electrical resistance of disordered one-dimensional lattices, Philosophical Magazine 21 (172) (1970) 863 -- 867. arXiv:http://dx.doi. org/10.1080/14786437008238472, doi:10.1080/14786437008238472. URL http://dx.doi.org/10.1080/14786437008238472 [22] A. Sen, C.-C. Kaun, Effect of electrode orientations on charge transport in alkanedithiol single-molecule junctions, ACS Nano 4 (11) (2010) 6404 -- 6408, pMID: 20936842. arXiv:http://dx.doi.org/10.1021/nn101840a, doi:10.1021/nn101840a. URL http://dx.doi.org/10.1021/nn101840a [23] M.-D. Fu, I.-W. P. Chen, H.-C. Lu, C.-T. Kuo, W.-H. Tseng, C.-h. Chen, Conductance of alkanediisothiocyanates: effect of headgroupelec- trode contacts, The Journal of Physical Chemistry C 111 (30) (2007) 11450 -- 11455. arXiv:http://dx.doi.org/10.1021/jp070690u, doi:10. 1021/jp070690u. URL http://dx.doi.org/10.1021/jp070690u [24] C. Li, I. Pobelov, T. Wandlowski, A. Bagrets, A. Arnold, F. Evers, Charge transport in single au -- alkanedithiol -- au junctions: coordination ge- ometries and conformational degrees of freedom, Journal of the American Chemical Society 130 (1) (2008) 318 -- 326, pMID: 18076172. arXiv:http: 12 //dx.doi.org/10.1021/ja0762386, doi:10.1021/ja0762386. URL http://dx.doi.org/10.1021/ja0762386 [25] S.-Y. Jang, P. Reddy, A. Majumdar, R. A. Segalman, Interpretation of stochastic events in single molecule conductance measurements, Nano Let- ters 6 (10) (2006) 2362 -- 2367, pMID: 17034112. arXiv:http://dx.doi. org/10.1021/nl0609495, doi:10.1021/nl0609495. URL http://dx.doi.org/10.1021/nl0609495 [26] T. A. Papadopoulos, I. M. Grace, C. J. Lambert, Control of electron trans- port through fano resonances in molecular wires, Phys. Rev. B 74 (2006) 193306. doi:10.1103/PhysRevB.74.193306. URL https://link.aps.org/doi/10.1103/PhysRevB.74.193306 [27] J. Zhou, B. Xu, Determining contact potential barrier effects on elec- tronic transport in single molecular junctions, Applied Physics Letters 99 (4) (2011) 042104. arXiv:http://dx.doi.org/10.1063/1.3615803, doi:10.1063/1.3615803. URL http://dx.doi.org/10.1063/1.3615803 [28] J. Zhou, F. Chen, B. Xu, Fabrication and electronic characterization of single molecular junction devices: A comprehensive approach, Jour- nal of the American Chemical Society 131 (30) (2009) 10439 -- 10446, pMID: 19722620. arXiv:http://dx.doi.org/10.1021/ja900989a, doi: 10.1021/ja900989a. URL http://dx.doi.org/10.1021/ja900989a 13 Figure 1: (a) The underlying 2D triangular lattice having basis vectors a1 and a2, cylindrical folding of which caters to 1D gold (m,n) chiral nanostructures with radius of (m2 + n2 − mn)1/2a1/2π. Top (left side) and lateral view (right side) for freestanding chiral structures of (b) Au(5,3)NT, (c) Au(7,3)NW, (d) Au(100)NW and (e) Au(111)NW. The contour plots for electron density of (f) Au(5,3)NT, (g) Au(7,3)NW, (h) Au(100)NW and (i) Au(111)NW with top as well as lateral view in the same order. The charge density appears to be minimal at the center of Au(5,3)NT, while it is strong in case of Au(7,3)NW, due to presence of the central strand in the latter. 14 Figure 2: The transmission profile for various alkanedithiol [ADT, HS − (CH2)N − SH] single molecule junctions, where the respective molecular moiety bridges either with (a-d) achiral leads of Au(100)NWs and Au(111)NWs or with (e-h) chiral leads of Au(5,3)NTs and Au(7,3)NWs. The molecular chain under study comprises (a,e) butanedithiol, (b,f) hex- anedithiol, (c,g) octanedithiol and (d,h) decanedithiol. The arrows indicate the first interact- ing states of the similar orbital character from the leads with the renormalized molecular levels (RMLs) of the junction molecule. The transmission eigenstates associated with the resonant peak near the Fermi level are shown for comparison with respect to different lead topology. 15 Figure 3: Logarithm of electronic conductance versus molecular length for the chiral Au(5,3)NT (orange) and Au(7,3)NW (blue) based single-molecule junctions (SMJs) compared to that for the achiral Au(100)NW (green) and Au(111)NW (magenta) based ones. The solid lines refer to the exponential decay in the electronic conductance of SMJs as a function of the molecular length. The horizontal dashed lines demonstrate the similarities in the conductance values for SMJs with different lead structures of varying length. In the inset are shown the contour plots of self-consistent local density of states (LDOS) for chiral Au(7,3)NW based SMJs with the increase in the number (N) of the methylene groups, CH2, in the alkanedithiol chain. 16 Figure 4: A comparative study of the electronic transmission behavior for various π-conjugated benzenedithiol single-molecule junctions formed by a set of (a-b) achiral and (c-d) chiral nanoleads. The transmission eigenstates having the isovalue of 0.4, associated with the prin- cipal resonant peaks near the Fermi level are also shown. 17 

1105.5791

1

1105

One nanometer thin carbon nanosheets with tunable conductivity and stiffness

We present a new route for the fabrication of ultrathin (~1 nm) carbon films and membranes, whose electrical behavior can be tuned from insulating to conducting. Self-assembled monolayers of biphenyls are cross-linked by electrons, detached from the surfaces and subsequently pyrolized. Above 1000K, the cross-linked aromatic monolayer forms a mechanically stable graphitic phase. The transition is accompanied by a drop of the sheet resistivity from ~10^8 to ~10^2 kOhm/sq and a mechanical stiffening of the nanomembranes from ~10 to ~50 GPa. The technical applicability of the nanosheets is demonstrated by incorporating them into a microscopic pressure sensor

cond-mat.mes-hall

cond-mat

1  Adv. Mater. 21 (2009) 1233-1237 DOI: 10.1002/adma.200803078 One nanometer thin carbon nanosheets with tunable conductivity and stiffness Andrey Turchanin1*, André Beyer1, Christoph Nottbohm1, Xianghui Zhang1, Rainer Stosch2, Alla Sologubenko3, Joachim Mayer3, Peter Hinze2, Thomas Weimann2, Armin Gölzhäuser1 1Fakultät für Physik, Universität Bielefeld, 33615 Bielefeld, Germany 2Physikalisch-Technische Bundesanstalt, 38116 Braunschweig, Germany 3Gemeinschaftslabor für Elektronenmikroskopie, RWTH Aachen, 52074 Aachen, Germany Key words: graphene, 2D carbon, nanomembranes, nanomechanics, nanostructures, self-assembly, nanosensors, pyrolysis E-mail: [email protected] Tel.: +49-521-1065376 Fax: +49-521-1066002   Abstract: 2  We present a new route for the fabrication of ultrathin (~1 nm) carbon films and membranes, whose electrical behavior can be tuned from insulating to conducting. Self-assembled monolayers of biphenyls are cross-linked by electrons, detached from the surfaces and subsequently pyrolized. Above 1000K, the cross-linked aromatic monolayer forms a mechanically stable graphitic phase. The transition is accompanied by a drop of the sheet resistivity from ~108 to ~102 k/sq and a mechanical stiffening of the nanomembranes from ~10 to ~50 GPa. The technical applicability of the nanosheets is demonstrated by incorporating them into a microscopic pressure sensor.   3  For centuries carbon allotropes have been identified with three-dimensional carbon phases like diamond and graphite. In the last few decades, nanoscopic zero- dimensional (fullerenes[1]) and one-dimensional (carbon nanotubes[2]) carbon allotropes generated a manifold of research due to their potential use in nanotechnology. The recent discovery of a two-dimensional carbon allotrope – graphene[3] – marks an important breakthrough in physics, since it has long been argued that free-standing atomically thin materials cannot exist at ambient conditions. The subsequent aim for novel applications of two-dimensional carbon ignited significant research efforts[4-13]. For example, it is highly desirable to have atomically thin carbon sheets with tunable electrical, mechanical, and optical properties as well as with controllable size, shape and chemical functionality. Nanoscale electronics[9], nanoelectromechanical systems (NEMS)[10], as well as nano- and biosensors[14] could particularly benefit from the incorporation of such two-dimensional carbon sheets in composite materials and devices[9]. However, methods currently used for graphene fabrication such as mechanical exfoliation of highly oriented pyrolytic graphite[3], epitaxial methods[6, 12], or reduction of graphene oxide[8], can only partly fulfill these demands. Thus, there is a great need for novel paths to two-dimensional carbon allotropes. Highly oriented pyrolytic graphite (HOPG) is the best ordered artificially made three- dimensional graphitic allotrope. It can be fabricated via pyrolysis of bulk aromatic polymers in the temperature range from 1000K to 3000K[15]. In analogy to this bulk transformation, we suggest that the pyrolysis of a molecular thin film of aromatic   4  molecules is a promising path for the generation of two-dimensional carbon. Biphenylthiols form densely packed self-assembled monolayers (SAMs) with a thickness of ~1 nm on gold surfaces[16]. Such aromatic SAMs could act as suitable precursors in a pyrolytic reaction. However, due to the low thermal stability of thiolates[17], they desorb from the surface at temperatures that are much lower (350-450K) than those required for pyrolysis. We have recently found that the temperature induced desorption of biphenylthiols on gold is inhibited[18], when biphenyls were cross-linked by electron irradiation in a molecular sheet[19]. In this report we show that vacuum pyrolysis can transform ~1 nm thick aromatic molecular nanosheets from an insulating to a conducting state. The resulting carbon nanosheets are atomically thin and mechanically stable as suspended membranes even at temperatures above 1200 K and their resistivity and stiffness are determined by the annealing temperature. To prepare carbon nanosheets (details in supporting online material (SOM)) biphenyl molecules are self-assembled on a substrate from solution and subsequently cross-linked by electron irradiation, Fig. 1A. Both size and shape of the nanosheets are determined by this initial exposure. Modern electron beam lithography and exposure tools allow the fabrication of sheets from macroscopic (cm2) down to nanometer sizes and in arbitrary shapes[20]. The nanosheets are then lifted from their surface and transferred to another solid substrate or holey structures, such as transmission electron microscope (TEM) grids, where the nanosheets become suspended free-standing nanomembranes. The transfer itself   5  is very simple. First, a polymeric transfer medium (“glue”) is applied to the sheet and the original substrate is dissolved. The hardened glue with the attached nanosheet is then placed onto another solid surface or a TEM grid and finally, the glue is dissolved (see SOM), leaving the nanosheet on the new substrate. Fig. 1B shows an optical micrograph of the section of a large (~5 cm2) carbon nanosheet that has been transferred from gold onto a silicon wafer with a ~300 nm thick silicon oxide layer. The color variation between the bare part of the silicon oxide surface and the part covered by nanosheet allows for visualization of the cross-linked biphenyl monolayer[21]. Within the nanosheet some dark lines are clearly visible. These lines are folds in the sheet that occurred during the transfer process. The thickness of the nanosheet has been determined by X-ray photoelectron spectroscopy and atomic force microscopy (AFM) to be ~1.2 nm (SOM), which is in good agreement with the height of a biphenyl molecule. Fig. 1C shows ~10 µm wide nanosheet lines that were written by electron beam lithography and then transferred onto silicon with a ~300 nm oxide layer. Compared to the cm2 sized sheet in Fig. 1B, the 10 µm wide nanosheet lines show almost no folds, indicating that small sheets have a lower tendency to wrinkle during the transfer process. Fig. 1D shows a scanning electron micrograph (SEM) of a nanosheet that has been transferred onto a TEM grid with 130x130 µm2 squared holes. The two holes on the left are covered by an intact homogeneous nanomembrane. In the upper right hole the membrane shows some folds, and in the lower right hole, the nanomembrane   6  has ruptured. Nanomembranes on TEM grids with small holes (10 µm) show very few such rupture defects, however, yield decreases with increasing hole size. We pyrolysed nanomembranes on TEM grids at temperatures from ~800K to ~1300K. Fig. 1E shows a TEM micrograph of a cross-linked biphenyl nanosheet on a gold grid that has been annealed at ~1100 K in ultra high vacuum (UHV). An intact nanosheet (with a few folds) that spans an 11x11 µm2 hole is clearly seen in Fig. 1E. Scanning Auger microscopy revealed that this suspended membrane consists only of carbon (SOM). This temperature stability is quite remarkable for a macroscopically large membrane with a thickness of only ~1 nm. In the next step, we explored the electrical properties of the heated nanosheets in suspended (membranes) and supported states (films). Fig. 2 shows the sheet resistivity as a function of the annealing temperature. The resistivity was determined at room temperature after the respective annealing steps. Nanosheets suspended on a gold grid were contacted by the tip of a scanning tunneling microscope; resistance was then determined by a two-point measurement in UHV. Fig. 2A shows a scanning electron micrograph of a tungsten tip touching a nanosheet that suspends over an 11x11 µm2 squared opening. Additional resistivity measurements were carried out under ambient conditions. To this end, nanosheets heated directly on gold substrate in UHV where transferred on silicon oxide and their sheet resistance was determined under ambient conditions by a four-point measurement (SOM). The sheet resistivity values measured in UHV and in ambient are in a very good agreement. A measurable electrical current is detected after   7  annealing at ~800 K. Here the sheet resistivity corresponds to ~108 k/sq. Upon annealing to temperatures between 800 and 1200K, we find linear current/voltage curves (Fig. 2 B,C,E). Increasing the annealing temperature to ~1200 K, drops the sheet resistivity to ~100 k/sq, which demonstrates the clear metallic nature of the film. This resistivity is only one order of magnitude higher than that of a defect free graphene monolayer[4], and ~100 times lower than the sheet resistivity of single chemically reduced graphene oxide sheets[22], which are currently most favored for mass production of graphene[8]. The structural transformations that occur upon annealing in the cross-linked aromatic monolayer were investigated by Raman spectroscopy and high resolution transmission electron microscopy (HRTEM). Again, nanosheets supported on silicon oxide substrates and films suspended on TEM grids were analyzed at room temperature after annealing. For annealing temperatures above 700 K, two peaks at ~1350 and ~ 1590 cm-1 are observed in the Raman spectrum (Fig. 3A). These bands are referred to as the so-called D- and G-peaks which are characteristic for sp2-bonded, honeycomb structured carbon allotropes[23]. Their positions, shapes and the intensity ratio I(D)/I(G) provide information about the degree of order in the carbon network[24]. At ~730K the D-peak has its maximum intensity at 1350 cm-1 while the G-peak has its maximum intensity at 1592 cm-1 and shows a shoulder at 1570 cm-1. At higher annealing temperatures, the shoulder in the G peak disappears. The band narrows and its position successively shifts to higher wave numbers reaching 1605 cm-1 at ~1200 K. Simultaneously, the ratio I(D)/I(G)   8  increases from ~0.75 to ~1 (Fig. 3B). The maximum of the D-peak almost remains at the same wave number while above ~950 K a shoulder appears at ~1180 cm-1. The observed temperature dependent changes in the Raman spectra are characteristic for a phase transition from an amorphous to a nanocrystalline carbon network[24]. Considering the thickness of the carbon nanosheet (1 nm), it is reasonable to attribute these changes to the formation of a nanosize graphene network. The Raman spectra also correlate very well with the successive decrease of the sheet resistivity for increasing annealing temperatures. The occurrence of structural ordering in the annealed sheets can be observed by HRTEM studies of suspended membranes. For non-annealed membranes, both high resolution imaging and selected area electron diffraction (SAED) show only the presence of amorphous material, Fig 3C, D. In annealed specimens, extended areas with curvy, nearly parallel fringes indicating the presence of graphitic material were found, Fig. 3E. The areas where fringes are observed, alternate with areas where they are not present. These observations clearly indicate that distinct structural changes occur in the nanosheet upon annealing and that the intrinsic properties of this two-dimensional material must vary accordingly[25, 26]. The line profiles across the regions with fringes, Fig.3E (1,2), give a periodicity of 0.35±0.03 nm, which is close to the interplanar spacing of the close-packed planes in graphite (0.342 nm). In our experiments, the scattered intensity modulations corresponding to the 0.35±0.03 nm periodicity of the fringes could not be found in the diffraction patterns from the investigated areas. Our evaluation showed that the corresponding   9  reciprocal distance still lies within the strong tails of the central beam of the diffraction pattern. However, the enlargement of the SAED pattern (Fig. 3F, SOM) taken from a much larger area as depicted in Fig.3E shows two distinct rings (marked 1 and 2) corresponding to the real space periodicities of 0.11±0.02 nm (ring 2) and 0.20±0.02 nm (ring 1). These can be interpreted as to correspond to the major indices (0-110) (1.23 Å spacing) and (1-210) (2.13 Å spacing) of highly in-plane oriented nanocrystalline graphitic sheets observed along the [0002] zone axis. The sharpness and intensity of the rings increase in specimens annealed at higher temperatures indicating progressing ordering in the nanomembrane. The structural transformation of a cross-linked aromatic monolayer is also reflected in the mechanical properties. To quantify these, we fabricated a nanomechanical pressure sensor in which the nanosheet acts as a membrane, cf. Fig. 4A. This rather simple device demonstrates the utilization of carbon nanomembranes as a nanomechanical transducer. Freely suspended nanosheets were mounted onto a sealed pressure cell, and a well defined pressure difference between both sides of the membrane was applied. The resulting membrane deflection was measured by AFM and used to determine the Young’s modulus and the residual strain of nanosheets by bulge tests[27]. Fig. 4B shows an AFM image of a nanosheet annealed at ~900 K without applied pressure. Although the membrane is pushed down ~15 nm by the tip, it remains intact. By applying a pressure of ~450 Pa to the sealed cell under the membrane, an upward deformation (bulging) occurs (Fig. 4C). This deformation is quantified by recording the AFM tip height at the membrane centre as function of the applied pressure.   10  Deformation datasets are presented in Fig. 4D which contains three successive measurement cycles. All measured data lie on one curve and no hysteresis is detectable, i.e. the deformation is elastic without any permanent change within the investigated strain range of up to 0.6 %. The absence of any hysteresis shows that the nanosheet does not slide on the silicon frame, presumably due to a sufficiently strong van der Waals interaction. Long-term stability was tested after five month, and no changes in the elastic properties could be observed. This demonstrates that the nanomembrane deflection can be utilized for pressure sensing. A model for the elastic deformation of thin membranes under tension contains two parts[27]: (i) membrane stretching leads to a cubic dependence of the pressure to the height, (ii) membrane tension at zero pressure due to residual stress results in a linear pressure-height-dependence. The experimental data fit very well to this model which is plotted in Fig. 4D. Curve fitting yields the Young’s modulus and the residual strain (see SOM). Fig. 4E shows both quantities as a function of the annealing temperature. Without thermal treatment the Young’s modulus is 12 GPa. This value is comparable to the Young’s modulus of multi-layered molecular/metallic nanocomposite membranes[28] that are thicker by an order of magnitude. Annealing leads to a systematic increase of the modulus with rising temperature up to 48 GPa at ~1000 K. This is in good agreement with an increasing graphitization, as the Young’s modulus of graphite varies from 39 GPa to 1.1 TPa[29], depending on its orientation. The formation of residual strain in the nanosheet is most likely related to structural transformations during the cross- linking process. Without annealing the nanosheet shows a residual strain of 0.8 %.   11  Annealing reduces the residual strain of the nanosheet to ~0.35 % above 800 K, which correlates with the onset of conductivity. Since nanomembranes are elastic and mechanically stable at ambient conditions they can be further utilized as sensitive diaphragms in various applications. Conducting nanomembranes may act as transducers in nanoelectromechanical systems (NEMS) and open an opportunity to build highly miniaturized pressure sensors that might eventually lead to microphones with nanometer dimensions. The possibility to chemically functionalize nanosheets by chemical lithography[30] further permits their use as highly sensitive chemical sensors that change their electromechanical characteristics upon the adsorption of distinct molecules. In conclusion, we have shown a simple method to produce ultrathin (~1 nm) conducting carbon films and membranes based on molecular self-assembly, electron irradiation and pyrolysis. Upon annealing, cross-linked aromatic monolayers undergo a transition to a mechanically stable graphitic phase. The above experiments demonstrate a plethora of applications that take advantage from the fact that size, shape and conductivity of the films and nanomembranes are easily controlled. Supporting Information: A detailed description of materials and methods, an analysis of spectroscopic and microscopic data, a description of models for evaluation of electrical and mechanical measurements.   12  Fig. 1. Fabrication scheme and micrographs of supported and suspended carbon nanosheets. (A) A ~1 nm thick self-assembled monolayer (SAM) of biphenyl molecules is irradiated by electrons. This results in a mechanically stable cross-linked SAM (nanosheet) that can be removed from the substrate and transferred onto other solid surfaces. When transferred onto transmission electron microscopy (TEM) grids, nanosheets suspend over holes. Upon heating to T>1000K in vacuum (pyrolysis), nanosheets transform into a graphitic phase. (B) Optical micrograph of the section of a ~5 cm2 nanosheet that was transferred from a gold surface to an oxidized silicon wafer (300 nm SiO2). Some folds in the large sheet are visible that originate from wrinkling during the transfer process. (C) Optical micrograph of a line pattern of 10 µm stripes of nanosheet. The pattern was fabricated by e-beam lithography and then transferred onto oxidized silicon. Note that the small lines are almost without folds. (D) Scanning electron micrograph of four 130x130 µm2 holes in a TEM grid after nanosheet (cross-linked biphenyl SAM) has been transferred onto the grid. Two left holes are covered by almost unfolded nanosheet. The upper right hole shows some folds, whereas in the lower right hole, the sheet has ruptured. (E) Transmission electron micrograph of a nanosheet transferred onto a TEM grid with 11x11 µm2 holes after pyrolysis at 1100K. The hole is uniformly covered with an intact nanosheet. Some folds within the sheet are visible. 13    Fig. 2. Room temperature resistivity of carbon nanosheets after annealing at different temperatures. (A) SEM image of the tungsten STM tip establishing an electrical and mechanical contact in the centre of the annealed biphenyl nanosheet suspended on the gold grid with ~11×11 µm2 squared openings, as employed for two-point resistivity measurements. A folded nanosheet (shown in false colour) was chosen for better visualisation. A boundary between the bare gold surface and the gold surface with a nanosheet can clearly be recognized in the lower left corner of the image. (B) and (C) Representative room temperature current vs. voltage data for two annealing temperatures in the two-point set-up of resistivity measurements in UHV. Each line corresponds to a measurement in a different window of the grid. (E) Room temperature current vs. voltage data (four-point set-up) of the nanosheets supported on a silicon oxide surface. The nanosheets were annealed on gold and then transferred to silicon oxide substrates. Each line corresponds to a measurement at a new position on the nanosheet. (D) Summary of the room temperature sheet resistivity as a function of annealing temperature. All samples were annealed for 30 min at the respective temperatures. 14    Fig. 3. Structural transformations of carbon nanosheets upon annealing. (A), Raman spectra of non-annealed biphenyl nanosheet and nanosheets annealed in vacuum on gold, transferred to silicon oxide substrates. The measurements for different annealing temperatures were performed under ambient conditions. (B), Changes in the position of the G-peak and the intensity ratio of I(D)/I(G) as a function of the annealing temperature. (C), (E) High-resolution phase contrast TEM images and SAED patterns from larger areas of non-annealed (D) and annealed (F) (~1300 K) biphenyl nanosheets. Two typical locations of line profiles (1) and (2), taken for evaluation of the periodicity of the graphitic fringes are depicted in (E). In addition, an enlargement of the diffraction (F) is given in SOM. The rings in (F) can be indexed as belonging to the (0-110) and (1-210) lattice plane spacings of highly in-plane oriented nanocrystalline graphitic sheets. 15    Fig. 4. Mechanical properties of carbon nanosheets upon annealing. (A), Schematic representation of the bulging test setup. The home-build pressure cell was mounted into an atomic force microscope (AFM) that measured the membrane deflection. AFM images of a membrane (topography, contact mode) without (B) and with (C) an applied pressure of 450 Pa. Scale bar, 10 µm. Line scans along the red lines are superimposed to the AFM images. (D), The Young’s modulus determination is presented for one representative membrane (annealed at ~900 K). First the deflection at the membranes centre is measured for different pressures and then these data are fitted by the displayed dependency which yields the modulus. (E), The Young’s modulus as function of annealing temperature. At higher temperatures the modulus shifts towards the value of graphite.   References 16  [1] H. W. Kroto, J. R. Heath, S. C. Obrien, R. F. Curl, R. E. Smalley, Nature 1985, 318, 162. [2] S. Iijima, Nature 1991, 354, 56. [3] K. S. Novoselov, A. K. Geim, S. V. Morozov, D. Jiang, Y. Zhang, S. V. Dubonos, I. V. Grigorieva, A. A. Firsov, Science 2004, 306, 666. [4] K. S. Novoselov, A. K. Geim, S. V. Morozov, D. Jiang, M. I. Katsnelson, I. V. Grigorieva, S. V. Dubonos, A. A. Firsov, Nature 2005, 438, 197. [5] Y. B. Zhang, Y. W. Tan, H. L. Stormer, P. Kim, Nature 2005, 438, 201. [6] C. Berger, Z. M. Song, X. B. Li, X. S. Wu, N. Brown, C. Naud, D. Mayo, T. B. Li, J. Hass, A. N. Marchenkov, E. H. Conrad, P. N. First, W. A. de Heer, Science 2006, 312, 1191. [7] T. Ohta, A. Bostwick, T. Seyller, K. Horn, E. Rotenberg, Science 2006, 313, 951. [8] S. Stankovich, D. A. Dikin, G. H. B. Dommett, K. M. Kohlhaas, E. J. Zimney, E. A. Stach, R. D. Piner, S. T. Nguyen, R. S. Ruoff, Nature 2006, 442, 282. [9] A. K. Geim, K. S. Novoselov, Nat. Mater. 2007, 6, 183. [10] J. S. Bunch, A. M. van der Zande, S. S. Verbridge, I. W. Frank, D. M. Tanenbaum, J. M. Parpia, H. G. Craighead, P. L. McEuen, Science 2007, 315, 490. [11] X. L. Li, X. R. Wang, L. Zhang, S. W. Lee, H. J. Dai, Science 2008, 319, 1229.   17  [12] J. Coraux, A. T. N'Diaye, C. Busse, T. Michely, Nano Lett. 2008, 8, 565. [13] M. J. Schultz, X. Zhang, S. Unarunotai, D.-Y. Khang, Q. Cao, C. Wang, C. Lei, S. MacLaren, J. A. N. T. Soares, I. Petrov, J. S. Moore, A. Rogers, PNAS 2008, 105, 7353. [14] C. C. Striemer, T. R. Gaborski, J. L. McGrath, P. M. Fauchet, Nature 2007, 445, 749. [15] T. Kyotani, N. Sonobe, A. Tomita, Nature 1988, 331, 331. [16] E. Sabatani, J. Cohenboulakia, M. Bruening, I. Rubinstein, Langmuir 1993, 9, 2974. [17] C. D. Bain, E. B. Troughton, Y. T. Tao, J. Evall, G. M. Whitesides, R. G. Nuzzo, J. Am. Chem. Soc. 1989, 111, 321. [18] A. Turchanin, M. El-Desawy, A. Gölzhäuser, Appl. Phys. Lett. 2007, 90, 053102. [19] W. Eck, A. Küller, M. Grunze, B. Völkel, A. Gölzhäuser, Adv. Mater. 2005, 17, 2583. [20] M. J. Lercel, H. G. Craighead, A. N. Parikh, K. Seshadri, D. L. Allara, Appl. Phys. Lett. 1996, 68, 1504. [21] P. Blake, E. W. Hill, A. H. C. Neto, K. S. Novoselov, D. Jiang, R. Yang, T. J. Booth, A. K. Geim, Appl. Phys. Lett. 2007, 91. [22] C. Gomez-Navarro, R. T. Weitz, A. M. Bittner, M. Scolari, A. Mews, M. Burghard, K. Kern, Nano Lett. 2007, 7, 3499. [23] W. H. Weber, R. Merlin, Eds., Raman Scattering in Materials Science, Springer, Heidelberg 2000.   18  [24] A. C. Ferrari, J. Robertson, Phys. Rev. B 2001, 64. [25] J. C. Meyer, A. K. Geim, M. I. Katsnelson, K. S. Novoselov, T. J. Booth, S. Roth, Nature 2007, 446, 60. [26] A. Fasolino, J. H. Los, M. I. Katsnelson, Nat. Mater. 2007, 6, 858. [27] J. J. Vlassak, W. D. Nix, J. Mater. Res. 1992, 7, 3242. [28] C. Y. Jiang, S. Markutsya, Y. Pikus, V. V. Tsukruk, Nat. Mater. 2004, 3, 721. [29] A. Bosak, M. Krisch, M. Mohr, J. Maultzsch, C. Thomsen, Phys. Rev. B 2007, 75. [30] A. Gölzhäuser, W. Eck, W. Geyer, V. Stadler, T. Weimann, P. Hinze, M. Grunze, Adv. Mater. 2001, 13, 806. SUPPORTING INFORMATION S1  Adv. Mater. 21 (2009) 1233-1237 DOI: 10.1002/adma.200803078 One nanometer thin carbon nanosheets with tunable conductivity and stiffness Andrey Turchanin1*, André Beyer1, Christoph Nottbohm1, Xianghui Zhang1, Rainer Stosch2, Alla Sologubenko3, Joachim Mayer3, Peter Hinze2, Thomas Weimann2, Armin Gölzhäuser1 1Fakultät für Physik, Universität Bielefeld, 33615 Bielefeld, Germany 2Physikalisch-Technische Bundesanstalt, 38116 Braunschweig, Germany 3Gemeinschaftslabor für Elektronenmikroskopie, RWTH Aachen, 52074 Aachen, Germany   E-mail: [email protected]   Materials and Methods Fabrication and transfer of nanosheets S2  For the preparation of 1,1’-biphenyl-4-thiol (BPT) SAMs, Fig. S1, we used 300 nm thermally evaporated Au on mica substrates (Georg Albert PVD-Coatings). The substrates were cleaned in a UV/ozone-cleaner (FHR), rinsed with ethanol and blown dry in a stream of nitrogen. They were then immersed in a ~10 mmol solution of BPT in dry, degassed dimethylformamide (DMF) for 72h in a sealed flask under nitrogen. Afterwards samples were rinsed with DMF and ethanol and blown dry with nitrogen. Cross-linking was achieved in high vacuum (<5*10-7 mbar) with an electron floodgun (Specs) at an electron energy of 100 eV and typical dose of 50 mC/cm2. Annealing of the cross-linked nanosheets on Au surfaces was conducted in UHV conditions in Mo sample holders with a resistive heater with the typical heating/cooling rates of ~150 K/h and the annealing time from 0.5 h to 3 h. Annealing temperature was controlled with a Ni/Ni-Cr thermocouple and two-color pyrometer (SensorTherm). Cross- linked biphenylthiol nanosheets on gold films were annealed in vacuum up to ~1200 K, however, the mica substrate is substantially damaged at temperatures above ~1000 K, leading to damage of the gold-film/nanosheet as well. In order to maximize defect free areas, the Au-film was cleaved from the mica by immersion in hydrofluoric acid (48%) for 5 min and transfered it to a clean quartz substrate. Transfer of the nanosheet after annealing follows the procedure as described below. Transfer of non-annealed and annealed nanosheets was conducted by cleaving the nanosheet from its substrate using a layer of polymethylmethacrylate (PMMA) for   stabilization. A ~500 nm thick layer of polymethylmethacrylate (PMMA) was spincoated S3  onto the sample and baked on a hotplate. The Au was cleaved from the mica by immersion in hydrofluoric acid (48%) for 5 min and etched away in an I2/KI-etch bath (~15 min). Afterwards the nanosheet/PMMA was transferred to a SiO2 substrate or TEM grid (Quantifoil, Plano) followed by dissolution of the PMMA in acetone to yield a clean nanosheet. With this method it is possible to obtain freestanding membranes of >100 µm in size. For TEM/SEM/STM measurements nanosheets were also annealed directly on TEM grids using either resistive or e-beam heaters. Spectroscopy and Microscopy X-ray photoelectron spectra were acquired with an Omicron Multiprobe spectrometer utilizing monochromatic Al K radiation under ultra high vacuum (UHV) conditions (~10- 10 mbar). Binding energies were calibrated with respect to the Au 4f7/2 peak at 84.0 eV1, resolution of the spectra corresponds to ~1eV. A constant pass energy mode of the energy analyzer was used. Auger spectroscopy and scanning electron microscopy (SEM) of the suspended membranes were conducted with a scanning Auger microscope (SAM) in connected UHV chamber (Omicron Multiscan). An electron energy of 3 kV and a constant retardation ratio mode of the energy analyser were utilized. Raman spectra were measured with a triple monochromator system Horiba Jobin-Yvon T64000 equipped with a liquid N2-cooled CCD detector and an Olympus BH2 microscope. Data was collected in back-scattering geometry with a spectral resolution of 2 cm-1 using 514.5 nm line of an Ar+ laser (as the excitation source). The use of an 80× objective led to a spatial resolution of ~2.5 μm and a laser power on the sample surface of 10 mW. Spectra were calibrated against the 520 cm-1 peak of the Si/SiO substrate. Transmission   electron microscopy was performed by Philips CM 200 FEG and FEI Titan T operated at S4  300 keV. Optical images were acquired with an Olympus BX51 microscope with a C5060 camera. AFM was done on an Ntegra system (NT-MDT) in contact mode with cantilevers by NT-MDT (Pt-coated, spring constant 0.1 N/m) as well as Olympus (0.02 and 0.08 N/m). Electrical and Mechanical Measurements Resistance measurements of suspended nanosheets on a gold grid were conducted in UHV by contacting with the tungsten tip of the scanning tunnelling microscope of an Omicron Multiscan Microscope. The sheet resistance of nanosheets transferred on SiO2 was determined by a four-point measurement using Suess probes and a Keithley SMU Source-Measure Unit (Model 236). Mechanical measurements of nanosheet membranes on silicon window structures were performed under ambient conditions with a home-built pressure cell in a NT-MDT NTEGRA Scanning Probe Microscope. A detailed description of the experimental procedures and data evaluation is given below.   Supporting Online Text Analysis of spectroscopy and microscopy data S5  X-ray photoelectron spectroscopy (XPS) measurements were conducted as described in the Materials and Methods. The monolayer thickness was calculated by assuming an exponential attenuation of the Au 4f7/2 (or Si 2p) signals with a photoelectron attenuation length of 36 Å (35 Å)2. The XP spectra of C1s, S2p and Auf peaks for pristine, e-beam cross-linked and annealed BPT monolayer on Au are presented in Fig. S2. No other elements have been identified in the widescan XP spectrum. In some pristine BPT samples an O1s signal was seen just at the noise level of the measurement. This signal disappears completely in UHV after electron irradiation and annealing. Electron irradiation and annealing of the samples was conducted in UHV in the spectrometer chamber. The effective thickness of a pristine monolayer at room temperature is calculated as ~ 1.0 nm. This is in good agreement with the AFM data, Fig. S3. It decreases to a value of ~ 0.7 nm after electron irradiation (50 eV, ~ 50 mC/cm2) and annealing (~ 1000 K), Fig. S2. This correlates with a partial decrease of the C1s intensity and the disappearance of the S2p signal (desorption of sulfur). The chemical composition of annealed suspended membranes on TEM grids was analysed with a scanning Auger microscope. Fig. S4 presents an Auger spectrum of a BPT nanosheet annealed at ~1300 K on a Quantifoil-on-Mo TEM grid by e-beam heating in vacuum for 5 min. For comparison the Auger spectra of highly oriented pyrolytic graphite (HOPG) and a hole in Quantifoil (space without nanosheet) are   presented. Besides the C KLL Auger line no other Auger transitions can be seen in the S6  spectrum of the annealed nanosheet. Thus it consists of only carbon. We observed that the annealing of the nanosheets on Quantifoil-on-Mo TEM grids by e-beam heating at temperatures above 1500 K results in the formation of precipitates in the membranes, Fig. S5a. These are most likely carbides of Mo and W and may result from the hot Mo and W parts of the e-beam heater (W filament, Mo bottom plate of the sample holder). Fig S5b shows an enlargement of the SAED in Fig 3f of the main text. The rings labelled (1) and (2) can be indexed as belonging to the (0-110) and (1-210) lattice plane spacings of highly in-plane oriented nanocrystalline graphitic sheets. Fig. S6 presents chemical analysis of the transferred nanosheets on oxidized Si wafers. The thickness of the non-annealed BPT nanosheet at room temperature is ~1.2 nm. Some residual contaminations (C, O, F, I) from the fabrication and transfer procedure are observed in the XP spectra. These contaminations disappear after annealing to ~600 K in UHV, correlating with a decrease of the film thickness to ~0.8 nm. Nanosheets heated first on Au and then transferred on oxidized Si did not show any F or I signals even without annealing. Analysis of resistivity measurements In situ resistivity measurements The resistivity of the monolayer was determined in ultra high vacuum with the aid of a scanning tunneling microscope (STM). For this measurement the nanosheet was prepared on a gold grid (1500 mesh) and contacted by the STM tip in the center of the ~11 µm wide openings. A bias V was then applied between the tip and the gold grid,   resulting in a linear current response. The sheet resistivity can be extracted from such S7  measurements by applying a simple model. First it is assumed that a homogenous film is contacted by two electrodes; one electrode, a circular dot, is placed at the centre of a ring-like second electrode that confines the measured area, cf. Fig. S7. The current- voltage-characteristics can be easily calculated for this setup if contact resistances are neglected. Thus, the sheet resistivity S is determined by  S 2  r/ dot r ln( ring V I ) with the inner radius of the ring electrode rring and the radius of the dot electrode rdot. In our case the contact area of the STM tip is not exactly known and the outer electrode is a square-like frame. However, as the ratio of radii enters only as the argument of a logarithmic function, the sheet resistivity is not drastically affected by the electrode geometry. We therefore approximated our setup with this model using a ring electrode radius rring of 5 µm and a dot electrode radius rdot of 100 nm. Assuming that the true contact area of the tip equals a circle with a radius of 1 Å, our approximation leads to an overestimation of the sheet resistivity by a factor of less than three. This overestimation is not critical in our contribution as we discuss changes in the resistivity of more than five orders of magnitude. Ex situ resistivity measurements A four-point probe measurement method was used to determine the sheet resistivity more accurately. The nanosheet was placed on a SiO2 surface. Four probe tips were equidistantly arranged in a line and contacted the film as shown in Fig. S8. A current I   was driven through the two outer needles and the voltage drop V between the two inner S8  needles was measured. In this setup the sheet resistance S is given by3 :  S  ln( 2 ) V I  4 . 532 V I  / square Analysis of mechanical measurements Bulging tests were employed to measure the Young’s modulus of nanosheets transferred onto silicon samples with micron-sized openings. These silicon samples with the suspended nanosheet on top were mounted on a self-made pressure cell, a hollow steel cylinder with two sideway openings for applying and measuring a pressure and one upward opening for connecting with one nanomembrane, cf. Fig. S9a. A layer of polydimethylsiloxane (PDMS) was used to estabilsh a gas-tight seal between the silicon sample and the pressure cell. The nitrogen gas supply and the differential pressure sensor (HCX001D6V, Sensortechnics) were connected with the cell as schematically shown in Fig. S9b. Deflection of the membrane was measured with a NT-MDT NTEGRA Atomic Force Microscope (AFM) in contact mode by employing a platinum-coated silicon cantilever (force constant: 0.1 N/m). The platinum coating reduces the adhesion between the tip and the monolayer. Scans for data acquisition were conducted with a scan-speed of 5 – 8 µm/s and a very low feedback gain of 0.01 – 0.02, whereas scans for imaging used a scan-speed of 15 µm/s and a feedback gain of 0.35. The latter setting yields an improved image quality but is less gentle to the nanomembranes, i.e. the probability of rupture is enhanced.   The deflection setpoint setting adjusts the force that the tip applies to the S9  nanomembrane. This force leads to an indentation as apparent by the 15 nm high step between the silicon frame and the nanomembrane in Fig. 4B of the main text. A systematic variation of the setpoint with the resulting step height is presented in Fig. S10. The linear dependence shows a vanishing step height for a deflection setpoint of zero. This setpoint corresponds to the deflection value of an unperturbed cantilever, e.g. far away from the sample. In other words the cantilever is not bent at the deflection value of zero and therefore it does not apply any force to the nanomembrane at this deflection value. The bulging measurements were performed with a slightly higher setpoint which led to a certain step height . This quantity was measured on non- pressurized membranes; it was modeled for arbitrary pressures and it was employed to correct the measured deflection of the nanomembranes as shown in the following. The deflection of the membrane h is given by the AFM height signal hAFM and the step height  by: h  AFMh  . Here all quantities are given in reference to the height level of the silicon frame. The step height  is defined as positive if the AFM height signal is below the unperturbed membrane deflection. In the example of Fig. 4B the deflection of the membrane h is zero and the step height  has a positve value of 15 nm. The step height  of non-pressurized membranes was measured and employed to correct the deflection h. However, this measurement was not possible on bulged membranes. Therefore the change of the step height  due to the increased tension in a bulged membrane was taken into account by the following calculation4-6    0  0 0  hE 2 C 13  2 a 2 S10  with the step height of non-pressurized membranes of 0. All other quantities are explained in the next section. Note, that one approximation of this correction scheme is to assume a constant step height, i.e.  = 0. This simplification results in an underestimation of the Young’s modulus and the residual stress of up to 9 % and 1.6 %, respectively for the presented datasets. Nevertheless, all data were treated with the full correction scheme, only the step height  was calculated with the Young’s modulus and the residual stress of the simplified scheme. The change of the membrane deflection due to the applied pressure was employed to determine the Young’s modulus and the residual stress. The deflection of the membranes centre hC is described by7 p  ( 1  Et 34 ,(ga)  ) b a 3 h C  b )(c 1 a t  0 2 a h C with the pressure p, the Young’s modulus E, the residual stress 0 , the membrane thickness t, the Poisson ratio  and the length of the membranes short-edge being 2a. The constants g and c1 are taken from the literature7 and depend only on the membrane’s aspect ratio b/a and in the case of g on the Poisson ratio  . The Young’s modulus and the residual stress were determined by fitting the equation above to the measured p(hC) data. Note that each fitting constant is a measure for just one quantity: the Young’s modulus E or the residual stress 0 . The dimension of the membrane was measured in scanning electron micrographs (SEM) which allowed a higher precession as the values extracted from AFM images. In the case of Fig. 4B-D the half length of the   membranes short edge a was determined to 15.93 µm. All calculations were carried out S11  with a thickness t of 1 nm and a Poisson ratio  of 0.35. The latter value is not known, so a typical value for polymers was used. The residual strain 0 was calculated from the residual stress and the Young’s modulus by5  0 ( 1 )  E 0 . The strain of the bulged membranes was determined from the membrane deflection hC by6  0 2 h C 2 2 a 3 . In all presented bulging measurements the strain did not exceed a value of 1.3 %. The accuracy of this Young’s modulus measurement will be discussed in two parts, first in terms of the comparability of the presented data then in terms of the absolute accuracy. The most striking point in this error analysis is the strong effect of an uncertainty in the membrane dimension due to its contribution as fourth power to the Young’s modulus. Therefore any relative measurement error in the membrane size results in a fourfold contribution to the uncertainty of the Young’s modulus. This and contributions from the fitting quality and the constant g were determined for each measurement and are given as error bars in Fig. 4 of the main text. Additional sources of uncertainty lead to a constant but unknown correction factor for all measurements. Therefore these error contributions are discussed separately in the following. All   uncertainties of the Young’s modulus related to the calibration of the AFM piezo, the S12  scale bar in the SEM and the pressure sensor sum up to 15 %. The accuracy of the constant g, given in Ref. 7, was estimated from the comparison with experimental data in Ref. 7 to 3 % which gives a contribution to the Young’s modulus uncertainty of 9 %. The monolayer thickness uncertainty was estimated to 15 %. Thus these error contributions add up to about 40 %. Supporting Figures and Legends S13    Fig. S1 1,1’-biphenyl-4-thiol (BPT) a S 2p b Fig. S2 XPS characterization of BPT nanosheets on Au surface. (a) XP spectra of pristine, electron irradiated and annealed BPT samples acquired with a monochromatic Al-K source at a detection angle of the electron analyzer of 18° (monochromatic Al K radiation). (b) Schematic representation of the detection angle in the XPS measurements.   S14  Fig. S3 Optical micrograph of an extremely large (~5 cm2) cross-linked BPT nanosheet transferred on a 300 nm silicon oxide layer on silicon (left). AFM micrograph of the nanosheet edge showing its thickness of ~1nm (right). a hole nanosheet b Fig. S4 Auger microscopy characterization of annealed suspended nanosheets. (a) SEM Image of the analyzed area. (b) Auger spectra of a BPT nanosheet annealed for 5 min on a Quantifoil-on-Mo TEM grid in vacuum and HOPG surface. Excitation energy 3 kV.   S15  Fig. S5 a) HRTEM micrograph of a BPT nanosheet annealed 5 min in vacuum on a Quantifoil-on-Mo TEM grid at ~1700 K. Scale bar, 5 nm. Inset: SAED of somewhat larger area. b) Enlarged SAED from Fig 3f (main text) The rings labelled (1) and (2) can be indexed as belonging to the (0-110) and (1-210) lattice plane spacings of highly in-plane oriented nanocrystalline graphitic sheets. Fig. S6 XPS of a BPT nanosheet transferred to an oxidized Si wafer before and after annealing at ~ 650 K. Monochromatic Al-K radiation, angle of the electron analyzer of 18° relative to the surface normal.   S16  Fig. S7 Schematic drawing of the 2-probe measurement (left) and SEM image (right) including the radii of the model that was used to determine the sheet resistivity. The nanosheet is contacted by a tungsten tip of an STM and a voltage is applied between the tip and the supporting gold grid to measure the current response. Fig. S8 Schematic drawing of the 4-probe measurement (left) and photograph of the actual setup (right). Four probes are equidistantly brought into contact with the nanosheet; a current is fed through the outer probes and the voltage drop is measured at the inner two probes.   S17  Fig. S9 Photograph of the pressure cell. b Schematic of the bulging test setup and photograph of the pressure cell with one sample mounted. Fig. S10 AFM height step at the border between the nanomembrane and the silicon frame as function of the deflection setpoint. A zero setpoint corresponds to the deflection of an unperturbed cantilever, e.g. far away from the sample.   S18  Supporting References and Notes S1. NIST X-ray Photoelectron Spectroscopy Database 20, (Version 3.4, Web Version, 2003). In NIST X-ray Photoelectron Spectroscopy Database 20, (Version 3.4, Web Version, 2003). S2. Briggs, D.; Grant, J. T., Surface Analyses by Auger and X-Ray Photoelectron Spectroscopy. SurfaceSpectra Limited: Chichester, 2003. S3. Plummer, J. D.; Deal, M. D.; Griffin, P. B., In Silicon VLSI Technology, Prentice Hall: Upper Saddle River, NJ, USA, 2000; p 114. S4. Morse, P. M., Vibration and sound. 2nd Ed. ed.; McGraw-Hill Book Company: New York, NY, 1948; p 176. S5. Small, M. K.; Nix, W. D. Journal of Materials Research 1992, 7, (6), 1553-1563. S6. Xiang, Y.; Chen, X.; Vlassak, J. J. Journal of Materials Research 2005, 20, (9), 2360-2370. S7. Vlassak, J. J.; Nix, W. D. Journal of Materials Research 1992, 7, (12), 3242-3249. 

1507.02833

1

1507

Dependence of the Energy Transfer to Graphene on the Excitation Energy

Fluorescence studies of natural photosynthetic complexes on a graphene layer demonstrate pronounced influence of the excitation wavelength on the energy transfer efficiency to graphene. Ultraviolet light yields much faster decay of fluorescence, with average efficiencies of the energy transfer equal to 87% and 65% for excitation at 405 nm and 640 nm, respectively. This implies that focused light changes locally the properties of graphene affecting the energy transfer dynamics, in an analogous way as in the case of metallic nanostructures. Demonstrating optical control of the energy transfer is important for exploiting unique properties of graphene in photonic and sensing architectures.

cond-mat.mes-hall

cond-mat

Dependence of the Energy Transfer to Graphene on the Excitation Energy Sebastian Mackowski, Izabela Kamińska Institute of Physics, Faculty of Physics, Astronomy and Informatics, Nicolaus Copernicus University, Grudziadzka 5, 87-100 Torun, Poland E-mail: [email protected] Keywords: graphene, photosynthetic complex, energy transfer, fluorescence Abstract Fluorescence studies of natural photosynthetic complexes on a graphene layer demonstrate pronounced influence of the excitation wavelength on the energy transfer efficiency to graphene. Ultraviolet light yields much faster decay of fluorescence, with average efficiencies of the energy transfer equal to 87% and 65% for excitation at 405 nm and 640 nm, respectively. This implies that focused light changes locally the properties of graphene affecting the energy transfer dynamics, in an analogous way as in the case of metallic nanostructures. Demonstrating optical control of the energy transfer is important for exploiting unique properties of graphene in photonic and sensing architectures. 1  Energy transfer is one of the most fundamental processes at the nanoscale [1,2]. Whenever a donor is placed sufficiently close to an acceptor, assuming their spectral properties and relative orientations do not inhibit it, they can couple via electrostatic interactions and the energy is funneled down to the acceptor [1]. Such a scheme evolved in natural photosynthesis [3] for efficient capturing and transport of the sunlight energy, and has been recently implemented in artificial light-harvesting assemblies [4,5]. The energy transfer between molecules with precisely designed optical spectra has also been useful in studying and understanding molecular mechanisms responsible for protein folding [6], intracellular transport [7], etc., as the efficiency of this process is extremely sensitive to the distance between a donor and an acceptor [1]. The key spectral signature of the energy transfer is a decrease of the emission intensity of a donor at the expense of an acceptor with simultaneous shortening of a donor fluorescence decay time [1]. In fact, from the reduction of the decay constant it directly related to the efficiency of the energy transfer, which in molecular assemblies is independent of the donor excitation wavelength, as the rates of the energy transfer are typically a few orders of magnitude slower than intra-molecular transitions [1]. As an example, the energy transfer takes place between emitters and metallic surfaces or nanostructures [8,9,10], which feature strong plasmonic oscillations of free electrons. However, as metals typically fluoresce very weakly, the energy transferred from a dipole placed in their vicinity is dissipated non-radiatively, mainly by heat. In such cases the interaction between an emitter and a metallic nanostructure can be probed by monitoring decrease of donor emission intensity and shortening of its fluorescence decay. The resonant character of plasmon excitations in metallic nanostructures implies strong wavelength dependence of both metal-enhanced fluorescence and fluorescence quenching [9,11,12]. The uniqueness of graphene, a two-dimensional sp2-hybridized carbon hexagonal lattice, has been advocated worldwide in the last decade, but predominantly in regard to its 2  thermal, electrical, and mechanical properties [13]. Only recently the optical properties of graphene have emerged as highly attractive from the point of view of potential applications in photonics and optoelectronics [14]. One of the remarkable facets of graphene is its uniform absorption, which extends over the whole visible range down to the infrared [15]. With the absence of fluorescence, it renders graphene as an exceptional acceptor in devices that utilize energy and/or electron transfer. Recently, energy transfer in graphene-based assemblies has been studied both experimentally and theoretically [16,17]. The two-dimensional character of graphene results in a weaker distance dependence of the energy transfer efficiency (~ d-4) as compared to a three-dimensional case (~ d-6) [16]. This difference has been verified experimentally by studying rhodamine dyes on graphene [18]. The results indicate that this coupling can be exceptionally strong, with the decay rates being enhanced by up to 90 times. In this work we aim at demonstrating that the efficiency of the energy transfer to graphene depends on the excitation wavelength. To this end a molecular system which features broad absorption, is a few-nm large, and fluoresces is required. Some of the naturally evolved photosynthetic complexes fulfill these conditions. For photosynthetic complexes deposited directly onto a graphene layer we observe decrease of the fluorescence intensity accompanied with reduction of the decay time. The key finding that emerges from these experiments is a strong dependence of the efficiency of the energy transfer on the excitation wavelength: the effect is much stronger for 405 nm than for 640 nm excitation. We conclude that energy quenching in graphene is driven not only by dipole-dipole interaction, but also a mechanism associated with light-induced oscillations of electrons in graphene. Indeed, exciting electrons in graphene has an effect of its dissipative efficiency, which opens avenues in interfacing electronic and plasmonic character of graphene in hybrid nanostructures and control the electronic dynamics of such systems with light. 3  In order to probe the interactions with graphene, we use peridinin-chlorophyll-protein (PCP) complex from algae Amphidinium carterae. It is a water-soluble light-harvesting complex, whose structure was determined with 1.7A resolution [19]. Optical spectra of PCP in solution are displayed in Fig. 1. The absorption spans over the whole visible range with the main band (from 350 to 550 nm) attributed mainly to peridinin absorption. Chlorophylls absorb through the Qy band around 670 nm and the Soret band at 437 nm [20]. The emission of PCP complex is associated with fluorescence of chlorophylls and occurs at 673 nm, with a 30% quantum yield and a decay time constant of 4 ns [21]. The decay time constant oft he fluorescence emisison is around 3 orders of magnitude shorter than any oft he intra-complex transfer times, within 3 ps, regardless of the excitation wavelength, the energy is transfered to the Qy band of chlorophyll molecules [20]. Graphene substrates were purchased from Graphene Supermarket®. We used 1cm × 1cm p-doped silicon wafers with a single layer graphene deposited using Chemical Vapor Deposition on a 285 nm thick silicon dioxide layer. The presence of graphene monolayer on the substrates was confirmed with Raman spectrum measurement. For optical experiments we used highly diluted (optical density of 0.009 at 671 nm, concentration less that 10 µM) aqueous solution of PCP complexes. Such a low concentration is very important as on the one hand it strongly reduces inner filter effect, but this also yields a thin layer of PCP complexes on a graphene surface. As a result, we minimize the fraction of PCP that is not coupled to graphene, thus takes no part in the energy transfer. Absorption and fluorescence spectra of PCP solution were obtained at room temperature using Cary 50 UV-Vis spectrophotometer (Varian) and Fluorolog 3 spectrofluorimeter (Jobin-Yvon). Spectrally- and time-resolved fluorescence measurements were performed using a home-built confocal fluorescence microscope described in detail in [22]. The sample was placed on a piezoelectric translation stage. We used pulsed laser excitation at 405 nm and 640 nm (repetition rate 20 MHz, average power 30 µW, power 4  density ~1MW/cm-2). Importantly, PCP can be efficiently excited at 405 nm (Soret band), and 640 nm (excited states of chlorophylls). The laser beam was focused on the sample by LMPlan 50x objective (Olympus) with a numerical aperture 0.5. Fluorescence was first filtered by a longpass filter (HQ665LP Chroma) and then the spectra were detected with Andor iDus DV 420A-BV CCD camera coupled to an Amici prism. Time-resolved measurements were performed by time-correlated single photon counting technique using an SPC-150 module (Becker & Hickl) with fast avalanche photodiode (idQuantique id100-20) as a detector. In order to select appropriate wavelength range, we used an additional bandpass filter (FB670/10 Thorlabs). When compared with the reference, fluorescence spectra for PCP on graphene (as shown in Fig. S1) indicate strong reduction of the emission intensity, an effect we can tentatively attribute to the energy transfer [23]. At the same time, the shape of the PCP emission spectrum on graphene remains unaffected, and is identical to previously published [21], which indicates that the protein is intact and that all the energy transfer pathways are active in the photosynthetic complex upon deposition on graphene. This observation implies graphene suitability for interfacing with functional biological systems. The assumption that the observed reduction of the emission intensity of PCP complexes deposited on graphene is due to the energy transfer from the chlorophylls in the photosynthetic complex to graphene is supported by a strong decrease of the fluorescence decay, as shown in Fig. 2a and 2b for 405 nm and 640 nm excitations, respectively. Three transients shown for each excitation wavelength were measured at different locations across the sample. For both excitation wavelengths fluorescence decays much faster than for the reference. In addition, the transients measured for PCP complexes on graphene, feature high inhomogeneity: there are decays that are relatively long, as well as comparable with the temporal resolution of the optical setup. While the results shown in Fig. 3a and 3b display transients close to average as well as borderline cases, we note that for the 405 nm excitation 5  the majority of measured decays is very fast, while exciting with 640 nm yields considerably longer decays. The distribution of fluorescence transients indicates that PCP complexes probed in every measurement couple slightly different to the graphene layer. It is expected as we have not incorporated any control of the separation between PCP complexes and graphene to make the interaction more uniform across the substrate. Furthermore, it has been shown that for graphene deposited on silica, the local structure of graphene is also quite inhomogeneous with islands of high and low mobility of carriers [24]. We might therefore assume that such a non- uniformity contributes also to the observed spreading of fluorescence transients, although the scale of these inhomogeneities is less than 100 nm, which is significantly less than spatial resolution of the fluorescence microscope. Regardless, it is striking that most of them exhibit almost monoexponential behavior. It could indicate that majority of PCP complexes with the laser spot is coupled to graphene with a comparable strength, although we clearly see distribution of lifetimes indicating the presence of sub-ensembles in our structure [1]. Each of the lifetimes would then be attributed to the PCP complexes coupled to graphene with different efficiency leading also to a different energy transfer rate. Certainly we do not observe long (~4 ns) component attributable to PCP complexes not coupled to graphene. We can conclude that the fluorescence experiments carried out for PCP deposited on a graphene surface demonstrate shortening of the fluorescence decay, which is accompanied with a decrease of the overall fluorescence intensity. Taken together, these observations strongly suggest that the energy absorbed by PCP complexes, both at 405 nm and 640 nm, after being transferred between and within the pigments comprising the light-harvesting complex, is efficiently dissipated into the graphene layer. In general such effects can also be attributed to electron transfer from photosynthetic complex to graphene, as observed recently in [25]. However in this case a charge-separating complex, that is Photosystem I, was used and the whole structure was immersed in properly chosen electrolyte in order to facilitate the 6  electron transfer. As the PCP complex is just a light-harvesting antenna, we can exclude such a scenario in our structure. As pointed out earlier, in molecular systems, where the interaction leading to the energy transfer takes place between two dipole moments, the decay constant of a donor in the presence of an acceptor is independent of the excitation wavelength used for exciting the donor. This is a reminiscence of the fact that light has no effect on the surrounding of the molecules participating in the energy transfer. In contrast, in the case of PCP complexes deposited on graphene, the fluorescence decay strongly depends on the excitation wavelength, as displayed in Fig. 4, where histograms of decay times measured for 405 nm and 640 nm excitations are compared. The average decay times are equal to 0.5 ns and 1.4 ns, respectively, what results in corresponding energy transfer efficiencies of 87% and 65%. Such a clear influence of the excitation wavelength on the energy transfer indicates that in addition to populating PCP excited states (the pigments embedded within the protein), light changes also the local surroundings, presumably the properties of graphene. In a way, although the analogy is certainly not complete, strong dependence of the donor decay on the excitation wavelength, is similar to metallic systems, where in order to facilitate the energy transfer it is necessary to excite particular resonance, i.e. plasmon resonance. On a final note, by integrating fluorescence transients we can estimate total emission intensity of PCP complexes, and therefore, obtain correlation between fluorescence decay and intensity. The result plotted in Fig. 5 shows that for PCP complexes deposited on graphene, excitation with both 405 nm and 640 nm results in strongly correlated values of fluorescence decay times and integrated intensities. Namely, shorter decay times are correlated with lower intensities. Correlation coefficients (Pearson’s coefficients) estimated for both excitation wavelengths are approximately 0.7. As both these quantities are spectral signatures of the energy transfer, correlation between them strongly suggests that decrease of emission intensity is indeed directly attributed to the energy transfer from PCP to graphene. 7  Although elaboration of the processes responsible for the observed excitation energy dependence of the energy transfer to graphene requires further experiments, in analogy to plasmonic nanostructures, we can propose a scenario that in principle can explain this effect. Electrons excited in graphene by light can oscillate, similarly as in a metallic nanoparticle, in a confined space defined by a monolayer of graphene on the one hand, and the size of the laser spot, on the other. As the latter is around 1 micron in a diameter, it is comparable to the wavelength of light used in the experiment. Therefore, locally excited graphene layer can be considered analogous to a metallic nanostructure with strong sensitivity of its dissipative properties on the excitation wavelength. Small variation of the excitation wavelength (less than 100 nm, characteristic of most organic dyes) is not sufficient to demonstrate this unique effect in a convincing way, therefore an emitter that can be excited across the wide range of wavelengths is required. In conclusion, we observe strong dependence of the energy transfer efficiency in hybrid assemblies composed of natural photosynthetic complexes and graphene on the excitation wavelength. High energy excitation (405 nm) results in energy transfer efficiency of 87%, while low energy one (640 nm) yields efficiency of 65%. This proof-of-concept result, in addition to paving a way towards designing novel graphene-based structures for photosynthetic energy harvesting and conversion, indicates that the energy transfer in a hybrid assembly can be controlled by light. We thank Prof. Eckhard Hofmann (Bochum University, Germany) for providing PCP complexes. Research was supported by the project number DEC-2013/10/E/ST3/00034 from the National Science Center (NCN) and the grant ORGANOMET No: PBS2/A5/40/2014 from the National Research and Development Center (NCBiR). 8  1. 2. 3. 4. 5. 6. 7. 8. 9. J. R. Lakowicz, Principles of Fluorescence Spectroscopy, Springer, Third Edit., 2006. K. E. Sapsford, L. Berti and I. L. Medintz, Angew. Chemie, 2006, 45, 4562–4589. R. E. Blankenship, Molecular Mechanisms of Photosynthesis, Wiley-Blackwell, 1987. I. Nabiev, A. Rakovich, A. Sukhanova, E. Lukashev, V. Zagidullin, V. Pachenko, Y. P. Rakovich, J. F. Donegan, A. B. Rubin and A. O. Govorov, Angew. Chemie, 2010, 49, 7217–7221. D. Kuciauskas, P. A. Liddell, S. Lin, T. E. Johnson, S. J. Weghorn, J. S. Lindsey, A. L. Moore, T. A. Moore and D. Gust, J. Am. Chem. Soc., 1999, 121, 8604–8614. T. Ha, T. Enderle, D. F. Ogletree, D. S. Chemla, P. R. Selvin and S. Weiss, Proc. Natl. Acad. Sci. U. S. A., 1996, 93, 6264–6268. G. Fink, L. Hajdo, K. J. Skowronek, C. Reuther, A. A. Kasprzak and S. Diez, Nat. Cell Biol., 2009, 11, 717–723. J. Y. Yan, W. Zhang, S. Duan, X. G. Zhao and A. O. Govorov, Phys. Rev. B - Condens. Matter Mater. Phys., 2008, 77, 1–9. E. Dulkeith, A. C. Morteani, T. Niedereichholz, T. A. Klar, J. Feldmann, S. A. Levi, F. C. J. M. van Veggel, D. N. Reinhoudt, M. Möller and D. I. Gittins, Phys. Rev. Lett., 2002, 89, 203002 (1–4). 10. K. Shimizu, W. Woo, B. Fisher, H. Eisler and M. Bawendi, Phys. Rev. Lett., 2002, 89, 117401 (1–4). 11. N. Czechowski, P. Nyga, M. K. Schmidt, T. H. P. Brotosudarmo, H. Scheer, D. Piatkowski and S. Mackowski, Plasmonics, 2012, 7, 115–121. 12. P. Bharadwaj, P. Anger and L. Novotny, Nanotechnology, 2006, 18, 044017 (1–5). 13. A. K. Geim, Science, 2009, 324, 1530–1534. 14. A. K. Geim and K. S. Novoselov, Nat. Mater., 2007, 6, 183–191. 15. R. R. Nair, P. Blake, A. N. Grigorenko, K. S. Novoselov, T. J. Booth, T. Stauber, N. M. R. Peres and A. K. Geim, Science, 2008, 320, 1308. 16. R. S. Swathi and K. L. Sebastian, J. Chem. Phys., 2009, 130, 086101 (1–3). 17. Z. Chen, S. Berciaud, C. Nuckolls, T. F. Heinz and L. E. Brus, ACS Nano, 2010, 4, 2964–2968. 18. L. Gaudreau, K. J. Tielrooij, G. E. D. K. Prawiroatmodjo, J. Osmond, F. J. García de Abajo and F. H. L. Koppens, Nano Lett., 2013, 13, 2030–2035. 19. E. Hofmann, P. M. Wrench, F. P. Sharples, R. G. Hiller, W. Welte and K. Diederichs, Science, 1996, 272, 1788–1791. 9  20. D. Zigmantas, R. G. Hiller, V. Sundstrom and T. Polivka, Proc. Natl. Acad. Sci. U. S. A., 2002, 99, 16760–16765. 21. S. Wörmke, S. Mackowski, T. H. P. Brotosudarmo, C. Jung, A. Zumbusch, M. Ehrl, H. Scheer, E. Hofmann, R. G. Hiller and C. Bräuchle, Biochim. Biophys. Acta - Bioenerg., 2007, 1767, 956–964. 22. B. Krajnik, T. Schulte, D. Piątkowski, N. Czechowski, E. Hofmann and S. Mackowski, Cent. Eur. J. Phys., 2011, 9, 293–299. See supplemental material at [URL will be inserted by AIP] for comparison of 23. fluorescence emission spectra of the studied samples. 24. K. M. Burson, W. G. Cullen, S. Adam, C. R. Dean, K. Watanabe, T. Taniguchi, P. Kim and M. S. Fuhrer, Nano Lett., 2013, 13, 3576–3580. 25. G. LeBlanc, E. Gizzie, G. K. Jennings, and D. Cliffel, Adv. Energy Mater., 2014, 4, 1301953. 10  Fig. 1. Optical spectra of PCP complexes in solution: red and black curve corresponds to absorption and emission, respectively. Fig. 2. Examples of normalized fluorescence transients measured for PCP complexes deposited on a single layer graphene measured for excitation (a) 405 nm and (b) 640 nm. Green lines correspond to the reference. Fig. 3. (a) Histograms of the decay times obtained for PCP on single layer graphene for excitation 405 nm (blue) and 640 nm (red). (b) Correlation between fluorescence intensity and the decay time measured for PCP on graphene for excitation 405 nm (blue) and 640 nm (red). 11 Fig. 1. 12 Fig. 2. 13 Fig. 3. 14 

1709.06043

1

1709

Landauer-B\"uttiker approach to strongly coupled quantum thermodynamics: inside-outside duality of entropy evolution

We develop a Landauer-B\"uttiker theory of entropy evolution in time-dependent strongly coupled electron systems. This formalism naturally avoids the problem of system-bath distinction caused by the strong hybridization of central system and surrounding reservoirs. In an adiabatic expansion up to first order beyond the quasistatic limit, it provides a clear understanding of the connection between heat and entropy currents generated by time-dependent potentials and shows their connection to the occurring dissipation. Combined with the work required to change the potential, the developed formalism provides a full thermodynamic description from an outside perspective, applicable to arbitrary non-interacting electron systems.

cond-mat.mes-hall

cond-mat

a Landauer-Buttiker approach to strongly coupled quantum thermodynamics: inside-outside duality of entropy evolution Anton Bruch,1 Caio Lewenkopf,2 and Felix von Oppen1 1Dahlem Center for Complex Quantum Systems and Fachbereich Physik, Freie Universitat Berlin, 14195 Berlin, Germany 2Instituto de F´ısica, Universidade Federal Fluminense, 24210-346 Niter´oi, Brazil (Dated: September 19, 2017) We develop a Landauer-Buttiker theory of entropy evolution in time-dependent strongly coupled electron systems. This formalism naturally avoids the problem of system-bath distinction caused by the strong hybridization of central system and surrounding reservoirs. In an adiabatic expansion up to first order beyond the quasistatic limit, it provides a clear understanding of the connection between heat and entropy currents generated by time-dependent potentials and shows their connection to the occurring dissipation. Combined with the work required to change the potential, the developed formalism provides a full thermodynamic description from an outside perspective, applicable to arbitrary non-interacting electron systems. Introduction.−Ongoing progress in nanofabrication raises interest in the thermodynamics of nanomachines [1, 2], describing the exchange of heat and work with their environment as well as their efficiencies. The laws of thermodynamics are extremely successful in charac- terizing machines consisting of a macroscopic number of particles by just a handful of parameters such as temper- ature and pressure. How these laws carry over to micro- scopic systems that consist of few particles and exhibit quantum behavior, is the central problem of the field of quantum thermodynamics. At small scales, the thermo- dynamic variables necessarily acquire strong fluctuations [3, 4] and the system-bath distinction becomes fuzzy [5– 8]. A crucial quantity in this regard is the entropy which links thermodynamics and information [9–12], describes irreversibility, and governs the efficiencies of various en- ergy conversion processes. Here, we put forward a formalism based on the Landauer-Buttiker scattering approach to describe the entropy evolution generated by (slow) time-dependent potentials in electronic mesoscopic systems strongly cou- pled to external reservoirs. A key advantage of the scat- tering approach is that it naturally avoids the system- bath distinction, a ubiquitous problem for theoretical treatments of the strong coupling regime [5, 13–17]. In an elementary thermodynamic transformation, an external agent performs work on a system by changing its Hamiltonian, constituting a single "stroke" of a quan- tum engine. For electronic nanomachines, this is achieved by changing the potential in a finite region which is cou- pled to electronic reservoirs. This type of machine can for instance be realized by a quantum dot connected to leads and subject to a time-dependent gate potential. If the gate potential is changed slowly, the coupling to the reservoir ensures thermal equilibrium at all times and the transformation occurs quasistatically. The change of the von-Neumann entropy S[ρ] = −Tr (ρ ln ρ) , (1) proportional to the heat dQ = T dS released into the reservoir at temperature T . This should be contrasted with the entropy evolution of a closed quantum system. Its purely unitary time evo- lution implies that the von-Neumann entropy remains unchanged at all times. Here we want to discuss the en- tropy evolution of simple electronic nanomachines, which combine fully coherent quantum dynamics with contact to baths and can involve strong coupling between sys- tem and reservoir. In this problem, quantum effects such as coherences, hybridization, and entanglement are ex- pected to become important. Such electronic nanoma- chines can be described by the Landauer-Buttiker formal- ism, that has been very successfully used to understand the conductance [18], electron pumping [19], heat trans- port and current noise [18, 20–22], entanglement creation [23, 24], and adiabatic reaction forces [25–28] in a variety of mesoscopic systems. One considers a scattering region connected to ideal leads, as depicted in Fig. 1, where non-interacting elec- trons propagate freely and under fully coherent quantum dynamics. Relaxation is accounted for by connecting the leads to electronic reservoirs at well defined temperatures and chemical potentials, which determine the distribu- tion of incident electrons. This allows one to calculate energy or particle currents in the leads by accounting for in- and outgoing electrons. Invoking energy and parti- cle conservation, these currents permit one to deduce the change of energy and particle number in the scattering re- gion from an outside perspective. Since the von-Neumann entropy S is conserved under coherent unitary dynam- ics, the change of entropy in the scattering region can also be inferred from the entropy currents carried by the scattered electrons. The subsequent relaxation processes occur in the bath. In the Landauer-Buttiker formalism the reservoirs are macroscopic and relax to equilibrium such that excitations entering a reservoir never return [18]. associated with the equilibrium state of the system is Several studies have investigated the thermodynamic 2 (5) Similarly, the energy current I E α in channel α reads I E α (t) = αα(
)(cid:9) . αα (t,
) − φin (cid:90) ∞ −∞
(cid:8)φout d
2π α −µαI N α = I E The heat current I Q α carried by the electrons in the leads is a combination of the particle current I N α into the corresponding reservoir with chemical potential µα and the energy current I E α . We can express the total heat current in terms of the diagonal elements of the distribution matrix φout (cid:8)φout(t,
) − φin(
)(cid:9) , (6) I Q tot(t) = (
− µ)trc d
2π (cid:90) ∞ −∞ where the trace runs over channel and lead space. Here, for simplicity we assume the same chemical potential µ in all reservoirs. To obtain the entropy current, we begin by consider- ing the entropy of a single incoming channel. For a given energy the channel can be either occupied or empty, ac- cording to fα(
), and contributes with σ [fα(
)] = −fα(
) ln [fα(
)] − (1 − fα(
)) ln [1 − fα(
)] (7) to the system entropy. By analogy with the particle cur- rent, Eq. (3), we write the incoming entropy current as I S in α = d
2π σ [fα(
)] . (8) (cid:90) ∞ −∞ FIG. 1. The scattering potential in the central region, e.g. a quantum dot, slowly changed by an external parameter X(t), driving a net heat and entropy current into the leads. properties of mesoscopic electronic systems using the sim- plest model that captures the difficulties of the strong coupling regime [6, 7, 29]: A single electronic level strongly coupled to a free electron reservoir and subject to a slowly varying gate potential. The thermodynamic transformations of this system were described from an inside perspective, i.e., in terms of the thermodynamic variables of the single level. To account for the hybridiza- tion in the strong coupling regime, these thermodynamic functions were treated with the help of the nonequilib- rium Green's function formalism. This approach was re- cently found to be limited to a small set of systems [8]. Entropy current carried by scattering states.−We con- sider a time-dependent scattering region connected to one or multiple ideal leads, in which the electrons propagate in transverse scattering channels, and leave the electronic spin degree of freedom implicit. Electrons in incoming and outgoing channels are described by annihilation op- erators a and b, related by the scattering matrix S,  b1(
) ... bN (
)  = (cid:90) d
(cid:48) 2π  a1(
(cid:48)) ... aN (
(cid:48))  . S(
,
(cid:48)) Here the subscript α = 1,··· , N labels the channels and leads. The leads are connected to electronic reservoirs, which determine the distribution of the incoming chan- † nels to be (cid:104)a β(
)aα(
(cid:48))(cid:105) = φin αβ(
)2πδ(
−
(cid:48)) in terms of a diagonal distribution matrix φin αβ(
) = δαβfα(
), where fα(
) is the Fermi distribution with temperature T and chemical potential µα. The particle current in channel α through any cross- section of the corresponding lead is obtained by account- ing for in- and outgoing electrons [18], (cid:90) ∞ −∞ d
2π (cid:8)φout αα(
)(cid:9) , αα (t,
) − φin I N α (t) = (3) where the one-dimensional density of states
α(
) = [2πvα(
)] −1 and the group velocity vα(
) compensate (we set  = 1). φout(t,
) is given by the Wigner transform φout αβ (t,
) = † β(
−
/2)bα(
+
/2) b . (4) (cid:90) d
2π e−iεt(cid:68) (cid:69) Hence, as expected [30], each of the incoming spin- resolved channels carries an entropy current of πT /6 to- wards the scattering region. (2) Scattering redistributes the electrons between the out- going channels, thereby modifying the entropy flow into the leads. The scattering-induced correlations between outgoing scattering states [20, 21] are encoded in the non- diagonal distribution matrix φout αβ (t,
) for the outgoing electrons. As we show below, the natural extension of Eq. (8) reads (cid:90) ∞ −∞ (cid:110) d
2π trc (cid:111) I S in(out)(t) = σ[φin(out)(t,
)] . (9) To motivate Eq. (9) we derive the non-interacting fermionic density matrix for a given distribution matrix † ¯φαβ = Tr[ρc βcα]. In the scattering setup the incoming operators describe particles of an equilibrium reservoir and the outgoing operators are linear functions of the incoming ones, cf. Eq. (2). Hence, all averages can be calculated via Wick's theorem and the single-particle cor- relations described by φ fully determine all expectation values. Our derivation exploits the maximum entropy prin- ciple that yields the most general density matrix given certain single-particle correlations [31]. (We obtain the 3 Entropy current induced by a dynamic scatterer.−The entropy and heat currents generated by a slowly changing scattering potential V [X(t)] are obtained by expanding the scattering matrix and the outgoing distribution ma- trix about the frozen configuration in powers of the ve- locity X [25–28]. Up to first order, the Wigner transform of the scattering matrix can be expressed in terms of the frozen scattering matrix S and its first order correction A, S(
, t) = S + XA. This expansion is well motivated in the regime where X(t) changes on a characteristic time scale much longer than the electronic dwell time in the scattering region. Accordingly, we write φout as φout (cid:39) I f + φout(1) + φout(2) , (16) where I is a unit matrix in channel and lead space and the superscript stands for the order in X. (We omit time and energy labels for better readability.) Similarly, we expand σ[φout(
)] up to second order about the uncorre- lated equilibrium φout(1) + φout(2)(cid:17) dσ [f ] (cid:16) φout(1)(cid:17)2 df (cid:16) . (17) σ[φout] (cid:39) Iσ [f ] + I + I 1 2 d2σ [f ] df 2 Note that the second order contribution proportional to −1 is always negative due to the d2σ [f ] /df 2 = (T ∂
f ) concavity of σ. By inserting the above expression in Eq. (15) we obtain (cid:90) ∞ −∞ d
2π trc (cid:26)
− µ T + 1 (cid:16) φout(1) + φout(2)(cid:17) φout(1)(cid:17)2 (cid:27) (cid:16) (cid:110) φout(1) + φout(2)(cid:111) , . 2T ∂
f (18) (19) where we have used that φin = I f (
). By the same token, Eqs. (6) and (16) give (14) (cid:90) ∞ −∞ I Q tot = (
− µ) trc d
2π These expressions nicely elucidate the connection be- tween heat and entropy currents, and the departure from dQ = T dS beyond the quasistatic limit. At first order in X, corresponding to the quasistatic regime, the en- tropy current is entirely given by the heat current over temperature I S(1) tot /T , i.e., the proposed form of the entropy current correctly connects to the quasistatic equilibrium. In contrast, at second order an additional negative correction appears tot = I Q(1) (cid:26)(cid:16) φout(1)(cid:17)2(cid:27) . (20) I S(2) tot = I Q(2) tot T + d
2π 1 2T ∂
f trc (cid:90) ∞ −∞ Since trc{(φout(1))2} contains all off-diagonal elements of φout(1), it encodes the correlations created by the dy- namic scatterer. These correlations determine by how S = σ[Λα] = tr (σ[Λ]) , (13) α I S tot = same result following the approach of Ref. [32].) The La- grangian for maximizing the von-Neumann entropy un- † der the constraints Trρ = 1 and ¯φαβ = Tr[ρc βcα] reads (cid:88) (cid:16) (cid:104) (cid:105) − ¯φαβ (cid:17) L = −Tr [ρ ln ρ] + λαβ Tr ρc † βcα − γ (Trρ − 1) , αβ (10) where Tr denotes the many-particle trace over all pos- sible occupations and γ as well as the λ's are Lagrange multipliers. It is convenient to diagonalize the Hermitian matrix ¯φ and introduce a rotated basis, namely ¯φ = U ΛU† and cα = Uαcdc , (11) (cid:88) c where U is a unitary matrix and Λαβ = Λαδαβ is diag- onal containing the real eigenvalues of ¯φ. In the rotated basis the Lagrangian L allows us to maximize the von- Neumann entropy with the given constraints. This yields the density matrix ρ = (1 − Λα) , (12) (cid:19)nα (cid:18) Λα 1 − Λα where nα is the occupation of mode α in the rotated basis. We calculate the entropy S of this density matrix by summing over all possible occupations in the rotated basis, (cid:89) α (cid:88) where the sum over the diagonal elements of Λ is in- cluded through the single-particle trace tr. Finally, ro- tating back to the original basis, Λ = U† ¯φU , we find the entropy in terms of the distribution matrix ¯φ, S = tr(cid:0)σ[ ¯φ](cid:1) . For a slowly changing scattering potential, we associate the entropy with the time-dependent distribution matrix φαβ(t,
) of the scattering states in Eq. (4), for which the single-particle trace represents an integral over energy and a trace trc over channel and lead indices. By combing in- and outgoing entropy currents we write the total entropy current into the leads as (cid:8)σ[φout(t,
)] − σ[φin(
)](cid:9) . (15) I S tot(t) = d
2π trc (cid:90) ∞ −∞ In the case of a static scatterer between two biased reser- voirs at zero temperature, the entropy current can be used to quantify the entanglement of outgoing electron- hole pairs created in a tunneling event. Indeed, we ver- ify that an immediate generalization of Eq. (15) repro- duces the quantum mutual information between outgoing scattering channels on the left and right as obtained in Ref. [23] (see [33] for details). much the entropy current in the leads is smaller than the corresponding heat current over temperature. This net inflow of entropy into the scattering region reflects the local dissipation-induced increase of entropy. We calculate φ explicitly within the gradient expansion [25–27]. Assuming that fα(
) = f (
), one writes φout(1) in terms of the frozen scattering matrix S, φout(1) (
, t) = i X∂
f S∂X S† . (21) Inserting φout(1) into the entropy current Eq. (20), we obtain the entropy current up to second order with W (2) = − X 2 2 (cid:90) ∞ −∞ . (22) (cid:0)∂X S†∂X S(cid:1) ≥ 0 . (23) I S tot = I Q tot T − W (2) T d
2π ∂
f (
)trc W (2) = γ X 2 is exactly the dissipated power Remarkably, that the external agent pumps into the system as a result W (2) was de- of the time-dependent system Hamiltonian. rived in Refs. [25–27] in terms of the friction coefficient γ of the back-action force that needs to be overcome by the external agent. Thus, from our outside perspective, dis- sipation leads to an inflow of entropy into the scattering region in addition to the heat-current contribution. We are now ready to discuss the inside-outside dual- ity of entropy evolution: We utilize the acquired knowl- edge about the entropy current (outside perspective) to draw conclusions about the evolution of the entropy s of the strongly coupled subsystem located in the scat- tering region (inside perspective). The direct calculation of the thermodynamic functions of such a subsystem has proven problematic in the past due to difficulties in tak- ing proper account of the coupling Hamiltonian and the presence of strong hybridization [6–8, 29]. These prob- lems are naturally avoided within the Landauer-Buttiker formalism. Since this formalism considers fully coherent unitary dynamics in both the leads and the scattering re- gion, the von-Neumann entropy associated with the scat- tering states is conserved in a scattering event. Hence, an additional inflow of entropy is reflected in an increased entropy s stored in the scattering region. As a result, the entropy is source-free ds dt + I S tot = 0 . (24) We can use this continuity equation and Eq. (22) to infer the evolution of s. Invoking energy and particle conser- vation, we identify Q = −I Q tot as the heat leaving the scattering region from the inside perspective. Thus, the entropy evolution can be expressed in terms of the ther- modynamic functions of the (strongly) coupled subsys- tem as ds dt = Q T + W (2) T . (25) 4 Therefore, dissipation leads to a local increase of entropy, which is provided by the scattered electrons. This con- stitutes the inside-outside duality of entropy evolution. Integrated over a full cyclic transformation of X, the entropy current needs to vanish, as it derives from a source-free thermodynamic state function, see Eq. (24). Averaged over a cycle, Eq. (22) thus implies that all extra energy pumped into the scattering region W (2) eventu- ally has to be released as heat into the leads I Q(2) tot = W (2) . (26) Application to the resonant level model.−To empha- size the advantage of the outside approach over calcu- lating the thermodynamic functions of a subsystem di- rectly, we connect here to the thermodynamics of the resonant level model derived earlier from an inside per- spective. This model consists of a single localized elec- tronic level HD = εd(t)d†d, which can be changed in time by an external agent. It is coupled to a free elec- † kck via a coupling Hamiltonian (cid:0)Vkd†ck + h.c.(cid:1) and was intensively studied in k
kc tron metal HB =(cid:80) HV =(cid:80) k the past [6, 8], with difficulties in Ref. [29] pointed out and overcome in Ref. [7]. The inside approach demands a splitting of the cou- pling Hamiltonian HV between effective system and bath, which strongly limits its applicability to the resonant level model in the wide band limit of energy-independent hybridization [7, 8]. In contrast, the here developed out- side approach yields the strong coupling thermodynamics for arbitrary non-interacting electron systems and fur- thermore reproduces the results for the resonant level. Deriving the distribution matrix φ for this model explic- itly, we show order by order that both the heat current I Q in Eq. (19) and the entropy current I S tot in Eq. (18) exactly reproduce the absorbed heat Q = −I Q tot and en- tropy change s = −I S tot from the inside perspective [7] (see Supplemental Material). Thereby we also explicitly confirm the inside-outside duality of entropy evolution: W (2) was shown to lead to a lo- The dissipated power cal increase of entropy for the resonant level in Ref. [7], and we demonstrate here that this is reflected in an ad- ditional inflow of entropy I S tot carried by the scattering states, leaving the entropy source-free, Eq. (24). Conclusion.−We developed a Landauer-Buttiker ap- proach to entropy evolution in strongly coupled fermionic systems, which considers fully coherent quantum dy- namics in combination with coupling to macroscopic equilibrium baths. This formalism naturally avoids the system-bath distinction and is applicable to arbitrary non-interacting electron systems. We showed that the en- tropy current generated by a dynamic scatterer depends on the correlations between different scattering channels, which are generated in the scattering event. At qua- sistatic order, the entropy current is just the heat current over temperature, while at next order the dissipation in- duced by the finite velocity transformation yields a net inflow of entropy into the scattering region. This inflow reflects the dissipation-induced local increase of entropy constituting the inside-outside duality of entropy evolu- tion. Acknowledgments.−We acknowledge fruitful discus- sions with A. Nitzan. This work was supported by the Deutsche Forschungsgemeinschaft (SFB 658), the Ger- man Academic Exchange Service (DAAD), and the Con- selho Nacional de Pesquisa e Desenvolvimento (CNPq). [28] M. Thomas, T. Karzig, and S. Viola Kusminskiy, Phys. Rev. B 92, 245404 (2015). [29] M. Esposito, M. A. Ochoa, and M. Galperin, Phys. Rev. Lett. 114, 080602 (2015). [30] J. B. Pendry, J. Phys. A: Math. Gen. 16, 2161 (1983). [31] E. T. Jaynes, Phys. Rev. 106, 620 (1957). [32] I. Peschel, J. Phys. A 36, L205 (2003). [33] See Supplemental Material. [34] M. Moskalets and M. Buttiker, Phys. Rev. B 69, 205316 (2004). [35] I. Peschel, Brazilian J. Phys. 42, 267 (2012). 5 [1] J. Millen and A. Xuereb, New J. Phys. 18, 011002 (2016). [2] J. P. Pekola, Nat. Phys. 11, 118 (2015). [3] C. Jarzynski, Annu. Rev. Condens. Matter Phys. 2, 329 (2011). [4] U. Seifert, Rep. Prog. Phys. 75, 126001 (2012). [5] P. Hanggi, G.-L. Ingold, and P. Talkner, New J. Phys. 10, 115008 (2008). [6] M. F. Ludovico, J. S. Lim, M. Moskalets, L. Arrachea, and D. S´anchez, Phys. Rev. B 89, 161306 (2014). [7] A. Bruch, M. Thomas, S. Viola Kusminskiy, F. von Op- pen, and A. Nitzan, Phys. Rev. B 93, 115318 (2016). [8] M. A. Ochoa, A. Bruch, and A. Nitzan, Phys. Rev. B 94, 035420 (2016). [9] R. Landauer, IBM J. Res. Dev. 5, 183 (1961). [10] A. B´erut, A. Arakelyan, A. Petrosyan, S. Ciliberto, R. Dillenschneider, and E. Lutz, Nature 483, 187 (2012). [11] J. V. Koski, T. Sagawa, O.-P. Saira, Y. Yoon, A. Kutvo- nen, P. Solinas, M. Mottonen, T. Ala-Nissila, and J. P. Pekola, Nat. Phys. 9, 644 (2013). [12] J. M. R. Parrondo, J. M. Horowitz, and T. Sagawa, Nat. Phys. 11, 131 (2015). [13] M. Campisi, P. Talkner, and P. Hanggi, Phys. Rev. Lett. 102, 210401 (2009). [14] S. Hilt and E. Lutz, Phys. Rev. A 79, 010101 (2009). [15] M. Esposito, K. Lindenberg, and C. Van den Broeck, New J. Phys. 12, 013013 (2010). [16] P. Strasberg, G. Schaller, N. Lambert, and T. Brandes, New J. Phys. 18, 073007 (2016). [17] M. Perarnau-Llobet, A. Riera, R. Gallego, H. Wilming, and J. Eisert, New J. Phys. 18, 123035 (2016). [18] M. Buttiker, Phys. Rev. B 46, 12485 (1992). [19] P. W. Brouwer, Phys. Rev. B 58, R10135 (1998). [20] M. Moskalets and M. Buttiker, Phys. Rev. B 66, 35306 (2002). [21] M. Moskalets and M. Buttiker, Phys. Rev. B 70, 245305 (2004). [22] J. E. Avron, A. Elgart, G. M. Graf, and L. Sadun, Phys. Rev. Lett. 87, 236601 (2001). [23] C. W. J. Beenakker, C. Emary, M. Kindermann, and J. L. van Velsen, Phys. Rev. Lett. 91, 147901 (2003). [24] P. Samuelsson, E. V. Sukhorukov, Phys. Rev. Lett. 92, 026805 (2004). and M. Buttiker, [25] N. Bode, S. Viola Kusminskiy, R. Egger, and F. von Oppen, Phys. Rev. Lett. 107, 036804 (2011). [26] N. Bode, S. Viola Kusminskiy, R. Egger, and F. von Oppen, Beilstein J. Nanotechnol. 3, 144 (2012). [27] M. Thomas, T. Karzig, S. V. Kusminskiy, G. Zar´and, and F. von Oppen, Phys. Rev. B 86, 195419 (2012). Supplemental Material Calculation of the outgoing distribution matrix in the gradient expansion 6 In the following we derive the adiabatic expansion for the outgoing distribution matrix Eq. (4) of the main text. With the expression of the outgoing operators b in terms of the incoming ones a via exact scattering matrix of the time-dependent problem S, Eq. (2) of the main text, we obtain † (cid:104)b β (
2) bα (
1)(cid:105) = (cid:88) (cid:90) d
3 (cid:90) d
4 2π 2π γδ (cid:10)S∗ γ(
3)Sαδ(
1,
4)aδ(
4)(cid:11) . βγ(
2,
3)a† We use that the incoming scattering states are uncorrelated equilibrium channels and get (cid:104)b † β (
2) bα (
1)(cid:105) = 2π γ † i (
1)aj(
2)(cid:105) = δij2πδ(
1 −
2)fi(
1) (cid:104)a (cid:88) (cid:90) d
3 Sαγ(
1,
3) fγ(
3,
4)S† γβ(
4,
2) with fγ(
3,
4) ≡ 2πδ(
3 −
4) f (
3). The Wigner transform of a convolution (cid:26) (cid:90) d
4 (cid:90) d
3 2π 2π takes the form of a Moyal product of Wigner transforms where C (ε, t) ∗ D (ε, t) = C (ε, t) exp G (
1,
2) = C(
1,
3)D(
3,
2) (cid:104) i 2 (cid:16)← ∂ ε φout αβ (
, t) = D (ε, t). Hence, we get ∂ t (cid:126)∂ε (cid:17)(cid:105) G (
, t) = C (ε, t) ∗ D (ε, t) (cid:126)∂t− ← (cid:90) d
e−iεt(cid:68) (cid:104)Sαγ(
, t) ∗ fγ(
, t) (cid:88) 2π b (cid:105) ∗ S† = † β(
−
/2)bα(
+
/2) γβ(
, t) . (cid:69) γ (cid:90) d
S(
, t) = e−i
tS(
+
2 ,
−
2 ) 2π (cid:27) , (27) (28) (29) (30) (31) (32) Expanding the exponential gives the different orders of velocity, which is called the gradient expansion. The Wigner transform of the incoming distribution in channel γ fγ(
3,
4) = 2πδ(
3 −
4) fγ(
3) is just the Fermi function of the associated reservoir fγ(
, t) = fγ(
). The Wigner transform of the full scattering matrix (33) (34) can be written as an expansion in powers of velocity (assuming X = 0)[25, 34] S(
, t) = St(
) + XAt(
) + X 2Bt(
) , In the setting of a quantum dot HD = (cid:80) Hamiltonian HT = (cid:80) where S is the frozen scattering matrix, A is the A-matrix, its first order correction, and B is its second order correction. All these matrices depend parametrically on time and from now on we drop their energy and time labels for better readability. The second order contribution to the scattering matrix B never contributes to the distribution matrix up to second order in X in absence of a bias fα = f∀α, as we show below. nhn,n(cid:48)(X)dn(cid:48) coupled to leads HL = (cid:80) ηcη via a coupling ηWηndn + h.c., the frozen scattering matrix S can be expressed in terms of the frozen retarded Green's function of the of the quantum dot GR and the coupling matrices W between the dot and the attached leads n,n(cid:48) d† η
ηc† η,n c† S = 1 − 2πiνW GRW † , where ν is the density of states in the leads. This formula is called the Mahaux-Weidenmueller formula. In the case of an energy-independent hybridization the A-matrix can be written as [25] A = πνW(cid:0)∂
GRΛX GR − GRΛX ∂
GR(cid:1) W † , where ΛX = ∂h(X)/∂X. (35) (36) At zeroth order only the frozen scattering matrix in the zeroth order gradient expansion contributes Zeroth order (cid:88) γ φout(0) αβ = SαγfγS † γβ . (cid:88) α = f∀α, this simplifies to γβ = f (cid:2)SS†(cid:3) SαγS † αβ . φout(0) αβ = f In the absence of voltage and temperature bias f in α = f∀α, we can order by order simplify the expression for the outgoing distribution matrix φout For f in αβ by invoking the unitarity of the scattering matrix at the corresponding order. At zeroth order this condition is just the unitarity of the frozen scattering matrix γ SS† = I , (39) which leads to the zeroth order outgoing distribution matrix diagonal in channel lead space, identical to the incoming distribution matrix φout(0) αβ = δαβf = φin αβ . (40) We derive the unitarity conditions for the first and second order below in Sec. . First order At first order there are contributions both from the zeroth order gradient expansion with the first order correction in Eq. (34), and the first order gradient expansion with the frozen scattering matrix S, which is to S(
, t), simplified by the fact that the incoming distribution function fγ has no time dependency. We obtain XAX t 7 (37) (38) (41) (42) (43) φout(1) αβ = X AαγfγS (cid:20) (cid:88) (cid:110) (cid:110)−∂X Sαγ∂
fγS ∂
Sαγ∂X S γ + + i 2 i 2 † γβ + SαγfγA † γβ † γβ − ∂X Sαγ∂
S † γβ (cid:111) fγ † γβ + Sαγ∂
fγ∂X S † γβ (cid:111)(cid:21) , (cid:20) (cid:88) (cid:88) γ i 2 γ + ∂
f AαγS † γβ + SαγA † γβ + i 2 ∂
Sαγ∂X S (cid:110) (cid:111) † γβ + Sαγ∂X S † γβ . (cid:111)(cid:21) † γβ − ∂X Sαγ∂
S † γβ (cid:110)−∂X SαγS † γβ + Sαγ∂X S † γβ (cid:111) φout(1) αβ = X∂
f i 2 = X∂
f i(cid:2)S∂X S†(cid:3) β=LR αβ (cid:110)−∂X SαγS (cid:88) where we used ∂tS = X∂X S. If we assume fα = f∀α we get φout(1) αβ X = f (cid:0)SS†(cid:1) = ∂X I = 0 . where we used ∂X This can be simplified by the unitarity condition at first order Eq. (49) Second order 8 At second order there are contributions by the zeroth order gradient expansion with X 2B in Eq. (34), the first order gradient expansion with XA and the second order gradient expansion with S. Using Eq. (50) we can simplify the expression to φout(2) αβ X 2 = 1 2
f(cid:2)∂X S∂X S†(cid:3) (cid:20) ∂2 αβ + ∂
f i 2 A∂X S† + S∂X A† + X S∂
S† + ∂
S∂2 (cid:0)∂2 i 2 X S† − ∂
∂X S∂X S† − ∂X S∂X ∂
S†(cid:1)(cid:21) . (44) αβ Unitarity condition at different orders (cid:90) d
The unitarity of the full scattering matrix(cid:88) 2π n Smn(
(cid:48),
)S† nk(
,
(cid:48)(cid:48)) = 2πδ(
(cid:48) −
(cid:48)(cid:48))δmk leads to different conditions at each order in velocity. Taking the Wigner transform of this expression leads to (cid:88) n 1δmk = Smn(
, t) ∗ S† nk(
, t) . (45) (46) We insert the adiabatic expansion of the scattering matrix Eq. (34) and consistently collect the terms order by order in the velocity. Zeroth order To zeroth order in the velocity we obtain the unitarity condition for the frozen scattering matrix (cid:88) n δmk = SmnS † nk . (cid:88) (cid:104) (cid:20) n 0 = X 2(cid:88) (cid:32) (cid:16) i 2 − 1 8 + AmnA † nk + SmnB,† nk(
) + BmnS (cid:33) † nk (cid:32) ∂Amn ∂
† ∂S nk ∂X − ∂Amn ∂X † nk ∂S ∂
+ i 2 ∂Smn ∂
† ∂A nk ∂X (cid:33) † ∂A nk ∂
− ∂Smn ∂X (cid:17)(cid:21) ∂2
Smn∂2 X S † nk + ∂2 X Smn∂2
S † nk − 2∂
∂X Smn ∂X ∂
S † nk . (50) First order Up to first order in the velocity Eq. (46) reads (cid:20) (cid:88) (cid:32) n i 2 δmk = + SmnS † nk + XAmnS † nk + XSmnA,† nk(
) ∂Smn ∂
† nk ∂S ∂t − ∂Smn ∂t † nk ∂S ∂
(cid:33)(cid:21) . Using that the frozen scattering matrix S is unitary and ∂tS = X∂X S this yields AX mnS † nk + Smn(
)AX,† nk n n ∂Smn ∂
† ∂S nk ∂X − ∂Smn ∂X † nk ∂S ∂
i 2 (cid:32) (cid:105) = −(cid:88) Second order Up to second order in the velocity one obtains upon using Eqs. (47) and (49) (47) (48) (49) (cid:33) . Application to the resonant level model 9 Here we derive the distribution matrix φ for the resonant level model introduced in the main text. In this case the scattering matrix can be reduced to a single element, the reflection coefficient, which can be obtained via the Mahaux-Weidenmueller formula Eq. (35) Here Γ is the decay rate of the dot electrons into the lead Γ = 2π(cid:80)
−
d(X) + iΓ/2 k Vk2 δ (ε − εk), and the A-matrix Eq. (36) vanishes [26]. The distribution matrix φ only contains a single element describing the occupation of the in- and outgoing scattering channel. The first order contribution φout(1) Eq. (43) takes the form . (51) S(
, t) = 1 − iΓ φout(1) (
, t) = − X∂
f ∂
d ∂X Add , (cid:18) ∂
d (cid:19)2 while the second order Eq. (44) can be simplified to where Add = Γ/(cid:0)[
−
d(X)]2 + Γ2/4(cid:1) is the spectral function of the dot electrons. φout(2) (
, t) = X 2 1 2 ∂2
f dd , ∂X A2 (52) (53) With this we show that the heat current in the leads I Q Eq. (6) of the main text is identical to the heat that leaves the extended level from the inside perspective Q order by order [7] (cid:90) ∞ (cid:90) ∞ −∞ −∞ I Q(1) = −
d I Q(2) = 1 2 ε2 d (
− µ)∂
f Add = − Q(1), (
− µ)∂2
f A2 dd = − Q(2) , d
2π d
2π (54) (55) where we wrote X ∂
d ∂X =
d to directly compare to the quantities in Ref. [7]. The inside-outside duality of entropy evolution Eq. (24) of the main text can be explicitly checked order by order tot Eq. (18) of the main text and compare it to the change of the entropy by inserting φout Eqs. (52) and (53) into I S s of the resonant level in Ref. [7] (cid:90) ∞ (cid:90) ∞ −∞ d
2π −∞ I S(1) = −
d I S(2) = ε2 d d
2π (
− µ) (cid:26) (
− µ) T ∂
f Add = − ds dt (1) 1 2 ∂2
f A2 dd + ∂
f 2T A2 dd T (cid:27) = − ds dt (2) . (56) (57) Thereby we reproduced the results of Ref. [7] with the here developed outside perspective. Entropy current and entanglement The developed form of the entropy current carried by the scattering states allows us to quantify the the correlations between different scattering channels created in the scattering event. If we regard the case of two attached leads (L and R), we can measure the correlations created by the scattering event in terms of the quantum mutual information between outgoing channels to the left and right I(L : R) = SL + SR − Stot , (58) where SL/R is the von-Neumann entropy of the reduced density matrix of the left (right) lead and Stot is the total entropy of the outgoing states, including correlations between L and R. In the case of a pure state of the composite system Stot = 0, and I(L : R) reduces to twice the entanglement entropy F = SL = SR created in the scattering event. as a measure of the correlations created per unit time. Here I M I = I S red,L + I S red,R − I S,out tot , (cid:8)σ[φout(t,
)](cid:9) , I S,out tot = d
2π trc (cid:90) ∞ −∞ (cid:90) ∞ −∞ (cid:110) d
2π trc (cid:111) is the outgoing component of the total entropy current Eq. (15) of the main text. The entropy current corresponding to the reduced density matrix of the electrons in the left (right) lead I S red,L(R) takes the form I S red,L(R) = σ[φout,L(R)(t,
)] , (62) with φout,L(R) being the submatrix of the distribution matrix φout defined on the left (right) subspace only. The simplest case to which we can apply this developed formalism is the case of a static scatterer between two reservoirs at zero temperature with an applied bias voltage, which was investigated as a device to create entangled electron-hole pairs in Ref. [23]. Considering two channels on each side (the authors of Ref. [23] consider a quantum Hall setup, in which the two channels can either represent two spin channels within the same Landau level or two different Landau levels), the scattering matrix, S = , (63) is a 4× 4 matrix, with the 2× 2 submatrices r,r(cid:48), t, t(cid:48) describing the reflection and transmission from the left or right respectively. Here we neglect the energy dependence of the scattering matrix in the bias window and drop the energy labels for better readability. For a static scatterer, the outgoing distribution matrix can be obtained by a simplified version of Eq. (32) φout(
) = S φin(
) S† . (64) Hence in the static case φout is obtained from φin by a unitary transformation given by the frozen scattering matrix S. At zero temperature, the incoming electrons are either fully occupied or empty at each energy fL = Θ(µ + eV −
) and fR = Θ(µ −
), and hence carry no entropy (cid:19) (cid:18) r t(cid:48) t r(cid:48) 10 (59) (60) (61) (65) (66) (67) (68) The reduced density matrix and the corresponding entropy of the outgoing states in the left and right lead can be obtained by a method developed by Peschel [32, 35], which takes the form of the argument presented in the main text, but confined to the subspace of interest. This results in the entropy of the reduced density matrix of subsystem A (cid:0)σ[φA](cid:1) . SA = trc where φA is the submatrix of the full distribution matrix φ, defined on subspace A only. Applied to outgoing scattering states analogous to the derivation in the main text, we obtain from Eq. (58) a mutual information current φout in Eq. (64) then shows that full outgoing entropy current also vanishes (cid:90) ∞ (cid:90) ∞ −∞ −∞ d
2π trc d
2π trc (cid:8)σ[φin(
)](cid:9) = 0 . (cid:8)σ[φout(
)](cid:9) = 0 , I S in tot = I S out tot = since the unitary transformation with the frozen scattering matrix leaves the incoming pure states pure at each energy. For the reduced entropy currents we calculate the outgoing distribution matrix from Eq. (64) φout = Θ(µ −
) I + Θ(µ + eV −
)Θ(
− µ)
, with
= (cid:18) rr† rt† (cid:19) tr† tt† . Inserting φout into the mutual information current Eq. (60) leads to I M I = = 2 eV 2π eV 2π eV 2π trcσ(cid:2)tt†(cid:3) eV 2π trcσ(cid:2)rr†(cid:3) + trcσ(cid:2)tt†(cid:3) = 2 (σ(T1) + σ(T2)) , 11 (69) (70) (71) where T1, T2 ∈ (0, 1) are the eigenvalues of the transmission matrix product t†t = I − r†r, which reproduces the entanglement entropy or entanglement of formation F = I M I /2 calculated in Ref. [23]. 

1005.5088

2

1005

Valley filter in strain engineered graphene

We propose a simple, yet highly efficient and robust device for producing valley polarized current in graphene. The device comprises of two distinct components; a region of uniform uniaxial strain, adjacent to an out-of-plane magnetic barrier configuration formed by patterned ferromagnetic gates. We show that when the amount of strain, magnetic field strength, and Fermi level are properly tuned, the output current can be made to consist of only a single valley contribution. Perfect valley filtering is achievable within experimentally accessible parameters.

cond-mat.mes-hall

cond-mat

Valley filter in strain engineered graphene T. Fujita,1, a) M. B. A. Jalil,2, 1 and S. G. Tan3, 1 1)Computational Nanoelectronics and Nano-device Laboratory, Electrical and Computer Engineering Department, National University of Singapore, 4 Engineering Drive 3, Singapore 117576 2)Information Storage Materials Laboratory, Electrical and Computer Engineering Department, National University of Singapore, 4 Engineering Drive 3, Singapore 117576 3)Data Storage Institute, A*STAR (Agency for Science, Technology and Research) DSI Building, 5 Engineering Drive 1, Singapore 117608 (Dated: 23 April 2018) We propose a simple, yet highly efficient and robust device for producing valley polarized current in graphene. The device comprises of two distinct components; a region of uniform uniaxial strain, adjacent to an out- of-plane magnetic barrier configuration formed by patterned ferromagnetic gates. We show that when the amount of strain, magnetic field strength, and Fermi level are properly tuned, the output current can be made to consist of only a single valley contribution. Perfect valley filtering is achievable within experimentally accessible parameters. I. INTRODUCTION As an atomically thin sheet of carbon atoms, graphene can be thought of as a flexible membrane. Very recently, the prospect of using strain to engineer the electronic properties of graphene1 has opened up new opportuni- ties and directions for graphene research. Strain essen- tially can be considered as a perturbation to the in-plane hopping amplitude, which induces a gauge potential in the effective Hamiltonian.1,2 Previously, strain engineer- ing of graphene has been theoretically applied e.g. in beam collimation1 and the quantized valley Hall effect.3 Experimentally, controllable strain has been induced in graphene via deposition onto stretchable substrates4,5 and free suspension across trenches.6 The valleys in graphene refer to the Dirac cones situ- ated at the six corners of the hexagonal Brillouin zone (BZ), of which there are two inequivalent types labeled K and K ′.7 The valley degree of freedom represents a spin-like quantity, and the study of manipulating and making use of valleys in technology has fittingly been termed valleytronics.8 A prerequisite for this technology is a simple and effective method for preparing valley po- larized currents. A number of such valley filters have been proposed in the literature.8 -- 10 The valley filter in Ref. 8 operates by passing electronic current through graphene nanoconstrictions. One caveat is that the filter is sensitive to the edge profile of the graphene sample, which is difficult to control practically. The filter in Ref. 10, on the other hand, exploits the valley-asymmetry of trigonal warping which is large for energies far away from the Dirac point. Large energy scales, however, enhance intervalley scattering and adversely affect the filter's effi- ciency. Lastly, the filter described in Ref. 9 is effective in bilayer graphene only under intense irradiation by high frequency light. In this letter, we propose an efficient and robust way a)Electronic mail: [email protected] to filter monolayer graphene's valleys through strain en- gineering. Significantly, we found that almost pure valley polarized currents can be attained within experimentally relevant parameters. Moreover, it operates in the low en- ergy limit, in which intervalley scattering is negligible.9 II. THEORY We consider the five-layered graphene device shown in Fig. 1, comprising of two cascaded components; (a) a region of uniform uniaxial strain, followed by (b) a mag- netic field configuration arising from a pair of pattered ferromagnetic (FM) gates. For clarity, the two compo- nents are analyzed separately in Sec. II A and II B, re- spectively. The combined structure is studied in Sec. II C. FIG. 1. (Color) The proposed valley filter comprising of two cascaded structures; (a) a region of length L1 of uniform uni- axial strain (strained bonds are highlighted in red), and (b) a magnetic barrier structure due to patterned FM top-gates separated by length L2. When the two stages are cascaded, it is shown that the output current is highly polarized in one of graphene's valleys. A. Valley-dependent tunneling through strained graphene structures We assume a region (length: L1) of uniform uniax- ial strain along the armchair direction (corresponding to the y-direction) of monolayer graphene as shown in Fig. 1(a). In the presence of strain, the effective Hamilto- ′ F F = vF ~σ · (cid:16)~p + v−1 nian reads HK = vF ~σ · (cid:16)~p − v−1 ~AS(cid:17) for the K-valley, ~AS(cid:17) for the K ′-valley where and HK ′ ~σ(′) = (σx, (−)σy).7 The sign difference between the in- duced gauge potentials ~AS ensures overall time-reversal (TR) symmetry. Explicitly, we have ~AS = δtx,1 where δt parametrizes the strain by its effect on the nearest neigh- bor hopping, t → t + δt (t ≈ 3 eV). To study the trans- mission across the strained region, we start by writing down the electronic wavefunctions in the three regions. Translational invariance along x permits solutions in the K-valley of the form Ψ(x, y) = exp (ikxx)Ψ(y), where e−iφ (cid:19) , e−iϕ (cid:19) , eiφ (cid:19) + Re−iky (cid:18) 1 eiϕ (cid:19) + Be−iqy (cid:18) 1 eiφ (cid:19) , ΨI(y) = eiky (cid:18) 1 ΨII(y) = Aeiqy (cid:18) 1 ΨIII(y) = T eiky(cid:18) 1 x + k2 = E2 = (kx ∓ δt)2 + q2. For the K ′- and k2 valley, we replace φ → −φ, ϕ → −ϕ. The phases φ and ϕ are parametrized as kx = E cos φ and kx ∓ δt = E cos ϕ, where the upper (lower) sign corresponds to val- ley K (K ′), and are related via the conservation of kx; E cos φ = E cos ϕ ± δt (in this paper we assume δt > 0 and E > 0). The transmission coefficient T is solved via wavefunction continuity, and determines the trans- mission probability T = T 2. In Fig. 2 we plot T as a function of φ [assuming δt = 25 meV and L1 = 250 nm], which reveals a remarkable valley-dependence.13 In particular, we find that transmission of opposite valleys is well separated in φ-space for low energies E ≤ δt. The behavior of T can be understood from the following. From k2 x + k2 = E2 we require kx ≤ E for non-vanishing transmission. Similarly, from (kx ∓ δt)2 + q2 = E2 we get −E ± δt ≤ kx ≤ E ± δt. Combining the two inequalities, φ must satisfy −1+δt/E ≤ cos φ ≤ 1 in the K-valley and −1 ≤ cos φ ≤ 1 − δt/E in the K ′-valley. Thus, electron transmission is restricted to valley-dependent windows of φ ∈ (cid:26) [0, arccos(−1 + δt/E)] ; K [arccos(1 − δt/E), π] ; K ′ . (1) Evidently, when δt = 0 (no strain), electron transmission spans the entire spectrum of φ ∈ [0, π] for both valleys. The onset of total separation occurs at E = δt (1), as confirmed in Fig. 2(c). This separation of the two valleys in φ-space forms the basis of our proposed valley filter. In the next section we design a simple structure which filters electrons in φ-space. 2 (b) (b) E=20 meV 0.2 0 φ/π 0.4 (d)(d) E=40 meV 0.6 0.8 1 0.6 0.8 1 (a) (a) E=15 meV K K' issi Transmission of valleys Transmission of valleys f valle separated in separated in φ-space 0.2 0 φ/π 0.4 0.6 0.8 1 (c) E=25 meV Transm on o ys 1 0.8 0.6  0.4 0.2 0 1 0.8 0.6  0.4 0.2 0 1 0.8 0.6  0.4 0.2 0 1 0.8 0.6  0.4 0.2 0 0.2 0 φ/π 0.4 0.6 0.8 1 0.2 0 φ/π 0.4 FIG. 2. (Color) Transmission probability T through the strained region in Fig. 1(a) as a function of φ (see inset of (a)) for various Fermi energies when the strain δt = 25 meV and L1 = 250 nm. T carries a strong valley-dependence; for the K-valley it is denoted by red, solid curves whilst for the K ′-valley it is denoted by blue, dashed curves. The shaded blue regions indicate the action (passable values of φ) of a matched φ-filter, which is discussed in Section II B. B. φ-space filtering via magnetic fields The required φ-filter is hinted from our analysis above. Fig. 2 shows that each of the valleys undergo separate φ- space filtering. The fact that the filtering characteristics are mirrored about φ = π/2 (equivalent to kx → −kx) re- flects the TR symmetry. For the current application, we require a gauge potential ~A which breaks TR, i.e. a mag- netic field. In Sec. II A, ~AS assumed a rectangular func- tion in the region of uniaxial strain; this can be matched by a magnetic vector potential ~A due to a pair of asym- metric, magnetic δ-barriers which can be formed via pat- terned FM gating structures (see Fig. 1(b)).11 For a mag- netic field configuration Bz(y) = B0lB [δ(y) − δ(y − L2)], where lB = p/eB0 and e < 0, the TR-breaking gauge potential reads ~A = B0lB [Θ(y) − Θ(y − L2)] x and en- ters the Hamiltonian in a valley-independent manner as H = vF ~σ(′) ·(cid:16)~p + e ~A(cid:17). Transmission across the magnetic barrier is identical to the red, solid curves in Fig. 2 for matched values v−1 F δt = eB0lB corresponding to a field strength of B0 = 1 e (cid:18) δt vF (cid:19)2 . (2) A strain of δt = 25 meV, for example, requires a field strength of B0 = 1 T. The pass band of the filter when B0 is matched to the strain (2) is shown in Fig. 2 as the blue, shaded regions. Perfect valley filtering can then be expected so long as EF ≤ δt, i.e. there is no mixing of valley currents in φ-space. Before proceeding, we note that the wavevector filtering property of magnetic fields is generic,12 and not limited to δ-type barriers, which are chosen above for simplicity. Wavevector filtering occurs just as effectively in graphene for uniform magnetic fields and other periodic structures.14 -- 16 C. Valley filtering in cascaded structure We analyzed the combined structure in Fig. 1, con- firming our predictions for fully valley polarized currents. The conductance of the device is GK(K ′) = G0Z π 0 dφ sin φT K(K ′) (EF , EF cos φ) , (3) is is the Fermi energy, T (E, kx) where EF the valley-dependent transmission probability and G0 = 2e2/h (EF Lx/vF ) [Lx is the width along x]. We define the valley polarization as 3 (a) (b) P = GK − GK ′ GK + GK ′ , (4) where P ≤ 1. Assuming a device with matched stages (2) and the following parameters; EF = δt = 25 meV, B0 = 1 T, and L1 = d = L2 = 4lB ≈ 100 nm, we obtain for the conductances GK = 0.310G0 and GK ′ = 1.31 × 10−3G0, which results in P > 0.99. These values represent experimentally relevant conditions; mag- netic field barriers of B0 ∼ 1 T may be systematically fab- ricated at these length scales,16 and strains in graphene of ≈ 1% have already been achieved4 (δt = 25 meV cor- responds to ≈ 0.8% strain). III. ANALYSIS AND DISCUSSION We analyze the robustness of P for unmatched filter stages keeping the Fermi level constant. In Fig. 3(a), we plot P for continuous values of the strain δt ∈ [15, 30] meV and magnetic field strength B0 ∈ [0.5, 1.5] T, for fixed EF = 25 meV and L1 = d = L2 ≈ 100 nm. As is evident, P remains large and is robust (> 0.99) for a large window of strain and magnetic field values. This is cou- pled with a significant conductance GK (normalized to G0) as shown in Fig. 3(b). We therefore expect the pro- posed valley filter to be relevant for practical valleytronic applications which demand a very high and robust valley polarization and an appreciable electronic current. The polarity of the filter (K → K ′) can be interchanged in one of two ways. Firstly, one could reverse the direction of the magnetic field Bz → −Bz, which flips the effect FIG. 3. (Color) Variation of (a) valley polarization P and (b) K-valley electronic conductance GK /G0 of the device shown in Fig. 1, for various values of strain δt and magnetic field strength B0. Parameter values used are Fermi energy EF = 25 meV and lengths L1 = d = L2 ≈ 100 nm. of the φ-filter. Alternatively, one may reverse the strain (δt → −δt), which interchanges the curves for K and K ′ in Fig. 2. Finally, it has been shown that strain can in- duce a bandgap in graphene. However, this occurs for large strains of ∼ 10 -- 20% and is due to the shifting of the valleys away from the BZ corners.17 For the small strains used here, we assume a negligible deformation of the valleys. 1V. M. Pereira and A. H. Castro Neto, Phys. Rev. Lett. 103, 046801 (2009). 2K.-I. Sasaki and R. Saito, Prog. Theor. Phys. Suppl. 176, 253 (2008). 3T. Low and F. Guinea, arXiv:1003.2717v1 (2010). 4Z. H. Ni et al., ACS Nano 2, 2301 (2008); T. M. G. Mohiuddin et al., Phys. Rev. B 79, 205433 (2009). 5T. Yu et al., J. Phys. Chem. C 112, 12602 (2008). 6W. Bao et al., Nature Nanotechnology 4, 562 (2009). 7A. H. Castro Neto et al., Rev. Mod. Phys. 81, 109 (2009). 8A. Rycerz, J. Tworzydo, and C. W. J. Beenakker, Nat. Phys. 3, 172 (2007); A. Rycerz, phys. stat. sol. (a) 205, 1281 (2008). 9D. S. L. Abergel and T. Chakraborty, Appl. Phys. Lett. 95, 062107 (2009). 10J. M. Pereira Jr, F. M. Peeters, R. N. Costa Filho, and G. A. 15L. Dell'Anna and A. De Martino, Phys. Rev. B 80, 155416 Farias, J. Phys.: Condens. Matter 21, 045301 (2009). 11A. Majumdar, Phys. Rev. B 54, 11911 (1996). 12A. Matulis, F. M. Peeters, and P. Vasilopoulos, Phys. Rev. Lett. 72, 1518 (1994). 13M. A. H. Vozmediano, M. I. Katsnelson, and F. Guinea, arXiv:1003.5179v1 (2010). 14M. R. Masir, P. Vasilopoulos, and F. M. Peeters, Appl. Phys. Lett. 93, 242103 (2008). (2009). 16S. Ghosh and M. Sharma, J. Phys.: Condens. Matter 21, 292204 (2009). 17V. M. Pereira, A. H. Castro Neto, and N. M. R. Peres, Phys. Rev. B 80, 045401 (2009); R. M. Ribeiro, V. M. Pereira, N. M. R. Peres, P. R. Briddon, and A. H. Castro Neto, New J. Phys. 11, 115002 (2009). 4 

1311.6310

1

1311

Strong plasmon reflection at nanometer-size gaps in monolayer graphene on SiC

We employ tip-enhanced infrared near-field microscopy to study the plasmonic properties of epitaxial quasi-free-standing monolayer graphene on silicon carbide. The near-field images reveal propagating graphene plasmons, as well as a strong plasmon reflection at gaps in the graphene layer, which appear at the steps between the SiC terraces. When the step height is around 1.5 nm, which is two orders of magnitude smaller than the plasmon wavelength, the reflection signal reaches 20% of its value at graphene edges, and it approaches 50% for step heights as small as 5 nm. This intriguing observation is corroborated by numerical simulations, and explained by the accumulation of a line charge at the graphene termination. The associated electromagnetic fields at the graphene termination decay within a few nanometers, thus preventing efficient plasmon transmission across nanoscale gaps. Our work suggests that plasmon propagation in graphene-based circuits can be tailored using extremely compact nanostructures, such as ultra-narrow gaps. It also demonstrates that tip-enhanced near-field microscopy is a powerful contactless tool to examine nanoscale defects in graphene.

cond-mat.mes-hall

cond-mat

Strong plasmon reflection at nanometer-size gaps in monolayer graphene on SiC Jianing Chen1, Maxim L. Nesterov2, Alexey Yu. Nikitin1,2, Sukosin Thongrattanasiri4,5, Pablo Alonso-González1, Tetiana M. Slipchenko2, Florian Speck6, Markus Ostler7, Thomas Seyller7, Iris Crassee8, Frank H. L Koppens9, Luis Martin- Moreno2, F. Javier García de Abajo 4,9, Alexey B. Kuzmenko8, Rainer Hillenbrand1,3,★ 1 CIC nanoGUNE Consolider, 20018 Donostia-San Sebastián, Spain 2 Instituto de Ciencia de Materiales de Aragón and Departamento de Física de la Materia Condensada, CSIC-Universidad de Zaragoza, E-50009, Zaragoza, Spain 3 IKERBASQUE Basque Foundation for Science, 48011 Bilbao, Spain 4 ICREA-Instituci ó Catalana de Recerca i Estudis Av an çats, Barcelona, Spain 5 Department of Physics, Kasetsart University, Bangkok 10900, Thailand 6 Department Physik, Universit ät Erlangen-Nürnberg, 91058 Erlangen, Germany 7 Institut für Physik - Technische Physik, Technisc he Universit ät Chemnitz, 09126 Chemnitz, Germany 8 Département de Physique de la Matie re Condensée, Université de Genève, 1211 Genève, Switzerland 9 ICFO-Institut de Ciéncies Fotoniques, Mediterranean Technology Park, 08860 Castelldefels, Barcelona, Spain ★ Email: [email protected] We employ tip-enhanced infrared near-field microscopy to study the plasmonic properties of epitaxial quasi-free-standing monolayer graphene on silicon carbide. The near-field images reveal propagating graphene plasmons, as well as a strong plasmon reflection at gaps in the graphene layer, which appear at the steps between the SiC terraces. When the step height is around 1.5 nm, which is two orders of magnitude smaller than the plasmon wavelength, the reflection signal reaches 20% of its value at graphene edges, and it approaches 50% for step heights as small as 5 nm. This intriguing observation is corroborated by numerical simulations, and explained by the accumulation of a line charge at the graphene termination. The associated electromagnetic fields at the graphene termination decay within a few nanometers, thus preventing efficient plasmon transmission across nanoscale gaps. Our work suggests that plasmon propagation in graphene-based circuits can be tailored using extremely compact nanostructures, such as ultra-narrow gaps. It also demonstrates that tip- enhanced near-field microscopy is a powerful contactless tool to examine nanoscale defects in graphene. 1 Graphene plasmons are electromagnetic waves propagating along graphene layers1-10. They exhibit a remarkable electrostatic tunability and the ability to strongly concentrate electromagnetic energy, potentially leading to new subwavelength-scale plasmonic and optoelectronic applications. One of the outstanding challenges towards the realization of graphene-based plasmonic circuits is the efficient on-chip manipulation of the plasmonic energy flow in CVD-grown graphene11,12 or in epitaxial graphene on SiC13-16. It has been recently reported that graphene plasmons are strongly reflected not only at graphene edges8,9 but also at defects of extremely subwavelength scale dimensions17 and at grain boundaries18. While uncontrolled plasmon reflections at naturally grown defects may represent major obstacles for the development of graphene plasmonic devices, the controlled structuring of graphene on the 1 nm scale may open in the near future promising avenues for steering plasmons with ultracompact reflecting elements. Regardless of a particular future technological realization, it is interesting from both fundamental and applied perspectives to study how efficient plasmons are reflected when the critical dimensions of reflecting elements are much smaller than the plasmon wavelength. Here we address this question by measuring plasmon reflection from nanometer-size gaps, which are expected to form at terrace steps in so-called quasi-free-standing monolayer graphene (QFMLG)15,19 on the Si-face of SiC. QFMLG is produced by forming a so-called buffer layer with (6√3 ×6 √3)R30 ° periodicity by annealing of the SiC surface, followed by decoupling of that layer through intercalation of hydrogen. This preparation results in significant p-type doping (~ 6 ×10 12 cm-2)19-21, which is essential for the existence of well-defined propagating plasmon modes. More information about the sample preparation is provided in the Methods section. The substrate terraces are atomically flat resulting in high-quality homogeneous graphene areas with a lateral extension of up to a few micrometers. The substrate terraces are separated by nanometer-size steps, which suggests that QFMLG is an interesting system to study graphene plasmon reflection at ultrasmall discontinuities (gaps in the graphene layer) without the need of patterning. Indeed, an intense terahertz absorption peak was recently observed in QFMLG22, indicating a strong influence of terrace steps on its far-field optical properties. In this paper we use scattering-type scanning near-field optical microscopy (s-SNOM)23,24, which was recently employed for interferometric plasmon imaging at graphene 2 edges,8,9 to directly observe plasmon reflection at the terrace steps. With this method we systematically study the plasmon reflection as a function of the step height and quantitatively analyze the data by comparing them with numerical simulations. a b d e ) m n ( t h g i e H G , 4 s / 4 s 2 1 0 1.5 1 0.5 -200 λ 0λ 0 c m n 6 0 ) . u . a ( 1 3 . 0 A B -100 0 Position x (nm) 100 200 Figure 1: Near-field imaging of graphene plasmons on SiC terraces. a) Schematics of the s-SNOM experiment. b) AFM topography image of quasi-free standing monolayer epitaxial graphene on a 6H-SiC substrate. Red arrows indicate fringes along a substrate step. c) Optical near-field amplitude (4th harmonics) at l0 = 9.3 mm recorded in the same area. d) AFM height profile across a step at positions marked as A and B in (b) and (c). e) Normalized near-field amplitude profiles s4/s4,G measured along the same line scans as in (d). s4,G is the near-field amplitude measured on graphene far away from the step. The principle of plasmon imaging with s-SNOM, which is based on atomic force microscopy (AFM), is sketched in Fig. 1a. A metal-coated AFM tip is illuminated with an infrared laser of the wavelength l0 and the backscattered light is recorded as a function of the tip position. In order to suppress background scattering from the tip 3 shaft and the sample, the tip is vibrated vertically and the detected signal is demodulated at a higher harmonic n of the tip vibration frequency, in this work at n = 4. In combination with pseudoheterodyne interferometric detection scheme24 we obtain background-free amplitude and phase signals, s4 and j4, of which we will only use the amplitude for the sake of brevity. Importantly, the tip converts the incident light into a strongly confined near field at the tip apex, which provides the necessary momentum to launch radially emanating plasmons in a graphene layer. When these plasmons are back-reflected at edges or defects, characteristic interference patterns are observed in the near-field images.8,9,18 A typical topography image of one of our samples is presented in Fig. 1b (measurements are obtained at room temperature), showing terraces separated by steps of up to a few nanometers height. In the simultaneously recorded infrared near- field amplitude image (Fig. 1c) we observe several fringes (intensity minima and maxima as shown by red arrows) parallel to the steps. This bears strong resemblance to the recently discovered fringe formation near graphene edges produced by interference between the tip-emitted and back-reflected plasmons8,9. The observation of fringes in Fig. 1c thus indicates that plasmons are launched by the tip, which are strongly back-reflected at the terrace steps. Notably, the main fringe is brightest, as the plasmons decay rapidly in propagation direction (i.e. perpendicular to the steps). The plasmon reflection on both sides of the step is therefore hallmarked by a pair of bright parallel lines (with the step in between) accompanied by a series of weaker ones. We note that the intensity and number of fringes on the two sides can be different, as for example observed in Fig. 1c. The fringes can be more intense on either side of the step. At position B, the more intense fringe is observed at the step edge, while at the position marked by the red arrows the more intense fringe is observed at the step corner. We conclude that the asymmetry of the plasmon fringes is not related to the step geometry. It might be caused by a variation of the electronic properties (for example the carrier mobility) of the graphene monolayer among the different terraces. 4 a 1µm 0 c 15 nm 0 λ =10.15µm 0 1.0 (a.u.) b d ) m n ( t h g i e H G , 4 s / 4 s 0 8 4 0 2 1 0 0 e ) . u . a ( 4 s 3 2 1 0 λ =9.3µm 0 λ =9.7µm 0 λ /2SP λ =10.15µm 0 0 400 800 Position x (nm) -1 Frequency (cm ) 1080 1040 1000 f 600 λ =9.7µm 0 0.8 (a.u.) 1000 ∆SS ∆SR λ =9.7µm 0 1000 Position x (nm) 400 ) m n ( P S 200 λ 100 9.2 9.6 10 Wavelength λ (µm) 0 Figure 2: Graphene plasmon dispersion on SiC substrate. a) AFM topography image of a 4x4 µ m area. Arrows mark selected low-he ight terrace steps (0.4 nm). b, c) Near-field amplitude images of the area shown in (a), recorded at 9.7 mm and 10.15 mm wavelength, respectively. The white rectangle in (b) indicates the position from where the near-field profiles in (e) were extracted. d) Profiles of topographic height (upper graph) and near-field amplitude (lower graph) extracted along the white dashed line in (a) and b. The vertical black dashed lines are guides to the eyes. The lower graph illustrates the definition of the values DSR and DSS, which are used to calculate the fringe visibility V = DSS/DSR. e) Near-field amplitude profiles along the long side and averaged over the short side of the white rectangle in (b). f) Plasmon wavelength as a function of the incident wavelength. Squares represent experimental values extracted from (e). The solid curve shows a calculation assuming a Fermi energy of EF = 0.34 eV. In Fig. 2, we analyze a larger area of another sample, which contains several terraces clearly visible in the AFM topography image (Fig. 2a). Figures 2b,c show the near- field amplitude images taken at two different wavelengths (9.7 and 10.15 µ m). Again, we observe fringes (most clearly the main one) parallel to the substrate steps. Notably, some terraces look much darker than the rest of the sample and do not show any plasmon interference fringes. From Raman images (not shown) we can identify these areas as graphene-free SiC. With increasing wavelength, the fringes broaden and their 5 spacing increases. This is more obvious in Fig. 2e, where we show near-field amplitude profiles perpendicular to a step in the area marked by the white rectangle in Fig. 2b for three different wavelengths (9.3, 9.7 and 10.15 µ m). Because the fringe spacing L corresponds to approximately the half of the plasmon wavelength lSP 8,9, we can plot lSP as a function of l0 (symbols in Fig. 2f). The solid curve corresponds to the theoretical dispersion relation1,4, which in the quasi-static limit (λSP << λ0) reduces l = ple l 22 2 2 , where EF is the Fermi energy with respect to to (2 11 /(( ( )) hc Ee F SP SiC 0 0 ) ) the Dirac point. For the simulations we used EF = 0.34 eV, which is close to the experimental value22. εSiC(λ0) is the dielectric function of SiC, which is strongly wavelength dependent due to the optical phonon in SiC. Note that for the actual doping level, the interband absorption in graphene starts at much higher photon energies (at 2EF ≈ 0.7 eV for vertical optical transition, but it is roughly around EF where the plasmon band enters the interband transitions region), and therefore we can legitimately neglect it in the above formula, which only accounts for the intraband Drude contribution. The experimental points follow rather closely the theoretical plasmon dispersion, corroborating that the fringes are a consequence of plasmon reflection at the steps. It is noteworthy that the amplitude of the main fringe is not the same for all values of the step height, as we conclude upon inspection of both the height and normalized near-field amplitude profiles, S = s4/ s4,G, taken along the white dashed lines in Figs 2a,b, respectively (Fig. 2d). s4,G is the near-field amplitude on graphene measured far away from the step. We also find that the fringe amplitude at graphene edges (i.e. at the boundaries between graphene-covered and graphene-free areas, DSR), is always higher than the fringe amplitude at steps between graphene-covered terraces (DSS). Because the fringe amplitude is nearly the same for all edges, we use DSR as a reference and plot the value V = DSS/DSR, hereafter referred to as fringe visibility, as a function of the step height hs. We show below that V is indicative of the plasmon reflectivity of the steps. The result of the analysis of a large number of different steps is shown in Fig. 3c (symbols). Two observations can be immediately made. First, all steps below 1 nm show a negligible fringe visibility, such as those marked by white arrows in Fig. 2a. Second, for all steps higher than hs > 1.5 nm we observe fringes, and their visibility increases from about 0.2 at hs = 1.5 nm to about 0.5 at hs = 5 nm. 6 Interestingly, for hs ~ 1 to 1.5 nm we find both zero and finite fringe visibility. The observation of two distinct types of fringe visibilities, either negligibly small or larger than 0.2 (indicated by black and red symbols in Fig. 3c, respectively), allows us to speculate that graphene continuously covers steps with a height below 1 nm, while for heights hs > 1.5 nm it is (electrically) disconnected. Because the fringe visibility V increases with increasing step height, we assume that a gap is formed at the step with a gap width corresponding to the step height. This assumption is supported by the recent observation that no buffer layer (and thus no graphene) is formed during the graphitization of nonpolar surfaces25, such as the surfaces of the terrace steps. Interestingly, we observe that some of the fringes are interrupted when the step height is about hs ~ 1.5 nm. This can be clearly seen for example at position A in Fig. 1c. Evidently, plasmons are not reflected at this step location. The step height at position A, however, is the same as at position B (Fig. 1d), where fringes indicate a strong plasmon reflection (Fig. 1e). The absence of plasmon reflection at A suggests that the graphene is continuous at this step location, thus allowing for nearly undisturbed propagation of the plasmons launched by the tip. Subsequently and in line with our observation at steps with hs > 1.5 nm, we attribute the appearance of strong fringes, for example at position B, to a discontinuity of the graphene directly at the step. In order to test our hypotheses and to achieve a more quantitative understanding of the plasmon reflection at the terrace steps, we performed a finite elements-based simulation of the near-field contrast observed in s-SNOM. To that end, the AFM tip is approximated by a conducting ellipsoid illuminated by a plane wave. We calculated the vertical component of the electric field just below the tip, which largely determines the tip scattered field and therefore can be compared with the experimental s-SNOM amplitude (for more details see Supporting Information). The normalized field amplitude, Ez,norm, can be plotted as a function of the lateral position of the ellipsoid. The results of such calculations are shown in Figs. 3a, b. The optical conductivity of graphene, which depends on doping and the relaxation time τ, was calculated within local (i.e., zero parallel wave vector) random-phase the approximation26-28, where we assumed │EF│ = 0.34 eV and τ = 0.05 ps, corresponding to a mobility of 1430 cm2/(V•s), close to the experimental values 19. In Fig. 3a, we show the calculated near-field profile (red solid line) when the tip scans 7 across a graphene edge (also discussed in Refs. 8 and 9), demonstrating an excellent agreement between our simulation and the experimental near-field profiles (symbols). We now apply the simulation model to compute near-field profiles perpendicular to a 1.5 nm high terrace step. We first assume that graphene does not cover the step, that is, there is a 1.5 nm wide vertical gap between two semi-infinite graphene sheets on the two terraces, as illustrated by the upper sketch in Fig. 3c. The calculated near-field profile (red solid line in Fig. 3b) exhibits strong near-field variations on both sides of the step, which resemble the experimental data (for example the red curve in Fig. 1e). In the second calculation, we assumed that graphene is continuous over the step, and for simplicity, we neglect any possible inhomogeneities near the step. The corresponding near-field profile (Fig. 3b, black curve) only exhibits mild signal variations, in agreement with the experimental data at some step positions, such as the black curve in Fig. 1e. The calculations thus confirm that the electrical connectivity of graphene at the step is the key factor determining the plasmon reflection, rather than the geometrical profile of the substrate step. For comparison with the experimental data, we show in Fig. 3c the calculated fringe visibility DSS/DSR as function of the step height hs. We consider two cases. First, the graphene covers the step (as illustrated in the bottom sketch in Fig. 3c). Second, the vertical section of the step is not covered by graphene (as illustrated in the top sketch of Fig. 3c), that is, the step represents a gap in the graphene with a width corresponding to the step height. We find that for all step heights the fringe visibility for discontinuous graphene (red curve in Fig. 3c) is much higher than for continuous one (black curve in Fig. 3c). The experimentally observed increase of V with the step height is qualitatively well reproduced in the simulation, and even quantitative agreement between both is found above hs ≈ 3 nm. At lower step heights, the calculations yield a higher fringe visibility than what is observed in the experiment. The deviation may be due to a possible variation of the carrier density or mobility near the edge, which is not taken into account in the calculations. A future closer study of the electronic properties around the steps can probably improve the agreement between experiment and theory. However, such a study would go beyond the scope of this work and we abstain from making more specific statements in this regard. 8 Graphene SiC b Graphene SiC ∆SR S∆S 0 200 400 Position x (nm) -200 -100 0 100 Position x (nm) 200 1 2 4 3 Step Height (nm) 5 a m r o n , z E 2 1 c R S ∆ / S S ∆ = V 0.8 0.6 0.4 0.2 0 0 d t n e i c i f f e o C n o i t c e l f e R 0.4 0.2 0 0 1 2 3 4 Step Height (nm) 5 Figure 3: Graphene plasmon reflection at SiC steps. a) Experimental (dots) and calculated (red solid line) near-field amplitude profiles across a graphene edge, s4/s4,G and, Ez,norm, respectively. A Fermi energy of EF = 0.34 eV and a relaxation time of 0.05 ps were assumed in the calculations. b) Calculated near-field amplitude profiles, Ez,norm, for l0 = 9.7 mm across a 1.5 nm high step using the same parameters as in (a). The red curve shows the result obtained when graphene is disconnected at the step, whereas the black curve refers to graphene continuously covering the step. In both (a) and (b), the horizontal and vertical dashed lines mark the averaged signal on graphene and the positions of graphene edges or steps, respectively. c) Experimental and theoretical fringe visibility V = DSS/DSR as a function of the step height. Each data point represents a set of measurements on various steps of similar height. Horizontal and vertical error bars correspond to the variation in measured step height and the fringe visibility, respectively. The red curve depicts the calculated V, assuming that graphene is disconnected at the step. The black curve is obtained when graphene covers the step with uniform conductivity. d) Plasmon reflection coefficient as a function of the step height for the two scenarios considered in (c), calculated analogous to ref. 17. It is important to connect the fringe visibility observed in s-SNOM experiments and the plasmon reflection coefficient,17 which cannot be directly measured but is 9 essential for theoretical analysis of plasmon propagation in structured graphene. To this end, we performed another series of calculations, where we consider a plasmon in the form of a plane wave propagating along the graphene sheet towards a step (instead of a radial wave created by the tip as considered above) as it is shown in the inset of Fig. 3d. We calculate the reflection coefficient, defined as the ratio between the intensity of reflected and incident plasmons, as a function of the step height (Fig. 3d) in the same situations as above (connected and disconnected graphene at substrate steps). Similar to the fringe visibility (Fig. 3c), the plasmon reflectivity exhibits a dramatically different behavior in the two scenarios. We observe certain similarity, although not a precise match, between the step height dependence of the two quantities. Therefore, we conclude that the fringe visibility is a reliable indicator of the plasmon reflection coefficient. The most important and striking result of our work is the experimental and theoretical demonstration that an ultranarrow graphene gap of only a few nanometers, which is almost two orders of magnitude smaller than the plasmon wavelength, can act as an efficient plasmon reflector and a reflectivity of about 50% can be achieved with gaps as small as 5 nm. To deeper rationalize this at first glance nonintuitive finding, we present in Fig. 4a and 4b the calculated spatial distribution of both x- (horizontal) and z-component of the electric fields of a plasmon wave reflecting from a graphene edge. Outside the graphene region (or inside a gap in the graphene layer), the antisymmetric field distribution of the vertical component, Ez, immediately cancels along x axis (Fig. 4c, black dashed curve) due to the absence of conductivity or charges. The horizontal component, Ex, does not cancel, however, it strongly decays within a few nanometer distance to the graphene edge along both x- and z- axes (Fig. 4c, green curves), which can be explained by the accumulation of a line charge along the extremely sharp graphene edge induced by the graphene plasmon. Beacuse of the rapid field decay in propagation direction of the plasmon, the induction of a mirror line charge in an adjacent graphene edge is strongly diminished already for gap sizes of a few nanometers. In other words, the capacitive coupling across nanometer-size graphene gaps is weak, and thus the transmission of graphene plasmons. This explains the relatively large reflection of graphene plasmons at nanometer-size gaps. Our results thus suggest that introducing nanoscale discontinuities, either by graphene patterning 10 or by tailoring the substrate topography, opens up promising pathways for the development of ultracompact graphene-based plasmonic devices and circuits. Figure 4: Spatial distribution of the electric field of a plasmon plane wave reflecting at a graphene edge. a) Instant snapshot of the x-component of the real part of the electric field Ex. b) The same as in (a), but for Ez. c), The dashed and solid curves indicate the decay of the fields along the x and z axes respectively. The fields are normalized to their value in the origin. The calculation is made for EF = 0.34 eV and the light wavelength l0 = 9.3 mm, which corresponds to lSP = 188 nm. Apart from its importance in plasmonics, this work has also implications in the field of epitaxial graphene. QFMLG has been shown to have advantages compared to regular epitaxial graphene on a buffer layer14,16. For example, the charge carrier mobility of QFMLG is almost independent of the temperature.20 Buffer-layer- elimination by interface hydrogenation of epitaxial graphene was demonstrated to improve the device performance in graphene field effect transistors (FETs).29 Here we have found direct evidence for the existence of discontinuities at step edges exceeding a critical height of about 1.5 nm. The absence of graphene overgrowth at large steps is in line with the observation that no buffer layer (and thus graphene after the hydrogen intercalation process) is formed during the graphitization of nonpolar surfaces25 to which the step surface belongs. The graphene overgrowth of steps smaller than 1 nm might be explained by the formation of graphene bridges connecting the terrace surfaces, owing to diffusion of carbon atoms accross the steps. Why the formation of these graphene bridges starts at step heights around 1.5 nm and whether it is a specific property of QFMLG on SiC(0001) are open questions, which have to await future 11 studies. We expect that the discontinuities in graphene have a strong impact on the electrical transport across the step edges30 compared to the charge transport within the terraces. Acknowledgements RH acknowledges support by the ERC Starting Grant 258461 (TERATOMO) and the National Project MAT2012-36580 from the Spanish Ministerio de Ciencia e Innovaci ón. M.L.N., A.Y.N., T.M.S. and L.M.M. ackno wledge the Spanish Ministry of Science and Innovation grant MAT2011-28581-C02. ST and F.J.G.A. acknowledge support from the Spanish MEC (contract No. MAT2010-14885). The work of I.C. and A.B.K. was supported by Swiss National Science Foundation (grant No. 200020-140710). F.H.L.K. acknowledges support by the Fundacicio Cellex Barcelona, the ERC Career integration grant 294056 (GRANOP) and the ERC starting grant 307806 (CarbonLight). We acknowledge support by the EC under Graphene Flagship (contract no. CNECT-ICT-604391). Declaration of competing financial interests: R.H. is co-founder of Neaspec GmbH, a company producing scattering-type scanning near-field optical microscope systems. All other authors declare no competing financial interests. Methods Quasi-free-standing graphene was obtained on the silicon face of SiC by graphitization at 1450 oC in Ar atmosphere. The first carbon layer (also called ‘buffer layer’), initially covalently bonded to the substra te, was transformed into p-doped quasi-free standing graphene by hydrogen intercalation at 600 oC15,19,20. This technique results in the average hole concentration close to ~6 ×10 12 cm-2, corresponding to a Fermi energy EF ≈ -0.3 eV with respect to the Dirac point.22 The charge mobility at room temperature in similar samples was found to be close to 1500 cm2/(V· s)19. 12 4 References 1 Hanson, G. W. Dyadic Green's functions and guided surface waves for a surface conductivity model of graphene. Journal of Applied Physics 2008, 103. 2 Vafek, O. Thermoplasma Polariton within Scaling Theory of Single-Layer Graphene. Physical Review Letters 2006, 97, 266406. 3 Wunsch, B., Stauber, T., Sols, F. & Guinea, F. Dynamical polarization of graphene at finite doping. New Journal of Physics 2006, 8, 318. Jablan, M., Buljan, H. & Soljaƒ çi ƒá, M. Plasmonic s in graphene at infrared frequencies. Phys. Rev. B 2009, 80, 245435. 5 Koppens, F. H. L., Chang, D. E. & García de Abajo , F. J. Graphene Plasmonics: A Platform for Strong Light-Matter Interactions. Nano Lett. 2011, 11, 3370-3377. 6 Nikitin, A. Y., Guinea, F., Garcia-Vidal, F. J. & Martin-Moreno, L. Fields radiated by a nanoemitter in a graphene sheet. Phys. Rev. B 2011, 84, 195446. 7 Vakil, A. & Engheta, N. Transformation Optics Using Graphene. Science 2011, 332, 1291-1294. 8 Chen, J. N. et al. Optical nano-imaging of gate-tunable graphene plasmons. Nature 2012, 487, 77-81. 9 Fei, Z. et al. Gate-tuning of graphene plasmons revealed by infrared nano- imaging. Nature 2012, 487, 82-85. 10 Grigorenko, A. N., Polini, M. & Novoselov, K. S. Graphene plasmonics. Nature Photonics 2012, 6, 749-758. 11 Li, X. S. et al. Large-Area Synthesis of High-Quality and Uniform Graphene Films on Copper Foils. Science 2009, 324, 1312-1314. 12 Bae, S. et al. Roll-to-roll production of 30-inch graphene films for transparent electrodes. Nature Nanotechnology 2010, 5, 574-578. 13 Berger, C. et al. Electronic confinement and coherence in patterned epitaxial graphene. Science 2006, 312, 1191-1196. 14 Emtsev, K. V. et al. Towards wafer-size graphene layers by atmospheric pressure graphitization of silicon carbide. Nat. Mater. 2009, 8, 203-207. 15 Riedl, C., Coletti, C., Iwasaki, T., Zakharov, A. A. & Starke, U. Quasi-Free- Standing Epitaxial Graphene on SiC Obtained by Hydrogen Intercalation. Physical Review Letters 2009, 103, 246804. 16 Tzalenchuk, A. et al. Towards a quantum resistance standard based on epitaxial graphene. Nat. Nano. 2010, 5, 186-189. 17 Garcia-Pomar, J. L., Nikitin, A. Y. & Martin-Moreno, L. Scattering of Graphene Plasmons by Defects in the Graphene Sheet. ACS Nano 2013, 7, 4988 –4994. 18 Fei, Z. et al. Electronic and plasmonic phenomena at graphene grain boundaries. Nature Nanotechnology 2013, DOI: 10.1038/NNANO. 2013.197. 19 Speck, F. et al. Quasi-free-standing nature of graphene on H-saturated SiC (0001). Mater.Sci. Forum 2010, 645-648, 629-632 20 Speck, F. et al. The quasi-free-standing nature of graphene on H-saturated SiC(0001). Appl. Phys. Lett. 2011, 99, 112016. 21 Ristein, J., Mammadov, S. & Seyller, T. Origin of Doping in Quasi-Free- Standing Graphene on Silicon Carbide. Physical Review Letters 2012, 108, 246104. 22 Crassee, I. et al. Intrinsic Terahertz Plasmons and Magnetoplasmons in Large Scale Monolayer Graphene. Nano Lett. 2012, 12, 2470-2474. 23 Keilmann, F. & Hillenbrand, R. Near-field microscopy by elastic light scattering from a tip. Philosophical Transactions of the Royal Society of London Series a- Mathematical Physical and Engineering Sciences 2004, 362, 787-805. 13 24 Ocelic, N., Huber, A. & Hillenbrand, R. Pseudoheterodyne detection for background-free near-field spectroscopy. Appl. Phys. Lett. 2006, 89, 101124. 25 Ostler, M. et al. Direct growth of quasi-free-standing epitaxial graphene on nonpolar SiC surfaces. Phys. Rev. B 2013, 88, 085408. 26 Falkovsky, L. A. & Varlamov, A. A. Space-time dispersion of graphene conductivity. European Physical Journal B 2007, 56, 281-284. 27 Hwang, E. H. & Das Sarma, S. Dielectric function, screening, and plasmons in two-dimensional graphene. Phys. Rev. B 2007, 75, 205418. 28 Wunsch, B., Stauber, T., Sols, F. & Guinea, F. Dynamical polarization of graphene at finite doping. New Journal of Physics 2006, 8, 318. 29 Robinson, J. A. et al. Epitaxial Graphene Transistors: Enhancing Performance via Hydrogen Intercalation. Nano Lett. 2011, 11, 3875-3880. 30 Ji, S. H. et al. Atomic-scale transport in epitaxial graphene. Nat. Mater. 2011, 11, 114-119. 14 Supporting Information J. Chen, M. L. Nesterov, A. Yu. Nikitin, S. Thongrattanasiri, P. Alonso- González, T. M. Slipchenko, F. Speck, M. Ostler, Th. Seyller, I. Crassee, F. Koppens, L. Martin-Moreno, F. J. Garc ía de Abajo, A .B. Kuzmenko, R. Hillenbrand In this document, we provide a detailed description of the numerical technique used to model the experimental data presented in the main text. We perform finite-element calculations (using COMSOL software), of the near-field profiles and directly compare them with the experimental profiles extracted from the s- SNOM measurements. Fig. S1. (a) Schematic of the model used in the simulations. (b) The near-field profile corresponding to the graphene edge. (c) Comparison between the model and experiment. The red curve presents our theoretical result, whereas the black curve with dots shows the experimentally measured profile. The parameters taken for the simulations are as follows: wavelength 9.3 mm, SiC refractive index nSiC = 1.684 + 0.0116i, Fermi level EF= -0.34 eV, and relaxation time t = 0.05 ps. A calculation which fully takes into account the sophisticated experimental setup (including an oscillating tip and the far-field demodulated signal1) is not feasible with currently available computers. For this reason, instead of calculating the oscillation of the tip and the demodulation of the signal, we consider a simplified system described by a few parameters as explained below. 15 = - A The model geometry consists of a 3D ellipsoidal metallic tip placed in the vicinity of the structure under study (see schematic in Fig. S1a). We illuminate the structure with a p-polarized plane wave at an angle of incidence of 45 degrees, so that its electric field has a finite component parallel on the tip. The tip is raster scanned along the x-direction at the fixed vertical distance D from the structure surface (being from the physical point of view an "average" distance for the real tip in the experiment). We record the z-component of the zE , at the point located right below the tip apex, at a near field (NF), distance δ. Since the experimentally observed signal presents the field scattered by the tip, we take the recording point close to the tip termination. The black dashed curve in Fig. S1 (a) displays the scan path of the tip across a graphene edge, which is used as a reference system. We record a set of NF ( ) zE x for different values of D and δ. After that, we select a profiles (NFPs) profile which provides the best fitting to the experimental background-free ( ) x . The fitting consists in the following linear transformation (see 4s signal Fig.S1 (b)): ( ) ( ) E x s x z 4 - s B A G where Gs is the measured signal and B is the calculated field amplitude, both in the graphene region of the sample far away from the defects. The parameter A takes into account an average background signal which disappears in the experiment due to the demodulation. We have to subtract this average signal because in the simulations we do not take the demodulation explicitly into account. Notice that the tip-sample separation D affects the value of the constant B in Eq. (A1) and also the visibility of the interference fringes (see Fig. S2a). Figure S2b illustrates that the shape of the , NFPs does not change much with the displacement of the recording point while the level B does. A comparison between theoretical and experimental NFPs is shown in Fig. S1c. One can see that the shape of the NF is perfectly captured by the model. A point to note is that the optical conductivity of graphene in these independent of coordinate x , i.e. calculations is assumed to be ( )x s = const . We find that the best combination for the studied sample is a graphene-tip separation of 12.5 nm and a NFP recording point position placed 2.5 nm below the ellipsoid, so that D = 12.5 nm and δ = 2.5 nm in Fig. S1. Once the constants A, D, and δ are found from this fitting, they are kept unchanged in the analysis of the other studied defect structures. (A1) , 16 Fig. S2. Near-field profiles for different tip-sample separations D and positions d of the point below the tip where the field are recorded. In (a) d = 5 nm, while in (b) D= 20 nm. The parameters taken for the simulations are as follows: wavelength 9.2 mm, SiC refractive index nSiC = 1.7 + 0.0127i, Fermi level EF=-0.4 eV, and relaxation time t = 0.1 ps. It is interesting to mention that the generated NFPs are not very sensitive to the shape of the elongated object (i.e., similar results are obtained for a cone, an elongated ellipsoid, etc.), which affects mainly the saturation constant B. As an example, we demonstrate in Fig. S3 that the NFPs for the cone- and spheroid-shaped tips coincide after performing the procedure specified in the r.h.s of Eq. (A1). Important requirements for the simulated tip should be its size (much larger than the studied defects) and the aspect ratio between longitudinal and transversal lengths. These two constraints guarantee the correct enhancement factor and sharpness of the tip termination. For all the simulations presented both in the manuscript and in this Supporting Information, we use an elongated ellipsoid of 900 nm in length and 100 nm in diameter. Fig. S3. Near field profiles calculated using the conical and spheroidal shapes of the tip for the graphene edge. The parameters taken for the simulations are the same as in Fig. S1. 1 Ocelic, N., Huber, A. & Hillenbrand, R. Pseudoheterodyne detection for background-free near-field spectroscopy. Appl. Phys. Lett. 89, 3 (2006). 17 

1303.6902

2

1303

Transient Charge and Energy Balance in Graphene Induced by Ultrafast Photoexcitation

Ultrafast optical pump-probe spectroscopy measurement on monolayer graphene observes significant optical nonlinearities. We show that strongly photoexcited graphene monolayers with 35 fs pulses quasi-instantaneously build up a broadband, inverted Dirac fermion population. Optical gain emerges and directly manifests itself via a negative conductivity at the near-infrared region for the first 200fs, where stimulated emission completely compensates absorption loss in the graphene layer. To quantitatively investigate this transient, extremely dense photoexcited Dirac-fermion state, we construct a two-chemical-potential model, in addition to a time-dependent transient carrier temperature above lattice temperature, to describe the population inverted electronic state metastable on the time scale of tens of femtoseconds generated by a strong exciting pulse. The calculated transient optical conductivity reveals a complete bleaching of absorption, which sets the saturation density during the pulse propagation. Particularly, the model calculation reproduces the negative optical conductivity at lower frequencies in the states close to saturation, corroborating the observed femtosecond stimulated emission and optical gain in the wide near-infrared window.

cond-mat.mes-hall

cond-mat

Transient Charge and Energy Balance in Graphene Induced by Ultrafast Photoexcitation Junhua Zhang,1 Jorg Schmalian,2 Tianqi Li,3 and Jigang Wang3 1Department of Physics, College of William and Mary, Williamsburg, Virginia 23187, USA 2Institute for Theory of Condensed Matter and Center for Functional Nanostructures, Karlsruhe Institute of Technology, Karlsruhe 76128, Germany 3Ames Lab and Department of Physics and Astronomy, Iowa State University, Ames, Iowa 50011, USA Abstract. Ultrafast optical pump-probe spectroscopy measurement on monolayer graphene observes significant optical nonlinearities. We show that strongly photoexcited graphene monolayers with 35fs pulses quasi-instantaneously build up a broadband, inverted Dirac fermion population. Optical gain emerges and directly manifests itself via a negative conductivity at the near-infrared region for the first 200fs, where stimulated emission completely compensates absorption loss in the graphene layer. To quantitatively investigate this transient, extremely dense photoexcited Dirac-fermion state, we construct a two-chemical-potential model, in addition to a time-dependent transient carrier temperature above lattice temperature, to describe the population inverted electronic state metastable on the time scale of tens of femtoseconds generated by a strong exciting pulse. The calculated transient optical conductivity reveals a complete bleaching of absorption, which sets the saturation density during the pulse propagation. Particularly, the model calculation reproduces the negative optical conductivity at lower frequencies in the states close to saturation, corroborating the observed femtosecond stimulated emission and optical gain in the wide near-infrared window. 3 1 0 2 l u J 0 3 ] l l a h - s e m . t a m - d n o c [ 2 v 2 0 9 6 . 3 0 3 1 : v i X r a Transient Charge and Energy Balance in Graphene Induced by Ultrafast Photoexcitation2 1. Introduction Despite the well-established linear optical properties in graphene, which is marked by a universal absorption A = πα = 2.3% ranging from near-infrared to visible light,[1, 2, 3] significantly less attention has been paid to the ultrafast nonlinear optical properties. Important for future photonic and optoelectronic applications,[4] carrier dynamics in graphene after being driven far out of equilibrium needs to be understood. Still, ultrafast spectroscopy studies are reported recently to show unusual properties.[5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21] Especially, the observation of nonlinear absorption when applying an ultrashort intense laser pulse to monolayer graphene reveals an extremely dense, quasithermal photoexcited-carrier state created by strong pumping on 10 fs time scale and metastable for several tens of femtoseconds, which implies a unique transient electronic state in the important emerging material graphene.[18] (v)2 When strongly driven out of equilibrium by coherent light, the excited carriers subsequently participate in several dynamical processes during relaxing back to its equilibrium. Among them are the carrier decoherence, thermalization, cooling, and electron-hole recombination. If the excitation pulse is short enough, by observing the responses followed from right after the pump, we can identify the typical time scales associated with theses processes. Facilitated by recent ultrafast spectroscopy measurements, certain progress has been made. It is recognized that the ultrafast carrier dynamics in graphene is different from that in metals and semiconductors. First of all, dimensional estimates for Dirac fermions in graphene yield the carrier decoherence time dc ∼ e4 τ −1 ω,[22] which becomes rather short, τdc ∼ 1fs, for pump photon energy on the order of 1eV. Rapid carrier-carrier scattering gives rise to an electronic thermalization time τth on the order of 10 fs.[12, 18] Time-resolved studies observe different cooling time scales with the shortest τc ∼ 100fs from electron-optical phonon coupling and longer electron-hole recombination time τr on the ps time scale.[5, 11, 10, 6, 12, 13, 23, 24, 25] Thus, when applying an ultrashort pump pulse of τp ∼ 10fs, the distinct time scales τth ≪ τc,r entails an intermediate electronic state purely determined by carrier-carrier scattering. Unlike in most semiconductors where τth ∼ 100fs > τp, the extremely short decoherence time and the rapid carrier-carrier scattering quickly deplete the phase space from the neighbourhood of optical excitation to the whole band to fully thermalize the carriers as illustrated in Fig. 1 (b). The analysis of Coulomb interaction between the Dirac-fermionic excitations shows a slow population imbalance relaxation,[22, 26] in particular at high pulse intensity due to the suppression of Auger processes for strong pulse excitation,[25] although there is no gap. Thus for τth < t < τc,r the number of photoexcited carriers in each branch of the Dirac cone decays rather slowly. Therefore, the fast decoherence and thermalization together with slow imbalance relaxation support a unique population-inverted electronic state adiabatically formed after the strong pump excitation, a quasi-stable "hourglass" state on the time scale of some tens of femtoseconds, as shown in Fig. 1 (c). Transient Charge and Energy Balance in Graphene Induced by Ultrafast Photoexcitation3 Figure 1. Schematic illustration of the formation of population inverted electronic state in the intermediate time τth < t < τc,r. (a) Photoexcited carriers generated by ∼10fs pump pulse; (b) The leading scattering processes of photo-excited carriers taking place in several femtoseconds: e + e → e + e, h + h → h + h, e + h → e + h, which quickly establish individual thermalization in electron and hole carriers sharing a common electronic temperature Te due to the electron-hole scattering events. (c) After the internal thermalization, the photoexcited carriers form a population inverted hourglass-like electronic state characterized by two chemical potentials and a common electron temperature. In this article, we focus on this unique transient electronic state and provide details on ultrafast optical pump-probe spectroscopy measurement on monolayer graphene to reveal significant ultrafast optical nonlinearities, including nonlinear absorption saturation and near-infrared stimulated emission. These properties arise from a broadband, inverted Dirac fermion population induced by 35fs pulse excitation. Optical gain emerges and directly manifests itself via a negative conductivity at the near- infrared region for the first 100s of fs, where stimulated emission completely compensates absorption loss in the graphene layer. To quantitatively investigate this transient, extremely dense photoexcited Dirac-fermion state, we construct a simple model of a quasi-thermalized distribution with one electron temperature but two distinct chemical potentials associated with the electron- and hole-band, respectively. We find this transient state associated with high electron temperature Te up to 3000-4000K, which causes a broadband distribution extending to high energy that naturally explains the observed blueshifted component in photoluminescence spectrum.[7, 8, 9] We further explore the phase space capacity and identify the maximal photoexcitation density restricted by phase space filling. and individual chemical potential of each band are calculated. To understand the observed nonlinear optical behavior and the measured large saturation density, we calculate the optical conductivity for this nonequilibrium electronic state. The results show that the available phase space cannot be completely filled but will be saturated at a lower photoexcitation density due to the balance between absorption and emission. This calculated saturation density is in excellent agreement with the experimental value.[18] Most interestingly, our model reproduces the experimental results that the nearly-saturated states created by a high-frequency pump are unstable to a low-frequency pulse through stimulated emission to bring the system Transient Charge and Energy Balance in Graphene Induced by Ultrafast Photoexcitation4 to a lower-level metastable state (illustrated in Fig. 2), resulting in an optical gain phenomenon within the first 100s of fs after photoexcitation. The excellent agreement between theory and experiment further corroborates that our simple model captures the feature of the transient state at early timescale (< 100fs) in the high excitation regime. Meanwhile, the comparison with the measured optical conductivity disfavors the equal-chemical-potential model to describe the transient states in graphene at the fs time scale, an outstanding issue debated in the community. Figure 2. Schematic illustration of (a) the femtosecond stimulated emission and optical gain when applying a low-frequency pulse to a high-density population-inverted state created by an intense high-frequency pump and (b) the resulting metastable low- density state. 2. Experimental Details 2.1. Spectroscopy measurement 3. The experimental setup is shown in Fig. In our experiment, the Ti:Sapphire amplifier with center wavelength 800nm, pulse width 35fs at 1kHz repetition rate is used. This further drives an optical parametric amplifiers tunable with tunable optical pump pulses covering 572-2400 nm allowing for both degenerate and non-degenerate pump/probe spectroscopy. The laser is further split into pump and probe paths. The pump beam, chopped as half harmonic of the laser repetition rate, directly excites the sample. The reflection of the probe beam, together with reference, is fed into an auto- balance detector, and the individual beams as well as the difference between them are picked up by three boxcar integrators. During the measurement, the pump fluence from few µJ/cm2 to mJ/cm2 level is finely controlled. This way we can record pump-induced differential reflectivity changes ∆R/R with ∼40fs time resolution and signal-to-noise ratio down to 5·10−5. Similar experimental setups and details are described elsewhere, e.g., see Refs. [27, 28, 29, 30, 31, 32]. Transient Charge and Energy Balance in Graphene Induced by Ultrafast Photoexcitation5 Figure 3. Experimental Schematics. HWP: half waveplate; PB: polarizing beam splitter, FL: focal length. 2.2. Samples Graphene was prepared from the thermal evaporation of SiC [33] with substrates used in the current experiments 6H-SiC(0001) purchased from Cree, Inc.. The samples were graphitized in UHV (P ∼ 1 × 10−10torr) by direct current heating of the sample to ∼1300°C measured with an infrared pyrometer with reading of the pyrometer adjusted to take account of the graphite emissivity reported in the literature [34]. The sample was not pretreated in a H2 atmosphere within a furnace which is a common practice because it excludes the formation of multi-step heights and easier control of thickness. The layer thickness (whether single layer G1 or bilayer G2) was controlled by the heating rate: faster one-step heating rates (within 2-3 seconds to reach 1300°C) result in large G1 domains while multiple heating steps with a slower rate (30seconds to reach 1350°C) result in samples with large G2 areas. Graphene thickness was identified using contrast thickness [35] and with step heights changes between different regions which were found to be combinations of only two steps ,i.e., 0.25nm (of SiC), and 0.33nm (of graphene ) as explained in Ref.[33]. Fig. 1c in Ref. [18] shows a large 2µm×2µm (left) G1 formed after heating with the fast rate. The atomic scale image is shown to the right with the 1 × 1 Transient Charge and Energy Balance in Graphene Induced by Ultrafast Photoexcitation6 unit cell seen with lattice constant 0.246nm and intensity modulation due to the 6√3 is also seen. The tunneling conditions are -0.5V, 1nA. The high intensity of the modulation and the resolution of the 6 atoms of the graphene ring indicate that this is predominantly G1(in excess of 90% of the area). The detailed growth conditions, characterizations and doping (∼0.4 eV for monolayer) of the obtained epitaxial graphene on SiC are extensively established by our papers [33, 36] and many others in the literature, e.g., [37, 38, 39]. 3. Threshold reflection coefficient and optical gain Considering graphene on a substrate with dielectric constant εs, the amplitude of the reflected and transmitted waves for a normal incident beam follow from Maxwell's equations along with the usual boundary conditions: r = t = 1 − ns − 4πσ (ω) /c 1 + ns + 4πσ (ω) /c 2 1 + ns + 4πσ (ω) /c , . (1) Here σ (ω) is the complex optical conductivity. The common reflection and transmission In case σ = 0 holds that R+T = 1. The presence of a finite conductivity in the graphene sheet leads to absorption coefficients are determined by R = r2 and T = ns(cid:12)(cid:12) t(cid:12)(cid:12) (1 + ns)2 (1 − T − R) . Ag = (2) 1 4 2 . where it is custom[40] to introduce the coefficient (1 + ns)2 /4 such that Ag corresponds to the absorption coefficient of a suspended graphene sheet. Following Ref.[40] we can introduce the reflection of the substrate (for σ = 0) and of the substrate with graphene Rs+g 1 + ns(cid:19)2 Rs =(cid:18) 1 − ns 1 + ns + 4πσ (ω) /c(cid:12)(cid:12)(cid:12)(cid:12) Rs+g =(cid:12)(cid:12)(cid:12)(cid:12) 4 θ (ω − 2µ) − 1 − ns − 4πσ (ω) /c i 2π e2 σeq (ω, T = 0) = (3) (4) . 2 ln(cid:18) ω + 2µ ω − 2µ(cid:19)2! . (5) For the complex optical conductivity of graphene in equilibrium and at T = 0 it holds Near the jump in the optical conductivity at ω = 2µ, the imaginary part of the conductivity has a logarithmic divergence which is smeared out in case of finite temperatures. Since σ (ω) is of order e2/, it holds that σ (ω) /c is of order of the finestructure constant of quantum electrodynamics αQED = e2/ (c) ≈ 1/137 ≪ 1. This allows for an expansion in σ (ω) /c. It follows Rs+g − Rs Rs = 4 n2 s − 1 4π c σ′ (ω) + O(cid:0)α2 QED(cid:1) (6) Transient Charge and Energy Balance in Graphene Induced by Ultrafast Photoexcitation7 Thus, the reflection coefficient to leading order in αQED is fully determined by the real part of the optical conductivity σ′ (ω) = Reσ (ω), the imaginary part only enters at higher orders. For the transmission and absorption coefficients follows in the same limit 4ns T = (1 + ns)2 − 4π σ′ (ω) + O(cid:0)α2 c This yields the result Ag = 8ns (ns + 1)3 QED(cid:1) . Rg+s − Rs Rs of Ref.[40]. 4π c σ′ (ω) + O(cid:0)α2 QED(cid:1) = 4 n2 s − 1 Ag Eq.6 enables us to determine a threshold value for the reflectivity that corresponds to a negative optical conductivity and thus to a behavior with optical gain. From Eq.6 it follows for the reflection after delay time τ : (7) (8) (9) ∆R/R ≡ = Rs+g (τ ) − Rs+g (0) Rs (σ′ (τ ) − σ′ (0)) 4π c 4 n2 s − 1 Using the experimentally established value σ′ (0) = e2/ (4) for the optical conductivity prior to the pulse, it follows that σ′ (τ ) < 0 if ∆R/R < ∆R/Rc where ∆R/Rc = − 4παQED n2 s − 1 . (10) With ns = 2.7 it follows ∆R/Rc = −1.4582%. If for some reason the dielectric constant of the substrate is larger that 2.7, this would only reduce the critical value of ∆R/R and we would only underestimate the regime where σ < 0. Given that our data yield the magnitude of ∆R/R as big as 1.9%, it follows that we have σ < 0 as long as ns > 2.41. In the literature, the uncertainty of ns = 2.7 is ±0.1. The smallest index of SiC is 2.55 in the THz range. These results demonstrate that our conclusion σ < 0 is robust. In Fig. 4, we experimentally determine the existence and value of the threshold ∆R/Rc = −1.4582% for zero conductivity in our sample. This further demonstrates unambiguously that the reflectivity geometry in current sample provides a direct measurement of the real part of conductivity σ of the graphene layer (or absorption), which directly accessing the gain/loss processes. This also demonstrates again our sample is graphene monolayer, consistent with conclusion from STM. Transient Charge and Energy Balance in Graphene Induced by Ultrafast Photoexcitation8 Figure 4. The differential reflectivity determined by the measurements with Rg+s or without Rs graphene monolayer on SiC substrate. The reflection from the zero conductivity in pumped garphene/SiC exactly corresponds to the case of bare SiC substrate. Consequently, the threshold ∆R/Rc for zero conductivity can be directly determined from the curve, which is consistent with value used in the manuscript ∼-1.46%. Using the same reasoning we can relate the reflectivity to the absorption coefficient Rs+g (τ ) = Rs + ns − 1 ns + 1 4 (1 + ns)2 A (τ ) and obtain Ag (τ ) − Ag,0 Ag,0 = Rs+g (τ ) − Rs+g (0) Rs+g (0) n2 s − 1 + 4Ag,0 4Ag,0 · (11) (12) that will be used in our analysis of the density of transient electrons and holes, where Ag,0 = Ag (τ = 0) = παQED. Transient Charge and Energy Balance in Graphene Induced by Ultrafast Photoexcitation9 4. Stimulated infrared emission and optical gain (a) Ultrafast ∆R/R at 1.55 eV pump and 1.33 eV probe for different Figure 5. pump fuences from 480 to 4759µJ/cm2(b) Experimental values (rectangles) vs. theory (line) for peak transient conductivity, showing negative conductivity above a threshold pump fluence. (c) The extracted peak transient transmission as function of the pump fluence clearly shows the positive transmission change, nonlinear saturation, and that the critical vaule ∆T T c (blue line) for zero conductivity indeed occurs. Here we provide a set of pump fluence dependence data at probe photon probe energy at 1.33 eV, as shown in Figs. 5a and 5b. Our conclusions w.r.t. stimulated emission and optical gain are based on the observed negative conductivity in strongly photoexcited graphene, which is fully consistent with the complementary data presented in Ref [18]. In addition, following the similar analysis Eqs. (6) and (7), the differential transmission of our sample can be extracted by the information of the differential reflectivity or the subsequently derived conductivity of the graphene sample. More importantly, there also exists a threshold value for the photoinduced differential transmission ∆T T c that corresponds to a zero optical conductivity, above which the optical gain has to emerge Transient Charge and Energy Balance in Graphene Induced by Ultrafast Photoexcitation10 because of the negative conductivity. ∆T T c = 2παQED ns + 1 (13) From Eq.6 and Eq.7 it follows for the differential transmission after delay time τ : ∆T T (τ ) = ∆R R (τ ) · 1 − ns 2 . (14) The extracted peak transient transmission as function of the pump Fluence, as shown in Fig. 5c, clearly shows the positive transmission change and, mostly critically, that the critical value (blue line) for zero conductivity indeed occurs. While those data for the transmission were obtained indirectly, from our reflectivity measurements, a direct measurement of the transmission would be an important confirmation of our results ideally using large area free standing graphene monolayer samples. 5. Analysis of the density of transient electrons/holes The amplitude of the time dependent absorption A as function of pump fluence can be derived from the measured differential reflectivity by applying the Fresnel equations in thin film limit [40, 41] ∆A(Ip) A0 = ∆Rg+s(Ip) Rg+s n2 s − 1 + 4Ag,0 4Ag,0 , · (15) where Rg+s and ∆Rg+s are the static reflectivity and pump-induced reflectivity changes for the graphene monolayer (g) on the substrate (s) with index ns = 2.7. Ag,0 is the absorption of graphene monolayer without pump, which takes a universal value of Ag,0 = π e2 c ≃ 0.023, as determined by the universal a.c. conductivity σ0 = e2 4 . Here the ∆A(Ip)/A0 is the relative differential absorption of graphene on the substrate: A0 = . Therefore Eq.(15) follows from Eq.(12). The peak amplitude A(Ip) = A0 + ∆A(Ip) gradually diminishes as increasing the pump fluence. From the measured transient saturation curve above, one can extract the density of photoexcited electrons(holes) in graphene after the propagation of a single laser pulse of 35 fs (τp) with pump fluence Ip (ns+1)2 Ag,0, which yields ∆A = ∆Ag Ag,0 A0 4 nex(Ip) = ∞ −∞ dt τp nex(t, Ip) = 1 ω ∞ −∞ dt τp I(t, Ip)A (t) , (16) where I(t, Ip) is the Gaussian pulse envelop I(t, Ip) = Ipq 4 ln 2 such that the total pulse fluence is Ip = ´ ∞ A0(1 + ∆A(t) A0 ), we have dt τp −∞ I(t, Ip). Since A(t) = A0 + ∆A(t) = π exph −4 ln 2 τ 2 p t2i, normalized nex(t, Ip) = I(t, Ip)A0 ω (cid:18)1 + ∆A(t) A0 (cid:19) . (17) Transient Charge and Energy Balance in Graphene Induced by Ultrafast Photoexcitation11 Applied to graphene where τth ≪ τp, A (t) is determined by the adiabatic dependence of the absorption on the pump fluence with Ipartial (t, Ip) = ´ t I(t′, Ip). Consequently, dt′ τp −∞ Eq.17 becomes (18) nex(t, Ip) = I(t, Ip)A0 ω (cid:18)1 + ∆A (Ipartial(t, Ip)) A0 (cid:19) . We determine A (Ipartial) experimentally from the reflectivity data of Fig. 6a, combined with Eq.15, as discussed above. Finally, from Eqs. (15)-(18) we have nex(Ip) = ∞ −∞ I(t, Ip)A0 ω ×(cid:20)1 + ∆Rg+s (Ipartial(t, Ip)) Rg+s n2 s − 1 + 4Ag,0 4Ag,0 · (cid:21) dt τp . (19) The result is shown in Fig. 6b, which clearly shows that using the actual absorption A (t), instead of A0, is crutial to understand the high density regime of fs dynamics in graphene discussed here. (a) The peak △R R peak as function of the pump fluence (black squares) Figure 6. measured by degenerate differential reflectivity at 1.55 eV for the graphene monolayer and the corresponding conductivity change (red solid dots). Blue arrow marks the linear threshold for zero conductivity ∆R/Rc = −1.4582% (see text). Dashed line: dependence (guide to the eyes). (b) The extracted transient fermion density at 40 fs (blue dots), as explained in the text, which is significantly lower than A0Ip/ω obtained from the universal conductivity (open circles), as illustrated in shadow area. 6. Theory 6.1. Model We construct the theoretical model based on the following considerations: the above analysis of the time scales associated with different dynamical processes shows that the Transient Charge and Energy Balance in Graphene Induced by Ultrafast Photoexcitation12 internal thermalization time τth of the electron system is much shorter than the cooling τc and the electron-hole recombination time τr; Simultaneous conservation of momentum and energy for Dirac-fermion scattering indicates the leading scattering processes are e + e → e + e, h + h → h + h, e + h → e + h in strong excitation regime with suppression of Auger processes.[25] For the intermediate time regime τth < t < τc,r, i.e., after absorbing photons but before losing energy to the lattice, photoexcited electrons and holes in graphene quickly establish separate thermal equilibrium through carrier-carrier scattering. This gives rise to sharply separated chemical potentials for the two bands, similar to the case of graphite thin film.[12] Due to the scattering between electron- and hole-carriers, e+ h → e+ h, the whole electronic system obtains a common temperature, the electron temperature that differs from the lattice temperature. Therefore we characterize this intermediate electronic state by two Fermi-Dirac distributions f±(k) = 1 (20) exph ε±(k)−µ± kBTe i + 1 for the upper (+) and lower (−) bands of the Dirac spectrum ε±(k) = ±vk with separate chemical potential µ± but the same electron temperature Te. Note that for the convenience of theoretical calculation, we use the upper- and lower-band electron picture instead of the electron-hole picture, which are related via f+ = fe, f− = 1 − fh and µ+ = µe, µ− = −µh. To solve for electron temperature and chemical potentials, we take into account the following conservation laws: 1) the total number of electrons before and after pump excitation is the same, 2) a pseudo-conservation law due to the slow population imbalance relaxation valid in strong excitation regime: the photoexcited carrier number in the intermediate state stays the same as that of right after the pump excitation, 3) and the above described adiabatic process requires that the absorbed photon energy is kept in the electron system until the formation of the quasi-thermal distribution (20). These three conditions are expressed as ntot = n+ + n− = n0 nex = n+ − n0 + = n0 + + n0 −, − − n−, nexω = u − u0, (21) (22) (23) where ntot = Ntot/L2 represents the total density of electrons in the system, nex = Nex/L2 refers to the density of photoexcited carriers, n± (n0 ±) indicate the electron densities in the intermediate (initially equilibrium) state, and u (u0) represents the intermediate (initial) energy density of the whole electron system while ω is the pump photon energy. Applying the distribution (20) to Eqs. (21)-(23) and taking into account the valley and spin degeneracy in graphene, we obtain the following expressions in terms Transient Charge and Energy Balance in Graphene Induced by Ultrafast Photoexcitation13 of fugacities z0 = e kBT 0 µ± kBTe with initial temperature T 0 e = 300K, µ0 e , z± = e (kBTe)2 δ = = g 2π g 2π (v)2 (cid:20)−Li2(−z+) + Li2(cid:18)− (v)2 (cid:20)−Li2(−z0) + Li2(cid:18)− e )2 (kBT 0 1 z−(cid:19)(cid:21) z0(cid:19)(cid:21) , 1 with δ referring to the initial doping density with respect to the neutrality point, and (24) nex = = g 2π g 2π nexω = as well as e )2(cid:2)−Li2(−z0)(cid:3)(cid:9) e )2(cid:20)−Li2(cid:18)− 1 1 1 (v)2(cid:8)(kBTe)2 [−Li2(−z+)] − (kBT 0 z−(cid:19)(cid:21) − (kBT 0 (v)2(cid:26)(kBTe)2(cid:20)−Li2(cid:18)− z−(cid:19)(cid:21) (v)2 (cid:20)−Li3 (−z+) − Li3(cid:18)− z0(cid:19)(cid:21) . (v)2 (cid:20)−Li3(cid:0)−z0(cid:1) − Li3(cid:18)− (kBTe)3 (kBT 0 e )3 − g π 1 1 g π 1 z0(cid:19)(cid:21)(cid:27) ,(25) (26) where g = 4 is the flavor index taking into account the valley and spin degeneracies, v represents the Fermi velocity v ≈ 1.1×106m/s ( 1 300c) in graphene, and the polylogarithm zn is defined by a power series Lis(z) = P∞ ns . Solving the three equations gives the transient electron temperature Te and the individual chemical potentials µ± = kBTe ln z± at a given photoexcitation density nex with initial temperature T 0 e and initial chemical potential µ0 associated with the equilibrium state before being excited. n=1 To perform numerical calculation, we introduce the dimensionless variables fex ≡ nex ¯n , x ≡ δ ¯n , te ≡ kBTe D , α± ≡ µ± D , Ω ≡ ω D , (27) π . Here we choose Λ such that πΛ2 = 1 with a choice for the upper momentum cutoff Λ to define the energy scale D = vΛ and the density scale ¯n = Λ2 2(2π)2/A0 where A0 = 33/2a2 0/2 is the area of the hexagonal unit cell. Note that these dimensionless units are solely introduced for computational convenience. None of our final expressions depends on the actual values of Λ, D or ¯n, as these quantities cancel in the final results (see for example Eq. (34)). In terms of the dimensionless variables, the equations are expressed as ftot − 1 = x g = 2 g = 1 1 t2 z−(cid:19)(cid:21) e(cid:20)−Li2(−z+) + Li2(cid:18)− z0(cid:19)(cid:21) , e(cid:1)2(cid:20)−Li2(−z0) + Li2(cid:18)− 2(cid:0)t0 e(cid:1)2(cid:2)−Li2(−z0)(cid:3)o e [−Li2(−z+)] −(cid:0)t0 e(cid:20)−Li2(cid:18)− z−(cid:19)(cid:21) −(cid:0)t0 e(cid:1)2(cid:20)−Li2(cid:18)− 1 g 2nt2 2(cid:26)t2 g and fex = = (28) (29) 1 z0(cid:19)(cid:21)(cid:27) , Transient Charge and Energy Balance in Graphene Induced by Ultrafast Photoexcitation14 as well as fex = g Ω(t3 −(cid:0)t0 1 e(cid:20)−Li3 (−z+) − Li3(cid:18)− z−(cid:19)(cid:21) z0(cid:19)(cid:21)). e(cid:1)3(cid:20)−Li3(cid:0)−z0(cid:1) − Li3(cid:18)− 1 (30) In the following analysis, either equation set (24)-(26) or the set (28)-(30) will be employed for convenience. 6.2. Characteristics of the Intermediate Electronic State We carry out our analysis for two cases: undoped graphene, the system at the charge neutrality point, and graphene on SiC substrate with a finite electron doping. It is easy to show that the hole-doped system is symmetric to the electron- doped system. By solving for the electron temperature Te and the chemical potentials µ± at different photoexcitation densities, we demonstrate the characteristics of this intermediate electronic state. i.e. 6.2.1. Neutral System The simplest case is the system at the neutrality point, i.e. δ = 0, possessing particle-hole symmetry. Equation (28) yields z0 = 1 (µ0 = 0) and z+ = 1 z− ≡ z (µ+ = −µ− ≡ µ), i.e., the lower-band chemical potential is always the (29) and (30) we opposite of the upper-band one in the neutral system. From Eq. obtain the expression for the dimensionless temperature te and the relation te = fex h(z) = (cid:16) fex , 12 e)2 π2 2 + (t0 −Li2(−z) !1/2 12(cid:17)3/2 2 + (t0 fexΩ 8 + (t0 e)2 π2 e)3 3 4ζ(3) (31) (32) with h(z) ≡ [−Li2(−z)]3/2 (33) −Li3 (−z) Since h(z) is monotonously increasing with an upper bound 3/√2 in large z limit, it 4 (cid:1)2 . That is to say, there exists a phase implies a maximum value of fex: f max space limit on the photoexcited carrier number: 16πv2 ω2. ex = (cid:0) 3Ω nmax ex = (34) 9 . For instance, at ω = 1.55eV the phase space capacity becomes nmax 1013cm−2. te −→ (f max ex = 9.7779 × In this limit, the electron temperature approaches zero as z → ∞, 4 ln z −→ 0. ln z = 3Ω )1/2 ex Transient Charge and Energy Balance in Graphene Induced by Ultrafast Photoexcitation15 For a 800nm pump with photon energy ω = 1.55eV, solving Eq. (31) and (32), we obtain Te as a function of nex plotted in Fig. 7. T (K) e low density regime n ex (10 cm ) 12 -2 n ex (10 cm ) 13 -2 Figure 7. Plot of the electron temperature Te varying with photoexcitation density nex in the neutral system. We can see that the system is rapidly heated up at lower densities, but is slowly cooled at higher densities. It shows that the electron temperature rises rapidly at low photoexcitation densities, but, instead of keeping heated up, it starts slowly dropping at higher densities and eventually approaches zero at a maximal density. The value of µ with respect to nex is plotted in Fig. 8. low density regime n ex (10 cm ) 12 -2 (eV) n ex (10 cm ) 13 -2 Figure 8. Plot of the upper-band chemical potential µ+(nex), in units of eV, in the neutral system. The lower-band chemical potential µ− = −µ+. Clearly, the upper- band chemical potential turns into negative at low densities before rising to positive at high densities. We clearly see that it turns into negative at low photoexcitation densities during the rapidly heating up time, but back to positive when temperature slowly decreasing. The down-turn behavior in electron temperature and the negative-to-positive transition in chemical potential signifies a crossover behavior that at small pump fluence the excited carriers form a hot and dilute classical gas, but with more carriers excited they gradually build up a quantum degenerate fermion system with temperature cooling Transient Charge and Energy Balance in Graphene Induced by Ultrafast Photoexcitation16 down in order to accommodate more electrons in the finite phase space. If the phase space could really be exhausted, the electron and the hole carriers would be pumped into zero temperature Fermi-Dirac distributions in which the carriers are closely packed with a sharp Fermi edge. Inspired by the analysis of the neutral system, we 6.2.2. Exhaustion of Phase Space see that phase space capacity is exhausted at zero electron temperature. To obtain an analytical estimate of the maximal available phase space at different electron-doping levels, we assume the initial temperature to be zero for convenience. Equations (21)-(23) are then simplified as δ = 1 π nmax ex = 1 1 1 π 1 1 π (v)2 h(cid:0)µmax + (cid:1)2 −(cid:0)µmax (v)2 h(cid:0)µmax + (cid:1)2 (v)2 h(cid:0)µmax − (cid:1)2i = −(cid:0)µ0(cid:1)2i = + (cid:1)3 +(cid:0)−µmax (v)2 (cid:0)µ0(cid:1)2 , − (cid:1)2 (v)2 (cid:0)µmax − (cid:1)3 −(cid:0)µ0(cid:1)3i . 2 3π 1 π 1 1 nmax ex ω = , (35) (36) (37) In terms of the dimensionless variables defined in (27), we find a relation between the maximal photoexcitation density and the doping level from the above equations f max ex = ex + x)3/2 + (f max ex 2 3Ωh(f max )3/2 − x3/2i . Solving this equation yields nmax ex at different doping densities as shown in Fig. 9. (38) n (max) ex (10 cm ) 13 -2 13 (10 cm ) -2 Figure 9. Plot of the maximum photoexcited carrier density nmax ex , when the phase space is completely filled, as a function of initial doping density δ. The maximum photoexcitation density decreases with increasing doping level, as expected. As seen from formula (34) for the neutral system, nmax v2 , the available phase space rises with increasing pump frequency. On the other hand, phase space capacity decreases with increasing initial electron doping density, as expected. ex ∼ ω2 Transient Charge and Energy Balance in Graphene Induced by Ultrafast Photoexcitation17 However, for photoexcitation density nex as an input parameter in our calculation, a critical question to ask is: does this maximal density nmax equal the saturation density nsat ex in real pumping process, or in another word, can phase space be completely filled? We will answer this question later. ex 6.2.3. Doped System Next we discuss the system away from the Dirac point with a finite electron doping, i.e., δ ≥ 0 or x = δ ¯n ≥ 0, which is often the case, e.g., in epitaxial graphene on SiC substrate. In this case, from Eqs. (28)-(30) we obtain the expression of te as a function of z+ te = fex 2 + (t0 e)2 [−Li2(−z0)] −Li2(−z+) !1/2 , then the coupled equations are reduced to − Li2(cid:18)− 1 z−(cid:19) = C1 [−Li2(−z+)] , with (cid:20)−Li3 (−z+) − Li3(cid:18)− C1(fex) = 1 − C2(fex) = (cid:16) fexΩ 4 + (t0 1 z−(cid:19)(cid:21)2/3 = C2 [−Li2(−z+)] , x , fex + 2 (t0 e)2 [−Li2(−z0)] e)3(cid:2)−Li3 (−z0) − Li3(cid:0)− 1 e)2 [−Li2(−z0)] fex 2 + (t0 z0(cid:1)(cid:3)(cid:17)2/3 (39) (40) (41) (42) . (43) Solving the two equations (40) and (41) we obtain z+ and z−, which in turn gives t via Eq. (39). Finally, the physical quantities are derived through kBTe = teD, µ+ = teD ln z+, µ− = teD ln z−. To show the numerical results, we choose the experimental system of graphene on SiC substrate with an initial electron doping δ = 1.17 × 1013cm−2, corresponding to an initial chemical potential µ0 = 0.4eV, being excited by the pump energy ω = 1.55eV. The phase space capacity is calculated from Eq. (38) to be nmax ex = 8.34 × 1013cm−2 at this doping level. The electron temperature Te (in units of Kelvin) changing with nex (in units of 1013cm−2) is plotted in Fig. 10. Compared to the undoped system, the evolution of electron temperature with photoexcitation density is smoother and the electron temperature is lower due to the finite initial carrier density. Transient Charge and Energy Balance in Graphene Induced by Ultrafast Photoexcitation18 T (K) e n ex (10 cm ) 13 -2 Figure 10. Plot of the electron temperature Te varying with photoexcitation density nex in epitaxial graphene on SiC substrate with initial chemical potential µ0 = 0.4eV. The non-monotonous behavior remains although the change of temperature with density is smoother compared to the neutral system. Figure 11 shows the upper- and lower-band chemical potential µ+ and µ− (in units of eV) varying with nex (in units of 1013cm−2). low density regime (eV) n ex (10 cm ) -2 9 n ex (10 cm ) 13 -2 Figure 11. Plot of the upper- and lower-band chemical potentials µ+ and µ−, in units of eV, varying with photoexcitation density nex in epitaxial graphene on SiC substrate with initial chemical potential µ0 = 0.4eV. The upper-band chemical potential remains positive although it drops a little bit at low densities, while the lower-band chemical potential drops rapidly at rather low densities followed by increasing separation from the upper-band chemical potential. Clearly, the upper- and lower-band chemical potentials are not symmetric as in the neutral system. Clearly, due to the large initial electron-doping, the low-density classical gas phase for the upper-band electrons is absent now, although there is a tiny presence for the lower band. And in the doped case, the upper- and lower-band chemical potentials are not symmetric, µ+ 6= −µ−, as they are in the neutral system. The separation between the two chemical potentials increases with photoexcitation density. 6.2.4. Broadband Distribution and Blueshifted Photoluminescence A direct conse- quence to the high electron temperature Te ∼ 3000 − 4000K and the slow population Transient Charge and Energy Balance in Graphene Induced by Ultrafast Photoexcitation19 relaxation is a broadband distribution of electron and hole excitations. This can be illustrated in the occupation number Ne,h(ε) = D(ε)fe,h(ε), where D(ε) = 2ε π(v)2 is the density of state at energy ε and fe,h(ε) are the electron and hole distribution functions with fe = f+, fh = 1 − f−. Figure 12 shows the electron and hole distribution at dif- ferent photoexcitation density in the neutral system and in the electron-doped system (µ0 = 0.4eV). occupation ( ) neutral system n ex = 4 10 cm ‰ 13 -2 n ex = 2 10 cm ‰ 13 -2 n ex = 0.5 10 cm ‰ 13 -2 energy (eV) doped system occupation ( ) n ex = 4 10 cm ‰ 13 -2 n ex = 2 10 cm ‰ 13 -2 n ex = 0.5 10 cm ‰ 13 -2 energy (eV) 2 Figure 12. Electron (blue line) and hole (green line) occupation number, in units of π(v)2 , distributed with respect to energy at different photoexcitation density states. We can see the broadband distribution of photoexcited carriers with a high- temperature tail up to almost 2 eV. The high temperature tail in the distribution extends the excited carriers well above the excitation energy 1.55eV up to 2-3eV. This coverage of higher energy states enables emission of photons with higher frequencies than that of the excitation photons, which exhibits blue-shift phenomena in the photoluminescence spectrum. This unusual blueshifted components have been observed in recent experiments[7, 8, 9] and is naturally explained in our model. jk = ev(cid:18)k k σz − k × ez k σy(cid:19) , (45) Transient Charge and Energy Balance in Graphene Induced by Ultrafast Photoexcitation20 6.3. Optical Conductivity Now we are back to the question raised before. In order to identify the saturation density, we need to study the optical responses of this electronic state at different photoexcitation densities. In this section, we calculate the optical conductivity for the nonequilibrium intermediate state using Keldysh technique. By analyzing its behavior at high densities, interesting optical properties are revealed. 6.3.1. General Formalism The low-energy noninteracting Hamiltonian of graphene can be written in the band representation as H0 = v g Xa=1 kXλ=± λkγ† a,λ(k)γa,λ(k) (44) with λ = + or − corresponding to the upper or lower band and γ† a,λ(k), γa,λ(k) are the operators that create or annihilate a quasiparticle of flavor a (spin and valley) at the 2-dimensional wavevector k = (kx, ky) in band λ. In the band representation the current vertex becomes where σy,z are Pauli matrices due to the chiral structure of the Dirac fermions in graphene. Note that we set  ≡ 1 during the derivation, but will recover it in the final results. In Keldysh formalism, the bubble diagram contributing to the optical conductivity gives the real part as Reσαβ(ω) = gπ ω dω′d2k (2π)2 Trhjαk Ak(ω′ + ω)jβk Nk(ω′) − jαk Ak(ω′)jβk Nk(ω′ + ω)i(46) for the direction α(β) = x, y in the 2D graphene layer with the definitions Ak(ω) = Nk(ω) = − i 2π i 2π (cid:16) Gret k (ω) − Gadv G< k (ω). k (ω)(cid:17) , (47) (48) Here the retarded and advanced Green's functions are matrices in band representation Gret/adv k (ω) = 1 ω±i0++¯µ−vk 0 ω±i0++¯µ+vk ! 0 1 (49) where + (−) sign associates with the retarded (advanced) Green's function, and the lesser Green's function is written as G< k (ω) = g< 0 k,+(ω) 0 k,−(ω) ! , g< (50) Transient Charge and Energy Balance in Graphene Induced by Ultrafast Photoexcitation21 g< k,±(ω) = 2πif (ε±(k) − µ±)δ(ω − ε±(k) + ¯µ), (51) 2 (µ+ + µ−) and the Fermi function f (x) ≡ with ¯µ = 1 . Note that distinct chemical potentials are employed in the distribution functions to characterize the nonequilibrium state but an average chemical potential is used in the spectral functions to avoid an artificial modification of the spectrum. ex/(kBTe)+1 1 Previous analysis (see Section 3) shows that the optical properties of graphene on an insulating substrate, such as reflection, transmission, and absorption, are fully determined by the real part of the optical conductivity Reσ(ω) to leading order in the fine-structure constant of quantum electrodynamics αQED = e2 137 ≪ 1. The imaginary part only enters at higher orders. Therefore, we make direct connections of the real part conductivity to the optical responses observable in experiments. c ≈ 1 From Eq.(46) we calculate the longitudinal conductivity which contains intraband and interband transitions. To show the transition processes specifically, introduce ak,λ(ω) ≡ δ(ω − λvk + ¯µ), and fλ(ω) ≡ f (ω − λδµ). It follows the intraband and interband conductivities Reσintra xx (ω) = gπ (ev)2 dω′d2k (2π)2 cos2 θXλ=± Reσinter xx (ω) = gπ (ev)2 dω′d2k k , sin θ = ky k . (2π)2 sin2 θXλ=± where cos θ = kx ak,λ(ω′)ak,λ(ω′ + ω) fλ(ω′) − fλ(ω′ + ω) ,(52) ω ak,λ(ω′)ak,¯λ(ω′ + ω) fλ(ω′) − f¯λ(ω′ + ω) ,(53) ω Intraband Transition First let us evaluate intraband conductivity. 6.3.2. straightforwardly obtained from Eq. (52) It is g (ev)2 0 ∂f (ω) kdk ∂ω (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)ω=vk−µ+ (2)2 δ(ω) ∞ − ln(cid:2)(1 + z+)(1 + z−1 − )(cid:3) kBTeδ(ω) e2  − ∂f (ω) ∂ω (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)ω=−vk−µ−   (54) Reσintra xx (ω) = = where we have recovered the factor  on the last line. The delta-function will be replaced by a Lorentzian δ(ω) → τ −1 ω2+(τ −1)2 for further discussions, which is not our concern here. In equilibrium state z+ = z− = z0 = eβµ0, the intraband conductivity becomes Reσintra e ln(2 + eβµ0 + e−βµ0)i δ(ω) which recovers the well- known expression in the neutral system at equilibrium Reσintra as expected. xx (ω) = 2 ln 2 e2  hkBT 0 xx (ω) = e2  kBT 0 e δ(ω), The most interesting observation from the intraband transition for the transient Transient Charge and Energy Balance in Graphene Induced by Ultrafast Photoexcitation22 electronic state is the modified Drude spectral weight D = e2  ln(cid:2)(1 + z+)(1 + z−1 − )(cid:3) kBTe, (55) which can be significantly enhanced by the high electron temperature. But it could also be reduced when the reduction of chemical potential dominates at low densities in the neutral system. Here we show the Drude weight change, normalized by the equilibrium value D0 = e2 e , at different nex in the neutral and electron- doped systems in Fig. 13.  ln [(1 + z0)(1 + 1/z0)] kBT 0 The results for the neutral system exhibits a drop in Drude spectral weight at low densities due to the large drop in chemical potential as shown in Fig. 8 but quickly followed by large enhancement at higher densities. In the electron-doped system, Drude weight is always increasing but with much less enhancement than in the neutral system. D/D 0 low density regime n ex (10 cm ) 11 -2 neutral system n ex (10 cm ) 13 -2 D/D 0 doped system n ex (10 cm ) 13 -2 Figure 13. Drude spectral weigh D, normalized by the equilibrium value D0, changes with photoexcitation density in neutral and doped (with initial chemical potential µ0 = 0.4eV) system. In neutral system it drops at low densities due to the large drop in chemical potential as shown in Fig. 8, but followed by a large enhancement at higher densities. In the doped system, it is always enhanced but with a much smaller enhancement factor than that of the neutral system. Interband Transition In order to understand the optical response at high 6.3.3. frequencies, as optical conductivity is dominated by interband transition for frequencies on the order of 1eV, we evaluate the interband conductivity from Eq.(53), Reσinter xx (ω) = = g (e)2 ω ω (2)4 hf (− 2 − µ−) − f ( 4kBTe (cid:19) + tanh(cid:18) ω − 2µ+ 4kBTe (cid:19)(cid:21) 2(cid:20)tanh(cid:18) ω + 2µ− 2 − µ+)i e2 4 1 (56) Transient Charge and Energy Balance in Graphene Induced by Ultrafast Photoexcitation23 xx (ω) = e2 where the probe photon frequency ω > 0 and we have reinserted the factor  on the last line. In equilibrium state, µ+ = µ− = µ0, the interband transition becomes Reσinter neutral system in equilibrium Reσinter xx (ω) = e2 response to different photon energies at various photoexcitation densities, unusual optical properties of the transient electronic state are found, which will be discussed in the following. e (cid:17)i which gives the expression for the 4 tanh(cid:16) ω e(cid:17), as expected. By studying the optical 4kBT 0 4 1 2htanh(cid:16) ω+2µ0 e (cid:17) + tanh(cid:16) ω−2µ0 4kB T 0 4kBT 0 Femtosecond Absorption Saturation and Perfect Transparency Let us first consider the optical response to the pump frequency. An interesting observation from the interband transition formula (56) arises due to the two density-dependent chemical potentials. As shown in Fig.8 and 11, with increasing photoexcitation density, the separation of the two chemical potentials also gets larger. Then at a certain photoexcitation density such that µ+ − µ− = ω, (57) the optical conductivity vanishes and the system becomes perfect transparent. For higher densities, µ+−µ− > ω such that optical conductivity turns into negative, which implies a stimulated emission to keep the photoexcitation density from rising. This indicates that the absorption reaches zero and the number of excited carriers saturates at this density nsat ex , which is stabilized by stimulated emission. To show the variation of optical conductivity with photoexcitation density, we again calculate for the neutral system and the electron-doped system with initial doping µ0 = 0.4eV by applying pump photon energy at ω = 1.55eV. We show the results in Fig.14 where they are normalized by the equilibrium value σ0 = e2 4 . Transient Charge and Energy Balance in Graphene Induced by Ultrafast Photoexcitation24 0 s / ) w ( s 0 s / ) w ( s 0 neutral system saturation n ex (10 cm ) 13 -2 doped system saturation n ex (10 cm ) 13 -2 Figure 14. Interband conductivity σ(ω = 1.55eV), normalized by the equilibrium value σ0 = e2 4 , varies with photoexcitation density nex in the neutral and doped systems. It reaches zero, which indicates complete bleaching of absorption, at 5.7 × 1013cm−2 in the neutral system and 4.8 × 1013cm−2 in the doped system, corresponding to the saturation densities. In both cases, the conductivity monotonously decreases due to the increasing electron temperature and separation of chemical potentials. The neutral system ex (theory) = 5.7× 1013cm−2 while the electron-doped system saturates at saturates at nsat roughly nsat ex (theory) = 4.8×1013cm−2. On the other hand, experimental measurement of ex (expt.) = 5.0×1013cm−2, as shown in Fig. 6b, which the electron-doped system gives nsat is in excellent agreement with the theoretical value 4.8 × 1013cm−2. This corroborates the correct description of the transient electronic state in our model. And it also answers the early posted question: the system saturates at a lower photoexcitation density before completely filling the available phase space, i.e., nsat ex < nmax ex . Transient Charge and Energy Balance in Graphene Induced by Ultrafast Photoexcitation25 0 s / ) w ( s 0 s / ) w ( s Loss Gain Loss Gain neutral system n ex (10 cm ) 13 -2 doped system n ex (10 cm ) 13 -2 Figure 15. The calculated interband conductivity σ(ω), normalized by the equilibrium value σ0 = e2 4 , at higher (ω = 1.7eV, green line), the pump (ω = 1.55eV, magenta line), and lower (ω = 1.2eV, blue line) frequency. It shows that in the vicinity of saturation density the optical conductivity at lower frequencies (but still high enough to mainly detect interband transition) becomes negative. This indicates a stimulated emission that enables an optical gain for the low-frequency probes. This theoretical calculation of an optical gain can serve as a test of our model, which have been used to simulate the optical differential reflectivity data performed in Fig. 5b (red line). The agreement is excellent. Femtosecond Stimulated Emission and Optical Gain It is easy to see that higher- frequency pump will saturate at higher density since it can open up more phase space. Then when applying a probe with frequency higher than the pump frequency, we will expect it can not detect the zero absorption, as long as it is still within the low-energy Dirac spectrum, as shown in Fig.15 (green line). However, if one applies a lower- frequency probe, but not too low such that it is still mainly detecting the change in interband transition, one would expect an optical gain in the vicinity of the saturation density, as the optical conductivity becomes negative in this regime as shown in Fig.15 (blue line, yellow region). The appearance of negative conductivity signifies a stimulated emission that drives the system to a lower density as illustrated in Fig.2. We stress that the transient conductivity is negative in a regime below the pump frequency but above a certain frequency below which intraband transition becomes dominant. Transient Charge and Energy Balance in Graphene Induced by Ultrafast Photoexcitation26 Comparison of Two Model Calculations with Pump-probe Spectroscopy Measurements Here we further compare the calculated optical conductivity from the distinct-µ model (µ+ 6= µ−) discussed above and the equal-µ model (µ+ = µ−) with the experimental value measured at probe photon energy ω = 1.55eV and 1.16eV for a pump energy at ω = 1.55eV. As shown in Fig.16, we compare the experimentally-extracted, transient conductivity at 40 fs [18] with the calculated conductivity σ(ω) as a function of the photoexcited carrier density nex for two probe photon energies 1.55 eV and 1.16 eV. The Fermi energy of the sample is ∼ 0.4 eV. The model calculation with the distinct chemical potentials reproduces the salient features of the experiment including nonlinear saturation and optical gain. Excellent agreement between experiment and theory also demonstrates a faithful representation of the transient state at 40 fs by the model described in the manuscript. The model calculation with the same chemical potential clearly fails to account for the experimental observations. For the degenerate scheme, our theory (black dashed line) yields σ → 0 and thus perfect transparency at nex = 4.8 × 1013cm−2. Once the system is driven into this regime, a balance between stimulated emission and absorption will lead to a transparency. For non- degenerate scheme by probing at 1.16 eV, our theory (black solid line) predicts a critical density 3.2 × 1013cm−2 for the transition from loss to gain. All of these results agree quantitatively with the experimental values 5.0 × 1013cm−2 and 3.4 × 1013cm−2, respectively. 0 σ / k a e p σ 1.0 0.5 0.0 0 ∆τ = 40fs, experiment 1.55 eV; 1.16 eV Loss Gain Theory Distinct µ Equal µ 2 1 4 Density (1013carriers/cm2) 3 5 Figure 16. Comparison of the calculation from the distinct-µ model (µ+ 6= µ−) and the equal-µ model (µ+ = µ−) with the experimentally measured transient optical conductivity at 40 fs after the 1.55eV pump at varying pump fluence. The probe photon energies are ω = 1.55eV (blue solid square) and 1.16eV (red solid square). Clearly, the distinct-µ model calculation of the conductivity at 1.55eV (black dashed line) and 1.16eV (black solid line) agrees quantitatively with the experimental data, in sharp contrast to the equal-µ model results (green dashed line for 1.55eV and green solid line for 1.16eV). For the probe photon at lower frequency 1.16eV in the non-degenerate scheme, the transient conductivity becomes negative above a critical density exhibiting a transition from optical loss to gain, predicted by distinct-µ model and substantiated by experiment. Transient Charge and Energy Balance in Graphene Induced by Ultrafast Photoexcitation27 7. Conclusions We have studied the electronic state in photoexcited graphene formed via rapid carrier- carrier scattering after strong photoexcitation but before energy relaxation that takes place on a longer time scale. We have provided evidence for the existence of pronounced femtosecond population inversion and broadband gain in strongly photoexcited graphene monolayers. These results clearly reveal the transient electron and hole potentials are separated on the time scale of 100s of fs. By characterizing the state in terms of two separate Fermi-Dirac distributions with a common electron temperature but distinct chemical potentials for the upper and lower bands, we showed that this intermediate electronic state is associated with high electron temperature Te up to 3000-4000K, which causes a broadband distribution extended to higher energy and is responsible for the observed blueshifted photoluminescence component. The analysis on the variation of electron temperature and chemical potentials with photoexcited carrier density in the neutral system clearly shows a crossover from hot dilute classical gas to dense quantum degenerate fermions. And unlike the phase space restriction in most semiconductors for a pump pulse on the order of 10fs, which is determined by the density of state at the optical excitation and the frequency width of the pulse, the fast depletion of phase space in graphene yields a broadband filling which significantly enlarge the accommodation of photoexcited carriers. Acknowledgments We thank Myron Hupalo and Michael Tringides for discussions. J.Z. acknowledges support by the Jeffress Memorial Trust, Grant No. J-1033. J.S. thanks the DFG Center for Functional Nanostructures. J.W. and T.L. acknowledge support by the the National Science Foundation (contract no. DMR-1055352). Work at Ames Laboratory was partially supported by the U.S. Department of Energy, Office of Basic Energy Science, Division of Materials Sciences and Engineering (Ames Laboratory is operated for the U.S. Department of Energy by Iowa State University under Contract No. DE- AC02-07CH11358). [1] R.R. Nair, P. Blake, A.N. Grigorenko, K.S. Novoselov, T.J. Booth, T. Stauber, N.M.R. Peres, and A.K. Gaim 2008 Science 320, 1308. [2] K.F. Mak, M.Y. Sfeir, Y. Wu, C.H. Lui, J.A. Misewich, and T.F. Heinz 2008 Phys. Rev. Lett. 101, 196405. [3] A.H. Castro Neto, F. Guinea, N.M.R. Peres, K.S. Novoselov, and A.K. Geim 2009 Rev. Mod. Phys. 81, 109. [4] F. Bonaccorso, Z. Sun, T. Hasan, and A.C. Ferrari 2010 Nature Photonics 4, 611. [5] D. Sun, Z.-K. Wu, C. Divin, X. Li, C. Berger, W.A. de Heer, P.N. First, and T.B. Norris 2008 Phys. Rev. Lett. 101, 157402. [6] H. Choi, F. Borondics, D.A. Siegel, S.Y. Zhou, M.C. Martin, A. Lanzara, and R.A. Kaindl 2009 Appl. Phys. Lett. 94, 172102. Transient Charge and Energy Balance in Graphene Induced by Ultrafast Photoexcitation28 [7] C.-H. Lui, K.F. Mak, J. Shan, and T.F. Heinz 2010 Phys. Rev. Lett. 105, 127404. [8] W.-T. Liu, S.W. Wu, P.J. Schuck, M. Salmeron, Y.R. Shen, and F. Wang 2010 Phys. Rev. B 82, 081408(R). [9] R.J. St?hr, R. Kolesov, J. Pflaum, and J. Wrachtrup 2010 Phys. Rev. B 82, 121408(R). [10] J.M. Dawlaty, S. Shivaraman, M. Chandrashekhar, F. Rana, and M.G. Spencer 2008 Appl. Phys. Lett. 92, 042116. [11] P.A. George, J. Strait, J. Dawlaty, S. Shivaraman, M. Chandrashekhar, F. Rana, and M.G. Spencer, 2008 Nano Lett. 8, 4248. [12] M. Breusing, C. Ropers, and T. Elsaesser 2009 Phys. Rev. Lett. 102, 086809. [13] R.W. Newson, J. Dean, B. Schmidt, and H.M. van Driel 2009 Optics Express 17, 2326. [14] Steve Gilbertson, Georgi L. Dakovski, Tomasz Durakiewicz, Jian-Xin Zhu, Keshav M. Dani, Aditya D. Mohite, Andrew Dattelbaum and George Rodriguez 2012 J. Phys. Chem. Lett. 3, 64-68. [15] K. M. Dani, J. Lee1, R. Sharma, A. D. Mohite1, C. M. Galande, P. M. Ajayan, A. M. Dattelbaum, H. Htoon, A. J. Taylor and R. P. Prasankumar 2012 Phys. Rev. B 86, 125403 [16] V. Ryzhii, M. Ryzhii and T. Otsuji 2007 J. Appl. Phys. 101, 083114 [17] S. Boubanga-Tombet, S. Chan, T. Watanabe, A. Satou, V. Ryzhii and T. Otsuji1, 2012 Phys. Rev. B 85, 035443 [18] T. Li, L. Luo, M. Hupalo, J. Zhang, M.C. Tringides, J. Schmalian and J. Wang, 2012 Phys. Rev. Lett. 108, 167401 [19] S Winnerl, et.al. 2013 J. Phys.: Condens. Matter 25, 054202 [20] Torben Winzer, Ermin Malic and Andreas Knorr 2012 arXiv:1209.4833 [21] B. Y. Sun, M. W. Wu 2012 arXiv:1302.3677 [22] L. Fritz, J. Schmalian, M. Muller, and S. Sachdev 2008 Phys. Rev. B 78, 085416. [23] W.-K. Tse and S. Das Sarma 2009 Phys. Rev. B 79, 235406. [24] R. Bistritzer and A. H. MacDonald 2009 Phys. Rev. Lett. 102, 206410. [25] T. Winzer, A. Knorr, and E. Malic 2010 Nano Lett. 10, 4839. [26] M.S. Foster and I.L. Aleiner 2009 Phys. Rev. B 79, 085415. [27] J. Wang, et al. 2004 Physica E 20, 412. [28] G.D. Sanders, et al. 2005 Phys. Rev. B 72, 245302. [29] J. Wang, et al. 2009 Appl. Phys. Lett. 94, 021101. [30] Jigang Wang, et al., 2010 Phys. Rev. Lett. 104, 177401. [31] Jigang Wang, et al., 2004 Journal of Modern Optics, 51, 2771. [32] Li, T. et.al., Nature in press, DOI 10.1038/nature11934 (2013). [33] M. Hupalo, E. H. Conrad, M. C. Tringides, 2009 Phys. Rev. B 80, 041401(R). [34] I. Forbeaux, J.-M. Themlin, and J.-M. Debever 2002 J. Appl. Phys. 92, 2479. [35] C. Riedl, U. Starke, J. Bernhardt, M. Franke , and K. Heinz 2007 Phys. Rev. B 76, 245406. [36] Myron Hupalo, Xiaojie Liu, Cai-Zhuang Wang, Wen-Cai Lu, Yon-Xin Yao, Kai-Ming Ho, Michael C. Tringides, et al., 2011 Advanced Materials, 23, 2082 [37] I. Gierz, C. Riedl, U. Starke, C. R. Ast, K. Kern 2008 Nano Lett, 8, 4603. [38] S. Y. Zhou, G.-H. Gweon, A. V. Fedorov, P. N. First, W. A. de Heer, D.-H. Lee, F. Guinea, A. H. Castro Neto, and A. Lanzara, 2007 Nat. Mater. 6, 770. [39] Isabella Gierz et al. 2010 Phys. Rev. B 81, 235408. [40] Kin Fai Mak, Matthew Y. Sfeir, Yang Wu, Chun Hung Lui, James A. Misewich, and Tony F. Heinz 2008 Phys. Rev. Lett. 101, 196405. [41] Born and E. Wolf, Principles of Optics (Cambridge University Press, Cambridge, UK, 1999). 

0911.4101

2

0911

Dynamic response of a mesoscopic capacitor in the presence of strong electron interactions

We consider a one dimensional mesoscopic capacitor in the presence of strong electron interactions and compute its admittance in order to probe the universal nature of the relaxation resistance. We use a combination of perturbation theory, renormalization group arguments, and quantum Monte Carlo calculation to treat the whole parameter range of dot-lead coupling. The relaxation resistance is universal even in the presence of strong Coulomb blockade when the interactions in the wire are sufficiently weak. We predict and observe a quantum phase transition to an incoherent regime for a Luttinger parameter $K<1/2$. Results could be tested using a quantum dot coupled to an edge state in the fractional quantum Hall effect.

cond-mat.mes-hall

cond-mat

Dynamic response of a mesoscopic capacitor in the presence of strong electron interactions Yuji Hamamoto,1, ∗ Thibaut Jonckheere,2 Takeo Kato,1 and Thierry Martin2, 3 1Institute for Solid State Physics, University of Tokyo, Kashiwa, Chiba 277-8581, Japan 2Centre de Physique Th´eorique, Case 907 Luminy, 13288 Marseille cedex 9, France 3Universit´e de la M´edit´erann´ee, 13288 Marseille cedex 9, France (Dated: June 1, 2018) We consider a one dimensional mesoscopic capacitor in the presence of strong electron interactions and compute its admittance in order to probe the universal nature of the relaxation resistance. We use a combination of perturbation theory, renormalization group arguments, and quantum Monte Carlo calculation to treat the whole parameter range of dot-lead coupling. The relaxation resistance is universal even in the presence of strong Coulomb blockade when the interactions in the wire are sufficiently weak. We predict and observe a quantum phase transition to an incoherent regime for a Luttinger parameter K < 1/2. Results could be tested using a quantum dot coupled to an edge state in the fractional quantum Hall effect. PACS numbers: 85.35.Gv, 73.21.La, 73.23.Hk, 73.43.Jn The dynamical response of mesoscopic conductors consti- tutes a mostly unexplored area of coherent quantum transport, which has recently led to groundbreaking experiments [1]. The mesoscopic capacitor [2] is one of its elementary build- ing blocks: a quantum dot influenced by an AC gate voltage, which is put in contact with an electron reservoir. It has been studied so far at the single electron level, with possible mean field generalizations [3]. Both the capacitance Cµ and the re- laxation resistance Rq, obtained from the low frequency ex- pansion of the admittance G(ω) ≈ −iωCµ + ω2C2 µRq, are fun- damentally affected by the quantum coherence of the device. At zero temperature, a single spin polarized channel yields a relaxation resistance Rq = h/(2e2), which is independent of the dot-reservoir connection. Ref. [1] has confirmed this re- sult for a quantum dot with weak charging energy. However, quantum dots with reduced size exhibit strong Coulomb blockade, and there is also a clear need to analyze whether electron-electron interactions in the lead are relevant. Here, taking rigorously these aspects into account, we prove that there is quantum phase transition from a coherent to an incoherent regime, where a relaxation resistance cannot be defined. For weak interactions, the universal behavior is re- covered even in the presence of strong Coulomb blockade. We consider a quantum dot (Fig. 1) connected to a reser- voir modelled by a Luttinger liquid lead, which allows to ac- count exactly for Coulomb blockade effects. We discuss sepa- rately the absence (Luttinger parameter K = 1) or the presence (K < 1) of interaction in the adjacent lead. This setup and the underlying physics is similar to that studied in Ref. [4], where attention was solely focused on the static occupation of a res- onant level. Here we show that below K = 1/2 the Kosterlitz Thouless type phase transition driven by the dot-lead tunnel- ing strength triggers a transition of dynamical transport from a coherent to an incoherent regime, hence provoking a devia- tion from the universal Rq = h/(2e2). We use a combination of analytical (perturbation theory, renormalization group) and numerical (quantum Monte Carlo) approaches to monitor the capacitance and the relaxation resistance over the whole range of dot-lead connection. The present results can be applied to carbon nanotube quantum wires as well as dots defined in the fractional quantum Hall effect (FQHE). The starting point is the Hamiltonian for a non-chiral, semi- infinite Luttinger liquid [5] where the dot region corresponds to the interval [0, L]: H =Z L −∞ 1 + vF dx ∂x!2 K2 ∂φ ∂x!2 2π  + vF ∂θ + kF L!#2 π2 EC "φ(x = 0) − πCVg e − V cos[2φ(x = 0)] . (1) The first part is the kinetic part, followed by the backscat- tering term at x = 0 (strength V), and finally the contri- bution from the charging energy with EC ≡ e2/(2C) (C is the geometrical capacitance). The canonically conjugated fields φ and θ satisfy the commutation relation [φ(x), θ(x′)] = (iπ/2)sgn(x − x′). V is the backscattering strength on the point contact. EC ≡ e2/(2C) denotes the charging energy The time dependent gate voltage oscillates around Vg. Us- ing the Matsubara imaginary time path integral formulation, the quadratic degrees of freedom away from x = 0 can be integrated out. The kinetic part of the effective action then reads S kin = (πKβ)−1Pωn ωn/(1 − e−2πKωn /∆) φ(ωn)2, where φ(ωn) is the Fourier transform of φ(τ) (now specified at x = 0), and ∆ ≡ πvF /L is the level spacing. The same action can be derived alternatively starting from a single chiral Luttinger liquid "loop", hence the relevance for the FQHE regime [6]. FIG. 1: Left: schematic view of the mesoscopic capacitor: a 1D quantum dot, capacitively coupled to a gate with time-dependent voltage Vg + δVg(t). Right: schematic phase diagram in the degener- ate case. Vc denotes the critical backscattering strength for K = 1/3. Within linear response in the oscillating gate voltage, the (imaginary frequency) admittance can be expressed as: G(iωn) = e2 h 2ωn π Z β 0 dτhφ(τ)φ(0)ieiωnτ. (2) C / m C The dynamical conductance is obtained by analytic continua- tion G(ω) = G(iωn → ω + iδ), while the capacitance reads: Cµ = e2 π2 β[h ¯φ2i − h ¯φi2] ¯φ ≡ 1 β Z β 0 dτφ(τ)! . (3) C / m C We start with a discussion of the weak barrier case, using perturbation theory in V/D (bandwidth D). The capacitance and relaxation resistance are derived as an expansion in orders of V, Cµ = C(0) q + .... Introducing µ + ... and Rq = R(0) q + R(2) µ +C(1) µ +C(2) q + R(1) 0.5 0.4 0.3 0.2 0.1 0 0.5 0.4 0.3 0.2 0.1 0 K = 1 Non-degenerate (N = 0) K = 1/3 Non-degenerate (N = 0) 3.5 3 2.5 2 1.5 1 0.5 100 10 1 2 K = 1 Charge degenerate (N = 1/2) K = 1/3 Charge degenerate (N = 1/2) T = 0.01 T = 0.04 T = 0.08 5 6 7 8 4 V 0 1 2 3 4 V 5 6 7 8 0 1 2 3 FIG. 2: (color online) Capacitance Cµ as a function of the backscat- tering strength V, obtained at temperature T = 0.04 with Monte Carlo computations (dashed lines). The solid lines show the pre- dictions of the perturbative calculations up to second order. We take EC /π2 as the unit of energy and use the parameters D ≡ 2πJ/β = 8π and ∆/(2K2) = π2. Temperature dependence is shown in the bot- tom right panel, where the vertical axis is measured on a logarithmic scale. The charging energy thus does not modify the value of relax- ation resistance, while electron interactions in the lead intro- duce a trivial factor 1/K. At zero temperature, the sums and integration of Eqs. (5)-(7) can be done analytically in certain cases. For example, when EC = 0 and K = 1, one has F+(0) = (∆/(2πD))2 (D ≫ ∆), and one can show that the result for the capacitance is Cµ = (e2/∆)(1 − 2r cos(2πN) + 2r2 cos(4πN)), with r = πV/D. This coincides with the development of the non-interacting formula found in Ref. [1] in powers of the re- flection coefficient r: Cµ = (e2/∆)(1 − r2)/(1 − 2r cos(2πN) + r2). In the more general case of non-zero EC, and K , 1, one has F+(0) ∼ (ET K/(πD))2K, and the integration of Eq. (7) has to be computed numerically. The perturbation theory thus proves the universality of the charge relaxation resistance in the weak barrier limit even in the presence of interactions. To study the non-perturbative regime, the path-integral Monte Carlo method is applied to the action for the discretized path φ(τ = jβ/J) ( j = 0, 1, · · · , J − 1). We estimate thermal average by generating discretized paths using local update in the Fourier space and the clus- ter update [7, 8]. The top (bottom) row of Fig. 2 shows the calculated capacitance Cµ as a function of V at T = 0.04 and K = 1 (K = 1/3). The left and right columns correspond to the non-degenerate case (N = 0) and the charge degenerate case (N = 1/2), respectively. With increasing V the Coulomb staircase becomes sharper, which results in the decrease (in- crease) in Cµ ≡ ∂hQi/∂Vg in the case of N = 0 (N = 1/2). The second-order perturbation theory, shown as solid lines, displays an excellent agreement for small V. Especially, it is remarkable that only for the case of K = 1/3 and N = 1/2 (the right bottom panel of Fig. 2), Cµ exhibits an abrupt increase at a finite V, signaling a possible transition. One can see that Cµ grows as ∝ 1/T in the large barrier region. (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) an = 1 πKβ ωn 1 − e−2πKωn/∆ + ECK π ! , one obtains to zeroth, first and second order: G(0)(iωn) = G(1)(iωn) = − ωn π 1 βan e2 h e2 h G(2)(iωn) = e2 h ωn π where we defined: cos(2πN) ωn π 1 2VpF+(0) 2V2F+(0) β2a2 n 1 I(ωn), β2a2 n F±(v) = exp −Xn ±2 β2an cos(ωnv) dv [cos(4πN)V+(ωn, v) + V−(ωn, v)] I(ωn) = Z β/2 0 V±(ωn, v) = 1 − F±(v) (1 ± cos(ωnv)) ∆ with ET = EC + 2K2 . N = kF L π + eVg 2ET The Pn in F± is limited by D. From Eqs. (5)-(7), the capaci- tance at low temperature becomes: C(0) µ = e2 h π ET C(1) µ = −C(0) µ C(2) µ = C(0) µ π2 ET π2 ET 2V pF+(0) cos(2πN) 2V2F+(0)I(ωn → 0) cos(4πN). It is clear from these expressions that the total capacitance Cµ is a periodic function of N, with period 1. Below we focus on the interval 0 ≤ N < 1. The results for the relaxation resis- tance, at low temperature, are simple since the computation of the first and second order contribution shows that they vanish: R(0) q + R(1) q + R(2) q = R(0) q = h 2e2 1 K . (15) To reveal the origin of the transition behavior, we next examine the strong barrier limit using an instanton method which was developed for the Kondo model [9]. Near the de- generacy point N = 1/2, the configuration of the bosonic field φ can be represented in the dilute instanton gas approximation φ(τ) ≃ π 2n Xj=1 s jΘ(τ − τ j) + π 2 (1 − s), (16) w i ( R where s j = s(−1) j−1, and s ∼ 1 denotes the separation be- tween the well minima (Θ is the step function). Inserting Eq. (16) in the full effective action, the partition function becomes: Z = ∞ 0 t2nZ β Xn=0 × exp 1 dτ2nZ τ2n 0 dτ2n−1 · · ·Z τ2 0 dτ1 2K Xj,k s jsk log τ j − τk τc − uXj (17) , s jτ j where t is the tunneling amplitude between the well minima, and τc is the short-time cutoff. u = (2N − 1)ET denotes the deviation from the degeneracy point. Note the similarity be- tween this partition function and that which was proposed in the context of dissipative Josephson junctions [10]. One can therefore identify the scaling equations [12]: dt dl = 1 − s2 ds2 dl 2K! t, = us(1 − 2t2) dus dl = −4s2t2, (18) (19) which are familiar in the context of a Kosterlitz Thouless transition in the two-dimensional XY model. No further ar- guments are needed when one deviates from the degeneracy point: since t is small, starting from u , 0, Eq. (19) predicts that u will further increase, leading the system further away from the degeneracy point. This means that φ will be trapped in an effective harmonic potential, and one thus recover the re- sult of Eq. (5), which is therefore universal. For the charge de- generate case N = 1/2, the transition corresponds to a Kondo type transition associated with the charge pseudo spin on the dot. Eqs. (18) determine the tendency of the dot-lead trans- mission as temperature is lowered; (t, s2) flows along one of the hyperbolic curves B2−4t2 = const., where B ≡ 1−s2/(2K). For K > 1/2, the tunneling strength always grows upon re- ducing the temperature, and the system reaches the Kondo fixed point where the dot is strongly coupled to the reservoir. An electron freely tunnels in and out of the dot irrespective of the initial tunneling strength. In particular at K = 1 this implies that the charge relaxation resistance is universal, i.e., Rq = h/(2e2), as a consequence of the unitary limit of the un- derlying Kondo model. On the other hand for K < 1/2, there is the possibility that at a critical, sufficiently weak transmis- sion t ("large" V), the RG flow always drives the system into a weak coupling configuration with specified charge. Then the charge fluctuation remains finite, i.e., h ¯φ2i − h ¯φi2 ≃ (πs/2)2, ) ) 2 e K ( / h ( / ) 0 = n 3 20 16 12 8 4 0 5 6 7 8 K = 1/5 1/3 1/1 1 2 3 5 6 7 4 V 6 5 4 3 2 1 0 FIG. 3: (color online) Extrapolated value R(iωn → 0) as a function of V at temperature T = 0.04 in the degenerate case. For D and ∆/(2K2), we use the same parameters as in Fig. 2. Inset: R(iωn → 0) for K = 1/3 at T = 0.01(◦), 0.04(△) and 0.08(⋄) in the vicinity of the location where the crossing occurs. so that the capacitance diverges as ∝ 1/T at low temperatures [see Eq. (3)]. This explains the transition observed for the capacitance in the strongly interacting case. We now describe the effect of the KT transition on the dy- namical properties. If ω ≪ 1/τRC holds (with RC time τRC), the charge relaxation resistance can be defined in the low- frequency expansion G(ω) = −iωCµ + ω2C2 µRq + O(ω3). How- ever, the validity of this expansion is not obvious, since the KT transition may influence τRC itself. Instead, we investigate the low-frequency resistance using R(iωn) ≡ 1 G(iωn) − 1 ωnCµ , (20) where G(iωn) and Cµ are defined in Eqs. (2) and (3), respec- tively. The extrapolation R(iωn → 0) gives the real part of the impedance in the low-frequency limit, hence τRC = R(0)Cµ. In Fig. 3, we plot R(0) for K = 1, 1/3 and 1/5 as a function of V at temperature T = 0.04. For K = 1, R(0) equals h/(2e2) irrespective of V, in agreement with the universal charge re- laxation resistance [2, 3]. For K = 1/3 and 1/5, the univer- sality is observed in the weak barrier region, whereas R(0) is abruptly enhanced with increasing V, reflecting the RG flow to the weak coupling regime due to the KT transition. The temperature dependence of R(0) for K = 1/3 is shown in the inset of Fig 3, which indicates that R(0) diverges as T → 0 in the strong barrier region. The KT transition plays a crucial role in the relevance of the universal charge relaxation resistance. If 2t + B > 0, the system scales to the weak barrier limit, where τRC is indepen- dent of temperature. If 2t + B < 0, on the other hand, the scaling equations (18) predict s2 → const. and t ∝ T −B, so that τRC roughly scales as ∝ T −1(T 2B + const.), which grows faster than the (thermal) coherence time τcoh ∝ 1/T as temper- ature is lowered. These observations suggest that if 2t + B > 0 coherent transport can be realized by lowering temperature to guarantee τRC < τcoh, while if 2t + B < 0 electronic transport in the dot decoheres before charge relaxation is achieved. In  ) 0 = n w i ( R ) 0 = n w i ( G 1 0.8 0.6 0.4 0.2 0 1 T = 0.08 0.04 0.01 102 T 101 m C ) 0 ( R 100 5 6 2 3 4 5 6 V 7 7 8 8 FIG. 4: (color online) Product G(iωn = 0)R(iωn = 0) for K = 1/3 as a function of the backscattering strength V, for different temperatures T . For D and ∆/(2K2), we use the same parameters as in Fig. 2. Inset: R(0)CµT as a function of V, for different T ; the crossing of the curves gives a good estimate of the transition point. the latter case, the quantum dot effectively acts as a reservoir and consequently the dynamical property of the system is gov- erned by transport through the point contact between the two "reservoirs". Therefore the V-dependent low-frequency resis- tance observed in the inset of Fig. 3 reflects the revival of the Landauer-type transport. To see this behavior more clearly, we plot in Fig. 4 the product G(iωn → 0)R(iωn → 0) for K = 1/3 as a function of V. In the strong barrier region, G(0) is finite and equal to [R(0)]−1, which is a familiar prop- erty of transport through a point contact. Upon decreasing V, however, G(0)R(0) is suppressed since G(0) decays to zero be- cause of charging up, although R(0) → Rq is finite. Moreover, we see that the coherent region G(0)R(0) = 0 extends to larger V upon lowering temperature. Finally, we determine the phase boundary of the coherent-incoherent transition by tracing the temperature dependence of the ratio τRC/τcoh = R(0)CµT . The above discussion suggests that there exists a critical backscat- tering strength Vc, below which τRC/τcoh decays to zero, while it diverges otherwise (see the right panel in Fig. 1). From the inset of Fig. 4, the critical value is estimated as Vc ≃ 7. In conclusion, the study of the mesoscopic capacitor in the presence of strong electron electron interaction shows that the relaxation resistance for a dot connected to Luttinger liquid is universal Rq ≡ h/(2e2K) as long as interactions are suffi- ciently weak. Below K < 1/2, this resistance is governed by the strength of the dot-lead coupling: at the charge degener- acy point, there is a critical coupling strength, governed by a KT type phase transition, below which the dot acts as an inco- herent reservoir and the low-frequency resistance exceeds the universal value. In this incoherent regime, the charge relax- 4 ation resistance cannot be defined anymore due to the diver- gence of the RC time. Results could be probed experimentally using quantum dots connected to an edge state in the FQHE regime. Another ex- perimental probe could use one dimensional quantum wires (non chiral Luttinger liquids) with the limitation that the oper- ating frequency would have to be larger than the inverse time of flight within the wire, in order to avoid renormalization effects due to eventual Fermi liquid leads connected to this wire [11]. Y.H. and T.K. are grateful to T. Fujii for valuable discus- sions. Y.H. acknowledges the support of the Japan Society for the Promotion of Science. This research was partially supported by JSPS and MAE under the Japan-France Inte- grated Action Program (SAKURA) and by Grant-in-Aid for Young Scientists (B) (No. 21740220) from the Ministry of Education, Science, Sports and Culture. It was also supported by ANR-PNANO Contract MolSpinTronics, No. ANR-06- NANO-27. The computation in this work was done using the facilities of the Supercomputer Center, Institute for Solid State Physics, University of Tokyo. ∗ Electronic address: [email protected] [1] J. Gabelli, G. Feve, J.-M. Berroir, B. Plac¸ais, A. Cavanna, B. Etienne, Y. Jin, and D. C. Glattli, Science 313, 499 (2006); J. Gabelli, G. Feve, T. Kontos, J.-M. Berroir, B. Plac¸ais, D.C. Glattli, B. Etienne, Y. Jin, M. Buttiker, Phys. Rev. Lett. 98, 166806 (2007). [2] M. Buttiker, H. Thomas, and A. Pretre, Phys. Lett. A 180, 364 (1993). [3] S. E. Nigg, R. L´opez, and M. Buttiker, Phys. Rev. Lett. 97, 206804 (2006). [4] A. Furusaki and K. A. Matveev, Phys. Rev. Lett. 88, 226404 (2002). [5] C. L. Kane and M. P. A. Fisher, Phys. Rev. Lett. 68, 1220 (1992); A. Furusaki and N. Nagaosa, Phys. Rev. B, 47, 3827 (1993). [6] In the FQHE regime, V denotes the strength of quasi-particle tunneling. [7] P. Werner and M. Troyer, Phys. Rev. Lett. 95, 060201 (2005). [8] Y. Hamamoto, K.-I. Imura, and T. Kato, Phys. Rev. B 77, 165402 (2008); Y. Hamamoto and T. Kato, ibid. 77, 245325 (2008). [9] P. W. Anderson, G. Yuval, and D. R. Hamann, Phys. Rev. B 1, 4464 (1970). [10] S. Chakravarty, Phys. Rev. Lett. 49, 681 (1982). [11] D. L. Maslov and M. Stone, Phys. Rev. B 52, R5539 (1995). [12] We assume that the local scatterer at x = 0 does not renormalize significantly the bulk interaction parameter K. 

1801.01153

1

1801

Coherent coupling of dark and bright excitons with vibrational strain

In many physical systems, there are specific electronic states called dark state that are protected from the rapid radiative decay imposed by the system symmetry. Although their long-lived nature indicates their potential for quantum information and spintronic applications, their high stability comes at the expense of optical accessibility. Breaking the symmetry by using magnetic and electric fields has been employed to hybridize dark and bright states thus making them optically active, but high-frequency and on-chip operation remains to be developed. Here we demonstrate the strain-induced coherent coupling of dark and bright exciton states in a GaAs mechanical resonator. The in-plane uniaxial strain breaks the rotational symmetry of the crystal, allowing the dark states to be optically accessible without any external fields. Such dark-bright coupling is tailored by the local strain distribution, which enables the coherent spin operation in the gigahertz regime and opens the way to on-chip excitonic quantum memories and circuits.

cond-mat.mes-hall

cond-mat

Coherent coupling of dark and bright excitons with vibrational strain Ryuichi Ohta1, Hajime Okamoto1, Takehiko Tawara1, Hideki Gotoh1, and Hiroshi Yamaguchi1 1NTT Basic Research Laboratories, NTT Corporation, 3-1 Morinosato Wakamiya, Atsugi-shi, Kanagawa 243-0198, Japan In many physical systems, there are specific electronic states called "dark states" that are protected from the rapid radiative decay imposed by the system symmetry 1-4. Although their long-lived nature indicates their potential for quantum information1,2,5,6 and spintronic7 applications, their high stability comes at the expense of optical accessibility. Breaking the symmetry by using magnetic8,9 and electric fields7,10 has been employed to hybridize dark and bright states thus making them optically active, but high-frequency and on-chip operation remains to be developed. Here we demonstrate the strain-induced coherent coupling of dark and bright exciton states in a GaAs mechanical resonator. The in-plane uniaxial strain breaks the rotational symmetry of the crystal, allowing the dark states to be optically accessible without any external fields. Such dark-bright coupling is tailored by the local strain distribution, which enables the coherent spin operation11,12 in the gigahertz regime and opens the way to on-chip excitonic quantum memories and circuits. Optical transition in two-level systems is generally governed by the conservation law of total angular momentum. To optically excite (relax) two-level systems, the total angular momentum must be conserved between the initial and final states. In the particular case of solid state systems, the optical transition obeys the spin polarizations of electrons and holes. For instance, electrons in the conduction band and holes in the heavy-hole band with antiparallel spins are optically active, and are called bright excitons, and those with parallel spins are optically inactive, and are called dark excitons. Although dark excitons are of great interest owing to their long lifetimes5-7,13, direct optical access is technically challenging14,15. A magnetic field is commonly used to break the rotational symmetry provoking the coherent coupling of dark and bright excitons8,9. Such hybridization allows the dark excitons to become optically active. However, a large magnetic field is only induced statically and uniformly, which limits the dynamic and local control of the dark excitons. A strain field breaks the rotational symmetries of the crystals, and causes intermixing of dark and bright excitons via deformation potentials. In contrast to a magnetic field, a strain field is dynamically and locally controlled with piezoelectric materials16,17. The strain effects for exciton states are described by the Pikus-Bir Hamiltonian18,19 as follows. 𝑃𝜖 + 𝑄𝜖 ∗ ∗ 𝐿𝜖 𝑀𝜖 0 ( 𝐿𝜖 𝑃𝜖 − 𝑄𝜖 0 𝑀𝜖 ∗ 𝑀𝜖 0 𝑃𝜖 − 𝑄𝜖 − 𝜀𝑒𝑥 −𝐿∗ 0 𝑀𝜖 −𝐿𝜖 𝑃𝜖 + 𝑄𝜖 − 𝜀𝑒𝑥) ( 𝐵𝑋𝐻𝐻⟩ 𝐵𝑋𝐿𝐻⟩ 𝐷𝑋𝐿𝐻⟩ 𝐷𝑋𝐻𝐻⟩) 𝑃𝜖 = (𝑎𝑐 − 𝑎𝑣)(𝜖𝑥𝑥 + 𝜖𝑦𝑦 + 𝜖𝑧𝑧) 𝑄𝜖 = −𝑏 (𝜖𝑥𝑥 + 𝜖𝑦𝑦 − 2𝜖𝑧𝑧) 2⁄ 𝐿𝜖 = 𝑑(𝜖𝑥𝑧 − 𝑖𝜖𝑦𝑧) 𝑀𝜖 = √3𝑏(𝜖𝑥𝑥 − 𝜖𝑦𝑦) 2⁄ (1) (2a) (2b) (2c) (2d) Here, BX (DX) stands for bright (dark) excitons, where the subscript of HH (LH) specifies a heavy-hole (light-hole) band. 𝑃𝜖, 𝑄𝜖, 𝐿𝜖, and 𝑀𝜖 are strain induced perturbations, which are derived from strain tensors (𝜖𝑖𝑗) and the deformation potentials of GaAs (𝑎𝑐, 𝑎𝑣, 𝑏, 𝑑). 𝜀𝑒𝑥 is the exchange energy of an electron and a hole. Uniaxial in-plane strain (𝜖𝑥𝑥 ≠ 𝜖𝑦𝑦) provides non-zero off-diagonal components 𝑀𝜖, which cause the inter-band mixing of BX and DX, whereas 𝑀𝜖 becomes negligible with isotropic in-plane strain (𝜖𝑥𝑥 = 𝜖𝑦𝑦), which can be caused by the lattice mismatches in heterojunctions. Therefore, uniaxial strain breaks the four-fold rotational symmetry of the zincblende structure, allowing the hybridization of different total angular momentum states, i.e. bright excitons and dark excitons as shown in Fig. 1(a). To generate time-varying uniaxial in-plane strain, we adopted a micro mechanical resonator based on GaAs, which greatly enhances the strain amplitude because of its quality factor. Figure 1(b) show scanning electron microscope images of this resonator. The sample preparation is detailed in Method. Figure 1(c) shows the calculated energies of the bright and dark excitons in the HH and LH bands with various mechanical displacements, where the strain tensors are calculated with the finite element method (FEM) (see SI). We take account of the contribution from the intrinsic isotropic strain induced by the epitaxial mismatch, which splits the exciton energies of the HH and LH bands when the cantilever is in the equilibrium position20. The applied uniaxial strain causes the additional energy splits and the mixing of the bright excitons in the LH band and the dark excitons in the HH band when the mechanical displacement exceeds 20 nm. Figure 2(a) shows the experimental setup used in this study, which consists of a Ti:Sa pump laser, a charge coupled device (CCD), and a Doppler interferometer. The exciton energies were characterized with a photoluminescence (PL) measurement, while the mechanical motion was simultaneously measured with the interferometer. The characterizations were performed at 7.2 K. The experimental setup is explained in detail in Method. Figure 2(b) shows a typical PL spectrum obtained at the mid-length of the resonator. The position dependence of the PL spectra is discussed in SI. The three higher (lower) energy peaks are attributed to donor (acceptor) bound excitons21. To induce time-varying strain at that position, the 2nd order flexural mode oscillating at 1.04 MHz was driven by a piezo actuator. Figure 2(c) shows the frequency response of this mechanical mode at various drive voltages, where the inset shows the strain distribution calculated by FEM. A mechanical Q of 32,000 is extracted from the linewidth. The vibrational amplitude at the resonance frequency and the corresponding uniaxial strain are shown in Fig. 2(d). The mechanical nonlinearity gradually saturates the amplitude at an actuation voltage higher than 6 mVpp, while the corresponding strain is high enough to cause coupling as estimated from Fig. 1(c). We employed stroboscopic PL measurement16,22,23 to investigate the effects of mechanically induced strain on the bound exciton energies. The pump laser was shaped into a periodic rectangular pulse with an acousto optical modulator (AOM), whose repetition rate was synchronized to the mechanical frequency. By changing the relative phase between the pump pulse and the mechanical motion, PL spectra under time-varying strain were obtained with a CCD. Figure 3(a) shows stroboscopic PL spectra of bound exciton states for a drive voltage of 12 mVpp. Figure 3 (b)-(g) are cross-sectional PL spectra of acceptor/donor bound excitons at 0 (d/g), 320 (c/f), and 480 ns (b/e), respectively. Each spectrum is fitted by a Lorentz function with the peaks labeled I to VI, and their positions are traced as dashed lines in Fig. 3(a). The vibrational strain sinusoidally modulates peaks I and II, whereas peak III is less affected. These behaviors reproduce the energy shifts of acceptor bound excitons characterized by static strain experiments24,25. The vibrational strain also modulates the energy difference of peaks V and VI, which originate from the bright exciton states bound by donors. The most interesting feature can be seen around 320 and 640 ns, where peak V shows the avoided crossing with peak IV. Peak IV is attributed to dark excitons in accordance with ref 26. In fact, its PL intensity vanishes when the energy is detuned from peak V. The avoided crossing is a clear sign that the bright and dark excitons are coherently coupled by mechanically induced strain. To investigate the mechanically hybridized exciton states quantitatively, we numerically calculated the energies and PL spectra of bound excitons with the Pikus-Bir Hamiltonian. We derived the values of 𝑃𝜖, 𝑄𝜖, 𝐿𝜖, and 𝑀𝜖 from the measured mechanical amplitudes and the calculated strain tensors with FEM. Figure 4(a)-(d) show the measured PL spectra and calculated dark and bright bound exciton energies (dashed lines) at drive voltages of 2, 6, 12, and 16 mVpp. Figure 4(e)-(h) show the calculated PL spectra, where the linewidths of the dark and bright excitons are 70 μeV (see SI). The good agreement between the experiments and the theoretical analysis confirms the hybridization of the dark and bright bound excitons via mechanical oscillation. The coupling strength (𝑔𝐷𝐵) is estimated to be about 50 μeV, which is of the same order as the linewidths. The 𝒌 ∙ 𝒑 model analysis indicates that 𝑔𝐷𝐵 can be further improved by tailoring 𝜖𝑥𝑥 and 𝜖𝑦𝑦. For instance, we can independently control 𝑄𝜖 and 𝑀𝜖 in a drum-shaped mechanical resonator, and increase 𝑔𝐷𝐵 to 1 meV, which corresponds to a cooperativity (𝐶 = 4𝑔𝐷𝐵 ) of 800 thus allowing the coherent manipulation of the two states (see SI). Moreover, the Pikus-Bir Hamiltonian suggests that the off-diagonal terms of 𝐿𝜖, namely the shear strain of 𝜖𝑥𝑧 and 𝜖𝑦𝑧, cause additional coupling between 𝐵𝑋𝐻𝐻⟩ and 𝐵𝑋𝐿𝐻⟩ ( 𝑔𝐵𝐵 ), and 𝐷𝑋𝐿𝐻⟩ and 𝐷𝑋𝐻𝐻⟩ ( 𝑔𝐷𝐷 ). The shear strain is generated by torsional oscillation and is concentrated near the side edges of the resonator. 2 𝛾𝐵𝛾𝐷 ⁄ We evaluated the additional exciton-exciton couplings in the 2nd order torsional mode of this resonator shown in Fig. 5(a). Figure 5(b) shows the calculated population of each bound exciton state, which describes the three avoided crossings representing the coherent couplings in multiple dark and bright excitons (see SI). In conclusion, we demonstrate the coherent coupling of dark and bright bound excitons and modulate their optical accessibility using time-varying strain in a mechanical resonator. The origin of the dark-bright coupling is the breaking of the rotational symmetry of the system. The good agreement between the experimental results and the 𝒌 ∙ 𝒑 theory calculation confirm that the vibrational strain causes the hybridization of the two states. Further improvement in the couplings (𝑔𝐷𝐵, 𝑔𝐵𝐵, and 𝑔𝐷𝐷) is possible by tailoring the strain distribution. Although the coherent coupling of dark and bright excitons has been demonstrated with a magnetic field, this mechanical approach has significant advantages as regards the modulation speed of the coupling strength. This device was operated at around 1 MHz, and this will be increased to several gigahertz with current MEMS technologies. In this regime, the coupling can be more rapidly switched than the decay of the bright excitons, which will provide the coherent control of the single spin in dark and bright excitons11,12 on a solid-state platform. This mechanical approach can be widely applied to any solid-state two-level system, such as quantum dots22,23,27,28, nitrogen-vacancy centers29,30, and impurities in semiconductors, and makes it possible to realize on-chip excitonic memories and circuits. Method The mechanical resonator consists of three layers of 100-nm-thick n-doped Al0.3Ga0.7As, undoped 400-nm-thick GaAs, and 5-period undoped Al0.3Ga0.7As /GaAs superlattices. They were grown by molecular-beam epitaxy on a semi-insulating GaAs substrate with a 3-μm-thick Al0.65Ga0.35As sacrificial layer. The mechanical resonator was fabricated by using photolithography, wet etching with H3PO4 and H2O2, and under-etching with HF. The fabricated resonator was 37 μm long, 20 μm wide, and 600 nm thick. The resonator was attached to a piezo actuator, which drove it electrically at the resonance frequency, and placed in a vacuum chamber in a closed-cycle cryostat (attoDRY 1100: attocube). Low temperature PL measurements were performed with a Ti:Sa laser (3900S: Spectra-Physics), a spectrometer (SP-2750: Princeton Instruments), and a CCD camera (PIXIS 400: Princeton Instruments). The wavelength was set at 780 nm, which is short enough to excite electrons and holes above the band gap. The mechanical motion was measured using a Doppler interferometer with a He:Ne laser (MLD_230: Neoark). The Ti:Sa and He:Ne lasers were focused on the resonator by using an objective lens with a numerical aperture of 0.82. Stroboscopic PL measurements were performed with an AOM (TEM-110: Brimrose), which shapes the pump laser to rectangular pulse with a duty ratio of 5 %. Acknowledgements We thank Dr. Y. Matsuzaki, Dr. M. Asano, Dr. S. Houri, Dr. K. Takata, and Dr. S. Adachi for discussions about the theoretical models and fruitful comments on the manuscript. This work is partly supported by MEXT KAKENHI (JP15H05869, JP16H01057) Author contributions R.O., H.O., and H.Y. designed the experiments. R.O. and H.O. fabricated the samples, and developed the experimental setup with support from T.T. and H.G. The measurements and calculations were performed by R.O.. R.O., H.O., T.T. and H.Y. analyzed the data. R.O. and H.Y. wrote the manuscript. All of the authors discussed the results and contributed to the manuscript. Reference 1. Fleischhauer, M. & Lukin, M. D. Dark-state polaritons in electromagnetically induced transparency. Phys. Rev. Lett. 84, 5094-5097 (2000). 2. Xu, X., Sun, Bo., Berman, P. R., Steel, D. G., Bracker, A. S. Gammon, D. & Sham, L. J. Coherent population trapping of an electron spin in a single negatively charged quantum dot. Nat. Phys. 4, 692-695 (2008). 3. Nirmal, M., Norris, D. J., Kuno, M. & Gawendi, M. G. Observation of the "dark exciton" in CdSe quantum dots. Phys. Rev. Lett. 75, 3728-3731 (1995). 4. Zhu, Z., Matsuzaki, Y., Amsuss, R., Kakuyanagi, K., Shimo-Oka, T., Mizuochi, N., Nemoto, K., Semba, K., Munro, W. J., & Saito, S. Observation of dark states in a super conductor diamond quantum hybrid system. Nat. Comm. 5, 3424 (2014). 5. Nazarkin, A., Netz, R., & Sauerbrey, R. Electromagnetically induced quantum memory Phys. Rev. Lett. 92, 043002 (2004). 6. Chanelière, T., Matukevich, D. N., Jenkins, S. D., Lan, S. -Y., Kennedy, T. A. B. & Kuzmich, A. Storage and retrieval of single photons transmitted between remote quantum memories. Nature 438, 833-836 (2005). 7. McFarlane, J., Dalgarno, P. A., Gerardot, B. D., Hadfield, R. H., Warburton, R. J., Karrai, K., Badolato, A. & Petroff, P. M. Gigahertz bandwidth electrical control over a dark exciton-based memory bit in a single quantum dot. Appl. Phys. Lett. 94, 093113 (2009). 8. Efros, Al. L., Rosen, M., Kuno, M., Nirmal, M., Norris, D. J., & Bawendi, M. Band-edge exciton in quantum dots of semiconductors with a degenerate valence band: Dark and bright exciton states. Phys. Rev. B 54, 4843-4856 (1996). 9. Bayer, M., Stern, O., Kuther, A. & Forchel, A. Spectroscopic study of dark excitons in InxGa1-xAs self-assembled quantum dots by a magnetic-field-induced symmetry breaking. Phys. Rev. B 61, 7273-7276 (2000). 10. Kraus, R. M., Lagoudakis, R. G., Rogach, A. L., Talapin, D. V., Weller, H., Lupton, J. M., & Feldmann, J. Room-temperature exciton storage in elongated semiconductor nanocrystals. Phys. Rev. Lett. 98, 017401 (2007). 11. Rubbmark, J. R., Kash, M. M., Littman, M. G. & Kleppner, D. Dynamical effects at avoided level crossings: A study of the Landau-Zener effect using Rydberg atoms. Phys. Rev. A 23, 3107-3117 (1981). 12. Forster, F., Petersen, G.,Manus, S., Hänggi, P., Schuh, D., Wegscheider, W., Kohler, S., & Ludwig, S., Characterization of qubit dephasing by Landau-Zener-Stückelberg- Majorana interferometry. Phys. Rev. Lett. 112, 116803 (2014). 13. Ye, Z., Cao, T., O'Brien, K., Zhu, H., Yin, X., Wang, Y., Louie, S. G., & Zhang, X., Probing excitonic dark states in single-layer tungsten disulphide. Nature 513, 214-218 (2014). 14. Poem, E., Kodriano, Y., Tradonsky, C., Lindner, N. H. Gerardot, B. D., Petroff, P. M. & Gershoni, D. Accessing the dark exciton with light. Nat. Phys. 6, 993-997 (2010). 15. Zhou, Y. et al. Probing dark excitons in atomically thin semiconductors via near-field coupling to surface plasmon polaritons. Nat. Nano. 12, 856-861 (2017). 16. Weiss, M. et al. Dynamic acoustic control of individual optically active quantum dot-like emission centers in heterostructure nanowires. Nano Lett. 14, 2256-2264 (2014). 17. Trotta, R., Wildmann, J. S., Zallo, E., Schmidt, O. G., & Rastelli, A. Highly entangled photons from hybrid piezoelectric-semiconductor quantum dot devices. Nano Lett. 14, 3439-3444 (2014). 18. Bahder, T. B. Eight-band k ∙ p model of strained zinc-blende crystals. Phys. Rev. B 41, 11992-12001 (1990). 19. Stier, O., Grundmann, M., & Bimberg, D. Electronic and optical properties of strained quantum dots modeled by 8-band k ∙ p theory. Phys. Rev. B 59, 5688-5701 (1999). 20. Huo, Y. H. et al. A light-hole exciton in a quantum dot. Nat. Phys. 10, 46-51 (2014). 21. Allen, D. G., Stanley, C. R., & Sherwin, M. S. Optically detected measurement of the ground-state population of an ensemble of neutral donors in GaAs. Phys. Rev. B 72, 035302 (2005). 22. Yeo, I. et al. Strain-mediated coupling in a quantum dot-mechanical oscillator hybrid system. Nat. Nano. 9, 106-110 (2014). 23. Montinaro, M. et al. Quantum Dot Opto-Mechanics in a Fully Self-Assembled Nanowire. Nano Lett. 14, 4454-4460 (2014). 24. Schmidt, M., Morgan, T. N., Schairer, W. Stress effects on excitons bound to shallow acceptors in GaAs. Phys. Rev. B 11, 5002-5007 (1975). 25. Karasyuk, V. A., Thewalt, M. L. W., & Springthorpe, A. J. Strain effects on bound exciton luminescence in epitaxial GaAs studied using a wafer bending technique. Phys. Stat. Sol. 210, 353-359 (1998). 26. Abe, Y. Electron-hole exchange energy in shallow excitons. J. J. Phys. Soc. 19, 818-829 (1964). 27. Wilson-Rae, I., Zoller, P., & Imamoglu, A. Laser cooling of a nanomechanical resonator mode to its quantum ground state. Phys. Rev. Lett. 92, 075507 (2004). 28. Nakaoka, T., Kakitsuka, T., Saito, T., Kako, S., Ishida, S., Nishioka, M., Yoshikuni, Y., Arakawa, Y. Strain-induced modifications of the electronic states of InGaAs quantum dots embedded in bowed airbridge structures. J. Appl. Phys. 94, 6812-6817 (2003). 29. Barfuss, A., Teissier, J., Neu, E., Nunnenkamp, A. & Maletinsky, P. Strong mechanical driving of a single electron spin. Nat. Phys. 11, 820-825 (2015). 30. Ovartchaiyapong, P., Lee, K. W., Myers, B. A., & Jayich, A. C. B. Dynamic strain-mediated coupling of a single diamond spin to a mechanical resonator. Nat. Comm. 5, 4429 (2014).  Figure 1 Schematic and SEM images of the mechanical resonator and energy diagram of bound exciton states. a, Schematic image of the mechanical resonator and the strain-induced coupling of dark and bright excitons. b, SEM image of the fabricated resonator (37 μm long, 20 μm wide, and 600 nm thick). c, Energy diagram of dark and bright bound excitons with mechanical displacement calculated by 𝒌 ∙ 𝒑 theory. In-plane strain enlarges the splitting of heavy-hole (HH) and light-hole (LH) bands, and generates coupling between dark and bright excitons.  Figure 2 Optical and mechanical properties of resonator. a, Experimental setup for characterizing the optical and mechanical properties of a resonator with bound excitons. PL spectra were obtained with a Ti:Sa pump laser, whose oscillating wavelength was 780 nm. Mechanical motion was measured using a Doppler interferometer with a He:Ne laser. The resonator was placed on a piezo actuator, which drove it at the resonance frequency. Experiments were performed in a vacuum at 7.2 K b, PL spectrum obtained at the mid-point of the resonator. It is fitted by seven Lorenz functions. Red (blue) peaks are attributed to acceptor (donor) bound excitons. The gray peak is considered to be background. c, Frequency response of the 2nd flexural mode of the resonator driven by the piezo actuator. The inset shows the strain distribution calculated by FEM. d, Voltage dependence of the vibrational amplitude and the corresponding in-plane strain of the resonator at the resonance frequency.  Figure 3 Stroboscopic PL spectra of acceptor and donor bound excitons. a, Wide-range stroboscopic PL spectrum. Peaks I to III (IV to VI) are attributed to acceptor (donor) bound excitons. The dashed lines are their fitted peak energies. b-g, Cross-sectional PL spectra of acceptor (b-d) and donor (e-g) bound excitons at 0 (d, g), 320 (c, f), and 480 ns (b, e), respectively. Peaks IV and V show the avoided crossing at 320 and 640 ns, respectively, indicating the coherent coupling of the two states. Peak IV is attributed to the dark exciton, which is optically inactive when it detuned from peak V.  Figure 4 Experimental and numerical spectra of dark and bright bound excitons. a-d, Stroboscopic PL spectra obtained with drive voltages of (a) 2, (b) 6, (c) 12, and (d) 16 mVpp. The dashed lines are the eigen energies of bound excitons calculated by using the Pikus-Bir Hamiltonian with measured parameters. e-h, Numerically calculated PL spectra at the corresponding vibrational amplitudes of a-d. The good agreement between the experimental and numerical spectra confirms that the vibrational strain causes the coherent coupling of the dark exciton in the HH band and the bright exciton in the LH band.  Figure 5 Exciton-exciton couplings in four bound states caused by torsional mechanical oscillation. a, The 2nd order torsional mechanical mode calculated with FEM. b, Displacement dependence of the dark and bright bound exciton states calculated using the Pikus-Bir Hamiltonian with the parameters from experiments. 𝐿𝜖 causes the coupling of 𝐵𝑋𝐻𝐻⟩ and 𝐵𝑋𝐿𝐻⟩, and 𝐷𝑋𝐿𝐻⟩ and 𝐷𝑋𝐻𝐻⟩ at 4 nm, while the coupling of 𝐷𝑋𝐻𝐻⟩ and 𝐵𝑋𝐿𝐻⟩ is provided by 𝑀𝜖 at 10 nm Supplementary Information Strain effects on dark and bright bound excitons The wave functions of electrons near the GaAs bandgap are derived from Kane's model, which considers the spin-orbit interaction in 𝒌 ∙ 𝒑 theory. 𝐻 = 𝒌2 2𝑚0 + 𝒌 ∙ 𝒑 𝑚0 + 𝒑2 2𝑚0 + 𝑉(𝒓) + ℏ 2𝑐2 (∇𝑉0) × 𝒑 ∙ 𝝈 + 4𝑚0 ℏ 2𝑐2 (∇𝑉0) × 𝒌 ∙ 𝝈 4𝑚0 The first, second, and third terms are standard 𝒌 ∙ 𝒑 Hamiltonians, the fourth term is the self-consisted periodic potential 𝑉(𝒓), and the fifth and sixth terms are the spin-orbit interaction. 𝑚0, 𝑐, and ℏ, are electron mass, light velocity, and Planck's constant, respectively. 𝝈 is the Pauli spin matrix with components 𝜎𝑥 = [ ] , 𝜎𝑦 = [ 0 1 1 0 0 −𝑖 𝑖 0 1 0 ] , 𝜎𝑧 = [ 0 −1 ] This Hamiltonian provides eight electron states with the following wave functions. 𝑈⟩ = 𝑆 ↑⟩ 𝐷⟩ = 𝑆 ↓⟩ 𝑢𝐸 𝑢𝐸 𝑈 ⟩ = 𝑢𝐻𝐻 𝐷 ⟩ = 𝑢𝐻𝐻 𝑈 ⟩ = 𝑢𝐿𝐻 𝐷 ⟩ = 𝑢𝐿𝐻 𝑈 ⟩ = 𝑢𝑆𝑂 𝐷 ⟩ = 𝑢𝑆𝑂 −1 √2 1 √2 −1 √6 1 √6 1 √3 1 √3 (𝑋 + 𝑖𝑌) ↑⟩ (𝑋 − 𝑖𝑌) ↓⟩ (𝑋 + 𝑖𝑌) ↓⟩ + √ 2 3 𝑍 ↑⟩ (𝑋 − 𝑖𝑌) ↑⟩ + √ 2 3 𝑍 ↓⟩ (𝑋 − 𝑖𝑌) ↑⟩ − (𝑋 + 𝑖𝑌) ↓⟩ + 1 √3 1 √3 𝑍 ↓⟩ 𝑍 ↑⟩ 𝑢𝐸⟩, 𝑢𝐻𝐻⟩, 𝑢𝐿𝐻⟩, 𝑢𝑆𝑂⟩ are the wave functions of electrons in conduction, heavy hole (HH), light hole (LH), and spin-orbits split-off bands, where 𝑆⟩ is an s-shaped function and 𝑋⟩, 𝑌⟩, and 𝑍⟩ are p-shaped functions, and 𝐷 ⟩ (𝑢𝐸 𝑈 ⟩ (𝑢𝐸 𝐷⟩ and 𝑢𝐻𝐻 𝐷⟩ and 𝑢𝐻𝐻 𝑈⟩ and ↑ (↓) indicates the spin up (down) state. Electron and hole pairs of 𝑢𝐸 𝑈⟩ and 𝑈 ⟩) form bright excitons (𝐵𝑋𝐻𝐻⟩), and those of 𝑢𝐸 𝑢𝐻𝐻 𝐷 ⟩) form dark excitons (𝐷𝑋𝐻𝐻⟩). The energies of the dark 𝑢𝐻𝐻 excitons are lower than those of the bright excitons due to the exchange energy of an electron and a hole ( 𝜀𝑒𝑥 ). We assume that the potential gradients of donors are much broader than the lattice constant so that the effective mass approximation well describes the bound exciton energiesS1. The strain effect of each state is characterized by the Luttinger-Kohn and 𝐷 ⟩, Pikus-Bir Hamiltonian for bases of (𝑢𝐸 𝑢𝑆𝑂 𝑈 ⟩, 𝑢𝐿𝐻 𝑈 ⟩, 𝑢𝑆𝑂 𝑈 ⟩, 𝑢𝐿𝐻 𝐷 ⟩, 𝑢𝐻𝐻 𝑈⟩, 𝑢𝐸 𝐷⟩, 𝑢𝐻𝐻 𝐷 ⟩) 𝐸𝑔 + 𝑟 0 0 0 0 0 0 0 0 ( 𝐸𝑔 + 𝑟 0 0 0 0 0 0 0 0 𝑝 + 𝑞 0 0 𝑙 0 0 𝑚 0 0 0 0 0 1 √2 − 0 0 𝑙 −√2𝑚 𝑝 − 𝑞 0 𝑚 √2𝑞 0 𝑝 − 𝑞 −𝑙 √ 3 2 𝑙∗ √ 3 2 𝑙 −√2𝑞 𝑚∗ −𝑙∗ 𝑝 + 𝑞 √2𝑚∗ − 1 √2 𝑙∗ 𝑙∗ √2𝑞 √ 3 2 𝑙 √2𝑚 𝑝 − ∆ 0 𝑙∗ 𝑚∗ 0 1 √2 − −√2𝑚∗ √ 3 2 𝑙∗ −√2𝑞 − 1 √2 𝑙∗ 0 𝑝 − ∆ ) 𝑟 = 𝑟𝑘 + 𝑟𝜖 𝑝 = 𝑝𝑘 + 𝑝𝜖 𝑞 = 𝑞𝑘 + 𝑞𝜖 𝑙 = 𝑙𝑘 + 𝑙𝜖 𝑚 = 𝑚𝑘 + 𝑚𝜖 𝑟𝑘 = 𝜀𝑐 + 𝑝𝑘 = 𝜀𝑣 − ∗ (𝑘𝑥 ℏ2 2𝑚𝑒 ℏ2 2𝑚0 2 + 𝑘𝑦 2 + 𝑘𝑧 2) 𝛾1(𝑘𝑥 2 + 𝑘𝑦 2 + 𝑘𝑧 2) 𝑞𝑘 = − 𝛾2(𝑘𝑥 2 + 𝑘𝑦 2 − 2𝑘𝑧 2) ℏ2 2𝑚0 ℏ2 2𝑚0 ℏ2 2𝑚0 𝑙𝑘 = 𝑚𝑘 = 𝛾32√3(𝑘𝑥 − 𝑖𝑘𝑦)𝑘𝑧 √3[𝛾2(𝑘𝑥 2 − 𝑘𝑦 2) − 𝑖2𝛾3𝑘𝑥𝑘𝑦] 𝑟𝜖 = 𝑎𝑐(𝜖𝑥𝑥 + 𝜖𝑦𝑦 + 𝜖𝑧𝑧) 𝑝𝜖 = 𝑎𝑣(𝜖𝑥𝑥 + 𝜖𝑦𝑦 + 𝜖𝑧𝑧) 𝑞𝜖 = 𝑏 2 (𝜖𝑥𝑥 + 𝜖𝑦𝑦 − 2𝜖𝑧𝑧) 𝑙𝜖 = −𝑑(𝜖𝑥𝑧 − 𝑖𝜖𝑦𝑧) 𝑚𝜖 = − √3 2 𝑏(𝜖𝑥𝑥 − 𝜖𝑦𝑦) 𝛾1,2,3 are Luttinger parameters and 𝑎𝑐, 𝑎𝑣, 𝑏, and 𝑑 are hydrostatic and shear deformation potentials. 𝜖𝑖𝑗 is the strain tensor component. 𝐸𝑔 is the bandgap energy of GaAs. The cross terms between conduction and valence bands are negligible around Γ-point (𝑘𝑥, 𝑘𝑦, 𝑘𝑧 = 0), and so the strain effects for the bound exciton states (𝐵𝑋𝐻𝐻⟩, 𝐵𝑋𝐿𝐻⟩, 𝐷𝑋𝐿𝐻⟩, 𝐷𝑋𝐻𝐻⟩) are described as follows. 𝐸𝑔 ∗ + 𝑃𝜖 + 𝑄𝜖 𝐿𝜖 𝐸𝑔 ∗ + 𝑃𝜖 − 𝑄𝜖 𝑀𝜖 0 ( ∗ ∗ 𝐿𝜖 𝑀𝜖 0 0 𝑀𝜖 −𝐿𝜖 ∗ + 𝑃𝜖 + 𝑄𝜖 − 𝜀𝑒𝑥) 𝐸𝑔 ∗ 𝐸𝑔 −𝐿∗ ∗ + 𝑃𝜖 − 𝑄𝜖 − 𝜀𝑒𝑥 0 𝑀𝜖 𝑃𝜖 = (𝑎𝑐 − 𝑎𝑣)(𝜖𝑥𝑥 + 𝜖𝑦𝑦 + 𝜖𝑧𝑧) 𝑄𝜖 = −𝑏(𝜖𝑥𝑥 + 𝜖𝑦𝑦 − 2𝜖𝑧𝑧) 2⁄ 𝐿𝜖 = 𝑑(𝜖𝑥𝑧 − 𝑖𝜖𝑦𝑧) 𝑀𝜖 = √3𝑏(𝜖𝑥𝑥 − 𝜖𝑦𝑦) 2⁄ ∗ takes account of the binding energies. 𝜀𝑒𝑥 is the exchange energy where 𝐸𝑔 of an electron and a hole. The diagonal terms of 𝑄𝜖 cause the energy splitting of HH and LH bands. The off-diagonal terms of 𝑀𝜖 cause the couplings of 𝐵𝑋𝐻𝐻⟩ and 𝐷𝑋𝐿𝐻⟩ (𝐵𝑋𝐿𝐻⟩ and 𝐷𝑋𝐻𝐻⟩). In the experiments, there are two kinds of strains, namely the static strain from the lattice mismatch of GaAs and Al0.3Ga0.7As, and the dynamic strain caused by the vibration. The former is an isotropic in-plane strain (𝜖𝑥𝑥 ≅ 𝜖𝑦𝑦), which causes the intrinsic energy difference between the LH and HH bands (𝜀𝐿𝐻 = 2 × 𝑄𝜖 ). However, the mixing terms from the intrinsic strain are negligible because 𝜖𝑥𝑥 and 𝜖𝑦𝑦 cancel each other out (𝑄𝜖 ≫ 𝑀𝜖 ). On the other hand, the mechanical motion in the cantilever structure as shown in Fig. 1(a) generates a uniaxial strain (𝜖𝑥𝑥 ≫ 𝜖𝑦𝑦), which provides both the energy splitting of the LH and HH bands and the mixing of the dark and bright excitons. We calculated the strain tensors of the 2nd flexural mode numerically using FEM at the measured position of the mechanical resonator. The parameters 𝜀𝑒𝑥, 𝑏, and 𝑑 were 2.9 × 10-4, -2.2, -5.4 eV based on ∗, 𝜀𝐿𝐻, and 𝑎 were extracted as 1.51486, 1.7 × 10-4, and references18,28, and 𝐸𝑔 -0.5 eV by fitting the experiments. The numerical solutions of the dark and bright exciton energies obtained with 𝒌 ∙ 𝒑 theory are shown in Fig. 1(d) and Fig. 4(a)-(d). Numerical calculation of PL spectra of bound excitons PL spectra are calculated from the population of bright excitons, which are described by the creation-annihilation operators in harmonic oscillator approximation. In this case, the creation-annihilation operator of each state is derived from the steady states of the Heisenberg equation of motion with the Pikus-Bir Hamiltonian. 𝜔𝐵𝐻 − 𝜔 ∗ ∗ 𝐿𝜖 𝑀𝜖 0 ( 𝐿𝜖 𝜔𝐵𝐿 − 𝜔 0 𝑀𝜖 ∗ 𝑀𝜖 0 𝜔𝐷𝐿 − 𝜔 0 𝑀𝜖 −𝐿𝜖 ( 𝑏𝐻𝐻 𝑏𝐿𝐻 𝑑𝐿𝐻 𝑑𝐻𝐻 ) = ( 𝜆𝐵𝐻 𝜆𝐵𝐿 𝜆𝐷𝐿 𝜆𝐷𝐻 ) ∗ 𝜔𝐷𝐻 − 𝜔) −𝐿𝜖 ∗+𝑃𝜖 + 𝑄𝜖 − 𝑖𝛾𝐵𝐻 ∗+𝑃𝜖 − 𝑄𝜖 − 𝑖𝛾𝐵𝐿 ∗+𝑃𝜖 − 𝑄𝜖 − 𝜀𝑒𝑥 − 𝑖𝛾𝐷𝐿 ∗+𝑃𝜖 + 𝑄𝜖 − 𝜀𝑒𝑥 − 𝑖𝛾𝐷𝐻 𝜔𝐵𝐻 = 𝐸𝑔 𝜔𝐵𝐿 = 𝐸𝑔 𝜔𝐷𝐿 = 𝐸𝑔 𝜔𝐷𝐻 = 𝐸𝑔 Here, 𝛾 , 𝑏(𝑑) , and 𝜆 are the linewidth, annihilation operator, and excitation of each state, respectively. We assume that the excitation ratios are the same in the four states. The PL intensities are derived from † 𝑏𝐿𝐻 , where we consider the collection efficiency of an † 𝑏𝐻𝐻 + 0.88 × 𝑏𝐿𝐻 𝑏𝐻𝐻 objective lens with a numerical aperture of 0.82. Two-thirds of the photons are emitted in the in-plane direction from LH band-excitons. Mechanical bending additionally modulates the PL intensity via the piezo potential effectS1, which is proportional to the magnitude of the in-plane strain. The piezo potential effect causes a potential gradient and modifies carrier distributions, because the magnitude of the in-plane strain varies linearly in the thickness direction. It reduces the PL intensity of all bound excitons, i.e. acceptor and donor bound excitons, at a large mechanical displacement. We take account of the time-varying intensity reduction from the experimentally obtained ensemble photon modulation from all bound excitons. Position dependence of PL spectra Figure S1(b) shows the position dependence of PL spectra obtained in the length direction of the cantilever, i.e. along the dotted line in Fig. S1(a). Around the clamping point of the resonator, the exciton peaks are broadened because of the large non-uniform strain imposed by the lattice-mismatched sacrificial layer. Such non-uniform strain is released around the middle of the resonator, and so sharp peaks originating from the bound exciton states are observed. On the basis of this position dependence of the PL spectra, we adopted the 2nd order flexural mode to efficiently induce in-plane strain for bound excitons, whereas the fundamental mode generates the strain at the bottom of the resonator where each exciton state is not resolved in the PL spectra. The PL spectra in Figs. 3 and 4 were obtained at a position 22 μm from the bottom. Tailoring coupling strength and energy shifts via strain engineering The 𝒌 ∙ 𝒑 model calculation indicates that the coupling strength and energy shifts can be independently manipulated by the strain distribution, i.e. the magnitude and anisotropy of the strain tensors. To enhance 𝑔𝐷𝐵, we need to maximize 𝑀𝜖 and inject a large vibrational amplitude. However, a large vibrational amplitude causes the energy detuning of dark and bright excitons, and so we also need to minimize 𝑃𝜖 and 𝑄𝜖 to involve the strong interaction in two states. A large 𝑀𝜖 and a small 𝑃𝜖 and 𝑄𝜖 are generated by the anisotropic strain, 𝜖𝑧𝑧 = 0, which result in the following diagonal and off-diagonal terms. 𝜖𝑥𝑥 = −𝜖𝑦𝑦 𝑀𝜖 = √3𝑏𝜖𝑥𝑥 𝑃𝜖 = 𝑄𝜖 = 0 Such anisotropic strain can be realized in a drum-shaped resonator as shown in Fig. S2(a), whereas a cantilever dominantly causes the uniaxial strain discussed in the main text. Here, we adopted two orthogonal modes, mode H and mode V as shown in Fig. S2(b)-(c), and chose their cross point where 𝜖𝑥𝑥 = −𝜖𝑦𝑦 and 𝜖𝑧𝑧 = 0. Figure S2(d)-(i) shows 𝜖𝑥𝑥, 𝜖𝑦𝑦, and 𝜖𝑧𝑧 in mode H and mode V calculated by FEM. The matrix elements (μeV/nm) are derived from the sum of two strain tensors and the parameters in the experiments. 𝑃𝜖 = −0.22 𝑄𝜖 = −0.75 𝐿𝜖 = 2.31 − 1.27𝑖 𝑀𝜖 = −45.3 Although numerical calculation does not perfectly cancel 𝑃𝜖 and 𝑄𝜖, 𝑀𝜖 is two orders of magnitude larger than they are. We calculated the PL spectrum from dark and bright excitons with the experimentally available vibrational amplitude as shown in Fig. S2(j). 𝑔𝐷𝐵 reaches 1 meV, which is much larger than the exciton linewidths (𝛾𝐵, 𝛾𝐷), the initial energy difference of dark and bright excitons ( 𝜀𝑒𝑥 − 𝜀𝐿𝐻 ), and the strain-induced energy detuning (2𝑄𝜖). The cooperativity (𝐶 = 4𝑔𝐷𝐵 ) is about 800, and thus allows the coherent manipulation of dark and bright bound exciton states at 2 𝛾𝐵𝛾𝐷 ⁄ the single quantum level. Exciton-exciton coupling in four bound exciton states The coherent couplings of the HH and LH bands emerge when the off-diagonal terms of 𝐿𝜖 become large, which is due to the shear strain of 𝜖𝑥𝑧 and 𝜖𝑦𝑧. Torsional mechanical vibration, as shown in Fig. 5(a), generates a large shear strain around the side edges of the resonator. We calculated the matrix elements at a position 100 nm from the side edge. 𝑃𝜖 = 0.46 𝑄𝜖 = 7.35 𝐿𝜖 = 9.99 + 0.18𝑖 𝑀𝜖 = −6.77 Here, 𝐿𝜖 is large enough to cause exciton-exciton coupling in the HH and LH bands. Instead of the PL intensity, we calculate the population of four bound † 𝑑𝐿𝐻 ) with strain-induced exciton states ( 𝑏𝐻𝐻 interactions. We assume that the excitation efficiencies for each state are the † 𝑑𝐻𝐻 + 𝑑𝐿𝐻 † 𝑏𝐻𝐻 + 𝑏𝐿𝐻 † 𝑏𝐿𝐻 + 𝑑𝐻𝐻 same, and that the correction efficiencies and piezo potential effects can be neglected. Here, 𝜀𝐿𝐻 is -1.6 × 10-4 eV. Avoided crossings between dark-dark and bright-bright excitons are observed together with those between dark-bright excitons. Reference S1. Sogawa, T., Sanada, H., Gotoh. H., Yamaguchi, H. & Santos, P. V. Spatially modulated photoluminescence properties in dynamically strained GaAs/AlAs quantum wells by surface acoustic wave. Appl. Phys. Lett. 100, 162109 (2012).  Figure S1 Position dependence of PL spectra. a, Top view of the SEM image of the resonator. b, PL spectra obtained along the dotted line in a. PL spectra are broadened and interfused around the bottom of the resonator due to the non-uniform strain imposed by the sacrificial layer. The non-uniform strain is released around the middle, therefore the PL peaks of the acceptor and donor bound excitons can be distinguished.  Figure S2 Strain tensors and dark-bright coupling in drum-shaped resonator. a, Schematic image of a drum-shaped resonator. The radius and thickness are 20 μm and 600 nm, respectively. b-i, Spatial distribution of amplitude (b,f), 𝜖𝑥𝑥 (c,g), 𝜖𝑦𝑦 (d,h), and 𝜖𝑧𝑧 (e,i) in mode H (b-e) and mode V (f-i). j, Calculated PL spectra from dark and bright bound excitons with the strain tensors at the circled position in b-i. It clearly shows the strong coupling with a 𝑔𝐷𝐵 of 1 meV, where the detuning of the dark and bright excitons is not significantly changed by mechanical displacement. 

1812.03359

5

1812

Circularly polarized thermal radiation from nonequilibrium coupled antennas

Circularly polarized light can be obtained by using either polarization conversion or structural chirality. Here we reveal a fundamentally unrelated mechanism of generating circularly polarized light using coupled nonequilibrium sources. We show that thermal emission from a compact dimer of subwavelength, anisotropic antennas can be highly circularly polarized when the antennas are at unequal temperatures. Furthermore, the handedness of emitted light is flipped upon interchanging the temperatures of the antennas, thereby enabling reconfigurability of the polarization state lacked by most circularly polarized light sources. We describe the fundamental origin of this mechanism using rigorous fluctuational electrodynamic analysis and further provide practical examples for its experimental implementation. Apart from the technology applications in reconfigurable devices, communication, and sensing, this work motivates new inquiries of angular-momentum-related thermal-radiation phenomena using thermal nonequilibrium, without applying magnetic field.

cond-mat.mes-hall

cond-mat

a Circularly polarized thermal radiation from nonequilibrium coupled antennas Chinmay Khandekar∗ and Zubin Jacob† Birck Nanotechnology Center, School of Electrical and Computer Engineering, College of Engineering, Purdue University, West Lafayette, Indiana 47907, USA (Dated: September 24, 2019) Circularly polarized light can be obtained by using either polarization conversion or structural chirality. Here we reveal a fundamentally unrelated mechanism of generating circularly polarized light using coupled nonequilibrium sources. We show that thermal emission from a compact dimer of subwavelength, anisotropic antennas can be highly circularly polarized when the antennas are at unequal temperatures. Furthermore, the handedness of emitted light is flipped upon interchanging the temperatures of the antennas, thereby enabling reconfigurability of the polarization state lacked by most circularly polarized light sources. We describe the fundamental origin of this mechanism using rigorous fluctuational electrodynamic analysis and further provide practical examples for its experimental implementation. Apart from the technology applications in reconfigurable devices, communication, and sensing, this work motivates new inquiries of angular momentum related thermal radiation phenomena using thermal nonequilibrium, without applying magnetic field. I. INTRODUCTION has light (CP) polarized Circularly recently acquired great attention in context of spin-controlled nanophotonics [1, 2], spintronics [3] and chiral quantum optics [4] where its spin angular momentum is harnessed for engineering spin-dependent light matter interactions at nanoscale. Given its fundamental and technological importance, there is a strong demand for CP light sources having high purity and compactness with/without reconfigurability of the polarization state. approach chirality at One approach to obtain CP light is by passing unpolarized light through a linear polarizer followed by an optimized polarization conversion device such as a metasurface [5 -- 7], which can preferentially convert it into right circularly polarized (RCP) light Another or left circularly polarized (LCP) light. structural utilizes more fundamental (geometric/material) the the source, examples of which include a long list of electroluminescent [8 -- 12] and photoluminescent [13 -- 16] CP light sources. Apart from these approaches, CP light generation via thermal radiation (incandescence) has been demonstrated experimentally using the same underlying mechanisms [17, 18]. These are few representative examples of all types of CP-light sources which are fundamentally based on either polarization conversion or structural chirality. level of In this work, we demonstrate a different mechanism of generating CP light based on near-field coupling between nonequilibrium sources. It is unrelated to polarization conversion or chirality and cannot be described using those concepts. This mechanism is useful because practically, it solves the challenging problem of reconfigurability of the polarization state lacked by most CP light sources. Fundamentally, it reveals an unforeseen connection between thermal nonequilibrium and angular momentum of light. A pair of anisotropic shaped, FIG. 1. subwavelength (dipolar), coupled antennas at unequal temperatures can emit strongly circularly polarized light in suitable directions (upper and lower hemispheres). Handedness/directionality of emitted polarized radiation is switched upon interchanging the temperatures of two antennas. thermal electrodynamic We consider a dimer system of two perpendicularly oriented, subwavelength, anisotropic shaped antennas depicted in figure 1. We analyze the spin angular momentum of emission from the dimer using fluctuational electrodynamics. We note that this approach [19] remains largely unexplored in the field of thermal radiation despite many decades of separate works on radiative heat [20, 21] and angular momentum of non-thermal light [22, 23]. The fluctuational reveals that suitable correlations (imaginary valued) between orthogonal components of thermally fluctuating sources are necessary for emission of CP light. We find that the strong correlations are realized through near field interactions between the antennas when they are held at unequal temperatures (nonequilibrium). On the other hand, antennas at equal temperatures emit very Interestingly, weakly polarized thermal the handedness for nonequilibrium antennas, of interchanging flipped emitted upon enabling the the antennas reconfigurability lacked is temperatures of the polarization state radiation. radiation analysis transfer of thus ColdHotRCP LCP Hot ColdRCP LCP xzysources by most CP light [17, 18]. We further explore the experimental feasibility of observing this phenomenon with dimers of antennas made of Silicon Carbide (SiC) and Indium Phosphide (InP), amongst many other plasmonic/polaritonic materials. The resonantly enhanced near-field interaction between the localized dipolar modes of the antennas, leads to reasonably strong mid-infrared CP light emission under temperature difference of ∆T ∼ 30K across separation of d ∼ 1µm between the antennas. This should be experimentally realizable in the near future since a significant experimental progress in measuring near-field radiative heat transfer has facilitated exploration of large temperature differences at nanoscale and has verified the validity of fluctuational electrodynamic theory [24 -- 26]. We further provide another conceptual example system based on plasmonic nanolaser, operated well below lasing threshold, where amplified spontaneous emission is described using fluctuational electrodynamics. That example demonstrates that the proposed mechanism is not limited to 'thermal' nonequilibrium but can also be implemented in other forms such as population inversion of gain atoms, paving the way for CP light emitting LEDs. solid-state designs In comparison to other that can emit CP thermal radiation [17, 18, 27, 28], the proposed mechanism offers a practical advantage of temperature-based polarization reconfigurability. The underlying fluctuational electrodynamic analysis sheds additional light on previous designs and expands the design space by incorporating nonequilibrium systems. In the field of thermal radiation, the proposed mechanism reveals a non-intuitive fundamental connection between thermal nonequilibrium and angular momentum of light. While almost all studies of angular momentum of light are primarily limited to non-thermal light [29 -- 31], few recent studies [19, 32] have started to explore related thermal radiation phenomena. But they require application of magnetic field. Our work suggests the possibility of exploring such phenomena using thermal nonequilibrium without applying magnetic field. In context of using nonequilibrium systems for shaping thermal radiation, other recent works have explored directional emissivity [33, 34], nontrivial thermal forces and torques [35] and enhanced emissivity from nonequilibrium antennas [36]. Our work conveys that one can take advantage of temperature-based for reconfigurability numerous applications sensing and detection technologies. nonequilibrium systems, in communications, in into following parts. We divide the paper In Sec. II, we demonstrate circular polarization (spin angular momentum) of thermal emission from the dimer system of Fig.1 using fluctuational electrodynamics. We describe the fundamental origin of CP light emission from nonequilibrium antennas and further provide general ii In Sec. design guidelines for systems more complicated than the proposed dimer. III, we explore the practical implementation of the proposed mechanism by considering realistic SiC and InP antenna dimers and a conceptual plasmonic nanolaser system as another viable option. Finally, we conclude by highlighting the fundamental and technological relevance of this work and future research directions in Sec. IV. II. DIMER OF ANTENNAS Theory: Thermal radiation consists of electromagnetic stochastic motion of fields generated by thermal, charges. These can be calculated from Maxwell's equations by introducing fluctuating currents provided they satisfy specific fluctuation-dissipation relations [20, 37]. To study thermal emission from a dimer of subwavelength antennas, we consider one variant of this fluctuational electrodynamic approach termed as finite dipole model where thermal sources are randomly fluctuating dipole moments of dipolar sources, satisfying specific fluctuation-dissipation relations [38 -- 40]. We consider the geometry depicted in Fig.1 consisting of two subwavelength (dipolar), anisotropic antennas placed near the origin in vacuum with their long axes oriented along arbitrary e1 and e2 directions. The antennas have temperatures T1, T2 and vacuum polarizabilities α1, α2 along their orientations respectively. The fluctuating dipole moments of these dipolar antennas in the absence of any interactions are given by p1e1 and p2e2 which satisfy the following fluctuation dissipation relations in the frequency domain [38 -- 40]: (cid:104)p∗ j (ω)pk(ω(cid:48))(cid:105) = 2ε0 ω Im{αj}Θ(ω, Tj)δ(ω − ω(cid:48))δjk (1) = = {j, k} {1, 2}, Θ(ω, T ) ω/2 + Here ω/[exp(ω/kBT ) − 1] is the mean thermal energy of a Harmonic oscillator of frequency ω. is the temperature of dipolar object and (cid:104)...(cid:105) denotes statistical ensemble average. When the dipoles are placed close to each other, the interactions between them lead to effective dipole moments: Tj pj = pj + ε0αj[Ejk(pk) · ej] (2) for j (cid:54)= k. Ejk(pk) denotes the electric field at dipole j due to dipole k. It is straightforward to calculate the far field thermal radiation as well as the near-field induced correlations of the effective dipole moments using the general expressions, (cid:20) (cid:114) ε0 µ0 eik0R (k0R) (cid:21) k3 [ej − (ej · eR)eR] 0 4πε0 ik0R − 1 (cid:21) (k0R)2 [ej − 3(ej · eR)eR] (cid:20) pj E(ω) = + H(ω) = k3 0 4πε0 eik0R (k0R) 1 + i k0R (eR × ej)pj (3) (4) for the electromagnetic fields at a point ReR due to a single dipole pjej at the origin with k0 = ω/c being the vacuum wavevector. The intensity flux of thermal radiation is given by the Poynting vector in the far field (Re{E∗(ω) × H(ω)}) and the degree of circular polarization is measured by the spin-angular-momentum density of thermal emission. By generalizing the definitions employed for non-thermal light in several works [22, 30, 41] to thermal radiation [19], we write the spectral energy density (cid:104)W (ω)(cid:105) and spin angular momentum density (cid:104)S(ω)(cid:105) of emitted light in vacuum as: 1 2 1 2ω (cid:104)W (ω)(cid:105) = (cid:104)S(ω)(cid:105) = (cid:104)ε0E∗(ω) · E(ω) + µ0H∗(ω) · H(ω)(cid:105) (5) Im{(cid:104)ε0E∗(ω) × E(ω) + µ0H∗(ω) × H(ω)(cid:105)} (6) Note the use of (cid:104)...(cid:105) which denotes the statistical ensemble average of physical quantities in context of thermally generated radiation. We further define a dimensionless vector quantity called as spectral thermal spin, ST (ω) = ω(cid:104)S(ω)(cid:105) (cid:104)W (ω)(cid:105) (7) whose magnitude in a given direction lies between [−1, 1] with −1 denoting pure LCP light and +1 denoting pure iii RCP light along that direction. In an actual experiment, this direction can be along the axis of a detector. Since the quantity given by Eq.(6) remains largely unexplored in the field of thermal radiation and may not be familiar to researchers working on thermal radiation and heat transfer topics, we point out its connection with closely related well-known concepts. In particular, one can also understand the spin of thermal radiation in more familiar language of Stokes polarimetry. The Stokes S3 parameter describes the circular polarization of the fields lying transverse to a given propagation direction. If one calculates the Stokes S3 parameters for thermally generated fields in all three orthogonal coordinate planes [42], one retrieves a form similar to equation (6). It is also well-known that circularly polarized laser light can impart angular momentum to small, absorptive particles in its path causing them to rotate about their own axis [43]. It is then meaningful to calculate the spin angular momentum of light [Eq. (6)] since optical torque on these particles is proportionate to it [44, 45]. We calculate the spectral thermal spin given by Eq. (7) for the example geometry of two dipolar antennas placed close to each other near the origin with their centers at x1 and x2 = x1 + ded. The fluctuating dipole moments and associated correlations such as (cid:104)p∗ j (ω)pk(ω)(cid:105) are calculated using equations (1-4). The calculation of electromagnetic fields and other related quantities in the far-field at a point ReR is simplified since only the leading order terms O( 1 k0R ) are important. The distances between dipoles and ReR in the far-field are Rj = ReR − xj = R − x · eR. Using simple algebraic manipulations, we derive the following expression for the spectral thermal spin in the far field at a point ReR: (cid:80) j,k=1,2 eik0(xj−xk)·eR(cid:104)p∗ 2Im{eik0(x1−x2)·eR(cid:104)p∗ 1(ω)p2(ω)(cid:105)}[(e1 × e2) · eR]eR j (ω)pk(ω)(cid:105)[(ej · ek) − (ej · eR)(ek · eR)] ST (ω) = (8) (cid:104)p∗ 1(ω)p2(ω)(cid:105) = 2ε0 ωD2 [(α2k3 (cid:104)p1(ω)2(cid:105) = 2ε0 ωD2 [Im{α1}Θ(ω, T1) + Im{α2}Θ(ω, T2)α1k3 0κ2], 0κ)Im{α1}Θ(ω, T1) + (α1k3 0κ)∗Im{α2}Θ(ω, T2)], D = 1 − α1α2k6 0κ2 (cid:104)p2(ω)2(cid:105) = (cid:104)p1(ω)2(cid:105)(1 ↔ 2), The dimensionless near field coupling κ between the two dipoles is given by: (cid:20) eik0d κ = (k0d)2[(e1 · e2) − (e1 · ed)(e2 · ed)] 4π(k0d)3 + (ik0d − 1)[(e1 · e2) − 3(e1 · ed)(e2 · ed)] (cid:21) Equation (8) is the central result of this work which offers many analytic insights regarding generation of CP light as discussed below. Necessary condition for emission of CP light: It follows from Eq.(8) that the spin of far-field thermal emission is always radial which we identify as radial thermal spin SR = ST · eR. Evidently SR = 0 in all directions (eR) when e1 = e2 (parallel dipoles) and Im{eik0(x1−x2)·eR(cid:104)p∗ 1(ω)p2(ω)(cid:105)} = 0 (no correlations). The latter condition can be used to find the necessary condition for emission of CP radiation from an arbitrary body. We note that if instead of two physically separate antennas, a single dipolar object with dipole moments pjej for j ∈ [x, y, z] is considered, then it follows that the correlations Im{(cid:104)p∗ j (ω)pk(ω)(cid:105)} between these orthogonal components are necessary to produce CP radiation in the far-field. Since it is not possible to have nonzero thermal spin with zero correlations between orthogonal components of underlying fluctuating sources, it follows that this is a necessary condition for generation of CP light. This condition can be generalized to arbitrary bodies by discretizing them into subvolumes much smaller than the emission wavelength and conceptualized as electric point dipoles. This approach is known as thermal discrete dipole approximation [46, 47]. It follows that the correlations Im{(cid:104)p∗ j (ω)pk(ω)(cid:105)} between the orthogonal components of the effective dipole moments of these subvolumes are necessary to emit CP light. This finding that the imaginary valued correlations are necessary for emission of CP light is also an important result of this work. It is insightful for designing CP thermal light sources. For instance, the environment surrounding a dipolar thermal emitter can be engineered such that the imaginary valued correlations between the effective dipole moment components are nonzero and consequently, light emitted is circularly polarized. We note that, for arbitrary bodies of non-dipolar nature, the necessary condition may not be sufficient since there can be cancellation of spin due to contribution from dipoles of many tiny subvolumes, requiring a full calculation to infer the circular polarization of total radiation emitted by such a body. Nonetheless, such calculations could be performed with advanced computational tools [47, 48] and using definition (6) above. In this work, we first focus on a simple dipolar dimer which is easier to understand, analyze and optimize. Optimum design for maximum purity CP light along normal direction: We now find a design that emits maximum purity (SR = ±1) CP thermal radiation. To simplify the optimization, we consider the example geometry of Fig.1 with e1 = ex, e2 = ey (antennas lying in the xy-plane) and focus on thermal emission in ez direction. Since the phase factor eik0(x1−x2)·ez = 1 1(ω)p2(ω)(cid:105)} for ez direction, radial spin SR ∼ Im{(cid:104)p∗ and useful analytical expressions can be obtained. Considering a practically relevant situation of equal iv vacuum polarizabilities α1 = α2 = α, one finds that the radial thermal spin SR ∼ Im[αk3 0κ)∗ Θ(ω,T2) Θ(ω,T1) ]. This yields SR = 0 when both the dipolar antennas are at equal temperatures T1 = T2. Therefore, we consider nonequilibrium configuration (T1 (cid:54)= T2) for which the radial thermal spin in the (normal) direction eR = ez is: 0κ + (αk3 SR = (1 + α2)(1 + Θ(ω,T2) Θ(ω,T1)} 2Im{α + α∗ Θ(ω,T2) Θ(ω,T1) ) − 2Re{α + α∗ Θ(ω,T2) Θ(ω,T1)} coupling κ) (k0) follows geometry (near-field the dimensionless, normalized polarizability Here, α = αk3 0κ is introduced to capture the dependence of radial spin SR on material properties (polarizability α), and in a concise manner. wavelength/wavevector that when the normalized polarizability It α = ±i, SR = ± Θ(ω,T1)−Θ(ω,T2) Θ(ω,T1)+Θ(ω,T2) . This gives high purity circular polarization (SR → ±1) when the ratio Θ(ω, T2)/Θ(ω, T1) is either very large or very small. We illustrate with practical examples further below that this dependence on both wavelength and temperature makes strong CP light feasible even when temperatures T1, T2 are not very different. In the following, not restricting ourselves to ez direction, we explore the dependence of handedness on various design parameters. Dependence on antenna temperatures: For small separation between the antennas (k0d < 1), it follows 1(ω)p2(ω)(cid:105)}. Under from Eq.(8) that SR ∝ Im{(cid:104)p∗ this condition, for any geometric configuration of two nonequilibrium coupled antennas of equal polarizabilities (α1 = α2), the thermal spin is flipped (SR → −SR) upon interchanging the temperatures of the antennas. While there is no specific advantage of using antennas of unequal polarizabilities (α1 (cid:54)= α2), similar flipping of handedness (sign of SR) is observed but the change in the magnitude of SR depends on the polarizabilities. As we show further below, it is difficult to realize high purity (SR ∼ 1) CP radiation at large separations due to decreased near-field interactions. But similar tunability of handedness based on temperatures is observed at large separations for weakly polarized light. Dependence on emission direction: We note that only one of the normalized polarizability conditions 0κ = ±i is true depending on the sign of near-field αk3 coupling κ since the condition Im{α} > 0 must hold true for real, passive (lossy) dipoles [49]. For the example configuration with horizontal (ex) and vertical (ey) dipoles with relative orientation ed = (−ex + ey)/ 8π(k0d)3 [(k0d)2 − 3 + Since Im{κ}/Re{κ} (cid:28) 1 and Re{κ} < 0 3ik0d]. for relevant separations k0d ≤ 1, it follows that α = −i is the physically permissible optimum normalized polarizability. Under this condition, radial thermal spin SR = +1 (RCP) when T2 (cid:29) T1 and SR = −1 (LCP) when T2 (cid:28) T1 for emission direction eR = ez 2, the coupling κ = eik0d √ v LCP/RCP emission in northern/southern hemisphere when T2 (cid:28) T1. This flipping of thermal spin SR upon inverting the direction follows from vectorial part of Eq.(8) where the radial spin SR → −SR when eR → −eR. As a consequence, the total (integrated over all directions) angular momentum of emitted radiation is zero as expected for a system lacking any intrinsic source of angular momentum. In context of emission from magneto-optic nanoparticles recently studied in Ref. [19], the cylcotron motion of electrons in presence of applied magnetic field is responsible for generating angular momentum intrinsically and consequently, the total angular momentum radiated by that particle is nonzero and lies along the direction of applied magnetic field (due to its spherical, isotropic shape). Dependence on polarizabilities: The top two figures of fig.2(a) illustrate the handedness distribution for polarizability αk3 0 = 0.22i which leads to the normalized polarizability α = −i, an optimum design for high purity CP light along ±ez direction. Bottom two figures of Fig.2(a) illustrate the change in the handedness distribution upon tuning the polarizability to αk3 0 = 0.2 + 0.22i. With this latter configuration, the maximum purity of the radial spin occurs along intermediate and multiple directions indicating that the direction of maximum purity CP emission depends on the polarizabilities and configuration of the antennas. This example shows that the condition α = ±i obtained above is not a unique and limiting configuration for observing maximum purity CP light from nonequilibrium antennas but other configurations can also be employed. For both these examples, despite complicated directional dependence, northern/southern hemispheres contain the same overall RCP/LCP emission when T2 (cid:29) T1, with handedness flipped upon T2 ↔ T1. Here, we introduce the notation ↔ to denote the interchange of two quantities in a concise manner. This allows us to summarize the above handedness dependence as: RCP ↔ LCP when T2 ↔ T1 or eR ↔ −eR keeping other parameters the same. overall for handedness Dependence on relative locations of antennas: northern/southern The hemispheres is determined by the relative spatial configuration of dipoles ed which affects the permissible optimum polarizability (α = +i or −i). In particular, for √ the configuration discussed above, ed = (−ex + ey)/ 2 leads to α = −i as the physically permissible optimum design along ±ez. In an alternative configuration with 2, α = i is the physically permissible ed = (ex + ey)/ value, and this leads to flipping of RCP ↔ LCP in previous configurations when all other parameters are kept the same. This is illustrated in Fig.2(b) where L-shaped configuration in the x − y plane with hot vertical and cold horizontal antennas leads to RCP/LCP emission in northern/southern hemisphere and opposite distribution is observed for its mirror image √ FIG. 2. Radial spin of thermal emission SR from a dimer of two dipolar antennas is analyzed. Both antennas of dimensionless vacuum polarizability αk3 0 having temperatures T1, T2 and orientations ex, ey respectively, are located at x1 = (0.1,−0.1, 0)/k0 and x2 = (−0.1, 0.1, 0)/k0. (a) demonstrates the distribution of SR for two different αk3 0. Top two figures correspond to optimum design for maximum purity (SR = ±1) along ±ez directions while bottom two figures illustrate maximum purity along some intermediate directions for a 0. The handedness flipping RCP ↔ LCP different value of αk3 is observed upon interchanging temperatures T1 ↔ T2 or inverting emission direction eR → −eR. (b) illustrates the dependence of spin upon the relative spatial orientation of antennas. A mirror image configuration of a given dimer leads to opposite handedness (RCP ↔ LCP flipping) in the same direction keeping all other parameters the same. (north pole). At eR = −ez (south pole), the opposite handedness is observed under the same conditions. Fig.2(a) demonstrates this dependence where the two dipoles are assumed to be located at x1 = (0.1,−0.1, 0) 1 k0 and x2 = (−0.1, 0.1, 0) 1 , having temperatures T1, T2. For top two figures, the vacuum polarizability is αk3 0 = 0.22i such that the optimum normalized polarizability α = −i is realized. This results in RCP/LCP emission in the northern/southern hemisphere when T2 (cid:29) T1 and k0 xzyxyzyxzxyzRadial spin of thermal emission from nonequilibrium antennasαk03 = 0.22 iαk03 = 0.2 + 0.22ixxxxyyyyRCPRCP LCP LCP a)b)Hot Cold-1-0.8-0.6-0.4-0.200.20.40.60.81SRRCPLCPT1T2T2T2T1T1T1T2Dependence on orientations of antennas: (in y − z-plane) counterpart. It further follows from Eq.(8) that RCP ↔ LCP is expected upon interchanging the dipole orientations (e1 ↔ e2) which has an important implication for any isotropic objects. For isotropic objects, the fluctuating dipole moments along all three directions are considered and the radial spin SR is calculated by considering all dipole pairs. It turns out that the radial spin vanishes for all directions due to spin cancellations from conjugate pairs in the same plane. For instance, SR emitted by the pair of dipoles (p1x, p2y) is always opposite in direction to that emitted by the pair (p1y, p2x) due to orientation flipping. Moreover, they are equal in magnitude because (cid:104)p∗ 1y p2x(cid:105) which is true for equal polarizabilities along both directions and provided that the interaction between dipoles is reciprocal. From this, it follows that the anisotropic shape of dipolar objects is necessary to produce CP light. Note that an anisotropic nanoparticle by itself does not emit CP light due to lack of correlations between orthogonal components of its fluctuating dipole moments. 1x p2y(cid:105) = (cid:104)p∗ Dependence on distance between antennas: While the dependence of thermal spin on distance d between the dipolar antennas is quite complicated 0κ = ±i [Eq.(8)], the optimum polarizability condition αk3 provides some useful insights. Since κ depends inversely on the distance k0d, it follows that the polarizabilities (αk3 0) and the separation distance (k0d) required for the optimum design are also inversely related. For small distances (k0d (cid:28) 1), the phase factor eik0d(ed·eR) ∼ 1 in Eq.(8). Under this condition, one can realize maximum purity CP thermal emission with small polarizabilities (αk3 0 < 1) using nonequilibrium antennas discussed above. As we show below, the emission from the antennas at equal temperatures in this regime is at best very weakly polarized. For large separation distances (k0d > 1) corresponding to negligible near-field interactions, the phase factor eik0d(ed·eR) matters and one can numerically optimize the design. However, since κ ∝ (k0d)3 , the 0 (cid:29) 1) and required polarizabilities are very large (αk3 quite difficult to realize with real, practical systems. Antennas at equal temperatures: For small separation distances k0d (cid:28) 1, the phase factor eik0d(ed·eR) ∼ 1 simplifies Eq.(8). It then follows that for two antennas of polarizabilities α1, α2 having radial spin SR ∝ equal temperatures (T1 = T2), Imκ[Imα1Reα2 − Imα2Reα1]. Thus, SR = 0 not only for equal vacuum polarizabilities (α1 = α2) but also when α2/α1 ∈ R where R stands for real numbers. Even after overcoming the difficulty of achieving α2/α1 /∈ R with engineered design of materials and geometrical shapes, it turns out that, such a dimer produces very weakly polarized radiation. This occurs because of its direct dependence on the dissipative (imaginary) part of 1 vi near field coupling (SR ∝ Im{κ}) which is very small (Im{κ} (cid:28) κ) for relevant separations k0d (cid:46) 1. For nonequilibrium antennas discussed above, the thermal spin SR is non-negligible since it also depends on Re{κ} because of unequal temperatures. A dimer of antennas at equal temperatures and having equal (or unequal) polarizabilities always produces unpolarized (or very weakly polarized) thermal radiation. A metasurface of We note that in order to produce strong CP thermal radiation with antennas at equal temperatures, a many body system can be considered where the antennas are anisotropic and arranged in a staggered manner so that imaginary-valued correlations between orthogonal dipole moments can be non-negligible through many-body interactions. such anisotropic equilibrium antennas is already explored in Ref. [18] where a Kagome lattice of silicon carbide nanorods is considered. In that work, the resulting emission is predicted to be CP based on the spin-split dispersion of underlying modes achieved with inversion asymmetric metasurface. While the prediction is based on a concrete underlying mechanism, it can be analyzed quantitatively by extending our fluctuational electrodynamic analysis to a many-body system of such dipolar antennas. Summary of fluctuational electrodynamic analysis: In the following, we point out the important aspects of CP thermal radiation from nonequilibrium coupled antennas by summarizing the dependence of radial spin SR upon a select few parameters analyzed above: 1. SR ↔ −SR when T1 ↔ T2: The polarization state or handedness of emitted light can be reconfigured by interchanging the temperatures. 2. SR → −SR when eR → −eR: Total angular momentum of emitted radiation is zero. 3. SR ↔ −SR when e1 ↔ e2: Cancellation of spin of thermal emission from isotropic dipolar particles. Anisotropic shape is required for generation of CP light 4. The maximum purity of CP light is max{SR} from = nonequilibrium antennas ± Θ(ω,T1)−Θ(ω,T2) Its magnitude depends only Θ(ω,T1)+Θ(ω,T2) . on the emission wavelength and the antenna temperatures while its direction depends on polarizabilities and geometric configuration of antennas. Thermal nonequilibrium enables strong CP thermal radiation from a compact dimer of antennas that produces very weakly polarized emission at equilibrium (equal temperatures). Strong CP thermal emission from systems at thermal equilibrium is possible by using optimized chiral absorber metasurfaces [27] or many-body configurations of antennas [18, 28]. However, III. PRACTICAL IMPLEMENTATION G(ω) = i 8π [rs − (1 − k2 p)rp]e2i vii along minor axis is given by (9) with the geometrical factor N(cid:48) = (1 − N )/2. It is well-known that both polarizabilities are resonantly enhanced (minimization of denominator) at two different frequencies. For the purpose of illustration, we focus on CP thermal emission at frequencies of dipolar resonant modes along major axes. At the corresponding resonant wavelength, the polarizabilities along minor axes are orders of magnitude weak and negligibly affect the dominant thermal emission from polarizabilities along major axes. The polarizability of these prolate ellipsoids is further influenced by the presence of the substrate. We account for these changes within the dipolar approximation using well-known Green's function technique [37, 51]. The modified effective polarizability of the dipolar antennas is αeff(ω) = α(ω)/[1− α(ω)k3 0G(ω)] where k0 = ω/c. The expression (cid:90) ∞ 0 kp(cid:113) 1 − k2 p dkp √ 1−k2 pds (11) is calculated numerically and it depends on the distance ds of the antenna dipole from the surface of the metallic heater. rs, rp denote the usual Fresnel reflection coefficients for light incident from vacuum onto the substrate. We assume the permittivity of the heater to be εm = −40 + 10i for calculation of Fresnel reflection coefficients. We pick this arbitrary value to reflect the fact that any highly reflective metallic surface such as tungsten, doped semiconductor, metal nitrides can be used in its place. The dependence of the results upon changes in the large negative substrate permittivity is negligible. We note that the dipolar analysis is not accurate for large field gradients and higher order multipoles contributing when the antennas are separated by small surface-to-surface (antenna surface) separations ds such that ds/Ra < 1. We avoid this regime for illustration purpose. Figure 3 describes the thermal emission spin (ST ) along the normal direction (ez) from the dimer pairs. It shows its dependence upon the separation between the antennas d and their temperatures, Tc = 300K (cold) and Th = Tc + ∆T (hot). We assume that the dimer pairs are separated by distance much larger than d and use the arrays to motivate an actual experiment. The magnitude of ST · ez for two different separations is shown by light and dark orange (a) and green (b) curves and that for two different ∆T is shown by solid and dashed lines. As depicted in the inset of Fig.3(a), we first consider a configuration where horizontal dipoles (in x− y plane) are heated to high temperature Th while the vertical dipoles are maintained at a lower temperature Tc. This results in the emission of RCP light along the normal ez direction. When the temperatures of the antennas are flipped leading to a configuration depicted dynamical reconfigurability of circular polarization state with such solid state designs is a major unsolved problem. Thermal nonequilibrium enables the reconfigurability by simply interchanging the temperatures. Furthermore, reconfigurability can be implemented at arbitrary operating temperatures unlike other potential options of reconfigurability such as phase change materials that are limited to certain operating temperatures due to Thermal material specific transition temperatures. nonequilibrium implemented in a dimer system of antennas can thus achieve three important features of efficient CP light sources namely, high purity, compactness and reconfigurability of the polarization state. In the following, we explore the experimental feasibility of this mechanism with suitable practical examples. As a potential experimental system to observe CP thermal emission from nonequilibrium antenna dimers, we consider a system depicted in the insets of Fig.3 comprising of two arrays of Silicon Carbide (SiC) dipolar nanoantennas fabricated on top of suitable micro-heaters. Silicon Carbide is considered a good material choice for studying thermal emission due to large quality factor (Q ∼ 103) phonon-polaritonic resonances which occur around room-temperature thermal wavelengths (λ ∼ 10µm). The localized surface polaritonic modes supported by these anisotropic antennas resonantly enhance the polarizabilities (α1, α2) as well as the interactions between the antennas. We consider the antennas having length, breadth and height of 0.4µm, 0.1µm and 0.1µm respectively. These can be approximated as prolate ellipsoids [49] having radii Ra = 200nm (major axis) and Rb = 50nm (minor axis). For the realistic examples considered here, the polarizability along the major axis is: α(ω) = πRaR2 b 4 3 (ε(ω) − 1) 1 + N [ε(ω) − 1] eccentricity of the ellipsoid ec =(cid:112)1 − (Rb/Ra)2 through where N is the geometrical factor dependent on the (9) the following relation [49]: (cid:18) N = 1 − e2 e2 c c − 1 + 1 2ec ln 1 + ec 1 − ec ε(ω) is the permittivity of SiC taken from reference [50] and is given below: ε(ω) = ε∞ (10) where ε∞ = 6.7, ωLO = 1.825×1014rad/s, ωTO = 1.494× 1014rad/s and Γ = 8.966 × 1011rad/s. The polarizability (cid:20) ω2 − ω2 ω2 − ω2 LO + iΓω TO + iΓω (cid:19) (cid:21) viii FIG. 3. Two arrays of horizontally and vertically orientated Silicon Carbide nanoantennas spatially separated by d and fabricated on top of two micro-heaters maintained at temperatures Tc = 300K (cold) and Th (hot). Length, breadth, height of each antenna are 0.4µm, 0.1µm, 0.1µm. respectively. (a) When horizontal antennas are at hotter temperature than the vertical antennas, resulting emission is RCP. (b) The handedness of emission is switched to LCP when the temperatures are interchanged. in the inset of Fig.3(b), the resulting emission is LCP keeping all other parameters the same. Since microheater temperatures can be tuned, this enables reconfigurability of the polarization state of the emitted radiation. As described earlier, for given temperatures Th and Tc, the maximum purity that can be reached upon design optimization is Θ(ω,Th)−Θ(ω,Tc) Θ(ω,Th)+Θ(ω,Tc) and depends on both wavelength and temperatures. At mid-infrared wavelength of 11.6µm, the proposed device emits strong circularly polarized light with thermal spin ST ∼ 0.65 for ∆T = 200K and ST ∼ 0.3 for ∆T = 50K. Note that the SiC antennas also exchange near-field radiative heat flux of the order of (cid:46) 1nW for these configurations (not shown). This near-field heat flux affects the temperatures of antennas. This effect can be monitored in a real experiment and can be analyzed using a separate thermal model. Here, we use reasonable values of steady state temperatures of the antennas for the calculations. Overall, this analysis of the thermal spin of emission from SiC antenna dimers indicates feasibility of observing reasonably strong CP light under temperature difference of ∆T ∼ 50K across separation of d (cid:38) 0.5µm between the antennas. One can also consider other Restrahlen materials [52] such as Indium Phosphide (InP), Gallium Arsenide (GaAs), Indium Antimonide (InSb) etcetera which support localized polaritons at longer wavelengths and can allow generation of CP light for larger separations. As an example, figure 4(a) demonstrates CP-light generation from a dimer of coupled nonequilibrium InP nanoantennas. The permittivity of InP is obtained from Ref. [50] and is described by the Lorentz oscillator model given in Eq.(10) with following parameteres. ε∞ = 9.61, ωLO = 6.498 × 1013rad/s, ωTO = 5.720 × 1013rad/s and Γ = 6.495 × 1011rad/s. Length, breadth and FIG. 4. (a) Thermal emission spin (ST ) for a dimer of InP antennas illustrates feasibility of observing mid-IR CP light with slightly larger gap size for similar ∆T = Th − Tc, compared to SiC antennas analyzed in figure 3. This figure shows that CP light generation from nonequilibrium antennas is not limited to one particular material choice. Figure (b) demonstrates a conceptual example of a dimer system of antennas where one of the antennas contains gain medium i.e. a plasmonic antenna enclosed in a dye-doped shell. Well below the lasing threshold, the incoherent amplified spontaneous emission (ASE) from the gain antenna coupled with another nearby antenna can lead to high- purity CP light at suitable separations. This example shows that the proposed mechanism is not limited to "thermal" nonequilibrium but can be potentially implemented in other forms such as population inversion of gain atoms. height of each antenna are assumed to be 0.8µm, 0.4µm, 0.4µm. respectively. Figure 4(a) shows the spectral thermal spin of emitted radiation when the antennas are separated by a fixed separation of d ≈ 1.1µm with temperatures Tc = 100K and Th (hot). As shown, the thermal emission spin SR ∼ 0.15 is obtained for a temperature difference of ∆T = 30K, indicating the feasibility of observing CP light with smaller ∆T and larger d compared to SiC antennas. This example shows that other materials, operating temperatures and dimer configurations can be explored to optimize emission of CP light from nonequilibrium antennas. While we focused so far on thermal nonequilibrium, other forms of nonequilibrium can be potentially implemented using the same dimer configuration. One example is amplified spontaneous emission (ASE) from one of the antennas containing active gain medium where nonequilibrium is in the form of population inversion of gain atoms. The incoherent ASE noise below loss compensation (lasing threshold) is described using fluctuational electrodynamic theory [53, 54]. It arises from the fluctuating dipole moments associated with the atomic transitions of gain atoms. These fluctuating sources have an effective temperature dependent on the population inversion and their correlations are described using specific fluctuation-dissipation relations [53, 55]. As depicted in the inset of fig.4(b), we consider a metallic gold, silver) antenna enclosed in a dye-doped (e.g. shell which acts as an amplifying medium. Such dLCPTcThxyλ (μm)ST # z 12.21211.811.611.40-0.1-0.2-0.3-0.4-0.5-0.60.6x + 0.3y 0.4x + 0.2yd (μm)502000.6x + 0.3y 0.4x + 0.2yd (μm)50200Th-Tc (K)TcxyThRCP d0.60.50.40.30.20.1011.411.611.81212.2ST # z λ (μm)Th-Tc (K)d (μm)dThTcyx1x + 0.4yTh-Tc (K)30100InP nanoantennas 00.050.10.150.20.250.329303132λ (μm)ST " z 1.31.51.71.92.100.20.40.60.81dyx120x + 60yd (nm)200x + 60yAmplified SpontaneousEmission λ (μm)ST " z intricate systems have been considered experimentally in context of plasmonic nanolasers [56, 57]. While large pump requirements for plasmonic lasing in these geometries make the experimental realization of lasing challenging [58], here we operate with lower gain values (well below lasing threshold) and consider its use for circularly polarized ASE from the dimer. For this conceptual example, we consider the gain nanoantenna to have an effective permittivity: ε(ω) = 1 − (cid:124) ω2 p + iΓωp ω2 p (cid:123)(cid:122) εr (cid:123)(cid:122) εg D0γ⊥ + (ω − ω21) + iγ⊥ (cid:125) (cid:124) (cid:125) (12) consisting of Drude part εr and a gain medium part εg. The transition frequency of gain atoms is ω21 while the strength of (pump-tunable) population inversion is characterized by D0. We choose the parameters ωp = 4.38 × 1015rad/s, Γ = ωp/200, ω21 = 1.2 × 1015rad/s, γ⊥ = ω21/100. The gain is chosen to be D0 = 0.15 such that Im{ε} > 0 at all wavelengths ensuring ASE regime well below loss-compensation. We assume the other nanoantenna to be characterized by the permittivity εr. Length, breadth and height of each antenna are 100nm, 25nm, 25nm respectively. The polarizabilities α1,2(ω) are calculated using Eq.(9). The fluctuating polarization associated with ASE from the gain antenna is described by the fluctuation dissipation relation [55]: Im{α1(ω)} Imεg(ω) Imε(ω) (cid:104)p∗ 1g(ω)p1g(ω(cid:48))(cid:105) = −n2ω21 n2 − n1 2ε0 ω δω,ω(cid:48) (13) where n2, n1 represent the populations in excited and ground states of gain atoms. We assume n2 = 0.95n, n1 = n − n2 with n as the total number of gain atoms. Since ASE is much larger than the thermal radiation from the antennas, it suffices to calculate the spectral thermal spin of ASE given by equation (8) and calculated using above correlations. Figure 4(b) demonstrates the spectral thermal spin for two different separations d between the antennas. For the above parameters, the highest purity CP light (SR = 1) is obtained at a wavelength λ = 1.55µm, for d = 134nm as shown by the dark blue curve. This conceptual example shows that the proposed mechanism for generating CP light using nonequilibrium antennas is not limited to systems under thermal nonequilibrium. Apart from the use of active gain medium, one can also consider use of biased semiconductors where the heat bath of underlying fluctuating currents has an effective chemical potential or temperature that can be tuned by changing the voltage bias [54, 59]. These other forms of nonequilibrium requiring more experimental [56, 57] and theoretical details can be explored in separate future works. Summary of potential experimental platforms: For mid-IR CP thermal emission based on thermal ix nonequilibrium, one important technical difficulty is that of maintaining large temperature difference across small gaps for high-purity CP light. However, we are confident that the proposed mechanism of CP-light generation can be experimentally implemented in the near future. In particular, there has been a significant experimental progress in the area of near-field radiative heat transfer which has facilitated exploration of thermal nonequilibrium at nanoscale and has also verified the validity of fluctuational electrodynamic theory [24, 25]. In particular, much larger temperature differences (∆T (cid:38) 100K) across small gaps d (cid:46) 1µm. have been experimentally probed in suitable geometries [60, 61]. Our analysis of SiC and InP antennas above provides an estimate that reasonably strong CP light (ST ∼ 0.2) can be observed with such dimers held at temperature difference as small as ∆T ∼ 30K and separated by a gap size of d ∼ 1µm. Such an experiment will be fundamentally important as it will reveal for the first time the non-intuitive connection between thermal nonequilibrium and angular momentum of light. In the long run, CP-light emission at near-IR wavelengths and other practical applications such as CP light emitting LEDs can be pursued by exploring other nonequilibrium systems considered above. In comparison to other CP light sources based on structural chirality or polarization conversion, reconfigurability is an important distinct advantage of the proposed mechanism. Related to the tunability of device temperature in context of CP thermal emission, recent works [36, 62, 63] have explored dynamical modulation of thermal emission in two-dimensional opto-electronic platforms where ultra-fast switching rates (cid:38) 100MHz have been realized. We therefore envision that in the long run, the proposed mechanism can be combined with such newly emerging concepts to build a compact, high purity and dynamically reconfigurable source of CP light. IV. CONCLUSION We demonstrate a mechanism of CP thermal radiation based on near-field coupling between nonequilibrium antennas. It is unrelated to geometric and material chirality or use of any polarization conversion device. We show that a simple dimer of coupled nonequilibrium antennas can facilitate a great degree of control over the polarization of emitted radiation upon tuning the temperatures, emission direction, relative orientations and positions of antennas. Through a rigorous fluctuational electrodynamic analysis of the dependence on these parameters, we reveal the fundamental origin and describe the general design guidelines for generating CP light Our analysis revealed that the imaginary valued correlations between orthogonal components of fluctuating sources from fluctuating thermal sources. are necessary for emission of CP light. In the context of thermal emission from coupled dipolar thermal sources, anisotropic shape is necessary to emit CP thermal radiation. While the computational design with such dipolar bodies is simple and convenient, one can also go beyond this regime with advanced computational tools and inquire about the circular polarization described by Eq.(6) of thermal emission from arbitrary, non-intuitive geometries [48]. We further explore the experimental feasibility of generating CP light from nonequilibrium antennas with realistic examples of SiC and InP antennas. A reasonably strong CP light from nonequilibrium antennas can be detected in the near future using these example systems. To show that the proposed mechanism is not limited to "thermal" nonequilibrium, we further provide a conceptual example of plasmonic nanolaser system that can emit circularly polarized amplified spontaneous emission by operating well below lasing threshold. Consideration of such nontrivial approaches is important for practical applications such as CP-light-emitting LEDs. The underlying approach of analyzing spin angular momentum property of thermal radiation remains largely unexplored in the field of thermal radiation [19]. This approach opens the door to numerous future studies of angular-momentum-related radiative heat transport phenomena. For instance, one immediate extension of the current work can be consideration of many-body configurations of anisotropic dipolar antennas for shaping angular-momentum properties of light, using thermal nonequilibrium without applying magnetic field. Our analysis also shows a spatial distribution of angular momentum radiated by nonequilibrium antennas. This implies that nontrivial torques on nanoscale bodies can be expected in the vicinity of these antennas. Interestingly, this is already probed by other recent work [35] which demonstrates thermal nonequilibrium enabled torques with temperature dependent sign and magnitude. This surprising fundamental connection between thermal nonequilibrium and angular momentum of in the context of not only nonequilibrium but also nonisothermal bodies. From the perspective of nanoscale thermometry, detection and sensing applications, one question that arises is how angular momentum of emitted radiation is influenced by the temperature gradients in nonisothermal bodies which requires an answer necessarily within fluctuational electrodynamic theory and yet remains unsolved. We leave these inquiries aside for future work. suggests new possibilities light V. ACKNOWLEDGMENTS We would like to thank Ryan Starko-Bowes and Aman Satija for helpful discussions. This work was supported x by the U.S. Department of Energy, Office of Basic Energy Science under award number DE-SC0017717 and the Lillian Gilbreth Postdoctoral Fellowship program at Purdue University (C.K.). ∗ [email protected] † [email protected] [1] B. Le Feber, N. Rotenberg, and L. Kuipers, Nanophotonic control of circular dipole emission, Nat. Commun. 6, 6695 (2015). [2] R. Mitsch, C. Sayrin, B. Albrecht, P. Schneeweiss, and A. Rauschenbeutel, Quantum state-controlled directional spontaneous emission of photons into a nanophotonic waveguide, Nat. Commun. 5, 5713 (2014). [3] I. Zuti´c, J. Fabian, and S.D. Sarma, Spintronics: Fundamentals and applications, Rev. Mod. Phys. 76, 323 (2004). [4] P. Lodahl, S. Mahmoodian, Stobbe, A. Rauschenbeutel, P. Schneeweiss, J. Volz, H. Pichler, and P. Zoller, Chiral quantum optics, Nature 541, 473 (2017). S. [5] H.L. Zhu, S.W. Cheung, K.L. Chung, and T.I. Yuk, Linear-to-circular polarization conversion using metasurface, IEEE Trans. Antennas. Propag. 61, 4615 (2013). [6] C. Pfeiffer, C. Zhang, V. Ray, L.J. Guo, and A. Grbic, High performance bianisotropic metasurfaces: asymmetric transmission of light, Phys. Rev. Lett. 113, 023902 (2014). [7] Y. Jiang, L. Wang, J. Wang, C.N. Akwuruoha, and reflective W. Cao, Ultra-wideband high-efficiency linear-to-circular on metasurface at terahertz frequencies, Opt. Express 25, 27616 (2017). polarization converter based [8] N. Nishizawa, K. Nishibayashi, and H. Munekata, Pure circular polarization electroluminescence at room temperature with spin-polarized light-emitting diodes, Proc. Natl. Acad. Sci. 114, 1783 (2017). [9] P. Asshoff, A. Merz, H. Kalt, and M. Hetterich, A spintronic source of circularly polarized single photons, Appl. Phys. Lett. 98, 112106 (2011). [10] D. Di Nuzzo, C. Kulkarni, B. Zhao, E. Smolinsky, F. Tassinari, S.C.J. Meskers, R. Naaman, E.W. and R.H. Friend, High circular polarization Meijer, self-assembly of of a light-emitting chiral conjugated polymer into multidomain cholesteric films, ACS Nano 11, 12713 (2017). electroluminescence achieved via [11] D. Zhao, H. He, X. Gu, L. Guo, K.S. Wong, J.W.Y. Lam, and B.Z. Tang, Circularly polarized luminescence and a reflective photoluminescent chiral nematic liquid crystal display based on an aggregation-induced emission luminogen, Adv. Opt. Mater. 4, 534 (2016). [12] Y.J. Zhang, T. Oka, R. Suzuki, J.T. Ye, and Y. Iwasa, light-emitting transistor, Electrically switchable chiral Science 344, 725 -- 728 (2014). [13] J. Kumar, T. Nakashima, and T. Kawai, Circularly polarized and supramolecular assemblies, J. Phys. Chem. Lett. 6, 3445 (2015). chiral molecules luminescence in [14] E. M. S´anchez-Carnerero, A.R. Agarrabeitia, F. Moreno, B. L. Maroto, G. Muller, M.J. Ortiz, and S. de la Moya, Circularly polarized luminescence from simple organic molecules, Chem.: Eur. J. 21, 13488 (2015). [15] K. Konishi, M. Nomura, N. Kumagai, S. Iwamoto, Y. Arakawa, and M. Kuwata-Gonokami, Circularly polarized light emission from semiconductor planar chiral nanostructures, Phys. Rev. Lett. 106, 057402 (2011). [16] A.A. Maksimov, I.I. Tartakovskii, E.V. Filatov, S.G. Tikhodeev, S.V. Lobanov, N.A. Gippius, C. Schneider, M. Kamp, S. Maier, S. Hofling, and V.D. Kulakovskii, Circularly polarized light emission from chiral spatially-structured planar semiconductor microcavities, Phys. Rev. B. 89, 045316 (2014). [17] S. L. Wadsworth, P. G. Clem, E. D. Branson, and G. D. Boreman, Broadband circularly-polarized infrared emission from multilayer metamaterials, Opt. Mater. Express 1, 466 -- 479 (2011). [18] N. Shitrit, I. Yulevich, E. Maguid, D. Ozeri, D. Veksler, V. Kleiner, and E. Hasman, Spin-optical metamaterial route to spin-controlled photonics, Science 340, 724 -- 726 (2013). [19] A. Ott, P. Ben-Abdallah, and S.A. Biehs, Circular heat and momentum flux radiated by magneto-optical nanoparticles, Phys. Rev. B. 97, 205414 (2018). [20] S.M. Rytov, Theory of electric fluctuations and thermal radiation, AFCRC-TR 59, 162 (1959). [21] L.D. Landau and E.M. Lifshitz, Course of theoretical physics, volume 5, Publisher: Butterworth-Heinemann 3 (1980). [22] S. M. Barnett, L. Allen, and M. J. Padgett, Optical angular momentum (CRC press, 2016). [23] L. Allen, M. W. Beijersbergen, R.J.C. Spreeuw, and J.P. Woerdman, Orbital angular momentum of light and the transformation of laguerre-gaussian laser modes, Phys. Rev. A 45, 8185 (1992). [24] B. Song, A. Fiorino, E. Meyhofer, and P. Reddy, Near-field radiative thermal transport: From theory to experiment, AIP Adv. 5, 053503 (2015). [25] K. Kim, B. Song, V. Fern´andez-Hurtado, W. Lee, W. Jeong, L. Cui, D. Thompson, J. Feist, M.T.H. Reid, F.J. Garc´ıa-Vidal, J.C. Cuevas, E. Meyhofer, and P. Reddy, Radiative heat transfer in the extreme near field, Nature 528, 387 (2015). [26] E. Tervo, E. Bagherisereshki, and Z. Zhang, Near-field radiative thermoelectric energy converters: a review, Front. Energy 12, 5 -- 21 (2018). [27] C. Wu, N. Arju, G. Kelp, J. A. Fan, J. Dominguez, E. Gonzales, E. Tutuc, and G. Shvets, Spectrally selective chiral silicon metasurfaces based on infrared fano resonances, Nat. Commun. 5, 3892 (2014). and H. Giessen, Interpreting chiral nanophotonic spectra: the plasmonic born -- kuhn model, Nano Lett. 13, 6238 (2013). [28] X. Yin, M. Schaferling, B. Metzger, I. Brener, [29] K.Y. Bliokh, A.Y. Bekshaev, and F. Nori, Optical momentum, spin, and angular momentum in dispersive media, Phys. Rev. Lett. 119, 073901 (2017). [30] K.Y. Bliokh, A.Y Bekshaev, and F. Nori, Dual electromagnetism: helicity, spin, momentum and angular momentum, New J. Phys. 15, 033026 (2013). [31] T. Van Mechelen Jacob, Universal spin-momentum locking of evanescent waves, Optica 3, 118 -- 126 (2016). and Z. xi [32] E. Moncada-Villa, V. Fern´andez-Hurtado, F.J. Garcia-Vidal, A. Garc´ıa-Mart´ın, and J.C. Cuevas, Magnetic field control of near-field radiative heat transfer and the realization of highly tunable hyperbolic thermal emitters, Phys. Rev. B. 92, 125418 (2015). [33] W. Jin, A.G. Polimeridis, and A.W. Rodriguez, Temperature control of thermal radiation from composite bodies, Phys. Rev. B. 93, 121403(R) (2016). [34] S.E. Han and D.J. Norris, Control of thermal emission by selective heating of periodic structures, Phys. Rev. Lett. 104, 043901 (2010). [35] M.T.H. Reid, O.D. Miller, A.G. Polimeridis, A.W. Rodriguez, E.M. Tomlinson, and S.G. Johnson, Photon torpedoes and rytov pinwheels: Integral-equation modeling of non-equilibrium fluctuation-induced forces and torques on nanoparticles, arXiv:1708.01985 (2017). [36] E. Sakat, L. Wojszvzyk, J.P. Hugonin, M. Besbes, C. Sauvan, and J. Greffet, Enhancing thermal radiation with nanoantennas to create infrared sources with high modulation rates, Optica 5, 175 -- 179 (2018). [37] L. Novotny and B. Hecht, Principles of nano-optics (Cambridge university press, 2012). [38] P. Ben-Abdallah, S.A. Biehs, and K. Joulain, Many-body radiative heat transfer theory, Phys. Rev. Lett. 107, 114301 (2011). [39] G. Domingues, S. Volz, K. Joulain, and J. Greffet, Heat transfer between two nanoparticles through near field interaction, Phys. Rev. Lett. 94, 085901 (2005). [40] K. Joulain, J. Mulet, F. Marquier, R. Carminati, and J. Greffet, Surface electromagnetic waves thermally excited: Radiative heat transfer, coherence properties and casimir forces revisited in the near field, Surf. Sci. Rep. 57, 59 -- 112 (2005). [41] M.V. Berry, Paraxial beams light, conference on singular optics 3487, spinning of International 6 -- 12 (1998). [42] T. Setala, M. Kaivola, and A.T. Friberg, Degree of polarization in near fields of thermal sources: effects of surface waves, Phys. Rev. Lett. 88, 123902 (2002). [43] O.V. Angelsky, A.Y. Bekshaev, P.P. Maksimyak, A.P. Maksimyak, I.I. Mokhun, S.G. Hanson, C.Y. Zenkova, and A.V. Tyurin, Circular motion of particles suspended in a gaussian beam with circular polarization validates the spin part of the internal energy flow, Opt. Express 20, 11351 (2012). [44] A. Canaguier-Durand, A. Cuche, C. Genet, and T.W. Ebbesen, Force and torque on an electric dipole by spinning light fields, Phys. Rev. A. 88, 033831 (2013). [45] M. Nieto-Vesperinas, Optical torque: Electromagnetic spin and orbital-angular-momentum conservation laws and their significance, Phys. Rev. A 92, 043843 (2015). [46] S. Edalatpour and M. Francoeur, The thermal discrete dipole approximation (t-dda) for near-field radiative heat transfer simulations in three-dimensional arbitrary geometries, J. Quant. Spectrosc. Radiat. Transf. 133, 364 -- 373 (2014). [47] S. Edalatpour, M. Cuma, T. Trueax, R. Backman, and M. Francoeur, Convergence analysis of the thermal discrete dipole approximation, Phys. Rev. E. 91, 063307 (2015). [48] MT Homer Reid and Steven G Johnson, Efficient computation of power, force, and torque in bem scattering calculations, IEEE Trans. Antennas Propag 63, 3588 -- 3598 (2015). [49] C.F. Bohren and D.R. Huffman, Absorption and scattering of light by small particles (John Wiley & Sons, 2008). [50] E.D. Palik, Handbook of optical constants of solids, Vol. 3 (Academic press, 1998). [51] B. Amorim, P.A.D. Gon¸calves, M.I. Vasilevskiy, and N.M.R. Peres, Impact of graphene on the polarizability of a neighbour nanoparticle: a dyadic greens function study, Appl. Sci. 7, 1158 (2017). [52] J.D. Caldwell, L. Lindsay, V. Giannini, I. Vurgaftman, and O.J. Glembocki, T.L. Reinecke, S.A. Maier, Low-loss, infrared and terahertz nanophotonics using surface phonon polaritons, Nanophotonics 4, 44 -- 68 (2015). [53] R. Matloob, R. Loudon, M. Artoni, S.M. Barnett, and J. Jeffers, Electromagnetic field quantization in amplifying dielectrics, Phys. Rev. A. 55, 1623 (1997). [54] C.H. Henry and R.F. Kazarinov, Quantum noise in photonics, Rev. Mod. Phys. 68, 801 (1996). [55] C. Khandekar, W. Jin, O.D. Miller, A. Pick, and A.W. Rodriguez, Giant frequency-selective near-field energy transfer in active -- passive structures, Phys. Rev. B. 94, 115402 (2016). [56] M.A. Noginov, G. Zhu, A.M. Belgrave, R. Bakker, V.M. Shalaev, E.E. Narimanov, S. Stout, E. Herz, T. Suteewong, and U. Wiesner, Demonstration of a spaser-based nanolaser, Nature 460, 1110 (2009). xii [57] K.G. Stamplecoskie, M. Grenier, and J.C. Scaiano, Self-assembled dipole nanolasers, J. Am. Chem. Soc. 136, 2956 -- 2959 (2014). [58] M. Premaratne and M.I. Stockman, Theory and technology of spasers, Adv. Opt. Photonics 9, 79 -- 128 (2017). [59] K. Chen, P. Santhanam, S. Sandhu, L. Zhu, and S. Fan, Heat-flux control and solid-state cooling by regulating chemical potential of photons in near-field electromagnetic heat transfer, Phys. Rev. B. 91, 134301 (2015). [60] M.P. Bernardi, D. Milovich, and M. Francoeur, Radiative heat transfer exceeding the blackbody limit between macroscale planar surfaces separated by a nanosize vacuum gap, Nat. Commun 7, 12900 (2016). [61] M. Ghashami, H. Geng, T. Kim, N. Iacopino, S.K. Cho, and K. Park, Precision measurement of phonon-polaritonic near-field energy transfer between macroscale planar thermal gradients, Phys. Rev. Lett. 120, 175901 (2018). structures under large [62] T. Mori, Y. Yamauchi, S. Honda, and H. Maki, An electrically driven, ultrahigh-speed, on-chip light emitter based on carbon nanotubes, Nano Lett. 14, 3277 (2014). [63] Y.D. Kim, Y. Gao, R.J. Shiue, L. Wang, O.B. Aslan, M.H. Bae, H. Kim, D. Seo, H.J Choi, S.H. Kim, A. Nemilentsau, T. Low, C. Tan, D.K. Efetov, T. Taniguchi, K. Watanabe, K.L. Shepard, T.F. Heinz, D. Englund, and J. Hone, Ultrafast graphene light emitters, Nano Lett. 18, 934 -- 940 (2018). 

1503.02134

1

1503

First-principles calculation of the thermoelectric figure of merit for [2,2]paracyclophane-based single-molecule junctions

Here we present a theoretical study of the thermoelectric transport through {[}2,2{]}para\-cyclo\-phane-based single-molecule junctions. Combining electronic and vibrational structures, obtained from density functional theory (DFT), with nonequilibrium Green's function techniques, allows us to treat both electronic and phononic transport properties at a first-principles level. For the electronic part, we include an approximate self-energy correction, based on the DFT+$\Sigma$ approach. This enables us to make a reliable prediction of all linear response transport coefficients entering the thermoelectric figure of merit $ZT$. Paracyclophane derivatives offer a great flexibility in tuning their chemical properties by attaching different functional groups. We show that, for the specific molecule, the functional groups mainly influence the thermopower, allowing to tune its sign and absolute value. We predict that the functionalization of the bare paracyclophane leads to a largely enhanced electronic contribution $Z_{\mathrm{el}}T$ to the figure of merit. Nevertheless, the high phononic contribution to the thermal conductance strongly suppresses $ZT$. Our work demonstrates the importance to include the phonon thermal conductance for any realistic estimate of the $ZT$ for off-resonant molecular transport junctions. In addition, it shows the possibility of a chemical tuning of the thermoelectric properties for a series of available molecules, leading to equally performing hole- and electron-conducting junctions based on the same molecular framework.

cond-mat.mes-hall

cond-mat

First-principles calculation of the thermoelectric figure of merit for [2,2]paracyclophane-based single-molecule junctions Marius Bürkle,1, ∗ Thomas J. Hellmuth,2, 3 Fabian Pauly,3 and Yoshihiro Asai1 1Nanosystem Research Institute (NRI) 'RICS', National Institute of Advanced Industrial Science and Technology (AIST), Umezono 1-1-1, Tsukuba Central 2, Tsukuba, Ibaraki 305-8568, Japan 2Institut für Theoretische Festkörperphysik, Karlsruhe Institute of Technology, 76131 Karlsruhe, Germany 3Department of Physics, University of Konstanz, 78457 Konstanz, Germany 5 1 0 2 r a M 7 ] l l a h - s e m . t a m - d n o c [ 1 v 4 3 1 2 0 . 3 0 5 1 : v i X r a Here we present a theoretical study of the thermoelectric transport through [2,2]paracyclophane- based single-molecule junctions. Combining electronic and vibrational structures, obtained from density functional theory (DFT), with nonequilibrium Green's function techniques, allows us to treat both electronic and phononic transport properties at a first-principles level. For the electronic part, we include an approximate self-energy correction, based on the DFT+Σ approach. This enables us to make a reliable prediction of all linear response transport coefficients entering the thermoelectric figure of merit ZT . Paracyclophane derivatives offer a great flexibility in tuning their chemical properties by attaching different functional groups. We show that, for the specific molecule, the functional groups mainly influence the thermopower, allowing to tune its sign and absolute value. We predict that the functionalization of the bare paracyclophane leads to a largely enhanced electronic contribution ZelT to the figure of merit. Nevertheless, the high phononic contribution to the thermal conductance strongly suppresses ZT . Our work demonstrates the importance to include the phonon thermal conductance for any realistic estimate of the ZT for off-resonant molecular transport junctions. In addition, it shows the possibility of a chemical tuning of the thermoelectric properties for a series of available molecules, leading to equally performing hole- and electron- conducting junctions based on the same molecular framework. I. INTRODUCTION a provide Nanostructured materials promising route towards high-performance thermoelectric energy conversion.1 -- 3 Nanostructuring does not only allow to improve the thermoelectric performance, but also enables novel applications like on-chip solutions for energy harvesting and refrigeration.2 Recently it became feasible to probe the thermoelectric properties and energy conversion in molecular junctions.4 -- 6 Within the different approaches to novel thermoelectric materials, molecular junctions are the ultimate level achievable regarding device miniaturization and provide a great flexibility to tailor thermoelectric properties, e.g. by chemical synthesis. Presently the applicability of organic thermoelectric materials is limited by the absence of high performance electron-type materials.7 The possibility to realize equally well performing hole- and electron-type (p- and n-type) molecular junctions and to enhance their efficiency8 -- 10 makes them very appealing for further scientific investigation. Irrespective of their practical use, they represent ideal systems to study thermoelectric transport at the atomic scale and to improve the understanding of the fundamental processes of nanoscale energy conversion. their On the theory side the combination of DFT-based elec- tronic structure calculations with nonequilibrium Green's function techniques allows an atomistic description of the electric and electronic thermal transport properties. With this approach important trends like the influence of the chemical and structural properties on the charge transport can be successfully captured.4,11 -- 13 Moreover the recently introduced DFT+Σ method, which improves the description of the level alignment at the metal- molecule interface, yields a quantitative agreement with experimental results at low computational costs.14 studies that Recent suggest computational antiresonances on molecular junctions15 -- 17 in the electronic transmission close to the Fermi energy can largely enhance the thermopower. The decrease of the electric conductance will be compensated by a simultaneous decrease of the electronic contribution to the thermal conductance. Consequently, this is reported to lead to an increase of the electronic contribution to the thermoelectric figure of merit ZelT . In such situations the overall thermoelectric figure of merit ZT is eventually limited by the phononic contribution to the thermal conductance (see Eqs. (18) and (19) for definitions of ZTel and ZT ). In Refs. 15 -- 17, however, this phonon thermal conductance is either neglected or just taken into account in an approximate manner. While it is speculated in Ref. 16 that the phonon thermal conductance might just play a minor role, it is suggested in Ref. 17 that it can have a significant contribution to the overall thermal conductance. Based on first-principles calculations of the relevant transport coefficients, we demonstrate in this work that it is indispensable for an accurate description of the energy conversion capabilities of molecular junctions to include also the phononic contribution. As an accurate total energy method, DFT provides the ideal tool to study phonon transport from first-principles and has already been applied successfully to various bulk and nanostructured systems.18 -- 20 However, for molecular junctions first-principles calculations of the phonon ther- mal conductance have up to now only been performed for 2 interactions.33 As basis set def2-SV(P)34 and the respec- tive Coulomb fitting basis35 are used. To ensure an accu- rate description of the vibrational properties, we use very tight convergence criteria. Total energies are converged to a precision of better than 10−9 a.u., while geometry optimizations are performed until the change of the max- imum norm of the Cartesian gradient is below 10−5 a.u.. In the construction of the contact geometries, as shown in Fig. 1b, we assume the Au atoms of the electrodes in (111) orientation to be fixed at their ideal fcc lattice posi- tions (lattice constant a0 = 4.08 Å). In contrast, the two Au apex atoms and the molecule are fully relaxed. While the precise metal-molecule-metal geometry will influence the thermoelectric properties, we focus here on the effect of the substituents. Due to the time-consuming DFPT calculations, we therefore restrict ourselves to one repre- sentative contact configuration. This contact geometry or extended central cluster (ECC)36 and its division into left (L), central (C) and right (R) regions is shown in Fig. 1b for the bare, unsubstituted [2,2]paracyclophane. B. Electronic transport In the Landauer-Büttiker picture the charge current I and the electronic thermal current Qel are given by37 -- 39 (1) (2) Here, τel(E) is the energy-dependent electron transmis- sion (e.g. from L through C into R, see Fig. 1b) and f (E, µ, T ) = {exp[(E − µ)/kBT ] + 1}−1 is the Fermi function, describing the occupation of the left (right) electron reservoir at the chemical potential µL (µR) and the temperature TL (TR). The energy of the electrons is measured relative to the average chemical potential µ = (µL + µR)/2. This is valid in the linear response regime, assuming ∆µ = µL − µR and ∆T = TL − TR to be infinitesimally small. In linear response the thermoelectric transport coeffi- cients, that is the electrical conductance G, thermopower S and electron thermal conductance κel, can be expressed as follows38,40,41 2e2 h G = K0, S = − K1 eT K0 , (cid:18) κel = 2 hT K2 − K 2 1 K0 (3) (4) (5) (cid:19) . 2e h 2 h dEτel(E) [f (E, µL, TL) I = −f (E, µR, TR)] , Qel = dE(E − µ)τel(E)× [f (E, µL, TL) − f (E, µR, TR)] . Figure 1. (a) Investigated paracyclophane derivatives with X = {H, NH2, OH, NO2, COCF3} in pseudo-para position. (b) Example of the used contact geometry, here with X = H. a few idealized systems.21 -- 24 Systematic studies for real- istic single-molecule junctions have yet to be provided. We present here a fully first-principles based study of the electron and phonon transport properties for a series of paracyclophane derivatives connected to gold electrodes. We focus on the question, how charge and phonon transport can be tuned chemically by intro- ducing different side-groups, which are either strongly electron-withdrawing or electron-donating. The chemical structure of the considered [2,2]paracyclophane deriva- tives is given in Fig. 1a, namely [2,2]paracyclophane, diamino-[2,2]paracyclophane, dihydroxy-[2,2]paracyclo- phane, dinitro-[2,2]paracyclophane and ditrifluoroacetyl- [2,2]paracyclophane, where we, for simplicity, just ana- lyze the pseudo-para isomers. For [2,2]paracyclophane the single-molecule electrical conductance was measured recently.25 II. FORMALISM AND METHODS A. Electronic, geometric and vibrational structure The electronic structure and the contact geometries are obtained at the DFT level, and the vibrational prop- erties in the framework of density functional perturba- tion theory (DFPT). Both procedures are implemented in the quantum chemistry software package TURBO- MOLE 6.4.26 -- 28 We employ the PBE functional29 -- 32 with empirical dispersion corrections to the total ground- state energy, accounting for long-range van der Waals  (cid:16)− ∂f (E) (cid:17) In the expressions e = e is the absolute value of the electron charge, h is the Planck constant, kB is the Boltz- mann constant, and T = (TL + TR)/2 is the average junction temperature. The coefficients in Eqs. (3)-(5) are defined as (E − µ)n, ∂E Kn = dEτel(E) (6) where the chemical potential µ ≈ EF is approximately given by the Fermi energy EF of the Au electrodes. The electronic transmission function is expressed within the Green's function formalism.36,37 The quasi- particle energies, entering the electronic Green's func- tions, are approximated by the Kohn-Sham eigenvalues, including a "self-energy correction". This self-energy cor- rection of the DFT+Σ method14 consists of two parts, one that accounts for the underestimation of the elec- tronic gap of the gas-phase molecule in local and semilo- cal approximations to DFT, the other one for long-range correlations in a junction that stabilize the charge on the molecule through image charges in the electrodes. De- tails regarding the self-energy correction can be found in Ref. 39. The electronic contact self-energies, on the other hand, are obtained as described in Ref. 36. To construct the electrode surface Green's function, we use 64 × 64 k-points in the transverse direction, which was found to be sufficient to obtain converged electronic transport co- efficients. C. Phononic transport Since phonons are chargeless, they can only carry a heat current. Similar to the electronic expression, this phononic heat current can be calculated in the Landauer- Büttiker picture42 -- 44 as dEEτph(E) [n(E, TL) − n(E, TR)] . (7) ∞ 0 Qph = 1 h ∞ 0 The corresponding thermal conductance due to the phonons is given in linear response by κph = 1 h dEEτph(E) ∂n(E, T ) ∂T , (8) where τph(E) is the phonon transmission and n(E, T ) = {exp(E/kBT )− 1}−1 is the Bose function, characterizing the phonon reservoirs in the left and right electrodes. Similar to the electronic transmission function, those of the phonons can also be computed within the Green's function formalism.37,43 -- 46 The system of inter- est here, namely a molecular junction or more gener- ally a nanoscale conductor, consists of an arbitrary, but finite-size scattering region (C), connected to two semi- infinite electrodes (L and R). For small displacements {Qξ} around the atomic equilibrium positions {R(0) ξ } the phononic Hamiltonian in the harmonic approximation can be written as (cid:88) p2 ξ + 1 2¯h2 H = 1 2 (cid:88) qξ qχKξχ. 3 (9) ξ ξχ Here we have introduced mass-weighted displacement op- erators qξ = (cid:112)Mξ Qξ and mass-scaled momentum op- erators pξ = Pξ/(cid:112)Mξ as conjugate variables, obey- trix Kξχ = ¯h2Hξχ/((cid:112)MξMχ), which is the second-order ing [qξ, pχ] = i¯hδξχ and [qξ, qχ] = [pξ, pχ] = 0, and ξ = (j, c) denotes a Cartesian component c = x, y, z of atom j at position (cid:126)Rj = (cid:126)R(0) j + (cid:126)Qj. The phononic system is characterized by its dynamical ma- mass-weighted derivative (or Hessian) of the total ground state energy with respect to the Cartesian atomic coor- dinates, Hξχ = ∂2 ξχE. These harmonic force constants are computed within DFPT. The local displacement basis allows us to partition the dynamical matrix into left (L), central (C), and right (R) parts  KLL KLC  . 0 K = KCL KCC KCR 0 KRC KRR (10) Under the assumption of vanishing coupling between L and R, KRL = KLR = 0, the transmission function for ballistic phonons can be written as43,46,47 τph(E) = Tr [Dr CC(E)ΛL(E)Da CC(E)ΛR(E)] , (11) where we have expressed the phonon transmission func- tion in terms of Green's functions solely defined on C. We use the general definition of the nonequilibrium phonon Green's function on the Keldysh contour Dξχ(τ, τ(cid:48)) = −i(cid:10)TC SC (qξ(τ )qχ(τ(cid:48)))(cid:11), (12) ´ where the displacement operators qξ(τ ) are in the inter- action picture, SC = exp[−i/¯h C dτ Hi(τ )] is the time evolution operator on the Keldysh contour, and TC is the corresponding time-ordering operator.45,47,48 To cal- culate the phononic heat current, we need the retarded and advanced Green's functions in energy space. They are obtained by a Fourier transform dtDr,a ξχ (t)eiEt/¯h (13) of the real-time functions, derived from Eq. (12) as45,47 ∞ −∞ Dr,a ξχ (E) = 1 ¯h (cid:104) Dr CC(E) = (cid:105) (E + iη)2 1 − KCC − Πr −Πr R(E) L(E) (14) CC(E)† and an infinitesimal quantity with Da η > 0. For non-interacting phonons in the electrodes CC(E) = Dr the corresponding contact self-energies can be calculated exactly as 4 Πr Y (E) = KCY dr Y Y (E)KY C. (15) Here, dr Y Y (E) is the surface Green's function of lead Y = {L, R}. Additionally, we have defined the spectral density of the electrodes as with Πa Y (E) = Πr Y − Πa Y ) ΛY = i(Πr Y (E)†. (16) Similar to our charge quantum transport approach36 we employ also for the phononic case a cluster-based pro- cedure. We calculate the dynamical matrix KECC of a large, but finite cluster consisting of the molecule of in- terest and parts of the electrodes. Using the partitioning of the ECC, given in Eq. (10) and depicted in Fig. 1b, we extract KCC, KLC and KRC from KECC. What re- mains to be calculated to determine τph(E) are the sur- face Green's functions of the electrodes dr Y Y (E). They are constructed for perfect semi-infinite crystals. The dy- namical matrix of the Au electrode is derived from those of a spherical cluster, see Fig. 2a. As in the crystal the atoms of the Au cluster are positioned on an fcc lattice with lattice constant a0 = 4.08 Å. We exploit that for large enough clusters the force constants from the cen- tral atom to its neighbors will be similar to the bulk. Following the procedure introduced in Ref. 36 for the charge transport, we construct the dynamical matrix of the crystalline electrode via these extracted bulk cou- plings. From this we compute surface and bulk Green's functions, dr Bulk(E), respectively, iteratively by means of the decimation technique.36,47,49 Y Y (E) and dr To check the quality of the extracted bulk parameters, we compare the bulk phonon density of states (DOS) against available experimental data. It can be directly obtained from the retarded bulk Green's function ρ(E) = − 2 π EIm (Tr [dr Bulk(E)]) . (17) The comparison of the DOS, calculated from parame- ters extracted from Au19, Au171 and Au333 clusters, with the experimental data of Ref. 50 is shown in Fig. 2b. For the smallest cluster Au19 the calculated DOS already resembles the general features of the experimental one. However, in the lower energy range from 0 to 15 meV the agreement is still poor. Increasing the cluster size to 171 atoms leads to a change especially in the low-energy range of the DOS. If we increase the cluster size further to 333 atoms, we observe only small modifications. Com- paring the DOS calculated for the parameters extracted from the Au333 cluster with the experiment,50 we see that the DFT results resemble the experimental DOS closely, but overestimate the vibrational energies by around 15%. By scaling the energy axis of the calculated DOS by a factor of 0.85, we basically obtain a perfect match with the experimental results. Regarding the scaling, we note Figure 2. (a) Shape of the three clusters, Au19, Au171 and Au333, used to extract the bulk parameters. (b) Bulk phonon density of states calculated with parameters extracted from the three cluster shown in (a) and experimental DOS from Ref. 50. The blue dashed curve is the DOS as obtained from the Au333 cluster scaled by 0.85 to fit the experimental vibra- tion energies. that in principle both the x- and y-axis of the DOS plot need to be scaled simultaneously with inverse factors to conserve the energy integral of the DOS. But since the experimental data is measured in arbitrary units, a scal- ing factor for the y axis is already used. We want to point out that, in order to obtain a reliable DOS as well as phonon transport properties, it is crucial to choose the broadening factor η sufficiently small. Oth- erwise the sharp features in the spectra are smeared out. Here, we used a broadening of η = 6.8 µeV, for which we obtained both a converged DOS and converged trans- mission spectra with respect to η. Due to the small η a large number of transverse k-points has to be chosen in the decimation procedure. We found the results to be converged for a k-point sampling of 512 × 512. After the DFPT step, this large number of k-points is the bottle neck of the transport calculations. However, the calcula- tion can be parallelized with respect to the k-points and after transforming the surface and bulk Green's functions back to real space, they can be conveniently stored on a hard drive and reused in later calculations. 5 Figure 3. Energy-dependent electronic transmission probabil- ity τel(E) relative to EF . D. Thermoelectric figure of merit The common measure for thermoelectric efficiency is given by the figure of merit ZT = GS2 κel + κph T, (18) which is determined from the thermoelectric transport coefficients in Eqs. (3)-(5) and (8) in the linear response regime. An upper bound for ZT in the limit of vanishing phonon thermal transport κph → 0 is given by the purely electronic contribution ZelT = S2G κel T = S2 L . (19) Here, we have introduced the Lorenz number L = κel/GT . With ZelT we can express the figure of merit in a slightly different form as ZT = ZelT 1 + κph/κel . (20) III. RESULTS AND DISCUSSION For the unsubstituted paracyclophane molecule (X = H), EF is located almost in the middle of the HOMO-LUMO gap (Fig. 3). The conductance is found to be GDFT+Σ = 0.004 G0 (without self-energy correc- tion GDFT = 0.07 G0), which compares reasonably well with the experimental data Gexp ≈ 0.01 G0, given in Ref. 25. Introducing the substituents leads to a shift of the molecular orbital energies relative to EF.51 For the electron-donating substituents NH2 and OH, the molecu- lar frontier orbital energies increase due to the increased Coulomb repulsion in the conjugated π-electron system of the benzene rings. This moves the HOMO resonance Figure 4. Temperature dependence of the thermoelectric transport coefficients, namely (a) electric conductance, (b) electron thermal conductance, and (c) thermopower. (d) Elec- tronic contribution ZelT to the figure of merit. around 1.1 eV closer to EF. The effect is slightly larger for NH2 than for OH (Fig. 3). The LUMO level is largely un- affected, which suggests that the two substituents mainly couple to the occupied π-orbitals. For the electron- withdrawing substituents NO2 and COCF3, we observe the opposite effect. The decreased Coulomb repulsion moves the LUMO resonance around 1.5 eV towards EF, with a slightly larger shift for COCF3 (Fig. 3). In Fig. 4a-c the temperature dependence of the thermo- electric transport coefficients is summarized. Before dis- cussing the individual features, it is worth stressing that due to the off-resonant transport and smooth transmis- sion functions around EF, the temperature dependence of the thermoelectric transport coefficients (Eqs. (3)-(5)) is dominated by the lowest order of the Sommerfeld ex- pansion of Eq. (6).53,54 Despite the substituent-induced movement of the or- bital energies, the transmission at EF remains roughly constant as compared to the unsubstituted molecule. Hence the influence of the substituents on the electric conductance G as well as the electron thermal conduc- tance κel ≈ LT G remains small. For NH2, G and κel are slightly increased. For the other three substituents a de- crease is observed (Fig. 4a,b). However, the slope of the transmission function at EF, and hence the thermopower S, changes substantially. For the unsubstituted molecule the transmission function is flat around EF, resulting in an almost vanishing thermopower (Fig. 4c). On the other hand, for all four substituents the absolute value of S is largely increased (Fig. 4c). Moreover the electron- 6 Figure 5. (a) Calculated and measured52 infrared spectra of [2,2]paracyclophane in the gas phase, showing the infrared-active molecular vibrations. (b) Phonon DOS of a Au (111) surface. (c) Phonon transmission τph(E) for [2,2]paracyclophane connected to Au electrodes. Green-shaded areas indicate the energy window, where the electrode phonon DOS is finite. Now we turn to the phononic contribution to the ther- mal conductance, which is often neglected in the cal- culations or treated in an approximate manner. Here, however, we determine also the phononic thermal con- ductance accurately from first-principles, as described in Sec. II C. Compared to the molecular vibration spec- trum (Fig. 5a), the electrode phonon density of states (DOS) is finite only in a narrow energy window (Fig. 5b). Even if the presence of the electrodes can rather strongly renormalize the molecular vibration spectrum and the character of the individual modes, simply speaking, just those molecular vibrations within the energy window de- fined by the electrode phonon DOS can contribute to the phonon transport (Fig. 5c). The temperature dependence of the phonon thermal conductance κph is displayed in Fig. 6. We obtain the expected ballistic behavior with a steep increase for low temperatures, as the phonon modes are getting occupied, while for higher temperatures κph approaches a constant value. Substituting the paracyclophane has a twofold effect. Firstly it influences the molecular vibration spec- trum directly, moving modes in and out of the energy window defined by the electrode phonon DOS. Secondly, it changes the binding geometry between the molecule and the Au electrode, modifying at the same time the cou- pling between the molecular vibrations and the electrode phonons. Both effects are, however, difficult to separate. Substituting with NO2 increases the phonon thermal con- ductance, while the other three substituents decrease it compared to the unsubstituted molecule. Eventually κph remains comparable for all five molecules (Fig. 6) and is, in the temperature range 10 K < T < 400 K, for all molecules at least one order of magnitude larger than the electronic thermal conductance (inset of Fig. 6). Since κph/κel > 10 for 10 K < T < 400 K, ZT is strongly suppressed according to Eq. (20) (Fig. 7). The largest ZT is obtained for NH2 substitution with ZT = 4 · 10−4 at room temperature (T = 300 K). The OH and COCF3 substituted paracyclophanes have al- most the same ZT in the whole temperature range with ZT = 0.9· 10−4 at T = 300 K, but the former one is hole- conducting while the latter one is electron-conducting. Figure 6. Temperature-dependent phonon thermal conduc- tance. Inset: Ratio κph/κel between phonon and electron thermal conductance. donating substituents NH2 and OH tune the slope of τel(E) at EF to be negative, giving a positive ther- mopower, characteristic for HOMO-dominated transport (p-type).5,55 -- 57 In contrast, the electron-withdrawing substituents give rise to LUMO-dominated transport (n- type), and hence the thermopower is negative (Fig. 4c). Ultimately, we can combine the thermoelectric trans- port coefficients to the electronic contribution to the fig- ure of merit ZelT , which is displayed in Fig. 4d. Due to the almost vanishing thermopower of the unsubsti- tuted molecule, ZelT remains very small. On the other hand, due to the quadratic dependence of ZelT on S, the enhanced thermopower for the substituted molecules increases ZelT largely. The NH2-substituted molecule shows the highest ZelT of around 4 · 10−3 at room tem- perature (T = 300 K). Overall ZelT remains rather small for the [2,2]paracyclophane derivatives studied here. The fact that it is possible to tune the system to be either p- or n-type by simply replacing the substituents is very encouraging, as this simple approach can be easily trans- ferred to other π-conducting systems. 7 the electronic contribution ZelT to the thermoelectric fig- ure of merit largely. However, for the studied off-resonant conduction, the phonon thermal conductance is, even at room temperature, at least one order of magnitude larger than the electronic one, leading to a significant suppres- sion of the overall ZT . Therefore we need to stress that it is necessary to include the phonon contribution to the thermal conductance for low-conducting molecular junc- tions in order to obtain accurate and reliable predictions of ZT . In our present study the largest ZT was found for the NH2-functionalized paracyclophane with ZT = 4 · 10−4 at T = 300 K. Even if the ZT remains rather small, we could demonstrate for a series of realistic molecules that it is possible to chemically adjust the transport properties and to enhance the thermoelectric figure of merit as well as to realize p- and n-conducting junctions using the same molecular framework. In the future more complex molecular structures and different functional groups could be used to enhance the thermoelectric figure of merit further. Beside the chem- ical tuning of mainly the electronic transport proper- ties, which was demonstrated in this work, limiting the phonon contribution to the thermal current is necessary to improve ZT . Nonlinear effects, leading to increased phonon-phonon scattering, could provide a convenient way to suppress the phonon transport. Such a discus- sion will be the topic of future work. ACKNOWLEDGMENTS MB and YA acknowledge fruitful discussions with H. Nakamura. This work was partly supported by a FY2012 (P12501) Postdoctoral Fellowship for Foreign Researchers from the Japan Society for Promotion of Sci- ence (JSPS) and by a JSPS KAKENHI, i.e. 'Grant-in-Aid for JSPS Fellows', grant no. 24·02501. YA is also thank- ful to another KAKENHI, i.e. Grant-in-Aid for Scientific Research on Innovation Areas "Molecular Architectonics: Orchestration of Single Molecules for Novel Functions" (#25110009). TJH would like to thank the Karlsruhe House of Young Scientists for financial support during his stay at the Nanosystem Research Institute. FP gratefully acknowledges funding from the Carl Zeiss foundation and the Junior Professorship Program of the Ministry for Science, Research, and Art of the state of Baden-Württemberg. Figure 7. Temperature dependence of the figure of merit ZT , including both electronic and phononic linear response trans- port coefficients. IV. CONCLUSIONS For a series of paracyclophane-based single-molecule junctions we computed the electrical and heat trans- port properties fully from first-principles. Combining DFT + Σ-based electronic structure and DFPT-derived harmonic force constants with the Green's function for- malism and the Landauer approach allowed us to make quantitative predictions of all linear response transport coefficients relevant for the thermoelectric figure of merit ZT . We showed that the transport properties can be tuned by functionalizing the paracyclophane molecule appropri- ately. The main influence of the functional groups was found to be on the junction thermopower, where both its absolute value and sign are affected. This is due to the fact that the Fermi energy lies right in the middle of the molecular HOMO-LUMO gap of the bare paracyclophane molecule, leading to a vanishing thermopower and con- sequently ZT . The electron-donating substituents NH2 and OH tune the thermopower to be positive, while the electron-withdrawing groups NO2 and COCF3 give rise to a negative value. The largest enhancement of the abso- lute value of the thermopower was found for NH2, fol- lowed by OH and COCF3. The latter two substituents yield a comparable absolute value of S, but a different sign. Naturally, the non-vanishing thermopower increases ∗ [email protected] 1 L. D. Hicks and M. S. Dresselhaus, Phys. Rev. B 47, 16631 (1993). 2 L. E. Bell, Science 321, 1457 (2008). 3 M. G. Kanatzidis, Chem. Mater. 22, 648 (2010). 4 P. Reddy, S.-Y. Jang, R. A. Segalman, and A. Majumdar, Science 315, 1568 (2007). 5 K. Baheti, J. A. Malen, P. Doak, P. Reddy, S.-Y. Jang, T. D. Tilley, A. Majumdar, and R. A. Segalman, Nano Lett. 8, 715 (2008). 6 W. Lee, K. Kim, W. Jeong, L. A. Zotti, F. Pauly, J. C. Cuevas, and P. Reddy, Nature 498, 209 (2013). 7 Q. Zhang, Y. Sun, W. Xu, and D. Zhu, Adv. Mater. 26, 6829 (2014). 30 J. C. Slater, Phys. Rev. 81, 385 (1951). 31 J. P. Perdew and Y. Wang, Phys. Rev. B 45, 13244 (1992). 32 J. P. Perdew, K. Burke, and M. Ernzerhof, Phys. Rev. Lett. 77, 3865 (1996). 33 S. Grimme, J. Comput. Chem. 27, 1787 (2006). 34 F. Weigend and R. Ahlrichs, Phys. Chem. Chem. Phys. 7, 8 S. K. Yee, J. A. Malen, A. Majumdar, and R. A. Segalman, 3297 (2005). 8 17 R. Stadler and T. Markussen, J. Chem. Phys. 135, 154109 381 (2008). 4, 5314 (2010). (2011). Nano Lett. 11, 4089 (2011). 9 C. Evangeli, K. Gillemot, E. Leary, M. T. Gonzalez, G. Rubio-Bollinger, C. J. Lambert, and N. Agrait, Nano Lett. 13, 2141 (2013). 10 J. R. Widawsky, W. Chen, H. Vázquez, T. Kim, R. Bres- low, M. S. Hybertsen, and L. Venkataraman, Nano Lett. 13, 2889 (2013). 11 L. Venkataraman, J. E. Klare, C. Nuckolls, M. S. Hybert- sen, and M. L. Steigerwald, Nature 442, 904 (2006). 12 F. Pauly, J. K. Viljas, and J. C. Cuevas, Phys. Rev. B 78, 035315 (2008). 13 A. Mishchenko, D. Vonlanthen, V. Meded, M. Bürkle, C. Li, I. V. Pobelov, A. Bagrets, J. K. Viljas, F. Pauly, F. Evers, M. Mayor, and T. Wandlowski, Nano Lett. 10, 156 (2009). 14 S. Y. Quek, L. Venkataraman, H. J. Choi, S. G. Louie, M. S. Hybertsen, and J. B. Neaton, Nano Lett. 7, 3477 (2007). 15 C. M. Finch, V. M. García-Suárez, and C. J. Lambert, Phys. Rev. B 79, 033405 (2009). 16 J. P. Bergfield, M. A. Solis, and C. A. Stafford, ACS Nano 18 A. Ward, D. A. Broido, D. A. Stewart, and G. Deinzer, Phys. Rev. B 80, 125203 (2009). 19 A. Calzolari, T. Jayasekera, K. W. Kim, and M. B. Nardelli, J. Phys.: Condens. Matter 24, 492204 (2012). 20 M. N. Luckyanova, J. Garg, K. Esfarjani, A. Jandl, M. T. Bulsara, A. J. Schmidt, A. J. Minnich, S. Chen, M. S. Dres- selhaus, Z. Ren, E. A. Fitzgerald, and G. Chen, Science 338, 936 (2012). 21 D. Nozaki, H. Sevinçli, W. Li, R. Gutiérrez, and G. Cu- niberti, Phys. Rev. B 81, 235406 (2010). 22 B. K. Nikolić, K. K. Saha, T. Markussen, and K. S. Thyge- sen, J. Comput. Electron. 11, 78 (2012). 23 H. Nakamura, T. Ohto, T. Ishida, and Y. Asai, J. Am. Chem. Soc. 135, 16545 (2013). 24 Y. Asai, J. Phys.: Condens. Matter 25, 155305 (2013). 25 S. T. Schneebeli, M. Kamenetska, Z. Cheng, R. Skouta, R. A. Friesner, L. Venkataraman, and R. Breslow, J. Am. Chem. Soc. 133, 2136 (2011). 26 TURBOMOLE 6.4, TURBOMOLE GmbH Karlsruhe, http://www.turbomole.com. TURBOMOLE is a develop- ment of University of Karlsruhe and Forschungszentrum Karlsruhe 1989-2007, TURBOMOLE GmbH since 2007. 27 P. Deglmann, F. Furche, and R. Ahlrichs, Chem. Phys. Lett. 362, 511 (2002). 28 P. Deglmann, K. May, F. Furche, and R. Ahlrichs, Chem. Phys. Lett. 384, 103 (2004). 29 P. A. M. Dirac, Proc. R. Soc. A 123, 714 (1929). 35 F. Weigend, Phys. Chem. Chem. Phys. 8, 1057 (2006). 36 F. Pauly, J. K. Viljas, U. Huniar, M. Häfner, S. Wohlthat, M. Bürkle, J. C. Cuevas, and G. Schön, New J. Phys. 10, 125019 (2008). 37 S. Datta, Electronic Transport in Mesoscopic Systems (Cambridge University Press, Cambridge, 1997). 38 U. Sivan and Y. Imry, Phys. Rev. B 33, 551 (1986). 39 L. A. Zotti, M. Bürkle, F. Pauly, W. Lee, K. Kim, and J. C. Cuevas, New W. Jeong, Y. Asai, P. Reddy, J. Phys. 16, 015004 (2014). 40 K. Esfarjani, M. Zebarjadi, and Y. Kawazoe, Phys. Rev. B 73, 085406 (2006). 41 K.-H. Müller, J. Chem. Phys. 129, 044708 (2008). 42 L. G. C. Rego and G. Kirczenow, Phys. Rev. Lett. 81, 232 (1998). 255503 (2006). 033408 (2006). 43 N. Mingo and L. Yang, Phys. Rev. B 68, 245406 (2003). 44 T. Yamamoto and K. Watanabe, Phys. Rev. Lett. 96, 45 J.-S. Wang, J. Wang, and N. Zeng, Phys. Rev. B 74, 46 Y. Asai, Phys. Rev. B 78, 045434 (2008). 47 J.-S. Wang, J. Wang, and J. T. Lü, Eur. Phys. J. B 62, 48 G. Stefanucci and R. van Leeuwen, Nonequilibrium Many- Body Theory of Quantum Systems: A Modern Introduction (Cambridge University Press, Cambridge, 2013). 49 F. Guinea, C. Tejedor, F. Flores, and E. Louis, Phys. Rev. B 28, 4397 (1983). B 8, 3493 (1973). 50 J. W. Lynn, H. G. Smith, and R. M. Nicklow, Phys. Rev. 51 M. Bürkle, L. A. Zotti, J. K. Viljas, D. Vonlanthen, A. Mishchenko, T. Wandlowski, M. Mayor, G. Schön, and F. Pauly, Phys. Rev. B 86, 115304 (2012). 52 "NIST Mass Spec Data Center, S. E. Stein, direc- tor, "Infrared Spectra" in NIST Chemistry WebBook, NIST Standard Reference Database Number 69, Eds. P. J. Linstrom and W. G. Mallard, National Insti- tute of Standards and Technology, Gaithersburg MD, 20899, http://webbook.nist.gov," (retrieved November 10, 2014)). 53 N. W. Ashcroft and N. D. Mermin, Solid State Physics (Hartcourt, Orlando, 1976). 54 G. Gómez-Silva, O. Ávalos-Ovando, M. L. de Guevara, and P. Orellana, J. Appl. Phys. 111, 053704 (2012). 55 M. Paulsson and S. Datta, Phys. Rev. B 67, 241403(R) (2003). 56 A. Tan, J. Balachandran, S. Sadat, V. Gavini, B. D. Duni- etz, S.-Y. Jang, and P. Reddy, J. Am. Chem. Soc. 133, 8838 (2011). 57 J. Balachandran, P. Reddy, B. D. Dunietz, and V. Gavini, J. Phys. Chem. Lett. 4, 3825 (2013). 

1603.03751

1

1603

Conductance Quantization at zero magnetic field in InSb nanowires

Ballistic electron transport is a key requirement for existence of a topological phase transition in proximitized InSb nanowires. However, measurements of quantized conductance as direct evidence of ballistic transport have so far been obscured due to the increased chance of backscattering in one dimensional nanowires. We show that by improving the nanowire-metal interface as well as the dielectric environment we can consistently achieve conductance quantization at zero magnetic field. Additionally, studying the sub-band evolution in a rotating magnetic field reveals an orbital degeneracy between the second and third sub-bands for perpendicular fields above 1T.

cond-mat.mes-hall

cond-mat

Conductance Quantization at zero magnetic field in InSb nanowires Jakob Kammhuber,† Maja C. Cassidy,† Hao Zhang,† Önder Gül,† Fei Pei,† Michiel W. A. de Moor,† Bas Nijholt,† Kenji Watanabe,‡ Takashi Taniguchi,‡ Diana Car,¶ Sébastien R. Plissard,†,¶,§ Erik P. A. M. Bakkers,†,¶ and Leo P. Kouwenhoven∗,† QuTech and Kavli Institute of Nanoscience, Delft University of Technology, 2628 CJ Delft, The Netherlands, Advanced Materials Laboratory, National Institute for Materials Science, 1-1 Namiki, Tsukuba, 305-0044, Japan, Department of Applied Physics, Eindhoven University of Technology, 5600 MB Eindhoven, The Netherlands, and CNRS-Laboratoire d'Analyse et d'Architecture des Systemes (LAAS), Université de Toulouse, 7 avenue du Colonel Roche, F-31400 Toulouse, France E-mail: [email protected] KEYWORDS: Quantum point contact, conductance quantization, nanowire, InSb, subband, orbital effects ∗To whom correspondence should be addressed †QuTech and Kavli Institute of Nanoscience, Delft University of Technology, 2628 CJ Delft, The Netherlands ‡Advanced Materials Laboratory, National Institute for Materials Science, 1-1 Namiki, Tsukuba, 305-0044, Japan ¶Department of Applied Physics, Eindhoven University of Technology, 5600 MB Eindhoven, The Netherlands §Current address: CNRS-Laboratoire d'Analyse et d'Architecture des Systemes (LAAS), Université de Toulouse, 7 avenue du Colonel Roche, F-31400 Toulouse, France 6 1 0 2 r a M 1 1 ] l l a h - s e m . t a m - d n o c [ 1 v 1 5 7 3 0 . 3 0 6 1 : v i X r a 1 Abstract Ballistic electron transport is a key requirement for existence of a topological phase tran- sition in proximitized InSb nanowires. However, measurements of quantized conductance as direct evidence of ballistic transport have so far been obscured due to the increased chance of backscattering in one dimensional nanowires. We show that by improving the nanowire- metal interface as well as the dielectric environment we can consistently achieve conductance quantization at zero magnetic field. Additionally, studying the sub-band evolution in a rotat- ing magnetic field reveals an orbital degeneracy between the second and third sub-bands for perpendicular fields above 1T. Semiconducting nanowires made from InAs and InSb are prime candidates for the investiga- tion of novel phenomena in electronic devices. The intrinsic strong spin-orbit interaction (SOI) and large g-factor combined with flexible fabrication has resulted in these materials being inves- tigated for applications in quantum computing,1, 2, 3 spintronics4, 5, 6 and Cooper pair splitters.7, 8 More recently, these nanowires have been investigated as solid-state hosts for Majorana zero modes (MZMs).9, 10, 11, 12 By bringing a one dimensional (1D) nanowire with strong SOI into close con- tact with a superconductor under an external magnetic field, a region with inverted band structure emerges, creating MZMs at its ends. Together with strong SOI and induced superconductivity, a key requirement for MZMs is quasi-ballistic electron transport along the length of the proximi- tized region in the nanowire, with a controlled odd number of occupied modes.13 In the absence of scattering, the motion of 1D confined electrons will be restricted to discrete energy bands resulting in quantized conductance plateaus.14, 15 Measurements of quantized conductance in the nanowires therefore provide direct evidence for controlled mode occupation, as well as ballistic transport in these nanowires. Although now routine in gate defined quantum point contacts (QPC) in two-dimensional elec- tron gases (2DEG),14, 15, 16, 17, 18 quantized conductance in one dimensional semiconductor nanowires is more difficult to achieve. In a 1D nanowire, scattering events along the electrons path to and through the constriction between the source and drain contacts have an increased probability of 2 reflection, obscuring the observation of quantized conductance.19 These scattering events may be due to impurities and imperfections in the crystal lattice, or due to surface states that create inho- mogeneities in the local electrostatic environment.20 A Schottky barrier between the nanowire and metallic contacts will result in additional backscattering events, further smearing out the quantized conductance plateaus. To date, quantized conductance in InSb nanowires has only been observed at high magnetic fields (> 4T), where electron backscattering is strongly suppressed.19 No quan- tization has been observed in InSb for magnetic fields below 1T, where the topological transition is expected to take place.9 Here we demonstrate conductance quantization in InSb nanowires at zero magnetic field. We have developed a robust fabrication recipe for observing quantized conductance by optimizing both the metal-nanowire contact interface and dielectric environment through the use of hexagonal boron nitride (hBN) as a gate dielectric. We study the evolution of the quantized conductance plateaus with both source-drain bias as well as magnetic field, and extract values for the Landé g-factor of the first three sub-bands in the nanowire. Additionally, we observe an orbital energy degeneracy of the second and third sub-bands at finite magnetic fields applied perpendicular to the nanowire. Figure 1a) shows a cross-sectional view of our devices. They consist of an intrinsic Si-substrate with local metallic gates made of Ti/Au (5/10nm), on top of which a sheet of hexagonal boron ni- tride (hBN) is mechanically transferred as the dielectric. The chemical stability, atomic flatness, and high breakdown voltage,21 together with the well established dry transfer mechanism22 makes hBN an ideal dielectric for our nanowire devices. InSb nanowires grown by metal-organic vapor phase epitaxy23, 24 (1 - 3µm long and 70 - 90nm diameter) are transferred deterministically with a micro-manipulator25 onto the hBN dielectric. Electrical contacts to the nanowire (evaporated Cr/Au (10/100nm), 150 - 400nm spacing) are defined by electron beam lithography. Before con- tact deposition, the surface oxide of the nanowires is removed using sulfur passivation26 followed by a short in situ He-ion mill. Residual sulfur from the passivation step also induces surface doping, which aids contact transparency. Further details of the fabrication are included in the supporting 3 Figure 1: (a) Cross-sectional schematic and (b) false color SEM image of a typical device. An InSb nanowire (blue) contacted by Cr/Au (yellow) is deposited on Ti/Au metal gates (grey) covered with hexagonal boron nitride (green) as insulating dielectric. (c) Pinch-off traces of four different devices each showing quantized conductance plateaus at high bias voltage (Vbias = 10mV). (d) Schematic diagram of the first five sub-bands in a nanowire. At zero magnetic field, each spin- degenerate sub-band below the Fermi level contributes a conductance of G0 = 2e2/h. Due to the rotational symmetry of the nanowires E2,E3 and E4,E5 are almost degenerate. (e) Sketch of the expected conductance steps as function of Vgate at high bias voltage showing suppression of the second and fourth plateaus due to orbital sub-band degeneracy. 4 G (2e2/h)Vgate135(e)E4,E5E2,E3kE(d)μDμS(c)135G (2e2/h)Vgate (V)0-2Vbias = 10mV500nmBZA+-(b)VbiasVgate(a)BoronnitrideTi/AuSiO2InSbCr/AuSiinformation. A top view scanning electron microscope image of a finished device is shown in Fig 1b). The samples are mounted in a dilution refrigerator with a base temperature of 15mK and measured using standard lock-in techniques at 73 Hz with an excitation VRMS = 70 µV. Voltage is applied to the outer contact and current measured through the grounded central contact, while the third, unused contact is left floating. We first characterize each device by sweeping the voltage on the underlying gate Vgate at fixed bias voltage Vbias = 10mV across the sample. Conductance is obtained directly from the measured current G = I/Vbias and an appropriate series resistance is subtracted in each case (see supporting information). Figure 1c) plots the conductance of the nanowire as function of gate voltage for four different devices fabricated on the same chip. Devices with both fine gates as well as wide back gates have been measured. We find that while fine gates allow more flexible gating, devices with wide back gates showed more pronounced conductance plateaus even after extensive tuning of the fine gates. Data from additional devices all fabricated on the same chip is included in the supporting information. As seen in Fig 1c) all devices show well defined plateaus at G0 and 3G0 but the plateaus at 2G0 and 4G0 appear smaller or even completely absent. Unlike QPC's formed in 2DEGs, nanowires possess rotational symmetry. This symmetry can give rise to additional orbital degeneracies in the energies for the 2nd and 3rd as well as the 4th and 5th sub-band (Fig 1d).27, 28 In conductance measurements at finite bias, sub-bands that are close in energy or degenerate will be populated at similar values in gate voltage giving a double step of 4 e2 suppressed plateaus at 2 and 4G0 (Fig 1e).29 h , which explains the h instead of 2 e2 To investigate this phenomenon in more detail, we measure the differential conductance G = dI/dVbias as function of gate voltage and bias voltage for one of these devices (corresponding to the green trace in Fig 1c). This data is shown in Fig 2a) as a color plot, with a line cut along zero bias voltage added in the bottom panel. At zero bias voltage an extended plateau is visible at 1G0, together with an additional small plateau at 2G0 which was not visible in the linear conductance data of Figure 1c). The existence of this small 2G0 plateau indicates that the device has a small, 5 but finite energy splitting between the second and third sub-band which was not resolved at high bias. At finite bias voltage the conductance will only be quantized in integer values of G0 if both µsource and µdrain occupy the same sub-band. This creates diamond shaped regions of constant conductance indicated by black dotted lines in Fig 2a). At the tip of the diamond the two dotted lines cross when Vbias is equal to the sub-band energy spacing ∆Esubband. From this we extract ∆Esubband and the lever-arm η of the bottom gate via η Vgate = ∆Esubband.30, A finite magnetic field breaks time reversal symmetry, lifting spin degeneracy and splitting the individual spin sub- bands En,↑/↓ by the Zeeman energy EZeeman = gµBB. Here µB denotes the Bohr magneton and g the Landé g-factor. Experimentally this splitting manifests as the appearance of additional half 2 · 2e2/h. At B = 4T we clearly observe this for the first sub-band as shown in Fig integer steps N 2b) where an additional plateau emerges at 0.5G0. Similarly, the second sub-band should also split into two plateaus at 1.5 and 2G0. However only the 2G0 plateau is visible, suggesting that the orbital degeneracy between E2,↑ and E3,↑ remains at finite magnetic field. Figure 2: Color-plot of the differential conductance G = dI/dVbias as function of Vbias and Vgate at (a) B = 0T and (b) BZ = 4T. A line cut along zero bias voltage is shown in the bottom panel. Plateaus appear as diamonds and are indicated by black dotted lines. The full evolution in magnetic field of the conductance and transconductance is shown in Fig 3a,b) and individual line traces of the conductance taken in steps of 1T are presented in Fig 3c). While the plateau at 1G0 remains very flat up to high magnetic fields, the second plateau at 2G0 increases in height for magnetic fields larger than 400mT. Around 1T two new plateaus emerge 6 (b)Vgate (V)Vbias (mV)(2e2/h)010-101300.5B = 4T120.5(2e2/h)(a)010-10130-0.50.5B = 0T12Vgate (V)Vbias (mV)12G (2e2/h)Figure 3: (a) Differential conductance G = dI/dVbias and (b) transconductance dG/dVgate as func- tion of magnetic field along BZ and Vgate taken at Vbias = 0mV. The level spacings plotted in d,e) are marked by arrows of corresponding color. Red dashed lines indicating the sub-band spacing in a,b) are drawn as guide to the eye. (c) Linecuts of a) in steps of 1T and offset by 200mV for clar- ity. (d) Energy level spacings of E1↓ − E1↑ (yellow), E2,3↑ − E1↓ (green) and E2,3↑ − E1↑ (orange) extracted from the 0.5 and 1G0 plateau in (a). A linear fit to E1↓ − E1↑ fixed at the origin gives the g-factor of the first sub-band g1 = 39± 1. (e) Energy spacing of E2,3↓ − E2,3↑ extracted from the 2G0 plateau with g2,3 = 40± 1. 7 (e)E (meV)BZ (T)02451015E2,3↓-E2,3↑ BZ (T)024E (meV)51015E1↓-E1↑E2,3↑-E1↓E2,3↑-E1↑(d)00.511.5Vgate (V)123G (2e2/h)BZ=0TBZ=5T(c)00.5024Vgate (V)BZ (T)01530dG/dVgate ((2e2/h)/V)(b)0.512024BZ (T)00.5Vgate (V)123G (2e2/h)(a)with similar slope at 0.5 and 2G0. These correspond to the lower energy spin sub-bands E1↑ and E2,3↑. Here we can clearly see experimentally that the non-degenerate orbital state at zero field transforms into a degenerate orbital state at finite field and that E2,3↑ remain degenerate over a magnetic field range of several Tesla. From the individual gate traces we convert the plateau width to energy by using the lever arm η extracted from Figure 2. This way we can directly extract the sub-band spacing E2↑ − E1↑ and the individual g-factors g1, g2,3 through a linear fit fixed at the origin to E1↓ − E1↑ and E2,3↓ − E2,3↑. We find g1 = 39± 1 g2,3 = 40± 1 and a constant sub-band spacing E2↑ − E1↑ ≈ 16meV . Figure 4: (a) Probability density of the first 5 sub-bands of a cylindrical nanowire. (b) Orientation of the nanowire with respect to the magnetic field axes. (c,d) Numerical simulations of the sub- band dispersion of a InSb nanowire in perpendicular (c) and parallel (d) magnetic field. Orbital degeneracy of sub-bands has previously been observed in metallic point contacts29 and recently also in passivated narrow InAs nanowires with highly symmetric conducting channels.27 However, the magnetoconductance of InSb nanowires may deviate significantly from the results found in InAs nanowires. In InAs, Fermi level pinning leads to conduction close to the nanowire surface31, 32 which strongly influences the sub-band dispersion in magnetic field.33, 34 InSb has no surface accumulation35 and the electron wave-function will be more strongly confined in the center of the nanowire. For cylindrical nanowires individual sub-band wave functions are given 8 (d)B(T)EB║E3E2E1E4012B┴B(T)E(c)E3E2E1E4012BZBYBX(b)(a)E1E2/3E4/5Figure 5: Transconductance dG/dVgate and differential conductance G for three different direc- tions of the magnetic field all taken at Vbias = 0mV. Green dashed lines indicating the sub-band spacing in a,c,e) are drawn as guide to the eye and red numbers label the height of the conductance plateaus. BZ is increased from 0− 2T and BX,Y from 0− 1T (a,b) Magnetic field aligned along BZ. (c,d) Magnetic field aligned along BX. (e,f) Magnetic field aligned along BY. 9 00.5Vgate (V)01G (2e2/h)23(f)BX(T)00.25Vgate (V)01-0.25(e)0.51200.5Vgate (V)01G (2e2/h)23(d)BY (T)00.25Vgate (V)01-0.25(c)20.51.51(b)Vgate (V)00.501G (2e2/h)23BZ (T)00.25Vgate (V)02-0.25(a)10.52015dG/dVgate ((2e2/h)/V)by Bessel functions with different orbital angular momentum along the wire (Fig 4a), and numeri- cal simulations of wires with a hexagonal cross-section show qualitatively similar results.28, 36 An additional magnetic field will add Zeeman splitting, but also causes orbital effects which can sub- stantially change the sub-band dispersion depending on the orientation of the field with respect to the nanowires axis.37 Numerical simulations of nanowires in a magnetic field show that orbital ef- fects strongly depend on the magnetic field orientation and can dominate the sub-band dispersion, leading to a decrease of the energy splitting between E2 and E3.37 Furthermore these simulations also show that the orbital effects can strongly influence the phase diagram of MZMs.37 Using the model of ref. 37 with the parameters of our device (wire radius: 35nm; g-factor: 40) we simulate this change in the sub-band dispersion for a magnetic field perpendicular (Fig 4c) and parallel (Fig 4d) to the nanowire. A perpendicular magnetic field causes E2,↑ and E3,↑ to shift higher and closer in energy, while a parallel magnetic field increases the energy splitting of the higher sub-bands E2 and E3, due to their different orbital angular momentum. Experimentally we test this by rotating the direction of the magnetic field, as shown in Fig 5. When aligning the magnetic field along BZ (almost perpendicular to the nanowire, Fig 5a,b), a small splitting appears at the beginning of the first plateau for fields above BZ = 0.6T, marking the onset of the 0.5-plateau. In contrast, in the second plateau the splitting only starts above 1T and the line-cuts (Fig 5b) show that the new plateau emerges at 2G0. Similarly, for a magnetic field along BX (Fig 5c,d), a new plateau emerges around BX = 0.6T in the first step but not in the second. However, for the magnetic field aligned along BY (mostly parallel to the nanowire) shown in Fig 5e,f) we do see a clear difference. Now two new plateaus emerge almost simultaneously around BY ≈ 0.75T, with the second plateau at 1.5 and not at 2G0, in agreement with the expected behavior due to orbital effects. In conclusion we achieved substantial improvements in electrical transport measurements of InSb nanowires by using a high quality hBN dielectric and clearly demonstrated conductance quan- tization at zero magnetic field, as well as degenerate sub-bands at magnetic field above 1T. In the future these, improvements will allow the more detailed investigation of features in the 1st plateau, 10 such as signatures of a helical gap,38, 39 or the presence of a 0.7 anomaly.40, 41, 42 The large SOI in our InSb nanowire strongly influences the electron dispersion relation and the tunability with magnetic field could add new insight into the underlying physics.43 We did not see any clear fea- tures related to the 0.7-anomaly in our devices. However, the 0.7 state becomes more pronounced at higher temperatures.40 A more detailed study of the temperature dependence of conductance quantization may reveal more information about the existence of this intriguing state in nanowire QPCs. Acknowledgement The authors thank M. Wimmer, P. Kim and A. Akhmerov for helpful discussions, S. Goswami for help with the hBN transfer, and D. van Woerkom for help with nanowire deposition. This work has been supported by funding from the Marie Curie ITN S3Nano, the ERC starting grant STATOPINS 638760, NWO/FOM and Microsoft Corporation Station Q. Supporting Information Available The supporting information contain a detailed fabrication recipe, a discussion of the subtracted series resistance, additional data of the main device as well as data of QPC devices fabricated with a SiO2 dielectric. This material is available free of charge via the Internet at http://pubs.acs.org/. References (1) Nadj-Perge, S.; Frolov, S. M.; Bakkers, E. P. A. M.; Kouwenhoven, L. P. Nature 2010, 468, 1084 -- 1087. (2) Van den Berg, J. W. G.; Nadj-Perge, S.; Pribiag, V. S.; Plissard, S. R.; Bakkers, E. P. A. M.; Frolov, S. M.; Kouwenhoven, L. P. Phys. Rev. Lett. 2013, 110, 066806. 11 (3) Petersson, K. D.; McFaul, L. W.; Schroer, M. D.; Jung, M.; Taylor, J. M.; Houck, A. A.; Petta, J. R. Nature 2012, 490, 380 -- 383. (4) Liang, D.; Gao, X. P. Nano Lett. 2012, 12, 3263 -- 3267. (5) Rossella, F.; Bertoni, A.; Ercolani, D.; Rontani, M.; Sorba, L.; Beltram, F.; Roddaro, S. Nat. Nanotechnol. 2014, 9, 997 -- 1001. (6) Žuti´c, I.; Fabian, J.; Das Sarma, S. Rev. Mod. Phys. 2004, 76, 323 -- 410. (7) Hofstetter, L.; Csonka, S.; Nygård, J.; Schönenberger, C. Nature 2009, 461, 960 -- 963. (8) Das, A.; Ronen, Y.; Heiblum, M.; Mahalu, D.; Kretinin, A. V.; Shtrikman, H. Nat. Commun. 2012, 3, 1165. (9) Mourik, V.; Zuo, K.; Frolov, S. M.; Plissard, S. R.; Bakkers, E. P. A. M.; Kouwenhoven, L. P. Science 2012, 336, 1003 -- 1007. (10) Deng, M. T.; Yu, C. L.; Huang, G. Y.; Larsson, M.; Caroff, P.; Xu, H. Q. Nano Lett. (11) Churchill, H. O. H.; Fatemi, V.; Grove-Rasmussen, K.; Deng, M. T.; Caroff, P.; Xu, H. Q.; Marcus, C. M. Phys. Rev. B 2013, 87, 241401. (12) Deng, M.; Yu, C.; Huang, G.; Larsson, M.; Caroff, P.; Xu, H. Sci. Rep. 2014, 4. (13) Lutchyn, R. M.; Stanescu, T. D.; Das Sarma, S. Phys. Rev. Lett. 2011, 106, 127001. (14) van Wees, B. J.; van Houten, H.; Beenakker, C. W. J.; Williamson, J. G.; Kouwenhoven, L. P.; van der Marel, D.; Foxon, C. T. Phys. Rev. Lett. 1988, 60, 848 -- 850. (15) Wharam, D. A.; Thornton, T. J.; Newbury, R.; Pepper, M.; Ahmed, H.; Frost, J. E. F.; Hasko, D. G.; Peacock, D. C.; Ritchie, D. A.; Jones, G. A. C. J. Phys. C 1988, 21, L209. (16) Chou, H. T.; Lüscher, S.; Goldhaber-Gordon, D.; Manfra, M. J.; Sergent, A. M.; West, K. W.; Molnar, R. J. Appl. Phys. Lett. 2005, 86. 12 (17) Többen, D.; Wharam, D. A.; Abstreiter, G.; Kotthaus, J. P.; Schaffler, F. Semicond. Sci. Technol. 1995, 10, 711. (18) Koester, S. J.; Brar, B.; Bolognesi, C. R.; Caine, E. J.; Patlach, A.; Hu, E. L.; Kroemer, H.; Rooks, M. J. Phys. Rev. B 1996, 53, 13063 -- 13073. (19) van Weperen, I.; Plissard, S. R.; Bakkers, E. P. A. M.; Frolov, S. M.; Kouwenhoven, L. P. Nano Lett. 2012, 13, 387 -- 391. (20) Gül, Ö.; van Woerkom, D. J.; van Weperen, I.; Car, D.; Plissard, S. R.; Bakkers, E. P. A. M.; Kouwenhoven, L. P. Nanotechnology 2015, 26, 215202. (21) Dean, C.; Young, A.; Meric, I.; Lee, C.; Wang, L.; Sorgenfrei, S.; Watanabe, K.; Taniguchi, T.; Kim, P.; Shepard, K.; Hone, J. Nat. Nanotechnol. 2010, 5, 722 -- 726. (22) Castellanos-Gomez, A.; Buscema, M.; Molenaar, R.; Singh, V.; Janssen, L.; van der Zant, H. S. J.; Steele, G. A. 2D Mater. 2014, 1, 011002. (23) Plissard, S. R.; Slapak, D. R.; Verheijen, M. A.; Hocevar, M.; Immink, G. W. G.; van Weperen, I.; Nadj-Perge, S.; Frolov, S. M.; Kouwenhoven, L. P.; Bakkers, E. P. A. M. Nano Lett. 2012, 12, 1794 -- 1798. (24) Car, D.; Wang, J.; Verheijen, M. A.; Bakkers, E. P. A. M.; Plissard, S. R. Adv. Mater. 2014, 26, 4875 -- 4879. (25) Flöhr, K.; Liebmann, M.; Sladek, K.; Günel, H. Y.; Frielinghaus, R.; Haas, F.; Meyer, C.; Hardtdegen, H.; Schäpers, T.; Grützmacher, D.; Morgenstern, M. Rev. Sci. Instrum. 2011, 82, 113705. (26) Suyatin, D. B.; Thelander, C.; Björk, M. T.; Maximov, I.; Samuelson, L. Nanotechnology 2007, 18, 105307. (27) Ford, A.; Kumar, S. B.; Kapadia, R.; Guo, J.; Javey, A. Nano Lett. 2012, 13 (28) van Weperen, I. Ph.D. thesis, TU Delft, 2014. (29) Krans, J. M.; Van Ruitenbeek, J. M.; Fisun, V. V.; Yanson, I. K.; De Jongh, L. J. Nature 1995, 375, 767 -- 769. (30) Kouwenhoven, L. P.; van Wees, B. J.; Harmans, C. J. P. M.; Williamson, J. G.; van Houten, H.; Beenakker, C. W. J.; Foxon, C. T.; Harris, J. J. Phys. Rev. B 1989, 39, 8040 -- 8043. (31) Scheffler, M.; Nadj-Perge, S.; Kouwenhoven, L. P.; Borgström, M. T.; Bakkers, E. P. A. M. J. Appl. Phys. 2009, 106. (32) Halpern, E.; Elias, G.; Kretinin, A. V.; Shtrikman, H.; Rosenwaks, Y. Appl. Phys. Lett. 2012, 100. (33) Holloway, G. W.; Shiri, D.; Haapamaki, C. M.; Willick, K.; Watson, G.; LaPierre, R. R.; Baugh, J. Phys. Rev. B 2015, 91, 045422. (34) Tserkovnyak, Y.; Halperin, B. I. Phys. Rev. B 2006, 74, 245327. (35) King, P. D. C.; Veal, T. D.; Lowe, M. J.; McConville, C. F. J. Appl. Phys. 2008, 104. (36) Vuik, A.; Eeltink, D.; Akhmerov, A. R.; Wimmer, M. arXiv:1511.08044 2015, (37) Nijholt, B.; Akhmerov, A. R. arXiv:1509.02675 2015, (38) Streda, P.; Šeba, P. Phys. Rev. Lett. 2003, 90, 256601. (39) Pershin, Y. V.; Nesteroff, J. A.; Privman, V. Phys. Rev. B 2004, 69, 121306. (40) Thomas, K. J.; Nicholls, J. T.; Simmons, M. Y.; Pepper, M.; Mace, D. R.; Ritchie, D. A. Phys. Rev. Lett. 1996, 77, 135 -- 138. (41) Bauer, F.; Heyder, J.; Schubert, E.; Borowsky, D.; Taubert, D.; Bruognolo, B.; Schuh, D.; Wegscheider, W.; von Delft, J.; Ludwig, S. Nature 2013, 501, 73 -- 78. 14 (42) Iqbal, M. J.; Levy, R.; Koop, E. J.; Dekker, J. B.; De Jong, J. P.; van der Velde, J. H. M.; Reuter, D.; Wieck, A. D.; Aguado, R.; Meir, Y.; van der Wal, C. H. Nature 2013, 501, 79 -- 83. (43) Goulko, O.; Bauer, F.; Heyder, J.; von Delft, J. Phys. Rev. Lett. 2014, 113, 266402. 15 Graphical TOC Entry 16 500nm010-10130-0.50.5B = 0T12Vgate (V)(2e2/h)Vbias (mV)BoronnitrideTi/AuSiO2InSbCr/AuSi

1112.1980

1

1112

Circuit Modeling of Tunneling Real-Space Transfer Transistors: Toward Terahertz Frequency Operation

High frequency operation of tunneling real-space transfer transistor (TRSTT) in the negative differential resistance (NDR) regime is assessed by calculating the device common source unity current gain frequency (fT) range with a small signal equivalent circuit model including tunneling. Our circuit model is based on an In0.2Ga0.8As and delta-doped GaAs dual channel structure with various gate lengths. The calculated TRSTT fT agrees very well with experimental data, limiting factor being the resistance of the delta-doped GaAs layer. By optimizing the gate dimensions and channel materials, we find fT in the NDR region approaches terahertz range, which anticipates potential use of TRSTT as terahertz sources.

cond-mat.mes-hall

cond-mat

> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 1 Circuit Modeling of Tunneling Real-Space Transfer Transistors: Toward Terahertz Frequency Operation Wen Huang, Student Member, IEEE, Xin Yu, Student Member, IEEE, Shi-Lin Zhang, Lu-Hong Mao, and Jean-Pierre Leburton, Fellow, IEEE Abstract—High frequency operation of tunneling real-space transfer transistor (TRSTT) in the negative differential resistance (NDR) regime is assessed by calculating the device common source unity current gain frequency (fT) range with a small signal equivalent circuit model including tunneling. Our circuit model is based on an In0 .2Ga0.8As and δδδδ-doped GaAs dual channel structure with various gate lengths. The calculated TRSTT fT agrees very well with experimental data, limiting factor being the resistance of the δδδδ-doped GaAs layer. By optimizing the gate dimensions and channel materials, we find fT in the NDR region approaches terahertz range, which anticipates potential use of TRSTT as terahertz sources. Index Terms—Tunneling real-space transistor transfer (TRSTT), negative differential resistance (NDR), small signal equivalent circuit model, unity current gain frequency I. INTRODUCTION T UNNELING real space transfer (TRST) is a quantum mechanical effect that arises between two channels of different mobilities in field effect transistors to achieve NDR controlled by the gate bias [1]. The effect was first demonstrated in pseudo-morphic AlGaAs/InGaAs MODFET [2], and investigated in several TRSTT structures in the last decades [3]–[5]. Recently, very clean TRST-induced NDRs with modulated peak to valley (P/V) ratios up to 4 at room temperature were demonstrated in pseudo-morphic GaAs Manuscript received December 7, 2011. This work is supported in part by the Beckman Institute for Advanced Science and Technology at the University of Illinois at Urbana-Champaign, the School of Electronic Engineering at the University of Electronic Science and Technology of China, and Tianjin University. Wen Huang is with Beckman Institute for Advanced Science and Technology a t the University of Illinois at Urbana-Champaign, Urbana, IL, 61801, USA and the School of Elec tronic Engineering at the University of Electronic Science and Technology of China, Chengdu, Sichuan, 611731, China (e-mail:[email protected]). Xin Yu is with the Department of Electrical and Computer Engineering at University of Illinois at Urbana-Champaign, Urbana, IL 61801 USA. (e-mail: [email protected]). Lu-Hong Mao and Shi-Lin Zhang are with the Department of Electronics and Information Engineering, Tianjin University, Tianjin 300072, China. J. P. Leburton is with the Department of Physics, Department of Electrical and Computer Engineering, and the Beckman Institute for Advanced Science and Technology at the University of Illinois at Urbana-Champaign, Urbana, IL, 61801, USA (phone: 217-333-6813; e-mail: [email protected]). /InGaAs MODFET structures (fig.1 inset) [6]. The devices however were relatively long and wide with large capacitance, so that fT was below 10 GHz. Therefore considerable room for improvement is expected with frequency operation far above 100 GHz, and hopefully up to the THz range [7]. The TRSTT operation frequency in the NDR region depends on two main factors: the tunneling time τt between the two channels, and the carrier transit time τSD from the source to drain in the dual channel structure. Usually, the tunneling time is estimated to be less than 0.5 ps [6], so the TRSTT performances are essentially determined by τSD. In this letter, based on devices similar to those in ref [6] we implement a small-signal equivalent circuit model accounting for tunneling between the TRSTT two channels, and for which fT in tunneling mode is calculated analytically in the common source circuit configuration. Our model predicts that fT in NDR region can reach THz range by optimizing the channel material, and shortening the gate length. II. TRSTT STRUCTURE AND MEASUREMENT Fig. 1. Experimental drain current vs. drain voltage under different gate voltages in TRSTT starting from Vgs=0 V to Vgs=0.6 V with a step of 0.1 V. The device structure is schematically shown in fig.1 inset. The heterostructures consist of a 0.8-µm intrinsic GaAs buffer layer, a 90-Å undoped In0 .2Ga0.8As channel layer, a 90-Å undoped GaAs spacer layer, followed by a silicon δ-doped layer with a 4×1012 cm-2 sheet density, and a 300-Å undoped GaAs cap layer. The gate dimension is 2×60 µm2, and the source-gate and gate-drain separations are 2 µm, each. Fig. 1 displays the device output characteristics with NDR > REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < > REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE CLICK HERE TO EDIT) < 2 under different gate biases. Noticeable gate gate leakage current is observed with the NDR onset, which is attributed to TRST [6]. observed with the NDR onset, which is attributed to TRST [6]. The channel electron mobility is measured to be 5604 cm2/Vs The channel electron mobility is measured to be 5604 cm by Hall test, and the sheet density of carrier of carrier in In0.2Ga0.8As channel is 9.02×1011 cm-2 at room temperature. at room temperature. In the absence of NDR i.e. TRST, fT at bias point C (V Vgs=0.6V, Vd=5V) is measured to be 8.9 GHz1. III. SMALL SIGNAL EQUIVALENT CIRCUIT W ITH IRCUIT WITH TUNNELING drain parasitic resistances for the 2 drain parasitic resistances for the 2-DEG InGaAs channel and the silicon δ-doped channel, respectively. doped channel, respectively. Rs is the total source parasitic resistance for the two channels. parasitic resistance for the two channels. Rt and Ct are the tunneling resistance and capacitance, with their product resistance and capacitance, with their product yielding the tunneling time. In our model, we assume tunneling yielding the tunneling time. In our model, we assume tunneling between the two channels occurs at a certain point right under between the two channels occurs at a certain point right under the gate in the 2-DEG channel. We define DEG channel. We define Lg1 and Lg2 as the effective lengths of the 2-DEG and DEG and δ-doped channels through which TRST electrons transit (see fig.2.a). Then which TRST electrons transit (see fig.2.a). Then Lg1/Lg determines the tunneling position under the gate. As the determines the tunneling position under the gate. As the tunneling distance is short, only a fraction of gate voltage drops tunneling distance is short, only a fraction of gate voltage drops across the tunneling barrier, although across the tunneling barrier, although generating a significant tunneling current, and correspondingly a negligible tunneling tunneling current, and correspondingly a negligible tunneling resistance Rt compared to the undoped GaAs cap layer. compared to the undoped GaAs cap layer. (a) (b) Fig. 2. TRSTT small-signal equivalent circuit in the common source configuration . (a). Functions of each circuit e lement with corresponding of each circuit element with corresponding position. (b). Schematic of the equivalent circuit. In Fig.2 (a) we show the small-signal equivalent circuit of signal equivalent circuit of tunneling effect. Under specific the TRSTT the TRSTT including including tunneling effect. Under specific combination of gate and source-drain biases, mo drain biases, most of electrons tunnel from the high mobility 2-dimensional electron gas dimensional electron gas (2-DEG) channel in the InGaAs material to the low mobility DEG) channel in the InGaAs material to the low mobility silicon δ-doped channel [1]–[2], where few electrons leak to the , where few electrons leak to the gate leakage. The TRSTT behavior in the tunneling mode can gate leakage. The TRSTT behavior in the tunneling mode can be described as two enhancement HEMTs operating in series be described as two enhancement HEMTs operating in series with the same gate bias. In the first one the 2 with the same gate bias. In the first one the 2-DEG is in the InGaAs channel and in the other one it is in the silicon InGaAs channel and in the other one it is in the silicon δ-doped channel. By following the paths the electrons flow, one can channel. By following the paths the electrons flow, one can build the equivalent circuit shown in fig.2 (b). quivalent circuit shown in fig.2 (b). Cgsp and Cgdp are source-gate and gate-drain geometrical capacitances whose drain geometrical capacitances whose values are determined by the dimensions of electrode layout. values are determined by the dimensions of electrode layout. Cgs1 and Cgs2 are the gate-source intrinsic capacitances that source intrinsic capacitances that control the source-drain current, while drain current, while gm1 and gm2 are the corresponding transconductances. Cgd is the drain-source intrinsic capacitance which describes the electron inflow into the electron inflow into the depletion layer when the electrode voltage changes. the depletion layer when the electrode voltage changes. Ri1 and Ri2 are the dual channel resistances, while ances, while Rd1 and Rd2 are the 1 In ref. [6], the mentioned fT=9 GHz is taken from the 6 9 GHz is taken from the 6 µm channel length sample used in our modeling and its accurate value is 8.9 GHz. value is 8.9 GHz. IV. UNITY CURRENT GAIN F FREQUENCY IN THE NDR REGION fT is defined when ห݅ௗ ݅௚ ห ൌ ห ൌ . For the sake of simplicity, Cgd and Cgdp are ignored here, because of their small values. From are ignored here, because of their small values. From the small-signal form of Kirchhoff’s signal form of Kirchhoff’s circuit laws one gets for the drain current, the gate current, and the voltage drops across the drain current, the gate current, and the voltage drops across Cgs1 and Cgs2 ,respectively. ݅ௗ ൌ ݃௠ଵݒଵ ଵ ൅ ݃௠ଶݒଶ ሻݒଵ ൅ ݆߱ܥ௚௦ଶݒଶ ൅ ݒଶ ݅௚ ൌ ݆߱ሺܥ௚௦ଵ ൅ ܥ௚௦௣ ሻ ܴ௚ ݒ௚ െ ሺ ሺ݅௚ ൅ ݅ௗ ሻܴ௦  ൅ ݆߱ሺܥ௚௦ ௚௦ଵ ൅ ܥ௚௦௣ ሻܴ௜ଵ ݒଵ ൌ (3) (2) (1) ݒଶ ൌ (4) ܴ௜ଶ ൅ ݒ௚ െ ൫݅௚ ൅ ൫ ൅ ݅ௗ ൯ܴ௦ ܴ௧ ܴ௧ ൅ ݃௠ଶܴ௧  ൅ ݆߱߬௧  ൅ ݆߱߬௧ ൰  ൅ ߬௧ ൅ ݆߱ܥ௚௦ଶ ൬ܴ௜ଶ ൅ ܴ௚  ൅ ݆߱߬ with ߱ ൌ ߨ்݂ , ߬௧ ൌ ܴ௧ ܥ௧ . Owing to the strong I vs. V . Owing to the strong I vs. V non-linearity, the fT determination determination in the NDR region would be tedious under small signal operation. Rather, tedious under small signal operation. Rather, one can estimate its value by calculating it on both sides of the NDR region i.e. at its value by calculating it on both sides of the NDR region i.e. at points A and B shown on fig. 1(a). By ignoring the gate leakage, one obtains the following equation obtains the following equation for drain currents in the two channels right before and after tunneling channels right before and after tunneling, ൌ ܫௗ஻ ݊௦ଵݍܹ௚ݒ௘ଵ ൅ ݊௦ଶݍܹ௚ ݒ௘ଶ ൌ ܫௗ஺ ሺ݊௦ଵ ൅ ݊௦ଶ ሻݍܹ௚ ݒ௘ଵ Here q is the electron charge, Here q is the electron charge, IdB and IdA are the drain currents at bias points A and B, ve1, , ve2 are the saturation velocity of electrons in both channels, W electrons in both channels, Wg is the gate width, and ns1 and ns2 are the carrier density in the 2 are the carrier density in the 2-DEG channel and δ-doped channel under the gate after tunnel channel under the gate after tunneling. Based on the definition gm1 and gm2, one gets ݃௠ଶ ൌ ߛ௏ ሺ െ ߛ௉ ሻ ݃௠ଵ ൌ ߛ௉ െ ߛ௏ ߛ௉ ሺ െ ߛ௏ ሻ ݃௠  ݃ ߛ௉ ሺ െ ߛ௏ ሻ ݃௠ (6) (5) 3 ்݂_ಳ ൌ (10) (7) (8) (9) ߲݊௦ଶ ߲ܸ௚௦ Fig. 3. (Color online) fT in the NDR region as a function of gate lengths for different tunneling positions and the optimized channel saturation ve locities. VI. CONCLUSION Our small signal equivalent circuit model for TRSTT shows that THz frequency operation in the NDR region of the devices is possible by shrinking the gate length, but also improving the mobility or saturation velocity in the 2DEG channel. As expected the critical factor is the δ-doped channel resistance that should remain significant compared to the 2DEG channel resistance if large NDR peak-to-valley ratios are required. In this context, it may be worth exploring other material systems offering optimum mobility/saturation velocity difference with large and abrupt NDRs. Finally, let us mention that a key assumption in our analysis was the TRST occurrence along the channel. For this purpose a physical model of quantum tunneling between channels under bias conditions is desirable. > REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < where we call ߛ௉ ൌ ܫௗ஻ܫௗ஺ and ߛ௏ ൌ ݒ௘ଶ ݒ௘ଵ the tunneling peak ratio and velocity difference ratio. Usually, the gate-source capacitance Cgs is defined as the net increase of positive charge in the depletion area by an incremental increase in gate-source voltage [8]. However for δ-doped TRSTT, the only positive charge is in the δ-doped layer. Instead, we obtain Cgs indirectly by calculating the variation of the carrier densities Q1 and Q2 in the two channels under the gate. In the tunneling mode, the total gate-source capacitance can be written as ܥ௚௦ ൌ ߲ܳଵ ߲ܸ௚௦ ൅ ߲ܳଶ ߲݊௦ଵ ߲ܸ௚௦ ൌ ݍܮ௚ଵܹ௚ ߲ܸ௚௦ ൅ ݍܮ௚ଶܹ௚ The drain currents in the two channels read ܫௗଵ ൌ ݍܹ௚ ݒ௘ଵ݊௦ଵ ൌ ݃ௗଵܸௗ௦  ܫௗଶ ൌ ݍܹ௚ ݒ௘ଶ݊௦ଶ ൌ ݃ௗଶܸௗ௦ where gd1, gd2 are the drain conductance of the 2-DEG and δ-doped channels, respectively. Combining (7), (8), one can get ݒ௘ଵ ݃௠ଵ ൅ ܮ௚ଶ ܥ௚௦ ൌ ܮ௚ଵ ݒ௘ଶ ݃௠ଶ One determines fT in the tunneling mode by solving (1)–(4), (6) and (9). In a first order approximation one can neglect Ri1, Ri2, Rs, Rd, Cgsp, and τt compared to other circuit elements to obtain a closed form for fT at point B as shown in (10) ݒ௘ଵ ߛ௉ ሺ െ ߛ௏ ሻ ߨൣܮ௚ଵሺߛ௉ െ ߛ௏ ሻ ൅ ܮ௚ଶ൫ െ ߛ௣ ൯൧ When no tunneling occurs, γP=1 and γv=0. The fT at point A can be derived from that of fT_B as shown in (11) ்݂_஺ ൌ ݒ௘ଵ ߨܮ௚ V. RESULTS AND DISCUSSION Based on the measured channel mobility, one can estimate the saturation velocity ve1 of the sample to be 1×107 cm/s. Inserting this value into (11), we get fT=7.96 GHz, which agrees well with the experimental value 8.9 GHz obtained for the device in saturation, in the absence of TRST1. In Fig. 3, we show the calculated fT as a function of gate length around the NDR region for different tunneling positions along the 2- DEG channel. The best fT can be achieved by shrinking the gate length to 40 nm with a tunneling position close to the drain. If we optimize the saturation velocity to 2×107 cm/s by using higher indium composition InGaAs material and maintain the values of other parameters in (10) and (11), fT in the NDR region can reach a range between 620 GHz and 800 GHz for Lg1/Lg=0.7, which indicates that high frequency response is obtained for tunneling closer to the drain than the source, in order to reduce the δ-doped channel resistance. This range of fT is shown in fig. 3 between the black solid line with solid circles and the red solid line. REFERENCES J. M. Bigelow and J. P. Leburton, “Tunneling real-space transfer induced by wave function hybridization in modulation-doped heterostructures,” Appl. Phys. Lett., vol. 57, no. 8, pp. 795–797, Aug. 1990. J. Laskar, J. M. Bigelow, J. P. Leburton, and J. Kolodzey, “Experiment and theoretical investigation of the dc and high-frequency characteristics of the negative differential resistance in pseudomorphic AlGaAs/InGaAs/GaAs MODFETs,” IEEE Trans. Electron Devices, vol. 39, no. 2, pp. 257–263, Feb. 1992. [3] C. L. Wu and W. C. Hsu, “Enhanced resonant tunneling real-space transfer in δ-doped GaAs/InGaAs ga ted dual-channel transistors grown by MOCVD,” IEEE Trans. Electron Devices, vol. 43, no. 2, pp. 207–212, Feb. 1996. [4] Y. W. Chen, Y. J. Chen, and W. C. Hsu, “Enhancement-mode In0.52Al0.48As/In0.6Ga0.4As tunneling real space transfer high electron mobility transistor,” J. Vac. Sci. Technol. B, Microelectron. Process. Phenom., vol. 22, no. 3, pp. 974–976, May 2004. [5] T. Sugaya and K. Komori, “InGaAs dual channel transistors with negative differential resistance,” Appl. Phys. Lett., vol. 88, p. 142 107, Apr. 2006. [6] Xin Yu, Lu-Hong Mao, Wei-Lian Guo, Shi-Lin Zhang, etc, “Monostable-Bistable Transition Logic Element Formed by Tunneling Real-Space Transfer Transistor With Negative Differential Resistance,” IEEE Electron Device Lett., vol. 31, no. 11, pp. 1224–1226, Nov. 2010. [7] K. Furuya, O. Numakami, N. Yagi, e tc, “Ana lysis of terahertz oscillator using negative differential resistance dual-channel transistor and integrated antenna,” Jpn. J. Appl. Phys., vol. 48, p. 04C146, Apr. 2009. [8] Peter H. Ladbrooke, MMIC Design: GaAs FETs and HEMTs. Norwood, MA: Artech House, 1989, pp. 105-109. VII. ACKNOWLEDGEMENT We are indebted to Dr. Elyse Rosenbaum for helpful discussions. (11) [1] [2] 

1212.5422

3

1212

Getting through the nature of silicene: sp2-sp3 two-dimensional silicon nanosheet

By combining experimental techniques with ab-initio density functional theory calculations, we describe the Si/Ag(111) two-dimensional system in terms of a sp2-sp3 crystalline form of silicon characterized by a vertically distorted honeycomb lattice. We show that 2D sp2-sp3 Si NSs are qualified by a prevailing Raman peak which can be assigned to a graphene-like E2g vibrational mode and that highly distorted superstructures are semiconductive whereas low distorted ones behave as semimetals.

cond-mat.mes-hall

cond-mat

Getting through the nature of silicene: sp2-sp3 two- dimensional silicon nanosheet Eugen io Cinquan ta *1, Emilio Sca lise2, Dan iele Ch iappe 1, Carlo Grazianetti1, 3, Bas van den Broek 2, M ichel Houssa 2, Marco Fanciu lli1,3 and Alessandro Mo lle 1 * Dr. E. C. Corresponding-Author: email: eugenio.cinquanta@mdm. imm.cnr. it Laboratorio MDM, IMM-CNR, via C. Olivetti 2, I-20864 Agrate Brianza (MB), Italy Semiconductor Physics Laboratory, Department of Physics and Astronomy, University of Leuven, B-3001 Leuven, Belgium Dipartimento di Scienza dei Materiali, Università degli Studi di Milano Bicocca, via R. Cozzi 53, I-20126, Milano (MI), Italy 1 2 3 Keywords: (silicene, silicon nanosheet, Raman spectroscopy, density-functiona l theory) 1 Abstract By combining experimental techniques with ab-initio density functional theory calculations, we describe the Si/Ag(111) two-dimensional system in terms of a sp2-sp3 crystalline form of silicon characterized by a vertically distorted honeycomb lattice. We show that 2D sp2-sp3 Si NSs are qualified by a prevailing Raman peak which can be assigned to a graphene -like E2g vibrational mode and that highly distorted superstructure s are semiconductive whereas low distorted ones behave as semimetals. 2 In the latest years a considerable effort has been devoted to identify the graphene -like allotrope of silicon, namely silicene, both from the experimental and theoretical point of view. 1-5 Although free-standing (FS) silicene has been hypothesized as a pure sp2 hybridization of silicon, its existence in ambient conditions has been argued due to strong electron correlation .6 On the other hand, many authors succeeded in epitaxially growing a two-dimensional (2D) Si nanosheet (NS) on the Ag(111),1,7-9 Ir(111)10 and ZrB2 11 surface with differently buckled sp2-sp3 arrangements. Compelling evidences of Dirac fermions and preliminary indications of band gap opening in a Si NS epitaxially grown on Ag(111) substrates1 makes it enormously attractive in terms of fundamental properties2-5 and technological transfer. 2, 11 So far epitaxial silicon NSs have be en referred to as silicene layers because their topography is consistent with a distorted, energetically stable, honeycomb lattice .1, 8 However, for a proper identification of these Si NSs as graphene -like 2D materials, it is essential to understand the arrangement of the bonds in the Si NSs (i.e . geometry and hybridization) and how the Si bonds influence their vibrational and electronic properties.1,12 To get through this aspect and elucidate the intimate nature of 2D Si NSs, various superstructures7,8,13-15 of a Si monolayer epitaxially grown on Ag(111) substrates have been scrutinized and selected by means of in situ scanning tunneling microscopy (STM) and, after non-reactive encapsulation,16 their vibrational modes have been unveiled by Raman spectroscopy. Experimental results have been then interpreted with ab -in itio Density-Functional Theory (DFT) calculation. Epitaxial 2D Si NSs on Ag(111) evidence a multi-phase character,7-9,13-14 namely a variety of reconstructed Si domains with characteristic periodic patterns including 4x4, √13X√13-II, and 2√3X2√3-II (where the periodicity is defined with respect to the Ag unit cell) which can be 3 recognized by STM7-9 or electron diffraction techniques. 1,13-14 Each superstructure corresponds to a honeycomb Si domain with characteristic buckling γ, i.e. the local vertical displacement induced by the underlying Ag atoms,14, 15 and follows a well-defined phase diagram dictated by the deposition temperature for a fixed deposition flux. 8, 13, 14, 17 In our experimental approach two characteristic growth regimes were identified: one at a lower deposition temperature (T=250°C) where the 4x4 and the √13X√13-II Si superstructures coexist on the Ag(111) surface, and a second one at a higher temperature (T=270°C) where a uniform Si layer with a characteristic 2√3X2√3-II superstructure is observed. The topography of the Si NS grown at 250°C is shown in Fig. 1(a) therein evidencing 2D Si domains with lateral size in between 20nm and 50nm and uniformly distributed throughout the Ag(111) surface. The atomically resolved STM magnification in Fig. 1(b) shows that Si atoms are sequentially placed either in between or on top of the Ag atoms, thus generating two characteristic surface patterns, a 4x4 (green contour in Fig. 1.b ) or a √13X√13-II buckled superstructure (red contour in Fig. 1.b). 14, 15 The 4x4 superstructure is defined with respect to the unit cell of the underlying Ag(111) surface. The calculated cell parameter of the 4x4 superstructure is 11.78 Å, i.e. nearly four times the cell parameter of the Ag(111) surface . The 4x4 supercell includes three hexagonal unit cells of silicene with cell parameter of 3.926 Å. This value is larger by about 1.5% than the calculated value for the FS silicene 3 thus indicating that the epitaxial silicene on Ag is slightly stretched. Other relevant features discriminating the epitaxial silicene from the FS one are: a) the more pronounced vertical stacking (0.71<γ<0.79 Å), and b) the non-uniform distribution of buckled atoms. The latter feature is a consequence of the substrate -induced breaking of the threefold rotational symmetry present in the FS silicene. In detail, only 6 of the 18 atoms building up the 4 4x4 superstructure are on a higher plane with respect to the lower one which is closer (~2 Å) to the Ag substrate and contains the remaining 12 atoms. 1, 7, 8, 13-15 The √13X√13-II reconstruction is a supercell of silicene made up of 14 atoms and rotated by about 5.2° relatively from its initial position in the 4x4 superstructure. The angle between the axis of the √13X√13-II superstructure and the 4x4 one is 13.9. Only 4 of 14 atoms are on the higher plane, forming a buckled structure with a vertical parameter very close to that of the 4x4 superstructure (γ0.79 Å).7, 8, 13-15 Figure 1.d reports the large scale topography of the 2√3X2√3-II Si NS superstructure grown at 270°C. For coverage 1 ML the 2√3X2√3-II domains extend over the whole Ag(111) terraces thus proving an improved continuity of the layer with respect to the more fragmented surface structure of the Si NS in Fig. 1(a). The atomic scale topography in Figure 1.e exhibits the characteristic surface patterns of the 2√3X2√3-II superstructure, as reported elsewhere. 7, 8, 14, 15 This superstructure is characterized by a unit cell with 12 planar and two buckled (γ 1 Å) atoms and it is misaligned from the 4x4 superstructure by an angle of 30°. 14, 15 Remarkably, for the three considered superstructures the bond length ranges f rom 2.34 to 2.39 Å for the 4x4 one, from 2.31 to 2.36 Å for the √13x√13-II one and from 2.28 and 2.37 for the 2√3x2√3-II, in all cases being quite close to the Si-Si bond length in sp3 diamond like Si (2.34 Å). DFT modeling of these superstructures generally results in a bond angle distribution peaked around 109° and 120° which respectively identify sp3 and sp2 hybridized Si atoms. It must be emphasized that the 2D Si NSs preserve the sp2 trigonal geometry (each Si atoms is coordinated with other 3 Si atoms) whereas the bond lengths and bond angles evidence a sp3-like nature because of the presence of buckled Si atoms stemming from substrate -induced local vertical distortions. More specifically, the substrate constraints concomitantly induce i) a non-uniformly 5 distributed vertical alignment of Si atoms, ii) a horizontal “in-plane” strain being tensile in character, and iii) a partial hybridization of the resulting unhybridized Si P z orbitals with the underlying Ag atoms.12 These features are expected to dramatically impact the vibrational properties of each single superstructure and they can be elucidated by means of Raman spectroscopy. Figure 2.a shows the Raman spectrum of the sample including the 4x4 and the √13X√13-II superstructures. The spectrum is characterized by an intense peak located at 516 cm-1 presenting an asymmetric and broad shoulder at lower frequency (440-500 cm-1). It is interesting to notice that this feature cannot be provided by a full sp3 nanocrystalline silicon (nc -Si), because only 2nm tailed nanocrystals would induce the observed redshift,18, 19 whereas the STM probed domains turn out to have a width ranging from 20 to 50 nm (see Figure 1.a). A derivation of the peak at 516 cm -1 from a fully sp3 stressed Si can be also ruled out because the periodicity of the Si lattice (0.54 nm) is by far larger than the Ag(111) surface cell parameter (0.29 nm). As a consequence, the resulting compressive strain should up-shift Raman peak of bulk silicon at 520 cm-1, thus not being consistent with the observed peak position.20 The two extra-features present at 300 and 800 cm-1 are provided by the Al2O3 capping layer. 21 In order to interpret the experimental spectrum in Fig. 2.a, the non-resonant Raman spectrum of the defect-free 2D Si NSs on Ag(111) has been simulated and reporte d in Fig. 2b and 2c. In detail, Fig. 2.b and 2.c show the vibrational modes of both √13X√13-II and 4x4 superstructures, respectively. Despite the sp2-sp3 nature of the two phases, a doubly degenerate E2g mode is predicted to take place at 505 cm-1 and 495 cm-1 for the √13X√13-II and 4x4 superstructures respectively. These frequencies are in pretty good agreement with the experimental feature at 6 516 cm-1, taking into account an underestimation of about 1-5% for the vibrational frequencies, typical for the implemented computation (see methods for details). 22, 23 As for the graphene G- peak, this mode is provided by the bond stretching of all sp2 silicon atom pair, thus reflecting the 2D-Si NSs sp2 character and being the fingerprint of a honeycomb lattice. On the other hand E2g mode frequency, 516 cm-1, is strictly dependent on the sp3-like Si-Si bond length (~2.34 Å), being present also in the FS silicene but with relatively much higher frequencies (~570 cm -1) as due to the shorter and full sp2 bond length (2.28 Å). 23 Although the presence of E2g modes clearly indicates the honeycomb nature of the 2D Si-NSs lattice, the broad and asymmetric shoulder in the 440-500 cm-1 is not compatible with a defect-free planar trigonal geometry. Hence it is interesting to elucidate the presence of this feature , especially considering the structural complexit ies characterizing 2D Si-NSs, as previously discussed (Fig. 1) . It is useful noticing that defect-free graphene, characterized by a planar honeycomb lattice, provides a Raman spectrum characterized by the presence of a E 2g mode (G peak at 1581 cm-1) plus the 2D peak at 2600 cm-1 activated by inter-valley electron-phonon scattering between K and K’ point of the first Brillouin Zone (FBZ). 25 For graphene , the presence of an additional D(A1g) peak at 1300 cm-1 is expected when double resonance processes take place , that is intra- valley electron-defect scattering at K in the FBZ. 24 Surprisingly, unlike FS silicene, 23 the calculated spectra of the Si superstructures are characterized by several vibrational modes having a non-vanishing Raman intensity (denoted as D, T and K in Fig. 2.b and 2.c) also assuming a defect-free configuration in the model. Interestingly, for the 4x4 case, two A 1g modes at 436 cm-1 and at 466 cm-1 (reported as D and T respectively in Fig. 2.c) are Raman-active and, surprisingly, the D one is the dominant feature. 7 Similarly, the √13X√13-II superstructure presents, beyond the E2g vibrational mode, a D (A1g) peak at 455 cm-1 plus a K (B2u) at 475 cm-1, as reported in Fig. 2.c. By comparing the calculated spectra reported in Fig. 2.b and 2.c, it can be noticed that the A1g mode at 436 cm-1 is the most intense Raman-active mode in case of the 4x4 superstructure , while the E2g mode becomes largely dominant for the √13X√13-II one. Since these two superstructures mutually differ in terms of their intimate buckling distribution, the observed intensity variation of the Raman features is reasonably related to their different atomic configuration. In particular, a reduced number of buckled bonds (as in the √13X√13-II) results in a strong suppression of the A1g modes intensity with respect to the E2g peak. The origin of A1g mode activation can be then associated to an intrinsic “disorder” related to the non-uniform substrate-induced buckling and to the mixed sp2-sp3 nature of the honeycomb Si lattice . More in details, for the 4x4 case , the D(A1g) mode comes form the breathing-like displacement of planar hexagons, while the T(A 1g) is related to the breathing-like displacement of non-planar hexagons (see Fig. S1 in Suppl. Information); for the √13X√13-II, the D(A1g) and K(B2u) modes arise from the breathing mode of the hexagonal rings having alternating up and down-standing atoms and to Kekule -distorted25 hexagonal rings respectively (see Fig. S1 in Suppl. Information). Since the experimental Raman spectrum integrates the contribute of the two phases weighted by their abundances ratio, the Raman spectrum of 2D sp2-sp3 Si NS is dominated by the E2g modes of the two superstructures, along with the asymmetric shoulder provided by the interplay of the disorder-activated modes (A1g and B2u). Remarkably, the experimental spectrum in Fig. 2a is well reproduced by the calculated spectrum of the √13X√13-II superstructure, rather than by the one of the 4x4 one. This asymmetry may come from the smaller amount of 4x4 superstructure domains with respect to the 8 √13X√13-II superstructure ones. Nonetheless, it is interesting to explore whether a Raman resonance, not implemented in the adopted DFT framework, might take place also in consideration of the measured band gap of 0. 6 eV for the 4x4 superstructure. In this case, resonance effects are expected to selectively amplify the E2g and A1g Raman-active modes. Indeed, the former is an intra -valley phonon scattering process at Γ and thus its resonance is strictly related to electronic transitions when the incident radiation is properly tuned with the direct band gap transition, with the consequent enhancement of the related Raman signal. To get through possible resonant behaviors, the Raman spectrum of the two configurations has been investigated as a function of the excitation energy. The case of the mixed 4x4/√13X√13-II is reported in Figure 3. The pronounced enhancement of E2g intensity with increasing excitation energy indicates a semiconductive character (formally similar to the one of sp3 silicon),26 being consistent with the reported presence of a direct gap in the 4x4 superstructure. 1 Interestingly, an additional hint at the multiphase character of the sample comes from the frequency dispersion of the band at 900 cm-1: the observed blueshift as a function of the excitation energy is similar to the one observed for the 2D peak of graphene, which is provided by the presence of the Kohn anomaly in the K point of the FBZ.24 Second order Raman feature of bulk-Si also places in between 900 and 1000 cm-1 as due to two transverse optical modes. When varying the exciting frequency, this band is affected by a change of the peaks intensity ratio because of Raman resonance26 rather than by frequency dispersion. The resonance behavior and the frequency disper sion can be rationalized by attributing a semiconducting character to the 4x4 superstructure consistently with ARPES outcomes 1 and a graphene-like character to the √13X√13-II one. However, it is not trivial to discriminate the two contributions as, even finely tuning the deposition parameters, it was not possible to isolate just 9 one of these two superstructures. In this effort, the 2√3X2√3-II superstructure has been successfully stabilized as single phase after carefully tailoring the growth conditions (see Fig. 1.d). Figure 4 compares the experimental Raman spectrum of the 2√3X2√3-II (Fig. 4.a), acquired with four excitation energies, with the calculated one (Fig. 4.b). As for the √13X√13-II superstructure, the Raman spectrum is characterized by the presence of a strong E2g mode at 521 cm-1, confirming both the sp2 hexagonal lattice symmetry and the sp3 like nature of its bonds lengths. Furthermore weak A1g modes are also active (see Fig. S2 in the Suppl. Information), thus indicating a lower amount of intrinsic -disorder with respect to the previously analyzed superstructures. Interestingly the Ultra -Violet (UV) Raman spectrum exhibits a quite low signal/noise ratio, thus reflecting a very low cross section at this frequency, that is opposed to the expected behavior of the sp3 diamond-like silicon.26 By observing the spectra reported in Fig. 4a, two remarkable facts must be underlined: a) no Raman resonance affects the E 2g mode, and b) a frequency dispersion characterizes the 900 cm-1 band. Both observations are not compatible with a full semiconductive sp3 structure, 26 but they rather reflect the presence of a Si honeycomb lattice with a characteristic graphene -like behavior, that is a non-resonant behavior and Kohn- anomaly related frequency dispersion. In summary, by combining experimental techniques with ab-initio DFT calculation, we describe the Si/Ag(111) 2D systems in terms of a sp2-sp3 form of silicon characterized by a vertically distorted honeycomb lattice provided by the constraint imposed by the substrate. The Raman spectrum reflects the multi-hybridized nature of the 2D Si NSs resulting from a buckling- induced distortion of a purely sp2 hybridized structure . This sp2-sp3 character provokes an intrinsic disorder which leads to the activation of Raman-inactive vibrational modes (A1g), whose intensity is a function of the amount of buckled Si atoms . For the 2D Si-NSs where two different 10 superstructures coexist (4x4 and √13X√13-II), the dependence of the Raman response with the excitation energy allows us to recognize a non-trivial semiconducting character. We then successfully isolate the 2√3X2√3-II superstructure which makes evidence of a low -distorted honeycomb lattice thus opening new interest for 2D sp2-sp3 silicon nanosheets with a graphene - like symmetry and electronic character. M e thods The synthesis of our sample is the same adopted in7 and briefly reported in the Supplementary Information. In order to prevent silicon oxidation we follow the same procedure showed in. 17 Ex-situ visible and Ultra-Violet (UV) Raman characterization was performed in a z backscattering geometry by using a Renishaw Invia spectrometer equipped with the 1.96 eV/633nm, 2.41eV/514nm, 2.56eV/488nm and 3.41eV/364nm line of an A r+ laser line focused on the sample by a 50x 0.75 N.A. Leica objective. The power at the sample was maintained below 1 mW in o rder to p reven t las er induced s amp le heat ing and hundreds of spectra have been acquired in order to get the highest signal/noise ratio. Calculations were carried out in the density-functional theory (DFT) framework, within the generalized gradient approximation (GGA), as implemented in the Quantum ESPRESSO package.27 The valence electrons of Si were treated using norm-conserving pseudopotentials 28 with a kinetic energy cutoff of 36 Ry. The k-point mesh was set to 4x4x1 during the structural relaxation of the silicene unit cell, which was performed until the average atomic force was lower than 10-4 Ry/Bohr, while a 25x25x1 Monkhorst-Pack grid was used for the phonon calculations. The grid was then reduced to 3x3x1 for the calculations of the silicene supercells. The phonon frequency was obtained by diagonalization of the dynamical matrix calculated by the density-functional perturbation theory (DFPT). 29 The non-resonance Raman tensors were 11 obtained within DFPT by second order response to an electric field as implemented in the Quantum ESPRESSO package. 30 12  Figure 1 : (a) Large scale STM characterization of silicene domains on Ag(111) and (b) t he two most abundant superstructures: green 4x4, red √13X√13-II;(c) 4x4 and √13X√13-II DFT-GGA relaxed superstructures. (d) Large scale STM characterization and (d) atomic scale topography of the 2√3X2√3 superstructure. (f) DFT-Local Density Approximation relaxed 2√3X2√3 superstructure. STM images were acquired at -1.4 V and 0.4 nA set point. Grey balls are Ag(111) atoms, blue balls are the low lying silicene atoms, yellow balls are the top lying silicene atoms. 13  Figure 2 (a) Experimental Raman spectrum of Al2O3-capped silicene acquired with the 633 nm laser lines. The inset shows the whole spectrum where at 300 and 800 cm-1 lie the Al2O3 features. (b) and (c) computed Raman spectra of the √13X√13-II and 4x4 superstructures respectively, obtained by the calculated vibrational spectra convoluted with a uniform Gaussian broadening having a full width at half maximum (FWHM) of 10 cm-1. 14 Figure 3 Raman spectrum of the coupled √13X√13-II and 4x4 superstructures acquired with four excitation energies. The inset shows the dispersion of the band at 900 cm-1 as a function of the excitation energy; spectra are vertically stacked for clarity. 15  Figure 4 (a) Raman spectrum of the 2√3X2√3 superstructure acquired with four different excitation energies. The inset shows the dispersion of the band at 900 cm-1 as a function of the excitation energy; UV spectrum is not shown because the signal/noise ratio is too low in this spectral region. Spectra are vertically stacked for clarity. (b) computed Raman spectra of the 2√3X2√3 superstructure obtained by the calculated vibrational spectra convoluted with a uniform Gaussian broadening having 10 cm-1 FWHM . 16 Acknowledgements The present research activity has been carried on within the framework of the EU project 2D - NANOLATTICES. The project 2D -NANOLATTICE acknowledge s the financial support of the Future and Emerging Technologies (FET) programme within the Seventh Framework P rogramme for Research of the European Commission, under FET -Open grant number: 270749. M.H., E.S. and B. v.d.B. greatly acknowledge Dr. B. Ealet (CNRS and Aix-Marseille University, CINaM) and Dr. G. Pourtois (IMEC and Chemistry Department, PLASMANT group, University of Antwerp). 17 [1]Vogt, P .; De Padova, P .; Quaresima, C.; Avila, J.; Frantzeskakis, E. ; Asensio, M. C.; Resta, A.; Ealet, B.; Le Lay G. Silicene: Compelling Experimental Evidence for Graphenelike Two-Dimensional Silicon, Phys. Rev. Lett. 2012 , 108 , 155501. [2]Ezawa, M. A topological insulator and helical zero mode in silicene under an inhomogeneous electric field, New J. Phys. 2012 , 14 , 033003. [3]Houssa, M.; Pourtois, G.; Heyns, M.M.; Afanas'ev V.V.; Stesmans , A. Electronic P roperties of Silicene: Insights from First-P rinciples Modeling, ECS Trans. 2010 , 33 , 185. [4] Drummond, N.D.; Zólyomi, V.; Fal'ko, V.I. Electrically tunable band gap in silicene , Phys. Rev. B 2012 , 85 , 075423. [5]Guzmán-Verri G. G.; Lew Yan Voon, L. C. Electronic structure of silicon-based nanostructures, Phys. Rev. B 2007 , 76 , 075131 . [6]Sheka, E. F. Why sp2 -like nanosilicons should not form: Insight from quantum chemistry, In t. J. Quan tum Chem. 2013 , 113 , 612–618. [7]Chiappe, D.; Grazianetti, C.; Tallarida, G.; Fanciulli M.; Molle, A. Local electronic properties of corrugated silicene phases. , Adv. Ma ter 2012 , 24 , 5088. [8]Feng, B.; Ding, Z.; Meng, S.; Yao, Y.; He , X.; Cheng, P .; Chen L.; Wu K. Evidence of Silicene in Honeycomb Structures of Silicon on Ag(111) , Nano Lett. , 2012 , 12, 3507. [9]Lin, C.L.; Arafune, R.; Kawahara, K.; Tsukahara, N.; Minamitani, E.; Kim, Y.; Takagi N.; Kawai M. Structure of Silicene Grown on Ag (111) , App l. Phys. Express 2012 , 5 , 045802. [10] Meng, L.; Wang, Y.; Zhang, L.; Du, S.; Wu, R.; Li, L.; Zhang, Y.; Li, G.; Zhou, H.; Hofer W. A.; Gao, H.-J. Buckled Silicene Formation on Ir(111) , Nano Lett. 2013 , 13 (2), 685-690. 18 [11]Fleurence , A.; Friedlein, R.; Ozaki, T.; Kawai, H.; Wang Y.; Yamada -Takamura, Y. Experimental Evidence for Epitaxial Silicene on Diboride Thin Films , Phys. Rev. Lett., 2012 , 108 , 245501. [12] Lin, C.-L.; Arafune, R.; Kawahara, K.; Kanno, M.; Tsukahara, N.; Minamitani, E.; Kim, Y.; Kawai, M.; Takagi, N. Substrate-Induced Symmetry Breaking in Silicene Phys. Rev. Lett. 2013 , 110 , 076801. [13] Arafune, R.; Lin, C. -L.; Kawahara, K.; Tsukahara, N.; Minamitani, E.; Kim, Y.; Takagi N.; Kawai, M. Structural transition of silicene on Ag(111) , Surf ace Science 2013 , 608 , 297. [14] Jamgotchian, H.; Colignon, Y.; Hamzaoui, N.; Ealet, B.; Hoarau, Y.; Aufray B.;. Bibérian, J. P ., Growth of silicene layers on Ag(111): unexpected effect of the substrate temperature, J. Phys.: Condens. Ma tter 2012 , 24, 172001. [15] Henriquez, H.; Vizzini, S.; Kara, A.; Lalmi B.; Oughaddou, H. Silicene structures on silver surfaces, J. Phys.: Condens. 2012 , 24 , 314211. [16]Molle. A.; Grazianetti, C.; Chiappe, D.; Cinquanta, E.; Cianci, E.; Tallarida, G.; Fanciulli, M. Hindering the Oxidation of Silicene with Non-Reactive Encapsulation, Adv. Func. Ma t. 2013 , DOI: 10.1002/adfm.201300354. [17]Gao J.; Zhao, J. Initial geometries, interaction mechanism and high stability of silicene on Ag(111) surface Sci. Rep . 2012 , 2 , 861. [18]Faraci, G.; Gibilisco, S.; Russo, P .; Pennisi, A.R. Modified Raman confinement model for Si nanocrystals, Phys. Rev. B 2006 , 3 , 033307. [19]Crowe, I.F.; Halsall, M. P .; Hulko, O.; Knights, A. P .; Gwilliam, R. M.; Wojdak M.; Kenyon, A. J. P robing the phonon confinement in ultrasmall silicon nanocrystals reveals a size - dependent surface energy, J. App l. Phys. 2011 , 109 , 083534. 19 [20]Holtz, M.; Carty, J.C.; Duncan, W.M. Ultraviolet Raman stress mapping in silicon, App l. Phys. Lett. 1999 , 74 , 2008. [21]Vali, R.; Hosseini, S.M. First-principles study of structural, dynamical, and dielectric properties of κ-Al2O3, Comp . Ma t. Sci. 2004 , 29 , 138. [22]Favot F.; Dal Corso, A. Phonon dispersions: Performance of the GGA approximation, Phys. Rev. B 1999 , 60 , 427-431. [23]Scalise, E., Houssa , M.; Pourtois, G.; van den Broek, B.; Afanas’ev V.; Stesmans, A. Vibrational properties of silicene and germanene , Nano Res,. 2013 , 6 , 19–28. [24] Ferrari, A.C.; Meyer, J. C.; Scardaci, V.; Casiraghi, C.; Lazzeri, M.; Mauri, F.; P iscanec, S.; Jiang, D.; Novoselov, K. S.; Roth, S.; Geim, A. K. Raman spectrum of graphene and graphene layers Phys. Rev. Lett. 2006 , 97 , 18740. [25]Shaik, S.; Shurki, A.; Danovich D.; Hiberty, P .C. Origins of the exalted b(2u) frequency in the first excited state of benzene , J. Am. Chem. Soc. 1996 , 118 , 666-671. [26] Renucci, J. B.; Tyte R. N.; Cardona , M. Resonant Raman scattering in silicon, Phys. Rev. B 1975 , 11 , 3885–3895. [27]Giannozzi, P . et a l. QUANTUM ESPRESSO: a modular and open-source software project for quantum simulations of materials , J. Phys.: Condens. Ma tter. 2009 , 21 , 395502. See also h t tp :/ /www.quan tum-es p res s o .o rg . [28]Bachelet, G.B.; Hamann D.R.; Schluter, M. P seudopotentials that work: From H to Pu, Phys. Rev. B 1982 , 26 , 4199. [29]Baroni, S.; de Gironcoli, S.; Dal Corso A.; Giannozzi, P . Phonons and related crystal properties from density-functiona l perturbation theory, Rev. Mod . Phys. 2001 , 73 , 515. 20 [30]Lazzeri M.; Mauri, F. First-principles calculation of vibrational Raman spe ctra in large systems: Signature of small rings in crystalline SiO2, Phys. Rev. Lett. 2003 , 90 , 036401. 21 Supporting Information Deposition process Growth experiments have been carried out in an ultra -high vacuum (UHV) system with base pressure in the 10-10mbar range incorporating three interconnected chambers for sample processing, chemical analysis and scanning probe diagnostics. The Ag(111) crystal was cleaned by several cycles of Ar+ ion sputtering (1 k eV ) and subsequent annealing a t around 530°C. Si was deposited from a heated crucible (EFM evaporator supplied by Omicron Nanotechnology GmbH) with the substrate at a temperature of 250°C and of 270°. The deposition rate was estimated to be around 6×10-2 ML s-1. STM topographies were obtained at room temperature using an Omicron STM setup equipped with a chemically etched tungsten tip. Atom displacements Figure S1 and S2 reports bond displacements related to Raman active modes of the 4x4 √13X√13-II and 2√3X2√3-II superstructures; next the E2g at 505 cm-1(Fig. 4.a), also the disorder-activated A1g modes at 436 cm-1, named D (by analogy with the D peak in graphene, see Fig. 4.b) and the T peak at 466 cm-1(as it is carried by low standing trimmers, see Fig. 4.c), are activated in the 4x4 silicene superstructures.. On the other hand, the D peak at 455 cm-1, together with a B2u at 475 cm-1 (named as K in Fig. 4.b), is present also in the calculated spectrum of the √13X√13-II. The 2√3X2√3-II superstructure presents the E2g mode at 526 cm-1 plus three A1g modes at 307 cm-1 (planar hexagons), 445 cm-1 and 430 cm-1 for hexagon with one buckled atom. 22  Figure S1 (a) E2g mode, (b) A1g mode of the planar hexagons and (c) A1g mode of the non-planar hexagons of the 4x4 superstructure. (d) E2g mode, (e) A1g mode and (f) B2u mode of the √13X√13-II superstructure. Light blue atoms are down-standing atoms, dark blue are up- standing atoms. Blue arrow indicates the forces acting on each atom, red arrows indicate forces providing each Raman-active mode. Yellow spots highlight hexagons providing A 1g and B2u modes 23 Figure S2 (a) E2g mode, (b) A1g mode of the planar hexagons, (c) and (d) A1g mode of the non- planar hexagons of the 2√3X2√3-II superstructure. Grey atoms are down-standing atom, blue are up-standing atoms. Green arrow indicates the forces acting on each atom. 24 

1303.3433

1

1303

Magnetic hysteresis loop as a probe to distinguish single layer from many layer graphitic structure

In this report we have pointed out that magnetic hysteresis loop can be used as a probe to distinguish a single layer from a many layer graphitic structure. Chemically we have synthesized graphitic oxide (GO) and reduced graphitic oxide (RGO) for this investigation. We observe ferromagnetic like hysteresis loops for both GO and RGO below a certain applied critical magnetic field and above this critical field we observe cross-over of the positive magnetization to negative magnetization leading to diamagnetic behaviour. This cross-over is more dominant for the case of many layer graphitic structure. Upon annealing of GO in air the critical cross-over field decreases and the magnetization increases for multilayer graphitic structure. Possible reasons for all these observations and phenomena is presented here.

cond-mat.mes-hall

cond-mat

Magnetic hysteresis loop as a probe to distinguish single layer from many layer graphitic structure. K. Bagani, B. Ghosh, M. K. Ray, N. Gayathri1, M. Sardar2 and S. Banerjee∗ Surface Physics Division, Saha Institute of Nuclear Physics, 1/AF Bidhannagar, Kolkata-700064, India 1Material Science Section, Variable Energy Cyclotron Centre, 1/AF Bidhannagar, Kolkata-700064, India 2Material Science Division, Indira Gandhi Center for Atomic Research, Kalpakkam 603 102, India Abstract In this report we have pointed out that magnetic hysteresis loop can be used as a probe to distinguish a single layer from a many layer graphitic structure. Chemically we have synthesized graphitic oxide (GO) and reduced graphitic oxide (RGO) for this investigation. We observe fer- romagnetic like hysteresis loops for both GO and RGO below a certain applied critical magnetic field and above this critical field we observe cross-over of the positive magnetization to negative magnetization leading to diamagnetic behaviour. This cross-over is more dominant for the case of many layer graphitic structure. Upon annealing of GO in air the critical cross-over field decreases and the magnetization increases for multilayer graphitic structure. Possible reasons for all these observations and phenomena is presented here. PACS numbers: 81.05.ue, 75.50.-y, 75.60.-d 3 1 0 2 r a M 4 1 ] l l a h - s e m . t a m - d n o c [ 1 v 3 3 4 3 . 3 0 3 1 : v i X r a ∗ email:[email protected] 1 I. INTRODUCTION: Low-dimensional graphitic structure such as graphene have attracted much research in- terest because of their unique electronic [1, 2], mechanical [3, 4], thermal [5] and magnetic [6 -- 9] properties to name few. Here in this paper we would like to investigate the magnetic behavior of graphitic oxide (GO) and reduced graphitic oxide (RGO) when it is evolving from single layer or very few layers to many layers tending towards bulk form. It is well known that graphite in a bulk form (3d-structure) exhibits diamagnetic behavior. But, it has been observed that when highly oriented pyrolytic graphite (HOPG) is irradiated with proton beam it exhibits ferromagnetic behaviour [9]. The observation of ferromagnetism in HOPG and its related materials such as graphene [10] is surprising because we know that in general the magnetic ordering is typically observed in materials with partially filled d or f shell electrons. Magnetism in zero-dimensional graphene nanofragments, one-dimensional graphene nanoribbons and defect-induced magnetism in graphene and graphite have been reviewed recently by Yazyev [11]. Many theoretical works already exists to explain unusual magnetic property in graphene related materials and are cited in our recent work [12]. In this report we would like to show that magnetic hysteresis as a function of field and tem- perature can be used as a probe to distinguish single layer graphitic structure from many layer graphitic structure tending towards bulk limit. II. EXPERIMENTAL DETAILS: Graphitic Oxide (GO) was synthesized using modified Hummers method [14]. To prepare GO we used graphite powder, conc. H2SO4, NaNO3 and KMnO4. The reaction of H2SO4 with strong oxidizing agents such as NaNO3 and KMnO4 results in increasing the interlayer separation of the carbon layers due to incorporation/intercalation of oxygen atoms. This reaction leads to formation of graphite oxide flakes. The solution was further ultrasonicated and centrifuged. Six different samples were collected at various stages: samples were filtered from the solution before reducing and we labelled it as sample A, after reducing with hy- drazine hydrate as sample B, after annealing at 150 oC and 450 oC as sample C and sample D respectively and finally samples collected from supernatant without reduction as sample E and upon reduction of the supernatant particles with hydrazine hydrate as sample F. The 2 supernatant and the filtered samples were washed with water and ethanol and was further collected and dried in vacuum drying oven at 800C to obtain dry powder samples. These samples were characterised by UV-VIS, FTIR, Raman spectroscopy, and SEM as described in [12]. Magnetic measurements were carried out using SQUID MPMS XL (Quantum Design). Magnetization data as a function of temperature and magnetic hysteresis data at three different temperatures 10 K, 100 K and 300 K were taken. Thermal gravimetric analysis (TGA) for qualitative estimation of oxygen content in the samples were also carried out. III. RESULTS: In fig. 1 we show magnetization versus temperature for sample A, B, C and D and the appropriate samples are labelled in the figure. We show zero field cooled (ZFC) and field cooled (FC) measurement at 100 Oe for all the samples while warming. We see that for all the samples there exist clear bifurcation in the ZFC-FC curve around room temperature (300K). The amount of bifurcation increases upon reduction with hydrazine (sample B) or upon annealing at high temperature (sample D). We also see that magnetization value (moments) increases upon annealing and upon chemical reduction of graphitic oxide (GO). We observe that upon annealing GO at 450 oC (sample D) the moment increases more than the chemically reduced sample (sample B). These are important observation and we would like to discuss these in discussion section below. In fig. 2 (a, b and c) we show the magnetization versus the applied field taken at 10 K, 100 K and 300 K for GO and RGO samples and we observe a distinctly sharp cross-over of magnetization from positive to negative value and vice versa in the negative direction of the applied field beyond a certain critical field for as prepared GO and annealed GO samples (sample A, C and D) in 100 K and 300 K data. The saturation moment of the as prepared GO (sample A) at T = 100 and 300K are 1.75 × 10−2 emu/gm and 1.2 × 10−2 emu/gm respectively. The crossover field, i.e., the field at which magnetization crosses over to negative values, is 3 Tesla and 1.2 Tesla at temperatures of 100 and 300 K. Beyond the crossover field, the magnetization is linear in field, with a diamagnetic susceptibility of χdia = 8 × 10−7 emu/gm/Oe and 1.325 × 10−6 emu/gm/Oe for temperature 100 K and 300 K respectively. This amounts to a 40 % increase in the diamagnetic susceptibility going from T=100 K to 300K. Diamagnetic suceptibilty 3 is expected to come from the π electrons/carriers from the relatively undistorted planar regions (sp2 bondend) free from functional groups like O or OH. The increase in diamagetic susceptibilty with increase in temperature indicates a gap in the electronic spectrum i.e., GO is electrically insulating with a small gap. The magnetization data of sample B (got after chemically reducing sample A, as described in the text) is interesting. We notice that the magnetization do not saturate even at H=7 Tesla at 100 K. The magnetic moment at H=7 Tesla is 0.105 emu/gm. The ratio of positive saturation magnetization at T=100 K of sample B and sample A is about 6. This is large increase in the magnetization of graphene oxide upon chemical reduction. At T=300 K, the crossover field of sample B is ∼ 5 Tesla. The corresponding value of crossover field of sample A (at T=300 K) is 1.2 Tesla. So the crossover field at 100 K for sample B must be higher than the maximum field used in the experiment (7 Tesla). It should be also noticed that crossover field is higher in sample B (5 Tesla) compared to sample A (1.2 Tesla). The diamagnetic susceptibility at high field of sample B at T=300 K, is 1.285 × 10−6 emu/gm/Oe. This means a 50% increase in diamagnetic susceptibilty of GO after chemical reduction. The chemical reduction process gets rid of O and OH groups from graphene surfaces leading to release of many π electron carriers which were bonded to the radicals. Reduction process also leads to restoration of planar sp2 bonded regions. Both of these probably causes substantial increase in π electron diamagnetism. Since diamagnetism of RGO is more than GO, the higher crosover field of RGO compared to GO can come only because of much higher para/ferro-magnetic moments in RGO compared to GO ( a factor of 6 ). The relevant question here is how the reduction process of GO can lead to such increase of moment? Now let us compare this observations with that of sample C and D (which were obtained by heating sample A at 150 oC and 450 oC). The diamagnetic susceptibility (χdia) and saturation magnetization (Msat) has the following order, χdia(A) ≈ χdia(C) < χdia(D) and Msat(A) ≈ Msat(C) < Msat(D) . The crossover field progressively decrease from A to C to D. Summarizing : (1) With increase in annealing temperature (going from A to C and D) there is an increase of para/ferro-magnetic moment. (2) Along with this there is also an increase in diamagnetic susceptibility. The increase in diamagnetic susceptibility after annealing can be roughly understood as due to removal of many O and OH groups from the surface and consequest recovery/increase in area of sp2 bonded planar zones. Here again the main puzzle is the reason for increase in net para/ferro-magnetic moment after annealing. So both chemical reduction done at 4 low temperatures and annealing at high temperatures increase net magnetic moment. We also observe the saturation magnetic moment (the value of magnetization at H=7 Tesla) and the crossover field for the reduced sample (B) is higher compared to sample C and D. In other words chemical reduction is more effective compared to annealing in generating extra magnetic moment on graphene oxide. In fig. 3(a,b and c) we show hysteresis loops magnified around origin for GO (sample A) and RGO (sample B) taken at 10 K, 100 K and 300 K respectively. We could clearly observe that upon reduction of GO the moment increases. The coercive field decreases with increase in temperature for both the samples A and B. The reduction of coercivity with increase in temperature is an important result as this is oppose to what we have reported recently for single layer graphene structure. This important aspect will also be discussed in the section below. In fig. 4(a, b and c) we show for further clarity that RGO have more moments than GO for sample collected by filtering (fig. 4(a)). The cross-over in the case of GO happens at lower temperature and lower fields fig. 4(b and c) than RGO. In fig. 5 we show magnetic hysteresis curves for GO and RGO (sample E and F) collected and dried from the supernatant liquid. Since the sample is collected from the graphitic suspension, one can assume that the particle sizes are smaller and contains much less number of layers tending to few layers or even tending to single layer (graphene). We can clearly see from the hysteresis curves for this supernatant samples the coercivity on the other hand increases upon increase in temperature contrary to all the above samples (samples A to sample D). This we have reported recently in another set of sample [13]. This behaviour (sample E and F) that we have found is reproducible. We have also observed that in the case of supernatant samples the moments of GO sample is more than chemically reduced samples unlike the above samples A to D. In fig. 6 we show TGA curves for sample A, B and D. We see that the sample D have the least oxygen content i.e., observation of least weight loss as we increase the temperature. This indicates that the sample D is the most reduced sample. IV. DISCUSSION: Let us point out the main features observed in the result section above and they are as follows: 5 (We seperate the set of samples as two types ie., (1) sample filtered from the solution (type I) and (2) Samples collected from supernatant (type II)). We observe: (1) For the case of type I the magnetization/moment of as-such collected GO samples after filtering are "less" than chemically reduced or annealed samples. (2) For the case of type II the magnetization/moment of as-such collected GO samples from the supernatant are "higher" than chemically reduced samples. (3) For the case of type I samples the coercivity of the magnetic hysteresis curve "de- creases" upon increasing the temperature. This is the expected behaviour of sparsely distributed (weak interaction) single domain magnetic moments (4) For the case of type II samples the coercivity of the magnetic hysteresis curve "in- creases" upon increasing the temperature. This is highly unusual, but as pointed out earlier is a very reproducible result [12]. These opposing behaviour observed in these two type of sample is being reported for the first time to our knowledge. For our convenience we shall discuss the type II sample first and then the type I samples Recently we have proposed that for the case of single layer graphitic structure such as graphene oxide and graphene (type II structure) the magnetic property such as coercivity of hysteresis curve is governed by the inherent presence of ripple/wrinkles in these single layer sheets. It has been observed that these ripples/wrinkles diminishes upon heating and this diminishing of ripple/wrinkles as a function of increasing temperature was invoked in reference [12] to explain the increase in coercivity of magnetization as a function of temperature. For the type II samples GO has more paramagnetic moment compared to RGO [13]. The origin of moments could be [12] due to the presence of OH groups and other defects such as uncompensated sites, vacancies etc.,. It is natural to assume that the reduction process removes many such small moment carrying OH clusters from GO. Upon reduction many OH groups are removed and hence we can consider less clustering of OH groups and isolated OH groups are scattered all over the sheet. Since GO has more OH groups than RGO, the net magnetic moment of GO is higher than RGO. An OH can bind to C atom on any sublattice. If there is an local imbalance in the number of OH group attched to C atoms on sublattice A and sublattice B then there is a local moment 6 formation by Lieb's theorem [15]. We have to remember that the local spin correlation between nearest neighbour C atoms is antiferromagnetic in nature (superexchange) due to large unscreened coulomb repulsion between the π electrons. So a local level cluster of OH groups on graphene surface will lead to some uncompensated but large moment around such a region. This was proposed by us to be the reason for magnetism in GO (sample E). In the reduction process many such OH groups will be removed leading to removal of many moments around them. This process also increases the diamagnetism of the materials because of the excess diamagnetic susceptibilty of the π electron, which were earlier bonded to an OH group. This was the reason for reduction of paramagnetic moment in sample F. The magnetism of all samples is composed of three parts: (a) Backgroud diamagnetism of the π electrons, (b) Paramagnetism at all temperatures coming from small moments due to vacancy, dangling bonds or very small OH clusters on graphene surface and (c) There are also some resonably large single domain magnetic moments, presumably coming from either exposed zigzag edges or large patches of OH groups on the surface. It is (c) which are responsible and important for magnetic irreversibilty and hysteresis. In the reduction process many OH groups are removed and hence net magnetic moment decreases. At low temperatures these isolated moments randomly placed in an insulating matrix with very little interaction between them due to highly crumpled structure behaves like paramagnets. The point number (2) and (4) mentioned above in the begining of this section has been explained in detail in reference [12] when the graphitic structure is like graphene oxide or graphene. The curious increase of coercivity with increase in temperature was argued by us as due to reduced average crumpling on single or few layer graphene oxide at higher temperature. Less crumpling leads to higher locally averaged magnetic anisotropy energy of such single domain large moments. Since coercivity of such materials is proportional to the average anisotropy energy of the single domain moments, the coercivity increases with increase in temperatures. We also pointed out the insulating nature of such materials with small gap ensuring an increase in number of carriers with increase in temperature. This liberated carriers can also bring in long range ferromagnetic correlation and coercivity can increase because of that. Now what happens to magnetization when we do not have single or few layer graphitic powder but many layered graphitic powder tending towards the bulk graphitic structure. First let us address the decrease in the coercivity in the hysteresis curve upon increasing the 7 temperature. This is an usual behaviour expected in Stoner - Wolfarth model [16], which applies to a collection of single domain non-interacting magnetic moments. We have recently shown that when the graphene is loosely held on HOPG, then there exist ripple and upon strong adherence of graphene with the bulk HOPG substrate the ripple vanishes [17]. Thus multilayer graphitic structure will not have any ripple formation like one generally observe in a single layer graphitic structure. Thus the many layer graphitic structure will behave more like sparsely distributed nanomagnetic system. That is why the coercivity has the usual decrease with increase in temperatures. Observation of increase in magnetization value upon chemical reduction or annealing at high temperature in air for the case of multilayer graphitic structure (sample B, C and D) is very interesting. Recently many effort has been made theoretically and experimentally to identify the structure of GO and it is now believed that the oxygen and the hydroxyl group are bonded to graphene in a definite fashion such that upon reduction or annealing (ie., removal of oxygen or hydroxyl) leads to unzipping of the graphene structure leading to frag- mentation of the graphene layers leaving many zigzag edges being exposed. It appears from our experimental observation that unzipping is more prominent for the case of multilayered graphitic structure than single layered structure and this is mainly due to development of strain upon reduction/unzipping in multilayered structure and the strain for the case of single layer is less than multilayer which may be obvious due to the nature of the structure (i.e., for single layer the strain developed across the sample is less than in multilayered sam- ples). These development of strain and unzipping of the layers leads to heavy fragmention of the bulk graphitic structure to large smaller fragments exposing large number of zig-zag edges. If we invoke the occurrence of magnetization due to zig-zag edges [18], then we can explain the increase of magnetization upon chemical reduction or annealing in air. We are working on this aspect of unzipping and shall be reporting elsewhere but however many prilimainary works have been reported by other [19 -- 21]. At present our discussion can only be speculative, but more theory and experiments are looking for ordered local arrangement of O and OH groups on graphene. There could be local low energy structures, where OH group is attached to a chain of C atoms. The OH groups are above and below the graphene surface over carbon atoms at A and B sublattices. For single layer GO, all such OH groups on both sides will be randomly removed, and it is unlikely to create a local imbalance in A and B sublattice sites to which OH is attached. For single or few layer graphene oxide, 8 when reduced there is only minute increase/decrease of moment. But a multilayer graphene when reduced may have uneuqal reduction on different sides of graphene surface. This will create a large extra moment. Another interesting observation by us is that the low field magnetization of sample D is more than sample B, whereas high field magnetization of B is higher than sample D. Since upon annealing at 450 oC, the graphitic flakes disintegrates into smaller sizes and hence the size of the graphitic flakes in sample D are much smaller that in sample A or sample B. This leads to many small segment of edges carrying moments. Hence, the low field magnetization is higher in sample D than sample B. Sample B has many large sized moments also, but they are blocked at low field. At high field these large moments align along the external field giving higher magnetization in sample B (chemically reduced sample). In other words, the moments at the edges align at low fields (sample D) and the moments due to patches in chemically reduced samples (sample B) align at higher fields. Finally we point out again that the crossover of magnetization towards negative values happens over a very small field interval. This almost discontinuous change of magnetiza- tion, looks like some kind of magnetic transition. More work is needed to understand this phenomena. V. CONCLUSION: In conclusion, we have shown for the first time that for "single or few layer" graphitic oxide (GO) and reduced graphitic oxide (RGO) the magnetization is larger for GO than RGO and for both these samples the magnetic coercivity increases with increase in temperature. Whereas, for multilayer GO or RGO the above two properties are opposite i.e., RGO have larger mangetization than GO and the magnitude of magnetic coercivity decrease with increase in temperature. [1] H. C. Neto, F. Guinea, N. M. R. Peres, K. S. Novoselov and A. K. Geim; Rev. Mod. Phys. 81, 109 (2009). A.K.Geim and K. S. Novoselov; Nature Mater. 6, 183 (2007). A. K. Geim and A. H. MacDonald; Phys. Today 608, 35(2007). 9 [2] S. Banerjee, M. Sardar, N. Gayathri, A. K. Tyagi, and Baldev Raj, Phys. Rev. B 72, 075418 (2005); S. Banerjee, M. Sardar, N. Gayathri, A. K. Tyagi, and Baldev Raj, Appl. Phys. Lett.88, 062111 (2006) [3] J. W. Suk, R. D. Piner, J. An, and R. S. Ruoff, ACS Nano, 4(11), 65576564 (2010) [4] I. W. Fran, D. M. Tanenbaum, A. M. vanderZande and P. L. McEuen, J. Vac. Sci. Technol. B 25(6), 2558(2007): A. R. Ranjbartoreh, B. Wang, X. Shen and G. Wang, J. Appl. Phys. 109, 014306(2011) [5] A. A. Balandin, S. Ghosh, Wenzhong Bao, I. Calizo, D. Teweldebrhan, F. Miao and C. N. Lau, Nano Lett., 8, 3(2008); K. Saito, J. Nakamura and A. Natori, Phys. ReV.B, 76, 115409(2007). [6] M. Sepioni, R. R. Nair, S. Rablen, J. Narayanan, F. Tuna, R. Winpenny, A. K. Geim and I. V. Grigorieva; Phys. Rev. Lett. 105 207205 (2010). [7] D. Jianga, B. G. Sumpter and S. Dai, J. Chem. Phys., 127, 124703(2007). [8] H. S. S. Ramakrishna Matte, K. S. Subrahmanyam, and C. N. R Rao,J. Phys. Chem. C 113 9982 (2009). [9] P.Esquinazi, D. Spemann, R. Ho hne, A. Setzer, K. H. Han, T. Butz; Phys. ReV. Lett. 91, 227201(2003). [10] Y. Wang, Y. Huang, Y. Song, X. Zhang, Y. Ma, J. Liang and Y. Chen; Nano Letters 9, 1, 220-224(2009). [11] O.V. Yazyev and L. Helm; Phys. Rev. B 75 125408 (2007), O.V. Yazyev, Rep. Prog. Phys. 73 056501 (2010) [12] K. Bagani, A. Bhattacharya, J. Kaur, A. Rai Chowdhury, B. Ghosh, M. Sardar and S. Baner- jee, arXiv:1302.3336v1 [cond-mat.mtrl-sci] 14 Feb 2013 [13] J. Kaur, A. Rai Chowdhury A. Bhattacharya, B. Ghosh, M. Sardar and S. Banerjee, AIP Conf. Proc. 1447, 1233 (2012); doi: 10.1063/1.4710457 [14] W. Hummers, R. Offeman; J. Am. Chem. Soc. 1958, 80, 1339 [15] E. H. Lieb, Phys. Rev. Lett. 62, 1201 (1989). [16] E. C. Stoner and E. P. Wohlfarth, Phil. Trans. R. Soc. Lond. A 240 599 (1948) [17] S. Panigrahi, A. Bhattacharya, S. Banerjee and D. Bhattacharyya, J.Phys.Chem.C 116, 4374- 4379(2012). [18] H. Kumazaki and D. S. Hirashima; J. Phys. Soc. Jpn., 77(4), 044705(2008), S. Bhowmick and V. B. Shenoy, J. Chem. Phys., 128, 244717(2008), J. Jung and A. H. MacDonald; Phys. Rev. 10 B 79, 235433 (2009). [19] D. Pan, J. Zhang, Z. Li, and M. Wu, Adv. Mater. 22, 734738(2010) [20] J. Li, K. N. Kudin, M. J. McAllister, R. K. Prudhomme, I. A. Aksay and R. Car, Phys. Rev. Lett, 96, 176101(2006) [21] T. Sun and S. Fabris, Nano Lett., 12, 1721(2012). 11 Figure Captions Fig. 1 ZFC and FC magnetization versus temperature for sample A, B, C and D and the appropriate samples are labelled in the figure. Fig. 2 Magnetization versus the applied field taken at (a) 10 K, (b) 100 K and (c) 300 K for GO and RGO samples and we observe a distinctly sharp cross-over of magnetization from positive to negative value and vice versa in the negative direction. Fig. 3 Hysteresis loops magnified around origin for GO (sample A) and RGO (sample B) taken at (a) 10 K, (b) 100 K and (c) 300 K respectively. Fig. 4 For further clarity to show that RGO have more moments than GO for sample collected by filtering. Fig. 5 Magnetic hysteresis curves for GO and RGO (sample E and F) collected and dried from the supernatant liquid. Fig. 6 TGA curves for sample A, B and D 12 Fig. 2(a) Fig. 1 Fig. 2(b) Fig. 2(c) Fig. 3(a) Fig. 3(b) Fig. 3(c) Fig. 4(a) Fig. 4(b) Fig. 4(c) Fig. 6 Fig. 5 

1911.00148

1

1911

A reduced-order modeling approach for electron transport in molecular junctions

To describe non-equilibrium transport processes in a quantum device with infinite baths, we propose to formulate the problems as a reduced-order problem. Starting with the Liouville-von Neumann equation for the density-matrix, the reduced-order technique yields a finite system with open boundary conditions. We show that with appropriate choices of subspaces, the reduced model can be obtained systematically from the Petrov-Galerkin projection. The self-energy associated with the bath emerges naturally. The results from the numerical experiments indicate that the reduced models are able to capture both the transient and steady states.

cond-mat.mes-hall

cond-mat

A reduced-order modeling approach for electron transport in molecular junctions Weiqi Chu∗,†,‡ and Xiantao Li∗,‡ †Department of Mathematics, University of California, Los Angeles, Los Angeles, CA ‡Department of Mathematics, Pennsylvania State University, University Park, PA E-mail: [email protected]; [email protected] Abstract To describe non-equilibrium transport processes in a quantum device with infinite baths, we propose to formulate the problems as a reduced-order problem. Starting with the Liouville-von Neumann equation for the density-matrix, the reduced-order technique yields a finite system with open boundary conditions. We show that with appropriate choices of subspaces, the reduced model can be obtained systematically from the Petrov-Galerkin projection. The self-energy associated with the bath emerges naturally. The results from the numerical experiments indicate that the reduced models are able to capture both the transient and steady states. 1 Introduction In the past decades, there has been significant progress in the investigation of molecular electronics and quantum mechanical transport, 1 -- 3 one emerging issue among which is the modeling of interfaces or junctions between molecular entities. 4 -- 7 The junctions encompass two sections: (i) a molecular core at the nanometer scale that bridges two metallic devices; 1 (ii) the surrounding areas from contacting materials. Notable examples include quantum dots, quantum wires, and molecule-lead conjunctions. The junctions play an essential role in determining the functionality and properties of the entire device and structure, such as photovoltaic cells, 8,9 intramolecular vibrational relaxation, 10 -- 13 infrared chromophore spec- troscopy, and photochemistry. 14 -- 17 At such a small spatial and temporal scale, modeling the transport properties and processes demands a quantum theory that directly targets the electronic structures. Such problems have been traditionally treated with the Landauer-Buttiker formalism, 18 -- 20 which aims at computing the steady-state of a system interacting with two or more macro- scopic electrodes, and the non-equilibrium Green's function (NEGF) approach, which, often based on the tight-binding (TB) representation, can naturally incorporate the external po- tential and predict the steady-state current. 21 This approach was later extended to the first-principle level 22 -- 24 using the density-functional theory (DFT). 25,26 Due to the dynamic nature and the involvement of electron excitations, one natural computational framework for transport problems is the time-dependent density-functional theory (TDDFT), 27 -- 31 which extends the DFT to model electron dynamics. This effort was initiated by Stefanucci and Almbladh, 29,32 and Kurth et al., 27 where the wave functions are projected into the center and bath regions. An algorithm was developed to propagate the wave functions confined to the center region so that the influence from the bath is taken into account. This is later treated by using the complex absorbing potential (CAP) method 33 by Varga. 34 One computational challenge from this framework is the computation of the initial eigenstates. Kurth et al. 27 addressed this issue by diagonalizing the Green's function. However, the normalization is still nontrivial, since the wave functions also have components in the bath regions. Another issue is that the CAP method is usually developed for constant external potentials. For time-dependent scalar potentials, a gauge transformation is usually needed to express the absorbing boundary condition, 35 and it is not yet clear how this can be implemented within CAP. 2 Another framework is based on the Liouville-von Neumann (LvN) equation 36,37 to com- pute the density-matrix operator directly. One advantage of the LvN approach is that the initial density-matrix can be obtained quite easily from the Green's function. Therefore diagonalization and normalization are not needed. To incorporate the influence of the bath, the LvN equation has been modified by adding a driving term at the contact regions accord- ing to the potential bias. This approach was later extended by Zelovich and coworkers, 38,39 which is again motivated by the CAP method. Despite the heuristic derivation, 38 these methods are still empirical in modeling the electron transport problem. In particular, the steady state and transient predicted by the driven LvN equation have not been compared with those from the full model. This paper follows the density-matrix-based framework. Rather than using the approach by S´anchez et al., 36 we derive the open quantum system using the reduced-order techniques that have been widely successful in many engineering applications. 40 -- 42 We first formulate the full quantum system as a large-dimensional dynamical system with low-dimensional in- put and output. This motivates a subspace projection approach, which has been the most robust method in reduced-order modeling. 40,41 In particular, we employ the Petrov-Galerkin projection, a standard tool in numerical computations, e.g., linear systems, eigenvalue prob- lems, matrix equations, and partial differential equations (PDEs). 43 -- 46 With appropriate choices of the subspaces, we obtain a reduced LvN equation, modeling an open quantum system where the computational domain only consists of the center and contact regions. We illustrate the procedure for a one-dimensional model system, as a first step to treat more realistic systems. The numerical results have shown that the reduced LvN equations can capture both the transient and the steady state solutions. The rest of the paper is organized as follows. In Sec. 2, we provide a detailed account of our methodology, including the mathematical framework and the derivation of the reduced models. In Sec. 3, we present results from some numerical experiments to examine the effectiveness of the derived models. Sec. 4 summarizes the methodology and provides an 3 outlook of future works. 2 Methods and algorithms 2.1 The density-matrix formulation Following the conventions from existing literature, 24,27,38,39 we consider a molecular junction, where a molecule is connected to two semi-infinite leads. More specifically, the physical domain for the entire system is denoted by Ω, divided into three parts, ΩL, ΩC, and ΩR representing respectively the left lead, the center region, and the right lead, as illustrated in Figure 1. Left Lead Extended Molecule (Center) Right Lead Figure 1: (Color online) Schematic representation of a two semi-infinite lead junction model consisting of two semi-infinite leads: left lead (L), right lead (R), and an extended molecule (C) in the center. We will start with the LvN equation, which for molecular conduction problems, has been proposed and implemented in a series of papers. 36 -- 39 The LvN equation governs the dynamics of the density-matrix operator ρ, which can be connected to the wave functions (e.g., the Kohn-Sham orbitals) as follows, ρ(r, r ′, t) =Xj nj ψj(r, t) ψj(r ′, t)∗, (1) with nj being the occupation numbers. The equation can be derived from a time-dependent Schrodinger equation (TDSE), and for the entire system Ω, it can be written as, i∂t ρ(t) = H(t)ρ(t) − ρ(t) H(t) = [ H(t), ρ(t)]. (2) 4 Here the bracket is the usual quantum commutator, which we will generalize as follows, [ A, B] := A∗ B − B ∗ A. (3) Here A∗ denotes the conjugate transpose (or Hermitian transpose of A). Notice that with this generalization, A or B can be non-Hermitian. Our goal is to derive an open quantum system for the density-matrix at the center region ΩC, where the influence from the leads is implicitly incorporated. For convenience, we first assume that the entire system (2) has been appropriately discretized in Ω so that ρ(r, r ′, t) is a matrix defined at certain grid points, here denoted by Ω∆ with ∆ indicating the grid size. Namely, ρ(r, r ′, t) is the density-matrix with r, r ′ ∈ Ω∆. This can be obtained by using a finite-difference scheme, especially in real-space methods. 47 As a result, one arrives at a matrix-valued infinite-dimensional system, and hence we will drop the notation from now on. A similar system can also be obtained using the TB approximation, where the wave functions are projected to atomic-centered orbitals, in which case, the LvN equation would contain the overlap matrix on the left hand side when the basis functions are not orthogonal. 39,48 However, it would not affect our following reduction method. Following the setup by Cini, 21 we treat the problem as an initial value problem (IVP), starting with an initial density ρ0 = feq(µ−H0) as an equilibrium density at t = 0. Such setup is particularly amenable for numerical computations. While it is challenging to compute the wave function in a subdomain, which in general requires solving nonlinear eigenvalue problems and normalization, 49 efficient algorithms are available to calculate the density- matrix in a sub-domain. 50 -- 52 These algorithms take advantage of the relation between the density-matrix and the Green's function, ρ = 1 2πiIC G(z)dz, G(z) = (zI − H)−1, (4) where the contour encloses all the occupied states. The restrictions of the density-matrix to a 5 finite subdomain can be obtained by E ∗ρE, where the operator E, with proper arrangement, can be written simply as E ∗ = [I, 0], with the identity operator I corresponding to the subdomain and the zero matrix corresponding to the exterior (bath). This observation, together with (4), reduces the problem to the computation of the following expression that we have slightly generalized the linear algebraic system to, [× 0](zI − H)−1  × 0   , (5) where the left and right vectors have finite supports. Although this amounts to solving an infinite-dimensional linear system, a finite number of unknowns are needed due to the multiplication by the sparse vector on the left and right. For one-dimensional (or quasi one- dimensional) systems, an iterative scheme can be used 53,54 to invert the block tri-diagonal matrix. For multi-dimensional problems, a discrete boundary element method 55 can be used. 56 We will refer to these algorithms in general as selective inversion. 51 Although our model works with the density-matrix, our primary interest is in the electric current induced by a time-dependent external potential that is switched on at t = 0+. Similar to the theory of linear response, 57 -- 59 we consider H(t) as a deviation from its initial value H0 and write H(t) = H0 + δH(t) with δH(t) being the applied potential from the leads. The response of the system due to the external potential could be represented in terms of the perturbed density, δρ(t) := ρ(t) − ρ0, δρ(0) = 0, which satisfies a response equation, i d dt δρ(t) = [H(t), δρ(t)] + Θ(t). (6) (7) Here Θ(t) = [δH(t), ρ0] is a non-homogeneous term that incorporates the influence from the external potential. 6 As is customary, 27,36,38,60 we neglect the direct coupling between the two leads and par- tition the density-matrix and the Hamiltonian operator in accordance with the partition of the domain indicated in Figure 1. In this case, Eq (7) translates to i d dt   δρLL δρLC δρLR δρCL δρCC δρCR δρRL δρRC δρRR   =     HLL HLC 0 HCL HCC HCR 0 HRC HRR   ,   δρLL δρLC δρLR δρCL δρCC δρCR δρRL δρRC δρRR     + Θ. (8) We are interested in the case when δH corresponds to scalar potentials in the leads, given by UL(t) and UR(t). Then the matrix function Θ(t) can be written as, Θ(t) =   0 UL(t)ρLC(0) (UL(t) − UR(t))ρLR(0) −UL(t)ρCL(0) 0 −UR(t)ρCR(0) −(UL(t) − UR(t))ρRL(0) UR(t)ρRC(0) 0   . (9) In practice, to mimic the infinite leads, one has to pick much larger regions ΩL/R to prevent the finite size effect, e.g., a recurrence. This makes a direct implementation using Eq (8) impractical and requires model reduction tools to reduce the complexity of the full problem. There are six unknown blocks in the density-matrix δρ : the blocks δρLL (and δρRR) are semi-infinite, and this is where an appropriate reduction is needed. It suffices to illustrate the reduction of the degrees of freedom in the left bath. A direct computation yields i d dt δρLL(t) = [HLL(t), δρLL(t)] + FL(t), (10) where HLL(t) = HLL(0) + δH LL(t) and δH LL(t) is the external potential imposed on the left lead. FL(t) represents the influence from the interior and can be extracted from (8), FL(t) = HLCδρCL(t) − δρLC(t)HCL + ΘLL(t). (11) 7 Now our key observation is that Eqs (10) and (11) constitute an infinite-dimensional control problem with control variables δρCL and output δρLL. In practice, only the entries in δρLL near the interface (between ΩL and ΩC) are needed. Such a large-dimensional dynamical system with low-dimensional input and output can be effectively treated by using the reduced-order techniques. 40,41,61 -- 63 2.2 General Petrov-Galerkin projection methods Motivated by the development of reduced-order modeling techniques 61,62,64 that have been widely used in control problems, 42 circuit simulation, 41 and microelectromechanical sys- tems, 63 etc., we propose a Petrov-Galerkin projection approach to derive a reduced model from the infinite-dimensional LvN Eq (10). The objective is to provide a reduced dynamics for the device region that captures both the transient and the steady state. The first ingredient is to pick an appropriate subspace where the approximate solution is sought. To start with, we pick an n-dimensional subspace VL spanned by a group of basis functions {ϕi}n i=1. The subspace can be expressed in a matrix form as VL = [ϕ1, ϕ2, · · · , ϕn]: VL = Range(VL). Throughout this paper, we will not distinguish a subspace VL and its matrix representation VL. In practice, the basis functions can be standard hat functions centered at certain grid points, as shown in Figure 2, or Gaussian-like functions that mimic atomic orbitals. ϕn · · · ϕ5 ϕ1ϕ2ϕ3ϕ4 Figure 2: A diagram of hat functions on ΩL that span a subspace VL with dimension n. ΩL With the subspace set up, one can seek a low-rank approximation of δρLL(t) as δeρLL in 8 the following form, where the n×n matrix DLL(t) represents the nodal values. This representation automatically δeρLL(t) := VLDLL(t)V ∗ L , (12) guarantees that the resulting density-matrix is Hermitian and semi positive-definite, as long as DLL has those properties. The residual error from this approximation can be directly deduced from the LvN equation (10) by subtraction, E(DLL, t) = iVL d dt DLL(t)V ∗ L − [H(t), VLDLL(t)V ∗ L ] − FL(t). (13) The second ingredient to determine DLL is by projecting the residual error to the orthogonal complement of a test subspace, WL, spanned by the columns of WL, that is W ∗ LE(DLL)WL = 0. (14) This yields a finite-dimensional system, and the reduction procedure described above is known in general as the Petrov-Galerkin projection, which has been a classical numerical method in the solutions of differential equations, 65 order-reduction problems, 40,41 and matrix equations. 66,67 The reduced equation from the Petrov-Galerkin projection Eqs (12) to (14) can be written as, i d dt DLL(t) = [eHLLML, DLL] − eFL(t), where the matrices are given by ML = (V ∗ L WL)−1 , eHLL(t) = V ∗ eFL(t) = M ∗ L HLL(t)WL, LW ∗ LFL(t)WLML. 9 (15) (16) Notice that in (15) we have used the generalized notation of commutators (3). At this point, we will keep the subspaces spanned by VL and WL at the abstract level, and the specific choices will be discussed in the next section. The same model reduction procedure can be applied to the right lead and it yields a similar finite-dimensional equation, i d dt DRR(t) = [eHRRMR, DRR] − eFR(t). (17) Eqs (15) and (17) are related by the non-homogeneous terms eFα(t), α = L, R that involve the evolution of δρCα and their Hermitian transpose. In the center region, no reduction is needed and we will retain this part of Eq (8). Therefore, we can construct a Petrov-Galerkin projection for the entire system, by gluing the subspaces as follows, V =   0 0 VL 0 InC 0 0 0 VR   , W =   WL 0 InC 0 0 0 0 0 WR We seek an approximate solution δρ(t) ≈ δeρ(t) := V D(t)V ∗, for the projected dynamics of Eq (7), such that, i d dt W ∗δeρ(t)W = W ∗(cid:0)[H(t), δeρ(t)] + Θ(t)(cid:1)W. Direct computations yield, i d dt D(t) = [Heff, D] + eΘ(t), 10   . (18) (19) (20) (21) where Heff is the reduced Hamiltonian, V ∗ L HLLWL (V ∗ L WL)−1 V ∗ L WL)−1 L HLC HCLWL (V ∗ HCC HCRWR (V ∗ 0 V ∗ RHRC V ∗ RHRRWR (V ∗ 0 RWR)−1 RWR)−1 Heff =   and eΘ(t) is given by eΘ(t) = M ∗W ∗Θ(t)W M = M ∗W ∗[δH(t), ρ0]W M. Here the matrix M is block-diagonal, M =   (V ∗ L WL)−1 0 0 0 InC 0 0 0 (V ∗ RWR)−1   .   , (22) (23) (24) It is worthwhile to point out that the subspaces can also be time-dependent. This offers the flexibility to pick subspaces that evolve in time. It should also be emphasized that our discussions regarding the Petrov-Galerkin projection is suitable for general cases and not limited to one-dimensional junction models, i.e., the typical lead-molecule-lead structures. With appropriate domain decomposition, it can be applied to high-dimensional systems with more general device structures. 2.3 The selection of the subspaces In this section, we discuss specific choices of the subspaces in the Galerkin-Petrov projection. Without loss of generality, we again start by considering the left lead ΩL. Let ΩΓL ⊂ ΩL be a subdomain in the left lead that is adjacent to the center region, as shown in Figure 3. ΩΓL and ΩΓR are often referred to as contact regions that have direct coupling with the interior. 52,68 In our case, we pick ΩΓR in such a way that the remaining component in the 11 lead has no coupling with the center region, i.e., Hi,j = 0 for i ∈ ΩC amd j ∈ ΩR − ΩΓR. This imposes a lower bound on the size of the contact region. Left Contact Region Extended Molecule (Center) Right Contact Region Figure 3: (Color online) A schematic representation of junction model with contact regions in green. In reduced-order modeling problems, the subspaces are often chosen based on how the input/control variables enter the large-dimensional system, e.g., see the review papers. 40,41 In our setting, we consider the dynamics in the left lead, given the density-matrix in the contact region. So we pick the basis VL so that V ∗ L acts as a restriction operator from ΩL to ΩΓL, V ∗ L = [0, InΓ,L], (25) where InΓ,L is an identity matrix with the dimension nΓ,L being the number of grid points in ΩΓL. The same procedure can be applied to the other lead region. When the subspaces are combined (cf. Eq (18)), we have, V =   0 InΓ,L 0 0 0 0 0 InC 0 0 0, 0 0 InΓ,R 0   . (26) The entire density-matrix is approximated as in (19). It is now clear that V is a restriction operator to an extended center domain, Ω eC = ΩΓL ∪ ΩC ∪ ΩΓR. Consequently, D in Eq (21) 12 becomes the density-matrix in eC, D(t) = δeρ(t)Ω eC . ×Ω eC (27) It remains to choose the subspaces WL/R. Motivated by the Green's function approach for quantum transport, 69 -- 71 we consider the test space, WL(ε) = (εI − HLL)−1VL, (28) where ε ∈ C is in the resolvent space of the Hamiltonian HLL. We require that Im(cid:0)ε(cid:1) < 0 to ensure the stability of the reduced models. In this case, it corresponds to the advanced Green's function as the imaginary part of ε goes to zero, lim Im(ε)→0− WL(ε) = GA L(ε)VL. (29) The selection of WR is similar. Intuitively, the subspace W obtained this way represents the solution of the corresponding TDSE with initial conditions supported in the extended device region eC. Combining the subspaces WL and WR, we have   WL W = . (30) 0 0 0 0 InC 0 0 0 WR 0 0 0   We notice in passing that unlike the basis VL amd VR, the basis WL and WR do not have compact support. We now examine the specific form of the reduced model (21). With the specific choices 13 of the subspaces (Eqs (26) and (30)), one can simplify the matrix M in Eq (24) as follows, MLL = (V ∗ L WL)−1 = εI − HΓL,ΓL(t) − ΣL(t, ε) =: εI − Heff,L(t, ε), (31) and similarly, MRR = εI − Heff,R(t, ε). Here Σα is the self energy 24,72 -- 74 contributed by the left (α = L) or right (α = R) lead, Σα(t, ε) = HΓα,α(εI − Hα,α(t))−1Hα,Γα, and Heff,α is the effective Hamiltonian associated with ΩΓα, 75 Heff,α(t, ε) = HΓα,Γα(t) + Σα(t, ε). Overall, the effective Hamiltonian Heff in (21) is simplified to, Heff(t) := Hc(t) + Σ(t, ε), where Hc is the Hamiltonian restricted in the extended center region Ω eC, Hc(t) := H(t)Ω eC ×Ω eC , and Σ is a block-wise diagonal matrix that incorporates the self-energies of two leads, ΣL(t, ε) 0 Σ(t, ε) =   0 0   . 0 0 0 0 ΣR(t, ε) (32) (33) (34) (35) (36) (37) The self-energy (33) involves the inverse of a large-dimensional (or infinite-dimensional) 14 matrix. Similar to the inversion in (5), it can be efficiently computed using a recursive algorithm, which has been well documented. 52,76,77 The self-energy only needs to be computed once for constant external potential and for periodic external potentials, it can be pre- computed for one period. Let ρc be the density-matrix restricted in the extended center region Ω eC, i.e., ρc(t) := ρ(t)Ω eC ×Ω eC = D(t) + ρc(0). The reduced model for this part of the density-matrix can now be written as, i d dt ρc(t) = [Heff(t), ρc(t)] + Θc(t), (38) (39) With our choice of the subspaces, the reduced dynamics is driven by the effective Hamil- tonian Heff. The non-homogeneous term Θc embodies the effect of the potential, Θc(t) = M ∗eV ∗ (ε∗I − H)−1 Θ(t) (εI − H)−1 eV M, where M is computed from Eq (31) and eV is in the form of   eV = VL −HLC 0 −HCΓLV ∗ L (ε − HLL)VL ε − HCC −HCΓRV ∗ R(ε − HRR)VR 0 −HRC VR (40) (41)   . The practical implementation of the reduced model hinges on the availability of efficient algorithms to compute (i) the self-energy (33); (ii) the initial density-matrix in the center and contact region; and (iii) the non-homogeneous term (40). The computation of the self-energy and the initial density-matrix, as previously discussed, can be computed using the selective inversion techniques, which is applicable for problems that can be cast into the form of (5) where the Green's function is accompanied by sparse vectors. As for the non-homogenous 15 term, we find that Vα and Hα,C, α = L, R have non-zeros elements only associated with those degrees of freedom in the domain eΩ, which implies the sparsity of eV . Upon closer inspection, we find that the product of inverse matrices in Θ(t), i.e., (ε∗I − H)−1 Θ(t) (εI − H)−1, can be written as a sum of single matrix inverses (partial fractions), provided that ε is in the resolvent of H and Im(ε) 6= 0. For example, we have, (zI − H)−1 (εI − H)−1 = 1 ε − z(cid:16) (zI − H)−1 − (εI − H)−1(cid:17). Consequently, all those blocks can be written in the general form (5), and one compute Θc efficiently by using the selective inversion techniques. 51 2.4 Properties of the reduced models 2.4.1 The Hermitian property of ρc(t) The projection method produces an approximation of the density-matrix in the extended center region, leading to an open quantum-mechanical model that can be subsequently used to predict the current. The influence from the infinite leads, through the self-energy, has been implicitly incorporated into the effective Hamiltonian. By taking the Hermitian of the reduced model (39), and noticing the anti-Hermitian property of the term eΘ, we find that c(0). As ρc(0) is Hermitian, and in light of the ρ∗ c also satisfies (39) with initial condition ρ∗ uniqueness of the solution, we obtain the Hermitian property for ρc(t). 2.4.2 The stability of the reduced models Next, let us turn to the analysis of stability. Since the stability of linear non-homogeneous system is implied by the stability of homogeneous system, we focus on the homogeneous case in Eq (39) to study its stability. The problem can be addressed as the stability of a finite system X(t), i d dt X(t) = A(t)X(t) − X(t)A∗(t), X(0) = ρc(0), (42) 16 where A = Hc + Σ∗. Since ρc(0) has an eigen-decomposition ρc(0) = Pℓ nℓψ0 difficult to verify that X(t) =Pℓ nℓψℓ(t)ψ∗ ℓ (t) is the solution of Eq (42) if ψℓ(t) satisfies ℓ ψ0∗ ℓ , it is not i d dt ψℓ(t) = A(t)ψℓ(t), ψℓ(0) = ψ0 ℓ . (43) It suffices to analyze the stability of Eq (43). There exists a decomposition A(t) = A1(t) + iA2(t), where A1, A2 are real-valued sym- metric matrices and A2 is determined from Σ due to the Hermitian property of Hc. Further computation yields, A2(t) =   L 0 eΦLΛL(t)eΦ∗ 0 0 0 0 0 0 eΦRΛR(t)eΦ∗ R   , (44) where eΦα = HΓα,αΦα and Φα is the eigenvectors of Hα,α. Thanks to the special form of Σ, one can compute that Λα is a real diagonal matrix, in the form, Λα = diag(λα 2 , · · · , λα 1 , λα n), with ℓ = Im(cid:18) 1 λα ε∗ − µα ℓ(cid:19) , (45) where µα ℓ is the eigenvalue of Hα,α. To ensure the stability, it is enough to require that A2 has only non-positive eigenvalues, 78 i.e., ℓ = Im(cid:18) 1 ε∗ − µα ℓ(cid:19) = λα Im(ε) ε∗ − µα ℓ 2 ≤ 0. (46) This confirms that when ε has negative imaginary part, the stability of (39) is guaranteed. 2.5 Higher order subspace projections The Galerkin-Petrov projection method can be extended to higher order, by expanding the subspaces VL/R and WL/R to higher dimensions. Here we provide two options to extend the 17 current subspaces. Expanding the contact region. One straightforward approach is to keep the choices of V and W according to (26) and (30), but increase the size of the region ΩΓ to increase the subspace. Through numerical tests, we observe that this is a rather simple alternative, and it captures steady state current with subspaces of relatively small dimensions nΓ. Block Krylov subspaces. Another approach, as motivated by the block Krylov tech- niques 79 for large-dimensional dynamical systems, is to expand the subspace VL to the block Krylov subspace, VL,m =(cid:2)VL HLLVL · · · H m−1 LL VL(cid:3) =: Km (HLL; VL) . The corresponding WL,m has a similar structure, WL,m =(cid:2)WL VL · · · H m−2 LL VL(cid:3) =: Km (HLL; WL) . (47) (48) The Krylov subspaces are composed of a generating matrix and a starting block. In order to keep the additional blocks full rank, we pick VL based on the interaction range in HLL. For example, if HLL is based on a one-dimensional nearest-neighbor Hamiltonian, then we pick nΓ = 1 to define VL, which would be a one-dimensional vector; We pick nΓ = 2 for a next nearest neighbor Hamiltonian, etc. 3 Numerical Experiments and Discussions To test the reduction method, we consider a one-dimensional two-lead molecular junction model within a TB setting. We follow the setup in Zelovich et al. 38 More specifically, in the computation, the leads are represented by two finite atomic chains with increasing lengths ( nL and nR respectively) to mimic an infinite dimensional system and eliminate the finite size effect. The extended molecule with length nC is represented by a finite atomic chain 18 coupled with both leads. Here, the atomic unit is used throughout the paper if not stated otherwise. Initially, the system is configured in thermodynamic equilibrium, with all single-particle levels occupied up to the Fermi energy εF = 0.3. The on-site energy is taken as α = 2, and the hopping integral between nearest neighbors is β = −1. At time t = 0+, a bias potential is switched on in the electrodes. With the computed density-matrix, we study the bond current through the molecular junction to monitor the dynamics, using the formula 38 I(t) = 2βIm[ρj,j+1(t)]. (49) For the time propagation of the density-matrix, we use the fourth-order Runge-Kutta scheme to solve the full model (2), as well as (39). We fix the size of the center region nC = 20 and simulate the system under two different types of external potentials: (1) constant biased potential: UL/R = ∓ δU 2 to mimic direct current (DC) circuit; (2) time-dependent potential: A sinusoidal signal in the left lead, UL = sin ωt, to mimic an alternating current (AC). In principle, the bath size needs to be infinite to model the two semi-infinite leads; but in computations, one can only treat a system of finite-size and expect the system to reach a steady state in the limit as the bath size goes to infinity. First, we examine such size effect by varying nL/nR and observing the current in the center region. More specifically, we run direct simulations using nL = nR = 200, 500, 1000, 2000. Our results (Figure 4) suggest that, for the constant potential case, the electric current gradually develops into a steady state until the propagating electronic waves reach the ends of the leads and get reflected toward the bridge. As we extend the leads size to nL = nR = 1000, the backscattering effect occurs much later and is no longer observed within the time window of our simulation. For the dynamic potential case, we observe periodic changes of the electric current. Size effects become insignificant when the size is increased to nL = nR = 500 over the duration of the simulation. We point out that this effort of using sufficiently large bath size is only 19 to generate a faithful result from the full model (2), to examine the accuracy of the reduced model (39). 0.1 0 t n e r r u C −0.1 0 0.1 0 t n e r r u C −0.1 0 NL=200 NL=500 NL=1000 NL=2000 500 1000 Time 1500 2000 NL=200 NL=500 NL=1000 500 1000 Time 1500 2000 Figure 4: (Color online) The finite size effect on the electric current. The figures show the time evolution of the currents through a junction coupled with leads of different lengths. Top: constant bias potential UL = −UR = 0.1. Bottom: dynamic potential UL = 0.2 sin(0.05t), UR = 0. Next we compute the transient current of the DC circuit (case 1) from the effective reduced models (39) and compare it with the current from the full model (2) to evaluate the accuracy of the reduction method. We also examine the different choices of increasing the subspaces (as discussed in section 2.5). In particular, in Figure 5 we show the numerical results from using the subspaces (26) and (30), and we choose the dimension nΓ from 1 to 10. First we notice that no recurrent phenomenon is observed, which can be attributed to the non-homogeneous term Θ(t) as well as the self-energy in Eq (39), since they take into account the influence from the bath. The results improve as we expand the subspace, Vα and Wα, α = L/R in Eq (18). The steady state current has already been well captured by the reduced model with dimensions nΓ = 2, while the transient results improve as we expand nΓ, and we arrive at a very satisfactory result when nΓ = 4. 20 t n e r r u C 0.1 0.08 0.06 0.04 0.02 0 0 0.1 t n e r r u C 0.05 0 0 50 Full NG=1 NG=2 Full NG=1 NG=2 NG=4 NG=10 40 50 10 20 30 100 Time 150 200 Figure 5: (Color online) The simulation of the DC circuit (constant bias UL = −UR = 0.1). The figure shows the time history of the current from the reduced model (39) with different subspace dimensions, compared to the result from the full model (2). The subspaces are chosen from (26) and (30) by extending the contact region ΩΓ with parameter ε = 0.3 − 0.1i. The inset shows the transient stage of the current. We also tested the Krylov subspaces according to (47) and (48). The subspaces can be expanded by increasing m. The steady state is well captured when m = 3, the transient requires higher order approximations. Our observation is that in order to achieve the same accuracy, we need larger subspaces than the previous approach. On the other hand, the Krylov subspace approach is more robust in the regime where Im(ε) is close to zero. t n e r r u C 0.1 0.08 0.06 0.04 0.02 0 0 0.1 t n e r r u C 0.05 0 0 50 Full Nsub=1 Nsub=2 Nsub=3 Full Nsub=1 Nsub=2 Nsub=3 Nsub=4 40 50 10 20 30 100 Time 150 200 Figure 6: (Color online) The results from the simulation of the DC circuit (constant bias UL = −UR = 0.1). The figure shows the time evolution of the current from the reduced model (39), generated by the block Krylov subspaces (47) and (48) for various choices of dimensions (Nsub=m), with parameter ε = 0.3 − 0.01i. The results are compared to the result from the full model (2). The inset shows the transient stage of the current. 21 Another important factor that plays a role in the reduced model is the selection of the parameter ε, which can be viewed as an interpolation point for the self-energy. Therefore, we study the dependence of ε in the reduced models, by observing the electric current at steady state for various different choices of ε. For the imaginary part, we require Im(ε) to be strictly less than zero to ensure that the self-energy (33) is well defined and (39) has the stability assurance. We start with Im(ε) = 0.1. When Im(ε) is further decreased (< 0.01), the electric current exhibits oscillations around the true value of the steady state. For the real part of ε, the optimal value appears around the Fermi energy. See Figure 7. This suggests that ε should be around the Fermi level with small imaginary part, although when the imaginary part is too small, the numerical robustness might be affected. 0.12 0.1 0.08 0.06 t n e r r u C e t a t s − y d a e t S 0.04 0 0.2 0.4 Real(ε) 0.6 True Im(ε)=−0.21 Im(ε)=−0.16 Im(ε)=−0.11 Im(ε)=−0.06 Im(ε)=−0.01 0.8 1 Figure 7: (Color online) The example of DC circuit with constant bias (UL = −UR = 0.1). The Figure shows the steady-state current predicted by the reduced model (39) using various choices of the parameter ε with nΓ = 1. Finally, we turn to the example of the AC circuit. Since a time-dependent external potential is imposed, Hc and Θc in Eq (39) are time-dependent as well. They need to be evaluated at each time step. Due to the periodic property, it suffices to pre-compute Hc(t) and Θc(t) within one time period. As shown in Figure 8, a periodic electric current has been reproduced by the reduced model (39), and the accuracy also improves as we expand the subspace size nΓ. The electric current is already well captured when nΓ = 4. 22 0.1 0.05 0 t n e r r u C −0.05 −0.1 0 Full NG=1 NG=2 NG=4 50 100 Time 150 200 Figure 8: (Color online) The example of an AC circuit with time-dependent potential UL(t) = 0.2 sin(0.05t), UR = 0. This figure displays the time evolution of the currents from the reduced models with different subspace dimensions nΓ, compared to that from the full model. The parameter ε = 0.3 − 0.1i is used. 4 Summary We have proposed to formulate the quantum transport problem in a molecular junction coupled with infinite baths as a reduced-order modeling problem. The goal is to derive a finite quantum system with open boundary conditions. Motivated by the works, 36,38,39 we work with the density-matrix, and obtain reduced Liouville-von Neumann equations for the center and contact regions. The reduced equations are derived using a systematic projection formalism, together with appropriate choices of the subspaces. Numerical experiments have shown that the reduced model is very effective in capturing the steady-state electric current as well as the transient process of the electric current. The accuracy increases as we expand the contact regions in the reduced model. In order to demonstrate the reduction procedure, we have considered a one-dimensional junction system. But the validity of the projection approach is not restricted to the one- dimensional system. It can be applied to general coupled system-bath dynamics that require model reduction due to the computational complexity. The extension to systems that are of direct practical interest is underway. Another possible extension is the data-driven im- plementation of reduced-order modeling. In this case, rather than computing the matrices 23 in the reduced models from the underlying quantum mechanical models, they are inferred from observations. 79,80 Self-consistency has not been included in the Liouville-von Neumann equation, especially the Coulomb potential, which in the linear response regime, leads to a dense matrix 81 from the Hartree term. This creates considerable difficulty for the reduce-order modeling since the partition (22) is no longer reasonable. However, the Coulomb and exchange correlation are known to be important for the Coulomb blockade phenomena. 82 This difficulty in the modeling of quantum transport has also been pointed out in. 27,83 In practice, this is often dealt with by solving Poisson's equation in a relatively larger domain with Dirichlet boundary conditions. 23 We will address this issue under the framework of reduced-order modeling in separate works. Acknowledgement This research was supported by NSF under grant DMS-1619661 and DMS-1819011. References (1) Aviram, A. Molecular electronics-science and technology. Angewandte Chemie 1989, 101, 536 -- 537. (2) Reed, M. A. Molecular-scale electronics. Proceedings of the IEEE 1999, 87, 652 -- 658. (3) Joachim, C.; Gimzewski, J. K.; Aviram, A. Electronics using hybrid-molecular and mono-molecular devices. Nature 2000, 408, 541. (4) Aradhya, S. V.; Venkataraman, L. Single-molecule junctions beyond electronic trans- port. Nature Nanotechnology 2013, 8, 399. 24 (5) Cahen, D.; Kahn, A.; Umbach, E. Energetics of molecular interfaces. Materials Today 2005, 8, 32 -- 41. (6) Cahen, D.; Kahn, A. Electron energetics at surfaces and interfaces: Concepts and experiments. Advanced Materials 2003, 15, 271 -- 277. (7) Hwang, J.; Wan, A.; Kahn, A. Energetics of metal-organic interfaces: New experiments and assessment of the field. Materials Science and Engineering: R: Reports 2009, 64, 1 -- 31. (8) Brabec, C. J. Organic photovoltaics: Technology and market. Solar Energy Materials and Solar Cells 2004, 83, 273 -- 292. (9) Br´edas, J.-L.; Norton, J. E.; Cornil, J.; Coropceanu, V. Molecular understanding of organic solar cells: The challenges. Accounts of Chemical Research 2009, 42, 1691 -- 1699. (10) Poulsen, J. A.; Rossky, P. J. Path integral centroid molecular-dynamics evaluation of vibrational energy relaxation in condensed phase. The Journal of Chemical Physics 2001, 115, 8024 -- 8031. (11) Potter, S.; Skinner, M. J. On transport integration: A contribution to better under- standing. Futures 2000, 32, 275 -- 287. (12) Everitt, K.; Egorov, S.; Skinner, J. Vibrational energy relaxation in liquid oxygen. Chemical Physics 1998, 235, 115 -- 122. (13) Nibbering, E. T.; Wiersma, D. A.; Duppen, K. Femtosecond non-Markovian optical dynamics in solution. Physical Review Letters 1991, 66, 2464. (14) Pshenichnikov, M. S.; Duppen, K.; Wiersma, D. A. Time-resolved femtosecond photon echo probes bimodal solvent dynamics. Physical Review Letters 1995, 74, 674. 25 (15) Joo, T.; Jia, Y.; Fleming, G. R. Ultrafast liquid dynamics studied by third and fifth order three pulse photon echoes. The Journal of Chemical Physics 1995, 102, 4063 -- 4068. (16) Becker, P.; Fragnito, H.; Bigot, J.; Cruz, C. B.; Fork, R.; Shank, C. Femtosecond photon echoes from molecules in solution. Physical Review Letters 1989, 63, 505. (17) Lang, M.; Jordanides, X.; Song, X.; Fleming, G. Aqueous solvation dynamics studied by photon echo spectroscopy. The Journal of Chemical Physics 1999, 110, 5884 -- 5892. (18) Landauer, R. Electrical resistance of disordered one-dimensional lattices. Philosophical Magazine 1970, 21, 863 -- 867. (19) Buttiker, M.; Imry, Y.; Landauer, R.; Pinhas, S. Generalized many-channel conductance formula with application to small rings. Physical Review B 1985, 31, 6207. (20) Buttiker, M. Four-terminal phase-coherent conductance. Physical Review Letters 1986, 57, 1761. (21) Cini, M. Time-dependent approach to electron transport through junctions: General theory and simple applications. Physical Review B 1980, 22, 5887. (22) Lang, N. Resistance of atomic wires. Physical Review B 1995, 52, 5335. (23) Taylor, J.; Guo, H.; Wang, J. Ab initio modeling of quantum transport properties of molecular electronic devices. Physical Review B 2001, 63, 245407. (24) Brandbyge, M.; Mozos, J.-L.; Ordej´on, P.; Taylor, J.; Stokbro, K. Density-functional method for nonequilibrium electron transport. Physical Review B 2002, 65, 165401. (25) Hohenberg, P.; Kohn, W. Inhomogeneous electron gas. Physical Review 1964, 136, B864. 26 (26) Kohn, W.; Sham, L. J. Self-consistent equations including exchange and correlation effects. Physical Review 1965, 140, A1133 -- A1138. (27) Kurth, S.; Stefanucci, G.; Almbladh, C.-O.; Rubio, A.; Gross, E. K. Time-dependent quantum transport: A practical scheme using density functional theory. Physical Review B 2005, 72, 035308. (28) Runge, E.; Gross, E. K. Density-functional theory for time-dependent systems. Physical Review Letters 1984, 52, 997. (29) Stefanucci, G.; Almbladh, C.-O. Time-dependent quantum transport: An exact formu- lation based on TDDFT. EPL (Europhysics Letters) 2004, 67, 14. (30) Burke, K.; Car, R.; Gebauer, R. Density functional theory of the electrical conductivity of molecular devices. Physical Review Letters 2005, 94, 146803. (31) Cheng, C.-L.; Evans, J. S.; Van Voorhis, T. Simulating molecular conductance using real-time density functional theory. Physical Review B 2006, 74, 155112. (32) Stefanucci, G.; Almbladh, C.-O. Time-dependent partition-free approach in resonant tunneling systems. Physical Review B 2004, 69, 195318. (33) Baer, R.; Seideman, T.; Ilani, S.; Neuhauser, D. Ab initio study of the alternating current impedance of a molecular junction. The Journal of Chemical Physics 2004, 120, 3387 -- 3396. (34) Varga, K. Time-dependent density functional study of transport in molecular junctions. Physical Review B 2011, 83, 195130. (35) Antoine, X.; Besse, C. Unconditionally stable discretization schemes of non-reflecting boundary conditions for the one-dimensional Schrodinger equation. Journal of Compu- tational Physics 2003, 188, 157 -- 175. 27 (36) S´anchez, C. G.; Stamenova, M.; Sanvito, S.; Bowler, D.; Horsfield, A. P.; Todorov, T. N. Molecular conduction: Do time-dependent simulations tell you more than the Landauer approach? The Journal of Chemical Physics 2006, 124, 214708. (37) Subotnik, J. E.; Hansen, T.; Ratner, M. A.; Nitzan, A. Nonequilibrium steady state transport via the reduced density matrix operator. The Journal of Chemical Physics 2009, 130, 144105. (38) Zelovich, T.; Kronik, L.; Hod, O. State representation approach for atomistic time- dependent transport calculations in molecular junctions. Journal of Chemical Theory and Computation 2014, 10, 2927 -- 2941. (39) Zelovich, T.; Kronik, L.; Hod, O. Molecule-Lead coupling at molecular junctions: Re- lation between the real-and state-space perspectives. Journal of Chemical Theory and Computation 2015, 11, 4861 -- 4869. (40) Bai, Z. Krylov subspace techniques for reduced-order modeling of large-scale dynamical systems. Applied Numerical Mathematics 2002, 43, 9 -- 44. (41) Freund, R. W. Krylov-subspace methods for reduced-order modeling in circuit simula- tion. Journal of Computational and Applied Mathematics 2000, 123, 395 -- 421. (42) Villemagne, C. d.; Skelton, R. E. Model reductions using a projection formulation. International Journal of Control 1987, 46, 2141 -- 2169. (43) Smith, G. D.; Smith, G. D. Numerical solution of partial differential equations: Finite difference methods; Oxford University Press, 1985. (44) Johnson, C. Numerical solution of partial differential equations by the finite element method ; Courier Corporation, 2012. (45) Lambert, J. D.; Lambert, J. Computational methods in ordinary differential equations; Wiley London, 1973; Vol. 23. 28 (46) Morton, K. W.; Mayers, D. F. Numerical solution of partial differential equations: An introduction; Cambridge University Press, 2005. (47) Beck, T. L. Real-space mesh techniques in density-functional theory. Reviews of Modern Physics 2000, 72, 1041. (48) Sankey, O. F.; Niklewski, D. J. Ab initio multicenter tight-binding model for molecular- dynamics simulations and other applications in covalent systems. Physical Review B 1989, 40, 3979. (49) Inglesfield, J. E. A method of embedding. J. Phys. C 1981, 14, 3795. (50) Kelly, P. J.; Car, R. Green's-matrix calculation of total energies of point defects in silicon. Phys. Rev. B 1992, 45, 6543. (51) Lin, L.; Yang, C.; Meza, J. C.; Lu, J.; Ying, L.; E, W. SelInv -- An algorithm for selected inversion of a sparse symmetric matrix. ACM Transactions on Mathematical Software (TOMS) 2011, 37, 40. (52) Williams, A.; Feibelman, P. J.; Lang, N. Green's-function methods for electronic- structure calculations. Physical Review B 1982, 26, 5433. (53) Godfrin, E. M. A method to compute the inverse of an n-block tridiagonal quasi- Hermitian matrix. Journal of Physics: Condensed Matter 1991, 3, 7843. (54) Pecchia, A.; Penazzi, G.; Salvucci, L.; Di Carlo, A. Non-equilibrium Green's functions in density functional tight binding: method and applications. New Journal of Physics 2008, 10, 065022. (55) Li, X. An atomistic-based boundary element method for the reduction of molecular statics models. Computer Methods in Applied Mechanics and Engineering 2012, 225- 228, 1 -- 13. 29 (56) Li, X.; Lin, L.; Lu, J. PEXSI-Σ: A Green's function embedding method for Kohn -- Sham density functional theory. Annals of Mathematical Sciences and Applications 2018, 3, 441 -- 472. (57) Gross, E.; Kohn, W. Local density-functional theory of frequency-dependent linear response. Physical Review Letters 1985, 55, 2850. (58) Dobson, J. F.; Bunner, M.; Gross, E. Time-dependent density functional theory be- yond linear response: An exchange-correlation potential with memory. Physical Review Letters 1997, 79, 1905. (59) Gunnarsson, O.; Lundqvist, B. I. Exchange and correlation in atoms, molecules, and solids by the spin-density-functional formalism. Physical Review B 1976, 13, 4274. (60) Li, X. Absorbing boundary conditions for time-dependent Schrodinger equations: A density-matrix formulation. The Journal of Chemical Physics 2019, 150, 114111. (61) Gugercin, S.; Stykel, T.; Wyatt, S. Model reduction of descriptor systems by interpo- latory projection methods. SIAM Journal on Scientific Computing 2013, 35, B1010 -- B1033. (62) Lucia, D. J.; Beran, P. S.; Silva, W. A. Reduced-order modeling: New approaches for computational physics. Progress in Aerospace Sciences 2004, 40, 51 -- 117. (63) Nayfeh, A. H.; Younis, M. I.; Abdel-Rahman, E. M. Reduced-order models for MEMS applications. Nonlinear Dynamics 2005, 41, 211 -- 236. (64) Decoster, M.; Van Cauwenberghe, A. A comparative study of different reduction meth- ods. Journal A 1976, 17, 125 -- 134. (65) Larsson, S.; Thom´ee, V. Partial differential equations with numerical methods; Springer Science & Business Media, 2008; Vol. 45. 30 (66) Jaimoukha, I. M.; Kasenally, E. M. Krylov subspace methods for solving large Lyapunov equations. SIAM Journal on Numerical Analysis 1994, 31, 227 -- 251. (67) Jbilou, K.; Riquet, A. Projection methods for large Lyapunov matrix equations. Linear Algebra and its Applications 2006, 415, 344 -- 358. (68) Do, V.-N. Non-equilibrium Green function method: Theory and application in simula- tion of nanometer electronic devices. Advances in Natural Sciences: Nanoscience and Nanotechnology 2014, 5, 033001. (69) Kadanoff, L.; Baym, G. Quantum Statistical Mechanics WA benjamin Inc. New York 1962, (70) Caroli, C.; Combescot, R.; Nozieres, P.; Saint-James, D. Direct calculation of the tun- neling current. Journal of Physics C: Solid State Physics 1971, 4, 916. (71) Datta, S. Quantum transport: Atom to transistor ; Cambridge University Press, 2005. (72) Popov, V. S. Tunnel and multiphoton ionization of atoms and ions in a strong laser field (Keldysh theory). Physics-Uspekhi 2004, 47, 855. (73) Danielewicz, P. Quantum theory of nonequilibrium processes, I. Annals of Physics 1984, 152, 239 -- 304. (74) Xue, Y.; Datta, S.; Ratner, M. A. First-principles based matrix Green's function ap- proach to molecular electronic devices: General formalism. Chemical Physics 2002, 281, 151 -- 170. (75) Meier, C.; Tannor, D. J. Non-Markovian evolution of the density operator in the pres- ence of strong laser fields. The Journal of Chemical Physics 1999, 111, 3365 -- 3376. (76) Sancho, M. L.; Sancho, J. L.; Rubio, J. Quick iterative scheme for the calculation of transfer matrices: application to Mo (100). Journal of Physics F: Metal Physics 1984, 14, 1205. 31 (77) Sancho, M. L.; Sancho, J. L.; Sancho, J. L.; Rubio, J. Highly convergent schemes for the calculation of bulk and surface Green functions. Journal of Physics F: Metal Physics 1985, 15, 851. (78) Brauer, F. Perturbations of nonlinear systems of differential equations. Journal of Math- ematical Analysis and Applications 1966, 14, 198 -- 206. (79) Ma, L.; Li, X.; Liu, C. Coarse-graining Langevin dynamics using reduced-order tech- niques. Journal of Computational Physics 2019, 380, 170 -- 190. (80) Benner, P.; Gugercin, S.; Willcox, K. A survey of projection-based model reduction methods for parametric dynamical systems. SIAM review 2015, 57, 483 -- 531. (81) Yabana, K.; Nakatsukasa, T.; Iwata, J. I.; Bertsch, G. F. Real-time, real-space im- plementation of the linear response time-dependent density-functional theory. Physica Status Solidi (B) Basic Research 2006, 243, 1121 -- 1138. (82) Kurth, S.; Stefanucci, G.; Khosravi, E.; Verdozzi, C.; Gross, E. Dynamical Coulomb blockade and the derivative discontinuity of time-dependent density functional theory. Physical Review Letters 2010, 104, 236801. (83) Ullrich, C. A. Time-dependent density-functional theory: Concepts and applications; OUP Oxford, 2011. 32 

1303.5784

2

1303

Electromagnetic Response of Weyl Semimetals

It has been suggested recently, based on subtle field-theoretical considerations, that the electromagnetic response of Weyl semimetals and the closely related Weyl insulators can be characterized by an axion term E.B with space and time dependent axion angle. Here we construct a minimal lattice model of the Weyl medium and study its electromagnetic response by a combination of analytical and numerical techniques. We confirm the existence of the anomalous Hall effect expected on the basis of the field theory treatment. We find, contrary to the latter, that chiral magnetic effect (that is, ground-state charge current induced by the applied magnetic field) is absent in both the semimetal and the insulator phase. We elucidate the reasons for this discrepancy.

cond-mat.mes-hall

cond-mat

Electromagnetic Response of Weyl Semimetals M.M. Vazifeh and M. Franz Department of Physics and Astronomy, University of British Columbia, Vancouver, BC, Canada V6T 1Z1 (Dated: February 7, 2014) It has been suggested recently, based on subtle field-theoretical considerations, that the electro- magnetic response of Weyl semimetals and the closely related Weyl insulators can be characterized by an axion term θE · B with space and time dependent axion angle θ(r , t). Here we construct a minimal lattice model of the Weyl medium and study its electromagnetic response by a combina- tion of analytical and numerical techniques. We confirm the existence of the anomalous Hall effect expected on the basis of the field theory treatment. We find, contrary to the latter, that chiral mag- netic effect (that is, ground-state charge current induced by the applied magnetic field) is absent in both the semimetal and the insulator phase. We elucidate the reasons for this discrepancy. When a three-dimensional topological insulator (TI) [1, 2, 25] undergoes a phase transition into an ordinary band insulator, its low-energy electronic spectrum at the critical point consists of an odd number of 3D massless Dirac points. Such 3D Dirac points have been experi- mentally observed in TlBi(S1−xSex )2 crystals [4] and in (Bi1−x Inx )2Se2 films [5]. In the presence of the time re- versal (T ) and inversion (P ) symmetries the Dirac points are doubly degenerate and occur at high-symmetry po- sitions in the Brillouin zone. When T or P is broken, however, each Dirac point can split into a pair of ‘Weyl points’ separated from one another in momentum k or energy E , as illustrated in Fig. 1. The resulting Weyl semimetal constitutes a new phase of topological quan- tum matter [6–14] with a number of fascinating physical properties including protected surface states and unusual electromagnetic response. The low energy theory of an isolated Weyl point is given by the Hamiltonian hW (k) = b0 + vσ · (k − b), (1) where v is the characteristic velocity, σ a vector of the Pauli matrices, b0 and b denote the shift in energy and momentum, respectively. Because all three Pauli matri- ces are used up in hW (k), small perturbations can renor- malize the parameters, b0 , b and v , but cannot open a gap. This explains why Weyl semimetal forms a stable phase [6]. Although the phase has yet to be experimen- tally observed there are a number of proposed candidate systems, including pyrochlore iridates [7, 8], TI multilay- ers [9–12], and magnetically doped TIs [13, 14]. The purpose of this Letter is to address the remarkable electromagnetic properties of Weyl semimetals. Accord- ing to the recent theoretical work [15–18], the universal part of their EM response is described by the topological (cid:90) θ-term, e2 8π2 (using  = c = 1 units) with the ‘axion’ angle given by θ(r , t) = 2(b · r − b0 t). dtdrθ(r , t)E · B , Sθ = (3) (2) FIG. 1: Low energy spectra in Dirac and Weyl semimetals. a) Doubly degenerate massless Dirac cone at the transition from a TI to a band insulator. Weyl semimetals with the individual cones shifted in b) momenta and c) energy. Panel d) illustrates the Weyl insulator which can arise when the excitonic instability gaps out the spectrum indicated in c). In all panels two components of the 3D crystal momentum k are shown. (4) ρ = j = This unusual response is a consequence of the chiral anomaly [19–21], well known in the quantum field the- ory of Dirac fermions. The physical manifestations of the θ-term can be best understood from the associated equa- tions of motion, which give rise to the following charge density and current response, e2 2π2 b · B , e2 2π2 (b × E − b0B ). Eq. (4) and the first term in Eq. (5) encode the anoma- lous Hall effect that is expected to occur in a Weyl semimetal with broken T [7–10]. The second term in Eq. (5) describes the ‘chiral magnetic effect’ [22], whereby a ground-state dissipationless current proportional to the applied magnetic field B is generated in the bulk of a Weyl semimetal with broken P . The anomalous Hall effect is known to commonly oc- cur in solids with broken time-reversal symmetry. In the present case of the Weyl semimetal its origin and magni- tude can be understood from simple physical arguments [7–10] applied to the bulk system as well as in the limit of decoupled 2D layers [18]. Understanding the chiral magnetic effect (CME) in a system with non-zero en- (5) 3 1 0 2 n u J 7 2 ] l l a h - s e m . t a m - d n o c [ 2 v 4 8 7 5 . 3 0 3 1 : v i X r a (cid:1)(cid:2)(cid:3)(cid:4)(cid:5)(cid:6)ergy shift b0 presents a far greater challenge. The issue becomes particularly intriguing in the case of a Weyl in- sulator, illustrated in Fig. 1d, which will generically arise due to the exciton instability in the presence of repul- sive interactions and nested Fermi surfaces. According to Ref. [15] CME should persist even when the chemical potential resides inside the bulk gap. At the same time, standard arguments from the band theory of solids dic- tate that filled bands cannot contribute to the electrical current [23]. We remark that using a different regular- ization scheme for the Weyl fermions Ref. [17] found that CME occurs in the semimetal but is absent in the insu- lator, while Ref. [18] concluded that it only occurs when b2 − b2 0 ≥ m2 D , where mD denotes the gap magnitude. Semiclassical considerations [24] on the other hand pre- dict a vanishing electrical current in the Weyl semimetal but non-zero ‘valley current’ proportional to B . CME, if present, could have interesting technological applications, as it constitutes a dissipationless ground state current, controllable by an external field. Disagree- ments between the various field-theory predictions, how- ever, raise important questions about the existence of CME in Weyl semimetals and insulators. The implied contradiction with one of the basic results of the band theory calls into question whether the results based on the low-energy Dirac-Weyl Hamiltonians are applicable to the real solid with electrons properly regularized on the lattice. In this Letter we undertake to resolve these questions by constructing and analyzing a lattice model of a Weyl medium. Using simple physical arguments and exact numerical diagonalization, we confirm the existence of the anomalous Hall effect as implied by Eqs. (4,5) when b (cid:54)= 0. We find, using the same model with b0 (cid:54)= 0, that CME does not occur in either the Weyl semimetal or insulator, in agreement with arguments from the band theory of solids which we review in some detail. Our starting point is the standard model describing a 3D TI in the Bi2Se3 family [25, 26], regularized on a simple cubic lattice, defined by the the momentum space Hamiltonian H0 (k) = 2λσz (sx sin ky − sy sin kx ) + 2λz σy sin kz respectively, and Mk =
− 2t (cid:80) + σxMk , (6) with σ and s the Pauli matrices in orbital and spin space, α cos kα . For λ, λz > 0 and 2t <
< 6t the above model describes a strong topological insulator with the Z2 index (1;000). In the following, we shall focus on the vicinity of the phase tran- sition to the trivial phase that occurs at
= 6t, via the gap closing at k = 0. It is easy to see that Weyl semimetal emerges when we add the following perturbation to H0 , H1 (k) = b0σy sz + b · (−σx sx , σx sy , sz ). (7) Nonzero b0 breaks P but respects T while b has the opposite effect. The two symmetries are generated as 2 FIG. 2: The band structure of the Weyl semimetal lattice (π , 0, π) → (0, 0, 0) → model, displayed along the path k : (0, 0, π) → (π , 0, π). a) Doubly degenerate 3D Dirac point when H1 = 0 and
= 6t. b) Momentum-shifted Weyl point for b = 0.9 and b0 = 0. c) Energy-shifted Weyl points for bz = 0 and b0 = 0.7. d) Weyl insulator with bz = 0 and b0 = 0.7 and the exciton gap modeled by taking
= 5.9t. In all panels we take λ = λz = 1.0, t = 0.5 and the energy is measured in units of λ. Red circles mark the location of the Dirac/Weyl points. follows, P : σxH (k)σx = H (−k) and T : syH ∗ (k)sy = H (−k). For simplicity and concreteness we focus on the case b = bz z , which yields a pair of Weyl points at k = ±(bz /2λz ) z . The band structure of H = H0 + H1 for various cases of interest is displayed in Fig. 2. We now address the anomalous Hall effect by directly testing Eq. (4). To this end we consider a rectangular sample of the Weyl semimetal with a base of (L × L) sites in the x-y plane and periodic boundary conditions, (cid:82) j infinite along the z -direction. The effect of the applied magnetic field is included via the standard Peierls substi- tution, t → t exp [2π i/Φ0 i A · dl], where Φ0 = hc/e is the flux quantum, A is the vector potential and the inte- gral is taken along the straight line between sites ri and rj of the lattice. For B = zB (x, y) we retain the transla- tional invariance along the z -direction and the Hamilto- nian becomes a matrix of size 16L2 for each value of kz . We find the eigenstates φn,kz (x, y) of H by means of ex- act numerical diagonalization and use these to calculate (cid:88) (cid:88) the charge density n∈occ kz φn,kz (x, y)2 . ρ(x, y) = e (8) Figure 3a displays ρ for the magnetic field configu- ration B (x, y) = Φ[δ(x − L/4) − δ(x + L/4)]δ(y), i.e. two flux tubes separated by L/2 along the x direction. In accord with Eq. (4) charge accumulates near the flux tubes, although ρ(x, y) is somewhat broadened compared to B (x, y). We expect the total accumulated charge per Out[56]=ZGMZ-4-2024ZGMZ-4-2024ZGMZ-4-2024ZGMZ-4-2024kEkEkEkEcbda3 (12) φn,kz Jz = nF [
n (kz )], † kz α ckz β . (11) c (cid:12)(cid:12)(cid:12)(cid:12)Az→0 (cid:88) The current operator is given by ∂H(Az ) ∂H αβ (kz ) = −e ∂ kz ∂Az kz (cid:12)(cid:12)(cid:12)(cid:12) ∂H (kz ) (cid:12)(cid:12)(cid:12)(cid:12) φn,kz (cid:28) (cid:29) (cid:88) This leads to the current expectation value Jz = −e ∂ kz n,kz where nF indicates the Fermi-Dirac distribution and
n (kz ) the energy eigenvalues of H (kz ). We note that Eq. (12) remains valid in the presence of the exciton con- densate as long as it is treated in the standard mean field theory. We have evaluated Jz from Eq. (12) for various system sizes, boundary conditions, field strengths and parameter values corresponding to energy- and momentum-shifted Weyl semimetals and insulators. In all cases we found Jz = 0 to within the numerical accuracy of our compu- tations, typically 6-8 orders of magnitude smaller than CME expected on the basis of Eq. (5). For an insulator, vanishing of Jz comes of course as no surprise. At T = 0 and using the fact that (cid:90) ∂kz (cid:104)φn,kz φn,kz (cid:105) = 0 one can rewrite Eq. (12) as (cid:88) Jz = −e ∂
n (kz ) dkz 2π ∂ kz n∈occ BZ (13) , which vanishes owing to the periodicity of
n (kz ) on the Brillouin zone. More generally, for a system at non-zero temperature and when partially filled bands are present (cid:90) (cid:88) we can rewrite Eq. (12) as BZ n ∂
n (kz ) ∂ kz Jz = −e nF [
n (kz )], dkz 2π (14) where the sum over n extends over all bands. By trans- forming the kz -integral in Eq. (14) into an integral over the energy it is easy to see that it identically vanishes for any continuous energy dispersion
n (kz ) that is periodic on the Brillouin zone and for any distribution function that only depends on energy. This reflects the well-known fact that one must establish a non-equilibrium distribu- tion of electrons to drive current in a metal, e.g. by apply- ing an electric field. Given these arguments we conclude that, as a matter of principle, CME cannot occur in a crystalline solid, at least when interactions are unimpor- tant and the description within the independent electron approximation remains valid. There are several notable cases when filled bands do contribute currents. A superconductor can be thought of as an insulator for Bogoliubov quasiparticles and yet it supports a supercurrent. This occurs because Bogoliubov quasiparticles, being coherent superpositions of electrons and holes, do not carry a definite charge and consequently the current cannot be expressed through Eq. (12). In FIG. 3: a) Charge density δρ(x, y) accumulated in the vicin- ity of the flux tubes Φ = 0.01Φ0 in the Weyl semimetal. b) Total accumulated charge per layer δQ near one of the flux tubes, in units of e/2π for indicated values of bz . Dashed lines represent the expectation based on Eq (9). We use λ = λz = t = 0.5,
= 3.0, L = 14 and Lz = 160 independent values of kz . Panels c) and d) show the charge accumulations as a function of bz in the presence of non-zero Dirac mass and b0 . Parameters as above except b0 = 0.1, 0.2, 0.3 in c) and
= 3.0, 2.9, 2.8, 2.7 for the curves in d) from left to right. (cid:18) bz (cid:19) Φ layer δQ to be proportional to the total flux, Φ0 2λz δQ = e π (9) , where we have restored the physical units. Fig. 3b shows that this proportionality holds very accurately when the flux through an elementary plaquette is small compared to Φ0 . [When the flux approaches Φ0 /2 we no longer expect Eq. (9) to hold because of the lattice effects.] We have also tested the effect of a non-zero Dirac mass, mD =
− 6t, and non-zero b0 on the anomalous Hall ef- 0 > b2 D + b2 fect. These terms compete with bz and for m2 z one expects the Hall effect to disappear [17, 18]. This is indeed what we observe in Figs. 3c,d. We have performed similar calculations for other field profiles B (x, y) reach- ing identical conclusions for the anomalous Hall effect. We now address the chiral magnetic effect, predicted to occur when b0 (cid:54)= 0. We consider the same sample ge- ometry as above, but now with uniform field B = zB . In order to account for possible contribution of the surface states we study systems with both periodic and open boundary conditions along x. To find the current re- sponse we introduce a uniform vector potential Az along the z -direction (in addition to Ax and Ay required to en- code the applied magnetic field). The second-quantized (cid:88) Hamiltonian then reads † H αβ (kz − eAz )c H(Az ) = kz α ckz β , kz (10) where α, β represent all the site, orbital and spin indices. (cid:10)(cid:13)(cid:10)(cid:10)(cid:13)(cid:7)(cid:8)(cid:10)(cid:7)(cid:14)(cid:7)(cid:8)(cid:10)(cid:15)(cid:16)(cid:14)(cid:7)(cid:8)(cid:17)(cid:7)(cid:7)(cid:8)(cid:17)(cid:1)(cid:2)(cid:18)(cid:15)(cid:19)(cid:16)(cid:20)(cid:15)(cid:16)(cid:1)(cid:21)(cid:3)(cid:22)(cid:3)(cid:7)(cid:11)(cid:7)(cid:8)(cid:10)(cid:7)(cid:11)(cid:7)(cid:8)(cid:7)(cid:23)(cid:7)(cid:8)(cid:7)(cid:7)(cid:7)(cid:8)(cid:7)(cid:23)(cid:7)(cid:8)(cid:10)(cid:7)(cid:7)(cid:8)(cid:7)(cid:7)(cid:7)(cid:8)(cid:7)(cid:23)(cid:7)(cid:8)(cid:10)(cid:7)(cid:7)(cid:8)(cid:10)(cid:23)(cid:7)(cid:8)(cid:17)(cid:7)(cid:7)(cid:8)(cid:17)(cid:23)(cid:7)(cid:8)(cid:9)(cid:7)(cid:1)(cid:2)(cid:3)(cid:4)(cid:2)(cid:5)(cid:5)(cid:6)(cid:7)(cid:8)(cid:9)(cid:7)(cid:8)(cid:10)(cid:11)(cid:7)(cid:8)(cid:10)(cid:11)(cid:7)(cid:8)(cid:9)(cid:12)(cid:7)(cid:8)(cid:7)(cid:7)(cid:8)(cid:10)(cid:7)(cid:8)(cid:17)(cid:7)(cid:8)(cid:9)(cid:7)(cid:8)(cid:13)(cid:7)(cid:8)(cid:23)(cid:7)(cid:8)(cid:24)(cid:7)(cid:8)(cid:7)(cid:7)(cid:8)(cid:10)(cid:7)(cid:8)(cid:17)(cid:7)(cid:8)(cid:9)(cid:7)(cid:8)(cid:13)(cid:7)(cid:8)(cid:23)(cid:7)(cid:8)(cid:24)(cid:1)(cid:21)(cid:4)(cid:2)(cid:12)(cid:7)(cid:7)(cid:8)(cid:10)(cid:7)(cid:8)(cid:17)(cid:7)(cid:8)(cid:9)(cid:7)(cid:8)(cid:13)(cid:7)(cid:8)(cid:23)(cid:7)(cid:8)(cid:24)(cid:7)(cid:7)(cid:8)(cid:10)(cid:7)(cid:8)(cid:17)(cid:7)(cid:8)(cid:9)(cid:7)(cid:8)(cid:13)(cid:7)(cid:8)(cid:23)(cid:2)(cid:12)(cid:1)(cid:21)(cid:4)4 These considerations thus explain the difference between the low-energy and lattice descriptions of Weyl semimet- als. In closing, we note that if correct, the time depen- dence of the axion angle implied by Eq. (3) would en- gender some peculiar consequences. One of them follows from the Witten effect [28, 29] whereby a unit magnetic monopole inserted into the axion medium carries a po- larization charge δQ = −e(n + θ/2π) with n integer. In the Weyl semimetal with a nonzero energy shift b0 this charge δQ would grow linearly with time according to Eq. (3). Although such ‘quantum time crystal’ behav- ior has been conjectured to arise in certain interacting systems [30] it is not clear by what mechanism it would occur in the ground state of a non-interacting semimetal. Our findings indeed confirm the absence of this behav- ior in the Weyl semimetal described by a natural lat- tice Hamiltonian. It remains an open question whether this fascinating phenomenon can be realized in another quantum system. Another interesting problem which we leave for future investigation is finding the regularization scheme for the low-energy theory that would match the results of our lattice calculation. The authors are indebted to I. Affleck, A.A. Burkov, M.P.A. Fisher, A. Grushin and I.F. Herbut for illumi- nating discussions and correspondence. The work was supported by NSERC and CIfAR. [1] J. E. Moore, Nature 464, 194 (2010). [2] M.Z. Hasan, C.L. Kane, Rev. Mod. Phys. 82 3045 (2010). [3] X.-L. Qi, S.-C. Zhang, Rev. Mod. Phys. 83, 1057 (2011). [4] S.-Y. Xu, et al., Science 332, 560 (2011). [5] M. Brahlek, et al., Phys. Rev. Lett. 109, 186403 (2012). [6] S. Murakami, New J. Phys. 9, 356 (2007). [7] X. Wan, A. M. Turner, A. Vishwanath, S. Y. Savrasov, Phys. Rev. B 83 205101 (2011). [8] W. Witczak-Krempa and Y.-B. Kim, Phys. Rev. B 85, 045124 (2012). [9] A.A. Burkov, L. Balents, Phys. Rev. Lett. 107 127205 (2011). [10] A.A. Burkov, M. D. Hook, L. Balents, Phys. Rev. B 84 235126 (2011). [11] A.A. Zyuzin, Si Wu, A.A. Burkov, Phys. Rev. B 85, 165110 (2012). [12] G.B. Hal´asz and L. Balents, Phys. Rev. B 85, 035103 (2012). [13] G.Y. Cho, arXiv:1110.1939. [14] C.-X. Liu, P. Ye, X.-L. Qi, arXiv:1204.6551. [15] A.A. Zyuzin, A.A. Burkov, Phys. Rev. B 86, 115133 (2012). [16] D.T. Son and N. Yamamoto, Phys. Rev. Lett. 109, 181602 (2012). [17] P. Goswami and S. Tewari, arXiv:1210.6352. [18] A.G. Grushin, Phys. Rev. D 86, 045001 (2012). [19] S. Adler, Phys. Rev. 177, 2426 (1969). [20] J.S. Bell and R. Jackiw, Nuovo Cimento 60A, 4 (1969). [21] H.B. Nielsen and M. Ninomiya, Phys. Lett. 130B, 389 FIG. 4: a) Chiral current Jz as a function of energy offset b0 for various values of the momentum cutoff Λ. The dashed line indicates the field theory prediction Eq. (5). b) The slope dJz /db0 in units of eη/2π as a function of cutoff Λ. Slope 1.0 is expected on the basis of Eq. (5). quantum Hall insulators non-zero σxy also implies non- vanishing current. In the standard Hall bar geometry, used in transport measurements, it is well known that the physical current is carried by the gapless edge modes, not through the gapped bulk. In the Thouless charge pump geometry the current indeed flows through the insulat- ing bulk but this requires a time-dependent Hamiltonian (the magnetic flux through the cylinder is time depen- dent). Our considerations leading to Eq. (12) are only valid for time-independent Hamiltonians. Finally, there are known cases [27] when the transition from Eq. (12) to (13) fails because the Hamiltonian is not self-adjoint on the space of functions that includes derivatives of φ. This can happen when the Hamiltonian is a differential opera- tor but in our case H (kz ) is a finite-size hermitian matrix with a smooth dependence on kz , which precludes any such exotic possibility. In any case, our numerical cal- culations addressed directly Eq. (12) so self-adjointness cannot possibly be an issue. Our considerations conclusively establish that anoma- lous Hall effect [7–10], quantitatively consistent with the prediction of the low-energy continuum theory [15–18], occurs in realistic Weyl semimetals defined on the lat- tice. The chiral magnetic effect [22], implied by the same considerations via Eq. (3), however runs afoul of the ba- sic results of the band theory and is found to be absent. Within the low-energy continuum theory the form of the axion angle given Eq. (3) can be expected on the basis of Lorenz invariance. In the real solid this symmetry is broken at the lattice scale so there is no fundamental reason why this form should hold beyond the low-energy approximation. To test this hypothesis we have evalu- ated current Jz from Eq. (12) with a momentum cutoff, i.e. limiting kz < Λ in the sum. As shown in Fig. (4) when Λ (cid:28) π one indeed obtains CME with a magnitude consistent with Eq. (5). The current however rapidly vanishes as the cutoff approaches the extent of the full Brillouin zone. We remark that imposing such a cutoff has no significant effect on the Hall effect calculation as long as Λ > bz /2λz because the entire Hall response comes from this region of the momentum space [7–10]. (cid:11)(cid:12)(cid:11)(cid:11)(cid:11)(cid:12)(cid:15)(cid:11)(cid:11)(cid:12)(cid:19)(cid:11)(cid:11)(cid:12)(cid:20)(cid:11)(cid:11)(cid:12)(cid:21)(cid:11)(cid:13)(cid:12)(cid:11)(cid:11)(cid:11)(cid:12)(cid:15)(cid:11)(cid:12)(cid:16)(cid:11)(cid:12)(cid:19)(cid:11)(cid:12)(cid:14)(cid:11)(cid:12)(cid:20)(cid:11)(cid:12)(cid:22)(cid:11)(cid:12)(cid:21)(cid:11)(cid:12)(cid:23)(cid:13)(cid:12)(cid:11)(cid:1)(cid:8)(cid:9)(cid:15)(cid:9)(cid:5)(cid:18)(cid:11)(cid:2)(cid:24)(cid:25)(cid:11)(cid:12)(cid:11)(cid:11)(cid:11)(cid:12)(cid:11)(cid:14)(cid:11)(cid:12)(cid:13)(cid:11)(cid:11)(cid:12)(cid:13)(cid:14)(cid:11)(cid:12)(cid:15)(cid:11)(cid:11)(cid:12)(cid:15)(cid:14)(cid:11)(cid:12)(cid:16)(cid:11)(cid:11)(cid:12)(cid:11)(cid:11)(cid:12)(cid:13)(cid:11)(cid:12)(cid:13)(cid:11)(cid:12)(cid:15)(cid:11)(cid:12)(cid:15)(cid:11)(cid:12)(cid:15)(cid:11)(cid:12)(cid:16)(cid:18)(cid:11)(cid:15)(cid:9)(cid:5)(cid:2)(cid:24)(cid:25)(cid:1)(cid:2)(cid:3)(cid:4)(cid:5)(cid:6)(cid:7)(cid:1)(cid:8)(cid:9)(cid:10)(cid:11)(cid:12)(cid:13)(cid:14)(cid:11)(cid:12)(cid:15)(cid:11)(cid:12)(cid:16)(cid:11)(cid:12)(cid:14)(cid:17)(cid:18)(1983). [22] K. Fukushima, D.E. Kharzeev, H.J. Warringa, Phys. Rev. D 78, 074033 (2008). [23] See e.g. N.W. Ashcroft and N.D. Mermin, Solid State Physics, (Saunders College, 1976). [24] J. Zhou, H. Jiang, Q. Niu, J. Shi, Chinese Phys. Lett. 30, 027101 (2013). [25] X.-L. Qi, S.-C. Zhang, Rev. Mod. Phys. 83, 1057 (2011). [26] L. Fu and E. Berg, Phys. Rev. Lett. 105 097001, (2010). [27] J.G Esteve, F. Falceto, C.G Canal, Phys. Lett. A 374, 819 (2010). [28] E. Witten, Phys. Lett. B 86, 283 (1979). [29] G. Rosenberg and M. Franz, Phys. Rev. B 82, 035105 (2010). [30] F. Wilczek, Phys. Rev. Lett. 109, 160401 (2012) 5 

1201.1837

1

1201

Simultaneous Determination of Conductance and Thermopower of Single Molecule Junctions

We report the first concurrent determination of conductance (G) and thermopower (S) of single-molecule junctions via direct measurement of electrical and thermoelectric currents using a scanning tunneling microscope-based break-junction technique. We explore several amine-Au and pyridine-Au linked molecules that are predicted to conduct through either the highest occupied molecular orbital (HOMO) or the lowest unoccupied molecular orbital (LUMO), respectively. We find that the Seebeck coefficient is negative for pyridine-Au linked LUMO-conducting junctions and positive for amine-Au linked HOMO-conducting junctions. Within the accessible temperature gradients (<30 K), we do not observe a strong dependence of the junction Seebeck coefficient on temperature. From histograms of 1000's of junctions, we use the most probable Seebeck coefficient to determine a power factor, GS^2, for each junction studied, and find that GS^2 increases with G. Finally, we find that conductance and Seebeck coefficient values are in good quantitative agreement with our self-energy corrected density functional theory calculations.

cond-mat.mes-hall

cond-mat

Simultaneous  Determination  of  Conductance  and   Thermopower  of  Single  Molecule  Junctions   Jonathan  R.  Widawsky1,  Pierre  Darancet2,  Jeffrey  B.  Neaton*2,  Latha   Venkataraman*1     1Department  of  Applied  Physics  and  Applied  Mathematics,  Columbia  University,  New  York,  NY   2Molecular  Foundry,  Lawrence  Berkeley  National  Laboratory,  Berkeley,  CA     AUTHOR  EMAIL  ADDRESS:  [email protected];  [email protected]   Abstract:  We  report  the  first  concurrent  determination  of  conductance  (G)  and  thermopower  (S)   of   single-­‐molecule   junctions   via   direct   measurement   of   electrical   and   thermoelectric   currents   using   a   scanning   tunneling   microscope-­‐based   break-­‐junction   technique.   We   explore   several   amine-­‐Au   and   pyridine-­‐Au   linked   molecules   that   are   predicted   to   conduct   through   either   the   highest   occupied   molecular   orbital   (HOMO)   or   the   lowest   unoccupied   molecular   orbital   (LUMO),   respectively.   We   find   that   the   Seebeck   coefficient   is   negative   for   pyridine-­‐Au   linked   LUMO-­‐conducting   junctions   and   positive   for   amine-­‐Au   linked   HOMO-­‐conducting   junctions.  Within   the   accessible   temperature   gradients   (<30   K),   we   do   not   observe   a   strong   dependence   of   the   junction   Seebeck   coefficient   on   temperature.   From   histograms   of   1000’s   of   junctions,  we   use   the  most   probable   Seebeck   coefficient   to   determine   a   power   factor,   GS2,   for   each   junction   studied,   and   find   that   GS2   increases   with   G.   Finally,   we   find   that   conductance   and   Seebeck   coefficient   values   are   in   good   quantitative   agreement   with   our   self-­‐energy   corrected  density  functional  theory  calculations.   Keywords:     Molecular   Thermopower,   Molecular   Conductance,   Density   Functional   Theory,   Single   Molecule      Understanding   transport   characteristics   of   single   metal-­‐molecule-­‐metal   junctions   is   of   fundamental   importance   to   the   development   of   functional   nanoscale,   organic-­‐based   devices1.   Much   work   has   been   performed   investigating   the   low-­‐bias   conductance   of   molecules   attached   to   gold   leads   using   a   variety   of   chemical   link   groups2-­‐7.     However,   conductance   measurements   provide   only   a   partial   description   of   electron   transport   through   molecules,   and   cannot   directly   probe   molecular   energy   level   alignment  with   the   electrode   Fermi   energy   or   level   broadening   due   to   electronic   coupling   to   the   leads.     Thus,   there   is   significant   value   in   developing   complimentary  methods   to   investigate   junction   electronic   structure   and   attain   a   more   complete   understanding   of   molecular-­‐scale   transport.     In   particular,   molecular   junction   thermopower   measurements   can   be   useful   in   determining   the   dominant   molecular   orbital  for  transport  and  the   identity  of  the  primary  charge  carriers8-­‐11.  In  a  single-­‐molecule   junction,  the   thermopower   or   Seebeck   coefficient   S   determines   the   magnitude   of   the   built-­‐in   potential   developed   across   a   material   (or   molecule)   when   a   temperature   difference   ΔT   is   applied.     With   the   additional   presence   of   an   external   voltage   bias   ΔV   across   the   junction,   following   Ref.   12,   Equation   (17),   the   total   current  I  in  this  case  is  simply     ,   where   G   is   the   electrical   conductance.   Eq.   (1)   applies   for   both   bulk   materials,   where   transport   is   (e.g.)   (1)       diffusive13,  and  in  single-­‐molecule  junctions,  where  transport  can  be  coherent14.    For  coherent  tunneling,   formula13,  14,   the   conductance   through   a   molecular   junction,   in   the   zero-­‐bias   limit,   can   be   given   by   the   Landauer     (2)   and  the  Seebeck  coefficient  by      where  T(E)   is  the  transmission  function,  T  [=  (T1  +  T2)/2]   is  the  average  temperature  of  the   leads,  Ef   is  the   (3)     Fermi  Energy  of   the   leads  and  kB   is   the  Boltzmann  constant.   In   this  coherent   tunneling   limit,   the  current   at   zero   external   bias   is   then   only   due   to   a   difference   in   temperature   between   the   two   leads,   and   depends   on   the   slope   of   the   transmission   function   at   the   Fermi   energy,   EF.   Thus,   from   the   sign   of   this   thermoelectric   current   (and   therefore   S),  we   can   potentially   deduce  whether   EF   is   closer   to   the   highest   occupied  molecular   orbital   (HOMO)   or   lowest   unoccupied  molecular   orbital   (LUMO)   resonance   energy,   assuming  a  simple  Lorentzian-­‐type  model.   Here,   we   present   a   study   of   thermopower   measurements   for   several   amine-­‐Au   linked   HOMO-­‐ conducting   and   pyridine-­‐Au   linked   LUMO-­‐conducting   single-­‐molecule   junctions.   In   contrast   to   previous   measurements  in  which  thermal  and  electronic  properties  are  not  measured  on  the  same  junctions8-­‐11  or   have  been  measured  simultaneously  on  a   junction  containing  a  few  molecules15,  we  determine  both  the   conductance,   G,   and   the   Seebeck   coefficient,   S,   concurrently   for   an   individual   molecular   junction.   Conductance  values  are  obtained  by  measuring  the  current  across  the  gold-­‐molecule-­‐gold  junction  at  an   applied   bias   voltage   of   10   mV.   The   Seebeck   coefficient   on   the   same   junction   is   determined   from   the   measured   thermoelectric   current   through   the   junction   held   under   a   temperature   gradient   while   maintaining   a   zero   (externally-­‐applied)   bias   voltage   across   the   junction.  We   find   that   amine-­‐terminated   molecules   have   S>0,   suggesting   that   the   HOMO   resonance   is   closest   to   EF,   while   pyridine-­‐terminated   molecules   have   S<0,   indicating   that   the   LUMO   resonance   is   closest.   We   also   find   that   a   diphenylphosphine-­‐terminated  alkane  has  an  S  near  zero,   indicating  that  EF   is  probably  very  close  to  the   middle   of   the   HOMO-­‐LUMO   gap.   We   compare   our   measurements   with   first   principles   transport   calculations   that   are   based   on   standard   density   functional   theory   (DFT)   that   is   extended   to   incorporate   self-­‐energy  corrections16,  17,  and  find  quantitative  agreement  with  both  conductance  values  and  Seebeck   coefficients.   Our   calculations   confirm   the   sign   of   S   in   the   case   of   each   junction,   and   its   expected   relationship   with   the   frontier   orbital   closest   to   EF,   but   also   reveal   a   complex   transmission   for   amine-­‐ linked   junctions,   with   implications   for   the   relationship   between   S   and   G.   In   particular,   for   amine-­‐linked   junctions,  T(E)  is  non-­‐Lorentzian  and  thus  S  varies  significantly  more  than  G  from  junction  to  junction.     Single   molecule   junctions   are   created   using   the   scanning   tunneling   microscope-­‐based   break   junction   technique   (STM-­‐BJ),   in   which   a   sharp   gold   tip   is   brought   in   and   out   of   contact   with   a   gold   substrate   in   an   environment   of   a   target   molecule3,   18.   The   molecules   used   in   this   study   are   4,4’-­‐ diaminostilbene   (1),   bis-­‐(4-­‐aminophenyl)acetylene   (2),   1,5-­‐bis(diphenylphosphino)pentane   (3),   4,4’-­‐ bipyridine   (4)   and   1,2-­‐di(4-­‐pyridyl)ethylene   (5).   The   target   molecules   are   deposited   onto   the   STM   substrate   by   thermal   evaporation   under   ambient   conditions   (except   for   (3)   which   is   deposited   from   an   acetone   solution)   and   thus   measurements   are   not   carried   out   in   solvent.   A   thermal   gradient   is   applied   using   a   Peltier   heater   to   controllably   heat   the   substrate   to   temperatures   ranging   from   room   temperature   to   60oC   while   maintaining   the   tip   close   to   room   temperature,   and   the   thermoelectric   current   through   these   junctions   is   measured   at   zero   applied   bias.   The   set-­‐up   is   allowed   to   come   to   thermal  equilibrium  for  about  one  hour  at  each  temperature  before  measurements  are  continued.   In  all   measurements   reported   here,   the   temperature   difference   (ΔT)   between   the   tip   and   substrate   is   set   at   approximately   one   of   three   values:   0K,   14K,   27K.   In   order   minimize   unaccountable   thermoelectric   voltages   across   the   leads,   a   pure   gold  wire   is   used   to   apply   the   voltage   to   the   hot   substrate   and   is   also   connected  to  the  cold  side  of  the  peltier.    A  schematic  of  the  circuitry  as  well  as  the  STM  layout  is  shown   in  Figure  1A.     Conductance  and  thermoelectric  current  measurements  are  carried  out  using  a  modified  version   of   the   STM-­‐BJ   technique19.   Briefly,   the   Au   STM   tip   is   brought   in   contact   with   the   heated   Au   substrate   while  applying  a  bias  voltage  of  10  mV  until  a  conductance  greater  than  5  G0   is  measured.  The  tip   is  first   retracted   from   the   substrate   by   2.4   nm   at   a   speed   of   15.8   nm/s,   then   held   fixed   for   50   ms,   and   finally   withdrawn   an   additional   0.8   nm   as   illustrated   by   the   piezo   ramp   shown   in   Figure   1B.   During   this   ramp,   the   current   through   the   junction   and   the   voltage   across   the   junction   is   continuously   measured   at   a   40   kHz   acquisition   rate.   The   applied   bias   is   set   to   zero   during   the  middle   25  ms   of   the   50  ms   period  when   the   tip/substrate   distance   is   fixed   (Figure   1C).   For   every   molecule,   and   tip/sample   pair,   over   3600   measurements   are   collected   at   each   of   the   three   ΔT’s   listed   above.   For   each   measurement,   the   data   during   the   50  ms   hold   period   are   analyzed   further,   as   described   in   detail   in   the   SI.   Briefly,   the   junction   conductance   during   the   first   and   last   12.5   ms   of   the   “hold”   period   is   determined.   If   both   values   are   found  to  be  within  a  molecule  dependent  range  as  determined  from  a  conductance  histogram,  the  trace   is   selected,   for   this   indicates   that   a   molecular   junction   is   sustained   during   the   entire   50   ms   “hold”   period.  Conductance  histograms  for  4  and  5,  show  two  peaks  due  to  two  different  binding  geometries17.   Here,   we   analyze   these  molecular   junctions   in   the   high   conductance   configuration.   The   analysis   of   4   in   the   low-­‐conducting   configuration   is   included   in   the   SI.     Typically,   about   10-­‐20%   of   the  measured   traces   are   selected   for   analysis,   since   only   these   have   a   molecule   bridging   the   tip   and   substrate   during   the   “hold”  period  of   the   ramp.  A  single  selected  measurement   for  1   is  shown   in  Figure  1B,  where  measured   current   is   in   red   and  measured   voltage   is   in   blue.   The   junction   thermoelectric   current   is   determined   by   averaging   the  measured   junction  current  during   the  middle  25  ms  of   the  “hold”  period  and   the   junction   conductance   is  determined  by  averaging   the  conductance  during   the   first  and   last  12.5  ms  of   the  “hold”   period.   This   allows   a   determination   of   junction   conductance   and   thermoelectric   current   for   each   individual  junction  formed,  and  thus  allows  a  simultaneous  determination  of  G  and  S  for  each  junction.   In  Figure  2,  we  show  histograms  of  conductance  values  and  average   thermoelectric  currents   for   1   and   4   determined   on   a   trace   by   trace   basis   for   measurements   at   three   different   ΔT   values.   Conductance   and   thermoelectric   current   distributions   for   other   molecules   studied   are   shown   in   the   SI.   We   see   that   with   a   ΔT   of   0   K,   the   thermoelectric   current   histogram   is   narrow   and   centered   about   zero   for   both   molecules,   implying   that   on   average,   no   current   flows   without   an   applied   temperature   difference   between   the   tip   and   substrate,   consistent  with   the   expression   for   current   in   Equation   1.   For   molecule   1,   we   find   that   with   a   finite   ΔT,   the   thermoelectric   current   (from   the   tip   to   the   substrate)   is   negative   while   for   4   we   measure   a   positive   thermoelectric   current.   We   also   see   that   the   peak   of   the   thermoelectric   current   distributions   for   1   (4)   shifts   to   lower   (higher)   value   with   increasing   ΔT,   thus   the   magnitude  of  the  thermoelectric  current  increases  with  increasing  ΔT.     The   Seebeck   coefficient   for   the   entire   system,   SMeasured,   is   determined   by   first   substituting,   in   Equation   (1),   the   conductance   and   thermoelectric   currents   determined   for   each   measured   trace   (SMeasured=I/GΔT).   Since   the   system   includes   a   section   of   gold   wire   which   is   maintained   under   the   opposite   thermal   gradient   (-­‐ΔT),   the   Seebeck   coefficient   of   the   Au-­‐molecule-­‐Au   junction   is   given   by   SJunction=   SAu  -­‐   SMeasured,   where   we   use   SAu   =   2 µV/K20.     In   Figure   3,   we   show   the   distribution   of   molecular   Seebeck   coefficients   (when   the   molecules   are   attached   to   Au   electrodes)   determined   for   all   five   compounds   studied   by   including  measurements   at   all   ΔTs.   These   distributions   are   fit   to   a   Gaussian   and   the   peak   positions   are   given   in   Table   1,   along   with   measured   conductance   values.   We   use   the   most   probable  molecular  junction  conductance  and  Seebeck  coefficients  to  determine  a  power  factor,  GS2,  for   these  systems,  which  are  also  given   in  Table  1.  We  see  first  that  the  amine-­‐terminated  molecules  (1  and   2)  have  a  positive  Seebeck  coefficients  indicating  HOMO  conductance,  while  the  4  and  5  have  a  negative   Seebeck   coefficient   (LUMO-­‐conducting).   For   molecule   3,   although   we   measure   a   small   positive   SJunction,   the  magnitude  is  small  enough  to  conclude  that  for  this  alkane,  EF  is  very  close  to  mid-­‐gap.       To   understand   these   measurements,   we   use   first-­‐principles   calculations   with   a   self-­‐energy   corrected,   parameter-­‐free   scattering-­‐state   approach   based   on   density   functional   theory   (DFT)17,  21,  22   to   determine   both   the   linear   response   conductance   (Eq.   2)   and   the   Seebeck   coefficient   (Eq.   3).   Eq.   (3)   assumes   that   T(E)   varies   smoothly   for   E   −   EF   <   kBT,   and   that   ΔT   is   small   compared   to   T14,   23.   Both   assumptions   hold   for   the   systems   studied   here   (ΔT   <   30   K).   Moreover,   since   the   measured   thermoelectric   current   is   found   to   be   approximately   linear   with   ΔT   for   the   small   values   of   ΔT   in   the   experiments,  we  expect  the  steady-­‐state  scattering  formalism  will  also  be  valid24.   We  model  the  electrodes  using  two  Au(111)  slabs  with  7  layers  of  gold  for  the  leads,  using  a  4×4   unit   cell.   To   reduce   computational   burden   for   molecule   3,   we   replace   the   phenyl   groups   with   methyl   groups,  a   simplification   that  has  been   shown  not   to  affect   conductance  experimentally25.  Previous  work   established   that   for   both   amine-­‐   and   pyridine-­‐Au   linked  molecular   junctions   that   the   amine   or   pyridine   group   binds   selectively   to   undercoordinated   atop   Au   sites21,   22.   Accordingly,   we   use   two   different   undercoordinated  binding  site  motifs,  consisting  of  either  a  trimer  of  gold  or  an  adatom  to  represent  the   tip  to  which  the  molecules  bind16,  21.  All  junction  geometries  are  fully  relaxed  within  DFT-­‐GGA  (PBE)  using   SIESTA26.  Transmission  functions  are  calculated  using  a  self-­‐energy  corrected  scattering-­‐states  approach,   “DFT+Σ”,  to  the  Landauer  formula  as  implemented  in  the  Scarlet  code27  (see  SI  for  details).     Junction   structures   for   molecules   1   and   4   are   shown   in   Figure   4.   To   calculate   the   junction   transmission,   we   augment   the   Kohn-­‐Sham   excitation   energies   with   a  model   self-­‐energy   correction   that   has   consistently   led   to   quantitative   agreement   for   both   conductance17,   21,   22   and   Seebeck   coefficient16.   Specifically,   we   correct   the   gas-­‐phase   gap   with   a   ΔSCF28   calculation,   and   correct   for   the   lack   of   static   non-­‐local   correlation   effects   through   an   electrostatic   “image   charge”   model,   following   prior   work17.   G   and   S   are   then   determined   from   the   transmission   and   its   derivative   at   EF,   via   Eq.   2   and   3.   Numerical   evaluation   of   the   derivative   of   T(E)   generally   requires   a   very   fine   k-­‐point   sampling.   To   minimize   sampling   errors,   we   fit   T(E)   around   EF   with   a   smooth   function,   and   take   its   derivative   analytically.   Comparing   these   two   approaches   for   one   junction,   1,4-­‐benzenediamine-­‐Au16,   we   find   a   less   than   5%   difference   in   S   obtained   from   numerically   differentiating   T(E)   on   a   24×24   k-­‐grid   and   fitting   a   T(E)   calculated   on   an   8×8   k-­‐grid.   For   all   the   results   presented   here,   T(E)   is   computed   on   a   16x16   k-­‐grid,   with  S  determined  from  the  analytic  derivative  of  a  fit  to  the  transmission  function  around  EF.   The   transmission   curves   for   1   and   4   are   shown   in   Figure   4   (other   molecules   are   shown   in   SI   Figure   S7)   and   the   calculated   values   for   G   and   SJunction   are   reported   in   Table   1.   We   find   that   for   amine-­‐ linked   junctions,   T(EF)   originates  with   a  HOMO-­‐derived   peak,   and   a  weakly-­‐transmitting   feature   formed   from  a  hybridization  of  Au-­‐d  and  N-­‐lone  pair  states,   resulting   in  a  positive  S.  The  calculated  G  are  within   15%   of   the   experiments,   and   do   not   vary   much   with   the   coordination   of   the   Au   binding   site,   in   agreement   with   past   work   showing   that   junction   geometry   does   not   affect   conductance   significantly21,   29.  For  the  pyridine-­‐linked   junctions,  transmission  at  Fermi  results  from  a  LUMO  derived  peak,  hence  S   is   negative.   The   calculated   G   is   within   a   factor   of   3   of   the   experiment   but   varies   significantly   with   the   Au   binding   site   coordination,   again   in   agreement   with   past   work30.   The   calculated   Seebeck   coefficients   for   both  series  are  within  a  factor  of  2  of  the  experimental  values.   For   the   amine-­‐linked   junctions,   hybridization   between   the   HOMO   resonance   and   Au   d-­‐states   ~1.8   eV   below   EF   (cf.   arrow   in   Figure   4A)   results   in   an   appreciable   energy-­‐dependent   coupling,   and   the   single   Lorentzian   model   breaks   down.   Increasing   the   binding   site   coordination   significantly   alters   the   lineshape  and  the  Seebeck  coefficients  (15%  variation  for  molecule  1),  but  results  in  just  a  modest  rise  in   the   density   of   states   at   EF   and   small   changes   in   the   conductance   (3%   variation).   Interestingly,   S   is  more   sensitive   to   HOMO-­‐Au   5d   hybridization   features   in   T(E)   than   G.   Because   these   features   may   be   underestimated   by   approximations   associated   with   DFT+Σ,   deviations   from   experiment   may   be   somewhat  greater  for  S  than  for  G  for  HOMO-­‐conducting  amine-­‐Au  junctions.  For  molecules  1  and  2,  the   simultaneous  measurement  of  G  and  S  allows  an  assessment  of   the  efficacy  of  Lorentzian  models.   If   the   transmission   function   has   a   simple   Lorentzian   form   for   a   molecule   symmetrically   coupled   to   both   electrodes8,  9,   S   and  G  determine   the   resonance  energy   relative   to   EF.  However,   for   the   amine-­‐Au   linked   junctions   studied   here,   a   single-­‐Lorentzian   model   (see   SI)   significantly   underestimates   the   resonance   energy.  For  molecule  1,   the   single-­‐Lorentzian  model  would  place   the  HOMO  at  –1.1  eV  while  our  DFT+Σ   calculations   show   that   HOMO   is   around   –2.3   eV.   Thus   the   non-­‐Lorentzian   behavior   of   amine-­‐linked   junctions  observed  here,  and  seen  with  some  other   linkers7  could  allow  for  the  tuning  of   their  S  without   greatly  affecting  G  through  the  position  of  the  d-­‐states,  for  example  using  transition  metal  contacts  with   d-­‐states  closer  to  EF.     For  pyridine-­‐linked   junctions,   the  calculated  DFT+Σ   transmission   function  has  a  Lorentzian   form,     LUMO   resonance   away   from   EF.   In   a   single   Lorentzian   model   in   the   weak-­‐coupling   limit,   (Γ/ΔE)2<<1     with   a   prominent   resonance   with   LUMO   character.   Increasing   the   binding   site   coordination   pushes   the   (where  Γ   is  an  energy-­‐independent  coupling  or   injection  rate,  and  ΔE  the  difference  between  the  LUMO   resonance   energy   and   EF),   and   the   Seebeck   coefficient   varies   more   slowly   with   ΔE   (as   1/ΔE)   than   the   conductance   (1/ΔE2).   This   is   indeed   what   we   find   in   our   calculations.   For   molecule   4,   S   has   a   +/-­‐5%   variation   for   the   different   binding   sites,   while   the   conductance   changes   by   +/-­‐25%.   This   Lorentzian-­‐like   behavior   is   further   validated   by   the   estimate   of   the   resonance   positions,   in   close   agreement   with   first-­‐ principles  calculations  (1.53  eV  determined  using  the  experimental  S  and  G  compared  with  1.47  eV  from   our  calculations  of  molecule  4).     In   conclusion,   we   have   demonstrated   that   we   can   determine   the   conductance   and   thermoelectric   current   concurrently   through   single-­‐molecule   junctions.   The   thermoelectric   currents   are   used   to   determine   a   Seebeck   coefficient   for   each   junction.   We   find   that   amine-­‐terminated   molecular   junctions  have  a  positive  Seebeck  coefficient  in  agreement  with  calculations  that  show  that  the  HOMO  is   the  molecular   resonance   that   is   closest   to   EF.   In   contrast,   pyridine-­‐terminated  molecular   junctions   have   a   negative   Seebeck   coefficient   and   conduct   through   the   LUMO.   These   experimental   results   are   in   good   quantitative   agreement   with   those   from   self-­‐energy   corrected   DFT   calculations,   which   also   reveal   a   complex,  non-­‐Lorentzian  form  for  transmission  for  amine-­‐linked  junctions.   Acknowledgements:    We  thank  M.  S.  Hybertsen,  H.  J.  Choi,  and  S.  Y.  Quek  for  discussions  and  S.  Berkley   for   help   with   measurements.     This   work   was   supported   in   part   by   the   EFRC   program   of   the   U.S.   Department  of  Energy   (DOE)  under  Award  No.  DE-­‐SC0001085  and  the  ACS-­‐PRF  program.  L.V.  thanks  the   Packard   Foundation   for   support.     Portions   of   this   work   were   performed   at   the   Molecular   Foundry   and   within   the   Helios   Solar   Energy   Research   Center,   both   were   supported   by   the   Office   of   Basic   Energy   Sciences   of   the   U.S.   Department   of   Energy   under   Contract   No.   DE-­‐AC02-­‐05CH11231.   We   acknowledge   NERSC  for  computing  resources.     References:   1.   2.   Nitzan,  A.;  Ratner,  M.  A.  Science,  2003,  300,  (5624),  1384-­‐1389.   Chen,   F.;   Li,   X.   L.;   Hihath,   J.;   Huang,   Z.   F.;   Tao,   N.   J.   Journal   of   the   American   Chemical   Society,   2006,  128,  (49),  15874-­‐15881.   3.   Venkataraman,  L.;  Klare,  J.  E.;  Tam,   I.  W.;  Nuckolls,  C.;  Hybertsen,  M.  S.;  Steigerwald,  M.  L.  Nano   Letters,  2006,  6,  (3),  458  -­‐  462.   4.   Mishchenko,   A.;   Zotti,   L.   A.;   Vonlanthen,   D.;   Burkle,   M.;   Pauly,   F.;   Cuevas,   J.   C.;   Mayor,   M.;   Wandlowski,  T.  Journal  of  the  American  Chemical  Society,  2011,  133,  (2),  184-­‐187.   5.   Martin,  C.  A.;  Ding,  D.;  Sorensen,  J.  K.;  Bjornholm,  T.;  van  Ruitenbeek,  J.  M.;  van  der  Zantt,  H.  S.  J.   Journal  of  the  American  Chemical  Society,  2008,  130,  (40),  13198-­‐13199.   6.   Schneebeli,   S.   T.;   Kamenetska,   M.;   Cheng,   Z.;   Skouta,   R.;   Friesner,   R.   A.;   Venkataraman,   L.;   Breslow,  R.  Journal  of  the  American  Chemical  Society,  2011,  133,  (7),  2136-­‐2139.   7.   Cheng,  Z.  L.;  Skouta,  R.;  Vazquez,  H.;  Widawsky,  J.  R.;  Schneebeli,  S.;  Chen,  W.;  Hybertsen,  M.  S.;   Breslow,  R.;  Venkataraman,  L.  Nature  Nanotechnology,  2011,  6,  (6),  353-­‐357.   8.   Malen,   J.  A.;  Doak,  P.;  Baheti,  K.;  Tilley,  T.  D.;  Segalman,  R.  A.;  Majumdar,  A.  Nano  Letters,  2009,   9,  (3),  1164-­‐1169.   9.   Reddy,  P.;  Jang,  S.  Y.;  Segalman,  R.  A.;  Majumdar,  A.  Science,  2007,  315,  (5818),  1568-­‐1571.   10.   Baheti,   K.;  Malen,   J.   A.;  Doak,   P.;   Reddy,   P.;   Jang,   S.   Y.;   Tilley,   T.  D.;  Majumdar,   A.;   Segalman,   R.   A.  Nano  Letters,  2008,  8,  (2),  715-­‐719.   11.   Malen,   J.   A.;   Yee,   S.   K.;  Majumdar,   A.;   Segalman,   R.   A.   Chem.   Phys.   Lett.,   2010,   491,   (4-­‐6),   109-­‐ 122.   12.   13.   14.   15.   Dubi,  Y.;  Di  Ventra,  M.  Reviews  of  Modern  Physics,  2011,  83,  (1),  131-­‐155.   Ashcroft,  N.  W.;  Mermin,  N.  D.,  Solid  State  Physics.   Paulsson,  M.;  Datta,  S.  Physical  Review  B,  2003,  67,  (24),  241403.   Tan,  A.;  Balachandran,   J.;  Sadat,  S.;  Gavini,  V.;  Dunietz,  B.  D.;   Jang,  S.-­‐Y.;  Reddy,  P.   Journal  of   the   American  Chemical  Society,  133,  (23),  8838-­‐8841.   16.   17.   18.   19.   Quek,  S.  Y.  Q.  S.  Y.;  Choi,  H.  J.;  Louie,  S.  G.;  Neaton,  J.  B.  ACS  Nano,  2011,  5,  (1),  551-­‐557.   Quek,  S.  Y.;  Choi,  H.  J.;  Louie,  S.  G.;  Neaton,  J.  B.  Nano  Letters,  2009,  9,  (11),  3949-­‐3953.   Xu,  B.  Q.;  Tao,  N.  J.  Science,  2003,  301,  (5637),  1221-­‐1223.   Widawsky,   J.   R.;   Kamenetska,   M.;   Klare,   J.;   Nuckolls,   C.;   Steigerwald,   M.   L.;   Hybertsen,   M.   S.;   Venkataraman,  L.  Nanotechnology,  2009,  (43),  434009.   20.   21.   Blatt,  F.  J.,  Thermoelectric  Power  of  Metals.  Plenum  Press,  New  York:  1976.   Quek,   S.   Y.;   Venkataraman,   L.;   Choi,   H.   J.;   Loule,   S.   G.;   Hybertsen,   M.   S.;   Neaton,   J.   B.   Nano   Letters,  2007,  7,  (11),  3477-­‐3482.   22.   Quek,   S.   Y.;   Kamenetska,   M.;   Steigerwald,   M.   L.;   Choi,   H.   J.;   Louie,   S.   G.;   Hybertsen,   M.   S.;   Neaton,  J.  B.;  Venkataraman,  L.  Nature  Nanotechnology,  2009,  4,  (4),  230-­‐234.   23.   24.   Butcher,  P.  N.  Journal  of  Physics:  Condensed  Matter,  1990,  2,  (22),  4869.   Dubi,  Y.;  Di  Ventra,  M.  Nano  Letters,  2009,  9,  (1),  97-­‐101.   25.   Parameswaran,   R.;   Widawsky,   J.   R.;   Vazquez,   H.;   Park,   Y.   S.;   Boardman,   B.   M.;   Nuckolls,   C.;   Steigerwald,   M.   L.;   Hybertsen,   M.   S.;   Venkataraman,   L.   Journal   of   Physical   Chemistry   Letters,   2010,   1,   2114-­‐2119.   26.   Soler,   J.   M.;   Artacho,   E.;   Gale,   J.   D.;   Garcia,   A.;   Junquera,   J.;   Ordejon,   P.;   Sanchez-­‐Portal,   D.   Journal  of  Physics:  Condensed  Matter,  2002,  14,  (11),  2745-­‐2779.   27.   28.   2004.   29.   30.   Choi,  H.  J.;  Marvin,  L.  C.;  Steven,  G.  L.  Physical  Review  B,  2007,  76,  (15),  155420.   Martin,  R.,  Electronic  structure:  basic   theory  and  practical  methods.  Cambridge  University  Press:   Li,  Z.;  Kosov,  D.  S.  Physical  Review  B,  2007,  76,  (3),  035415.   Kamenetska,  M.;  Quek,  S.  Y.;  Whalley,  A.  C.;  Steigerwald,  M.  L.;  Choi,  H.  J.;  Louie,  S.  G.;  Nuckolls,   C.;   Hybertsen,   M.   S.;   Neaton,   J.   B.;   Venkataraman,   L.   Journal   of   the   American   Chemical   Society,   2010,   132,  (19),  6817-­‐6821.          Table 1 Molecule 1 2 3 4 5 GEXP (10-3 G0) 0.63 0.57 0.39 0.68 0.24 SEXP (µV/K) 13.0 (7.0) 9.7 (6.1) 1.1 (4.1) -9.5 (4.3) -12.3 (9.1) [GS2]EXP (10-18 W/K2) 8.25 4.16 0.037 4.76 2.81 GDFT+Σ (10-3 G0) 0.58 0.62 0.70 0.2 0.07 Table Caption SDFT+Σ (µV/K) 5.89 4.71 0.33 -7.88 -12.11 Table 1: List of molecular conductance, G, and Seebeck coefficient, S (with the HWHMs included in parentheses), determined experimentally and theoretically. Also shows the experimentally determined power factor, GS2. Figure Captions Figure 1: (a) Top panel: Simplified diagram illustrating measurement of thermoelectric current (IT). Bottom panel: Schematic of the STM-BJ set-up. (b) Top panel: Piezo ramp used, including a “hold” portion between 150 and 200 ms. Middle panel: External applied voltage across the leads which drops to zero during the center of the “hold” portion. Bottom panel: Sample trace for molecule 1. The measured current is shown in red and the voltage measured across the junction is shown in blue. Note: The voltage is applied across the junction in series with a 10 kΩ resistor. Figure 2: (a) Average measured conductance histograms for molecules 1 (top) and 4 (bottom), for the three ΔT’s (ΔT=0 K, green; ΔT=14 K, blue; and ΔT=27 K, red). (b) Average thermoelectric current histograms for molecules 1 (top) and 4 (bottom), for the three ΔT’s (ΔT=0 K, green; ΔT=14 K, blue; and ΔT=27 K, red). For 1, the thermoelectric current shifts left with increasing ΔT, while for 2, the thermoelectric current shifts right. Figure 3: Histograms of Seebeck coefficient for all molecules 1–5. Histograms are fit with Gaussians (red). 1 and 2 exhibit positive S while 4 and 5 exhibit negative S. The Seebeck coefficient for 3 is close to zero. Figure 4: Upper panel: The optimized geometries for junctions with molecules 1 (a) and 4 (b). Lower panel: Transmission curves shown on a log scale for both molecules calculated using DFT+Σ. Arrow indicates the position of the Au-d states. Insets: Transmission curves around EF on a linear scale. Figure 1 Figure 2 Figure 3 Figure 4 

1610.08402

2

1610

Time-resolved measurements of surface spin-wave pulses at millikelvin temperatures

In this work, we experimentally investigate the propagation of pulsed magnetostatic surface spin-wave (magnon) signals in an yttrium iron garnet (YIG) waveguide at millikelvin temperatures. Our measurements are performed in a dilution refrigerator at microwave frequencies. The excellent signal-to-noise ratio afforded by the low-temperature environment allows the propagation of the pulses to be observed in detail. The work gives insight both into low-temperature magnon dynamics in YIG and the potential application of systems of propagating magnons to solid-state quantum information processing.

cond-mat.mes-hall

cond-mat

Time-resolved measurements of surface spin-wave pulses at millikelvin temperatures A. F. van Loo,1 R. G. E. Morris,1 and A. D. Karenowska1 1Clarendon Laboratory, Department of Physics, University of Oxford, OX1 3PU, Oxford, United Kingdom (Dated: April 8, 2019) We experimentally investigate the propagation of pulsed magnetostatic surface spin-wave (magnon) signals in an yttrium iron garnet (YIG) waveguide at millikelvin temperatures. Our measurements are performed in a dilution refrigerator at microwave frequencies. The excellent signal-to-noise ratio afforded by the low-temperature environment allows the propagation of the pulses to be observed in detail. The evolution of the envelope shape as the spin-wave travels is found to be consistent with calculations based on the known dispersion relation for YIG. We ob- serve a temperature-dependent shift of the ferromagnetic resonance frequency below 4K which we suggest is due to the low-temperature properties of the substrate below the film, gallium gadolinium garnet. Our measurement and the accompanying calculations give insight into both low-temperature magnon dynamics in YIG and the feasibility of the use of propagating magnons in solid-state quan- tum information processing. The unusual and highly tunable dispersion of propa- gating magnons in yttrium iron garnet (Y3Fe5O12; YIG) has recently excited the interest of the superconduct- ing circuit quantum electrodynamics (QED) community. The frequencies at which superconducting quantum de- vices typically operate overlap with the band in which magnons can be excited in YIG and, as superconduct- ing quantum bit (qubit) structures are generally sensitive to electromagnetic fields, magnons and superconducting qubits can be made to communicate. So far, magnons in spheres of YIG have been shown to couple strongly to three-dimensional cavities [1 -- 4], reentrant cavities [5, 6], and superconducting qubits via such cavities [7]. Fur- thermore, magnons in YIG films have been shown to interact with superconducting coplanar waveguides and three-dimensional resonators [8, 9]. In this work, we investigate the propagation of mag- netostatic surface magnons in a YIG waveguide made from a high-purity monocrystalline film grown by liquid- phase epitaxy on a gallium gadolinium garnet (GGG) substrate. Our measurements are performed in a dilution refrigerator at approximately 20 mK. At this tempera- ture, thermal excitations of gigahertz-frequency magnons and photons are negligible. We previously investigated a similar system [10], focusing on achieving signal limits equivalent to the propagation of a single magnon. In this paper, we study the propagation of the excitations in significantly greater detail, comparing our experimen- tal measurements with the predictions of theory. Un- derstanding low-temperature magnon propagation is an essential step toward the integration of systems of propa- gating magnons with circuit QED systems. Work in this area sets its sights not only on the development of new quantum devices but also on the use of the sophisticated technology and methods developed for superconducting quantum computing to investigate the physics of single magnons. To excite propagating magnons in a waveguide, a mag- netic bias field must be applied. The dynamics of the particular modes that can be observed depend on the ori- entation of the bias field relative to the waveguide axis: in this study, we focus on magnetostatic surface spin waves (MSSWs; also called "Damon-Eshbach modes") that propagate perpendicular to an in-plane magnetic field. The dispersion relation for MSSWs [11] can be written as (cid:114) (1 − e−2kd) ω2 M 4 ω = ωH (ωH + ωM ) + (1) where ωM = −γµ0MS and ωH = −γB. Here, d is the film thickness, γ is the gyromagnetic ratio, µ0 is the vacuum permeability, B is the applied magnetic field, and MS −1 in our film the saturation magnetization (197.4 kA m at approximately 20 mK). From the dispersion relation, the group and phase velocities (vgr and vph respectively) can be derived: = d(cid:0)(2ωH + ωM )2 − 4ω2(cid:1) (cid:104) (−2ω+2ωH +ωM )(2(ω+ωH )+ωM ) √ 4 √ −2d ω2 ω2 ω2 M (2) (3) (cid:105) vgr = ∂ω ∂k vph = ω k = log The YIG waveguide used in our experiments is 9 µm thick, 2 mm wide, and approximately 15 mm long. To prevent coherent reflections of spin waves from the ends, both are cleaved at a 45 degree angle. The waveguide is affixed to a PCB in close contact with two lithographed stripline antennae [Fig. 1(a)] which are 50 µm wide and 6 mm apart. The complete assembly is housed in a copper sample box attached to the cold plate of a dilution refrigerator. A superconducting magnet is connected to the 4K plate of the refrigerator and posi- tioned so that the homogeneity of the field it produces 2 FIG. 1. (a) The YIG waveguide is attached to a printed cir- cuit board with two inductive antennae on its surface (an in- put and an output). The complete experimental assembly is housed in a copper box attached to the cold plate of a dilution refrigerator. (b) Measured transmission of a continuous-wave signal through the sample in a magnetic bias field of 167 mT. The dashed line indicates the position of the ferromagnetic resonance frequency. is maximal at the position of the waveguide. The orien- tation of the field relative to the waveguide [Fig. 1(a)] is such that only MSSWs are excited. Because of the non- reciprocal nature of these excitations, they are only able to travel in one direction on the film surface adjacent to the antennae: from the input port to the output port of the experiment [11, 12]. To ensure that room-temperature noise cannot couple into the waveguide, the line connected to the input an- tenna is heavily attenuated inside the refrigerator [Fig. 2]. The output antenna is connected to a low-temperature HEMT amplifier via two isolators. The amplified signal is filtered and down-converted to an intermediate frequency of 500 MHz at room temperature. A fast data acquisition card (Spectrum M4i-2234-x8) records the resulting volt- age as a function of time at a rate of 2.5 GHz for 800 ns. Typically 211 to 216 such traces are recorded and aver- aged, although much higher numbers are can be used if desired (e.g. when the input signal is reduced to very low levels). Through this step, incoherent and inelastically scattered signals cancel out, leaving only the coherently scattered radiation at the drive frequency. After a dig- ital down-conversion and filtering step, the envelope of the transmitted microwave signal is recovered. When a microwave-frequency signal is applied to the input antenna, there are two routes by which it can reach the output: via the vacuum in the sample box as mi- crowave photons at the speed of light c, and as MSSWs in the magnetic waveguide at the group velocity deter- mined by eq. 2. The MSSW group velocity is typically several orders of magnitude slower than the speed of light. To perform an initial characterization of the YIG film, we connect the cold setup [center part in Fig. 2] to a network analyzer and measure the transmission (S21) of the experimental system between 4 and 8 GHz. Results from such a measurement are shown in Fig. 1(b). The low-frequency limit of the magnon passband is the ferro- FIG. 2. The microwave setup used in our experiments can be divided into three parts. In the up-conversion section, short microwave pulses are created by mixing the output of a continuous-wave microwave source with envelope shapes gen- erated by an arbitrary-waveform generator. Inside the dilu- tion refrigerator, the input line to the sample is heavily atten- uated to thermalize the signal, such that the electronic noise temperature is decreased to a level similar to the phonon noise temperature. The down-conversion system allows the recovery of the envelope of the signal transmitted through the waveguide. magnetic resonance (FMR) frequency (around 7.4 GHz in Fig. 1(b)); this is the k = 0 mode of the magnon system in which the spins throughout the YIG film precess in- phase. The oscillations in the signal transmittance as a function of frequency are caused by interference between the vacuum and magnon signals. Having performed our initial characterization in the continuous-wave regime, we investigate the dynamics of the magnetic system by driving it with short microwave pulses. The difference in propagation speed between the MSSW and the signal propagating through the vacuum here works to our advantage, allowing us to separate these two responses in time for sufficiently short pulses. The short microwave pulses also have a relatively wide spectral bandwidth, and as such allow the investigation of a range of magnetostatic surface spin-waves with dif- ferent wave numbers at a single magnetic field. Figure 2 shows the experimental setup used to generate and mea- sure the transmittance of short microwave pulses through the YIG sample. Figure 3a shows the spin-wave output signal in re- sponse to the application of a square 30 ns microwave pulse of constant carrier frequency (7 GHz) as the mag- netic bias field is swept across the magnon band. In oder to investigate magnons with a variety of k-values, we can either keep the bias field constant and vary the carrier frequency, or keep the carrier frequency constant and sweep the bias field. Here we choose to sweep the mag- net current at a fixed carrier frequency to eliminate the effects that the frequency-dependent attenuation of the microwave lines and components have on the data. The horizontal stripe of high signal intensity from 0 to 30 ns is due to direct electromagnetic interaction between the two antennae. The magnon signal begins to appear at LO10mK100mK4Kw300KRFSpectrumM4i-2234Up-conversionColdwsetupDownconversionArbitraryWaveGenerator70KLowpassIQRFLO-20dB-20dB-20dBSample50Ω50ΩIQRFLO50ΩBandpassapproximately 70 ns. For each vertical slice, the differ- ence in shape between the initial pulse and the magnon response is due to the different frequency components within the pulse traveling at different velocities, given by the dispersion relation. For each slice, therefore, we sample the dispersion relation with a simultaneous range of frequencies given by the spectrum of our input pulse, which is limited by the 300 MHz output bandwidth of our arbitrary-waveform generator. The field at which FMR occurs is 155 mT. For clarity, the FMR position is in- dicated in Fig. 3. When the field is reduced below that corresponding to the FMR frequency (moving from right to left along the horizontal axis in Fig. 3), the carrier fre- quency in the pulse excites modes of increasingly higher k which have a lower group velocity [Eq. 2] and therefore take longer to reach the output antenna. As the field is reduced further, the region of the dispersion relation sampled by the input pulse is shifted. The power used for these measurements is on the order of 105 photons per pulse at the input antenna. We previously studied the possibility of reducing the power to the single-magnon power level in our setup for both thin films [10] and YIG spheres[13]. Figure 3(b) shows the results of a simulation of the response of the experimental system calculated on the basis of Eq. 1. These calculations are performed as fol- lows: Like the experimental pulse, the simulated input signal is constructed by our multiplying a continuous- wave carrier signal with a 30 ns square envelope. A very small offset is added to the square pulse (before multi- plication) to account for imperfect mixer calibration in the experiment. This input signal is Fourier transformed and, for each frequency component, the phase velocity is calculated. The arrival time of each frequency com- ponent at the output antenna is then determined from the phase velocity and the distance between the anten- nae. A delay in time is equivalent to a phase shift in Fourier space, and thus the k-dependent travel time is implemented by multiplying the Fourier component by e−iωd/vp , d being the inter-antenna distance (6 mm for the sample in Fig. 3), and vp the phase velocity of the magnon at that particular frequency. The propagation loss of surface waves is expected to be a function of time only [14]. Therefore, waves of different k values expe- rience different amounts of loss when traveling the same inter-antenna distance because of their different phase ve- locities. This is taken into account in the simulations by multiplying the Fourier components by e−d/(vpTL), where TL is the characteristic frequency-independent [14] loss time. The time-dependent signal at the output antenna is reconstructed via an inverse Fourier transform, after which the analog and digital signal processing applied to the data as described above is mimicked. To avoid over- fitting, the antennae and sample box are treated as being ideal, save for a k-dependent coupling factor between the antennae and the YIG, which we take, according to [14], 3 FIG. 3. Measurements and simulations of pulsed propagating magnon signals. (a) Experimental data showing the transmit- ted signal amplitude due to a square 30 ns microwave pulse with a carrier frequency of 7 GHz. The magnetic bias field (horizontal axis) is swept across the magnon band. (b) Re- sults of a simulation under the same conditions. to be of the form J0(kw/2), where J0 is the zeroth-order Bessel function. The ratio between antenna-YIG cou- pling and antenna-antenna coupling is tuned manually to resemble the data. As predicted by eq. 2, the speed of the magnons re- duces as their wavenumber increases, resulting in an upward curve of the magnon signal on the left side of Figs. 3(a) and 3(b). For these data, the bandwidth of the magnon signal (measured by the range of magnetic fields for which we observe a magnon signal at a con- stant frequency) is not limited by the bandwidth of the antenna. Rather, the width of the measurable magnon band is limited by the propagation loss rate of magnons traveling in YIG. This was confirmed by measuring a sim- ilar system with a smaller inter-antenna distance that is found to have a higher signal bandwidth. Fitting our simulations to the data, we find that the characteristic loss time TL = 0.85 ± .05 ns. Note that this does not mean that magnons typically decay within this time: TL is referenced to the phase rather than the group velocity and is used to model the different amounts of loss that  -- by virtue of the different phase velocities -- the differ- ent frequency components are subject to. The typical timescale for the pulse to lose 1/e of its energy is approx- imately 30 ns. Overall, it can be seen that there is good agreement between experiment and theory, especially re- garding the structure and shape of the magnon response to the left of the FMR in the plot. There are, however, also some differences that warrant discussion. One difference between simulation and experiment is the vertical line close to the FMR frequency seen only in the theoretical plot. This line is an artifact of the sim- ulation that comes about from the phase velocity tend- ing to infinity as k reduces to zero. In the real physi- cal system the phase velocity diverges close to the FMR frequency, producing a signal between the magnon and the directly-coupled pulses. We also see some ringing in the experiments, appearing as low-amplitude oscilla- tions following the directly-coupled signal. This feature is not exactly reproduced in the simulations and occurs as a result of unmodeled nonidealities of the antennae and sample box. A further difference exists at magnetic fields above 155 mT: in Fig. 3(a) there is a structure in the measurement that is not reproduced in the simula- tions. This signal disappears when the magnetic bias field is reversed, confirming it must be due to a surface mode rather than a (reciprocal) backward or forward volume excitation. We have investigated several effects as poten- tial origins of this signal, including the finite width of the waveguide and the inhomogeneity of the magnetic bias field. To exclude the effect of a finite waveguide width as an explanation, we used a modified dispersion relation given in Ref. [15] to calculate the phase velocity in our simulations. An inhomogeneous magnetic bias field was implemented in our model by performing the simulations for a small range of fields simultaneously, mimicking the effects of different parts of the YIG film being subject to different bias fields. Neither of these alterations to the simulations produced the signal observed in Fig. 3 at fields above 155 mT. Effects due to the presence of a ground plane close to the sample, which can be calcu- lated from Ref. [16], are expected to be very small for our experimental setup, and likewise did not produce the observed signal when they were included in the simula- tions. A possible explanation for this sub-FMR signal is that the inhomogeneous magnetic fields close to the edges of the waveguide cause localized modes [17]. It should be noted that evidence of this signal is also found in the continuous-wave experiment [to the left of the FMR line in Fig. 1b], and similar signals have also been reported in room-temperature data in the literature (e.g., Ref. [18]). Data such as shown in Fig. 3(a) were obtained at sev- eral carrier frequencies and powers across the measure- ment bandwidth of our system (between 4 and 8 GHz). Figure 3 is representative in both the agreements and disagreements between the experiments and simulations. Our microwave setup can be used to inject pulses 4 of arbitrary envelope into the experimental system. In Fig. 4(a), we see the output signal in response to a Gaus- sian input with σ = 12 ns, and in Fig. 4(b), a train of three 20 ns pulses separated by 10 ns. In Fig. 4(b), the signal produced by the direct coupling of the last of the input pulses from the input antenna to the output an- tenna overlaps with the first magnon pulse. The result is an interference pattern akin to that seen in Fig. 1(b). The clear separation between the individual pulses in the train underlines the promising potential of magnon sys- tems as a platform for the transmission of information signals. FIG. 4. Experimental data showing the propagation of (a) a Gaussian pulse with σ = 12 ns and (b) a train of three 20 ns pulses separated by 10 ns. When comparing the FMR frequency with the fre- quency predicted by Eq. 1, we find a slight difference be- tween observation and theory. Repeating measurements similar to those shown in Fig. 1(b) for different magnetic fields, we find that the frequency difference as a function of the magnetic field is well fitted with a linear function. This shift in frequency could be due to the field at the YIG being slightly different from the field that we apply, or the YIG FMR frequency being different from that pre- dicted by Eq. 1. If we assume that the field at the YIG is different from the applied field, we can use the dif- ference between the expected and observed FMR to ex- tract the offset magnetic field needed to explain this fre- quency difference. The total magnetic field is then given by Btot = Bmag + αBmag + β, with α = −0.052 ± 0.002 and β = −0.0112 ± 0.0002T. These values for α and β are extracted from data taken at base temperature (ap- proximately 20 mK). The same procedure is used over a range of temperatures up to 10 K in our dilution refriger- ator. The results are shown in Fig. 5. As can be seen in the inset, the values for α and β do not measurably vary up to 0.5 K. Beyond that, the magnitude of both fitted parameters reduces, seemingly converging toward zero at higher temperatures. The value of the YIG's saturation magnetization is expected to not change significantly at these temperatures [19, 20], and is here taken to be con- stant at 197.4 kA/m, . The values in the horizontal axes of Fig. 3(a) and Fig. 4 were adjusted such that Btot is shown rather than Bmag. Our prime suspect for the measured deviation of the FMR frequency is the substrate for the YIG film, GGG. While GGG is a convenient magnetically inert material to use for spin-wave experiments at room temperature, at cryogenic temperatures down to 1.5 K, it is param- agnetic, and displays Curie-Weiss susceptibility [21]. In this regime the short relaxation time of the paramag- netic GGG dampens spin-wave propagation of YIG de- posited on its surface. The exchange interaction be- tween neighboring spins in paramagnetic GGG is approx- imately 1.5 K/kB [22], and its behavior is therefore ex- pected to be different below this temperature but is not well documented. There are suggestions that GGG un- dergoes a spin glass transition around 200 mK [23] or en- ters a complex magnetic state involving locally correlated spin loops [24]. While it is beyond the scope of this work to establish the low-temperature behaviour of GGG, as- suming the observed shift in FMR to be caused by this material, we can make a few observations. First, α being nonzero should not surprise us as GGG is known to be- come paramagnetic at lower temperatures. The nonzero value of β, however, is more interesting: this would im- ply a magnetic ordering of the GGG substrate. Note that our measurements were not extended to zero field as we cannot measure the FMR frequency outside 4 − 8 GHz, the band of our measurement setup. We suggest that the relatively high measured loss ex- perienced by the magnon signal during propagation is related to the observed low-temperature magnetic prop- erties of GGG. Coupling YIG to a magnetic material with higher spin-wave damping certainly would be expected to result in such an effect. It should be noted that we ignore anisotropy in these considerations. Effects due to anisotropy are generally expected to be too small to account for the irregularities we record in our measure- ments, but as far as we are aware, this has not been experimentally confirmed at these temperatures. In conclusion, in this work we present a detailed study of the propagation of pulsed magnetostatic surface spin- 5 FIG. 5. At each temperature, the deviation of the observed FMR frequency from the theoretical value can be fitted as a linear function of the applied magnetic field. The offset and gradient of these linear functions are plotted here for a range of different temperatures. wave signals in an yttrium iron garnet waveguide at mil- likelvin temperatures. Calculations using a simple model based on the dispersion relation of MSSWs agree well with our experimental findings. Our investigations con- firm that trains of short spin-wave pulses can readily be excited and detected inductively in YIG waveguides grown on gallium gadolinium garnet substrates. We ob- serve a temperature- and field-dependent deviation from the theoretically expected FMR frequency, which we at- tribute to the low-temperature magnetic ordering in the substrate GGG. As well as being of interest in their own right, these results are an important step toward the combination of systems of propagating magnons with mi- crowave superconducting quantum circuit technology. The authors acknowledge Bob Watkins for preparing the waveguide and the EPSRC for funding this research with Grant EP/K032690/1. [1] Y. Tabuchi, S. Ishino, T. Ishikawa, R. Yamazaki, K. Us- ami, and Y. Nakamura, Phys. Rev. Lett. 113, 083603 (2014). [2] J. Bourhill, N. Kostylev, M. Goryachev, D. L. Creedon, and M. E. Tobar, Phys. Rev. B 93, 144420 (2016). [3] X. Zhang, C.-L. Zou, L. Jiang, and H. X. Tang, Phys. Rev. Lett. 113, 156401 (2014). [4] D. Zhang, X.-M. Wang, T.-F. Li, X.-Q. Luo, W. Wu, and J. Q. You, Npj Quantum Inf. 1, 15014 F. Nori, (2015). [5] M. Goryachev, W. G. Farr, D. L. Creedon, Y. Fan, M. Kostylev, and M. E. Tobar, Phys. Rev. Applied 2, 054002 (2014). [6] N. Kostylev, M. Goryachev, bar, Appl. Phys. Lett. 108, and M. E. To- (2016), 062402 0246810Temperature (K)0.060.050.040.030.020.010.00Gradient Gradient121086420Offset (mT)Offset0.00.20.40.60.0580.0540.0500.04612108http://dx.doi.org/10.1063/1.4941730. 4886 (1978). [7] Y. Tabuchi, S. Ishino, A. Noguchi, T. Ishikawa, R. Ya- mazaki, K. Usami, and Y. Nakamura, Science (2015), 10.1126/science.aaa3693. [16] T. Yukawa, J. ichi Yamada, K. Abe, and J. ichi Ikenoue, Jpn. J. Appl. Phys. 16, 2187 (1977). [17] K. Y. Guslienko, R. W. Chantrell, and A. N. Slavin, 6 [8] H. Huebl, C. W. Zollitsch, J. Lotze, F. Hocke, M. Greifen- stein, A. Marx, R. Gross, and S. T. B. Goennenwein, Phys. Rev. Lett. 111, 127003 (2013). Jiang, 119, and H. X. (2016), [9] X. Zhang, C. Zou, L. Phys. Tang, http://dx.doi.org/10.1063/1.4939134. J. Appl. 023905 Phys. Rev. B 68, 024422 (2003). [18] A. A. Serga, A. V. Chumak, and B. Hillebrands, Journal of Physics D: Applied Physics 43, 264002 (2010). [19] E. E. Anderson, Phys. Rev. 134, A1581 (1964). [20] P. Hansen, P. Roeschmann, and W. Tolksdorf, J. Appl. Phys. 45, 2728 (1974). [10] A. D. Karenowska, A. D. Patterson, M. J. Peterer, E. B. Magn´usson, and P. J. Leek, arXiv:1502.06263 (2015). [11] B. A. Kalinikos, Sov. Phys. J. 24, 718 (1981). [12] A. Gurevich and G. Melkov, Magnetization Oscillations [21] P. Schiffer, A. P. Ramirez, D. A. Huse, P. L. Gammel, U. Yaron, D. J. Bishop, and A. J. Valentino, Phys. Rev. Lett. 74, 2379 (1995). [22] P. Schiffer, A. P. Ramirez, D. A. Huse, and A. J. and Waves (CRC Press, Inc, 1996). Valentino, Phys. Rev. Lett. 73, 2500 (1994). [13] R. G. E. Morris, A. F. van Loo, S. Kosen, and A. D. [23] P. P. Deen, O. Florea, E. Lhotel, and H. Jacobsen, Phys. Karenowska, Scientific Reports 7, 11511 (2017). Rev. B 91 (2015), 10.1103/physrevb.91.014419. [14] D. D. Stancil and A. Prabhakar, Spin Waves (Springer Nature, 2009). [15] T. W. O'Keeffe and R. W. Patterson, J. Appl. Phys. 49, [24] J. A. M. Paddison, H. Jacobsen, O. A. Petrenko, M. T. Fernandez-Diaz, P. P. Deen, and A. L. Goodwin, Science 350, 179 (2015). 

1810.12040

1

1810

Optimized electrical control of a Si/SiGe spin qubit in the presence of an induced frequency shift

Electron spins confined in quantum dots are an attractive system to realize high-fidelity qubits owing to their long coherence time. With the prolonged spin coherence time, however, the control fidelity can be limited by systematic errors rather than decoherence, making characterization and suppression of their influence crucial for further improvement. Here we report that the control fidelity of Si/SiGe spin qubits can be limited by the microwave-induced frequency shift of electric dipole spin resonance and it can be improved by optimization of control pulses. As we increase the control microwave amplitude, we observe a shift of the qubit resonance frequency, in addition to the increasing Rabi frequency. We reveal that this limits control fidelity with a conventional amplitude-modulated microwave pulse below 99.8%. In order to achieve a gate fidelity > 99.9%, we introduce a quadrature control method, and validate this approach experimentally by randomized benchmarking. Our finding facilitates realization of an ultra-high fidelity qubit with electron spins in quantum dots.

cond-mat.mes-hall

cond-mat

Optimized electrical control of a Si/SiGe spin qubit in the presence of an induced frequency shift K. Takeda,1 J. Yoneda,1 T. Otsuka,1, 2, 3 T. Nakajima,1 M. R. Delbecq,1, 4 G. Allison,1 Y. Hoshi,5 N. Usami,6 K. M. Itoh,7 S. Oda,8 T. Kodera9 and S. Tarucha1, 10 1. RIKEN, Center for Emergent Matter Science (CEMS), Wako-shi, Saitama, 351-0198, Japan 2. JST, PRESTO, 4-1-8 Honcho, Kawaguchi, Saitama 332-0012, Japan 3. Research Institute of Electrical Communication, Tohoku University, 2-1-1 Katahira, Aoba-ku, Sendai, 980-8577, Japan 4. Laboratoire Pierre Aigrain, Ecole Normale Supérieure-PSL Research University, CNRS, Université Pierre et Marie Curie-Sorbonne Universités, Université Paris Diderot-Sorbonne Paris Cité, 24 rue Lhomond, 75231 Paris Cedex 05, France 5. Advanced Research Laboratories, Tokyo City University, 8-15-1 Todoroki, Setagaya- ku, Tokyo 158-0082, Japan 6. Graduate School of Engineering, Nagoya University, Nagoya 464-8603, Japan 7. Department of Applied Physics and Physico-Informatics, Keio University, Hiyoshi, Yokohama 223-8522, Japan 8. Department of Physical Electronics and Quantum Nanoelectronics Research Center, Tokyo Institute of Technology, O-okayama, Meguro-ku, Tokyo 152-8552, Japan 9. Department of Electrical and Electronic Engineering, Tokyo Institute of Technology, O-okayama, Meguro-ku, Tokyo 152-8552, Japan 10. Department of Applied Physics, The University of Tokyo, Hongo, Bunkyo-ku, Tokyo, 113-8656, Japan Correspondence: K. Takeda ([email protected]) or S. Tarucha ([email protected] tokyo.ac.jp) 1 Abstract Electron spins confined in quantum dots are an attractive system to realize high-fidelity qubits owing to their long coherence time. With the prolonged spin coherence time, however, the control fidelity can be limited by systematic errors rather than decoherence, making characterization and suppression of their influence crucial for further improvement. Here we report that the control fidelity of Si/SiGe spin qubits can be limited by the microwave-induced frequency shift of electric dipole spin resonance and it can be improved by optimization of control pulses. As we increase the control microwave amplitude, we observe a shift of the qubit resonance frequency, in addition to the increasing Rabi frequency. We reveal that this limits control fidelity with a conventional amplitude-modulated microwave pulse below 99.8%. In order to achieve a gate fidelity > 99.9%, we introduce a quadrature control method, and validate this approach experimentally by randomized benchmarking. Our finding facilitates realization of an ultra-high fidelity qubit with electron spins in quantum dots. Introduction Electron spins confined in semiconductor quantum dots provide an excellent platform for scalable solid-state quantum computing [1]. Quantum operations including single- spin rotation [2-4] and two-spin entanglement control [5-7] have been realized in the past. The control fidelities for single- [8-12] and two-qubit gates [13-16] have been largely improved by recent technical advancements in extending the spin coherence time. The single-qubit control fidelities have already reached the level close to or exceeding the threshold value required for implementing fault-tolerant logical qubits in the surface code structure [8, 10-16]. As the qubit performance improves, one needs to challenge the simplified view that relates spin qubit control fidelity solely to the ratio between the dephasing rate and the operation speed, since unitary errors such as pulse-induced effects can also be relevant. This problem has never been addressed, however, for quantum-dot qubits with a single electron spin 1/2 forming a natural two level system, in contrast to some other qubit systems where it is widely recognized (e.g. a.c. Stark shift and state leakage for transmons [17-19]). Such an approach may facilitate rapid single-qubit gates with fidelities high enough for fault-tolerant universal quantum operations [20], where multiple single-qubit gates are commonly involved for a two-qubit gate implementation. In addition, it is also important for precise qubit error metrology based on quantum tomography, which usually relies on single-qubit control for precise state preparation 2 and measurement. Here we report the observation and correction of microwave pulse induced systematic qubit errors in quantum-dot spin qubits. The spin qubit used in this work is defined in Si/SiGe quantum dots with a cobalt micro-magnet [21]. When the microwave burst is applied, in addition to the expected spin rotation, we observe an unexpected shift of the spin resonance frequency. While the frequency shift is typically an order of magnitude smaller than the Rabi frequency (𝑓Rabi ), it is much larger than the spin resonance linewidth and therefore causes a systematic error in the qubit rotation axis. This will limit the single-qubit control fidelity to 99.8 % according to our numerical simulations with realistic experimental parameters. To mitigate this problem and achieve high- fidelity, we introduce a quadrature microwave control which corrects the phase error of the qubit. The improvement of the qubit fidelity is experimentally confirmed by randomized benchmarking [22]. Results The quantum dots used here are formed by locally depleting a two-dimensional electron gas in an undoped Si/SiGe heterostructure using lithographically defined electrostatic gates (Fig. 1a). We measure two devices, A and B, with a nominally identical structure except for the quantum well materials to characterize sample-to-sample dependence. The quantum well in device A has a natural isotopic composition [10] and for device B it consists of isotopically enriched silicon with approximately 800 ppm 29Si [12]. An on-chip cobalt micro-magnet induces the magnetic field gradient across the quantum dot [21]. A nearby sensor quantum dot coupled to a radio-frequency tank circuit allows rapid measurement of the quantum dot charge configuration [23]. All measurements were performed at an electron temperature of approximately 120 mK in a dilution refrigerator with an in-plane external magnetic field 𝐵ext. The spin state is read out in a single-shot manner using an energy-selective spin-to-charge conversion [24]. We use a quantum dot formed in the left (right) side of the device for device A (B). The expected lithographical dot position is shown as the blue (red) circle in Fig. 1a. Figure 1b shows the pulse sequence for the spin control. First, a spin-down electron is prepared by applying gate voltages such that only the spin-down electron can tunnel into the dot. Next, the gate voltages are pulsed such that the electron confined in the dot is pushed deep in Coulomb blockade. Then, a microwave burst with a frequency of 𝑓MW is applied to gate C to induce electric dipole spin resonance (EDSR). Finally, the gate 3 voltages are pulsed back to the spin readout position where only a spin-up electron can tunnel out to the reservoir. When the microwave burst is applied to the gate, the electrons confined in the dot oscillate spatially in the slanting magnetic field induced by the micro-magnet, resulting in an effective oscillating magnetic field 𝐵AC perpendicular MM. At the condition where ℎ𝑓MW = 𝑔𝜇B𝐵0 (𝑔 is to the static magnetic field 𝐵0 = 𝐵ext + 𝐵𝑧 the electron 𝑔 -factor and 𝜇B is the Bohr magneton), EDSR takes place. The ∗ ~ 1.8 μs for device A inhomogeneous dephasing time of each qubit is estimated to be 𝑇2 ∗ ~ 20 μs for device B [12] from the Gaussian decay of the Ramsey fringe [10] and 𝑇2 H ∼ 11 μs (the associated amplitude. In addition, device A has a Hahn echo decay time 𝑇2 measurement result is available in Supplementary Section 2) and device B has a Hahn echo decay time 𝑇2 H ∼ 99 μs [12]. The effect of strong EDSR microwave pulses can be readily observed in the microwave frequency dependence of the Rabi oscillations. Figure 1c shows the Rabi oscillation measured in device A with 3 different microwave amplitudes. 𝑃↑ is the spin-up probability obtained by averaging 500 to 1,000 single-shot measurement outcomes. The applied microwave burst has a rectangular envelope with an amplitude that is denoted by 𝐴MW = 0.3√𝑃(𝑓MW)/𝑃0(𝑓MW), where 𝑃(𝑓MW) is the microwave power and 𝑃0(𝑓MW) is the microwave power corresponding to 𝑓Rabi = 10 MHz. The definition results in a normalized microwave amplitude of 𝐴MW = 0.3 at 𝑓Rabi = 10 MHz. For the smallest microwave amplitude (𝐴MW = 0.1), the resonance frequency is almost at the center of the image (𝑓MW = 15.748 GHz, indicated by the red arrows). However, when 𝐴MW is increased to 0.3, the center resonance frequency moves to higher frequencies. This frequency shift is further enhanced by increasing the microwave amplitude (∼5 MHz frequency shift for 𝐴MW = 0.6). To quantify the resonance frequency shift Δ𝑓 more precisely, we perform a modified Ramsey interference measurement with an off-resonance microwave burst (Fig. 2a). It is worth noting that this measurement can also check whether the shift occurs only on resonance or not. During the waiting time 𝑡w between two resonant Xπ/2 pulses, we apply an additional off-resonance microwave burst at a frequency of 𝑓MW = 𝑓res − 180 MHz, where 𝑓res = 𝑔𝜇B𝐵0/ℎ is the bare qubit resonance frequency in the weak driving limit. When the qubit precession frequency shifts due to the off-resonance microwave burst, the oscillation period of the Ramsey fringe changes. Figure 2b shows the frequency shift Δ𝑓 for device B measured for various 𝐴MW. Each data point is obtained by fitting the Ramsey oscillations using a sinusoidal function 𝑃↑(𝑡) = 𝐴sin (2πΔ𝑓 + 𝜂) + 𝐵 with 𝐴, 4 𝐵, 𝜂 and Δ𝑓 as fitting parameters as shown in Fig. 2c (the data for device A is available 𝑏 in Supplementary Section 3). We find that an empirical power-law relation Δ𝑓 = 𝑎𝐴MW fits well with the experimental data for both devices, however, the fitting parameters 𝑎 and 𝑏 are distinctively different between them. This may indicate that the frequency shift is related to some uncontrolled sample dependent parameters (e.g. local confinement potentials, defects etc.). We obtain the exponents 𝑏 = 1.39 ± 0.02 for device A (data shown in Fig. S3) and 𝑏 = 0.59 ± 0.03 for device B. Moreover, it is found that Δ𝑓 is positive (𝑎 > 0) for device A, while it is negative (𝑎 < 0) for device B. An additional striking feature of the frequency shift is observed in the post microwave burst response. We find that, even after the microwave burst is turned off, the qubit resonance frequency shift remains and causes an additional qubit phase accumulation. To quantify this, the qubit phase accumulated after a microwave burst is extracted from a Hahn echo type measurement. Here we utilize a modified Hahn echo sequence which consists of two π/2 pulses, a π pulse and an additional 200 ns off-resonance microwave burst (Fig. 3a). The off-resonance microwave burst is interleaved in between the π pulse and the second π/2 pulse. The phase of the second π/2 pulse is modulated by 𝜙 to extract the echo phase 𝜃(𝑡d). The post pulse delay time 𝑡d indicates the time interval between the off-resonance microwave burst and the second π/2 pulse. The evolution time between the π/2 pulses and the π pulse is fixed to 20 s to cancel out the unwanted phase fluctuation caused by quasi-static noise. Figure 3b shows the post pulse time dependence of the echo signal. Figure 3c shows the extracted echo phase evolution after the microwave burst application. For 𝐴MW = 0 , the black solid line shows an average of the blue data points, while for 𝐴MW = 0.15, the black solid curve shows a fitting curve with an exponential function 𝜃(𝑡d) = 𝐶exp(−𝑡d/𝜏) + 𝐷 with 𝐶, 𝜏, and 𝐷 as fitting parameters, giving a characteristic decay time of 𝜏 = 6 s. For both cases, the offset at 𝑡d = 0 is mainly caused by the post-pulse phase accumulation due to the on- resonance pulses. From the measured qubit phase accumulation 𝜃(𝑡d), the temporal post microwave burst frequency shift Δ𝑓(𝑡d) = (1/2π)(d𝜃(𝑡d)/d𝑡d) can be obtained (Fig.3d). The green points show numerical derivative obtained from the data points in Fig. 3c. The black solid line shows an exponential fitting curve. Although the single exponential function fits the measured phase data well for 𝑡d ≥ 0.3 s, Δ𝑓(𝑡d = 0) ∼ −80 kHz derived from the single exponential dependence extrapolation does not match the value estimated from the fitting curve to the continuous-wave response derived from Fig. 2b (Δ𝑓(𝑡d = 0) ∼ −320 kHz with 𝐴MW = 0.15). We also note that the similar frequency shift as observed here was also measured in a different Si/SiGe spin qubit device with micro- 5 magnet [16] and in a phosphorous donor electron spin qubit, albeit with values several orders of magnitude smaller [25]. There may be several physical origins for the frequency shift and among them we find that heating caused by the microwave burst may explain the exponential delayed response of the frequency shift (see Supplementary Sections 4 and 5). Since the thermal expansion is different between silicon and germanium, the increase of the lattice temperature can cause a change of the strain in the quantum well [26]. The strain caused by the metallic gate electrodes [27] may also be temperature dependent. In any case, the strain variation modifies the potential shape for the confined electron and the center quantum dot position. Because of the magnetic field gradient, the quantum dot position shift results in the local magnetic field or the resonance frequency shift. Since it takes some time to cool down the system to the base temperature after turning off the microwave burst, the frequency shift occurs during and even after the microwave burst application. However, this does not explain the discontinuous frequency shift between the continuous-wave response in Fig. 2b and the exponential decay in Fig. 3c because there should be no abrupt change in the system temperature before and after turning off the microwave pulse. Although the detailed physical mechanism will not affect the qubit fidelity optimization described in what follows, further investigation is needed to fully explain the observed frequency shift. Now we turn to the qubit control fidelity. The observed resonance frequency shift affects the control fidelity because it is much larger than the fluctuation of resonance frequency for our device (𝜎 ∼ 20.6 kHz for device B). Therefore, here we discuss the qubit control optimization in the presence of such a microwave amplitude dependent frequency shift. The simplest way to cancel the frequency shift effect may be to keep the microwave amplitude always constant by applying off-resonance microwave even when the qubit is idle [16]. In this way, the qubit frequency shift during the control stage is kept constant and we can choose the shifted qubit resonance frequency as the rotating frame frequency. However, this method causes too much additional heating of the device which may be harmful for the qubit control because we need a relatively large microwave power to realize the qubit rotation faster than the dephasing time. In addition, due to the limited bandwidth of the microwave modulation circuit, creation of the smooth shaped pulse is difficult for this type of control including abrupt frequency switching. We therefore investigate a way to cancel out the unwanted qubit phase accumulation by 6 quadrature microwave control [17, 19, 28]. The technique was originally proposed for cancelling the microwave induced frequency shift (a.c. Stark shift) and the state leakage of transmon qubits. Because spin qubits generally have a well-defined two-level system and the state leakage is negligible, the quadrature control can be used to just correct the microwave induced frequency shifts. In this case, in contrast to the transmon qubit case where the single quadrature parameter has to be set to an optimal point to balance the compensation of two infidelity sources, one quadrature parameter can be used to fully compensate the influence of the frequency shift. To calculate the single-qubit time evolution, here we consider the rotating frame Hamiltonian of the system written as follows: 𝐻(𝑡) = − ℏ 2 (𝑋(𝑡)𝜎x + 𝑌(𝑡)𝜎y + 𝑍(𝑡)σz), (1) where 𝑋(𝑡) and 𝑌(𝑡) are the EDSR microwave control amplitudes, 𝑍(𝑡) is the frequency shift caused by the XY control, and ℏ is the reduced Planck's constant. The rotating frame frequency and 𝑓MW are set at the qubit resonance frequency during the free evolution with 𝑋(𝑡) = 𝑌(𝑡) = 0 . Here we consider the pulse optimization for a Gaussian π/2 rotation 𝑋(𝑡) = 𝐴Xexp (−𝑡2/2𝜎2) and the quadrature derivative control 𝑌(𝑡) = 𝛼π/2𝜎(d𝑋(𝑡)/d𝑡) truncated at ±2𝜎 . 𝐴X is the microwave control amplitude normalized with the ideal π/2 control amplitude 𝐴π/2 = π/(𝜎 ∫ exp(−𝑡2/2) 𝑑𝑡 that the quadrature coefficient 𝛼 has to be adjusted independently for π and π/2 ). Note 2 −2 pulses. The microwave induced frequency shift is calculated from the power-law relation 𝑍(𝑡) = 𝑎(𝑋(𝑡)2 + 𝑌(𝑡)2)𝑏/2 ( 𝑡 ∈ [−2σ, 2σ] ), i.e. it is assumed to be dominated by the instantaneous response and the slowly changing part is ignored. The partial optimization still works reasonably well to mitigate the qubit control errors because the slow delayed response is several times smaller than the fast response. Figure 4a shows a plot of the averaged qubit control fidelity 𝐹 of Xπ/2 gate calculated using the equation 𝐹(𝑈, ℇ) = 1/2 + (1/12) ∑ , where 𝑈 = exp (𝑖π𝜎𝑥/4) is the ideal process matrix and ℇ is the actual quantum operation [29]. Here we plot 𝐹 −1 = 1/4𝜎 ranging from 1 to 20 MHz, which is a for the gate clock frequency 𝑡π/2 reasonable operation range for device B. In this qubit operation range, 𝐹 is limited to Tr(𝑈𝜎𝑗𝑈†ℇ𝜎𝑗) 𝑗=𝑥,𝑦,𝑧 approximately 99.8 % because of the unwanted phase accumulation due to the frequency shift. In Fig. 4b, we calculate 𝐹 at 𝑡π/2 −1 = 20 MHz (corresponds to 𝑓Rabi = 5 MHz for rectangular microwave burst) as a function of π/2 quadrature coefficient 𝛼π/2. The model predicts a gate fidelity higher than 99.999 % with an optimized parameter set at 𝐴X = 1.00 and 𝛼π/2 = −0.173. (The graphical Bloch sphere representation of the qubit 7 evolution is depicted in Fig. S6.) We experimentally confirm the effectiveness of the quadrature control using an interleaved randomized benchmarking technique (Fig. 4c). Only device B is used for this measurement as the influence of the frequency shift is too subtle to observe experimentally in device A. The Xπ/2 interleaved randomized benchmarking is used to characterize the fidelity of Xπ/2 gate and 𝑓MW is set to the free evolution frequency calibrated by the Ramsey fringe. Figures 4d and 4e show the Xπ/2 interleaved randomized benchmarking sequence fidelity 𝐹 at a fixed number of Clifford gates, 𝑚 = 122, measured for various values of 𝛼π/2 and 𝐴X. The sequence fidelity is defined as 𝐹 = 𝑃↑ sequence designed to obtain ↑⟩(↓⟩) as an ideal final state. To clarify the parameter ) is the measured spin-up probability for the , where 𝑃↑ (𝑃↑ ↑⟩ − 𝑃↑ ↓⟩ ↑⟩ ↓⟩ dependence of 𝛼π/2 and 𝐴X, the other parameters (microwave frequency and amplitude, 𝛼 for other Clifford gates) are adjusted to maximize the sequence fidelity. We find that the sequence fidelity is maximized at 𝛼π/2 = −0.18, which is in reasonable agreement with the value derived from the theory. The small deviation may come from the post pulse effect. From a separate measurement using the same device and the quadrature control, we obtain a single gate fidelity as high as 99.93 % [12] and this is well above the upper limit given by the microwave burst induced frequency shift. Discussion We have reported the shift of resonance frequency of electron spin qubits in Si/SiGe quantum dots with increasing applied microwave burst amplitude and quadrature control method to cancel out the qubit control error cause by the frequency shift. Although part of the observed frequency shift may be explained by the effect of heating, the overall physical origin remains unknown and full characterization needs further investigation. Nevertheless, for the purpose of practical optimization of quadrature compensation pulse presented in this work, the Ramsey-based measurement of the amplitude dependence described in Fig. 2 is sufficient. We anticipate that the full understanding of the frequency shift mechanism will allow for further optimizations beyond what is presented in this work, such as the prediction of the frequency shift from the device parameters and the minimization of the frequency shift itself by the device design. Methods In both devices, the quantum dot is formed by locally depleting the two-dimensional electron gas in an undoped Si/SiGe heterostructure. A 250 nm thick cobalt micro-magnet 8 is deposited on top of the accumulation gate to induce a stray magnetic field across the quantum dot. The sample is cooled down using a dilution refrigerator to a base electron temperature of approximately 120 mK (unless otherwise noted) which is estimated from the transport linewidth. Further details about the devices and the measurement setup are described in Supplementary information, Refs. 10 (device A), and 12 (device B). For both devices, the valley splitting is confirmed by magneto-spectroscopy measurement to be larger than the Zeeman splitting. Therefore, the physics in this work is mainly described by a conventional single-valley picture, although there may be a small fraction of the population in the excited valley state due to initialization errors. Data availability The data that support the findings of this study are available from the corresponding author upon reasonable request. Acknowledgements This is a post-peer-review, pre-copyedit version of an article published in npj Quantum Information. The final authenticated version is available online at: https://doi.org/10.1038/s41534-018-0105-z. We thank the Microwave Research Group in Caltech for technical support. This work was supported financially by Core Research for Evolutional Science and Technology (CREST), Japan Science and Technology Agency (JST) (JPMJCR15N2 and JPMJCR1675) and the ImPACT Program of Council for Science, Technology and Innovation (Cabinet Office, Government of Japan). K.T. acknowledges support from JSPS KAKENHI grant number JP17K14078. J.Y., T.O. and T.N. acknowledge support from RIKEN Incentive Research Projects. T.O. acknowledges support from Precursory Research for Embryonic Science and Technology (PRESTO) (JPMJPR16N3), JSPS KAKENHI grant numbers JP16H00817 and JP17H05187, Advanced Technology Institute Research Grant, the Murata Science Foundation Research Grant, Izumi Science and Technology Foundation Research Grant, TEPCO Memorial Foundation Research Grant, The Thermal and Electric Energy Technology Foundation Research Grant, The Telecommunications Advancement Foundation Research Grant, Futaba Electronics Memorial Foundation Research Grant and Foundation for Promotion of Material Science and Technology of Japan (MST) Foundation Research Grant. K.M.I. acknowledges support from JSPS KAKENHI grant number JP26220602 and JSPS Core-to-Core Program. T.K. acknowledges support from JSPS KAKENHI grant numbers JP26709023 and JP16F16806. S.T. acknowledges 9 support from JSPS KAKENHI grant numbers JP26220710 and JP16H02204. Competing interests The authors declare that they have no competing interests. Author contribution K.T. and J.Y. performed the measurements and analyzed the data. K.T., and T.O. fabricated the samples. Y. H., N. U., and K. M. I. supplied the isotopically enriched Si/SiGe heterostructure. K.T. wrote the article with inputs from the rest of the authors. M.R.D., G.A., T.N., T.K., and S.O. contributed to the sample fabrication, measurement, and data analysis. S.T. supervised the project. References [1] Loss, D. & DiVincenzo, D. P. Quantum Computation with Quantum Dots. Phys. Rev. A 57, 120 (1998). [2] Koppens, F. H. L. et al. Driven coherent oscillations of a single electron spin in a quantum dot. Nature 442, 766-771 (2006). [3] Nowack, K. C., Koppens, F. H. L., Nazarov, Y. V. & Vandersypen, L. M. K. Coherent Control of a Single Electron Spin with Electric Fields. Science 318, 1430-1433 (2007). [4] Pioro-Ladriere, M. et al. Electrically driven single-electron spin resonance in a slanting Zeeman field. Nat. Phys. 4, 776-779 (2008). [5] Brunner, R. et al. Two-qubit gate of combined single-spin rotation and interdot spin exchange in a double quantum dot. Phys. Rev. Lett. 107, 146801 (2011). [6] Petta, J. R. et al. Coherent manipulation of coupled electron spins in semiconductor quantum dots. Science 309, 2180-2184 (2005). [7] Shulman, M. D. et al. Demonstration of Entanglement of Electrostatically Coupled Singlet-Triplet Qubits. Science 336, 202-205 (2012). [8] Veldhorst, M. et al. An addressable quantum dot qubit with fault-tolerant control fidelity. Nat. nanotechnol. 9, 981-985 (2014). [9] Eng, K. et al. Isotopically enhanced triple-quantum-dot qubit. Sci. Adv. 1, e1500214 (2015). [10] Takeda, K. et al. A fault-tolerant addressable spin qubit in a natural silicon quantum dot. Sci. Adv. 2, e1600694 (2016). [11] Kawakami, E. et al. Gate fidelity and coherence of an electron spin in a Si/SiGe quantum dot with micro-magnet. Proc. Natl. Acad. Sci. 113, 11738-11743 (2016). [12] Yoneda, J. et al. A quantum-dot spin qubit with coherence limited by charge noise 10 and fidelity higher than 99.9%, Nat. Nanotechnol. 13, 102 -- 106 (2018). [13] Veldhorst, M. et al. A two-qubit logic gate in silicon. Nature 526, 410-414 (2015). [14] Nichol, J. M. et al. High-fidelity entangling gate for double-quantum-dot spin qubits. npj Quantum Information 3, 3 (2017). [15] Zajac, D. M. et al. Resonantly driven CNOT gate for electron spins, Science 359, 439-442 (2018). [16] Watson, T. F. et al. A programmable two-qubit quantum processor in silicon, Nature 555, 633 -- 637 (2018). [17] Lucero, E. et al. Reduced phase error through optimized control of a superconducting qubit. Phys. Rev. A 82, 042339 (2010). [18] Chow, J. M. et al. Optimized driving of superconducting artificial atoms for improved single-qubit gates. Phys. Rev. A 82, 040305(R) (2009). [19] Chen, Z. et al. Measuring and Suppressing Quantum State Leakage in a Superconducting Qubit. Phys. Rev. Lett. 116, 020501 (2016). [20] Fowler, A. G. Stephens, A. M. & Groszkowski, P. High-threshold universal quantum computation on the surface code. Phys. Rev. A 80, 052312 (2009). [21] Tokura, Y. Van Der Wiel, W. G., Obata, T. & Tarucha, S. Coherent single electron spin control in a slanting zeeman field. Phys. Rev. Lett. 96, 047202 (2006). [22] Knill, E. et al. Randomized benchmarking of quantum gates. Phys. Rev. A 77, 012307 (2008). [23] Reilly, D. J., Marcus, C. M., Hanson, M. P. & Gossard, A. C. Fast single-charge sensing with a rf quantum point contact. Appl. Phys. Lett. 91, 162101 (2007). [24] Elzerman, J. M. et al. Single-shot read-out of an individual electron spin in a quantum dot. Nature 430, 431-435 (2004). [25] Freer, S. et al. A single-atom quantum memory in silicon, Quantum Science and Technology 2, 015009 (2017). [26] Park, J. et al. Electrode-stress-induced nanoscale disorder in Si quantum electronic devices. APL Mater. 4, 066102 (2016). [27] Popescu, D. P., Eliseev, P. G., Stintz, A. & Malloy, K. J. Temperature dependence of the photoluminescence emission from InAs quantum dots in a strained Ga 0.85 In 0.15 As quantum well. Semicond. Sci. Technol. 19, 33-38 (2004). [28] Motzoi, F., Gambetta, J. M., Rebentrost, P. & Wilhelm, F. K. Simple Pulses for Elimination of Leakage in Weakly Nonlinear Qubits. Phys. Rev. Lett. 103, 110501 (2009). [29] Nielsen, M. A. A simple formula for the average gate fidelity of a quantum dynamical operation. Phys. Lett. A 303, 249-252 (2002). 11  Figures and tables Figure 1. Device structure and Rabi oscillation frequency shift. a Scanning electron microscope image of the device. The scale bar represents 200 nm. The gate electrode geometry is nominally identical for both devices A and B. Three of the gate electrodes (R, L, and C) are connected to the 50 ohm coaxial lines. The blue (red) circle shows the estimated position of the quantum dot for device A (B). b Pulse sequence used for the Rabi oscillation measurement. The initialization and readout are done at the same gate voltage condition where only the spin-down electron can tunnel into the dot. The compensation stage to make the pulse d.c. voltage offset to zero (used only for device A) is omitted for simplicity. c Rabi oscillation measured with different microwave amplitudes at 𝐵ext = 0.51 T (device A). The red arrows show the center resonance frequency positions. As 𝐴MW is increased, in addition to the increase of 𝑓Rabi, the center resonance frequency increases as well. 12  Figure 2. Resonance frequency shift measurements (device B). a Schematic showing the modified Ramsey sequence. During the waiting time 𝑡w, an off-resonance microwave burst with a rectangular envelope is applied to observe the microwave induced frequency shift. b Resonance frequency shift Δ𝑓 measured as a function of the off-resonance microwave amplitude 𝐴MW. The red points show the experimental data and the black solid line shows a power-law fitting Δ𝑓 = 𝑎𝐴MW Ramsey fringe oscillations measured under the conditions indicated by the arrows in 𝑏 with 𝑏 = 0.59. c Fig. 2b. The black solid lines show sinusoidal fitting curves. 13  Figure 3. Post pulse frequency shift measurement (device B). a Schematic showing the modified Hahn echo sequence used to obtain the post microwave burst response. The interval between each π/2 pulse and the π pulse is fixed at 20 s. b Measured echo signal shift as a function of 𝑡d at 𝐴MW = 0.15. c Extracted echo phase shift 𝜃 after turning off the microwave burst. The circles show the data obtained by fitting the echo signal with a sinusoidal function 𝑃↑(𝜙) = −𝐸cos(𝜙 + 𝜃(𝑡d)) + 𝐹 with 𝐸(> 0), 𝐹, and 𝜃(𝑡d) as fitting parameters. The error bars represent one standard deviation of uncertainty. The black solid lines show fitting curves. d Transient frequency shift derived from the echo phase accumulation at 𝐴MW = 0.15. The black solid line shows a derivative of the exponential fitting curve Δ𝑓(𝑡d) = (1/2π)(d𝜃(𝑡d)/d𝑡d) in Fig 3c. 14  Figure 4. Qubit fidelity analysis and optimization using a quadrature microwave control technique. a Calculated qubit operation clock 𝑡π/2 −1 dependence of the averaged qubit fidelity 𝐹 of Xπ/2 gate. b Average qubit fidelity 𝐹 as a function of the control amplitude and the quadrature control coefficient 𝛼π/2. The gate time is set at 𝑡π/2 = 50 ns. c Schematic showing the pulse sequence for the randomized benchmarking measurement. In between the randomly chosen Clifford gates, an Xπ/2 gate is interleaved to characterize its fidelity. d Interleaved randomized benchmarking fidelity for Xπ/2 gate measured as a function of X control amplitude 𝐴X. 𝐴X is directly proportional to the microwave voltage amplitude. The number of random Clifford gates is fixed at 𝑚 = 122 and 𝑘 = 32 gate sets are used for the measurements. The grey scattered points show the sequence fidelity for each random Clifford gate set and the red points show the sequence fidelity averaged over all 32 random gate sets. The light blue band shows standard error of the mean at each 𝐴X. e Interleaved randomized benchmarking fidelity for Xπ/2 gate measured as a function of the quadrature coefficient απ/2. The number of random Clifford gates is fixed at 𝑚 = 122. The grey scattered points show the sequence fidelity for each random Clifford gate set and the red points show the sequence fidelity averaged over all 32 random gate sets. The light blue band shows standard error of the mean at each απ/2. 15 Supplementary information for "Optimized electrical control of a Si/SiGe spin qubit in the presence of an induced frequency shift" K. Takeda,1 J. Yoneda,1 T. Otsuka,1, 2, 3 T. Nakajima,1 M. R. Delbecq,1, 4 G. Allison,1 Y. Hoshi,5 N. Usami,6 K. M. Itoh,7 S. Oda,8 T. Kodera9 and S. Tarucha1, 10 1. RIKEN, Center for Emergent Matter Science (CEMS), Wako-shi, Saitama, 351-0198, Japan 2. JST, PRESTO, 4-1-8 Honcho, Kawaguchi, Saitama 332-0012, Japan 3. Research Institute of Electrical Communication, Tohoku University, 2-1-1 Katahira, Aoba-ku, Sendai, 980-8577, Japan 4. Laboratoire Pierre Aigrain, Ecole Normale Supérieure-PSL Research University, CNRS, Université Pierre et Marie Curie-Sorbonne Universités, Université Paris Diderot-Sorbonne Paris Cité, 24 rue Lhomond, 75231 Paris Cedex 05, France 5. Advanced Research Laboratories, Tokyo City University, 8-15-1 Todoroki, Setagaya- ku, Tokyo 158-0082, Japan 6. Graduate School of Engineering, Nagoya University, Nagoya 464-8603, Japan 7. Department of Applied Physics and Physico-Informatics, Keio University, Hiyoshi, Yokohama 223-8522, Japan 8. Department of Physical Electronics and Quantum Nanoelectronics Research Center, Tokyo Institute of Technology, O-okayama, Meguro-ku, Tokyo 152-8552, Japan 9. Department of Electrical and Electronic Engineering, Tokyo Institute of Technology, O-okayama, Meguro-ku, Tokyo 152-8552, Japan 10. Department of Applied Physics, The University of Tokyo, Hongo, Bunkyo-ku, Tokyo, 113-8656, Japan 1 S1 Device structure and micro-magnet field simulation Both devices A and B are fabricated on top of undoped Si/SiGe heterostructures. The details of the Si/SiGe heterostructures are available in Refs. 10 (device A) and 12 (device B). Figure S1a shows the layer stack of our Si/SiGe quantum dot device. The surface of the heterostructure is covered by a 10 nm thick Al2O3 insulator formed by atomic layer deposition. The ohmic contacts are fabricated by phosphorus ion implantation. The quantum dot confinement gates and the accumulation gate are formed by electron-beam lithography and metal deposition. The accumulation gate and depletion gate electrodes are separated from each other by another 50 nm thick Al2O3 insulator layer. A 250 nm thick cobalt micro-magnet is deposited on top of the accumulation gate to induce a stray magnetic field around the quantum dot. The micro-magnet magnetic field simulation shown here is performed by using the Mathematica package Radia [30]. The following micro-magnet field simulation results are based on the geometry of device A (60 nm thick SiGe spacer layer), but device B (40 nm thick SiGe spacer layer) has a very similar geometry and therefore the conclusions are quantitatively applicable to both devices. Figures S1a and S1b show the device layer stack and the micro-magnet geometry. The blue point in Fig. S1b shows the ideal quantum dot position, however, due to the limited alignment precision in electron-beam lithography and the non-ideality of the gate voltage confinement, we speculate that there may be about ± 50 nm uncertainty in the quantum dot position. Figs. S1c and S1d MM/d𝑧 distribution. Around the expected position show the calculated slanting field d𝐵y of the quantum dot (𝑥 = −0.05 m and 𝑧 = 0 m), d𝐵y MM/d𝑧 ∼ 0.7 T/um is obtained. S2 Echo measurement for device A Figure S2a shows the Hahn echo measurement sequence. Figure S2b shows the echo signal as a function of evolution time 𝑡evolve . By fitting the normalized echo decay 𝐻 and an exponent α as fitting 𝐶(𝑡evolve) using 𝐶(𝑡evolve) = exp(−(𝑡evolve/𝑇2 parameters, we obtain 𝑇2 𝐻 = 11 μ𝑠 and α = 1.2. 𝐻)𝛼) with 𝑇2 S3 Frequency shift measurement for device A The pulse sequence used for the measurement is depicted in Fig. 2a in the main text. First, a Xπ/2 pulse (−7.4 MHz detuned from the center resonance frequency) is applied 2 to rotate the spin state to the equator of the Bloch sphere. Next the frequency shift is induced by an off-resonance microwave burst (−67.4 MHz detuned from the center resonance frequency) with a duration of 𝑡w. Finally the accumulated qubit phase is projected to the z-axis by the second Xπ/2 pulse (−7.4 MHz detuned). The −7.4 MHz frequency offset is intentionally added to observe a finite frequency oscillation at 𝐴MW = 0. Figure S3a shows the measured oscillations of the Ramsey fringe as a function of off- resonance microwave amplitude 𝐴MW. Figure S3b shows the frequency shift extracted by fitting the observed Ramsey data as a function of 𝐴MW. The red points show the data 𝑏 with 𝑎 and and the black solid line shows an empirical fitting curve with Δ𝑓 = 𝑎𝐴MW 𝑏 as the fitting parameters. The exponential function fits well to the measured data, however, as mentioned in the main text, it is found that the fitting constants have strong sample-to-sample variation. S4 Possible reasons for the frequency shift 1. Bloch-Siegert shift The most conventional reason for the microwave induced qubit resonance frequency shift may be the Bloch-Siegert shift, which is well-known as a special case of a.c Stark shift for strongly driven two-level systems [31]. The frequency shift we observed is obviously inconsistent with this mechanism because in this case the frequency shift should follow 2/𝐸Z where 𝐸Z = 𝑔𝜇B𝐵0 is the Zeeman splitting. In the quadratic relation Δ𝑓 = 𝑓Rabi addition, the value of the shift is quantitatively too small (device A: Δ𝑓~6 kHz for 𝑓Rabi =10 MHz and 𝐸Z = 16 GHz, device B: 𝛥𝑓~2 kHz for 𝑓Rabi =5 MHz and 𝐸Z = 18 GHz). 2. Rectification due to the anharmonicity of the confinement Another possible reason is the quantum dot motion rectification due to the anharmonicity of the confinement potential. Figure S1e shows the calculated in-plane stray magnetic field (𝐵z) distribution in the two-dimensional electron gas (2DEG) plane. Although the micro-magnet is designed to be robust against a relatively large MM/d𝑧, not for the misalignment [33], the robustness only stands for the slanting field d𝐵y Zeeman splitting or the magnetic field 𝐵z. As shown in Fig. S1f, misalignment of the micro-magnet along the z-direction can cause the relatively large variation of the in- plane longitudinal field gradient d𝐵z MM/d𝑧. For ±50 nm misalignment of the micro- magnet position, d𝐵z MM/d𝑧 can be as large as 0.3 T/m. In addition, the direction of electric field by gate C at the quantum dot position is not perfectly parallel to the z-axis 3 due to the device geometry and therefore the transverse field gradient d𝐵z S1g) may also contribute to the gate voltage dependence of 𝐵z. Here, using the value of MM/d𝑧 = 0.3 T/ μm , we numerically simulate the rectification effect with a one- d𝐵z 4 dimensional model assuming quartic anharmonic potential 𝑈(𝑧) = ∑ 𝑘=2 . Although MM/d𝑥 (Fig. 𝑎𝑘𝑧𝑘 it is difficult to directly measure the coefficients 𝑎3, 𝑎4 experimentally, we can estimate these parameters by the Rabi frequency saturation caused by the anharmonicity. In both devices A and B, the anharmonicity suppresses 𝑓Rabi to ~19 MHz at 𝐴MW = 0.6, which is approximately 1 MHz smaller than the value expected from the linear relation observed at lower microwave amplitudes [10, 12]. Figure S4a shows simulated Rabi frequency suppression Δ𝑓Rabi(𝑎3, 𝑎4) = 𝑓Rabi(0,0) − 𝑓Rabi(𝑎3, 𝑎4) and Fig. S4b shows the exponent of the resonance frequency shift obtained by fitting the Rabi frequency deviation with a power-law relation Δ𝑓 = 𝑎(AMW)𝑏 . In the proper parameter range where Δ𝑓Rabi(𝑎3, 𝑎4) is close to the experimental value 1 MHz (the black region in Fig. S4a), we obtain 𝑏~2, which results in a quadratic shift (Fig. S4b). There is a clear discrepancy between the simulated and the measured exponents 𝑏 (𝑏 = 0.59 ± 0.03 for device B and 𝑏 = 1.39 ± 0.02 for device A). Therefore, we rule out the rectification effect from the possible reasons for the frequency shift. 3. External microwave setup The frequency shift can also occur if there is a frequency shift of the microwave signal. Such an unintentional microwave frequency change might be caused by a large output parameter change of the microwave signal generator (Keysight E8267D is used for all measurements). However, when we change 𝐴MW in the experiment, rather than changing the output condition (power) of the signal generator, we change the output amplitude of an arbitrary waveform generator used for I/Q modulation. This modulation scheme causes hardly any change of the frequency shift of the microwave signal. In addition, we monitor the output microwave signal (after passing all active components) using a spectrum analyzer and there is no noticeable shift of the frequency when the I/Q modulation amplitude or the source power is changed. The microwave circuit contains some additional passive components (attenuators, cables etc.), but those will not affect the frequency of microwave signal. As for the delayed response of the frequency shift, the reflection of the microwave due to impedance mismatch is a possible cause. The microwave reflection can cause delayed transient microwave output with the time scale depending on the length of the reflection path. In our measurement setup, the dominant source for the microwave reflection seems 4 to be the room-temperature modulation circuit as the I/Q mixer and the microwave amplifier(s) have much poorer VSWRs as compared to those for the other components. At the output of the microwave modulation circuit, we measured transient signal with a time scale < 10 ns using a real-time oscilloscope (Keysight DSOX92004Q with a sampling rate of 80 GSa/s). Although it may slightly affect the experimental results, the time scale is much shorter than we observed in Fig. 3c in the main text (a characteristic decay time 𝜏 ∼ 6 μs) and cannot explain the experimental result. 4. Heating due to microwave burst Here, we consider a Hamiltonian with harmonic confinement as follows: 𝐻 = 2 2 + 𝑝𝑣 𝑝𝑢 2𝑚∗ + 1 2ℏ2 𝐸orb 𝑢 2 𝑚∗(𝑢 2 + 𝑣2) − 𝑒𝐸⃗ ⋅ ( 𝑣 ) , (1) where 𝑚∗ is the transverse effective mass of electron in strained silicon, 𝑒 is the elementary charge, ℏ is the reduced Planck's constant, 𝑝𝑢 and 𝑝𝑣 are the momentum operators, 𝑢 and 𝑣 are the position operators, 𝐸orb is the orbital spacing of the quantum dot, and 𝐸⃗ is the in-plane electric field. In what follows, we set the electric field as 𝐸⃗ = (𝐸, 0)T for simplicity, but it does not affect the conclusion due to the symmetry of the Hamiltonian (note that we have assumed a symmetric in-plane confinement). From simple mathematics, the dot position offset 𝑢0 caused by 𝐸⃗ can be derived as follows: 𝑢0 = ℏ2𝑒𝐸 2 𝑚∗𝐸orb . (2) According to Ref. 32, the potential change caused by the temperature change can be dealt as the change of the effective mass. In such a case, the induced dot position shift δ𝑢0 for a small change of effective mass δ𝑚∗ can be written as follows: δ𝑢0 = 𝑢0(𝑚∗ + 𝛿𝑚∗) − 𝑢0(𝑚∗) ∼ ℏ2𝑒𝐸 𝑚∗𝐸orb 2 ( 𝛿𝑚∗ 𝑚∗ ) . (3) Then, the position shift results in a frequency shift when combined with the micro- magnet field gradient as follows: δ𝑓 = 𝛾e MM d𝐵z d𝑢 δ𝑢0 ∼ 𝛾e ( MM d𝐵z d𝑢 ) ℏ2𝑒𝐸 𝑚∗𝐸orb 2 ( δ𝑚∗ 𝑚∗ ) , (4) where 𝛾e = 28 GHz/T is the gyromagnetic ratio for electron spin in silicon. By using a MM/d𝑢 = 0.3 T/μm, 𝐸 = 0.1 MV/m, 𝑚∗ = 0.19𝑚e (𝑚e is the electron parameter set d𝐵z rest mass), and 𝐸orb = 0.5 meV, in addition to the referred effective mass change caused by the strain effect δ𝑚∗/𝑚e = 2 × 10−6 × δ𝑇 𝐾−1, where δ𝑇 is the temperature change [32], we can obtain an estimate for the frequency shift, 5 δ𝑓 ∼ 14 × δ𝑇 kHz/K. (5) According to this equation, the measured electron temperature increase of a few hundred mK under the microwave burst will cause a few tens of kHz frequency shift, which is not too far from the measured results in Fig. 3d (we assume thermal equilibrium between the lattice and electrons). However, we stress that this result just shows a rough estimation of the frequency shift value and is not enough to deal with the detailed properties like the device dependent exponent and sign. Simulations including the local electrostatics and strain of the devices will be needed for further investigation. S5 Measurement of electron temperature under microwave burst For both devices examined in this study, the microwave burst is applied to a gate (C gate in Fig. 1), which has an electrically open end. Ideally, the voltage applied on the open end circuit is reflected back to the source and there should be no power consumption at the sample end. However, the high frequency signals (15-20 GHz) used for the EDSR measurements in this work will be easily dissipated. For instance, the stray capacitances to the neighboring conductors (the surrounding two-dimensional electron gas and metallic gates) can result in finite a.c. current flowing through the device. To confirm this, we perform a measurement of electron temperature under the microwave burst. Device A is used for the measurement and the microwave frequency is 20 MHz detuned from the spin resonance frequency. Fig. S5a shows the dot-to-reservoir transitions under several different microwave excitation amplitudes. As can be seen in the data, the line width is broadened with the increased microwave amplitude. From the Fermi-Dirac fitting curve, we can extract 𝑇e at each 𝐴MW. Fig. S5b shows 𝑇e extracted for a wider range of 𝐴MW. The measurement is limited to a relatively low-power range (𝐴MW = 0.012 corresponds to 𝑓Rabi = 400 kHz) because the charge sensor sensitivity rapidly decreases at the higher microwave amplitudes and the reliable estimation of the line width becomes difficult. We find that the line width increases linearly as a function of the microwave amplitude. From this measurement, although at higher temperatures some cooling mechanisms will suppress the linear temperature increase, we roughly estimate that the electron temperature increases by a few Kelvin when a microwave burst for EDSR in the MHz range is applied. We note that this measurement is done at a decreased base electron temperature with a modified setup ( 𝑇e ∼ 35 mK in this measurement whereas 𝑇e ∼ 120 mK for the others). 6 References [30] Chubar, O., Elleaume, P., & Chavanne, J., A 3D Magnetostatics Computer Code for Insertion Devices, SRI97 Conference August 1997, J. Synchrotron Rad. 5, 481-484 (1998). [31] Bloch, F. & Siegert, A., Magnetic Resonance for Nonrotating Fields. Phys. Rev. 57, 522 (1940). [32] Richard, S., Cavassilas, N., Aniel, F., & Fishman, G. Strained silicon on SiGe: Temperature dependence of carrier effective masses. J. Appl. Phys. 94, 5088 (2003). [33] Yoneda, J. et al. Robust micromagnet design for fast electrical manipulations of single spins in quantum dots. Appl. Phys. Express 8, 84401 (2015). 7  Figure S1. Micro-magnet design and simulation results. a Schematic layer sequence of the device structure. The external magnetic field is applied along the positive z-direction. b Schematic of the micro-magnet design. The grey area shows the micro-magnet pattern and the quantum dot location is represented by the blue box. c Simulated out-of-plane MM/d𝑧 as a function of the positions in the 2DEG plane 𝑥 and slanting magnetic field d𝐵y 𝑧. The black circle shows the expected quantum dot position. The designed quantum dot position is 𝑥 = −0.05 m and 𝑧 = 0 m. d Line cut of the out-of-plane slanting field MM/d𝑧 along the z-axis at the dot position 𝑥 = −0.05 m. e Simulated in plane stray d𝐵y magnetic field 𝐵z MM as a function of the positions in the 2DEG plane 𝑥 and 𝑧. The black circle shows the expected quantum dot position. The designed quantum dot position is 𝑥 = −0.05 m and 𝑧 = 0 m. f Line cut of d𝐵z 𝑥 = −0.05 m. g Line cut of d𝐵z MM/d𝑥 along the x-axis at 𝑧 = 0 m. MM/d𝑧 along the z-axis at the dot position 8 Figure S2. Spin echo measurement for device A. a Schematic of the echo measurement pulse sequence. b Measured echo data. The red circles show the data points and the black solid line shows a fitting curve. 9 Figure S3. Modified Ramsey measurement result for device A. a Ramsey fringe measurement result at 𝐵ext = 0.5045 T and 𝑓MW = 15.600 GHz. b Measured resonance frequency shift Δ𝑓 as a function of the off-resonance microwave amplitude 𝐴MW. The red points show the experimental data and the black solid line shows a power-law fitting 𝑏 with 𝑏 = 1.39. Δ𝑓 = 𝑎𝐴MW 10  Figure S4. Simulation results for the effects of anharmonicity. a Rabi frequency suppression Δ𝑓Rabi calculated as a function of the third and fourth order coefficients. The microwave amplitude is set to 𝐴MW = 0.6, which corresponds to 𝑓Rabi = 20 MHz if a linear relation is assumed. The black coloured area shows the conditions where the calculated values are close to the experimental value Δ𝑓Rabi~1 MHz. b Calculated exponents 𝑏 as a function of the third and fourth order term strengths. 𝑏~2 is obtained for all parameter value used in the simulation. c A typical microwave amplitude dependence of the frequency shift with parameters 𝑎3 = 0.3 and 𝑎4 = 0.85 (the black circle in Fig. S4b). 11 Figure S5. Measurement of electron temperature under microwave excitation. a Dot-to- reservoir transition measured at four different microwave amplitudes. The circles show the measured data. The solid lines show fitting curves with a Fermi-Dirac function Δ𝑉rf = 𝑉 2 tanh ( 𝜀 2𝑘B𝑇e ) with 𝑉 and 𝑇e as fitting parameters. A linear background is subtracted from each of the data sets and the fitting curves. b Electron temperature as a function of the microwave amplitude. The circles show the electron temperatures extracted from the dot-to-reservoir transition line widths and the linear fitting curve 𝑇e ∝ 𝐴MW is obtained by using the six data points from the largest 𝐴MW. 12 Figure S6. Bloch sphere representations of qubit trajectory for Xπ/2 pulse for an initial state of spin down. a Optimized control with quadrature control α𝜋/2 = −0.173 and 𝐴X = 1.00. b Standard control without quadrature control α𝜋/2 = 0 and 𝐴X = 1.00. 13 

1507.05921

1

1507

Probing molecular dynamics at the nanoscale via an individual paramagnetic center

Understanding the dynamics of molecules adsorbed to surfaces or confined to small volumes is a matter of increasing scientific and technological importance. Here, we demonstrate a pulse protocol using individual paramagnetic nitrogen vacancy (NV) centers in diamond to observe the time evolution of 1H spins from organic molecules located a few nanometers from the diamond surface. The protocol records temporal correlations among the interacting 1H spins, and thus is sensitive to the local system dynamics via its impact on the nuclear spin relaxation and interaction with the NV. We are able to gather information on the nanoscale rotational and translational diffusion dynamics by carefully analyzing the time dependence of the NMR signal. Applying this technique to various liquid and solid samples, we find evidence that liquid samples form a semi-solid layer of 1.5 nm thickness on the surface of diamond, where translational diffusion is suppressed while rotational diffusion remains present. Extensions of the present technique could be adapted to highlight the chemical composition of molecules tethered to the diamond surface or to investigate thermally or chemically activated dynamical processes such as molecular folding.

cond-mat.mes-hall

cond-mat

Probing  molecular  dynamics  at  the  nanoscale  via  an  individual   paramagnetic  center     T.  Staudacher1,2,  N.  Raatz3,  S.  Pezzagna3,  J.  Meijer3,  F.  Reinhard1,  4,  C.  A.  Meriles5,*,  and  J.  Wrachtrup1     13rd  physics  Institute  and  IQST,  University  of  Stuttgart,  70569  Stuttgart,  Germany   2Max  Planck  Institute  for  Solid  State  Research,  70174  Stuttgart,  Germany   3Department  of  Nuclear  Solid  State  Physics,  Institute  for  Experimental  Physics  II,  University  of   Leipzig,  Linnéstr.  5,  04103  Leipzig,  Germany   4TUM  –  Technical  University  of  Munich,  Walter  Schottky  Institut,  Am  Coulombwall  4,  85748   Garching,  Germany   5Department  of  Physics,  CUNY-­‐City  College  of  New  York,  160  Convent  Ave,  New  York,  NY  10031,  USA       Abstract     *For  correspondence:  [email protected]     Understanding  the  dynamics  of  molecules  adsorbed  to  surfaces  or  confined  to  small  volumes  is  a   matter  of  increasing  scientific  and  technological  importance.  Here,  we  demonstrate  a  pulse   protocol  using  individual  paramagnetic  nitrogen  vacancy  (NV)  centers  in  diamond  to  observe  the   time  evolution  of  1H  spins  from  organic  molecules  located  a  few  nanometers  from  the  diamond   surface.  The  protocol  records  temporal  correlations  among  the  interacting  1H  spins,  and  thus  is   sensitive  to  the  local  system  dynamics  via  its  impact  on  the  nuclear  spin  relaxation  and  interaction   with  the  NV.  We  are  able  to  gather  information  on  the  nanoscale  rotational  and  translational   diffusion  dynamics  by  carefully  analyzing  the  time  dependence  of  the  NMR  signal.  Applying  this   technique  to  various  liquid  and  solid  samples,  we  find  evidence  that  liquid  samples  form  a  semi-­‐ solid  layer  of  1.5  nm  thickness  on  the  surface  of  diamond,  where  translational  diffusion  is   suppressed  while  rotational  diffusion  remains  present.  Extensions  of  the  present  technique  could   be  adapted  to  highlight  the  chemical  composition  of  molecules  tethered  to  the  diamond  surface  or   to  investigate  thermally  or  chemically  activated  dynamical  processes  such  as  molecular  folding.         1 Main  text   Nuclear  magnetic  resonance  (NMR)  is  among  the  most  versatile  tools  for  investigating  the  dynamics   of  molecular  processes  down  to  the  atomic  level.  It  is  widely  used  in  physical  and  life  sciences,  but   has  been  limited  to  large  sample  quantities  due  to  the  low  sensitivity  of  conventional  detection   methods1.  Performing  NMR  detection  at  the  nanoscale  can  substantially  expand  the  microscopist’s   toolbox,  potentially  allowing  for  non-­‐destructive  imaging  of  complex  macromolecules  and/or   studying  the  dynamics  of  diverse  biochemical  systems.  One  route  to  performing  nanoNMR  is   magnetic  resonance  force  microscopy  (MRFM),  already  used  to  image  small  organisms  with   nanometer  resolution2.  Typical  operating  conditions  of  MRFM  however  require  ultralow   temperatures  and  high  vacuum2,  3,  which,  unfortunately,  are  incompatible  with  most  molecular   processes  of  interest.   An  alternate  approach  to  NMR  at  the  nanoscale  makes  use  of  individually  addressable  paramagnetic   centers  near  the  surface  of  a  solid-­‐state  host  to  probe  sample  spin  species  in  its  immediate  vicinity.   Perhaps  the  most  prominent  example  is  the  nitrogen-­‐vacancy  (NV)  center,  a  spin-­‐1  defect  in  the   diamond  lattice  formed  by  a  substitutional  nitrogen  atom  and  an  adjacent  vacancy4.  Recently,  single   NV  centers  separated  only  a  few  nanometers  from  the  diamond  surface  were  used  to  detect  the   NMR  signal  associated  with  the  random  magnetic  spin  noise  of  a  nanoscale  proton  ensemble  under   ambient  conditions5,  6.  Subsequent  studies  extended  this  initial  work  to  other  spin  species  and   demonstrated  improved  detection  sensitivity,  attaining  the  limit  of  a  few  nuclear  spins7,  8,  9,  10,  11.   Furthermore,  by  articulating  NV  magnetometry  with  scanning  microscopy,  it  has  been  possible  to   image  the  spatial  distribution  of  nuclear  spins  within  a  polymeric  phantom  with  about  10  nm   resolution12,  13.         In  this  study,  we  use  shallow  NVs  to  probe  mesoscale  proton  ensembles  from  different  organic   substances  deposited  on  the  diamond  surface.  We  resort  to  a  form  of  correlation  spectroscopy  and   reconstruct  the  equivalent  of  a  nuclear  ‘free-­‐induction-­‐decay’  (FID),  which,  unlike  the  NMR   2 counterpart,  does  not  require  nuclear  spin  pre-­‐polarization.  This  pseudo  FID—below  referred  to  as   ‘correlation  signal’—has  a  limited  decay  time  governed  by  the  NV  spin-­‐lattice  relaxation  time  T1   (typically  longer  than  the  NV  coherence  lifetime  T2),  which  allows  us  to  attain  spectral  resolution   superior  to  that  possible  with  standard  magnetometry  techniques.  Upon  applying  this  scheme  to   solid-­‐  and  liquid-­‐state  substances  we  find  substantial  differences  in  the  correlation  signal  envelope,   which  we  associate  with  the  presumably  dissimilar  molecular  dynamics  governing  these  systems.  In   particular,  we  observe  long-­‐lived  1H  signals  from  oil  molecules,  which  we  interpret  in  terms  of  an   interplay  between  molecular  tumbling  and  self-­‐diffusion.     Figure  1.  Schematics  of  the  experimental  setup  and  basic  detection  protocol.  (a)  An  organic  sample  is   brought  into  contact  with  the  diamond  surface,  and  shallow  NV  centers  are  used  as  NMR  detectors.  (b)   XY8-­‐N  multi-­‐pulse  sequence.  A  change  in  the  NV  response  is  observed  when  the  inter-­‐pulse  separation  𝝉   relative  to  each  other.  (c)  Repeating  the  XY8-­‐N  sequence  for  multiple  pulse  spacings  𝝉  yields  an  effective   proton  spins  is  attained  by  choosing  the  pulse  spacing  𝝉  equal  to  half  the  Larmor  period.  The  correlation   signal  shows  a  decaying  oscillating  behavior  as  a  function  of  the  1H  spin  evolution  interval  𝝉!.  (e)  The  Fourier   transform  of  the  time  domain  signal  yields  the  sample  NMR  spectrum.       matches  half  the  Larmor  period.  The  blue/red  hue  indicates  different  MW  phases,  which  are  shifted  90°   CW  spectrum  of  the  nuclear  spin  noise.  (d)  Nuclear  spin  correlation  protocol.  Selective  detection  of  the     Fig.  1a  depicts  a  typical  NMR  experiment  using  NV  centers.  The  setup  consists  of  a  home-­‐built   confocal  microscope,  which  excites  single  NV  centers  in  the  illumination  volume  via  a  532  nm  laser.   Green  light  initializes  the  NV—a  spin-­‐1  system—into  the  mS=0  level  of  its  ground  state  triplet,  which   features  a  zero-­‐field  splitting  of  2.87  GHz14.  The  spin-­‐state  dependent  back  fluorescence  of  the  NV   center  is  collected  via  the  imaging  objective  and  is  focused  onto  a  single  photon  detector.  A   moveable  electromagnet  provides  a  static  magnetic  field  of  about  ~25  mT  and  a  nearby  coplanar   3 waveguide  (CPW)  is  used  for  the  application  of  microwave  irradiation  at  the  frequency  of  the   𝑚!=0↔𝑚!=−1  transition  (~2.2  GHz).  For  the  experiments  herein  we  use  single  NV  centers   produced  via  2.5  keV  15N+  ion  implantation  into  a  type-­‐IIa  [100]  diamond  crystal  (Section  A  of  the   Supplementary  Information).  The  dipolar  coupling  to  nuclear  spins  external  to  the  diamond  lattice  is   strong  enough  to  imprint  the  NV  response  with  a  signature  that  originates  from  the  statistical   nuclear  spin  polarization5,  6.  The  detection  volume  is  roughly  defined  by  the  NV  distance  to  the   surface  —  about  5  nm  in  the  present  case  —  which  approximately  corresponds  to  ~103  protons  for   typical  organic  samples5,  6,  10.   A  common  way  to  detect  the  NMR  signal  via  the  NV  center  is  based  on  a  quantum  lock-­‐in   algorithm15,  16,  which  is  implemented  through  an  XY8-­‐N  dynamical  decoupling  sequence6  -­‐  12.  Here  a   train  of  equidistant  𝜋-­‐pulses  is  used  to  selectively  enhance  the  NV  detection  sensitivity  at  a   frequency  determined  by  the  inverse  pulse  spacing  𝜏!!.  As  shown  in  Fig.  1b,  the  XY8-­‐N  sequence  is   embedded  within  a  Ramsey  protocol  (comprising  two  π/2-­‐pulses),  so  as  to  convert  the  integrated   effect  of  the  nuclear  spins  —  in  the  form  of  an  accumulated  NV  phase  shift  —  into  a  change  in  the   NV  fluorescence.  The  result  of  such  a  measurement  is  an  effective  CW  spectrum  of  the  sample  spins   (Fig.  1c),  whose  linewidth  is  ultimately  limited  by  the  XY8-­‐N  coherence  time  𝑇!!" .     The  coherence  lifetimes  of  shallow  NVs  are  often  times  shorter  than  the  characteristic  time  scales   governing  nuclear  spins,  thus  complicating  our  ability  to  gather  detailed  spectroscopic  information   on  the  structure,  chemical  composition,  or  dynamics  of  the  system  under  investigation.  One  way  to   circumvent  this  limitation  is  the  nuclear  spin  detection  protocol  of  Fig.  1d17  designed  to  exploit  the   typically  longer  NV  spin-­‐lattice  relaxation  times  𝑇!!" .  The  pulse  sequence  comprises  two  XY8-­‐N   trains  separated  by  a  variable  interval  𝜏;  in  each  of  them  the  interpulse  separation  is  kept  in  sync   with  the  sample  spin  Larmor  precession,  i.e.,  we  choose  𝜏=  !!𝜏!.  During  the  evolution  time  𝜏  the   magnetization  information  is  stored  in  the  longitudinal  NV  spin  component  while  the  sample  nuclear   spins  are  allowed  to  evolve.  The  underlying  idea  is  that  if  the  nuclear  spin  coherence  loss  is   4 sufficiently  slow,  the  phases  𝜙!,𝜙!  picked  up  by  the  NV  during  the  two  consecutive  XY8-­‐N   nuclear  spin  coherences  lasting  up  to  hundreds  of  microseconds  —  the  typical  𝑇!!"  time  of  the  NVs   conventional  NMR,  which,  however,  does  not  require  nuclear  spin  (pre-­‐)polarization.  Interestingly,   interrogations  are  correlated  with  each  other.  The  latter  results  in  a  signal  similar  to  an  FID  in   we  use  here    (Section  F  of  the  Supplementary  Information)—  can  be  probed  with  this  technique,   thus  allowing  us  to  better  discriminate  between  different  sample  dynamics.  In  this  light,  our   technique  can  be  considered  an  alternative  to  double  resonance  schemes  already  used  to   reconstruct  FID-­‐like  signals  from  1H  spins  near  shallow  NVs5.     Fig.  2a  shows  the  NV  response  in  the  presence  of  a  solid  organic  film  as  a  function  of  the  evolution   time  𝜏.  This  system  —  a  complex  mixture  of  long-­‐chain  polymers  hereby  referred  to  as  Sample  A  —   is  formed  by  the  adhesive  used  to  affix  the  diamond  crystal  to  the  sample  holder    (Merckoglas®).   Similar  to  a  conventional  FID  the  correlation  signal  oscillates  over  tens  of  microseconds  to  gradually   decay  to  zero.  We  find  that  the  decay  is  reasonably  described  by  an  exponential,  and  has  a   characteristic  time  constant  𝑇!"##! ~20  µμs.  After  a  cosine  transformation,  we  find  a  peak  centered  at   the  1H  Larmor  frequency  (~1.06  MHz)  exhibiting  a  Lorentzian  linewidth  of  ~30  kHz,  in  agreement   with  that  expected  for  static  protons  in  a  typical  solid-­‐state  organic  system18.  Of  note,  the  correlation       Figure  2.  Time-­‐resolved  NMR  of  near-­‐surface  1H  spins.  (a)  XY8-­‐3  correlation  signal  (circles)  from  protons   in  a  solid  polymeric  mixture  (Sample  A)  in  contact  with  the  diamond  surface.  The  yellow  trace   corresponds  to  an  exponentially-­‐damped  sinusoid  with  time  constant  of  21  µs  and  frequency  equal  to  ~1   MHz.  The  vertical  axis  indicates  the  contrast  in  a  scale  relative  to  the  NV  Rabi  amplitude.  (b)  Comparison   between  the  1H-­‐NMR  spectra  obtained  with  an  XY8-­‐10  sequence  (blue  squares)  and  after  Fourier   transformation  of  the  XY8-­‐3  correlation  protocol  shown  in  (a)  (yellow  circles).  The  blue  and  yellow  traces   respectively  indicate  Lorentzian  fits  to  each  data  set,  accompanied  by  their  corresponding  FWHM  values.   Both  curves  are  vertically  displaced  for  clarity.       5 signal  amplitude  is  only  a  small  fraction  (~10  %)  of  the  maximum  possible  fluorescence  contrast  (30   %  between  spin  states  mS  =  0  and  mS  =  ±1)14.  On  the  other  hand,  the  resulting  spectral  linewidth  is   about  a  factor  2  smaller  than  that  obtained  with  an  XY8-­‐10  sequence,  which  highlights  the   limitations  inherent  to  sensing  protocols  governed  by  the  coherence  lifetime  of  the  probe  NV  (Fig.   2b).     Since  the  time  scale  probed  in  Fig.  2a  is  still  considerably  shorter  than  𝑇!!"  (Section  F  of  the   Supplementary  Information),  a  natural  question  is  whether  this  technique  can  be  exploited  to   investigate  longer-­‐lived  nuclear  spin  coherences  arising,  for  example,  from  alternate  forms  of   motional  narrowing.  A  first  step  in  this  direction  is  shown  in  Fig.  3  where  we  compare  representative   correlation  signals  from  different  organic  systems,  including  that  in  Fig.  2  as  well  as  a  softer     Figure  3.  XY8-­‐3  correlation  signals  for  three  different  organic  samples.  The  data  corresponding  to  both   solid  samples  (Samples  A  and  B,  respectively  orange  and  blue  traces)  match  an  exponentially-­‐damped   sinusoid  with  time  constants  of  ~21  µs  and  ~29  µs.  By  contrast,  the  correlation  signal  from  protons  in   Sample  C  exhibits  a  long-­‐lived  tail  that  outlives  the  exponential  decay  at  early  times  (green  trace  in  the   figure  inset).  The  overall  response  can  be  reproduced  semi-­‐quantitatively  via  a  model  comprising  a  1.5  nm   layer  of  adsorbed  molecules  rotating  about  fixed  positions  and  an  outer  section  of  self-­‐diffusing  fluid.  Best   agreement  with  the  experimental  observations  is  attained  assuming  translational  and  rotational  diffusion   constants  of  0.3  nm2/µs  and  0.05  rad2/  µs  for  the  outer  and  inner  layer,  respectively  (red  sinusoid  trace  in   the  main  figure,  and  red  envelope  in  the  figure  insert).  Given  the  relatively  low  fluorescence  contrast,  only   portions  of  the  signal  were  measured.  Though  obtained  with  different  NVs,  each  curve  must  be  considered   representative  of  the  NV  response  for  the  corresponding  proton  ensemble  under  study  (Section  B  of  the   SI).     6 polymeric  film  (Polydimethylsiloxane,  a.k.a  PDMS)  and  a  drop  of  immersion  oil  (Fluka  Analytical,   10976)  in  direct  contact  with  the  diamond  surface  (Sections  A  and  B  of  the  Supplementary   Information).  Below  we  refer  to  the  latter  two  systems  as  Samples  B  and  C,  respectively.  In  all  three   cases  we  observe  periodic  oscillations  reflecting  the  already  highlighted  precession  of  proton  spins   about  the  applied  magnetic  field.  However,  the  time  over  which  these  oscillations  last  and,  perhaps   more  importantly,  the  way  the  signal  envelope  changes  over  time  differ  significantly  in  each  case.  In   particular  we  find  that  the  correlation  signal  from  Sample  C  exhibits  a  long-­‐lasting  tail  extending  to   about  80  µs  (possibly  limited  by  NV  spin  lattice  relaxation,  Section  F  of  the  Supplementary  Material).   Unlike  Sample  A,  this  behavior  cannot  be  captured  by  a  single-­‐exponential  envelope  (see  the  lower   insert  in  Fig.  3)  and  thus  points  to  differing  nuclear  spin  relaxation  mechanisms.  Sample  B,  on  the   other  hand,  shows  a  somewhat  intermediate,  longer-­‐lived  signal,  which,  nonetheless,  does  not   depart  from  the  single-­‐exponential  response  observed  in  Sample  A.     To  gain  a  more  quantitative  understanding  of  the  mechanisms  at  play,  we  start  by  modeling  the  NV   response  in  a  way  that  accommodates  the  different  dynamics  governing  solid  and  liquid  samples.   Using  a  semi-­‐classical  approximation  to  describe  the  NV  interaction  with  the  nuclear  spin  bath   (Section  C  of  the  Supplementary  Information)  and  assuming  all  molecules  move  independently,  we   write  the  correlation  signal  as       𝜙!,!0,𝜙!,!𝜏 !,! !,!  ~cos2𝜋𝜈!2𝑁!"!𝜏+𝜏 𝑆𝜏,𝜏~ 𝐴!,!𝜏,𝜏  𝑒!! !!"##!,! ,                                (1)   where  the  sum  extends  over  all  proton  spins  i  in  the  j-­‐th  molecule  and  𝜙!,!  denotes  the   corresponding  contribution  to  the  accumulated  NV  phase,  𝐴!,!  is  the  resulting  signal  amplitude,  𝜈!  is   the  nuclear  Larmor  frequency,  and  𝑁!"!  is  the  total  number  of  π-­‐pulses  in  each  XY8-­‐N  train.  The   signal  amplitude  𝐴!,!  is  proportional  to  the  rms-­‐field  generated  by  the  proton  spins  at  the  positon  of   time  𝜏.  In  the  case  of  a  solid,  molecules  occupy  fixed  positions  and  nuclear  spin  relaxation  is   the  NV  center  and  varies  with  different  molecule  positions  and/or  orientations  over  the  evolution   7 dominated  by  the  nuclear  dipolar  interactions,  presumably  homogeneous  throughout  the  sample.  In   uniform  value,  and  Eq.  (1)  converges  to  an  exponentially  damped  sinusoid17,  the  case  observed  for    describing  the  spin  relaxation  of  nuclear  moment  𝜇!,!  approaches  a   this  limit,  the  rate  1 𝑇!"##!,! both  solid  samples.  For  Sample  A,  we  find  𝑇!"##! ~21  µs  which  corresponds  to  a  decay  dominated  by   static  dipolar  couplings,  of  order  𝜇! Δ𝑟!~30  kHz  for  typical  inter-­‐proton  distances  Δ𝑟~0.1  nm  in   organic  samples.  The  slightly  longer  coherence  lifetime  in  Sample  B,  𝑇!"##! ~29  µs,  possibly  originates   from  an  enhanced  mobility  of  the  molecular  mobility  in  this  material  (see  below,  and  Section  G  of   the  Supplementary  Information).     Fluidic  systems,  on  the  other  hand,  differ  from  solids  in  that  molecules  experience  a  markedly   distinct  dynamics  dominated  by  fast  tumbling  and  self-­‐diffusion.  Both  mechanisms  contribute  to   average  out  the  inter-­‐nuclear  spin  couplings  and  thus  lead  to  longer  nuclear  spin  coherence   lifetimes.  However,  given  the  nanoscale  detection  volume  of  a  shallow  NV  —  roughly  restricted  to  a   half-­‐sphere  of  diameter  equal  to  the  NV  depth6,19  —  molecules  interacting  with  the  probe   paramagnetic  center  may  exchange  with  the  bulk  of  the  system  at  some  arbitrary  time  during  the   protocol.  This  situation  is  somewhat  reminiscent  of  that  found  in  fluorescence  correlation   spectroscopy  (FCS),  where  molecules  diffuse  in  and  out  of  the  detection  volume20,  21,  22.  In  the   present  case,  we  simplify  the  problem  by  assuming  that  molecules  occupy  ‘frozen’  positions  during   each  interrogation  interval,  thus  ‘instantaneously’  tagging  the  NV  with  a  phase  shift  corresponding              𝑆𝜏,𝜏~cos2𝜋𝜈!2𝑁!"!𝜏+𝜏 to  the  molecule  positions  at  times  0  and  𝜏.  In  this  limit,  we  rewrite  the  NV  response  as         ,                      2   where  𝐴!,!! ≡𝐴!,!0,0 ,  𝑝!,!𝜏  denotes  the  conditional  probability  of  nuclear  spin  (i,  j)  remaining   within  the  detection  volume  over  the  correlation  interval  𝜏,  and  𝑇!!,! relaxation  time.  The  correlation  decay  of  Eq.  (1)  𝑒!! !!"##!,! 𝐴!,!!  𝑝!,!𝜏  𝑒!! !!!"𝑒!! !!!,!  is  the  nuclear  spin  transverse    is  expressed  as  the  product  of  the  NV   !,! 8 Figure  4.  Dynamics  of  the  fluid  near  the  solid-­‐ liquid  interface.  (a)  We  assume  a  layered  model   comprising  a  static  adsorption  film  where   translational  diffusion  of  the  molecules  is   prohibited,  and  a  mobile,  outer  section  in  which   molecules  self-­‐diffuse  as  in  a  bulk  liquid.  (b)   Simulated  correlation  signal  for  varying  sample   conditions.  For  the  case  of  a  rigid  polymer  the   signal  decay  is  dominated  by  inter-­‐nuclear  dipolar   coupling  (blue  trace).  In  the  opposite  limit  of  an   unrestricted  bulk  liquid,  the  signal  amplitude   quickly  decreases  due  to  molecules  leaving  the   detection  volume  during  the  inter-­‐sequence  time   (green  trace).  Combining  the  two  regimes  via  an   adsorption  layer  separating  the  diamond  surface   from  the  bulk  liquid  yields  an  intermediate   response  (orange  trace).  Our  experimental   observations  are  best  described  by  allowing   molecules  in  this  layer  to  rotate  about  their   equilibrium  positions  (red  trace).  The  translational   self-­‐diffusion  coefficient  used  for  the  bulk  fluid  is   0.3  nm2/µs  (green,  yellow,  and  red  traces);  the   rotational  diffusion  coefficient  in  the  adsorbed   layer  is  0.05  rad2/µs  (red  trace).  (c)  For  a  fixed   adsorbed  layer  thickness  (1.5  nm),  we  increase  the   translational  self-­‐diffusion  coefficient  in  the  outer   section  of  the  fluid  and  the  rotational  diffusion   constant  on  the  surface  from  0.3  nm2/µs  –  0.05     rad2/µs  (red  trace),  to  0.6  nm2/µs  –  0.05  rad2/µs   (green  trace),  1.0  nm2/µs  –  0.10  rad2/µs  (yellow   trace),  2.5  nm2/µs  –  0.25  rad2/µs  (light  blue  trace),   and  10  nm2/µs  –  1.00  rad2/µs  (dark  blue  trace).   Note  the  two  sections  of  the  graph,  dominated   either  by  molecular  diffusion  or  rotation.    remain     ,  the  conditional  probability  𝑝!,!𝜏 ,  and  the  NMR  decay  of  the  sample  spins   decay    𝑒!! !!!" .    For  simplicity,  we  assumed  in  Eq.  (2)  that  the  signal  amplitude  𝐴!,!,  and    𝑇!!,! 𝑒!! !!!,! unchanged  after  self  diffusion  during  𝜏,  which  is  consistent  with  a  bimodal  distribution  where   molecules  either  self-­‐diffuse  or  not  depending  on  their  location  at  time  zero.   Although  the dynamics of molecules at a solid-liquid interface are not fully understood, several studies suggest that solids induce order in adjacent fluids23. The boundary condition typically invoked is one where the liquid is (nearly) static over the surface24. In particular, recent AFM experiments in water indicate that the transition to bulk fluid dynamics is abrupt and takes place on the nm range, depending on the surface hydrophobicity25. To numerically calculate 𝑆𝜏,𝜏 in Eq. (2) we divide the 9 detection volume in two layers (Fig. 4a): Molecules adjacent to the surface rotate about fixed positions, while molecules in the outer layer rotate and physically diffuse away from the NV. The set of conditional probabilities 𝑝!,!𝜏 for the mobile layer is calculated from a Monte Carlo run assuming a bulk diffusion coefficient for the fluid. To bring the number of free parameters to a minimum, we tie the rotation rate of bulk molecules to the diffusion constant via the Stokes-Einstein and Debye- Stokes-Einstein equations; comparison between the observed and calculated signals is carried out by systematically varying the rigid layer thickness and self-diffusion constant of the fluid (Sections D and E in the Supplementary Material) To illustrate the impact of the dynamics of molecular diffusion and rotation on the correlation signal, we use Eq. (2) to calculate the envelope governing the decay of 𝑆𝜏,𝜏 assuming different boundary conditions. For example, the green trace in Fig. 4b shows the response anticipated in the case where the sample dynamics corresponds to that of a bulk fluid up to the very surface of the diamond crystal, i.e., we assume the layer of adsorbed molecules has negligible thickness. Using a diffusion constant 𝐷!"#~0.3×10!!" m2/s (comparable to that anticipated for this sample) we find that the calculated signal envelope somewhat reproduces the trend in our experiment, namely, it exhibits a long lasting tail that outlives the exponential decay at early times (green trace in Fig. 4b). A numerical analysis shows that this tail stems from the longer nuclear spin coherences inherent to the mobile molecules of a fluid. Quantitatively, however, the calculated envelope does not agree well with the experimental data: First we note that the relative contribution of the long-lasting signal is comparatively small (because most molecules leave the detection volume when 𝜏 is sufficiently long). Further, compared to the case where no motion is present (infinite rigid layer, blue trace in Fig. 4b), we find a much faster initial decay (also absent in the experimental signal of Fig. 3). Changing the diffusion constant to greater or smaller values either accelerates the initial decay or further reduces the tail, thus pointing to the inadequacy of the underlying model (Section E of the Supplementary Information). We can, however, reproduce the experimental data when we assume the presence of a 1.5 nm thick adsorption layer (orange trace in Fig. 4b), particularly if molecules within this layer are allowed to rotate about their equilibrium positions (red trace in Fig. 4b). The observed and calculated correlation 10 signals for the fluidic sample are superimposed in Fig. 3b (red trace). The result captures reasonably well both the initial decay rate and the amplitude of the long-lived contribution. From the transverse relaxation rate 1 𝑇!"##!"#$ ≈3  kHz assigned to adsorbed protons we determine the rotational correlation time 𝜏!!"# ≈3.3 µs, much longer than in the outer segment of the sample (𝜏!!"#~190 ns as calculated from the Stokes-Einstein equation for 𝐷!"#  ~  0.3×10!!" m2/s, Section C of the Supplementary Information). We caution, however, that this value must be interpreted as an upper limit given the likely impact of NV spin-lattice relaxation on the correlation signal of Sample C (Section F of the Supplementary Information). In light of the above discussion, it is important to consider the time window the present technique is sensitive to. For example, Fig. 4c shows that doubling the diffusion constant from 0.3x10-12 m2/s to 0.6x10-12 m2/s leads to a significant change of the predicted signal envelope, which more quickly converges to the longer-lived tail produced by the adsorbed nuclei. In particular, molecules diffusing distances greater than 10 nm on a microsecond scale — corresponding to self diffusion coefficients of order ~10-11 m2/s or greater — leave a negligible imprint on the correlation signal. Correspondingly, fluidic systems such as water (Dwater ~ 2.3×10-9 m2/s) would be detectable only through the formation of a semi-mobile layer adjacent to the diamond surface. We emphasize that the sensitivity to the local system dynamics is not immediately indicative of accuracy in the derived parameters, which here must be understood as moderate given the crude assumptions of our model (spherical versus linear molecules, isotropic versus anisotropic diffusion, etc.; see Section E of the Supplementary Information). Along the same lines, we mention that it is difficult to ascertain whether the thin adsorbed layer observed in our experiments strictly originates from molecules in the fluid. In particular, recent studies have shown the presence of a protonated layer of comparable thickness possibly formed by water molecules or other hydrocarbons adsorbed upon ambient exposure12, 13. Controlled preparation of the diamond surface combined with NV-based NMR of spin species other than protons (e.g., 19F or 31P) can provide the means to more precisely separate contributions from sample molecules structured near the solid-liquid interface. Finally, deeper NVs (e.g., 15-20 nm from the surface) could be used to increase the relative contribution from more distant nuclear spins not 11 adsorbed on the diamond surface, though at the expense of an overall lower detection sensitivity and spatial resolution as well as longer interrogation times19. On a final note, we hypothesize that the impact of the diamond crystal on the near-surface dynamics of the organic system is not restricted to fluidic samples but more in general, extends to most soft condensed matter systems. In particular, comparison between NV-detected and inductively detected NMR signals — omitted here for brevity — reveals longer coherence lifetimes for nuclear spins within the bulk of solid Samples A and B (200 µs and ~500 µs, respectively), which we interpret as indicative of a more restricted dynamics near the diamond crystal. Naturally, this conclusion relies on the connection between the nuclear spin coherence lifetime and the molecular mobility in these systems, a notion we confirm by examining control samples engineered to exhibit a variable degree of rigidity (Section G of the Supplementary Information). In summary, the results herein introduce a new strategy to nanoscale nuclear spin sensing where the signal is recorded in the form of an ‘FID’ without the need for nuclear spin polarization or molecular labeling. The inherently small detection volume — of order 100 nm3 in the present case — makes this form of sensing ideal to investigate dynamical processes on mesoscales, hard to access with other experimental techniques. For example, our approach could be exploited to more clearly expose the role of nanoscale surface roughness on the dynamics of flow, or to experimentally test differing boundary conditions invoked at the liquid-solid interface23, 24 (e.g., no slip, multilayer locking, etc). Extensions articulating the present technique with known NMR protocols (e.g., homo- or heterospin decoupling sequences26) can be used to shed light on the chemical composition of adsorbed films in cases where the dynamics are insufficient to suppress inter-nuclear couplings. In particular, nuclear spin manipulation in the form of radio-frequency multipulse sequences can be applied during 𝜏 without deleterious effects on the NV response. Likewise, spin swap schemes — designed to exchange the spin states of the NV and its 14N (or 15N) host during the correlation interval 17 — provide a route to probe slow dynamical processes on time scales exceeding the NV longitudinal relaxation time6. Studies in this regime — more susceptible to the limited fluorescence contrast affecting the present correlation protocol — can benefit from the enhanced photon collection 12 efficiency recently demonstrated for NVs within engineered diamond nanostructures27, 28, 29. More in general, the present technique opens interesting opportunities for the investigation of chemical or biochemical systems affected by compositional heterogeneities or local aggregation. Cell membranes in particular could serve as a fascinating research platform, since molecular diffusion is non-Brownian and mostly restricted to nanoscale fluidic pockets (presumably) separating more rigid structures30. By the same token, experiments as a function of temperature and/or the composition of the bulk fluid can help explore various thermally- or chemically-activated processes including, for example, protein folding or the dynamics of molecular motors tethered to the diamond surface. References   1. 2. 3. 4. 5. 6. 7. 8. 9. Glover,  P.  &  Mansfield,  P.  Limits  to  magnetic  resonance  microscopy.  Reports  on  Progress   in  Physics  65,  1489  (2002).   Degen,   C.   L.,   Poggio,   M.,   Mamin,   H.   J.,   Rettner,   C.   T.   &   Rugar,   D.   Nanoscale   magnetic   resonance  imaging.  PNAS  106,  1313-­‐1217  (2009).   Rugar,   D.,   Budakian,   R.,   Mamin,   H.   J.   &   Chui,   B.   W.   Single   spin   detection   by   magnetic   resonance  force  microscopy.  Nature  430,  329-­‐332  (2004).   Gruber,   A.   et.   al.,   Scanning   confocal   optical   microscopy   and   magnetic   resonance   on   single  defect  centers.  Science  276,  2012-­‐2014  (1997).   Mamin,  H.  J.  et.  al.,  Nanoscale  nuclear  magnetic  resonance  with  a  nitrogen-­‐vacancy  spin   sensor.  Science  339,  557-­‐560  (2013).     Staudacher,   T.   et.   al.,   Nuclear   magnetic   resonance   spectroscopy   on   a   (5-­‐nanometer)³   sample  volume.  Science  339,  561-­‐563  (2013).   Taylor,  J.  M.  et.  al.,  High-­‐sensitivity  diamond  magnetometer  with  nanoscale  resolution.   Nature  Physics  4,  810-­‐816  (2008).   Müller,  C.  et.  al.,  Nuclear  magnetic  resonance  spectroscopy  with  single  spin  sensitivity.   Nature  Communications  5,  4703  (2014).   Sushkov,   A.   O.   et.   al.,   Magnetic   resonance   detection   of   individual   proton   spins   using   13 quantum  reporters.  Physical  Review  Letters  113,  197601  (2014).   10. Loretz,  M.,  Pezzagna,  S.,  Meijer,  J.  &  Degen,  C.  L.  Nanoscale  nuclear  magnetic  resonance   with   a   1.9-­‐nm-­‐deep   nitrogen-­‐vacancy   sensor.   Applied   Physics   Letters   104,   033102   11. 12. 13. 14. 15. 16. 17. 18. 19. 20. (2014).   DeVience,   S.   et.   al.,   Nanoscale   NMR   spectroscopy   and   imaging   of   multiple   nuclear   species.  Nature  Nanotechnology  (2015).   Rugar,   D.   et.   al.,   Proton   magnetic   resonance   imaging   using   a   nitrogen-­‐vacancy   spin   sensor.  Nature  Nanotechnology  (2014).   Häberle,  T.,  Schmid-­‐Lorch,  D.,  Reinhard,  F.  &  Wrachtrup,  J.  Nanoscale  nuclear  magnetic   imaging  with  chemical  contrast.  Nature  Nanotechnology  (2014).   Manson,  N.  B.,  Harrison,  J.  P.  &  Sellars,  M.  J.  Nitrogen-­‐vacancy  center  in  diamond:  Model   of   the   electronic   structure   and   associated   dynamics.   Physical   Review   B   74,   104303   (2006).   Kotler,  S.,  Akerman,  N.,  Glickman,  Y.,  Keselman,  A.  &  Ozeri,  R.  Single-­‐ion  quantum  lock-­‐in   amplifier.  Nature  473,  61-­‐65  (2011).   De  Lange,  G.,  Ristè,  D.,  Dobrovitski,  V.  V.  &  Hanson,  R.  Single-­‐spin  magnetometry  with   multipulse  sensing  sequences,  Physical  Review  Letters  106,  080802  (2011).   Laraoui,  A.  et.  al.,  High-­‐resolution  correlation  spectroscopy  of  13C  spins  near  a  nitrogen-­‐ vacancy  centre  in  diamond.  Nature  Communications  4,  1651  (2013).   C.P.  Slicher,  Principles  of  Magnetic  Resonance,  Springer  Series  in  Solid-­‐State  Sciences  1,   Springer,  1996.   Meriles,   C.   A.   et.   al.,   Imaging   mesoscopic   nuclear   spin   noise   with   a   diamond   magnetometer.  Journal  of  Chemical  Physics  133,  124105  (2010).   Eigen,   M.   &   Rigler,   R.   Sorting   single   molecules:   application   to   diagnostics   and   evolutionary  biotechnology.  Proceedings  of  the  National  Academy  of  Sciences  91,  5740-­‐ 5747  (1994).   14 21. 22. 23. 24. 25. 26. 27. 28. 29. 30. Bacia,  K.,  Kim,  S.  A.  &  Schwille,  P.  Fluorescence  cross-­‐correlation  spectroscopy  in  living   cells.  Nature  Methods  3,  83-­‐89  (2006).   Kim,   S.   A.,   Heinze,   K.   G.   &   Schwille,   P.   Fluorescence   correlation   spectroscopy   in   living   cells.  Nature  Methods  4,  963-­‐973  (2007).   Thompson,   P.   A.   &   Trojan,   S.   M.   A   general   boundary   condition   for   liquid   flow   at   solid   surfaces.  Nature  389,  360-­‐362  (1997).     Batchelor,   G.   K.   An   Introduction   To   Fluid   Dynamics   (Cambridge   University   Press,   Cambridge,  1967).   Ortiz-­‐Young,  D.,  Chiu,  H.-­‐C.,  Kim,  S.,  Voitchovsky,  K.  &  Riedo,  E.  The  interplay  between   apparent   viscosity   and   wettability   in   nanoconfined   water.   Nature   Communications   4,   2482  (2013).   Ernst,  R.  R.,  Bodenhausen,  G.  &  Wokaun,  A.  Principles  Of  Nuclear  Magnetic  Resonance  In   One  And  Two  Dimensions  (Clarendon  Press,  Oxford,  1987).   Momenzadeh,   S.   A.   et.   al.,   Nano-­‐engineered   diamond   waveguide   as   a   robust   bright   platform   for   nanomagnetometry   using   shallow   nitrogen   vacancy   centers.   Nano   letters   (2014).   Li,  L.  et.  al.,  Three  megahertz  photon  collection  rate  from  an  NV  center  with  millisecond   spin  coherence.  arXiv:1409.3068  (2014).   Neu,   E.   et.   al.,   Photonic   nanostructures   on   (111)-­‐oriented   diamond.   Applied   Physics   Letters  104,  153108  (2014).   Ritchie,   K.   et.   al.,   Detection   of   non-­‐brownian   diffusion   in   the   cell   membrane   in   single   molecule  tracking.  Biophysical  Journal  88,  2266-­‐2277  (2005).     Acknowledgements   We  thank  Dr.  D.  Pagliero  and  Dr.  Y.  Li  for  assistance  with  some  of  the  experiments.  T.S.   15 acknowledges  support  from  the  IMPRS-­‐AM.  C.A.M  acknowledges  support  from  the  National  Science   Foundation  through  grants  NSF-­‐1401632  and  NSF-­‐1309640.  This  work  was  supported  by  the  German   Science  Foundation:  SFB/TR21,  FOR1694,  EU  SQUTEC,  SIQS,  and  the  Max  Planck  Society.       Author  contributions   N.  R.,  S.  P.  and  J.  M.  implanted  the  diamond  sample  for  the  preparation  of  shallow  NV  centers.  C.  A.   M.  and  F.  R.  designed  the  experiment,  and  T.  S.  performed  the  experiments  and  analyzed  the  data.  J.   W.  supervised  the  project.  All  authors  contributed  to  the  writing  of  the  manuscript  and  gave   approval  for  its  final  version.     16    Supplementary  Information     Probing  molecular  dynamics  at  the  nanoscale  via  an  individual  paramagnetic  center         T.  Staudacher1,2,  N.  Raatz3,  S.  Pezzagna3,  J.  Meijer3,  F.  Reinhard1,  4,  C.  A.  Meriles5,*,  and  J.  Wrachtrup1     13rd  physics  Institute  and  Research  Center  SCoPE,  University  of  Stuttgart,  70569  Stuttgart,  Germany   2Max  Planck  Institute  for  Solid  State  Research,  70174  Stuttgart,  Germany   3Jan  Meijer   4TUM  –  Technical  University  of  Munich,  Walter  Schottky  Institut,  Am  Coulombwall  4,  85748   Garching,  Germany   5Department  of  Physics,  CUNY-­‐City  College  of  New  York,  160  Convent  Ave,  New  York,  NY  10031,  USA           *For  correspondence:  [email protected]       1 A. Materials  and  Methods   Diamond  samples   We  use  Type  IIa  CVD  grown  (100)-­‐oriented  diamond  samples  from  Element6.  The  NV  centers  used  in   this  study  were  generated  via  2.5keV  15N+  ion  implantation  into  the  diamond  substrates.  The   substrates  were  subsequently  annealed  at  240°C  for  2h,  followed  by  an  8h  annealing  at  850°C   temperature,  both  in  high  vacuum.  Afterwards  the  diamonds  were  boiled  in  a  1:1:1  mixture  of   H2SO4,  HClO4  and  HNO3,  in  order  to  remove  any  residual  graphitic  contaminations  from  the  diamond   surfaces.     Sample  preparation   We  measured  the  correlation  signals  for  three  different  organic  samples,  two  organic  polymers  (A,   B),  and  a  liquid  sample  (C).   For  the  measurements  of  the  solid  compounds  we  spin  coated  a  visibly  thick  layer  of  the  respective   material  onto  the  diamond  surface.  Sample  A  is  a  “liquid  coverslide”  material  dissolved  in  toluene   (Merckoglas,  Merck),  which  has  a  refractive  index  similar  to  glass  and  immersion  oil.  We  dilute  the   Merckoglas  base  solution  with  toluene  in  a  1:1  ratio  prior  to  spin  coating.  For  the  analysis  we  treat   Merckoglas  as  a  typical  organic  polymer,  such  as  Poly(methyl  methacrylate)  (PMMA),  and  assume  a   comparable  proton  density.     Sample  B  is  a  film  of  Polydimethylsiloxane  (PDMS).  For  this  purpose  a  droplet  of  PDMS  (Sylgard  184   Silicone  Elastomer,  Dow  Corning  in  a  10:1  mixing  ratio  of  the  base  to  the  curing  agent)  was  spin-­‐ coated  onto  the  diamond  surface.  Afterwards  the  PDMS  is  annealed  by  placing  the  diamond  onto  a   hotplate  for  2h  at  80°C.  We  note  that  the  signal  to  noise  ratio  under  the  PDMS  coating  is  much   smaller  than  for  the  other  two  samples,  and  requires  longer  measurement  times.  This  is  due  to  the   larger  mismatch  of  refractive  indices,  and  the  accompanied  loss  of  fluorescence  signal.   The  measurements  of  the  liquid  sample  C  were  performed  by  covering  the  diamond  surface  with   immersion  oil  (Fluka  Analytical,  10976).     2 Before/after  the  measurements,  the  samples  were  removed  by  rinsing  the  diamond  in  a  solvent   while  sonicating.  The  diamond  is  subsequently  boiled  in  the  above-­‐mentioned  1:1:1  acid  mixture  for   multiple  hours  to  remove  any  organic  residues  of  the  sample  or  solvent  from  the  surface.     To  ensure  the  observed  NMR  signal  (mainly)  originates  from  the  sample  rather  than  from  adsorbates   due  to  exposure  to  the  ambient  environment,  the  diamond  is  removed  from  the  acid  and   immediately  covered  with  the  respective  sample,  so  as  to  minimize  the  exposure  time  to  the   ambient.     B. Signal  reproducibility   All  experiments  were  carried  out  with  the  same  diamond  sample,  and  thus  with  the  same  set  of   shallow  NVs.  However,  because  the  diamond  crystal  must  be  physically  removed  from  the   microscope  for  surface  coating,  it  is  difficult  to  use  the  same  individual  NV  for  testing  nuclear  spins   from  different  sample  films.  To  circumvent  this  complication,  we  collected  data  from  multiple  (50  to   100)  individual  NVs  exposed  to  the  same  sample  film  (see  below).  Among  these  only  a  few  can  be   dynamically  decoupled  for  a  time  sufficiently  long  to  see  the  nuclear  spin  signature  via  an  XY8-­‐N   sequence  (a  likely  result  of  the  NV  depth  dispersion  and  heterogeneity  of  the  local  concentration  of   paramagnetic  impurities).  We  find  that  not  every  NV  center  exhibiting  a  nuclear  spin  induced  dip   under  the  XY8-­‐N  sequence  also  shows  detectable  nuclear  spin  correlation.  The  reason  for  this  is  still   not  fully  understood  but  it  may  relate  to  the  fortuitous  overlap  between  the  proton  spin  signature   and  the  fourth  harmonic  of  the  13C-­‐induced  dip  (see  below,  Section  H):  The  latter  can  be  easily   mistaken  by  the  former  in  an  XY8-­‐N  sequence  but  not  in  the  correlation  protocol  (where  the  nuclear   spin  Larmor  frequency  is  directly  probed).  Also  worth  noting  is  the  small  relative  amplitude  of  the   correlation  signal  (≈10%  of  the  maximum  possible  fluorescence  contrast  between  the  NV  spin   states),  which  leads  to  a  correspondingly  low  SNR  and  makes  detection  difficult.  Within  these   experimental  limitations,  Figs.  2  and  3  in  the  main  text  show  representative  data  sets  from  the  NVs   3 that  did  display  a  sizeable  correlation  signal.  Among  the  latter,  excellent  reproducibility  was   observed  between  data  sets  in  all  the  samples  we  explored,  as  shown  immediately  below.     Sample  A  –  Merckoglas   We  examined  a  set  of  52  NV  centers,  out  of  which  10  could  be  dynamically  decoupled  to  coherence   times  of  ≈100  µμs,  allowing  us  to  detect  the  proton  signal  via  an  XY8-­‐10  sequence.  Only  4  of  these   NVs  show  a  detectable  signal  after  an  XY8-­‐3  correlation  sequence.     Fig.  S1  shows  three  of  those  NV  centers,  the  one  presented  in  the  main  text  (NV20)  and  two     Figure  S1.  Signal  reproducibility  for  Sample  A.  (a)  Proton  correlation  signal  as  detected  from  an  NV  center   referred  to  as  NV  19  (open  circles)  for  Sample  A  (Merckoglas).  The  solid  trace  is  the  envelope  of  our  model   adapted  to  a  fully  static  proton  bath.  (b)  Fourier  transform  (magnitude  mode)  of  the  signal  in  (a)  centered   at  the  proton  Larmor  frequency  𝜈!  along  with  a  Lorentzian  fit  (solid  green  trace).  (c,d)  Same  as  in  (a,b)  for   NV  20.  (e,f)  Same  as  in  (c,d)  but  for  NV  26.     4 additional  ones.  We  find  a  reproducible  correlation  signal,  described  by  an  exponentially  damped   sinusoid  centered  around  the  proton  Larmor  frequency.  The  decay  envelope  (here  maintained   unchanged  for  all  NVs  in  the  figure)  is  qualitatively  captured  by  the  presented  model,  with  a  decay   time  of  20  µs.     Sample  B  –  PDMS   For  the  PDMS  coating  we  examined  a  set  of  85  NV  centers,  out  of  which  13  showed  proton  signal  via   an  XY8-­‐10  sequence.  Only  2  of  these  13  NVs  reveal  a  noticeable  signal  after  an  XY8-­‐3  correlation   sequence.     The  two  correlation  curves  shown  in  Fig.  S2  show  the  curve  presented  in  Fig.  3  of  the  main  text   (NV1)  and  a  second  curve  for  a  second  NV  (NV78).  As  explained  above,  experiments  with  this  sample   suffered  from  poor  collection  efficiency  and  required  particularly  long  integration  times,  the  reason   why  the  second  data  set  was  measured  only  for  evolution  times  <20  µμs.  Both  measured  signals  are   overlayed  with  the  same  calculated  decay  envelope,  assuming  a  slow  molecular  dynamics  of  the   PDMS  molecules  as  described  in  the  main  text  (see  also  Section  D  below).   Figure  S2.  XY8-­‐3  correlation  signal  under  a  coating  of  sample  B  for  two  different  NV  centers.  The  time   domain  data  is  compared  with  a  decay  envelope  calculated  assuming  partial  motional  narrowing  due  to   slow  molecular  dynamics  as  described  above.         Sample  C  –  Immersion  oil   For  the  immersion  oil  a  set  of  83  NV  centers  was  measured.  Among  them,  nine  NVs  exhibited  proton   signal  after  an  XY8-­‐10  sequence.  Five  of  those  NVs  showed  an  XY8-­‐3  correlation  signal.  We  measured   5 two  of  those  NVs  over  a  longer  evolution  time  window  (>80µμs,  NV6  is  the  one  shown  in  the  main   text  and  related  SI  sections).  As  shown  in  Fig.  S3,  we  measure  virtually  identical  responses   characterized  by  a  fast  initial  decay  and  a  long-­‐lived  tail  as  highlighted  in  the  main  text.  Remarkably,   the  two  data  sets  are  so  consistent  that  the  same  set  of  parameters  (adsorbed  layer  thickness,   translational  diffusion  constant,  and  rotational  diffusion  constant)  can  be  used  to  describe  both   observations.       Figure  S3.  Signal  reproducibility  in  the  presence  of  a  liquid.  (a)  Proton  correlation  signal  from  a  shallow   NV  (referred  to  as  NV  6,  open  circles)  in  the  presence  of  immersion  oil.  The  yellow  and  light  blue  traces  are   the  envelopes  of  an  exponentially  decaying  sinusoidal  and  the  presented  model,  respectively.  The   exponential  fit  considers  the  initial  decay  𝜏≤25  µμs  whereas  the  model  considers  contributions  from   molecules  in  the  bulk  liquid  and  semi-­‐mobile  molecules  from  an  adsorbed  layer  as  described  in  the  main   text.  (b)  Zoomed  image  of  the  data  points  within  the  shaded  area  in  (a);  the  light  blue  trace  is  the  segment   of  a  fit  to  a  sinusoidal  function  at  the  proton  Larmor  frequency  scaled  by  the  envelope  in  (a).  (c,d)  Same  as   in  (a,b)  but  for  a  different  NV  (identified  as  NV  11  in  the  image).  In  both  cases  we  use  the  same   parameters,  namely,  the  adsorbed  film  thickness  as  well  as  the  translational  and  rotational  diffusion   coefficients.       C. Modeling  the  correlation  signal  envelope   The  envelope  of  the  correlation  signal  is  described  by  (see  Eq.  (2)  of  the  main  text)   6 𝑝!,!𝜏  𝑒!! !!!,! where  𝑝!,!𝜏  is  the  conditional  probability  that  the  i-­‐th  nuclear  spin  of  the  j-­‐th  molecule  remains  in   the  detection  volume  over  the  evolution  time  𝜏,  and  𝑇!!,!  is  the  transverse  relaxation  time  of  said   !,! ,       (S1)   nuclear  spin.  Here  we  neglect  any  additional  contribution  to  the  envelope  arising  from  longitudinal   relaxation  of  the  NV  electron  spin  (see  below  Section  F).  This  is  a  reasonable  assumption  as  typical   longitudinal  relaxation  times  are  in  the  order  of  hundreds  of  µs,  while  we  focus  here  on  effects  on   timescales  of  a  few  tens  of  µs,  i.e.  the  time  range  of  our  measurements.  Before  describing  how  we   model  the  probability  and  relaxation  time  in  detail,  we  introduce  below  the  underlying  assumptions   and  simplifications  of  the  presented  theory.     First,  we  assume  that  the  value  of  the  sum  in  Eq.  (S1)  is  time  independent,  i.e.  the  total  number  of   protons/molecules  inside  the  detection  volume  stays  approximately  constant  over  time.  Therefore   the  amplitude  and  decay  components  of  the  correlation  signal  depend  solely  on  the  spatial   distribution  of  the  molecular  mobility  with  respect  to  the  surface  distance.  Further,  we  neglect  any   hydrodynamic  interactions  among  molecules  other  than  those  implicit  when  assigning  a  candidate   translation  or  rotation  diffusion  coefficient.   The  molecules  are  treated  as  rigid  spheres  of  radius  𝑎,  and  their  molecular  mobility  varies  on  the   diffusion  constants  𝐷!(𝑟)  and  𝐷!(𝑟)  for  the  translational  and  rotational  diffusion  respectively,  which   length  scale  of  the  detection  volume.  This  heterogeneity  is  described  by  position  dependent   takes  into  account  the  possible  presence  of  adsorbate  molecules  on  the  diamond  surface.  We   estimate  that  the  transition  between  adsorbate  and  free  molecules  is  sharp  and  that  the  latter  obey   classical  diffusion  equations,  as  well  as  the  Stokes-­‐Einstein-­‐(Debye)  equations  for  the  diffusion   constants     𝐷!= 𝑘!𝑇6𝜋𝜂𝑎    ;                                                S2   7 𝐷!= 𝑘!𝑇8𝜋𝜂𝑎!    ;                                              (S3)   where  𝑘!  is  the  Boltzmann  constant,  𝑇  is  the  medium  temperature,  and  𝜂  is  the  viscosity  of  the   sample  medium.     We  further  assume  that  the  diffusion  heterogeneity  is  restricted  to  the  z-­‐axis  perpendicular  to  the   diamond  surface.  The  rotational  diffusion  constant  can  have  a  finite  value  at  the  surface,  while  the   translational  diffusion  constant  goes  to  zero.  These  values  stay  approximately  constant  until  a   distance  𝑧!"#  to  the  surface,  which  we  define  as  the  thickness  of  the  adsorbate  layer.  Beyond  this   layer,  the  diffusion  constants  are  related  to  each  other  via  𝐷!= !!!!𝐷!.  We  assign  to  both  diffusion   constants  a  similar  transition  profile,  which  we  model  as  (Fig.  S4a)   𝐷𝑧 = 𝐷!"#$%&’,𝑧≤𝑧!"# (𝐷!"#$−𝐷!"#$%&’)   1−exp −!!!!"#! +𝐷!"#$%&’,      𝑧>𝑧!"#  ,   (S4)   scale  small  compared  to  the  radial  size  of  the  detection  volume.   experiments  measuring  the  transition  of  surface  to  bulk  fluid  dynamics1.  We  find  that  the  exact  value   where  𝐷!"#$,and  𝐷!"#$%&’  are  the  respective  values  of  the  diffusion  constants  in  the  bulk  of  the   volume  and  at  the  surface,  and  𝜁  is  a  transition  parameter,  which  we  keep  fixed  at  an  empirical  value   of  𝜁=0.25  nm.  We  note  that  this  value  is  in  the  same  order  of  magnitude  as  that  found  for  AFM   of  𝜁  has  a  negligible  impact  on  the  results  of  the  simulation,  as  long  as  the  transition  takes  place  on  a   The  coherences  of  the  nuclear  spin  bath  characterized  by  the  transverse  relaxation  time  𝑇!!,! the  conditional  probability  𝑝!,!𝜏  .   Nuclear  spin  relaxation  time  𝑻𝟐     The  nuclear  spin  dynamics  mainly  depends  on  position  dependent  correlation  times  of  motion  𝜏!.   calculated  using  established  means  known  from  classical  NMR2,  3,  4.    In  the  following  we  will  briefly   outline  the  calculation  steps  of  the  relaxation  time,  before  describing  the  model  we  use  to  estimate    can  be   8 The  only  interaction  considered  among  the  nuclei  is  homonuclear  dipolar  coupling,  so  that  one  can   distinguish  different  time  regimes  by  the  ratio  of  the  rate  of  nuclear  spin  reorientation  (∝  𝜏!!!)   versus  the  dipolar  coupling  constant  𝜔!.     Among  the  interactions  responsible  for  the  correlation  decay  one  can  distinguish  between   intramolecular  interactions  inside  a  molecule,  and  intermolecular  interactions  among  spins  of   different  molecules.  In  the  interior  of  a  molecule,  the  variation  of  the  dipolar  coupling  between  spins   arises  almost  exclusively  from  the  rotation  of  the  molecule  (neglecting  distance  variations  due  to   vibrations),  while  for  the  interactions  between  spins  in  different  molecules  their  relative  translation   must  be  considered.     1H-­‐1H  interaction   The  homonuclear  dipolar  coupling  between  two  proton  spins  is  described  by  the  Hamiltonian   𝐻!"#=  !!ℏ!!!  !!"!!!!!  (𝐼!𝐼!−3(!!!  !!)(!!!  !!) !!!! ),                        (S5)   where  𝜇!  is  the  magnetic  constant,  ℏ  is  the  reduced  Planck  constant,  𝛾!"  is  the  gyromagnetic  ratio   of  a  proton  spin,  𝐼!  are  the  spin  vector  operators  of  the  i-­‐th  proton,  and  𝑟!!  is  the  distance  vector   between  two  proton  spins.  Squaring  Eq.  (S5),  averaging  over  the  azimuthal  and  polar  angles,  and   taking  the  trace  over  the  spin  matrices  yields  an  expression  for  the  rms-­‐dipolar  coupling  strength5,  6   𝜔!!=  !"!   !!  !!  ℏ!!  !!"!!!!!    ,                                                              (S6)   where  we  use  a  representation  of  𝑟!!  in  spherical  coordinates.  In  Eq.  (S6)  𝑟!!  is  the  average   distance  to  the  nearest  nuclear  neighbor,  which  is  dependent  on  the  proton  density  𝜌  and  is   distributed  according  to  𝑃𝑟!! =4𝜋𝜌  𝑟!!! exp(−!!!𝑟!!! 𝜌)    .                              (S7)   This  results  in  an  average  proton-­‐proton  distance  of     9    𝑟!! ≈!!  𝜌!!!  .     𝜔!!=63   !!  !!ℏ!!  𝛾!"!𝜌.                   (S8)        (S9)   Combining  Eq.  (S8)  and  Eq.  (S6)  yields  an  expression  for  the  average  interaction  strength     The  relaxation  rate  of  the  axial  magnetization  component  for  a  2-­‐spin  cluster  due  to  their  mutual   dipolar  interaction  is  calculated  via5,  6,  7   𝑓!"#=!!  !!ℏ!!  𝛾!!! !!!! 𝜌      .                                                          (S10)   Since  we  assume  a  constant  proton  density  throughout  the  detection  volume,  the  average   interaction  strength  and  dipolar  relaxation  rate  are  therefore  position  independent.  Using  a  typical   value  for  the  proton  density  in  organic  compounds  of  𝜌=50  nm!!  for  all  samples,  yields  a   fluctuation  rate  of  𝑓!"#≈20  kHz,  and  average  interaction  strength  of   𝜔!!≈45  kHz.     Translational  diffusion  of  1H  spins   The  translational  diffusion  of  molecules  generates  a  small  fluctuating  magnetic  field.  The  fluctuation   rate  due  to  translational  diffusion  is  given  by5   𝑓!"#$%𝑟 =2𝐷!𝑟   !!  ! ! .       (S11)   Rotational  diffusion  of  1H  spins   For  a  spherical  molecule  of  radius  𝑎  rotating  in  a  liquid  of  viscosity  𝜂  the  rotational  fluctuation  rate  is   described  by  Stokes’  law2   𝑓!"#𝑟 = !  !!! !!!!  !(!).     𝑓!"#𝑟 =6𝐷!𝑟 .     Using  the  Stokes-­‐Einstein  relation  (Eq.  (  S3))  we  find     (S12)   (S13)           10 Combined  dynamics  and  relaxation  rate   In  order  to  describe  the  nuclear  spin  relaxation  dynamics  of  the  sample  molecules  we  use  an   autocorrelation  function  of  the  fluctuating  magnetic  field  B(t)  experienced  by  the  nuclear  spins   𝐵𝑡𝐵(𝑡!) =  𝐵! exp(−!!!!!! ),                        (S14)   where  𝐵!  is  the  rms  field  strength  experienced  by  the  protons  spins,  and  𝜏!  is  the  characteristic   .  The  spectral  density  𝐽(𝜔,𝜏!)  is  twice  the  Fourier  transform   𝜏!(𝑟)= 𝑓!"#+𝑓!"#$%(𝑟)+𝑓!"#(𝑟) !! correlation  time  of  the  magnetic  fluctuations,  which  includes  all  above  mentioned  mechanisms,  i.e.     Figure  S4.  Calculation  of  the  nuclear  spin  relaxation  process.  (a)  Translational  and  rotational  diffusion   constants  with  increasing  distance  to  the  diamond  surface.  The  rotational  constant  has  a  finite  value  at  the   surface,  and  both  constants  are  linked  via  the  Stokes-­‐Debye-­‐Einstein  relations  in  the  outside  layer.  (b)  The   varying  diffusion  constants  result  in  different  fluctuation  rates  throughout  the  detection  volume,  the   dipolar  interaction  is  constant,  as  we  assume  a  constant  proton  density  throughout  the  volume.  (c)  The   relaxation  time  is  dominated  by  the  strong  dipolar  interactions  in  the  static  adsorption  layer  but  quickly   prolongs  with  faster  molecular  dynamics.  (d)  Integrated  NMR  decay  over  the  detection  volume,  for   different  molecular  interactions.  The  NMR  signal  quickly  decays  in  the  absence  of  molecular  dynamics   (blue  curve),  gets  longer  in  the  presence  of  translational  diffusion  (yellow  curve),  and  is  maximum  if  also   rotational  diffusion  is  allowed  (green,  red  curve).    Same  parameters  as  in  the  main  manuscript:  molecule   radius  of  a  =  0.5  nm,  proton  density  𝝆  =  50  nm-­‐3,  adsorption  layer  thickness  zads  =  1.5  nm,  surface   translational  diffusion  constant  of  DT,surface  =  0.0  nm2/µs,  bulk  translational  diffusion  constant  of  DT,bulk  =  0.3   nm2/µs,  a  rotational  diffusion  constant  on  the  surface  of  DR,surface  =  0.05  rad2/µs,  and  a  rotational  diffusion   constant  in  the  bulk  of  DR,bulk  =  3/4a2  DT,bulk.     11 of  Eq.  (S14)  4   𝐽𝜔,𝜏! =2𝐵!   !!!!!!!  .   !! The  normalized  spectral  density  𝐽(𝜔,𝜏!)  is     𝐽𝜔,𝜏! = !! !!!!!!!  .                 (S15)                                              (S16)   The  transverse  relaxation  is  caused  by  the  dephasing  of  the  proton  spins  due  to  their  mutual  dipolar   interactions.    In  the  case  of  intramolecular  (homonuclear)  dipole-­‐dipole  relaxation,  the  transverse   𝑇!!!𝑟  = !!" 𝜔!!!  3𝐽0,𝜏!𝑟 +5𝐽𝜔!,𝜏!𝑟 +2𝐽2𝜔!,𝜏!𝑟 relaxation  rate  𝑇!!!  between  two  spin  ½  nuclei  is  given  by3,  4   ,                        (S17)   where  𝜔!  is  the  nuclear  Larmor  precession  frequency,  and   𝜔!!  is  the  average  dipole-­‐dipole   Probability  𝐩𝐢,𝐣𝛕       We  model  the  probability  𝑝!,!𝜏  using  a  semiquantitative  geometric  model.  Simply  put,  we  draw  a   ‘diffusion  sphere’  around  a  starting  position  𝑟  with  a  radius     𝑟!"##(𝑟,𝜏)= 6  𝐷!𝑟  𝜏,   coupling  constant,  given  by  Eq.  (S6).     (S18)   which  is  the  rms-­‐diffusion  distance  in  three  dimensions  for  Brownian  motion.  We  only  consider   translational  diffusion  here,  and  the  translational  diffusion  constant  𝐷!𝑟  only  depends  on  the   starting  position  𝑟  (Fig.  S4a,  Fig.  S5a).  The  conditional  probability  to  find  the  molecule  inside  the   detection  volume  (𝑉!")  after  the  time  𝜏  is  then  approximately  given  by  the  ratio  of  the  overlap   between  the  diffusion  volume  (𝑉!"!!)  and  the  detection  volume,  normalized  to  the  detection  volume   (Fig.  S5b):   12 Figure  S5.  Geometric  model  for  signal  loss  due  to  translational  diffusion  (a-­‐c)  The  detection  volume   (green)  can  be  approximately  described  by  a  heterogeneous  hemisphere  comprising  a  static  near-­‐surface   layer  and  a  mobile  outer  segment.  The  probability  of  a  molecule  staying  within  the  detection  volume   during  𝜏  depends  on  the  assumed  translational  diffusion  constant  in  the  outer  section  of  the  hemisphere.     This  probability  (growing  from  (a)  to  (c))  is  described  by  the  overlap  (red)  of  the  diffusion  sphere  with  the   detection  volume.  (d)  For  a  translational  diffusion  constant  of  DT,bulk  =  0.3  nm2/µs,  we  find  that  the   probability  of  molecules  staying  inside  the  detection  volume  quickly  diminishes  (red  curve),  unless  a   sizeable  adsorption  layer  is  present.   𝑝!,!𝜏  =  !!"##∩!!" !!" .       (S19)   translates  to  an  expected  volume  of  (5  nm)³.      The  detection  volume  is  simplified  as  a  hemisphere  with  radius  𝑅!"=4  nm,  which  roughly   The  overlap  𝑉!"##∩𝑉!"  can  be  calculated  via  the  intersection  of  two  spheres   𝑉!"#$%&$’  =  !  !!"  !  !!"##  !  ! !(!!!!!!!"!  !!"##  !  !!!"!!!"## !) where  𝑟  is  the  absolute  value  of  the  starting  position  vector  𝑟,  and  𝑟!"##  is  obtained  from  via  Eq.   inside  the  detection  volume  (𝑝!,!𝜏  =1),  or  if  the  diffusion  sphere  becomes  larger  than  the   detection  volume.  In  the  latter  case  the  probability  𝑝!,!𝜏  is  determined  by  the  ratio  of  the  diffusion   (S18).  The  intersection  volume  will  have  a  negative  value  if  either  the  diffusion  sphere  is  completely   !"  !  ,              (S20)   sphere  volume  to  the  detection  volume.     13 Since  the  detection  volume  is  only  a  hemisphere,  we  have  to  subtract  any  intersection  or  diffusion   volume  contributions  extending  below  the  surface  before  calculating  the  value  for  the  probability.   The  volume  beneath  the  surface  can  be  calculated  as  the  volume  of  a  sphere  cap  (𝑉!"#).  Therefore   the  volume  below  the  surface  can  be  calculated  as   𝑉!"#$%=𝑉!"#−𝑉!"#$#,      with  𝑉!"#=  !!ℎ!  (3𝑟!"##−ℎ),  where  ℎ=𝑟!"##−𝑧  is  the  height  of  the  cap,  and  𝑧  is  the  z-­‐ component  of  the  position  vector  𝑟.  The  volume  of  the  piece  which  would  be  subtracted  is   (S21)     approximately  given  by       𝑉!"#$#≈ !!!  𝑉!"#,       where  𝑏=   𝑟!"##! −𝑧!      is  the  radius  of  the  circular  base  of  the  cap  volume,  and  𝑐=𝑎+𝑏−𝑅!"= 𝑟!−𝑧!+   𝑟!"##! −𝑧!−𝑅!".     (S22)     comparison  to  the  experimental  data   D. Monte  Carlo  (MC)  simulation  of  the  molecular  dynamics  in  the  detection  volume  and   For  the  simulation  we  pick  random  starting  positions  𝑟  inside  the  detection  volume,  then  calculate   the  probability  to  remain  within  this  volume  for  varying  𝜏  as  described  above,  i.e.  varying  the  radii  of   the  diffusion  sphere.  For  each  𝜏  we  calculate  an  average  correlation  time  𝑇!!,! decay  exp(−𝜏/𝑇!!,! ).  We  do  an  ensemble  average  over  multiple  starting  positions  to  get  the   envelope  of  the  correlation  signal.  We  vary  the  values  of  the  adsorption  layer  thickness  𝑧!"#,  the   rotational  diffusion  constant  at  the  surface  𝐷!,!"#$%&’,  and  the  translational  diffusion  constant  in  the   bulk  𝐷!,!"#$  ,  and  overlay  the  results  with  the  experimental  data  to  obtain  quantitative  estimates  of   highest  and  the  lowest  z-­‐value  in  the  diffusion  sphere.  The  probability  is  then  multiplied  by  the  NMR    between  the   14 the  molecular  mobility  in  the  samples.   We  find  good  qualitative  agreement  between  the  simulation  and  the  experimental  data.  For  Sample   C  (immersion  oil,  Fig.  S6)  the  model  manages  to  reproduce  both  significant  features,  the  initial  fast   decay  (attributed  to  the  molecules  diffusing  outside  the  detection  volume),  and  the  long  lasting   signal  tail  (described  by  the  NMR  decay  of  the  molecules  inside  the  adsorption  layer).  The   ‘amplitude’  of  the  long  lasting  decay  is  set  by  the  thickness  of  the  adsorption  layer  (Fig.  S5d),  from   which  a  value  of  1.5  nm  is  estimated.  This  result  is  in  good  agreement  with  prior  experimental   observations8,  9,  and  would  correspond  to  approximately  1-­‐2  layers  of  molecular  adsorbates  on  the   surface.  The  translational  diffusion  constant  mainly  influences  the  fast  initial  drop  of  the  signal,  and   we  find  best  agreement  for  a  value  of  𝐷!,!"#$  ~0.3    nm2/µs.  We  note  that  this  is  in  the  order  of   magnitude  that  one  would  expect  from  Eq.  (S2),  which  yields  a  value  of  𝐷!,!"#$  ~  1.0  nm2/µs,   corresponding  to  a  molecule  of  radius  a  =  0.5  nm,  a  medium  of  viscosity  𝜂=  437  mPa.s  (as  listed  in   the  compound  datasheet),  and  room  temperature  conditions.  The  deviations  could  be  readily   explained  by  varying  the  molecule  radius  a,  as  the  immersion  oil  contains  various  different  molecules   of  different  chain  length  etc.,  which  are  treated  identically  in  the  simulations.         Figure  S6.  Comparison  of  the  MC  simulation  with  the  immersion  oil  data.  A  scaled  simulation   (red/green  trace)  is  overlaid  over  the  experimental  data  points  (grey  squares).  The  blue  curve  is  a  fit  to  a   sinusoidal  using  the  simulation  as  an  envelope.  The  parameters  used  for  the  simulation  were  a  molecule   radius  of  a  =  0.5  nm,  proton  density  𝝆  =  50  nm-­‐3,  adsorption  layer  thickness  zads  =  1.5  nm,  bulk   translational  diffusion  constant  of  DT,bulk  =  0.3  nm2/µs,  and  a  rotational  diffusion  constant  on  the  surface   of  DR,surface  =  0.05  rad2/µs.  The  red  (green)  segment  of  the  envelope  highlights  the  part  of  the  decay   dominated  by  translational  (rotational)  diffusion  from  molecules  in  the  bulk  fluid  (adsorbed  to  the   diamond  surface).  The  amplitude  of  the  green  segment  depends  on  the  thickness  zds  assumed  for  the   adsorbed  layer.       15 Our  model  also  manages  to  describe  the  decay  times  of  the  purely  exponential  signal  envelopes  for   the  solid  polymers  quite  well.  Here  we  assume  no  translational  diffusion  throughout  the  detection   volume,  i.e.  the  whole  volume  is  treated  as  an  ‘adsorption  layer’.  We  then  vary  the  value  of  the   rotational  diffusion  constant  DR,surface  until  we  obtain  a  reasonable  match  with  the  experimental  data.    In  the  case  of  Merckoglas  (sample  A)  we  can  only  reproduce  the  signal  shape  by  removing  the   rotational  diffusion  as  well,  i.e.  molecules  remain  static  over  the  course  of  the  measurement.  In  this   case  the  probability  𝑝!,!𝜏  becomes  unity,  and  the  signal  decay  is  mainly  governed  by  NMR   relaxation  due  to  inter-­‐  and  intramolecular  dipolar  interactions,  yielding  a  fast  single  exponential   decay  (Fig.  S7a)  as  expected  for  a  solid.   Figure  S7.  Simulated  correlation  envelopes  for  the  solid-­‐state  samples  (a)  The  Merckoglas  data  can  be   readily  described  by  a  completely  static  bath  of  molecules,  i.e.  by  a  purely  dipolar  NMR  relaxation,  yielding   a  single  exponential  decay  envelope  (yellow  curve),  in  good  agreement  with  Eq.  1  in  the  main  manuscript.   (b)  The  PDMS  data  can  be  readily  reproduced  by  either  assuming  slow  molecular  rotation  with  no  center-­‐ of-­‐mass  translation  (blue  curve,  DR,surface  =  0.001rad2/µs),  or  by  assuming  a  completely  static  layer  with  a   (~20%)  reduced  proton  density  of  40  nm-­‐3  (green  curve).     For  PDMS  the  correlation  signal  also  decays  exponentially  but  the  decay  time  (of  order  ~30  µs)  is   found  to  be  longer  than  anticipated  for  an  ensemble  of  static,  dipolarly-­‐coupled  protons  (at  least   assuming  a  standard  proton  density  of  𝜌  =  50  nm-­‐3).  We  can  reproduce  the  observed  behavior  by   assuming  a  rotational  diffusion  constant  of  DR,surface  =  0.001  rad2/µs  throughout  the  detection   volume.  The  latter  is  supported  by  prior  NMR  experiments  showing  that  PDMS  can  exhibit  significant   molecular  dynamics  at  room  temperature,  thus  leading  to  partial  motional  narrowing10,  11.  The  self-­‐ diffusion  depends  on  the  PDMS  chain  length/molecular  weight,  i.e.,  on  the  mixing  ratio  of  the   16 components.  Even  if  partly,  this  enhanced  mobility  can  be  the  reason  of  the  line  narrowing  captured   in  the  non-­‐negligible  value  of  the  rotational  diffusion  constant  in  the  simulation.  For  completeness,   we  note  that  the  observed  data  set  can  also  be  reproduced  by  a  static  ensemble  of  protons  with  a   ~20%  lower  density  than  normal  (40  nm-­‐3  as  opposed  to  50  nm-­‐3,  lighter  green  trace  in  Fig.  S7b).   However,  in  light  of  the  experimental  evidence  mentioned  above,  this  alternative  seems  less  realistic   (see  also  Section  G).       E. Model  precision  and  accuracy   As  described  above  the  key  parameters  in  our  model  are  the  bulk  translational  diffusion  coefficient   𝐷!,!"#$  (which  governs  the  initial  signal  decay  for  a  liquid  sample),  the  rotational  diffusion  coefficient   in  the  semi-­‐mobile  adsorption  layer  𝐷!,!"#$%&’    (describing  the  decay  time  of  the  long  lasting  signal   tail),  and  the  thickness  of  the  adsorption  layer  𝑧!!"  (which  sets  the  amplitude  of  this  long  signal   component).  Fig.  S6  schematically  visualizes  the  role  of  each  parameter  on  the  resulting  decay   envelope.     As  shown  in  Fig.  S8  for  Sample  C,  we  find  that  the  modeled  response  of  NV  6  is  quite  sensitive  to  the   values  assigned  to  these  parameters.  For  example,  Fig.  S8a  shows  the  initial  decay  of  the  correlation   One  can  qualitatively  estimate  a  confidence  interval  (light  blue  shaded  area)  of   signal  due  to  the  translational  diffusion  of  the  sample  molecules  for  varying  translational  diffusion   coefficients  𝐷!,!"#$.  Here  the  values  for  the  rotational  diffusion  inside  the  semi-­‐mobile  layer   (𝐷!,!"#$%&’=0.05  rad!/µμs)  and  the  adsorption  layer  thickness  (z!"#=1.5  nm)  are  kept  constant.   𝐷!,!"#$=  [0.2!!!!!,0.4!!!!!]  for  which  the  model  parameters  reproduce  the  envelope  shape  to  a   measurement,  allowing  us  to  distinguish  changes  of  the  diffusion  coefficient  within  ∼30%.       comparable  extent.  Beyond  this  interval  the  calculated  response  notably  departs  from  the   17 Fig.  S8b  shows  the  long  decay  component  of  the  correlation  signal  for  varying  rotational  diffusion   coefficients  𝐷!,!"#$%&’.  Here  we  keep  the  values  for  the  translational  diffusion  in  the  bulk  layer   (𝐷!,!"#$=0.3  nm!/µμs)  and  the  adsorption  layer  thickness  (z!"#=1.5  nm)  constant.  Over  the   measured  proton  evolution  time  we  find  that  the  correlation  amplitude  stays  approximately   constant,  with  no  distinct  signal  decay.  Therefore  we  can  only  estimate  a  lower  bound  for  the   rotational  diffusion  coefficient  for  which  we  can  reproduce  this  behavior.  We  find  that  this  signal   shape  is  reproduced  by  rotational  diffusion  coefficients  larger  than  𝐷!,!"#$%&’=  0.05!"#!!! ,  and  that  a   notable  deviation  from  the  data  can  be  observed  for  coefficients  below  𝐷!,!"#!"#$=  0.02!"#!!! .   Therefore  within  the  measurement  region  the  model  can  distinguish  changes  below  0.05!"#!!!  with  a   precision  of  ∼60%.     Figure  S8.  Variation  of  the  model  parameters  and  influence  in  the  obtained  decay  envelope.  (a)  Variation     of  the  translational  diffusion  coefficient  DT,bulk  ,  the  respective  value  is  noted  in  units  of  [nm2/µμs].  (b)   value  is  noted  in  units  of  [rad2/µμs]..  (c)  Variation  of  the  adsorption  layer  thickness  zads,  the  respective  value   Variation  of  the  rotational  diffusion  coefficient  in  the  semi-­‐mobile  adsorption  layer  DR,surface,  the  respective   is  noted  in  units  of  [nm].  (d)  Influence  of  anomalous  diffusion  on  the  Initial  signal  decay.  In  (a)  through  (d)   the  data  points  correspond  to  NV  6  and  the  diamond  coating  is  sample  C  (immersion  oil).  In  all  cases  only   the  noted  parameter  is  changed  while  the  others  have  a  fixed  value  coincident  with  that  noted  in  the  main   text.       18 Fig.  S8c  shows  the  variation  in  the  amplitude  of  the  long  decay  component  of  the  correlation  signal   for  varying  adsorption  layer  thicknesses.  In  this  case  the  translational  diffusion  in  the  bulk  layer   (𝐷!,!"#$=0.3  nm!/µμs)  and  the  rotational  diffusion  in  the  adsorption  layer  (𝐷!,!"#$%&’=  0.05!"#!!! )   are  kept  constant.  We  can  estimate  a  confidence  interval  of  z!"#=[1.25  nm,1.75  nm],  which   results  in  comparable  decay  responses  corresponding  to  a  precision  of  ∼17%.     The  relatively  good  precision  of  our  model  in  discriminating  between  different  numerical  values  of   the  fit  parameters  must  be  distinguished  from  the  model  absolute  accuracy  which,  in  light  of  the   various  simplifications  (Section  C),  should  be  considered  only  moderate.  For  example,  one  of  the   underlying  assumptions  is  that  the  sample  molecules  are  spherical  and  that  they  diffuse  according  to   a  free  Brownian  motion.  The  effect  of  anisotropic  diffusion  due  to  molecular  non-­‐sphericity  is   difficult  to  estimate.  We  can,  however,  qualitatively  examine  the  effects  of  non-­‐Brownian  motion  by   considering  the  case  of  anomalous  diffusion.  In  this  case,  the  diffusion  radius  is  given  by   𝑟!"##!"#$!%#&’(𝑟,𝜏)= Γ  𝐷!𝑟  𝜏!,     (S23)   to  cellular  transport  processes).     where  Γ  is  a  scaling  factor,  and  𝛼  characterizes  the  diffusion  process.  In  the  case  of  a  typical   (Brownian)  diffusion  𝛼=1,  while  𝛼>1  is  often  associated  with  super-­‐diffusion  processes  (e.g.,  due   We  set  Γ=6  (corresponding  to  3D  Brownian  diffusion)  since  the  exact  value  only  affects  the   Brownian  diffusion  to  super-­‐diffusion  as  we  increase  the  exponent  𝛼,  while  keeping  the  other  values   constant  (𝐷!,!"#$=0.3  nm!/µμs,  𝐷!,!"#$%&’=  0.05!"#!!! ,    z!"#=1.5  nm).       regime,  as  the  faster  signal  decay  can  be  partly  compensated  by  reducing  the  value  of  Γ  and/or   magnitude  of  the  diffusion  constant  but  not  the  decay  envelope.  Fig.  S8d  shows  the  transition  from   qualitatively  similar  to  free  diffusion.  This  leads  to  a  certain  ambiguity  in  determining  the  diffusion   The  change  to  super-­‐diffusion  leads  to  a  faster  signal  decay,  while  maintaining  an  overall  shape   19 𝐷!,!"#$.       F. Role  of  NV  spin-­‐lattice  relaxation   The  observation  of  a  correlation  signal  depends  on  the  NV  ability  to  store  the  phase  picked  up  during   the  first  interrogation  segment  throughout  the  evolution  interval  𝜏.  A  question  of  interest  is,   therefore,  whether  NV  spin-­‐lattice  relaxation  𝑇!!"  must  be  taken  into  account  in  our  modeling.     While  we  did  not  specifically  determine  𝑇!!"  for  the  NVs  we  used  in  our  correlation  measurements,   observations  from  virtually  identical  NVs  are  presented  in  Fig.  S9.  The  plotted  data  points   correspond  to  the  difference  between  pump-­‐probe  and  inversion-­‐recovery  protocols.  Out  of  the  25   NVs  we  studied,  we  determine  an  average  spin-­‐lattice  relaxation  time  of  ~350  µs  with  a  maximum   subtracting  the  NV  signals  from  pump-­‐probe  and  inversion  recovery  protocols  (both  differ  only  on  the   Figure  S9.  Impact  of  NV  spin-­‐lattice  relaxation  on  the  correlation  signal.  (a)  We  determine  𝑇!(!") application  of  a  π-­‐pulse  immediately  after  the  pump  laser  pulse).  (b)  Representative  𝑇!(!") (squares);  the  solid  line  is  a  fit  to  an  exponential  decay  with  time  constant  𝑇!(!")=  301  µs.  For   the  long-­‐lived  section  of  the  curve  decays  with  a  characteristic  time  𝑇!"##(!"#$)~250  µs.  (c)  Out  of  the  25   comparison,  the  red  trace  reproduces  the  (extrapolated)  envelope  of  the  correlation  signal  for  Sample  C;      by    measurement   NVs  investigated,  most  NVs  spin-­‐relax  with  a  characteristic  time  of  ~300  µs.  The  data  sets  in  the  plot  are   to  be  considered  representative  of  the  NV  ensemble.  (d)  Histogram  with  the  distribution  of  NV  spin-­‐ lattice  relaxation  times.       20 dispersion  between  𝑇!!"  values  of  about  100  µs.  These  relaxation  times  are  substantially  longer   than  the  time  constants  describing  the  decay  of  the  correlation  signal  in  the  solid  samples  (~20  µs   and  ~30  µs  for  Samples  A  and  B,  respectively),  confirming  the  above  conclusion  that  NV  relaxation   can  be  ignored  in  these  two  cases.  Sample  C,  on  the  other  hand,  deserves  some  special   consideration:  The  first  segment  of  the  correlation  signal  (governed  by  molecular  diffusion  and   meaning  that  no  corrections  are  required  in  our  numerical  estimates  of  the  bulk  diffusion  constant   decaying  on  a  time  scale  of  𝑇!"##(!!!"#)~15  µs)  is  clearly  insensitive  to  NV  spin  lattice  relaxation,   𝐷!,!"#$  and  the  adsorbed  layer  thickness  𝑧!"#.  This  is,  however,  not  the  case  for  the  longer-­‐lived   segment  of  the  correlation  signal,  which  decays  on  a  time  scale  𝑇!"##(!"#$)~250  µs  comparable  to  𝑇!!"   𝐷!,!"!  must  be  interpreted  as  a  lower  bound.     (Fig.  S10b).  Therefore,  our  estimate  of  the  rotational  diffusional  constant  in  the  adsorbed  layer     G. Comparison  with  bulk  NMR  data   Relevant  information  on  the  sample  dynamics  near  the  diamond  surface  can  be  gained  by  comparing   the  measured  correlation  signals  with  bulk  1H  NMR  data.  Fig.  S10  shows  the  NMR  proton  signal  for   Sample  A  (Merckoglas)  and  Sample  B  (PDMS)  upon  application  of  a  π/2-­‐pulse  at  600  MHz   (corresponding  to  a  14.1  T  magnetic  field).  In  both  cases  we  find  that  the  NV-­‐detected  correlation   signals  (Figs.  S1  and  S2)  are  shorter-­‐lived  than  their  inductively  detected  counterparts  (1H  NMR  FIDs   in  Figs.  S10a  and  S10c).  The  difference  is  starker  in  the  case  of  Sample  B,  where  the  nuclear  spin   coherence  time  is  seen  to  persist  beyond  0.5  ms.  We  interpret  these  differences  as  a  manifestation   of  the  singular  molecular  dynamics  near  the  diamond  surface,  where  motion  is  likely  more  restricted   than  in  the  bulk.     Indirect  support  for  this  idea  is  provided  by  the  results  of  Figs.  S10b  and  S10d  where  we  expose  the   importance  of  molecular  packing  in  the  dynamics  of  both  Sample  A  and  Sample  B:  In  the  first  case   21 Figure  S10.  High-­‐field  NMR  of  the  solid  samples.  (a)  1H  NMR  free  induction  decay  from  a  dried  1:1  solution   of  Merckoglas  in  toluene  (as  used  in  our  NV  experiments).  (b)  Same  as  in  (a)  but  for  dried  Merckoglas   (without  added  toluene).  (c)  1H  NMR  signal  from  bulk  PDMS;  the  elastomer/curing  agent  ratio  is  identical   (10:1)  to  that  used  in  our  NV  experiments.  (d)  Same  as  in  (c)  but  for  ‘rigid’  PDMS  (6:1  mixing  ratio).  All  NMR   experiments  were  carried  out  at  14.1  T  (600  MHz  proton  frequency)  under  ambient  conditions.       (Fig.  S10b),  we  let  a  sample  of  ‘as-­‐purchased’  Merckoglas  dry  to  form  a  solid  crust,  which  we  then   analyze  via  inductive  NMR.  This  system  differs  from  the  one  used  in  Figs.  S10a  only  in  its  preparation   protocol:  Unlike  the  practice  in  our  NV  experiments—where  Mercoglas  is  dissolved  in  toluene  to   form  a  1:1  solution  prior  to  gluing  the  diamond  crystal  to  the  sample  holder—no  solvent  was  added.   The  result  is  an  1H  FID  substantially  longer  than  that  in  Fig.  S10a,  the  increased  mobility  likely   originating  from  the  reduced  number  of  interstitial  molecules  after  solvent  evaporation.   Interestingly,  we  observe  the  converse  effect  in  Sample  B  where  we  shorten  the  FID  duration  by   bringing  the  ratio  between  PDMS  and  the  curing  agent  to  a  value  lower  than  that  used  for  our  NV   experiments.  The  latter  leads  to  a  more  rigid  form  of  the  resulting  polymer  and  thus  a  shorter-­‐lived   proton  FID.     For  completeness,  Fig.  S11  presents  1H  NMR  data  corresponding  to  Sample  C  (immersion  oil),  where   we  find  a  proton  spin  transverse  relaxation  time  of  order  25  ms,  limited  by  field  inhomogeneity  (an   22 Figure  S11.  1H  NMR  signal  from  Sample  C.  (a)  1H  free  induction  decay  from  a  sample  of  the  immersion   oil  used  in  our  NV  experiments.  The  proton  spin  transverse  relaxation  time  is  approximately  25  ms,   limited  by  field  inhomogeneity.  (b)  FID  from  protons  in  acetone.  All  experiments  were  carried  out  at  14.1   T  (600  MHz  proton  frequency)  under  ambient  conditions.     FID  from  acetone  is  included  as  a  reference).       H. Origin  of  the  signal   Recently  it  was  pointed  out  that  the  XY8-­‐N  noise  spectroscopy  is  prone  to  measure  spurious  effects   of  higher  Larmor  frequency  harmonics12.  Especially  the  4th  harmonic  of  the  13C  nuclei  intrinsic  in  the   diamond  lattice  is  spectrally  located  very  close  to  the  proton  Larmor  frequency,  and  can  be  easily   misinterpreted  as  the  signal  of  external  protons.    Our  protocol  is  immune  to  this  problem  because  it   directly  records  the  proton  spin  precession  during  𝜏,  thus  leading  to  resonance  spectra  at  the   specific  nuclear  Larmor  frequency13.  In  particular,  any  residual  effect  from  carbon  spins  during  the   XY8-­‐N  segments  of  the  correlation  protocol  would  lead  to  a  peak  at  the  carbon  spin  resonance   frequency,  which  can  be  easily  separated  from  proton-­‐induced  contributions.       I. References     1. 2. Ortiz-­‐Young,  D.,  Chiu,  H.-­‐C.,  Kim,  S.,  Voitchovsky,  K.  &  Riedo,  E.  The  interplay  between   apparent   viscosity   and   wettability   in   nanoconfined   water.   Nature   Communications   4,   2482  (2013).   Abragam,  A.  The  Principles  Of  Nuclear  Magnetism  (Clarendon  Press,  Oxford,  1961).   23 3. 4. 5. 6. 7. 8. 9. Bloembergen,   N.,   Purcell,   E.   &   Pound,   R.   V.   Relaxation   effects   in   nuclear   magnetic   resonance  absorption.  Physical  Review  73,  679  (1948).   Levitt,  M.  H.  Spin  Dynamics:  Basics  Of  Nuclear  Magnetic  Resonance  (John  Wiley  &  Sons,   Hoboken,  2008).   Hall,  L.  T.  Principles  And  Applications  Of  Quantum  Decoherence  In  Biological,  Chemical,   And  Condensed  Matter  Systems  (The  University  of  Melbourne,  Melbourne,  2013).   Steinert,   S.   et.   al.,   Magnetic   spin   imaging   under   ambient   conditions   with   sub-­‐cellular   resolution.  Nature  Communitcation  4,  1607  (2013).   Ziem,   F.,   Götz,   N.,   Zappe,   A.,   Steinert,   S.   &   Wrachtrup,   J.   Highly   sensitive   detection   of   physiological  spins  in  a  microfluidic  device,  Nano  letters  13,  4093-­‐4098  (2013).   Rugar,   D.   et.   al.,   Proton   magnetic   resonance   imaging   using   a   nitrogen-­‐vacancy   spin   sensor.  Nature  Nanotechnology  (2014).   Häberle,  T.,  Schmid-­‐Lorch,  D.,  Reinhard,  F.  &  Wrachtrup,  J.  Nanoscale  nuclear  magnetic   imaging  with  chemical  contrast.  Nature  Nanotechnology  (2014).   10. Kimmich,   R.,   Unrath,   W.,   Schnur,   G.   &   Rommel,   E.   NMR   measurement   of   small   self-­‐ diffusion  coefficients  in  the  fringe  field  of  superconducting  magnets.  Journal  of  Magnetic   Resonance  91,  136-­‐140  (1991).   11. Litvinov,   V.   M.   &   Spiess,   H.   W.   2H   NMR   study   of   molecular   motions   in   polydimethylsiloxane  and  its  mixtures  with  aerosils.  Die  Makromolekulare  Chemie  192,   3005-­‐3019  (1991).   Loretz,   M.   et.   al.,   Spurious   harmonic   response   of   multipulse   quantum   sensing   sequences.  arXiv:1412.5768  (2014).   Laraoui,  A.  et.  al.,  High-­‐resolution  correlation  spectroscopy  of  13C  spins  near  a  nitrogen-­‐ vacancy  centre  in  diamond.  Nature  Communications  4,  1651  (2013).   24 12. 13.   

1905.13088

2

1905

Intrinsically Undamped Plasmon Modes in Narrow Electron Bands

Surface plasmons in 2-dimensional electron systems with narrow Bloch bands feature an interesting regime in which Landau damping (dissipation via electron-hole pair excitation) is completely quenched. This surprising behavior is made possible by strong coupling in narrow-band systems characterized by large values of the "fine structure" constant $\alpha=e^2/\hbar \kappa v_{\rm F}$. Dissipation quenching occurs when dispersing plasmon modes rise above the particle-hole continuum, extending into the forbidden energy gap that is free from particle-hole excitations. The effect is predicted to be prominent in moir\'e graphene, where at magic twist-angle values, flat bands feature $\alpha\gg1$. The extinction of Landau damping enhances spatial optical coherence. Speckle-like interference, arising in the presence of disorder scattering, can serve as a telltale signature of undamped plasmons directly accessible in near-field imaging experiments.

cond-mat.mes-hall

cond-mat

Intrinsically Undamped Plasmon Modes in Narrow Electron Bands Massachusetts Institute of Technology, 77 Massachusetts Ave, Cambridge, MA02139, USA Cyprian Lewandowski, Leonid Levitov Surface plasmons in 2-dimensional electron systems with narrow Bloch bands feature an interest- ing regime in which Landau damping (dissipation via electron-hole pair excitation) is completely quenched. This surprising behavior is made possible by strong coupling in narrow-band systems characterized by large values of the "fine structure" constant α = e2/κvF. Dissipation quench- ing occurs when dispersing plasmon modes rise above the particle-hole continuum, extending into the forbidden energy gap that is free from particle-hole excitations. The effect is predicted to be prominent in moir´e graphene, where at magic twist-angle values, flat bands feature α (cid:29) 1. The ex- tinction of Landau damping enhances spatial optical coherence. Speckle-like interference, arising in the presence of disorder scattering, can serve as a telltale signature of undamped plasmons directly accessible in near-field imaging experiments. Landau damping, a process by which collective mode decays into electron-hole pairs, is often taken to be an integral attribute of graphene plasmon excitations [1 -- 5]. Here, we predict extinction of this dissipation mechanism in materials with narrow electron bands, such as twisted bilayer graphene (TBG) [6 -- 10]. Intrinsically undamped plasmons in narrow-band materials arise due to large fine structure parameter values α = e2/κvF: strong interac- tions push plasmon dispersion into the energy gap above the particle-hole (p-h) continuum as illustrated in Fig. 1. In this region, plasmons become decoupled from p-h pair excitations. Dissipation quenching, which is a surpris- ing manifestation of strong coupling physics, is a robust effect that persists up to room temperature and is in- sensitive to disorder (Figs. 1 and 2). Collective charge modes, which are damping free, are of keen interest for quantum information science as a vehicle to realize dis- sipationless photon-matter coupling, high-Q resonators, single-photon phase shifters and other missing compo- nents for photon-based quantum information processing toolbox [15]. Although extinction of Landau damping is a general effect present in all narrow electron bands, our analysis will focus on TBG flat bands, a system of high current interest [16 -- 20], in which undamped plasmons can be directly probed. Fig. 1 depicts plasmon mode for a narrow-band model that mimics the key features of the TBG band. Mode dis- persion (red line) and its damping are of a conventional form at energies less than the bandwidth, ω (cid:46) W . At lowest energies, plasmon mode is positioned outside the p-h continuum, as expected; this suppresses the T = 0 Landau damping, but does not protect the mode from decaying into p-h excitations through disorder scatter- ing or from the conventional T > 0 Landau damping [1, 2, 21 -- 25]. At higher energies, ω ∼ 2EF (marked by arrows in Fig. 1), the mode plunges into p-h contin- uum and is Landau-damped at 2EF (cid:46) ω (cid:46) 2W , even at T = 0. However, an interesting change occurs after the mode rises above the p-h continuum. In the forbid- den gap region, ω > 2W , it becomes damping-free, since at these energies there are no free p-h pairs into which (a) Electron loss function Im (−1/ε(ω, q)) for a FIG. 1. narrow-band toy model (the hexagonal tight-binding model) [Eq. (10)]. Parameter values are chosen to mimic TBG bands (bandwidth W = 3.75 meV, lattice periodicity LM = 13.4 nm, Fermi energy in the conduction band at EF ≈ 1.81 meV); log scale is used to clarify the relation between different fea- tures. Arrows mark the interband p-h continuum edges. Plas- mon dispersion (red line) is fitted with ωp(q) = (cid:112)βqq [Eq. (1)] (dashed line). The difference between Landau-damped (b) and undamped behavior (c) is illustrated by line cuts of plasmon resonances at the locations marked in (a), taken at temperatures T /EF = 0, 0.075, 0.1, 0.2, 0.3, 0.4. Resonances broaden with T in (b) and are T independent in (c) (the residual resonance width models extrinsic damping due to phonons and disorder [11 -- 14]). Resonances at the 3 lowest T values in (c) are slightly offset for clarity. (d) Speckle pat- tern in scanning near-field microscopy signal [4, 5] S(r) [Eq. (3)] due to undamped plasmons; optical coherence is manifest in Fourier spectrum Sk2 (inset). Results shown are for plas- mon momentum q0 = qM /2 ≈ 0.14 nm−1, where qM is the distance between points M and Γ, and disorder is modeled as 40 randomly placed point defects. 9 1 0 2 t c O 1 ] l l a h - s e m . t a m - d n o c [ 2 v 8 8 0 3 1 . 5 0 9 1 : v i X r a Intrinsicallyundamped plasmons0.50.550.604080120160T growsa)b)c)WParticle-holecontinuum00.050.10246810T growsWavenumberWavenumber -10-505-101d)200100-200-1000y (nm)2001000-100-2002EF2Wx (nm)2q0plasmon could decay. This behavior is manifest in the T dependence of the resonances, which are washed out with increasing temperature at ω (cid:46) W but remain sharp at ω > W even at T ∼ EF (Fig. 1 b and c). As we will see, mode dispersion has a square root form characteristic of 2-dimensional (2D) plasmons [26, 27], ωp(q) =(cid:112)βqq, (1) with a weak q dependence in βq [Eq. (14)]. This expression, however, is valid not just at low energies, 0 < ω (cid:46) W , but also at higher energies, ω (cid:29) W , where the mode is undamped. While the dispersion in Eq. (1) is of the conventional 2D plasmon form, we emphasize that here it takes on a different role, as it describes the plasmon mode at frequencies much higher than the car- rier bandwidth, extending to ωp ∼ √ αW (cid:29) W, α ∼ 20 − 30, (2) where the high-α values correspond to flat bands in magic-angle moir´e graphene. Also, unlike the conven- tional plasmons, the dispersion in Eq. (1) is not limited to longest wavelengths. Indeed, as illustrated Fig. 1a, it extends to fairly high wavenumbers on the order of the mini Brillouin zone size. The wavelengths of these plasmons are only 2 to 3 times greater than the moir´e superlattice period. Such short wavelengths are of considerable interest for plas- monics and are within resolution of the state-of-the-art scanning near-field microscopy techniques [4, 5] (cur- rently as good as 10 nm [28, 29]). In addition to mea- suring plasmon dispersion, these techniques can be used to directly visualize the qualitative change in the damp- ing character and strength. Enhanced optical coherence will manifest itself in striking speckle-like interference as illustrated in Figs. 1d and 2. Indeed, because of the absence of Landau damping at the energies of interest, ω > W , and also because these energies are smaller than carbon optical phonon energies, the dominant dissipation mechanism is likely to be elastic scattering by disorder. At low energies, where plasmon mode coexists with p-h continuum, disorder scattering merely assists Landau damping, allowing plasmons to de- cay into p-h pairs by passing some of their momentum to the lattice. However, at the energies above p-h contin- uum, ω > W , since the decay into pairs is quenched, dis- order will lead to predominantly elastic scattering among plasmon excitations. Such scattering preserves optical coherence and is expected to produce speckle patterns in spatial near-field images as illustrated in Fig. 1d. To model this behavior we consider the signal S(r), excited by the scanning tip and measured at the same location. Monochromatic plasmon excitation at energy E is scattered by impurities or defects and on returning to the tip, produces signal S(r) = J0 d2r(cid:48)GE(r − r(cid:48))η(r(cid:48))GE(r(cid:48) − r), (3) (cid:90) 2 FIG. 2. (a-d) Speckle patterns arising due to optical coher- ence of undamped plasmons in scanning near-field microscopy signal S(r) [Eq. (3)] at various ratios of the incoherent to co- herent damping δ/q0. Insets show the corresponding square of the speckle pattern's Fourier transform amplitude Sk2. In all panels, for clarity of comparison, we set the plasmon momentum as in Fig. 1d (q0 = qM /2 ≈ 0.14 nm−1) and vary only the ratio δ/q0. The disorder is taken as 40 randomly placed Dirac delta functions. where η(r) is the disorder potential, J0 is excitation am- plitude, and GE(r) is the Green's function of the plasmon excitation (see supplemental information). The spatial signal (Fig. 1d) exhibits a characteristic speckle pat- tern familiar from laser physics. In graphene plasmon- ics, speckle-like interference provides a direct manifesta- tion of optical coherence enhancement in the absence of Landau damping. Accordingly, the Fourier transform of Sk2 that features a ring-like structure; the ring radius is k = 2q0, where q0 is the plasmon excitation wavenum- ber (Fig. 1d inset). Simple calculation, described in supplemental information, predicts power spectrum that sharply peaks at the ring: the image, Sk =(cid:82) d2rS(r)e−ikr, yields power spectrum Sk2 ∼ ηk2 k2 − 4(q0 − iδ)2 , (4) where δ is a parameter characterizing extrinsic damping due to phonon scattering and other inelastic processes. In the fully coherent regime (δ = 0) the quantity Sk2 exhibits a power law singularity at the ring, k = 2q0. As the amount of incoherent scattering increases, the peak is gradually washed out. This behavior is illustrated in Fig. 2. We note that recent work [19] analyzed interband plas- mon excitations in TBG, which are dominated by polar- -101a)b)d)c)2q02q02q02q0y (nm)y (nm)x (nm)x (nm)ization of the bands above the flat band and are distinct from the flat-band plasmons analyzed here. Recent ex- periment [20] reported observation of plasmons in TBG; however, their appeal for constructing intrinsically pro- tected collective modes remained unnoticed in graphene literature. Also, plasmons in narrow bands were ana- lyzed in the context of high-Tc superconductivity [30], finding that plasmon mode can rise above the flat band. However, in cuprates, unlike moir´e graphene, the narrow band is not separated from higher bands by a forbidden energy gap, and thus, the mode studied in ref. [30] will plunge into a higher band before acquiring an undamped character. Next, we present analysis of the hexagonal-lattice toy model that mimics the key features of Landau- damped and intrinsically undamped modes in TBG. The hexagonal-lattice tight-binding model possesses the same symmetry and the same number of subbands as the flat band in TBG. We match the energy and length scales by choosing the width of a single band W and the hexag- onal lattice period LM identical to the parameters in TBG: W = 3.75 meV and LM = a/2 sin(θ/2) is the moir´e superlattice periodicity. For the magic angle value θ = 1.05◦, using carbon spacing a = 0.246 nm, this gives LM = 13.4 nm. To ensure that a unit cell of the toy model can accomodate 4 electrons just as the moir´e cell does in TBG, we make the toy model 4-fold degenerate. Comparison with plasmons for the actual TBG model, presented below, will help us to identify the features that are general as well as those which are a specific property of TBG. Our nearest-neighbor tight-binding Hamiltonian is Htoy = , hk = W 3 eik·ej , (5) (cid:19) (cid:18) 0 hk h∗ k 0 (cid:88) ej (cid:18)seiϕk (cid:19) 1 √ with the hopping matrix element W/3 to nearest neigh- 3, bors at positions ej = (cos(2πj/3), sin(2πj/3))LM / √ j = 0, 1, 2. Here, W is the bandwidth measured from Dirac point, and the nearest neighbor distance LM / 3 is chosen such that the lattice period of the hexagonal toy model matches the moir´e superlattice period. Corre- sponding energies Es,k and eigenstates Ψs,k are then Es,k = shk, Ψs,k = 1√ 2 , (6) where ϕk = arg hk and the band index s = ± labels the conduction and valence band. Plasmons can be obtained from the nodes of the com- plex dielectric function, describing the dynamical re- sponse of a material to an outside electric perturbation: ε(ω, q) = 1 − VqΠ(ω, q). (7) Here, Vq = 2πe2/κq is the Coulomb interaction in a medium with a background dielectric constant κ, and 3 Π(ω, q) is the electron polarization function. The relation in Eq. (7) is exact as long as the polarization function is defined as an exact microscopic density-density pair cor- relator given by a sum of all irreducible bubble diagrams. As such, this relation can yield useful information about plasmon dispersion, even when electron interactions are strong. Similar to the conventional analysis of plasmons in 2D systems, here a simplification occurs in the small-q limit, regardless of whether the random-phase approximation (RPA) is used to evaluate Π(ω, q). Indeed, since the Coulomb potential diverges at small q, zeros of ε(ω, q) are found when the polarization function is small. How- ever, at small q, this quantity vanishes as λq2/ω2, a behavior that is a consequence of the general symme- try requirements (namely, gauge invariance demanding that spatially uniform external potential does not per- turb density) [31]. This immediately yields a q1/2 scaling for plasmon frequency at small-enough q. Below, we use the RPA approach to estimate the pref- actor λ and to demonstrate that the mode ω ∼ q1/2 ex- tends far above the TBG p-h continuum. To compare with other systems, we recall the familiar "classical ac- celeration" behavior found for particles with parabolic dispersion: Π(ω, q) = nq2/mω2, where n is the charge density and m is the electron band mass [31]. For a more general band dispersion, the ratio n/m is replaced by the band Fermi energy, λ ∼ EF /2 [1 -- 3]. Interactions have no impact on the behavior of Π(ω, q) for the parabolic band case; however, for nonparabolic bands, the band mass m must change to an effective value m∗ described by Landau Fermi-liquid renormalization [32]. In our case, the scaling relation Π(ω, q) ≈ λq2/ω2 features different values of λ for low and high energies, ω (cid:46) EF and ω > 2W . To see this, we start with the RPA expression for polarization function (fs,k+q − fs(cid:48),k)F ss(cid:48) Es,k+q − Es(cid:48),k − ω − i0 k+q,k . (8) (cid:88) Here, summation (cid:80) Π(ω, q) = 4 k,s,s(cid:48) k denotes integration over the Bril- louin zone (BZ), the indices s,s(cid:48) run over the electron bands and the factor of 4 in front of the summation ac- counts for the 4-fold degeneracy of the toy model. Here, fs,k is the equilibrium distribution 1/(eβ(Es,k−EF) + 1), and F ss(cid:48) k+q,k describes band coherence factors. For our toy model, k+q,k = (cid:104)Ψs,k+qΨs(cid:48),k(cid:105)2 = F ss(cid:48) 1 + ss(cid:48) cos(ϕk+q − ϕk) 2 , (9) where Ψs,k are pseudospinors given in Eq. (6). As we now show, an analytic expression for plasmon dispersion can be obtained, describing both the Landau- damped and the undamped cases in a unified way. We (cid:88) first rewrite Eq. (8) by performing a standard replace- ment k + q → −k in the term containing fs,k+q followed by −k − q,−k → k + q, k justified by the k → −k time- reversal symmetry. This gives k,k+q(Es(cid:48),k − Es,k+q) F ss(cid:48) (Es,k+q − Es(cid:48),k)2 − (ω + i0+)2 . (10) The behavior of this expression at small q, which will be of interest for us, can be found in a closed form. In the small-q limit the coherence factors behave as Π(ω, q) = 8 k,s,s(cid:48) fs(cid:48),k k+q,k ≈ 1, F s=−s(cid:48) F s=s(cid:48) k+q,k ≈ 1 4 (q · ∇kϕk)2 (11) The values O(1) for intraband transitions and O(q2) for interband transitions might suggest that the polarization function is dominated by the intraband transitions. How- ever, as we now show, the interband and intraband con- tributions are of the same order of magnitude. s = s(cid:48), can be rewritten by noting that, upon integra- (cid:80) tion over k, only the even-k part of series ex- pansion Es,k+q − Es,k survives, giving Π1(ω, q) ≈ k,s fs,k (Es,k+q + Es,k−q − 2Es,k). Expanding in intraband contributions, Indeed, the 4 ω2 small q, we have Π1(ω, q) ≈ 4 ω2 fs,k(q · ∇k)2Es,k (12) (cid:88) k,s for parabolic band Ek = k2/2m As a sanity check, this yields the familiar "classical acceleration" result Π(ω, q) = nq2 mω2 . The interband contributions, s = −s(cid:48), can be simpli- fied by noting that Es,k+q ≈ −Es(cid:48),k, giving Π2(ω, q) ≈ 4 fs,k Es,k (q · ∇kϕk)2 s,k − (ω + i0)2 . 4E2 (13) (cid:88) k,s As a sanity check, at T = 0 the imaginary part of Π2, describing interband transitions, is nonzero only for 2EF < ω < 2W , as expected. The real part of Π2 is nega- tive at small ω and positive at large ω because the valence band contribution dominates over that of the conduction band. Plasmon dispersion ωp is given by the solution of the equation ε(ω, q) = 0 with Π = Π1 + Π2. Comparing the ω dependence of Π1 and Π2, we see that at small frequen- cies, ω < 2EF, the intraband contribution Π1 dominates. This gives the dispersion in Eq. (1) with βq = β0 + β1q + O(q2) (14) where the leading term β0 = 4αvFEF/ originates from Π1 (see supplemental information), and the subleading q-dependent contribution is due to Π2. Negative sign of Π2 translates into β1 < 0, softening the dispersion at 4 low frequencies. This behavior, which holds the limit ω < 2EF, agrees with refs. [1, 2, 27]. In the same manner, we can obtain the dispersion at high frequencies, ω > 2W (the intrinsically undamped regime). The analysis is again simplified by noting that, since α = e2/κvF (cid:29) 1, the relevant values of q are small compared to the Brillouin zone size, and thus, the small-q limit considered above is sufficient to describe this behavior. Taking both the intraband and interband con- tributions in the asymptotic form Π1 = λ1q2/ω2, Π2 = λ2q2/ω2 where λ1 ≈ 2EF/2π, λ2 ≈ 2(W − EF)/2π (see supplemental (1) with β = 2πe2 κ (λ1 + λ2). The first term is identical to β0 found at low frequencies, the second term is of a positive sign, λ2 > 0, describing stiffening of the plasmon dispersion due to interband transitions. information), yields Eq. In the undamped regime, plasmon frequency peaks at q values on the order of Brillouin zone scale. The peak value of ωp, given in Eq. (2), can be found by estimating the energy differences Es,k+q − Es(cid:48),k in Eq. (10) as W and noting that the coherence band factor for large q is in general non-vanishing and of order 1. This gives, for the practically interesting case of EF ∼ W , βq = 2(cid:112)αvFW q/ provided that vFq saturates the result ωp ∼ √ αW , which agrees with the dispersion mation) indicate that ωp =(cid:112)βqq relation from Eq. (1) ωp = at W . Indeed, the estimated values of β0,β compared with the fitted curve in Fig. 1a (see supplemental infor- √ is a good approximation for the plasmon dispersion at both small and large q. The dielectric function of the 2-band toy model faith- fully reproduces all of the qualitative features expected for the TBG bandstructure. However, we find that, de- spite matching the bandwidth W and lattice period to those of TBG, the resulting plasmon dispersion extends to much higher energies then those that will be found below for the actual TBG bandstructure. This is simply because the 2-band model does not account for the effects of interband polarization of higher electron bands, which renormalize the dielectric constant down and soften the plasmon dispersion. We account for this in the toy model case by rescaling the effective fine structure constant such that the resulting plasmon dispersion is comparable in magnitude with the TBG result. Specifically, in Fig. 1a, we use an effective background dielectric constant κ = 12.12, which is 4 times larger than the dielectric con- stant κ = 3.03 corresponding to an air/TBG/hexagonal boron nitride (hBN) heterostructure. Next, we turn to the analysis of plasmons in TBG flat bands at an experimentally relevant magic angle value θ = 1.05◦ [16 -- 18]. To accurately describe the TBG band structure and eigenstates, we use the effective continuum Hamiltonian HTBG introduced in ref. [33]. The full dis- cussion of the band structure details can be found in the supplemental material; here, we only discuss 2 relevant 5 tum k and band/valley/spin composite label n, which diagonalize the continuum Hamiltonian (see supplemen- tal materials). The integral in Eq. (15) is carried over the moir´e unit cell Ω. After the polarization function is evaluated, we can determine the dielectric function and identify TBG's col- lective modes from poles of 1/ε(ω, q) as above. An exam- ple of a TBG's dielectric function at approximately half- filling of the electron band, EF = 0.289 meV, is shown in Fig. 3; fixed q line cuts and zeros of ε(ω, q) are illustrated in the supplemental materials. In discussing the figure, it is helpful to contrast it with the calculation for the hexagonal-lattice toy model shown in Fig. 1a. We again see a well-defined intrinsically undamped plasmon mode ωp (red in Fig. 1a) positioned above the p-h continuum; the mode resides inside the band gap 2W < ωp < W +∆, which peaks at ωp ≈ 8.5 meV before decreasing and be- coming almost flat ωp ≈ 6.5 meV at large momenta. In agreement with the analytic considerations above, we see the interband continuum extending from 2EF to 2W , but since EF = 0.289 meV is extremely small, it makes the conventional (Landau-damped) part of plasmon dis- persion ω < 2EF invisible on the figure. There are several unique aspects of the TBG plas- mon dispersion compared with the behavior of generic narrow-band plasmons discussed above. To analyze the dispersion at ωp > 2W , we proceed just as in the toy model case, rewriting the TBG polarization function in a slightly different form of Eq. (10), where the indices n, m and the band coherence factor are modified as described above. To proceed further analytically, we need to analyze Eq. (10) in the long-wavelength limit. However, unlike the 2-band toy model, where the only characteristic energy scale was the bandwidth W , the TBG band structure fea- tures an additional energy scale, namely, the gap between the flat bands and the rest of the energy spectrum. This impacts the small-q series expansion of the polarization function, as now the energy difference En − Em between the occupied and unoccupied states can be larger than ω. To account for such contributions in the series ex- pansion, we split the summation over TBG bands into 2 parts, depending on whether ω or the energy difference En − Em is the largest energy scale in the denominator of Eq. (10). This yields an approximate expression for the dielectric function ε(ω, q) ≈ 1 + A(q) − B(q) ω2 , where we defined 2 auxiliary functions: A(q) = fm,k F nm k+q,k En,k+q − Em,k (16) (17) (cid:88)(cid:48) k,n,m 8πe2 κq (cid:88)(cid:48)(cid:48) k,n,m 8πe2 κq FIG. 3. Electron loss function Im (−1/ε(ω, q)) for TBG bandstructure. The Fermi energy value EF = 0.289 meV corresponds to electron band half-filling, and the average background dielectric constant is κ = 3.03 (typical of an air/TBG/hBN heterostructure). Log scale is used to clar- ify the relation between different features. Arrows mark the approximate interband p-h continuum edges obtained for the effective bandwidth W ≈ 2 meV (see text). Plasmon disper- sion (red line) at small q is fitted with ωp(q) = (cid:112)βqq [Eq. (1)] (dashed line), demonstrating a significant deviation from the typical 2D plasmon dispersion at large q. In the calcula- tion, we used both flat bands and the next conduction/valence nonflat bands and verified that higher bands do not alter the quantitative and qualitative behavior. energy scales: flat-band bandwidth W and the gap ∆ between the flat bands and the rest of the band struc- ture. With regard to W value, we note that, technically, the bandwidth of the flat-bands, as predicted by the con- tinuum mode HTBG, is on the order of W ≈ 3.75 meV. However, the bandwidth scale relevant for the interband and intraband excitations is actually closer to W ≈ 2 meV, because most of the states in the band lie below 2 meV. In addition, since the states with energies outside −2 meV < E < 2 meV are small k, their contribution to polarization function [Eqs. (12) and (13)), evaluated at small q, is small. We also note that, while the bandgap as predicted by the continuum model is ∆ ≈ 11.75 meV, the actual gap is still a subject of debate [34]. The definition of the polarization function for the TBG continuum model is essentially identical to that of the tight binding toy model [Eq. (8)]. Now, however, we must account explicitly for the valley and spin degrees of freedom, for a larger number of electron bands, and for different coherence factors. Accordingly, we promote the band indices s,s(cid:48) in Eq. (8) to composite labels n,m, which label all electron bands, spins σ and valleys ξ; this makes the additional factor of 4 in front of Eq. (8) re- dundant. The toy model coherence factors are replaced by the TBG coherence band factors F nm k+q,k, which are given by (cid:12)(cid:12)(cid:12)(cid:12)(cid:90) Ω (cid:12)(cid:12)(cid:12)(cid:12)2 F nm k+q,k = d2rΨ † n,k+q(r)eiq·rΨm,k(r) , (15) and B(q) = where Ψn,k(r) are the Bloch wavefunctions for momen- fm,kF nm k+q,k(En,k+q − Em,k). (18) -8-6-4-20242EFParticle-holecontinuumKMK'ΓIntrinsically undamped plasmonsHere the band summations (cid:80)(cid:48) and (cid:80)(cid:48)(cid:48) run over bands such that ω2 > (En,k+q − Em,k)2 and ω2 < (En,k+q − Em,k)2, respectively: for example, at large momenta, as seen in Fig. 3, the plasmon mode lies in the gap be- tween the flat and non-flat bands, and hence, the B(q) summation extends only over the flat bands, whereas the summation in A(q) includes all of the remaining combi- nations of band indices. This allows us to write a closed form expression for the plasmon dispersion as p ≈ B(q) ω2 1 + A(q) , (19) which must hold for both small and large q. We consider these 2 limits separately. At small q, the matrix element of the Bloch wavefunc- tions, just as in the toy model case, favors the overlap between states from the same band. At the same time, there are fewer states in the A(q) satisfying the condi- tion ω2 > (En,k+q−En,k)2, and hence, A(q) vanishes for small q. This amounts to the plasmon dispersion ωp from p ≈ B(q), and by comparison with Eq. (19) reducing to ω2 Eq. (12), we similarly expect a conventional 2D plasmon dispersion ωp =(cid:112)βqq with βq given by the series from Eq. (14). As we see in Fig. 3, the ωp =(cid:112)βqq dispersion is a valid description only at very small q compared to the Fig. 1a, which can be traced back to higher bands softening the plasmon dispersion through the A(q) term in Eq. (19). To determine how high the plasmon mode rises above the p-h continuum, we consider large q values compara- ble to the reciprocal lattice vector. The arguments sim- ilar to those in the toy model show that, since α (cid:29) 1, we have A(q) (cid:29) 1. The dependence on the e2/κq ra- ωp ≈(cid:112)B(q)/A(q) ∼ √ tio, therefore, cancels between the A(q) and B(q) func- tions, resulting in the value of the plasmon dispersion W ∆ ≈ 6.6 meV being dictated only by the continuum model's band structure parame- ters. This lack of explicit dependence on α suggests that, after the doping is such that α (cid:29) 1, the large-q value of ωp ≈ √ W ∆ becomes insensitive to doping (and hence, the toy model, where ωp ∼ √ Fermi velocity). This behavior is different from that in αW at large q. The rela- tively more weak dependence on α in the TBG case is due to interband polarization involving higher bands, which significantly alters the effective dielectric constant. The weak q dependence at large q is in agreement with the properties of interband plasmons described in ref. [19]. We also note that, although plasmons above the p- h continuum are kinematically protected from p-h exci- tation, which makes them undamped at the RPA level, there exist relaxation pathways through higher-order pair production in which several electron-hole pairs are emit- ted with total energy exceeding W , as well as phonon- assisted processes. For conventional plasmons these pro- cesses were analyzed in ref. [35]. The role of these effects 6 for plasmon lifetimes in TBG will be a subject of future work. Before closing, we note that suppressing damping has always been central to the quest for tightly-confined low- loss surface plasmon excitations. An early approach uti- lized surface electro-magnetic modes traveling at the edge of an air/metal boundary [36], in which dissipation is low because most of the mode field resides outside the metal; however, the field confinement scale, set by optical wave- length, was fairly large. Next came surface plasmons propagating in high-mobility 2D electron gases in semi- conductor quantum wells and monolayer graphene [14], which can provide deep-subwavelength confinement [3]. However, plasmons in these systems are prone to a variety of dissipation mechanisms, with Landau damping usu- ally regarded as the one that sets the fundamental limit on possible plasmon wavelengths and corresponding life- times. The possibility to overcome this fundamental lim- itation in narrow-band systems, such as moir´e graphene, creates a unique opportunity for graphene plasmonics. Damping-free plasmons can enable novel interference phenomena, dissipationless photon-matter coupling, and other interesting behaviors. It is also widely expected that low-dissipation plasmons can lead to unique applica- tions for photon-based quantum information processing [15]. Furthermore, reduced damping has more immedi- ate consequences, as it translates into enhanced optical coherence that can be directly probed by scanning near- field microscopy, as discussed above, providing a clear signature of the undamped collective modes. We thank Ali Fahimniya for useful discussions. This work was supported, in part, by the Science and Tech- nology Center for Integrated Quantum Materials, NSF Grant No. DMR-1231319; and Army Research Office Grant W911NF-18-1-0116. [1] B. Wunsch, T. Stauber, F. Sols, F. Guinea, Dynamical polarization of graphene at finite doping. New J. Phys. 8, 318 (2006). [2] E. H. Hwang, S. Das Sarma, Dielectric function, screen- ing, and plasmons in two-dimensional graphene. Phys. Rev. B 75, 205418 (2007). [3] H. Buljan, M. Jablan, M. Soljacic, Damping of plasmons in graphene. Nat. Photon. 7, 346-399 (2013). [4] J. Chen, M. Badioli, P. Alonso-Gonzalez, S. Thon- grattanasiri, F. Huth, J. Osmond, M. Spasenovic, A. Centeno, A. Pesquera, P. Godignon, A. Zurutuza, N. Camara, F J. Garcia De Abajo, R. Hillenbrand, F. H. L. Koppens, Optical Nano-Imaging of Gate-Tunable Graphene Plasmons, Nature 487, 77 (2012). [5] Z. Fei, A. S. Rodin, G. O. Andreev, W. Bao, A. S. McLeod, M. Wagner, L. M. Zhang, Z. Zhao, M. Thiemens, G. Dominguez, M. M. Fogler, A. H. Cas- tro Neto, C. N. Lau, F. Keilmann, D. N. Basov, Gate- Tuning of Graphene Plasmons Revealed by Infrared 7 and shear viscosities of the two-dimensional electron liq- uid in a doped graphene sheet, Phys.Rev. B 88, 195405 (2013). [24] M. Polini, R. Asgari, G. Borghi, Y. Barlas, T. Pereg- Barnea, and A. H. MacDonald, Plasmons and the spec- tral function of graphene, Phys. Rev. B 77, 081411(2008). [25] H. Yan, T. Low, W. Zhu, Y. Wu, M. Freitag, X. Li,F. Guinea, P. Avouris, and F. Xia, Damping pathways of mid-infrared plasmons in graphene nanostructures, Na- ture Photonics 7, 394 EP (2013). [26] F. Stern, Polarizability of a Two-Dimensional Electron Gas, Phys. Rev. Lett. 18, 546 (1967). [27] A. Principi, M. Polini, and G. Vignale, Linear response of doped graphene sheets to vector potentials, Phys. Rev. B 80 ,075418 (2009). [28] M. Cohen, R. Shavit, and Z. Zalevsky, Observing Optical Plasmons on a Single Nanometer Scale, Scientific Reports 4, 4096 EP (2014). [29] H. Duan, A. I. Fernandez-Dominguez, M. Bosman, S. A. Maier, and J. K. W. Yang, Nanoplasmonics: Classical down to the Nanometer Scale, Nano Letters 12, 1683 (2012). [30] E. A. Pashitskii, Plasmon mechanism of high- temperature superconductivity in cuprate metal-oxide compounds, JETP 76, 425 (1993). [31] G. Mahan, Many-Particle Physics. (Springer US, Boston, 2000) [32] L. S. Levitov, A. V. Shtyk, M. V. Feigelman, Electron- electron interactions and plasmon dispersion in graphene. Phys. Rev. B 88, 235403 (2013). [33] M. Koshino, N. F. Q. Yuan, T. Koretsune, M. Ochi, K. Kuroki, and L. Fu, Maximally Localized Wannier Or- bitals and the Extended Hubbard Model for Twisted Bi- layer Graphene, Phys. Rev. X 8, 031087 (2018). [34] S. L. Tomarken, Y. Cao, A. Demir, K. Watanabe, T. Taniguchi, P. Jarillo-Herrero, and R. C. Ashoori, Elec- tronic compressibility of magic angle graphene superlat- tices, Phys. Rev. Lett. 123, 046601 (2019). [35] E. G. Mishchenko, M. Y. Reizer, L. I. Glazman, Plas- mon attenuation and optical conductivity of a two- dimensional electron gas. Phys. Rev. B 69, 195302 (2004). [36] H. Raether, Surface plasmons on smooth surfaces. In: Surface Plasmons on Smooth and Rough Surfaces and on Gratings. Springer Tracts in Modern Physics, vol. 111. (Springer, Berlin, Heidelberg,1988) [37] S. L. Adler, Quantum Theory of the Dielectric Constant in Real Solids, Phys. Rev. 126, 413 (1962). [38] N. Wiser, Dielectric Constant with Local Field Effects Included, Phys. Rev. 129, 62 (1963). Nano-Imaging, Nature 487, 82 (2012). [6] E. J. Mele, Commensuration and interlayer coherence in twisted bilayer graphene, Phys. Rev. B 81, 161405(R) (2010). [7] P. San-Jose, J. Gonz´alez, F. Guinea, Non-Abelian gauge potentials in graphene bilayers. Phys. Rev. Lett. 108, 216802 (2012). [8] R. Bistritzer, A. H. MacDonald, Moir´e bands in twisted double-layer graphene, PNAS 108 (30) 12233-12237 (2011). [9] J. M. B. Lopes dos Santos, N. M. R. Peres, and A. H. Castro Neto, Graphene Bilayer with a Twist: Electronic Structure, Phys. Rev. Lett. 99, 256802 (2007). [10] S. Fang, and E. Kaxiras, Electronic structure theory of weakly interacting bilayers, Phys. Rev. B 93, 235153, (2016). [11] T. Langer, J. Baringhaus, H. Pfn´ur, H. W. Schumacher, and C. Tegenkamp, Plasmon damping below the Landau regime: the role of defects in epitaxial graphene, New Journal of Physics 12 , 033017 (2010). [12] F. J. Garcia de Abajo, Graphene Plasmonics: Challenges and Opportunities, ACS Photonics 1, 135 (2014). [13] G. X. Ni, A. S. McLeod, Z. Sun, L. Wang, L. Xiong, K. W. Post, S. S. Sunku, B. Y. Jiang, J. Hone, C. R. Dean, M. M. Fogler, and D. N. Basov, Fundamental limits to graphene plasmonics, Nature 557, 530 (2018). [14] A. Woessner, M. B. Lundeberg, Y. Gao, A. Prin- cipi, P. Alonso-Gonzalez, M. Carrega, K. Watanabe, T. Taniguchi, G. Vignale, M. Polini, J. Hone, R. Hillen- brand, and F. H. L. Koppens, Highly confined low- loss plasmons in graphene-boron nitride heterostructures, Nature Materials 14, 421 EP (2014). [15] M. Gullans, D. E. Chang, F. H. L. Koppens, F. J. Gar- cia de Abajo, and M. D. Lukin, Single-Photon Nonlinear Optics with Graphene Plasmons, Phys. Rev. Lett. 111, 247401 (2013) [16] Y. Cao, V. Fatemi, A. Demir, S. Fang, S. L. Tomarken, J. Y. Luo, J. D. Sanchez-Yamagishi, K. Watanabe, T. Taniguchi, E. Kaxiras, R. C. Ashoori, and P. Jarillo- Herrero, Correlated insulator behaviour at half-filling in magic-angle graphene superlattices, Nature 556, 80 EP (2018). [17] Y. Cao, V. Fatemi, S. Fang, K. Watanabe, T. Taniguchi, E. Kaxiras, and P. Jarillo-Herrero, Unconventional super- conductivity in magic-angle graphene superlattices, Na- ture 556, 43 EP(2018). [18] M. Yankowitz, S. Chen, H. Polshyn, Y. Zhang, K. Watan- abe, T. Taniguchi, D. Graf, A. F. Young, and C. R. Dean, Tuning superconductivity in twisted bilayer graphene, Science 363, 1059 (2019). [19] T. Stauber and H.Kohler, Quasi-Flat Plasmonic Bands in Twisted Bilayer Graphene, Nano Lett. 16, 6844 (2016). [20] F. Hu, S. R. Das, Y. Luan, T. F. Chung, Y. P. Chen, and Z. Fei, Real-Space Imaging of the Tailored Plasmons in Twisted Bilayer Graphene, Phys. Rev. Lett. 119, 247402 (2017). [21] N. K. Emani, T. F. Chung, X. Ni, A. V. Kildishev, Y. P.Chen, and A. Boltasseva, Electrically tunable damping of plasmonic resonances with graphene, Nano Letters 12, 5202 (2012). [22] T. Low and P. Avouris, Graphene Plasmonics for Tera- hertz to Mid-Infrared Applications, ACS Nano 8, 1086 (2014). [23] A. Principi, G. Vignale, M. Carrega, and M. Polini, Bulk SUPPLEMENTAL INFORMATION SPATIAL SPECKLE PATTERNS IN NEAR-FIELD OPTICAL MICROSCOPY Here we elaborate on the analysis connecting Eq. (3) and the speckle patterns shown in Fig. 1d and Fig. 2a-d. For an in-depth discussion of the near-field optical mi- croscopy measurement technique and quantitative mod- eling of the detected signal we refer the reader to refs. [4, 5, 14]. As argued in the main text, we can estimate the strength of the measured signal in the near-field opi- cal microscopy by evaluating the equal-point correlation function Eq. (3), which here we restate for convenience: S(r) = J0 d2r(cid:48)GE(r − r(cid:48))η(r(cid:48))GE(r(cid:48) − r). (S.1) (cid:90) It describes an amplitude of plasmon excitation, which traveled from the tip at position r to a disorder at po- sition r(cid:48) and was then reflected back towards the tip at r. Here the Green's function GE(r) of the plasmon ex- citation of wavenumber q0 is taken in the limit rq0 (cid:29) 1 as GE(r) ≈ eiq0r(cid:112)2πr e−δr, (S.2) which describes radially propagating waves in 2D. The factor e−δr describes damping due to extrinsic effects such as phonons and other inelastic processes. Upon sub- stitution of the Green's function into (S.1), the measured signal S(r) is given by S(r) = J0 d2r(cid:48)η(r(cid:48)) ei2q0r−r(cid:48)e−2δr−r(cid:48) 2πr − r(cid:48) . (S.3) (cid:90) This expression, which is a convolution of two functions, will generate a product under Fourier transform. For purposes of Fig. 1d and Fig. 2a-d we evaluate the above convolution numerically by using the convolution theorem, that is first performing a fast Fourier trans- form of both terms individually, multiplying them and then carrying out an inverse Fourier transform. The in- set of Fig. 1d and Fig. 2a-d is the intermediate step of this process, but we can also determine it analytically by evaluating the Fourier transform of the signal S(r) Sk = d2re−ik·rS(r). (S.4) (cid:90) As expected by the convolution theorem the expression factorizes into a product of two separate factors (cid:90) d2r(cid:48)η(r(cid:48))e−ik·r(cid:48)(cid:90) Sk = 8 where the first factor is nothing but the Fourier har- monic of η(x) and the second factor is the r-r(cid:48) influ- ence function, simplified by performing a variable change r − r(cid:48) → r. To evaluate the integral over d2r we first integrate over r and then carry out angular integration using the identity(cid:90) 2π dθ 1 a + b cos θ = 2π√ a2 − b2 . (S.6) 0 After substituting a = k, b = 2k0 + 2δi this gives Eq. (4) of the main text. BEHAVIOR OF THE INTRABAND AND INTERBAND POLARIZATION FUNCTIONS Here we discuss the behavior of the intraband and in- terband polarization functions Π1(ω, q) and Π2(ω, q) of the toy model, defined in Eqs. (12), (13) of the main text. In particular, we estimate the coefficients λ1 and λ2 describing the small-q behavior of Π1 and Π2, defined in the paragraph beneath Eq.(14). We are mostly inter- ested in high frequency values ω > 2W describing the intrinsically undamped regime. We start with the quantity λ2 describing the contribu- tion of intraband transitions. At small q, the interband coherence factor from Eq. (9) is non-negligible only in proximity of the points K and K(cid:48). Near these points a linear dispersion Es,k = svF k, with s = ±1, is a good approximation for the bandstructure. In that limit, the small-q interband coherence band factor from Eq. (11) becomes k+q,k ≈ 1 F s=−s(cid:48) 4 (q · ∇kϕk)2 ≈ 1 4 q2 k2 sin2 θ, (S.7) where θ is the angle between k and q. The quantity Π2(ω, q) is therefore given by: Π2(ω, q) = − 8q2 ω2 fs,ksvF k sin2 θ k2 . (S.8) (cid:88) k,s In the above we used the linear dispersion approximation Es,k = svF k for the whole band and accounted for the K and K(cid:48) points through an additional factor of 2. This gives q2 ω2 + 2W π q2 ω2 = Π2(ω, q) ≈ − 2EF π q2 ω2 , (S.9) with the first and second terms originating from the con- duction band and the valence band respectively. This gives (W − EF ) 2 π e−ik·r+i2q0re−2δr d2r 2πr , λ2 = 2(W − EF )/π, (S.10) (S.5) which takes positive values since −W < EF < W . Next, we proceed to estimate the λ1. Without loss of generality, we place the Fermi energy in the conduction band. In this case, the interband contribution to the polarization function is non-vanishing only in the con- duction band. This can be seen by going back to the Eq. (12), which for s = s(cid:48) = −1 and small q vanishes: Π1(ω, q) ≈ − 8 ω2 = − 8 ω2 f−1,k(E−1,k − E−1,k+q) (S.11) (E−1,k − E−1,k+q) = 0, (S.12) since f−1,k = 1 for all k in the valence band. It is there- fore sufficient to focus on the contribution of the par- tially filled (conduction) band. To be consistent with the λ2 analysis above we replace the dispersion energy as E1,k = vF k. The intraband contribution to the polariza- tion function Π1(ω, q) is then (cid:88) (cid:88) k k (cid:88) k Π1(ω, q) ≈ 8q2 ω2 giving f1,kvF sin2 θ k = 2 π EF q2 ω2 , (S.13) λ1 = 2EF /π. (S.14) As argued in the main text [see discussion below Eq. (14)], this result remains unchanged for frequencies ω < 2EF and, therefore, β0 = 4αvF EF . (S.15) Going back to the ω > 2W regime, and using λ1 and λ2 derived above, gives the square-root plasmon dispersion ωp = βq with √ β = 2πe2 κ (λ1 + λ2) = 4αvF W. (S.16) √ Therefore, at small ω < 2EF the dispersion behaves as √ ωp = 2 αvF EF q, becoming enhanced at high energies ω > 2W , ωp = 2 αvF W q by a factor(cid:112)W/EF . To complete the analysis of the polarization function behavior, now we focus on frequencies in the region 2EF < ω < 2W . Working again in the small-q limit we find that, as pointed out earlier, only the interband con- tribution to the polarization function Π2(ω, q) develops an imaginary part, whereas the intraband polarization function Π1(ω, q) is real-valued, given by the Eq. (S.13). To determine the form of Π2(ω, q) in the interband p-h continuum energy range, we approximate the coherence factor as in Eq. (S.7) to obtain Π2(ω, q) ≈ 8 fs,k svF k F k2 − (ω + i0)2 × q2 k2 sin2 θ. 4v2 (S.17) (cid:88) k,s 9 Here we used the linear approximation to the energy dis- persion Es,k = svF k, accounting for the fact that, be- cause of the behavior of the coherence factors, only the states near the Dirac point contribute to Π2. As always, we account for the K and K(cid:48) points by an additional factor of 2. After carrying out integration over d2k we arrive at: Π2(ω, q) ≈ −i 2q2 ω Θ(ω − 2EF )Θ(ω − 2W ). (S.18) Here Θ(x) is the Heaviside function, which ensures that the imaginary part is non-zero only in the particle-hole continuum region 2EF < ω < 2W . The dielectric func- tion in this region is therefore ε(ω, q) = 1 − β0 (S.19) , q ω2 + i β0π EF q ω which shows that the collective mode ωp in the 2EF < ωp < 2W region has a damped square-root dispersion ωp ≈(cid:112)β0q − i πβ0 2EF q. (S.20) The imaginary part, which scales linearly with q, de- scribes damping due to particle-hole pair production. We finish the discussion of the collective modes by com- paring the analytically predicted dispersion with the nu- merical result in Fig. 1a. While the simulated dispersion q, the agreement between the simulation and ωp = (cid:112)βqq dis- closely follows the square-root dependence ωp ∝ √ persion is drastically improved if the two first terms β0 and β1 from the series expansion in Eq. (14) are used for a fitting. Although the terms β0 and β1 could in prin- ciple be computed by carrying out an expansion of the polarization function in Eq. (10) in powers of q and then evaluating the resulting integrals numerically, we instead treat β0 and β1 as free parameters and fit them to the simulated dispersion. This approach yields values β0 = 0.96 × 103 meV2 nm, β1 = −103 meV2 nm2. (S.21) The best-fit β0 value is close to β0 = 4αvF EF ≈ 0.86 × 103 meV2 nm predicted from Eq. (S.15). We also see that, since β1 is negative, the plasmon dispersion is in- deed softened by interband polarization, in agreement with the argument given in the main text [see Eq. (14)]. TWISTED BILAYER GRAPHENE - DETAILS OF THE MODEL Here we describe in detail the model for twisted bilayer graphene (TBG) bandstructure used in the main text. We use the effective continuum Hamiltonian introduced in ref. [33], adopting notations and numerical values used in ref. [33]. The continuum approach is made possible by the small values of the twist angle θ by which the two graphene layers in TBG are rotated relative to one another. We start by taking two AA-stacked graphene layers and ro- tating the layer 1 and the layer 2 around the B-sites by −θ/2 and θ/2 respectively. For the "magic" value of θ = 1.05◦, the moir´e real-space lattice constant is LM = a/2 sin(θ/2) ≈ 13.4 nm. This is two orders of magnitudes greater than the graphene's lattice constant a = 0.246 nm, justifying the use of the continuum ap- proach. In momentum space this real-space rotation translates into two graphene Brillioun zones rotated by angle θ rel- ative to each other. Both BZs are centered at the same Γ point but the K (and K(cid:48)) points of the two layers are sep- arated by a small momentum 4π/(3LM ). As the moir´e periodicity LM is much greater than the lattice constant a, we can ignore the intervalley mixing between the two valleys K and K(cid:48) of the original graphene layers - here labeled by ξ = −1, 1. The total Hamiltonian of the sys- tem becomes therefore block diagonal in the valley index. The blocks H (ξ) describing each of the two valleys take the form (cid:18)H1 U† (cid:19) U H2 H (ξ) = (S.22) in the basis of (A1, B1, A2, B2) sites. The matrices Hl (l = 1, 2) correspond to the intralayer Hamiltonians of the layers. The latter, due to the lengthscale separation between LM and a, can be approximated by performing the standard kp expansion around the points K and K(cid:48). This procedure gives 2×2 Dirac Hamiltonians centered at the K(l) ξ points (cid:104) Hl = −v R (±θ/2) (k − K(l) ξ ) (S.23) where k is a momentum in the BZ of the original graphene layers, and R (ϕ) is the 2 × 2 rotation matrix (cid:105) · (ξσx, σy) , (cid:19) (cid:18)cos ϕ − sin ϕ sin ϕ cos ϕ R (ϕ) = (S.24) that accounts for rotation of the BZ of the original graphene layers. The signs ± in Eq. (S.23) correspond to the layers l = 1 and 2, respectively. The energy scale for the Hamiltonians Hl is v/a = 1 , K(l)−1, which denote the 2.1354 eV. The vectors K(l) Dirac points K and K(cid:48) of the layers, are given by K(1) ξ = −ξ 4π 3a R (−θ/2) , K(2) ξ = −ξ 4π 3a R (θ/2) 0 0 (S.25) respectively. We stress that, while k alone has length close to ∼ 4π/3a, the difference k − K(l) is small, since k is always located near the vicinity of the K(l) ξ points. ξ (cid:18)1 (cid:19) 10 This makes the linear expansion from Eq. (S.23) a well defined approximation. More quantitatively, the expressions in Eq. (S.23), found by Taylor expanding the graphene tight-binding Hamiltonian, are valid for momenta close enough to the Dirac points of the two layers, k−K(l) ξ a (cid:28) 1. For θ (cid:28) 1 this condition is obeyed in the entire mini Brillouin zones of the TBG superlattice. In the analysis below the moir´e superlattice BZ is de- fined as in the inset of Fig. 3, with the two reciprocal lattice vectors 1 = − 2π√ GM 3LM , GM 2 = 4π√ 3LM . (S.26) (cid:19) (cid:18) 1√ 3 (cid:18)1 (cid:19) 0 1 = GM We denote the reciprocal lattice vector length as GM = GM 3LM . Matrix U is the effective moir´e interlayer coupling given by: (cid:18) u u(cid:48) u(cid:48) u √ 2 = 4π/ (cid:19) (cid:18) u (cid:18) u u(cid:48)νξ + + u(cid:48)ν−ξ (cid:19) (cid:19) u(cid:48)ν−ξ u u(cid:48)νξ u U = eiξGM 1 ·r+ eiξ(GM 1 +GM 2 )·r, (S.27) (cid:88) where we introduced a notation for the phase factor ν = ei2π/3. The interlayer couplings u and u(cid:48) are taken as u = 0.0797 eV and u(cid:48) = 0.0975 eV to match values in ref. [33]. To determine the energy bands and the eigenstates we take the Bloch wavefunction ansatz for a valley ξ as ΨX ξ,n,k(r) = ξ,n,k(G)ei(k+G)·r C X (S.28) G with X labeling the spinor components X = A1, B1, A2, B2. The band index, labeled by n and k, is the Bloch wave vector in the BZ of the original graphene layers. Here G runs over all possible integer combinations of the reciprocal lattice vectors, G = m1GM 2 with integer m1 and m2. As discussed in ref. [33], the low- energy states are expected to be dominated by states near the original Dirac points. We therefore take only not-too- large indices m1 and m2 that satisfy the condition 1 + m2GM k + G − M ξ ≤ zGM , (S.29) (cid:18)1 (cid:19) (cid:16) , M ξ = 1 2 where z is a conveniently chosen number of order one [ref. [33] uses z = 4], and M ξ are the "mean" Dirac point locations K(1) ξ + K(2) ξ = − 4π 3a ξ cos(θ/2) , (S.30) given by the midpoint between the K (or K(cid:48)) points of the two layers. (cid:18)1 (cid:19) 0 (cid:17) ELECTRON LOSS FUNCTION FOR THE TBG BANDSTRUCTURE local field effects [37, 38]. Although these effects are often small, they require careful examination and thus will be a subject of a future work. 11 Fig. 4 details the behavior of the electron loss function for TBG, depicted in Fig. 3 of the main text. Panels a and b show constant-momentum q linecuts of the real and imaginary parts of the dielectric function ε(ω, q). The finite width of the plasmon resonance in the loss function in Fig. 4c is due to the infinitesimal imaginary part of ω + i0 in the polarization function in Eq. (8) replaced with ω + iγ, with a suitably chosen small γ introduced for illustration purposes. Strong electron-electron interactions in the narrow electron bands lead to large dielectric function values, as can be seen in Fig. 4. For energies ω < 2W the dielectric function imaginary and real parts take values a few orders of magnitude higher than those of graphene monolayer. The origin of these large values can be traced to the high effective fine structure constant (or, equiv- alently, low Fermi velocity) in the flat electron bands, as discussed in the main text. To see this in more de- tail, we recall the Thomas-Fermi expression for the long- wavelength static dielectric function of graphene [2] ε(ω = 0, q → 0) = 1 + qT F /q (S.31) with the Thomas-Fermi momentum qT F = N αkF , where N is the degeneracy factor N = 8 (2 spins, 2 layers, 2 valleys). For illustration purposes, taking a fine structure constant α ∼ 30 and Fermi momentum kF ∼ K, for the momentum q ∼ K/2 (red line in the Fig. 4) Eq. (S.31) predicts a dielectric function value ε ∼ 480, which is in good agreement with the simulation results. Above ω > 2W the dielectric function rapidly decreases until ω > 20 meV where the contributions of higher electron bands start to dominate. At these energies, plasmon dispersion is strongly af- fected by the presence of higher electron bands. At small q plasmon dispersion is predominantly due to intra- band transitions, and is thus insensitive to other electron bands. At large q the situation changes. In the absence of higher electron bands the zeros of the dielectric func- tion would occur at much larger energy scales ωp ∼ 40 meV. However, as argued in Eq. 19 in the main text, higher electron bands push plasmon dispersion down with the large-q zeros of the dielectric function on the order W ∆ ≈ 6.6 meV. Here W is the flat-band band- width and ∆ is the band gap as defined in the main text. The independence of this value of α is the behavior to be expected for large enough α, such that plasmon dis- persion extends above the p-h continuum. The indepen- dence of ωp of α at α (cid:29) 1 is a characteristic feature of interband plasmons. ωp ≈ √ Lastly, we note that our simulation is expected to be accurate only for q inside the TBG Brillouin zone. When q approaches zone boundary it is necessary to consider FIG. 4. Imaginary (a) and real (b) parts of the TBG's di- electric function ε(ω, q) for several momenta values marked by the colored lines in the inset of (a). The zoom-in in panel (b) shows the positions of plasmon resonances found from ε(ω, q) = 0. The inset in (a) is a replica of the loss function shown in Fig. 3; higher-resolution linecuts at the selected mo- menta are presented in (c). The curves in (a-c) were smoothed with an equal-weighted moving filter. 50010001500024680.010.020.03a)b)c)0500100015006789-101

1601.07755

2

1601

Spin-Cherenkov effect in a magnetic nanostrip with interfacial Dzyaloshinskii-Moriya interaction

Spin-Cherenkov effect enables strong excitations of spin waves (SWs) with nonlinear wave dispersions. The Dzyaloshinskii-Moriya interaction (DMI) results in anisotropy and nonreciprocity of SWs propagation. In this work, we study the effect of the interfacial DMI on SW Cherenkov excitations in permalloy thin-film strips within the framework of micromagnetism. By performing micromagnetic simulations, it is shown that coherent SWs are excited when the velocity of a moving magnetic source exceeds the propagation velocity of the SWs. Moreover, the threshold velocity of the moving magnetic source with finite DMI can be reduced compared to the case of zero DMI. It thereby provides a promising route towards efficient SW generation and propagation, with potential applications in spintronic and magnonic devices.

cond-mat.mes-hall

cond-mat

Spin-Cherenkov effect in a magnetic nanostrip with interfacial Dzyaloshinskii-Moriya interaction Jing Xia,1, 2 Xichao Zhang,1, 2 Ming Yan,3 Weisheng Zhao,4, 5, ∗ and Yan Zhou1, 2, 6, † 1School of Electronics Science and Engineering, Nanjing University, Nanjing 210093, China 2Department of Physics, University of Hong Kong, Hong Kong, China 3Department of Physics, Shanghai University, Shanghai 200444, China 4Fert Beijing Institute, Beihang University, Beijing 100191, China 5School of Electronic and Information Engineering, Beihang University, Beijing 100191, China 6Center of Theoretical and Computational Physics, University of Hong Kong, Hong Kong, China (Dated: July 12, 2021) Spin-Cherenkov effect enables strong excitations of spin waves (SWs) with nonlinear wave dispersions. The Dzyaloshinskii-Moriya interaction (DMI) results in anisotropy and nonreciprocity of SWs propaga- tion. In this work, we study the effect of the interfacial DMI on SW Cherenkov excitations in permalloy thin-film strips within the framework of micromagnetism. By performing micromagnetic simulations, it is shown that coherent SWs are excited when the velocity of a moving magnetic source exceeds the prop- agation velocity of the SWs. Moreover, the threshold velocity of the moving magnetic source with finite DMI can be reduced compared to the case of zero DMI. It thereby provides a promising route towards efficient SW generation and propagation, with potential applications in spintronic and magnonic devices. PACS numbers: 75.30.Ds, 75.78.Cd, 85.70.-w, 85.75.-d The Cherenkov radiation of light occurs when a charged par- ticle moves faster than the light speed within a medium [1]. This effect is named after Pavel Alekseyevich Cherenkov, who studied this phenomenon experimentally. Cherenkov radia- tion is used frequently in particle identification detectors in particle physics [2]. The Cherenkov effect is analogous to the sonic boom produced by shock waves propagating away from an aircraft, if its speed exceeds the sound velocity. Similar to the Doppler effect observed in different physical system [3 -- 7], the Cherenkov effect belongs to one of the fundamental phenomena induced by the radiation of moving sources. The Cherenkov-like effect of spin waves (SWs) has been theoreti- cally studied in ferromagnets, which can be used to excite co- herent SWs without the necessity of external alternating mag- netic field or current [8, 9]. Recently, the study of the influ- ence of the antisymmetric exchange interaction, namely the Dzyaloshinskii-Moriya interaction (DMI), on magnetic exci- tations such as domain walls and vortex is one of the hottest topics in nanomagnetism and spintronics [10 -- 20]. DMI is an antisymmetric interaction induced by spin-orbital coupling due to broken inversion symmetry in lattices or at the inter- face of magnetic films [12], which has been measured for both magnetic interfaces [21, 22] and bulk materials [23, 24]. DMI facilitates the creation of topologically protected spin textures in chiral magnetic materials, i.e. magnetic skyrmions, which are favorable information carriers in the next-generation data storage and spin logic devices. On the other hand, theoret- ical [13, 14, 16] and experimental [10, 15, 25] studies have demonstrated that DMI leads to an asymmetrical spin-wave dispersion. DMI has also been measured in a wide range of materials including permalloy [22, 26, 27]. Specifically, the asymmetry in the formation of a vortex state in a permalloy ∗ [email protected] † [email protected] Figure 1. Schematics of the micromagnetically modeled system. An external rectangular shape field pulse (H), 12 nm wide in the x-direction and 100 nm long in the y-direction, is applied to a 12- µm-long, 100-nm-wide, and 10-nm-thick magnetic strip with a mag- nitude of 10 mT in the z-direction and constant speed of vh. The color scale represents the out-of-plane component of the magnetiza- tion mz, which has been used throughout this paper. nanodisk has been studied by micromagnetic simulations with interfacial DMI [22]. In this paper, the influence of interfacial DMI on the Spin- Cherenkov effect (SCE) in permalloy strip are studied by mi- cromagnetic simulations. Pictorial illustrations of the setup is shown in Fig. 1, where a rectangular moving magnetic field pulse is applied to the permalloy strip with a magnitude of 10 mT along the +z-direction. Our numerical results show that the interfacial DMI leads to a reduction of the threshold veloc- ity of moving source, i.e. the minimum SW phase velocity, for coherent SW excitation in the absence of external ac magnetic field. Therefore it provides a promising route to reducing SCE threshold to facilitate experimental realization of such effect in magnetic medium. 6 1 0 2 r p A 2 1 ] l l a h - s e m . t a m - d n o c [ 2 v 5 5 7 7 0 . 1 0 6 1 : v i X r a Results SW excitation via SCE in permalloy strips. The response of the magnetization distribution to the pulse velocity vh is shown in Fig. 2. A localized magnetic field of constant mag- nitude traveling along the wire axial direction with velocity of vh, is applied to mimic the interacting force with the magneti- zation. The magnetization dynamics is strongly dependent on the pulse velocity vh. For D = 0 mJ m−2 and vh = 500 m s−1 and 900 m s−1 (see Fig. 2(a1-a2)), the moving magnetic field pulse only causes a distortion of the magnetization dis- tribution traveling with the pulse. There is no SW excitation by the moving pulse. As the velocity of the magnetic pulse increases, SW excitations are observed for vh = 1050 m s−1, 1100 m s−1, and 1200 m s−1, as shown in Fig. 2(a3-a5). At vh = 1050 m s−1, the SWs proceeding the source and lagging the source are well-distinguished with different wave lengths. Upon the application of a moving dc magnetic field pulse, the system reaches a dynamic equilibrium and the excited spin waves comprise of two branches, giving rise to the SCE as reported in Ref. 9. For the permalloy nanostrip with interfacial DMI, similar magnetization dynamics occur as the moving field pulse is applied. For the case of D = 1 mJ m−2, there is no SW excitation by the moving magnetic field with the speed of vh = 500 m s−1 and 900 m s−1. The Cherenkov emission of SWs emerge at vh = 1050 m s−1, 1100 m s−1, and 1200 m s−1 in the permalloy strip as shown in Fig. 2(b3-b5). The SWs excited in permalloy nanostrip are distorted due to the presence of DMI. The interfacial DMI term for any in-plane direction u can be expressed as Dz × u [28], which may be treated as an effective field transverse to the magnetic track, resulting in the distorted spin waves. As D = 2 mJ m−2, spin wave is excited when vh = 900 m s−1, indicating that the presence of interfacial DMI leads to the decreasing of the threshold velocity of the moving field source for spin wave excitation. It should be noticed that the SWs with more sig- nificant distortions are observed at vh = 1050 m s−1, 1100 m s−1, and 1200 m s−1 as shown in Fig. 2(c3-c5). Moreover, the difference between the SW branches proceeding and lagging the source becomes even more obvious with increasing DMI strength for a given pulse velocity. Figure 3 shows the numerically determined SW phase ve- locity vp(k) and group velocity vg(k) for D = 0 mJ m−2. The inset of Fig. 3 shows the SW dispersion in the permalloy strip, which agrees well with the analytical results for zero DMI (proportional to k2). In Fig. 3, the phase velocity vp and group velocity vg of the SWs are extracted from the dispersion relation of vp(k) = ω/k and vg(k) = dω/dk. The minimum of the phase velocity occurs around 1000 m s−1, meaning that no SW is excited by the moving field pulse in the permalloy strip for the pulse velocities below the minimum v0. In other words, the velocity v0 is the threshold velocity of SCE, below which there is no coherent SWs excitation. As the field pulse moves at a velocity larger than v0, there are two SW modes excited with different group velocity vg but equal phase ve- locity vp. The SW packet with larger (cid:126)k (vg > vp) moves in the front of the source and leaves the one with smaller (cid:126)k (vg < vp) behind. The field pulses with different velocity vh 2 Figure 2. Snapshots of the z-component of the magnetization in a magnetic strip in the vicinity of the field pulse traveling at a constant speed. (a1) D = 0 mJ m−2 and vh = 500 m s−1, (a2) D = 0 mJ m−2 and vh = 900 m s−1, (a3) D = 0 mJ m−2 and vh = 1050 m s−1, (a4) D = 0 mJ m−2 and vh = 1100 m s−1, (a5) D = 0 mJ m−2 and vh = 1200 m s−1, (b1) D = 1 mJ m−2 and vh = 500 m s−1, (b2) D = 1 mJ m−2 and vh = 900 m s−1, (b3) D = 1 mJ m−2 and vh = 1050 m s−1, (b4) D = 1 mJ m−2 and vh = 1100 m s−1, (b5) D = 1 mJ m−2 and vh = 1200 m s−1, (c1) D = 2 mJ m−2 and vh = 500 m s−1, (c2) D = 2 mJ m−2 and vh = 900 m s−1, (c3) D = 2 mJ m−2 and vh = 1050 m s−1, (c4) D = 2 mJ m−2 and vh = 1100 m s−1, (c5) D = 2 mJ m−2 and vh = 1200 m s−1. are applied and the corresponding responses of the magneti- zation distribution are shown in Fig. 2(a1-a5). The calculated k are shown with color stars in Fig. 3. As vh is below the critical velocity, there is no SW excitation by the moving field pulse. When vh = 1050 m s−1, there are two SW modes observed in the permalloy strip. The spin waves formed in front of and behind the pulse exhibit different characteristics in Fig. 2(a3), and the corresponding k values are shown in Fig. 3. The Cherenkov emission of SWs can also be observed for the case of vh = 1100 m s−1 and 1200 m s−1. The disper- sions are in an excellent agreement with the curves of vp(k) and vg(k). The curves of vp(k) and vg(k) extracted from the numerical SW dispersion relation ω(k), which is obtained by 3 Figure 3. Numerically determined phase velocity vp(k) and group velocity vg(k) of SWs in the magnetic strip which is 12 µm long, 100 nm wide, and 10 nm thick as D = 0 mJ m−2. The value of vp(k) and vg(k) are extracted from the SW dispersion rela- tion ω(k) shown in the inset. vp(k) has a minimum v0 at k0, where the two curves vp(k) and vg(k) cross. The colored stars represent the wave vectors of the SW tails excited by the moving field pulse applied to the strip at the corresponding speed. The colored horizon- tal line, indicating the speed of the field pulse, connect the two SW branches. Figure 4. Numerically determined phase velocity vp(k) and group velocity vg(k) of SWs in the magnetic strip, with a length of 12 µm, a width of 100 nm, and a thickness of 10 nm when D = 2 mJ m−2. The value of vp(k) and vg(k) are extracted from the SW dispersion relation ω(k) shown in the inset. vp(k) has a min- imum v0 at k0, where the curve vp(k) crosses the curve vg(k). The colored stars show the wave vectors of the SW tails excited by the moving field pulse applied to the strip at the corresponding speed. The colored horizontal line, indicating the speed of the field pulse, connects the two SW branches. applying a localized ac field and extracting the wavelength of the excited spin waves, can predict the Cherenkov excitation of SWs precisely. Influence of DMI on SCE in permalloy. Figure 4 shows the numerically determined phase velocity vp(k) and group velocity vg(k) with D = 2 mJ m−2, where vp(k) and vg(k) are calculated from the SW dispersion relation ω(k) shown in the inset. Similar to the case of D = 0 mJ m−2, there exists a critical velocity v0 for the Cherenkov excitation of SWs. The Cherenkov excitation of SWs can be observed for vh > v0 (887 m s−1). We also investigate the evolution of the SW branches by varying vh for D = 2 mJ m−2. At vh = 1050 m s−1, the Cherenkov excitation of SWs is observed with two SW modes formed in the permalloy strip. The calculated k marked with color stars agrees very well with the numerically determined curve of vp(k). The curves of vp(k) and vg(k) extracted from the SW dispersion relation ω(k) can predict the Cherenkov excitation of SWs precisely with the presence of finite DMI. Figure 5 shows the minimum velocity v0 of the Spin- Cherenkov excitation for D = 0 mJ m−2, 1 mJ m−2, 2 mJ m−2 and 3 mJ m−2, as well as the corresponding dispersion relation. Only the branches of positive k are shown. The crit- ical velocity for the SCE decreases linearly with increasing DMI in the investigated DMI range. As shown in the inset of Fig. 5, the dispersion relation for D = 1 mJ m−2 devi- ates from the case of D = 0 mJ m−2, resulting in a smaller critical velocity v0. v0 = 1000 m s−1 for D = 0 mJ m−2 whereas v0 = 944 m s−1 for D = 1 mJ m−2. The criti- cal velocity further drops to 887 m s−1 as D increases to 2 mJ m−2, indicating that the Cherenkov emission of SWs in permalloy strip with finite DMI can be excited more easily than the case of zero DMI. The SW dispersion relations with- out DMI (D = 0 mJ m−2) is a parabolic function of SW vec- tor k when k is large [9, 13, 29, 30]. In Refs. 14 and 25, the SW dispersion relation is given analytically when DMI is in- cluded, indicating that the interfacial DMI results in the asym- metric dispersion. Such asymmetric dependence due to the interfacial DMI depends on the spin wave vector k and equi- librium magnetization distributions. Different with the typi- cal Damon-Eshbach spin waves studied in Refs. 14 and 25, the SWs in our case propagate along the ±x-directions while the equilibrium magnetization is along the +x-direction, cor- responding to the back-volume mode. The SW dispersion is 0 − (kD∗mz0)2 when the DMI is con- given by ω = γ0 sidered. Here D∗ = 2D and ω0 is the angular frequency in the absence of DMI. mz0 is the z-component of the magneti- zation. The SW dispersion relation remains symmetric, when the interfacial DMI is included. The SCE threshold can be calculated by dω k , which decreases with DMI [9]. dk = ω (cid:112)ω2 µ0MS Discussion The Cherenkov emission of spin waves has been numerically studied by considering the effect of DMI in permalloy strip. The resonant Spin-Cherenkov effect can be excited as the ve- locity of moving magnetic field pulse exceeds a certain thresh- where M is the magnetization, Heff is the effective field, γ0 is the Gilbert gyromagnetic ration, and α is the Gilbert damping coefficient. The effective field reads as follows 4 (2) The average energy density E as a function of M is given by 0 . Heff = −µ−1 ∂E ∂M E = A[∇( M MS (n · M )2 M 2 S − µ0M · H M · Hd(M ) + EDM, )]2 − K − µ0 2 (3) Figure 5. The minimum velocity v0 for the Spin-Cherenkov exci- tation as D = 0 mJ m−2, 1 mJ m−2, 2 mJ m−2, and 3 mJ m−2. The insets shows the corresponding dispersion relations, which are obtained by applying a localized ac magnetic field and extracting the wavelength of the excited spin waves. old velocity. The Spin-Cherenkov effect threshold can be re- duced in the presence of finite DMI. By further tuning the ma- terial parameters and geometries, resonant spin waves can be excited through Spin-Cherenkov effect at much lower thresh- old velocity of the moving dc field source. On the other hand, it is also feasible to increase the interaction region of the mov- ing dc field source on the magnetic strip in real experiments. Indeed, we have investigated the Spin-Cherenkov effect with different sizes of the moving dc field source (see Supplemen- tary Figure 1), as well as with different thicknesses of the mag- netic strip (see Supplementary Figure 2). The Spin-Cherenkov effect in finite-DMI system might be interesting for funda- mental physics and also promising for potential applications in spintronic and magnonic devices, easing the experimental complexity and difficulty of applying an external ac magnetic field or current for resonant SW excitations. Methods Modeling and simulation. The micromagnetic simu- lations are performed using the Object Oriented Micro- Magnetic Framework (OOMMF) software including the interface-induced Dzyaloshinskii-Moriya interaction (DMI) extension module [19 -- 21, 31 -- 33]. The three-dimensional time-dependent magnetization dynamics is controlled by the Landau-Lifshitz-Gilbert (LLG) ordinary differential equa- tion [34 -- 36] dM dt = −γ0M × Heff + α MS (M × dM dt ), (1) where A and K are the exchange and anisotropy energy con- stants, respectively. H and Hd(M) are the applied and magne- tostatic self-interaction fields while MS = M(r) is the spon- taneous magnetization. The EDM is the energy density of the interfacial DMI of the form [21, 31, 36, 37] EDM = D M 2 S (Mz −Mx ∂Mx ∂x ∂Mz ∂x + Mz − My ∂My ∂y ∂Mz ∂y ), (4) where the Mx, My, Mz are the components of the magnetiza- tion M and D is the interfacial DMI constant. The five terms at the right side of equation (3) correspond to the exchange energy, the anisotropy energy, the applied field (Zeeman) en- ergy, the magnetostatic (demagnetization) energy and the in- terfacial DMI energy, respectively. The typical material parameters of permalloy, µ0MS = 1 T, exchange constant A = 1.3 × 10−11 J m−1, and zero anisotropy are adopted [8, 9, 22, 26]. Considering that the value of the effective interfacial DMI constant in the Py/Pt bi- layers has been estimated to be within the range of 1.0 ∼ 2.2 mJ m−2 in Refs. 27 and 38, the interfacial DMI constant D is varied from 0 to 3 mJ m−2 in this paper. A rectangular shape field pulse is applied to 12-µm-long, 100-nm-wide, and 10-nm-thick magnetic strip with a magnitude of 10 mT in the +z-direction and a 12 nm width in the x-direction, as shown in Fig. 1. The moving field pulse can be realized, for exam- ple, with a laser beam scanning over the surface of magnetic thin films [39]. The results with different widths of the mov- ing field pulse are shown in Supplementary Figure 1. The results with different thicknesses of the magnetic strip are shown in Supplementary Figure 2. For simplicity, the uni- form field along the film thickness is assumed. All samples are discretized into cells of 3 nm × 5 nm × 5 nm in the simu- lation. Gilbert damping coefficient α is set to be 0.02 and the value for Gilbert gyromagnetic ratio γ0 equals 2.211 × 105 m A−1 s−1. Initially, the magnetization orients along the +x- direction due to the shape anisotropy. The absorbing boundary condition has been implemented at both ends of the nanos- trip, which effectively avoids the spurious spin wave reflec- tions [40]. [1] Cerenkov, P. A. Visible radiation produced by electrons moving in a medium with velocities exceeding that of light. Phys. Rev. 52, 378-379 (1937). [2] Jelley, J. V. Cerenkov radiation and its applications. Br. J. Appl. [3] Berger, H. Complex Doppler effect in dispersive media. Am. J. Phys. 6, 227-232 (1955). Phys. 44, 851-854 (1976). [4] Hu, X., Hang, Z., Li, J., Zi, J. & Chan, C. T. Anomalous Doppler effects in phononic band gaps. Phys. Rev. E 73, 015602 (2006). [5] Lisenkov, I. V. & Nikitov, S. A. The complex Doppler effect in double negative media. J. Commun. Technol. Electron. 56, 687-689 (2011). [6] Vlaminck, V. & Bailleul, M. Current-induced spin-wave Doppler Shift. Science 322, 410-413 (2008). [7] Sekiguchi, K. et al. Time-domain measurement of current- induced spin wave dynamics. Phys. Rev. Lett. 108, 017203 (2012). [8] Yan, M., Andreas, C., Kákay, A., García-Sánchez, F. & Her- tel, R. Fast domain wall dynamics in magnetic nanotubes: sup- pression of Walker breakdown and Cherenkov-like spin wave emission. Appl. Phys. Lett. 99, 122505 (2011). [9] Yan, M., Kákay, A., Andreas, C. & Hertel, R. Spin-Cherenkov effect and magnonic mach cones. Phys. Rev. B 88, 220412 (2013). [10] Zakeri, K. et al. Asymmetric spin-wave dispersion on Fe(110): direct evidence of the Dzyaloshinskii-Moriya interaction. Phys. Rev. Lett. 104, 137203 (2010). [11] Nagaosa, N. & Tokura, Y. Topological properties and dynamics of magnetic skyrmions. Nat. Nanotech. 8, 899-911 (2013). [12] Fert, A., Cros, V. & Sampaio, J. Skyrmions on the track. Nat. Nanotech. 8, 152-156 (2013). P. [13] Cortes-Ortuno, D. & Landeros, the Dzyaloshinskii-Moriya interaction on the spin-wave spec- tra of thin films. J. Phys.: Condens. Matter. 25, 156001 (2013). Influence of [14] Moon, J.-H. et al. Spin-wave propagation in the presence of interfacial Dzyaloshinskii-Moriya interaction. Phys. Rev. B 88, 184404 (2013). [15] Cho, J. et al. Thickness dependence of the interfacial Dzyaloshinskii-Moriya interaction in inversion symmetry bro- ken systems. Nat. Commun. 6, 7635 (2015). [16] Kravchuk, V. P. Influence of Dzialoshinskii-Moriya interaction on static and dynamic properties of a transverse domain wall. J. Magn. Magn. Mater. 367, 9-14 (2014). [17] Wang, W. et al. Magnon-driven domain-wall motion with the Dzyaloshinskii-Moriya interaction. Phys. Rev. Lett. 114, 087203 (2015). [18] Zhou, Y. & Ezawa, M. A reversible conversion between a skyrmion and a domain-wall pair in junction geometry. Nat. Commun. 5, 4652 (2014). [19] Zhang, X., Ezawa, M. & Zhou, Y. Magnetic skyrmion logic gates: conversion, duplication and merging of skyrmions. Sci. Rep. 5, 9400 (2015). [20] Zhang, X. et al. All-magnetic control of skyrmions in nanowires by a spin wave. Nanotechnology 26, 225701 (2015). [21] Rohart, S. & Thiaville, A. Skyrmion confinement in ultrathin film nanostructures in the presence of Dzyaloshinskii-Moriya interaction. Phys. Rev. B 88, 184422 (2013). [22] Im, M. Y. et al. Symmetry breaking in the formation of mag- netic vortex states in a permalloy nanodisk. Nat. Commun. 3, 983 (2012). [23] Mühlbauer, S. et al. Skyrmion lattice in a chiral magnet. Science 323, 915-919 (2009). [24] Huang, S. X. & Chien, C. L. Extended skyrmion phase in epitaxial FeGe(111) thin films. Phys. Rev. Lett. 108, 267201 (2012). 5 [25] Di, K. et al. Direct observation of the Dzyaloshinskii-Moriya interaction in a Pt/Co/Ni film. Phys. Rev. Lett. 114, 047201 (2015). [26] Chen, S. J. et al. Effect of Dzyaloshinskii-Moriya interaction on the magnetic vortex oscillator driven by spin-polarized current. J. Appl. Phys. 117, 17B720 (2015). [27] Stashkevich, A. A. et al. Experimental study of spin-wave dis- persion in Py/Pt film structures in the presence of an interface Dzyaloshinskii-Moriya interaction. Phys. Rev. B 91, 214409 (2015). [28] Boulle, O. et al. Domain wall tilting in the presence of the Dzyaloshinskii-Moriya interaction in out-of-plane magnetized magnetic nanotracks. Phys. Rev. Lett. 111, 217203 (2013). [29] You, C.-Y. & Kim, N.-H. Critical Dzyaloshinskii-Moriya inter- action energy density for the skyrmion states formation in ultra- thin ferromagnetic layer. Curr. Appl. Phys. 15, 298-301 (2015). [30] You, C.-Y. Curie temperature of ultrathin ferromagnetic layer with Dzyaloshinskii-Moriya interaction. J. Appl. Phys. 116, 053902 (2014). [31] 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. Nanotech. 8, 839-844 (2013). [32] Donahue, M. J. & Porter, D. G. OOMMF User's Guide, Ver- sion 1.0 Interagency Report NISTIR 6376 (National Institute of Standards and Technology, Gaithersburg, MD, 1999). [33] Boulle, O., Buda-Prejbeanu, L. D., Jué, E., Miron, I. M. & Gaudin, G. Current induced domain wall dynamics in the pres- ence of spin orbit torques. J. Appl. Phys. 115, 17D502 (2014). [34] Gilbert, T. L. A Lagrangian formulation of the gyromagnetic equation of the magnetization field. Phys. Rev. 100, 1243 (1955). [35] Landau, L. & Lifshitz, E. On the theory of the dispersion of magnetic permeability in ferromagnetic bodies. Physik. Z. Sow- jetunion 8, 153-169 (1935). [36] Thiaville, A., Rohart, S., Jué, É., Cros, V. & Fert, A. Dynam- ics of Dzyaloshinskii domain walls in ultrathin magnetic films. Europhys. Lett. 100, 57002 (2012). [37] Bogdanov, A. N. & Yablonskii, D. A. Thermodynamically sta- ble "vortices" in magnetically ordered crystals. The mixed state of magnets. Zh. Eksp. Teor. Fiz. 95, 178-182 (1989). [38] Nembach, H. T., Shaw, J. M., Weiler, M., Jué, E. & Silva, T. J. Linear relation between Heisenberg exchange and interfacial Dzyaloshinskii-Moriya interaction in metal films. Nat. Phys. 11, 825-829 (2015). [39] Vorobev, P. V. & Kolokolov, I. V. Cherenkov emission of magnons by a slow monopole. JETP Lett. 67, 910-912 (1998). [40] Venkat, G., Franchin, M., Fangohr, H. & Prabhakar, A. Mesh size and damped edge effects in micromagnetic spin wave simu- lations. arXiv 1405.4615, http://arxiv.org/abs/1405.4615 (2014) (Accessed: 1st June 2014). Acknowledgements X.Z. was supported by JSPS RONPAKU (Dissertation Ph.D.) Program and was partially supported by the Scien- tific Research Fund of Sichuan Provincial Education De- partment (Grant No. 16ZA0372). M.Y. acknowledges the support by National Natural Science Foundation of China (Project No. 11374203). Y.Z. acknowledges the support by National Natural Science Foundation of China (Project No. 1157040329), the Seed Funding Program for Ba- sic Research and Seed Funding Program for Applied Re- search from the HKU, ITF Tier 3 funding (ITS/171/13 and 6 ITS/203/14), the RGC-GRF under Grant HKU 17210014, and University Grants Committee of Hong Kong (Contract No. AoE/P-04/08). W.S.Z. acknowledges the support by the projects from the Chinese Postdoctoral Science Founda- tion (No. 2015M570024), National Natural Science Founda- tion of China (Projects No. 61501013, No. 61471015 and No. 61571023), Beijing Municipal Commission of Science and Technology (Grant No. D15110300320000), and the In- ternational Collaboration Project (No. 2015DFE12880) from the Ministry of Science and Technology of China. X.Z. thanks M.J. Donahue for useful discussions. Author Contributions Y.Z. conceived the problem. Y.Z. and W.S.Z. coordinated the project. X.Z. carried out the numerical simulations. J.X. per- formed the theoretical analysis with the input from M.Y. All authors discussed the results and prepared the manuscript. Additional Information Supplementary information accompanies http://www.nature.com/srep this paper at Correspondence and requests for materials should be ad- dressed to W.Z. and Y.Z. Competing Financial Interests The authors declare no competing financial interests. 

1507.06420

1

1507

Magneto-transport in closed and open mesoscopic loop structures: A review

Magneto-transport properties in closed and open loop structures are carefully reviewed within a tight-binding formalism. A novel mesoscopic phenomenon where a non-vanishing current is observed in a conducting loop upon the application of an Aharonov-Bohm flux $\phi$ and we explore its behavior in the aspects of quantum phase coherence, electron-electron correlation and disorder. The essential results are analyzed in three different parts. First, we examine the behavior of persistent current in different branches of a zigzag carbon nanotube within a Hartree-Fock mean field approach using the second quantized formulation. The phase reversal of persistent current in several branches as a function of Hubbard interaction is found to exhibit interesting patterns. Our numerical results suggest a filling-dependent metal-insulator transition in a zigzag carbon nanotube. Next, we address the behavior of persistent current in an ordered-disordered separated nanotube keeping in mind a possible experimental realization of shell-doped nanowire which can provide a strange electronic behavior rather than uniformly doped nanowires. Finally, we focus our attention on the behavior of persistent current in an open loop geometry where we clamp an ordered binary alloy ring between two ideal semi-infinite electrodes to make an electrode-ring-electrode bridge. From our investigation we propose that under suitable choices of the parameter values the system can act as a $p$-type or an $n$-type semiconductor.

cond-mat.mes-hall

cond-mat

Magneto-transport in closed and open mesoscopic loop structures: A review Paramita Dutta1, ∗ and Santanu K. Maiti2, † 1Condensed Matter Physics Division, Saha Institute of Nuclear Physics, Sector-I, Block-AF, Bidhannagar, Kolkata-700 064, India 2Physics and Applied Mathematical Unit, Indian Statistical Institute, 203 Barrackpore Trunk Road, Kolkata-700 108, India Magneto-transport properties in closed and open loop structures are carefully reviewed within a tight-binding formalism. A novel mesoscopic phenomenon where a non-vanishing current is observed in a conducting loop upon the application of an Aharonov-Bohm flux φ and we explore its behavior in the aspects of quantum phase coherence, electron-electron correlation and disorder. The essential results are analyzed in three different parts. First, we examine the behavior of persistent current in different branches of a zigzag carbon nanotube within a Hartree-Fock mean field approach using the second quantized formulation. The phase reversal of persistent current in several branches as a function of Hubbard interaction is found to exhibit interesting patterns. Our numerical results suggest a filling-dependent metal-insulator transition in a zigzag carbon nanotube. Next, we address the behavior of persistent current in an ordered-disordered separated nanotube keeping in mind a possible experimental realization of shell-doped nanowire which can provide a strange electronic behavior rather than uniformly doped nanowires. Finally, we focus our attention on the behavior of persistent current in an open loop geometry where we clamp an ordered binary alloy ring between two ideal semi-infinite electrodes to make an electrode-ring-electrode bridge. From our investigation we propose that under suitable choices of the parameter values the system can act as a p-type or an n-type semiconductor. PACS numbers: 73.23.-b, 73.23.Ra., 71.23.An I. INTRODUCTION The abundant progress in nano-science and technology has stimulated us to fabricate different artificial nano- structures over the last few decades whose dimensions are comparable to and even smaller than the mean free paths or wavelengths of electrons such that they can move through the samples without randomizing their phase memories1. The manifestation of phase coherence length is one of the most striking aspects in the meso- scopic regime which can be obtained by lowering the temperature of the samples below sub-Kelvin tempera- tures since the scatterings due to electron-electron and electron-phonon interactions are highly suppressed in this temperature regime. Such experimentally accessible low- temperatures also make the system energy levels behave like discrete states which play a pivotal role in appear- ing several quantum-mechanical phenomena. Therefore, a mesoscopic system which can be modeled as a phase- coherent elastic scatterer2 provides us the opportunity to explore various novel quantum-mechanical effects be- yond the atomic realm3. The experimental realization of different quantum-mechanical incidents like, universal conductance fluctuations, non-local current-voltage re- lationship, new Onsager reciprocity relations, Coulomb blockade in micro-tunnel junctions4 -- 8, Anderson local- ization9, quantum Hall-effect10, Aharonov-Bohm (AB) oscillations11 -- 14, to name a few, has raised the popular- ity of the mesoscopic world among the scientists and en- gineers. Another reason behind this popularity can be attributed to the tailor-made geometries which look very simple but have high potential from the application per- spective. In addition, several other fluctuation patterns of the conductance are easily reproducible simply by tun- ing external parameters like, magnetic field, electric field, Fermi level, etc. The existence of dissipationless current, the so-called persistent current, in a mesoscopic normal metal ring pierced by an AB flux φ is one of such remarkable ef- fects which reveals the importance of phase coherence of electronic wave functions in low-dimensional quantum systems. The alluring question of persistent current in normal-metal rings threaded an AB flux was first ex- plored during 1960's15. Later, in 1983, Buttiker et al. have revived the interest regarding this topic16. They have predicted that a small isolated normal metal ring penetrated by a slowly varying magnetic flux carries an equilibrium current which does not decay and circulates within the sample. But its experimental realization was somewhat difficult because at that time it was a chal- lenging task to confine magnetic flux in such a small re- gion like nanoscale ring. However, a few years later Levy et al.17 have given first experimental evidence of persis- tent current in a mesoscopic metallic ring. They have observed the oscillations with period φ0/2 (φ0 = ch/e, the elementary flux-quantum) while measuring persis- tent current in an ensemble of 107 independent Cu rings. Similar oscillations with period φ0/2 have also been re- ported not only for an ensemble of disconnected 105 Ag rings18 but also for an array of 105 isolated GaAs-AlGaAs rings19. Later, many other theoretical20 -- 29 as well as ex- perimental30 -- 33 attempts have been done to explore the actual mechanisms of persistent current in single-channel rings and multi-channel cylinders. However, a contro- versy still persists among the experimental observations and theoretical estimates of persistent current ampli- tudes. All the experimentally measured current ampli- tudes were found to be one and two orders of magnitude larger than the theoretically predicted results, except in the case of nearly ballistic GaAs-AlGaAs rings30 where φ0-periodic persistent currents have been observed with amplitude I0 ∼ evF /L (vF and L are the Fermi velocity and circumference of the ring, respectively), which is very close to the value obtained from the free electron the- ory at absolute zero temperature (T = 0 K). In a recent work, Bluhm et al.33 have examined magnetic response of 33 individual cold mesoscopic gold rings, considering one ring at a time, using a scanning SQUID technique. Their results agree well with the theoretically estimated value20 in a single ballistic ring30 and an ensemble of 16 nearly ballistic rings34. But, the amplitudes of persistent current in single-isolated-diffusive gold rings35 are still two orders of magnitude larger than the theoretical cal- culations. This discrepancy initiated intense theoretical activity, and it is generally believed that the electron- electron correlation plays an important role in the disor- dered diffusive rings36 -- 38. An explanation based on the perturbative calculation in presence of interaction and disorder has been proposed and it seems to give a quan- titative estimate closer to the experimental results, but still it is less than the measured currents by an order of magnitude, and the interaction parameter used in the theory is not well understood physically. The behavior of persistent current in an isolated loop geometry enclosing a magnetic flux is highly sensitive to the location of Fermi level of the system where the role of magnetic flux is essentially to destroy the time-reversal symmetry and as a result, the degeneracies among the current-carrying states get removed. Depending on the Fermi level and the direction of the magnetic flux cur- rent flows in either direction revealing diamagnetic or paramagnetic nature. Since the discovery persistent cur- rent have been studying in different kinds of mesoscopic loop structures. Among them single channel mesoscopic rings20,39 were mainly in the focus of attention in spite of having topological simplicity they carry a high poten- tial from the application point of view. Not only single isolated rings but also array of such rings have been used to study the nature of persistent current. A very few works on multi-channel closed loop systems have been discussed40 -- 42. In case of multi-channel mesoscopic cylin- ders it has been noticed that the typical single-level cur- rent gets reduced with increasing number of conducting channels M . The correlations in the energy spectrum governs that the ratio of the total and single-level cur- rents is proportional to M 40. √ the non-decaying charge Similar to persistent current in such isolated closed loop geometries, current is also observed in open loop systems43 -- 51, where rings/cylinders are coupled to source and drain elec- trodes. In 1985, Buttiker has introduced a conceptually simple and elegant approach43 to investigate the effect of 2 an electron reservoir on persistent current in a loop pen- etrated by a magnetic flux. The reservoir is considered as a source of dissipation and the inelastic scattering pro- cesses take place only within this reservoir to which the loop is attached via a single current lead while the other scatterings occur in the lead are assumed to be elastic. So, there exists a complete spatial separation between the elastic and inelastic scatterings. The reservoir being a sink of electrons maintained the chemical potential of the loop to a fixed value. This modifies the statistical mechanical treatment of the system to a different ensem- ble, the grand canonical ensemble where close systems (in absence of lead and reservoir) belong to canonical ensemble average. This establishes a remarkable differ- ence between the study of closed and open loop systems from the statistical point of view. Experimental verifica- tion of persistent current in open system was first done in 199330. They have examined both by Mailly et al. closed and open systems and detected a periodic signal carrying the signature of persistent current which was in agreement with the theory. After those pioneering works last two decades have witnessed several approaches to reveal the properties of persistent currents in open sys- tems. Unlike the behavior of the other physical quantity such as conductance, persistent current in open system is sensitive to the direction of the transport current and this property is helpful for recognizing this current from the other currents (noise) associated with experimental measurements. In open systems, persistent current can appear even in absence of external magnetic field52 as the lead-current, the current from the source to drain, plays the role of the driving force. Jayannavar et al.52,53 have shown theoretically the flow of persistent current in an open metallic loop connecting to two electron reser- voirs. They have taken the lengths of the two arms of the loop unequal which results in a circulating current in the loop. Instead of setting the arm-lengths unequal one can also introduce any local scatterer anywhere in the geometry to establish a circular flow within the loop ge- ometry in non-equilibrium situation. On the other hand, in an equilibrium condition i.e., when the chemical poten- tials of both leads are same, persistent current can still arise due to evanescent modes. In a work by Jayannavar et al. the phenomenon of persistent current due to two non-classical effects, AB effect and quantum mechanical tunneling has been studied in detail53,54. Although the studies involving persistent current in single-channel rings and multi-channel rings55 or cylin- ders, both in the closed and open loop shapes, have al- ready generated a wealth of literature there is still need to look deeper into the problem to address several signifi- cant issues those have not been well explored before as for examples the understanding of the distribution of persis- tent currents in different channels in presence of electron- electron interaction and disorder which inspect the net response of the full system and also the sensitivity of persistent current on disorder in partially disordered sys- tems like shell-doped nanowires which can provide a un- familiar electronic behavior rather than uniformly doped nanowires. In the present review we essentially concen- trate on these issues. In the first part we address magneto-transport prop- erties in a zigzag carbon nanotube, formed by rolling up a graphite ribbon in the cylindrical form56, with its de- tailed energy band structure in presence electron-electron interaction within a Hartree-Fock (HF) mean field ap- proach using the second quantized formulation. Since the isolation of a single layer graphene by Novoselov et al.57 intense and diverse research is going on to explore elec- tronic transport in this system. Graphene, a single layer of carbon atoms tightly packed into a two-dimensional honey-comb lattice, has drawn attention of scientists in various disciplines due to its unconventional and fasci- nating electronic properties arising particularly from the linear dispersion relation around the Dirac points of the hexagonal Brillouin zone. These unique properties can be understood in terms of the Dirac Hamiltonian58 since it actually describes the physics of electrons near the Fermi level of the undoped material. The carriers in graphene effectively behave as massless relativistic particles within a low energy range close to Fermi energy and these mass- less Dirac Fermions59 evince various phenomena in this energy range. The bipartite character of the wonderful lattice structure of graphene strongly influences its intrin- sic properties and makes graphene a wonderful testbed for quantum field theory and mathematical physics as well as condensed matter theory. In the last few years ex- tensive studies on persistent current in carbon nanotubes have been performed and many interesting physical phe- nomena have been explored60 -- 62. Persistent current in a carbon nanotube is highly sensitive to its radius, chirality, deformation, etc. In a recent experiment it has also been observed that the Fermi energy of a carbon nanotube can be regulated nicely by means of electron or hole doping, which can induce a dramatic change in persistent cur- rent62. It is well established that in a conventional multi- channel mesoscopic cylinder electron transport strongly depends on the correlation among different channels as well as the shape of Fermi surface. Therefore, we might expect some interesting features of persistent current in a carbon nanotube due to its unique electronic structure. Here we discuss the behavior of persistent current in dif- ferent branches together with the total current of a zigzag carbon nanotube in presence of an AB flux. Most of the works available in literature20,60 -- 67 generally investigate magnetic response of the entire system, but a complete knowledge of magnetic response in individual branches provide much better insight to the problem68. The phase reversal of persistent current in different branches as a function of electron-electron correlation is found to ex- hibit interesting pattern. Our detailed numerical analysis suggests a filling-dependent metal-insulator (MI) transi- tion in a zigzag carbon nanotube. In the second part, we concentrate on the behavior per- sistent current in an ordered-disordered separated nan- otube considering a possible experimental realization of 3 shell-doped nanowires which may provide several un- usual electronic behavior rather than uniformly doped nanowires and have potential applications in nanoscale electronic and optoelectronic devices. In the last part we investigate persistent current to- gether with average density of states (ADOS) in an open loop geometry where an ordered binary alloy ring threaded by a magnetic flux is clamped between two ideal semi-infinite metallic electrodes, commonly known as source and drain electrodes, followed by the character- istic properties of an isolated ordered binary alloy ring. Inclusion of some foreign atoms in anyone of the two arms of the ring provides some interesting patterns in ADOS and from our numerical analysis we propose that under suitable choices of the parameter values the system can act as a p-type or an n-type semiconductor. Throughout the review we choose c = e = h = 1 for numerical calculations and restrict ourselves at absolute zero temperature. II. MAGNETO-TRANSPORT IN A ZIGZAG CARBON NANOTUBE This section is devoted to reveal the magnetic re- sponse of interacting electrons in a zigzag carbon nan- otube enclosing a magnetic flux within a HF mean field approach. Following the description of energy spectra for FIG. 1: (Color online). Schematic view of a zigzag graphite nano-ribbon with Nx and Ny number of atomic sites along the x and y directions, respectively. both non-interacting and interacting cases we investigate the energy-flux characteristics, persistent current in in- dividual branches of the system and also the net current of the entire system. A. The model and the mean field scheme We begin by referring to Fig. 1 where a graphite nano- ribbon of zigzag edges is shown. The filled magenta (large) and yellow (small) circles correspond to two dif- ferent sub-lattices, namely, A and B, respectively. Nx and Ny are the number of atomic sites along the x and y directions, respectively. In order to elucidate magnetic response of a nanotube we roll up the graphite ribbon along x direction using periodic boundary condition and allow to pass a magnetic flux φ (measured in unit of el- ementary flux quantum φ0 = ch/e) along the axis of the tube as shown in Fig. 2. Our model quantum system is illustrated by the nearest-neighbor tight-binding (TB) framework which captures most of the essential proper- 4 m and n. The site indexing is schematically shown in Fig. 3 for better viewing. The factor θ (= 2πφ/Nx), the so-called Peierl's73 phase factor, is introduced into the above Hamiltonian to incorporate the effect of magnetic flux applied along the axis of the tube. Decoupling of interacting Hamiltonian: Using the gener- alized HF approach, the so-called mean-field approxima- tion, we decouple the TB Hamiltonian into three different parts two of which correspond to two different values of σ, ↑ and ↓, with modified site-energies. The decoupled Hamiltonian is expressed as, H MF = H↑ + H↓ + H 0 where, H↑ = U m,n,↓(cid:105)na m,n,↑ + (cid:104)nb m+1,n,↓(cid:105)nb (2) m+1,n,↑(cid:1) m+1,n,↓(cid:1) m,n (cid:88) (cid:0)(cid:104)na (cid:16) (cid:88) (cid:88) (cid:0)(cid:104)na (cid:16) (cid:88) (cid:88) m+1,n,↑(cid:105)(cid:104)nb m,n m,n + t † m,n,↑bm−1,n,↑e−iθ + a a † m,n,↑bm+1,n,↑eiθ m,n † m,n,↑bm,n+1,↑ + h.c. + a (cid:17) , m,n,↑(cid:105)na m,n,↓ + (cid:104)nb m+1,n,↑(cid:105)nb † † m,n,↓bm−1,n,↓e−iθ + a m,n,↓bm+1,n,↓eiθ a + t m,n † m,n,↓bm,n+1,↓ + h.c. + a H 0 = −U m,n,↑(cid:105)(cid:104)na + (cid:104)nb (cid:0)(cid:104)na (cid:17) , m,n,↓(cid:105) m+1,n,↓(cid:105)(cid:1) . (3) where, H↑ and H↓ represent the up-spin and down-spin Hamiltonians, respectively. H 0 is a constant term which gives the energy shift. Here, na m,n,σ are the number operators associated with the A and B sites, re- spectively. m,n,σ and nb m,n,σ(cid:105) and (cid:104)nb Self-consistent procedure: In order to get the energy eigenvalues of the interacting Hamiltonian we go through a self-consistent procedure considering initial guess val- ues of (cid:104)na m,n,σ(cid:105). With these initial values, the up and down spin Hamiltonians are diagonalized nu- merically and a new set of values of (cid:104)na m,n,σ(cid:105) are calculated. These steps are repeated until a self- consistent solution is achieved. m,n,σ(cid:105) and (cid:104)nb Finding the ground state energy: After getting the self- consistent solution we determine the ground state energy (E0) at absolute zero temperature (T = 0 K) for a par- ticular filling by taking the sum of individual states upto the Fermi level (EF ) for both up and down spin electrons. The expression for ground state energy reads, (cid:88) (cid:88) E0 = Ei,↑ + Ei,↓ + H0 (4) i i where, i runs over the states up to the Fermi level. Ei,↑'s and Ei,↓'s are the single particle energy eigenvalues ob- FIG. 2: (Color online). A graphite nanotube threaded by a magnetic flux Φ. H↓ = U H = t ties of the tube nicely69 -- 72. To incorporate the effect of electron-electron interaction in the Hamiltonian we em- ploy a HF mean field approximation. In Wannier basis, the Hamiltonian of an zigzag nanotube takes the form, m,n,σbm−1,n,σe−iθ (cid:88) (cid:0)a† (cid:16) (cid:88) m,n,σbm+1,n,σ eiθ + a† † † m,n,↑am,n↑a m,n,↓am,n,↓ a m,n,σ + a† + U m,n † m+1,n,↑bm+1,n,↑b + b † m+1,n,↓bm+1,n,↓ m,n,σbm,n+1,σ (cid:1) + h.c. (cid:17) (1) † where, a† m,n,sigma) is the creation operator for an up spin or down spin electron of spin associated with m,n,σ (b FIG. 3: (Color online). Schematic view of different atomic sites with their co-ordinates. A (B) type of atoms at the position (m,n) and the corresponding annihilation operator is denoted by am,n (bm,n). t is the nearest-neighbor hopping integral and U is the strength of on-site Hubbard interaction. The co- ordinates of the lattice sites are denoted by two integers, tained by diagonalizing the up and down spin Hamilto- nians H↑ and H↓, respectively. B. Energy band structure To make this present communication a self contained study let us first start with the energy band structure of a finite width zigzag nano-ribbon. Non-interacting case: To establish the energy dispersion relation of a zigzag nano-ribbon we find an effective dif- ference equation analogous to the case of an infinite one- dimensional chain. This can be done by proper choice of 5 unit cells. a is the length of each side of a hexagonal ben- zene like ring. Finally, we solve Eq. 7 to get the desired energy dispersion relation (E vs. kx) of the ribbon. As illustrative example in Fig. 5, we display the varia- tion of energy levels (blue curves) as a function of wave vector kx for a finite width zigzag nano-ribbon consider- ing Ny = 4. At E = 0, nearly flat bands appear in the spectrum. The electronic states corresponding to those almost flat bands are strongly localized near the zigzag edges of the tube. The existence of these edge states have also been reported earlier by some other groups58,74,75. With this energy band structure of a finite width nano- ribbon we now pay attention on the variation of energy levels of a nanotube. In order to get the nanotube from the ribbon we apply periodic boundary condition along the x-direction76 which results quantized values of kx, and the quantized wave numbers are expressed from the FIG. 4: (Color online). Unit cell configuration of a zigzag nano-ribbon. a unit cell from the nano-ribbon. The schematic view of the unit cell configuration with Ny pairs of B-A atoms in a zigzag nano-ribbon is shown in Fig. 4. With this arrangement, we express the effective difference equation of the nano-ribbon in the form, (E I − Eσ)ψj,σ = T ψj+1,σ + T †ψj−1,σ (5) where,  ψj1B,σ ψj1A,σ ψj2B,σ . . ψjNyA,σ  . ψj,σ = (6) E and T are the site-energy and nearest-neighbor hopping matrices of the unit cell, respectively. I is a (2Ny × 2Ny) identity matrix. According to our convention, the translational invariance of the nano-ribbon exists along the x-direction and we can write ψj,σ in terms of the Bloch waves and then Eq. 5 takes the form, (EI − Eσ) = T eikxΛ + T †e−ikxΛ √ 3a is the horizontal separation between two where, Λ = filled magenta or yellow circles situated at two successive (7) FIG. 5: (Color online). Energy levels (blue curve) as function of kx for a finite width zigzag nano-ribbon considering Ny = 4. The discrete eigenvalues (filled black circles) of a nanotube with Nx = 12 and Ny = 4, in absence of AB flux φ, are superimposed. Here we set U = 0. relation kx = 4πnx/NxΛ, where nx is an integer lies within the range: −Nx/4 ≤ nx < Nx/4. Plugging the quantized values of kx in Eq. 7 we can easily determine the eigenvalues of a finite sized nanotube. In Fig. 5 we represent the discrete energy eigenvalues (filled black cir- cles) for a zigzag nanotube considering Nx = 12 and Ny = 4. It is to be noted that the results displayed in Fig. 5 correspond to the case when AB flux φ is set equal to zero. With these parameter values of the nanotube kx gets six quantized values (−π/Λ, −2π/3Λ, −π/3Λ, 0, π/3Λ and 2π/3Λ), and therefore, total 48 energy values are obtained since Ny is fixed at 4. Interacting case: In the presence of e-e interaction energy levels get modified significantly depending on the filling of electrons. The results calculated for a particular value of U are presented in Fig. 6 where we set Ny = 4. In the half-filled band case, a gap opens up at the Fermi energy77 which is consistent with the DFT calculations78 and the gap increases with the value of U . A careful investigation also predicts that the full energy band gets shifted by the factor U/2. -Π0Π-3.203.2kxssE
t6 for the interacting case (U (cid:54)= 0) we employ a mean-field scheme where the interacting Hamiltonian (Eq. 1) are de- coupled (for a particular filling) into two non-interacting Hamiltonians corresponding to up and down spin elec- trons and then diagonalize both H↑ and H↓ to get the energy eigenvalues of the system. Since in our case we set Nx = 10 and Ny = 7, we get total 70 independent energy levels and due to their overlaps individual energy levels are not clearly distinguished from the spectra given in Fig. 7. For identical filling factor of up and down spin electrons the energy levels are exactly similar both for H↑ and H↓ (see Fig. 7(b)), and therefore, one energy spectrum cannot be separated from the other. At E = 0, the energy levels become almost flat for a wide range of φ, and, near φ = ±φ0/2 they vary slowly with φ as shown in Fig. 7(a). In Fig. 7(b) the variation of energy levels with φ for a zigzag nanotube with the same parameter values sated above are plotted considering Hubbard in- teraction. Here we choose U = 1.5. Both for the up and down spin Hamiltonians the eigenvalues are exactly identical and they overlap with each other. In absence of FIG. 8: (Color online). Energy gap (∆E) as a function of on-site Hubbard interaction strength U for a zigzag nanotube with Nx = 10 and Ny = 7 in the half-filled band case when φ is set at φ0/2. interaction there is no gap in the center (around E = 0) of the band spectrum (Fig. 7(a)). As soon as the in- teraction is taken into consideration a finite gap opens up in the band center (Fig. 7(b)) and it increases with the strength of interaction (U ) as shown in Fig 8. This energy gap is consistent with the energy gap obtained in the E-kx diagram (Fig. 6(b)). All these energy levels vary periodically with φ providing φ0 (= 1 in our chosen unit) flux-quantum periodicity. At half-integer or integer multiples of φ0, energy levels have either a maximum or a minimum (see Fig. 7). Both the energy spectra take a complicated look as there are many crossings among the energy levels particularly in the regions away from E = 0 (Fig. 7). In Fig. 9 we present the variation of ground state en- ergy of a carbon nanotube with zigzag edges as a func- tion of magnetic flux φ for the half-filled band case. Here (a)-(d) represent the four different cases corresponding to four different values of electronic correlation strength U = 0, 0.5, 1 and 1.5, respectively. The energy levels evince one flux-quantum periodicity, as expected, and FIG. 6: (Color online). Energy levels as function of kx for a finite width zigzag nano-ribbon considering Ny = 4 and U = 1.4, where (a) and (b) correspond to the one-third- and half-filled cases, respectively. C. Energy-flux characteristics In this sub-section we examine the energy-flux charac- teristics of a zigzag nanotube. The results for a zigzag nanotube considering Nx = 10 and Ny = 7 are dis- played in Fig. 7, where (a) and (b) represent U = 0 and U = 1.5, respectively. In the absence of Hubbard FIG. 7: (Color online). Energy-flux characteristics of a half- filled zigzag nanotube with Nx = 10 and Ny = 7. (a) U = 0 and (b) U = 1.5. interaction (U = 0), energy levels are obtained simply by diagonalizing the non-interacting Hamiltonian and the nature of the energy spectrum becomes independent of the total number of electrons Ne in the system. While, -2Π-Π0Π2Π-2.700.45
a
3.60kxssE
t-2Π-Π0Π2Π-2.500.70
b
3.90kxssE
t-1-0.500.51-3.200
a
3.20ΦE-1-0.500.51-2.450.75
b
3.95ΦE00.511.5200.40.8UDE7 For a particular eigenstate ψp,σ(cid:105) persistent current in n-th channel becomes, n,σ = − e I p Nx (cid:104)ψp,σvn,σψp,σ(cid:105) (11) where, the eigenstate ψp,σ(cid:105) is written as, (cid:88) (cid:0)αp m,n +βp ψp,σ(cid:105) = m,n,σm, n, σ(cid:105) + βp m+1,n,σm + 1, n, σ(cid:105) + βp m−1,n,σm − 1, n, σ(cid:105) m,n+1,σm, n + 1, σ(cid:105)(cid:1) , (12) where, m, n, σ(cid:105)'s are the Wannier states and αp m,n,σ and βp m,n,σ's are the corresponding coefficients. Simplifying Eq. 11, we get the final relation of persistent charge cur- rent for n-th zigzag channel as, I p n,σ = (cid:88) m m,n,σe−iθ (cid:0)βp ∗ m−1,n,σe−iθ(cid:1) . m+1,n,σαp m+1,n,σeiθ − βp ∗ iet Nx −αp ∗ +αp ∗ m,n,σβp m,n,σβp m−1,n,σαp m,n,σeiθ (13) their energies get increased with U . D. Evaluation of persistent current in the second quantized form Following the second quantized formulation79,80, we es- timate persistent current in individual zigzag paths of a nanotube, threaded by an AB flux φ. This is an ele- gant and nice way of studying the response in separate branches of any quantum network. At first, we express the basic equation of current oper- ator Iσ corresponding to spin σ in terms of the velocity operator vσ (= xσ) as, Iσ = − 1 Nx e xσ (8) where, xσ is the displacement operator. The velocity operator is obtained from the relation, vσ = 1 i [xσ, Hσ] . (9) We use this expression to find the velocity operator of an electron with spin σ in a zigzag channel n (say) in the form, vn,σ = † m+1,n,σam,n,σe−iθ † m−1,n,σam,n,σeiθ t i − a† + a† m b (cid:16) (cid:88) m,n,σbm−1,n,σe−iθ(cid:1) . m,n,σbm+1,n,σeiθ − b  Ny(cid:88) (cid:0)αp ∗ n=1,3,... it 2Ny With the above prescription we can also evaluate persis- tent current in individual armchair paths (along y direc- tion) of the nanotube. Finally, it takes the form, (10) Ny(cid:88) m,n,σeiθ(cid:1) + Ny−1(cid:88) (cid:1) + (cid:0)βp ∗ (cid:0)βp ∗ n=2,4,... (cid:1) I p m−1,m,σ = m,n,σβp m−1,n,σe−iθ − βp ∗ m−1,n, σαp m,n,σαp m−1,n,σe−iθ − αp ∗ m−1,n,σ × βp m,n,σeiθ + βp ∗ m−1,n,σαp m−1,n−1,σ − αp ∗ m−1,n−1,σβp m−1,n,σ m,n+1,σαp m,n,σ − αp ∗ m,n,σβp m,n+1,σ where, an armchair channel (m− 1, m) is constructed by (m−1)-th and m-th lines according to our indexing. It is noteworthy to mention that all the calculations are done at absolute zero temperature (T = 0 K). Now, the net persistent current driven by electrons of spin σ in a par- ticular channel n for a nanotube described with Fermi energy EF can be determined by taking the sum of indi- vidual contributions from the lowest energy eigenstates upto the Fermi level. Therefore, we have, In,σ = I p n,σ. (15) (cid:88) p n=2,4,... (14) (cid:88) Taking the contributions from all possible channels n, both up and down spin (σ) electrons, the total persistent current in the nanotube can be expressed as, IT = In,σ. (16) n,σ To judge the accuracy of the persistent current calculated from the present scheme we determine persistent current in some other ways as available in literature. Proba- bly the simplest way of determining persistent current is the case where first order derivative of ground state en- ergy with respect to AB flux φ is taken into account24,39. Therefore, we can write, IT = −c ∂E0(φ) ∂φ (17) where, E0(φ) is the total ground state energy for a par- ticular electron filling. But, in our present method, the so-called second quantized approach, there are some ad- FIG. 9: (Color online). Ground state energy level as a func- tion of φ for a zigzag nanotube in the half-filled band case considering Nx = 20 and Ny = 8. (a), (b), (c) and (d) corre- spond to U = 0, 0.5, 1 and 1.5, respectively. vantages compared to other available procedures. With the second quantized formulation we can easily determine current in any branch of a complicated network and the evaluation of individual responses in separate branches helps us to elucidate the actual mechanism of electron transport in a more transparent way. E. Current-flux characteristics Now in this subsection we focus our attention on the behavior of persistent current in a zigzag nanotube. Here, also we adopt the unit where c = h = e = 1, fix t = −1 and measure all the physical quantities in unit of t. To illustrate the behavior of persistent current in sep- arate branches of a zigzag carbon nanotube we show in Fig. 10 the variation of persistent current in individual zigzag paths as a function of flux φ for the half-filled case considering Nx = 20 and Ny = 7, where (a)-(g) correspond to 1st-7th zigzag channels of the tube, re- spectively. The Hubbard correlation strength U is fixed at 1.2. Currents carried by up and down spin electrons are displayed in each of these figures simultaneously and they are exactly superposed on each other as the mag- nitudes and behaviors are exactly similar to each other. All these current profiles exhibit φ0 flux-quantum peri- odicity. There are no similarity in magnitudes of per- sistent currents corresponding to separate channels but also another similarity carried by them. Interestingly we 8 observe that I1↑ is exactly identical to I7↑, and, similarly for the (I2↑, I6↑) and (I3↑, I5↑) pairs. I4,↑, the current in the middle channel, becomes the isolated one since we have chosen Ny = 7. This is true for any zigzag nan- otube with odd Ny. For a tube with even Ny, currents are pairwise identical. Summing up the individual cur- rents in seven zigzag channels we get separately the net persistent current carried by up and down spin electrons in the nanotube which is presented in Fig. 10(h) and the total current is obtained by taking both contribu- tions IT↑ and IT↓ and it is depicted in Fig. 10(i). Now the total current derived from the conventional method where first order derivative of the ground state energy is taken into account, is displayed in Fig. 11. These two net currents calculated from the two different schemes are exactly identical to each other. It emphasizes that the net contribution of persistent current in a zigzag carbon nanotube comes only from the individual zigzag channels, not from the armchair paths. To justify it in Fig. 12 we present the variation of persistent current in an armchair path as a function of φ for a half-filled zigzag nanotube considering Nx = 20 and Ny = 7, which clearly shows zero current for the entire range of φ. In Fig. 7 we have already observed a few almost flat energy levels. These energy levels contribute a very little to the persistent current while, on the other hand, the energy levels with larger slopes provide large persistent current. Also, at the minima or maxima points persis- tent current becomes zero which is quite obvious since the current is obtained by taking the first order derivative of the eigen energy with respect to flux φ (Eq. 17). This pe- culiar nature of the energy levels invokes the current am- plitude to become filling dependent and we elaborate it in this section. To clarify this feature in Fig. 13 the current- flux characteristics of a zigzag nanotube, with Nx = 14, Ny = 6 and U = 1, is depicted where four different fig- ures correspond to the four different cases of electron fillings, Ne = 10, 20, 30, and 82. In all these cases, per- sistent current varies periodically with flux φ, exhibiting φ0 flux-quantum periodicity. Also, we observe that there are multiple kinks in the current profiles at different val- ues of φ when the number of electrons remains lower than that required for half-filling of the band. These kinks are associated with the multiple crossings of energy levels, as shown in Figs. 13(a)-(c) where the number of electrons are respectively, 10, 20, and 30. This is quite analogous to the feature of persistent current observed in conven- tional multi-channel mesoscopic cylinders. The behavior of persistent current gets significantly modified when the nanotube becomes half-filled or nearly half-filled. As il- lustrative example Fig. 13(d) is depicted. As we have already shown a few results for half-filled band case, here we choose Ne = 82 i.e., the nanotube is very near to the half-filled band condition. For this choice of parameter values it is examined that all the kinks disappear making the variation of persistent current quite smoother. This is analogous to the behavior of persistent current observed in traditional single-channel mesoscopic rings. For the a-1-0.50.501-106.1-106.4-106.7ΦE0b-1-0.50.501-97.45-97.75-98.00ΦE0c-1-0.50.501-88.75-89.05-98.00ΦE0d-1-0.50.501-80.25-89.45-89.65ΦE09 FIG. 10: (Color online). Persistent current in individual zigzag paths as a function of φ for a half-filled zigzag nanotube (Nx = 20 and Ny = 7) with U = 1.2, where, (a)-(g) correspond to 1st-7th zigzag channels of the tube, respectively. The net current corresponding to both up and down spin electrons are displayed in (h) while in (i) total persistent current is shown. FIG. 11: (Color online). Total persistent current obtained in a traditional derivative approach (Eq. 17) as a function of φ for the same parameter values mentioned in Fig. 10. FIG. 12: (Color online). Persistent current in an armchair path as a function of φ for a zigzag nanotube with the same parameter values as mentioned in Fig. 10. cases when the nanotube is far away from half-filling, current amplitudes are quite comparable to each other (see Figs. 13(a)-(c)). On the other hand, when the tube is nearly half-filled current amplitude remarkably gets suppressed and this suppression is very much clear from Fig. 13(d). The reason behind this enormous reduction of current amplitude can be understood clearly when we look at the E-φ characteristics of Fig. 7. At half-filling or very close to half-filling, the top most filled energy level lies in the nearly flat region i.e., around E = 0 (see Fig. 7(a)) and it contributes a little to the current. More- over, when U (cid:54)= 0, there is gap in the midband region. Now, for a particular filling we find the net persistent current taking the sum of individual contributions from a-1-0.50.501-.0030.003ΦI1,I1b-1-0.50.501-.0060.006ΦI2,I2c-1-0.50.501-.0090.009ΦI3,I3d-1-0.50.501-0.0100.01ΦI4,I4e-1-0.50.501-.0090.009ΦI5,I5f-1-0.50.501-.0060.006ΦI6,I6g-1-0.50.501-.0030.003ΦI7,I7h-1-0.50.501-0.0400.04ΦIT,ITi-1-0.50.501-0.0800.08ΦIT-1-0.50.501-.080.08ΦIT-1-0.50.501-1.001.0ΦI10 A. The model The schematic diagram of the model quantum ordered- disordered separated nanotube is illustrated in Fig. 14, where M co-axial rings are vertically attached and each ring contains N atomic sites. The cylinder is subjected to an AB flux φ. For a comparative study we will explore the behavior of persistent currents in a fully disordered and an ordered-disordered separated cylinder simultane- ously83. In order to get an ordered-disordered separated FIG. 14: (Color online). Schematic diagram of a mesoscopic cylinder having 6 rings where each ring contains 8 atomic sites. cylinder disorders are introduced into M/2 number of rings of the lower half of the system (green sites) whereas the remaining part (magenta sites) becomes the ordered one. On the other hand, full disordered system is ob- tained by putting random disorder in all the atomic sites. A non-interacting single-band tight-binding (TB) Hamil- tonian is used to describe the model quantum system which reads as, (cid:2)
m,nc† (cid:88) m,n H = m,ncm,n + tr(eiθr c† +tv(eiθv c† m,ncm+1,n + h.c.)(cid:3) m,ncm,n+1 + h.c.) (18) where,
m,n is the on-site energy. For the impurity sites, the site energies
m,n are selected randomly from a 'Box' distribution function of width W = 1. tr and tv refer to the intrachannel and interchannel nearest-neighbor cou- plings, respectively. Here, (m,n) is the co-ordinate of a lattice point where, m and n run from 1 to M and N , re- spectively. To incorporate the effect of magnetic flux we consider θr = 2πφ/N with the threading magnetic flux φ measured in unit of the elementary flux-quantum φ0. The phase factor along the vertical direction θv is taken as zero. c† m,n (cm,n) is the creation (annihilation) oper- ator of an electron at the site (m, n). At absolute zero temperature, the total persistent current of the cylinder is determined from the relation of Eq. 17, where E0 is the ground state energy of the system which is obtained FIG. 13: (Color online). Current-flux characteristics of a zigzag nanotube with Nx = 14 and Ny = 6, where (a), (b), (c) and (d) correspond to Ne = 10, 20, 30 and 82, respectively. U is fixed at 1. the lowest filled energy levels, and, in this process only the contribution which comes from the highest occupied energy level finally survives and the rest disappear due to their mutual cancellations. It leads to the enormous re- duction of persistent current amplitude in the half-filled or nearly half-filled band case. This feature is indepen- dent of the size of the nanotube. Before we end this section we can emphasize that, the current amplitude in a zigzag nanotube is highly sensitive to the electron filling and this phenomenon can be uti- lized in designing a high conducting to a low conducting switching operation and vice versa. III. AN ORDERED-DISORDERED SEPARATED NANOTUBE IN PRESENCE OF A MAGNETIC FLUX This section illustrates the behavior of persistent cur- rent in an ordered-disordered separated mesoscopic cylin- der. Persistent current in conventional disordered sys- tems have already been reported by many physicists and Anderson type localization9 due to disorder is not a new one. But, recent breakthroughs in the growth of semicon- ductor nanowires have opened up great opportunities to revolutionize technologies in nanoscale electronics81. Re- cent experiments on nanowires have already yielded some results in contrast to the phenomenon of strong localiza- tion due to doping. For instance, Cui et al.82 reported that the carrier mobility in boron-doped and phoporus- doped silicon nanowires under low dopant concentration is extremely low compared to bulk silicon but the con- ductance becomes diffusive and about five orders of mag- nitude larger with heavy doping. This motivates us to investigate whether these kinds of shell doped nanowires give rise to new features in the context of persistent cur- rent or not.
a
0.51-2.202.2ΦI
b
0.51-2.202.2ΦI
c
0.51-2.202.2ΦI
d
0.51-.060.06ΦIby taking the sum over lowest Ne (number of electrons) energy levels. B. Band structure In this sub-section we briefly discuss the energy band spectra of a mesoscopic cylinder. To find an analytic form of the energy levels we take a very small sized cylinder having only two rings with 10 sites in each ring. For this smallest possible size of the cylinder, expression of the energy levels takes the form, (cid:20) 2π(n + φ) (cid:21) E = ±tv + 2trCos Nsφ0 (19) where, Ns(= M xN ) is the total number of atomic sites of the cylinder. In Fig. 15 we plot the variation of energy levels as a function of magnetic flux φ. The numerical values of the parameters are taken as tr = tv = −1 and a = 1, where a is the lattice parameter. Two energy bands consisting of N number of energy levels within the boundary E = −1 to +3 and −3 to +1 are found to overlap in the region E = −1 to +1 (in unit of tr). The width of the overlap region can be tuned by tuning the FIG. 15: (Color online). Energy-flux characteristics for a mesoscopic cylinder with 2 rings where each ring contains 10 atomic sites. vertical hopping strength (tv). Each energy level exhibits one flux-quantum quantum periodicity. The variation of the energy levels of an ordered multi-channel system are already discussed in the previous section. Effect of the disorder can be attributed to the removal of degeneracy of the energy levels which can be realized from Fig. 16. The results are calculated numerically for a particular disorder configuration. In presence of impurity gap opens up at the crossing points and width of this gap increases with the increase of the disorder strength making the slope of the energy levels smaller and smaller. This makes a remarkable change in magnitude of persistent current. We discuss it in the forthcoming sub-section. Not only the full disordered cylinder here we also present the energy spectra of an ordered-disordered sep- arated cylinder in Fig. 17, where (a) and (b) correspond to two different values of the disorder strengths, W = 2.5 and 5, respectively. Here, the nature of the variation of 11 energy levels are much more different than that in the case of full disordered system. In order-disordered sepa- rated nanotube the energy levels become flatter mainly from the boundary of the spectra. C. Results and discussion In order to study the effect of disorder on persistent current in Fig. 18 we plot the typical persistent current amplitude I at a particular value of φ as a function of the disorder strength W for an ordered-disordered separated cylinder. Figure (a) corresponds to the results for a half- filled cylinder consisting of 10 rings with 12 sites in each FIG. 16: (Color online). Energy-flux characteristics for a mesoscopic cylinder with the same parameter values men- tioned in Fig. 15 in presence of disorder at all atomic sites (W = 3.5). ring and the magnetic flux φ = 0.4. On the other hand, figure (b) represents the results for a half-filled cylinder having 8 rings (10 atomic sites in each ring) in presence of magnetic flux φ = 0.2. Here we compute the root mean square of the current amplitude taking the average over 30 random disordered configurations. We observe that for a full disordered cylinder the current amplitude de- creases with the rise of impurity strength W (red curve), while in the case of ordered-disordered separated cylinder the current amplitude initially decreases upto a certain value of W after which it slowly increases with the rise of FIG. 17: (Color online). E-φ curves for an ordered-disordered mesoscopic cylinder with 2 rings and 10 atomic sites in each ring. the impurity strength (green curve). This phenomenon can be explained as follows. In the presence of disorder -2-1012-3.203.2ΦE-2-1012-3.203.2ΦE-2-1012-4.20
a
4.2ΦE-2-1012-4.20
b
4.4ΦEcurrent amplitude decreases due to the localization of the energy levels which is the so-called Anderson localization. The more impurity strength results the more reduction in current amplitude. This is true only for a full disor- dered system. But, for the ordered-disordered separated system there is a different physical picture. As we gradu- ally tune the disorder strength towards the higher value, the ordered and disordered regions get decoupled from each other and after a critical limit of W , (say Wc), the cylinder behaves as composed of two completely decou- pled regions. At this situation only the ordered region contributes to the current, and thus, the current ampli- tude gets enhanced with the disorder strength W . The critical value Wc depends on the system size and other 12 A. The model Let us concentrate on the simplest model of a binary alloy ring as shown in Fig. 19, where the ring consists of two different types of atoms placed alternately in a regular pattern. They are characterized by two different on-site potential energies, namely, α and β. The ring is subjected to an AB flux φ. Within a non-interacting single-band TB framework we illustrate the model of bi- FIG. 19: (Color online). A binary alloy ring, threaded by a magnetic flux φ, is composed of two different types of atomic sites, viz, α and β those are represented by filled blue and red circles, respectively. FIG. 18: (Color online). I-W characteristics of an ordered- disordered separated cylinder in the half-filled band case. (a) M = 10, N = 12 and φ = 0.4 and (b) M = 8, N = 10, and φ = 0.2. The red and green color correspond to full disordered and half-disordered cylinder. parameter values like magnetic flux. As for example, it is 3 and 2.5 for two different system sizes as shown in Figs. 18(a) and (b), respectively. To summarize, in this section we have explored the behavior of persistent current in a disordered and an ordered-disordered separated cylinder in presence a mag- netic flux. Most interestingly, we see that the current amplitude shows an anomalous behavior with the in- crease of impurity strength for an ordered-disordered sep- arated cylinder in contrast to the completely disordered one. This study may be helpful for exploring localization- delocalization transition in shell-doped nanotubes. In this section we undertake an analysis of the band structure and persistent current in an isolated binary al- loy ring (no source and drain electrodes) enclosing a mag- netic flux φ84 IV. A BINARY ALLOY RING WITHOUT EXTERNAL ELECTRODES E =
α +
β 2 ± + 4 t2 r cos2(ka) (21) (cid:88) nary alloy ring and the TB Hamiltonian reads, H R = † l cl + treiθc † l cl+1 + tre−iθc † l+1cl) (20) (
lc l where, the on-site energy,
l takes two values
α and
β corresponding to two different sites α and β, respectively. tr is the nearest-neighbor hopping integral. The phase factor θ = 2πφ/N of the Hamiltonian takes an account of the effect of the magnetic flux φ threaded by the ring which is measured in unit of the elementary flux-quantum † φ0. c l (cl) is the creation (annihilation) operator of an electron at the site l. Here, l runs from 1 to N , where N is the total number of sites in the binary ring. B. Energy spectrum Before addressing the main points i.e., the character- istic features of persistent current in an ordered binary alloy ring, let us have an idea about the energy band structure of the system. The analytical expression of en- ergy dispersion relation for the ordered binary alloy ring is as follows, (cid:115)(cid:18)
α −
β (cid:19)2 where, a is the lattice spacing and k is the wave vec- tor. The periodic boundary condition of the ring sets the quantized values of k in presence of the AB flux φ as, k = 2π N a n + φ φ0 (22) 2 (cid:18) (cid:19)
a


01.534.5600.20.61.0WI
b
01.534.5600.20.81.4WItrαβΦ13 E = ±(cid:112)
2 + 4 t2 Ne(cid:88) where, n is an integer and it is restricted within the range: −N/2 ≤ n < N/2. Throughout our manuscript we con- sider
α = −
β =
and Eq. 21 is modified according to this condition as, particular filling Ne, we take the sum of individual con- tributions from the lowest Ne energy eigenstates as we do our calculations at absolute zero temperature. The expression is, r cos2(ka). (23) I = In (25) In Fig. 20 we plot the energy levels as a function of flux φ, obtained from Eq. 23, for a 40-site binary alloy ring considering
= 1 and tr = 1. Looking at Fig. 20, two different sets of energy levels are noticed to form two quasi-bands separated by a finite energy gap. This gap, on the other hand, is tunable by the parameter values de- scribing the TB Hamiltonian Eq. 20. The origin of two different sets of energy levels is also clearly understood from Eq. 23. The energy levels of Fig. 20 have either a maximum or a minimum at half-integer or integer mul- tiples of flux-quantum and it results vanishing nature of n=1 Following this way, we plot the variation of persistent cur- rent for an ordered 120-site binary alloy ring in Fig. 21. The values of the parameters considered in this figure are FIG. 20: (Color online). Energy-flux characteristics of an ordered binary alloy ring (N = 40) considering
α = −
β = 1 and tr = 1. φ0 is set at 1. persistent current at these specific values of φ, since the current is obtained by taking the first order derivative of energy E(φ) with respect to flux φ. All these energy levels vary periodically providing φ0 flux-quantum peri- odicity, φ0 being 1 in our chosen unit (c = h = e = 1). C. Persistent current Our task of calculating persistent current for individual energy eigenstates is now easier as we know the energy eigenvalues of the ring as a function of flux φ. It is simply the first order derivative of energy with respect to flux φ. Therefore, for an n-th energy eigenstate we can write the expression for the current as, (cid:18) 4πt2 r N aφ0 (cid:19) (cid:113) sin(cid:2) 4π N a (n + φ/φ0)(cid:3) rcos2(cid:2) 2π N a (n + φ/φ0)(cid:3) 1 + 4t2 In = ± FIG. 21: (Color online). Current-flux characteristics of an ordered binary alloy ring (N = 120) considering
α = −
β = 1 and tr = 1. The blue and purple colors in (a) correspond to Ne = 15 and 35, while in (b) they correspond to Ne = 10 and 30, respectively. The lattice spacing a is set at 1 and we choose φ0 = 1.
= 1 and tr = 1. In both figures, Fig. 21(a) and (b), the current profiles show saw-tooth like variation as a func- tion of flux φ, similar to that of traditional single-channel mesoscopic rings85. The purple and blue colors corre- spond to two different fillings of the band. For odd num- ber of electrons the results are shown in Fig. 21(a) while, in Fig. 21(b) the results are given for even number of electrons. The sharp transitions at half-integer (for odd Ne) or integer (for even Ne) multiples of flux-quantum (φ0) in persistent current appears due to the crossing of energy levels at these respective values of φ. Quite inter- estingly, we also examine that the current shows always diamagnetic response irrespective of the filling factor. V. A BINARY ALLOY RING WITH EXTERNAL (24) ELECTRODES where, +ve or −ve sign appears in the current expression depending on the choice of n i.e., in which sub-band the energy level exists. To get total persistent current I for a Upto now we have discussed the basic features of per- sistent current in different isolated mesoscopic systems. In this section we extend our analysis of persistent cur- rent to an open system where a binary alloy ring83 is -2-1012-2.402.4ΦE
a
-1-0.500.51-.080.08ΦI
b
-1-0.500.51-.080.08ΦIclamped between two semi-infinite one-dimensional (1D) electrodes. The behavior of persistent current in presence of transport current will also be analyzed. A. The model Let us start by referring to Fig. 22 where a binary alloy ring, threaded by a magnetic flux φ, is attached to two semi-infinite one-dimensional metallic electrodes, namely, left-lead and right-lead, via two atomic sites la- beled as µ and ν. The total numbers of identical pairs of α-β sites in the binary ring is N1. Now we incorporate some foreign atoms denoted by γ having on-site poten- tials
γ in any one of the two arms of the ring. Here, we choose the upper arm in this purpose. There are N2 iden- tical foreign atomic sites, embedded together in a small (cid:88) (cid:16) m≥1 (cid:17) (cid:125) + t0 (cid:124) † b mbm+1 + h.c. (cid:123)(cid:122) 14 (cid:17) (cid:125) † mbm−1 + h.c. (cid:16) b (cid:88) m≤0 fashion as, H L = t0 (cid:124) and, (cid:123)(cid:122) (cid:16) left lead right lead (cid:17) † 0cµ + τRb † 1cν (27) + h.c. (28) H T = τLb † where b m (bm) are the creation (annihilation) operator of an electron at the site m of the leads and t0 represents the nearest-neighbor hopping strength within these leads. We set the site energy for the identical sites in the leads to zero and due to this reason in Eq. 27 we have omitted the site-energy terms. Here, τL is the coupling strength between the left lead and the ring, while it is τR for the other case. B. Wave-guide theory To find transmission probability across the ring and also to calculate persistent current in such an open ring geometry we adopt the wave-guide theory45,49. In the present sub-section we describe the formulation very briefly. We begin with the Scrodinger equation Hψ(cid:105) = Eψ(cid:105), where ψ(cid:105) is the stationary wave function of the entire system. In the Wannier basis it (ψ(cid:105)) can be expressed as, (cid:88) (cid:124) m≤0 (cid:123)(cid:122) (cid:88) (cid:124) m≥1 (cid:123)(cid:122) (cid:125) left lead right lead (cid:88) (cid:123)(cid:122) (cid:124) l Cll(cid:105) (cid:125) ring (cid:125) Bmm(cid:105) + Bmm(cid:105) + (29) FIG. 22: (Color online). A binary alloy ring in presence of some identical foreign atoms (labeled as γ sites), threaded by an AB flux φ, is attached to two semi-infinite one-dimensional metallic leads. µ and ν are the connecting sites. ψ(cid:105) = portion of the ring. These γ sites are often referred to as impurity sites in the present manuscript. We enumer- ate the atomic sites of the two side-attached leads in a particular way, as shown in Fig. 22. A single-band non- interacting TB framework is used to describe the entire system. For the full system we can partition the total Hamiltonian as a sum of three terms like, H = H R + H L + H T (26) and where, the co-efficients Bm and Cl correspond to the probability amplitudes in the respective sites. Keeping in mind the periodicity of the ordered binary ring we write the wave functions associated with the electrons as a plane wave and the wave amplitudes in the left and right leads are, Bm = eikm + re−ikm, for m ≤ 0 Bm = teikm, for m ≥ 1 (30) (31) where, H R, H L and H T represent the Hamiltonians for the ring, leads (left and right) and coupling between the ring and leads, respectively. The ring Hamiltonian (HR) takes the form exactly similar to Eq. 20, but instead of two possible on-site potentials here
l has three possibil- ities for three different atomic sites (α, β and γ). By means of some external gate voltage Vg, the site energy
γ can be tuned and accordingly the site-energies are changed. Thus we express the on-site energy of a sin- gle γ atom like,
γ =
0 γ is the site energy in absence of any external potential. The other two terms of Eq. 26, H L and H T , can also be written in a similar γ + Vg, where
0 where, r and t are the reflection and transmission ampli- tudes, respectively. k is the wave number and it is related to the energy E of the incident electron by the expression E = 2t0 cos k. The lattice spacing a is set equal to 1. In order to find out the transmission amplitude t, we have to solve the following set of coupled linear equations. EB0 = t0B−1 + τLCµ (E −
l)Cl = treiφCl+1 + tre−iφCl−1 + τLB0δl,µ + τRB1δl,ν EB1 = t0B2 + τRCν (32) γτRLtr0−1−21t023µνΦβαRight leadLeft leadτwhere, the co-efficients B0, B−1, B1 and B2 can be easily expressed in terms of r and t by using Eqs. 30 and 31, and they are in the form: B0 = 1 + r B−1 = B0eik − 2i sin k B1 = teik B2 = te2ik (33) Thus, for a particular value of E we can easily solve the set of linear equations and find the value of t. Finally, the transmission probability across the ring becomes T (E) = t2. (34) 15 behavior is elaborated in Figs. 23(b) and (c), where T - E and ρ-E spectra are superimposed with each other for two different numbers of impurity sites (N2). In the band center, several energy levels appear and these lev- els do not provide any contribution to the transmission of electrons. They are almost localized in nature. Mainly the states within the two bands are responsible for elec- tron transmissions. The number of the almost localized energy levels increases with the increase of the number of the impurity sites, and for sufficiently large number of impurities they form a quasi-energy band of localized Now, to compute the persistent current between any two neighboring sites in the binary ring we use the following relation, (cid:16) l Cl+1e−iφ(cid:17) † Il,l+1 = 2etr N Im C . (35) C. Transmission and average density of states Throughout our calculations we set
= 1,
0 γ = 0, tr = 1,
0 = 0 and t0 = 2. The energy scale is measured in unit of tr. Two-terminal transmission probability T (orange color) as a function of injecting electron energy E for some typical binary alloy rings considering different num- ber of impurity sites is displayed in Fig. 23 where (a), (b) and (c) correspond to three different numbers of impurity atoms. Figure (a) represents the transmission spectrum of the binary alloy ring with 28 pairs of α-β atoms but without any foreign impurity atoms while (b) and (c) correspond to N2 = 12 and 22, respectively, keeping N1 fixed to 28. The average density of states (ADOS) is also superimposed in each spectrum. In all these cases the magnetic flux φ is set equal to zero. The results are quite interesting. In absence of any impurity site the transmis- sion spectrum is characterized by two bands separated by a finite gap and the gap between these two bands are also tunable by the parameter values describing the sys- tem. Looking back at Fig. 20 the band splitting for the ordered binary alloy ring is easily understood and we pre- dict that the spectrum is just the finger print of the sys- tem energy levels. This transmission spectrum exactly overlaps with the ADOS profile (ρ-E spectrum) which ensures that electronic transmission takes place through all the energy eigenstates of the binary alloy ring and they are extended in nature. The situation becomes re- ally interesting when some additional impurities (γ sites) are introduced in any part of the binary alloy ring. Due to the inclusion of such atomic sites some energy levels appear within the band of extended regions those are no longer extended, but they are almost quasi-localized and do not contribute to the electronic transmission. This FIG. 23: (Color online). Transmission probability (orange color) and average density of states (dark-blue color) for some typical binary alloy rings with fixed number of α-β pairs (N1 = 28), but different values of impurity sites N2, where (a) N2 = 0, (b) N2 = 12 and (c) N2 = 22. For these three cases we set µ = 1 and choose ν = 29, 35 and 40, respectively. Other parameters are:
γ = 0, τL = τR = 1 and φ = 0. states. The location of the localized energy band can be shifted towards the edge of extended regions simply by tuning the site energy of these foreign atoms, and this can be done by means of applying an external gate voltage Vg. We utilize this feature to make the binary alloy ring behave like an extrinsic semiconductor, either p-type or n-type by tuning the Fermi level to an appropriate place. We plot Fig. 24 to explore the semiconductor-like be- havior of the binary ring. The variation of transmission probability and the ADOS are shown in this figure. Here, we consider a binary ring with 60 identical pairs of α-β
a
-4-202401.12.2ETΡ
b
-4-202401.12.2ETΡ
c
-4-202401.12.2ETΡsites and 44 number of impurity sites. Two different cases are exhibited in Fig. 24(a) and (b) for two different values of the on-site potential energies of the γ atoms. Figure (a) represents the transmission spectrum of the binary alloy ring when
γ is set at 0.6 while the other figure (Fig. 24(b)) corresponds to
γ = −0.6. In presence of the impurity atoms we get almost quasi-energy bands (ADOS spectra) and quite interestingly we observe that when
γ is fixed at 0.6, a localized energy band for a wide range of energy is formed along the left edge of the extended region (Fig. 24(a)). Now, if the Fermi level is set around E = −1.7, then many electrons in the localized region below the Fermi level can jump easily, even at much low temperature since the energy gap is almost zero, to the extended regions and can contribute to the current. As a 16 the current. But, the absence of electrons are realized by holes in the extended regions which can carry current and the system behaves like a p-type carriers. Before we end this discussion, we can emphasize that by setting the Fermi level in appropriate places our model quantum sys- tem can be tailor-made to use as a p-type or an n-type semiconductor. At the end of this sub-section we would like to mention a few points. To establish the fact that how such a geome- try can be utilized as a p-type or an n-type semiconductor with appropriate choice of the Fermi level we have taken a particular set of parameter values. As we are doing a model calculation it is very easy to take some specific values of the parameters and use in our numerical cal- culations, but all these physical phenomena are exactly invariant with the change of the parameter values. Only the numerical values will be altered. These features are also exactly valid even for a non-zero value of magnetic flux φ. The physical picture will be much more appealing FIG. 24: (Color online). Transmission probability (orange color) and ADOS (dark-blue color) as a function of energy for a binary alloy ring (N1 = 60) with 44 impurity sites (N2 = 44), where (a) and (b) correspond to
γ = 0.6 and −0.6, respectively. Other parameters are: φ = 0, τL = τR = 1.5, µ = 1 and ν = 83. result large number of excess electrons become available in the conduction band region which behave as n-type carriers. This movement however depends on the local- ized energy levels and also the available extended energy states. On the contrary, when the site-energies of the γ atoms are tuned to another value, say, −0.6 an exactly opposite behavior is obtained. In this case the wide band of localized states is formed in the right edge of the ex- tended region (see Fig. 24(b)). Now, if the Fermi level is tuned to appropriate place, around E = 1.7 in this case, then the electrons from the filled extended levels below the Fermi level hop to the nearly empty localized levels, and these electrons do not contribute anything to FIG. 25: (Color online). Persistent current (IU ) in the upper arm (red and violet color) as a function of energy E for a binary alloy ring (N1 = 20) in the absence of impurity atoms (N2 = 0), where (a) φ = 0 and (b) φ = φ0/4. Other parame- ters are: τL = τR = 0.8, µ = 1 and ν = 21. For this ordered ring, persistent current in the lower arm (IL) is exactly iden- tical to IU . ADOS profile (gray and light-blue color) is also superimposed in each spectrum. if we consider larger rings with more impurity sites and it gives us the confidence to propose an experiment in this way. In a recent work Bellucci et al.86,87 have done a de- tailed study of magneto-transport properties in quantum rings considering tunnel barriers in the presence of mag- netic field and shown how metal-to-insulator transition takes place in such a geometry. They have also estab- lished that by controlling the strength and the positions of the barriers, the energy shift can be done in a tunable way. This is quite analogous to our present study, and
a
-4-202401.12.2ETΡ
b
-4-202401.12.2ETΡ
a

a
-3-1.501.5300.20.4EIUΡ
b

b
-3-1.501.53-13013EIUΡso, an experiment in this regard will be challenging. D. Persistent current in presence of transport current Now we focus on the characteristic features of persis- tent current in the binary alloy ring in presence of exter- nal bias voltage. We also study the role of magnetic field in this context. In Fig. 25 upper arm current, IU , in the upper arm of an ordered binary alloy ring as a function of energy E is presented where (a) and (b) correspond to φ = 0 and φ0/4, respectively. Due to conservation of current at the junction points, current between any two sites are equal to each other. We refer upper arm current to the bond current between any two nearest-neighbor atomic sites of the upper arm. The ring is symmetrically coupled to the side attached leads i.e., the upper and lower arms have identical length, and accordingly, the current IL in the lower arm becomes exactly identical to 17 mind. Additionally, all the values of the current are pos- itive i.e. they are in the same phase ensuring the fact that this current is completely due the transport current between the two leads fixed at two different chemical po- tentials as there is only one driving force, the external bias. This phenomenon depicted in Fig. 25(a) correspond to zero magnetic flux. As well as the magnetic field with flux density φ is switched on, not only both the positive and negative values of current appear indicating opposite phases but also they appear in regular alternative fash- ion as shown in Fig. 25(b). Moreover, the magnitudes are also different in these two cases indicating the effect quantum interference of mesoscopic regime. Therefore, by measuring persistent current we can directly estimate the nature of the current i.e., whether it is paramagnetic FIG. 27: (Color online). Persistent current as a function of magnetic flux for a binary alloy ring (N1 = 16) in presence of impurity sites (N2 = 12) with Vg = 0.5 Other parameters are: τL = τR = 0.8, µ = 1 and ν = 23. or diamagnetic in nature and also predict the character- istics of energy eigenstates, which are somewhat interest- ing in the study of electron transport. Needless to say, magnetic field plays an important role in this context. Already stated in introduction the current can persists in this kind of loop geometry if we set the lengths of two arms unequal or put any kind of static disorder in path of the electron or in any other way we can disturb the wave function or create any path difference between the wave functions corresponding to different arms. In our study we incorporate impurities in the upper arm of the binary ring as mentioned earlier. The results are dis- played in Fig. 26 where (a) and (b) represent the cases corresponding to the upper and lower arms of the ring, respectively. To plot the curve we consider a typical bi- nary alloy ring with N1 = 16 and N2 = 12. With the inclusion of impurity sites, the symmetry between the two arms is broken, and therefore, the currents in the upper and lower arms are no longer identical to each other as shown from the spectra (Figs. 26(a) and (b)). So, there must be a circulating current within the ring which is the so-called persistent current in presence of transport current. Unlike to the ordered binary alloy ring (Fig. 25), here all the energy eigenstates are not ex- tended in nature. Some localized energy levels appear in the band of extended energy states due to the presence FIG. 26: (Color online). Persistent current as a function of energy for a binary alloy ring (N1 = 16) in presence of impu- rity sites (N2 = 12) for a finite value of φ (φ = 0.3), where (a) and (b) correspond to the currents for the upper (pink color) and lower (violet color) arms respectively. Other parameters are:
γ = 0.5, τL = τR = 0.8, µ = 1 and ν = 23. ADOS profile (green and light-blue color) is superimposed in each spectrum. the current obtained in the upper arm. Similarly, we de- fine the lower arm current. Here, again the band-splitting put its mark on the current density profile. Finite current is available for two wide range of energies, separated by a finite gap, associated with the energy levels of the ring those are clearly visible from the ADOS profiles (green and light-blue color). The quite exact superposition of the current profile and the AVDOS profile invokes the extended natures of all these energy eigenstates in our -3-1.501.53-10010EIUΡ-3-1.501.53-10010EILΡ-1-0.500.51-.040.04ΦIof impurity sites in the ring. This is clearly visible from the spectra since for these impurity levels no current is available. We also check the periodicity of the persis- tent current. It is shown in Fig. 27. Persistent current exhibits φ0 flux-quantum periodicity. Thus, calculating persistent current we can emphasize the nature of energy eigenstates very nicely, and this idea can be utilized to reveal the localization properties of energy eigenstates in any complicated geometry. VI. CONCLUDING REMARKS To conclude, in the present article we have made a de- tailed investigation of magneto-transport in both closed and open mesoscopic systems. We have started with the discussion of a zigzag nanotube pierced by a mag- netic flux using a generalized Hartree-Fock mean field ap- proach. Based on the tight-binding model we explore the effect of the Hubbard interaction on the energy levels of the tube. After describing different numerical results we have established the second quantized form to evaluate persistent current in individual paths of a zigzag carbon nanotube and based on this formulation we have also pre- sented the numerical results for the distribution of persis- tent current in different branches of the nanotube. From the current-flux characteristics we have emphasized that the current amplitude in the zigzag nanotube is highly sensitive to the electron filling and this phenomenon can be utilized in designing a high conducting to a low con- ducting switching device and vice versa. Following the 18 study of nanotube we have explored the persistent cur- rent in an ordered-disordered separated cylinder pene- trated by a magnetic flux. Most interestingly, we have seen that the current amplitude shows an anomalous be- havior with the increase of impurity strength. This study may be helpful for exploring localization-delocalization transition in shell-doped nanotubes. To discuss per- sistent current in open system i.e., system with side- attached electrodes we have considered a binary alloy ring in presence of a magnetic flux φ, as an illustrative ex- ample. Within a single-band non-interacting TB frame- work persistent current and band structure of an isolated ordered binary alloy ring have been analyzed. Then, we have explored the magneto-transport properties of a bi- nary alloy ring in presence of external electrodes. The effect of impurities have also been addressed. Quite in- terestingly we have noticed that in the presence of some foreign atoms, those are not necessarily be random, in any part of the ring, some quasi-localized energy levels appear within the band of extended energy levels. The lo- cations of these almost localized energy levels can also be regulated by means of some external gate voltage. This leads to a possibility of using such a system as a p-type or an n-type semiconductor by fixing the Fermi level in appropriate places. Before we end, we would like to mention that in the present review we have addressed some important aspects of quantum transport through some low-dimensional model quantum systems. Several other important quan- tum phenomena in such meso- and nano-scale systems have also been reported in recent reviews88 -- 93. ∗ Electronic address: [email protected] † Electronic address: [email protected] 1 S. Datta, Electronic transport in mesoscopic systems, Cam- bridge University Press, Cambridge (1997). 2 P. Singha Deo and A. M. Jayannavar, Phys. Rev. B 50, 11629 (1994). 3 Quantum coherence in mesoscopic systems, NATO Ad- vanced Study Instute Series B: Physics 254, ed. by B. Kramer, Plenum, New York (1991). 4 Mesoscopic phenomenon in solids, ed. by B. L. Altshular, P. A. Lee, and R. A. Webb, Noth-Holland (1991). 5 C. W. J. Beenakker and H. van Houten, Solid state physics: Semiconductor heterostructures and nanostructures 44, ed. by H. Ehrenreich and D. Turnbull, Academic Press (1991). 6 S. Washburn and R. A. Webb, Adv. Phys. 35, 375 (1986). 7 N. Kumar and A. M. Jayannavar, Phys. Rev. B 32, 3345 (1985). 13 O. Hod, E. Rabani, and R Baer, Acc. Chem. Res. 39, 109 (2006). 14 F. Nichele, Y. Komijani, S. Hennel, C. Gerl, W. Wegschei- der, D. Reuter, A. D. Wieck, T. Ihn, and K. Ensslin, New J. Phys. 15, 033029 (2013). 15 N. Bayers and C. N. Yang, Phys. Rev. Lett. 7, 46 (1961). 16 M. Buttiker, Y. Imry, and R. Landauer, Phys. Lett. A 96, 365 (1983). 17 L. P. Levy, G. Dolan, J. Dunsmuir, and H. Bouchiat, Phys. Rev. Lett. 64, 2074 (1990). 18 R. Deblock, R. Bel, B. Reulet, H. Bouchiat, and D. Mailly, Phys. Rev. Lett. 89, 206803 (2002). 19 B. Reulet, M. Ramin, H. Bouchiat, and D. Mailly, Phys. Rev. Lett. 75, 124 (1995). 20 H. F. Cheung, Y. Gefen, E. K. Reidel, and W. H. Shih, Phys. Rev. B 37, 6050 (1988). 21 V. Ambegaokar and U. Eckern, Phys. Rev. Lett. 65, 381 8 P. Singha Deo and A. M. Jayannavar, Mod. Phys. Lett. B (1990). 7, 1045 (1993). 9 P. W. Anderson, Phys. Rev. 109, 1492 (1958). 10 K. v. Klitzing, G. Dorda, and M. Pepper, Phys. Rev. Lett. 45, 494 (1980). 11 Y. Aharonov and D. Bohm, Phys. Rev. 115, 485 (1959). 12 A. van Oudenaarden, M. H. Devoret, Y. V. Nazarov, and 22 A. Schmid, Phys. Rev. Lett. 66, 80 (1991). 23 U. Eckern and A. Schmid, Europhys. Lett. 18, 457 (1992). 24 S. K. Maiti, Solid State Commun. 150, 2212 (2010). 25 S. K. Maiti, M. Dey, S. Sil, A. Chakrabarti, and S. N. Karmakar, Europhys. Lett. 95, 57008 (2011). 26 S. K. Maiti, J. Chowdhury, and S. N. Karmakar, Phys. J. E. Mooij, Nature 391, 768 (1998). Lett. A 332, 497 (2004). 19 27 S. K. Maiti, J. Chowdhury, and S. N. Karmakar, Solid State Commun. 135, 278 (2005). 28 S. K. Maiti, J. Appl. Phys. 110, 064306 (2011). 29 S. K. Maiti, Phys. Status Solidi B 248, 1933 (2011). 30 D. Mailly, C. Chapelier, and A. Benoit, Phys. Rev. Lett. 70, 2020 (1993). 59 K. S. Novoselov, A. K. Geim, S. V. Morozov, D. Jiang, M. I. Katsnelson, I. V. Griorieva, S. V. Dubonos, and A. A. Firsov, Nature 438, 197 (2007). 60 K. Sasaki, S. Murakami, R. Saito, and Y. Kawazoe, Phys. Rev. B 71, 195401 (2005). 61 R. B. Chen, B. J. Lu, C. C. Tsai, C. P. Chang, F. L. Shyu, 31 E. M. Q. Jariwala, P. Mohanty, M. B. Ketchen, and R. A. and M. F. Lin, Carbon 42, 2873 (2004). Webb, Phys. Rev. Lett. 86, 1594 (2001). 62 M. Szopa, M. Marganska, and E. Zipper, Phys. Lett. A 32 R. Deblock, R. Bel, B. Reulet, H. Bouchiat, and D. Mailly, 299, 593 (2002). Phys. Rev. Lett. 89, 206803 (2002). 33 H. Bluhm, N. C. Koshnick, J. A. Bert, M. E. Huber, and K. A. Moler, Phys. Rev. Lett. 102, 136802 (2009). 34 W. Rabaud, L. Saminadayar, D. Mailly, K. Hasselbach, A. Benoit, and B. Etienne, Phys. Rev. Lett. 86, 3124 (2001). 35 V. Chandrasekhar, R. A. Webb, M. J. Brady, M. B. Ketchen, W. J. Gallagher, and A. Kleinsasser, Phys. Rev. Lett. 67, 3578 (1991). 36 M. Abraham and R. Berkovits, Phys. Rev. Lett. 70, 1509 (1993). 63 S. K. Maiti, Physica E 31, 117 (2006). 64 S. K. Maiti, Solid State Phenom. 155, 87 (2009). 65 S. Bellucci and P. Onorato, Physica E 41, 1393 (2009). 66 P. A. Orellana and M. Pacheco, Phys. Rev. B 71, 235330 (2005). 67 L. K. Castelano, G. -Q. Hai, B. Partoens, and F. M. Peeters, Phys. Rev. B 78, 195315 (2008). 68 P. Dutta, S. K. Maiti, and S. N. Karmakar, Eur. Phys. J. B 85, 126 (2012). 69 M. P. L. Sancho, M. C. Munoz, and L. Chico, Phys. Rev. 37 G. Bouzerar, D. Poilblanc, and G. Montambaux, Phys. B 63, 165419 (2001). Rev. B 49, 8258 (1994). 38 T. Giamarchi and B. S. Shastry, Phys. Rev. B 51, 10915 (1995). 39 S. K. Maiti, J. Chowdhury, and S. N. Karmakar, J. Phys.: Condens. Matter 18, 5349 (2006). 70 H. Lin, J. Lagoute, V. Repain, C. Chacon, Y. Girard, J. -S. Lauret, F. Ducastelle, A. Looiseau, and S. Rousset, Nature Mater. 9, 235 (2010). 71 S. Sorella and E. Tosatti, Europhys. Lett. 19, 699 (1992). 72 R. Heyd, A. Charlier, and E. McRae, Phys. Rev. B 55, 40 H. F. Cheung, Y. Gefen, E. K. Reidel, and W. H. Shih, 6820 (1997). Phys. Rev. B 37, 6050 (1988). 41 O. Entin-Wohlman and Y. Gefen, Europhys. Lett. 8, 477 73 R. E. Peierls, Z. Phys. 80, 763 (1933). 74 K. Wakabayashi, M. Fujita, H. Ajiki, and M. Sigrist, Phys. (1989). 42 H. Bary-Soroker, O. Entin-Wohlman, and Y. Imry, Phys. Rev. B 82, 144202 (2010). 43 M. Buttiker, Phys. Rev. B 32, R1846 (1985). 44 E. Akkermans, A. Auerbach, J. E. Avron, and B. Shapiro, Phys. Rev. Lett. 66, 76 (1991). 45 J. -B. Xia, Phys. Rev. B 45, 3593 (1992). 46 P. A. Mello, Phys. Rev. B 47, 16358 (1993). 47 P. A. Orellana, M. L. Ladr´on de Guevra, M. Pacheco, and A. Latg´e, Phys. Rev. B 68, 195321 (2003). Rev. B 59, 8271 (1999). 75 L. Brey and H. A. Fertig, Phys. Rev. B 73, 235411 (2006). 76 K. Wakabayashi and K. Harigaya, J. Phys. Soc. Jpn. 72, 998 (2003). 77 J. Fern´andez-Rossier, Phys. Rev. B 77, 075430 (2008). 78 Y. Son, M. L. Cohen, and S. G. Louie, Nature (London) 444, 347 (2006). 79 S. K. Maiti, S. Saha, and S. N. Karmakar, Eur. Phys. J. B 79, 209 (2011). 80 S. Saha, S. K. Maiti, and S. N. Karmakar, Eur. Phys. J. B 48 A. M. Jayannavar and P. Singha Deo, Phys. Rev B 49, 85, 283 (2012). 13685 (1994). 49 Y. -J. Xiong and X. -T. Liang, Phys. Lett. A 330, 307 81 J. Zhong and G. M. Stocks, Nano Lett. 6, 128 (2006). 82 Y. Cui, X. F. Duan, J. T. Hu, C. M. Lieber, J. Phys. Chem. (2004). B 104, 5213 (2000). 50 T. P. Pareek, P. Singha Deo, and A. M. Jayannavar, Phys. 83 P. Dutta, S. K. Maiti, and S. N. Karmakar, AIP Conf. Rev. B 52, 14657 (1995). 51 S. Bellucci and P. Onorato, Physica E 41, 1393 (2009). 52 P. Singha Deo and A. M. Jayannavar, Phys. Rev. B 51, 10175 (1995). 53 A. M. Jayannavar, P. Singha Deo, and T. P. Pareek, Phys- ica B 212, 261 (1995). 54 P. Singha Deo and A. M. Jayannavar, Mod. Phys. Lett. B 7, 1045 (1993). 55 P. Dutta, S. K. Maiti, and S. N. Karmakar, arXiv:1307.2103v2. 56 O. O. Kit, T. Talliner, L. Mahadevan, J. Timonen, and P. Koskinen, Phys. Rev. B 85, 085428 (2012) 57 A. K. Geim and K. S. Novoselov, Nat. Mater. 6, 183 (2007). 58 A. H. C. Neto, F. Guinea, N. M. R. Peres, K. S. Novoselov, and A. K. Geim, Rev. Mod. Phys. 81, 109 (2009). Proc. 1536, 581 (2013). 84 P. Dutta, S. K. Maiti, and S. N. Karmakar, Phys. Lett. A 367, 1567 (2012). 85 S. K. Maiti, Nanotechnol. Rev. 1, 255 (2012). 86 S. Bellucci and P. Onorato, Physica E 41, 1393 (2009). 87 S. Bellucci and P. Onorato, Eur. Phys. J. B 73, 215 (2010). 88 E. L. Pankratov and E. A. Bulaeva, Rev. Theor. Sci. 1, 58 (2013). 89 M. Lipkind, Rev. Theor. Sci. 1, 145 (2013). 90 S. Celasun, Rev. Theor. Sci. 1, 319 (2013). 91 M. Ibrahim and H. Elhaes, Rev. Theor. Sci. 1, 368 (2013). 92 N. A. Zimbovskaya, Rev. Theor. Sci. 2, 1 (2014). 93 P. Di Sia, Rev. Theor. Sci. 2, 146 (2014). 

1608.02560

2

1608

Note on "Quantum superconducting criticality in graphene and topological insulators"

We correct our previous conclusion regarding the fate of a charged quantum critical point across the superconducting transition for two dimensional massless Dirac fermion. Within the leading order $\epsilon$ expansion, we now find that the requisite number of four-component Dirac fermion flavors ($N_f$) for the continuous phase transition through a charged critical point is $N_f>18.2699$. For $N_f\geq1/2$, the critical number of bosonic flavors for this transition is significantly reduced as compared to the value determined in the absence of the Dirac fermions in the theory.

cond-mat.mes-hall

cond-mat

Note on "Quantum superconducting criticality in graphene and topological insulators" Bitan Roy,1 Vladimir Jurici´c,2 and Igor F. Herbut3 1Condensed Matter Theory Center and Joint Quantum Institute, University of Maryland, College Park, Maryland 20742-4111, USA 2Nordita, Center for Quantum Materials, KTH Royal Institute of Technology and Stockholm University, Roslagstullsbacken 23, 10691 Stockholm, Sweden 3 Department of Physics, Simon Fraser University, Burnaby, British Columbia, Canada V5A 1S6 We correct our previous conclusion regarding the fate of a charged quantum critical point across the superconducting transition for two dimensional massless Dirac fermion. Within the leading order
expansion, we now find that the requisite number of four-component Dirac fermion flavors (Nf ) for the continuous phase transition through a charged critical point is Nf > 18.2699. For Nf ≥ 1/2, the critical number of bosonic flavors for this transition is significantly reduced as compared to the value determined in the absence of the Dirac fermions in the theory. We here take the opportunity to correct an error in the expression from Eq. (18) of the "Supplementary Materials" (SM) of Ref. [1]. The correct expression should read as (cid:90) (cid:18) (cid:19) (4a) = −(−g)(ie)2 ddq (2π)d 1 q2 δµν − qµqν q2 /k − /q (k − q)2 P+ (cid:48) − /q /k + /k (k + k(cid:48) − q)2 γ5γν, γ5γµ (1) where /k = kµγµ. The evaluation of this expression has already been shown in the SM (notice the extra minus sign in the above expression) of the original article, which then leads to a slightly modified renormalization condition for the Yukawa coupling in Eq. (10) of the main part of the paper ZΨZ 1/2 Φ g0µ−
/2 − 3e2g 1
= g. (2) Notice that instead of the expression as given in the paper, the above equation contains a negative sign in the term ∼ e2g. The renormalization factors ZΨ and ZΦ are given by Eq. (8) in the original paper. As a result, the correct form of the β-function for the Yukawa coupling is βg2 = −
g2 + (Nf + 1)g4 − 6e2g2, (3) which differs in the value of the coefficient of the last term as compared to Eq. (14) of the original article and is in agreement with Ref. [2] when we set Nf = 1/2 (as the authors of Ref. [2] addressed superconducting criticality for the surface states of topological insulators). Rest of the flow equations remains unchanged and together they support a pair of charged fixed points located at (cid:32) (cid:0)e2∗, g2∗, λ± ∗(cid:1) = ∆1 ±(cid:112)∆2 2XY W 1 + ∆2 (cid:33) , 3 4Y , 9 + 2Y 2XY , (4) (5) analogous to Eq. (19) of the original paper, where X = Nf + 1, Y = Nf + Nb, W = Nb + 4 and ∆1 = XY + 18X − Nf (9 + 2Y ), ∆2 = −4W(cid:2)54X 2 − 2Nf (4.5 + Y )2(cid:3) . We thus obtain a different function determining the critical number of the bosonic flavors (N c Note that expressions for ∆1 and ∆2 is now slightly modified due to the change in renormalization condition in Eq. (2). b ) as a function of the number of four-component fermionic flavors (Nf ) for the second order quantum phase transition through a charged fixed point, as shown in Fig. 1. In particular, the quantum-critical point in the supersymmetric theory with Nf = 1/2 and Nb = 1 becomes unstable when both fermions and bosons are coupled to the gauge field. The transition then turns into the first order one. However, for sufficiently large number of fermionic flavors the transition through the charged fixed point is continuous for arbitrary number of bosonic flavors (Nb). Due to this modification in Eq. (2) the flow equation for the Ginzburg-Landau parameter κ2 = λ/(2e2) from b for a Eq. (20) of the original article is changed accordingly. The flow equation of κ2 gives κ2 given Nf . + > 0 when Nb > N c This result does not affect any other conclusions of the paper. However, few comments are due and they are stated below. 6 1 0 2 p e S 4 1 ] l l a h - s e m . t a m - d n o c [ 2 v 0 6 5 2 0 . 8 0 6 1 : v i X r a 2 FIG. 1: Critical number of bosonic flavors (N c b ) for the continuous quantum phase transition through a charged fixed point as a function of the number of four-component fermionic flavors (Nf ). For Nb > N c b the transition is continuous, while otherwise it is of the first order in nature (see the blue curve). When Nf > 18.2699, the quantum phase transition through the charged fixed point is continuous for any number of bosonic flavors. 1. Notice from Fig. 1 that upon incorporating the coupling between fluctuating gauge fields and massless Dirac fermions, the critical number of bosonic flavors (N c b ) for continuous phase transition through a charged critical point reduces drastically even for sufficiently small number of fermionic flavors Nf . In particular, when Nf ≥ 0.5 b ∼ 183 determined in the absence the critical number of bosonic flavors drops to a value of N c of massless Dirac fermions in the theory [3]. Thus coupling of Dirac fermions with the gauge field increases the propensity of a continuous phase transition through a charged critical point. b (cid:39) 4 − 5 from N c fermionic flavor number for which N c b = 1. From the leading order
-expansion we find Nf ≈ 16.55 for N c 2. Since only one bosonic field couples with the fluctuating gauge field, it is worth determining the requisite b = 1. 3. Although Nf ≈ 16.55 seems too large to realize the existence of a charged critical point in any physical systems, it should be noted that upon incorporating next to leading order corrections (∼
2, for example) in the flow equations the above mentioned value of Nf can change quite dramatically. Such analysis for our problem is not presently available. Nevertheless, it is worth pointing out that N c b becomes negative (suggesting continuous phase transition through charged critical point for any value of Nb) even in the absence of massless Dirac fermions, when corrections ∼
2 are taken into account and
is set to be unity (for d = 3) at the end of the anaysis [3]. Thus it remains as an open and interesting question to investigate the variation of critical fermion number for which a charged critical point can be accessed even for Nb = 1, once the higher order corrections are taken into account. [1] B. Roy, V. Jurici´c, I. F. Herbut, Phys. Rev. B 87, 041401(R) (2013). [2] N. Zerf, C. H. Lin, and J. Maciejko, arXiv: 1605.09423. [3] I. F. Herbut, A Modern Approach to Critical Phenomena (Cambridge University Press, Cambridge, 2007). 0510152004812NfNbc

1303.2258

1

1303

Thermal properties of fluorinated graphene

Large scale atomistic simulations using the reactive force field approach (ReaxFF) are implemented to investigate the thermomechanical properties of fluorinated graphene (FG). A new set of parameters for the reactive force field potential (ReaxFF) optimized to reproduce key quantum mechanical properties of relevant carbon-fluor cluster systems are presented. Molecular dynamics (MD) simulations are used to investigate the thermal rippling behavior of FG and its mechanical properties and compare them with graphene (GE), graphane (GA) and a sheet of BN. The mean square value of the height fluctuations $< h^2>$ and the height-height correlation function $H(q)$ for different system sizes and temperatures show that FG is an un-rippled system in contrast to the thermal rippling behavior of graphene (GE). The effective Young's modulus of a flake of fluorinated graphene is obtained to be 273 N/m and 250 N/m for a flake of FG under uniaxial strain along arm-chair and zig-zag direction, respectively.

cond-mat.mes-hall

cond-mat

Thermal properties of fluorinated graphene S. K. Singh1, S. Goverapet Srinivasan2, M. Neek-Amal1, S. Costamagna1,3, Adri C. T. van Duin2, and F. M. Peeters1 1Universiteit Antwerpen, Department of Physics, Groenenborgerlaan 171, BE-2020 Antwerpen, Belgium. 2 Department of Mechanical and Nuclear Engineering, The Pennsylvania State University, University Park, Pennsylvania 16801, USA. 3Facultad de Ciencias Exactas Ingenier´ıa y Agrimensura, Universidad Nacional de Rosario and Instituto de F´ısica Rosario, Bv. 27 de Febrero 210 bis, 2000 Rosario, Argentina. (Dated: October 29, 2018) Large scale atomistic simulations using the reactive force field approach (ReaxFF) are imple- mented to investigate the thermomechanical properties of fluorinated graphene (FG). A new set of parameters for the reactive force field potential (ReaxFF) optimized to reproduce key quantum mechanical properties of relevant carbon-fluor cluster systems are presented. Molecular dynamics (MD) simulations are used to investigate the thermal rippling behavior of FG and its mechanical properties and compare them with graphene (GE), graphane (GA) and a sheet of BN. The mean square value of the height fluctuations hh2i and the height-height correlation function H(q) for different system sizes and temperatures show that FG is an un-rippled system in contrast to the thermal rippling behavior of graphene (GE). The effective Young's modulus of a flake of fluorinated graphene is obtained to be 273 N/m and 250 N/m for a flake of FG under uniaxial strain along arm-chair and zig-zag direction, respectively. I. INTRODUCTION The fascinating properties of single layer graphene (GE) have triggered a broad interest in the solid state physics community [1 -- 5]. Despite its high electron mo- bility [6], the zero band gap defies its employment in nano transistors where it is desirable to have a large on-off ra- tio between conducting and non-conducting states. A band gap can be induced by the addition of adatoms, which changes locally the hybridization of the carbon (C) atoms, but also modifies the electron mean free-path affecting the electron transport properties. Hydrogen (H) and fluor (F) are two well tested candidates [7 -- 10] that leads to a large band gap opening. Graphane (GA, hydrogenated graphene) and fluorographene (FG) have been studied both experimentally and theoretically [11 -- 14] to engineer the band gap. When H or F atoms are attached to the C atoms of GE, the bonds transit from an sp2 to an sp3 hybridiza- tion, which turns the conjugated graphitic C-C bonds into single C-C bonds. In the fully covered cases both GA and FG are insulating materials [7, 8] and the structure changes locally the planar shape of GE into an angstrom scale out-of-plane buckled shaped membrane [15] known as chair configuration [16, 17]. From its potential applications in nanotechnology, FG is a more favourable material than GA since the C- F bonds are energetically more stable than the C-H bonds [13, 15, 17, 18]. Fluorographene has a very large temperature-dependent resistance and when the fluor content is increased it induces large changes in the elec- tron transport [19]. As in GE, it is expected that temper- ature also affects strongly the lattice structure and the mechanical properties of FG. According to the Mermin-Wagner theorem, [20], ther- mal excited ripples in two dimensional-like materials (GE, bilayer GE, GA and FG) has to play an impor- tant role in the stability of the membrane. While in GE and bilayer GE the corrugations are well described within the theory of two dimensional continuous mem- branes [21, 22], for GA instead, we recently found that the angstrom scale buckling of the carbon layer of GA prevents the formation of intrinsic long wavelength ther- mal ripples for temperatures up to at least 900 K [23]. Since the C atom has a higher (lower) electronegativity than H (F), it will take (give) away charge from the H (F) atom and consequently transforms the resulting C-H and C-F covalent bonds into sp3 bonds. Therefore, it is expected that similar rippling effects as in GA will occur in FG although the C-F bonds are somewhat stronger than the C-H bonds. The latter is due to the larger amount of charge that is shifted from C to F as com- pared to the one from H to C [13]. However in order to simulate large scale FG samples an appropriate force field is needed which describes the true chemical bond in C-F. Indeed, the absent of such a suitable interatomic potential for C-F restricted most of the recent studies to ab-initio calculations of their electronic properties using a small computational unit cell. ReaxFF potential serves to describe both bond and non-bond interactions in solids. Recently, such poten- tials were parameterized and were well tested for dif- ferent kind of structures, e.g. hydrocarbons [24], car- bon allotropes [25], etc. In this study we present a new set of parameters for ReaxFF, appropriate for structures with C-F bonds. Using molecular dynamics (MD) sim- ulations over large scale samples we study the thermal corrugations of FG and compare the results with those found for GA, GE and BN. We show that fully covered FG follows the same trend as GA and does not develop long-wavelength ripples or significant corrugation. The bending rigidity κ of FG is found to be larger than the one of GE, GA and BN. Furthermore, κ turns out to be temperature independent. Our results indicate that long- wavelength ripples are instead present in partial covered FG samples with a larger amplitude as compared to GA. The paper is organized as follows. In Sec. II, we intro- duce a new set of parameters for the ReaxFF potential of the C-F covalent bond. Then, in Sec. III using the introduced parameters, we analyze the thermal rippling behavior. Here, we consider both fully and partially cov- ered graphene sheets by F atoms. All the results are compared with those previously found for graphane. We also estimate the effective Young's modulus of FG flakes. We conclude the paper in Sec. IV. II. REAXFF POTENTIAL FOR FLUOROGRAPHENE ReaxFF [24] is a general bond-order dependent poten- tial that uses a relationship between bond distance and bond order on the one hand and a relationship between bond order and bond energy on the other hand to de- scribe bond formation and dissociation. Many body in- teractions such as the valence angle and torsional inter- actions are formulated as function of bond order so that their energy contributions vanish smoothly upon bond dissociation. Non bonded interactions, namely Coulomb and van der Waals interactions, are calculated between every pair of atoms irrespective of their connectivity. Ex- cessively close range interactions are avoided by shield- ing. ReaxFF uses the geometry dependent charge calcu- lation scheme (EEM scheme) of Mortier et al [26]. The system energy in ReaxFF consists of a sum of terms: Esys = Ebond + Eunder + Eover + Elp + Eval + Epen+ Etors + Econj + EvdW aals + ECoulomb. A detailed description of each of these terms and their functional forms can be found in the original work [24]. The reactive force field for C/F contain- ing systems was developed by parameterizing the poten- tial against DFT data obtained at the B3LYP/6-31g** level of theory (which is implemented in Schrodinger [27] which is an electronic structure packages) for various quantities such as fluorine and carbon atom charges in H3C−CF2−CH3, C−F and C−C bond lengths in H3C−CF2−CH3 and H3C−CF(CH3)−CH3, F−F bond length in the F2 molecule, potential energy curve for C-F bond dissociation in H3C−CF2−CH3, F-C-F an- gle bending in H3C−CF2−CH3, C-C-F angle bending in H3C−CF2−CH3 and F-C-C-F dihedral twisting in F2C=CF2 along with various chemical reactions involv- ing fluoroalkanes and fluoroalkenes. The results of the force field training are presented in Figs. 1(a)-(d) and in Table II. Fig. 2 depicts the geometrical quantities rele- vant to Figs. 1(a)-(d). It can be seen from Table II that ReaxFF reproduces closely the DFT based equilibrium ReaxFF DFT (a) 200 150 100 ) l o m / l a c k ( R E 50 2 (b) 80 60 40 20 ) l o m / l a c k ( R E 0 1 80 60 40 20 0 80 ) l o m / l a c k ( R E 3 2 4 C-F distance (Å) 5 0 80 100 140 C-C-F angle (deg.) 120 160 (c) ) l o m / l a c k ( R E 100 140 F-C-F angle (deg.) 120 160 50 40 30 20 10 0 0 (d) 50 100 F-C-C-F torsion angle (deg.) 150 200 FIG. 1: (Color online) Comparison between DFT (solid squares) and ReaxFF (open circles) results for: (a) C-F bond dissociation in H3C-CF2-CH3 (b) C-C-F angle bending in H3C-CF(CH3)-CH3, (c) F-C-F angle bending in H3C-CF2- CH3 ,and (d) F-C-C-F dihedral twisting in F2C=CF2. geometries for various compounds. ReaxFF predicts F2 dissociation energy of 36.6 kcal/mol, in very good agree- ment with the DFT value of 37 kcal/mol. Fig. 1(a) shows that ReaxFF based potential energy curve for the C-F bond dissociation in H3C−CF2−CH3 closely follows the DFT based potential energy curve. ReaxFF is able to predict very precisely the equilibrium C-F bond length (see Table II) and the C-F bond dissociation energy. Sim- ilarly the force field can closely reproduce the DFT based potential energy curve and the equilibrium geometry (see Table II) for C-C-F angle bending and the C-F-C angle bending as shown in Figs. 1(b)-(c). Figure 1(d) shows the variation of the potential energy upon F-C-C-F dihe- dral angle twisting. Though ReaxFF predicts the correct trend, the torsional rotation barrier in ReaxFF is around 18 kcal/mol lower than that predicted by DFT. Over- all, the ReaxFF force field for C/F systems can closely reproduce the DFT based energies and geometries for a number of molecules and reactions. This force field will now be employed in large scale fully reactive molecular dynamics simulation of C/F containing systems. In the next section we study the thermal structural fluctuations and mechanical properties of a single layer of FG using large scale atomistic simulations employing the presented ReaxFF parameters. These parameters were implemented in the large-scale atomic/molecular mas- sively parallel simulator package LAMMPS [28, 29]. TABLE I: Comparison of equilibrium geometrical parameters between ReaxFF and DFT. DFT ReaxFF F2 bond length C-F bond length in H3C-CF2-CH3 1.43A 1.4012A 1.3841A 1.4057A C-F bond length in H3C-CF(CH3)-CH3 1.3841A 1.4158A Non-bonding C-F distance in CF2 dimer 2.00 A 2.4471A 105.65o 107.2197 o 106.2 o 109.9625 o C-C-F angle in H3C-CF(CH3)-CH3 F-C-F angle in H3C-CF2-CH3 3 (a) (c) (b) (d) FIG. 2: (Color online) (Color online) The molecules used for parameterizing the ReaxFF force field in this study. (a) C-F bond in H3C-CF2-CH3 (b) F-C-F angle in H3C-CF2-CH3 (c) C-C-F angle in H3C-CF(CH3)-CH3, and (d) F-C-C-F dihedral angle in F2C=CF2. The atoms constituting the geometric parameters are represented by balls while the rest of the atoms are represented by sticks. F atoms are colored brown, C atoms are colored green and H atoms are colored blue. III. RESULTS A. Thermal rippling behavior of FG In order to study the rippling behavior of FG we con- sidered a square shaped computational unit cell of FG with both armchair and zigzag edges in the x and y di- rections. Partial fluor contents of 10%, 50%, 70%, 90% and the fully covered 100% case (with a total number of N = 17280 atoms) were studied. In our simulation we employed the NPT ensemble with P =0 using the Nos´e- Hoover thermostat and varied the temperature from 10 K to 900 K. Figure 3 shows the obtained buckled shape of fully fluorinated sample after relaxation which is in agree- ment with recent DFT results [13]. One would expect that the thermal excited ripples in FG can be described by membrane theory for a 2D con- tinuous membrane [30]. This theory, described in a series of related works [23, 31 -- 33], is supposed to be universal and independent of the atomic scale details of the mem- brane. The main predictions of this theory are as fol- FIG. 3: (Color online) (a) Side view of the buckled structure, known as chair configuration, of fully fluorinated graphene. The averaged bond-angle, and C-C and C-F distances, are respectively 109.5◦, dC−C =1.58 A and dC−F =1.41 A at room temperature. lows. Let h be the out-of plane displacement of a given atom of a sheet, then the Fourier transform of the height- height correlation function is in the harmonic approxima- tion given by H(q) = hh(q)2i = N kBT κS0q4 , (1) where q is the wave-vector, N is the number of atoms, S0 is the surface area per atom, κ is the bending rigidity of the membrane, and kB is the Boltzmann constant. In the large wavelength limit, anharmonic couplings between bending and stretching modes are important re- sulting in a renormalization of the q-dependent behavior H(q) = N kBT κS0q4−η (2) where η is an universal scaling exponent which is about ≈ 0.85 [34 -- 36]. In order to compare our results for FG with other two dimensional materials, we used a modified Tersoff potential (which is an ordinary defined potential in the LAMMPS package [29]) according to the parameters pro- posed by Sevik et al for the h-BN sheet [37]. To simulate GE and GA we have used the AIREBO potential [38] which is suitable for simulating hydrocarbons. Recently, we found that in GA, H(q) acquires a strong renormalization for small wave-vectors q and the layer remains almost flat even for temperatures as high as 900 K [23]. Here we will analyze the thermal rippling be- havior of FG and compare it with GA. A comparison with GE and BN single layers which behave as 2D mem- branes [31, 39] will also be presented. H(q) for FG was AIREBO κ(eV ) 1.165 10.19 -- ReaxFF hh2i(A2) κ(eV ) hh2i(A2) 1.307 0.070 -- 1.16 7.26 5.07 0.627 0.074 0.073 GE GA GF TABLE II: Comparison of AIREBO and ReaxFF for the bending rigidity and hh2i for GE, GA and FG at 300 K. calculated following the steps described in our previous work [23]. Starting from a pure GE sheet, the variation of the height-height correlation function H(q) at room temper- ature for different partial fluor contents is shown in Fig. 4 (a). The curves were shifted for a better comparison. We found that while for 10 to 90% coverage, H(q) follows Eq. (2) up to small q−values which is similar to the case of GE [32]. But, for fully FG at q∗ ≈ 0.2A−1, H(q) de- viates from the harmonic law (solid line) and approaches roughly a constant value similar to was previously found for GA [23]. In the inset of Fig. 4(a) we show the square average of the out-of-plane fluctuations hh2i at 300 K. Notice that the out of plane fluctuations for partially covered samples are considerably larger for FG than for GA. The temperature dependence of H(q) for fully flu- orinated graphene is shown in Fig. 4(b). Irrespective of temperature, the short wave-length limit of H(q) tends always approximately to a constant value. The charac- teristic q-value where H(q) deviates from the harmonic approximation result decreases with increasing tempera- ture. The renormalization of H(q) for long wavelengths in- dicates the suppression of large out-of-plane height fluc- tuations. In Fig. 5(a) we compare the behavior of hh2i against temperature for GE, BN (panel (a.1)), FG and GA (panel (a.2)). Notice that hh2i increases from 0.7 A2 up to 4 A2 in BN and from 0.7 A2 up to 2 A2 in GE when temperature is varied from 10 K up to 900 K. Due to the absence of long wave-length ripples, hh2i remains approximately constant for GA and FG, and the varia- tions are smaller than those for BN and GE, over the same temperature range. The temperature dependence of the bending rigidity κ, computed from the harmonic part of H(q) is shown in Fig. 4(b). Note that the larger magnitude for GA and FG is a consequence of the smaller corrugations present in these materials. We also find the opposite temperature dependence for BN and GE when compared with GA and FG. In this sense GE and BN behave anomalously. The corresponding bending rigid- ity and hh2i at room temperature for GE, GA and FG are listed in Table II. Density functional calculations for band structure of FG (and GA) [40] shows that the acoustic out-of-plane modes (ZA) in FG (and GA) are different from that of GE. The most important difference from GE is the de- coupled optical and acoustic bands in FG and GA close (a) 0% 10% 50% 70% 90% 100% 0.125 ) s t i n u . b r a ( ) q ( H ) s t i n u . b r a ( ) q ( H 0.125 4 F H 0 20 40 60 80 100 % Atoms 2 > h < 30 20 10 0 0.25 0.5 q [Å-1] 0.25 0.5 q [Å-1] 1 (b) 900K 700K 500K 300K 100K 1 2 FIG. 4: (Color online) (a) Log-log plot of H(q) for different coverage of F atoms at T=300 K. The solid lines show the harmonic q−4 behavior valid in the limit of large q values. Note the strong deviation, starting roughly at q∗ ≈ 0.2A−1 in the limit of long wave-lengths, for the case of fully fluorinated graphene. The variation of hh2i is shown in the inset. (b) H(q) for fully fluorinated FG at different temperatures. to the Γ point. The light H atom contributes to the high- est phonon frequencies which is not the case for F atoms. It is also seen that the ZA modes close to the Γ point for FG (and GA) are not well fitted by a quadratic func- tion in contrast to the GE case. This is clear indication of anharmonicity. In summary, the atomistic details of the structure of FG is more complicated and therefore more details of this structure should be included in any continuum theory. Before ending this section, note that as we discussed in our previous work [23], the scaling with the system size present in GE is no longer valid for FG and GA (hh2i in FG and GA is almost constant irrespective to the system size). The lower wavelengths (q) adopted for the calculation of H(q) are equal to qx−min = 2π and lx ) 2 Å ( > 2 h < BN GE (a.1) GA FG (a.2) 4 2 0 (b.1) (b.2) ) V e ( κ 1.5 1.25 1 0.15 0.1 0.05 15 10 5 0 300 600 900 0 T (K) 300 600 900 T(K) FIG. 5: (Color online) Variation of (a) hh2i, and (b) κ against temperature for FG (open squares), GA (filled squared), GE (open circles), and BN (stars). qy−min = 2π and represent the 'computational' cut-offs ly of possible large wavelength ripples where lx and ly are the dimension of the system. Notice that deviation from the harmonic behavior takes place at larger value of q and hence this effect can not be a finite size effect and is instead an intrinsic phenomenon of the material. B. Effective Young's modulus In order to study the mechanical stiffness we consider an FG flake with dimension lx × ly = 15 × 15 nm2. Be- fore the stretching process, the sample is equilibrated for 5 ps (i.e. 50,000 time steps). Stretching direction is al- ways along x and the uniaxial strain is applied within the NPT ensemble [41] where the pressure is slowly increased, i.e. 2 GPas/ps. In this section the lateral edges (in the y direction) were taken as the arm-chair direction hav- ing both free (FBC) and periodic boundary conditions (PBC). We kept temperature fixed at T=10 K. The total strain energy per atom of the strained flake can be written as a function of the imposed strain (ǫ) ET (ly, ǫ) = E0 + S0 ly γ(ǫ) + S0 2 Y ǫ2 (3) where E0 is the energy of the infinite planar undeformed flake, γ(ǫ) is the excess edge energy, and Y is Young's modulus (Y ) of the flake. For nano-ribbons with no lateral edges we have γ = 0 (assuming that the longitudinal edges which are fixed make no contribution). This is due to the fact that free edges increase the energy due to buckling and bond-order changes [42]. Recently Lu et al used the Brenner poten- tial [43] in molecular dynamics simulations and studied the excess edge energy of graphene nano-ribbons as a 5 function of width and chirality [42]. Our systems are dif- ferent from those of Ref. [42]. In contrast to Ref. [42] we are not interested in effects due to the edge energy effect and the size dependence. We rewrite Eq. (3) in the following as an effective Young's modulus which qualita- tively gives a good description of the mechanical stiffness of all the examined 2D materials. Nevertheless our re- sults are in qualitative agreement with those reported by Lu at al, i.e. increasing of total energy for the FBC case as compared to a nano-ribbon. Assuming a quadratic relation for γ(ǫ) = ly 2 ǫ2 valid for small ǫ, the simplest method to estimate Young's modulus is by fitting the quadratic function to the total energy (per area): ET = E0 + 1 2 Yef f ǫ2, (4) where Yef f is the effective Young's modulus of the sys- tem. Using aforementioned fitting process we found Yef f respectively for a flake with arm-chair and zig-zag FG, to be 273 N/m and 250 N/m. Notice that the experimental result is 100 N/m for not perfect FG [8] while the DFT result is 250 N/m [18]. The latter disagreement between theory and experiment may be explained due to the fact that in experimental samples the fluor-to-carbon ratio is larger than unity, i.e. 1.1 [8], because of the presence of defects. Such defects become active regions which can adsorb the free F (and even H) atoms. Therefore, in the defected parts more F atoms will be found which is responsible for a F/C ratio larger than one. In order to understand the effect of the different bound- ary conditions, we depict in Fig. 6 the variation of ET per atom with ǫ for flakes with both FBC (dashed lines) and PBC, i.e. nano-ribbon (solid lines). It is seen that for flakes with FBC the free edges result in an increase of the energy. The inset shows the difference between two curves, i.e. ∆ET = EF BC − EP BC which is posi- tive. Because the free boundaries have many dangling bonds which are not saturated by F atoms it results in extra energy. This can also occur in other systems, e.g. graphene [44]. Notice that for the studied low tempera- ture here, i.e. T=10 K we do not expect that bond recon- struction at the edges is important. Notice that even by saturating all the bonds by F, still the change in the bond order term in ReaxFF (due to different chemical environ- ment of the boundary atoms) results in higher energy as compared to PBC. Furthermore, both FBC and PBC results exhibit a quadratic behavior which is an indication of the harmonic regime for the applied uniaxial stain. As is clear from the inset of Fig. 6, the difference between the two curves is increasing with applied strain. This is due to the devia- tion from equilibrium for the C-F bonds, C-C-F (F-C-F) bond angles, and the dihedral angles (F-C-C-F torsion angle) of the free edge atoms. The larger the strain (and the larger the length of ribbon), the larger the deviation from equilibrium for the bonds and the angles. In the PBC case there is no such edge effect but nevertheless because of the fixing of the two longitudinal ends (the  is under tension from the arm-chair direction. The fixed longitudinal ends do not have any effect in our results be- cause both FBC and PBC have the same contributions. 6 0.04 0.06 0.02 Strain IV. CONCLUSIONS ) t m o a V e ( / 0.12 0.10 0.08 0.06 0.04 0.02 0.00 T E 0.00 ) m o t / a V e ( T E -5.52 -5.56 -5.60 -5.64 -5.68 -5.72 -5.76 0.00 0.01 0.02 0.03 Strain 0.04 0.05 0.06 0.07 FIG. 6: (Color online) (a) Variation of total energy against uniaxial strain for FG subjected to free boundary condition (FBC) and periodic boundary condition (PBC) for the lateral edges, i.e. the dashed and solid curves respectively. The inset shows the difference between the two energy curves. edges which are under uniaxial stress) the energy varia- tion of the PBC system should be different from that of an infinite FG which is periodic in both directions while it We provided a new set of parameters for the ReaxFF potential for the C-F covalent bond and tested it on var- ious molecules. Subsequently, molecular dynamics sim- ulations were used to investigate the thermal rippling behavior and the mechanical response of fluorographene (FG) under uniaxial stress. The obtained results are compared with those for graphene (GE), graphane (GA) and hexagonal boron nitride sheet (BN). We found that fluorographene remains a flat sheet similar to graphane even at high temperature, i.e. up to 900 K. The bending rigidity of FG is found to be independent of tempera- ture and its Young's modulus is in good agreement with experiment. Acknowledgments. This work is supported by the ESF- Eurographene project CONGRAN, the Flemish Science Foundation (FWO-Vl) and the Methusalem Foundation of the Flemish Government. SGS and ACTvD acknowl- edge support by the Air Force Office of Scientific Re- search (AFOSR) under Grant FA9550-10-1-0563. [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, Nat. Mater. 6, 183 (2007). [3] C. Berger, Z. Song, T. Li, A. Y. Ogbazghi, R. Feng, Z. Dai, A. N. Marchenkov, E. H. Conrad, P. N. First, and W. A. de Heer, Science 312, 1191 (2006). [4] M. I. Katsnelson, K. S. Novoselov, and A. K. Geim, Nat. Phys. 2, 620 (2006). A. K. Stubos, V. Georgakilas, K. Safarova, D. Jancik, C. Trapalis, and M. Otyepka, Small 6, 2885 (2010). [11] S. H. Cheng, K. Zou, F. Okino, H. R. Gutierrez, A. Gupta, N. Shen, P.C. Eklund, J. O. Sofo, and J. Zhu, Phys. Rev. B 81, 205435 (2010). [12] F. Withers, M. Dubois, and A. K. Savchenko, Phys. Rev. B 82, 073403(2010). [13] O. Leenaerts, H. Peelaers, A. D. Hern´andez-Nieves, B. Partoens, and F. M. Peeters, Phys. Rev. B 82, 195436 (2010). [5] H. S¸ahin, R. T. Senger, and S. Ciraci, J. Appl. Phys. [14] D. K. Samarakoon, Z. Chen, C. Nicolas, and X. Q. Wang, 108, 074301 (2010). Small 7, 965 (2011). [6] A. H. Castro Neto, F. Guinea, N. M. R. Peres, K. S. Novoselov, and A. K. Geim, Rev. Mod. Phys. 81, 109 (2009). [7] D. C. Elias, R. R. Nair,T. M. G. Mohiuddin, S. V. Mo- rozov, P. Blake, M. P. Halsall, A. C. Ferrari, D. W. Boukhvalov, M. I. Katsnelson, A. K. Geim, and K. S. Novoselov, Science 323, 610 (2009). [8] R. R. Nair, W. Ren, R. Jalil, I. Riaz, V. G. Kravets, L. Britnell, P. Blake, F. Schedin, A. S. Mayorov, S. Yuan, M. I. Katsnelson, H. M. Cheng, W. Strupinski, L. G. Bulusheva, A. V. Okotrub, I. V. Grigorieva, A. N. Grig- orenko, K. S. Novoselov, and A. K. Geim, Small 6, 2877 (2010). [9] J. T. Robinson, J. S. Burgess, C. E. Junkermeier, S. C. Badescu, T. L. Reinecke, F. K. Perkins, M. K. Zalautd- niov, J. W. Baldwin, J. C. Culbertson, P. E. Sheehan, and E. S. Snow, Nano Lett. 10, 3001 (2010). [10] R. Zboril, F. Karlicky, A. B. Bourlinos, T. A. Steriotis, [15] D. W. Boukhavalov, M. I. Katsnelson, and A. I. Lichten- stein, Phys. Rev. B 77, 035427 (2008). [16] M. H. F. Sluiter and Y. Kawazoe, Phys. Rev. B 68, 085410 (2003). [17] J. O. Sofo, A. S. Chaudhari, and G. D. Barber, Phys. Rev. B 75, 153401 (2007). [18] H. S¸ahin, M. Topsakal, and S. Ciraci, Phys. Rev. B 83, 115432 (2011). [19] F. Withers, S. Russo, M. Dubois, and M. F. Craciun, Nanoscale Res. Lett. 6, 526 (2011). [20] N. D. Mermin and H. Wagner, Phys. Rev. Lett. 17, 1133 (1996). [21] K. V. Zakharchenko, J. H. Los, M. I. Katsnelson, and A. Fasolino, Phys. Rev. B 81, 235439 (2010). [22] J. C. Meyer, A. K. Geim, M. I. Katsnelson, K. S. Novoselov, T. J. Booth, and S. Roth, Nature (London) 446, 60 (2007). D. A. Kirilenko, A. T. Dideykin, and G. Van Tendeloo, Phys. Rev. B 84, 235417 (2011). 7 [23] S. Costamagna, M. Neek-Amal, J. H. Los, and F. M. solino, Phys. Rev. B 72, 214102 (2005). Peeters, Phys. Rev. B 86, 041408 (2012). [24] A. C. T. van Duin, S. Dasgupta, F. Lorant, and W. A. Goddard, J. Phys. Chem. A 105, 9396 (2001). [36] J. H. Los, M. I. Katsnelson, O. V. Yazyev, K. V. Za- kharchenko, and A. Fasolino, Phys. Rev. B 80, 121405 (2009). [25] J. E. Mueller, A. C. T. van Duin, and W. A. Goddard [37] C. Sevik, A. Kinaci, J. B. Haskins, and T. C¸ agin, Phys. III, J. Phys. Chem. C 114, 4939 (2010). Rev. B 84, 085409 (2011). [26] W. J. Mortier, S. K. Ghosh, and S. Shankar, J. Am. [38] S. J. Stuart, A. B. Tutein, and J. A. Harrison, J. Chem. Chem. Soc. 108, 4315 (1986). Phys. 112, 6472 (2000). [27] Jaguar, version 7.8, Schrodinger, (LLC, New York, NY, [39] S. K. Singh, M. Neek-Amal, S. Costamagna, and F. M. 2011). [28] http : //lammps.sandia.gov [29] S. J. Plimpton, J. Comp. Phys. 117, 1 (1995). [30] D. Nelson, T. Piran, and S. Weinberg, Statistical Mechan- ics of Membrane and Surface (Word Scientific, Singapore, 2004). [31] A. Fasolino, J. H. Los, and M. I. Katsnelson, Nat. Mater. 6, 858 (2007). [32] S. Costamagna and A. Dobry, Phys. Rev. B 83, 233401 (2011). [33] R. Rold´an, A. Fasolino, K. V. Zakharchenko, and M. I. Katsnelson, Phys. Rev. B 83, 174104 (2011). Peeters, Submitted to Phys. Rev. B (2012). [40] H. Peelaers, A. D. Hernandez-Nieves, O. Leenaerts, B. Partoens, and F. M. Peeters, Appl. Phys. Lett. 98, 051914 (2011). [41] Y. Zheng, N. Wei, Z. Fan, L. Xu, and Z. Huang, Nan- otechnology 22, 405701 (2011). [42] Q. Lu, W. Gao, and R. Huang, Phys. Rev. B 81, 155410 (2010); ibid J. Phys.: Condens. Matter 19, 054006 (2011). [43] D. W. Brenner, O. A. Shenderova, J. A. Harrison, S. J. Stuart, B. Ni, and S. B. Sinnott, J. Phys.: Condens. Matter 14, 783 (2002). [34] P. Le Doussal and L. Radzihovsky, Phys. Rev. Lett. 69, [44] P. Koskinen, S. Malola, and Hakkinen, Phys. Rev. Lett. 1209 (1992). 101, 115502 (2008). [35] J. H. Los, L. M. Ghiringhelli, E. J. Meijer, and A. Fa- 

1708.01230

1

1708

Adhesion, Stiffness and Instability in Atomically Thin MoS2 Bubbles

We measured the work of separation of single and few-layer MoS2 membranes from a SiOx substrate using a mechanical blister test, and found a value of 220 +- 35 mJ/m^2. Our measurements were also used to determine the 2D Young's modulus of a single MoS2 layer to be 160 +- 40 N/m. We then studied the delamination mechanics of pressurized MoS2 bubles, demonstrating both stable and unstable transitions between the bubbles' laminated and delaminated states as the bubbles were inflated. When they were deflated, we observed edge pinning and a snap-in transition which are not accounted for by the previously reported models. We attribute this result to adhesion hysteresis and use our results to estimate the work of adhesion of our membranes to be 42 +- 20 mJ/m^2.

cond-mat.mes-hall

cond-mat

Adhesion, Stiffness and Instability in Atomically Thin MoS2 Bubbles. David Lloyd1, Xinghui Liu2, Narasimha Boddeti2, Lauren Cantley1, Rong Long2, Martin L. Dunn3 and J. Scott Bunch1,4* 1Boston University, Department of Mechanical Engineering, Boston, MA 02215 USA 2University of Colorado, Department of Mechanical Engineering, Boulder, CO 80309 USA 3Singapore University of Technology and Design, Singapore, 487372 4Boston University, Division of Materials Science and Engineering, Brookline, MA 02446 USA *e-mail: [email protected] Abstract We measured the work of separation of single and few-layer MoS2 membranes from a SiOx substrate using a mechanical blister test, and found a value of 220 ± 35 mJ/m2. Our measurements were also used to determine the 2D Young's modulus (E2D) of a single MoS2 layer to be 160 ± 40 N/m. We then studied the delamination mechanics of pressurized MoS2 bubbles, demonstrating both stable and unstable transitions between the bubbles' laminated and delaminated states as the bubbles were inflated. When they were deflated, we observed edge pinning and a snap-in 1 transition which are not accounted for by the previously reported models. We attribute this result to adhesion hysteresis and use our results to estimate the work of adhesion of our membranes to be 42 ± 20 mJ/m2. KEYWORDS: adhesion, MoS2, adhesion hysteresis, Young's modulus, friction, bubble Adhesive forces play an important role in shaping the mechanical behavior of atomically thin materials such as graphene or molybdenum disulfide, MoS2. These forces keep the material clamped to the substrate, and also influence how the membrane folds1, slides2, and peels3. An understanding of adhesion in these materials is important in the fabrication of nanoelectromechanical systems4, flexible electronic devices5, graphene origami1,6, graphene separation membranes7, and stacked heterostructures formed from 2D materials. Atomically thin crystals may also provide a fruitful system in which to study novel features of friction and adhesion present only at the nanoscale2,8–10. In terms of device performance, adhesive forces determine the maximum strain 2D materials can support which is important in designing stretchable electronic devices11 and pressure sensors12. The study of bubbles formed by atomically thin sheets has proven to be useful for discovering the adhesive and mechanical properties of these materials, and has allowed measurements of the adhesion energies13, friction coefficient14, and Young's modulus of graphene and other 2D materials15. In particular, Koenig et al. used a mechanical blister 2 test to measure the adhesion energy between graphene and SiOx of ~450 mJ/m2. Like graphene, atomically thin MoS2 is a mechanically exceptional material16, whilst also being piezoelectric11,17 and a direct gap semiconductor with a highly strain sensitive band gap18–21. A good understanding of the mechanical stiffness and adhesion to the substrate is therefore of particular importance to this material which has applications involving the interplay between adhesive and tensile forces. In this paper, we measure the work of separation (sometimes referred to as the adhesion energy) between MoS2 and the substrate by employing the same geometry as used in our previous work7,13,22, in which we suspend mechanically exfoliated or chemical vapor deposition (CVD) grown membranes over cylindrical microcavities etched into a silicon oxide (SiOx) substrate (Fig. 1a and 1b). The devices are then placed in a pressure chamber filled with a gas of pressure p0, which gradually leaks into the cavities through the SiOx substrate until the internal pressure pint reaches that of the chamber (pint = p0). We used either N2, Ar, H2 or He gas which allowed us to choose a convenient leak rate of the gas into the microcavities. When the devices are removed from the pressure chamber the internal pressure (pint) is greater than the external pressure (pext = 1 atm), and this pressure difference (Δp = pint – pext > 0) causes the membrane to bulge up (Fig. 1c and 1d). For each charging pressure p0 we measure the deflection δ and radius a of the bubble using an atomic force microscope (AFM) after which the devices are returned to the pressure chamber at a higher p0 and the process is repeated. We fabricated devices of 1-3 layer thickness by mechanical exfoliation, and made monolayer devices from CVD grown MoS2 using a PMMA transfer method (see supporting information for details). We 3 transferred 6 different growths to produce CVD samples N1-6, with each containing many individual devices. The SiOx substrates were O2 plasma cleaned prior to transfer. As can be seen in Fig. 1d-f, increasing p0 causes δ to increase with a initially remaining pinned at the radius of the cylindrical microcavity, a0. After a critical pressure is reached (p0 ~ 600 kPa), the force from the pressure difference across the membrane overcomes the adhesive forces keeping the membrane clamped to the substrate, and delamination occurs in the form of a snap-out transition of the radius from 4.4 μm to 6 μm. After the snap-out transition, both a and δ continue to gradually increase as p0 is increased. We begin by using our values for p0, δ and a to determine the Young's modulus of MoS2 with a formula developed in Hencky's model for clamped pressurized membranes23, which relates the pressure difference across the membrane Δp to the deflection δ and radius a by the formula, ∆𝑝 = 𝐾(𝑣)𝐸2𝐷𝛿3 𝑎4 (1) with a Poisson's ratio ν = 0.2916, numerical constant K(υ) = 3.54 and a two dimensional Young's modulus E2D equal to the bulk Young's modulus multiplied by the thickness of the material. The pressure difference, Δp, is calculated from p0 by assuming isothermal expansion of a fixed number of ideal gas molecules from the initial volume of the cavity (V0) to its final volume (V0 + Vb), such that p0V0 = pint(V0+Vb). From Hencky's model, the 4 volume created beneath the bubble can be found from the device geometry using the expression Vb = C(υ)πa2δ, and a numerical constant C(ν)= 0.522. We measured the E2D of 3 CVD samples (N1-3), and of exfoliated monolayer and trilayer flakes containing 2 and 16 devices respectively. Fig. 2a shows a plot of Δp against K(v) δ3/a4 for each of our CVD monolayer and bilayer devices in sample N2, including linear fits which are used to determine E2D for each device. The E2D of each device in these samples is plotted in Fig. 2b. In Fig. 2c we plot the mean E2D for each sample divided by the number of layers n in the membranes in order to compare estimates for the E2D of a single MoS2 layer. Error bars represent the standard deviation. For our exfoliated devices we find an average E2D per layer of 190 ± 35 N/m, and for our CVD grown MoS2 monolayers we find an average E2D of 128 ± 20 N/m. There is a low variance of E2D within each CVD grown sample, however there is a significant difference between the average E2D for each CVD sample. The discrepancy between CVD and exfoliated samples and among different CVD samples may be due to differences in defect densities24,25 which occur during CVD growth, as an increased sulfur vacancy density26 is predicted to lower E2D in MoS2 27. The average of all our exfoliated and CVD grown samples is 160 ± 40 N/m, which falls within the same range of values as found in previous studies16,28,29, which we plot in Fig. 2c for comparison. 5 We next determined the work of separation, Γsep, using our values for p0, δ and a, and a free energy model described in detail by others30,31. Briefly, we can write the total free energy of the system F as, 𝐹 = (𝑝𝑖𝑛𝑡−𝑝𝑒𝑥𝑡)𝑉𝑏 4 + Г𝜋(𝑎2 − 𝑎𝑜 2) − 𝑝𝑜𝑉𝑜 𝑙𝑛 [ 𝑉𝑜+𝑉𝑏 𝑉𝑜 ] + 𝑝𝑒𝑥𝑡𝑉𝑏 (2) where V0 is the initial volume of the cavity, Vb is the additional volume created as the bubble expands. Γ is the adhesion energy, which is equal to Γsep in the case of delamination. The first two terms represent the elastic strain energy and the work to separate the membrane from the substrate respectively, and the final two terms account for the isothermal expansion of the gas. When a device is removed from the pressure chamber, the bubble volume expands until the free energy of the system F reaches a local minimum. We minimize F with respect to a by setting dF/da = 0 and using the relationship p0V0 = pint(V0 + Vb). This yields the expression for the work of separation: 𝛤𝑠𝑒𝑝 = 5𝐶 4 ( 𝑝0𝑉0 𝑉0+𝑉𝑏(𝛿,𝑎) − 𝑝𝑒𝑥𝑡) 𝛿 (3) with the constant C(ν) = 0.522 for ν = 0.2916. Using this expression, we can determine Γsep of each device using the charging pressure of the pressure chamber p0, and δ and a of 6 the bubble measured using an AFM. We can also substitute the pressure terms in Eq. 3 with Hencky's result in Eq. 1 which yields, 𝛤𝑠𝑒𝑝 = 5 4 C𝐾𝐸2𝐷 ( 4 𝛿 𝑎 ) (4) which holds for all devices which have started to delaminate (a > a0). This allows Γsep to be determined from δ and a without knowing p0, which avoids the long waiting times required for devices to reach equilibrium in the pressure chamber. For our exfoliated devices we calculated Γsep using Eq. 4 (using the mean value of E2D = 190 N/m per layer we found earlier for exfoliated samples), and used Eq. 3 to calculate Γsep for our CVD devices where p0 was well known. We find no significant difference in Γsep between single and few layer samples, or CVD and exfoliated samples (Fig. 3). By averaging over all samples we find the mean work of separation to be Γsep = 220 ± 35 mJ/m2, which is close to the value of 170 ± 30 mJ/m2 measured for many layer MoS2 32 and is in the same range of values as found for graphene13,33–36. The devices shown in Fig. 1d-f exhibit unstable delamination, whereby a discontinuously increases from the initial radius a0 when p0 ≳ 600 kPa. The etched depth of the 7 microcavities in that case was d = 1500 nm. We also fabricated devices with cavity depths of d = 650 nm, and again performed measurements of δ and a at increasing p0 (Fig. S9) using the method described earlier. With this cavity depth, the devices show no snap-out transition, and rather stably delaminate with a continuously increasing from a0. The difference in behavior in these two cases has been observed and modeled by others31,37, and Bodetti et al found that the transition from unstable to stable delamination occurs when the parameter S = 2Vb/V0 satisfies the condition S >1 just before the point of delamination31. Reducing the well depth decreases the volume of the cavity relative to the volume of the bubble which increases S. By making various device geometries and finding S from AFM measurements we confirmed empirically that this transition occurs in the range 0.74 < S < 1.11, and we obtained the same value for Γsep for both stable and unstable delamination (see supplementary info for details). After the devices with d = 1500 nm (on sample N2) had been delaminated to their largest radii, they were left out in ambient conditions to deflate over the course of ~48 hours. During this time AFM scans captured δ and a as the number of gas molecules N decreases from the initial value of N0 (= p0V0/kbT). AFM cross sections of a bubble are shown in Fig. 4a during the inflation (increasing N0) and deflation (decreasing N) of the device. Initially as the device is inflated, δ increases and a remains pinned at a0. When p0 ≳ 600 kPa the snap-out transition occurs and a jumps to a larger value, after which both a and δ increase together as N0 increases. When devices are left to deflate, δ decreases from an initial value of δ0, however a now does not change from its radius at the beginning of deflation, which we refer to as the 'pinned radius' ap. After the deflection of the devices 8 reaches a critical value δ = δc the devices undergo a snap-in transition where the radius jumps from ap to a0, and δ continues to decrease to zero. Values for δ and a throughout this process are shown in Fig. 4b, which shows devices deflating at a number of different ap. Videos of the snap-out and snap-in transitions can be seen in the supporting information. We can interpret this using the result derived in Eq. 4, which requires that after delamination the ratio δ /a remains constant, with the magnitude of this ratio being proportional to Γsep 1/4. We plot the line corresponding to this formula in Fig. 4b (upper dashed line) with the values of E2D and Γsep determined earlier, and find our data for increasing N0 follows this trend very well. This formula is independent of whether N is increasing or decreasing, so when our devices are left to deflate we should expect δ and a to return along the same path as during inflation described by Eq. 4. As can be seen in Fig. 4a however, there is a significant difference in the geometry of the bubbles during inflation and deflation, which suggests some element of our system is irreversible. We attribute the difference between inflation and deflation we see in our data to the widely observed phenomenon of adhesion hysteresis34,38,39, whereby the energy required to separate the membrane from the surface Γsep is greater than the energy returned to the 9 system as the membrane re-adheres Γadh, with Γadh < Γsep. After making this simple modification (see supporting information for more details), our model now predicts that the device should remain pinned at radius ap until a snap-in transition occurs at a critical deflection determined by 𝛤𝑎𝑑ℎ = 5 4 C𝐾𝐸2𝐷 ( 𝛿𝑐 𝑎𝑝 4 ) (5) We perform a linear fit of our measurements of δc and ap (lower dashed line in Fig. 4b) which yields an estimate of the work of adhesion for this sample to be Γadh = 14 ± 5 mJ/m2. Multiple measurements of Γadh with the same device show that this measurement is repeatable over many cycles (Fig. S3b in the supporting information). We performed measurements on a total of 5 CVD grown samples (N2-6) and found the mean work of adhesion for all our samples to be 42 ± 20 mJ/m2, with Γadh < Γsep in every device. Fig. 4c shows a comparison between the works of separation and adhesion for 3 of these samples (N2-4). Γadh varied noticeably between samples, with sample means falling in the range 14 - 63 mJ/m2 (Fig. S5 in the supporting information). Our measurements of Γadh show that as little as one tenth of the energy required to separate the membrane from the substrate (Γsep ~ 220 mJ/m2) is recovered as the membrane at the edge of the bubble re-adheres to the substrate. We used Raman spectroscopy to measure the membrane strain distribution around our devices before and after snap-in (see supporting information for details), and found that whilst some energy was dissipated in the form of residual strain transferred to the membrane, this can only 10 account for <10% of the dissipation that produces a difference between Γadh and Γsep. This strain may also dissipate some energy through frictional sliding as the membrane changes its length on the surface of the substrate14. Adhesion hysteresis is a commonly observed phenomenon40 which has previously been observed in nano-indentation measurements of graphene34, and the fraction of the energy dissipated in our system is comparable with the hysteresis observed in elastomers41. The behavior of our devices is also analogous to the related phenomenon of contact angle hysteresis seen in liquid bubbles39, and constant contact area pinning during unloading has been seen previously between two adhered solid spheres42. Surface roughness and chemical heterogeneity on the surface can produce contact angle and adhesion hysteresis40,43, and a further contribution in our system could be the finite time over which deflation occurs. This could mean that the membrane does not have time during the measurement to re-conform fully to the surface or re-make the bonds which were made before the device delaminated44,45. This would result in the system being in a transient non-equilibrium state during the measurement, which is a common cause of thermodynamic irreversibility and adhesion hysteresis40,46,47. Our method of finding Γsep also involves subjecting the membranes to high external pressures prior to measurement, which could improve their conformation to the substrate and thereby enhance Γsep relative to Γadh. 11 We have measured the work of separation of single and few layer MoS2 fabricated by CVD and mechanical exfoliation, and found a value of Γsep = 220 ± 35 J/m2. We also measured the Young's modulus, and found that E2D = 160 ± 40 N/m for a single MoS2 layer. Bulge testing provides a complimentary method to nanoindentation to determine E2D, and our results are in the same range of values as reported in previous studies. We demonstrated snap-out and snap-in instabilities, which mechanically amplify small changes in pressure and could be used for pressure sensing. Finally we observed bubble edge pinning, analogous to contact angle hysteresis observed in liquids, and used Raman spectroscopy to provide evidence that the trapping of strain energy after the snap-in transition can account for some but not all of the hysteresis. We measured a Γadh which was significantly lower than Γsep, which may affect the performance of nanomechanical switches made from atomically thin materials48,49. The distinction between Γadh and Γsep we have observed here is an important consideration in the analysis of bubbles formed under atomically thin crystals15,50,51, and in the design of folded 3D structures made from 2D sheets1,6. Supporting information Supporting information includes details of CVD growth and characterization, the effect of membrane pre-tension, the full set of work of adhesion and separation data, the free energy model including the effect of adhesion hysteresis, contact angle measurements of a bubble during deflation, the trapping of strain around the edge of the devices, the effect of membrane slipping on our E2D calculations, data from devices which exhibit stable delamination, additional measurements of deflating devices, the full set of Young's modulus data, and videos of the snap transitions. 12 Acknowledgments: This work was funded by the National Science Foundation (NSF), grant no. 1054406 (CMMI: CAREER, Atomic Scale Defect Engineering in Graphene Membranes), a grant to L. Cantley by the NSF Graduate Research Fellowship Program under grant no. DGE- 1247312, and a BUnano Cross-Disciplinary Fellowship to D. Lloyd. We thank Chuanhua Duan for use of the high speed camera. References (1) Cranford, S.; Sen, D.; Buehler, M. J. Appl. Phys. Lett. 2009, 95, 123121 1–4. (2) Li, S.; Li, Q.; Carpick, R. W.; Gumbsch, P.; Liu, X. Z.; Ding, X.; Sun, J.; Li, J. Nature 2016, 539, 2–7. (3) Annett, J.; Graham, L. W. Nature 2016, 535 (7611), 271–275. (4) Bunch, J. S.; Van Der Zande, A. M.; Verbridge, S. S.; Frank, I. W.; Tanenbaum, D. M.; Parpia, J. M.; Craighead, H. G.; McEuen, P. L. Science 2007, 315, 490–493. (5) Akinwande, D.; Petrone, N.; Hone, J. Nat. Commun. 2014, 5:5678. (6) Ebbesen, T. W.; Hiura, H. Adv. Mater. 1995, 305 (6), 582–586. (7) Koenig, S. P.; Wang, L.; Pellegrino, J.; Bunch, J. S. Nat. Nanotechnol. 2012, 7, 728–732. (8) Carpick, R. W.; Salmeron, M. Chem. Rev. 1997, 97, 1163–1194. (9) Lee, C.; Li, Q.; Kalb, W.; Liu, X.; Berger, H.; Carpick, R. W.; Hone, J. Science 2010, 328, 76–80. (10) Koren, E.; Lörtscher, E.; Rawlings, C.; Knoll, A. W.; Duerig, U. Science 2015, 348 (6235), 679–683. (11) Wu, W.; Wang, L.; Li, Y.; Zhang, F.; Lin, L.; Niu, S.; Chenet, D.; Zhang, X.; Hao, Y.; Heinz, T. F.; Hone, J.; Wang, Z. L. Nature 2014, 514 (7523), 470–474. (12) Smith, A. D.; Niklaus, F.; Paussa, A.; Schro der, S.; Fischer, A. C.; Sterner, M.; Wagner, S.; Vaziri, S.; Forsberg, F.; Esseni, D.; O stling, M.; Lemme, M. C. ACS Nano 2016, 10, 9879−9886. (13) Koenig, S. P.; Boddeti, N. G.; Dunn, M. L.; Bunch, J. S. Nat. Nanotechnol. 2011, 6 (9), 543–546. 13 (14) Kitt, A. L.; Qi, Z.; Remi, S.; Park, H. S.; Swan, A. K.; Goldberg, B. B. Nano Lett. 2013, 13, 2605−2610. (15) Khestanova, E.; Fumagalli, L.; Geim, A. K.; Grigorieva, I. V. Nat. Commun. 2016, 7, 12587. (16) Cooper, R. C.; Lee, C.; Marianetti, C. A.; Wei, X.; Hone, J.; Kysar, J. W. Phys. Rev. B 2013, 87 (3), 035423. (17) Zhu, H.; Wang, Y.; Xiao, J.; Liu, M.; Xiong, S.; Wong, Z. J.; Ye, Z.; Ye, Y.; Yin, X.; Zhang, X. Nat. Nanotechnol. 2014, 10 (2), 151–155. (18) He, K.; Poole, C.; Mak, K. F.; Shan, J. Nano Lett. 2013, 13 (6), 2931–2936. (19) Conley, H. J.; Wang, B.; Ziegler, J. I.; Haglund, R. F.; Pantelides, S. T.; Bolotin, K. I. Nano Lett. 2013, 13 (8), 3626–3630. (20) Castellanos-Gomez, A.; Roldan, R.; Cappelluti, E.; Buscema, M.; Guinea, F.; Van Der Zant, H. S. J.; Steele, G. A. Nano Lett. 2013, 13 (11), 5361–5366. (21) Lloyd, D.; Liu, X.; Christopher, J. W.; Cantley, L.; Wadehra, A.; Kim, B. L.; Goldberg, B. B.; Swan, A. K.; Bunch, J. S. Nano Lett. 2016, 16, 5836−5841. (22) Bunch, J. S.; Verbridge, S. S.; Alden, J. S.; Van Der Zande, A. M.; Parpia, J. M.; Craighead, H. G.; McEuen, P. L. Nano Lett. 2008, 8 (8), 2458–2462. (23) Fichter, W. NASA Tech. Pap. 1997, 3658, 1–24. (24) Zandiatashbar, A.; Lee, G.; An, S. J.; Lee, S.; Mathew, N.; Terrones, M.; Hayashi, T.; Picu, C. R.; Hone, J.; Koratkar, N. Nat. Commun. 2014, 5, 3186. (25) López-Polín, G.; Gómez-Navarro, C.; Parente, V.; Katsnelson, M. I.; Pérez-Murano, F.; Gómez-Herrero, J. Nat. Phys. 2015, 11, 26–31. (26) Hong, J.; Hu, Z.; Probert, M.; Li, K.; Lv, D.; Yang, X.; Gu, L.; Mao, N.; Feng, Q.; Xie, L.; Zhang, J.; Wu, D.; Zhang, Z.; Jin, C.; Ji, W.; Wu, D.; Zhang, Z.; Jin, C.; Ji, W.; Zhang, X.; Yuan, J.; Zhang, Z. Nat. Commun. 2015, 6, 6293. (27) Gan, Y.; Zhao, H. Phys. Lett. A 2014, 378 (38-39), 2910–2914. (28) Bertolazzi, S.; Brivio, J.; Kis, A. ACS Nano 2011, 5 (12), 9703–9709. (29) Liu, K.; Yan, Q.; Chen, M.; Fan, W.; Sun, Y.; Suh, J.; Fu, D.; Lee, S.; Zhou, J.; Tongay, S.; Ji, J.; Neaton, B.; Wu, J. Nano Lett. 2014, 14, 5097−5103. (30) Wan, K.-T.; Mai, Y.-W. Acta Met. mater. 1995, 43 (11), 4109–4115. (31) Boddeti, N. G.; Koenig, S. P.; Rong, L.; Xiao, J.; Bunch, J. S.; Dunn, M. L. J. Appl. Mech. 2013, 80 (4), 040909. (32) Deng, S.; Gao, E.; Xu, Z.; Berry, V. ACS Appl. Mater. Interfaces 2017, 9, 7812−7818. (33) Cao, Z.; Wang, P.; Gao, W.; Tao, L.; Suk, J. W.; Ruoff, R. S.; Akinwande, D.; Huang, R.; Liechti, K. M. Carbon 2014, 69, 390–400. 14 (34) Suk, J. W.; Na, S. R.; Stromberg, R. J.; Stauffer, D.; Lee, J.; Ruoff, R. S.; Liechti, K. M. Carbon 2016, 103, 63–72. (35) Zong, Z.; Chen, C.; Dokmeci, M. R.; Wan, K. J. Appl. Phys. 2010, 107, 026104. (36) Akinwande, D.; Brennan, C. J.; Bunch, J. S.; Egberts, P.; Felts, J. R.; Gao, H.; Huang, R.; Kim, J. S.; Li, T.; Li, Y.; Liechti, K. M.; Lu, N.; Park, H. S.; Reed, E. J.; Wang, P.; Yakobson, B. I.; Zhang, T.; Zhang, Y. W.; Zhou, Y.; Zhu, Y. Extrem. Mech. Lett. 2017, 13, 42–72. (37) Wang, P.; Liechti, K. M.; Huang, R. J. Appl. Mech. 2016, 83 (7), 071002. (38) Shull, K. R. Mater. Sci. Eng. 2002, 36, 1–45. (39) Israelachvili, J. N. Intermolecular and Surface Forces; 2011. (40) Chen, Y. L.; Helm, C. A.; Israelachvili, J. N. J. Phys. Chem. 1991, 95, 10736–10747. (41) Yu, Y.; Sanchez, D.; Lu, N. J. Mater. Res. 2015, 30 (18), 2702–2712. (42) Maugis, D.; Barquins, M. Appl. Phys 1978, 11, 1989–2023. (43) de Gennes, P. G. Rev. Mod. Phys. 1985, 57 (3), 827–863. (44) Kim, S.; Choi, G. Y.; Ulman, A.; Fleischer, C. Langmuir 1997, 7463 (4), 6850–6856. (45) Tian, K.; Gosvami, N. N.; Goldsby, D. L.; Liu, Y.; Szlufarska, I.; Carpick, R. W. Phys. Rev. Lett. 2017, 076103, 1–6. (46) Israelachvili, J.; Berman, A. Isr. J. Chem. 1995, 35, 85–91. (47) Qian, L.; Yu, B. Adhesion Hysteresis. Encyclopedia of Tribology; 2013; pp 29–32. (48) Shi, Z.; Lu, H.; Zhang, L.; Yang, R.; Wang, Y.; Liu, D.; Guo, H.; Shi, D.; Gao, H.; Wang, E.; Zhang, G. Nano Res. 2012, 5 (2), 82–87. (49) Liu, X.; Suk, J. W.; Boddeti, N. G.; Cantley, L.; Wang, L.; Gray, J. M.; Hall, H. J.; Bright, V. M.; Rogers, C. T.; Dunn, M. L.; Ruoff, R. S.; Bunch, J. S. Adv. Mater. 2014, 26, 1571– 1576. (50) Algara-Siller, G.; Lehtinen, O.; C.Wang, F.; Nair, R. R.; Kaiser, U.; A.Wu, H.; Geim, A. K.; Grigorieva, I. V. Nature 2015, 519, 443–447. (51) Dollekamp, E.; Bampoulis, P.; Poelsema, B.; Zandvliet, H. J. W.; Kooij, E. S. Langmuir 2016, 32, 6582−6590. 15  Fig. 1 Fig.1 a) Microscope image of a delaminated device (scale bar is 5μm). b) Device schematic. c) AFM image and d) AFM cross sections. e) Deflection δ and f) radius a plotted against input pressure p0. Inset microscope images show a device before and after snap-out (scale bar is 5μm). 16 Fig. 2 Fig. 2 a) Plots for CVD monolayer and bilayer devices (different symbols/colors represent each device), with linear fits (dashed lines) used to find E2D. b) E2D for each device in our exfoliated samples, and three of our CVD samples (N1-3) c) E2D divided by number of layers n for each sample. Data points and error bars represent the mean and standard deviation respectively for each sample. Results from nanoindentation measurements in references 16, 28 and 29 are plotted for comparison. Fig. 3 Fig.3 Work of separation of membranes of 1 to 3 layer thickness. The data includes measurements of CVD monolayer devices from three separate growths and transfers (N2-4). Several devices are measured per sample, with data points and error bars representing the means and standard deviations respectively. For samples with fewer than 3 measurements the data points represent each device measured. The dashed line marks the mean of the 6 samples. 17 Fig. 4 Fig. 4 a) AFM cross sections of a device during inflation (increasing N) and deflation (decreasing N). Arrows mark the snap transitions. b) δ and a of devices during inflation and deflation. Different colors represent different devices on sample N2. More data can be found in the supplementary information which is not shown here for reasons of clarity. Red and blue arrows mark snap-out and snap-ins respectively. The upper and lower dashed lines correspond to solutions to Eq. 4 and Eq. 5 respectively. c) A comparison between the works of separation and adhesion for samples N2-4. Data points and error bars represent the means and standard deviations respectively of all the devices measured on each sample. 18 Supporting Information Adhesion, Stiffness and Instability in Atomically Thin MoS2 Bubbles. David Lloyd1, Xinghui Liu2, Narasimha Boddeti2, Lauren Cantley1, Rong Long2, Martin L. Dunn3 and J. Scott Bunch1,4* 1Boston University, Department of Mechanical Engineering, Boston, MA 02215 USA 2University of Colorado, Department of Mechanical Engineering, Boulder, CO 80309 USA 3Singapore University of Technology and Design, Singapore, 487372 4Boston University, Division of Materials Science and Engineering, Brookline, MA 02446 USA *e-mail: [email protected] 19 1. Growth and characterization Devices were grown by chemical vapor deposition (CVD) according to a method described in an earlier paper1. The devices were transferred over the etched microcavities using a PMMA dry transfer method. Immediately prior to transfer the SiOX wafers were O2 plasma cleaned for 15 mins to remove any surface contamination. Before annealing off the PMMA layer at 340 oC, the devices were left in a vacuum desiccator for > 3 days to allow any gas trapped in the microcavities to leak out. Monolayers were identified by their optical contrast, and their Raman and photoluminescence (PL) spectra (Fig. S1). The separation between the E1 2g and A1g Raman modes was 20.3 cm-1, and the A exciton peak in the PL spectrum was located at 1.88 eV, which demonstrates that the membrane was single layered2,3. The E1 strain. 2g peak position is later used to determine the residual membrane Fig. S1 a) The Raman and b) PL spectrum of a suspended single layer MoS2 device with zero pressure difference across the membrane. 2. The effect of membrane pre-tension Even when there is no pressure difference across the membrane there is usually a residual pre-strain observed in suspended devices, due either to the transfer procedure or the membrane sticking to the sidewalls of the cavity4. We can estimate the pre-tension in our membranes by using photoluminescence spectroscopy. In an earlier paper1 we showed that the band-gap in monolayer MoS2 reduces when biaxial strain is applied, at a rate of - 20 99 meV/%. We took a PL spectrum of a device with no pressure difference across the membrane (Fig. S1), meaning any observed strain would correspond to the pre-strain. We can then convert this to a pre-tension using the formula5, 𝜎0 = 𝐸2𝐷𝜀0 1−𝜈 (S1) Our devices have a pre-strain of ε0 < 0.002 which corresponds to a pre-tension of σ0 < 0.2 N/m, which is comparable to previously reported values for atomically thin membranes in this geometry4,6. Campbell 1956 [5] showed that when the non-dimensional parameter, 𝑃 = 1/2 ∆𝑝𝑎𝐸2𝐷 3/2 𝜎0 (S2) satisfies the condition P > 100, Hencky's formula in Eq. 1 is correct to within 5%. Most of our data points were taken in a high enough pressure range to satisfy this condition. For instance for the data presented in Fig. 2a, P = 100 when Δp = 350 kPa. Since nearly all of our data was taken with Δp > 350 kPa we use Eq. 1 to calculate E2D, and neglect the effect of the pre-tension. 3. Work of separation The full set of data used to produce means and standard deviations of each sample in Fig. 3 of the main text is shown in Fig. S2. Each data point represents the measured value of Γsep for an individual device of a given sample. 21 Fig. S2. All Γsep data used to calculate means and standard deviations of each sample in Fig. 3. 4. Free energy model including adhesion hysteresis We can interpret the results described in Fig. 4 of the main text using the free energy model described in Eq. 1. Taking the derivative of F with respect to a, and substituting the pressure terms for the Hencky's result in Eq. 3 yields, 𝑑𝐹 𝑑𝑎 = 2𝜋𝑎 [𝛤 − 5 4 𝐶𝐾𝐸2𝐷 ( 4 𝛿 𝑎 ) ] (S3) Setting this formula equal to zero to find the radius at which the free energy is minimized leads to, 𝛤𝑠𝑒𝑝 = 5 4 𝐶𝐾𝐸2𝐷 ( 4 𝛿 𝑎 ) (S4) The constants C and K depend only on the Poisson's ratio ν, and their values for various 2D materials are tabulated in Table S1. 22  Poisson's Ratio ν MoS2 Graphene hBN Black Phosphorus 0.29 0.16 0.22 0.4 K(ν) 3.54 3.09 3.28 4.07 C(ν) 0.522 0.524 0.523 0.519 Table S1. Values for constants C(ν) and K(ν) for several 2D crystals, calculated using Hencky's solution. We plot the relationship described by Eq. S4 in Fig. S4a with a value of Γsep ~ 220 mJ/m2 and find our data fits this relationship very well. This formula is independent of whether N is increasing or decreasing, so when our devices are left to deflate we should expect δ and a to return along the same path as during inflation, and described by Eq. S4. We can explain the difference between inflation and deflation we see in our data as a result of adhesion hysteresis, whereby the energy required to separate the membrane from the surface Γsep is greater than the energy returned to the system as the membrane re- adheres Γadh, with Γadh < Γsep. For changes of the device radius Δa, we now have: 𝛤 = { 𝛤𝑎𝑑ℎ , ∆𝑎 < 0 𝛤𝑠𝑒𝑝, ∆𝑎 > 0 (S5) As the device inflates and Δa >0, the free energy of the system is minimized according to 1/4. When deflating the radius of the device will only Eqs. 4 and 5, with δ0/ap ~ Γsep decrease when dF/da >0 for Δa <0 (with Γ=Γadh), in order for the free energy to be minimized. From examining Eq. S3 and considering that Γadh < Γsep, this will only occur when δ has decreased from δ0 to below the critical value of δ = δc after which the device radius can reduce in the form of a snap-in transition. Since the radius cannot decrease until δc is reached, the bubble edge remains pinned at ap. The critical deflection δc marks the point where dF/da = 0 for Δa <0 (i.e. Γ=Γadh), and from using Eq. S3 we can see that this occurs when the relationship, 𝛤𝑎𝑑ℎ = 5 4 𝐶𝐾𝐸2𝐷 ( 4 ) 𝛿𝑐 𝑎𝑝 (S6) is satisfied. This corroborates with what we see in Fig S3a, in which the value of δc is roughly proportional to ap for the devices measured. We can estimate the value of Γadh by 23 fitting this relationship to the values of δc and ap of devices just before the snap in transition occurs, and we plot this line of best fit in Fig. S3a which corresponds to Γadh ~ 14 mJ/m2. We checked the repeatability of our measurements of Γadh by repeating the experiment 6 times on a single device, which resulted in a mean and standard deviation of 13 mJ/m2 and 5 mJ/m2 respectively (Fig. S3b). These arguments are best seen graphically in terms of the free energy landscape plotted as a function of radius in Fig S4b, c and d. In the absence of adhesion hysteresis, as the pressure inside the device decreases and the devices deflate, the free energy minima moves to a smaller radius (Fig. S4b). The path taken by our devices is shown in Fig. S4c, and clearly shows the devices not following the local minima in the free energy. By introducing adhesion hysteresis into the model (Fig. S4d), ΔF is calculated using Γsep for Δa >0 and Γadh for Δa <0, which results in the device radius remaining trapped in a local minima as the device deflates. The radius only changes when dF/da >0 for Δa <0 which only happens when Eq. S6 (Eq. 5 in the main text) is satisfied. Fig. S3 a) Data for devices measured on sample N2, showing the values of δc and ap just before snap-in used to calculate Γadh. Each color/symbol represents a different device. b) Multiple measurements of a single device at a number of different pressures showing repeatability. Dashed line represents the mean adhesion value. 24 Fig. S4 a) As devices delaminate the ratio δ/a remains constant according to Eq. S4. b) The free energy landscape if there is no adhesion hysteresis. The device radius is that which minimizes the free energy, and the grey dots mark the path we would expect the device to take. c) The actual path our devices take, which appears to not minimize the free energy. d) The modified free energy landscape if Γadh < Γsep. As the device reduces its radius its free energy is determined by the dashed lines. The device is now trapped in a free energy minima and snap-in only occurs when the gradient of the dashed line is greater than zero. To see if the work of adhesion varied between samples fabricated with the same method of CVD growth and transfer, we performed measurements of 5 different CVD samples (N2-6) with at least 4 devices measured per sample. Monolayer devices were delaminated and left to deflate, and AFM measurements of δc and ap taken just before snap-in were used to calculate the work of adhesion using Eq. S6. The data is presented in Fig. S5a, with each data point representing a measurement of Γadh in a single device of a given sample. The mean and standard deviations of each sample are shown in Fig. S5b. Between different samples there is considerable variation in the mean work of adhesion, 25 which suggests that factors such as the cleanliness of substrate or membrane which can vary from sample to sample may play significant roles in adhesion hysteresis. A few of the devices measured did not snap in completely from radius ap to a0, but rather initially snapped in to an intermediate radius followed by a second snap in to a0 (Fig. S9b). All the transitions between these states were unstable and occurred in less than one second. Fig. S5 a) Work of adhesion for every device measured in each sample. b) Mean and standard deviations of the work of adhesion in each sample. The dashed line represents the mean of the 5 samples. 5. Contact angle of bubbles during deflation Instead of analyzing the snap-out and snap-in data in terms of δ and a, an analogous method is to measure the contact angle θc between the membrane and the substrate (see Fig. S6 inset) using an AFM. In Fig. S6 we plotted the contact angle against the radius of a device as it is inflated (black) and then left to deflate (red). As the device is inflated the contact angle increases until a critical value, at which point the device delaminates with the contact angle remaining constant. When the device is left to deflate the contact angle 26 decreases at constant radius until another critical contact angle is reached, at which point the device undergoes the snap-in transition. Fig. S6. The contact angle of a device during inflation (black) and deflation (red). 6. Strain trapping around the edge of the membrane To investigate a possible mechanism for the observed adhesion hysteresis we used Raman spectroscopy to measure the strain distribution around our devices. The peak positions of the Raman modes in monolayer MoS2 are known to be sensitive to strain1,7, so by measuring how these peaks shift at different locations around the device we can build up an image of how strain is distributed. For these measurements we used the E1 estimate the strain (Fig. S1), since it has a peak position which is strain sensitive and independent of doping effects. 2g peak to Fig. S7a shows an AFM image of a device delaminated to ap ~ 7.5 μm, which was then left to deflate and undergo the snap-in transition. A Raman map was then taken after snap-in (Fig S7b), with the strain calculated from the position of the E1 2g peak using the reported shift rate of ~ 5 cm-1 / %1,7. A region of ε ~ 0.5% can be clearly seen around the circumference of where the delaminated bubble was before snap-in. This strain likely originates from the pressure induced radial strain at the edge of the bubble, which for 27 these devices is ~1.5% (Fig. S8d). Using this upper bound of ε ~ 1.5% and the formula for the isotropic strain membrane energy density8, U = ½ E2D ε2, we can estimate the energy stored in the strained regions to be U ~ 20 mJ / m2, which can account for some but not all the energy dissipation which produces a difference between Γadh and Γsep. The presence of strain in the membrane also implies some contribution of energy dissipation through friction as the membrane changes its length on the surface of the substrate9. Fig. S7 a) AFM image (amplitude channel) of a delaminated device before the snap-in transition. The position of the microcavity is marked by a dashed circle. Below is a cross section of the device. b) Strain map of the same device after the snap-in transition when the device has fully deflated. Strain is calculated using the peak shift in the E1 2g Raman mode at each point. Each pixel is 1 x 1 μm and corresponds to a single Raman scan. 2g peak position as a function of distance (Fig. S8a and S8b). In order to observe the process by which this strain becomes 'trapped' in the membrane around the device, we took Raman line scans over a cross section of a device as it deflated and plotted the E1 Before each Raman scan we found the corresponding geometry of the device by taking an AFM image (Fig. S8c). Across the delaminated bubble region (marked by dashed lines) the peak shift abruptly increases at the edge of the bubble, followed by a gradual increase towards the center of the device. In Fig. S8d we used Hencky's solution to find the predicted strain profile across the device for its initial geometry (Fig. S8c red line) before deflation. In the model, the strain jumps from zero to purely radial tensile strain at the edge of the device, with the tangential component gradually increasing from zero to be 28 equal to the radial component at the center. The E1 contributions of both the radial and tangential strain, so this model explains the profile seen in Fig. S8a. 2g peak position depends on Fig. S8b shows that a region of strain extends ~1.5 μm outside the edge of the bubble in the initial Raman scan (red line). As the device deflates and the radius remains pinned the peak shift across the delaminated region of the membrane reduces as it becomes less strained, however the region of strain outside bubble remains roughly constant throughout deflation. These results show that the ring of strain in Fig. S7b is formed when the device initially delaminates, and that this strain does not relax as the device deflates and eventually snaps in. Fig. S8 a) Raman line scans over a device over time as it deflates. Dashed vertical lines mark the edge of the delaminated bubble. b) A zoomed in version of a) focusing on the edge region of the device. c) AFM cross sections of the device at each time, using the same color scheme as in a). d) Radial (εr) and tangential (εθ) components of the strain as a function of radius for this device's 29 initial geometry before deflating, calculated using Hencky's model with values for δ and a taken from the red curve in c). 7. The effect of the slipping of the membrane on E2D calculations The strain at the edge of the bubble introduces extra slack into the membrane of bubble, which may affect our measurements of E2D. We can estimate the effect this has on our measurements by integrating the strain over the strained region at the edge of the bubble in Fig. S8b to find the total extra slack, ∆𝐿, added to the bubble membrane. We can write the slack added to the membrane as, ∆𝐿 = ∫ 𝜀(𝑥)𝑑𝑥 𝑥1 0 (S7) The initial measurement in Fig. S8b (red line color and labeled '87 mins') shows that the peak shift linearly decreases from ~ 5.5 cm-1 around the edge of the device to ~0 cm-1 at 1.5 μm outside the device radius, so we take x1 = 1.5 μm. To find 𝜀(𝑥) we take ε ~ 1.5 % at the edge of the device (Fig. S8d) and use the linear strain profile seen in Fig. S8b, which leads to ε(x) ~( 0.015/1.5) x μm-1. This gives ΔL ~ 11 nm over a device radius of 6.5 μm. This reduces the pre-strain by ~ 0.0017 which is about the same as the initial pre- strain. We therefore take this change to be negligible in to the pressure range we are studying due to the arguments made in section 2. 8. Stable delamination devices Devices of well depth d ~ 650 nm were fabricated that exhibited stable delamination (Fig. S9). These devices showed the same hysteric behavior as our other devices. To calculate the work of separation of these devices we used Eq. 4 in the main text with AFM measurements of δ and a, and used the mean E2D of all our CVD devices of 128 N/m. We measured 3 devices over 4 different pressures, and found a value of Γsep = 207 ± 19 mJ/m2. We also measured the work of adhesion of the device shown in Fig. S9b, which we found to be Γadh = 40 mJ/m2. 30 Fig. S9 a) Stable delamination with increasing pressure. b) A device which delaminates stably with increasing pressure, but shows adhesion hysteresis upon deflation. This device snapped in to an intermediate step before fully re-laminating to the substrate. 9. Additional snap-in data Fig. S10 shows the complete data set for our snap-in measurements presented in Fig. 4b of the main text. This data was taken using an AFM in tapping mode. To confirm that the forces from to the AFM tip were not affecting our results, we measured the snap-in of a device as it deflated by using solely optical measurements. We took sequential PL spectra at the center of the device as it deflated, where the membrane is under biaxial strain. In an earlier paper1 we found that the PL peak red-shifts under biaxial strain by -99 meV/%, so PL measurements allow us to measure the biaxial strain ε in the device. We can also measure the radius a of the device as it deflates using an optical microscope. Using these values for a and ε we can estimate the deflection of the device using the formula, 𝜀 = 𝜎(𝜈) ( 2 𝛿 𝑎 ) (S8) where σ(ν) is a numerical constant which depends only on Poisson's ratio ν , and in this case 𝜎 = 0.709. We measured a deflating device using the non-contact optical method, after which we re-inflated the device to the same pressure and used the AFM to measure the geometry of the device as it deflated. We compare the results of these two methods in Fig. S10b, and find very similar results in the two cases. The device appears to snap-in at a slightly lower δ in the AFM measurements, however this is likely due to the long scan 31 times (~3 min) required to take a PL spectrum meaning that we couldn't measure the device right at the moment before snap-in. Fig. S10 a) Complete data containing all data points of results presented in Fig. 4b in the main text. Each color represents a different device. b) Comparison of snap-in transitions measured optically or by AFM. For optical measurements a is determined using an optical microscope with a 100x objective, and δ is determined from the PL peak position and Eq. S8. 10. Young's modulus Fig. S11 shows the complete data set used to calculate the Young's modulus for each device in Fig. 2b in the main text. 32 Fig. S11 a) CVD monolayer devices from sample N1 and N2. b) Exfoliated monolayer and trilayers devices. Dashed lines are plotted for each of the sample means reported in Fig. 2c of the main text. Different color/symbols represent different devices. 11. Videos of snap transitions Video 1 shows the snap-in transition of a deflating device taken with a high speed camera. The snap-in transition occurs faster than the frame rate of the camera (0.5 ms). Video 2 shows a device in a pressure chamber with a quartz window, allowing us to observe a delaminated device as the chamber pressure is increased and decreased (video speed is 4x). For the first half of the video the external pressure is increased, with the delaminated device snapping-in at ~6 s. During the second half of the video the pressure is decreased, with the device snapping-out at ~30 s. References (1) Lloyd, D.; Liu, X.; Christopher, J. W.; Cantley, L.; Wadehra, A.; Kim, B. L.; Goldberg, B. B.; Swan, A. K.; Bunch, J. S. Nano Lett. 2016, 16, 5836−5841. (2) Lee, C.; Yan, H.; Brus, L. E.; Heinz, T. F.; Hone, J.; Ryu, S. ACS Nano 2010, 4 (5), 2695–2700. (3) Mak, K. F.; Lee, C.; Hone, J.; Shan, J.; Heinz, T. F. Phys. Rev. Lett. 2010, 105 (13), 2–5. (4) Bunch, J. S.; Verbridge, S. S.; Alden, J. S.; Van Der Zande, A. M.; Parpia, J. M.; Craighead, H. G.; McEuen, P. L. Nano Lett. 2008, 8 (8), 2458–2462. (5) Campbell, J. D. Quart. Journ. Mech. Appl. Math 1956, IX, 84–93. (6) Wang, L.; Travis, J. J.; Cavanagh, A. S.; Liu, X.; Koenig, S. P.; Huang, P. Y.; George, S. M.; Bunch, J. S. Nano Lett. 2012, 12, 3706–3710. (7) Conley, H. J.; Wang, B.; Ziegler, J. I.; Haglund, R. F.; Pantelides, S. T.; Bolotin, K. I. Nano Lett. 2013, 13 (8), 3626–3630. (8) Gould, P. L. Introduction to Linear Elasticity 3rd ed.; 2013. (9) Kitt, A. L.; Qi, Z.; Remi, S.; Park, H. S.; Swan, A. K.; Goldberg, B. B. Nano Lett. 2013, 13, 2605−2610. 33 

1501.01877

1

1501

Visible Light Assisted Gas Sensing with TiO2 Nanowires

Sensing response of individual single-crystal titania nanowires configured as chemiresistors for detecting reducing (CO, H2) and oxidizing (O2) gases is shown to be sensitive to visible light illumination. It is assumed that doping of the TiO2 nanowires with C and/or N during carbon assisted vapor-solid growth creates extrinsic states in the band gap close to the valence band maximum, which enables photoactivity at the photon energies of visible light. The inherently large surface-to-volume ratio of nanowires, along with facile transport of the photo-generated carriers to/from the nanowires surface promote the adsorption/desorption of donor/acceptor molecules, and therefore open the possibility for visible light assisted gas sensing. The photo-catalytic performance of TiO2 nanowire chemiresistors demonstrates the prospect of combining light harvesting and sensing action in a single nanostructure.

cond-mat.mes-hall

cond-mat

Visible Light Assisted Gas Sensing with TiO2 Nanowires Jie Zhang1, Evgheni Strelcov1*†, and Andrei Kolmakov1** 1Department of Physics, Southern Illinois University Carbondale, Illinois, USA Abstract: Sensing response of individual single-crystal titania nanowires configured as chemiresistors for detecting reducing (CO, H2) and oxidizing (O2) gases is shown to be sensitive to visible light illumination. It is assumed that doping of the TiO2 nanowires with C and/or N during carbon assisted vapor-solid growth creates extrinsic states in the band gap close to the valence band maximum, which enables photoactivity at the photon energies of visible light. The inherently large surface-to-volume ratio of nanowires, along with facile transport of the photo-generated carriers to/from the nanowire’s surface promote the adsorption/desorption of donor/acceptor molecules, and therefore open the possibility for visible light assisted gas sensing. The photo-catalytic performance of TiO2 nanowire chemiresistors demonstrates the prospect of combining light harvesting and sensing action in a single nanostructure. Introduction: Titania (TiO2) is a well-known photocatalyst used in water photoelectrolysis applications for converting sunlight into a chemical fuel (hydrogen), for conversion of CO2 to hydrocarbon fuels, for photodegradation of toxic molecules to simple harmless products, self-cleaning glass and fabric,1, 2 photo-induced superwetting,3 etc.4, 5 However, *Present address: Center for Nanophase Materials Sciences, Oak Ridge National Laboratory, Oak Ridge, Tennessee ** Present address: Center for Nanoscale Science and Technology, NIST, Gaithersburg, MD †Corresponding author: [email protected] 1 pristine TiO2 is a wide band gap semiconductor with the band gap in excess of 3 eV. Its photocatalytic activation therefore requires ultraviolet irradiation that comprises only a small fraction (~5%) of the solar constant.6 Thus, it is desirable to develop a TiO2-based photocatalyst active in the visible light (45% of the sun’s radiation energy).7 Considerable efforts have been made recently to extend the photoresponse of the TiO2 systems farther into the visible-light region via compositional doping with a wide range of transition-metal cations.8, 9 These works revealed a disadvantage of the cationic dopants: they can result in localized d-electronic levels deep in the band gap of TiO2, which serve as recombination centers for photogenerated charge carriers. A large part of the existing literature agrees that anionic nonmetal dopants, such as carbon,10 sulfur,11 and nitrogen,12 are more appropriate for extending the TiO2 photocatalytic activity into the longer wavelength region. Accordingly, the reported doping methods include chemical approaches, such as sol-gel reaction synthesis, electrochemical doping and oxidation of titanium nitride, as well as physical methods such as magnetron co- sputtering and ion implantation.9, 13, 14 Historically, photo-assisted metal oxide gas sensors have been known for a long time.15-17 Pristine TiO2 nanowires (NWs), having a high surface-to-volume ratio and photocatalytically active in the UV region, are one of the most studied platforms for detection of reducing gases. In this report, TiO2 nano- and meso-wires responsive to visible light were synthesized via vapor-solid (VS) technique with unintentional carbon and/or nitrogen doping during the carbothermal evaporation. Experimental 2 N-type TiO2 nanowires 50-1000 nm wide and hundred microns long were synthesized via vapor-solid carbothermal method using a slightly modified protocol described previously.18-20 Namely, a precursor mixture of TiO2 and graphite powders (1:1 by volume) was heated in a tube furnace at 1100 °C in an Ar carrier gas atmosphere (30 sccm, 200 Torr) containing traces of air at 10-2 Torr. Pure Ti (99.99%) polycrystalline samples placed downstream in the 750~800 °C zone served as collector substrates. Ti surface becomes oxidized by residual oxygen during the temperature ramp. Newly formed micron thick polycrystalline TiO2 skin provides plenty of nucleation sites for preferential 1D titania NW growth (Fig 1a). The morphology and crystalline structure of the grown TiO2 NWs were characterized by Scanning Electron Microscopy (SEM) and X-Ray Diffraction analysis (XRD), as shown in Figure 1. The nanowires collected from the substrates were placed on the oxide side of a Si/(300 nm)-SiO2 wafer. Ti (20 nm)/Au (200 nm) micro-pad contacts were thermal-vapor deposited through a shadow mask (Fig. 2 inset). The Si substrate (p-doped) back gate electrode was grounded during the measurements. 3 Figure 1. a) SEM micrograph of the synthesized TiO2 nanostructures; b) XRD pattern of nanowires reveals their rutile structure. The photocatalytic and gas sensing measurements on nanostructures were conducted in a custom-made high vacuum variable-temperature probe station equipped with a microheater. Prior to measurements, the sample was UV cleaned and annealed in a vacuum at 10-6 Torr for approximately 30 min at 350 °C. Pure oxygen (10-4 Torr) was admitted to the vacuum chamber as a background gas mimicking oxidizing conditions of the ambient air. Computer-controlled solenoid valve pulses introduced reducing gases H2 (4·10-4 Torr) and CO (2·10-4 Torr) into the chamber. The resultant changes in the source- drain current were measured as a function of time at a bias VDS = 6 V. For photocatalytic measurements, a halogen lamp with a total light flux to the sample area of less than 0.2 4 W/cm2 was used. Optical filters were added to block the ultraviolet and infrared spectrum regions of the light flux, eliminating artifacts related to the near UV and IR light components. Response of the nanostructures to reducing gases was measured as a function of the light illumination. A similar device fabricated from a SnO2 individual nanowire was measured in the same conditions as the TiO2 nanostructure for comparison. Results and Discussion The post-annealed crystalline structure of TiO2 NWs as determined from the XRD pattern of Figure 1c is rutile. This polymorph has superior light scattering properties beneficial from the perspective of effective light harvesting, as compared with anatase.21 Figure 2. I-V curves of the TiO2 nanowire device: The blue curve was recorded in the dark; the red curve shows electrical behavior under illumination with higher conductivity. The inset presents an SEM micrograph of the device. The electrical behavior of titania NWs is shown in Figure 2. The blue and red IV curves were obtained in the absence and presence of illumination, respectively. Both of the curves appear to be non-linear, presumably due to the presence of a Schottky barrier at 5 the nanowire-electrode contact. The IV curve recorded under illumination shows higher conductivity and deviates less from linearity than the one recorded in the dark. Several mechanisms can be responsible for this behavior. Firstly, visible-light assisted photoexcitation of electrons can be due to the existence of the filled impurity levels within the bandgap. These impurity levels become partially ionized and contribute new carriers in to conduction band under irradiation conditions. The electric field induced at the Schottky contact and initial band banding near nanowire surface both facilitate spatial separation of the hot electron-hole pairs and depresses the recombination rate, thus leading to a reduction of the Schottky barrier, band flattening and facilitated transport. Alternatively, energy of the visible light can be sufficient to induce photoinjection of electrons from the metallic electrode into the nanowire, which likewise increases overall conductivity. The gas sensing properties of the fabricated TiO2 NW microsensor were measured at 350 °C. At this elevated temperature the introduced background oxygen gas molecules easily dissociate and chemisorb onto the surface at the vacancy sites containing localized electrons and form O- ions, according to:16 𝑂2(𝑔𝑎𝑠) → 𝑂2(𝑎𝑑𝑠) (1) 𝑂2(𝑎𝑑𝑠) + 𝑒− → 𝑂2(𝑎𝑑𝑠) − (2) − 𝑂2(𝑎𝑑𝑠) + 𝑒− → 2𝑂(𝑎𝑑𝑠) − (3) Concurrently, energy bands bend upward decreasing the electron concentration in the near surface region and electron depletion layer is formed near the semiconductor surface. 6 A typical response of the TiO2 NW device to admission of CO (2·10-4 Torr) in to oxygen reach backround is shown in Figure 3. The lower blue curve was recorded with light “OFF”, and the upper yellow ones were recorded under the visible light illumination. As can be seen, the overall background current as well as gas sensing response are elevated during illumination compared to the dark values similar to the IV curve data of Figure 2. In order to compare the gas responses (ΔI) for both conditions more clearly, the background currents (I0) were subtracted from the measured total current in the Figure 4. When CO was admitted to the chamber, a distinguishable decrease in nanowire resistance was observed within several seconds (Fig. 4). The NW’s resistance recovered back to its original level upon interrupting CO inflow. The red and blue curves of Figure 4 show sensor’s response to CO with and without light illumination, respectively. Approximately 10% higher response was achieved with the current intensity of visible light irradiation, as compared to that in the dark environment, despite the fact that the rise and recovery times are similar for both conditions. The TiO2 NW response to H2 (4·10-4 Torr) exhibits similar result with approximately 25% higher response under illumination than in the dark, but again without a change in the response and recovery times (Fig 4b). These results demonstrate that the visible light photo activated TiO2 NW is a good candidate material for photocatalysis and gas sensing. 7 Figure 3. Response of individual TiO2 NW to carbon monoxide (4·10-4 Torr pulses) in the dark (blue curves) and under illumination (yellow curves) as measured at 350 °C. There exist a few mechanisms that could explain these results. Herein, two scenarios are considered. First, the photo-assisted chemical sensing mechanism of a semiconductor material is discussed. The CO and H2 molecules react with the adsorbed oxygen on the nanowire surface: 22 𝑂− + 𝐶𝑂(𝑎𝑑𝑠) → 𝐶𝑂2(𝑑𝑒𝑠) + 𝑒− (5) 𝑂− + 𝐻2(𝑎𝑑𝑠) → 𝐻2𝑂(𝑑𝑒𝑠) + 𝑒− (6) This process releases captured electrons back into the nanowire’s surface leading to a reduction in the depletion layer width and a drop in the sensor’s resistance. Carbon and/or Nitrogen doped TiO2 has donor impurities levels in the bandgap, which can be photon excited by visible light irradiation to produce electron-hole pairs on the surface. The flattening of the bands promotes delivery of the photoexited electrons to the surface, where they can contribute to reduction of the preadsorbed oxygen species following Eq. (3). Thus, photoabsorption of the visible light increases the preadsorbed density of ionic 8 oxygen (O-) at the TiO2 nanowire surface, providing more active sites for further oxidation of CO and H2. Figure 4. Response of individual TiO2 NW to a) carbon monoxide and b) hydrogen (4·10-4 Torr pulses) in the dark and under illumination as measured at 350 °C. Another plausible mechanism – Schottky diode – corroborates with previous studies on gas detection by metal-semiconductor structures (e.g. Au-carbon nanotube).23-25 As the reducing gas molecules (H2, CO) adsorb on the Au/TiO2 interface, the work function of Au electrode is reduced, leading to a decrease in Schottky barrier (SB) for carrier injection to TiO2. Consequently, the electrons that were either photogenerated on the 9 dopant levels or photoexcited in the metal, can overcome the barrier, increasing the source-drain current:24 𝐼𝐷𝐶~𝑇2𝑒− 𝑞𝜙𝐵 𝑘𝑇 (7), where ϕB is the Schottky barrier height, k is the Boltzmann constant, T is the absolute temperature, and q is the elementary charge. In this case, the sensitivity of the SB modulation24 is given by 𝑆 = Δ𝑅 𝑅0 𝑞∆𝜙𝐵 𝑘𝑇 − 1 (8) ≈ 𝑒 Thus, because of the exponential dependence of the response on the changes in the SB height, a very high sensitivity can be achieved (as one shown in Fig. 4a and 4b). Therefore, the gas sensing response could be promoted by photoactivity on the TiO2 NW surface in the visible light region. Figure 5. Response of individual SnO2 NW to hydrogen (4·10-4 Torr pulses) in the dark and under illumination (yellow region) as measured at 350 °C. 10 SnO2 18 and ZnO could be a suitable alternative to TiO2. A microsensor from an individual SnO2 NW was fabricated in the same configuration and covered with contact micropads similar to those used for the TiO2 NW device. The response of the SnO2 NW to H2 (2·10-4 Torr) was measured at 350 °C both in the dark and under visible light illumination. It appeared that its response is insensitive to illumination (Fig. 5), which gives less support to the proposed Schottky diode mechanism. Conclusion: Electrical measurements on individual TiO2 single crystal nanowires imply that light induced electron-hole pair formation contribute to the increased electrical conductivity under visible light illumination. Furthermore, gas sensing (CO, H2) measurements taken under both visible light irradiation and in the dark, indicate photo-activated chemical oxidization on the surface of TiO2 nanowires. However, the precise mechanism of photo- assisted gas sensing is still unclear. It is speculated that TiO2 single crystal nanowires were unintentionally doped with C and/or N during thermal growth process, which helped to expand their photoactivity to visible light. Future efforts will be focused on modifying the synthesis protocol to produce nanowires with higher sensing efficiency and controlled doping level. Acknowledgements: We would like to thank Dr. Y. Lilach for writing data acquisition LabView codes, Ms. K. Winspec for assistance with NW fabrication, and Mr. C. Watt for hardware setup. 11 References: Bozzi, A.; Yuranova, T.; Kiwi, J. J. Photochem. Photobiol. A-Chem. 2005, 172, Khan, S. U.; Al-Shahry, M.; Ingler, W. B. Science 2002, 297, (5590), 2243-2245. Umebayashi, T.; Yamaki, T.; Itoh, H.; Asai, K. Applied Physics Letters 2002, 81, Diwald, O.; Thompson, T. L.; Zubkov, T.; Goralski, E. G.; Walck, S. D.; Yates, J. T. Sclafani, A.; Herrmann, J.-M. Journal of Photochemistry and Photobiology A: Choi, W.; Termin, A.; Hoffmann, M. R. The Journal of Physical Chemistry 1994, Fujishima, A.; Honda, K. Nature 1972, 238, (5358), 37-38. Fujishima, A.; Rao, T. N.; Tryk, D. A. Journal of Photochemistry and Fujishima, A.; Zhang, X. T. C. R. Chim. 2006, 9, (5-6), 750-760. Lee, K.; Kim, Q.; An, S.; An, J.; Kim, J.; Kim, B.; Jhe, W. Proceedings of the 1. (1), 27-34. 2. 3. National Academy of Sciences 2014, 111, (16), 5784-5789. 4. 5. Photobiology C: Photochemistry Reviews 2000, 1, (1), 1-21. Diebold, U. Applied physics A 2003, 76, (5), 681-687. 6. 7. Sun, H.; Bai, Y.; Cheng, Y.; Jin, W.; Xu, N. Industrial & engineering chemistry research 2006, 45, (14), 4971-4976. 8. Chemistry 1998, 113, (2), 181-188. 9. 98, (51), 13669-13679. 10. 11. (3), 454-456. 12. The journal of physical chemistry B 2004, 108, (19), 6004-6008. 13. Energy Reviews 2007, 11, (3), 401-425. 14. 15. 78, (1–3), 73-77. 16. 2003, 4, 48-54. 17. Chemical 2000, 65, (1–3), 260-263. 18. 2005, 5, (4), 667-673. 19. 2019-2022. 20. Materials Chemistry 2004, 14, (4), 440-450. 21. 1042-1048. 22. (37), 12804-12807. 23. 1003-1009. Yang, T.-Y.; Lin, H.-M.; Wei, B.-Y.; Wu, C.-Y.; Lin, C.-K. Rev. Adv. Mater. Sci Comini, E.; Cristalli, A.; Faglia, G.; Sberveglieri, G. Sensors and Actuators B: Kolmakov, A.; Klenov, D.; Lilach, Y.; Stemmer, S.; Moskovits, M. Nano Letters Lilach, Y.; Zhang, J.-P.; Moskovits, M.; Kolmakov, A. Nano letters 2005, 5, (10), Zhang, Z.; Yates Jr, J. T. Journal of the American Chemical Society 2010, 132, Ruths, P. F.; Ashok, S.; Fonash, S. J. IEEE Trans. Electron Devices 1981, 28, (9), Ni, M.; Leung, M. K.; Leung, D. Y.; Sumathy, K. Renewable and Sustainable Zaleska, A. Recent Patents on Engineering 2008, 2, (3), 157-164. Comini, E.; Faglia, G.; Sberveglieri, G. Sensors and Actuators B: Chemical 2001, Rao, C.; Gundiah, G.; Deepak, F. L.; Govindaraj, A.; Cheetham, A. Journal of Lee, D.-U.; Jang, S.-R.; Vittal, R.; Lee, J.; Kim, K.-J. Solar Energy 2008, 82, (11), 12 Peng, N.; Zhang, Q.; Chow, C. L.; Tan, O. K.; Marzari, N. Nano letters 2009, 9, Hu, Y.; Zhou, J.; Yeh, P.-H.; Li, Z.; Wei, T.-Y.; Wang, Z. L. Advanced Materials 24. (4), 1626-1630. 25. 2010, 22, (30), 3327-3332. 13 

1704.02838

4

1704

Chirality blockade of Andreev reflection in a magnetic Weyl semimetal

A Weyl semimetal with broken time-reversal symmetry has a minimum of two species of Weyl fermions, distinguished by their opposite chirality, in a pair of Weyl cones at opposite momenta $\pm K$ that are displaced in the direction of the magnetization. Andreev reflection at the interface between a Weyl semimetal in the normal state (N) and a superconductor (S) that pairs $\pm K$ must involve a switch of chirality, otherwise it is blocked. We show that this "chirality blockade" suppresses the superconducting proximity effect when the magnetization lies in the plane of the NS interface. A Zeeman field at the interface can provide the necessary chirality switch and activate Andreev reflection.

cond-mat.mes-hall

cond-mat

Chirality blockade of Andreev reflection in a magnetic Weyl semimetal N. Bovenzi,1 M. Breitkreiz,1 P. Baireuther,1 T. E. O'Brien,1 J. Tworzyd(cid:32)lo,2 I. Adagideli,3 and C. W. J. Beenakker1 1Instituut-Lorentz, Universiteit Leiden, P.O. Box 9506, 2300 RA Leiden, The Netherlands 2Institute of Theoretical Physics, Faculty of Physics, University of Warsaw, ul. Pasteura 5, 02–093 Warszawa, Poland 3Faculty of Engineering and Natural Sciences, Sabanci University, Orhanli-Tuzla, 34956 Istanbul, Turkey (Dated: April 2017) A Weyl semimetal with broken time-reversal symmetry has a minimum of two species of Weyl fermions, distinguished by their opposite chirality, in a pair of Weyl cones at opposite momenta ±K that are displaced in the direction of the magnetization. Andreev reflection at the interface between a Weyl semimetal in the normal state (N) and a superconductor (S) that pairs ±K must involve a switch of chirality, otherwise it is blocked. We show that this "chirality blockade" suppresses the superconducting proximity effect when the magnetization lies in the plane of the NS interface. A Zeeman field at the interface can provide the necessary chirality switch and activate Andreev reflection. I. INTRODUCTION Spin-momentum locking is a key feature of topological states of matter: In both topological insulators and topo- logical semimetals the massless quasiparticles are gov- erned by a Hamiltonian H± = ±vFp·σ that ties the direc- tion of motion to the spin polarization.1–4 In a topologi- cal insulator the ± sign distinguishes spatially separated states, e.g., the opposite edges of a quantum spin-Hall insulator along which a spin-up electron moves in oppo- site directions.5 In a topological semimetal the ± sign distinguishes Weyl cones in the band structure. A mag- netic Weyl semimetal has the minimum number of two Weyl cones centered at opposite points ±K in the Bril- louin zone, containing left-handed and right-handed Weyl fermions displaced in the direction of the magnetization.6 It is the purpose of this paper to point out that the switch in chirality between the Weyl cones forms an ob- stacle to Andreev reflection from a superconductor with conventional, spin-singlet s-wave pairing, when the mag- netization lies in the plane of the normal-superconductor (NS) interface. The obstruction is illustrated in Fig. 1. Andreev reflection is the backscattering of an elec- tron as a hole, accompanied by the transfer of a Cooper pair to the superconductor. For a given spin band and a given Weyl cone, electrons and holes move in the same direction,7 so backscattering must involve either a switch in spin band (σ (cid:55)→ −σ) or a switch in Weyl cone (K (cid:55)→ −K), but not both. This is at odds with the requirement that zero spin and zero momentum is transferred to the Cooper pair. This "chirality blockade" of Andreev reflection is spe- cific for the conical dispersion in a Weyl semimetal, and it does not appear in other contexts where spin-momentum locking plays a role. In a quantum spin-Hall insulator, there is no need to switch the chirality because the hole can be reflected along the same edge as the incident electron.8 There is a formal similarity with graphene,9 where Andreev reflection switches between valleys at ±K, but there σ is an orbital pseudospin and the real spin is not tied to the direction of motion. FIG. 1: Andreev reflection (AR) from a superconductor in a quantum spin-Hall insulator (top panel) and in a Weyl semimetal (bottom panel.) The red and blue wedges desig- nate electron and hole quasiparticles (Weyl fermions) moving towards or away from the interface (solid versus dashed ar- rows indicate v in the ±x direction). The orientation of the wedge distinguishes the polarization σ = ±1 of the spin band and the color indicates the chirality C = sign (vσ). Andreev reflection switches σ and v, which is blocked if it must also switch C. We will show that the chirality blockade can be lifted by breaking the requirement of zero-spin transfer with a Zeeman field. We also discuss the subtle role played by inversion symmetry, by contrasting a scalar with a pseudoscalar pair potential.10 The absence of the chiral- ity blockade for pseudoscalar pairing explains why it did not appear in the many previous studies of Andreev re- flection in a Weyl semimetal.11–19 The outline of this paper is as follows. In the next sec- tion, we introduce the model of an NS junction between a Weyl semimetal and a conventional superconductor. The 8 × 8 Bogoliubov-De Gennes Hamiltonian is block- diagonalized in Sec. III, after which the chirality blockade of Andreev reflection is obtained in Sec. IV. In the next section V, we show how to remove the blockade by a spin- 7 1 0 2 l u J 7 2 ] l l a h - s e m . t a m - d n o c [ 4 v 8 3 8 2 0 . 4 0 7 1 : v i X r a active interface or by an inversion-symmetry breaking in- terface. As an experimental signature, we calculate the conductance of the NS junction in Sec. VI. To eliminate the effects of a lattice mismatch, we consider in Sec. VII the NS junction between a Weyl semimetal and a Weyl superconductor - which shows the same chirality block- ade for a scalar spin-singlet pair potential. More general pairing symmetries (spin-triplet and pseudoscalar spin- singlet) are considered in an Appendix. The Josephson effect in an SNS junction is studied in Sec. VIII. We con- clude in Sec. IX. II. MODEL OF A WEYL SEMIMETAL – CONVENTIONAL SUPERCONDUCTOR JUNCTION We study the junction between a Weyl semimetal in the normal state (N) and a conventional (spin-singlet, s- wave) superconductor (S), by first considering separately the Hamiltonians in the two regions and then modeling the interface. Throughout the paper, we take the configuration of Fig. 1 (bottom panel), with the magnetization along z in the plane of the NS interface at x = 0. An out-of-plane rotation of the magnetization by an angle α does not change the results for isotropic Weyl cones, provided that the Fermi surfaces of opposite chirality are not coupled upon reflection at the interface. The geometric condition for this is cos α ≥ kF/K, with kF the Fermi wave vector and (0, 0,±K) the location of the two Weyl points. We assume kF/K (cid:28) 1 in order to have well-resolved Weyl cones, and then there is a broad range of magnetization angles α over which our analysis applies. A. Weyl semimetal region The Weyl semimetal in the region x > 0 has the generic Hamiltonian20–22 HW(k) = τz(σxtx sin kx + σyty sin ky + σztz sin kz) + mkτxσ0 + βτ0σz − µWτ0σ0, mk = m0 + t(cid:48) x(1 − cos kx) + t(cid:48) y(1 − cos ky) + t(cid:48) z(1 − cos kz). (2.1b) The units are normalized by  ≡ 1 and lattice constant a0 ≡ 1. The Pauli matrices τα and σα refer to orbital and spin degrees of freedom (with τ0, σ0 the 2 × 2 unit matrix). The Weyl points are at k = (0, 0,±K), with (2.1a) K 2 ≈ β2 − m2 z + t(cid:48) t2 0 zm0 (2.2) 2 While time-reversal symmetry is broken by the mag- netization β, the inversion symmetry of the material is preserved: τxHW(−k)τx = HW(k). (2.3) The presence of inversion symmetry plays a crucial role when superconductivity enters, because the pair poten- tial couples electrons and holes at opposite momentum. To describe the superconducting proximity effect we add the electron-hole degree of freedom ν, with electron and hole Hamiltonians related by the operation of time- reversal: H (e) W(−k)σy. (2.4) W (k) = HW(k), H (h) two Hamiltonians The Bogoliubov-De Gennes (BdG) Hamiltonian are incorporated in the W (k) = σyH∗ (cid:33) (cid:32) HW = H (e) W 0 −H (h) 0 W = νzτz(σxtx sin kx + σyty sin ky + σztz sin kz) + mkνzτxσ0 + βν0τ0σz − µWνzτ0σ0. (2.5) (2.6) Electron-hole symmetry is expressed by νyσyH∗ W(−k)νyσy = −HW(k). Note that the electron-hole symmetry operation squares to +1, as it should in symmetry class D (fermions without spin-rotation or time-reversal symmetry). B. Superconducting region (cid:19) (cid:18)p2/2m − µS The region x < 0 contains a conventional spin-singlet s-wave superconductor (real pair potential ∆0), with BdG Hamiltonian HS = ∆0 . ∆0 −p2/2m + µS (2.7) For a chemical potential µS (cid:29) µW, the momentum components py, pz parallel to the NS interface at x = 0 can be neglected relative to the perpendicular component px. We expand px = ±pF + kx around the Fermi momen- tum pF = mvF (with µS = p2 F/2m), by carrying out the unitary transformation HS (cid:55)→ e−iτzpFxHSeiτzpFx = vFkxνzτzσ0 + ∆0νxτ0σ0 + O(k2 x). (2.8) Left-movers and right-movers in the x-direction are dis- tinguished by the τ degree of freedom, and we have in- serted a σ0 Pauli matrix to account for the spin degener- acy in S. Electron-hole symmetry in S is expressed by S(−kx)νyτxσy = −HS(kx). νyτxσyH∗ (2.9) displaced by the magnetization β in the z-direction. The mass term mk ensures that there are no other states near the Fermi energy, so that we have the minimal number of two Weyl cones of opposite chirality. There is an additional τx Pauli matrix, in comparison with the corresponding symmetry relation (2.6) in N, to account for the switch from +pF to −pF. (The electron- hole symmetry operation still squares to +1.) C. Interface transfer matrix III. BLOCK-DIAGONALIZATION OF THE WEYL HAMILTONIAN 3 The wave functions ψW and ψS on the two sides of the NS interface at x = 0 are related by a transfer matrix, ψS = (tx/vF)1/2MψW, M = , (2.10) (cid:19) (cid:18)Me 0 0 Mh which ensures that particle current is conserved across the interface. We assume that the interface does not couple electrons and holes,23 hence the block-diagonal structure, and we also assume that M is independent of energy. The symmetry relations (2.6) and (2.9) imply that the electron and hole transfer matrices are related by Mh = τxσyM∗ e τ0σy. Particle current conservation is expressed by (cid:104)ψSvFνzτzσ0ψS(cid:105) = (cid:104)ψWtxνzτzσxψW(cid:105), (2.11) (2.12) where we have also linearized HW in kx. The resulting restriction on the electron transfer matrix is M† e τzσ0Me = τzσx. (2.13) (cid:104) Eq. (2.11) then implies that the hole transfer matrix Mh satisfies the same restriction. It is helpful to factor out the unitary matrix Ξ0, Me ≡ ΞΞ0, Ξ0 = exp τx(σ0 − σx) i π 4 , (2.14) † 0τz = Ξ2 with Ξ0τzΞ (2.13) we have a quasi-unitarity restriction 0 = σx, because now instead of Eq. Ξ−1 = τzΞ†τz (2.15) (cid:105) For the mode-matching calculations at the NS interface it is convenient to block-diagonalize HW in the τ degree of freedom, by means of the unitary transformation24 (cid:18)iτyσzΩθ 0 (cid:19) 0 Ωθ , (3.1) HW = UHWU†, U = Ωθ = exp(− 1 2 iθτyσz), with a k-dependent angle θ ∈ (0, π) defined by cos θ = − tz sin kz Mk , sin θ = mk Mk , (3.2) (cid:113) Mk = m2 k + t2 z sin2 kz. Note that U satisfies U(k) = νyσyU∗(−k)νyσy, (3.3) because k (cid:55)→ −k maps θ (cid:55)→ π − θ, so the electron-hole symmetry relation (2.6) for HW is preserved upon the unitary transformation. The transformed Hamiltonian, HW(k) = νzτz(σxtx sin kx + σyty sin ky) + Mkν0τzσz + βν0τ0σz − µWνzτ0σ0, (3.4) is block-diagonal in τ . The Weyl cones are in the τ = −1 block, which has low-energy states near k = (0, 0,±β/tz) when Mk ≈ β. The τ = +1 block is pushed to higher energies of order 2β. The unitary transformation changes the wave function in N as ψW = UψW, and hence the matching equation (2.10) becomes that is satisfied by the unit matrix. The corresponding factorization of the hole transfer ψS = (tx/vF)1/2MU† ψW. (3.5) matrix is Mh = τxσy(ΞΞ0)∗τ0σy, (2.16) IV. ANDREEV REFLECTION as required by the electron-hole symmetry (2.11). For later use we give the inverse M−1 h = (σyΞ0σy)(τzσyΞTτzσy)τx, (2.17) in view of the quasi-unitarity (2.15). (The superscript T denotes the transpose of a matrix.) As an aside, we note that if the interface preserves time-reversal symmetry, we have the additional restric- tion Ξ = τxσyΞ∗τxσy. Inversion symmetry is expressed by Ξ = τxΞ−1τx. (2.18) (2.19) At excitation energies E below the superconducting gap ∆0, an electron incident on the superconductor from the Weyl semimetal is reflected, either as an electron (normal reflection, with amplitude ree) or as a hole (An- dreev reflection, with amplitude rhe). We calculate these reflection amplitudes, initially restricting ourselves to normal incidence on the NS interface, in order to sim- plify the formulas. The angular dependence is included in Sec. VI, when we calculate the conductance. We include the energy dependence of the reflection am- plitudes, but since we assume only the low-energy states in the τ = −1 block are propagating our analysis is re- stricted to E (cid:46) β. Typically β (cid:39) 100 meV is much larger than ∆0 (cid:39) 0.1 meV, so this covers the relevant energy range. 4 A. Effective boundary condition at the NS interface As in the analogous problem for graphene,25 the ef- fect of the superconducting region x < 0 on the Weyl semimetal region x > 0 can be described by an effective boundary condition on the wave functions in the limit x → 0 from above, indicated as x = 0+. ∂ ∂x According to the Hamiltonian (2.8), the propagation of the wave function into the superconductor at energy E is governed by the differential equation ψ(x) =(cid:0)iEνz + ∆0νy (cid:1)τzσ0ψ(x) ≡ XSψ(x). (4.1) vF The eigenvalues of XS are ±(cid:112)∆2 0 − E2. To ensure a de- caying wave function in the S region x < 0 for E < ∆0, the state ψS at x = 0− should be a linear superposition of the four eigenvectors with positive eigenvalue. This is expressed by the boundary condition νxψS = exp(iανzτzσ0)ψS, α = arccos(E/∆0) ∈ (0, π/2). (4.2) If we decompose ψS = (ψe, ψh) into electron and hole components, the boundary condition can be written as from Section III, when both Weyl points are in the τz = −1 band. The incident electron wave function ψe,inc = (0, 0, 1, 1) has σx = +1 in the τz = −1 band, so that its ve- locity txνzτzσx is in the negative x-direction. The re- flected wave function ψreflected = ( ψe,refl, ψh,refl) con- tains an electron component ψe,refl = ree(0, 0, 1,−1) with σx = −1, and a hole component ψh,refl = rhe(0, 0, 1, 1) with σx = +1, both waves propagating in the positive x-direction. The reflected waves are related to the inci- dent wave by the normal reflection amplitude ree and the Andreev reflection amplitude rhe. At the interface the propagating modes in the τz = −1 band may excite evanescent modes in the τz = +1 band. Their wave function ψevan in N is an eigenstate of νzσy with eigenvalue +1, so that the Hamiltonian (3.4) pro- duces a decay for x → ∞. The electron and hole compo- nents of the evanescent mode are ψe,evan = a(1, i, 0, 0) and ψh,surf = b(1,−i, 0, 0), with unknown amplitudes a, b. The boundary condition (4.4) then equates the vectors  b−ib  = T  a ia  . rhe rhe 1 + ree 1 − ree (4.6) ψh(0−) = exp(iατz)ψe(0−). (4.3) There is no dependence on the chemical potential µW in the Weyl semimetal for normal incidence. This is a special case of the more general relation between electron and hole wave functions at an NS interface de- rived in App. A. The combination of Eqs. (3.5) and (4.3) gives on the Weyl semimetal side of the NS interface the relation ψh(0+) = T ψe(0+), T = −iΩθM−1 h exp(iατz)MeΩ † θτyσz, (4.4) which can be worked out as T = − iΩθ(σyΞ0σy)(τzσyΞTτzσy)τx exp(iατz)Ξ · Ξ0Ω † θτyσz = U † θ τxσxΞTτyσy exp(iατz)ΞUθ, Uθ ≡ Ξ0Ω † θτyσz, (4.5a) (4.5b) substitution of upon † 0. τyσz(σyΞ0σy)τyσz = Ξ Eq. (2.17) and using B. Reflection amplitudes an We consider incident mode ψincident = (ψe,inc, ψh,inc) without a hole component, ψh,inc = 0, and initially take the simplest case of normal incidence, when ky = 0 and kz = ±K is at one of the two Weyl points. (The dependence on the angle of incidence is included later on.) We work in the transformed basis For an inactive interface, with Ξ = 1, we have T = τyσz cos α − iτxσy sin α, and we find ree = −ie−2iα, rhe = 0, (4.7) (4.8) i.e. fully suppressed Andreev reflection at all energies (and also at all angles of incidence, see Sec. VI). For E < ∆0 the incident electron is reflected as an electron with unit probability, without any transfer of a Cooper pair into the superconductor. For E > ∆0 the angle α = −i arcosh (E/∆0) is imaginary and the incident electron is partly transmitted through the NS interface - but still without any Cooper pair transfer. V. ACTIVATION OF ANDREEV REFLECTION Andreev reflection can be restored by a suitably chosen interface potential. We examine two types of interfaces, one that breaks time-reversal symmetry by a Zeeman coupling to the spin, and another that breaks inversion symmetry by a tunnel coupling to the orbital degree of freedom. A. Spin-active interface We consider an interface with a Zeeman Hamiltonian Hinterf = gµBB · σ on the S side, which gives a transfer matrix Ξ = exp(cid:2)i((cid:96)/vF)τzHinterf (cid:3) = exp(cid:2)iγτz(n · σ)(cid:3), (5.1) with γ = gµBB(cid:96)/vF, n a unit vector in the B-direction, and (cid:96) is the thickness of the interface layer. The su- perconducting coherence length ξ = vF/∆0 is an up- per bound on (cid:96), and hence γ (cid:46) EZeeman/∆0, with EZeeman = gµBB the Zeeman spin splitting. Depending on the direction of the field, we find the , , (5.2a) Andreev reflection amplitudes Hinterf = Bxσx ⇒ rhe = − 2 cos α sin 2γ sin θ sin2 2γ sin2 θ + e2iα Hinterf = Byσy ⇒ rhe = 2i sin α sin 2γ cos θ sin2 2γ cos2 θ − e2iα Hinterf = Bzσz ⇒ rhe = − 2i cos α sin 2γ (5.2c) sin2 2γ + e2iα At the Fermi level (E = 0 ⇒ α = π/2), we have rhe = 0 for B in the x-direction or in the z-direction, while a field in y-direction activates the Andreev reflection. For m0 (cid:28) β (cid:28) tz we may approximate K ≈ β/tz (cid:28) 1, sin θ ≈ β/2tz (cid:28) 1 and cos θ ≈ ∓1. The Andreev reflection probability Rhe = rhe2 at the Fermi level for B in the y-direction is then given by (5.2b) . Rhe = 4 sin2 2γ (1 + sin2 2γ)2 . (5.3) It oscillates with γ, reaching a maximum of unity when γ = 1 4 π modulo π/2. We next consider interfaces that break inversion sym- metry rather than time-reversal symmetry. A poten- tial barrier on the S side of the interface couples ±kF, and thereby switches the τz index. This is modeled by a tunnel Hamiltonian of the form Hinterf = Vbarrierτα with α ∈ {x, y}, which preserves time-reversal symmetry (Hinterf = τxσyH∗ interf τxσy). The choice Hinterf = Vbarrierτx gives the transfer matrix Ξ = e−γ(cid:48)τy , γ(cid:48) = Vbarrier(cid:96)/vF (cid:46) Vbarrier/∆0. (5.4) This preserves inversion symmetry [see Eq. (2.19)], and does not activate Andreev reflection: rhe = 0 for all E. If instead we take the Hamiltonian Hinterf = Vbarrierτy, we have Ξ = eγ(cid:48)τx . Inversion symmetry is broken, and we find activated Andreev reflection: 2i sin α sinh 2γ(cid:48) cos θ rhe = sin2 α sinh2 2γ(cid:48) sin2 θ + (sin α cosh 2γ(cid:48) − i cos α)2 . (5.5) At the Fermi level, and for m0 (cid:28) β (cid:28) tz, the Andreev reflection probability is Rhe = 4 sinh2 2γ(cid:48) cosh4 2γ(cid:48) . (5.6) B. Inversion-symmetry breaking interface N (E) = 2 It reaches a maximum of unity for γ(cid:48) = 1 0.441, decaying to zero for both smaller and larger γ(cid:48). 2 ln(1 + 2) = √ 5 VI. CONDUCTANCE OF THE NS JUNCTION The reflection probabilities Ree = ree2 and Rhe = rhe2 determine the differential conductance dI/dV = G(eV ) of the NS junction, per unit surface area, accord- ing to29 (cid:90) dky (cid:90) dkz 2π 2π G(E) = e2 h (1 − Ree + Rhe). (6.1) The reflection amplitudes ree and rhe, as a function of energy E and transverse momentum components ky, kz, follow from the solution of Eq. (4.6), suitably generalized to include an arbitrary angle of incidence. We consider an incident electron near the Weyl point at k = (0, 0, K), with K ≈ β/tz (cid:28) 1. [The other Weyl cone at −K gives the same contribution to the conductance and we may set θ = 0 in the transfer matrix (4.5).] We take µW, E > 0 so the electron is above the Fermi level at energy µW + E in the upper half of the Weyl cone. The Andreev reflected hole is below the Fermi level at energy µW − E, which drops into the lower half of the Weyl cone when E > µW. For brevity we denote qx = txkx, qy = tyky, qz = tzkz − β. We normalize the conductance by the total number N (E) of propagating electron modes in the Weyl cones at energy E above µW, given by (cid:90) dqy (cid:90) dqz 2πty 2πtz Θ(cid:2)(E + µW)2 − q2 y − q2 z (cid:3) = (E + µW)2 2πtytz . (6.2) (The prefactor 2 sums the contributions from the two Weyl cones.) The low-energy Hamiltonian HK follows upon projec- tion of the Hamiltonian (3.4) on the τ = −1 band and expansion around the Weyl point, HK = −νz(σxqx + σyqy) − ν0σzqz − µWνzσ0. (6.3) The x-component of the momentum is −qx and +qx for the incident and reflected electron, and q(cid:48) x for the hole, with (cid:113) qx = x = sign (E − µW) q(cid:48) (E + µW)2 − q2 (cid:113) y − q2 z , (E − µW)2 − q2 y − q2 z . (6.4) Only real qx contribute to the wave vector integration in Eq. (6.1), and when q(cid:48) x becomes imaginary one should set Rhe ≡ 0. Substitution of the corresponding spinors into Eq. (4.6) (normalized to unit flux) gives the mode matching con- 6 FIG. 3: Differential conductance of the NS junction, calcu- lated from Eqs. (6.1) and (6.5), for the spin-active interface of Sec. V A (dashed curves, for Hinterf = EZeemanσy with γ = π/4), and for the inversion-symmetry breaking inter- √ face of Sec. V B (solid curves, for Hinterf = Vbarrierτy with 2)). For eV (cid:29) ∆0, all curves tend to the γ(cid:48) = 1 normal-state interface conductance of 0.8 N e2/h. 2 ln(1 + superconducting, forming a Weyl superconductor.3,4 In this section we study how the chirality blockade man- ifests itself in an NS junction between the normal and superconducting state of Weyl fermions. To make con- tact with a specific microscopic model, we consider the heterostructure approach of Burkov and Balents6, which can describe both a Weyl semimetal and a Weyl superconductor.26,27 A. Heterostructure model For the Weyl semimetal, we start from a multilayer heterostructure, composed of layers of a magnetically doped topological insulator (such as Bi2Se3), separated by a normal-insulator spacer layer with periodicity d. Its Hamiltonian is6,28,31 H(k) = vFτz(−σykx + σxky) + βτ0σz + (mkτx − τytz sin kzd)σ0, z + tz cos kzd. mk = t(cid:48) (7.1) The Pauli matrices σi act on the spin degree of freedom of the surface electrons in the topological insulator layers. The τz = ±1 index distinguishes the orbitals on the top and bottom surfaces, coupled by the t(cid:48) z hopping within the same layer and by the tz hopping from one layer to the next. Magnetic impurities in the topological insulator layers produce a perpendicular magnetization, leading to an exchange splitting β. The two Weyl points are at FIG. 2: Zero-bias conductance of the NS junction, calculated from Eq. (6.6), for the spin-active interface (dashed curve) and for the inversion-symmetry breaking interface (solid curve). The conductance is normalized by the number of modes N from Eq. (6.2). For the inactive interface the conductance vanishes. dition(cid:115)  qx(E + µW − qz) x(E − µW − qz) q(cid:48)  = T b−ib (q(cid:48) x + iqy)rhe (E − µW − qz)rhe a ia  . qx − iqy + (qx + iqy)ree (E + µW − qz)(1 − ree)  = (6.5) (cid:40) For the inactive interface, when Ξ = 1, the Andreev reflection amplitude vanishes at all energies for all an- gles of incidence. Andreev reflection is activated by the spin-active interface or by the inversion-symmetry- breaking interface, as discussed in Sec. V. At the Fermi x = −qx) we recover the results (5.3) and level (E = 0, q(cid:48) (5.6) multiplied by the factor q2 z ) that accounts for the deviation from normal incidence. The resulting zero-bias conductance is given by x + q2 x/(q2 lim V →0 dI dV = 16 3 N (0) × e2 h sin2 2γ/(1 + sin2 2γ)2, sinh2 2γ(cid:48)/ cosh4 2γ(cid:48), (6.6) as plotted in Fig. 2, with γ = EZeeman(cid:96)/vF (cid:46) EZeeman/∆0 in the spin-active interface Hamiltonian Hinterf = EZeemanσy, and γ(cid:48) = Vbarrier(cid:96)/vF (cid:46) Vbarrier/∆0 in the inversion-symmetry breaking case Hinterf = Vbarrierτy. The voltage-dependent differential conductance is plot- ted in Fig. 3. The conductance vanishes at eV = µW < ∆0, when the hole touches the Weyl point. (The same feature appears at the Dirac point in graphene30.) VII. WEYL SEMIMETAL – WEYL SUPERCONDUCTOR JUNCTION So far we have considered the junction between a Weyl semimetal and a superconductor formed from a conven- tional metal. A doped Weyl semimetal can itself become 7 represents zero-momentum pairing of spin-up and spin- down electrons within the same conducting surface of each topological insulator layer (inversion-symmetric, spin-singlet, intra-orbital pairing). The BCS pairing interaction (7.6) corresponds to a scalar pair potential in the spin and orbital degrees of freedom. We restrict ourselves to that pairing symme- try in this section. Other BCS pair potentials (spin- triplet and pseudoscalar spin-singlet) are considered in Appendix B. To describe a NS interface at x = 0, we set ∆(x) = 0 for x > 0 and ∆(x) = ∆0 for x < 0 (see Fig. 4). We also adjust the chemical potential µ(x), from a small value µW for x > 0 to a large value µS for x < 0. For the other parameters we take x-independent values. B. Mode matching at the NS interface We can now follow the mode-matching analysis of the preceding sections, with one simplification and one com- plication. The simplification is that, because we have the same Weyl Hamiltonian on the two sides of the NS inter- face, we no longer need an interface matrix to conserve current across the interface. The complication is that the block-diagonalization in the τ degree of freedom on the N side of the interface introduces off-diagonal blocks in the pair potential on the S side. The unitary transformation that achieves this partial (cid:18)τyσzΩθU0 0 (cid:19) 0 ΩθU0 , (7.7) H = VHV†, V = Ωθ = exp(− 1 2 iθτyσz), with U0 from Eq. (7.3). The kz-dependent angle θ is defined by (cid:113) cos θ = (tz sin kzd)/Mk, sin θ = mk/Mk, Mk = m2 k + t2 z sin2 kzd. For closely-spaced Weyl points (when tz − t(cid:48) tzd) we may approximate sin θ ≈ 0, cos θ ≈ 1. z (cid:28) β (cid:28) The transformed Hamiltonian is H(k) = vFνzτz(σxkx + σyky) + Mkν0τzσz + βν0τ0σz − µνzτ0σ0 + ∆, ∆ ≡ V∆V† = ∆(x)νxτyσz. This has the same block-diagonal form (3.4) on the N side x > 0 of the interface (where ∆ = 0), but on the S side x < 0 the transformed pair potential ∆ is off-diagonal in the τ degree of freedom.32 We again assume µS (cid:29) µW so that in S we may neglect the transverse wave vector component ky and take kz at the Weyl point, where Mk = β. The wave equation in S (7.8) (7.9) + mkτxσ0 + βτ0σz, U0 = exp[− 1 4 iπ(τ0 + τx)σz]. (7.3) block-diagonalization is FIG. 4: Cross-section through a layered Weyl semimetal- superconductor junction, based on the heterostructure model6,26 of alternating topological insulator (TI) layers and normal (N) or superconducting (S) spacer layers. In this model the orbital τ degree of freedom refers to the conducting top and bottom surfaces of the TI layers. k = (0, 0, π/d ± K), with K 2 ≈ β2 − (tz − t(cid:48) z)2 d2tzt(cid:48) z . (7.2) They are closely spaced near the edge of the Brillouin zone for tz − t(cid:48) z (cid:28) β (cid:28) tzd. To make contact with the generic Weyl Hamiltonian (2.1), we note the unitary transformation U0H(k)U † 0 = vFτz(σxkx + σyky) − τzσztz sin kzd We will make use of this transformation later on. Following Meng and Balents26, the spacer layer may have a spin-singlet s-wave pair potential ∆, with a uni- form phase throughout the heterostructure (which we set to zero, allowing us to take ∆ real). The pair potential in- duces superconductivity in the top and bottom surfaces of the topological insulator layers, as described by the BdG Hamiltonian H(k) = vFνzτz(−σykx + σxky) + βν0τ0σz + νz(mkτx − τytz sin kzd)σ0 − µνzτ0σ0 + ∆, (7.4) ∆ = ∆(x)νxτ0σ0. It acts on eight-component Nambu spinors Ψ with ele- ments Ψ = (ψ+↑, ψ+↓, ψ−↑, ψ−↓, ψ∗ +↓,−ψ∗ +↑, ψ∗ −↓,−ψ∗ −↑), (7.5) where ± refers to the top and bottom surface and (cid:108) refers to the spin band. The pair potential ∆ in Eq. (7.4) is diagonal in the τ and σ degrees of freedom. The corresponding BCS pairing interaction, (cid:105) † † † +↓(−k) + c −↓(−k) −↑(k)c (cid:88) (cid:104) † +↑(k)c c HBCS = ∆ k + H.c., (7.6) corresponding to the Hamiltonian (7.9) then reads 8 ∂ ∂x ψ(x) = XSψ(x), x < 0, vF (7.10) XS = i(Eνz + µSν0)τzσx − βνz(τ0 + τz)σy − ∆0νyτxσy. As derived in Appendix A, the decaying eigenvectors for E < ∆0 and x → −∞ satisfy νxτyσzψ = exp(iανzτzσx)ψ, (7.11) with α = arccos(E/∆0) ∈ (0, π/2). The corresponding boundary condition on ψ = (ψe, ψh) is ψh(0) = T ψe(0), T = eiατzσx τyσz. (7.12) Because ψ(x) is now continuous across the interface, we do not need to distinguish 0+ and 0− as we needed to do in Sec. IV A. Substitution of T into the mode matching equation (6.5) gives rhe ≡ 0; fully suppressed Andreev reflection at all energies and all angles of incidence. This is the chirality blockade. VIII. FERMI-ARC MEDIATED JOSEPHSON EFFECT While the conductance of a single NS interface is fully suppressed by the chirality blockade, the supercurrent through an SNS junction is nonzero because of overlap- ping surface states (Fermi arcs) on the two NS interfaces. We have calculated this Fermi-arc mediated Josephson effect (see Appendix C), and we summarize the results. The Fermi arcs connect the Weyl cones of opposite chirality33. As they pass through the center of the Bril- louin zone, the chirality blockade is no longer operative and the Fermi arcs acquire a mixed electron-hole char- acter. At kz = 0, the surface states are charge neutral Majorana fermions.24 The Fermi arcs are bound to the NS interface over a distance of order vF/β, so a coupling of the two NS inter- faces is possible if their separation L (cid:46) vF/β. For larger L, the critical current is suppressed ∝ exp(−L/ξarc), with ξarc (cid:39) vF/β the penetration depth of the surface Fermi arc into the bulk (see Fig. 5). IX. DISCUSSION In conclusion, we have shown that Andreev reflection at the interface between a Weyl semimetal and a spin- singlet s-wave superconductor is suppressed by a mis- match of the chirality of the incident electron and the reflected hole. Zero-momentum (s-wave) pairing requires that the electron and hole have opposite chirality, while singlet pairing requires that they occupy opposite spin bands, and these two requirements are incompatible, as illustrated in Fig. 1. FIG. 5: Critical current density jc of the SNS junction as a function of the separation L of the NS interfaces for different values of β, calculated from the Hamiltonian (7.4) for µ = 0, vF = tz = t(cid:48) z = d = 1. The dashed lines indicate the exponential decay ∝ e−cβL/vF with c = 1.7. We have identified two mechanisms that can remove the chirality blockade and activate Andreev reflection. The first mechanism, a spin-active interface, has the same effect as spin-triplet pairing: it enables Andreev reflec- tion by allowing an electron and a hole to be in the same spin band. The second mechanism, inversion-symmetry breaking either at the interface or in the pair potential, is more subtle, as we now discuss. Consider the single-cone Weyl Hamiltonian centered at k = (0, 0, +K), H+ = vxkxσx + vykyσy + vz(kz − K)σz. (9.1) By definition, its chirality is C = sign (vxvyvz). For the second Weyl cone centered at k = (0, 0,−K) of opposite chirality, we can take either H− = −vxkxσx − vykyσy − vz(kz + K)σz or − = vxkxσx + vykyσy − vz(kz + K)σz, H(cid:48) (9.2) (9.3) or some permutation of x, y, z, but either all three signs or one single sign of the velocity components must flip. The first choice satisfies inversion symmetry, H−(−k) = H+(k), while the second choice does not. In Fig. 6 we show the spin-momentum locking in the pair of Weyl cones HW = (H+, H−) and H(cid:48) −) with and without inversion symmetry. We see that the chirality blockade can be removed by breaking inversion symme- try. W = (H+, H(cid:48) This explains why Uchida, Habe, and Asano11 (who, with Cho, Bardarson, Lu, and Moore34, fully appreci- ated the importance of spin-momentum locking for su- perconductivity in a Weyl semimetal) did not find any Synergy Grant. 9 Appendix A: Derivation of the boundary condition at a Weyl semimetal – Weyl superconductor interface Equation (4.2) gives the effective boundary condition at the NS interface between a Weyl semimetal and a con- ventional superconductor. Here we generalize this to the interface between a Weyl semimetal and a Weyl super- conductor. We allow for a more general pairing symme- try than considered in the main text, and in Appendix B we apply the boundary condition to spin-triplet pairings and to a pseudoscalar spin-singlet pairing. As discussed in the related context of graphene25, the local coupling of electrons and holes at the NS inter- face that is expressed by the effective boundary condition holds under three conditions: (i) The chemical potential µS in the superconducting region is the largest energy scale in the problem, much larger than the superconduct- ing gap ∆0 and much larger than the chemical potential µN in the normal region; (ii) the interface is smooth and impurity-free on the scale of the superconducting coher- ence length vF/∆0; and (iii) there is no lattice mismatch at the NS interface. We start from Eq. (7.10), which governs the decay of the wave function in the superconducting region, ∂ ∂x ψ(x) = XSψ(x), x < 0, vF XS = (iµSτzσx + YS), YS = iνzτzσx(E − ∆). (A1) We have omitted the β term, which anticommutes with the µS term and can be neglected in the large-µS limit. We seek a boundary condition on ψ at x = 0 that ensures decay for x → −∞. In the most general case, ∆ is an Hermitian 8 × 8 matrix that satisfies the electron-hole symmetry relation νyσy ∆∗νyσy = − ∆. (A2) We make the following four additional assumptions: 1. ∆ anticommutes with νz (so it is fully off-diagonal in the electron-hole degree of freedom); 2. ∆ commutes with τzσx (anticommuting terms do not contribute to the spectrum of XS in the large- µS limit, so they may be ignored); 3. ∆ is independent of the momentum perpendicular to the NS interface (it may depend on the parallel momentum); FIG. 6: Illustration of the spin-momentum locking for states at the Fermi energy in a pair of Weyl cones at k = (0, 0,±K). The arrows indicate the direction of the spin polarization for a momentum eigenstate at ky = 0, as a function of kx and kz. The left column is for the Hamiltonian HW = (H+, H−) with inversion symmetry, the right column is for H(cid:48) W = (H+, H(cid:48) −) without inversion symmetry. Andreev reflection (AR) along the x-direction on a superconductor with zero-momentum spin-singlet pairing is blocked for HW (the red and blue ar- rows point in the same direction, so the spin is not inverted, as it should be for spin-singlet pairing), while it is allowed for H(cid:48) W (red and blue arrows point in opposite directions). suppression of Andreev reflection at normal incidence on the NS interface. Their two-band model of a Weyl semimetal20,35,36 has the same spin texture as H(cid:48) W - hence it breaks inversion symmetry and does not show the chirality blockade. The relevance of inversion sym- metry also explains why no chirality blockade appeared in Refs. 12–14, where a pseudoscalar pair potential was used that breaks this symmetry (see Appendix B 2). The chirality blockade suppresses the superconducting proximity effect, but since it can be lifted in a controlled way by a Zeeman field (see Fig. 2), it offers opportu- nities for spintronics applications. In the geometry of Fig. 1, a magnetic field in the y-direction, in the plane of the NS interface and perpendicular to the magnetiza- tion, activates Andreev reflection when the Zeeman en- ergy EZeeman becomes comparable to the superconduct- ing gap ∆0. (To prevent pair-breaking effects from this Zeeman field one can use a thin-film superconductor with strong spin-orbit coupling37.) For a typical Zeeman en- ergy of 1 meV/Tesla and a typical gap of 0.1 meV, a 100 mT magnetic field can then activate the transfer of Cooper pairs through the NS interface. This provides a phase-insensitive alternative to the phase-sensitive con- trol of Cooper pair transfer in a Josephson junction. Acknowledgments We have benefited from discussions with Tobias Meng. This research was supported by the Netherlands Organi- zation for Scientific Research (NWO/OCW) and an ERC 4. ∆ squares to a scalar ∆2 0 (this assumption is not essential, but allows for a simple closed-form an- swer). Under these conditions XS and YS commute, so they can 0 − be diagonalized simultaneously. Moroever, Y 2 E2, hence a decaying wave function for E < ∆0 is an 0 − E2, eigenfunction of YS with eigenvalue +(cid:112)∆2 S = ∆2 (cid:113) YSψ = (cid:18) (cid:113) (cid:18) 0 − E2 ψ. ∆2 (cid:113) (A3) (cid:19) ψ (cid:19) 0 − E2 ∆2 ψ We rearrange this to obtain a relation between the elec- tron and hole components of ψ = (ψe, ψh): − iνzτzσx ∆ψ = −iEνzτzσx + ⇒ ∆ψ = ⇒ ∆ψ = ∆0 exp(iανzτzσx)ψ, 0 − E2 νzτzσx ∆2 E + i (A4) with α = arccos (E/∆0) ∈ (0, π/2). For a superconduct- ing phase ϕ we can decompose ∆ = ∆0(νx cos ϕ − νy sin ϕ)χ, (A5) with χ a 4 × 4 Hermitian matrix that squares to unity and commutes with τzσx. We thus arrive at the desired boundary condition, eiϕχψh(0) = eiατzσx ψe(0). (A6) In a more general geometry, with a unit vector n in the x–y plane perpendicular to the NS interface and pointing from N to S, we can write the boundary condition as ψh(0) = T ψe(0), T = e−iϕ exp(cid:2)−iατz(n · σ)(cid:3)χ. (A7) This was derived for subgap energies E < ∆0. The boundary condition still holds by analytic continuation for E > ∆0, when α = −i arcosh (E/∆0) is imaginary, provided that there is no particle current incident on the NS interface from the superconducting side. Appendix B: Generalizations to other pairing symmetries The pair potential ∆ = ∆0νxτ0σ0 in the Meng-Balents Hamiltonian (7.4) represents inversion-symmetric, spin- singlet, intra-orbital pairing, appropriate for the het- erostructure model of Fig. 4. Other types of pair- ing may be relevant for Weyl semimetals with intrinsic superconductivity.27,31 We calculate the corresponding Andreev reflection probabilities. 1. Spin-triplet pair potential For the three s = x, y, z spin-triplet pairings, the rela- tionship between the pair potential ∆s in the Hamilto- nian (7.4) and the transformed pair potential ∆s in the Hamiltonian (7.9) is ∆s = ∆0νxτyσs ⇒ ∆s = −∆0νyχs, χx = −τ0σx cos θ − τyσy sin θ, χy = −τ0σy cos θ + τyσx sin θ, χz = τxσz cos θ − τzσ0 sin θ. 10 (B1a) (B1b) (B1c) (B1d) Each χs squares to unity but only χx and χz commute with τzσx. The s = y pairing anticommutes and does not open a gap in the large-µS limit. For the s = x and s = z pairings we can read off the electron-hole coupling matrix Ts from Eq. (A7), Ts = −ieiατzσxχs, (B2) and then derive the Andreev reflection amplitude by solv- ing Eq. (6.5). The result for normal incidence is 2 sin α cos θ cos2 θ − e2iα , rhe = rhe = − 2i cos α sin θ sin2 θ + e2iα for s = x, (B3a) , for s = z. (B3b) More generally, for any angle of incidence, we have at the Fermi level (when E = 0 ⇒ α = π/2) the Andreev reflection probabilities Rhe = Fk2 v2 x µ2 − v2 Fk2 y 4 cos2 θ (1 + cos2 θ)2 , for s = x, (B4) Rhe = 0, for s = z. 2. Pseudoscalar spin-singlet pair potential (cid:88) (cid:104) The pairing interaction H(cid:48) BCS = ∆ † † +↓(−k) − c +↑(k)c c k + H.c. (cid:105) † † −↓(−k) −↑(k)c (B5) differs from HBCS in Eq. (7.6) by a π phase shift of the pair potential on the top and bottom surfaces. The cor- responding pair potential in the BdG Hamiltonian (7.4) is ∆(cid:48) = ∆0νxτzσ0. (B6) It anticommutes with τx and thus changes sign upon in- version, representing a pseudoscalar pairing in the clas- sification of Ref. 10. Bednik, Zyuzin, and Burkov27 obtain the pseudoscalar pairing (B5) in a model where the pairing interaction is intrinsic to the Weyl semimetal, rather than proximity- induced as in the multilayer structure of Fig. 4. (The τ degree of freedom then refers to a molecular orbital instead of to a heterostructure layer.) The change from scalar to pseudoscalar pairing has drastic consequences for Andreev reflection: The trans- formed pair potential in Eq. (7.9), ∆(cid:48) ≡ V∆(cid:48)V† = −∆0νxτ0σ0, (B7) − µνzτ0σ0 − ∆0νxτ0σ0, is diagonal rather than off-diagonal in the τ degree of freedom. We can therefore project the transformed Hamiltonian, H(cid:48)(k) = vFνzτz(σxkx + σyky) + Mkν0τzσz + βν0τ0σz (B8) onto the τ = −1 subband without losing the pair poten- tial. There is now no chirality blockade. The Andreev reflection amplitude is rhe = − E/∆0 + i 1 − E2/∆2 0, (cid:113) (B9a) at normal incidence for any energy, rhe = , (B9b) ikx(cid:113) x + k2 k2 y at the Fermi level for any angle of incidence. The projected Hamiltonian, H(cid:48) τ =−1 = − vFνz(σxkx + σyky) + (β − Mk)ν0σz − µνzσ0 − ∆0νxσ0, (B10) is essentially the one studied in Refs. 12–14. This ex- plains why no chirality blockade was obtained in those studies of Andreev reflection in a Weyl semimetal. 3. Comparison with tight-binding model simulations 11 we solved the scattering problem at the NS interface nu- merically, using the Kwant toolbox.38 Equation (7.4) is linear in kx and ky, and a straight- forward discretization, by replacing kx (cid:55)→ sin kx, ky (cid:55)→ sin ky, would suffer from fermion doubling. To avoid this, we follow Ref. 22 and add quadratic terms in kx and ky to the mass term mk, resulting in the tight-binding Hamil- tonian H(k) = νzτz(−σy sin kx + σx sin ky) + βν0τ0σz + νz(mkτx − τy sin kz)σ0 − µνzτ0σ0 + ∆, mk = 3 + cos kz − cos kx − cos ky. (B11a) (B11b) For simplicity we have set the Fermi velocity vF and the hopping energies tz, t(cid:48) z equal to unity, and we have taken the same lattice constant d = a ≡ 1 parallel and perpen- dicular to the layers. The Weyl points are at k = (0, 0, π ± K), where (1 − cos K)2 + sin2 K = β2 ⇒ K = arctan β(cid:112)4 − β2 (cid:32) 2 − β2 (cid:33) . (B12) Near the Weyl point, the normal-state dispersion is (E + µW)2 = k2 qz = (π − K − kz) cos θ, cos2 θ = 1 − β2/4. y + q2 z , x + k2 (B13) To test these analytical formulas, we have discretized the eight-orbital Hamiltonian (7.4) on a cubic lattice, and The analytical results for the Andreev reflection prob- ability at the Fermi level (E = 0), as a function of the transverse momenta ky and qz, are: Rhe = 4 − β2 (2 − β2/4)2 y − q2 W − k2 µ2 W − k2 µ2 y z Rhe = 0, W − k2 µ2 W − q2 µ2 Rhe = z y − q2 z , s = x triplet pairing, ∆ = ∆0νxτyσx, s = z triplet pairing, ∆ = ∆0νxτyσz, , pseudoscalar singlet pairing, ∆ = ∆0νxτzσ0, Rhe = 0, scalar singlet pairing, ∆ = ∆0νxτ0σ0. (B14a) (B14b) (B14c) (B14d) In Fig. 7 we compare the analytics with the numerical simulation, and find good agreement without any fit pa- rameter. All of this is for a magnetization in the plane of the NS interface. If the magnetization is rotated out of the plane by an angle α, the Andreev reflection probabil- ity for scalar pairing shows the threshold behavior dis- cussed in Sec. II, see Fig. 8. The threshold angle given by cos αc = kF/K ≈ µW/β for an isotropic Weyl cone is in reasonable approximation with the numerical re- sult, with some deviations because the Weyl cone of the Hamiltonian (B11) has a significant anisotropy. Appendix C: Calculation of the Fermi-arc mediated Josephson effect We calculate the supercurrent flowing through an SNS junction in response to a phase difference φ between the superconducting pair potentials. As explained in Sec. 12 FIG. 8: Threshold dependence of the chirality blockade on the direction of the magnetization. The horizontal axis shows the out-of-plane rotation angle α of the magnetization, the vertical axis shows the Andreev reflection probability at the Fermi level for normal incidence. The data points are cal- culated numerically from the tight-binding model (B11) with a scalar pair potential (parameters µW = 0.18, µS = 0.2, ∆0 = 0.9, β = 0.85). The dashed vertical line is the thresh- old angle αc = arccos (µW/β) = 78◦ expected for an isotropic Weyl cone. The resulting Hamiltonian (cid:0)kxσx + kyσy − Mkσz (cid:1) + βν0τ0σz H = vFνzτz − µνzτ0σ0 + ∆0(νx cos ϕ − νy sin ϕ)τ0σ0 (C3) is diagonal in τ . We may therefore replace τz by the variable τ = ±1 and τ0 by 1. At the NS interfaces x = ±L/2 we have the boundary condition (A7), ψh(±L/2) = T ±1 ψe(±L/2), T = e−iφ/2e−iατ σx , α = arccos(E/∆0). (C4) Integration of Hψ = Eψ, with ψ = (ψ+, ψ−) the two ν- components of the wave function, gives the x-dependence in the N region, ψ±(x) = ex Ξ±ψ±(0), −L/2 < x < L/2, Ξ± = iτ σx Mk ± τ β + σzky. − σy µ ± E vF vF (C5a) (C5b) A bound state in the SNS junction, a socalled Andreev level, appears at energies when25 det(cid:0)1 − e−L Ξ+ T eL Ξ− T(cid:1) = 0. (C6) We assume that the separation L of the NS interfaces is small compared to the superconducting coherence length ξ = vF/∆0. In this short-junction regime the energy dependence of Ξ± can be neglected and only the energy dependence of T needs to be retained.39 FIG. 7: Andreev reflection probability at the Fermi level of a Weyl semimetal – Weyl superconductor interface, for four different pairing symmetries in the superconductor (scalar and pseudoscalar spin-singlet, and s = x or s = z spin-triplet). The left panel shows the analytical results (B14) for ky = 0, β = 0.5, µW = 0.1. The right panel shows the results from a numerical simulation of the tight-binding model (B11), with additional parameters µS = 0.4, ∆0 = 0.55. There are two Weyl points at kz = π ± K, only one of which is shown (the other gives the same results). VIII, because of the chirality blockade of Andreev reflec- tion this supercurrent is due entirely to overlapping Fermi arcs on the two NS surfaces. It is exponentially small when the distance L of the NS interfaces is large com- pared to the decay length vF/β of the surface states into the bulk. This is the key difference between the present calculation for the Weyl semimetal Josephson junction and a similar calculation for a graphene Josephson junc- tion in Ref. 25. 1. Andreev bound states We start from the Hamiltonian (7.4), H = vFνzτz(−σykx + σxky) + βν0τ0σz + νz(mkτx − τytz sin kzd)σ0 − µνzτ0σ0 + ∆0(νx cos ϕ − νy sin ϕ)τ0σ0, (C1) generalized to account for a complex pair potential ∆0eiϕ. In the N region x < L/2 we set ∆0 = 0, while in the S regions x > L/2 we have a nonzero gap ∆0 and a phase ϕ equal to φ/2 for x > L/2 and equal to −φ/2 for x < −L/2. We carry out a (partial) block-diagonalization by means of the unitary transformations H (cid:55)→ WVHV†W†, with V defined in Eq. (7.7) and W defined by (cid:18)iτyσz 0 (cid:19) 0 τ0σ0 W = . (C2) defining (cid:1) and Introducing the vector notation σ = (cid:0)σx, σy, σz (cid:1), (cid:0)iτ µ,−Mk ± τ β, vF sin ky d± =(cid:0)dx, d±,y, dz det(cid:0)e−iφ/2ed−·σ − eiφ/2 eiασx ed+·σ eiασx(cid:1) = 0. the bound-state condition can be written as (cid:1) = L vF (C7) (C8) 13 To simplify the equations we define sinhc x = sinh x x . The identity ed·σ =σ0 cosh d + (d · σ) sinhc d, allows us to evaluate the determinant Eq. (C8) as d · d, d = √ γ2 0 = γ2 1 + γ2 2 + γ2 3 , where γ0 = e−iφ/2 cosh d− − eiφ/2(cid:0)cos 2α cosh d+ γ1 = e−iφ/2dx sinhc d− − eiφ/2(cid:0)i sin 2α cosh d+ + i sin 2α dx sinhc d+ (cid:1), (cid:1), + cos 2α dx sinhc d+ (C12b) γ2 = e−iφ/2d−,y sinhc d− − eiφ/2d+,y sinhc d+, γ3 = e−iφ/2dz sinhc d− − eiφ/2dz sinhc d+. The phase dependence of the bound-state energy can be solved exactly from Eq. (C11) when the Fermi level is near the Weyl points, µ (cid:28) vF/L: (C12d) (C12c) (cid:113) 1 E(φ) = ∆0 2 + p(φ) 1 + (d− · d+) sinhc d− sinhc d+ p(φ) = 2 cosh d− cosh d+ sin2(φ/2) cosh d− cosh d+ , − (C9) (C10) (C11) FIG. 9: Current-phase relationship of the Josephson current density for various values of β and L. The extrema are close to π/2 and 3π/2, indicated by the dashed lines. This is cal- culated from Eq. (C16) for vF = tz = t(cid:48) z = d = 1. We take µ (cid:28) vF/L and substitute Eq. (C13), to arrive at (C12a) j(φ) = (cid:90) π (cid:90) π e∆0 8π2 dky −π dkz −π sin φ (cid:113) 1 . 2 + p(φ) (C16) (The integrand is symmetric in τ = ±, so the sum over τ has been omitted in favor of an overall factor of 2.) cosh d− cosh d+ We take parameters vF = tz = t(cid:48) z = d = 1, when Mk = (1 + cos kz)2 + sin2 kz = 2 cos(kz/2). (C17) (cid:113) The current-phase relationship is close to sinusoidal, see Fig. 9. The critical current can then be accurately ap- proximated by jc ≈ j(π/2). This is plotted as a func- It decays ∝ exp(−L/ξarc), with tion of L in Fig. 5. ξarc (cid:39) vF/β the penetration depth of the surface Fermi arc into the bulk. (C13a) (C13b) where the d± are taken at µ = 0. The energy levels are doubly degenerate in τ = ±1. This degeneracy is lifted by a finite chemical potential: The first-order correction δE± to the bound state energy reads δE± = ± Lµ∆0 2vF (cid:12)(cid:12)(cid:12) tanh d+ d+ (cid:12)(cid:12)(cid:12)(cid:113) 1 − tanh d− d− 2 − p(φ). (C14) 2. Josephson current In the short-junction limit only the Andreev levels con- tribute to the supercurrent density,39 according to j(φ) = − e  dky 2π dkz 2π dE(φ) dφ . (C15) (cid:90) π (cid:88) τ =±1 −π (cid:90) π −π 1 M. Z. Hasan and C. L. Kane, Topological insulators, Rev. Mod. Phys. 82, 3045 (2010). 2 X.-L. Qi and S.-C. Zhang, Topological insulators and su- perconductors, Rev. Mod. Phys. 83, 1057 (2011). 3 M. Z. Hasan, S.-Y. Xu, I. Belopolski, and S.-M. Huang, Discovery of Weyl fermion semimetals and topological Fermi arc states, Annu. Rev. Condens. Matter Phys. 8, 289 (2017). 4 B. Yan and C. Felser, Topological materials: Weyl semimetals, Annu. Rev. Condens. Matter Phys. 8, 337 (2017). 5 M. Konig, H. Buhmann, L. W. Molenkamp, T. L. Hughes, C.-X. Liu, X.-L. Qi, and S.-C. Zhang, The quantum spin Hall effect: Theory and Experiment, J. Phys. Soc. Jpn. 77, 031007 (2008). 6 A. A. Burkov and L. Balents, Weyl semimetal in a topo- logical insulator multilayer, Phys. Rev. Lett. 107, 127205 (2011). 7 A quick way to see that electrons and holes in a given spin band and Weyl cone move in the same direction, is to note that their dispersion relations are related by Eelectron(k) = −Ehole(−k), so a linear dispersion E = vFk remains the same. 8 S. Hart, H. Ren, T. Wagner, P. Leubner, M. Muhlbauer, C. Brune, H. Buhmann, L. W. Molenkamp, and A. Yacoby, Induced superconductivity in the quantum spin Hall edge, Nature Phys 10, 638 (2014). 9 C. W. J. Beenakker, Andreev reflection and Klein tunneling in graphene, Rev. Mod. Phys. 80, 1337 (2009). 10 Z. Faraei and S. A. Jafari, Superconducting proximity in three dimensional Dirac materials: Odd-frequency, pseudo- scalar, pseudo-vector and tensor-valued superconducting orders, arXiv:1612.06327. 11 Shuhei Uchida, Tetsuro Habe, and Yasuhiro Asano, An- dreev reflection in Weyl semimetals, J. Phys. Soc. Jpn. 83, 064711 (2014). 12 Wei Chen, Liang Jiang, R. Shen, L. Sheng, B. G. Wang, and D. Y. Xing, Specular Andreev reflection in inversion- symmetric Weyl semimetals, EPL 103, 27006 (2013). 13 U. Khanna, D. K. Mukherjee, A. Kundu, and S. Rao, Chi- ral nodes and oscillations in the Josephson current in Weyl semimetals, Phys. Rev. B 93, 121409(R) (2016). 14 K. A. Madsen, E. J. Bergholtz, and P. W. Brouwer, Joseph- son effect in a Weyl SNS junction, Phys. Rev. B 95, 064511 (2017). 15 Y. Kim, M. J. Park, and M. J. Gilbert, Probing uncon- ventional superconductivity in inversion-symmetric doped Weyl semimetal, Phys. Rev. B 93, 214511 (2016). 16 M. Salehi and S. A. Jafari, Sea of Majorana fermions from pseudo-scalar superconducting order in three dimensional Dirac materials, arXiv:1611.03122. 17 U. Khanna, A. Kundu, and S. Rao, 0–π transitions in a Josephson junction of an irradiated Weyl semimetal, Phys. Rev. B 95, 201115(R) (2017). 18 L. Aggarwal, S. Gayen, S. Das, R. Kumar, V. Suss, C. Felser, C. Shekhar, and G. Sheet, Mesoscopic supercon- ductivity and high spin polarization coexisting at metallic point contacts on Weyl semimetal TaAs, Nature Commun. 8, 13974 (2017). 19 He Wang, Huichao Wang, Yuqin Chen, Jiawei Luo, Zhu- jun Yuan, Jun Liu, Yong Wang, Shuang Jia, Xiong-Jun 14 Liu, Jian Wei, and Jian Wang, Discovery of tip induced unconventional superconductivity on Weyl semimetal, Sci- ence Bull. 62, 425 (2017). 20 Kai-Yu Yang, Yuan-Ming Lu, and Ying Ran, Quantum Hall effects in a Weyl semimetal: Possible application in pyrochlore iridates, Phys. Rev. B 84, 075129 (2011). 21 G. Y. Cho, Possible topological phases of bulk magnetically doped Bi2Se3: Turning a topological band insulator into Weyl semimetal, arXiv:1110.1939. 22 M. M. Vazifeh and M. Franz, Electromagnetic response of Weyl semimetals, Phys. Rev. Lett. 111, 027201 (2013). 23 It is justified to ignore electron-hole coupling in the in- terface matrix (2.10), because this describes the transmis- sion through a length of order 1/kF in S, which is smaller than the electron-hole coupling length vF/∆0 by a factor EF/∆0 (cid:29) 1. 24 P. Baireuther, J. Tworzyd(cid:32)lo, M. Breitkreiz, I. Adagideli, and C. W. J. Beenakker, Weyl-Majorana solenoid, New J. Phys. 19, 025006 (2017). 25 M. Titov and C. W. J. Beenakker, Josephson effect in bal- listic graphene, Phys. Rev. B 74, 041401(R) (2006). 26 T. Meng and L. Balents, Weyl superconductors, Phys. Rev. B 86, 054504 (2012). 27 G. Bednik, A. A. Zyuzin, and A. A. Burkov, Superconduc- tivity in Weyl metals, Phys. Rev. B 92, 035153 (2015). 28 H. Zhang, C.-X. Liu, X.-L. Qi, X. Dai, Z. Fang, and S.- C. Zhang, Topological insulators in Bi2Se3, Bi2T e3 and Sb2T e3 with a single Dirac cone on the surface, Nature Phys. 5, 438 (2009). 29 G. E. Blonder, M. Tinkham, and T. M. Klapwijk, Transi- tion from metallic to tunneling regimes in superconducting microconstrictions: Excess current, charge imbalance, and supercurrent conversion, Phys. Rev. B 25, 4515 (1982). 30 C. W. J. Beenakker, Specular Andreev reflection in graphene, Phys. Rev. Lett. 97, 067007 (2006). 31 L. Fu and E. Berg, Odd-parity topological superconductors: Theory and application to CuxBi2Se3, Phys. Rev. Lett. 105, 097001 (2010). 32 We should note that our Eq. (7.9), with a pair potential that is off-diagonal in τ , disagrees with Eqs. 5–8 of Ref. 26, which have a fully block-diagonal Hamiltonian in the layer degree of freedom. We do agree on the Hamiltonian before the unitary transformation [our Eq. (7.4) and Eqs. 1–3 in Ref. 26] and we have traced the discrepancy to an algebraic error. 33 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). 34 G. Y. Cho, J. H. Bardarson, Y.-M. Lu, and J. E. Moore, Superconductivity of doped Weyl semimetals: Finite- momentum pairing and electronic analog of the 3He-A phase, Phys. Rev. B 86, 214514 (2012). 36 The Hamiltonian H(cid:48) 35 R. Okugawa and S. Murakami, Dispersion of Fermi arcs in Weyl semimetals and their evolutions to Dirac cones, Phys. Rev. B 89, 235315 (2014). W = (H+, H(cid:48) −) is called a two-band model because near the Weyl points Eqs. (9.1) and (9.3) can be combined into a single 2 × 2 matrix of the form z − K 2)σz. No additional or- vxkxσx + vykyσy + (vz/2K)(k2 bital degree of freedom is needed, as in the four-band model (2.1). The inversion-symmetry breaking remains hidden unless there are spin-dependent processes that couple the Weyl cones, as in Andreev reflection. 37 H. Nama, H. Chena, T. Liub, J. Kima, C. Zhang, J. Yong, T. R. Lemberger, P. A. Kratz, J. R. Kirtley, K. Moler, P. W. Adams, A. H. MacDonald, and C.-K. Shih, Ultra- thin two-dimensional superconductivity with strong spin- orbit coupling, PNAS 113, 10513 (2016). 38 C. W. Groth, M. Wimmer, A. R. Akhmerov, and X. Wain- tal, Kwant: A software package for quantum transport, New J. Phys. 16, 063065 (2014). 39 C. W. J. Beenakker and H. van Houten, Josephson current through a superconducting quantum point contact shorter than the coherence length, Phys. Rev. Lett. 66, 3056 (1991). 15 

1709.09513

3

1709

Geometric Dynamics of Magnetization: Electronic Contribution

To give a general description of the influences of electric fields or currents on magnetization dynamics, we developed a semiclassical theory for the magnetization implicitly coupled to electronic degrees of freedom. In the absence of electric fields the Bloch electron Hamiltonian changes the Berry curvature, the effective magnetic field, and the damping in the dynamical equation of the magnetization, which we classify into intrinsic and extrinsic effects. Static electric fields modify these as first-order perturbations, using which we were able to give a physically clear interpretation of the current-induced spin-orbit torques. We used a toy model mimicking a ferromagnet-topological-insulator interface to illustrate the various effects, and predicted an anisotropic gyromagnetic ratio and the dynamical stability for an in-plane magnetization. Our formalism can also be applied to the slow dynamics of other order parameters in crystalline solids.

cond-mat.mes-hall

cond-mat

Geometric Dynamics of Magnetization: Electronic Contribution Bangguo Xiong,1, ∗ Hua Chen,2, 3 Xiao Li,1 and Qian Niu1, 4 1Department of Physics, The University of Texas at Austin, Austin, TX 78712, USA 2Department of Physics, Colorado State University, Fort Collins, CO 80523, USA 3School of Advanced Materials Discovery, Colorado State University, Fort Collins, CO 80523, USA 4ICQM and CICQM, School of Physics, Peking University, Beijing 100871, China To give a general description of the influences of electric fields or currents on magnetization dynamics, we developed a semiclassical theory for the magnetization implicitly coupled to electronic degrees of freedom. In the absence of electric fields the Bloch electron Hamiltonian changes the Berry curvature, the effective magnetic field, and the damping in the dynamical equation of the magnetization, which we classify into intrinsic and extrinsic effects. Static electric fields modify these as first-order perturbations, using which we were able to give a physically clear interpretation of the current-induced spin-orbit torques. We used a toy model mimicking a ferromagnet-topological- insulator interface to illustrate the various effects, and predicted an anisotropic gyromagnetic ratio and the dynamical stability for an in-plane magnetization. Our formalism can also be applied to the slow dynamics of other order parameters in crystalline solids. PACS numbers: 75.78.-n, 75.60.Jk, 75.76.+j Introduction-Magnetization dynamics is convention- ally described by the phenomenological Landau-Lifshitz- Gilbert (LLG) equation, in which the effective magnetic field and the damping factor can be associated with var- ious mechanisms such as dipolar interaction, exchange coupling, electron-hole excitations, etc., through micro- scopic theories. The recently discovered current-induced spin-orbit torques emerge as current-dependent modifi- cations to the LLG equation, and can be consequently categorized as field-like and damping-like torques1–6. In systems with strong spin-orbit coupling and broken inver- sion symmetry, e.g. GaMnAs, heavy-metal/ferromagnet bilayers and magnetically doped topological insulator heterostructures, magnetization switching using electric current alone through the spin-orbit torque has been achieved experimentally6–9. Theoretical studies of spin- orbit torques have mostly adopted s − d type couplings between transport electrons and those contributing to magnetization1–4,6, or a self-consistent-field picture based on the spin density functional theory5,10. Then the spin- orbit torques can be understood as the modification to the effective exchange fields proportional to the current- induced spin densities in inversion symmetry breaking systems, known as the Edelstein effect2,11. However, in general neither the size of the exchange field nor its de- pendence on order parameter (magnetization) direction is known a priori 12,13. It is thus more desirable to de- velop a theoretical framework that does not explicitly depend on the details of the coupling between transport electrons and the magnetization. In this Rapid Communication, we provide a semiclassi- cal framework for the dynamics of magnetization implic- itly coupled to electronic degrees of freedom, based on the wave-packet method. We found that the Bloch elec- trons yield a Berry curvature Ωmm, acting as a magnetic field in the magnetization space, while the gradient of the electronic free energy with respect to the magnetization acts as a static electric field in the magnetization space, in agreement with previous adiabatic theory of magne- tization dynamics14. These two fields thus govern the dynamics of magnetization as that of Lorentz force to a charged particle. In addition, we identified an extrinsic contribution to the magnetization dynamics, correspond- ing to the Gilbert damping in the LLG equation, which is not included in the adiabatic theory. A static elec- tric field enters the magnetization equation of motion by modifying the Berry curvature Ωmm, the effective field, and the damping factor as a first-order perturbation. In particular, the modification to the effective field includes a part proportional to the Berry curvature Ωmk and hav- ing a geometric nature. We used a simplified model for the ferromagnet-topological-insulator interface to illus- trate the various effects, and showed that the gyromag- netic ratio is renormalized anisotropically and that an in-plane magnetization can be dynamically stable under moderate electric fields. Formulation and general results-We start from a general Hamiltonian of Bloch electrons implicitly depending on the order parameter m, He(q; m), where q is the crys- tal momentum. External electromagnetic fields are de- scribed by the scalar and vector potentials (φ, A) that en- ter the Hamiltonian through minimum coupling ( = 1, e = e), H = He(q + eA; m) − eφ. (1) Following Ref. 15, a wave packet is constructed with cen- ter position x and center physical momentum k from the Bloch eigenstates of the local electronic Hamiltonian. The Lagrangian of a single wave packet reads as L = x·[k−eA(x, t)]+ k·Ak + m·Am−[ε−eφ(x, t)], (2) with Aλ = i(cid:104)u∇λu(cid:105) (λ = k or m) the Berry connec- tions of the Bloch state u(cid:105), and ε the wave packet en- ergy. For notational simplicity we have dropped the band index. The Lagrangian depends on (x, k) of the wave packets and magnetization m. Thus a set of coupled equations of motion for all three variables can be derived from the Lagrangian principle16: k = −eE, ∂ε ∂k x = (cid:18) [dk]f + k · Ωkk + m · Ωmk, m · Ωmm + k · Ωkm + (3) (4) = 0, (5) (cid:19) ∂ε ∂m the Berry Ωλiλj curvatures where = −2Im(cid:104)∂u/∂λi∂u/∂λj(cid:105), λ = k or m. Eq. 5 is ob- tained by summing over all occupied states, and f is the distribution function for the electrons. Note the magnetization dynamics enters the electron equations of motion through Ωkk in Eq. 4, and the terms in the square brackets of Eq. 5 can be viewed as conjugates of the right hand side of Eq. (4), by interchanging k and m. This is a manifestation of the reciprocity between charge pumping due to magnetization precession and electric-current-induced spin-orbit torque. The nonequilibrium response of the electrons to an external electric field and/or a dynamical m is ac- counted for using the semiclassical Boltzmann equation, according to which the deviation of the distribution function from the equilibrium Fermi-Dirac distribution f0[ε(k, m)] is δf = −τ ∂f0 ∂ε k · ∂ε ∂k + m · ∂ε ∂m , (6) (cid:18) (cid:19) where we have assumed a grand canonical ensemble with fixed temperature and chemical potential. τ is the re- laxation time which we take as a constant for simplic- ity. Generalization to including more specific scattering mechanisms is straightforward but involved, and does not necessarily provide additional insight on the main issues considered in this work. The equations (3-6) complete our semiclassical descrip- tion of coupled magnetization and electron dynamics in the presence of external electric fields, though they can be easily extended to including magnetic fields and other perturbations. In the absence of electric fields, k = 0, and we can obtain from Eq. (6) and Eq. (5) the following equations of motion of the magnetization, m · ( ¯Ωmm + ηmm) − H = 0, (7) in getting which we have ignored higher order m2 terms by assuming that the magnetization dynamics is slow compared to typical electronic time scales. The Berry curvature ¯Ω, the damping coefficient η and the effective field H in the equation above are respectively [dk]f0Ωmm, [dk] ∂f0 ∂ε ∂ε ∂m ∂ε ∂m , ¯Ωmm = ηmm = −τ H = − ∂G ∂m , (8) (9) (10) 2 ´ where G is the free energy of the electron system. For non-interacting electrons G = −β−1 [dk] ln[1 + e−β(ε−µ)] for a single band, where β = 1/kBT . Interac- tion effects may be included in G through different levels of approximations, which will also modify the way mag- netization appears in G. At this point we will leave G as a general electron free energy depending on m implicitly. We only consider the transverse modes ( m perpendic- ular to m) of the magnetization dynamics in this work, although Eq. 7 can be used for the longitudinal mode as well. The magnetization is thus described by the polar angle θ and the azimuthal angle φ. Eq. (7) can then be converted to the familiar form of the LLG equation, m = −γm × (H − ηmm · m) , (11) where the gyromagnetic ratio γ is related to the Berry curvature through ¯Ω = m/γm2, (12) where ¯Ωi = εijk ¯Ωjk/2 is the vector form of the Berry cur- vature tensor. Expressions similar to Eq. (7), but without the damping term, have been derived using the adiabatic theory17. Since the damping term is explicitly depen- dent on the relaxation time, which is ultimately due to dissipative microscopic processes such as electron-phonon scattering and electron-impurity scattering, we call it ex- trinsic contribution to the magnetization dynamics. Note Eq. 9 suggests η is positive definite, which means it al- ways leads to energy dissipation through Eq. 11. The remaining terms are intrinsic contributions from the elec- tron degrees of freedom. In particular, from Eq. (7) one can see that the two intrinsic terms are formally similar to the Lorentz force of a charged particle, with the anti- symmetric part of Ωmm (or equivalently the vector form Ωm) analogous to the magnetic field and H playing the role of the electric field. Electric fields enter our formalism through the equa- tion of motion for k [Eq. (3)], which makes the 2nd term in the integrand of Eq. 5 nonzero and also contributes to the nonequilibrium distribution function δf in Eq. (6). After some algebra, we arrive at the same equation as Eq. (7), but with H, ¯Ωmm, and ηmm acquiring the fol- lowing corrections proportional to the electric field: [dk] Ωkmf0 − τ ∂ε ∂k ∂ε ∂m H E = eE · ¯ΩE mimj = eτ E · (cid:20) ∂ε ∂k [dk] Ωmimj − Ωkmi ηE mimj = eτ E · [dk] Ωkmi (cid:18) (cid:18) (cid:18) (cid:19) (cid:19) (cid:19) ∂f0 ∂ε (cid:21) ∂f0 ∂ε ∂f0 ∂ε A S , (13) (14) , (15) ∂ε ∂mj ∂ε ∂mj where subscript S (A) means the part of Ωkmi that is symmetric (antisymmetric) under i ↔ j. We next discuss the physical meanings of these results in detail. ∂ε ∂mj For the correction to the effective field, H E, the first term in Eq. 13 has a geometric nature and is an intrinsic contribution from the Fermi sea electrons. It is of Ωmt type, where the time variation is due to the momentum change of a single wave packet driven by E: ∂t = k· ∂k = −eE·∂k. We note there is a nice identity connecting Ωmt and the "magnetic field" in magnetization space Ωm: ∂tΩm + ∇m × Ωmt = 0. (16) Since Ωmt = Ωmk · (−eE) is a correction to the static effective electric field H (Eq. 10) in the magnetization space, above equation is a magnetic analog of the Fara- day's law for charged particles. The 2nd term in Eq. 13 is extrinsic since it is proportional to τ , and does not have an electromagnetism analog. H E also provides new insights on the charge pumping effect of a nonzero m18. Since P ≡ H E· m has the mean- ing of power density and is proportional to E, there is an electric current induced by m as jp = ∂(H E · m)/∂E. The change of the polarization density ("pumping") af- ter m completes a closed path in its configuration space is obtained by integrating jp over this period. A finite charge pumping thus corresponds to a nonzero work den- sity, and is related to the curl of H E in the magnetization space through W = jp · Edt = H E · dm ∇m × H E · dσm, = (17) where we have used the Stokes theorem, and dσm is the infinitesimal area in the magnetization space. Thus in order to have finite charge pumping, H E must not be conservative, i.e., it cannot be written as a gradient of certain scalar free energy. mm and ηE We now move on to ¯ΩE mm, which are all Fermi surface contributions due to the non-equilibrium part of the distribution function δf . They are important in magnetic metals and should be discussed on an equal footing as H E for current-induced effects on magnetiza- tion dynamics. In the form of Eq. 11, ¯ΩE mm renormal- izes the gyromagnetic ratio as γ(cid:48) = γ/(1 + γ/γE), where γE ≡ 1/m· ¯ΩE, while ηE mm modifies the damping tensor as η(cid:48) = η + ηE. It is interesting to note that ηE does not have to be positive definite. A negative definite total damping will make the free energy minima dynamically unstable while the maxima dynamically stable. Thus in addition to the potential of switching the magnetization between different easy directions, a suitably chosen elec- tric field can in principle switch the magnetization be- tween easy and hard directions, which provides a new mechanism (though volatile) for current driven reading and writing processes in magnetic memory devices. Before ending this section, we translate our results Eq. (13-15) into the commonly used spin-orbit torque language. For small electric fields they can be converted to additional terms added to the right hand side of the LLG equation Eq. 11: m = −γm × (H − ηmm · m) − γτso, (18) 3 so +τ γ where τso = τ H so with the separate terms being so +τ η so = m × H E, τ H so = −γ/γEm × (H + ηγm × H), τ γ so = γηEm × (m × H). τ η (19) (20) (21) For the special s − d type coupling, H E is propor- tional to the spin density response to electric fields since ∂ H/∂m ∼ s, in agreement with previous studies2,5,8,11, though our formalism is not limited to this coupling form. Morever, there are additional torques τ γ so that cannot be directly explained using spin density response to electric fields. They can, however, always be classified into either field-like or damping-like torques depending on whether there is a sign change upon m → −m. Model example-As a concrete example, we consider a 2D toy model that can be used to describe the interface between a ferromagnetic insulator and a 3D topological insulator (TI)19–21: so and τ η H(m) = v(−kyσx + kxσy) + Jm · σ, (22) where m is the 2D magnetization of the ferromagnet, σ is the Pauli matrix vector for the spin operators, v is the Fermi velocity of the Dirac surface electrons of the TI, and J is the exchange coupling strength between m and σ. The exchange coupling opens a gap proportional to the z component of m. We consider zero temperature and set the chemical potential µ = 0. The Berry curva- ture of the lower band is calculated similarly as the (cid:126)k · (cid:126)σ model16 ¯Ωs θφ = α2 sin 2θ 8πa2 sgn(α), (23) where α = Jma/v is the exchange energy measured in typical scales of the kinetic energy
0 = v/a (a is the lattice constant). Using relation Eq. (12), the Berry curvature gives an anisotropic gyromagnetic ratio γs(θ) = 4πma2 α2 cos θ sgn(α). (24) We should note that the ferromagnet by itself has a gyro- magnetic ratio, denoted as γf , and the overall gyromag- netic ratio γ is corrected as γ−1 = γf −1 + γs −1, (25) or equivalently γ = γf · 1 1 + γf /γs(θ) . (26) The variation of γ for m moving across the Bloch sphere is shown in Fig. 1(a). On the equator (θ = π/2), γ = γf ; at the north and south poles, γ = γf /(1 + γfα2/4πma2sgn(α)). This angular dependence of gy- romagnetic ratio should be able to be detected by ferro- magnetic resonance experiments in such systems. The free energy density at zero temperature is calcu- lated by integrating the energy of the lower bands. Ig- noring a constant term, we get Gs = −J0m2 z (27) where J0 =
0kcα2/4πm2a and kc is the momentum cut- off. Gs has two minima at the north and south poles, as shown in Fig. 1(b). Thus the surface states provide a perpendicular magnetic anisotropy for the ferromagnet. For simplicity we ignored the magnetic anisotropy energy of the ferromagnet itself. For nonzero mz there is no con- tribution from the surface state electrons to η because of the finite gap, and if the intrinsic damping of the ferro- magnet is ignorable the magnetization should move along equal-energy lines without driving forces, along the direc- tions determined by −γm × H [Eq. (11)], as illustrated in Fig. 1(b). 4 be interesting to detect experimentally. Moreover, since GN − GS ∝ mx, they cannot be connected by a constant energy shift across the equator. The electric field thus shifts the two free energy minima at the north and the south poles in opposite directions, and distorts the equal energy lines in the vertical direction, as shown in Fig. 2. In addition, the opposite signs of GN and GS very close to the equator make half of the equator dynami- cally stable, as can be seen from the arrows pointing to the equator from both above and below in Fig. 2. Specif- ically, if we still assume a vanishing intrinsic damping of the ferromagnet, when the magnetization is very close to the equator with φ ∈ (π, 2π), or more generally when it is between the two critical trajectories determined by GN/S = −eEα/4πa, it will follow the equal energy lines and end up on the half equator with φ ∈ (0, π). Con- versely, for a magnetization outside of the region between the two critical trajectories, i.e., GN/S < −eEα/4πa, it will keep precessing around one of the free energy min- ima. When there is a small damping, the size of the attraction area around the half equator reduces because energy is dissipated during evolution. FIG. 1. (Color online) (a) Renormalized gyromagnetic ratio γ through coupling to the topological surface states. (b) Con- tour plot of free energy Gs in the absence of electric fields. The arrows indicate the directions of magnetization motion. Parameters: γf = 2ma2/, α = 1 We now consider the effect of an electric field along x direction on the magnetization dynamics. For nonzero mz all Fermi surface contributions in Eqs. (13-15) are zero, and the only finite term is the Fermi sea contribu- tion in H E: H E = − eEα 4πma sgn(mz)x. (28) It has constant magnitude but opposite directions de- pending on the sign of mz. The curl of H E is thus zero everywhere except on the equator, which also means nonzero charge is pumped by magnetization dynamics when the precession axis is in plane22. Based on our dis- cussion in the previous section we can only define free energy functions separately for the north (N) and the south (S) hemispheres as GN and GS but not globally: GN = −J0m2 z + GS = −J0m2 eEα 4πma z − eEα 4πma mx, mx. (29a) (29b) On each hemisphere, the 2nd term in the free energy im- plies a magnetization-dependent polarization, which will FIG. 2. (Color online) Contour plot of free energies GN and GS in the presence of an electric field along x. Parameters: γf = 2ma2/, α = 1, eEα/4πJ0 = 0.4. of strong case In the limiting electric fields eEα/4πa > 2J0m2, the critical trajectories disappear on the Bloch sphere and the magnetization will always evolve to the stable half equator. Since without the magnetic field the magnetization has a perpendicular anisotropy due to the topological surface states, electric fields can lead to dynamical switching between easy (out- of-plane) and hard (in-plane) directions. This mechanism is unique to the FM/TI system and is independent of the easy-hard-axes switching due to a negative-definite damping tensor discussed in the last section. Since the electric field enters our formalism only through its modification on momentum [Eq. (3)], our the- ory can be straightforwardly generalized to other time- varying perturbations that influence wave-packet dynam- ics in similar ways, which will give both Fermi-surface contributions and Fermi-sea contributions through the Berry curvature Ωmt. For example, a potential appli- cation is the magnetization dynamics driven by sound wave23,24. Separately, our formalism can be applied to the slow dynamics of other order parameters in crys- talline solids, and to its dependence on electromagnetic fields through the electron degrees of freedom. We acknowledge useful discussions with A. H. Mac- Donald, R. Cheng, Y. Gao, H. Zhou. This work is supported by National Basic Research Program of China (Grant No. 2013CB921900), DOE (DE-FG03- 02ER45958, Division of Materials Science and Engineer- ing), NSF (EFMA-1641101), and the Welch Foundation (F-1255). 5 ∗ [email protected] 1 A. Manchon and S. Zhang, Phys. Rev. B 78, 212405 (2008). 2 I. Garate and A. H. MacDonald, Phys. Rev. B 80, 134403 (2009). 3 D. A. Pesin and A. H. MacDonald, Phys. Rev. B 86, 014416 (2012). 4 X. Wang and A. Manchon, Phys. Rev. Lett. 108, 117201 (2012). 86 EP (2014). 13 K. M. D. Hals and A. Brataas, Phys. Rev. B 88, 085423 (2013). 14 Q. Niu, X. Wang, L. Kleinman, W.-M. Liu, D. M. C. Nicholson, and G. M. Stocks, Phys. Rev. Lett. 83, 207 (1999). 15 G. Sundaram and Q. Niu, Phys. Rev. B 59, 14915 (1999). 16 D. Xiao, M.-C. Chang, and Q. Niu, Rev. Mod. Phys. 82, 5 F. Freimuth, S. Blugel, and Y. Mokrousov, Phys. Rev. B 1959 (2010). 90, 174423 (2014). 6 H. Kurebayashi, J. Sinova, D. Fang, A. C. Irvine, T. D. Skinner, J. Wunderlich, V. Nov´ak, R. P. Campion, B. L. Gallagher, E. K. Vehstedt, L. P. Zarbo, K. V´yborn´y, A. J. Ferguson, and T. Jungwirth, Nat Nano 9, 211 (2014). 7 A. Chernyshov, M. Overby, X. Liu, J. K. Furdyna, Y. Lyanda-Geller, and L. P. Rokhinson, Nature Physics 5, 656 EP (2009). 8 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, and D. C. Ralph, Nature 511, 449 (2014). 9 Y. Fan, P. Upadhyaya, X. Kou, M. Lang, S. Takei, Z. Wang, J. Tang, L. He, L.-T. Chang, M. Montazeri, G. Yu, W. Jiang, T. Nie, R. N. Schwartz, Y. Tserkovnyak, and K. L. Wang, Nat Mater 13, 699 (2014). 10 Z. Qian and G. Vignale, Phys. Rev. Lett. 88, 056404 (2002). 11 V. Edelstein, Solid State Communications 73, 233 (1990). 12 A. Brataas and K. M. D. Hals, Nature Nanotechnology 9, 17 Q. Niu and L. Kleinman, Phys. Rev. Lett. 80, 2205 (1998). 18 F. Freimuth, S. Blugel, and Y. Mokrousov, Phys. Rev. B 92, 064415 (2015). 19 I. Garate and M. Franz, Phys. Rev. Lett. 104, 146802 (2010). 20 T. Yokoyama, J. Zang, and N. Nagaosa, Phys. Rev. B 81, 241410 (2010). 21 Y. Tserkovnyak and D. Loss, Phys. Rev. Lett. 108, 187201 (2012). 22 H. T. Ueda, A. Takeuchi, G. Tatara, and T. Yokoyama, Phys. Rev. B 85, 115110 (2012). 23 A. V. Scherbakov, A. S. Salasyuk, A. V. Akimov, X. Liu, M. Bombeck, C. Bruggemann, D. R. Yakovlev, V. F. Sapega, J. K. Furdyna, and M. Bayer, Physical Review Letters 105, 117204 (2010). 24 M. Weiler, L. Dreher, C. Heeg, H. Huebl, R. Gross, M. S. Brandt, and S. T. B. Goennenwein, Physical Review Let- ters 106, 117601 (2011). 

1709.05853

1

1709

A New Numerical Method for $\mathbb{Z}_2$ Topological Insulators with Strong Disorder

We propose a new method to numerically compute the $\mathbb{Z}_2$ indices for disordered topological insulators in Kitaev's periodic table. All of the $\mathbb{Z}_2$ indices are known to be derived from the index formulae which are expressed in terms of a pair of projections introduced by Avron, Seiler, and Simon. For a given pair of projections, the corresponding index is determined by the spectrum of the difference between the two projections. This difference exhibits remarkable and useful properties, as it is compact and has a supersymmetric structure in the spectrum. These properties make it possible to numerically determine the indices of disordered topological insulators highly efficiently. The method is demonstrated for the Bernevig-Hughes-Zhang and Wilson-Dirac models whose topological phases are characterized by a $\mathbb{Z}_2$ index in two and three dimensions, respectively.

cond-mat.mes-hall

cond-mat

a Typeset with jpsj3.cls <ver.1.1> Letter A New Numerical Method for Z2 Topological Insulators with Strong Disorder Yutaka Akagi1 ∗, Hosho Katsura1, and Tohru Koma2 1Department of Physics, Graduate School of Science, The University of Tokyo, Hongo, Tokyo 113-0033, Japan 2Department of Physics, Gakushuin University, Mejiro, Toshima-ku, Tokyo 171-8588, Japan We propose a new method to numerically compute the Z2 indices for disordered topological insulators in Kitaev's periodic table. All of the Z2 indices are known to be derived from the index formulae which are expressed in terms of a pair of projections introduced by Avron, Seiler, and Simon. For a given pair of projections, the corresponding index is determined by the spectrum of the difference between the two projections. This difference exhibits remarkable and useful properties, as it is compact and has a supersymmetric structure in the spectrum. These properties make it possible to numerically determine the indices of disordered topological insulators highly efficiently. The method is demonstrated for the Bernevig-Hughes-Zhang and Wilson-Dirac models whose topological phases are characterized by a Z2 index in two and three dimensions, respectively. KEYWORDS: Topological insulators, Time reversal symmetry, Z2 topological invariant, disordered systems Since the discovery of Z2 topological insulators by Kane and Mele,1) many methods have been proposed to compute the Z2 index. In particular, Fu and Kane found that the calculation of the Z2 index can be consid- erably simplified in a system with inversion symmetry.2) However, for disordered systems, a numerical determi- nation of the Z2 index is still very challenging. Roughly speaking, three types of numerical methods have been proposed so far for this problem: • The first one was proposed by Kane and Mele them- selves,1) and later some modifications were introduced. Basically, for a given model, the Z2 index is defined by a certain Pfaffian with twisted boundary conditions.3) The methods were applied to class AII models4–6) in two and three dimensions with or without a certain inversion symmetry.7–9) • The second one is based on a scattering matrix the- ory.10) The Z2 indices which are defined by the scattering matrices were numerically computed for models in the classes, AII and DIII, in two and three dimensions.10, 11) • The third one was proposed by Loring and Hastings.12) The Z2 indices are defined by Bott indices which are introduced as an obstruction to approximating almost commuting matrices. For models in the class AII in two and three dimensions, the Z2 indices were numerically computed.12, 13) For systems with chiral symmetry, Lor- ing and Schulz-Baldes14) proposed a numerical method to obtain the values of Bott indices. In this paper, we propose an alternative method to numerically calculate the Z2 indices for disordered topo- logical insulators in arbitrary dimensions. The method is based on the index formulae which were derived in Refs. [15, 16]. and has the following two advantages: (i) There is no need to take an average of the Z2 index over random variables in a model. (ii) The Z2 index is de- termined by the discrete spectrum of a certain compact operator with a supersymmetric structure. The latter makes it possible to numerically determine the Z2 index highly efficiently. Our method is demonstrated for Bernevig-Hughes- Zhang (BHZ)17–19) and Wilson-Dirac20–22) models whose topological phases are characterized by a Z2 index of the class AII in two and three dimensions, respectively. In consequence, the method enables us to determine all of the values of the Z2 indices of the strong and weak topological insulators, and the normal insulator phases in the phase diagrams. These values of the Z2 indices com- pletely coincide with the predictions in previous studies using a reliable transfer-matrix method.18, 19, 22) To begin with, we introduce two Dirac operators as15, 16) Da(x) := x1 + ix2 − (a1 + ia2) x1 + ix2 − (a1 + ia2) in two dimension (2D), and Da(x) := 1 x − a (x − a) · σ (1) (2) in three dimensions (3D), where x = (x1, · · · , xd) ∈ Zd is the position operator and a = (a1, · · · , ad) ∈ Rd\Zd is a vector for d = 2, 3 [see Fig. 1(a)]. The three-component vector σ is defined by σ = (σ1, σ2, σ3) whose components are given by Pauli matrices σi, each of which acts on an auxiliary Hilbert space C2. The whole Hilbert space is given by the tensor product of the auxiliary C2 and the original Hilbert space for the Hamiltonian of tight- binding models which we will consider shortly. Next, we define the Z2 index for an infinite-volume system which is a tight-binding model on a square Z2 or cubic lattice Z3.23) Let PF be the projection operator onto the states below the Fermi energy EF. The differ- ence of two projections is defined as16, 24) A :=(PF − D∗ aPFDa PF − DaPFDa in 2D in 3D, (3) where D∗ a is the adjoint of the Dirac operator Da. Then, ∗E-mail address: [email protected] 1 Letter Y. Akagi, H. Katsura, and T. Koma Hamiltonian (7) belongs to the symmetry class AII. We set the Fermi energy EF = 0 which is located at the cen- ter of the energy gap. In the following, we write λi for the i-th eigenvalue of Ω of Eq. (6) in descending order includ- Ω is the operator A(Λ) ing multiplicity.25) We note that the spectrum of A(Λ) included in the interval [−1, 1]. Before showing our numerical results, we abbreviate the topological and the ordinary insulator phases as TI and OI, respectively, in the phase diagram,18, 19) and write ν for the value of the Z2 index. Figure 1(c) and (d) show, respectively, λ1 and λ1 − λ2 as a function of the mass ∆ and the strength W of disorder. In the region TI, our numerical results are satisfactory because λ1 ≃ 1 and λ1 − λ2 6= 0. Actually, these imply ν = 1, i.e., the phase is topological as we expected. In the region OI, λ1 is sig- nificantly different from 1, and thus ν = 0. These results show that the Z2 index enables us to distinguish TI from OI. In OI phase, one notices λ1 ≃ λ2 . 0.2 as seen in Fig. 1(d). This degeneracy is nothing but a consequence of the two symmetries,15, 16) the time-reversal symmetry of the Hamiltonian and the supersymmetric structure of the operator A. This kind of degeneracy is very useful to determine the Z2 index. In the region W . 6, our results are in good agreement with the previous results19) which were obtained by using a transfer-matrix method. In the region with a sufficiently large W , Anderson localization is expected to occur. For the intermediate critical region between these two regions, our method is not under con- trol because the existence of a significantly nonvanishing spectral or mobility gap cannot be expected. In fact, our numerical results in this region show large fluctuations in both λ1 and λ1 − λ2. The second example is the Wilson-Dirac model22) with disorder on the cubic lattice Z3. The Hamiltonian is writ- ten as H3D = H0 + Hhop + Hdis, (8) where H0 = Xx Xk=1,2,3(cid:2) it 2 c† x+ek αkcx− m2 2 c† x+ek βcx +h.c.(cid:3) 2 J. Phys. Soc. Jpn. the Z2 index ν is defined as ν = dim ker (A − 1) modulo 2. (4) When the Fermi energy EF lies in a spectral gap or a mobility gap, the Z2 index is known to be well defined.16) In the following, we will consider such situations. Now we describe our numerical scheme for calculating the Z2 index. Let Λ and Ω be two finite regions satisfying Ω ⊂ Λ ⊂ Rd and a ∈ Ω as in Fig. 1. First, we approxi- mate the Fermi projection PF in the infinite volume by the corresponding Fermi projection in the finite region Λ. More precisely, the approximate one is given by P (Λ) F := XEn≤EF ni hn , (5) where ni are eigenstates of the tight-binding Hamilto- nian H on Λ which we consider, and we have denoted by En the corresponding eigenvalues. To avoid the presence of gapless edge/surface states, we impose periodic bound- ary conditions for the Hamiltonian in practical numerical calculations. For the operator A in Eq. (3), we replace the Fermi projection PF by P (Λ) F , and write A(Λ) for the approximate one. Further, we restrict the operator A(Λ) to the subregion Ω as A(Λ) Ω := χΩA(Λ)χΩ, (6) where χΩ is the characteristic function of the subregion Ω. Under the above gap assumption, the operator A in Eq. (3) is compact even in the infinite volume limit. Therefore, the spectrum of A is discrete and has no ac- cumulation point except for zero. This implies that an eigenstate of A is localized if the corresponding eigen- value is not equal to zero. Let λ 6= 0 be an eigenvalue of A. From these observations, it is clear that the eigen- value λ can be approximated by an eigenvalue λ′ of the approximate operator A(Λ) Ω if the subregion Ω is suffi- ciently large. In addition to this, the cutoff function χΩ in Eq. (6) enables us to remove the boundary effects due to the finite size of the region Λ. As the first demonstration, we compute the Z2 index of the BHZ model17–19) with disorder on a square lattice Z2. The Hamiltonian is written as H2D =Xx Xk=1,2(cid:2)c† xtkcx+ek +h.c.(cid:3)+Xx c† x(τ0 ⊗ǫx)cx, (7) 2 g2τ3 ⊗ s1 − i 2 g2τ0 ⊗ s2 + i where cx = [cx1↑, cx1↓, cx2↑, cx2↓]T , and cxiα is the an- nihilation operator of an electron with spin α and or- bital i at site x. The hopping matrices, t1 and t2, are given by t1 = g1τ0 ⊗ s3 − i 2 g3τ2 ⊗ s3, and t2 = g1τ0 ⊗ s3 − i 2 g3τ2 ⊗ s0, where τk and sk (k = 1, 2, 3) are Pauli matrices for the or- bital and the spin, respectively. Here, g1, g2 and g3 are real parameters. The on-site potential ǫx is given by ǫx = diag[∆ − 4g1 + W + x ], where W + and W − x are a random potential whose distribution is uniform in the interval [−W/2, W/2] with a positive pa- rameter W . e1 and e2 are the unit vectors in the x and y directions, respectively. As is well known,17–19) this x , −∆ + 4g1 + W − x + (m + 3m2)Xx Hhop = Xx Xk=1,2,3(cid:2)t0c† Hdis = Xx vxc† xcx. c† xβcx, x+ek cx + h.c.(cid:3), (9) (10) (11) Here, cx = [cx1↑, cx1↓, cx2↑, cx2↓]T , the vector ek is the unit vector in k = x, y, z direction, and αk = τ1 ⊗ sk, β = τ3 ⊗ s0; vx is the on-site random potential whose distribution is uniform in the interval [−W/2, W/2] with a positive parameter W . This Hamiltonian (8) belongs to the symmetry class AII for W 6= 0 or t0 6= 0. In the following, we will treat the case with t0 6= 0. Similarly, we abbreviate the weak, strong topological, the ordinary insulator and the diffusive metal phases as WTI, STI, OI, and M, respectively,22) and write ν for J. Phys. Soc. Jpn. (a) (b) Letter (a) Y. Akagi, H. Katsura, and T. Koma 3 Ω (cid:16105)x a λ1 λ2(cid:16089) (cid:16105)x a λ1 Λ (d) 12 10 8 W 6 (c) 12 10 8 W 6 4 2 0 OI ν=0 TI ν=1 -2 -1 0 2 3 4 1 Δ 4 2 0 OI ν=0 TI ν=1 -2 -1 0 2 3 4 1 Δ Fig. 1. (Color online). (a) The location of a is chosen to be the center of the finite lattice Λ of linear size L. (b) The subregion Ω is chosen so that its center is the same as a and it does not overlap with the boundary of Λ. (c) and (d), respectively, show λ1 and λ1 − λ2 in the BHZ model as a function of the mass ∆ and the disorder strength W . The obtained value ν of the Z2 index is indicated in both OI and TI phases. The values of the parameters used are EF = 0, g1 = g2 = 1, g3 = 0, and the system size is L2 = 1600. The curves of the phase boundaries with the dots are plotted by using the results in Ref. [19]. the value of the Z2 index.26) Figure 2 shows λ1 and λ1 − λ2 as a function of the mass parameter m0/m2 and the strength W/m2 of disorder. In the region OI, ν = 0 because λ1 . 0.8. In the region STI, λ1 ≃ 1 and λ1 − λ2 6= 0, and hence ν = 1. In the region WTI, λ1 = λ2 ≃ 1 but λ3 is significantly different from 1 [see Fig. 3(b)], and hence ν = 0 because the multiplicity of the eigenvalue λ ≃ 1 is two. As for the region M, we can- not expect our method to be effective because the spec- tral or mobility gap is expected to vanish if the metal- lic character of the spectrum is true. To summarize, as seen in Fig. 2, our numerical results for the Z2 index are in good agreement with the predictions of Ref. [22]. In particular, the phase boundaries between WTI and STI, and between STI and OI are considerably sharp. More- over, although our method is expected to be useless in the metallic phase, there do not appear fluctuations like those in the two-dimensional case. Although the Z2 index vanishes in both the OI and WTI phases, there is a definite difference between them as follows: OI phase yields no eigenvalue λ ≃ 1 while WTI phase yields the eigenvalue λ ≃ 1 which is doubly degenerate. If the eigenvalue λ ≃ 1 is related to surface states, one can expect, from our numerical results, that the multiplicity of λ ≃ 1 coincides with the number of Dirac cones which appear on the surface of the system. In fact, it was numerically observed in a generalized Kane- Mele model on a diamond lattice that WTI (STI) phase exhibits two surface Dirac cones (odd number of Dirac cones).27) M M WTI ν=0 (b) STI ν=1 OI ν=0 WTI ν=0 STI ν=1 OI ν=0 Fig. 2. (Color online). (a) and (b), respectively, show λ1 and λ1 − λ2 in the Wilson-Dirac-type model as a function of the mass parameter m0/m2 and the disorder strength W/m2. The numer- ical values ν of the Z2 index are indicated in the phases WTI, STI and OI. The parameters used are EF = 0, t = 2, t0 = 0.01, and the system size is L3 = 1728. The curves of the phase boundaries with the dots are plotted by using the results in Ref. [22]. Figure 3 shows W/m2 and m0/m2 dependences of the eigenvalues λi, i = 1, 2, . . . , 5, of the operator A(Λ) Ω . Ac- cording to Ref. [22], STI emerges for W/m2 . 7 and for m0/m2 = −1.0. One can see that λ1 is significantly dif- ferent from λ2, and that λ2 and λ3 are degenerate, and similarly, λ4 and λ5 are degenerate. As mentioned in the case of two dimensions, this even degeneracy is a con- sequence of the two symmetries,15, 16) the time-reversal symmetry of the Hamiltonian and the supersymmetric structure of the operator A. For the region W/m2 & 7 in Fig. 3(a), the diffusive metallic phase appears. In this region, one can see that the above double degeneracy in the spectrum of A(Λ) Ω is lifted due to the vanishing of the spectral or mobility gap in the metallic phase. The inset of Fig. 3(a) shows system-size dependences of λ1 and λ2 for fixed parameters m0/m2 = −1.0 and W/m2 = 1.0. As the system size increases, λ1 and λ2 converge to 1 and about 0.7, respectively. Thus in the infinite-volume limit, we can clearly conclude that the Z2 index ν is unity. One can perform a similar analysis and extract the eigenvalues of A in the infinite-volume limit for other values of the parameters. As seen in Fig. 3(b), the double degeneracy of λ ≃ 1 appears in the region of WTI where the corresponding parameter satisfies m0/m2 . −2. Clearly, one can see that λ3 is separated from λ1 ≃ λ2 ≃ 1, and hence the Z2 index ν is equal to zero. Thus, if the first and the sec- ond eigenvalues, λ1 and λ2, are degenerate, information 4 J. Phys. Soc. Jpn. Letter Y. Akagi, H. Katsura, and T. Koma (a) 1.0 0.8 0.6 λ 0.4 0.2 0.0 0 (b) 1.0 0.8 0.6 λ 0.4 0.2 0.0 λ 1.0 0.8 0.6 0.4 0.2 0.0 0.000 λ1 λ2 0.002 0.004 2 4 6 W/m 2 8 1/L3 10 λ1 λ2 λ3 λ4 λ5 λ1 λ2 λ3 λ4 λ5 -3.0 -2.5 -2.0 -1.5 -1.0 -0.5 0.0 0.5 1.0 m 0 /m 2 Fig. 3. (Color online). (a) W/m2 and (b) m0/m2 dependences of the eigenvalues λi, i = 1, 2, . . . , 5, of the operator A(Λ) Ω for the system size L3 = 2744. (a) and (b) correspond to cross-sections of Fig. 2 for m0/m2 = −1.0 and W/m2 = 1.0, respectively. The inset in (a) shows λ1 and λ2 as a function of the inverse volume of the system (1/L3) for fixed parameters, m0/m2 = −1.0 and W/m2 = 1.0. The green (yellow) line is a fit to the numerical data of λ1 (λ2). about the third eigenvalue λ3 is useful to determine the Z2 index. Summary: We have presented our new method for nu- merically calculating the Z2 indices of topological insula- tors with strong disorder. Our method has the following two advantages: (i) There is no need to take an average of the Z2 index over random variables in a model. (ii) The Z2 index is determined by the discrete spectrum of a cer- tain compact operator with a supersymmetric structure. These properties make it possible to numerically deter- mine the Z2 index highly efficiently. In order to check the effectiveness of our method, we have demonstrated that all of the numerical values of the Z2 indices com- pletely coincide with the predictions in previous studies using a reliable transfer-matrix method18, 19, 22) for the two-dimensional Bernevig-Hughes-Zhang and the three- dimensional Wilson-Dirac models. Thus, the strong topo- logical insulator phases can be characterized by the Z2 index in the index formulae,15, 16) and can be clearly dis- tinguished from other phases. We believe that the good agreement between the two different approaches is one of the steps toward the understanding of the nature of Z2 topological insulators although we cannot definitely compare our method with other approaches mentioned at the beginning of the present paper. Finally, we remark that the generalization of our method to models in other symmetry classes in arbitrary dimensions is straightfor- ward. Acknowledgements The authors acknowledge helpful discussions with Ken-Ichiro Imura, Takahiro Misawa, Tomi Ohtsuki and Synge Todo. This work was supported by JSPS KAKENHI Grant Nos. JP15K17719, JP16H00985, JP17K14352. 1) C. L. Kane and E. J. Mele, Z2 Topological Order and the Quantum Spin Hall Effect, Phys. Rev. Lett. 95, 146802 (2005). 2) L. Fu and C. L. Kane, Topological insulators with inversion symmetry, Phys. Rev. B 76, 045302 (2007). 3) Q. Niu, D. J. Thouless, and Y.-S. Wu, Quantized Hall con- ductance as a topological invariant, Phys. Rev. B 31, 3372 (1985). 4) A. P. Schnyder, S. Ryu, A. Furusaki, and A. W. W. Ludwig, Classification of topological insulators and superconductors in three spatial dimensions, Phys. Rev. B 78, 195125 (2008). 5) A. Kitaev, Periodic table for topological insulators and su- perconductors, AIP Conference Proceedings 1134, 22 (2009). 6) S. Ryu, A. P. Schnyder, A. Furusaki, and A. W. W. Ludwig, Topological insulators and superconductors: tenfold way and dimensional hierarchy, New J. Phys. 12, 065010 (2010). 7) A. M. Essin and J. E. Moore, Topological insulators beyond the Brillouin zone via Chern parity, Phys. Rev. B 76, 165307 (2007). 8) H.-M. Guo, Topological invariant in three-dimensional band insulators with disorder, Phys. Rev. B 82, 115122 (2010). 9) B. Leung and E. Prodan, Effect of strong disorder in a three- dimensional topological insulator: Phase diagram and maps of the Z2 invariant, Phys. Rev. B 85, 205136 (2012). 10) I. C. Fulga, F. Hassler, and A. R. Akhmerov, Scattering theory of topological insulators and superconductors, Phys. Rev. B 85, 165409 (2012). 11) B. Sbierski and P. W. Brouwer, Z2 phase diagram of three- dimensional disordered topological insulators via a scattering matrix approach, Phys. Rev. B 89, 155311 (2014). 12) T. A. Loring and M. B. Hastings, Disordered topological in- sulators via C ∗-algebras, EPL (Europhysics Lett.) 92, 67004 (2010). 13) T. A. Loring, K-theory and pseudospectra for topological in- sulators, Ann. Phys. 356, 383 (2015). 14) T. A. Loring and H. Schulz-Baldes, Finite volume calculation of K-theory invariants, Preprint arXiv:1701.07455 (2017). 15) H. Katsura and T. Koma, The Z2 index of disordered topolog- ical insulators with time reversal symmetry, J. Math. Phys. 57, 021903 (2016). 16) H. Katsura and T. Koma, The Noncommutative Index The- orem and the Periodic Table for Disordered Topological Insulators and Superconductors, Preprint arXiv:1611.01928 (2016). 17) B. A. Bernevig, T. L. Hughes, and S.-C. Zhang, Quantum Spin Hall Effect and Topological Phase Transition in HgTe Quantum Wells, Science 314, 1757 (2006). 18) A. Yamakage, K. Nomura, K.-I. Imura, and Y. Kuramoto, Z2 Topological Anderson Insulator, J. Phys.: Conf. Ser. 400, 042070 (2012). 19) A. Yamakage, K. Nomura, K.-I. Imura, and Y. Kuramoto, transition Criticality of driven by disorder, Phys. Rev. B 87, 205141 (2013). the metal-topological insulator 20) K. G. Wilson, Confinement of quarks, Phys. Rev. D 10, 2445 (1974). 21) X.-L. Qi, T. L. Hughes, and S.-C. Zhang, Topological field theory of time-reversal invariant insulators, Phys. Rev. B J. Phys. Soc. Jpn. 78, 195424 (2008). 22) K. Kobayashi, T. Ohtsuki, and K.-I. Imura, Disordered Weak and Strong Topological Insulators, Phys. Rev. Lett. 110, 236803 (2013). 23) We note that a tight-binding model on an arbitrary lattice in d dimension can be mapped onto the model on Zd with suitably chosen hopping integrals. 24) J. Avron, R. Seiler, and B. Simon, The Index of a Pair of Projections, J. Func. Anal. 120, 220 (1994). 25) Here, A(Λ) Da(x) ⊗ I4 as A(Λ) = PF − D∗ is expressed in terms of PF and Da(x) := aPF Da, where I4 is the 4- Letter Y. Akagi, H. Katsura, and T. Koma 5 dimensional identity matrix and Da should be thought of as a matrix of dimension Λ. 26) In this case, the operator A(Λ) is written as A(Λ) = a(x) ⊗ I4 ⊗ σi, and Di PF − Da PF Da, where PF := PF ⊗ I2, Da(x) := P3 i=1 Di a(x) := (xi − ai)/x − a. Here, I2 and I4 are the 2- and 4-dimensional identity matrices, respec- tively. Each operator Di a should be thought of as a matrix of dimension Λ. 27) L. Fu, C. L. Kane, and E. J. Mele, Topological Insulators in Three Dimensions, Phys. Rev. Lett. 98, 106803 (2007). 

1102.0794

2

1102

Optical Forces Between Coupled Plasmonic Nano-particles near Metal Surfaces and Negative Index Material Waveguides

We present a study of light-induced forces between two coupled plasmonic nano-particles above various slab geometries including a metallic half-space and a 280-nm thick negative index material (NIM) slab waveguide. We investigate optical forces by non-perturbatively calculating the scattered electric field via a Green function technique which includes the particle interactions to all orders. For excitation frequencies near the surface plasmon polariton and slow-light waveguide modes of the metal and NIM, respectively, we find rich light-induced forces and significant dynamical back-action effects. Optical quenching is found to be important in both metal and NIM planar geometries, which reduces the spatial range of the achievable inter-particle forces. However, reducing the loss in the NIM allows radiation to propagate through the slow-light modes more efficiently, thus causing the light-induced forces to be more pronounced between the two particles. To highlight the underlying mechanisms by which the particles couple, we connect our Green function calculations to various familiar quantities in quantum optics.

cond-mat.mes-hall

cond-mat

Optical Forces Between Coupled Plasmonic Nano-particles near Metal Surfaces and Negative Index Material Waveguides C. Van Vlack,∗ P. Yao,† and S. Hughes Queen's University, Dept. of Physics, Kingston Ontario, Canada K7L 3N6 (Dated: October 29, 2018) We present a study of light-induced forces between two coupled plasmonic nano-particles above various slab geometries including a metallic half-space and a 280-nm thick negative index material (NIM) slab waveguide. We investigate optical forces by non-perturbatively calculating the scattered electric field via a Green function technique which includes the particle interactions to all orders. For excitation frequencies near the surface plasmon polariton and slow-light waveguide modes of the metal and NIM, respectively, we find rich light-induced forces and significant dynamical back-action effects. Optical quenching is found to be important in both metal and NIM planar geometries, which reduces the spatial range of the achievable inter-particle forces. However, reducing the loss in the NIM allows radiation to propagate through the slow-light modes more efficiently, thus causing the light-induced forces to be more pronounced between the two particles. To highlight the underlying mechanisms by which the particles couple, we connect our Green function calculations to various familiar quantities in quantum optics. PACS numbers: 42.50.-p, 78.67.Bf, 73.20.Mf, 78.67.Pt I. INTRODUCTION Since the proposal of using intense laser light to trap atoms1 and particles2 by Ashkin, scientists have achieved a remarkable ability to manipulate matter via optical forces3. This has lead to a plethora of light-matter force manipulation techniques, from Gaussian beam optical traps4, that are now routinely used in laboratories5, to near-field nanometric tweezers6 which have motivated an entire field involving plasmonics to enhance and manip- ulate optical forces7 -- 9. The use of plasmonic structures allow strong evanescent field enhancements near the plas- mon resonances of the system, and can efficiently trap particles with a size smaller than the wavelength of illu- mination. Examples include a patterned substrate with metallic particles10,11, metallic near-field tips6, or close to nano-apertures in thin metallic films12. In the latter case, Juan et. al.12 recently examined the trapping of 50 nm and 100 nm polystyrene particles near a nano-aperture in a thin sheet of gold and illuminated with light very near to the transmission cutoff wavelength of the aperture. It was observed that as the particles were trapped in this geometry, the scattering induced by the particle worked to enhance the trapping, causing a self-induced "back- action". This back-action illustrates that when consider- ing the optical forces on nano-particles (NPs) near reso- nances, the NPs can interact non-perturbatively with the system and thus any models or theoretical descriptions must include NP coupling in a self-consistent way. Similar to field-enhancements using plasmonic struc- tures, it is also possible to obtain radiation enhancements using metamaterial structures which can be engineered with a negative index of refraction (ε < 0, µ < 0). Vese- lago introduced the concept of a negative refraction in 196813, and, in 2000, Smith et. al.14 experimentally real- ized such a material. This breakthrough initiated the field of "metamaterials" in which patterned materials with unit cells smaller than the wavelength of illumina- tion were created to tune the effective material responses. Soon after, predictions were made of super-lensing15, cloaking devices16,17 and "left-handed" waveguides18, all prompted by the ability to create negative index materi- als (NIMs). Negative index materials also posses interesting quan- tum optics and quantum electrodynamics (QED) prop- erties. For instance, Kastel and Fleischhauer19 exam- ined an idealized (non-absorbing) negative index mate- FIG. 1. (Color online) Schematic of the geometry. We con- sider the optical forces on two NPs with various resonance frequencies and radii of 0.015λ in air above a planar geom- etry. We study both an infinite half-space of silver, and a 280 nm metamaterial slab which supports slow light modes (see text for details). The NPs are illuminated with a plane wave perpendicular to the interface, with the electric field polarization along the axis between the particles. rial placed on top of a mirror with an atom above the en- tire structure. They found that depending on the height of the atom it was possible to obtain complete suppres- sion of spontaneous emission. Additionally, they exam- ined atoms separated by a perfect negative index slab (n = −1) and showed that it was possible to obtain per- fect subradiance and superradiance. Yao et. al.20 exam- ined the enhancement of the spontaneous emission rate (Purcell Factor21) above a NIM slab which supports a collection of slow-light modes in the region of negative index18,22; related Purcell factor enhancements in NIM waveguides have been found by Xu et al.23 and Li et al.24. In the slow light region, very large field enhance- ments can be obtained similar to the enhancements found in plasmonic structures. However, there were two impor- tant differences between the plasmonic and NIM systems: (i) the effects of quenching through non-radiative energy transfer, and (ii) the role of the quasi-static approxima- tion, i.e., the neglect of dynamic retardation effects. In plasmonic systems, it is possible to obtain strong Pur- cell factor enhancements however most of the emission is absorbed by the metal manifesting in non-radiative quenching25. This optical quenching is a result of the intrinsic material loss which, in principle, can be tuned in metamaterial systems. For metals, the quasi-static approximation is also typically used26,27 for small ob- jects approaching the surface (kh < 1, where k is the wavevector in the background medium containing the ob- ject and h is the height above the surface), but for NIMs, such an approximation does not necessarily hold even for kz ∼ 0.0320. is that it is possible to obtain large Purcell factor enhancements in the optical region of spectrum, which also translates to larger spatial distances. Recent experimental results28 have shown the possibilities of engineering the sponta- neous emission rate using metamaterials composed of silver nano-wires embedded in PMMA. The nano-wire structure was shown to exhibit an in-plane anisotropic hyperbolic dispersion curve which have been shown to exhibit a negative refractive index29. By exciting dye molecules embedded on top of the metamaterial the de- cay rate of the dye molecules was shown to be enhanced by a factor of 6 at λ = 800 nm compared to dye deposited on silver and gold films. This was even greater than ex- pected when compared to a semi-classical model which predicted an enhancement of 1.830. Another potential advantage of NIMs, In this work, we theoretically investigate the optical forces on NPs above metallic and NIM planar geome- tries by self-consistently including the interaction of the NPs with the scattered electromagnetic field. We ex- ploit a photon Green function technique that can be used to introduce multiple particles within the dipole approximation31 or, if required, by using the full cou- pled dipole method32 -- 34, where the latter discretizes the NPs as small polarizable subunits. Green function and coupled dipole approaches have been very successful in examining light scattering35 and optical forces31,34 in di- 2 electric particles located in evanescent fields above glass, and for computing optical forces between metallic par- ticles in free space36 and above glass26. Green function approaches have also been used to successfully model op- tical trapping using near field optics37. In addition to computing the light-induced forces, we also analyze the properties of the Green function, both with and without the particles, which contains all the key electromagnetic interactions of the system, including coupling to surface plasmon polaritons (SPP) of the metallic half-space, the localized surface plasmons (LSP) of the particles, and evanescent coupling to slow light modes (SLMs) of the NIM slab. To help clarify the underlying physics, we also make direct comparisons with various well known concepts in quantum optics, such as the Purcell Factor, the Lamb shift, and real and virtual photon exchange; all of these effects are relevant for understanding the ensuing coupling dynamics between the particles. Our paper is organized as follow. In Sec. II we de- scribe the theory of light induced-scattering from NPs close to multi-layered surfaces, including, II A -- the self- consistent calculation of the Green function and the elec- tric field, and II B -- the calculation of the force from the total electric field. We exemplify our theoretical results for silver half-space in Sec. III, and for the NIM slab in Sec. IV. In Sec. V, we conclude. II. THEORY A. Green Function Calculation For a spherical particle with radius a, and in the limit where kBa << 1, the "bare" (i.e., no radiative coupling) polarizability is given by the Clausius-Mossotti relation, α0 (ω) = 4πεBa3 ε (ω) − εB ε (ω) + 2εB , (1) where ε(ω) is the particle dielectric constant (relative electric permittivity) imbedded in a homogeneous mate- rial with a background dielectric constant, εB, assumed to be real. Here kB = ω√εB/c is the wavevector in the background material. It was shown by Draine33 that in order to satisfy the optical theorem, this polarizability must be corrected to include the homogeneous-medium contribution to radiative reaction: α (ω) = α0 (ω) 1 − 3α0(ω)M(ω) 4πεB a3 , (2) where the self-induction term, M (ω), can be calculated exactly for a spherical particle38,39, and for kBa << 1 can be approximated as M = 2i(akB)3/9. Alternatively, this term comes naturally from the homogeneous contribution of the Green function. Our system is initially characterized by an initial electric field E(0) (r; ω), and an initial Green function (0) (r, r′; ω), in the absence of any particles. The field, G E(0) (r; ω), can be of any form we wish (e.g., Gaussian beam, plane wave including reflections from surface), and we define it as the field prior to adding any scatterers. We subsequently introduce N particles into the system where the ith particle is at position ri, with polarizability αi (ω). The total electric field -- excitation field plus par- ticle scattered field -- can be calculated self-consistently from E(N ) (r; ω) = E(0) (r; ω) + N Xi=1 αi (ω) G (0) (r, ri; ω) · E(N ) (ri; ω) . (3) Similarly, the Green function of the system after adding N particles can be calculated via the Dyson equation40, (N ) G (r, r′ω) = G N (0) (r, r′; ω) + Xi=1 (0) αi(ω)G (N ) (r, ri; ω) · G (ri, r′; ω) , (4) which is now the total Green function of the medium, including the response of the NPs. These equations form the basis of the coupled dipole method32, and, impor- tantly, they apply to any general inhomogeneous and lossy media. To help clarify the underlying physics we will fur- ther assume particles with a size much smaller than the wavelength, and consider each NP within the dipole ap- proximation; however it should be noted that at very short inter-particle distances this approximation eventu- ally breaks down41. Thus we will restrict the distances to regimes where the dipole approximation is expected to be a good approximation; in this way, we include the particle dipoles exactly, while essentially dealing with spatially- averaged particle quantities. Using the Dyson equation, we can also rewrite the right hand side of Eq. (3) to be given only in terms of the excitation field38, E(0) (r; ω). One has E(N ) (r; ω) = E(0) (r; ω) + N Xi=1 αi (ω) G (N ) (r, ri; ω) · E(0) (ri; ω) , (5) (N ) where G includes the particle(s) response. The Green function, G, can also be separated into homo- geneous (direct), Ghom, or scattered (indirect), Gscatt, contributions. Since Re[Ghom(r, r′ → r)] diverges, in what follows below we will consider the non-divergent Re[Gscatt(r, r′)] when r = r′, as this is the only rele- vant photonic contribution. While Re[Ghom(r, r)] can give a very small vacuum Lamb shift, this effect can be already included by simply redefining the resonance fre- quency of the particles. In addition, since α(ω) includes the effect of Im[Ghom(r, r)] through the self-induction 3 term, then we need only consider Gscatt(r, r), i.e., when r = r′. For r 6= r′, we consider the full G(r, r) = Gscatt(r, r′) + Ghom(r, r′). Frequently, it is possible to (N ) solve Eqs. (3-5), perturbatively, by considering all G and E(N ) on the RHS to be the unperturbed quantities, and E(0). When the system i.e., approximated by G constituents are far separated this can hold; however, as the particles come closer together and nearer the planar surface, we will show that dynamical coupling become important and cannot be neglected. (0) To examine the mechanisms of light-induced forces, it is useful to consider the Green function of the planar surface with and without the inclusion of the particle(s). We define the complex local density of states (LDOS) as ρ(N ) m (r; ω) = G(N ) mm (r, r; ω) Im [Ghom mm (r, r; ω)] , (6) where G(N ) mm (r, r; ω) is given by Eq. (4). Although the LDOS only depends on Im[ρ(N ) m as a complex quantity for ease of notation. For example, the imaginary part of the Green function in a homogeneous lossless material at r = r′ is given by m ], we introduce ρ(N ) Im(cid:2)Ghom mm (r, r; ω)(cid:3) = ω3√εB 6πc3 , (7) which is related to the homogeneous-medium LDOS. When the homogeneous material contains loss, then both the real and imaginary part of Ghom formally diverge as r → r′ instead of just the real part of Ghom (i.e., as in the case of a lossless material). Consistent with our dis- cussions and notation above, when ρ(N ) m (r; ω) = 0, this means that the LDOS at that point is equal to the homo- geneous density of states, as the scattered contribution is zero; thus the total ρtot m = ρm + 1, and we will focus on ρm. Since we will consider the force interaction between two particles, it is also useful to introduce a complex non-local density of states (NLDOS), ρ(N ) mn (r, r′; ω) = G(N ) Im [Ghom mn (r, r′; ω) mm (r′, r′; ω)] , (8) which describes light propagation between the two space points r and r′ 6= r. Many of these quantities are useful also for connecting to the quantum optical properties of the optical NPs42,43. This is a key strength of the Green function approach over brute-force numerical electromagnetic techniques such as, e.g., FDTD (finite-difference time domain)44. In the case of the complex LDOS, the real part of Eq. (6) can describes frequency shifts (Lamb shifts) of an emitter caused by the environment, whereas the imaginary part describes the material-dependent spontaneous emission. For the photonic Lamb shift, one has δω(N )(r, ω) = − d · Re[G (N ) scatt(r, r; ω)] · d ¯hε0 , (9) for an emitter at position r. For calculations of light- induced optical forces, the real part of the LDOS can therefore manifest itself as resonance shifts of the local electric field. The variation of the imaginary part of the LDOS causes gradient forces on the particle as it moves through the field. In terms of non-local QED interactions between two particles, the real part of the NLDOS describes vir- tual (instantaneous) photon exchange between two points or emitters and the imaginary part describes real (dy- namic) photon exchange. Virtual photon exchange man- ifests itself in well known processes like Forster coupling, which, for a homogeneous medium, exhibit an R−3 scal- ing with inter-particle separation (R) in free space, and real photon exchange manifests itself in dipole-dipole coupling which exhibits an R−1 dependence45. The effect of the real part of the NLDOS on classical optical forces manifests itself in the scattered contribution to the force and the imaginary part would contribute in a manner similar to radiation pressure. For a non-homogeneous medium, we stress that the photon coupling mechanisms 4 are considerably more complicated than simple Forster coupling. In addition, we fully include dynamical retar- dation effects through the frequency-dependence of the response functions. The above prescriptions are relatively straightforward provided one knows the bare Green functions without any particles. For the calculation of the initial planar Green function, we use a well established multilayer tech- nique46, which is outlined in Appendix A and further nu- merical details are detailed by Paulus et. al.47. Although we specialize our study for two NPs, the generalization to any arbitrary number of NPs is straightforward with no change in theoretical formalism. B. Light-Induced Forces Using the electric field E(N ) (ri; ω) at the NP position ri, the time-averaged total force on a particular NP from a time-harmonic electromagnetic wave is48 (ω is implicit), hF (ri)i = ε0 4T Z T /2 −T /2h(cid:16)αiE(N ) (ri) + α∗ i (cid:16)E(N ) (ri)(cid:17)∗(cid:17) · ∇(cid:16)E(N ) (ri) +(cid:16)E(N ) (ri)(cid:17)∗(cid:17) + (cid:16)αi E(N ) (ri) + α∗ i (cid:16) E(N ) (ri)(cid:17)∗(cid:17) ×(cid:16)B(N ) (ri) +(cid:16)B(N ) (ri)(cid:17)∗(cid:17)i , (10) where ε0 is the permittivity of free space, B is the mag- netic field and T is the period of the time-harmonic ra- diation. Upon carrying out the integration, and using B = 1/iω ∇ × E, and E = −iωE, RenαhE(N ) + εjkl εlmnE(N ) j (cid:17)∗ k ∂k(cid:16)E(N ) n (cid:17)∗io , k ∂m(cid:16)E(N ) hFji = ε0 2 (11) where εjkl is the Levi-Civita tensor. Using the relation εjkl εlmn = δjmδkn − δjnδkm, we obtain the desired force (ri)(cid:17)∗i . (ri)∇(cid:16)E(N ) RehαiE(N ) 2 Xj=x,y,z F (ri) = ε0 j j (12) This light-induced force describes the force on the parti- cle due to the electric field and its interaction with the planar geometry as well as the scattering due to the other NPs in the system. This is different from the usual gra- 2 α0∇E2 , which is only ap- dient force given by F = 1 plicable when α is real and the phase of the field varies slowly in space. III. SILVER HALF-SPACE To calculate the optical forces on the NPs, we consider the geometry shown in Fig. 1, with two NPs in air above a planar structure. For silver, we consider a half-space geometry or an optically thick slab with a permittivity given via the Drude model, ε (ω) = εr − ω2 pe ω2 + iωγ . (13) Here εr = 6 is the permittivity as ω → ∞, ωpe = 9.87 eV is the electric plasma frequency, and γ = 51 meV is the damping rate due to collisions49,50. For a metallic half- space, the characteristic surface plasmon polariton (SPP) frequency is given by Re [ε (ω = ωSP P )] = −εB. Below this frequency, SPPs are confined to the interface and can only be coupled to by breaking the symmetry of the sys- tem (e.g., via a grating coupler). For a silver/air interface the SPP is located at ωSP P = 3.73 eV (λSP P = 332 nm). Surface plasmon polaritons are well known for their abil- ity to enhance the electric field in their vicinity, but these enhancements are also associated with high losses mean- ing that coupling between objects or the far field can be suppressed/quenched. For a spherical NP, the localized surface plasmon res- onance is at Re [ε (ω)] = −2εB [see Eq. (1)] which occurs at ωLSP = 3.49 eV (λLSP = 355 nm) using the param- eters for silver and air. To investigate coupling between the NPs and the resonances of the metallic half-space, we will fix εr and γ to the parameters for bulk silver de- spite deviations from bulk-like behavior for small NPs51. ) 3 a 0 ǫ ( / α 400 200 0 −200 4 n 2 0 2 (a) (b) x50 2.5 3 Energy [eV] 3.5 4 FIG. 2. (Color online) Particle and silver half-space response functions. For both the silver half-space and the silver NP permittivity we use the Drude model [Eq. (13)], with γ = 51 meV and εr = 6. For the silver half-space, the plasma frequency is given by ωpe = 9.87 eV. To tune the LSP of the NP to the SPP we set the plasma frequency at ωpe = 10.56 eV; and to tune the LSP of the NP to be off-resonant with the SPP, we set the plasma frequency at ωpe = 7.47 eV. (a) Bare polarizability [Eq. (1)] of a silver NP with radius a = 5 nm (= 0.015 λSP P ) with the LSP resonance tuned to the SPP of the silver half-space, ωSP P = ωLSP = 3.73 eV (solid lines), and with the LSP resonance tuned to be off resonant with the SPP of the silver half-space, ωLSP = 2.63 eV (dashed lines). Gray- light curve indicates real parts and red-dark curve indicates imaginary parts. (b) Refractive index of the silver half-space, gray-light curve indicates real part (scaled by a factor of 50) and red-dark curve indicates imaginary part. For small NPs, the localized surface plasmon of the NP becomes strongly dependent on size52 and shape53 and can be further detuned by adding dielectric or metallic coatings54. We shall use this as motivation to tune the LSP of our NPs to be either on resonance with the SPP (ωLSP = ωSP P = 3.73 eV, ωpe = 10.56 eV), or off res- onance with the SPP (ωLSP = 2.63 eV, ωpe = 7.47eV). The bare polarizability and metal half-space permittivity response functions are shown in Fig. 2. In the following, we consider the geometry shown in Fig. 1, where NPs with tunable LSPs are located above a planar structure. We first use particles with a radius of 0.015 λSP P (= 5 nm) so as to have a reasonable sized particle for which the dipole approximation will apply. We vary the height, h, of the particles above the half- space (measured from the center of the particles) while keeping both particles at the same height for simplicity. Also, we vary the separation, s, between the particles (along the y direction and measured from their centers) and the frequency of light illumination. To illuminate the particles, we use a homogeneous excitation field, which is a solution to the scattering problem (incident light plus scattered light) without any NPs; we choose the polariza- 5 tion to be along the direction of the particles (y direction) to maximize the effect of the coupling between the parti- cles. To excite surface plasmons, which only exist for TM polarization, the symmetry of the system must be bro- ken to allow coupling between the surface plasmon and the particles, so only the scattered field from the NPs can excite the SPP. The incident intensity is 1 W/µm2, which we choose only as a convenient reference -- the force scales linearly with the incident intensity as can be seen from Eq. (12). For comparison, we note that the earth's gravitational force on a 5 nm silver particle is 54 fN; for the intensities considered below, the gravitational force is smaller by a factor of 10−3, and thus can be safely neglected. In Fig. 3 (a) we examine the y component of the LDOS of the half-space, which will dominate the forces for the particular illumination scheme that we have selected. We vary the height [r = (0, 0, h)] keeping in mind that we cannot approach closer to the structure than our particle radius which is indicated by the gray shaded region. We consider ω = ωSP P = ωLSP and add the first particle at r1 = r and the second particle at r2 = (0, 0.05 λSSP , h). When there are no particles in the system, we can see that the imaginary part of the LDOS (red-dark dashed curve) diverges, which would lead to an infinite LDOS for an ex- cited emitter; although this effect may seem surprising, such a divergence always happens above a lossy struc- ture20,55 [Eq. (A10)]. However, the inclusion of the par- ticle where we calculate the LDOS (red-dark chain curve) acts to renormalize the LDOS for distances < 0.08 λ. The addition of a second particle at the same height as the first, but separated by 0.05 λ (center to center), further renormalizes the LDOS (red-dark solid curve), though it becomes apparent that the effect of the silver half-space becomes negligible for heights greater than about 0.06 λ when there are two particles. For the real part of the y component of the LDOS (gray-light curves), and with no particle in the system (dashed), we see that there is a minimum as the par- ticle approaches the half-space which would give a blue shift for an emitter placed close to the surface. Includ- ing a particle at this location (gray-light chain curve) reduces the blue shift and including the second particle (gray solid curve) causes a change in sign which means an emitter would be shifted to the red. For both the real and imaginary part of the LDOS these shifts are only seen by going beyond the perturbative limit and solving Eq. (4) exactly. These results emphasize the pronounced back-action effects that occur in describing the electro- magnetic properties of the medium. Figure 3 (b) shows the yy component of the NLDOS in a similar manner to the LDOS described above, with ω = ωSP P = ωLSP , r′ = r1 = (0, 0, h), and we consider r = r2 = (0, 0.05 λSP P , h). With no particles in the sys- tem, the imaginary part of the NLDOS (red-dark dashed curve) becomes large but remains finite as the surface is approached, implying that photons are very easily trans- ferred from r′ to r (off scale). However once a particle is ωLSP iv 200 (a) (b) iii 6 ωSP P ii ωLSP = ωSP P i ) r ( y ρ m I ) r ( y ρ m I 100 0 150 100 50 0 2 ) r ( y ρ ) ′ r , r ( y y ρ 100 50 0 −50 −100 0 140 120 100 80 60 40 20 0 −20 0 0.05 0.1 h/λ (a) LDOS 0.15 0.2 0.05 0.1 h/λ (b) NLDOS 0.15 0.2 y y (r; ω), ρ(0) (Color online) (a) Complex LDOS, ρ(N) (r′, ω), and FIG. 3. (b) NLDOS, ρ(N) yy (r, r′, ω), for a silver half-space as a func- tion of height at ω = ωSP P . For the LDOS, r = (0, 0, h), and for the NLDOS, r′ = (0, 0, h), r = (0, 0.05 λ, h). Gray- light curves represent the real part and red-dark curves rep- resent the imaginary part. The dashed line has no parti- cles [ρ(0) yy (r, r′; ω)], the chain line has one parti- cle located at the position the LDOS/NLDOS is calculated [ρ(1) yy (r, r′; ω) and r1 = (0, 0, h)], and the solid line has two particles -- one where the LDOS/NLDOS is calcu- lated aand the other at the same height as the first but sep- arated by 0.05 λ in the y direction [ρ(2) yy (r, r′; ω), r1 = (0, 0, h), and r2 = (0, 0.05 λ, h)]. The gray shaded re- gion indicates region where particles would overlap the planar structure. (r; ω), ρ(2) (r; ω), ρ(1) y y added (red-dark chain curve), the NLDOS reduces dras- tically as the particle breaks the symmetry of the sys- tem allowing quenching to occur. Interestingly, the addi- tion of the second particle (red-dark solid curve) further breaks the symmetry of the system and allows light to couple to more channels in the planar structure, which further enhances the quenching effect and reduces the NLDOS. This dramatic reduction is a result of the non- perturbative coupling between the particles and the half- 2.5 3 3.5 4 Energy [eV] y y y FIG. 4. (Color online) Im[ρ(N) (r, ω)] for a silver half-space as a function of frequency at the position r = (0, 0, 0.05 λSP P ). The dashed line has no particles, Im[ρ(0) (r; ω)], the chain line has one particle, Im[ρ(1) (r; ω)], and the solid line has two par- ticles, Im[ρ(2) (r; ω)] at locations r1 = (0, 0, 0.05λSP P ), and r2 = (0, 0.05λSP P , 0.05λSP P ). (a) The particle LSPs are red detuned by ∆ω = 1.1 eV compared to the SPP frequency (see Fig. 2). (b) The particle LSPs are at the SPP frequency. Arrows indicate the particular scenarios we are examining in force graphs: 'i' -- Fig. 5(a), 'ii' -- Fig. 5(b), 'iii' -- Fig. 5(c) and 'iv' -- Fig. 5(d). y space which is theoretically described through the self- consistent solution of Eq. (4). For heights greater than 0.15 λ, light propagates purely via virtual photon prop- agation as the real part (gray-light curve) approaches a finite value (the homogeneous Green function) for both zero (dashed), and one (chain) particle. For two particles this happens even closer to the surface (≈ 0.1 λ) due to the additional scattering events. Real photon propaga- tion occurs when the half-space begins to interact with the system and multiple paths are possible for a photon to reach r from r′. The quasi-static approximation is often invoked for particles very close to a surface or to each other55, and this approximation holds for the imaginary part of the LDOS as the surface is approached at the SPP frequency; the real part deviates significantly in this limit. However, when the incident frequency is detuned from the SPP (ω = 2.63 eV), the quasi-static approximation again be- comes valid. Similar results are found for the NLDOS, except when the inter-particle separation is greater than 0.07 λ and the quasi-static approximation again breaks down. This means that for resonance interactions one must be very careful about applying a quasi-static ap- proximation. It is also useful to examine the LDOS as a function of frequency for fixed particle position, as is shown in λ / h 0.25 0.2 0.15 0.1 0.05 0 0 3 ] N p [ ) F ( g o l 2.5 2 1.5 1 0.5 0 −0.5 −1 0.05 0.1 0.15 0.2 0.25 s/λ λ / h 0.25 0.2 0.15 0.1 0.05 0 0 0.05 0.1 0.15 0.2 0.25 s/λ 7 1 0 −1 −2 −3 −4 ] N p [ ) F ( g o l (a) Force lines on particle 2, at region-'i' on Fig. 4 (b) Force lines on particle 2, at region-'ii' on Fig. 4 λ / h 0.25 0.2 0.15 0.1 0.05 0 0 1 0.5 0 −0.5 −1 −1.5 ] N p [ ) F ( g o l −2 −2.5 −3 0.05 0.1 0.15 0.2 0.25 s/λ λ / h 0.25 0.2 0.15 0.1 0.05 0 0 0.05 0.1 0.15 0.2 0.25 s/λ 3 2 1 0 −1 −2 ] N p [ ) F ( g o l (c) Force lines on particle 2, at region-'iii' on Fig. 4 (d) Force lines on particle 2, using at region-'iv' on Fig. 4 FIG. 5. (Color online) Silver half-space system showing the coupled nano-particle forces. (a) Two dimensional graphs showing inter-particle forces as a function of height and particle separation; the particles are illuminated from above with a plane wave polarized in the y direction at the SPP frequency of the silver half-space and with the NP LSP tuned to be resonant with the SPP frequency. Colorbar indicates magnitude in log scale and arrows indicate directionality of the forces. One particle remains at x1 = y1 = 0 and the second particle is moved in the y direction. The two particles are then both varied in the z direction such that they are always in the same plane. The arrows describe the force on the second particle which is not at the origin. (b) As in (a), but now the NP is far red detuned from the SPP frequency to a LSP resonance of ω = 2.63 eV. (c) As in (a), but the incident field is at a frequency far detuned from the SPP frequency, ω = 2.63 eV. The NP LSP frequency of the particle is resonant with the SPP frequency. (d) As in (c), but with the NP ia also far red detuned to having a LSP resonance at ωLSP = 2.63 eV. Figs. 4(a-b) for different NP detunings (see Fig. 2 and Eqn. 13). In both figures, the dashed lines correspond to ρ(0), the chain lines correspond to ρ(1), and the solid lines correspond to ρ(2). For simplicity we only focus on the imaginary part to examine the effects of the inter- actions. The particles both have their height fixed at h = 0.05 λSP P , and their separation is s = 0.05 λSP P . In Fig. 4(a) we consider a NP that is detuned by ∆ω = 1.1 eV, and in Fig. 4(b) the NP is on resonance with the SPP. In Fig. 4(a), the SPP is visible at ω = 3.73 eV in the LDOS when the particles are far detuned from this resonance, but the particles have a negligible effect on the SPP. Additionally, we see the resonances of the par- ticles interacting and produce a doublet feature caused by photon exchange effects. As the LSP resonances are moved towards the SPP resonance, the high frequency coupled LSP peak merges into the SPP resonance and acts to broaden it as well as detune it. The NLDOS be- havior (not shown) mirrors the effects seen here where the NPs strongly renormalize the NLDOS in the regime of the NP LSP regardless of where the LSP is with re- spect to the SPP. Similar effects for the LDOS and the NLDOS are seen in cavity-QED systems where the non- perturbative coupling between atoms or quantum dots causes additional photon exchange oscillations on top of the vacuum Rabi oscillations56 (the latter occur in sys- tems with suitably small dissipation). With the LDOS and NLDOS calculations acting as a guide, we can now examine the light-induced force calculations for the geometry shown in Fig. 1 and de- scribed above. Four excitation regimes of interest shown in Figs. 5(a)-(d), corresponding to the regions highlighted in Figs. 4(a)-(b). We plot in log scale the intensity and use arrows to indicate the direction of the force. In Fig. 5(a) we illuminate at the SPP frequency and tune the LSP resonance of the particle to be at the same value. The particle separations and heights are both varied up to 0.25 λ (= 83 nm). For an inter-particle separation greater than 0.15 λ, the particles cannot feel each other except when the magnitude of the force is much less than 1 pN, which happens at a height of ≈ 0.125λ and is due to the single particle interaction with the surface. The particles would thus be pushed away and then trapped in stationary positions ≈ 0.125λ above the surface and at a separation of 0.185 λ. Interestingly, as the particles get closer to each other their interaction can still be negligi- ble compared to the particle-surface interaction if their height is smaller than their inter-particle separation; this is caused by quenching which reduces the transfer of ra- diation between the two NPs. If the inter-particle sepa- ration is sufficiently close, and greater than their height above the surface, then the particles strongly optically couple to each other and to the half-space -- as can be seen by the fact that force still varies as the height of the particle varies. The vector force topology seen in this graph manifests itself through the coupling between the systems constituents as their separation varies. This dynamic coupling is very similar to the self-induced back- action demonstrated by Juan et al. 12. In Fig. 5(b), we consider a similar excitation scenario as in Fig. 5(a), except that the LSP of the NP is far red detuned, with ωLSP = 2.63 eV. In this case, we notice that the silver half-space dominates the response and the particles are continually drawn to the surface unless s < 0.15 λ. When s < 0.15 λ the particles can couple to each other and are drawn together, however the effects of the surface seem to be negligible for h > 0.1 λ. In Fig. 5(c) we tune our illumination to ω = 2.63 eV, but keep our LSP resonant with the SPP; note that 0.25 λ =118 nm. For particle separations greater than 0.08λ, the effects of the surface dominate, and for sep- arations below 0.08 λ, the inter-particle effects dominate -- though they still sensitively depend upon height. Both Fig. 5(b) and (c) exhibit very weak inter-particle cou- pling and very weak particle-surface coupling, so that the perturbative expression for Eq. (3) would hold. This is highlighted by the fact that the magnitude of the par- ticle forces are much lower than when we illuminate on the LSP resonance. For our final force example in Fig. 5(d), we examine the case when the LSP and the illumination are both far detuned (ω = 2.63 eV, cf ωSP P = 3.73 eV) from the SPP resonance. We observe three useful coupling regimes: i) When the particles are very close to the surface (< 0.05 λ), and the inter-particle separation is greater than 0.1 λ, we see that the surface completely 8 dominates the forces and the particles are pulled towards the half-space. ii) When the particles are very close to each other (< 0.1 λ), then the inter-particle interaction dominates but this is again mediated by the half-space as there is a height dependence. iii) In the remaining region, we can see that both inter-particle coupling and surface-particle coupling is present where the half-space dominates when h < s and particle-particle coupling is more dominant when h > s. This trend does not con- tinue indefinitely as for h > 0.2 λ we see the inter-particle forces become weaker for equivalent separations. The role of electromagnetic quenching is also minimal as we are so far from the "lossy" SPP resonance. It is worth mentioning again, that the use of the gradi- ent force instead of Eq. (12) would predict an entirely dif- ferent answer. Additional calculations (not shown) show that for the case of Fig. 5(a), the gradient force topog- raphy is completely different, with an additional node along a vertical line at s/λ ≈ 0.16 and no variation of force direction above h/λ = 0.1. Thus using the gradient force for such a strongly perturbed system is generally not valid. IV. NEGATIVE INDEX MATERIAL SLAB WAVEGUIDE We next examine a 280-nm NIM metamaterial slab which supports a negative index in the frequency re- gion ω = 0.78 − 0.92 eV. The relevant NIM and NP response functions are shown in Fig. 6. The possible benefits of using metamaterials is the ability to tune the material properties by engineering the constituents of the unit cell. Negative index metamaterials can be produced with very low loss in the microwave regime, however scaling to the visible has proven to be quite a challenge as the materials become very lossy57, though continued progress is being made with new designs58. We will use NIM parameters that are close to experimental state-of- the-art for communications wavelengths, yet still have a respectable figure-of-merit: F OM = Re(n)/Im(n). Reported figure-of-merits are F OM = 2.0 at 1.8 µm59 and F OM = 0.5 at 780 nm60; for our calculations, F OM ≈ 1.0 at ω = 0.78 eV. The permittivity is given by Eq. (13) with εr = 1, ωpe = 2.03 eV and γ = 8.3 meV. The permeability is given by the Lorentz model61, µ (ω) = 1 + ω2 pm ω2 0 − ω2 − iωγ , (14) with the magnetic plasma frequency ωpm = 0.69 eV and the atomic resonance frequency ω0 = 0.78 eV. Detailed descriptions of the exact corresponding com- plex band structure and Purcell effect are given by Yao et al.20 for the same parameters as given above. Here, we briefly point out a few features of interest for this study. At ω0, the dispersion curves of all of the leaky slow light modes (SLMs) of the system converge at a sin- gle frequency, ωSLM = ω0; thus particles near the slab ) 3 a 0 ǫ ( / α 400 200 0 (a) −200 0.75 20 n 0 (b) −20 0.75 0.8 0.85 Energy [eV] 0.9 0.95 9 100 80 60 40 20 0 ) r ( y ρ 0.8 0.85 Energy [eV] 0.9 0.95 −20 0 0.05 0.1 h/λ (a) LDOS 0.15 0.2 FIG. 6. (Color online) Particle and NIM slab response func- tions. For both the NIM and the NP permittivity we use the Drude model [Eq. (13)], and for the NIM we model the permeability with a Lorentzian [Eq. (14)]. (a) Bare polar- izability [Eq. (1)] of a nanoparticle with the LSP resonance tuned to the SLM of the 280 nm thick metamaterial slab, ωSLM = ωLSP = 0.78 eV (solid lines) (ωpe = 2.22 eV), and with the LSP resonance tuned to be off resonant with the LSP of the 280 nm metamaterial slab, ωLSP = 0.89 eV (dashed lines) (ωpe = 2.53 eV). In both cases γ = 11 meV. Gray- light curves indicates real parts and red-dark curves indicates imaginary parts. (b) Refractive index of the slab, where the gray-light curve indicates real part and the red-dark curve in- dicates imaginary part. Parameters are εr = 1, ωpe = 2.03 eV, γ = 8.3 meV, ωpm0.69 eV and ω0 = 0.78 eV. can couple to many different slow light modes at this frequency. The slow light frequency regime gives rise to an enhancement in the LDOS and correspondingly to an increased Purcell effect. An increase in the LDOS is also seen near the SPP modes of the metallic surfaces, but the effects of quenching in metallic systems reduces the amount of light that escapes to the far field and the typical propagation distances of SPPs are limited by the material loss. However, slow light modes could, in prin- ciple, propagate for much longer distances. Also, the typical scaling laws associated with the quasi-static ap- proximation are not reached, even very close to the slab20 (because of the strong magnetic resonance). Thus for our study, we are never really in the quasi-static regime above a NIM slab and we must consider retardation effects. It is also worth noting that NIM slabs support both TE (transverse electric) and TM (transverse magnetic) SPPs, which is in contrast to metallic surfaces that only support TM SPPs. The SPP modes in NIMs will not be discussed here, as their general properties are similar to the SPP of metals and at higher frequencies (ωT E SP P = 0.92 eV and ωT M SP P = 1.43 eV)20. For the metamaterial system, we consider a particle with a scaled radius of 0.015λSLM = 24 nm, and we tune our NP to be in the frequency regime of our peak LDOS ) ′ r , r ( y y ρ 140 120 100 80 60 40 20 0 −20 0 0.05 0.1 h/λ (b) NLDOS 0.15 0.2 y (Color online) (a) Complex LDOS, ρ(N) (r′, ω), FIG. 7. and (b) NLDOS, ρ(N) yy (r, r′, ω), for a 280 nm thick meta- material slab as a function of height at ω = ωSLM . For the LDOS r = (0, 0, h), and for the NLDOS r′ = (0, 0, h), r = (0, 0.05λ, h). Gray-light curves represent the real part and red-dark curves represent the imaginary part. The dashed line has no particles [ρ(0) yy (r, r′; ω)], the chain line has one particle located at the position the LDOS/NLDOS is calculated [ρ(1) yy (r, r′; ω) and r1 = (0, 0, h)], and the solid line has two particles; one where the LDOS/NLDOS is calculated at the other at the same height as the first but separated by 0.05λ in the y direction [ρ(2) yy (r, r′; ω), r1 = (0, 0, h), and r2 = (0, 0.05λ, h)]. Gray shaded region in- dicates region where particles would overlap the surface. (r; ω), ρ(1) (r; ω), ρ(0) (r; ω), ρ(2) y y y associated with the SLMs (at ω0, see Fig. 6); practically, such tuning may be achieved, e.g., by using nano-shell structures62. Additionally, we reduce the NP damping rate to γ = 11 meV to examine the SLM features which would otherwise be obscured. For the metamaterial slab, we expect large enhancements of LDOS at the slow-light modes frequency similar to Ref. [20], however it is not obvious what the inter-particle coupling effects will be, nor the role of inter-particle coupling from the waveg- uide modes. Similar to the metal half-space case [Figs. 3 (a-b)], we first examine the LDOS and the NLDOS in Figs. 7(a-b) as a function of height. In Fig. 7(a), the real (gray-light curve) and imaginary parts (red-dark curve) of the LDOS again diverge as the slab is approached [Eq. (A10)] when no particles are in the system, however there is a change in sign of the real part compared to the metallic case, indicating the Lamb shift would be a red shift instead of a blue shift. The imaginary part (Purcell factor) reaches a value of 100 at 0.02λSLM = 31.7 nm compared to the metallic case which reaches a value of 100 at 0.05 λSP P = 16.6 nm. Thus the metamaterial gives an equivalent enhancement at twice the distance in absolute units. Introducing a particle at the location of the LDOS [r1 = r = (0, 0, h)] renormalizes both the real and imaginary parts of the LDOS at small h/λ and re- moves some of the divergence behavior for small distances close to the slab -- similar to the metallic case. We also see that the maximum of the real part is no longer lo- cated closest to the surface. The addition of the second particle [r2 = (0, 0.05λ, h)] increases the imaginary part to a constant for h > 0.08 λ to almost exactly the same value as for the metallic case which is due to the fact that we are in the quasi-static limit for the homogenous inter- action between the particles and the slab no longer plays a role. The real part dips slightly below zero indicating that there can be either a blue or a red shift depending on the height of the particles and stays below zero for h > 0.025 λ. For the NLDOS [Fig. 7(b)], the real part (gray-light curve) follows a very similar trend as in the metallic case, where the zero particle case (dashed curve) is reduced as the slab is approached but is finite. Including the first particle (chain) drastically decreases the real part and thus the virtual photon exchange and the second parti- cle (solid) further reduces it. At closest approach the real part is small but still greater in magnitude to the metallic case by a factor of 20. The imaginary part with no particles (red-dark dashed curve) qualitatively follows the metallic case however when a particle is included in the system (red-dark chain curve), instead of reducing the real photon transfer there is an increase. The in- clusion of the second particle (solid curve) reduces the effect again but we still are able to increase coupling be- tween the particles compared to the metallic case. This transfer can be further increased as it crucially depends on the metamaterial loss used in the effective permittiv- ity and permeability. Obtaining lower losses is possible by improving metamaterial fabrication techniques which would result in less lossy slow-light propagation modes. A comparison of the LDOS in terms of frequency for the NIM slab is shown in Fig. 8 for different NP detun- ings. Again, the particles both have their height fixed at z = z′ = 0.05 λSLM = 79 nm, and their separation is 0.05 λSLM . In Fig. 8(a) we consider a NP that is blue detuned by ∆ω = 0.11 eV, and Fig. 4(b), the NP is on resonance with the slab SLMs. When the NP is detuned from the slow-light modes (non-resonant case), we see that there is still a large enhancement of the LDOS at ωSLM ωLSP 10 ii ωLSP = ωSLM i (a) iv (b) 100 50 0 150 100 50 ) r ( y ρ m I ) r ( y ρ m I 0 0.75 iii 0.8 0.85 0.9 Energy [eV] y (Color online) Im[ρ(N) y (r; ω)], the chain line has one particle, Im[ρ(1) (r, ω)] for a 280 nm thick FIG. 8. metamaterial slab as a function of frequency at the posi- tion r = (0, 0, 0.05λSLM ). The dashed line has no particles, Im[ρ(0) (r; ω)], and the solid line has two particles, Im[ρ(2) (r; ω)], at locations r1 = (0, 0, 0.05 λSLM ), and r2 = (0, 0.05λSLM , 0.05λSLM ). (a) The particle LSPs are blue detuned by ∆ω = 0.11 eV compared to the slab SLM frequency (see Fig. 6). (b) The particle LSPs are at the slab SLM frequency. Arrows indicate the particular scenarios we are examining in force graphs: 'i' -- Fig. 9(a), 'ii' -- Fig. 9(b), 'iii' -- Fig. 9(c) and 'iv' -- Fig. 9(d). y y the NP resonance compared with the zero particle case, but this is essentially the homogeneous space coupling due to the particles and is only slightly altered by the presence of the slab. When the NP is tuned to the ωSLM resonance there is a greater enhancement than in free space but the coupling is dominated by inter-particle in- teractions. If we include only one particle, it is evident that the inter-particle effects dominate the spectrum as the LDOS more closely follows the zero particle case. To illustrate the effect on light-induced forces, we again consider four different cases that are highlighted in Figs. 8(a-d). We illuminate with a plane wave at the slow-light resonance frequency of the metamaterial slab (ωSLM = 0.78 eV), first with the NP on-resonance [Fig. 9 (a)], and then with the NP off-resonance [Fig. 9(b)]. We then illuminate off resonance (ω = 0.89 eV) tuning the nanoparticle to be on-resonance with the slow-light modes [Fig. 9(c)], and then to be off-resonant and at the same frequency of the illumination [Fig. 9(d)]. All figures show the log of the magnitude of the force in intensity scale and arrows indicate directionality. When the NP and the slow-light modes are both on- resonance with the incident radiation [Fig. 9(a)], we see a very similar situation as when the NP was on resonance with the SPP [Fig. 5(a)]. For s < 0.2 λ there is a divi- λ / h 0.25 0.2 0.15 0.1 0.05 0 0 0.05 0.1 0.15 0.2 0.25 s/λ 5 4 3 2 1 0 −1 ] N p [ ) F ( g o l λ / h 0.25 0.2 0.15 0.1 0.05 0 0 11 3 ] N p [ ) F ( g o l 2.5 2 1.5 1 0.5 0 −0.5 −1 0.05 0.1 0.15 0.2 0.25 s/λ (a) Force lines on particle 2, at region-'i' on Fig. 8 (b) Force lines on particle 2, at region-'ii' on Fig. 8 λ / h 0.25 0.2 0.15 0.1 0.05 0 0 0.05 0.1 0.15 0.2 0.25 s/λ 3 2 1 0 −1 −2 −3 ] N p [ ) F ( g o l λ / h 0.25 0.2 0.15 0.1 0.05 0 0 0.05 0.1 0.15 0.2 0.25 s/λ 5 4 3 2 1 0 −1 ] N p [ ) F ( g o l (c) Force lines on particle 2, at region-'iii' on Fig. 8 (d) Force lines on particle 2, using at region-'iv' on Fig. 8 FIG. 9. (Color online) Metamaterial slab system showing the coupled nano-particle forces. (a) Two dimensional graphs showing inter-particle forces as a function of height and particle separation; the particles are illuminated from above with a plane wave polarized in the y direction at the SLM frequency of the slab (ωSLM = 0.78 eV). The NP LSP is also tuned to be resonant with the SLM frequency. Here one particle remains at x1 = y1 = 0 and the second particle is moved in the y direction. The two particles are then both varied in the z direction such that they are always in the same plane. The arrows describe the force on the second particle which is not at the origin. (b) As in (a), but now the NP is blue detuned from the SLM frequency to a LSP resonance of ω = 0.89 eV. (c) As in (a), but the incident field is at detuned to ω = ωLSP = 0.89 eV. (d) As in (c), but with the NP ia also detuned to having a LSP resonance at ω = 0.89 eV. sion along the line s ≈ 2 h where below this line the slab dominates the forces, above this line the inter-particle in- teraction dominates the forces and around which we see a combination of the two. As the height gets above 0.1 λ we see that this does not continue indefinitely and the inter- particle interaction becomes weaker and shorter ranged as the slab no longer enhances the coupling between the two. Particles that have a small initial separation will be pushed away from each other and to a height of ≈ 0.08 λ. Note the forces here are an order of magnitude greater than in the metallic case, which is mostly due to the po- larizability scaling with the particle size but these could in principle be further tuned by improving the loss in these structures. When the NP is tuned off-resonance from both ωSLM and the incident frequency [Fig. 9(b)], the long range coupling is lost compared to Fig. 9(a). For s > 0.12λ, the force is dominated by slab interactions and are con- tinually drawn to the surface of the slab. For s < 0.1 λ, and h < 0.1 λ we see that the particles and the slab are all interacting which results in the particles being pulled towards the slab and together however the interaction range is short. When h > 0.1 λ the particles essentially only interact with each other and are mostly drawn to- gether. Figures 9(c-d) show light-induced force calculations with the incident radiation detuned from the slow-light mode frequency to ω = 0.89 eV. In Fig. 9(c) the NP LSP is tuned to be resonant with the SLMs and off-resonant with the radiation. For s > 0.1 λ the particles are un- λ / h 0.25 0.2 0.15 0.1 0.05 0 0 0.05 0.1 0.15 0.2 0.25 s/λ 5 4 3 2 1 0 −1 ] N p [ ) F ( g o l FIG. 10. msterial loss, (γ′ = γ/10) (Color online) As in Fig. 9(a) but with a low affected by each other and are repelled from the slab, except when they are almost touching. For s < 0.1λ the slab dominates over the inter-particle interaction for h < 0.1 λ which is in contrast to Fig. 9(b). For h > 0.1 λ the slab interaction weakens and the particles begin to interact and repel each other. Next, we examine the case where both the radiation and the NP are detuned to 0.89 eV and away from the SLMs, shown in Fig. 9(d). We see that, similar to Fig. 5(d), there are three different interaction regions, i) h < 0.08 λ and s > 0.1 λ, ii) h < 0.08 λ and s < 0.1 λ, iii) and h > 0.08 λ. In the first region, the slab domi- nates the force by pulling the particle when it is almost touching and pushing the particle away when it is slightly higher h > 0.02 λ until the point where the vertical force becomes negligible and the particles are attracted to each In region ii), the particles are other at h ≈ 0.08 λ. causing a dramatic renormalization of the Green func- tion which leads to very strong, position dependent par- ticle interactions which are pushing the particles away from the slab and each other until they get to h ≈ 0.1 λ, s ≈ 0.1 λ. Finally, above h = 0.1 λ the particle interac- tion dominates but is mediated by the height above the slab and causes the particles to essentially be repelled to a fixed separation of s ≈ 0.08 λ, as the separation in- creases the slab starts to draw the particles towards it again however the particles will still be pulled towards s ≈ 0.08 λ. To investigate the influence of metamaterial loss on the inter-particle forces, we show the same scenario as Fig. 9 (a) where the LSP and radiation is resonant with the SLMs, but we now decrease the material loss by a factor of 10, thus γ = 0.83 meV. We show the resulting force in Fig. 10, and see a number of important differences. First, where there was once a fixed height at which the particles would be attracted to, this height now varies with inter-particle separation. Below this dividing line in the region where h < 0.08 λ and s > 0.18 λ there are 12 60 50 40 30 20 10 0 −10 −20 0 0.05 0.1 0.15 0.2 0.25 s/λ ) ′ r , r ( y y ρ (a) NLDOS for a nominal material loss, γ = 8.3 meV ) ′ r , r ( y y ρ 60 50 40 30 20 10 0 −10 −20 0 0.05 0.1 0.15 0.2 0.25 s/λ (b) NLDOS for a lower material loss, γ = 0.83 meV (Color online) Complex NLDOS, ρ(N) yy (r, r′, ω), for FIG. 11. a 280 nm thick and metamaterial slab with regular loss (a) or with low-loss (b) as a function of separation [r′ = (0, 0, 0.05λ), r = (0, s, 0.05λ)] at ω = ωSLM . Gray-light curves represent the real part and red-dark curves represent the imaginary part. The dashed line has no particles [ρ(0) yy (r, r′; ω)], the chain line has one particle located at the position the NL- DOS is calculated [ρ(1) yy (r, r′; ω) and r1 = (0, 0, 0.05λ)], and the solid line has two particles; one where the NLDOS is cal- culated at the other at the same height as the first but sep- arated along the y direction [ρ(2) (r, r′; ω), r1 = (0, 0, 0.05λ), and r2 = (0, s, 0.05λ)]. Gray shaded region indicates region where particles would overlap each other. y still inter-particle forces where in the regular loss case these forces have since died away. Finally, in the regular loss case, as the particles are moved vertically, along the line s = 0.04 λ we see that the particles are attracted to each other close the slab, h < 0.125 λ, but are repulsive at higher distances. This is contrasted in the low loss case where there is a division at s = 0.045 λ, below which the particles are always attracted and above which the particles are always repelled. To further examine how the loss alters the long-range coupling effects, we plot the NLDOS in Fig. 11 for both regular and low loss metamaterial slabs at a fixed height, h = 0.05 λ, and vary the separation between r′ = (0, 0, 0.05λ) and r = (0, s, 0.05 λ). In both cases we see that the real part (gray-light curve) diverges at low s when there are no particles in the system (dashed line), due to the homogeneous part of the Green function. The addition of particles once more renormalizes these val- ues. We also see that in the low loss case, the real part plateaus between s = 0.08 λ and s = 0.0125 λ, whereas in the nominal loss case this simply decays, so material loss has a large influence on the light-induced forces The imaginary part of the NLDOS (red-dark curves) varies slowly towards zero in the regular loss case, however we see that the imaginary part of the NLDOS in the low loss case oscillates around a value of -2, with a much larger amplitude. The increase in the NLDOS allows the par- ticles to couple much farther in Fig. 10 than in Fig. 9 (a). Thus for decreasing material losses, the NPs can be coupled over longer distances where this coupling is mediated by the slow light waveguide modes. V. CONCLUSIONS We have introduced a theoretical formalism to com- pute the Green function response of small particles within the vicinity of multi-layered geometries. We have applied this theory to calculate the non-perturbative force inter- actions between two NPs in the vicinity of surface plas- mon polariton modes for a metal half-space geometry, and in the vicinity of a slow-light NIM waveguide. Both planar structures facilitate a large local density of states, and non-local photon interactions between the particles. We have found that both structures exhibit rich but sim- ilar force maps despite the different mechanisms for in- creasing the LDOS. When both particle and slab (metal and NIM) are on resonance with the incident illumina- tion, the particles will be pushed away from each other 13 and pushed to a fixed height above the slab (Figs. 5(a) and 9(a)). Such an effect would aid in preventing the ag- gregation of NPs. When the particle and illumination are off resonance with the slab then the particles are pulled to a specific height and pulled towards each other up to a fixed distance (Figs. 5(d) and 9(d)) which would enable the creation of dimers. In all the other cases the most likely scenarios are the particles being pulled towards the slab or the particles being pushed away from the slab. For metallic surfaces, the material parameters are largely fixed, limiting some of the engineering available to such structures, however metamaterials in principle have the ability to have their intrinsic parameters tuned by changing the basic unit cell. Such tunability will al- low simplified geometries such as the planar structures to aid in the creation of long range optical forces for the trapping and localization of small NPs. The same structures also exhibit rich and fundamentally interest- ing QED phenomena offering applications for radiative decay engineering of embedded quantum light sources. ACKNOWLEDGMENTS This work was supported by the National Sciences and Engineering Research Council of Canada and the Canadian Foundation for Innovation. We gratefully ac- knowledge Mark Patterson for assistance with the multi- layered Green function calculations. Appendix A: Planar Green function The Green function above a planar structure can be written in terms of its angular spectrum42 which involves decomposing the wavevector k = kxx + ky y + kzz into its various components and integrating over each contri- bution. The real-space homogeneous Green function Ghom (r, r′) = − (A1) where kB = ω√εB/c is the wavevector in the background material, and the z-component is given by kz = B − k2 (cid:0)k2 . The matrix f hom is given by f homeikx(x−x′)+iky(y−y ′)+ikzz−z ′dkxdky, y(cid:1)1/2 x − k2 δ (R) + −∞ iω2 8π2c2εB Z ∞ −∞Z ∞ zz εB f hom = 1 kz , (A2)   B − k2 k2 −kxky k2 ∓kxkz ∓kykz k2 x −kxky ∓kxkz y ∓kykz B − k2 B − k2 z   where the upper sign is used when z > z′ and the lower sign is used when z < z′. The scattered part of the Green function in a multilayer environment (no particles) can be written similarly in terms of s and p polarized contributions, Gscatt (r, r′) = iω2 8π2c2εB Z Z ∞ −∞hf s scatt + f s scatt = f rs (kx, ky) y(cid:1) kz (cid:0)k2 x + k2 p scatti eikx(x−x′)+iky(y−y ′)+ikz(z−z ′)dkxdky,    −kxky 0  , 0 0 −kxky k2 x 0 k2 y 0 (A3) (A4) p scatt = f rp (kx, ky) k2 B (cid:0)k2 y(cid:1) x + k2 kBk2 x kxkykz   −kx(cid:0)k2 kxkykz kBk2 y kx(cid:0)k2 ky (cid:0)k2 x + k2 y(cid:1) −(cid:0)k2 x + k2 y(cid:1) x + k2 y(cid:1) y(cid:1) /kz x + k2 y(cid:1) −ky(cid:0)k2 x + k2 14 (A5)   . s scatt and f p Here the matrices (or dyadics) f scatt are given in terms of the reflection coefficients, rs/p, for s and p polarization above the multilayer. For the three-layer ge- ometry of a slab with height h considered here, the upper, background layer having εB = ε1, µB = µ1, the middle layer having ε2, µ2 and the lower layer having ε3, µ3 these reflection coefficients are 12 + , (A6) rs/p = rs/p 23 e2iβh 23 e2iβh 12 ts/p ts/p 1 − rs/p where β = ±(cid:0)ω2ε2µ2/c2 − k2 x + k2 Re(cid:0)ω2ε2µ2/c2(cid:1) > Re(cid:0)k2 is the z com- ponent of the wavevector in the middle layer when The sign of β depends on whether or not the refractive index of the middle layer is positive (upper) or negative , β = 21 rs/p 21 rs/p y(cid:1)1/2 x − k2 y(cid:1)1/2 (lower)63,64. For Re(cid:0)ω2ε2µ2/c2(cid:1) < Re(cid:0)k2 i(cid:0)k2 for both positive and negative index materials. The single-layer reflection and transmis- sion coefficients are y − ω2ε2µ2/c2(cid:1)1/2 y(cid:1)1/2 x + k2 x + k2 . rs ij = µjkiz − µikjz µjkiz + µikjz , rp ij = εjkiz − εikjz εjkiz + εikjz (A7) ts ij = 2µjkiz µjkiz + µikjz , tp ij = 2εjkiz εjkiz + εikjz . (A8) Solutions of Eqs. (A1) and (A3) for real dielectrics can be difficult due to poles close or along the path of inte- gration, however this can be solved by numerically inte- grating around the poles in the complex plane which lie in a known region47. For lossy NIMs the poles along the integration path are found to be in the lower part of the complex plane, whereas for lossy positive index materi- als the poles are located in the upper half of the complex plane20. The solution described by Paulus et. al.47 is more complicated for materials which are able to sup- port negative index modes and surface plasmons as the location of the poles in the complex plane are essentially given by the complex band structure of the material65. The largest contributions to the Green function are no x + k2 y(cid:1)1/2 longer confined to the region where Re(cid:0)ω2ε2µ2/c2(cid:1) > Re(cid:0)k2 and careful attention must be paid to the integrand. This is trivial for a small number of cal- culations but can be cumbersome when many locations are required. As an example, for a two particle force cal- culation at a single point, the above calculations required 14 separate Green function calculations when employing the dipole approximation. When considering the Green function in the quasi- static approximation, the homogeneous space Green function66 is given by Ghom,QS (r, r′) = 1 4πεBR3 (cid:18) 3RR R2 − I(cid:19) , (A9) where I is the unit dyadic (diagonal terms are unity and non-diagonal terms are zero). The total Green function above a half-space in the quasi-static approximation in- volves the direct contribution from Eq. (A9) as well as the scattered contribution from an image source located beneath the surface55, GQS (r, r′) = Ghom,QS (r, r′) ∓ ε2 − ε1 ε2 + ε1 Ghom,QS (r, r′′) . (A10) Here the minus sign is for x/y directed dipoles and the plus sign is for z directed dipoles. The location of the image charge is given by r′′ which is related to r′ via, x′ = x′′, y′ = y′′, and z′ = −z′′ when the surface of the half-space is located at z = 0. The scattering part of the Green function is then given by, Gscatt,QS (r, r′) = ∓ ε2 − ε1 ε2 + ε1 Ghom,QS (r, r′′) . (A11) ∗ [email protected] † Current address: Department of Optics and Optical En- gineering, University of Science and Technology of China, 230026, P. R. China 1 A. Ashkin, Phys. Rev. Lett. 40, 729 (1978). 2 A. Ashkin, Science 210, 1081 (1980). 3 D. G. Grier, Nature 424, 21 (2003). 4 J. P. Barton, D. R. Alexander, and S. A. Schaub, J. Appl. Phys., 66, 4594 (1989). 5 K. C. Neuman and S. M. Block, Rev. Sci. Instrum., 75, 6 L. Novotny, R. X. Bian, and X. S. Xie, Phys. Rev. Lett. 79, 645 (1997). 7 J. A. Schuller, E. S. Barnard, W. Cai, Y. C. Jun, J. S. White, and M. L. Brongersma, Nat. Mat. 9, 193 (2010). 8 R. Quidant and C. Girard, Laser and Photonics Review 2, 47 (2008). 9 K. Dholakia and P. Zem´anek, Rev. Mod. Phys. 82, 1767 (2010). 10 M. Righini, A. S. Zelenina, C. Girard, and R. Quidant, Nat. Phys. 3, 477 (2007). 2787 (2004). 11 A. N. Grigorenko, N. W. Roberts, M. R. Dickinson, and 40 O. J. F. Martin, A. Dereux, and C. Girard, J. Opt. Soc. 15 Y. Zhang, Nat. Phot. 2, 365 (2008). 12 M. L. Juan, R. Gordon, Y. Pang, F. Eftekhari, and R. Quidant, Nat. Phys. 5, 915 (2009). 13 V. G. Veselago, Sov. Phys. Usp. 10, 509 (1968). 14 D. R. Smith, W. J. Padilla, D. C. Vier, S. C. Nemat-Nasser, and S. Schultz, Phys. Rev. Lett. 84, 4184 (2000). 15 J. B. Pendry, Phys. Rev. Lett. 85, 3966 (2000). 16 J. B. Pendry, D. Schurig, and D. R. Smith, Science 312, 1780 (2006). 17 D. Schurig, J. J. Mock, B. J. Justice, S. A. Cummer, J. B. Pendry, A. F. Starr, and D. R. Smith, Science 314, 977 (2006). 18 I. V. Shadrivov, A. A. Sukhorukov, and Y. S. Kivshar, Phys. Rev. E 67, 057602 (2003). 19 J. Kastel and M. Fleischhauer, Phys. Rev. A 71, 011804 (2005). 20 P. Yao, C. Van Vlack, A. Reza, M. Patterson, M. M. Dignam, and S. Hughes, Phys. Rev. B 80, 195106 (2009). 21 E. M. Purcell, Phys. Rev. 69, 681 (1946). 22 W. T. Lu, S. Savo, B. Didier, F. Casse, and S. Sridhar, Microwave Opt. Technol. Lett. 51, 2705 (2009). 23 J.-P. Xu, Y.-P. Yang, Q. Lin, and S.-Y. Zhu, Phys. Rev. A 79, 043812 (2009). 24 G.-X. Li, J. Evers, and C. H. Keitel, Phys. Rev. B 80, 045102 (2009). 25 P. Anger, P. Bharadwaj, and L. Novotny, Phys. Rev. Lett. 96, 113002 (2006). 26 P. C. Chaumet and M. Nieto-Vesperinas, Phys. Rev. B 62, 11185 (2000). 27 K. Joulain, R. Carminati, J.-P. Mulet, and J.-J. Greffet, Phys. Rev. B 68, 245405 (2003). 28 M. A. Noginov, H. Li, Y. A. Barnakov, D. Dryden, G. Nataraj, G. Zhu, C. E. Bonner, M. Mayy, Z. Jacob, and E. E. Narimanov, Opt. Lett. 35, 1863 (2010). 29 V. A. Podolskiy and E. E. Narimanov, Phys. Rev. B 71, 201101 (2005). 30 Z. Jacob, I. Smolyaninov, and E. Narimanov, arXiv:0910.3981v2 (2009). 31 P. C. Chaumet and M. Nieto-Vesperinas, Phys. Rev. B 64, 035422 (2001). 32 E. M. Purcell and C. R. Pennypacker, Astrophys J. 186, 805 (1973). 33 B. Draine, Astrophys J. 333, 848 (1988). 34 P. C. Chaumet and M. Nieto-Vesperinas, Phys. Rev. B 61, Am. A 11, 1073 (1994). 41 V. D. Miljkovi´c, T. Pakizeh, B. Sepulveda, P. Johansson, and M. Kall, J. Phys. Chem. C 114, 7472 (2010). 42 L. Novotny and B. Hecht, Principles of Nano-Optics (Cambridge, 2006). 43 W. Vogel, , and G. Welsch, Quantum Optics (Wiley-VCH, 2006). 44 A. Taflove and S. C. Hagness, Computational Electro- dynamics: The Finite-Difference Time-Domain Method (Artech House, 2005). 45 P. Thomas, M. Moller, R. Eichmann, T. Meier, T. Stroucken, and A. Knorr, Phys. Status Solidi B 230, 25 (2002). 46 J. E. Sipe, J. Opt. Soc. Am. B 4, 481 (1987). 47 M. Paulus, P. Gay-Balmaz, and O. J. F. Martin, Phys. Rev. E. 62, 5797 (2000). 48 P. C. Chaumet and M. Nieto-Vesperinas, Opt. Lett. 25, 1065 (2000). 49 Y. Liu, G. Bartal, and X. Zhang, Opt. Exp. 16, 15439 (2008). 50 P. B. Johnson and R. W. Christy, Phys. Rev. B 6, 4370 (1972). 51 W. P. Halperin, Rev. Mod. Phys. 58, 533 (1986). 52 J. M. McMahon, S. K. Gray, and G. C. Schatz, Phys. Rev. Lett. 103, 097403 (2009). 53 J. Ye, P. Van Dorpe, W. Van Roy, G. Borghs, and G. Maes, Langmuir 25, 1822 (2009). 54 J. Zhang, Y. Tang, L. Weng, and M. Ouyang, Nano. Lett. 9, 4061 (2009). 55 P. Gay-Balmaz and O. J. F. Martin, Opt. Comm. 184, 37 (2000). 56 P. Yao and S. Hughes, Opt. Exp. 17, 11505 (2009). 57 A. Boltassevaa and V. M. Shalaev, Metamaterials 2, 1 (2008). 58 S. P. Burgos, R. de Waele, A. Polman, and H. A. Atwater, Nat. Mat. 9, 407 (2010). 59 S. Zhang, W. Fan, K. J. Malloy, S. R. J. Brueck, N. C. Panoiu, and R. M. Osgood, J. Opt. Soc. Am. B 23, 434 (2006). 60 G. Dolling, M. Wegener, C. M. Soukoulis, and S. Linden, Opt. Lett. 32, 53 (2007). 61 A. Reza, M. M. Dignam, and S. Hughes, Nature 455, 312 (2008). 62 R. Bardhan, N. K. Grady, T. Ali, and N. J. Halas, ACS 14119 (2000). 35 P. C. Chaumet, A. Rahmani, F. de Fornel, and J.-P. Du- four, Phys. Rev. B 58, 2310 (1998). 36 J. R. Arias-Gonz´alez and M. Nieto-Vesperinas, J. Opt. Soc. Am. A 20, 1201 (2003). 37 P. C. Chaumet, A. Rahmani, and M. Nieto-Vesperinas, Nano 4, 6169 (2010). 63 H.T. Dung, S.Y. Buhmann, L. Knoll, D.G. Welsch, S. Scheel, and J. Kastel, Phys. Rev. A 68, 043816 (2003). 64 S. A. Ramakrishna and O. J. F. Martin, Opt. Lett. 30, 2626 (2005). 65 A. Archambault, T. V. Teperik, F. Marquier, and J. J. Phys. Rev. Lett. 88, 123601 (2002). 38 O. J. F. Martin and N. B. Piller, Phys. Rev. E 58, 3909 Greffet, Phys. Rev. B 79, 195414 (2009). 66 P. Gay-Balmaz and O. J. F. Martin, Appl. Opt. 40, 4562 (1998). 39 A. Yaghjian, Proc. IEEE 68, 248 (1980). (2001). 

1910.09481

1

1910

Generation of exchange magnons in thin ferromagnetic films by ultrashort acoustic pulses

We investigate generation of exchange magnons by ultrashort, picosecond acoustic pulses propagating through ferromagnetic thin films. Using the Landau-Lifshitz-Gilbert equations we derive the dispersion relation for exchange magnons for an external magnetic field tilted with respect to the film normal. Decomposing the solution in a series of standing spin wave modes, we derive a system of ordinary differential equations and driven harmonic oscillator equations describing the dynamics of individual magnon mode. The external magnetoelastic driving force is given by the time-dependent spatial Fourier components of acoustic strain pulses inside the layer. Dependencies of the magnon excitation efficiencies on the duration of the acoustic pulses and the external magnetic field highlight the role of acoustic bandwidth and phonon-magnon phase matching. Our simulations for ferromagnetic nickel evidence the possibility of ultrafast magneto-acoustic excitation of exchange magnons within the bandwidth of acoustic pulses in thin samples under conditions readily obtained in femtosecond pump-probe experiments.

cond-mat.mes-hall

cond-mat

Generation of exchange magnons in thin ferromagnetic films by ultrashort acoustic pulses V.Besse, A.V. Golov, V.S. Vlasov1,2,∗ A. Alekhin1, D. Kuzmin3,4, I.V. Bychkov3,4, L.N. Kotov3, and V.V. Temnov1† 1IMMM CNRS 6283, Le Mans Universit´e, 72085 Le Mans cedex, France 2Syktyvkar State University named after Pitirim Sorokin, 167001, Syktyvkar, Russian Federation 3Chelyabinsk State University, 454001 Chelyabinsk, Russian Federation and 4South Ural State University (National Research University), 454080 Chelyabinsk, Russian Federation Abstract We investigate generation of exchange magnons by ultrashort, picosecond acoustic pulses prop- agating through ferromagnetic thin films. Using the Landau-Lifshitz-Gilbert equations we derive the dispersion relation for exchange magnons for an external magnetic field tilted with respect to the film normal. Decomposing the solution in a series of standing spin wave modes, we derive a system of ordinary differential equations and driven harmonic oscillator equations describing the dynamics of individual magnon mode. The external magnetoelastic driving force is given by the time-dependent spatial Fourier components of acoustic strain pulses inside the layer. Dependen- cies of the magnon excitation efficiencies on the duration of the acoustic pulses and the external magnetic field highlight the role of acoustic bandwidth and phonon-magnon phase matching. Our simulations for ferromagnetic nickel evidence the possibility of ultrafast magneto-acoustic excita- tion of exchange magnons within the bandwidth of acoustic pulses in thin samples under conditions readily obtained in femtosecond pump-probe experiments. PACS numbers: Valid PACS appear here 9 1 0 2 t c O 1 2 ] l l a h - s e m . t a m - d n o c [ 1 v 1 8 4 9 0 . 0 1 9 1 : v i X r a ∗ [email protected] † [email protected] 1 I. INTRODUCTION The discovery of ultrafast laser-induced demagnetization in ferromagnetic nickel in 1996 [1] opened new booming field of femtosecond laser manipulation of magnetization [2 -- 21]. Depending on a mechanism, magnetization dynamics induced by a femtosecond laser pulse can be sorted into the following categories: (i) thermal effects [3, 13], (ii) magne- tooptical effects [22], (iii) magnetoacoustic effects [15, 18, 19, 23 -- 27] and (iv) spin transfer torque [28 -- 31]. Despite of the facts that the spin transfer torque is an efficient mechanism to drive coherent magnetization dynamics in ferromagnet through ultrashort pulses of spin- polarized electrons propagating in noble metals for several tens of nanometers before they lose their spin polarization [32], acoustic pulses can propagate over much larger distances before they vanish. Picosecond acoustic pulses generated by femtosecond laser excitations [25, 33 -- 35] can drive the small-angle precession of ferromagnetic resonance (FMR) in various ferromagnetic samples [15, 19, 23, 24]. Experimental configurations using fs-laser excited periodic acoustic transients [21, 26, 27] can be used to resonantly enhance the amplitude of FMR precession angle. Previous experimental work on ultrafast magnetoacoustics by Kim and co-workers [20, 21] demonstrated the possibility to control the ferromagnetic res- onance (FMR) [36] by a series of ultrashort, picosecond acoustic pulses. The theoretical treatment based on the phenomenological analysis of Landau-Lifshitz-Gilbert (LLG) equa- tions accounting for some selected terms of the thermodynamic free energy density F (the magnetocrystalline and magnetoelastic anisotropies, the demagnetization energy and the external magnetic field energy) was sufficient to interpret their experimental observations. Though, it did not allow to investigate the possibility to acoustically excite other elementary magnetic excitations such as exchange magnons [3]. While searching for fingerprints of elastic magnon excitation, Bombeck et al. [24] reported on the magnetic-field dependence of two closely spaced frequencies in the magneto-optical response of an elastically driven 200 nm thick magnetic semiconductor (Ga,Mn)As. One of the modes was claimed to be the low-order exchange mode, but the theoretical intepretation rooted on a very specific spatial profiles of exchange magnons, strongly dependent on the unknown magnetic boundary conditions. Kim and Bigot also observed the conspicuous beating of magneto-acoustic signals in a free-standing 300 nm thick ferromagnetic nickel film [21], but explained the beating within the framework of the magnetoacoustic coherent control of FMR excitations, i.e. without involving any exchange magnons. Clearly, there 2 is a need for a simple and transparent theory describing the magnetoelastic generation of exchange magnons using picosecond acoustic pulses. In this article we develop such magnetoelastic theory and predict the possibility of ul- trafast magnetoelastic excitation of exchange magnons by ultrashort acoustic pulses in thin ferromagnetic films. This theory is based on the Landau-Lifshitz-Gilbert (LLG) equations where ultrashort acoustic strain pulses propagating through a ferromagnetic thin film modify its magnetoelastic energy and drive precessional dynamics of standing modes of exchange magnons, which can be also described by a simple equation of a driven harmonic oscillator. Being applied to a 30-nm nickel film excited by picosecond pulses of longitudinal acous- tic phonons, our theory demonstrates the excitation of exchange magnons with frequencies within the bandwidth of the acoustic pulses. II. EXPERIMENTAL GEOMETRY AND MAIN EQUATIONS Our numerical simulations are conducted for ferromagnetic nickel on which the pioneer- ing ultrafast laser-induced demagnetization [1] and the most recent magneto-acoustic ex- periments [19 -- 21] have been performed by the group of Jean-Yves Bigot. We report a numerical study of exchange magnons in a 30-nm nickel film excited by picosecond acoustic pulses. Typical experimental configuration for magneto-acoustic [19 -- 21] and magnonic [37] measurements is presented in Fig. 1 (a). It utilizes a rotating permanent magnet placed on the top of a ferromagnetic sample which allows to apply external magnetic field with a magnitude up to a few hundreds of millitesla at an arbitrary angle ξ with respect to the surface normal. The equilibrium direction m0 of the magnetization vector M = M0m0 is usually non-collinear with the external magnetic field due to the magnetic anisotropies and points at an angle θ with respect to the surface normal. Acoustic pulses with duration τ and spatial width csτ smaller than the sample thickness L propagate through the film and locally alter the direction of the effective magnetic field Heff thereby driving precessional motion of the magnetization. The resulting magnetization dynamics can be represented as a sum of different magnon modes: homogeneous precession of the magnetization (FMR) and exchange-coupled non-uniform magnon modes. 3 (a) Ferromagnetic layer FMR precession FMR+magnon precession (b) s(t) 800 600 400 Acoustic pulse Magnon precession n=2 n=1 n=0 (FMR) s(t) z ) z H G ( y c n e u q e r F 0.3 Tesla, ξ=45 1.0 Tesla, ξ=45 3.0 Tesla, ξ=45 6.5 Tesla, ξ=45 Acoustic phonons ° ,θ =75.2 ° ,θ =60.2 ° ,θ =50.7 ° ,θ =47.7 ° ° ° ° n=2 n=1 n=3 FMR 200 0 0 1 2 3 4 5 6 7 8 9 Normalized wavevector kL/π FIG. 1. (a) Picosecond pulses of longitudinal acoustic phonons propagating at the speed of sound cs through a ferromagnetic layer of thickness L can excite simultaneously homogeneous (FMR, corresponding to the magnon mode with n=0) and non-uniform magnetization precession (magnon modes with n>0) around the equilibrium direction tilted by an angle θ with respect to the film normal. The external magnetic field is tilted by an angle ξ. (b) Magnon dispersion can be tuned by the amplitude of an external magnetic field. The crossing points of magnon and phonon dispersions are marked with black dots. III. MAGNETIZATION DYNAMICS EXCITED BY ACOUSTIC PULSES The Landau-Lifschitz-Gilbert (LLG) equations ∂m ∂t = −γµ0m (t, z) × Heff (t, z) + αm (t, z) × ∂m ∂t (1) represent the most common tool to model the spatio-temporal dynamics of the unit mag- netization vector m (t, z) driven by the effective magnetic field Heff(z, t). Here, the α is the dimensionless phenomenological Gilbert damping parameter and γ demotes the gyro- magnetic ratio. The effective magnetic field is a functional derivative of the free energy density Hef f = − 1 µ0M0 ∂F ∂m + 1 M0 3 Xp=1 ∂ ∂xp , ∂U ∂xp(cid:17) ∂(cid:16) ∂m 4 (2) where M0 is a saturation magnetization. For the purposes of this investigation we define the free density energy of a ferromagnetic thin film as a sum F = Fz + Fd + Fex + Fme(t, z) , which includes the Zeeman conribution Fz = −µ0M0m · H with an external magnetic field H, the demagnetizing field energy Fd = 1 2 µ0M 2 0 m · N · m determined by the demagnetization tensor N = 0 0 0 0 0 0 0 0 1     for a thin film geometry, the exchange energy Fex = 1 2 M0 3 Xp=1 ∂xp(cid:19)2 D(cid:18) ∂m characterized by the exchange stiffness D, and the magnetoelastic energy Fme(t, z) = b1m2 zǫzz(z, t) , (3) (4) (5) (6) (7) (8) In the latter term, the magnetoelastic constant b1 couples the normal magnetization compo- nent mz to the dynamic strain ǫzz(t, z) propagating in the z-direction (here our consideration is limited to a single non-zero strain component ǫzz). While neglecting a weak dependence of the exchange stiffness D on the applied strain, here we assume that the magnetoelastic interactions are driven solely by the last time-dependent magnetoelastic term Fme(z, t) in Eq. (3). Given that the free energy density in Eq. (3) represents a superposition of different terms, the effective magnetic field in Eq. (9) also appears to be a sum of the corresponding contri- butions: Heff = H + Hd + Hex + Hme(z, t) (9) 5 where H is the external magnetic field, Hd is the demagnetization field, Hex is the exchange field and the time-dependent magnetoelastic field Hme(t, z). After calculation of the func- tional derivatives from Eqs. (4), (5), (7) and (8) and their sum, we obtain the total effective magnetic field: Heff,x = D Heff,y = D Heff,z = D ∂2mx ∂z2 + H cos ξ , ∂2my ∂z2 ∂2mz ∂z2 + H sin ξ − M0mz − , 2b1 µ0M0 mzǫzz(z, t) . (10) (11) (12) Now, when the driving force for the LLG equations Eq.(1) is known, we first analyze the dissipation free case, i.e. α = 0: ∂m ∂t = −γµ0m × Hef f . (13) Second, we introduce a small dynamic perturbation s(z, t) ≪ m0 of the magnetization vector oscillating around the equilibrium magnetization direction m0 = (sin θ, 0, cos θ): mx = cos θ + sx(z, t) , my = sy(z, t) , mz = sin θ + sz(z, t) . (14) (15) (16) By substituting Eqs. (14-16) in Eqs. (10-13) and keeping only the linear terms in si, we obtain the following system of differential equations: ∂2sy ∂z2 + (H cos ξ − M0 cos θ)sy = 2b1 µ0M0 ǫzz(t, z)sy cos θ , (17) ∂2sz ∂z2 ] − (H cos ξ − M0 cos θ)sx + 1 γµ0 ∂sx ∂t − D cos θ 1 γµ0 ∂sy ∂t + D[cos θ ∂2sx ∂z2 − sin θ 2b1 µ0M0 +(H sin ξ + M0 sin θ)sz = − ǫzz(z, t) (sx cos θ + sz sin θ + sin θ cos θ) , 1 γµ0 ∂sz ∂t + D sin θ ∂2sy ∂z2 − (H sin ξ)sy = 0 . (18) (19) In real experiments, strains are usually small ǫzz(z, t) < 0.01, which allows us also to neglect the mixed terms ∝ siǫzz(z, t) in comparison with other terms linear in si and ǫzz and further 6 simplify these equations to: 1 γµ0 ∂sx ∂t + D[cos θ 1 γµ0 ∂sy ∂t − D cos θ ∂2sx ∂z2 − sin θ ∂2sy ∂z2 + (H cos ξ − M0 cos θ)sy = 0 , ∂2sz ∂z2 ] − (H cos ξ − M0 cos θ)sx + +(H sin ξ + M0 sin θ)sz = − 2b1 sin θ cos θ µ0M0 ǫzz(z, t) , 1 γµ0 ∂sz ∂t + D sin θ ∂2sy ∂z2 − (H sin ξ)sy = 0 . (20) (21) (22) The magnetoelastic term ∝ b1 sin θ cos θǫzz(z, t) on the right-hand side of Eq. (21) drives the magnetization dynamics. It becomes zero for in-plane (θ = 90◦) or out-of-plane (θ = 0) static magnetization directions. This observation highlights the importance of a tilted magnetic configuration for magnetoelastic studies. Substituting the plane waves si = ciexp(iωt − ikz) in Eq. (20, 21, 22), where the afore- mentioned time-dependent magnetoelastic driving term is neglected, leads to the following secular equation iω −A12(k) 0 −A21(k) iω −A23(k) 0 −A32(k) iω (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) with coefficients Aij(k) defined as: = 0 , (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) A12(k) = −γµ0(cid:2)(cid:0)Dk2 − M0(cid:1) cos θ + H cos ξ(cid:3) A21(k) = −A12 A23(k) = −γµ0(cid:2)(cid:0)Dk2 + M0(cid:1) sin θ + H sin ξ(cid:3) A32(k) = γµ0(cid:2)Dk2 sin θ + H sin ξ(cid:3) . (23) (24) (25) (26) (27) The secular equation provides the dispersion relation for the magnon modes at the frequency ω propagating in z-direction with the wave vector k: ω(k) = p−A12A21 − A23A32 . (28) At k = 0 and some particular orientations of the magnetic field, i.e. in-plane versus out- of-plane, this dispersion relation is reduced to well-known Kittel equations for the FMR frequency [36]. For large non-zero k, when exchange interactions dominate, the dispersion relation becomes quadratic ω(k) ∼ Dk2. This dispersion relation can be tuned both by the amplitude and the direction of the external magnetic field, as shown in Fig. 1. As noted 7 previously, the tilted orientation of the external magnetic field is crucial for magnetoelastic interactions. This is why we have conducted numerical simulations for an external magnetic field tilted by 45◦ with respect to the film normal and inspected the results as a function of the magnitude of the external magnetic field. The dependence on the external magnetic field is shown in Fig. 1(b). For small magnetic field the magnon dispersion crosses the acoustic dispersion twice: at low frequency slightly above the FMR frequency and at very high frequency of order of several hundred GHz corre- sponding to large k-vectors. In strong magnetic fields the magnon dispersion is upshifted and both frequencies get closer and merge in a single point when the parabolic magnon dispersion touches the linear acoustic dispersion and the phase-matching is fulfilled over a wide range of frequencies. In case of monochromatic excitations the analysis of phase-matching condi- tions would be sufficient to predict the elastically driven magnetization dynamics. However, here we consider ultrashort acoustic pulse possessing very broad frequency spectrum and propagating through a ferromagnetic samples with a thickness of a few tens of nanometers, which requires the analysis beyond the phase-matching conditions. In a ferromagnetic film with a finite thickness, only discrete number of magnonic modes are supported and survive on a larger time scale after the excitation. These modes are created by interference of two counter-propagating spin waves with wave vectors ±kn. Discretization of k-vector is determined by the boundary conditions for the dynamic magnetization at the interfaces between the magnetic layer and the adjacent material. Without the loss of generality we choose free boundary conditions = 0 , (29) ∂si ∂z (cid:12)(cid:12)(cid:12)(cid:12)z=0,L which result in cosine-like magnon eigenmodes ∝ cos(knz) with kn = πn/L and the dynamic magnetization s(z, t) = N Xn=0 s(n)(t) cos knz (30) is represented as a sum over all contributing magnon modes; N is the number of modes with non-zero amplitudes used in numerical calculations. We substitute this expression in the system of equations Eq. (20, 21, 22), where we now keep the time-dependent magnetoelastic driving force represented in a form of the Fourier series. Due to orthogonality of the magnon eigenmodes, integration of the equations over the film thickness from 0 to L leads to the system of decoupled ordinary differential equations for time-dependent amplitudes s(n)(t) of 8 all magnon modes: ds(n) x dt ds(n) y dt ds(n) z dt = A(n) 12 s(n) y , = A(n) 21 s(n) x + A(n) 23 s(n) z + = A(n) 32 s(n) y , 2γb1 sin θ cos θ M0 ǫ(n) zz (t) , (31) (32) (33) with A(n) ij = Aij(kn). The magnetization dynamics are driven by the time-dependent Fourier components of the elastic strain pulses ǫ(n) zz (t) = 1 L Z L 0 ǫzz(z, t) cos(knz)dz . (34) The dynamic strain component ǫzz,n acts on the nth magnon mode as an external driving force. Ultrashort, picosecond acoustic pulses with the length shorter than the film thickness produce a multitude of non-zero Fourier components. These Fourier components ǫ(n) zz (t) are time-dependent: they change not only when an acoustic pulse enters or leaves a ferromagnetic film, but also when it propagates through the sample. The physical insight in the mechanisms of the magnetoelastic interactions in thick films can be obtained from the theoretical analysis of the integrals ǫ(n) 0 ǫzz(z − cst) cos(knz)dz accounting for the propagation of acoustic pulses through the ferromagnetic layer at the speed of sound cs, where the dominant role L R L zz = 1 of phase-matching conditions between magnons and phonons can be elucidated. In this paper we are going to study a special case of a ferromagnetic thin film and it comes out that in order to excite high-frequency exchange magnons of the order n, the Fourier spectra of acoustic pulses (in k-space, along the propagation direction z) should possess non-zero components at the respective wavevector kn. As we are going to see in the next section, this condition is fulfilled for picosecond acoustic pulses generated by ultrashort laser pulses. The system of equations (31,32,33) can be reduced to an equation of a harmonic oscillator for each magnon mode. Taking the time-derivative of Eq. (32) and using Eqs. (31, 33) it is easy to obtain the following equation: d2s(n) y dt2 + ω2 ns(n) y = 2γb1 sin θ cos θ M0 dǫ(n) zz (t) dt , (35) where ωn = ω(kn) = q−A(n) 12 A(n) 21 − A(n) 23 A(n) 32 is the magnon frequency given by Eq. 28. Tak- ing into account that the most common experimental configuration for the polar magneto- optical Kerr effect measures the z-component of the magnetization vector, it is useful to 9 derive an equation for s(n) z . For that we extract s(n) y from Eq. (33) and substitute it in Eq.( 35): z d3s(n) dt3 + ω2 n ds(n) z dt = 2A(n) 32 γb1 sin θ cos θ M0 dǫ(n) zz (t) dt Integration over time leads to the usual oscillator equation: z d2s(n) dt2 + ω2 ns(n) z = 2A(n) 32 γb1 sin θ cos θ M0 ǫ(n) zz (t) (36) (37) By adding the damping term and the explicit expression for A(n) 32 from Eq. (27) we arrive at the most important analytical equation in this paper: z d2s(n) dt2 + αωn ds(n) z dt + ω2 ns(n) z = µ0γ2b1 sin(2θ)(Dk2 M0L n sin θ + H sin ξ) Z L 0 ǫzz(z, t) cos(knz)dz . (38) The magnon decay constant αωn, determined by the Gilbert damping parameter α, is in- cluded in a phenomenological way based on the fact that the magnon lifetime 1/(αωn) is inversely proportional to its frequency, this scaling verified up to the THz frequencies [30]. In spite of some possible deviations of this approximation for the in-plane geometry of the external magnetic field, this approximation appears to be accurate for high frequency ex- change magnons under the tilted magnetic field [37]. The time-dependent driving force on the right hand side displays a non-trivial dependence both on the tilt angles θ and ξ as well as the magnon order n. Moreover, this equation as well as the entire theory are valid for arbitrary acoustic strains ǫzz(z, t). IV. RESULTS OF NUMERICAL SIMULATIONS Here we apply the developed theory to calculate the magnetization dynamics in a poly- crystalline nickel film induced by picosecond acoustic pulses. Since the frequency interval between the neighbouring magnon modes is mainly determined by the film thickness and their widths in the Fourier spectra are given by inverse lifetimes, here we consider a rather small film thickness of L=30 nm. It allows individual magnon modes to be resolved in frequency domain. Moreover, the velocity of longitudinal acoustic waves in polycrystalline nickel cs=5.6 nm/ps provides a rather short acoustic travel time through the sample of L/cs=5.4 ps, which is short enough compared to the characteristic magnon lifetimes. It makes it possible to observe the magnon dynamics after the acoustic pulse escaped from the ferromagnetic layer. The damping-free system of ordinary differential equations (31-33) is 10 solved by the 4th-order Runge-Kutta method. At each Runge-Kutta integration time step ∆t all magnon components s(n) are multiplied by exp(−αωn∆t) providing magnon life times i nearly identical to those obtained from the numerical solutions of the LLG equations (1). In order to calculate the magnetization dynamics mi(z, t), we are summing up over all magnon modes using Eq. (30) truncated at N=20 as the amplitudes of higher magnon modes are negligibly small for all simulations discussed in this manuscript. For numerical calculations for ferromagnetic nickel we used nickel the Gilbert damping parameter α = 0.05 [3] and magnetoelastic coupling constant b1 = 107 J/m3 [38]. Being aware of the importance of the acoustic bandwidth from the analysis of Eqs. (31- 33), we have performed the numerical simulations for three values of the acoustic pulse duration: 1, 2 and 3 ps corresponding to their spatial width in nickel of 5.6, 11.2 and 16.8 nm, respectively. Ultrashort acoustic pulses with 2-3 ps duration can be routinely generated in thin metal samples excited by femtosecond laser pulses [33, 34, 39]. Even shorter acoustic pulses with ultimate pulse durations in the deeply subpicosecond range have been observed in form of acoustic solitons resulting from the nonlinear propagation effects of ultrashort acoustic pulses in crystalline solids at cryogenic temperatures [40]. Whereas subpicosecond soliton strains can get as high as 2 × 10−3, acoustic pulses generated in ferromagentic thin films can reach 10−2 = 1% amplitudes [34]. As such, large amplitude ultrashort acoustic pulses can be routinely obtained in every laboratory for femtosecond laser spectroscopy. At present we do not discuss the details of acoustic injection in a ferromagnetic thin film and just assume it is sandwiched between two acoustically matched nonmagnetic materials allowing for an ultrashort acoustic pulse to be injected through the front interface (z=0) at zero time and leave it through the back interface (z = L) 5.4 picoseconds later. We have analyzed the magnetization dynamics at both interfaces for different amplitudes of the external magnetic field. In case of a thin sample, the magnon dynamics at both interfaces are quite similar. For this reason, in this paper we focus on the data obtained for the normal component of the magnetization sz(t) at the back interface and their Fourier spectra calculated for three acoustic pulse durations τ =1, 2 and 3 ps (with pulse shapes given by a half of the period of the cosine function) and four values of an external magnetic field of 0.3, 1, 3, and 6.5 Tesla, respectively (see Fig. 2). For small value of the magnetic field µ0H = 0.3 T the Fourier spectra for all three pulse durations consist of a strong FMR peak at 11.6 GHz frequency and high-frequency magnon peaks. Being much weaker as compared to the FMR, the peaks for two lowest exchange modes at f1=24.2 GHz and f2=59.2 GHz 11 0.02 ) t ( z s 0 -0.02 0.05 ) t ( z s 0 -0.05 0.05 ) t ( z s 0 τ = 1 ps τ = 2 ps τ = 3 ps 0.3 T 0 200 400 600 800 Delay time (ps) 40 20 ) ω ( z s 0 0 40 ) ω ( z s 20 1.0 T 0 100 200 300 400 Delay time (ps) ) ω ( s z f0 f1 f2 τ = 1 ps τ = 2 ps τ = 3 ps 0.3 T 20 40 60 80 100 Frequency (GHz) f3 f0 f1 f2 1.0 T 50 100 150 200 Frequency (GHz) f0 f1 f4 100 200 Frequency (GHz) f1f0 300 f6 0 0 20 10 0 0 4 3.0 T 3.0 T 50 100 150 200 Delay time (ps) -0.05 0 0.05 ) t ( z s 0 -0.05 6.5 T 0 10 20 30 40 50 Delay time (ps) ) ω ( z s 2 6.5 T 0 0 200 400 Frequency (GHz) 600 FIG. 2. Variations of the magnetization component at the back interface sz(t, z = L) and their Fourier spectra obtained for three different acoustic pulse durations τ =1, 2 and 3 ps and four values of an external magnetic field µ0H=0.3, 1, 3 and 6.5 T, respectively. The amplitude of acoustic pulses is 0.5%, the magnetic field is tilted by ξ=45◦. are still visible in the Fourier spectra. As the magnetic field is increased, all frequencies are shifted up and the magnon amplitudes increase relatively to the FMR frequency. At the magnetic field of µ0H = 3.0 T, the amplitude of he first magnon f1 = 93.2 GHz almost reaches the strength of the FMR mode oscillating at f0 = 81.8 GHz. This observation correlates very well with the concept in Fig. 1(b) showing that the magnon dispersion crosses the phonon dispersion precisely at the frequency of the first magnon mode, i.e. the phonon- 12 0.02 ) t ( z s 0 -0.02 0.05 ) t ( z s 0 -0.05 0.05 ) t ( z s 0 -0.05 ) t ( z s 0.04 0.02 0 -0.02 -0.04 0 0 0 0 f0 f3 f7 τ = 1 ps τ = 2 ps τ = 3 ps 0.3 T 0.4 0.2 ) ω ( z s τ = 1 ps τ = 2 ps τ = 3 ps 0.3 T 10 20 Delay time (ps) 30 0 0 f0 0.4 500 Frequency (GHz) f7 1000 f9 1.0 T ) ω ( 1.0 T 30 z s 0.2 0 0 1 f0 ) ω ( z s 0.5 0 0 4 ) ω ( z s 2 0 0 3.0 T 30 6.5 T 30 10 20 Delay time (ps) 10 20 Delay time (ps) 10 20 Delay time (ps) 500 Frequency (GHz) 1000 f9 3.0 T 500 1000 Frequency (GHz) f0 f8 6.5 T 500 1000 Frequency (GHz) FIG. 3. Variations of the magnetization component at the back interface sz(t, z = L) at their Fourier spectra for three different acoustic pulse durations τ =1, 2 and 3 ps and four values of an external magnetic field µ0H=0.3, 1, 3 and 6.5 T, respectively. The shaded area represent the acoustic pulses and their spectra, respectively. Acoustic pulses injected at zero delay time escape from the nickel layer at 5.4 ps, as indicated by the dashed vertical lines. The amplitude of acoustic pulses is 0.5%, the magnetic field is tilted by ξ=45◦. magnon phase-matching condition is fulfilled. For µ0H = 6.5 T, the phonon and magnon dispersions cross at two points, which are close to the frequencies of the 3rd and the 5th magnon modes, and are almost parallel over a larger frequency range. The multitude of higher order magnons is excited in this case. The mechanisms of magnon excitation and the 13 acoustic pulse width can be understood, if one looks at the time dependencies and the Fourier spectra in Fig. 3. At µ0H = 3.0 T, the magnetization dynamics starts before at the delay time around 2 ps, i.e. well before the arrival time of an acoustic pulse at the back interface at 5.4 ps, marked by a black vertical line in all time dependencies in Fig. 3. This initial part of magnetization dynamics oscillates at the characteristic time scale of approximately 1...2 ps. The corresponding Fourier spectrum on the right hand side (note the difference in the vertical scaling as compared to Fig. 2) indeed contains several spectrally overlapping high frequency magnons peaking around the 7th magnon mode at f7 = 573 GHz. This frequency range corresponds to the the second crossing point between phonon and magnon dispersions in Fig. 1(b) at small magnetic fields. These high frequency magnons are characterized by a higher group velocity. Therefore they arrive at the back interface earlier than the acoustic pulse that has generated them. This interpretation is consistent with the results for higher magnetic fields. Notably at µ0H = 6.5 T the entire magnetization dynamics can be seen as a chirped magnon pulse consisting of the high frequency components arriving at short delay times and low frequency components arriving later. The role of the acoustic pulse duration becomes evident from an obvious visual correlation of the excited magnon spectra and the Fourier spectra of ultrashort acoustic pulses marked as colored shaded area on the right hand side of Fig. 3: the shorter the pulses the broader their spectrum, the more efficiently they excite high frequency magnons. As discussed earlier, this conclusion is in line with the results of our analytical considerations suggesting that the external driving force for magnons is proportional to the Fourier component of an acoustic pulse at the magnon frequency. At this point it is worth mentioning that the dependence of FMR amplitude on the acoustic pulse duration is opposite: longer acoustic pulses possess a larger pulse area [41] resulting in a larger amplitude of an acoustically generated FMR precession. This behavior is evident in the Fourier spectra of Fig. 2, both for the FMR and low-frequency exchange magnons. V. SUMMARY AND CONCLUSIONS The main result of this paper is that ultrashort acoustic pulses propagating through thin ferromagnetic samples must excite not only the FMR precession but also high-order exchange magnons falling within the spectral bandwidth of acoustic pulses. Although the efficiency of magnon excitation is naturally enhanced when phonon-magnon phase-matching conditions 14 are fulfilled, our simulations show that exchange magnons with measurable amplitudes get excited even when phonon and magnon dispersion do not cross. It is rather the acoustic band- width than phase matching that determines the excitation efficiency of exchange magnons. The optimum conditions for ultrafast magnetoelastic generation of exchange magnons can be elucidated from systematic numerical simulations as a function of multiple physical pa- rameters such as the exchange stiffness, Gilbert damping, sample thickness, acoustic pulse duration etc. Such analysis could be quite helpful in a view of the experiments evidencing an ultrafast optical excitation of the exchange magnons in ferromagnetic thin films [3, 37], where possible contributions of magnetoelastic excitation of exchange magnons are masked by the dominant mechanism of ultrafast demagnetization [1]. Applications of our simple theory to the experimental investigations on thick ferromagnetic films, characterized by a quasi-continuous magnon spectrum and suggesting the physical interpretation in terms of propagating magnon pulses, can be envisaged [21, 24]. ACKNOWLEDGEMENTS Funding through Agence Nationale de la Recherche under grant "PPMI-NANO" (ANR- 15-CE24-0032 and DFG SE2443/2), Strat´egie internationale NNN-Telecom de la R´egion Pays de La Loire, Alexander von Humboldt Stiftung, PRC CNRS-RFBR "Acousto-magneto- plasmonics" (grant number 17-57-150001) and Act 211 Government of the Russian Federation (contract 02.A03.21.0011) is greatfully acknowledged. REFERENCES [1] E. Beaurepaire, J.-C. Merle, A. Daunois, and J.-Y. Bigot, Phys. Rev. Lett. 76, 4250 (1996), URL https://link.aps.org/doi/10.1103/PhysRevLett.76.4250. [2] B. Koopmans, M. Van Kampen, J. Kohlhepp, and W. De Jonge, Phys. Rev. Lett. 85, 844 (2000), URL https://link.aps.org/doi/10.1103/PhysRevLett.85.844. 15 [3] M. Van Kampen, C. Jozsa, J. Kohlhepp, P. LeClair, L. Lagae, W. De Jonge, and B. Koopmans, Phys. Rev. Lett. 88, 227201 (2002), URL https://link.aps.org/doi/10.1103/PhysRevLett.88.227201. [4] L. Guidoni, E. Beaurepaire, and J.-Y. Bigot, Phys. Rev. Lett. 89, 017401 (2002), URL https://link.aps.org/doi/10.1103/PhysRevLett.89.017401. [5] Q. Zhang, A. V. Nurmikko, A. Anguelouch, G. Xiao, and A. Gupta, Phys. Rev. Lett. 89, 177402 (2002), URL https://link.aps.org/doi/10.1103/PhysRevLett.89.177402. [6] M. Vomir, L. Andrade, L. Guidoni, E. Beaurepaire, and J.-Y. Bigot, Phys. Rev. Lett. 94, 237601 (2005), URL https://link.aps.org/doi/10.1103/PhysRevLett.94.237601. [7] J.-Y. Bigot, M. Vomir, L. Andrade, and E. Beaurepaire, Chemical physics 318, 137 (2005), URL http://www.sciencedirect.com/science/article/pii/S030101040500248X. [8] A. Kimel, A. Kirilyuk, P. Usachev, R. Pisarev, et al., Nature 435, 655 (2005), URL https://doi.org/10.1038/nature03564. [9] G. Malinowski, F. Dalla Longa, J. Rietjens, P. Paluskar, R. Huijink, H. Swagten, and B. Koop- mans, Nature Physics 4, 855 (2008), URL https://doi.org/10.1038/nphys1092. [10] J.-Y. Bigot, M. Vomir, and E. Beaurepaire, Nature Physics 5, 515 (2009), URL https://doi.org/10.1038/nphys1285. [11] U. Bovensiepen, Nature Physics 5, 461 (2009), URL https://doi.org/10.1038/nphys1322. [12] I. Radu, G. Woltersdorf, M. Kiessling, A. Melnikov, U. Bovensiepen, J.- U. Thiele, and C. H. Back, Phys. Rev. Lett. 102, 117201 (2009), URL https://link.aps.org/doi/10.1103/PhysRevLett.102.117201. [13] E. Carpene, E. Mancini, D. Dazzi, C. Dallera, E. Puppin, and S. De Silvestri, Phys. Rev. B 81, 060415 (2010), URL https://link.aps.org/doi/10.1103/PhysRevB.81.060415. [14] C. Boeglin, E. Beaurepaire, V. Halt´e, V. L´opez-Flores, C. Stamm, N. Pontius, H. Durr, and J. Bigot, Nature 465, 458 (2010), URL https://doi.org/10.1038/nature09070. [15] A. V. Scherbakov, A. S. Salasyuk, A. V. Akimov, X. Liu, M. Bombeck, C. Bruggemann, D. R. Yakovlev, V. F. Sapega, J. K. Furdyna, and M. Bayer, Phys. Rev. Lett. 105, 117204 (2010), URL https://link.aps.org/doi/10.1103/PhysRevLett.105.117204. [16] I. Radu, K. Vahaplar, C. Stamm, T. Kachel, N. Pontius, H. Durr, T. Ostler, J. Barker, R. Evans, R. Chantrell, et al., Nature 472, 205 (2011), URL https://doi.org/10.1038/nature09901. 16 [17] D. Rudolf, L.-O. Chan, M. Battiato, R. Adam, J. M. Shaw, E. Turgut, P. Maldonado, S. Mathias, P. Grychtol, H. T. Nembach, et al., Nature communications 3, 1037 (2012), URL https://doi.org/10.1038/ncomms2029. [18] M. Bombeck, J. V. Jager, A. V. Scherbakov, T. Linnik, D. R. Yakovlev, X. Liu, J. K. Furdyna, A. V. Akimov, and M. Bayer, Phys. Rev. B 87, 060302 (2013), URL https://link.aps.org/doi/10.1103/PhysRevB.87.060302. [19] J.-W. Kim, M. Vomir, and J.-Y. Bigot, Phys. Rev. Lett. 109, 166601 (2012), URL https://link.aps.org/doi/10.1103/PhysRevLett.109.166601. [20] J.-W. Kim, M. Vomir, and J.-Y. Bigot, Scientific reports 5, 8511 (2015), URL https://doi.org/10.1038/srep08511. [21] J.-W. Kim and J.-Y. Bigot, Phys. Rev. B 95, 144422 (2017), URL https://link.aps.org/doi/10.1103/PhysRevB.95.144422. [22] A. Kirilyuk, A. V. Kimel, and T. Rasing, Rev. Mod. Phys. 82, 2731 (2010), URL https://link.aps.org/doi/10.1103/RevModPhys.82.2731. [23] L. Thevenard, E. Peronne, C. Gourdon, C. Testelin, M. Cubukcu, E. Charron, S. Vincent, A. Lemaıtre, and B. Perrin, Phys. Rev. B 82, 104422 (2010), URL https://link.aps.org/doi/10.1103/PhysRevB.82.104422. [24] M. Bombeck, A. Salasyuk, B. Glavin, A. Scherbakov, C. Bruggemann, D. Yakovlev, V. Sapega, X. Liu, J. Furdyna, A. Akimov, et al., Phys. Rev. B 85, 195324 (2012), URL https://link.aps.org/doi/10.1103/PhysRevB.85.195324. [25] V. V. Temnov, Nature Photonics 6, 728 (2012), URL https://doi.org/10.1038/nphoton.2012.220. [26] J. Janusonis, C.-L. Chang, T. Jansma, A. Gatilova, V. Vlasov, A. Lomonosov, V. Temnov, and R. Tobey, Phys. Rev. B 94, 024415 (2016), URL https://link.aps.org/doi/10.1103/PhysRevB.94.024415. [27] C. Chang, A. Lomonosov, J. Janusonis, V. Vlasov, V. Temnov, and R. Tobey, Phys. Rev. B 95, 060409 (2017), URL https://link.aps.org/doi/10.1103/PhysRevB.95.060409. [28] P. Nemec, E. Rozkotov´a, N. Tesarov´a, F. Troj´anek, E. De Ranieri, K. Olejn´ık, J. Ze- men, V. Nov´ak, M. Cukr, P. Mal´y, et al., Nature physics 8, 411 (2012), URL https://doi.org/10.1038/nphys2279. [29] A. J. Schellekens, K. C. Kuiper, R. de Wit, and B. Koopmans, Nature Communications 5, 4333 (2014), URL https://doi.org/10.1038/ncomms5333. 17 [30] I. Razdolski, A. Alekhin, N. Ilin, J. P. Meyburg, V. Roddatis, D. Diesing, U. Bovensiepen, and A. Melnikov, Nature communications 8, 15007 (2017), URL https://doi.org/10.1038/ncomms15007. [31] A. Alekhin, I. Razdolski, N. Ilin, J. P. Meyburg, D. Diesing, V. Roddatis, I. Rungger, M. Sta- menova, S. Sanvito, U. Bovensiepen, et al., Phys. Rev. Lett. 119, 017202 (2017), URL https://link.aps.org/doi/10.1103/PhysRevLett.119.017202. [32] A. Alekhin, I. Razdolski, M. Berritta, D. Burstel, V. V. Temnov, D. Diesing, U. Bovensiepen, G. Woltersdorf, P. M. Oppeneer, and A. Melnikov, Journal of Physics: Condensed Matter 31, 124002 (2019), URL https://doi.org/10.1088/1361-648X/aafd06. [33] C. Thomsen, H. T. Grahn, H. J. Maris, and J. Tauc, Phys. Rev. B 34, 4129 (1986), URL https://link.aps.org/doi/10.1103/PhysRevB.34.4129. [34] V. V. Temnov, C. Klieber, K. A. Nelson, T. Thomay, V. Knittel, A. Leitenstorfer, D. Makarov, M. Albrecht, and R. Bratschitsch, Nature Communications 4, 1468 (2013), URL https://doi.org/10.1038/ncomms2480. [35] V. V. Temnov, I. Razdolski, T. Pezeril, D. Makarov, D. Seletskiy, A. Mel- nikov, and K. A. Nelson, Journal of Optics 18, 093002 (2016), URL https://iopscience.iop.org/article/10.1088/1361-648X/aafd06. [36] M. Farle, Reports on Progress in Physics 61, 755 (1998), URL https://iopscience.iop.org/article/10.1088/0034-4885/61/7/001/pdf. [37] R. Salikhov, A. Alekhin, T. Parpiiev, T. Pezeril, D. Makarov, R. Abrudan, R. Meck- enstock, F. Radu, M. Farle, H. Zabel, et al., Phys. Rev. B 99, 104412 (2019), URL https://link.aps.org/doi/10.1103/PhysRevB.99.104412. [38] M. Getzlaff, Fundamentals of Magnetism (Springer, 2008). [39] K. J. Manke, A. A. Maznev, C. Klieber, V. Shalagatskyi, V. V. Temnov, D. Makarov, S.- H. Baek, C.-B. Eom, and K. A. Nelson, Applied Physics Letters 103, 173104 (2013), URL https://doi.org/10.1063/1.4826210. [40] P. J. S. van Capel, E. P´eronne, and J. Dijkhuis, Ultrasonics 56, 36 (2015), URL http://www.sciencedirect.com/science/article/pii/S0041624X14002868. [41] O. Kovalenko, T. Pezeril, and V. V. Temnov, Phys. Rev. Lett. 110, 266602 (2013), URL https://link.aps.org/doi/10.1103/PhysRevLett.110.266602. 18 

1701.09038

1

1701

Atomic-scale imaging of few-layer black phosphorus and its reconstructed edge

Black phosphorus (BP) has recently emerged as an alternative 2D semiconductor owing to its fascinating electronic properties such as tunable bandgap and high charge carrier mobility. The structural investigation of few-layer BP, such as identification of layer thickness and atomic-scale edge structure, is of great importance to fully understand its electronic and optical properties. Here we report atomic-scale analysis of few-layered BP performed by aberration corrected transmission electron microscopy (TEM). We establish the layer-number-dependent atomic resolution imaging of few-layer BP via TEM imaging and image simulations. The structural modification induced by the electron beam leads to revelation of crystalline edge and formation of BP nanoribbons. Atomic resolution imaging of BP clearly shows the reconstructed zigzag (ZZ) edge structures, which is also corroborated by van der Waals first principles calculations on the edge stability. Our study on the precise identification of BP thickness and atomic-resolution imaging of edge structures will lay the groundwork for investigation of few-layer BP, especially BP in nanostructured forms.

cond-mat.mes-hall

cond-mat

Atomic-scale imaging of few-layer black phosphorus and its reconstructed edge Yangjin Lee1, Jun-Yeong Yoon1, Declan Scullion3, Jeongsu Jang1, Elton J. G. Santos3,4, Hu Young Jeong2,*, and Kwanpyo Kim1,* 1Department of Physics, Ulsan National Institute of Science and Technology (UNIST), Ulsan 689-798, South Korea. 2UNIST Central Research Facilities (UCRF), Ulsan National Institute of Science and Technology (UNIST), Ulsan 689-798, South Korea. 3School of Mathematics and Physics, Queen's University Belfast, Belfast, BT95AG, United Kingdom. 4School of Chemistry and Chemical Engineering, Queen's University Belfast, Belfast, BT95AL, United Kingdom. *Address correspondence to K.K. ([email protected]) or H.Y.J. ([email protected]) Keywords: Black phosphorus, Phosphorene, Nanoribbon, Edge structure, Transmission electron microscopy Abstract Black phosphorus (BP) has recently emerged as an alternative two-dimensional semiconductor owing to its fascinating electronic properties such as tunable bandgap and high charge carrier mobility. The structural investigation of few-layer black phosphorus, such as identification of layer thickness and atomic-scale edge structure, is of great importance to fully understand its electronic and optical properties. Here we report atomic-scale analysis of few-layered BP performed by aberration corrected transmission electron microscopy (TEM). We establish the layer-number-dependent atomic resolution imaging of few-layer BP via TEM imaging and image simulations. The structural modification induced by the electron beam leads to revelation of crystalline edge and formation of BP nanoribbons. Atomic resolution imaging of BP clearly shows the reconstructed zigzag edge structures, which is also corroborated by van der Waals first principles calculations on the edge stability. Our study on the 1 precise identification of BP thickness and atomic-resolution imaging of edge structures will lay the groundwork for investigation of few-layer BP, especially BP in nanostructured forms. 1. Introduction A century after its discovery[1-3], black phosphorus (BP) has regained much attention as an alternative two-dimensional (2D) material owing to its promising electrical, optical, and chemical properties[4-19]. As a layered structure, BP has the largely tunable bandgap as a function of the number of layers (0.35 – 2.0 eV), which can bridge the missing bandgap range from the currently available various 2D materials[9, 15, 16, 20]. BP also poses various interesting electrical, mechanical, and optical properties, such as large tunability by strain[20-22] and high in-plane anisotropy[11, 16, 23, 24]. Especially, researchers have recently demonstrated the high charge carrier mobility from few- layer BP[5, 6, 8, 15, 16] opening up various interesting electronic applications[5, 25, 26] and fundamental studies[10, 11]. Atomic-scale structural analysis of few-layer BP is essential to fully understand its electronics and optical properties. The various defects[27, 28] have a profound effect on charge carrier dynamics, which becomes more important for the few-layer form of BP. BP nanoribbons also have various interesting properties, including edge-type-dependent electronic properties and special edge states, as shown by recent theoretical studies[22, 29, 30]. Until now, only a few experimental results on the structural characterization of BP using various microscopy techniques have been reported[9, 31-33]. Although these reports provide general structural analysis on BP, atomic-scale imaging and analysis of structural modification of BP are mainly unexplored at this point. Here we report atomic-scale analysis of few-layered BP performed by aberration corrected transmission electron microscopy (TEM). Previously, aberration-corrected TEM imaging has been applied to various 2D materials including graphene[34-36], hexagonal boron nitride (h-BN)[37-39], and other transition metal dichalcogenides[40-42]. We establish the layer-number-dependent atomic 2 resolution imaging of few-layer BP via TEM experiments and image simulations. In addition, we find that the electron beam can be utilized to form BP nanoribbons with crystalline edge structures. TEM imaging reveals that the BP edge shows the reconstructed edge configuration, which is also confirmed by first principles calculations with van der Waals dispersion force method. Our study on the precise identification of BP thickness and atomic resolution imaging of BP edges will lay the groundwork for investigation of electrical and optical properties of BP nanostructures. 2. Methods 2.1. Black phosphorus sample preparation and thickness characterizations Black phosphorus (BP) crystals were purchased from Smart Elements (purity, ~99.998%). We use two different methods to prepare TEM samples. For determination of the flake thickness via independent characterization tools such as atomic force microscopy, we rely on sample preparation method 1 (See Supplementary Figure S1). We mechanically exfoliate BP crystals onto thin poly-methyl-methacrylate (PMMA)-coated SiO2(300nm)/Si substrate. PMMA was spin-coated for one minute with 6000 revolutions per minute. The thin flakes were firstly identified by optical microscopy under the reflection mode. Consequently, the thickness of thin BP flakes was measured using an atomic force microscope (AFM) (DI-3100, Veeco, USA). After measurement of BP thickness (sometimes we skip the AFM imaging to avoid the possible degradation to the samples), TEM samples were fabricated using the direct transfer method[43] and PMMA layer was removed by acetone. Finally, the transmission mode in an optical microscope was used to determine the number of layers[9]. Except for AFM analysis, all processes (optical microscopy, BP exfoliation, TEM grid fabrication, and remove the polymer layer) were performed in the N2 filled glove box. Another sample preparation method (sample preparation method 2) was mainly used for atomic-resolution TEM imaging. For this purpose, BP crystals were exfoliated onto SiO2/Si wafers using conventional mechanical exfoliation method. Exfoliated BP flakes were transferred to Quantifoil Au TEM grids using direct transfer method[43]. Gentle plasma cleaning with H2 and O2 gas environment was sometimes carried out using plasma cleaner (Advanced plasma system, Gatan, USA) for 5 minutes with 10 W input power to thin BP flakes and remove surface residues. To minimize the oxidation of BP specimens, BP samples were immediately loaded into TEM chamber after sample preparation process. 2.2. TEM characterizations and image simulations 3 The atomic resolution imaging of BP was performed with a FEI Titan G2 operated at 80 kV, which is equipped with image Cs aberration corrector and monochromator. For the TEM time series, the exposure time of 0.2 seconds together with the processing time of 1.3 seconds were used. This results in the image frame time of 1.5 second per image. All the TEM image simulations were performed using MacTempas software. The crystal structure of BP with a=3.31 Å, b=10.48 Å and c=4.37Å (space group Cmca[44]) were used. The b-axis is the layer stacking direction. The normal image calculations with simulation parameters (convergence angle = 0.10 mrad, Cs = - 12 μm, mechanical vibration = 0.5 Å) in MacTempas were used. The image simulations with relevant B2 coefficient of 220nm were also performed for validity check-up. 2.3. Theoretical calculations: Van der Waals Ab Initio Calculations for the (PAW)[51, 52] latter, and norm-conserving The calculations reported here are based on ab initio density functional theory using the SIESTA method[45] and the VASP code[46, 47]. The generalized gradient approximation[48] along with the optB88-vdW[49, 50] functional was used in both methods, together with a double-ζ polarized basis set in Siesta, and a well-converged plane-wave cutoff of 500 eV in VASP. Projected augmented wave method (NC) Troullier–Martins pseudopotentials[53] for the former, have been used in the description of the bonding environment for P. NC pseudopotentials include non-linear-core corrections (NLCC)[54] to correctly account for the weak interactions between core and valence densities. The pseudocore radii rNLCC(ao) (in Bohrs) together with the different l channels rl(ao) have been optimized and the values are: rs(ao)=1.83, rp(ao)=1.83, rd(ao)=1.83, rf(ao)=1.83, rNLCC(ao)=1.45. The shape of the NAOs was automatically determined by the algorithms described in[45]. The cutoff radii of the different orbitals were obtained using an energy shift of 50 meV, which proved to be sufficiently accurate to describe the geometries and the energetics. Atomic coordinates were allowed to relax until the forces on the ions were less than 0.04 eV/Å under the conjugate gradient algorithm. Further relaxations (0.01 eV/ Å) do not change appreciably the energetics and geometries. The lattice constants for the monolayer unit cell were optimized and found to be a=3.297 Å, b=22.1220 Å, c=4.655 Å in SIESTA which is in good agreement with the results obtained using VASP, a=3.295 Å, b=22.1219 Å, c=4.535 Å. To model the system studied in the experiments, we created large supercells containing up to 136 atoms to simulate the interface between different nanoribbon layers and edge reconstructions in the phosphorene. To avoid any interactions between supercells in the non-periodic direction, a 20 Å vacuum space was used in all calculations. In addition to this, a cutoff energy of 120 Ry was used to resolve the real-space grid 4 used to calculate the Hartree and exchange-correlation contribution to the total energy. For the phosphorene sheets, the Brillouin zone was sampled with a 10x8x1 grid under the Monkhorst-Pack scheme[55], which gives similar results as those using a finer 17x15x1 k-sampling. In addition to this we used a Fermi-Dirac distribution with an electronic temperature of kBT = 20 meV. 3. Results and discussion We first study the crystal structure of BP using relatively thick (20nm or thicker) flakes. Supplementary Figure S2(a) shows a low-magnification TEM image of a typical BP flake. Without tilting samples, BP usually exhibits the crystal direction viewed along zone axis [010] as shown in Figure S2(b) and S2(d). The Fourier transform of the image clearly shows the diffraction signal with lattice parameters which are consistent with previous results[2] (Figure S2(e) and Table S1). Some transferred BP flakes display folded edge structures and this allows us to observe BP at different crystallographic directions, even without tilting of samples. Supplementary Figure S2(h) shows a TEM image around the flake edge where the crystal structure at zone axis [100] is clearly observed. At [100] zone axis, the puckered layered structure of BP can be clearly observed. The interlayer distance of BP is found to be 5.27 Å, which is consistent with previously reported results[2] (Table S1). With tilting of the specimens, atomic resolution imaging at extra zone axes is also performed as shown in Supplementary Figure S3. Especially, atomic resolution TEM images from [101] zone axis can differentiate AB and AA stacking and our observation indicates that bulk BP has mainly AB stacking (Supplementary Figure S3(f)-(j)). Our first-principles total energy calculations for different stacking also confirm that AB stacking is the most stable stacking configuration (Supplementary Figure S4). The precise and facile identification of BP thickness is a prerequisite for investigation of various properties. We combine optical microscope (OM) imaging, atomic force microscopy (AFM), TEM imaging, and selected area electron diffraction (SAED) to precisely determine the thickness of thin flakes. The exfoliated and pre-identified BP flakes on PMMA/SiO2(300nm)/Si substrate are 5 characterized by AFM and then transferred to a Quantifoil holey TEM grid as shown in Figure 1(a)- 1(c). (See Methods sections for details.) The suspended BP flakes reduce the transmittance at visible light range, which can be easily checked by transmission-mode of optical microscopy. The intensity along the dashed line in Figure 1(c) shows that the flake reduces the transmittance by about 17% and 23% compared to the region without a flake (Figure 1(d)). This shows the possibility of using the optical transmittance as a tool to determine the thickness of flakes, which will be discussed later in detail. The pre-characterized flakes are then investigated by SAED and atomic resolution TEM imaging for precise confirmation of sample thickness. Figure 1(e) shows a bright-field TEM image of the pre-identified flake. The SAED pattern is consistent with the expected crystal direction along zone axis [010] as shown in the inset of Figure 1(e). The intensity ratio (I101/I200) of diffraction patterns can be utilized for thickness determination (Figure 1(f)) similar to previous reports on 2D materials [9, 56]. By comparison with diffraction intensity simulation results (Supplementary Table S2), we confirm that the thinner area of flake has 5 layers, which is consistent with AFM result shown in Figure 1(b). In the pre-characterized region of 5-layer thickness, atomic resolution TEM imaging and comparison with image simulation are performed. Especially, the comparison of the intensity modulation and its pattern along c-axis of BP is performed between observed TEM images and simulation results (Figure 1(g) and 1(h)). Since the phase contrast TEM image depends on the number of BP layers and defocus value, we perform a series of image simulations along [010] zone axis as a function of the layer number as well as the defocus value, where we assume the AB stacking (Supplementary Figure S5). We find that the experimentally-observed and simulated TEM image intensity pattern matches very well whereas the absolute magnitude of modulation differs by a factor of two. This factor of two discrepancy is often called as the Stobb's factor and was observed in other two dimensional materials such as graphene[34, 57, 58]. Recent studies have demonstrated that the discrepancy can be caused by the neglect of the detector modulation-transfer function (MTF) during the image simulation [59, 60]. Due to the limitation in the simulation software, we did not consider the 6 effect of MFT for our image simulation, which can be the main reason for the observed discrepancy. After taking into account this factor, the simulation intensity pattern with defocus value of 6 nm match well with 5 layers. We confirm that the analysis is valid under imaging conditions with some residual aberrations (second-order coma B2) relevant to our experiments (Supplementary Note S1, Figure S6 and Table S2). We find that the optical transmittance of flakes is one of reliable and facile parameters for determination of flake thickness as confirmed by aforementioned complimentary characterizations as shown in Figure 2. This is of great importance since this allow us to prepare TEM samples of known thickness without AFM imaging (and resulted degradation by exposure to ambient conditions) as all the sample preparation and optical characterization process can be performed inside N2-filled glove- box. In the range of 3~9 layer thickness of BP flakes we find that the optical transmittance reduces 3.3% per layer (Figure 2), which is also confirmed by SAED analysis of flakes (Supplementary Figure S7, S8, and Table S3). This is consistent with a recent report[9]. With the established procedure of atomic resolution imaging/simulation, we investigate a BP specimen where the thickness is not homogeneous in the different locations due to electron-beam- induced sputtering (Figure 3(a)). We assign the layer numbers and defocus values for the observed TEM images by comparison with simulated images through the intensity modulation and its pattern along c-axis of BP. Generally, the magnitude of intensity modulation increase as the thickness increases for a given focus value (Supplementary Figure S9 and Table S4). We can reproduce the experimentally- observed intensity patterns by simulation results with high accuracy as shown in Figure 3(b) and 3(c). The TEM images obtained from various locations of the specimen show good agreements with simulation results after the Stobb's factor correction[34, 57, 58] as shown in Figure 3(c). Through this procedure, we determine that the BP specimen imaged in Figure 3(a) has a thickness ranging from three to seven layers. This result confirms that the atomic resolution imaging alone can be utilized to reliably determine the BP thickness. One thing to note is that even and odd layer numbers produce 7 distinct image patterns (Supplementary Figure S5). Simulations with even number of layers (for example, double layer) show the intensity modulation with the half of usual bulk lattice parameters due to the symmetric AB stacking. On the other hand, with an odd number of layers (monolayer and triple layer), the simulated images display the intensity modulation with the periodicity of usual lattice parameters of bulk BP (Supplementary Figure S5). Now we start our discussion on the edge structure of BP. The investigation of edge structure of BP is an important topic as it significantly influences various physical properties of BP, especially for BP in nanostructured forms. There are a few recent studies on BP edges and nanoribbons mainly by theoretical calculations[29, 30]. Previously, TEM imaging has been adapted to study edge structures of various other 2D materials including graphene and hexagonal boron nitride (h-BN), which has revealed edge-specific bonding and reconstruction[35, 61-64]. On the other hand, there is only a limited number of experimental studies on BP edges by any imaging technique[31]. As shown in Supplementary Figure S10, as-prepared BP specimens exhibit amorphous edge structure, where the amorphous edge regions of several nanometers are always observed. This amorphous edge is possibly due to residual oxide layer. The plasma-treatment of a specimen can be used to reduce the amorphous edge region of the flake but this process still leaves some amorphous region (Supplementary Figure S10). The edge structure of BP crystals can be structurally modified by e-beam irradiation during TEM imaging. Previously, the electron-beam-induced sputtering has been used for thinning down the specimen as well as cleaning[35, 37]. Remarkably, the crystalline edge structure can be obtained via this method. Figure 4(a) and 4(b) shows the changes of sample over 48-second e-beam exposure. The atoms at the amorphous edge (indicated by yellow arrows in Figure 4(a)) can be preferentially sputtered out, exposing the crystalline edge structure. The crystal direction of the exposed edge shows zigzag (ZZ) edge direction. Figure 4(c) is the zoom-in image of BP edge where the periodic edge structure over five unit-cells is clearly observed. Moreover, the image pattern at the edge shows a 8 higher intensity modulation compared to the basal plane. This strongly suggests that there is a reconstructed edge formation. To have a better understanding on atomic-scale structure of ZZ edge, we calculate the relaxed edge structures with various possibilities using first principles calculations with van der Waals interactions (See Methods sections). As shown in Figure 5, we find that reconstructed ZZ edges (type 1 and type 2) exhibit similar edge formation energies compared to zigzag (ZZ) or armchair (AC) edge configurations. This result is quite distinct from the graphene edge case, where the edge formation energy strongly depends on the edge type[63, 65, 66]. To compare with experimental TEM images, a series of image simulations assuming different ZZ edge structures including usual ZZ edge termination (Supplementary Figure S11) and two types of reconstructed edge configurations are undertaken (Supplementary Figure S12 and S13). Since the observed area has three-layer thickness, which is determined by the previous image pattern and intensity modulation analysis (Figure 3(b)), we focus on the simulated images from triple-layer. By comparison, we find that the observed ZZ edge is consistent with the reconstructed zigzag edge type-1 (RZZ1) edge (Figure 4(d)). The usual ZZ edge without reconstruction (Figure 4(g) and 4(h)) and RZZ2 edge (figure 4(i) and 4(j)) are not consistent with the observed image. We note that the observed RZZ1 edge was theoretically studied together with some experimental evidence but the direct atomic resolution edge imaging was not previously performed[31]. Finally, we discuss the sample thinning and BP nanoribbon formation induced by electron beam irradiation. Figures 6(a)-(e) show a time series of structural modification of BP under electron beam. The same series of images are overlaid with different colors in Figures 6(f)-(j) for easy identification of structural changes. Different colors indicate triple-layer region (blue), thicker area (pink) and amorphous regions (yellow). The sample thinning with electron-beam sputtering is observed from Figure 6(f) and 6(g); the region overlaid with pink color (thicker area) is gradually replaced by the blue region (three layers). 9 Electron-beam is one of useful ways to manipulate the materials at nanoscale and we demonstrate that BP nanoribbons can be formed by prolonged e-beam exposure. Figure 6(d) and 6(i) clearly show that the formation of approximately 4nm-wide BP nanoribbons with crystalline edge. After prolonged e-beam irradiation, the BP nanoribbon is amorphized to form amorphous BP nano- constriction with a less than 2 nm neck width. Consequently, the nano-constriction breaks down (See Supplementary Movie S1). The electron-beam-induced structural modification for BP seems more pronounced compared to graphene and this may be related to low energy barriers for structural phase transformation of BP[67]. The identifications of low-energy defect structures and sputtering mechanisms, such as knock-on damage and chemical etching, are important experimental issues during TEM imaging. We are currently performing the calculations of various low-energy defect structures as well as knock-on damage threshold for phosphorene. We note that a nanoribbon formation on relatively thick samples was recently reported[68]. 4. Conclusion In conclusion, the atomic-scale structure of few-layered BP and its reconstructed ZZ edges were investigated by Cs-corrected TEM and imaging simulation. The precise and facile characterization methods of BP thickness demonstrated in our study will lead to various fundamental studies, such as measurements of layer-number-dependent electrical and optical properties. We also demonstrate that electron beam irradiation can be used to form BP nanoribbons as well as to expose crystalline reconstructed ZZ edge for the first time. Further TEM analysis on BP is expected to shed light on various defect structures and structural degradation mechanisms. Acknowledgements This work is supported by Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Education (NRF-2014R1A1A2058178). H.Y.J. was supported by Creative Materials Discovery Program through the National Research Foundation 10 of Korea (NRF) funded by the Ministry of Science, ICT and Future Planning (NRF- 2016M3D1A1900035). D.S. thanks the studentship from the EPSRC-DTP award. E.J.G.S. acknowledges the use of computational resources from the UK national high performance computing service, ARCHER, for which access was obtained via the UKCP consortium and funded by EPSRC grant ref EP/K013564/1; and the Extreme Science and Engineering Discovery Environment (XSEDE), supported by NSF grants number TG-DMR120049 and TG-DMR150017. The Queen's Fellow Award through the startup grant number M8407MPH is also acknowledged. 11 References [1] [2] [3] [4] 523 Bridgman P W 1914 J. Am. Chem. Soc. 36 1344 Hultgren R, Gingrich N S and Warren B E 1935 J. Chem. Phys. 3 351 Keyes R W 1953 Phys. Rev. 92 580 Ling X, Wang H, Huang S, Xia F and Dresselhaus M S 2015 Proc. Natl. Acad. Sci. U. S. A. 112 4 [5] l. 9 372 Li L, Yu Y, Ye G J, Ge Q, Ou X, Wu H, Feng D, Chen X H and Zhang Y 2014 Nat. Nanotechno [6] [7] Liu H, Neal A T, Zhu Z, Luo Z, Xu X, Tománek D and Ye P D 2014 ACS Nano 8 4033 Xia F, Wang H and Jia Y 2014 Nat. Commun. 5 4458 [8] 4 103106 Koenig S P, Doganov R A, Schmidt H, Castro Neto A H and Özyilmaz B 2014 Appl. Phys. Lett. 10 [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] 80 [19] 96 [20] [21] [22] [23] [24] Castellanos-Gomez A et al. 2014 2D Mater. 1 025001 Li L et al. 2015 Nat. Nanotechnol. 10 608 Wang X et al. 2015 Nat. Nanotechnol. 10 517 Youngblood N, Chen C, Koester S J and Li M 2015 Nat. Photon. 9 247 Yuan H et al. 2015 Nat. Nanotechnol. 10 707 Kim J et al. 2015 Science 349 723 Das S, Zhang W, Demarteau M, Hoffmann A, Dubey M and Roelofs A 2014 Nano Lett. 14 5733 Qiao J, Kong X, Hu Z X, Yang F and Ji W 2014 Nat. Commun. 5 4475 Park C M and Sohn H J 2007 Adv. Mater. 19 2465 Sun J, Lee H-W, Pasta M, Yuan H, Zheng G, Sun Y, Li Y and Cui Y 2015 Nat. Nanotechnol. 10 9 Kang J, Wood J D, Wells S A, Lee J-H, Liu X, Chen K-S and Hersam M C 2015 ACS Nano 9 35 Rodin A S, Carvalho A and Castro Neto A H 2014 Phys. Rev. Lett. 112 176801 Fei R and Yang L 2014 Nano Lett. 14 2884 Han X, Morgan Stewart H, Shevlin S A, Catlow C R A and Guo Z X 2014 Nano Lett. 14 4607 Fei R, Faghaninia A, Soklaski R, Yan J-A, Lo C and Yang L 2014 Nano Lett. 14 6393 Jiang J-W and Park H S 2014 Nat. Commun. 5 4727 [25] Lett. 14 6424 Wang H, Wang X, Xia F, Wang L, Jiang H, Xia Q, Chin M L, Dubey M and Han S-j 2014 Nano [26] Zhu W, Yogeesh M N, Yang S, Aldave S H, Kim J S, Sonde S, Tao L, Lu N and Akinwande D 2 12 015 Nano Lett. 15 1883 [27] 46801 [28] [29] [30] [31] [32] [33] [34] 2 [35] [36] 7 209 [37] [38] [39] Ziletti A, Carvalho A, Campbell D K, Coker D F and Castro Neto A H 2015 Phys. Rev. Lett. 114 0 Wang V, Kawazoe Y and Geng W T 2015 Phys. Rev. B 91 045433 Tran V and Yang L 2014 Phys. Rev. B 89 245407 Guo H Y, Lu N, Dai J, Wu X J and Zeng X C 2014 J. Phys. Chem. C 118 14051 Liang L, Wang J, Lin W, Sumpter B G, Meunier V and Pan M 2014 Nano Lett. 14 6400 Liu X L, Wood J D, Chen K S, Cho E and Hersam M C 2015 J. Phys. Chem. Lett. 6 773 Favron A et al. 2015 Nat. Mater. 14 826 Meyer J C, Kisielowski C, Erni R, Rossell M D, Crommie M F and Zettl A 2008 Nano Lett. 8 358 Girit Ç Ö et al. 2009 Science 323 1705 Warner J H, Margine E R, Mukai M, Robertson A W, Giustino F and Kirkland A I 2012 Science 33 Alem N, Erni R, Kisielowski C, Rossell M D, Gannett W and Zettl A 2009 Phys. Rev. B 80 Meyer J C, Chuvilin A, Algara-Siller G, Biskupek J and Kaiser U 2009 Nano Lett. 9 2683 Jin C, Lin F, Suenaga K and Iijima S 2009 Phys. Rev. Lett. 102 195505 [40] Lett. 109 035503 Komsa H-P, Kotakoski J, Kurasch S, Lehtinen O, Kaiser U and Krasheninnikov A V 2012 Phys. Rev. [41] [42] [43] [44] Zhou W et al. 2013 Nano Lett. 13 2615 Azizi A et al. 2014 Nat. Commun. 5 4867 Meyer J C, Girit C O, Crommie M F and Zettl A 2008 Appl. Phys. Lett. 92 123110 Lange S, Schmidt P and Nilges T 2007 Inorg. Chem. 46 4028 [45] ns. Matter 14 2745 José M S, Emilio A, Julian D G, Alberto G, Javier J, Pablo O and Daniel S-P 2002 J. Phys.: Conde [46] [47] [48] [49] [50] 125112 [51] [52] Kresse G and Hafner J 1993 Phys. Rev. B 48 13115 Kresse G and Furthmüller J 1996 Phys. Rev. B 54 11169 Perdew J P, Burke K and Ernzerhof M 1996 Phys. Rev. Lett. 77 3865 Jiří K, David R B and Angelos M 2010 J. Phys.: Condens. Matter 22 022201 Thonhauser T, Cooper V R, Li S, Puzder A, Hyldgaard P and Langreth D C 2007 Phys. Rev. B 76 Blöchl P E 1994 Phys. Rev. B 50 17953 Kresse G and Joubert D 1999 Phys. Rev. B 59 1758 13 [53] [54] [55] Troullier N and Martins J L 1991 Phys. Rev. B 43 1993 Louie S G, Froyen S and Cohen M L 1982 Phys. Rev. B 26 1738 Monkhorst H J and Pack J D 1976 Phys. Rev. B 13 5188 [56] 2007 Solid State Commun. 143 101 Meyer J C, Geim A K, Katsnelson M I, Novoselov K S, Obergfell D, Roth S, Girit C and Zettl A [57] [58] [59] [60] 134 94 [61] [62] Hÿtch M J and Stobbs W M 1994 Ultramicroscopy 53 191 Howie A 2004 Ultramicroscopy 98 73 Thust A 2009 Phys. Rev. Lett. 102 220801 Krause F F, Müller K, Zillmann D, Jansen J, Schowalter M and Rosenauer A 2013 Ultramicroscopy Chuvilin A, Meyer, J. C., Algara-Siller, G., & Kaiser, U. 2009 New Journal of Physics 11 083019 Suenaga K and Koshino M 2010 Nature 468 1088 [63] un. 4 2723 Kim K, Coh S, Kisielowski C, Crommie M F, Louie S G, Cohen M L and Zettl A 2013 Nat. Comm Alem N et al. 2012 Phys. Rev. Lett. 109 205502 Koskinen P, Malola S and Häkkinen H 2008 Phys. Rev. Lett. 101 115502 Wassmann T, Seitsonen A P, Saitta A M, Lazzeri M and Mauri F 2008 Phys. Rev. Lett. 101 096402 Zhu Z and Tománek D 2014 Phys. Rev. Lett. 112 176802 Masih Das P et al. 2016 ACS Nano 10 5687 [64] [65] [66] [67] [68] 14  Figure 1. Complementary thickness determination of a BP flake. (a) Optical microscope image of a BP flake on PMMA/SiO2/Si substrate. Scale bar, 2μm. (b) AFM image of the same flake. The height profile is along the white dashed line. Scale bar, 2μm (c) Transmission-mode optical microscope image of the BP flake after transfer to a TEM grid. Scale bar, 2μm (d) Intensity profile along the red arrow in panel c. (e) Bright-field TEM image of the same BP flake. Scale bar, 1μm. (f) Electron diffraction intensity profile along the yellow arrow in inset. Inset shows the diffraction pattern acquired at the 5- layer region. (g) TEM image (top) and simulated image (bottom) from 5- layer with the defocus value of 6nm. (h) Intensity profiles from TEM image (black solid line) and simulation image (red and blue solid line with circle) along the c-axis of crystal. Blue line shows the simulation profile after normalization (×0.5). 15  Figure 2. Optical transmittance of suspended BP flakes as a function of layer thickness. The observed reduction of transmittance is 3.3±0.3% per layer. Inset images show exemplary investigated samples (left: transmission-mode optical microscope image, right: TEM images of the same area). 16  Figure 3. Atomic resolution TEM imaging and simulation of BP. (a) TEM image of BP showing inhomogeneous thickness at different locations. Scale bar, 2 nm. The area inside dashed squares is used for intensity profile analysis. (b) TEM images (left) acquired at different sample locations and simulated images (right) from the chosen thickness and defocus values. Scale bar, 0.5nm. (c) Intensity profiles from TEM images (solid lines) and simulation images (dashed line) along the c-axis. Black, red, blue, cyan, and pink colors indicates the data from region A, B, C, D, and D'. D' indicate the same location of D at different defocus value. All the simulation plots are normalized (×0.5). All experimental intensity profiles are the average from 5 unit cells. 17  Figure 4. Atomic resolution TEM images of BP edges. (a) TEM image of amorphous structure at BP edge sites as indicated by yellow arrows. Scale bar, 1nm. (b) Crystalline BP edge produced by e-beam irradiation. The image was acquired after 48-second e-beam irradiation from panel a. Atoms at amorphous edge are etched by e-beam and consequently crystalline edge is exposed. The red box is the field of view for panel c. Scale bar, 1nm. (c) Zoom-in image of the BP edge. Scale bar, 0.2nm. (d) Calculated atomic model of reconstructed ZZ edge configuration 1 (RZZ1) of triple-layer BP. The atomic layer at the bottom (red) is overlapped with the top layer (blue). (e) TEM image simulation of the atomic model in panel d. (f) The same simulation image with atomic model overlay. (g) Atomic model of regular ZZ edge configuration without reconstruction. (h) TEM image simulation of regular ZZ edge structure. (i) Atomic model of reconstructed ZZ configuration 2 (RZZ2) of triple-layer BP. (j) TEM image simulation of the atomic model in panel i. Defocus value for all the simulation images is -2nm. 18  Figure 5. BP Edge formation energy calculations. (a) Atomistic models used for edge formation energy calculations. Pristine zigzag (ZZ), armchair (AC), reconstructed zigzag 1 (RZZ1), and reconstructed zigzag 2 (RZZ2) edge configurations are shown. (b) Edge formation energy for different edge configurations, which is given by Ef = (1/2L)(Eribb – NPEbulk), where Eribb is the total energy of a ribbon with NP atoms in the supercell, and Ebulk total energy per atom in monolayer phosphorene. L is the ribbon edge length. 19 Figure 6. Fabrication of a BP nano-constriction by electron beam irradiation. (a-e) A time series of TEM image of BP under electron beam irradiation. Interval time of frame is 1.5 seconds. Scale bar, 1nm. (f-j) The same TEM images with color overlay. Different colors indicate triple layer (blue), thicker (pink), and amorphous (yellow) regions. 20 

1208.3997

1

1208

Towards the chemical tuning of entanglement in molecular nanomagnets

Antiferromagnetic spin rings represent prototypical realizations of highly correlated, low-dimensional systems. Here we theoretically show how the introduction of magnetic defects by controlled chemical substitutions results in a strong spatial modulation of spin-pair entanglement within each ring. Entanglement between local degrees of freedom (individual spins) and collective ones (total ring spins) are shown to coexist in exchange-coupled ring dimers, as can be deduced from general symmetry arguments. We verify the persistence of these features at finite temperatures, and discuss them in terms of experimentally accessible observables.

cond-mat.mes-hall

cond-mat

Towards the chemical tuning of entanglement in molecular nanomagnets 1 Dipartimento di Fisica, Universit`a di Modena e Reggio Emilia, Italy and I. Siloi1,2 and F. Troiani2 2 S3 Istituto Nanoscienze-CNR, Modena, Italy (Dated: September 12, 2018) Antiferromagnetic spin rings represent prototypical realizations of highly correlated, low- dimensional systems. Here we theoretically show how the introduction of magnetic defects by controlled chemical substitutions results in a strong spatial modulation of spin-pair entanglement within each ring. Entanglement between local degrees of freedom (individual spins) and collective ones (total ring spins) are shown to coexist in exchange-coupled ring dimers, as can be deduced from general symmetry arguments. We verify the persistence of these features at finite temperatures, and discuss them in terms of experimentally accessible observables. PACS numbers: 03.67.Bg,75.50.Xx,75.10.Jm From a magnetic perspective, most molecular nano- magnets (MNs) can be essentially regarded as spin clus- ters with dominant exchange interaction [1]. As such, they represent prototypical examples of correlated, low- dimensional quantum systems. It is thus tempting to consider how the wide tunability of their physical prop- erties, enabled by chemical synthesis, can allow to con- trol quantum entanglement [2, 3]. In particular, differ- ent forms of entanglement can be possibly fine-tuned by chemical processing of well defined molecular building blocks, such as the controlled substitution of single mag- netic ions or the growth of supramolecular bridges, both of which have been recently demonstrated in Cr-based wheels [4, 5]. The former process results in the intro- duction of magnetic defects in otherwise homogeneous molecules [6], thus affecting correlations between their constituent spins. The latter one can instead induce weak exchange couplings between two or more MNs [7], so as to entangle their total spins [8]. The effect of magnetic defects on entanglement has been widely investigated in infinite systems of 1/2 spins [2]. With respect to these, MNs present some relevant dif- ferences, that make them peculiar model systems: they typically consist of s > 1/2 spins, are finite systems, and are characterized by point symmetries rather than translational invariance. The role played by these fea- tures [9], as well as by the specific magnetic defects and supramolecular structures mentioned above, represents a specific reason of interest for the study of quantum en- tanglement in MNs. Here we refer to a prototypical class of molecules, namely the heterometallic octahedral Cr7M wheels, obtained by replacing one of the Cr ions in the parent Cr8 molecule with different transition metals M [10, 11]. We theoretically show how the magnetic defects spatially modulate spin-pair entanglement, and identify a clear dependence of such effect on the impurity spin sM . We finally investigate the interplay between intra- and inter-molecular entanglement in dimers of Cr7M rings, whose compatibility can be deduced from general sym- metry arguments. All these features are fully captured by local observables, such as exchange energy of individual spin pairs and partial spin sums, that are now available in direct geometry inelastic neutron scattering [12, 13]. The dominant term in the spin Hamiltonian of the Cr7M rings is the antiferromagnetic exchange between 0.4 0.2 0.0 1.5 1.0 0.5 0.0 h t W / W 0.0 0.4 T / J 0.8 1.2 1.6 EF ( i,i+1) N ( i,i+1) (a) 0 1 2 T / J (b) 4 3 FIG. 1: (color online) (a) Negativity N and entanglement of formation EF for nearest neighboring spins of the Cr8 ring, as a function of temperature. The dotted lines are the gaus- sian fits of the reported points. (b) Temperature dependence of the entanglement witnesses WS (red) and WH (blue), nor- malized to the respective threshold values: W th S = 12 and W th H = −18J. The shaded areas correspond to the tempera- ture ranges where the two EWs detect entanglement. nearest neighbors [14]: H = J N −2 X k=1 sk · sk+1 + J ′ sN · (sN −1 + s1), (1) where sk<N = sCr = 3/2, sN = sM, and N = 8. Addi- tional terms, accounting for local crystal field and dipo- lar interactions, are typically two orders of magnitude smaller, and can be safely neglected in the present con- text. The spin sM introduced by the chemical substi- tution determines the total spin S of the ring ground state. In particular, the spin values for the elements M=Zn, Cu, Ni, Cr, Fe, Mn, are sM = 0, 1/2, 1, 3/2, 2, 5/2, corresponding to S = 3/2, 1, 1/2, 0, 1/2, 1, respectively [10, 11]. In order to investigate the entanglement properties of the Cr7M molecules, we compute entanglement witnesses and measures [3, 15]. An entanglement witness (EW) is an observable whose expectation value can exceed a 0.6 0.5 0.4 0.3 0.2 0.1 0.0 ) i , 1 - i ( N Cr7Cu Cr7Zn Cr7Fe Cr8 Cr7Ni Cr7Mn sM>3/2 sM<3/2 1 2 3 4 5 i (spin pair) 6 7 8 FIG. 2: (color online) Negativity of the neighboring spin pairs in the Cr7M rings. The values refer to the ground state, with M = S, of the spin Hamiltonian H. The inset shows a pic- torial representation of the entanglement localization induced by the defect M: the shaded areas highlight the most entan- gled spin pairs. given threshold only in the presence of (a particular form of) entanglement. For a ring consisting of N exchange- coupled spins s (with J > 0), the Hamiltonian itself can be regarded as an EW, being hHi ≥ −JN s2 for any fully separable density matrix ρ [16, 17]. The violation of such inequality implies the presence of entanglement within the system, and in particular between pairs of neighboring spins. We generalize the above criterion to the case of a ring with a spin sN 6= sk<N and two different exchange constants (Eq. 1): WH ≡ hHi ≥ −(N − 2)Js2 Cr − 2J ′sCrsM ≡ W th H , (2) being J, J ′ > 0. The full separability of the density ma- trix also implies a lower bound for the variances of the αi−hSαi2] ≥ N s. total-spin projections [18]: Pα=x,y,z[hS 2 In rotationally invariant spin Hamiltonians, the above quantity can be identified with the magnetic susceptibil- ity, up to a multiplicative constant [19]. We modify the above criterion to account for the presence of the mag- netic defect (Eq. 1): WS ≡ X α=x,y,z [hS 2 αi−hSαi2] ≥ (N−1)sCr+sM ≡ W th S . (3) We note that, unlike WH , WS depends on correlations between all spin pairs, and allows to detect forms of en- tanglement that don't show up in the two-spin density matrices. While EWs represent a practical means for the detec- tion of entanglement, its quantification requires entangle- ment measures [3]. In particular, entanglement between the spin pairs (si, sj) is hereafter quantified by the neg- ativity N of their reduced density matrices ρij. This is defined as: N (ρij ) = (Pk λk − λk)/2, where λk are the eigenvalues of the partially transposed density matrix ρTj ij . Homometallic ring -- The parent molecule of the het- erometallic Cr7M rings is Cr8 [20]. Its energy spectrum is 2 characterized by an S = 0 ground state, separated from the first excited multiplet by a gap ∆ ≃ 0.559J [21]. As for the case of an 8 qubit ring [22, 23], spin-pair entan- glement in the absence of an applied magnetic field is limited to nearest neighbors: we find in fact that N = 0 at all temperatures for any other spin pair. The tempera- ture dependence of N (ρi,i+1) presents instead a finite and nearly constant value for T /J . ∆, followed by a smooth decay at higher temperatures [Fig. 1 (a), black squares]. The threshold value of the temperature at which N van- ishes is given by: TN ≃ 1.58J. The same qualitative be- havior is found for the entanglement of formation EF [2], here reported as a benchmark (red squares). The witness WH , like N , only reflects entanglement between neigh- boring spins, and in fact provides a similar value for the threshold temperature: WH < W th H for T < TH ≃ 1.5J [Fig. 1 (b)]. The witness WS reveals instead further forms of entanglement, that persist up to higher tem- peratures, being WS < W th S for T < TS ≃ 2.75J. We finally note that values of WH /J lower than -20.65 and -21.8 imply the presence in the system state of three- and five-spin entanglement, respectively [24]. In the Cr8 ring, these thresholds are exceeded for T /J lower than 1.05 and 0.74. Heterometallic rings -- The chemical substitution of one Cr with an M ion introduces a magnetic defect in the molecule. This consists in the replacement of an sCr with an sM 6= 3/2 spin, and - in principle - in the modification of the exchange coupling between s8 and its nearest neighbors (Eq. 1). In the following we focus on the case J ′ = J, which is compatible with the current estimates [11], and allows to isolate the dependence on sM of spin-pair entanglement. In particular, we compute the negativity N (ρi−1,i) to quantify the ground-state en- tanglement, and the local witnesses WHi : WHi ≡ Jhsi−1 · sii ≥ −Jsi−1si ≡ W th Hi . (4) Violation of the above inequality implies entanglement specifically between si−1 and si. The witnesses WHi , that can be regarded as the local versions of WH = P8 i=1 WHi , are used hereafter to estimate the threshold tempera- tures up to which entanglement persists in the various molecules. The negativity corresponding to the ground state of H is reported in Fig. 2. The translational invariance that characterizes the case of Cr8 (black squares) is re- placed by a strong spatial modulation. The negativity of the spin pairs with i = 1, 3, 6, 8 is an increasing function of sM , while for i = 2, 7 the opposite occurs. For each molecule, the oscillations of N (ρi−1,i) as a function of i are paralleled by those of WHi (not shown). In fact, they can be intuitively explained in terms of competi- tion between the non-commuting exchange operators of H, since the state that maximizes entanglement between si and si−1 coincides here with the one that minimizes their exchange energy [25]. In particular, large values of the impurity spin tend to minimize WH1 (and thus to maximize entanglement of sM with s1), at the expense of the competing term WH2 (and of entanglement between s1 and s2). This in turn favors exchange interactions and quantum correlations between s2 and s3, and so on. The  J / H T 2.0 1.8 1.6 1.4 1.2 1.0 0.5 1.0 1.5 sM (a) 2.5 2.0 (b) 0.0 0.5 1.0 1.5 2.0 2.5 sM T S / J 1.6 1.5 1.4 2.8 2.4 2.0 J / S T Zn Ni Fe Cu Cr Mn 1.6 2 4 n 6 (c) 8 1.000 0.995 0.990 0.985 ) ( i N (a) 0.0 0.1 ) ( i N 1.000 0.995 0.990 0.985 (c) i=4 i=2 0.4 0.5 i=3 i=4 i=3 i=1 0.2 0.3 1- 2 i=2 i=1 0.4 1- 2 0.0 0.2 0.6 0.0 0.2 0.4 1- 2 3 Cr7Ni 0.2 D i ( ) 0.1 (b) 0.0 0.0 0.1 0.2 0.3 0.4 0.5 1- 2 Cr7Cu 0.3 0.2 D i ( 0.1 ) (d) 0.6 0.0 FIG. 3: (color online) Threshold temperatures of the local witnesses WHi as a function of the impurity spin sM: (a) M-Cr pair (i = 1); (b) three inequivalent Cr-Cr pairs i = 2, 3, 4 (blue, orange, green). The grey curve, reported in both panels, gives the threshold temperatures of the global witness WH. (c) Threshold temperatures for the witnesses WSn as a function of the spin number n and of the impurity spin sM. inverse applies to rings with sM < sCr. These general trends are confirmed by the finite- temperature results. In Fig. 3 (a,b) we report the tem- peratures below which the witnesses WHi violate the in- equalities Eq. 4, thus detecting entanglement in the cor- responding spin pair. The spin pairs that exhibit the strongest dependence on sM are the ones that include or in particular, TH1 in- are next to the magnetic defect: creases linearly with sM [panel (a), red squares], while TH2 decreases monotonically [panel (b), blue squares]. The other two inequivalent spin pairs also display a monotonic - though weaker - dependence on sM (green and orange squares). These different behaviors cannot be appreciated in TH (sM) (grey curve in both the upper panels). In fact, the threshold temperature of the global witness WH increases monotonically for finite values of the impurity spin, and systematically underestimates the persistence of entanglement, being THi > TH for at least some i in all molecules. In order to bridge local (two-spin) and global witnesses, we introduce the EWs WSn , corresponding to the vari- ances of partial spin sums: n l=1 α=x,y,z X l=1 Sn , (5) sl ≡ W th [h(Sn,α)2i − hSn,αi2] ≥ WSn ≡ X where Sn = Pn sl and thus WSn ≡ WS. Their thresh- old temperatures TSn are reported in Fig. 3(c) for even values of n. The most prominent feature is the signifi- cant increase of TSn with n, for all the Cr7M molecules. The same feature shows up within n−spin subensembles with odd n (not shown). Entanglement for TH . T . TS is thus progressively averaged away by the partial traces that are performed to derive the reduced density ma- trices ρn = Trn+1,...,8{ρ} of decreasing spin number n. Based on calculations performed for the 8-qubit ring, we conjecture that the density matrix ρ(T ) in such temper- ature range is given by the mixture of terms that include FIG. 4: (color online) Negativity of neighboring spin pairs belonging to Cr7M ring dimers, with M=Ni (a) and M=Cu (c): ∆Ni = 1−Ni(α)/Ni(0). (b,d) Trace distance Di between the reduced density matrices ρA i−1,i(0). The correspondence between spin pairs and colors is reported in panels (a,c). i−1,i(α) and ρA two-spin entanglement between different pairs. In other words, ρ(T ) would be a mixed 2-producible state [26], whereas multi-spin entanglement is limited to T . J (see discussion on the Cr8 molecule). As to the dependence on the impurity spin, this is most significant for n = 2 and n = 4. In particular, TS2 decreases monotonically with sM , like TH2 (that refers to the same spin pair), while TS8 = TS displays the opposite behavior. Ring dimers -- The Cr7M rings are composite quan- tum systems, whose subsystems are represented by the constituents magnetic ions. However, they can also be re- garded as building blocks of supramolecular assemblies, consisting of molecule dimers or oligomers [5]. Here, the intermolecular exchange typically mixes states belonging to the ground S multiplet of each ring, and can thus entangle collective degrees of freedom, such as the total spin projections of the MNs [8]. In the following, we investigate the interplay between individual- and collective-spin entanglement in a proto- typical state of a Cr7Ni-ring dimer: AB(α)i = α1/2,−1/2i− (1− α2)1/2− 1/2, 1/2i, (6) ΨN i where A and B label the two rings (with SA/B = 1/2), and the components MA, MBi are labelled after the val- ues of the ring-spin projections (SA z and SB z ). Entan- glement between these collective spins can be quanti- ABi) = α√1 − α2. fied by the negativity: NAB ≡ N (ΨN i The state ΨABi thus varies from a maximally entangled (NAB = 1/2) to a factorizable state (NAB = 0) as α ranges from 1/√2 to 1. Entanglement between two in- dividual spins, sA j , of ring A is quantified by the negativity of the two-spin reduced density matrix ρA ij : i and sA ij(α) = α2ρ1/2 ρA ij + (1 − α2)ρ−1/2 ij . (7) ij Here, ρMA denotes the two-spin density matrix corre- sponding to the ground state of the single-ring Hamilto- nian (Eq. 1) with total-spin projection MA. ij ij and ρ−1/2 Entanglement between the total spin projections of A and B (α < 1) results in a mixing of ρ1/2 (Eq. 7). This generally tends to reduce entanglement between si and sj - due to the convexity of entanglement mea- sures [15] - making quantum correlation between individ- ual and collective degrees of freedom mutually exclusive. As shown in Fig. 4, such reduction is however very lim- ited in the case of exchange-coupled rings. In fact, as α varies from 1/√2 to 1, the relative change of the nega- tivity N (ρi−1,i) is below 2% for all the spin pairs (panel a), in spite of the fact that the reduced density matri- ces ρi−1,i vary significantly with α (b). Such variation is quantified by the trace distance [27] between ρij(α) and the reference state ρij(1): Dij(α) = (1/2)Trq[ρij (α)−ρij(1)]†[ρij(α)−ρij(1)]. (8) Entanglement between individual spins within each Cr7Ni ring (sA j ), is thus fully compatible with that between collective spins (SA z and SB z ), such as that induced by weak intermolecular exchange. i and sA The same applies to different ring dimers, such as that formed by two Cr7Cu rings (SA = SB = 1). Here, we consider the prototypical state ΨCu AB(α)i = [(1 − α2)/2]1/2(1,−1i + − 1, 1i) − α0, 0i. (9) This passes from maximally entangled (NAB = 1) to fac- torized (NAB = 0), as α varies from 1/√3 to 1, being NAB = αp2(1 − α2)+(1−α2)/2. In this same range, the negativity displays a very limited decrease for all pairs of neighboring spins (panel c), also for significant values of the trace distance (d). Symmetry arguments suggest that such compatibility between intra- and inter-molecular entanglement is more 4 In fact, the entanglement witnesses WHi are general. scalar operators. As results from the Wigner-Eckart the- orem [28], their expectation value is thus identical for all the states that form an irreducible representation of the rotational group, such as the eigenstates belonging to the ground S multiplet of the spin-ring Hamiltonian H (Eq. 1). As a consequence, if any of the inequalities Eqs. 2,4 is violated by a single-ring ground state, it's also vi- olated by ring-dimer singlet states such as ΨN i and ΨCu z are maximally en- tangled and the single-ring density matrix is a mixture of all the ground multiplet states. The same argument applies to the witnesses WSn , provided that their state is rotationally invariant: in this case, hSn,αi = 0, and WSn can be identified with the expectation value of the scalar operator S2 n. AB(1/√3)i, where SA AB(1/√2)i z and SB In conclusion, we have shown how the introduction of magnetic defects in the heterometallic Cr7M rings intro- duces a strong spatial modulation of pairwise entangle- ment, that persists at finite temperatures. This suggests that suitably combined chemical substitutions can rep- resent an effective means for engineering entanglement in molecular systems. Besides, we quantitatively show that in ring dimers entanglement between individual and collective spins can coexist, and deduce the generality of such property from symmetry arguments. The discussed features are fully captured by local observables acting as entanglement witnesses, such as the exchange energy of spin pairs and the variances of partial spin sums, that are accessible to direct geometry inelastic neutron scattering. We thank M. Affronte and V. Bellini for useful discus- sions. We acknowledge financial support from PRIN of the Italian MIUR. [1] D. Gatteschi, R. Sessoli, and J. Villain, Molecular nano- magnets (Oxford University Press, 2007). [2] L. Amico, R. Fazio, A. Osterloh, and V. Vedral, Rev. Mod. Phys. 80, 517 (2008). [10] R. Caciuffo, T. Guidi, G. Amoretti, S. Carretta, E. Liv- iotti, P. Santini, C. Mondelli, G. Timco, C. A. Muryn, and R. E. P. Winpenny, Phys. Rev. B 71, 174407 (2005). [11] V. Bellini and M. Affronte, J. Phys. Chem. B 114, 14797 [3] R. Horodecki, P. Horodecki, M. Horodecki, and (2010). K. Horodecki, Rev. Mod. Phys. 81, 865 (2009). [12] C. Brukner, V. Vedral, and A. Zeilinger, Phys. Rev. A [4] M. Affronte, S. Carretta, G. A. Timco, and R. E. P. 73, 012110 (2006). Winpenny, Chem. Commun., p. 1789 (2007). [13] C. Stock, R. A. Cowley, J. W. Taylor, and S. M. Ben- [5] G. A. Timco, S. Carretta, F. Troiani, F. Tuna, R. J. Pritchard, C. A. Muryn, E. J. L. McInnes, A. Ghirri, A. Candini, P. Santini, et al., Nature Nanotech. 4, 173 (2009). [6] F. K. Larsen, E. J. L. McInnes, H. E. Mkami, J. Over- gaard, S. Piligkos, G. Rajaraman, E. Rentschler, A. A. Smith, G. M. Smith, V. Boote, et al., Angew. Chem. Int. Ed. 42, 101 (2003). [7] V. Bellini, G. Lorusso, A. Candini, W. Wernsdorfer, T. B. Faust, G. A. Timco, R. E. P. Winpenny, and M. Affronte, Phys. Rev. Lett. 106, 227205 (2011). [8] A. Candini, G. Lorusso, F. Troiani, A. Ghirri, S. Car- retta, P. Santini, G. Amoretti, C. Muryn, F. Tuna, G. Timco, et al., Phys. Rev. Lett. 104, 037203 (2010). [9] F. Troiani, Phys. Rev. A 83, 022324 (2011). nington, Phys. Rev. B 81, 024303 (2010). [14] F. Troiani, A. Ghirri, M. Affronte, S. Carretta, P. San- tini, G. Amoretti, S. Piligkos, G. Timco, and R. E. P. Winpenny, Phys. Rev. Lett. 94, 207208 (2005). [15] O. Guhne and G. T´oth, Phys. Rep. 474, 1 (2009). [16] C. Brukner and V. Vedral, arXiv:quant-ph/0406040 (2004). [17] G. T´oth, Phys. Rev. A 71, 010301 (2005). [18] M. Wie´sniak, V. Vedral, and C. Brukner, New J. Phys. 7, 258 (2005). [19] S. Ghosh, T. F. Rosenbaum, G. Aeppli, and S. N. Cop- persmith, Nature 425, 48 (2003). [20] J. van Slageren, R. Sessoli, D. Gatteschi, A. A. Smith, M. Helliwell, R. E. P. Winpenny, A. Cornia, A.-L. Barra, A. G. M. Jansen, E. Rentschler, et al., Chem.-Eur. J. 8, 277 (2002). 052316 (2004). [21] O. Waldmann, T. Guidi, S. Carretta, C. Mondelli, and [26] O. Guhne, G. T´oth, and H. J. Briegel, New J. Phys. 7, A. L. Dearden, Phys. Rev. Lett. 91, 237202 (2003). 229 (2005). [22] K. M. O'Connor and W. K. Wootters, Phys. Rev. A 63, 052302 (2001). [23] M. C. Arnesen, S. Bose, and V. Vedral, Phys. Rev. Lett. 87, 017901 (2001). [24] F. Troiani and I. Siloi, arXiv:1206.4928v3 (2012). [25] C. M. Dawson and M. A. Nielsen, Phys. Rev. A 69, [27] M. Nielsen and I. L. Chuang, Quantum Computation and Quantum Information (Cambridge University Press, 2000). [28] B. Tsukerblat, Group Theory in Chemistry and Spec- troscopy (Academic Press, New York, 1994). 5 

1212.5283

1

1212

Effects of localized trap-states and corrugation on charge transport in graphene nanoribbons

We investigate the role played by electron traps on adiabatic charge transport for graphene nanoribbons in the presence of an acoustically induced longitudinal surface acoustic wave (SAW) potential. Due to the weak longitudinal SAW-induced potential as well as the strong transverse confinement by a nanoribbon, minibandsof sliding tunnel-coupled quantum dots are formed so that by varying the chemical potential to pass through the minigaps, quantized adiabatic charge transport may be obtained. We analyze the way that the minigaps may be closed, thereby destroying the likelihood of current quantization in a nanoribbon. We present numerical calculations showing the effects due to electron traps which lead to localized-trap energy levels within the minigaps. Additionally, for comparison, we present results for the minibands of a corrugated nanoribbon in the absence of a SAW.

cond-mat.mes-hall

cond-mat

Effects of localized trap-states and corrugation on charge transport in graphene nanoribbons Oleksiy Roslyak1, Upali Aparajita1, Godfrey Gumbs1,2, and Danhong Huang3 1Department of Physics and Astronomy, Hunter College, City University of New York 695 Park Avenue, New York, NY 10065, USA 2Donostia International Physics Center (DIPC), P de Manuel Lardizabal, 3Air Force Research Laboratory (ARFL/RVSS), Kirtland Air Force Base, NM 87117, USA 4, 20018 San Sebastian, Basque Country, Spain and (Dated: April 6, 2019) We investigate the role played by electron traps on adiabatic charge transport for graphene nanoribbons in the presence of an acoustically induced longitudinal surface acoustic wave (SAW) potential. Due to the weak longitudinal SAW-induced potential as well as the strong transverse confinement by a nanoribbon, minibandsof sliding tunnel-coupled quantum dots are formed so that by varying the chemical potential to pass through the minigaps, quantized adiabatic charge trans- port may be obtained. We analyze the way that the minigaps may be closed, thereby destroying the likelihood of current quantization in a nanoribbon. We present numerical calculations showing the effects due to electron traps which lead to localized-trap energy levels within the minigaps. Additionally, for comparison, we present results for the minibands of a corrugated nanoribbon in the absence of a SAW. I. INTRODUCTION A considerable amount of research work has been carried out so far on the design and improvement of electronic devices which are based on the use of quantized adiabatic charge transport. [1 -- 10] Moreover, under a surface-acoustic wave (SAW), the inelastic capture and tunneling escape effects on the non-adiabatic transport of photo-excited charges in quantum wells was also investigated. [11] The underlying challenge is to produce a device with an accuracy for the quantized current of one part in 108 on the plateaus. When this goal is achieved, one application of this device would be in metrology for standardizing the unit of current. At the present time, a SAW is launched on a piezoelectric heterostructure, such as GaAs/AlGaAs, and GHz single/few-electron pumps have been gaining close scrutiny due to the fact that the measured currents lie within the nanoamp range, high enough for the measured current to be suitable as a current standard. However, these pumps have so far been capable of delivering electrons/holes in each cycle of a sliding dynamic quantum dot (QD), giving rise to a quantized current with an accuracy of one part in 106 as reported in Refs. [3 -- 8]. Interestingly, in Ref. [12] a measurement was carried out of the noise accompanying a 3-GHz SAW pump. It was observed in this experiment that the current near the lowest plateau, corresponding to the transfer of one electron per SAW cycle, is dominated by shot noise. However, away from the plateau, the noise is attributed to electron traps in the material. There have been some attempts to increase the flatness of the plateaus by applying magnetic fields. [13, 14] Some time ago, a proposal was put forward by Thouless [10] which would make use of quantized adiabatic charge transport. This adiabatic approach involves the use of a one-dimensional (1D) electron system subjected to a slowly- sliding periodic potential. Relatively simple analysis indicates that in such a 1D system minigaps are generated in instantaneous electronic spectra as a function of the SAW amplitude. With the use of a gate, the chemical potential can be varied by applying a voltage to the gate. Consequently, when the chemical potential lies within a minigap, there will be an integral multiple of electron charge transported across the system during a single time period. [1] In other words, by combining with the strong transverse confinement of a nanoribbon, the weal longitudinal SAW potential has induced a series of dynamic (sliding) tunnel-coupled QDs whose impenetrable"wall is constructed through destructive interference of the electronic wave functions around a minimum of the SAW potential. In principle, such an adiabatic- transport device could provide an important application, like a current standard. Talyanskii, et al. [1] investigated the physical mechanisms of quantized adiabatic charge transport in carbon nanotubes for a SAW to produce a periodic potential required for miniband/minigap formation. In the presence of a SAW, the scattering effects from impurities embedded in a 1D electronic system are expected to play an important role on the flatness of a current plateau. The current quantization should be completely smeared out when the level broadening from impurity scattering becomes comparable to the minigaps of dynamic tunnel-coupled QDs. On the other hand, we can also simulate localized electron traps by superposing a series of negative δ-potentials onto a SAW potential within each spatial period. Consequently, we expect a set of localized trap states occurring within the minigaps of dynamic QDs. This provides an escape channel for the QD-confined electrons being carried 2 1 0 2 c e D 0 2 ] l l a h - s e m . t a m - d n o c [ 1 v 3 8 2 5 . 2 1 2 1 : v i X r a 2 by the SAW. This trap mechanism is quite different from the impurity one [1] where a spatial average with respect to the distribution of impurities within a dynamic QD is inevitable due to a SAW. In this paper, we consider a 1D Dirac-like electron gas in a graphene nanoribbon in the presence of a SAW. We will introduce two mechanisms for miniband formation. First, the nanoribbon is modulated by a longitudinal potential from a SAW. Secondly, the nanoribbon is periodically corrugated. We notice that the second mechanism does not lead to a quantized current but instead produces traps for Dirac electrons, thereby limiting electron mobility. Our numerical calculations reveal that localized electron trap states are an effective mechanism to adversely affect the adiabatic transport because the localized-trap levels lying within the minigaps are very sensitive to the phase of either the SAW or the corrugation-induced potential. Varying the weight or the position of the δ-potential leads to different positions of localized trap levels within the minigaps of the nanoribbon. Therefore, these inevitable fluctuations of the trap potential in a realistic system would most likely impede the current quantization. The rest of the paper is organized as follows. In Sec. II, we present the formalism for calculating band structure with localized trap states for nanoribbons in the presence of a SAW. In Sec. III, numerical results for nanoribbons in the absence/presence of a SAW and those for corrugated nanoribbons in the absence of a SAW are presented to demonstrate and explain the localized trap states within the minigaps. The conclusions drawn from these results are briefly summarized in Sec. IV. II. MINIBAND STRUCTURE WITH LOCALIZED TRAP STATES The work done by Talyanskii, et al. [1] on quantum adiabatic charge transport focused on the coupling between a semimetallic carbon nanotube and a SAW. The electron backscattering from the SAW potential is used to induce a miniband spectrum. The electron interactions enhance the minigaps thereby improving current quantization. The effect due to impurities in the carbon nanotube is averaged by a SAW potential. For the cases of a semimetallic carbon nanotube, semiconducting carbon nanoribbon with applied SAW potential and corrugated nanoribbon, the energy levels are given by the spectra of discretized 1D Dirac Hamiltonian (see the Appendix A for detailed derivations)  H = . . . . . . . . . . . . . . . an−1 b∗ n−1 c∗ . . . bn−1 a∗ n−1 0 an . . . 0 c∗ bn . . . cn 0 n n−1 . . . . . . . . . 0 0 0 0 cn+1 0 . . . . . . . . . . . . . . . . . . 0 0 0 n cn−1 b∗ n a∗ 0 0 c∗ 0 n 0 cn an+1 b∗ 0 n+1 bn+1 a∗ c∗ . . . . . . . . . n+1 n+1 . . . . . . . . . . . . . . . . . . . . .  2N×2N . (1) The eigenvalue problem is defined within the spatial interval 0 < x < 2π/k and assumes periodic boundary conditions. In this notation, k stands for either the wave number kSAW of the SAW potential or the wave number kc of an effective potential induced by the corrugation. The discretization of the Hamiltonian is provided by the mesh xn = nδx with n ∈ 0 . . . N − 1 and δx = 2π/kN . For either the carbon nanotube or graphene nanoribbon, the parameters for the Hamiltonian matrix in Eq. (1) are given by an = 0 , −iα(xn) , ivF 2δx , bn = ∆ e cn = (2) where we have introduced the SAW and impurities combined phase (cid:34) an = vF −√ ∆c(x)[2 + C2k2 c sin(2kcxn) 2C2k3 c cos(2kcxn) + C2k2 bn = ivF k(m) c ]3/2 −2iαc(xn) , y C2k3 4∆3 c(x) c + i e ivF (cid:35) sin(2kcxn) , (4) x(cid:90) 0 α(x) = du [VSAW (u) − Vtrap(u)] = λ cos(kSAW x) + V0 θ(x − x0) , 3 (3) with λ = 2A/(vF kSAW ) the normalized SAW amplitude, vF is the Fermi velocity of graphene. Additionally, V0 = Vtrap/(vF kSAW ) and x0 denote the normalized trap-potential amplitude and position of the short-range dynamic trap for electrons, respectively, and the trap is sliding together with the SAW potential. The mass term involving ∆ is the original energy gap for the system in the absence of a SAW. In case of a nanotube, ∆ may be generated by a magnetic field. For a nanoribbon, the gap is structural for the semiconducting nanoribbon vF k(m) with k(m) being the transverse electron wave number due to finite size across the ribbon. For nanoribbons, the explicit form for the phase introduced in Eq. (3) can be found from the Appendix A. y y As far as the minigaps are concerned, the effect due to the SAW potential on the nanoribbon may be compared with corrugation. A sinusoidal corrugated semiconducting ribbon can be mapped on to a flat ribbon. The mapping introduces an additional σ1 term into the Dirac equation, as described in Appendix A. This yields where the corrugation-induced gap is given by ∆c(xn) =(cid:112)1 + C2k2 ∆c(xn) 2δx cn = , of the corrugation, and k(m) y is the quantized wave number across the nanoribbon. c cos2(kcxn), with C being the normalized amplitude III. NUMERICAL SIMULATION AND DISCUSSIONS In our numerical calculations, all the energies in Figs. 1 and 2, such as ε, ∆ and Eg, are normalized to vF kSAW . The SAW potential amplitude A is also normalized to vF kSAW . Additionally, all the energies in Fig. 3, such as ε and ∆, are measured in units of vF kc, and the corrugation amplitude C is normalized to 1/kc. Besides, the transverse wave number k(m) in all the plots is scaled by 2π/3a0. In this way, we are able to draw some universal conclusions concerning the effects due to minigaps. y In Fig. 1, we compare the energy band structure of nanoribbons for two values of ∆ in the absence of electron traps. Two values of k(m) y were chosen (light and dark colors) to describe the two lowest energy levels (see the discussions in the Appendix A). The minigaps are generated by a sliding dynamic QD and they oscillate as a function of the SAW amplitude A, as may be verified using perturbation theory, vanishing at values close, but generally not equal, to the roots of Bessel functions. Increasing or decreasing the value of ∆ results in a shift of the nodes on the graph as evidenced by comparing our results in Fig. 1. Therefore, ∆ determines not only the magnitude of the original gap in the absence of a SAW but also the size of the minigaps in the presence of a SAW. Higher energy minigaps are partially closed by the energy levels corresponding to a larger value of k(m) (not shown here). y We now introduce electron traps into our nanoribbon by superposing a negative δ-potential onto the SAW potential so as to simulate a short-range Coulomb interaction. In this case, the position of the trap is fixed in the moving SAW frame of reference, which is quite different from embedded impurities in a nanostructure. In the moving SAW frame, the embedded impurities are moving against the dynamic QDs created by both the transverse dimension of the nanostructure and the longitudinal SAW potential. This results in an average of the impurity effects with respect to these dynamic QDs in the longitudinal direction. As seen in the results presented in Fig. 2, localized trap states occur within the minigaps once the weight of the trap potential V0 becomes larger than ∆/(vF kSAW ). Relative energy value of these trap states in the presence of SAW is sensitive to the position x0 of the trap within a dynamic QD. If we set λ = V0, then the contribution to α(x) from the trap located in the nodes of the SAW potential is fully compensated by the cosine term in Eq. (3). As a result of this compensation, the localized trap states will disappear from the gap and minigap regions. If we extend the single-trap model employed in this paper to a uniform distribution of traps, the fluctuations in the phase term α(x) [see Eq. (3)] would fill up the entire minigap region with a delocalized 4 trap band. Consequently, the adiabatic approximation may not be applicable. In other words, to satisfy the adiabatic assumption, one must have dominance of the SAW potential, i.e., λ (cid:29) V0 must be satisfied. We compare the results for SAW-based dynamic QDs in Figs. 1 and 2 with those for static QDs created by corru- gation on a graphene nanoribbon in the absence of a SAW and electron traps. This we do by displaying in Fig. 3 the minigaps induced by the corrugation. We find from the figure that minigaps only exist for finite but small values of the corrugation amplitude C. This means that these minigaps are generally much less than those induced by a SAW. As a matter of fact, the existence of non-vanishing diagonal terms an given in Eq. (4), effectively mitigates the phase fluctuations in the off-diagonal terms bn. This keeps the minigaps open and the energy spectra robust even after traps have been introduced to cause a fluctuation in the phase term αc(x). Finally, let us assume that a narrow channel is formed within a two-dimensional electron-gas layer lying in the xy- plane. We will neglect the finite thickness of the quantum well for the heterostructure in the z-direction and consider the electron motion as strictly two dimensional. We will employ one of the simplest models for the gate-induced or etched [15] confining electrostatic potential. In this way, a 1D channel is formed on the two-dimensional electron-gas layer and the dynamics of massive electrons can be modeled by a discretized 1D Schrodinger equation. However, from numerical results (not shown here), we find no evidence of the minigaps for this model, i.e., the minigaps are the characteristics of Dirac fermions. IV. CONCLUDING REMARKS In conclusion, we have calculated in this paper the energy band structure for graphene nanoribbons, embedded with a single electron trap, upon which a SAW is launched. Our results show that localized trap states appear in the minigaps. More importantly, the location of the trap state-energy level is determined by the positions of the trap with respect to the phase of the sinusoidal SAW. Consequently, the adiabatic approximation might not be appropriate whenever the minigap is less or comparable with the weight of a short-range δ-potential for the trap (see Fig. 2). On the other hand, in Fig. 1, where there are no electron traps, a larger value of ∆ in the energy spectrum leads to a substantial increase in the number of minigaps as well as the ballistic current quantization. Periodic corrugation of the nanoribbon may be used instead of a SAW as a mechanism for inducing minigaps. Those are expected to be less sensitive to the presence of charged impurities or electron trap potentials. Acknowledgments The authors are very grateful to Professor Leonid Levitov for helpful suggestions and critical comments during the course of this work. His critical remarks have undoubtedly helped to strengthen the presentation of this paper. This research was supported by the contract # FA 9453-07-C-0207 of AFRL. DH would like to thank the Air Force Office of Scientific Research (AFOSR) for its support. [1] V. I. Talyanskii, D. S. Novikov, B. D. Simons, and L. S. Levitov, Phys. Rev. Lett. 87, 276802 (2001). [2] L. P. Kouwenhoven, A. T. Johnson, N. C. van der Vaart, C. J. P. M. Harmans, and C. T. Foxon, Phys. Rev. Lett. 67, 1626 (1991); M. W. Keller, A. L. Eichenberger, J. M. Martinis, and N. M. Zimmerman, Sci. 285, 1706 (1999). [3] J. M. Shilton, V. I. Talyanskii, M. Pepper, D. A. Ritchie, J. E. F. Frost, C. J. B. Ford, C. G. Smith, and G. A. C. Jones, J. Phys.: Condens. Matt. 8, L531 (1996). [4] V. I. Talyanskii, J. M. Shilton, M. Pepper, C. G. Smith, C. J. B. Ford, E. H. Linfield, and D. A. Ritchie, Phys. Rev. B 56, 15180 (1997). [5] G. R. Aızin, G. Gumbs, and M. Pepper, Phys. Rev. B 58, 10589 (1998). [6] G. Gumbs, G. R. Aızin, and M. Pepper, Phys. Rev. B 60, R13954 (1999). [7] G. Gumbs, G. R. Aızin, and M. Pepper, Phys. Rev. B 57, 1654 (1998). [8] M. D. Blumenthal, B. Kaestner, L. Li, S. Giblin, T. J. B. M. Janssen, M. Pepper, D. Anderson, G. Jones, and D. A. Ritchie, Nat. Phys. 3, 343 (2007). [9] B. Kaestner, V. Kashcheyevs, S. Amakawa, M. D. Blumenthal, L. Li, T. J. B. M. Janssen, G. Hein, K. Pierz, T. Weimann, U. Siegner, and H. W. Schumacher, Phys. Rev. B 77, 153301 (2008). [10] D. J. Thouless, Phys. Rev. B 27, 6083 (1983). [11] D. H. Huang, G. Gumbs, and M. Pepper, J. Appl. Phys. 103, 083714 (2008). [12] A. M. Robinson and V. I. Talyanskii, Phys. Rev. Lett. 95, 247202 (2005). [13] S. J. Wright, M. D. Blumenthal, G. Gumbs, A. L. T. Thorn, S. N. Holmes, T. J. B. M. Janssen, M. Pepper, D. Anderson, G. A. C. Jones, and C. N. Nicoll, Phys. Rev. B 78, 233311 (2008). [14] S. J. Wright, A. L. Thorn, M. D. Blumenthal, S. P. Giblin, M. Pepper, T. J. B. M. Janssen, M. Kataoka, J. D. Fletcher, G. A. C. Jones, C. A. Nicoll, G. Gumbs, and D. A. Ritchie, J. Appl. Phys. 109, 102422 (2011). [15] J. Cunningham, V. I. Talyanskii, J. M. Shilton, and M. Pepper, Phys. Rev. B 62, 1564 (2000). [16] V. Atanasov and A. Saxena, Phys. Rev. B 81, 205409 (2010). 5 Appendix A: Energy Band Calculations in Absence of Electron Traps 1. Carbon Nanotubes 6 The electron eigenstates in a semi-metallic nanotube are described by a 1D Dirac equation. For simplicity, a noninteracting system is considered here. Under the stationary approximation, the single particle energy spectrum ε(k) is obtained from the following perturbed 1D Dirac equation ε(k) ψα(x) = −ivF ε(k) ψβ(x) = ivF ∂ψα(x) ∂x ∂ψβ(x) ∂x + ∆ ψβ(x) + A sin(kx) ψα(x) , + ∆ ψα(x) + A sin(kx) ψβ(x) . (A1) (A2) In this notation, k represents the electron wave number along the nanotube, vF is the Fermi velocity of Dirac electrons, A is the SAW amplitude, ∆ is the energy gap of the system in the absence of a SAW, α and β label the two sublattices of graphene from which the nanotube is rolled. In addition, we require k = kSAW to satisfy momentum conservation, where kSAW is the wave number of a SAW propagating along the nanotube. To explore the miniband structure due to quantum confinement in the radial direction, a gauge transformation is implemented and is defined by (cid:21) (cid:20) ψα(x) ψβ(x) (cid:20) ψ(cid:48) (cid:21) (cid:20) e(i/2)λ cos(kx) ψ(cid:48) e(−i/2)λ cos(kx) ψ(cid:48) β(x) α(x) (cid:21) = e(i/2)σ3λ cos(kx) α(x) ψ(cid:48) β(x) = , (A3) where λ = 2A/(vF k) is the normalized SAW amplitude. Substituting Eq. (A3) into Eqs. (A1) and (A2), we obtain By introducing the following identities ε(k) e(i/2)λ cos(kx) ψ(cid:48) +∆ e(−i/2)λ cos(kx) ψ(cid:48) ε(k) e(−i/2)λ cos(kx) ψ(cid:48) +∆ e(i/2)λ cos(kx) ψ(cid:48) −ivF −A sin(kx) ψ(cid:48) ∂ ∂x ivF −A sin(kx) ψ(cid:48) (cid:105) (cid:105) β(x) β(x) , α(x) α(x) . (cid:104) (cid:104) ∂ ∂x ∂ ∂x β(x) = −ivF e(i/2)λ cos(kx) ψ(cid:48) α(x) + A sin(kx) e(i/2)λ cos(kx) ψ(cid:48) e(−i/2)λ cos(kx) ψ(cid:48) α(x) = ivF β(x) + A sin(kx) e(−i/2)λ cos(kx) ψ(cid:48) (cid:104) (cid:105) β(x) − ivF e(i/2)λ cos(kx) ∂ψ(cid:48) (cid:105) (cid:104) α(x) + ivF e(−i/2)λ cos(kx) ∂ψ(cid:48) e(−i/2)λ cos(kx) ψ(cid:48) e(i/2)λ cos(kx) ψ(cid:48) β(x) ∂x α(x) = β(x) = ∂ ∂x , α(x) ∂x , (A4) (A5) (A6) (A7) (A8) (A9) Eqs. (A5) and (A 1) may be simplified as Furthermore, by employing the basis set Ψ(cid:48)(x) ≡ (cid:110) compact matrix form, given by ε(k) ψ(cid:48) ε(k) ψ(cid:48) β(x) = −ivF α(x) = ivF −iλ cos(kx) ψ(cid:48) + ∆ e + ∆ eiλ cos(kx) ψ(cid:48) β(x) . α(x) , ∂ψ(cid:48) β(x) ∂x ∂ψ(cid:48) α(x) ∂x ψ(cid:48) α(x), ψ(cid:48) (cid:111) β(x) , the above equations can be rewritten into a ε(k) I Ψ(cid:48)(x) = H Ψ(cid:48)(x) , H = ivF σ3 ∂ ∂x + ∆ σ1 e −iλσ3 cos(kx) . 7 (A10) (A11) We will solve the eigenvalue problem within the spatial interval 0 < x < 2π/k and introduce the N -point mesh xn = n δx, where n ∈ 0 . . . N − 1 and δx = 2π/kN . In this way, the derivative on the mesh can be approximated by ∂ Ψ(cid:48)(xn)/∂x → [ Ψ(cid:48)(xn+1) − Ψ(cid:48)(xn−1)]/(2δx). Especially, on this spatial mesh, the Hamiltonian in Eq. (A11) may be projected into the matrix given in Eq. (1). 2. Graphene Nanoribbons Here, we consider an armchair graphene nanoribbon lying along the x-direction. The total number of carbon atoms (in both sublattices) across the ribbon is assumed to be M . The armchair edges mix up the graphene K and K(cid:48) valleys so that the wave function becomes where the transverse wave number is given by Ψm(x, y) = eik(m) y y ΨK, m(x) + e −ik(m) y k(m) y = 2πm 2L + a0 + 2π 3a0 , y ΨK(cid:48), m(x) , (A12) (A13) m = 0, ±1, ±2, ··· is an integer, L is the nanoribbon width, and a0 = in graphene. Nanoribbons with width L/a0 = 3M + 1 give rise to the following relation 3a/2 (a ≈ 1.42A) is the size of the unit cell k(m) y = 2πm 6M a0 + 3a0 + 2π 3a0 = 2π 3a0 , (A14) and it is clear that the minimum energy occurs at k(−2M−1) next miniband corresponds to k(−2M ) L/a0 = 3M (upper Eq.) or L/a0 = 3M − 1 (lower Eq.), the two minimal values of k(m) = 0. Therefore, such nanoriibons are metallic. The = (2π/3a0) [1/(2M + 1)]. On the other hand, for nanoriibons having width are found to be y y y k(−2M ) k(−2M ) y y = (2π/3a0) [1/(6M + 1)] = (2π/3a0) [−1/(6M − 1)] and and y k(−2M−1) k(−2M +1) y = (2π/3a0) [−2/(6M + 1)] , = (2π/3a0) [2/(6M − 1)] . Those nanoribbons are semiconducting with the energy gap determined by (cid:18) (cid:19) ∆(M ) = 2πvF 3a0 1 6M ± 1 . The x-component of the wave function in Eq. (A12) may be determined by (cid:18) ε(k) I ΨK(K(cid:48)), m(x) = H ΨK(K(cid:48)), m(x) , H = ±vF −iσ1 ∂ ∂x + ik(m) y σ2 + A sin(kx) I , (cid:19) (A15) (A16) (A17) (A18) where the ± signs correspond to K and K(cid:48) valleys. Additionally, the Hamiltonian in Eq. (A18) can be transformed into the form in Eq. (A11) after applying the following unitary transformation √ (cid:18) 2M + 1 + m (cid:19) 2M + 1 (cid:20) e−iα(x) −eiα(x) e−iα(x) eiα(x) (cid:21) U = , (A19) where α(x) = −A cos(kSAW x)/(vF kSAW ). As a result, the transformed Hamiltonian takes the form H(cid:48) = U† ⊗ H ⊗ U (cid:34) −i∂/∂x y e−i2α(x) ik(m) −ik(m) y ei2α(x) i∂/∂x (cid:35) . = ±vF Formally, this Hamiltonian is equivalent to that in Eq. (A11) after we applying the following substitutions 2α(x) → −λ cos(kSAW x) vF k(m) y → ∆ . 8 (A20) (A21) The valley sign ± does not change anything due to the mirror symmetry in the energy dispersion relation ε(k) with respect to ±k. 3. Corrugated Nanoribbons We now turn to the case of a corrugated graphene nanoribbon whose modulation is sinusoidal with amplitude C and wavelength 2π/kc along the x-axis. The model Hamiltonian for such a corrugated graphene nanoribbon has been given in Ref. [16] as H = −ivF σ1 ∂ ∂x + vF ky σ2 − ivF σ1 K(x) df (x)/dx 2 ∆c(x) , (A22) where ∆c(x) =(cid:112)1 + [df (x)/dx]2 f (x) = C sin(kcx) . and K(x) = −d2f (x)/dx2 1 + [df (x)/dx]2 , After applying the following unitary transformation (cid:20) e−iαc(x) −eiαc(x) e−iαc(x) eiαc(x) (cid:21) (cid:20) eiαc(x) eiαc(x) −e−iαc(x) e−iαc(x) (cid:21) , and U† = 1√ 2 U = 1√ 2 where αc(x) = (cid:34) = 1 2 c 0 C2k3 4 x(cid:90) (cid:112)2 + C2k2 (cid:34) sin (2kcu) c cos2 (kcu)]3/2 du [1 + C2k2 √ 2 c cos (2kcx) + C2k2 c (cid:35) , − 1(cid:112)1 + C2k2 (cid:35) c H(cid:48) = vF ∆c(x) −ikye2iαc(x) G(x) − i∂/∂x ikye−2iαc G∗(x) + i∂/∂x ∆c(x) , (A23) the transformed Hamiltonian becomes where the complex function is defined by −√ G(x) = ∆c(x) [2 + C2k2 c sin(2kcx) 2C2k3 c cos(2kcx) + C2k2 c ]3/2 C2k3 4∆3 c(x) c + i sin(2kcx) . By making use of the finite-difference method for calculating ∂/∂x along with the following basis set ϕ = {An−1, Bn−1, An, Bn, An+1, Bn+1} , the transformed Hamiltonian matrix in Eq. (A23) may be projected as  H = vF G(xn−1) ikye−2iαc(xn−1) i ∆c(xn) 2δx 0 0 0 −ikye2iαc(xn−1) G∗(xn−1) −i ∆c(xn−1) 2δx 0 0 −i ∆c(xn) 2δx 0 0 G(xn) ikye−2iαc(xn) i ∆c(xn+1) 2δx 0 0 i −ikye2iαc(xn) ∆c(xn−1) 2δx G∗(xn) 0 −i ∆c(xn+1) 2δx 0 0 −i ∆c(xn) 2δx 0 G(xn+1) ikye−2iαc(xn+1) 0 0 0 i ∆c(xn) 2δx −ikye2iαc(xn+1) G∗(xn+1) The above Hamiltonian matrix has the same form as Eq. (1). 9 (A24)  . 10 (Color online) Scaled electron energy spectrum ε(k)/(vF kSAW ) in the absence of electron traps as a function of the FIG. 1: normalized SAW amplitude A/(vF kSAW ) for semiconducting nanoribbons subjected to an acoustically induced SAW potential. In this figure, we choose ∆/(vF kSAW ) = 1.0 (upper panel) and 1.5 (lower panel). Only the eigen-spectra arising from the two lowest dispersion curves in the absence of a SAW are displayed. Higher subbands contribute significantly at larger SAW amplitude. The lighter shaded regions arise from the lowest energy dispersion curves, whereas the darker shaded regions are associated with the energy dispersions of the next subband. 11 (Color online) Scaled electron energy spectrum ε(k)/(vF kSAW ) in the presence of electron traps as a function FIG. 2: of the normalized SAW amplitude A/(vF kSAW ) for semiconducting nanoribbons subjected to an acoustically induced SAW potential. In this figure, we chose the energy gap ∆/(vF kSAW ) = 1.0 and the trap weight V0 = Vtrap/(vF kSAW ) = 1.0 for all four panels. As in Fig. 1, only the eigen-spectra due to the two lowest dispersion curves in the absence of the SAW are shown. Here, the position x0kSAW /(2π) of electron traps, moving with the SAW potential, is located at: 1/3 (upper-left panel), 1/5 (lower-left panel), 1/7 (upper-right panel), and 1/11 (lower-right panel). Two induced trap-state energy levels within the energy gap are indicated by arrows in the lower-left panel for emphasis. 12 (Color online) Minigap spectrum ε(k)/(vF kc), induced by a corrugation potential, as a function of the normalized FIG. 3: modulation amplitude Ckc. In this figure, the two plots correspond to the two lowest values of the quantized transverse wave number k(m) . The graph at the top comes from the lowest quantized energy subband, whereas the graph at the bottom is due to the first-excited subband. y 

1210.5272

2

1210

Distilling one, two and entangled pairs of photons from a quantum dot with cavity QED effects and spectral filtering

A quantum dot can be used as a source of one- and two-photon states and of polarisation entangled photon pairs. The emission of such states is investigated from the point of view of frequency-resolved two-photon correlations. These follow from a spectral filtering of the dot emission, which can be achieved either by using a cavity or by placing a number of interference filters before the detectors. The combination of these various options is used to iteratively refine the emission in a "distillation" process and arrive at highly correlated states with a high purity. So-called "leapfrog processes" where the system undergoes a direct transition from the biexciton state to the ground state by direct emission of two photons, are shown to be central to the quantum features of such sources. Optimum configurations are singled out in a global theoretical picture that unifies the various regimes of operation.

cond-mat.mes-hall

cond-mat

Distilling one, two and entangled pairs of photons from a quantum dot with cavity QED effects and spectral filtering Elena del Valle Physik Department, Technische Universitat Munchen, James-Franck-Strasse, 85748 Garching, Germany E-mail: [email protected] Abstract. A quantum dot can be used as a source of one- and two-photon states and of polarisation entangled photon pairs. The emission of such states is investigated from the point of view of frequency-resolved two-photon correlations. These follow from a spectral filtering of the dot emission, which can be achieved either by using a cavity or by placing a number of interference filters before the detectors. The combination of these various options is used to iteratively refine the emission in a "distillation" process and arrive at highly correlated states with a high purity. So-called "leapfrog processes" where the system undergoes a direct transition from the biexciton state to the ground state by direct emission of two photons, are shown to be central to the quantum features of such sources. Optimum configurations are singled out in a global theoretical picture that unifies the various regimes of operation. PACS numbers: 78.67.Hc, 42.50.Ct, 03.67.Bg, 03.65.Yz 2 1 0 2 v o N 9 1 ] l l a h - s e m . t a m - d n o c [ 2 v 2 7 2 5 . 0 1 2 1 : v i X r a Distilling one, two and entangled pairs of photons from a quantum dot 2 1. Introduction Quantum dots have proven in the recent years to be excellent platforms for single photon sources [1, 2, 3], spin manipulation and coherent control at the exciton [4, 5, 6, 7, 8], or biexciton level [9, 10, 11, 12, 13, 14], or for entangled photon-pair generation [15, 16, 17]. The achievement of strong coupling between a single quantum dot and a cavity mode [18, 19, 20] impulsed even further these possibilities by increasing their efficiency and output-collectability [21, 22] to the point of reaching new regimes such as microlasing [23] or two-photon emission [24]. The cavity mode can also serve as a coupler between two distant dots [25, 26]. Cavity QED effects are thus a powerful resource to exploit the quantum features of a quantum dot [27, 28, 29, 30]. There is an alternative way to control, engineer and purify the emission of a quantum emitter which relies on extrinsic components at the macroscopic level, in contrast with the intrinsic approach at the microscopic level that supplements the quantum dot with a built-in microcavity. Namely, one can use spectral filtering. This approach is "extrinsic" in the sense that the filters are placed between the system which emits the light and the observer who detects it. As such, it belongs more properly with the detection part. The filter can in fact be modelling the finite resolution of a detector that is sensible only within a given frequency window. In this text, to keep the discussion as simple as possible, we will assume perfect detectors and describe the detection process through spectral filters (this means that the detector has a better resolution than the one imparted by the filter in front of it). Each filter is theoretically fully specified by its frequency of detection and linewidth. We will assume Lorentzian spectral shapes, which corresponds to the case of most interference filters. Commonly used spectral filter of this type are the thin-film filters and Fabry-Perot interferometers (in the figures, we will sketch such filters as dichroic bandpass filters, with different colours to imply different frequencies.) Since they rely on interference effects, they are basically cavities in weak coupling. This reinforces the main theme of this text which is to investigate cavity effects on a quantum emitter. The cavity itself can be, again, intrinsically part of the heterostructure itself, all packaged on-chip, or extrinsically due to the external filters. Combining these features, such as, filtering the emission of a cavity-QED system, we arrive to the notion of "distillation" where the emitter sees its output increasingly filtered by consecutive sequences to finally deliver a highly correlated quantum state of high purity. While the idea is general and could be applied to a wealth of quantum emitters, we concentrate here on a single quantum dot, sketched as a little radiating pyramid in Fig. 1. Theoretically, it will be described as a combination of two two-level systems, representing two excitons of opposite spins. Such a system will be used for the generation of photons one by one or in pairs, with various types of quantum correlations. The four- level system formed by the two possible excitonic states (corresponding to orthogonal polarisations) and the doubly occupied state, the biexciton, is ideal to switch from one type of device to the other by simply selecting and enhancing the emission at the different Distilling one, two and entangled pairs of photons from a quantum dot 3 Figure 1. Sketch of the various schemes investigated in this text to study two- photon correlations from the light emitted by a quantum dot (left). The various detection configurations are, from left to right: 1, the direct emission of the dot (gp2qrσHs), 2, the enhanced and filtered emission of the dot by a cavity mode (gp2qras); p2q Γ rσHspω; ωq), 4, the two-photon spectroscopy 3, the filtered emission of the dot (g p2q Γ rσHspω1; ω2q), 5, the filtered emission of the dot-in-a-cavity emission of the dot (g p2q p2q Γ raspω1; ω2q) Γ raspω; ωq), 6, the two-photon spectroscopy of the dot-in-a-cavity (g (g and 7, the tomographic reconstruction of the density matrix for the polarisation- entangled photon pairs, θΓpω1, ω2; ω3, ω4q. intrinsic resonances [31, 32, 33]. Figure 1 gives a summary of the various filtering and detection schemes that will be applied, with the "naked" dot on the left. Its emission will be considered both from within or without a cavity, with various numbers of filters interceding. We will assume the microcavity both in the weak and strong-coupling regimes. The latter system has been extensively studied [31, 32, 33, 34, 35] and will be revisited here in the light of its spectral filtering [36] and distillation. The rest of the text is organised as follows. In Sec. 2, we present the system and its basic properties and we introduce the two-photon spectrum which is the counterpart at the two-photon level of the photoluminescence spectrum at the single-photon one. In Sec. 3 we provide the first application of two-photon distillation, achieved via a cavity mode weakly coupled to the dot transitions or through spectral filtering. In Sec. 4, we compare the cavity filtering in the weak coupling regime with the enhancement of the emission in the strong coupling regime. In Sec. 4.1, we go one step further in the distillation of the two-photon emission and filter it from the cavity emission as well. In Sec. 5, we consider one of the most popular applications of the biexciton structure, the generation of polarisation entangled photon pairs. In Sec. 6 we draw some conclusions. 2. Two-photon spectrum from the quantum dot direct emission The system under analysis consists of a quantum dot that can host up to two excitons with opposite spins. The corresponding orthogonal basis of linear polarisations, Horizontal (H) and Vertical (V), reads tGy ,Hy ,Vy ,Byu, where G stands for the ground state, H and V for the single exciton states and B for the biexciton or doubly occupied state. The four level scheme that they form is depicted in Fig. 2(a). The Hamiltonian of the system reads ( " 1): HyxH ` VyxV ` ωB ByxB , Hdot " ¯ (1) ´ ωX ` δ 2 ¯ ´ ωX ´ δ 2 where we allow the excitonic states to be split by a small energy δ, as is typically the case experimentally, by the so-called fine structure splitting [24]. The biexciton at FilterFilterFilterFilterFilterFilterDistilling one, two and entangled pairs of photons from a quantum dot 4 Figure 2. (a) Level scheme of the quantum dot investigated, modelled as a system able to accommodate two excitons Vy and Hy in the linear polarisation basis, with an energy splitting δ between them and which, when present jointly, form a biexciton By with binding energy χ. The excitons decay radiatively at a rate γ. When placed in a cavity (with linear polarisation H), two extra decay channels are opened for the H- polarisation: through the one-photon cascade at rate κ1P, and the two-photon emission at rate κ2P. In the sketch, the cavity is placed at the two-photon resonance ωa " ωB{2. (b) PL spectra from the quantum dot system in H-polarisation (orange) consisting of two peaks at ωBH and ωH, and in V-polarisation (green) consisting of two peaks at ωBV and ωV. (c) PL spectrum when the dot is placed inside a cavity, both for the case of strong (g " κ, solid line) and weak (g Ñ 0, dotted line) coupling. A new peak appears at the centre from the two-photon emission. Parameters: P " γ, χ " 100γ, δ " 20γ, κ " 5γ. We consider ωX Ñ 0 as the reference frequency. ωB " 2ωX ´ χ is far detuned from twice the exciton thanks to the binding energy, χ, typically the largest parameter in the system. We include the dot losses, at a rate γ, and an incoherent continuous excitation (off-resonant driving of the wetting layer), at a rate P , in both polarisations x "H, V, in a master equation: ` P 2 ´ LxyxG`LByxx Btρ " irρ, Hdots` ´ LGyxx`LxyxB ¯ ÿ " γ 2 x"H,V ¯ı pρq , (2) where Lcpρq " 2cρc: ´ c:cρ ´ ρc:c is in the Lindblad form. We assume in what follows an experimentally relevant situation, χ " 100γ, δ " 20γ, and study the steady state under P " γ, in which case all levels are equally populated (the populations read ρG " γ2{pP `γq2, ρH " ρV " P γ{pP `γq2 and ρB " P 2{pP `γq2). The photoluminescence spectra of the system, Spωq, are shown in Fig. 2(b) for the H- and V-polarised emission with orange and green lines respectively. The four peaks are well separated thanks to the binding energy and fine structure splitting, corresponding to the four transitions depicted in panel (a) with the same colour code: ωBH " ωB ´ ωH " ωX ´ χ ´ δ{2 " ´110γ , ωH " ωX ` δ{2 " 10γ , ωBV " ωB ´ ωV " ωX ´ χ ` δ{2 " ´90γ , ωV " ωX ´ δ{2 " ´10γ , (3) (4) (5) (6) ----~~~~(a)Energy(b)-150-100-50050(c)Distilling one, two and entangled pairs of photons from a quantum dot 5 with ωX Ñ 0 as the reference and FWHM γBH " γBV " 3γ ` P , γH " γV " 3P ` γ. We concentrate on the H-mode emission because it has the largest peak separation and allows for the best filtering but all results apply similarly to the V polarisation. The emission structure at the single-photon level is very simple: two Lorentzian peaks are observed corresponding to the upper and lower transitions, " ı Spωq " 1 π ρB pγBH{2q2 ` pω ´ ωBHq2 ` ρH γBH{2 γH{2 pγH{2q2 ` pω ´ ωHq2 . (7) The second-order coherence function of the H-emission in the steady state reads: gp2qrσHspτq " xσ` Hp0qσ` HpτqσHpτqσHp0qy{xσ` HσHy2 where the H-photon destruction operator is defined as: σH " s1 ` s2 , with s1 " HyxB and s2 " GyxH . (8) (9) In Eq. (8), we have specified the channel of emission in square brackets since this will be an important attribute in the rest of the text. The quantum-dot described with the spin degree of freedom, exhibits uncorrelated statistics in the linear polarisation: gp2qrσHspτq " 1 . (10) One recovers the expected antibunching of a two-level system [37], by turning to the intrinsic two-level systems composing the quantum dot, namely, the spin-up and spin- down excitons: gp2qrσÒsp0q " gp2qrσÓsp0q " 0. The Pauli exclusion principle that Ö " 0, one has holds for the spin σÖ, breaks in the linear polarisation, i.e., while σ2 H " GyxB ‰ 0. We can find a simple explanation for this if we write the total σ2 correlations in terms of the four contributions (different from zero): xs` i p0qs` j pτqsjpτqsip0qy , (11) Hp0qσ` xσ` HpτqσHpτqσHp0qy " which are given by (τ ě 0): ÿ i,j"1,2 gp2qrs1; s1spτq " p1 ´ e´pγ`Pqτqp1 ` γ e´pγ`Pqτq , P e´pγ`Pqτq , gp2qrs2; s2spτq " p1 ´ e´pγ`Pqτqp1 ` P γ gp2qrs1; s2spτq " p1 ´ e´pγ`Pqτq ` e´pγ`Pqτp2 ` γ P gp2qrs2; s1spτq " p1 ´ e´pγ`Pqτq2 , in their normalised form, gp2qrsi; sjspτq " xs` j sjyq. As shown in Fig. 3(c) and (d) with pale grey lines, all these functions are antibunched except for gp2qrs1; s2spτq, which corresponds to the natural order in the H-cascaded emission of two photons and is, consequently, bunched. It compensates fully the other three terms, leading to total correlations of 1 for all τ . j pτqsjpτqsip0qy{pxs` i siyxs` i p0qs` ` P γ (12) q , Fig. 2 provides a clear picture o physical grounds of how such a system can be used as a quantum emitter but it lacks even a qualitative picture of how quantum correlations are distributed. Fig. 2(b) merely shows where the system emits light but Distilling one, two and entangled pairs of photons from a quantum dot 6 Γ pω1; ω2, τq [38]. This is the extension of the Glauber second p2q nothing on how correlated is this emission. All these crucial features are revealed in the two-photon spectrum g order correlation function -- which quantifies the correlations between photons in their arrival times -- to frequency. By specifying both the energy and time of arrivals of the photons, one provides an essentially complete description of the system. Γ denotes the linewidth of the frequency window of the filter over which this joint characterisation is obtained. The corresponding time-resolution is given by its inverse, 1{Γ. It is a necessary variable without which nonsensical or trivial results are obtained. The two-photon spectrum unravels a large class of processes hidden in single-photon spectroscopy and can be expected to become a standard tool to characterise and engineer quantum sources. The computation of such a quantity has remained a challenging task for theorists since the mid-eighties [39, 40, 41], until a recent workaround [36] has been found which allows an exact numerical computation. It will be applied here for the first time to the case of biexciton emission and used to understand, characterize and enhance various processes useful for its quantum emission. A detailed discussion on even more fundamental emitters is given in Ref. [38]. Experimentally, the two-photon spectrum corresponds to the usual Hanbury Brown -- Twiss setup to measure second-order correlations through photon counting, with filters or monochromators being placed in front of the detectors to select two, in general different, frequency windows. The technique has been amply used in the laboratory [42, 43, 44, 45, 15, 16, 46, 22, 47, 48, 49] but lacking hitherto a general theoretical description, the global picture provided here has not yet been achieved experimentally. Note finally that when considering correlations between equal frequencies, ω1 " ω2, the result is equivalent to placing a single filter before measuring the correlations of the outcoming photon stream [36]. Figure 3(a) shows the much richer landscape provided by the frequency resolved Γ rσHspω1; ω2, τq, p2q second-order coherence function, that is, the two-photon spectrum g in contrast with the one-photon spectrum Spωq and the colour-blind second order correlations gp2qrσHs, Eq. (10). It is shown at zero delay (τ " 0) with the sensor linewidths taken to filter the full peaks (Γ " 5γ). In such a case, one can see well defined regions of enhancement and suppression of the correlations: subpoissonian values (ă 1) are coloured in blue, Poissonian (" 1) in white and superpoissonian (ą 1) in red. This figure is the backbone of this text. We now discuss in turns these different regions where the quantum-dot operates as a quantum source with different properties. 3. Distilling single photons and photon pairs Single-photon source: When the filters are tuned to the same frequency [diagonal black line in the pω1; ω2q space in Fig. 3(a)], there is a systematic enhancement of the bunching as compared to the surrounding regions due to the two possibilities of detecting identical photons [38]. Despite this feature, that is independent of the system dynamics, when both frequencies coincide with one of the dot transitions, pωH; ωHq or Distilling one, two and entangled pairs of photons from a quantum dot 7 Figure 3. (a) Two-photon spectrum of a quantum dot (with a biexciton structure), with Γ " 5γ. The density plot shows how the correlations between photons are distributed depending on their frequency of emission, from subpoissonian ΓrσHspω1; ω2q ă 1, in blue) to superpoissonian (ą 1, in red) passing by Poissonian (g2 (" 1, in white). The blue "butterflies" on the diagonal are typical of two-level systems. The antidiagonal corresponds to leapfrog processes with direct emission of two photons through an intermediate virtual state. (b) Cuts from the density plot along the diagonal ω2 " ω1 (in black) and the antidiagonal ω2 " ´χ ´ ω1 (in red). The diagonal also corresponds to applying a single filter. (c -- e) Comparison of the τ -dynamics for three cases of interest: (c) bunching at the heart of the two blue butterflies on the diagonal at pωBH; ωBHq (solid dark blue) and pωH; ωHq (dashed clear blue), (d) cross-correlation of the peaks at pωBH; ωHq (solid) and vice-versa, pωH; ωBHq (dotted), showing the typical cascade behaviour and (e) the strong bunching at pωB{2; ωb{2q where two- photon emission is optimum. In (c) and (d) we also plot with pale grey lines the second order correlations of the effective corresponding operators, Eqs. (12). All scales are logarithmic. Parameters: P " γ, χ " 100γ, δ " 20γ. pωBH; ωBHq, there is a dip in the correlations. This is more clearly shown in the cut at equal frequencies, the black line in Fig. 3(b). The blue butterfly shape that is observed in the two-photon spectrum locally around each of the dot transitions is characteristic of an isolated two-level system [38]. This zero delay information is complemented by the antibunched τ dynamics, shown in Fig. 3(c). The two dot resonances, upper and lower, coincide in this case due to the symmetric conditions P " γ but they are typically different (the upper level being less antibunched at low pump). Filtering and detection makes impossible to have a perfect antibunching, getting closest to the ideal correlations -150-100-50050-150-100-50050(a)Filter-150-100-50050(b)0.51.05.010.050.0100.01.000.303.001.000.500.300.70-3-2-10123110100(c)(d)(e)Distilling one, two and entangled pairs of photons from a quantum dot Γ rσHspωH; ωH, τq « gp2qrs1; s1spτq from Eqs. (12), at around Γ « χ{2. At this point, the p2q g peaks are maximally filtered with still negligible overlapping of the filters. It is possible to derive a useful expression for the filtered correlations at τ " 0, with Γ ď χ{2, in the limit: 8 Γ rσHspωH; ωHq " p2q lim χÑ8 g typically relevant in experiments. 2pP ` γq2p3P ` γ ` Γqp2γ ` Γq γpP ` γ ` Γqp2P ` 2γ ` Γqp3P ` γ ` 3Γq , (13) All this shows that one can recover or optimise the quantum features of a single photon source (antibunching) in a system whose total emission is uncorrelated, by frequency filtering photons from individual transitions. Consequently, the system should exhibit two-photon blockade when probed by a resonant laser at frequency ωL in resonance with the lower transition, ωL " ωH. The antibunched emission of each of the four filtered peaks of the spectrum, has been observed experimentally [15]. Cascaded two-photon emission: When the filtering frequencies match both the upper and lower quantum dot transitions, i.e., pωBH; ωHq, the correlations are close to one at zero delay, like for the total emission gp2qrσHs (which is exactly one). However, although the latter is uncorrelated, since it remains equal to one at all τ , the filtered cascade emission is not uncorrelated since it is close to unity only at zero delay and precisely because of strong correlations that are, however, of an opposite nature at positive and negative delays, i.e., showing enhancement for τ ą 0 when photons are detected in the natural order that they are emitted, and suppression for τ ă 0 when the order is the opposite. This is depicted in Fig. 3(d) where the solid line corresponds to pωBH; ωHq and the dotted line to exchanging the filters, pωH; ωBHq. As this also corresponds to detecting the photons in the opposite time order, the two curves are exact mirror image of each other. The identification of the upper and lower transition photons with frequency-blind operators [gp2qrs1; s2spτq in Eqs. (12)] provides crossed correlations different to our exact and general frequency resolved functions, specially at τ " 0, as shown in Fig. 3(d), where there is a discontinuity for the approximated functions. The frequency resolved functions have the typical smooth cascade shape that been observed experimentally [45, 15]. The dynamics at large τ , converges to the approximated functions only for Γ « χ{2. simultaneous two-photon emission, the Simultaneous two-photon emission: For strongest feature lies on the antidiagonal (red line) in Fig. 3(a), which is also shown as the solid red line in Fig. 3(b). The strong bunching observed here, when both frequencies are far from the system resonances ωBH and ωH, corresponds to a two-photon deexcitation directly from the biexciton to the ground state without passing by an intermediate real state. This two-photon emission from a Hamiltonian, Eq. (1), that does not have a term to describe such a process is made possible via a virtual state that arises in the quantum dynamics and that can be revealed by the spectral filtering. As the intermediate virtual state has no fixed energy and only the total energy ω1 ` ω2 " ωB needs be conserved, Distilling one, two and entangled pairs of photons from a quantum dot 9 the simultaneous two-photon emission is observed on the entire antidiagonal (except, again, when touching a resonance, in which case the cascade through real states takes over). We call such processes "leapfrog" as they jump over the intermediate excitonic state [38]. The largest bunching is found at the central point, ω1 " ω2 " ωB{2 " ´χ{2, and at the far-ends ω1 ! ωBH and ω1 " ωH. Among them, the optimal point is that where also the intensity of the two-photon emission is strong. The frequency resolved Mandel Q parameter takes into account both correlations and the strength of the filtered signal [50]: " a SΓpω1qSΓpω2q g ı Γ pω1; ω2q ´ 1 p2q QΓpω1; ω2q " (14) (where also for the single-photon spectra, the detection linewidth, Γ, is taken into account [51]). As expected, QΓrσHspω1; ω2q becomes negligible at very large frequencies, far from the resonances of the system, and reaches its maximum at the two-photon resonance, pωB{2; ωB{2q (not shown). This latter configuration is therefore the best candidate for the simultaneous and, additionally, indistinguishable, emission of two photons. The bunching is shown in Fig. 3(e). The small and fast oscillations are due to the effect of one-photon dynamics with the real states but are unimportant for our discussion and would be difficult to resolve experimentally. While the bunching in such a configuration has not yet been observed experimentally, recently, Ota et al. successfully filtered the two-photon emission from the biexciton with a cavity mode [24], which corroborates the above discussion. 4. Filtering and enhancing photon-pair emission from the quantum dot via a cavity mode Large two-photon correlations are the starting point to create a two-photon emission device. When they have been identified, the next step is to increase their efficiency by enhancing the emission at the right operational frequency. The typical way is to Purcell enhance the emission through a cavity mode with the adequate polarisation and strongly coupled with the dot transitions (at resonance). Theoretically this amounts to adding to the master equation (2) a Hamiltonian part, that accounts for the free cavity mode (ωa) and the coupling to the dot (with strength g), Hcav " ωaa:a ` gpa:σH ` aσ: Hq , (15) along with a Lindblad term κ 2Lapρq, that accounts for the cavity decay (at rate κ). By placing the cavity mode at the two-photon resonance, ωa " ωB{2, the virtual leapfrog process becomes real as it finds a real intermediate state in the form of a cavity photon. The deexcitation of the biexciton to ground state is thereby enhanced at a rate κ2PR « p4g2{χq2{κ, producing the emission of two simultaneous and indistinguishable cavity photons at this frequency [31, 32, 33]. There is as well some probability that the cavity mediated deexcitation occurs in two steps, through two different cavity photons at frequencies ωBH and ωH, at the same rate κ1PR « 4g2κ{χ2. The two alternative paths Distilling one, two and entangled pairs of photons from a quantum dot 10 are schematically depicted with curly blue arrows in Fig. 2(a). The cavity being far from resonance with the dot transitions, the ratio of two- versus one-photon emission can be controlled by an appropriate choice of parameters [32]. We set g " κ " 5γ, to be in strong coupling regime and have κ2P " 0.2γ ą κ1P " 0.05γ, but with a coupling weak enough for the system to emit cavity photons efficiently. The cavity parameters are such that κ ą 2P and the pump does not disrupt the two-photon dynamics [33]. The two-photon emission indeed dominates over the one-photon emission as seen in Fig. 2(c), where the cavity spectrum is plotted with a solid line: the central peak, corresponding to the simultaneous two-photon emission, is more intense than the side peaks produced by single photons. A better cavity (smaller κ) does not emit the biexciton photons right away outside of the system, but spoils the original (leapfrog) correlations and leads to smaller correlations in the cavity emission gp2qras " xa:a:aay{xa:ay2 Ñ 2. This is shown in Fig. 4(a) with a blue solid line. Our previous choice κ " 5γ, is close to that which maximises bunching (vertical line in Fig. 4(a)). A weak coupling due to small coupling strength (g Ñ 0), plotted with a blue dotted line, recovers the case of a filter, discussed in the previous Section. The cavity spectrum in this case, plotted with a blue dotted line in Fig. 2(c), is no longer dominated by the two-photon emission. Regardless of the coupling strength g, the system goes into weak-coupling at large enough κ, so both blue lines, solid and dotted, converge to the same curve at κ Ñ 8. The cavity then filters the whole dot emission and recovers the total dot correlations for the H-mode, gp2qras Ñ gp2qrσHs " 1. Note that while the bunching in gp2qras is better in weak-coupling or with a filter, this is at the price of decreasing the enhancement of the emission and, therefore, the efficiency of the quantum device, as the total Mandel Qras parameter shows in Fig. 4(b). 4.1. Distilling the two-photon emission from the cavity field Γ raspω1; ω2q for a cavity with κ " 5γ and ωa " ωB{2. p2q In view of the preceding results, we now consider the possibility to further enhance the two-photon emission by filtering the cavity photons from the central peak in Fig. 2(c). In this That is, we study g way, first, the cavity acts as a filter, extracting the leapfrog emission where it is most correlated, but also enhances specifically the two-photon emission and, second, the filtering of the cavity emission selects only those photons that truly come in pairs. Such a chain is alike to a "distillation" process where the quantum emission is successively refined. The results are plotted with red lines in Fig. 4, for the case κ " 5g pinpointed by circles on the blue lines. The filtered cavity emission is indeed generally more strongly correlated at the two-photon resonance than the unfiltered total cavity emission, plotted Γ raspωB{2; ωB{2q ě gp2qras. This is so for all Γ ą κ for the cavity in weak- p2q in blue: g coupling, where the distillation always enhances the correlations. In strong coupling, the filter must strongly overlap with the peak (Γ " κ). This is because the side peaks are prominent in weak-coupling (κ2P ă κ1P), and the filtering efficiently suppress their Distilling one, two and entangled pairs of photons from a quantum dot 11 Figure 4. (a) Second order correlations of a cavity mode embedding a quantum dot in weak (dotted) and strong (solid) coupling. Both the full, colour-blind, cavity emission (in blue) gp2qras and the frequency-resolved correlations at the two-photon resonance p2q Γ raspωB{2; ωB{2q are considered, with the former relating to the bottom axis (in red) g (the cavity linewidth κ) and the latter to the upper axis (the detector linewidth Γ) at κ " 5γ (indicated by a circle). (b) Mandel Q parameter in the same configuration as in (a), supplementing the information of the previous panel with the intensity of the emission. (c) Same as (a) but now as a function of the cavity frequency ωa (bottom axis) for the colour-blind correlations (in blue) and of the detection frequency ω (upper axis) for the frequency-resolved correlations. Parameters: (a) the strong coupling is for g " 5γ and the weak coupling is in the limit of vanishing coupling g Ñ 0 where the cavity is fully equivalent to a filter. Parameters: P " γ, χ " 100γ, δ " 20γ. detrimental effect, whereas in strong-coupling, two-photon correlations are already close to maximum thanks to the dominant central peak as seen in Fig. 2(c), and it is therefore important for the filter to strongly overlap with it. In all cases, at large enough Γ, the Γ raspω1; ω2q " gp2qras. In p2q full cavity correlations are recovered as expected: Fig. 4(a), this means that the red lines converge to the value projected by the circle on the blue lines at κ " 5γ. Here, again, filtering enhances correlations but reduces the number of counts, as shown in Fig. 4(b). limΓÑ8 g In Fig. 4(c), we do the same analysis as in Fig. 3(b) where there was no distillation. We address the same cases but now as a function of frequency, fixing the cavity decay rate κ " 5γ and the filtering linewidth at Γ " 10γ. In blue, we consider the cavity QED case in weak- and strong-coupling, without filtering. Since the weak-coupling limit is identical to the single filter case, note that the blue dotted line in Fig. 4(c) is identical to (a)(b)(c)scwcscFilterwcFilter0.51510501005000.51510501005000.11101001030.111010010310311010050020.11101001030.1110100110100500210310-1210-1010-810-610-40.0110.11101001030.111010010-1210-1010-810-610-40.011Distilling one, two and entangled pairs of photons from a quantum dot 12 p2q Γ raspω1; ω2q for a quantum dot in a cavity in the Figure 5. Two-photon spectra g (a) strong, g " κ, and (b) weak coupling regime, when the cavity is at the two-photon resonance, ωa " ωB{2. The bunching regions are strengthened by the cavity (the total, colour-blind correlations are gp2qras « 21). A horizontal and vertical structure also emerges at the two-photon resonance, betraying the emergence of real states. These are stronger the stronger the coupling. The same logarithmic scale as Fig. 3(a) applies, so all three figures can be compared directly. Parameters: κ " 5γ, P " γ, χ " 100γ, δ " 20γ and Γ " 10κ. the black solid line in Fig. 3(b). Off-resonance, the cavity acts as a simple filter due to the reduction of the effective coupling. The stronger coupling to the cavity has an effect only when involving the real states, where it spoils the correlations, less bunched at the two-photon resonance ωa " ωB{2, and less antibunched at the one-photon resonances, ωa " ωBH or ωa " ωH. This shows again that useful quantum correlations are obtained in a system where quantum processes are Purcell enhanced and quickly transferred outside, rather than stored and Rabi-cycled over within the cavity. The same is true for the red line, further filtering the output. Finally, comparing solid lines together, we see again that there is little if anything to be gained by filtering in strong-coupling, whereas in weak-coupling, the enhancement is considerable. As a summary, the filtering of the weakly coupled cavity (red dotted), provides the strongest correlations (at the cost of the available intensity), corresponding to distilling the photon pairs out of the original dot spectrum without any additional enhancement. In Fig. 5, we show the full two-photon spectra for a quantum dot in a cavity, in both (a) strong- and (b) weak-coupling. The same colour code and logarithmic scale is used as in Fig. 3(a), for comparison. The antibunching regions on the diagonal and the bunching ones on the antidiagonal are qualitatively similar to the filtered dot emission, but antibunching is milder and less extended while correlations are, respectively, weaker -150-100-50050-150-100-50050Filter-150-100-50050(a)(b)wcFilterscFilterDistilling one, two and entangled pairs of photons from a quantum dot 13 (stronger) at the central point due to the saturated (efficient) distillation in strong (weak) coupling. Another striking feature added by the cavity is the appearance of an additional pattern of horizontal and vertical lines at ωa " ωB{2. While diagonal and anti-diagonal features correspond to virtual processes, horizontal and vertical stem from real processes, that pin the correlations at their own frequency. Therefore, the new features are a further illustration that the two-photon emission becomes a real resonance of the cavity-dot system, in contrast with Fig. 3(a) where it was virtual. The effect is more pronounced in the strong-coupling regime since this is the case where the new state is better defined. Another qualitative difference between Fig. 5 and Fig. 3(a) is in the regions surrounding the cascade configuration, pωBH; ωHq, which has changed shape around two antibunching spots. This is due to the fact that the single-photon cascade is much less likely to happen through the cavity mode than the direct deexcitation of the dot, as κ1P " 0.05γ ! γ. Even if the first photon from the biexciton emission decays through the cavity, the second will most likely not. This new two antibunching spots are slightly pushed to the left of the red line by the leapfrog bunching line and the presence of the V-polarised resonances. 5. Distilling entangled photon pairs One of the most sophisticated applications of the biexciton structure in a quantum dot is as a source of polarisation entangled photon pairs [52, 15, 16, 53, 54, 55, 56, 57, 58, 21, 59, 60, 35, 61]. Without the fine-structure splitting, δ " 0, the two possible two-photon deexcitation paths are indistinguishable except for their polarisation degree of freedom (H or V), producing equal frequency photon pairs pωBX; ωXq, with ωBX " ωB´ ωX. This results in the polarisation entangled state: ` HpωBXq, HpωXqy ` e´iφ VpωBXq, VpωXqy ´ The splitting δ provides "which path" information [62] that spoils indistinguishability, producing a state entangled in both frequency and polarisation [34]: HpωBHq, HpωHqy ` e´iφ1 VpωBVq, VpωVqy ψy " 1? 2 (17) (16) . ¯ . ψ1y " 1? 2 Although these doubly entangled states are useful for some quantum applications [63, 64], it is typically desirable to erase the frequency information and recover polarisation- only entangled pairs. Among other solutions, such as canceling the built-in splitting externally [16, 65], filtering has been implemented with ω1 " ωBX and ω2 " ωX to make the pairs identical in frequencies again [34, 15, 54], at the cost of increasing the randomness of the source (making it less "on-demand"). Recently, the cavity filtering of the polarisation entangled photon pairs with ωa " ωB{2 has been proposed by Schumacher et al. [35], taking advantage of the additional two-photon enhancement [32]. Let us revisit these effects in the light of the previous results. The properties of the output photons can be obtained from the two-photon state density matrix, θpτq, reconstructed in the basis tH1, H2y ,H1, V4y ,V3, H2y ,V3, V4yu, Distilling one, two and entangled pairs of photons from a quantum dot 14 denoting by xiy with x "H or V and i " 1, 2, 3, 4, the state xpωiqy. The frequencies ωi are, in general, different. The second photon is detected with a delay τ with respect to the first one (detected in the steady state). Let us express this matrix in terms of frequency resolved correlators, as is typically done in the literature [53, 55]. However, in contrast to previous approaches, we do not identify the photons with the transition from which they may come from (using the dot operators Gyxx, xyxB with, again, x standing for either H or V) but with their measurable properties, that is, polarisation, frequency and time of detection (for a given filter window). This is a more accurate description of the experimental situation where a given photon can come from any dot transition and any transition can produce photons at any frequency and time with some probability. We describe the experiments by considering four different filters, that is, including all degrees of freedom of the emitted photons in the description. Each detected filtered photon corresponds to the application of the filter operator ςj with j "H1, H2, V3, V4 corresponding to its coupling to the H or V dot transitions with ωi frequencies. Γpτq (the prime refers to the lack of normalisation) Then, the two-photon matrix θ1 corresponding to a tomographic measurement is theoretically modelled as: xnH1p0qnH2pτqy h.c. h.c. h.c. Γpτq " θ1 xnH1p0qrς` H2ςV4spτqy xnH1p0qnV4pτqy H1ςV3sp0qnH2pτqy xrς` H1ςV3sp0qrς` xrς` V4ςH2spτqy xnV3p0qnH2pτqy h.c. h.c. ‹‹‹‚ , (18) h.c. H2ςV4spτqy H1ςV3sp0qrς` xrς` xrς` H1ςV3sp0qnV4pτqy xnV3p0qrς` H2ςV4spτqy xnV3p0qnV4pτqy i ςi [we have dropped the frequency dependence in the notation, writing Γpω1, ω2; ω3, ω4, τq]. Since a weakly coupled cavity mode behaves as a where ni " ς` Γpτq instead of θ1 θ1 filter, this tomographic procedure is equivalent to considering the four dot transitions coupled to four different cavity modes with the corresponding polarisations, central frequencies and decay rates [35, 61]. Unlike in other works where for various reasons and particular cases, some of the elements in θ1 here we keep the full matrix with no a priori assumptions since, in general, it may not reduce to a simpler form due to the incoherent pumping, pure dephasing, frequency filtering and fine-structure splitting. Γpτq are set to zero or considered equal, There are essentially two ways to quantify the degree of entanglement from Γpτq. The most straightforward is to consider the τ -dependent the density matrix θ1 matrix directly, which merely requires normalisation at each time τ , yielding θΓpτq " Γpτq{Trrθ1 Γpτqs. The physical interpretation is that of photon pairs emitted with a θ1 delay τ , that is to say, within the time-resolution 1{Γ of the filter or cavity [34, 61]. In particular, the zero-delay matrix, θΓp0q, represents the emission of two simultaneous photons [59, 60]. The second approach is closer to the experimental measurement which In this case, one considers the integrated quantity ΘΓpτmaxq " averages over time. ş τmax 0 15 Γ pτmaxq and CΓpτq, respectively. Distilling one, two and entangled pairs of photons from a quantum dot p Γpτq dτq{N , that averages over all possible emitted pairs from the system [15, 53]. θ1 It is also normalised (by N ), but after integration, so that the two approaches are not directly related to each other and present alternative aspects of the problem, discussed in detail in the following. Without the cutoff delay τmax, the integral diverges due to the continuous pumping. The degree of entanglement of any bipartite system represented by a 4 4 density matrix θ, can be quantified by the concurrence, C, which ranges from 0 (separable states) to 1 (maximally entangled states); [66]. High values of the concurrence require high degrees of purity in the system [67], being, for instance, impossible to extract any entanglement from a maximally mixed state (in which case all the four basis states occur with the same probability). The filtered density matrices, ΘΓpτmaxq and θΓpτq, provide each their own concurrence that we will denote C int We begin by considering the standard cascade configuration by detecting photons at the dot resonances, i.e., ω1 " ωBH, ω2 " ωH, ω3 " ωBV, ω4 " ωV, as sketched in Γ pτmaxq and the density plot the inset of Fig. 6(a). The upper density plot shows C int below shows the time-resolved concurrence CΓpτq, as a function of τmax or τ and δ, for Γ pτmaxq and CΓpτq, are qualitatively different except Γ " 2γ. The two concurrences, C int at τ " τmax " 0 where they are equal by definition. They also have in common that the maximum concurrence is not achieved at zero delay [it is most visible as the darker red area around pγτmax, δ{γq « p0.4, 0q]. This is because this filtering scheme relies on the real-states deexcitation of the dot levels, and thus exhibits the typical delay from the cascade-type dynamics of correlations [see Fig. 3(d)]. The major departure Γ pτmaxq is strongly dependent on δ, while CΓpτq between the two is that the decay of C int is not. With no splitting, at δ " 0, the ideal symmetrical four-level structure efficiently produces the entangled state ψy and the concurrence is maximum both for the integral and the time-resolved forms. The decay time of CΓpτq is the simplest to understand as, when filtering full peaks, it is merely related to the reloading time of the biexciton, of the order of the inverse pumping rate, „ 2{P [61]. The asymmetry due to a nonzero splitting in the four-level system causes an unbalanced dynamics of deexcitation via the H and V polarisations. The entanglement in the form ψy is downgraded to ψ1y. The fact that the concurrence CΓpτq is not affected shows that it accounts for the degree of entanglement in both polarisation and frequency, which is oblivious to the "which Γ pτmaxq is path" information that the last variable provides. On the other hand, C int suppressed by the splitting as fast as τmax ą 2π{δ (this boundary is superimposed to the density plot). That is, as soon as the integration interval is large enough to resolve Γ pτmaxq, therefore, accounts for the degree of entanglement in it. The integrated C int polarisation only, and is destroyed by the "which path" information provided by the different frequencies. The exact mechanism at work to erase the entanglement is shown in the lower row of Fig. 6, which displays the upper right matrix element rθΓpτqs1,4 of the density matrix, that is the most responsible for purporting entanglement. The modulus λ4us, where tλ1, λ2, λ3, λ4u are the eigenvalues ; Its definition reads C " rmaxt0, in decreasing order of the matrix θT θT , with T an antidiagonal matrix with elements t´1, 1, 1,´1u. λ1´? λ2´? λ3´? ? Distilling one, two and entangled pairs of photons from a quantum dot 16 Figure 6. Concurrence for three different schemes of entanglement distillation (as sketched above each column): (a) filtering the four different dot resonances in their cascade through the real states, (b) filtering at two different frequencies ω1 " ωBX and ω2 " ωX, degenerate in both polarisation decay paths, (c) filtering at the same frequency ωB{2, i.e., at the two-photon resonance. The upper panels show the time- Γ pτmaxq as a function of its cutoff τmax and the fine-structure integrated concurrence C int In (a) the line τmax " 2π{δ bounding the region with entanglement is splitting δ. superimposed. The intermediate panels show the instantaneous concurrence CΓpτq as a function of delay and splitting. The lowest panels show the modulus and phase of the off-diagonal element rθΓpτqs1,4 typically used in the literature to quantify entanglement. All these plots show that as far as the degree of entanglement is concerned, the leapfrog emission is the best configuration and is optimum at the two-photon resonance. The concurrence colour code is blue for 0, white for 0.5 and red for 1. In the modulus density plots, the colour code is blue for 0, white for 0.25 and red for values ě 0.5. In the phase density plots, the colour code is black for values approaching ´π and white for π. Parameters: P " γ, χ " 100γ, Γ " 2γ. 0.00.51.01.52.00.00.51.01.52.00.00.51.01.52.0(a)05101520(b)(c)0.00.51.01.52.005101520051015200.01.02.00.01.02.00.00.51.01.52.00.00.51.01.52.00.01.02.00.01.02.00.01.02.00.01.02.0IntegratedTime-resolvedVHFilterHVFilterHVVHFilterHVDistilling one, two and entangled pairs of photons from a quantum dot 17 is shown on the left (with blue meaning 0 and red meaning ě 0.5) and the phase on the right (in black and white). The time-resolved concurrence CΓpτq is very similar to the modulus of rθΓpτqs1,4 which justifies the approximation often-made in the literature CΓpτq " 2prθΓpτqs1,4´rθΓpτqs2,2q with rθΓpτqs2,2 small [53]. Although each photon pair is entangled, it is so in the state ψ1y with a phase, φ1 " ´π ` δτ , that accumulates with τ at a rate δ [34], but this does not matter as far as instantaneous entanglement is concerned, and this is why the splitting does not affect CΓpτq. On the other hand, when integrating over time, the varying phase, that completes a 2π-cycle at intervals 2π{δ, randomises the quantum superposition and results in a classical mixture. This is Γ pτmaxq for τmax ą 2π{δ. And this is how why the splitting does completely destroy C int the system restores the "which path" information: given enough time, if the splitting is large enough, the photons loose their quantum coherence due to the averaging out of the relative phase between them because of their distinguishable frequency. Another possible configuration proposed in the literature [34, 15, 54] is sketched in the inset of Fig. 6(b). Photons are detected at the frequency that lies between the two polarisations, ω1 " ω3 " ωBX, ω2 " ω4 " ωX. For small splittings, the entanglement production scheme still relies on the cascade and real deexcitation of the dot levels. Γ pτmaxq is similar to (a) (decaying with δ) but slightly Therefore, the behaviour of C int improved by the fact that the contributions to the density matrix are more balanced by filtering in-between the levels. For splittings large enough to allow the formation of leapfrog emission in both paths, δ " Γ, there is a striking change of trend and Γ pτmaxq remains finite at longer delays τmax when increasing δ. This is a clear sign that C int the entanglement relies on a different type of emission, namely simultaneous leapfrog photon pairs, rather than a cascade through real states. Accordingly, the intensity is reduced as compared to that obtained at smaller δ or to the non-degenerate cascade in (a), but the bunching is stronger, as evidenced by the two-photon spectrum, and results in a larger degree of entanglement than at δ " 0. A similar result is obtained qualitatively with the time-resolved concurrence, a strong resurgence of entanglement with δ, indicating again very high correlations in this configuration. Note finally that the phase becomes much more constant with increasing δ, resulting in the persistence of the correlation in the time-integrated concurrence. This is because the two-photon emission is through the leapfrog processes. Since the latter, by definition, involve intermediate virtual states that are degenerate in frequency, this is a built-in mechanism to suppress the splitting and not suffer from the "which path" information as when passing through the real states. Finally, a configuration proposed more recently in the literature [35] is the two- photon resonance, with four equal frequencies, ω1 " ω2 " ω3 " ω4 " ωB{2, as sketched in the inset of Fig. 6(c). This provides a two-photon source in both polarisations that can be enhanced via two cavity modes with orthogonal polarisations [32]. Remarkably, in this case, the splitting has almost no effect on the degree of entanglement, that is maximum. Here, the leapfrog mechanism plays at its full extent: the virtual states, on top of being degenerated and thus immune to the splitting, remain always far, and Distilling one, two and entangled pairs of photons from a quantum dot 18 are therefore protected, from the real states. This results in the exactly constant phase (black panel on the lower right end of Fig. 6). As a result, both C int remain large. The only drawback of this mechanism is, being virtual-processes mediated, a comparatively weaker intensity. Γ pτmaxq and CΓpτq We conclude with a more detailed analysis of what appears to be the most suitable scheme to create a robust entanglement, the leapfrog photon-pair emission. The target state is always ψy, Eq. (16), due to the degeneracy in the filtered paths. The concurrence is shown as a function of the first photon frequency, ω1, in Fig. 7 (the second photon has the energy ω2 " ωB ´ ω1 to conserve the total biexciton energy). The time-integrated Γ p1{γq [resp. CΓp0q] is shown in red (resp. blue) (resp. instantaneous) concurrence C int lines, both for a large splitting δ " 20γ in strong tones and for δ " 0 in softer tones. In the first panel, (a), the frequency ω1 of the filters are varied. This figure shows that concurrence is very high (with high state purity) when the filtering frequencies are far from the system resonances: ωBH{BV and ωH{V. These are shown as coloured grid lines to guide the eye. In this case, the real states are not involved and the leapfrog emission is efficient in both polarisations. The concurrence otherwise drops down when ω1 is resonant with any one-photon transition, meaning that photons are then emitted in a cascade in one of the polarisations, rather than simultaneously through a leapfrog process. Moreover, if at least one of the two deexcitation paths is dominated by the real state dynamics, this brings back the problem of "which path" information, that spoils indistinguishability and entanglement. There is only one exception to this general rule, namely, the δ " 0 integrated case (soft red line) which has a local maximum at ωBX, i.e., when touching its resonance in the natural cascade order. This is because the paths are anyway identical and the integration includes the possibility of emitting the second photon with some delay (up to τmax). For the case of δ ‰ 0, if this is large enough, it is still possible to recover identical paths while filtering the leapfrog in the middle points, ω1 " ωBX and ω2 " ωX, to produce entangled pairs. Overall, the optimum configuration is, therefore, indeed at the middle point ω1 " ω2 " ωB{2 (two-photon resonance), where the photons are emitted simultaneously, with a high purity, and entanglement degrees are also identical in frequency by construction. Entanglement is also always larger in the simultaneous concurrence (blue lines) as this is the natural choice to detect the leapfrog emission, which is a fast process. This comes at the price, expectedly, of decreasing the total number of useful counts and increasing the randomness of the source. Note finally that the blue curves are symmetric around ωB{2, but the red ones are not, given that in the integrated case, the order of the photons with different frequencies is relevant. The left-hand side of the plot, ω1 ă ωB{2, corresponds to the natural order of the frequencies in the cascade, ω1 ă ω2. The opposite order, being counter-decay, is detrimental for entanglement. Of all leapfrog configurations, those in panels (b) and (c) of Fig. 6 are therefore the optimum cases to obtain high degrees of entanglement. Let us conclude with a study on their dependence on the filter linewidths, Γ, in Figs. 7(b) and (c). First, we observe that in the limit of small linewidths for the filters, i.e., in the region Γ ă 1{τmax, the Distilling one, two and entangled pairs of photons from a quantum dot 19 Figure 7. Concurrence computed with the instantaneous emission at τ " 0 (blue) or integrated up to τmax " 1{γ (red). (a) Plotted for a fixed filter linewidth Γ " 5γ as a function of the filters frequencies pω1; ωB ´ ω1q. The concurrence is high unless a real state is probed and a cascade through it overtakes the leapfrog emission. The optimum case is the two-photon resonance. (b -- c) Plotted as a function of the filter linewidth Γ for the two cases that maximise entanglement in (a), that is at (b) the degenerate cascade configuration pωBX; ωXq and (c) the biexciton two-photon resonance pωB{2; ωB{2q. The concurrence of the ideal case at δ " 0 is also plotted as a reference with softer solid lines. Parameters: P " γ, χ " 100γ, δ " 20γ. ωX " 0 is set as the reference frequency. simultaneous and time-integrated concurrences should converge to each other, since the time resolution becomes larger than the integration time. Therefore, the integrated emission provides the maximum entanglement when the frequency window is small enough to provide the same results than the simultaneous emission. Decreasing Γ below P (which in this case is also P " γ) results in a time resolution in the filtering larger than the pumping timescale 1{P . Therefore, photons from different pumping cycles start to get mixed with each others. As a result C drops for Γ ă γ in all cases. In the limit Γ ! P , the emission is completely uncorrelated and C " 0. This is an important difference from the cases of pulsed excitation or the spontaneous decay of the system from the biexciton state. In the absence of a steady state, entanglement is maximised with the smallest window, Γ Ñ 0 [15, 54]. Two opposite behaviours are otherwise observed in these figures when increasing the filter linewidth and Γ ą 1{τmax. The simultaneous emission gains in its degree of entanglement whereas the time-integrated In the limit Γ Ñ 8, this disparity is easy to one loses with increasing linewidth. understand. We recover the colour blind result in all filtering configurations, (b) or from 1 corresponding to ψy at τ " 0, to 0 (c), that is, the decay of entanglement: corresponding to a maximally mixed state at large τ . Therefore, CΓp0q Ñ Cp0q " 1 and Γ pτmaxq Ñ C intpτmaxq " 0 (for our particular choice of τmax and P ). C int -150-100-500500.00.20.40.60.81.00.11101001000FilterFilter(a)(b)(c)VHFilterVHHV0.11101001000ConcurrenceDistilling one, two and entangled pairs of photons from a quantum dot 20 The decrease in C int Γ pτmaxq when increasing the filtering window, has been discussed in the literature for the case in Fig. 7(b) [15, 54], and it has been attributed to a gain of "which path" information due to the overlap of the filters with the real excitonic levels. In the lights of our results, when Γ ą δ the real state deexcitation takes over the leapfrog Γ pτmaxq is suppressed indeed due to such a gain of "which path" information, as and C int Γ pτmaxq decreases with Γ discussed previously. However, we find another reason why C int in all cases, based on the leapfrog emission: the region 1{P ă Γ ă δ for case (b) and the region 1{P ă Γ ă χ for case (c). The maximum delay in the emission of the second photon in a leapfrog processes is related to 1{Γ, due to its virtual nature. Therefore, the initial enhancement of entanglement starts to drop at delays τ « 1{Γ, after which the emission of a second photon is uncorrelated to the first one (not belonging to the Γ pτmaxq with same leapfrog pair). For a fixed cutoff τmax, this leads to a reduction of C int Γ. Broader filters have a smaller impact on the case of real state deexcitation [see the zero splitting case in Fig. 7(b), plotted in soft red]: since the system dynamics is slower than the filtering (γ, P ă Γ), the filter merely emits the photons faster and faster after Γ pτmaxq with Γ receiving them from the system. This results in a mild reduction of C int until the detection becomes colour blind and it drops to reach its aforementioned limit of zero. 6. Conclusions In summary, we have characterised the emission of a quantum dot modelled as a system able to accommodate two excitons of different polarisation and bound as a biexciton. Beyond the usual single-photon spectrum (or photoluminescence spectrum), we have presented for the first time the two-photon spectrum of such a system, and discussed the physical processes unravelled by frequency-resolved correlations and how they shed light on various mechanisms useful for quantum information processing. We relied on the recently developed formalism [36] that allows to compute conveniently such correlations resolved both in time and frequency. This describes both the application of external filters before the detection or due to one or many cavity modes in weak-coupling with the emitter. Filters and cavities have their respective advantages, and when combined, can realize a distillation of the emission, by successive filtering that enhance the correlations and purity of the states. We addressed three different regimes of operation depending on the filtering scheme, namely as a source of single photons, a source of two-photon states (both through cascaded photon pairs and simultaneous photon pairs) and as a source of polarisation entangled photon pairs, for which a form of the density matrix that is close to the experimental tomographic procedure was proposed. In particular, so- called leapfrog processes -- where the system undergoes a direct transition from the biexciton to the ground state without passing by the intermediate real states but jumping over them through a virtual state -- have been identified as key, both for two- photon emission and for entangled photon-pair generation. In the latter case, this allows Distilling one, two and entangled pairs of photons from a quantum dot 21 to cancel the notoriously detrimental splitting between the real exciton states that spoils entanglement through a "which path" information, since the intermediate virtual states have no energy constrains and are always perfectly degenerate. Entanglement is long- lived and much more robust against this splitting than when filtering at the system resonances. At the two-photon resonance, degrees of entanglement higher than 80% can be achieved and maintained for a wide range of parameters. Acknowledgements The author acknowledges support from the Alexander von Humboldt Foundation. References [1] S. Strauf, N. G. Stoltz, M. T. Rakher, L. A. Coldren, P. M. Petroff, and D. Bouwmeester. High- frequency single-photon source with polarization control. Nat. Photon., 1:704, 2007. [2] H. S. Nguyen, G. Sallen, C. Voisin, Ph. Roussignol, C. Diederichs, and G. Cassabois. Ultra- coherent single photon source. Appl. Phys. Lett., 99:261904, 2011. [3] C. Matthiesen, A. N. Vamivakas, and M. Atature. Subnatural linewidth single photons from a quantum dot. Phys. Rev. Lett., 108:093602, 2012. [4] S. J. Boyle, A. J. Ramsay, A. M. Fox, and M. S. Skolnick. Beating of exciton-dressed states in a single semiconductor InGaAs/GaAs quantum dot. Phys. Rev. Lett., 102:207401, 2009. [5] A. N. Vamivakas, Y. Zhao, C.-Y. Lu, and M. Atature. Spin-resolved quantum-dot resonance fluorescence. Nat. Phys., 5:198, 2009. [6] A. J. Ramsay. A review of the coherent optical control of the exciton and spin states of semiconductor quantum dots. Semicond. Sci. Technol., 25:103001, 2010. [7] C. M. Simon, T. Belhadj, B. Chatel, T. Amand, P. Renucci, A. Lemaitre, O. Krebs, P. A. Dalgarno, R. J. Warburton, X. Marie, and B. Urbaszek. Robust quantum dot exciton generation via adiabatic passage with frequency-swept optical pulses. Phys. Rev. Lett., 106:166801, 2011. [8] Yanwen Wu, I. M. Piper, M. Ediger, P. Brereton, E. R. Schmidgall, P. R. Eastham, M. Hugues, M. Hopkinson, and R. T. Phillips. Population inversion in a single InGaAs quantum dot using the method of adiabatic rapid passage. Phys. Rev. Lett., 106:067401, 2011. [9] G. Chen, T. H. Stievater, E. T. Batteh, X. Li, D. G. Steel, D. Gammon, D. S. Katzer, D. Park, and L. J. Sham. Biexciton quantum coherence in a single quantum dot. Phys. Rev. Lett., 88:117901, 2002. [10] T. Flissikowski, A. Betke, I. A. Akimov, and F. Henneberger. Two-photon coherent control of a single quantum dot. Phys. Rev. Lett., 92:227401, 2004. [11] S. Stufler, P. Machnikowski, P. Ester, M. Bichler, V. M. Axt, T. Kuhn, and A. Zrenner. Two- photon Rabi oscillations in a single InxGa1´xAs/GaAs quantum dot. Phys. Rev. B, 73:125304, 2006. [12] I. A. Akimov, J. T. Andrews, and F. Henneberger. Stimulated emission from the biexciton in a single self-assembled II-VI quantum dot. Phys. Rev. Lett., 96:067401, 2006. [13] S.J. Boyle, A.J. Ramsay, , A.M. Fox, and M.S. Skolnick. Two-color two-photon rabi oscillation of biexciton in single InAs/GaAs quantum dot. Physica E, 42:2485, 2010. [14] T. Miyazawa, T. Kodera, T. Nakaoka, K. Watanabe, N. Kumagai, N. Yokoyama, and Y. Arakawa. Two-photon control of biexciton population in telecommunication-band quantum dot. Appl. Phys. Express, 3:064401, 2010. [15] N. Akopian, N. H. Lindner, E. Poem, Y. Berlatzky, J. Avron, D. Gershoni, B. D. Gerardot, and P. M. Petroff. Entangled photon pairs from semiconductor quantum dots. Phys. Rev. Lett., 96:130501, 2006. Distilling one, two and entangled pairs of photons from a quantum dot 22 [16] R. M. Stevenson, R. J. Young, P. Atkinson, K. Cooper, D. A. Ritchie, and A. J. Shields. A semiconductor source of triggered entangled photon pairs. Nature, 439:179, 2006. [17] R. Hafenbrak, S. M. Ulrich, P. Michler, L. Wang, A. Rastelli, and O. G. Schmidt. Triggered polarization-entangled photon pairs from a single quantum dot up to 30K. New J. Phys., 9:315, 2007. [18] J. P. Reithmaier, G. Sek, A. Loffler, C. Hofmann, S. Kuhn, S. Reitzenstein, L. V. Keldysh, V. D. Kulakovskii, T. L. Reinecker, and A. Forchel. Strong coupling in a single quantum dot -- semiconductor microcavity system. Nature, 432:197, 2004. [19] T. Yoshie, A. Scherer, J. Heindrickson, G. Khitrova, H. M. Gibbs, G. Rupper, C. Ell, O. B. Shchekin, and D. G. Deppe. Vacuum Rabi splitting with a single quantum dot in a photonic crystal nanocavity. Nature, 432:200, 2004. [20] E. Peter, P. Senellart, D. Martrou, A. Lemaıtre, J. Hours, J. M. G´erard, and J. Bloch. Exciton- photon strong-coupling regime for a single quantum dot embedded in a microcavity. Phys. Rev. Lett., 95:067401, 2005. [21] A. Dousse, J. Suffczy´nski, A. Beveratos, O. Krebs, A. Lemaıtre, I. Sagnes, J. Bloch, P. Voisin, and P. Senellart. Ultrabright source of entangled photon pairs. Nature, 466:217, 2010. [22] K. Hennessy, A. Badolato, M. Winger, D. Gerace, M. Atature, S. Gulde, S. Falt, E. L. Hu, and A. Imamo¯glu. Quantum nature of a strongly coupled single quantum dot -- cavity system. Nature, 445:896, 2007. [23] M. Nomura, N. Kumagai, S. Iwamoto, Y. Ota, and Y. Arakawa. Laser oscillation in a strongly coupled single-quantum-dot -- nanocavity system. Nat. Phys., 6:279, 2010. [24] Y. Ota, S. Iwamoto, N. Kumagai, and Y. Arakawa. Spontaneous two-photon emission from a single quantum dot. Phys. Rev. Lett., 107:233602, 2011. [25] A. Laucht, J.M. Villas-Boas, S. Stobbe, N. Hauke, F. Hofbauer, G. Bohm, P. Lodahl, M.-C. Amannand, M. Kaniber, and J. J. Finley. Mutual coupling of two semiconductor quantum dots via an optical nanocavity. Phys. Rev. B, 82:075305, 2010. [26] 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. Garc´ıa, and P. A. Postigo. Optical coupling of two distant InAs/GaAs quantum dots by a photonic-crystal microcavity. Phys. Rev. B, 81:193301, 2010. [27] L. Schneebeli, M. Kira, and S. W. Koch. Characterization of strong light -- matter coupling in semiconductor quantum-dot microcavities via photon-statistics spectroscopy. Phys. Rev. Lett., 101:097401, 2008. [28] J. Kasprzak, S. Reitzenstein, E. A. Muljarov, C. Kistner, C. Schneider, M. Strauss, S. Hofling, A. Forchel, and W. Langbein. Up on the Jaynes-Cummings ladder of a quantum-dot/microcavity system. Nat. Mater., 9:304, 2010. [29] F.P. Laussy, E. del Valle, M. Schrapp, A. Laucht, and J. J. Finley. Climbing the Jaynes -- Cummings ladder by photon counting. arXiv:1104.3564, 2011. [30] A. Majumdar, M. Bajcsy, and J. Vuckovic. Probing the ladder of dressed states and nonclassical light generation in quantum-dot -- cavity QED. Phys. Rev. A, 85:041801(R), 2012. [31] E. del Valle, S. Zippilli, F. P. Laussy, A. Gonzalez-Tudela, G. Morigi, and C. Tejedor. Two-photon lasing by a single quantum dot in a high-Q microcavity. Phys. Rev. B, 81:035302, 2010. [32] E. del Valle, A. Gonzalez-Tudela, E. Cancellieri, F. P. Laussy, and C. Tejedor. Generation of a two-photon state from a quantum dot in a microcavity. New J. Phys., 13:113014, 2011. [33] E. del Valle, A. Gonzalez-Tudela, and F. P. Laussy. Generation of a two-photon state from a quantum dot in a microcavity under incoherent and coherent continuous excitation. Proc. SPIE, 8255:825505, 2012. [34] T. M. Stace, G. J. Milburn, and C. H. W. Barnes. Entangled two-photon source using biexciton emission of an asymmetric quantum dot in a cavity. Phys. Rev. B, 67:085317, 2003. [35] S. Schumacher, J. Forstner, A. Zrenner, M. Florian, C. Gies, P. Gartner, and F. Jahnke. Cavity- assisted emission of polarization-entangled photons from biexcitons in quantum dots with fine- Distilling one, two and entangled pairs of photons from a quantum dot 23 structure splitting. Opt. Express, 20:5335, 2012. [36] E. del Valle, A. Gonzalez-Tudela, F. P. Laussy, C. Tejedor, and M. J. Hartmann. Theory of frequency-filtered and time-resolved N -photon correlations. Phys. Rev. Lett., 109:183601, 2012. [37] B. R. Mollow. Power spectrum of light scattered by two-level systems. Phys. Rev., 188:1969, 1969. [38] A. Gonzalez-Tudela, F. P. Laussy, C. Tejedor, M J. Hartmann, and E. del Valle. Two-photon spectra of quantum emitters. In preparation., 2012. [39] H. F. Arnoldus and G. Nienhuis. Photon correlations between the lines in the spectrum of resonance fluorescence. J. phys. B.: At. Mol. Phys., 17:963, 1984. [40] L. Knoll and G. Weber. Theory of n-fold time-resolved correlation spectroscopy and its application to resonance fluorescence radiation. J. phys. B.: At. Mol. Phys., 19:2817, 1986. [41] J. D. Cresser. Intensity correlations of frequency-filtered light fields. J. phys. B.: At. Mol. Phys., 20:4915, 1987. [42] A. Aspect, G. Roger, S. Reynaud, J. Dalibard, and C. Cohen-Tannoudji. Time correlations between the two sidebands of the resonance fluorescence triplet. Phys. Rev. Lett., 45:617, 1980. [43] C. A. Schrama, G. Nienhuis, H. A. Dijkerman, C. Steijsiger, and H. G. M. Heideman. Destructive interference between opposite time orders of photon emission. Phys. Rev. Lett., 67, 1991. [44] R. Centeno Neelen, D. M. Boersma, M. P. van Exter, G. Nienhuis, and J. P. Woerdman. Spectral filtering within the Schawlow -- Townes linewidth as a diagnostic tool for studying laser phase noise. Opt. Commun., 100:289, 1993. [45] E. Moreau, I. Robert, L. Manin, V. Thierry-Mieg, J. M. G´erard, and I. Abram. Quantum cascade of photons in semiconductor quantum dots. Phys. Rev. Lett., 87:183601, 2001. [46] D. Press, S. Gotzinger, S. Reitzenstein, C. Hofmann, A. Loffler, M. Kamp, A. Forchel, and Y. Yamamoto. Photon antibunching from a single quantum dot-microcavity system in the strong coupling regime. Phys. Rev. Lett., 98:117402, 2007. [47] M. Kaniber, A. Laucht, A. Neumann, J. M. Villas-Boas, M. Bichler, M.-C. Amann, and J. J. Finley. Investigation of the nonresonant dot-cavity coupling in two-dimensional photonic crystal nanocavities. Phys. Rev. B, 77:161303(R), 2008. [48] G. Sallen, A. Tribu, T. Aichele, R. Andr´e, L. Besombes, C. Bougerol, M. Richard, S. Tatarenko, K. Kheng, and J.-Ph. Poizat. Subnanosecond spectral diffusion measurement using photon correlation. Nat. Photon., 4:696, 2010. [49] A. Ulhaq, S. Weiler, S. M. Ulrich, R. Rossbach, M. Jetter, and P. Michler. Cascaded single-photon emission from the Mollow triplet sidebands of a quantum dot. Nat. Photon., 6:238, 2012. [50] G. Bel and F. L. H. Brown. Theory for wavelength-resolved photon emission statistics in single- molecule fluorescence spectroscopy. Phys. Rev. Lett., 102:018303, 2009. [51] J.H. Eberly and K. W´odkiewicz. The time-dependent physical spectrum of light. J. Opt. Soc. Am., 67:1252, 1977. [52] O. Benson, C. Santori, M. Pelton, and Y. Yamamoto. Regulated and entangled photons from a single quantum dot. Phys. Rev. Lett., 84:2513, 2000. [53] F. Troiani, J. I. Perea, and C. Tejedor. Analysis of the photon indistinguishability in incoherently excited quantum dots. Phys. Rev. B, 73:035316, 2006. [54] E. A. Meirom, N. H. Lindner, Y. Berlatzky, E. Poem, N. Akopian, J. E. Avron, and D. Gershoni. Distilling entanglement from random cascades with partial "which path" ambiguity. Phys. Rev. A, 77:062310, 2008. [55] G. Pfanner, M. Seliger, and U. Hohenester. Entangled photon sources based on semiconductor quantum dots: The role of pure dephasing. Phys. Rev. B, 78:195410, 2008. [56] J. E. Avron, G. Bisker, D. Gershoni, N. H. Lindner, E. A. Meirom, and R. J. Warburton. Entanglement on demand through time reordering. Phys. Rev. Lett., 100:120501, 2008. [57] R. Johne, N. A. Gippius, G. Pavlovic, D. D. Solnyshkov, I. A. Shelykh, and G. Malpuech. Entangled photon pairs produced by a quantum dot strongly coupled to a microcavity. Phys. Rev. Lett., 100:240404, 2008. Distilling one, two and entangled pairs of photons from a quantum dot 24 [58] P. K. Pathak and S. Hughes. Generation of entangled photon pairs from a single quantum dot embedded in a planar photonic-crystal cavity. Phys. Rev. B, 79:205416, 2009. [59] A. Carmele, F. Milde, M.-R. Dachner, M. B. Harouni, R. Roknizadeh, M. Richter, and A. Knorr. Formation dynamics of an entangled photon pair: A temperature-dependent analysis. Phys. Rev. B, 81:195319, 2010. [60] A. Carmele and A. Knorr. Analytical solution of the quantum-state tomography of the biexciton cascade in semiconductor quantum dots: Pure dephasing does not affect entanglement. Phys. Rev. B, 84:075328, 2011. [61] A. N. Poddubny. Effect of continuous and pulsed pumping on entangled photon pair generation in semiconductor microcavities. Phys. Rev. B, 85:075311, 2012. [62] M. O. Scully, B.-G. Englert, and H. Walther. Quantum optical tests of complementarity. Nature, 351:111, 1991. [63] W. Chuan and Z. Yong. Quantum secret sharing protocol using modulated doubly entangled photons. Chinese Phys., 18:3238, 2009. [64] H. Wang, S. Liu, and J. He. Thermal entanglement in two-atom cavity QED and the entangled quantum otto engine. Phys. Rev. E, 79:041113, 2009. [65] R. Trotta, E. Zallo, C. Ortix, P. Atkinson, J. D. Plumhof, J. van den Brink, A. Rastelli1, and O. G. Schmid. Universal recovery of the energy-level degeneracy of bright excitons in ingaas quantum dots without a structure symmetry. Phys. Rev. Lett., 109:147401, 2012. [66] W. K. Wootters. Entanglement of formation of an arbitrary state of two qubits. Phys. Rev. Lett., 80:2245, 1998. [67] W. J. Munro, D. F. V. James, A. G. White, and P. G. Kwiat. Maximizing the entanglement of two mixed qubits. Phys. Rev. A, 64:030302, 2001. 

1601.03429

1

1601

Spin Chern Pumping from the Bulk of Two-Dimensional Topological Insulators

Topological insulators (TIs) are a new quantum state of matter discovered recently, which are characterized by unconventional bulk topological invariants. Proposals for practical applications of the TIs are mostly based upon their metallic surface or edge states. Here, we report the theoretical discovery of a bulk quantum pumping effect in a two-dimensional TI electrically modulated in adiabatic cycles. In each cycle, an amount of spin proportional to the sample width can be pumped into a nonmagnetic electrode, which is attributed to nonzero spin Chern numbers $C_{\pm}$. Moreover, by using a half-metallic electrode, universal quantized charge pumping conductivities $-C_{\pm}e^2/h$ can be measured. This discovery paves the way for direct investigation of the robust topological properties of the TIs.

cond-mat.mes-hall

cond-mat

a Spin Chern Pumping from the Bulk of Two-Dimensional Topological Insulators M. N. Chen1, L. Sheng1,∗ R. Shen1, D. N. Sheng2, and D. Y. Xing1† 1National Laboratory of Solid State Microstructures, Department of Physics, and Collaborative Innovation Center of Advanced Microstructures, Nanjing University, Nanjing 210093, China 2 Department of Physics and Astronomy, California State University, Northridge, California 91330, USA Topological insulators (TIs) are a new quantum state of matter discovered recently, which are characterized by unconventional bulk topological invariants. Proposals for practical applications of the TIs are mostly based upon their metallic surface or edge states. Here, we report the theoretical discovery of a bulk quantum pumping effect in a two-dimensional TI electrically modulated in adiabatic cycles. In each cycle, an amount of spin proportional to the sample width can be pumped into a nonmagnetic electrode, which is attributed to nonzero spin Chern numbers C±. Moreover, by using a half-metallic electrode, universal quantized charge pumping conductivities −C±e2/h can be measured. This discovery paves the way for direct investigation of the robust topological properties of the TIs. PACS numbers: 72.25.-b, 73.43.-f, 73.23.-b, 75.76.+j I. INTRODUCTION Topological transport phenomena have been attract- ing a great deal of interest, because they exhibit univer- sal properties that are insensitive to perturbations and independent of material details. A classical example of such a transport phenomenon is the integer quantum Hall (IQH) effect in two-dimensional (2D) electron systems, first discovered in 1980, [1] which is characterized by an integer quantization of the Hall conductivity in unit of e2/h. The IQH effect has been observed in a large va- riety of materials, ranging from traditional semiconduc- tors, to oxides, [2] graphene, [3] and topological insula- tors (TIs). [4] Laughlin [5] interpreted the IQH effect in terms of an adiabatic charge pump. Thouless, Kohmoto, Nightingale, and Nijs [6] established a relation between the quantized Hall conductivity of the IQH system and a topological invariant, the first Chern number. Thouless and Niu [7, 8] also related the amount of charge pumped in a 1D charge pump to the Chern number. A variant of the IQH effect, the quantum spin Hall (QSH) effect, was proposed recently, [9, 10] which has been experimentally realized in HgTe quantum wells [11] and InAs/GaSb bilayers. [12] Extension of the idea of the QSH effect has led to the discovery of 3D TIs. [13 -- 16] A QSH system, which is also called a 2D TI, has an insulat- ing band gap in the bulk and a pair of gapless helical edge states at the sample boundary. When the electron spin is conserved, a QSH system can be viewed as two indepen- dent IQH systems without Landau levels. [17] Different from the charge, the spin does not obey a fundamental conservation law. In general, when the spin conservation is absent, unconventional topological invariants, either the Z2 index [18] or the spin Chern numbers, [19 -- 21] are needed to describe the QSH systems. The time-reversal ∗ [email protected] † [email protected] (TR) symmetry is considered to be a prerequisite for the QSH effect, which protects both the Z2 index and gap- less nature of the edge states. However, based upon the spin Chern numbers, it was shown that the bulk topo- logical properties remain intact even when the TR sym- metry is broken. [21] This finding evokes interest to pur- sue direct investigation and possibly utilization of the robust topological properties of the TIs, besides using their symmetry-protected gapless edge states which are more fragile in realistic environments. Unlike the first Chern number underlying the IQH sys- tems, which is embedded into the Hall conductivity, up to now the topological invariants in the TIs have not been directly observable. Several experimental methods were proposed, but have not been realized. One was to measure the topological magnetoelectric effect, [22, 23] for which experimental complexities exist. [23] Fu and Kane [24] put forward an abstract 1D model, in which the spin pumping was related to the Z2 index in the limit of weak coupling. However, how this fictitious model could be implemented is still unknown. Furthermore, from the viewpoint of application, generalization of the idea of the Z2 pump to higher dimension is meaningless, because according to the Z2 theory, [24] only the states at the TR-invariant point of the Brillouin zone can con- tribute to the spin pumping, and so the pumping rate cannot be enhanced by increment of dimension. In a re- cent work, [25] the more general case of finite coupling between the pump and electrode is investigated by us- ing the scattering matrix method. It was found that the spin pumping in the model of Fu and Kane can survive finite scattering of magnetic impurities, and so may be attributed to the spin Chern numbers rather than the Z2 index. Some other authors [26, 27] proposed to pump quantized charge through the helical edge states by pre- cessing a magnet covering the edge of a 2D TI, so that the number of gapless edge channels can be counted through electrical measurement. This method is indirect, in the sense that the topological invariants are intrinsic prop- erties of the bulk electron wavefunctions, which do not immediately determine the charge pumping in the edge channels. Here we predict an intriguing bulk topological pumping effect, directly driven by nonzero spin Chern numbers, in a QSH system electrically modulated in adiabatic cycles. As a consequence of the topological spectral flows of the spin-polarized Wannier functions (SPWFs) in the bulk of the system, spin can be pumped into a nonmagnetic electrode continuously without net charge transfer. The total amount of spin pumped per cycle is proportional to the (cross-section) width of the sample, and insensitive to the material parameters and spin-mixing effect due to the Rashba spin-orbit coupling. This electrical spin pump es- tablishes a basis, on which spintronic applications taking advantage of the robust topological properties of the TIs can be developed. Especially, if a half-metallic electrode with spin polarization parallel (or antiparallel) to the z- axis is used, a quantized charge pumping conductivity, −C+e2/h (or −C−e2/h), can be measured by electrical means, demonstrating a way to observe the spin Chern numbers C± directly. II. SPIN CHERN NUMBERS AND SPWFS Let us consider a 2D model Hamiltonian HP = H0+H1 with H0 = vF [kx sz σx − (ky + eA(t)) σy] − M (t)σz . (1) Here (−e) is the electron charge, k is the 2D momentum, A(t) = A0 sin(ω0t) is the vector potential of an ac elec- tric field −E0 cos(ω0t) applied along the y direction with A0 = E0/ω0 and frequency ω0 > 0 being designated, and M (t) = M0 cos(ω0t). This model can describe both the QSH materials, the HgTe quantum wells, [28 -- 30] and InAs/GaSb bilayers, [31] in the linear order in momen- tum. For definity of discussion, we confine ourselves to the HgTe quantum wells, for which sα with α = x, y, z are the Pauli matrices for spin, and σα for the electron and hole bands. As will be discussed below, the time- dependent mass term M (t) can be induced by varying the voltages of the dual gates. H1 represents the Rashba spin-orbit coupling [32] H1 = R0 2 (1 + σz)[sykx − sx(ky + eA(t))] . (2) To the linear order in momentum, the Rashba spin-orbit coupling is nonvanishing only in the electron band. [32] Within the adiabatic approximation, for a bulk sam- ple there exists a finite energy gap between the con- duction and valence bands for ω0t 6= π/2 or 3π/2. At ω0t = π/2 and 3π/2, the conduction and valence bands touch at kx = 0 and ky = kc y = eA0 = eE0/ω0. To clarify the topological properties underlying the spin/charge pumping, we consider ky as a parameter, and calculate the spin Chern numbers C± in the standard way, [21] on the torus of the two variables kx ∈ (−∞,∞) y or −kc y with kc 2       6  
 t 4  2 0 -6 -3 0 3 6 Wannier centers <x>/a  FIG. 1. Plot of the centers of mass of the SPWFs (horizontal axis) as functions of ω0t (vertical axis). The parameters are taken to be ky = 0.4kc y, M0 = vFeA0 = R0/a0 = 0.1t0, V0 = 0.3t0, and d = a0, with a0 as the lattice constant and t0 = vF/a0 (v′ F = vF) as the hopping integral of the tight- binding Hamiltonian. and t ∈ [0, T ) with T = 2π/ω0 as the period. The spin Chern numbers are obtained as C± = ±sgn(E0M0) , (3) for ky < kc the band touching points ky = ±kc points. y, and vanish elsewhere. Not surprisingly, y serve as the critical We now consider a system consisting of a pump for x < 0, an electrode for x > d, and a potential barrier in between. The total Hamiltonian of the system reads HP HE + V0 σz HE (x < 0) (0 < x < d) (x > d) , (4) H =  where HP has been given above, and HE = v′ Fkxsz σx is the Hamiltonian of the electrode. A possible experi- mental setup for realizing this Hamiltonian is explained in Appendix A in more details. In the barrier region, the term V0 σz opens an insulating gap of size 2V0, which accounts for contact deficiencies between the pump and electrode. The crucial role of the nonzero spin Chern numbers in the spin/charge pumping process can be visu- alized by using the SPWFs, which were first introduced in Ref. [25]. We construct a tight-binding Hamiltonian for the effective 1D system at any given ky according to Eq. (4), and diagonalize the total Hamiltonian of the pump and electrode numerically. Following the same procedure as calculating the spin Chern numbers, [21] the space oc- cupied by electrons is partitioned into two spin sectors after diagonalizing the spin operator sz in the occupied space. By definition, the states in the two spin sectors are essentially the maximally spin-polarized states. Then we construct the Wannier functions [33, 34] for the spin- up and spin-down sectors, respectively, which are called the SPWFs. w The evolution of the centers of mass of the SPWFs for ky = 0.4kc y and R0 = 0.1vF is shown in Fig. 1. We see that the Wannier centers for the spin-up sector move right and those for the spin-down sector move left, each center shifting on average a lattice constant per cycle. Within the adiabatic approximation, time t ∈ [0, T ) plays the same role as the momentum of an additional dimen- sion, [24] namely, kt ∈ [0, T ). Therefore, when ky is considered as a parameter, the evolution of the Wannier functions of the effectively 1D system related to vari- ous kx with time t can be understood from the static properties of a 2D system associated with various kx and kt. In the general theory, [34] the relationship between the Chern number and the spectral flows of the Wannier functions in a 2D system has been established. Accord- ing to this theory, the average displacement of each of the centers of the SPWFs in the spin-up (spin-down) sector with changing kt (or t) from 0 to T , in units of the lattice constant, must equal to the spin Chern number C+ = 1 (C− = −1). Therefore, the nontrivial transfer of the SP- WFs observed in Fig. 1 is a direct manifestation of the nonzero spin Chern numbers C± = ±1 in the pump (for E0M0 > 0). More interestingly, we see that such spec- tral flows can go across the finite barrier (V0d > 0), and extend into the electrode, even though the barrier and electrode are topologically trivial. Physically, because the system needs to recover its original eigenstates when each cycle ends, the nontrivial spectral flows of the SP- WFs in the TI need to constitute closed loops through formation of edge states at the boundary, [35] or extend into the electrode. However, localized edge states can not exist at the finite barrier due to quantum tunneling effect, so the transfer of the spectral flows of the SPWFs 3 into the electrode occurs. This result will be further con- firmed by direct calculation based upon the scattering matrix theory in the next section. The SPWFs are just another equivalent representation of the occupied space, and so the counter spectral flows of the Wannier centers in the two spin sectors represent the true movements of the electrons. If the Rashba spin-orbit coupling were neglected, the Wannier functions would be the eigenstates of sz. The nontrivial spectral flows indicate that at the given ky, in each cycle a spin-up electron goes from the pump into the electrode, and a spin-down electron moves oppositely. Therefore, no net charge transfer occurs but a quantized spin of 2(/2) is pumped into the electrode. When the small Rashba spin-orbit coupling is turned on, while the topological spectral flows remain intact, as seen from Fig. 1, the spin polarizations of the Wannier functions are no longer fully parallel to the z-axis, and may also vary with time. As a consequence, the amount of spin pumped per cycle will deviate from the quantized value. III. THE PROCESS OF SPIN CHERN PUMPING A. Spin pumping for a nonmetallic electrode In general, the amount of the spin pumped can be con- veniently calculated by using the scattering matrix for- mula. [36, 37] The z-component of the spin pumped per cycle is given by [36, 37] ∆sz(ky) =  4πiIT dt(cid:16)r∗ ↑↑ dr↑↑ dt − r∗ ↓↓ dr↓↓ dt − r∗ ↓↑ dr↓↑ dt + r∗ ↑↓ dr↑↓ dt (cid:17) , (5) where rαβ (α, β =↑,↓) is the reflection amplitude for an electron at the Fermi energy incident from the spin-β channel of the electrode and reflecting back into the spin-α channel. In the following calculations, the Fermi energy is set to be zero (EF = 0), and the Rashba spin-orbit coupling is treated as a perturbation. As shown in Appendix B, to the linear order in R0, we obtain r↑↑ = − cos(2θ) + i[sh(2γ0d) − sin(2θ)ch(2γ0d)] ch(2γ0d) − sin(2θ)sh(2γ0d) + O(ǫ2) , (6) (8) r↓↑ = ǫ 2 sin(2θ)[1 − cos(2θ)] ch(2γ0d) − sin(2θ)sh(2γ0d) + O(ǫ2) , (7) and r↓↓ = r↑↑2θ→(π−2θ) and r↑↓ = −r↓↑2θ→(π−2θ), where γ0 = V0/v′ F and 2θ = arg[vF(ky + eA(t)) + iM (t)]. We note that the dimensionless quantity ǫ = R0/vF ap- pears as the small expansion parameter. Since r↓↑ and r↑↓ are always real, the contributions from the third and ↑↑dr↑↑ − r∗ ↓↓dr↓↓(cid:1) . This expression has a geometric explanation: the amount of spin pumped per cycle equals to the difference be- tween the areas enclosed by the directional trajecto- ries of r↑↑ and r↓↓ on the complex plane, multiplied by /2π. Due to the relation r↓↓ = r↑↑2θ→(π−2θ), yielding r↓↓ = −Re(r↑↑)+iIm(r↑↑), the two terms in Eq. (8) make fourth terms in Eq. (5) vanish. Consequently, ∆sz(ky) =  4πiIT (cid:0)r∗ 6 Im 1 4 -1 0 ) 0 ( - ) t ( 2 0 Re 1 d ) = ( 0.4 k* ) , 0 ) + , 0) (0.8k% & ) 0 . 1 , k# $ .4 0 ( -1 , g ( (k , 0) (1.2k! " 0 2 4 6 t ' 4 2.0 s - Re 1 R =0.0 0 0.05v , 0 . 1 v Im / 1.9 -1 0 1 0.0 -1 0.5 d . 1.0 FIG. 2. Argument of the complex reflection amplitude, ϕ(t) = arg(r↑↑), as a function of ω0t for four sets of (ky, γ0d). The other parameters are taken to be R0 = 0 and vFeA0 = M0 with kc y = eA0. Inset: trajectories of r↑↑ in a cycle on the complex plane. FIG. 3. ∆sz(ky) (in unit of /2) as a function of γ0d for ky = 0.4kc y and three different values of R0. The other parameters are taken to be the same as in Fig. 2. Inset: the trajectory of r↑↑ in a cycle for R0 = 0.1vF and γ0d = 1.0, with the unit circle indicated by the dotted line. an equal contribution, so that we can focus on the first term. While the expression (6) for r↑↑ is independent of the Rashba spin-orbit coupling, as will be shown soon, a combination of Eqs. (6) and (7) allows us to evaluate the amount of spin pumped up to the second order in R0/vF. We first consider the case of R0 = 0. From Eq. (6), it is easy to show r↑↑(ky) = 1. In Fig. 2, we plot the argument ϕ(t) of r↑↑(ky) as a function of ω0t for several parameter sets. For either γ0d = 0 (ideal contact) or 1.0 (strong potential barrier), ϕ(t) always increments 2π in a cycle as long as ky < kc y. In this case, the trajectories of r↑↑(ky) always form a unit circle on the complex plane, oriented counterclockwise, as shown in the inset of Fig. 2, suggesting ∆sz(ky) =  (for E0M0 > 0). For ky > kc y, however, the situation is quite different. ϕ(t) does not change after going through a cycle, and the trajectory of r↑↑(ky) does not enclose a finite area, so that ∆sz(ky) = 0. Apparently, the present result conforms to the spin Chern numbers given by Eq. (3) and the spectral flows of the SPWFs. Next we study the correction to ∆sz(ky) due to nonzero Rashba spin-orbit coupling. By expressing r↑↑ = ρeiϕ = p1 − δρ2ei(ϕ(0)+δϕ) in the polar coordinate sys- tem, where ϕ(0) is the argument at R0 = 0, and δϕ and δρ2 stand for the second-order corrections to ϕ and ρ2, re- spectively, due to the Rashba spin-orbit coupling, Eq. (8) becomes ∆sz(ky) = (/2π)[HT (1 − δρ2)dϕ(0) +HT dδϕ] + O(ǫ3). We notice that δϕ is a small quantity fluctuating around 0 and periodic in time, δϕt=0 = δϕt=T , so that HT dδϕ = 0. Using the identity δρ2 = r↓↑2, we then ob- tain ∆sz(ky) = (/2π)HT (1 − r↓↑2)dϕ(0), where dϕ(0) can be calculated from Eq. (6) and r↓↑ has been given by Eq. (B26). This is an expression for ∆sz(ky) accurate to the second order in R0/vF. At γ0d = 0, we obtain ∆sz(ky) as ∆sz(ky) ≃  2"1 − 5 32(cid:18) R0 vF(cid:19)2# (C+ − C−) , (9) for ky < kc y, and ∆sz(ky) = 0 elsewhere. As expected, nonzero Rashba spin-orbit coupling causes ∆sz(ky) to deviate from its quantized value, i.e., (C+ − C−)/2. In real materials, R0 is usually much smaller (by an order of magnitude or more) than vF, [32] so that the deviation is less than a percent for an ideal connection between the pump and electrode. For γ0d 6= 0, ∆sz(ky) can be evaluated numerically and its calculated result is plotted in Fig. 3 as a function of γ0d for three different strengths of the Rashba spin- orbit coupling. For R0 = 0, ∆sz(ky) is quantized to , independent of γ0d. For R0 = 0.05vF and 0.1vF, weak potential barrier (γ0d ≪ 1) has little effect on ∆sz(ky). This is reasonable as the leading-order correction of small γ0d must be O(ǫ2γ0d). Appreciable deviations from the quantized value occur for strong potential barrier (e.g., γ0d ≃ 1). We note that δϕ does not affect the orbit of r↑↑, as it represents a variation in the tangent direction of the orbit, and the orbit can be determined by r↑↑ = Inset shows the trajectory of r↑↑ on the complex plane for R0 = 0.1vF and γ0d = 1.0. For such a strong potential barrier, the orbit of r↑↑ deviates from the unit circle visibly. The above result suggests that improving the contact quality between the pump and electrode is helpful for obtaining a nearly integer- quantized value of the pumped spin. By summing over ky between −kc for the total spin pumped per cycle p1 − r↓↑2eiϕ(0) y, we obtain y and kc . ∆Sz = σs(2E0Ly/ω0) , (10) j j w D g with σs ≃ e 4π "1 − 5 32(cid:18) R0 vF(cid:19)2# (C+ − C−) , (11) for a good contact (γ0d ≪ 1). ∆Sz is in scale with width Ly of the pump. By noting that ω0 is proportional to the number of cycles per unit time, σs can be considered as the spin pumping conductivity. B. Charge pumping for a half-metallic electrode Now we discuss a possible way to experimentally ob- serve the spin Chern numbers, by using a half-metallic electrode, in which conducting channels for electron spin antiparallel to the spin polarization are absent. We first consider the case, where the spin polarization of the elec- trode is parallel to the z axis. The Hamiltonian of the F(ky)kx sz σx+V1(1−sz)σz/2. electrode is taken as HE = v′ In this case, as shown in Appendix C, r↑↑ is still given by Eq. (6) but r↓↑ ≡ 0. It follows that for any ky between −kc y, the charge pumped per cycle is integer- quantized and equal to ∆q(ky) = (−e)C+. Similarly, for the spin polarization of the electrode antiparallel to the z axis, the charge pumped is equal to ∆q(ky) = (−e)C−. Therefore, the total charge pumped per cycle is given by y and kc with ∆Q = σc(2E0Ly/ω0) , σc = −C± e2 h , (12) (13) where the spin Chern number C+ (C−) is taken for the spin polarization of the electrode parallel (antiparallel) to the z axis. We emphasize that Eqs. (12) and (13) ob- tained above are valid for finite Rashba spin-orbit cou- pling and finite potential barrier between the pump and electrode, indicating that the quantized charge pumping is robust against small perturbations. Experimentally, ∆Q can be obtained by measuring the electrical current in the electrode, and from Eqs. (12) and (13), C± can be evaluated, yielding an experimental method to mea- sure the spin Chern numbers directly. The sign inversion of ∆Q with reversing the spin polarization of the half- metallic electrode, as indicated by Eqs. (12) and (13), is a hallmark of the present spin Chern charge pump, which can be used to distinguish it from the conventional Thou- less charge pump [7, 8]. We have used the single-electron approximation, where the electron interaction is not taken into account. In particular, in the half-metallic electrode case, one of the spin channel is blocked at the boundary, which naturally induces some charge and spin accumulations, and consequently changes the potential profile. However, owing to the screening effect, the change in the potential profile is expected to be localized at the boundary, 5 which in effect modifies the potential barrier between the pump body and the electrode. As has been shown above, the charge pumping effect is independent of the existence and details of the potential barrier, and so we believe that the pumping effect will survive the charge and spin accumulations. IV. DISCUSSION Up to now, all the results obtained from the scatter- ing matrix formula are apparently in complete agreement with the spin Chern numbers given by Eq. (3). These re- sults cannot be explained within the framework of the Z2 theory. [24] While one can define a Z2 index at the TR-invariant point ky = 0, the effective 1D Hamilto- nian given by Eqs. (1) and (2) for any given nonzero ky does not preserve the TR symmetry, as its TR partner is at −ky, making the Z2 index invalid. The Z2 theory predicted that the TR symmetry is crucial for the topo- logical spin pumping, [24] suggesting that only the states at the TR-invariant point ky = 0 can contribute to the spin pumping. This clearly contradicts the present re- sult that all the states with ky < kc y contribute equally, which is obtained directly from the scattering matrix for- mula. This point is also evidenced by the fact that the total amount of charge or spin pumped per cycle is in proportion to the sample width Ly. For the same reason, the pumping effect found in this work is also essentially different from that via edge states in Refs. [26, 27], where the amount of spin or charge pumped per cycle is pro- portional to the number of the gapless edge channels. In conclusion, our work uncovers a bulk topological pumping effect due to direct transfer of the SPWFs be- tween the pump and electrode, without the participation of edge states. This measurable effect reveals the bulk topological properties of the system that are neither cap- tured by the Z2 index nor reflected by the number of gapless edge channels. It can be accurately described by the spin Chern numbers. This spin Chern pump may lay the foundation for direct experimental study and possi- bly utilization of the robust topological properties of the TIs. The previous experimental work [38] evidenced the dif- ficulty of modulating in time the properties of an open quantum dot without generating undesired bias voltages due to stray capacitances. This problem might not be significant in our pumping setup, where a much larger bulk sample of the TI can be used and the stray ca- pacitances can be greatly reduced. Moreover, a possible way around the obstacle is to use the ac Josephson ef- fect to induce periodically time-dependent Andreev re- flection amplitudes in a hybrid normal-superconducting system. [39] Concrete design of a spin Chern pump based upon the Josephson effect will await future work. While the proposed spin pumping scheme may have the advan- tage of low noises, its practical application in spintronic devices still relies on the discovery of new TIs with bulk band gaps much greater than room temperature, which determine the temperature range where the spin Chern pumping effect can survive. Currently, precessing mag- netization is a feasible method to generate robust spin currents in spintronic devices at room temperature. [40] ACKNOWLEDGMENTS This work was supported by the State Key Pro- gram for Basic Researches of China under grants num- bers 2015CB921202, 2014CB921103 (LS), 2011CB922103 and 2010CB923400 (DYX), the National Natural Sci- ence Foundation of China under grant numbers 11225420 (LS), 11174125, 91021003 (DYX) and a project funded by the PAPD of Jiangsu Higher Education Institutions. We also thank the US NSF grants numbers DMR-0906816 and DMR-1205734 (DNS). Appendix A: A POSSIBLE EXPERIMENTAL REALIZATION OF THE SPIN CHERN PUMP In what follows we expand on the model setup and possible experimental realization of the spin Chern pump in more details. 1. The pump A possible experimental realization of the Hamilto- nian Eq. (1) for the pump is illustrated in Fig. 4. A HgTe/CdTe quantum-well heterostructure with dual gates (top and bottom) is placed between two conduc- tive plates. It is known that when the width of the quantum well (thickness of the HgTe film) is above a critical size dc = 6.3nm, [28, 29] the band structure is inverted, characterized a negative mass term −M0σz in the Hamiltonian, corresponding to the QSH state. If the width of the quantum well falls below dc, the band struc- ture will be aligned in a "normal" way with a positive mass term M0 σz, corresponding to a normal insulator. As has been discussed in Refs. [30] and [31], the topolog- ical phase transition between the QSH phase and normal insulator can also be tuned by applying a gate voltage, which effectively reduces the width of the quantum well. It is assumed that the quantum well under considera- tion has a width somewhat greater than dc, and so has a negative mass term −M0σz initially. With increasing the gate voltage, the electron mass increases, and can invert its sign. Usually, increasing the gate voltage may also adjust the carrier density. Nevertheless, it has been shown [30, 31] that, by using dual gates and properly tun- ing their voltages V1(t) and V2(t), it is generally possible to change the electron mass in the desired manner to be 6 nj LJ sϭ;ƚͿ sϮ;ƚͿ ĚdĞ ,ŐdĞ ĚdĞ ¥ h;ƚͿ FIG. 4. A schematic view of a experimental setup for realiza- tion of a 2D spin Chern pump. A CdTe/HgTe/CdTe quantum well heterostructure, with dual gates on its top and bottom, is placed between two conductive plates. When the voltages of the gates and plates are adiabatically modulated in proper cycles, spin or charge can be pumped into electrodes coupled to the quantum well along the x direction. −M0 cos(ω0t), while keeping the electron Fermi energy still in the band gap. The effect of the conductive plates is easily understood. When a voltage drop U (t) is applied across the plates, a uniform electric field E(t) = E(t)y will be generated in the space between the two plates. The electrons in the quantum well experience a vector potential A(t) = A(t)y with A(t) defined as E(t) = −∂A(t)/∂t. If the electric field is chosen to be E(t) = −E0 cos(ω0t), one gets A(t) = A0 sin(ω0t) with A0 = E0/ω0, as desired. We point out that the exact time dependencies of M (t) and A(t) are not essential for realizing the spin Chern pump, provided that they have the same periodicity and a constant relative phase shift. 2. The nonmagnetic electrode The pumping effect is insensitive to material details of the electrode. The electrode is taken to be a nor- mal metal with a 2D parabolic Hamiltonian HE = −E0 + p2/2m. When E0 is sufficiently large, for a given py, we can linearize the effective 1D Hamiltonian HE at the right and left Fermi points px = ±mv′ F(ky) with F(ky) = q2m(EF + E0) − k2 v′ y/m. A Pauli matrix σx is introduced to describe the two branches. To be consis- tent with the form of the Hamiltonian in the pump, we use σx = 1 and −1, respectively, to represent the right- moving and left-moving branches for sz = 1 and oppo- sitely for sz = −1. As a result, the Hamiltonian of the electrode becomes HE = v′ F(ky)kxsz σx at EF = 0, where ky = py and kx = px∓ mv′ F(ky). The spin pumping effect is usually dominated by small ky, so that we can further approximate v′ F(ky) ≃ v′ F, with purpose to minimize the number of adjustable parameters in the model. F(ky = 0) ≡ v′ 3. The barrier For the present Dirac-like Hamiltonian, an ordinary potential barrier has a very weak effect on the elec- tron transmission due to the Klein paradox. There- fore, we take the Hamiltonian for the barrier to be HB = HE + V0 σz. The inclusion of potential V0 σz opens up an insulating energy gap of size 2V0 around the Fermi level, which presumably is more efficient for describing the contact deficiencies and structural mismatch between the pump and electrode. 4. The half-metallic electrode The half metal, e.g., CrO2, La2/3Sr1/3MnO3, etc., is a substance that acts as a conductor to electrons of one spin orientation, but as an insulator to those of the other spin orientation. From the viewpoint of the electronic structure, one of the spin subbands is metallic, whereas the Fermi level falls into an energy gap of the other spin subband. To simulate the half-metallic electrode, HE is F(ky)kxsz σx + V1(1 ∓ sz)σz/2, where taken to be HE = v′ ∓ stands for the spin polarization of the electrode parallel and antiparallel to the z-axis, respectively. The second term opens an energy gap of size 2V1 around the Fermi level for electron spin antiparallel to the spin polariza- tion of the electrode, without affecting the other spin subband. As a result, the electron density of states is fully spin-polarized at the Fermi energy. V1 is set to be infinity in the final result. 7 Appendix B: CALCULATION OF THE REFLECTION AMPLITUDES FOR A NONMAGNETIC ELECTRODE 1. Electron wavefunctions in the pump and potential barrier We now solve the scattering problem for an electron at the Fermi energy incident from the electrode. The Fermi energy will be taken to be EF = 0, which is in the band gap of the pump. Therefore, the incident electron will be fully reflected back into the electrode. The Rashba spin-orbit coupling is treated as a perturbation, and the result will be calculated to the linear order in the small quantity ǫ = R0/vF. The wavefunctions of the pump (x < 0), barrier (0 < x < d) and electrode (x > d) are denoted by ΨP (x), ΨB(x), and ΨE(x), respectively. We have two boundary conditions: ΨP (0−) = ΨB(0+) and ΨB(d − 0+) = ΨE(d + 0+). We use ↑ and ↓ to represent the eigenstates of sz, and +1 and −1 to represent those of σz. On the basis ↑, +1i, ↑,−1i, ↓, +1i, and ↓,−1i, the Hamiltonian of pump (the Eqs. (1) and (2) in the manuscript) can be expanded as a 4 × 4 matrix HP =  −M (t) vF(kx − iky) R0(ikx − ky) 0 vF(kx + iky) R0(−ikx − ky) M (t) 0 0 0 −M (t) vF(−kx − iky) 0 0 vF(−kx + iky) M (t)   (B1) where ky = ky + eA(t). For energy E = EF = 0, the eigen-equation is obtained from Eq. (B1) [M 2(t) + v2 F k2]2 − M 2(t)R2 0 k2 = 0 , (B2) with k2 = k2 x + k2 eigenfunctions are y, and up to a normalization factor, the eikxx/ , (B3)   A1M (t)/vF −A1(kx − iky) A2M (t)/vF A2(kx + iky)   F where A1 = −(ikx + ky)M (t) and A2 = [M 2(t) + k2]/R0. We need to solve kx from the eigen-equation v2 Eq. (B2). We notice that the equation is a 4th-degree polynomial of kx with real coefficients, so complex con- jugate roots must appear in pairs. Moreover, Eq. (B2) is even in kx, so positive and negative roots appear in pairs. In combination, Eq. (B2) must have four roots of the form kx = a + ib, a− ib, −a + ib, and −a− ib. By substitution of the four roots into Eq. (B3), we can in principle obtain four different eigenfunctions. For the present scattering problem, we only need the two eigenfunctions that are decaying into the pump, which correspond to the two roots with negative imaginary parts. For R0 = 0, it is easy to obtain for the roots for the two y, which are two-fold degenerate. The corresponding two decaying eigenfunctions are given by decaying modes: kx = −iη with η = qM 2(t) + k2 ϕ+(x) = ↑i ⊗(cid:18) sin θ ϕ−(x) = ↓i ⊗(cid:18) cos θ i cos θ (cid:19) eηx , −i sin θ (cid:19) eηx , (B4) (B5) ky + iM (t)]. For R0 6= 0, we write where 2θ = Arg[vF the roots of kx as kx = −iη + δkx, and also write k2 x as k2 solve for δk2 x = −(η)2 + δk2 x from the eigen-equation Eq. (B2) x. To the second order in ǫ, we can be δk2 x =(cid:18)±i R0 vF + R2 0 2v2 F(cid:19) (η)2 sin2(2θ) + O(ǫ3) . (B6) ϕ1,2(x) = Noticing that the expression for A2 given below Eq. (B3) has a factor R0 in the denominator, we keep δk2 x to the second order, for the purpose to calculate A2 to the linear order. By using the relation δk2 x = −2iηδkx +O(ǫ2), we derive from Eq. (B6) δkx = ∓ R0 2vF η sin2(2θ) + O(ǫ2) . (B7) With these relations, we obtain for A1 and A2 A1 = −vF(η)2 sin(2θ)[1 + cos(2θ)] (η)2 sin3(2θ) + O(ǫ2) , R0 2 ± i (B8) (B9) and A2 =(cid:18)±i + R0 2vF(cid:19) vF(η)2 sin2(2θ) + O(ǫ2) . (B10) We can always eliminate any common factor that appears in all the four components of Eq. (B3), whenever possible. By eliminating a common factor 2vF(η)2 sin(2θ) cos θ, we rewrite A1 and A2 as A1 = − cos θ ± i R0 2vF sin θ sin(2θ) + O(ǫ2) , (B11) and A2 =(cid:18)±i + R0 2vF(cid:19) sin θ + O(ǫ2) . (B12) Then the two decaying wavefunctions can be derived to 8   eη∓x , (B13)   vF vF sin2 θ(cid:17) − cos2 θ sin θ(cid:16)1 ∓ i R0 sin2 θ(cid:17) −i cos3 θ(cid:16)1 ∓ 2i R0 2vF(cid:17) ±i sin2 θ cos θ(cid:16)1 ∓ i R0 ± sin3 θh1 ∓ i(cid:0) 1 vFi 2 + cos2 θ(cid:1) R0 sin2(2θ)(cid:3). 2vF where η∓ = η(cid:2)1 ∓ R0 Some remarks are in order. With respect to the wave- functions at R0 = 0, namely, Eqs. (B4) and (B5), nonzero R0 leads to nonperturbative change in the wavefunctions Eq. (B13), in the sense that Eq. (B13) will not recover Eqs. (B4) and (B5) in the limit R0 → 0. This is rea- sonable, just as what always happens in the degenerate perturbation theory of the quantum mechanics. For the present problem, the wavefunction ΨP (x) in the pump is always expressed as an arbitrary linear superposition of the two decaying modes: ΨP (x) = B1ϕ1(x) + B2ϕ2(x). By defining ϕ+(x) = −[ϕ1(x) + ϕ2(x)] and ϕ−(x) = −i[ϕ1(x) − ϕ2(x)], we can rewrite ΨP (x) as ΨP (x) = D1ϕ+(x) + D2ϕ−(x). The final result for the reflection amplitudes depends only on ΨP (x = 0−), so we explicitly write out the expression for ΨP (x = 0−) as follows ΨP (x = 0−) = D1ϕ+(0−) + D2ϕ−(0−) , (B14) where sin θ i cos θ − sin2 θ cos2 θ cos θ R0 2vF 2 + cos2 θ(cid:1) sin3 θ i(cid:0) 1 cos2 θ R0 vF ϕ+(0−) =  ϕ−(0−) =  cos2 θ sin θ R0 vF 2i cos3 θ R0 vF cos θ −i sin θ , (B15) (B16)     Now we see that in the limit R0 → 0, the total wave- function ΨP (0−) will go back to the form of a superposi- tion of the two decaying wavefunctions given in Eqs. (B4) and (B5). In conclusion, while small Rashba spin-orbit coupling may cause a nonperturbative change of the in- dividual decaying wavefunctions, it modifies the "space" spanned by the two decaying modes in a perturbative manner. It is this "space" which determines the final result of the reflection amplitudes. This is the physical reason why in the final result, the Rashba spin-orbit cou- pling modifies the reflection amplitudes in a perturbative manner. The wavefunction in the potential barrier can be written as ΨB(x) = C1√2 ↑i ⊗(cid:18) 1 −i(cid:19) eγ0x + C2√2 ↑i ⊗(cid:18) 1 i (cid:19) e−γ0x + C3√2 ↓i ⊗(cid:18) 1 i (cid:19) eγ0x + C4√2 ↓i ⊗(cid:18) 1 −i(cid:19) e−γ0x ,(B17) where γ0 = V0/v′ F. 2. An electron incident from the spin-up channel For an electron incident from the spin-up channel, the wavefunction in the electrode is given by ΨE(x) = + 1 √2 ↑i ⊗(cid:18) 1 r↓↑√2 ↓i ⊗(cid:18) 1 −1(cid:19) + −1(cid:19) . 1(cid:19) r↑↑√2 ↑i ⊗(cid:18) 1 (B18) First, matching the wavefunctions Eqs. (B17) and (B18) at x = d, one obtain 9 The forms of the wavefunctions in the pump and barrier remain to be the same. By some algebra, we arrive at ΨB(0+) = 1 √2  r↑↓Γ−(2γ0d) r↑↓Γ+(2γ0d) Γ+(2γ0d) + r↓↓Γ−(2γ0d) Γ−(2γ0d) − r↓↓Γ+(2γ0d)   . (B28) Equating Eq. (B14) with Eq. (B28), we obtain sin θ cos θ (cid:18) −i cos2 θ − sin θ cos θ −i sin2 θ (cid:19)(cid:18) Γ− −Γ+ (cid:19) Γ−(2γ0d) − r↓↓Γ+(2γ0d)(cid:19) + O(ǫ2) =(cid:18) Γ+(2γ0d) + r↓↓Γ−(2γ0d) i R0 vF r↑↓ cos3 θ sin θ (B29) C1 = C2 = C3 = C4 = 1 [(1 − i) + r↑↑(1 + i)]e−γ0d , 2 1 [(1 + i) + r↑↑(1 − i)]eγ0d , 2 r↓↑ (1 + i)e−γ0d , 2 r↓↑ 2 (1 − i)eγ0d . (B19) (B20) It follows from Eq. (B29) r↓↓ = cos(2θ) − i[sh(2γ0d) − sin(2θ)ch(2γ0d)] ch(2γ0d) − sin(2θ)sh(2γ0d) + O(ǫ2) , (B30) (B21) and (B22) r↑↓ = − 1 2 sin(2θ)[1 + cos(2θ)] ch(2γ0d) − sin(2θ)sh(2γ0d)(cid:18) R0 vF(cid:19) + O(ǫ2) . (B31) In the next step, we will match wavefunctions at x = 0. Substituting Eqs. (B19-B22) into Eq. (B17), we can write Eq. (B17) at x = 0+ as ΨB(0+) = 1 √2  Γ+(2γ0d) + r↑↑Γ−(2γ0d) −Γ−(2γ0d) + r↑↑Γ+(2γ0d) r↓↑Γ−(2γ0d) −r↓↑Γ+(2γ0d)   where Γ(ξ) = ch(ξ) ± ish(ξ). Now equating Eq. (B14) with Eq. (B23), we obtain sin θ cos θ (cid:18) i sin2 θ i cos2 θ (cid:19)(cid:18) Γ− −Γ+ (cid:19) − sin θ cos θ −Γ−(2γ0d) + r↑↑Γ+(2γ0d) (cid:19) + O(ǫ2) =(cid:18) Γ+(2γ0d) + r↑↑Γ−(2γ0d) i R0 vF r↓↑ cos θ sin3 θ (B24) It follows from Eq. (B24) r↑↑ = − cos(2θ) + i[sh(2γ0d) − sin(2θ)ch(2γ0d)] ch(2γ0d) − sin(2θ)sh(2γ0d) +O(ǫ2) , (B25) and r↓↑ = ǫ 2 sin(2θ)[1 − cos(2θ)] ch(2γ0d) − sin(2θ)sh(2γ0d) + O(ǫ2) . (B26) 3. An electron incident from the spin-down channel 4. A verification of the result (B23) The total Hamiltonian of the system is invariant under the transformation (−isy σz)H(−ky)(isy σz) = H(ky) , (B32) so the corresponding transformation of Eq. (B18) = 1 1(cid:19) √2 ↓i ⊗(cid:18) 1 (−isy σz)ΨE(x)ky →−ky r↑↑ky→−ky + √2 r↓↑ky→−ky √2 − −1(cid:19) ↓i ⊗(cid:18) 1 1 (cid:19) , ↑i ⊗(cid:18) 1 (B33) must also be an eigenstate of H(ky). This result tells that an electron incident from the spin-down channel will be reflected into the spin-down channel with amplitude r↑↑ky →−ky , and also into the spin-up channel with am- plitude −r↓↑ky →−ky . Comparing it with Eq. (B27) and noticing that ky → −ky is equivalent to 2θ → (π − 2θ), we find immediately the following relations The reflection amplitudes for an electron incident from the spin-down channel can be solved similarly. Now the wavefunction in the electrode is given by and ΨE(x) = 1 √2 ↓i⊗(cid:18) 1 1(cid:19)+ r↓↓√2 ↓i⊗(cid:18) 1 −1(cid:19)+ r↑↓√2 ↑i⊗(cid:18) 1 1(cid:19) . (B27) r↓↓ = r↑↑2θ→(π−2θ) , r↑↓ = −r↓↑2θ→(π−2θ) . (B34) (B35) The reflection amplitudes given in Eqs. (B25), (B26), (B30), and (B31) apparently satisfy these relations. Appendix C: CALCULATION OF THE REFLECTION AMPLITUDES FOR A HALF-METALLIC ELECTRODE To simulate the half-metallic electrode, the Hamilto- nian of the electrode is taken to be HE = v′ F(ky)kxsz σx + V1(1 ∓ sz)σz/2. Consider first the case, where the spin polarization of the electrode is parallel to the z-axis. Now the Hamiltonian of the potential barrier becomes HB = HE +(cid:18) 0 0 V1 σz (cid:19) . 0 (C1) 10 up channel has some probability to be reflected into the spin-down channel from the pump, but the reflected wave will decay quickly to 0 within the barrier, and has no chance to reach the electrode. Therefore, r↓↑ ≡ 0 and the wavefunction in the electrode becomes ΨE(x) = 1 √2 ↑i ⊗(cid:18) 1 −1(cid:19) + r↑↑√2 ↑i ⊗(cid:18) 1 1 (cid:19) . (C2) For simplicity, we will take V1 → ∞ limit in the final result. In this case, an electron incident from the spin- The wavefunction in the potential barrier can be written as ΨB(x) = C1√2 ↑i ⊗(cid:18) 1 −i(cid:19) eγ0x + C2√2 ↑i ⊗(cid:18) 1 i (cid:19) e−γ0x + C3√2 ↓i ⊗(cid:18) 1 −i(cid:19) e−γ1x , (C3) F and γ1 = V1/v′ where γ0 = V0/v′ F → ∞. The wave- function in the pump remains to be same. Repeating the same calculation as in sec. II, it is straightforward to obtain r↑↑ = − cos(2θ) + i[sh(2γ0d) − sin(2θ)ch(2γ0d)] ch(2γ0d) − sin(2θ)sh(2γ0d) +O(ǫ2) . (C4) This expression is identical in form to Eq. (B25) ob- tained in Sec. II for a nonmagnetic electrode. How- ever, an important difference is r↓↑ ≡ 0. As a result, r↑↑2 ≡ (1−r↓↑2) ≡ 1 up to any order in ǫ. This means that O(ǫ2) in Eq. (C4) must be a correction only to the argument of r↑↑, which as discussed in the manuscript, will not modify the orbit of r↑↑ on the complex plane. The orbit is always a unit circle for ky < kc y, and the amount of charge pumped per cycle by the ky state is quantized to ∆q(ky) = −eC+. Similarly, for the spin po- larization of the electrode antiparallel to the z-axis, the charge pumped per cycle is ∆q(ky) = −eC−. [1] K. Klitzing, G. Dorda, and M. Pepper, Phys. Rev. Lett. 45, 494 (1980). [12] I. Knez and R. R. Du, Frontiers of Phys. 7, 200 (2012). [13] J. E. Moore and L. Balents, Phys. Rev. B 75, 121306 (R) [2] A. Tsukazaki, A. Ohtomo, T. Kita, Y. Ohno, H. Ohno, (2007). and M. Kawasaki, Science 315, 1388 (2007). [3] Y. Zhang, Y. W. Tan, H. L. Stormer, and P. Kim, Nature [14] L. Fu and C. L. Kane, Phys. Rev. B 76, 045302 (2007). [15] M. Z. Hasan, and C. L. Kane, Rev. Mod. Phys. 82, 3045 438, 201-204 (2005). (2010). [4] C. Brune, C. X. Liu, E. G. Novik, E. M. Hankiewicz, H. Buhmann, Y. L. Chen, X. L. Qi, Z. X. Shen, S. C. Zhang, and L. W. Molenkamp, Phys. Rev. Lett. 106 126803 (2011). [5] R. B. Laughlin, Phys. Rev. B 23, 5632 (1981). [6] D. J. Thouless, M. Kohmoto, M. P. Nightingale, and M. den Nijs, Phys. Rev. Lett. 49, 405 (1982). [7] Q. Niu, and D. J. Thouless, J. Phys. A: Math. Gen. 17, 2453 (1984). [8] D. J. Thouless, Phys. Rev. B 27, 6083 (1983). [9] C. L. Kane, and E. J. Mele, Phys. Rev. Lett. 95, 226801 (2005). [16] X. L. Qi and S. C. Zhang, Physics Today 63, 33 (2010). [17] F. D. M. Haldane, Phys. Rev. Lett. 61, 2015 (1988). [18] C. L. Kane and E. J. Mele, Phys. Rev. Lett. 95, 146802 (2005). [19] D. N. Sheng, Z. Y. Weng, L. Sheng, and F. D. M. Hal- dane, Phys. Rev. Lett. 97, 036808 (2006). [20] E. Prodan, Phys. Rev. B 80, 125327 (2009). [21] Y. Yang, Z. Xu, L. Sheng, B. G. Wang, D. Y. Xing, and D. N. Sheng, Phys. Rev. Lett. 107, 066602 (2011). [22] X. L. Qi, T. L. Hughes, and S. C. Zhang, Phys. Rev. B 78, 195424 (2008). [23] A. M. Essin, J. E. Moore, and D. Vanderbilt, Phys. Rev. [10] B. A. Bernevig,and S. C. Zhang, Phys. Rev. Lett. 96, Lett. 102, 146805 (2009). 106802 (2006). [11] 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). [24] L. Fu, and C. L. Kane, Phys. Rev. B 74, 195312 (2006). [25] C. Q. Zhou, Y. F. Zhang, L. Sheng, R. Shen, D. N. Sheng, and D. Y. Xing, Phys. Rev. B 90, 085133 (2014). 11 [26] X. L. Qi, T. L. Hughes, and S. C. Zhang, Nature Physics 4, 273 (2008). [27] F. Mahfouzi, B. K. Nikoli´c, S. H. Chen, and C. R. Chang, [33] G. H. Wannier, Rev. Mod. Phys. 34, 645 (1962). [34] X. L. Qi, Phys. Rev. Lett. 107, 126803 (2011). [35] H. Li, L. Sheng, and D. Y. Xing, Phys. Rev. Lett. 108, Phys. Rev. B 82, 195440 (2010). 196806 (2013). [28] B. A. Bernevig, T. L. Hughes, and S. C. Zhang, Science [36] M. Buttiker, H. Thomas, and A. Pretre, Z. Phys. B 94, 314, 1757-1761 (2006). 133 (1994). [29] M. Konig, H. Buhmann, L. W. Molenkamp, T. L. Hughes, C. X. Liu, X. L. Qi, and S. C. Zhang, J. Phys. Soc. Jpn. 77, 031007 (2008). [30] W. Yang, K. Chang, and S. C. Zhang, Phys. Rev. Lett. 100, 056602 (2008). [31] C. X. Liu, T. L. Hughes, X. L. Qi, K. Wang, and S. C. Zhang, Phys. Rev. Lett. 100, 236601 (2008). [32] D. G. Rothe, R. W. Reinthaler, C. X. Liu, L. W. Molenkamp, S. C. Zhang, and E. M. Hankiewicz, New J. Phys. 12, 065012 (2010). [37] P. W. Brouwer, Phys. Rev. B 58, R10135 (1998). [38] M. Switkes, C. M. Marcus, K. Campman, and A. C. Gos- sard, Science 283, 1905 (1999). [39] F. Giazotto, P. Spathis, S. Roddaro, S. Biswas, F. Tad- dei, M. Governale, and L. Sorba, Nature Physics 7, 857 (2011). [40] G. E. W. Bauer, and Y. Tserkovnyak, Physics 4, 40 (2011). 

1105.5279

2

1105

Transmission through Biased Graphene Strip

We solve the 2D Dirac equation describing graphene in the presence of a linear vector potential. The discretization of the transverse momentum due to the infinite mass boundary condition reduced our 2D Dirac equation to an effective massive 1D Dirac equation with an effective mass equal to the quantized transverse momentum. We use both a numerical Poincare Map approach, based on space discretization of the original Dirac equation, and direct analytical method. These two approaches have been used to study tunneling phenomena through a biased graphene strip. The numerical results generated by the Poincare Map are in complete agreement with the analytical results.

cond-mat.mes-hall

cond-mat

ucd-tpg:1103.05 Transmission through Biased Graphene Strip H. Bahloulia,b, E.B. Choubabia,c, A. El Mouhafida,c and A. Jellal∗a,c,d aSaudi Center for Theoretical Physics, Dhahran, Saudi Arabia bPhysics Department, King Fahd University of Petroleum & Minerals, cTheoretical Physics Group, Faculty of Sciences, Chouaıb Doukkali University, Dhahran 31261, Saudi Arabia PO Box 20, 24000 El Jadida, Morocco dPhysics Department, College of Sciences, King Faisal University, PO Box 9149, Alahsa 31982, Saudi Arabia Abstract We solve the 2D Dirac equation describing graphene in the presence of a linear vector potential. The discretization of the transverse momentum due to the infinite mass boundary condition reduced our 2D Dirac equation to an effective massive 1D Dirac equation with an effective mass equal to the quantized transverse momentum. We use both a numerical Poincar´e Map approach, based on space discretization of the original Dirac equation, and direct analytical method. These two approaches have been used to study tunneling phenomena through a biased graphene strip. The numerical results generated by the Poincar´e Map are in complete agreement with the analytical results. PACS numbers: 73.63.-b; 73.23.-b; 11.80.-m Keywords: Dirac, Graphene, Tunneling, Linear Potential ∗[email protected] -- [email protected] 1 Introduction Graphene, a single layer of carbon atoms laid out in a honeycomb lattice, is one of the most interesting electronic systems discovered in recent years [1,2]. It differs from conventional two dimensional electron gas (2DEG) systems in that the low energy physics is governed by a massless Dirac Hamiltonian rather than the more common form used for semiconductors, characterized by an effective mass and a band gap. The tunneling of Dirac fermions in graphene has already been verified experimentally [3], which in turn has spurred an extraordinary amount of interest in the investigation of the electronic trans- port properties in graphene based quantum wells, barriers, p-n junctions, transistors, quantum dots, superlattices, ··· etc. The electrostatic barriers in graphene can be generated in various ways [4, 5], by applying a gate voltage, cutting it into finite width nanoribbons and using doping or otherwise. On the other hand, magnetic barrier could in principle be realized with the creation of magnetic dots. In the case of graphene, results of the transmission coefficient and the tunneling conductance were already reported for the electrostatic barriers [4 -- 10] and magnetic barriers [11 -- 13]. The fact that in an ideal graphene sheet the carriers are massless gives rise to Klein paradox, which allows particles to tunnel through any electrostatic potential barriers, that is the wavefunction has an oscillatory tail outside the electrostatic barrier region. Hence this property excludes the possibility to confine electrons using electrostatic gates, as in usual semiconductors. Thus to enable the fabrication of confined structures, such as quantum dots, we need to use other type of potential coupling such as the scalar potential coupling [14]. However, in our present work we ensure confinement of our fermions in the y-direction by using infinite mass confinement, which requires infinite mass at the boundary of the y-strip and results in a specific quantization of the y-component of the momentum [14]. For the solution of the electrostatic problem at hand, we proceed in two complementary ways to study the tunneling of Dirac fermions through a biased graphene strip. First, we implement our recent developed Poincar´e map [15], which is very handy and efficient for numerical computation. Second, we use an analytical approach to solve the effective 1D Dirac equation in the presence of an electrostatic barrier. Comparison between the results generated by both approaches shows complete agreement. The paper is organized as follows. In section 2, we describe our theoretical model Hamiltonian and apply the Poincar´e map approach, based on the space discretization of the effective 1D Dirac equation. In section 3, we expose the direct analytical approach to solve the same problem. In section 4, we proceed to discuss the numerical implementation of our approaches to a specific model potential, the linear potential which generates a static electric field, and make a comparative study between the two approaches. 2 Poincar´e map Before we embark on the two approaches mentioned above, we would like to describe mathematically our system of massless Dirac fermions within a strip of graphene characterized by a very large length scale, and a width W in the presence of the applied linear potential V (x) between x = 0 and x = L. So our system is composed of three major regions: the extremes (I) and (III) contain intrinsic graphene free of any external potentials and an intermediate region (II) subject to the applied linear potential 1 V (x). Graphene band structure has two Fermi points, each with a two-fold band degeneracy, and can be described by a tight binding Hamiltonian describing two interlacing honeycomb sublattices. At low energies this Hamiltonian can be can be described by a continuum approximation to the original tight binding model which reduces to the two dimensional Dirac equation with a four-component envelope wavefunction whose components are labeled by a Fermi-point pseudospin = ±1. Specifically, the Hamiltonian for one-pseudospin component for the present system can be written as where vF ≃ 9.84× 106m/s is the Fermi velocity and ~σ = (σx, σy) are the Pauli matrices. Hereafter we set our units such that vF =  = 1. The linear potential V (x) has the following form H = vF ~σ · ~p + V (x) (1) V (x) =( −F x + V0, 0, 0 < x < L otherwise (2) where F = V0 Figure 1 below. L is the strength of the static electric field. This potential configuration is shown in V(x) V0 Ψ0 A0 0 I Ψn n L II ΨN +2 BN +2 AN +2 N+1 x III Figure 1: Discretization of the linear potential V (x). Our system is supposed to have a finite width W with infinite mass boundary conditions for the wavefunction at the boundaries y = 0 and y = W along the y-direction [8,14]. This boundary condition results in a quantization of the transverse momentum along the y-direction, which gives ky = π W (cid:18)l + 1 2(cid:19) , l = 0, 1, 2··· . (3) One can therefore assume a spinor solution of the following form ψj(x, y) = (cid:16)φj eikyy where the superscript j = I, II, III, indicates the space region while the subscripts indicate the two spinor components. Thus our problem reduces to an effective 1D problem whose Dirac equation can be written as 1(x), φj 2(x)(cid:17)† d dx + ky V (x) − ε − d dx + ky V (x) − ε !  2 φj 1(x) −iφj 2(x)   = 0. (4) Due to the space dependence of the potential V (x) we make the following transformation on our spinor components to enable us to obtain Schrodinger like equations for each component, χj 1 + ψj and χj 2(cid:17), which obey the coupled stationary equations. These are 2i(cid:16)ψj 2(cid:16)ψj 1 − ψj 2 = 1 1 = 1 2(cid:17) dχj 1,2(x) dx ± i (V (x) − ǫ) χj 1,2(x) ∓ ikyχj 2,1(x) = 0. (5) Each spinor component χj equation 1,2 can be shown to satisfy the following uncoupled second order differential d2 dx2 χj 1,2 (x) +(cid:18)±i d dx V (x) + [V (x) − ε]2 − ky 2(cid:19) χj 1,2 (x) = 0. (6) In this section we will apply the Poincar´e map approach to solve the above effective 1D Dirac equation. In this approach we start by subdividing the potential interval L into N + 1 regions (Figure 1). In every n-th region we approximate the linear potential by a constant value Vn = V (xn) where xn = nh and h = L N +1 . Hence, the Dirac equation in each region (n), defined by h(n − 1) < x < hn, can be easily solved for the piece-wise constant potential. For simplicity, we chose the incident wave propagating from right to left and apply the continuity of the spinor wavefunctions at the boundary separating adjacent regions. The general solutions of equation (6) in the n-th region where V (x) = Vn are given by n ! e−iknx + Bn 1 ψn = An 1 −z∗ zn ! eiknx (7) with kn = q(ε − Vn)2 − k2 = sgn (ε − Vn) kn+iky√k2 n+k2 In order to obtain the relationship between ψn+1 and ψn we apply continuity of ψ at the boundary x = xn (Figure 2). This leads to y, the complex number zn is defined by zn = 1 z ∗ . n y Mn(xn) An Bn ! = Mn+1(xn) An+1 Bn+1 ! . Also Mn+1(xn+1) and Mn+1(xn) are related by Mn+1(xn+1) = Mn+1(xn)Sn+1, Sn+1 = e−ihkn+1 0 0 eihkn+1 ! . (8) (9) Figure 2: Solutions of the 1D Dirac equation in two consecutive regions, continuity of spinors is applied at x = xn. 3 Using the above results we can write the desired Poincar´e map as ψn+1(xn+1) = τnψn(xn) (10) where we have defined a simplified notation by ψn = ψn(xn) and τn = Mn+1(xn+1)Sn+1M −1 or more explicitly n+1(xn+1), τn = z∗ 1 (1+n) + z(1+n) z∗ (1+n)eik(1+n) + z(1+n)e−ik(1+n) −e−ik(1+n) + e−ik(1+n) −e−ik(1+n) + e−ik(1+n) (1+n)e−ik(1+n) + z(1+n)e−ik(1+n) ! . z∗ (11) To make use of the above Poincar´e map in solving our scattering problem we need to define our incident, reflected and transmitted waves. For x ≤ 0 where V = 0 (region I), we can use for our transmitted spinor evaluated at n = 0, the suitably normalized form 0 ! . ψ0 = 1 −z∗ (12) This is juste the value of the transmitted wave at the zeroth site, x = 0 (n = 0), which is given by ψL = A0 1 0 ! e−ik0x. −z∗ (13) On the other side, for x ≥ h(N + 1) where V = 0 (region III), we have both incident and reflected spinor waves. Just outside the potential region on the right hand side in the (N + 2)-th region the spinor wave can be written as ψR = AN +2 1 0 ! e−ik0x + BN +2 1 −z∗ z0 ! eik0x. (14) Hence to evaluate the transmission amplitude all we need is to find AN +2 using the above recursive scheme. Our strategy now is to express AN +2 in terms of ψN +1 and ψN +2, the two end point spinors. This can be easily done using our previous relationships and leads to AN +2 = eihk0(N +2) 2(1 − e2ihk0)(cid:16) 1 From the above notation we can easily define the transmission amplitude as follows −z0 (cid:17)(cid:16)ψN +2 − eihk0ψN +1(cid:17) . t = 1 AN +2 . (15) (16) Summing up, our numerical procedure requires first that we iterate the Poincar´e map (10) to obtain the end point spinors, ψN +1 and ψN +2, in terms of the normalized transmitted spinor. These spinors will then be injected in (15) and (16) to determine the transmission amplitude. The transmission coefficient is given by T = t2. The numerical implementation of this scheme in the case of linear vector potential will be done in section 4. Before closing this section, we would like to point out that transfer matrix methods have been used heavily in the context of transport in graphene [11] and graphene superlatices [16]. However, the Poincar´e map, which can be of great interest in application related to disordered choatic systems, applies only to discretized systems and has been applied in its present form only recently to the Dirac equation [15]. 4 3 Analytical method Let us now solve analytically the effective 1D Dirac equation or equivalently equation (6) in the presence of an electrostatic barrier (region II). Our objective is to find the transmission coefficient for a Dirac fermion scattered by a linear potential and then compare our results with those found in previous section using the Poincar´e map method. Before we proceed further, we would like to mention that the transmission through a trapezoidal barrier in graphene was analytically calculated by Sonin [17]. However, the exact solution of the Dirac equation in uniform electric field in terms of confluent hypergeometric functions was found long time ago by Sauter [18]. The solution of equation (6) in region I and III are given by φI(x) = 1 z ! eikxx + r 1 −z∗ ! e−ikxx, φIII(x) = t 1 z ! eikxx (17) where r and t are the reflection and transmission amplitudes, respectively. The wave vector kx = y. In region II the y and the complex number z is defined by z = sgn(ε)(kx + iky)/qk2 general solution can be expressed in terms of the parabolic cylinder function [19, 20] as qε2 − k2 x + k2 χII 1 (x) = αDν−1 r 2 F eiπ/4(F x + E)! + βD−ν −r 2 F e−iπ/4(F x + E)! (18) ik2 y ky where ν = 2F , E = ε− V0, α and β are constants. Substituting (18) in (5) gives the other component ky h2(E + F x)D−ν(cid:16)−q 2 F e−iπ/4(F x + E)(cid:17)i 2 (x) = − β χII √2F e−iπ/4Dν(cid:16)q 2 − α F e−iπ/4(F x + E)(cid:17) + √2F eiπ/4D−ν+1(cid:16)−q 2 F eiπ/4(F x + E)(cid:17) . The components of the spinor solution of the Dirac equation (1) in region II can be obtained from (18) and (19) where φII 1 + iχII 2 and φII 2 (x) = χII 2 . This results in 1 (x) = χII (19) 1 − iχII ψII(x) = α a+(x) a−(x) ! + β b+(x) b−(x) ! where the function a±(x) and b±(x) are given by (20) (21) a±(x) = Dν−1 r 2 b±(x) = ± 1 ky √2F ky eiπ/4Dν r 2 e−iπ/4(F x + E)! F F eiπ/4(F x + E)! ∓ √2F e−iπ/4D−ν+1 −r 2 (−2iE ± ky − 2iF x)D−ν −r 2 F F e−iπ/4(F x + E)! . 1 ky ± eiπ/4(F x + E)! The coefficients r, α, β and t are determined from the continuity of the spinor wavefunctions at the boundaries x = 0, L, that is ψI(x = 0) = ψII(x = 0) and ψII(x = L) = ψIII(x = L). The transmission coefficient through the linear potential is obtained from T = t2 where the corresponding amplitude t is obtained from the aforementioned boundary conditions. It is given by t = [b+(0) + zb−(0)] [a−(L) − za+(L)] − [a+(0) + za−(0)] [b−(L) − zb+(L)] e−ikxL(cid:2)1 + z2(cid:3) [b+(L)a−(L) − b−(L)a+(L)] . (22) 5 4 Results and discussion In this section we implement our previous Poincar´e map and analytical approaches to a nanoribbon system subject to an electric potential of strength V0 = 10, 20 and a field region of length L = 3, 10 so that the resulting static electric field strength is given by F = V0/L = 10/3, 2, respectively. In Figure 3 we show the transmission as a function of energy for a transverse momentum ky = 1. The solid lines corresponds to the exact transmission derived in section 3 and given by equation (22) while the dashed lines are generated by our Poincar´e map for N = 200 iterations, the agreement is just perfect. T 1.0 0.8 0.6 0.4 0.2 L = 3 V0 = 10 L = 10 L = 3 T 1.0 0.8 0.6 0.4 0.2 V0 = 20 L = 10 5 10 15 Ε 20 10 20 30 Ε 40 Figure 3:Transmission coefficients T versus energy εfor L = 3, 10, V0 = 10, 20 and ky = 1. Figure 3 shows the concordance between the results generated by the analytical and Poincar´e map method we adapted. We note that below a certain critical energy ε = ky the transmission is almost zero, then it starts oscillations whose frequency increases with L, the size of the region subject to the electric field. The transmission increases with L and reaches unity for energies above V0 + 2ky. T 1.0 0.8 0.6 0.4 0.2 0.0 0 L = 3 Ε = 10 L = 10 5 10 15 20 V0 25 T 1.0 0.8 0.6 0.4 0.2 0.0 0 L = 3 L = 10 Ε = 20 10 20 30 V0 40 Figure 4:Transmission coefficients T versus V0 for L = 3, 10, ε = 10, 20 and ky = 1 Figure 4 shows the transmission as a function of the strength of the applied voltage, total transmission is observed for small values of V0 less than the energy of the incident fermion. It then decreases sharply for V0 > ε − 2ky until it reaches a relative minimum and then begins to increase in an oscillatory manner. We notice in both Figures 3 and 4 that the amplitude of oscillations and period increase as we decrease the size of the electric field region, L. 6 T 1.0 0.8 0.6 0.4 0.2 0.0 0 ky = 1 ky = 2 ky = 3 10 20 30 Ε 40 Figure 5:Transmission coefficients T versus energy ε for L = 3, V0 = 20 and different values of ky. Figure 5 shows that the effect of the transverse momentum ky on transmission, is antagonistic to that of length L. But it should be pointed out that the number of oscillations increases as ky decreases and the curves for different values of ky do not intersect. Now we would point out that our effective 1D massless Dirac equation is equivalent to a massive one with an effective mass equal to the transverse quantized wave vector ky. For this purpose we would like to consider a unitary transformation, which enable us to map the effective 1D (equation (4)) into a 1D massive Dirac equation. Such a unitary transformation does not affect the energy spectrum π 4 σ2. Thus, the or the physics of the problem. We choose a rotation by π transformed Hamiltonian and wavefunction read 2 about the y-axis, U = ei = 0, ψj1,2(x) = U ψj1,2(x) (23) V (x) − ε + ky − d dx d dx V (x) − ε − ky !  ψj 1(x) ψj 2(x)   which is identical to a 1D massive Dirac equation with an effective mass m = ky. To check the validity of this assertion numerically we show in Figure 6 the transmission as a function of energy as generated by the exact analytical result (22), the Poincar´e map (16) and the 1D massive Dirac equation with an effective mass m∗ = ky in (23). We see from this figure that the three curves coincide to the point that we cannot even distinguish between them. This lead us to include an inset in Figure 6 showing each figure translated for ease of comparison purposes. T 1.0 0.8 0.6 0.4 0.2 0.0 0 Exact 1 DM 2 DMS 10 20 10 30 20 30 40 40 50 50 Ε 60 Ε 60 Figure 6 :Transmission coefficients T (ε) for L = 1, V0 = 40 and ky = m∗ = 9π 10 . This last figure confirms, numerically, the equivalence between a one-dimensional system of Dirac fermions with mass and a two-dimensional system of massless Dirac fermions constrained along the 7 y-direction by an infinite mass boundary condition, hence forming a graphene nanoribbon. The transverse component of the wave vector, ky, played the role of an effective mass [21] in the resulting effective 1D Dirac equation. To close this section we would like to mention that our present work could be extended to handle a system of 2D Dirac fermions with mass m as done in reference [22]. Once confined to a strip along the y. y-direction we will end up with an effective 1D Dirac equation with effective mass meff =qm2 + k2 This might be considered as a simple extension, which is useful to model an underlying substrate. Acknowledgments The generous support provided by the Saudi Center for Theoretical Physics (SCTP) is highly appre- ciated by all Authors. AJ and (EBC, AE) acknowledge partial support by King Faisal University and KACST, respectively. We also acknowledge the support of KFUPM under project RG1108-1-2. We would to express our deep appreciation for the very constructive comments made by the referee. References [1] A. K. Geim and K. S. Novoselov, Nat. Mat. 6, 183 (2007). [2] A. H. C. Neto, F. Guinea, N. M. R. Peres, K. S. Novoselov and A. K. Geim, Rev. Mod. Phys. 81, 109 (2009). [3] N. Stander, B. Huard and D. G. Gordon, Phys. Rev. Lett. 102, 026807 (2009). [4] M. I. Katsnelson, K. S. Novoselov and A. K. Geim, Nature Phys. 2 620 (2006). [5] H. Sevin¸cli, M. Topsakal and S. Ciraci, Phys. Rev. B 78, 245402 (2008). [6] L. Dell'Anna and A. De Martino, Phys. Rev. B 79, 045420 (2009). [7] S. Mukhopadhyay, R. Biswas and C. Sinha, Phys. Status Solidi B 247, 342 (2010). [8] J. Tworzydlo, B. Trauzettel, M. Titov, A. Rycerz and C. W. J. Beenakker, Phys. Rev. Lett. 96, 246802 (2006). [9] A. D. Alhaidari, H. Bahlouli and A. Jellal, Relativistic Double Barrier Problem with Three Sub- Barrier Transmission Resonance Regions, arXiv:1004.3892. [10] K. S. Novoselov, E. McCann, S. V. Morozov, V. I. Falko, M. I. Katsnelson, U. Zeitler, D. Jiang, F. Schedin and A. K. Geim, Nature Phys. 2, 177 (2006). [11] M. R. Masir, P. Vasilopoulos and F. M. Peeters, New J. Phys. 11, 095009 (2009). [12] E. B. Choubabi, M. El Bouziani and A. Jellal, Int. J. Geom. Meth. Mod. Phys.7, 909 (2010). [13] A. Jellal and A. El Mouhafid, J. Phys. A: Math. Theo. 44, 015302 (2011). [14] M. V. Berry and R. J. Modragon, Proc. R. Soc. London Ser. A 412, 53 (1987). 8 [15] H. Bahlouli, E. B. Choubabi and A. Jellal, Solution of One-dimensional Dirac Equation via Poincar´e Map, arXiv:1105.4741, to appear in Europhys. Lett (2011). [16] Y.P. Bliokh, V. Freilikher, S. Savel´ev and F. Nori, Phys. Rev. B 79, 075123 (2009). [17] E.B. Sonin, Phys. Rev. B 79, 195438 (2009). [18] F. Sauter, Zeitschrift fur Physik 69, 742 (1931). [19] M. Abramowitz and I. Stegum, Handbook of Integrabls, Series and Products, (Dover, New York, 1956). [20] L. Gonzalez-Diaz and V. M. Villalba, Phys. Lett. A 352, 202 (2006). [21] A. D. Alhaidari, A. Jellal, E. B. Choubabi and H. Bahlouli, Mass Generation via Space Com- pactification in Graphene, arXiv:1010.3437. [22] M. Barbier, F.M. Peeters, P. Vasilopoulos and J. Milton Pereira, Phys. Rev. B 77, 115446 (2008). 9 

1612.02187

2

1612

Field emission: the theoretical link between voltage loss, reduction in field enhancement factor, and Fowler-Nordheim-plot saturation

With a large-area field electron emitter, when an individual post-like emitter is sufficiently resistive, and current through it sufficiently large, then voltage loss occurs along it. This Letter provides a simple analytical and conceptual demonstration that this voltage loss is directly and inextricably linked to a reduction in the field enhancement factor (FEF) at the post apex. A formula relating apex-FEF reduction to this voltage loss was obtained in the paper by E. Minoux, O. Groening, K. B. K. Teo, S. H. Dalal, L. Gangloff, J.-P. Schnell, L. Hudanski, I. Y. Y. Bu., P. Vincent, P. Legagneux, G. A. J. Amaratunga, and W. I. Milne [Nano Lett. 5, 2135 (2005)], by fitting to numerical results from a Laplace solver. This Letter derives the same formula analytically, by using a "floating sphere" model. The analytical proof brings out the underlying physics more clearly, and shows that the effect is a general phenomenon, related to reduction in the magnitude of the surface charge in the most protruding parts of an emitter. Voltage-dependent FEF-reduction is one cause of "saturation" in Fowler-Nordheim plots. Another is a voltage-divider effect, due to measurement-circuit resistance. An integrated theory of both effects is presented. Both together, or either by itself, can cause saturation. Experimentally, if saturation occurs but voltage loss is small (< 20 V, say), then saturation is more probably due to FEF-reduction than voltage division. In this case, existing treatments of electrostatic interaction ("shielding") between closely spaced emitters may need modification. Other putative causes of saturation exist, so the present theory is a partial story. Its extension seems possible, and could lead to a more general physical understanding of the causes of FN-plot saturation.

cond-mat.mes-hall

cond-mat

arXiv: 1612.02187v2: Corrected Revised version: 10 March 2017 The theoretical link between voltage loss, reduction in field enhancement factor, and Fowler- Nordheim-plot saturation Richard G. Forbesa) Advanced Technology Institute & Department of Electrical and Electronic Engineering, University of Surrey, Guildford, Surrey, GU2 7XH, United Kingdom (Submitted 6 December 2016; revised 5 March 2017 With a large-area field electron emitter, when an individual post-like emitter is sufficiently resistive, and current through it sufficiently large, then voltage loss occurs along it. This Letter provides a simple analytical and conceptual demonstration that this voltage loss is directly and inextricably linked to a reduction in the field enhancement factor (FEF) at the post apex. A formula relating apex- FEF reduction to this voltage loss was obtained in the paper by E. Minoux, O. Groening, K. B. K. Teo, S. H. Dalal, L. Gangloff, J.-P. Schnell, L. Hudanski, I. Y. Y. Bu., P. Vincent, P. Legagneux, G. A. J. Amaratunga, and W. I. Milne [Nano Lett. 5, 2135 (2005)] by fitting to numerical results from a Laplace solver. This Letter derives the same formula analytically, by using a "floating sphere" model. The analytical proof brings out the underlying physics more clearly, and shows that the effect is a general phenomenon, related to reduction in the magnitude of the surface charge in the most protruding parts of an emitter. Voltage-dependent FEF-reduction is one cause of "saturation" in Fowler-Nordheim (FN) plots. Another is a voltage-divider effect, due to measurement-circuit resistance. An integrated theory of both effects is presented. Both together, or either by itself, can cause saturation. Experimentally, if saturation occurs but voltage loss is small (< 20 V, say), then saturation is more probably due to FEF-reduction than voltage division. In this case, existing treatments of electrostatic interaction ("shielding") between closely spaced emitters may need modification. Other putative causes of saturation exist, so the present theory is a partial story. Its extension seems possible, and could lead to a more general physical understanding of the causes of FN-plot saturation. 1 ______________________________________ a)Electronic mail: [email protected] The last twenty years have seen much interest in possible applications of large-area field electron emitters (LAFEs), especially those based on carbon nanotubes (CNTs), or, more recently, carbon nanofibers (CNFs) (e.g., Refs 1-3]). One form of ideal LAFE can be visualised as a regular array of near-identical post-like emitters, standing upright on a flat cathode plate, called here the "emitter plate". To model a single emitter, the "hemisphere-on-cylindrical post" (HCP) physical model is often used. This takes the emitter as a cylindrical classical conductor, of radius r and length ℓ, capped by a conducting hemisphere also of radius r, as illustrated in Fig. 1a. Points "a" and "c" label the emitter apex and a point on the circle of join between cap and cylinder, respectively. For simplicity, all model surfaces (both post and emitter plate) are given the same work function φ. FIG. 1. (a) The hemisphere-on-cylindrical-post (HCP) physical model for a field emitting carbon nanotube or nanofiber. (b) The related "floating sphere" model used as an approximation to the HCP model. These diagrams are not to scale: normally the ratio of height ℓ to radius r is much greater than is shown. The quantity ΔΦ is the difference in electrostatic potential between point "c" and the emitter plate (EP). For other nomenclature, see text. 2  It is assumed that, in the absence of the emitter, there would be a uniform classical electrostatic field EM (the macroscopic field) in space above the emitter plate. In this paper, as in Ref. 4, classical electrostatics is used, and the positive field direction is taken as from the emitter plate into vacuum; thus, for a field electron emitter, values of fields and charges below are negative. As in Ref. 4, the reference zero for electrostatic potential Φ is taken to be the potential ΦEP at a point on the emitter plate far distant from the emitter location. When the HCP-model emitter is present, then the total field Ea at its apex is enhanced relative to the macroscopic field, and a macroscopic apex field enhancement factor (FEF) γa is defined by: γa ≡ Ea / EM . (1) There is no known exact analytical method of determining γa, but many approximate analytical methods, and also numerical methods, have been used to estimate γa (for example, see Refs 4-8). Physically, nearly all these derivations take the whole emitter plate to be at constant electrostatic potential ΦEP= 0, and take the whole post surface to be at the same constant potential ΦEP. In particular, they make Φa = Φc = 0. This is equivalent to assuming that the effects of any current through the emitter may be disregarded, and that the Fermi level (and hence the thermodynamic voltage9) may be taken constant throughout the emitter and emitter plate. This can be called the small- current electrostatic approximation (SCEA). The related apex-FEF value is denoted here by γa sc. However, when the emitter is sufficiently resistive and the current sufficiently high, then voltage loss ("Fermi-level variation") occurs along the post. (The term "voltage loss" refers to the fact that the voltage between the emitter apex and the counter-electrode will be less than the measured voltage provided by the high-voltage generator.) A simple analytical model is used below to show clearly that this voltage loss (Vd) along the post is directly and inextricably associated with a reduction in apex FEF, with both effects being caused by a reduction in the magnitude of the electric charge in the emitter apex region. This voltage loss is associated with a non-zero difference ΔΦ in electrostatic 3 potential between point "c" and the emitter plate. The link between voltage loss and FEF reduction was pointed out by Groening et al.10 in their 1999 paper. In a 2005 paper11,Minoux et al. found a relationship between apex-FEF reduction and voltage loss along the emitter by solving Laplace's equation numerically, and fitting a formula to the results. For the case of no contact resistance at the plate/emitter interface, Minoux et al. inferred a relation that (in the notation used here) can be written γa = γa sc [1 – Vd/ΕMℓ] . (2) The discussion here reaches the same formula, but the simple analytical model used here brings out the underlying physics more clearly. Further, the argument here can be generalised qualitatively, to apply to emitters of any shape across which current-related voltage loss occurs. In the SCEA, a simple method of deriving a formula for the apex FEF γa sc uses the so-called "Floating Sphere at Emitter Plate Potential" model, as illustrated in Fig. 1b, taking ΔΦ=0. This model, and its relation to electron thermodynamics, were recently reviewed4. Several levels of mathematical approximations are possible, all of which provide qualitative understanding and get qualitative trends correct, but none of which yields precisely accurate estimates of γa sc. A particularly simple mathematical approximation (Approach II in Ref. 4) is used here to demonstrate the physics of what is happening. The methodology behind this kind of modelling is as follows. First place charges and dipoles at appropriate locations and choose their values so that the potentials at one or two specified locations (here "a" and "c") have the values desired physically. Then use these charges (and the macroscopic field) to estimate the total field at the emitter apex "a", and hence the apex FEF. With the SCEA, the method proceeds as follows4-6. First, a dipole is placed at the centre of the floating sphere, of strength p such that p/(4πε0r2) = rEM. This ensures that, when the sphere is immersed in the field ΕM, its surface is an equipotential, and in particular that Φc = Φa. A charge is then placed at the sphere centre, of strength qsc such that 4  ΔΦ ≡ (Φc – ΦEP) = qsc/4πε0r – ΕMℓ = 0 . (3) This ensures that the floating sphere is at potential ΦEP (if image contributions are neglected). At the emitter apex, this "sphere charge" qsc creates a field contribution Εa,q given, using (3), by Ea,q = qsc/4πε0r2 = (ℓ/r) EM . (4) In a fuller treatment4,6 there would be other contributions, associated with the images of the sphere charge and dipole in the emitter plate, but––for typical experimental values of the ratio (ℓ/r)–– often 100 or more––the resulting contributions to the total apex field Ea are negligible in comparison with the sphere contribution Ea,q. This leaves just the contributions to Ea resulting from the sphere charge, the sphere dipole and the macroscopic field. However, under normal circumstances, the Ea,q term is much larger than the other two (see Ref. 4) and we reach the well-known result sc = Εa/EM ≈ Ea,q /EM = (ℓ/r) . γa (5) In the case where significant current flows, and conditions are such that a significant voltage loss Vd occurs, the above treatment is easily modified, as follows. For simplicity, it is assumed that the whole voltage loss occurs along the cylinder of the HCP model. In classical electromagnetism, the apex of a field electron emitter is more positive that the emitter plate, and Vd is positive. Because the work function is taken the same for all surfaces, this corresponds to an electrostatic potential difference (PD) ΔΦ along the cylinder surface given by ΔΦ = Vd . It is useful to write ΔΦ as a fraction k of the (positive) PD (–EMℓ) induced between "c" and the emitter plate by the macroscopic field; hence ΔΦ = Vd = –kEMℓ . (6) 5  To allow this PD to be present in the floating-sphere model, the magnitude of the (negative) sphere charge has to be reduced. Equation (3) above has to be replaced, and the sphere charge needs to be changed (by a positive amount Δq) from qsc to the value q given via ΔΦ = q/4πε0r – ΕMℓ = – kEMℓ , q/4πε0r = (1–k)EMℓ . (7) (8) From the same argument as before, that the apex field Ea is dominated by the field contribution due to the sphere charge q (provided, in this case, that k is not very close to unity) it follows that, when the SCEA is abandoned, the apex field and FEF are normally given adequately by Ea ≈ q/4πε0r2 = (1–k)(ℓ/r) EM ≈ (1–k) γa sc EM , γa = Ea /EM ≈ (1–k) γa sc = γa sc {1 – Vd/(–ΕMℓ)} ≡ Θfrγa sc . (9) (10) where the correction factor Θfr [= γa/γa sc = 1–Vd/(–ΕMℓ)] is to be attributed to "FEF reduction (fr)". Since EM is negative for a field electron emitter, eq. (10) is the same result as eq. (2) above found11 by fitting to numerical simulation. However, it is explicitly clear here that the sphere-charge q appears in expressions both for the apex field and for the potential difference between the cylinder ends, and hence for the "voltage loss along the emitter". It follows that FEF reduction and current- induced voltage loss along the emitter are directly and inextricably linked, physically. The use of a simple analytical model has led to an explicit formula. However, it is clear that–– qualitatively––the effect is a general one, applicable to an emitting protrusion of any shape, and not dependent on the size of the emitter apex or on the precise local geometry or nature of emission sites. When current through the protrusion leads to significant voltage loss (in accordance with Ohm's law), 6 then this voltage loss is associated with a reduction in the magnitude of the charge in the most vacuum facing parts of the protrusion, and hence with reductions in the magnitude of the local barrier field and FEF there. Although the floating-sphere model allows the basic physics of current-induced FEF-reduction to be displayed, allows eq. (2) to be retrieved, and allows the above qualitative conclusions to be drawn, it needs to be emphasised that it is not (and is not intended to be) a quantitatively accurate model. In particular, it will not deal accurately with situations where the emitter is cone-shaped rather than post- shaped, or where the electrical resistance is non-uniformly distributed (as would occur if most of the resistance is across a poor contact between the emitter and the substrate). In such cases, more sophisticated modelling in needed. Thus, Minoux et al. find11 that, when most of the resistance is in the contact, a correction factor α has to be included, and the r.h.s. of eq. (2) becomes γa sc[1–αVd/ΕMℓ] (they find α=0.92). FIG. 2. Schematic diagram that illustrates the definition of the (positive) quantity ΔL given by eq. (15). A practical context in which FEF-reduction issues arise is the explanation of "saturation" in Fowler-Nordheim (FN) plots, which causes the plots to adopt the kinked form illustrated schematically in Fig. 2. It is readily shown that eq. (10) leads to this effect. This is true for any Fowler-Nordheim-type (FN-type) equation. The exponent –GGB of the "general form" FN-type equation12 is 7  –GGB = –νF GB bφ3/2/Ea , (11) where b is the second FN constant13, and νF GB is the relevant barrier form correction factor12. The (negative) macroscopic field EM is related to the (positive) voltage Vp applied between a counter-electrode ("anode", with field electron emission) and the emitter plate, by EM = –Vp/ζM. Here, ζM is the relevant (positive) macroscopic conversion length. In a given practical arrangement, ζM is a constant that depends on the system geometry; in planar-parallel-plate geometry ζM is adequately given by the plate separation. Thus, the relationship between Ea and Vp becomes Ea = (γa/ζM)Vp = (Θfrγa sc/ζM)Vp . (12) A complication arises if the emitter plate is itself sufficiently resistive that the voltage Vp between the plate front surface (facing vacuum) and the anode is not equal to the measured voltage Vm. In this case, a "voltage divider (vd) effect" will occur12, and the two voltages will be related by Vp = ΘvdVm , where Θvd is a correction factor that is current dependent and will lie in the range 0<Θvd≤1. In this more general case, the relationship between Ea and Vm can be written Ea = (Θγa sc/ζM)Vm ≡ cmVm , (13) where the total correction factor Θ = ΘvdΘfr , and cm [≡Θγa sc/ζM] is an auxiliary parameter introduced to simplify equation presentation. It follows that, in FN coordinates of type ln{im/Vm 2} vs 1/Vm, the general-form FN-type equation12 can be written L ≡ ln{im/Vm 2} = ln{Af aφ–1cm 2} – νF GBbφ3/2/cmVm , (14) where a is the first FN constant13, and Af is the emitter's formal emission area12 as defined in the context of this equation. 8 In circumstances where Θ effectively has the value 1 (which is usually the case for low measured voltages), cm has the constant value cm sc = γa sc/ζM. However, from eq. (12), if at higher measured voltages, Θ becomes progressively smaller, then cm will become progressively smaller. As illustrated in Fig. 2, let ΔL denote the difference between the quantity L(Θ=1) evaluated using the SCEA and the quantity L(Θ<1). It is readily shown that ΔL is predicted adequately by: ΔL ≈ 2ln{cm sc/cm} + (νF GBbφ3/2/cm scVm){(cm sc/cm) –1} . (15) Clearly, as Vm increases (and 1/Vm decreases), then cm progressively decreases below cm sc, and (cm sc/cm) becomes progressively greater than 1. The effect is that both terms in eq. (15), and hence ΔL, progressively increase as (1/Vm) decreases. This is the effect often called "saturation". The actual calculation of values for cm (or of the shape of saturated FN plots) is non-trivial and is outside the scope of this letter. The treatment here, in terms of Θ=ΘvdΘfr, shows that saturation can be caused either by a voltage- divider effect or by FEF reduction, or by both acting together. Both effects are due to significant series resistance in the measurement circuit. The former occurs when the resistance is in the emitter plate itself (for example, if the plate is poorly conducting silicon, or if some other form of ballast resistance has been designed in), the latter when the resistance is in the emitting protrusion or the contact between protrusion and plate. Most past discussions of series-resistance effects on FN plots have concentrated on the voltage- divider explanation. However, simulations of voltage-divider effects by the present author and Deane (see Ref. 12) did not lead to plausible-looking simulated plots. By contrast, the simulations11 of Minoux et al, using the FEF-reduction explanation, did generate plausible-looking FN plots. This leads the author to suspect that, for LAFEs, FEF reduction may usually be the more plausible explanation. The following illustrative argument is a further indicator. Consider a resistive emitting post with φ=4.5 eV, radius r= 10 nm, length ℓ = 1 µm, and take the small-current FEF γa sc to be 100. Such a post 9 would field emit significantly at a barrier field Ea of around –4 V/nm, and (if no series-resistance effects occur) a macroscopic field EM around –40 V/µm. The related value of (–EMℓ) would be 40 V. To get a value Θfr=0.5, one would need a voltage loss (along the post) of 20 V. By contrast, an illustrative value for Vm (assuming a typical value of ζM as 25 µm) might be 2000 V. To get a value Θvd=0.5 by the voltage-divider effect, one would need a voltage loss of 1500 V across the series resistance associated with the emitter plate and the rest of the path to the high-voltage generator. The large difference between 1500 V and 20 V makes it look plausible that, in many cases where series resistance can be presumed responsible for saturation, and especially with LAFEs, the detailed cause is more likely to be FEF reduction than the voltage-divider effect. On the other hand, with other situations, for example flash memory devices where electrons are field-injected into an oxide layer, presumably from metallic nanoprotrusions on metal electrodes, the voltage-divider explanation may look more relevant.14 Of course, field electron emission into vacuum from metallic emitters with a good conducting path to the high-voltage generator represents a situation where effectively Θ=1 over the whole working range, and the SCEA applies throughout the range. Experimental estimates of the total voltage loss across the whole series resistance are possible in principle, either by putting a metal probe in direct contact with the emitter, or by retarding potential energy analysis of field emitted electrons (e.g., Refs. 10,15,16). The measurement of a relatively small total voltage loss associated with the saturated part of the im(Vm) characteristics would be an indication that saturation is probably due to a FEF-reduction effect. Minoux et al. used a direct- contact probe in an experiment of this type, finding Vd values between 0 and 4 V. With LAFEs, the quantitative theory above applies to single emitters that do not significantly interact electrostatically with adjacent emitters, because they are sufficiently well spaced. However, FEF-reduction effects will also occur when electrostatic interaction (usually called "shielding") takes place between emitters. Recently, many papers discussed electrostatic interactions between closely- spaced emitters (e.g., Refs 4,7,8), all using the SCEA. In physical situations where the SCEA is not valid, numerical results relating to electrostatic interactions may need modifying. 10 The theory above does not cover all putative causes of saturation-like effects. Other putative causes include voltage-dependent relative changes in work-function or operative work function, either as a result of adsorbate behaviour (perhaps influenced by joule heating), or––with semiconductors–– as a result of the kind of field penetration and band-bending effects that occur in the Modinos17 "zero- current approximation". Hopefully it may be possible, as some future point, to extend the present theory to cover some or all of these effects, by relaxing the constant-work-function assumption made (for simplicity) in this Letter. Finally, I suggest that kinked FN plots as indicated in Fig, 2, should be regarded, not as evidence of some sort of anomaly, but as the actual current-voltage characteristics (presented in FN coordinates) of an electronic circuit device that is basically a form of diode. If you need a device with different characteristics, then try modifying its structure or the materials from which it is built. (In fact, this is what was done by Minoux et al., who eliminated their voltage-loss and FEF-reduction effects by thermal processing of their carbon nanotubes, thereby reducing their resistivity.) I thank the University of Surrey for provision of facilities, and thank Professor S.R.P. Silva for reinforcing my view that field electron emitters, especially LAFEs, could usefully be thought of as electronic-circuit elements with measured current-voltage characteristics. 1M. T. Cole, M. Mann, K. B. K. Teo, and W. I. Milne, "Engineered carbon nanotube field emission devices", Chap. 5 in: W. Ahmad and M. J. Jackson (eds) Emerging Nanotechnologies for Manufacturing (Second Edition) (Elsevier, 2015). 2Y. Li, Y. Sun, and J. T. W. Yeo, Nanotechnology 26, 242001 (2015). 3N. Shimoi and S.-I. Tanaka, ACS Appl. Mater. Interfaces 5, 768 (2013). 4R. G. Forbes, J. Appl. Phys. 120, 054302 (2016). 5R. G. Forbes, C. J. Edgcombe, and U. Valdrè, Ultramicroscopy 97, 57 (2003). 6A. I. Zhbanov, E. G. Pogorelov, Y.-C. Chang and Y.-G. Lee, J. Appl. Phys. 110, 114311 (2011). 7J. R. Harris, K. L. Jensen, and D. A. Shiffler, AIP Advances 5, 087182 (2015). 11 8W. W. Tang, D. A. Shiffler, J. R. Harris, K. L. Jensen, K. Golby, M. LaCour, and T. Knowles, AIP Advances 6, 095007 (2016). 9"Thermodynamic voltage" is voltage V as defined in terms of a difference ΔΕF of local Fermi levels, by V = –ΔΕF/e, where e is the elementary positive charge, and as measured by galvanometric and digital voltmeters. 10O. Gröning, O. M. Küttel, P. Gröning, and L. Schlapbach, J. Vac. Sci. Technol. B 17, 1970 (1999). 11E. Minoux, O. Groening, K. B. K. Teo, S. H. Dalal, L. Gangloff, J.-P. Schnell, L. Hudanski, I. Y. Y. Bu., P. Vincent, P. Legagneux, G. A. J. Amaratunga, and W. I. Milne, Nano Lett. 5, 2135 (2005). 12R. G. Forbes, J. H. B. Deane, A. Fischer, and M. S. Mousa, Jordan J. Phys. 8, 125 (2015); arXiv:1504.06134v7. 13R. G. Forbes and J. H. B. Deane, Proc. R. Soc. Lond. A 467, 2927 (2011). See related electronic supplementary material on Royal Society website for details of universal constants used in field emission. 14E. Miranda, Electronics Lett. 40, 1153 (2004). 15M. J. Fransen, Th. L. van Rooy, and P. Kruit, Appl. Surf. Sci. 146, 312 (1999). 16M. S. Mousa and T. F. Kelly. Surf. Interface Anal. 36, 444 (2004). 17A. Modinos, Field, Thermionic and Secondary Electron Emission Spectroscopy (Plenum, New York, 1984; reprinted by: Springer, New York, 2013). 12 

1607.04036

2

1607

Nature of excitons bound to inversion domain boundaries: Origin of the 3.45-eV luminescence lines in spontaneously formed GaN nanowires on Si(111)

We investigate the 3.45-eV luminescence band of spontaneously formed GaN nanowires on Si(111) by photoluminescence and cathodoluminescence spectroscopy. This band is found to be particularly prominent for samples synthesized at comparatively low temperatures. At the same time, these samples exhibit a peculiar morphology, namely, isolated long nanowires are interspersed within a dense matrix of short ones. Cathodoluminescence intensity maps reveal the 3.45-eV band to originate primarily from the long nanowires. Transmission electron microscopy shows that these long nanowires are either Ga polar and are joined by an inversion domain boundary with their short N-polar neighbors, or exhibit a Ga-polar core surrounded by a N-polar shell with a tubular inversion domain boundary at the core/shell interface. For samples grown at high temperatures, which exhibit a uniform nanowire morphology, the 3.45-eV band is also found to originate from particular nanowires in the ensemble and thus presumably from inversion domain boundaries stemming from the coexistence of N- and Ga-polar nanowires. For several of the investigated samples, the 3.45-eV band splits into a doublet. We demonstrate that the higher-energy component of this doublet arises from the recombination of two-dimensional excitons free to move in the plane of the inversion domain boundary. In contrast, the lower-energy component of the doublet originates from excitons localized in the plane of the inversion domain boundary. We propose that this in-plane localization is due to shallow donors in the vicinity of the inversion domain boundaries.

cond-mat.mes-hall

cond-mat

Nature of excitons bound to inversion domain boundaries: Origin of the 3.45-eV luminescence lines in spontaneously formed GaN nanowires on Si(111) Carsten Pfuller,∗ Pierre Corfdir,† Christian Hauswald,‡ Timur Flissikowski, Xiang Kong,§ Johannes K. Zettler,¶ Sergio Fern´andez-Garrido, Pınar Dogan,∗∗ Holger T. Grahn, Achim Trampert, Lutz Geelhaar, and Oliver Brandt Paul-Drude-Institut fur Festkorperelektronik, Leibniz-Institut im Forschungsverbund Berlin e.V., Hausvogteiplatz 5 -- 7, 10117 Berlin, Germany 6 1 0 2 t c O 9 1 ] l l a h - s e m . t a m - d n o c [ 2 v 6 3 0 4 0 . 7 0 6 1 : v i X r a We investigate the 3.45-eV luminescence band of spontaneously formed GaN nanowires on Si(111) by pho- toluminescence and cathodoluminescence spectroscopy. This band is found to be particularly prominent for samples synthesized at comparatively low temperatures. At the same time, these samples exhibit a peculiar morphology, namely, isolated long nanowires are interspersed within a dense matrix of short ones. Cathodolumi- nescence intensity maps reveal the 3.45-eV band to originate primarily from the long nanowires. Transmission electron microscopy shows that these long nanowires are either Ga polar and are joined by an inversion domain boundary with their short N-polar neighbors, or exhibit a Ga-polar core surrounded by a N-polar shell with a tubular inversion domain boundary at the core/shell interface. For samples grown at high temperatures, which exhibit a uniform nanowire morphology, the 3.45-eV band is also found to originate from particular nanowires in the ensemble and thus presumably from inversion domain boundaries stemming from the coexistence of N- and Ga-polar nanowires. For several of the investigated samples, the 3.45-eV band splits into a doublet. We demonstrate that the higher-energy component of this doublet arises from the recombination of two-dimensional excitons free to move in the plane of the inversion domain boundary. In contrast, the lower-energy component of the doublet originates from excitons localized in the plane of the inversion domain boundary. We propose that this in-plane localization is due to shallow donors in the vicinity of the inversion domain boundaries. I. INTRODUCTION The promising optoelectronic properties of spontaneously formed GaN nanowires (NWs) reported in the pioneering works of Yoshizawa et al. [1] and S´anchez-Garc´ıa et al. [2] have triggered world-wide research activities that have led to the demonstration of light-emitting [3] and light-harvesting devices [4, 5] based on group-III-nitride NWs. Despite this progress, several open questions still exist regarding the spontaneous formation of GaN NWs and their structural and optical properties. In particular, a prominent band at 3.45 eV has been widely reported in the low-temperature photolu- minescence (PL) spectra of GaN NWs grown by plasma- assisted molecular beam epitaxy (PAMBE) on Si(111) [6 -- 11]. The origin of this band in GaN NWs has been a subject of a lively debate for almost two decades. In bulk GaN, two different recombination mechanisms are known to manifest themselves by luminescence lines at about 3.45 eV: first, the two-electron satellite (TES) of the donor-bound exciton tran- sition [(D0,XA)] [12 -- 15] and, second, excitons bound to in- version domain boundaries [16]. The intensity of the TES transitions in bulk GaN is about two orders of magnitude lower than that of the related (D0,XA) line [15]. In contrast, the 3.45-eV band in GaN NWs is often prominent and sometimes even dominates the near band-edge PL spectrum. Nevertheless, this band was ascribed to the TES by Corfdir et al. [9], who proposed that the distortion of the (D0,XA) wave function near the NW surface would lead to a strong enhancement of this transi- tion. The same group substantiated this hypothesis by inves- tigating the evolution of the 3.45-eV band with NW diameter and NW density [17]. However, investigations of the Fermi level pinning in GaN NWs [18] as well as polarization- resolved PL and magneto-optical experiments [11] later re- futed the interpretation of the 3.45-eV band as an enhanced TES transition. Inversion domain boundaries (IDBs) may give rise to in- tense PL lines in GaN films at 3.45 eV [16, 19, 20]. An IDB denotes a boundary between Ga- and N-polar GaN for which two different stacking sequences have been proposed by Northrup et al. [21]. The IDB∗ notation refers to the spe- cific atomic structure at which each atom remains fourfold coordinated by exclusively forming Ga-N bonds across the boundary. In contrast, the unstarred IDB indicates a struc- ture where the formation of Ga-Ga or N-N bonds would oc- cur. The IDB∗ has an exceptionally low formation energy and does not induce electronic states in the band gap, thus facilitating the radiative recombination of excitons bound to these defects [16] (again in contrast to the unstarred IDB, which entails electronic states in the band gap). Robins et al. [7] suggested that the 3.45-eV band observed in PL spectra of GaN NWs is related to the presence of IDB∗s in GaN NWs just as in the bulk. This suggestion, however, was not supported by investigations of the microstructure of GaN NWs at that time. In fact, transmission electron microscopy (TEM) performed on isolated GaN NWs invariably demon- strated the absence of extended defects in the NW volume [9, 22 -- 25]. Several groups therefore favored point defects as the ori- gin for the 3.45-eV emission. In early work, in which GaN NWs were assumed to elongate axially via a Ga- induced vapor-liquid-solid mechanism, Ga interstitials were proposed as likely candidates for these point defects [6]. Later, GaN NW growth was understood to proceed under N-rich conditions, which were suggested to result in an en- hanced formation of Ga vacancies near the NW surface [8]. Brandt et al. [10] discussed the dependence of the 3.45-eV band on excitation density in terms of both planar defects such as the IDB∗ and abundant point defects and eventually favored the latter for being most consistent with the whole set of available data. However, the observation of strong NW-to-NW variations in the intensity of the 3.45-eV emis- sion from single NWs [26] contradicted this conclusion as well. Recently, Auzelle et al. [27] presented definitive exper- imental evidence for the presence of IDB∗s in GaN NWs grown on AlN-buffered Si(111). Subsequently, the same authors correlated µ-PL experiments with high-resolution scanning TEM performed on single GaN NWs grown on AlN-buffered Si [28]. They observed a systematic correla- tion between the presence of IDB∗s in the GaN NW and tran- sitions at 3.45 eV in its µ-PL spectrum and thus concluded that these transitions are caused by exciton recombination at IDB∗s. Finally, they proposed that the 3.45-eV band is indicative for the presence of IDB∗s in GaN NWs also for other substrates. It is not uncommon to observe mixed polarities in AlN films grown on Si(111) [29], and the coexistence of N- and Ga-polar NWs on such a film is therefore not an actual sur- prise. The spontaneous formation of GaN NWs directly on Si(111), however, is largely believed to occur on an amor- phous SiNx interlayer formed during the (unintentional or intentionally promoted) nitridation of the Si substrate by the N plasma [22, 30]. The polarity of GaN NWs formed directly on such a nitridated Si(111) surface has been de- bated for a long time. In earlier work, the GaN NWs were mostly reported to be Ga polar [17, 29, 31 -- 34], while recent studies (which also include ensemble investigations with far better statitics) indicate that GaN NWs grow predominantly or even exclusively N polar [35 -- 40]. In any case, whether IDB∗s form for GaN NWs on nitridated Si(111), and if so, at which density, are open questions. In this paper, we report a comprehensive investigation of the structural and optical properties of GaN NWs on Si(111) fabricated with or without intentional substrate nitridation and within a wide range of substrate temperatures. This in- vestigation focuses on the nature of the 3.45-eV band that is observed for all samples but with varying intensity. In Sec. II, we briefly describe the samples under investiga- tion and the experimental setups used for our study. Low- temperature cathodoluminescence (CL) spectroscopy and TEM are employed in Sec. III to investigate the origin of the 3.45-eV band. In accordance with the findings of Auzelle et al. [28], we attribute this band to the presence of IDB∗s in our NWs. In Sec. IV, we present an in-depth investi- gation of the 3.45-eV band comprising time-resolved and polarization-resolved PL spectroscopy accompanied by the- oretical considerations based on data published by Fiorentini [41]. The observed doublet structure of the 3.45-eV band is identified to be due to the recombination of localized and delocalized states at the IDB∗. The paper closes with a sum- mary and conclusion in Sec. V. 2 sample substrate reference TABLE I. List of the investigated samples. The substrate temper- ature TS during growth, the year of fabrication, and the reference containing growth details are also given. TS (◦C) 825 ◦C 720 ◦C 780 ◦C 800 ◦C 820 ◦C 835 ◦C 865 ◦C 875 ◦C 750 ◦C 780 ◦C AlN/SiC(0001) nitridated Si(111) nitridated Si(111) nitridated Si(111) nitridated Si(111) nitridated Si(111) nitridated Si(111) nitridated Si(111) nitridated Si(111) year 2011 2009 2009 2009 2009 2013 2013 2013 2009 2007 [37] [43] [43] [43] [43] [44] [44] [44] [30] [45] A B1 B2 B3 B4 B5 B6 B7 C D Si(111) II. EXPERIMENTAL The GaN NW ensembles studied in this work (see Tab. I for an overview) were grown by PAMBE on AlN-buffered SiC(0001) (sample A) or Si(111) (samples B1 -- B7, C, and D). They were fabricated over the course of six years and in four different PAMBE systems (as indicated by the letters in their names). The Si(111) substrates were intentionally nitri- dated prior to NW growth for all samples except for sample D. We have intentionally chosen NW ensembles synthesized over a wide range of substrate temperatures TS between 720 and 875 ◦C. All samples were grown under N-rich condi- tions with the Ga and N fluxes adjusted accordingly to ac- count for the increased Ga desorption at elevated tempera- tures [42]. Due to the wide range of conditions, the differ- ent NW samples exhibited incubation times between a few minutes and hours as well as different NW densities, aver- age NW diameters, and coalescence degrees. Further details on the growth conditions can be found elsewhere (sample A [37], B1 -- B4 [43], B5 -- B7 [44], C [30], and D [45]). The morphological and structural properties of the NWs were investigated by scanning electron microscopy (SEM) and TEM. For TEM, cross-sectional specimens were pre- pared by mechanical grinding, polishing, and subsequent Ar-ion polishing. The TEM images were recorded using an acceleration voltage of 300 kV. The polarity of the NWs was determined using convergent beam electron diffrac- tion (CBED) complemented by subsequent dark-field TEM imaging. SEM and CL spectroscopy were carried out in a field- emission instrument equipped with a CL system and a He- cooling stage. A photomultiplier tube was used for the ac- quisition of monochromatic images and a charge-coupled device (CCD) camera for recording CL spectra. Throughout the experiments, the acceleration voltage and the probe cur- rent of the electron beam were set to 5 kV and 0.75 nA, re- spectively. The spectral resolution amounted to 8 meV. Due to the strong quenching of the CL intensity of GaN NWs under the electron beam [46], the irradiation time was kept to a minimum. Since the (D0,XA) transition is more prone to quenching than the 3.45-eV band [18], the bichromatic CL images shown in this work consist of two superimposed monochromatic false-color images recorded first at 3.47 and then at 3.45 eV. Control experiments performed in reverse order confirmed the validity of our results. All PL experiments were performed in backscatter geom- etry with the samples mounted in liquid He-cooled cryostats offering continuous temperature control from 5 to 300 K. All measurements were conducted with NW ensembles except for the polarization-resolved PL experiments, where single NWs had been dispersed onto bare Si(111) wafers prior to the measurement. For continuous-wave PL spectroscopy, the 325-nm line of a HeCd laser was used to excite the sam- ples. The laser was focused to a spot with a diameter of 1 µm by a near-ultraviolet microscope objective with a nu- merical aperture of 0.65. The photoluminescence signal was collected by the same objective and dispersed by a spectrom- eter with an energy resolution of 0.25 to 1 meV. The dis- persed signal was detected by a CCD camera. Polarization- resolved measurements were carried out using a half-wave plate followed by a linear polarizer [47]. Time-resolved PL measurements were performed by focusing the second har- monic (325 nm) of an optical parametric oscillator pumped by a femtosecond Ti:sapphire laser (pulse width and repeti- tion rate of 200 fs and 76 MHz, respectively). The PL signal was dispersed by a monochromator and detected by a streak camera operating in synchroscan mode. The energy and time resolutions are 2 meV and 20 ps, respectively. III. IDB∗s AS THE ORIGIN OF THE 3.45-eV BAND IN GaN NWs ON Si(111) In this section we will establish the correlation between IDB∗s and the 3.45-eV band observed for GaN NWs on Si(111). Prior to a systematic investigation of our GaN NW ensembles on Si(111), we examine sample A which we already know to contain a very high density of IDBs or, as we will discuss later, most likely IDB∗s. This sam- ple was part of our study devoted to the role of substrate polarity in the formation of GaN NWs [37]. For this study, we attempted to induce the formation of GaN NWs on sub- strates with well-defined polarity, namely, AlN/SiC(0001) and AlN/SiC(000¯1). In the present work, we discuss further aspects of the sample synthesized at a substrate temperature of 825 ◦C on SiC(0001), which is here referred to as sam- ple A. The polar nature of the SiC substrate is known to determine the polarity of group-III-nitride layers deposited on it [37], and the AlN buffer was consequently found to be Al polar. Initiating the deposition of GaN at conditions typical for the growth of NWs resulted in a highly faceted Ga-polar GaN layer interspersed with sparse vertical NWs [SEM images of sample A can be found in Figs. 2(c) and 2(d) in Ref. 37]. We have found the majority of these NWs to be N polar due to a Si-induced polarity flip at the interface between AlN and GaN. Consequently, IDB∗s form upon co- alescence between the Ga-polar matrix and the N-polar NWs 3 FIG. 1. Low-temperature (10 K) PL spectrum of a GaN NW en- semble grown on AlN/SiC(0001) (sample A). In addition to the (D0,XA) transition at 3.469 eV, several very narrow lines around 3.45 eV are observed. The high-resolution TEM image in the in- set shows the central part of a GaN NW of sample A revealing a Ga-/N-polar core/shell structure. The arrows denote the IDB∗s be- tween the Ga-polar core and the N-polar shell. [37], and the sample is thus characterized by an exception- ally high fraction of NWs containing an IDB∗. The PL spectrum of sample A is shown in Fig. 1. The (D0,XA) transition is observed at 3.469 eV, but the spec- trum also exhibits several intense and narrow lines around 3.45 eV. The slight redshift of the (D0,XA) transition with respect to the one usually observed for GaN NWs (3.471 eV) as well as its comparatively large line width suggests that it mainly originates from the Ga-polar GaN layer. The strong transitions observed around 3.45 eV are consistent with the high fraction of NWs with an IDB∗ found in our previous FIG. 2. Low-temperature (10 K) PL spectra of samples B1, B2, B3, and B4 grown on Si(111). The spectra are normalized to the (D0,XA) transition. The substrate temperature used for the growth of each sample is specified in the figure. 3.423.443.46sample AGaN5 nmNNormalized PL intensityEnergy (eV)(D0,XA)(IDB*,X)10 K3.443.453.463.473.48Normalized PL intensityEnergy (eV) B1, 720 °C B2, 780 °C B3, 800 °C B4, 820 °C10 K4 FIG. 3. Bird's eye view SEM images of samples (a) B4 and (b) B1 grown at TS = 820 and 720 ◦C, respectively. Samples grown at higher substrate temperatures such as the ones depicted in (a) exhibit uniform NW lengths. In contrast, for GaN NWs fabricated at lower substrate temperatures [cf. (b)], a dense matrix of short NWs is interspersed by a few long and thin ones. analysis [see Fig. 4(b) in Ref. 37 for an example]. More- over, further investigations of sample A by TEM conducted in the course of the present work revealed in addition the existence of Ga-/N-polar core/shell NWs analogous to those reported by Auzelle et al. [28] (see the high-resolution TEM image displayed in the inset of Fig. 1). All these findings are in agreement with those reported in Ref. 28 and strongly suggest that the lines observed around 3.45 eV in Fig. 1 are in fact due to excitons bound to IDB∗s. Note that we ascribe the corresponding transitions to the IDB∗ rather than to the unstarred IDB because of the former's low formation energy and particularly the absence of dangling bonds, promoting the radiative decay of the exciton bound to it. We will there- fore label these transitions in all what follows as (IDB∗,X). An interesting finding in this context is the fine structure of the 3.45-eV band visible in Fig. 1. The origin of the distinct narrow lines in the PL spectrum will be addressed in Sec. IV. A prominent band at 3.45 eV is commonly also observed for GaN NWs grown on Si(111) [6 -- 11]. Figure 2 shows the evolution of the low-temperature (10 K) PL spectrum from GaN NW ensembles formed on Si(111) (samples B1 -- B4) under identical conditions except for the substrate tem- perature. The linewidth of the (D0,XA) transition decreases with TS increasing from 720 to 820 ◦C, indicating a progres- sive reduction of micro-strain induced by NW coalescence [44, 48, 49]. In parallel, the intensity of the 3.45-eV band with respect to that of the (D0,XA) line also decreases with increasing TS. If this band is also related to the presence of IDB∗s in GaN NWs grown directly on Si(111), its evolution seems to suggest that NWs fabricated at a higher TS exhibit a lower density of IDB∗s. In the following, we investigate whether this decrease in the relative intensity of the 3.45- eV band with increasing TS is correlated with changes in the morphology of the NW ensemble. Furthermore, we examine the spatial distribution of the different spectral components for various NW ensembles. Figures 3(a) and 3(b) display SEM bird's eye view im- ages of samples B4 and B1, which were grown at TS = 820 and 720 ◦C, respectively. The morphology of these sam- ples is representative for NW ensembles grown at high and low TS. For substrate temperatures higher than 750 ◦C, the NW ensemble usually exhibits a NW density of about 5× 109 cm−2 together with a uniform height distribution as shown in Fig. 3(a) [50]. In contrast, for temperatures sig- nificantly lower than this value, the ensemble morphology is characterized by a dense (1× 1010 cm−2) matrix of short and highly coalesced NWs interspersed by long NWs with a much lower density of 1× 108 cm−2 [cf. Fig. 3(b)]. Top- view SEM images (not shown here) indicate that most of the long NWs are attached to short NWs. To identify a possible correlation between the peculiar morphology of sample B1 and its intense 3.45-eV band, we record the spatial intensity distribution of the two distinct PL bands by CL spectroscopy. Figure 4(a) shows the superposi- tion of an SEM image with a bichromatic CL map recorded at 3.47 and 3.45 eV from sample B1 at 10 K. Clearly, the 3.45-eV band (color coded in green) originates almost ex- clusively from long NWs, whereas the (D0,XA) line (color- coded in magenta) stems mostly from the top part of short NWs. Figure 4(b) shows the same measurement for sam- ple C, which is another typical representative for a GaN NW ensemble grown at comparatively low TS. The spatial dis- tribution of the emission at 3.47 and 3.45 eV is the same as for sample B1, as indeed for all samples exhibiting the morphology characteristic for growth at low substrate tem- peratures. Note that in Figs. 4(a) and 4(b) many NWs appear not to emit at all (i.e., with the applied linear intensity scale they neither appear magenta nor green). This is consistently 1 µm1 µm(b)(a)sample B1sample B45 FIG. 4. Bird's eye view SEM images superimposed with bichromatic CL maps of samples (a) B1 and (b) C acquired at 10 K. NWs exhibiting an intense emission at 3.47 [corresponding to the (D0,XA) transition] and 3.45 eV appear magenta and green (linear intensity scale), respectively, while those emitting at both energies appear white. observed for samples grown at low TS and implies that only a few NWs actually contribute to the CL signal. The strong quenching of the CL under electron irradiation furthermore enhances this phenomenon [51]. To clarify the reason for this spatial distribution, we have performed CBED as well as TEM imaging on several NW FIG. 5. Representative dark-field TEM images of sample C show- ing the columnar matrix of short N-polar NWs and the long NWs to be either [(a) and (b)] Ga polar or [(c) and (d)] to exhibit a Ga- /N-polar core/shell structure. An inverted diffraction vector g also inverts the contrast between the Ga- and N-polar material. samples grown at low TS. Using CBED, it is straightforward to determine the polarity of the short NWs constituting the columnar matrix in these NW ensembles, since their effec- tive diameter is large due to the high degree of coalescence. The polarity of this matrix was found to be exclusively N polar for both samples B1 and C (not shown here). For the long NWs with diameters below 50 nm, the polarity cannot be reliably determined by CBED. However, since we know the polarity of the columnar matrix, we can instead employ dark-field TEM and exploit the fact that opposite polarities induce a contrast inversion in images recorded by this tech- nique. In addition, inverting the diffraction vector g should also invert the contrast for both polarities. Representative dark-field micrographs are displayed in Fig. 5. The oppo- site contrast between the long NW and its short neighbor in Fig. 5(a) recorded with g = 0002 as well as the contrast in- version in Fig. 5(b) recorded with g = 000¯2 demonstrate that the long NW is Ga polar. The situation is more complex in Figs. 5(c) and 5(d). Here, the shell of the long NW has the same polarity as the adjacent material, but it clearly has a Ga-polar core. These peculiar polarity core/shell NW struc- tures seem to be identical to those observed in Fig. 1 and in Ref. 28, suggesting that the mechanism giving rise to the formation of such structures is a general one and does not depend on the substrate. The long NWs are either Ga polar [Figs. 5(a) and 5(b)] or have a core/shell structure with a Ga-polar core surrounded by a N-polar shell [Figs. 5(c) and 5(d)]. In the former case, planar IDB∗s are formed at the coalescence boundaries be- tween the long Ga-polar NWs and the N-polar columnar ma- trix. In the latter case, the Ga-/N-polar core/shell structure results in the formation of an IDB∗ tube. IDB∗s thus exist either at the junctions between the long NWs and the sur- rounding columnar matrix or directly within the long NWs themselves. It is thus very plausible that the 3.45-eV emis- 1 µm1 µmsample B1sample C10 K10 K(b)(a)3.45 eV(D0,XA)N polarGa polargg-g50 nm-gGa polarN polar(d)(c)(b)(a)6 FIG. 6. (a) Low-temperature (10 K) PL spectrum of sample B5 grown at TS = 835 ◦C. The line labeled (I1,X) is due to excitons bound to I1 basal-plane stacking faults. (b) Bird's eye view SEM image of sample B5 superimposed with a bichromatic CL map acquired at 10 K. Magenta (green) areas imply that the related NWs emit dominantly at 3.471 eV (3.45 eV). The circles indicate the locations at which the spectra in panel (c) have been recorded. (c) Low-temperature (10 K) CL spectra of individual NWs of sample B5. The spectra have been recorded after acquisition of the CL map in panel (b) and have been shifted vertically for clarity. sion, which arises almost exclusively from the long NWs [see Fig. 4], originates from the (IDB∗,X) complex. We have shown so far that NW ensembles grown at a lower TS exhibit both short and long NWs and that the op- tical transition at 3.45 eV is due to exciton recombination at IDB∗s in long thin NWs. However, with increasing TS, the NW ensembles are getting more homogeneous in diam- eter and length [Fig. 3(a)], and the intensity of the transition at 3.45 eV strongly decreases in comparison to that of the (D0,XA) as depicted in Fig. 2. Figure 6(a) shows for sam- ple B5 grown at TS = 835 ◦C that the intensity of the line at 3.45 eV for NW ensembles grown at high substrate temper- atures can be two orders of magnitude smaller than that of the (D0,XA) line. In the bulk, the TES of the (D0,XA) is also centered at 3.45 eV and its intensity is about two orders of magnitude smaller than that of the (D0,XA) transition [15]. Consequently, the question arises whether the weak 3.45-eV band observed in Fig. 6(a) is related to IDB∗s at all. Figure 6(b) shows a bichromatic CL map of sample B5 and Fig. 6(c) depicts CL spectra taken at 10 K on individ- ual NWs. Clearly, the (D0,XA) and the 3.45-eV emission lines do not coincide spatially, ruling out the standard two- electron satellites as a possible origin of the 3.45-eV band and suggesting instead that the 3.45-eV band also arises from the presence of IDB∗s in these high-TS NW ensembles. Remarkably, the density of NWs with dominant (IDB∗,X) transitions is comparable to that observed for low-TS ensem- bles (cf. Fig. 4), demonstrating that the density of IDB∗s does not change significantly with substrate temperature. This finding seems to contradict the evolution of the rela- tive intensity of the (IDB∗,X) band with TS as depicted in Fig. 2. However, plotting the same data on an absolute in- tensity scale as done in Fig. 7 reveals that the intensity of the (IDB∗,X) band is actually not significantly reduced with increasing TS. The decrease of the relative intensity of the (IDB∗,X) band with increasing TS is instead caused by the drastic increase in the (D0,XA) emission intensity, reflect- ing the reduced concentration of nonradiative point defects at high TS [44, 52]. FIG. 7. Low-temperature (10 K) PL spectra of samples B1, B2, B3, and B4. The samples have been measured side-by-side under identical conditions, and the spectra are displayed on an absolute intensity scale. The substrate temperature used for the growth of each sample is specified in the figure. 3.423.443.463.483.423.443.463.48Normalized PL intensityEnergy (eV)(I1,X)(D0,XA)(IDB*,X)sample B510 K10 K(c)(a)Normalized CL intensityEnergy (eV) 1 2 3 4sample B5sample B5(b)10K1 µm(D0,XA)3.45 eV21343.443.453.463.473.4810 KPL intensity (arb. units)Energy (eV) B1, 720 °C B2, 780 °C B3, 800 °C B4, 820 °C7 FIG. 9. Polarization map of the near-band-edge luminescence of a dispersed NW of sample B7 recorded at 40 K. The NW axis, which is parallel to the c axis of GaN, is oriented along 90◦ as indicated by the dashed line. amplitudes within and outside of the quantum well, respec- tively, leading to a large electron-hole overlap and thus mak- ing the radiative recombination of the (IDB∗,X) an efficient process. To get quantitative information on the properties of the electronic state associated with the IDB∗, we calculated the wavefunction and the energy of an exciton in the poten- tial profile obtained by Fiorentini [41] using the variational approach described in Ref. 54. The result of our calculations is shown in Fig. 8. The calculations yield an electron-hole overlap, defined as the absolute square of the overlap inte- gral between the electron ((cid:104)Ψe) and hole (Ψh(cid:105)) wavefunc- tions, of (cid:104)ΨeΨh(cid:105)2 = 0.5. Therefore, despite the type-II band alignment across the IDB∗ plane, the (IDB∗,X) transi- tion possesses a large oscillator strength in agreement with the high intensity observed experimentally [cf. Figs. 4 and 6(b)]. For isotropic electron and hole masses of 0.2m0 and 1.0m0,[55] with m0 denoting the mass of the free electron, the energy of this transition ∆X is found to be 3.445 eV, i. e., close to the experimental value. 7+ and Γv Fiorentini [41] also predicted a significant mixing be- 9 valence bands due to the IDB∗. To ver- tween the Γv ify this prediction, we have performed polarization-resolved PL experiments at 40 K on single NWs from sample B7 dis- persed on a Si substrate. A typical polarization-resolved PL map taken on a single NW is shown in Fig. 9. In agree- ment with previous reports [11, 47], the free A exciton (XA) and the (D0,XA) transitions are polarized perpendicular to the NW axis (⊥ c), whereas the (IDB∗,X) band is polarized parallel to the NW axis ((cid:107) c). The three lowest optical tran- sitions in strain-free bulk GaN obey selection rules such that 9 transition is allowed only for light polarized ⊥ c, 7×Γv the Γc 7+ transition for both light polarized (cid:107) c and ⊥ c 7 × Γv the Γc 7− transition mostly for light polarized (cid:107) c. 7 × Γv , and the Γc The (IDB∗,X) band is polarized (cid:107) c as evident in Fig. 9, thus suggesting that it originates from a pure Γc 7− transition, i. e., the C exciton [56]. For a NW with sub-wavelength di- ameter, however, we have to bear in mind that the dielectric 7 ×Γv FIG. 8. (a) Electronic potential across an IDB∗ in GaN as calcu- lated in Ref. 41 (solid line) and including the attractive potential ex- erted by the holes as derived from effective-mass calculations (dot- ted line). (b) Band profile (dotted lines), electron- and hole prob- ability density (solid lines), and their energy levels (dashed lines) as a function of the distance to the IDB∗ as derived from effective- mass calculations [see (a)]. The values for the bandgap EG, the transition energy ∆X, and the absolute square of the overlap inte- gral between the electron and hole wavefunctions (cid:104)ΨeΨh(cid:105)2 are also given. IV. OPTICAL PROPERTIES OF THE (IDB∗,X) BAND We have established in the previous section that the 3.45- eV transition in GaN NWs on Si(111) is related to IDB∗s regardless of the substrate temperature. In this section, we investigate the optical properties of excitons bound to IDB∗s in more detail. In particular, we critically examine the con- sistence of our experimental results with the properties ex- pected theoretically for this particular bound exciton state. We focus on samples which exhibit a detailed fine structure in their PL spectra. One basic property of the (IDB∗,X) transition can be de- duced directly from the fact that the IDB∗ is a planar defect, which laterally extends over several tens of nm and verti- cally spans the entire NW length. In analogy to the result in Ref. 53 for basal-plane stacking faults, we hence expect the density of states of the (IDB∗,X) to be two-dimensional. As a result of the large number of states available, the (IDB∗,X) transition should be difficult to saturate even for high exci- tation conditions. Indeed, the (IDB∗,X) transition has been observed to scale linearly with excitation density even after the higher-energy (D0,XA) transition has started to saturate [6, 10, 16]. Using density-functional theory, Fiorentini [41] calcu- lated the electronic potential in the vicinity of an IDB∗ based on the stacking sequence proposed by Northrup et al. [21]. This potential, depicted in Fig. 8, acts as a barrier for elec- trons and as a quantum well for holes: an IDB∗ can therefore be seen as a type-II quantum well that binds holes. Elec- trons are maintained in the surrounding of the IDB∗ due to the Coulomb interaction with the holes. In addition, since the type-II quantum well formed by the IDB∗ is extremely thin, the electron and hole wavefunctions have significant -1.0-0.50.00.51.03.43.63.84.04.2-10-505100.00.53.54.0 DX = 3.445 eVEnergy (eV)z (nm)EG = 3.504 eV(a)Energy (eV) z (nm)(b)= 0.53.403.423.443.463.48090180270360Normalized PL intensityXAEnergy (eV)Polarization angle (deg)0.00.20.40.60.81.0(I1,X)(IDB*,X)(D0,XA)sample B740 K8 that between the (D0,XA) and the XA transitions. Figure 11 shows the evolution of the (IDB∗,X) doublet with increasing excitation density and temperature for sam- ple B6 and B7, respectively. At low temperatures and exci- tation densities, the (IDB∗,X) doublet is dominated by the lower energy transition [(IDB∗,X)L] at 3.452 eV. However, with increasing excitation density [Fig. 11(a)] or tempera- ture [Fig. 11(b)], the higher energy transition [(IDB∗,X)H] takes over and eventually dominates the PL spectrum. At around 40 K, carriers start to escape from the IDB∗, which manifests itself in a quenching of the (IDB∗,X) doublet [59]. As noted in Ref. 47, the excitation power and temperature dependences of the (IDB∗,X) doublet are similar to the ones observed for the (D0,XA) and XA transitions. In view of the fact that IDB∗s act as quantum wells, we thus attribute the high-energy line of the (IDB∗,X) doublet to the recombina- tion of excitons free to move along the IDB∗ plane and the low-energy one to excitons localized within this plane. Localization within the plane of a quantum well usually occurs at well width fluctuations or due to alloy disorder [60], which clearly cannot be the origin of the intra-IDB∗ lo- calization observed here. Following the results reported for the localization of excitons in I1 basal-plane stacking faults [61], we propose that the short-range potential of shallow donors such as Si and O distributed in the vicinity of the IDB∗ induce the localization of excitons within the IDB∗ plane. The excitation dependence of the intensity ratio at 10 K between the (IDB∗,X)H and (IDB∗,X)L lines with that between the XA and the (D0,XA) in Fig. 11(c) is consistent with this idea. Both ratios remain nearly constant for low excitation powers, but increase together for powers higher than 18 µW. This finding indicates that the density of local- ized states within the IDB∗s is comparable to the equiva- lent density of donors in NWs and thus confirms that intra- IDB∗ exciton localization occurs due to donors. Note that we computed the characteristic extent of the exciton wave- function perpendicular to the IDB∗ plane to be 5 nm. As- suming that the diffusion length of excitons in an IDB∗ is 100 nm [62, 63], a donor density of 1016 cm−3 is found to be sufficient to localize excitons within the IDB∗ plane, a value close to those reported for unintentionally n-doped GaN NWs [26]. In ensemble measurements, the (IDB∗,X)L band typically has a line width of several meV, whereas single NWs ex- hibit numerous sharp lines in this region [10, 47]. The small differences in transition energies originate from the varying distances between the involved donor and the IDB∗ plane [61]. This effect can also be seen very clearly for sample A in Fig. 1. Due to the low NW density of 5 × 108 cm−2 of this sample, only a small number of NWs is probed si- multaneously, and the individual narrow lines can be re- solved even in an ensemble measurement. The line width of the sharp (IDB∗,X)L lines in Fig. 1 and also in Fig. 3 of Ref. 47 is resolution limited, demonstrating that these lines stem from the radiative decay of bound excitons. In con- trast, ensemble spectra such as the ones shown in Figs. 10 and 11 contain contributions from about 103 NWs, and the individual transitions can no longer be resolved, but blend FIG. 10. (a) Low-temperature (40 K) PL spectra of sample B6, B7, and D grown at 865, 875, and 780 ◦C, respectively. The spec- tral windows containing the (IDB∗,X) and the (D0,XA) have been normalized independently to the strongest transition. contrast strongly suppresses emission polarized perpendic- ular to the NW axis and thereby artificially enhances the component polarized (cid:107) c [47]. The strong polarization of the (IDB∗,X) band along the NW axis ((cid:107) c) thus suggests a reversal in the order of the Γv 9 valence bands in the IDB∗, in agreement with the theoretical result of Fioren- tini [41]. Note also that the (I1,X) band is polarized ⊥ c, demonstrating that the opposite behavior observed for the (IDB∗,X) transition is a consequence of the peculiar poten- tial induced by the IDB∗ and not a characteristic of excitons bound to planar defects in general. 7 and Γv 7 and Γv The reordering between Γv 9 states proposed in Ref. 41 and observed in Fig. 9 is also consistent with the magneto-optical behavior of the (IDB∗,X) transition re- ported in Ref. 11. The Land´e factor g was found to be close to zero for the (IDB∗,X), while for the (D0,XA) transition in GaN NWs it was observed to be 1.75, a value similar to that reported for the bulk [11]. As both the electron and hole Land´e factors are extremely sensitive to valence band mix- ing [57], it is in fact not surprising to measure very different values of g for the (D0,XA) and the (IDB∗,X) lines. All properties of the (IDB∗,X) transition discussed so far are consistently described by the electronic potential com- puted by Fiorentini [41]. However, this model does not pro- vide any explanation for the fact that the (IDB∗,X) band of- ten exhibits a doublet structure. As shown first by Calleja et al. [6], the band at 3.45 eV actually consists of two lines centered at about 3.449 and 3.455 eV [47, 58]. This finding is confirmed by the PL spectra of samples B6, B7, and D at 40 K as shown in Fig. 10. For these three samples, the lower and higher energy component of the doublet is cen- tered at about 3.452 and 3.458 eV, respectively. Note that the splitting between these two lines is almost identical to 3.453.463.473.48XA(D0,XA)(IDB*,X)L(IDB*,X)HPL intensity (arb. units)Energy (eV) B6 B7 D40 K9 FIG. 11. (a) Low-temperature (10 K) PL spectra of sample B6 for various excitation powers. (b) Evolution of the PL spectra of sample B7 with temperature. The spectra have been shifted vertically for clarity. (c) Intensity ratio RIDB∗ between the (IDB∗,X)H and and (IDB∗,X)L lines (squares) and RNBE between the XA and (D0,XA) transitions (circles) as a function of excitation power. The intensities have been obtained from a fit of the spectra in (a) with Voigt functions. together to an (IDB∗,X)L band with a line width of several meV [10]. Note that Schuck et al. [19] observed a fine struc- ture of the (IDB∗,X) band in bulk GaN already in 2001. The (IDB∗,X)H band exhibits a linewidth of a few meV as ex- pected for delocalized states. Finally, we have performed time-resolved PL experiments FIG. 12. Low-temperature (10 K) PL transients of the (D0,XA) and XA transitions as well as of the (IDB∗,X)L and (IDB∗,X)H lines of sample B6. The arrows denote the respective rise times, and the solid lines are exponential fits. to obtain quantitative information on the capture efficiency of excitons by IDB∗s and on the intra-IDB∗ localization pro- cess. Figure 12 shows PL transients of sample B6 measured at 10 K for the XA and (D0,XA) transitions as well as for the (IDB∗,X)H and (IDB∗,X)L lines. While the initial increase of the XA PL intensity is almost instantaneous, it takes 90 ps for the (D0,XA) PL intensity to reach a maximum. The lat- ter time corresponds roughly to the characteristic time for the capture of excitons by donors with a density of about 5×1016 cm−3 [64]. In agreement with the results reported in Ref. 64, the XA and (D0,XA) PL decay in parallel at longer times as a result of the quasi-thermalization between these exciton states. Using an exponential fit, the effective decay time for the XA and (D0,XA) is found to be 195 ps, very sim- ilar to values obtained in previous reports [52, 65, 66]. This decay is much faster than expected for the radiative decay of the (D0,XA) complex and is due to the nonradiative decay of the XA at point defects [66]. Analogously to the (D0,XA) and XA transitions, the initial increase of the (IDB∗,X)L intensity is delayed compared to that of the (IDB∗,X)H. Interestingly, the rise time of the (IDB∗,X)H intensity is only 47 ps. Therefore, the capture of excitons by IDB∗s is more efficient than their trapping by neutral donors due to the IDB∗s' large capture cross- section. This situation differs from observations for the rise time of the (I1,X) transition at 10 K [59], which is lim- ited by the inefficient transport of excitons from one donor to the next [65]. For time delays longer than 600 ps, the 3.453.473.453.473.49(IDB*,X)H(IDB*,X)L10 Ksample B7PL intensity (arb. units)Energy (eV)10 mW2.7 mW525 µW77 µW18 µW(a)sample B6(D0,XA)XAXB(IDB*,X)H(IDB*,X)L(b)PL intensity (arb. units)Energy (eV)10 K20 K40 K80 K100 KXA(D0,XA)10-1101103RIDB*, RNBEExcitation power (µW)10 Ksample B6(c)0200400600800PL intensity (arb. units)Time (ps) (IDB*,X)L (IDB*,X)H (D0,XA) XA10 Ksample B6(IDB∗,X)L and (IDB∗,X)H decay in parallel, again demon- strating quasi-thermalization between these states. Finally, for time delays longer than 1 ns, the PL decay of the cou- pled XA and (D0,XA) states slows down and asymptotically approaches that of the (IDB∗,X) doublet (not shown), in- dicating full thermalization between the XA, (D0,XA) and (IDB∗,X) states. This process is the origin of the biexpo- nential PL decay reported for the (D0,XA) in Refs. 9 and 64. V. SUMMARY AND CONCLUSIONS The 3.45-eV band observed in low-temperature PL and CL spectra of spontaneously formed GaN NWs on Si(111) arises from planar or tubular IDB∗s. While the former are due to the coalescence of adjacent Ga- and N-polar NWs, the latter form at the interface of Ga-/N-polar core/shell NWs. The intensity ratio between the (IDB∗,X) and the (D0,XA) transitions decreases with increasing TS. This decrease is a consequence of the reduction in the density of nonradiative point defects with increasing TS, leading to an increase in the absolute intensity of the (D0,XA) line. In contrast, the absolute intensity of the (IDB∗,X) band is neither directly governed by TS nor by the presence or absence of an inten- tional nitridation step. The same applies to the abundance of IDB∗s observed in spatially resolved, bichromatic CL maps. These results confirm the idea of Auzelle et al. [28] that the 3.45-eV luminescence band in GaN NWs signifies the presence of IDB∗s regardless of the substrate. In fact, we now know with certainty that IDB∗s occur in GaN NW en- sembles synthesized by PAMBE on AlN-buffered Si(111) [7, 28], on nitridated Si(111), and on SiC(0001). The ex- ceptionally low formation energy of IDB∗s is obviously an important factor promoting their frequent occurrence. How- ever, a prerequisite for the formation of an IDB∗ is the si- multaneous presence of both Ga- and N-polar material. At present, it remains entirely unclear why an apparently con- stant fraction of the GaN nuclei on all of these different sub- strates are Ga polar. We also do not understand how Ga- polar NWs can evolve from these nuclei despite our inability to synthesize them intentionally on cation-polar substrates [37] and at substrate temperatures at which GaN(0001) usu- ally decomposes. Finally, it is unclear how the peculiar Ga- /N-polar core/shell NWs form, which have been observed by different groups and on different substrates. The lack of knowledge regarding these apparently universal and basic phenomena demonstrates that the nucleation and formation of GaN NWs in PAMBE are still far from being completely 10 understood. Concerning the electronic and optical properties of IDB∗s, it is helpful to imagine them as a thin type-II quantum well that binds holes. The Coulomb attraction exerted by these holes is strong enough to bind electrons, and the resulting (IDB∗,X) state decays radiatively and thus gives rise to in- tense light emission. The change in the symmetry of the fundamental hole state in IDB∗s strongly modifies the polar- ization and the behavior of the (IDB∗,X) when subjected to magnetic fields. Donor atoms distributed in the vicin- ity of the IDB∗ plane localize the exciton within the IDB∗ plane. As a result of this localization, the (IDB∗,X) band resolves into a doublet with the lines at low and high energy being associated with localized and free (IDB∗,X) states, respectively. Their excitonic nature is manifested by the observation of a fine structure of the low energy line con- sisting of sharp, resolution-limited peaks. Finally, while the capture of excitons by IDB∗s takes less than 50 ps, quasi- thermalization between the near-band edge excitons and the (IDB∗,X) takes much longer, resulting in the nonexponen- tial decay usually observed for the (D0,XA) transition in GaN NWs at low temperatures. Analogously to stacking faults [59], IDB∗s do not suffer from fluctuations in layer thickness or composition as con- ventional quantum wells and thus offer unique possibilities for the study of low-dimensional excitons [47]. Particularly interesting in this context are the tubular IDB∗s formed in Ga-/N-polar core/shell NWs and intersections of these tubu- lar IDB∗s with stacking faults forming perfect crystal-phase quantum rings. Since the electronic states associated with all of these quantum structures are shallow [59], they should be of little practical relevance for conventional devices. How- ever, we envisage that their exceptionally well-defined prop- erties make them ideal model systems for an understanding of quantum effects important for a future generation of op- toelectronic devices. VI. ACKNOWLEDGMENTS The authors thank Pierre Lefebvre for fruitful discus- sions, Vincent Consonni and Caroline Ch`eze for provid- ing additional samples, and Uwe Jahn for a critical read- ing of the manuscript. P. C. acknowledges funding from the Fonds National Suisse de la Recherche Scientifique through project 161032. This work was partly supported by the Ger- man BMBF joint research project MONALISA (Contract no. 01BL0810), by the Deutsche Forschungsgemeinschaft within SFB 951, and by Marie Curie RTN PARSEM (Grant No. MRTN-CT-2004-005583). ∗ [email protected] † C. P. and P. C. contributed equally to this work. ‡ Present address: DILAX Intelcom GmbH, Alt-Moabit 96b, 10559 Berlin, Germany. § Previously at Paul-Drude-Institut fur Festkorperelektronik. ¶ Present address: LayTec AG, Seesener Str. 10 -- 13, 10709 Berlin, Germany. ∗∗ Present address: Department of Electrical and Electronics Engineering, Faculty of Engineering, Mugla Sıtkı Koc¸man University, Kotekli, 48000, Mugla, Turkey. [ 1 ] M. Yoshizawa, A. Kikuchi, M. Mori, N. Fujita, and K. Kishino, Jpn. J. Appl. Phys. 36, L459 (1997). [ 2 ] M. A. S´anchez-Garc´ıa, E. Calleja, E. Monroy, F. Sanchez, F. Calle, E. Munoz, and R. Beresford, J. Cryst. Growth 183, 23 (1998). [ 3 ] A. Kikuchi, M. Kawai, M. Tada, and K. Kishino, Jpn. J. Appl. Phys. 43, L1524 (2004). [ 4 ] J. Kamimura, P. Bogdanoff, J. Lahnemann, C. Hauswald, L. Geelhaar, S. Fiechter, and H. Riechert, J. Am. Chem. Soc. 235, 10242 (2013). [ 5 ] M. F. Cansizoglu, S. M. Hamad, D. P. Norman, F. Keles, E. Badraddin, T. Karabacak, and H.-W. Seo, Appl. Phys. Ex- press 8, 042302 (2015). [ 6 ] E. Calleja, M. S´anchez-Garc´ıa, F. S´anchez, F. Calle, F. Naranjo, E. Munoz, U. Jahn, and K. H. Ploog, Phys. Rev. B 62, 16826 (2000). [ 7 ] L. H. Robins, K. A. Bertness, J. M. Barker, N. A. Sanford, and J. B. Schlager, J. Appl. Phys. 101, 113506 (2007). [ 8 ] F. Furtmayr, M. Vielemeyer, M. Stutzmann, A. Laufer, B. K. Meyer, and M. Eickhoff, J. Appl. Phys. 104, 074309 (2008). [ 9 ] P. Corfdir, P. Lefebvre, J. Risti´c, P. Valvin, E. Calleja, and B. Deveaud-Pl´edran, J. A. Trampert, J.-D. Gani`ere, Appl. Phys. 105, 013113 (2009). [ 10 ] O. Brandt, C. Pfuller, C. Ch`eze, L. Geelhaar, and H. Riechert, Phys. Rev. B 81, 45302 (2010). [ 11 ] D. Sam-Giao, R. Mata, G. Tourbot, J. Renard, A. Wysmołek, B. Daudin, and B. Gayral, J. Appl. Phys. 113, 043102 (2013). [ 12 ] B. Skromme, H. Zhao, B. Goldenberg, H. S. Kong, M. T. Leonard, G. E. Bulman, C. R. Abernathy, and S. J. Pearton, MRS Proc. 449, 713 (1996). [ 13 ] J. A. J. Freitas, W. Moore, B. Shanabrook, G. Braga, S. Lee, S. Park, and J. Han, Phys. Rev. B 66, 233311 (2002). [ 14 ] A. Wysmołek, K. P. Korona, R. Stepniewski, J. M. Bara- nowski, J. Błoniarz, M. Potemski, R. L. Jones, D. C. Look, J. Kuhl, S. S. Park, and S. K. Lee, Phys. Rev. B 66, 245317 (2002). [ 15 ] P. Paskov, B. Monemar, A. Toropov, J. P. Bergman, and A. Usui, Phys. Status Solidi C 4, 2601 (2007). [ 16 ] M. A. Reshchikov, D. Huang, F. Yun, P. Visconti, L. He, H. Morkoc¸, J. Jasinski, Z. Liliental-Weber, R. J. Molnar, S. S. Park, and K. Y. Lee, J. Appl. Phys. 94, 5623 (2003). [ 17 ] P. Lefebvre, S. Fern´andez-Garrido, J. Grandal, J. Risti´c, M.- and E. Calleja, Appl. Phys. Lett. 98, A. S´anchez-Garc´ıa, 083104 (2011). [ 18 ] C. Pfuller, O. Brandt, F. Grosse, T. Flissikowski, C. Ch`eze, V. Consonni, L. Geelhaar, H. T. Grahn, and H. Riechert, Phys. Rev. B 82, 45320 (2010). [ 19 ] P. Schuck, M. D. Mason, R. D. Grober, O. Ambacher, A. P. Lima, C. Miskys, R. Dimitrov, and M. Stutzmann, Appl. Phys. Lett. 79, 952 (2001). [ 20 ] R. Kirste, R. Collazo, G. Callsen, M. R. Wagner, T. Kure, J. Sebastian Reparaz, S. Mita, J. Xie, A. Rice, J. Tweedie, and A. Hoffmann, J. Appl. Phys. 110, 093503 Z. Sitar, (2011). [ 21 ] J. E. Northrup, J. Neugebauer, and L. T. Romano, Phys. Rev. Lett. 77, 103 (1996). [ 22 ] A. Trampert, J. Risti´c, U. Jahn, E. Calleja, and K. Ploog, in Microscopy of Semiconducting Materials 2003, IOP Conf. Ser. No. 180, edited by A. G. Cullis and P. A. Midgley (Insti- tute of Physics, Bristol, 2003) p. 167. 11 [ 23 ] K. A. Bertness, N. A. Sanford, J. M. Barker, J. B. Schlager, A. Roshko, A. V. Davydov, and I. Levin, J. Electron. Mater. 35, 576 (2006). [ 24 ] L. Cerutti, J. Risti´c, S. Fern´andez-Garrido, E. Calleja, A. Trampert, K. H. Ploog, S. Lazi´c, and J. M. Calleja, Appl. Phys. Lett. 88, 213114 (2006). [ 25 ] H. Sekiguchi, T. Nakazato, A. Kikuchi, and K. Kishino, J. Cryst. Growth 300, 259 (2007). [ 26 ] C. Pfuller, O. Brandt, T. Flissikowski, C. Ch`eze, L. Geelhaar, H. T. Grahn, and H. Riechert, Nano Res. 3, 881 (2010). [ 27 ] T. Auzelle, B. Haas, A. Minj, C. Bougerol, J.-L. Rouvi`ere, A. Cros, J. Colchero, and B. Daudin, J. Appl. Phys. 117, 245303 (2015). [ 28 ] T. Auzelle, B. Haas, M. Den Hertog, J.-L. Rouvi`ere, B. Daudin, and B. Gayral, Appl. Phys. Lett. 107, 051904 (2015). [ 29 ] M. D. Brubaker, I. Levin, A. V. Davydov, D. M. Rourke, N. A. Sanford, V. M. Bright, and K. A. Bertness, J. Appl. Phys. 110, 053506 (2011). [ 30 ] V. Consonni, M. Hanke, M. Knelangen, L. Geelhaar, and H. Riechert, Phys. Rev. B 83, 035310 A. Trampert, (2011). [ 31 ] D. Cherns, L. Meshi, I. Griffiths, S. Khongphetsak, S. V. Novikov, N. Farley, R. P. Campion, and C. T. Foxon, Appl. Phys. Lett. 92, 121902 (2008). [ 32 ] F. Furtmayr, M. Vielemeyer, M. Stutzmann, J. Arbiol, S. Estrad´e, F. Peir`o, J. R. Morante, and M. Eickhoff, J. App. Phys. 104, 034309 (2008). [ 33 ] R. Armitage and K. Tsubaki, Nanotechnol. 21, 195202 (2010). [ 34 ] L. Geelhaar, C. Cheze, B. Jenichen, O. Brandt, C. Pfueller, S. Muench, R. Rothemund, S. Reitzenstein, A. Forchel, P. Komninou, G. P. Dimitrakopulos, T. Kehagias, T. Karakostas, L. Lari, P. R. Chalker, M. H. Gass, and H. Riechert, IEEE J. Sel. Top. Quant. Elec. 17, 878 (2011). [ 35 ] K. Hestroffer, C. Leclere, C. Bougerol, H. Renevier, and B. Daudin, Phys. Rev. B 84, 245302 (2011). [ 36 ] F. Schuster, F. Furtmayr, R. Zamani, C. Mag´en, J. R. Morante, J. Arbiol, J. A. Garrido, and M. Stutzmann, Nano Lett. 12, 2199 (2012). [ 37 ] S. Fern´andez-Garrido, X. Kong, T. Gotschke, R. Calarco, L. Geelhaar, A. Trampert, and O. Brandt, Nano Lett. 12, 6119 (2012). [ 38 ] S. D. Carnevale, T. F. Kent, P. J. Phillips, A. T. M. G. Sarwar, C. Selcu, R. F. Klie, and R. C. Myers, Nano Lett. 13, 3029 (2013). [ 39 ] O. Romanyuk, S. Fern´andez-Garrido, P. Jir´ıcek, I. Bartos, L. Geelhaar, O. Brandt, and T. Paskova, Appl. Phys. Lett. 106, 021602 (2015). [ 40 ] A. Minj, A. Cros, N. Garro, J. Colchero, T. Auzelle, and B. Daudin, Nano Letters 15, 6770 (2015), pMID: 26380860. [ 41 ] V. Fiorentini, Appl. Phys. Lett. 82, 1182 (2003). [ 42 ] S. Fern´andez-Garrido, J. Grandal, E. Calleja, M. A. S´anchez- Garc´ıa, and D. Lopez-Romero, J. Appl. Phys. 106, 126102 (2009). [ 43 ] P. Dogan, O. Brandt, C. Pfuller, A.-K. Bluhm, L. Geelhaar, and H. Riechert, J. Cryst. Growth 323, 418 (2011). [ 44 ] J. K. Zettler, C. Hauswald, P. Corfdir, M. Musolino, L. Geel- and S. Fern´andez-Garrido, haar, H. Riechert, O. Brandt, Cryst. Growth Des. 15, 4104 (2015). [ 45 ] C. Ch`eze, L. Geelhaar, A. Trampert, and H. Riechert, Appl. Phys. Lett. 97, 43101 (2010). [ 46 ] E. M. Campo, G. S. Cargill, M. Pophristic, and I. Ferguson, MRS Internet J. Nitride Semicond, Res. 9, 8 (2004). [ 47 ] P. Corfdir, F. Feix, J. K. Zettler, S. Fern´andez-Garrido, and O. Brandt, New J. Phys. 17, 033040 (2015). [ 48 ] B. Jenichen, O. Brandt, C. Pfuller, P. Dogan, M. Knelangen, and A. Trampert, Nanotechnology 22, 295714 (2011). [ 49 ] S. Fern´andez-Garrido, V. M. Kaganer, C. Hauswald, B. Jenichen, M. Ramsteiner, V. Consonni, L. Geelhaar, and O. Brandt, Nanotechnology 25, 455702 (2014). [ 50 ] K. K. Sabelfeld, V. M. Kaganer, F. Limbach, P. Dogan, O. Brandt, L. Geelhaar, and H. Riechert, Appl. Phys. Lett. 103, 133105 (2013). [ 51 ] J. Lahnemann, T. Flissikowski, M. Wolz, L. Geelhaar, H. T. Grahn, O. Brandt, and U. Jahn, Nanotechnology 27, 455706 (2016). [ 52 ] M. Sobanska, K. P. Korona, Z. R. Zytkiewicz, K. Klosek, and G. Tchutchulashvili, J. Appl. Phys. 118, 184303 (2015). [ 53 ] P. Corfdir, C. Hauswald, O. Marquardt, T. Flissikowski, J. K. Zettler, S. Fern´andez-Garrido, L. Geelhaar, H. T. Grahn, and O. Brandt, Phys. Rev. B 93, 115305 (2016). [ 54 ] P. Corfdir and P. Lefebvre, J. Appl.Phys. 112, 053512 (2012). [ 55 ] I. Vurgaftman and J. R. Meyer, J. Appl. Phys. 94, 3675 [ 56 ] P. Misra, O. Brandt, H. T. Grahn, H. Teisseyre, M. Siekacz, C. Skierbiszewski, and B. Łucznik, Appl. Phys. Lett. 91, 141903 (2007). [ 57 ] L. M. Roth, B. Lax, and S. Zwerdling, Phys. Rev. 114, 90 (2003). (1959). 12 [ 58 ] P. Lefebvre, "Surface-Related Optical Properties of GaN- Based Nanowires," in Wide Band Gap Semiconductor Nanowires 1, edited by V. Consonni and G. Feuillet (John Wiley & Sons, Inc., 2014) pp. 59 -- 79. [ 59 ] P. Corfdir, C. Hauswald, J. K. Zettler, T. Flissikowski, J. Lahnemann, S. Fern´andez-Garrido, L. Geelhaar, H. T. Grahn, and O. Brandt, Phys. Rev. B 90, 195309 (2014). [ 60 ] C. Weisbuch, R. Dingle, A. Gossard, and W. Wiegmann, Solid State Commun. 38, 709 (1981). [ 61 ] P. Corfdir, P. Lefebvre, J. Risti´c, J.-D. Gani`ere, B. Deveaud-Pl´edran, Phys. Rev. B 80, 153309 (2009). and [ 62 ] P. Corfdir, J. Levrat, A. Dussaigne, P. Lefebvre, H. Teis- seyre, I. Grzegory, T. Suski, J.-D. Gani`ere, N. Grandjean, and B. Deveaud-Pl´edran, Phys. Rev. B 83, 245326 (2011). [ 63 ] G. Nogues, T. Auzelle, M. Den Hertog, B. Gayral, and B. Daudin, Appl. Phys. Lett. 104, 102102 (2014). [ 64 ] C. Hauswald, T. Flissikowski, T. Gotschke, R. Calarco, L. Geelhaar, H. T. Grahn, and O. Brandt, Phys. Rev. B 88, 075312 (2013). [ 65 ] P. Corfdir, J. Risti´c, P. Lefebvre, T. Zhu, D. Martin, A. Dus- and B. Deveaud- saigne, J.-D. Gani`ere, N. Grandjean, Pl´edran, Appl. Phys. Lett. 94, 201115 (2009). [ 66 ] C. Hauswald, P. Corfdir, J. K. Zettler, V. M. Kaganer, K. K. Sabelfeld, S. Fern´andez-Garrido, T. Flissikowski, V. Con- sonni, T. Gotschke, H. T. Grahn, L. Geelhaar, and O. Brandt, Phys. Rev. B 90, 165304 (2014). 

1803.01144

1

1803

Inducing and controlling rotation on small objects using photonic topological materials

Photonic topological insulator plates violate Lorentz reciprocity which leads to a directionality of surface-guided modes. This in-plane directionality can be imprinted via an applied magnetic field. On the basis of macroscopic quantum electrodynamics in nonreciprocal media, we show that two photonic topological insulator surfaces are subject to a tuneable, magnetic-field dependent Casimir torque. Due to the directionality, this torque exhibits a unique $2\pi$ periodicity, in contradistinction to the Casimir torques encountered for reciprocal uniaxial birefringent media or corrugated surfaces which are $\pi$-periodic. Remarkably, the torque direction and strength can be externally driven in situ by simply applying a magnetic field on the system, and we show that this can be exploited to induce a control the rotation of small objects. Our predictions can be relevant for nano-opto-mechanical experiments and devices.

cond-mat.mes-hall

cond-mat

Inducing and controlling rotation on small objects using photonic topological materials Frieder Lindel1, George W. Hanson2, Mauro Antezza3,4, and Stefan Yoshi Buhmann1,5 1 Physikalisches Institut, Albert-Ludwigs-Universitat Freiburg, Hermann-Herder-Strasse 3, 79104 Freiburg, Germany 2 Department of Electrical Engineering, University of Wisconsin-Milwaukee, 3200 N. Cramer St., Milwaukee, Wisconsin 53211, USA 3 Laboratoire Charles Coulomb, UMR 5221 Universit´e de Montpellier and CNRS, F-34095 Montpellier, France 4 Institut Universitaire de France, 1 rue Descartes, F-75231 Paris Cedex 05, France 5 Freiburg Institute for Advanced Studies, Albert-Ludwigs-Universitat Freiburg, Albertstrasse 19, 79104 Freiburg, Germany (Dated: March 6, 2018) 8 1 0 2 r a M 3 ] l l a h - s e m . t a m - d n o c [ 1 v 4 4 1 1 0 . 3 0 8 1 : v i X r a Photonic topological insulator plates violate Lorentz reciprocity which leads to a directionality of surface- guided modes. This in-plane directionality can be imprinted via an applied magnetic field. On the basis of macroscopic quantum electrodynamics in nonreciprocal media, we show that two photonic topological insulator surfaces are subject to a tuneable, magnetic-field dependent Casimir torque. Due to the directionality, this torque exhibits a unique 2π periodicity, in contradistinction to the Casimir torques encountered for reciprocal uniaxial birefringent media or corrugated surfaces which are π-periodic. Remarkably, the torque direction and strength can be externally driven in situ by simply applying a magnetic field on the system, and we show that this can be exploited to induce a control the rotation of small objects. Our predictions can be relevant for nano-opto- mechanical experiments and devices. PACS numbers: 31.30.J-, 31.30.jf, 73.43.-f, 78.68.+m The Casimir force was originally proposed as an attractive force between two perfectly conducting plates due to a re- duced virtual photon pressure in the space between the plates [1]. In macroscopic quantum electrodynamics (QED) the Casimir force was further generalized to bodies of arbitrary shape and material by realizing that its existence stems from fluctuating charge carriers within the material [2]. Subse- quently the Casimir force for objects consisting of anisotropic materials or possessing anisotropic surfaces like birefrigent plates [3], magnetodielectric metamaterials [4] or corrugated metals [5] was studied. Since all those materials have a dis- tinguishable axis in the plane of the plates it is natural to ask whether the Casimir energy depends on the relative angle be- tween the two axes when bringing two anisotropic surfaces together. It turns out that indeed one obtains a torque when the angle between the axes is different from zero or π [3, 5– 8]. Another way to understand this phenomena is to realize that photons do not only carry linear momentum but also a nonvanishing angular momentum which can therefore be car- ried from the vacuum to the objects due to broken rotational symmetry [3]. More recently there have been several promis- ing proposals for experiments with the goal to measure the Casimir torque between birefrigent materials [9–11]. An example of a material which is able to break rotational symmetry and which is of great interest at the moment is pro- vided by topological insulators (TI) [12]. Topological insula- tors behave like regular insulators in their bulk but they pos- sess conducting surface states. Originally they were proposed for electronic states but in more recent years it was shown that they also exist in so called photonic topological insula- tors (PTI) [13–16] as for example magnetized plasma [17– 19]. One of the most striking features of TIs is that there exist unidirectional waves on the surfaces of these materials which turn out to be immune to backscattering [20, 21]. Due to this directionality of the edge states PTIs do not only have a dis- tinguishable axis as i.e. birefrigent materials but their axes possesses also a direction, that means an axis that is sensi- tive to whether an electromagnetic wave is traveling in one or the other direction along the axis. This feature has been of great interest and was used to construct devices like direc- tional wave guides [21], optical isolators or circulators. Now the natural question arises what quantum optical effects emerge when working with PTIs. This has been done by pre- vious authors before: they studied the influence of the pres- ence of a PTI on the entanglement of a two level system [22]; in Ref. [23, 24] the normal and lateral Casimir-Polder force acting on an atom close to a vacuum/PTI interface was ana- lyzed; a huge anisotropic thermal magnetoresistance was ob- tained in the near field radiative heat transfer between two spherical particles consisiting of a PTI in Ref. [25]; but also the Casimir force has already been studied for two infinite half spaces consisting of PTIs, namely InSb, in Ref. [26]. All those works showed that there exist interesting new features in quantum optics arising from the interplay of the quantized electromagnetic field with PTIs. However, the unidirectional features of PTIs have not yet been seen manifest in Casimir energies and torques between two macroscopic objects. One of the main challenges when working with PTIs is that one has to take nonreciprocal material response into account, e.g. arising from the unidriectional surface states, which has been done in Ref. [27]. Optical effects for nonreciprocal ma- terials have been studied in recent years leading to some stun- ning results like a persistent directional heat current between three objects at thermal equilibrium [28, 29]. The expression for the Casimir force has also been generalized for the case of nonreciprocal materials and has been applied to PTIs in Ref. [26]. In this Letter, we want to show how the unidirecitonality and nonreciprocity of PTI plates manifest itself in the Casimir force and torque. Therefore we will show in the following that, in addition to a normal component of the Casimir force, there exists a non-negligible Casimir torque whose magnitude 2 electron mass. Furthermore, as depicted in Fig. 1, we define the two applied magnetic fields B+ and B− laying in the xy− plane by their absolute values B+ and B− and by the angles between B+ and B− and the x−axis, namely Φ+ and Φ−. Al- though we choose a specific PTI model here, our findings are general and can be applied to other specific PTI realizations. We first have to derive a general expression for the Casimir torque of our system. Normally this is done by directly calu- lating the Casimir energy E to obtain the Casimir torque from it. Nevertheless we are going to first calculate the Casimir force and derive an expression for E from it, since in our setup the permittivity tensor in Eq. (1) breaks the isotropy in the xy−plane and therefore there could possibly exist a lateral Casimir Force. This lateral component of the Casimir force would not show up by directly calculating the Casimir energy. To calculate the Casimir force F acting on body one we cannot use the final result of Ref. [26] (Eq. (31) in Ref. [26]), due to the broken isotropy in the xy−plane. Nevertheless we can use the intermediate result Eq. (18) found in the same reference: dξ dA (cid:90) F = − ¯h 2π ∞(cid:90) (cid:26)2ξ 2 S(cid:104) G(1)(cid:0)r,r(cid:48),iξ(cid:1)(cid:105) (cid:20)ξ 2 c2 G(1)(cid:0)r,r(cid:48),iξ(cid:1) + c2 −Tr ∂V 0 · −→ ∇ × S(cid:104) ∇ (cid:48)(cid:21) ∇ × G(1)(cid:0)r,r(cid:48),iξ(cid:1)×←− G(1)(cid:0)r,r(cid:48),iξ(cid:1)(cid:105)×←− (cid:27) + 2 −→ 1 ∇ (cid:48) . r(cid:48)→r (2) by S [G (r,r(cid:48),ω)] = (1/2)(cid:2)G (r,r(cid:48),ω) + GT (r(cid:48),r,ω)(cid:3). Fur- Here we introduced the symmetrization S of a tensor defined thermore ξ = −iω where ω is the frequency of the electro- magnetic wave, ∂V is any infinite planar surface in the vac- uum gap between the two planar bodies and dA its surface element, r and r(cid:48) are arbitrary points on the surface of body one. Most importantly G(1) (r,r(cid:48),iξ ) is the scattering Greens tensor [31] of our setup. G(1) (r,r(cid:48),iξ ) can be expressed using the reflection coefficients of the vacuum/PTI interfaces cal- culated in the supplementary material ?? and the different components of the wave vector k = (k(cid:107),kz = iκ)T satisfying k2 = −ξ 2/c2 where c is the speed of light in vacuum. The full expression of G(1) (r,r(cid:48),iξ ) and its derivation can also be found in the supplementary material ??. Note that in Ref. [23] an expression for G(1) (r,r(cid:48),iξ ) in case of a setup with only one PTI half space has already been calculated. Using these results we find first of all, as expected, that there is no lateral force in the ground state of the system and thus the x and y components of F are zero. Furthermore in the sup- plementary material ?? we find a general expression for the normal component of F per unit area A which we define as f ≡ Fz/A, where Fz is the z−component of F. This result is not shown here since we are only interested in the near field behavior of our system. Thus we want to analyze f further un- der the assumption ξ /c (cid:28) k(cid:107) which is often referred to as the nonretarded limit which becomes valid for small separations L, compare Ref. [31]. Under this assumption the reflection Figure 1. We consider a setup as shown here consisting of two semi infinite half spaces filled with a PTI. Additional in each half space there is a applied external magnetic field B+ and B−. The fields are in the xy−plane and are parametrised by the angles Φ+ and Φ− which are the angles between B+, B− and the x−axis, respectively. Furthermore, as shown, we define the angle ϕ as the angle between the parallel component k(cid:107) of the wave vector and the x−axis. and direction is tuneable by the external magnetic fields. Fur- thermore, due to the directionality of the topological surface states, we find that this torque is 2π-periodic with respect to the relative angle between the two bias magnetic fields, in sharp distinction from the π-periodicity occurring for recip- rocal bianisotropic media. We also discuss how the tuneabil- ity of the Casimir torque can be exploited in nanomechanical schemes to induce rotations. We study a setup consisting of two semi infinite half spaces separated by vacuum with separation L and filled with PTIs where a bias magnetic field is applied to each half space (see Fig. 1). The PTI is implemented by a magnetized plasma with an applied bias magnetic field B as an example of a gyrotropic material which is described by a permittivity tensor of the form [30] εxx εεεε = 0 εzz 0 εyz 0 0 −εyz εzz L − ω2 ω2 −iΓω + ω2 T εzz(ω) = 1 + εxxω = 1 +  , with εyz(ω) = iωcω2 p c − (ω + iγ)2) ω(ω2 ω2 p (ω + iγ) c − (ω + iγ)2) ω(ω2 , T T − ω2 − ω2 p ω (ω + iγ) (1) T − ω2 + L − ω2 ω2 −iΓω + ω2 ωp =(cid:112)nq2 if B points in x-direction. This is easily generalized for arbitrary directions of the magnetic field by simply rotat- ing εεεε. The plasma and cyclotron frequencies are given by e/(m(cid:63)ε0) and ωc = Bqe/m(cid:63), respectively, where qe is the electron charge, m(cid:63) its reduced mass, n is the free electron density and γ is the free carrier damping constant. Furthermore Γ represents the phonon damping constant and ωL and ωT are the longitudinal and transverse optic-phonon frequency, respectively. Throughout this paper we will use the following values for the material constants of InSb which have been measured in Ref. [30]: ωL = 3.62 · 1013 rad/s, ωT = 3.39· 1013 rad/s, Γ = 5.65· 1011 rad/s, γ = 3.39· 1012 rad/s, n = 1.07· 1017 cm−3, m(cid:63) = 0.022· me where me is the ���������Body 1Body 2, where s,s (cid:39) r± s,p (cid:39) r± s,p (cid:39) 0 and coefficients simplify significantly to r± pp(ω) (cid:39) −2 + D∓ 2iεyz sin(ϕ − Φ±) D =(cid:112)2εzz(εzz + εxx + (εxx − εzz)cos [2(ϕ − Φ±)]). r± 2 + D∓ 2iεyz sin(ϕ − Φ±) (3) Here ϕ is defined by k(cid:107) = k(cid:107) (sin(ϕ),cos(ϕ))T with k(cid:107) ≡ k(cid:107) (compare Fig. 1) and the "±" indicates the reflection at the lower and upper half space, respectively. Note that this reflec- tion coefficient is not real even when evaluated at imaginary frequencies ω = iξ due to the terms proportional to εyz. But since this term also flips sign under k(cid:107) → −k(cid:107) the Schwarz reflection principle G(1) (r,r(cid:48),iξ ) = G(1)(cid:63) (r,r(cid:48),iξ ) is obeyed pp(k(cid:107),iξ ) = rpp(−k(cid:107),iξ ). which according to [32] implies r(cid:63) Thus it is ensured that the Greens tensor and therefore the Casimir force is real. Finally using the previous result of the reflection coefficients in the nonretarded limit f simplifies to pp(iξ )r− r+ pp(iξ )r− 1− r+ pp(iξ )e−2κL pp(iξ )e−2κL (cid:2)r+ pp(iξ )(cid:3) = − ¯h 16π3L3 dξ dϕ Li3 pp(iξ )r− (4) f =− ¯h 16π3L3 dξ dϕ dκ κ2 ∞(cid:90) 0 2π(cid:90) ∞(cid:90) 0 ∞(cid:90) 2π(cid:90) 0 0 0 where Li3 is the polylogarithm of order three. As in the reciprocal case [31] and for a magnetized plasma with a bias magnetic field perpendicular to the interface [26] we find a simple f ∝ 1/L3 behavior in the nonretarded limit. Therefore we can easily calculate the Casimir energy E per unit area in the nonretarded limit from Eq. (4) by integrating f with respect to L under the boundary condition E → 0 if L → ∞ and eventually find E = L f /2. From this result we can now calculate the Casimir torque T = −∂ E/(∂∆Φ) where ∆Φ = Φ− − Φ+. pp ppr− (cid:2)r+ 0 dϕ Li3 Next, we want to analyze the previous results for the Casimir force, energy and torque. To this end, in Fig. 2, we display the dimensionless Hamaker constant defined by (cid:3) at a fixed gap distance 0 (dξ /ωp)(cid:82) 2π H ≡(cid:82) ∞ L = 100 nm. Note that one can easily retrieve f , E and T from H via f = −ωp¯hH/16π3L3, E = −ωp¯hH/32π3L2 and T = (ωp¯h/32π3L2)∂ H/(∂∆Φ). Before discussing the qual- itative features of these results let us mention a few things about the magnitude of the torque. As is depicted in Fig. 2 (c) the Casimir torque at zero temperature for two semi in- finite PTI half spaces reaches the same order of magnitude as the one for quartz or calcite half spaces kept parallel to a barium titanate half space. Those examples for birefrin- gent plates have been studied in Ref. [9] and we have used the same model for the permittivity including the same val- ues for the constants measured in Ref. [33] to reproduce these results. More concretely this means that the torque for the PTI setup reaches a maximal torque of about 67 pN/m with B = 5 T whereas the quartz (calcite) - barium titanate setup reaches 22 pN/m (317 pN/m). Nevertheless, we remark that 3 Figure 2. Dimensionless Casimir energy, torque and force. In (a) we plot the dimensionless Hamaker constant H as a function of ∆Φ for different values of B = 1,2,3,4,5 T (dots, squares, diamonds, tri- angles, upside down triangles). In (b) we used again the ε-tensor in Eq. (1) with the only exception that we set εyz = 0 to obtain a descrip- tion of a normal anisotropic medium (B = 5 T). The Casimir torque T (∆Φ) is plotted in (c) for different magnetic fields where the same shapes correspond to the same value of B as in (a). Additionally we plotted the Casimir torque for the material model of plot (b) (dotted line) and for a setup where the two PTI half spaces are replaced by one barium titanate and one quartz half space (dashed line). Finally we plotted the Casimir torque as a function of B in (d) for differ- ent values of ∆Φ = 0,π/4,π/2,5π/4,3π/2 (dots, squares, triangles, diamonds, upside down triangles). In all plots we used L = 100 nm. the Casimir torque between two corrugated metals is three or- ders of magnitude larger [5]. But note that these torques are all periodic under a rotation of π of one of the plates around its normal component since their distinguished axes are not directional. Thus, now we want to study the qualitative characteristics of the torque for our setup and we see in Fig. 2 (a) that H(∆Φ) is 2π periodic with a maximum (minimum) when the two mag- netic fields B± point in the same (opposite) direction. This re- sult is therefore qualitatively different from the ones observed when dealing with birefringent half spaces with one in-plane optical anisotropy [9] (compare the dashed line in Fig. 2 (c)) or corrugated metals [5]. Heuristically, this new periodicity can be explained by the fact that our material model does not only have a distinguished axis, but this axis also has a direc- tion. The angle-dependence of the Casimir energy can be un- derstood in more detail by studying the contributions of dif- ferent surface-plasmon polaritons (SPPs) which dominate in the nonretarded limit. To this end, we take a closer look at the spectral decomposition of the Casimir energy evaluated at real frequencies H(ω) ≡(cid:82) 2π H = (cid:82) ∞ (cid:2)r+ 0 (dω/ωp) H(ω) and therefore E, f ∝ (cid:82) ∞ which allows us to see which surface modes contribute the most to the Casimir energy. The total Casimir en- ergy is simply the integral over this spectral energy density, 0 dω H(ω), as can be seen from contour-integral techniques and using 0 dϕ Im(cid:2)Li3 pp(ω)(cid:3)(cid:3) pp(ω)r− ●●●●●●●●●●●●●●●●●■■■■■■■■■■■■■■■■■◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼0.00.51.01.52.03.43.53.63.73.8(a)0.00.51.01.52.03.1963.1983.2003.202(b)●●●●●●●●●●●●●●●●●■■■■■■■■■■■■■■■■■◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼0.00.51.01.52.0-6-4-20246(c)●●●●●●●●●●●■■■■■■■■■■■◆◆◆◆◆◆◆◆◆◆◆▲▲▲▲▲▲▲▲▲▲▲▼▼▼▼▼▼▼▼▼▼▼012345-0.4-0.20.00.20.4(d)4 (6) ζ is the Riemann zeta function: H(ω) ≈ ζ (3)(cid:0)Im(cid:2)r+ pp(ω)(cid:3)Re(cid:2)r∓ ucts Im(cid:2)r± pp(ω)(cid:3)Re(cid:2)r− of Im(cid:2)r+ pp(ω)(cid:3)Re(cid:2)r− pp(ω)(cid:3) pp(ω)(cid:3)(cid:1) pp(ω)(cid:3)Re(cid:2)r+ +Im(cid:2)r− pp(ω)(cid:3) in Figs. 3(a)–(c). The symmetric pp(ω)(cid:3), re- pp(ω)(cid:3) and Im(cid:2)r− pp(ω)(cid:3)Re(cid:2)r+ and antisymmetric coupled SPPs have parallel and antiparallel electric-field vectors on the two surfaces, their frequencies are given by the minimum or maximum of the the two resonances We can clearly see how H(ω,ϕ) is built up from the prod- spectively. As illustrated by the solid lines in Figs. 3(a)– (c), the symmetric coupled SPPs give the dominant positive contribution to the Casimir energy (left peak) while the an- tisymmetric coupled SPPs give a smaller negative contribu- tion (right dip). As further seen in the figures, the difference between the positive and negative contributions is quite pro- nounced for ∆Φ = 0 where the SPPs of the two plates and sub- sequently also the symmetric and antisymmetric SPPs have their largest possible splitting in frequency space, Fig. 3(a), leading to a large net Casimir energy. The splitting is reduced for larger angles, Fig. 3(b) until eventually the single-plate SPPs coincide for ∆Φ = π and the two coupled SPPs become very close in frequency and similar in magnitude. For this case, we have a smaller Casimir energy. Our observations remain valid for general combinations of magnetic-field and wave vector directions and thus also when integrating over all wave-vector directions ϕ ∈ [0,2π] to ob- tain the total spectral energy density H(ω). As seen from Fig. 3(d), this energy density has quite a complex profile as it is the sum over contributions from many SPPs with different resonant frequencies. Nevertheless, we again find that positive and negative contributions are most different in magnitude for ∆Φ = 0 (dots), leading to a large total Casimir energy. As the angle difference increases towards ∆Φ = π (squares), the positive and negative contributions become more similar in magnitude and the total Casimir energy decreases. The advantage of having an in situ tuneability of the torque can be exploited for nanomechenical schemes. Therefore we are going to show how one can set a disk consisting of InSb into rotation via a time dependent bias magnetic field. For instance, we consider a setup as depicted in Fig. 4 (a), where two disks with the same size as considered in Ref. [10] namely with radius r = 20 µm and thickness d = 20 µm consisting of InSb which has a mass density of ρInSb = 5.59g/cm3 [35] are held at a fixed distance from another. Ad- ditionally, in the upper disk there is a constant magnetic field B+ applied which is attached to the disk e.g. via magnetic coating [36]. This disk is free to rotate around the z-axis and therefore the angle between B+ and the x-axis, namely Φ+ ≡ θ (t) may change over time due to the rotation of the disk. The lower disk is fixed and thus not free to rotate. Furthermore in the lower disk there exists a magnetic field which is not at- tached to it and whose angle with the x-axis Φ−(t) is tune- able. Therefore the relative angle between the two magnetic fields is given by ∆Φ(t) = Φ−(t)−θ +(t). The dependence of the torque T on the relative angle between the applied mag- netic fields is very well approximated by T (∆Φ) ∼= T0 sin(∆Φ) Figure 3. Contributions of surface-plasmon polaritons (SPPs) to the Casimir energy. (a) Total spectral energy density H(ω) with ∆Φ = 0,π/4,π/2,π (dots, triangles, diamonds, squares). (b)-(c) Single- pp(ω)(cid:3) (dashed lines) plate SPPs Re(cid:2)r± ϕ − Φ+ = π/2 and ∆Φ = 0,π/2,π ((b),(c),(d)). Re(cid:2)r+ pp(ω)(cid:3) and pp(ω)(cid:3) corresponds to the curves with the resonance at higher Im(cid:2)r+ frequencies whereas Re(cid:2)r− pp(ω)(cid:3) correspond to the pp(ω)(cid:3) (dotted lines) and Im(cid:2)r± pp(ω)(cid:3) and Im(cid:2)r− and angle-resolved spectral energy density H(ω,ϕ) (solid lines) with once with a lower resonance frequency (Ω+ > Ω− in this case); the two single-plate SPP frequencies Ω±(ϕ − Φ±) as given by Eq. (5) are indicated as vertical lines. For all plots we have used B = 3 T. the Schwartz reflection principle as well as the fact that limω→∞ ω2G(1) = 0. Central ingredient to the spectral Casimir energy density are the SPPs of the individual plates which are resonances of pp(ω). The the respective reflection coefficients r+ frequencies of the SPPs are easily found by setting the denom- inators in the reflection coefficients (3) to zero which upon us- ing Eq. (1) and neglecting the photon contribution leads to the dispersion relations pp(ω) and r− (cid:16)(cid:113) Ω±(ϕ − Φ±) = 1 4 6ω2 c + 8ω2 p + 2ω2 ±2ωc sin(cid:2)ϕ − Φ±(cid:3)(cid:1) c cos [2(ϕ − Φ±)] (5) in the lossless limit, as also found in Ref. [23]. The SPP fre- quencies hence depend on the angle between the wave vector and the respective magnetic field, ϕ − Φ±. Thus, for each plate, there is a whole manifold of SPPs at different frequen- cies for different directions, in contrast to the case of simple metal plates [34]. The single-plate SPPs are illustrated by the dotted and dashed lines in Figs. 3(a)–(c) for selected combina- tions of magnetic-field and wave vector directions. In particu- lar, we see that the SPP frequencies of the two plates coincide in the special case ∆Φ = π, Fig. 3(c). When the two plates are brought in close proximity, as is the case in the nonretarded limit considered, the single plate SPPs combine to form symmetric and antisymmetric coupled SPPs [34]. Mathematically, this can be seen from the Casimir spectral energy density using Li3 [z] ≈ ζ (3)z for z (cid:28) 1 where 0.40.60.81.0-100102030(a)ΔΦ=00.40.60.81.0-20-1001020(b)ΔΦ=π/20.40.60.81.0-30-20-100102030(c)ΔΦ=π●●●●●●●●●●●●●●●●●●■■■■■■■■■■■■■■■■■■◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆◆▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲0.20.40.60.81.01.21.4-50050(d)neglecting finite size effects and friction is given by sin(cid:2)Φ−(t)− θ +(t)(cid:3) . d2θ dt2 = 2T0 lr2ρInSb 5 (7) Figure 4. In (a) one can find the setup under consideration. As shown, there are two disks with radius r = 20 µm and thickness d = 20 µm consisting of InSb and held at a distance of L = 100 nm from each other. Furthermore B+ is a static magnetic field whose angle with the x-axis Φ+ only changes if the whole disk rotates. The rotation of the upper disk is described by the angle θ (t) ≡ Φ+(t). In the lower disk there is a magnetic field B− whose direction described by the angle Φ−(t) may change over time although the plate is fixed. In (b) we show the numerical solution to Eq.(7) with B+ = B− = 5 T for the cases Φ−(t) = 0.75πt,0.4πt,0.1πt,2πt,0 (dot, square, dia- mond, triangle, upside down triangle) with solid lines. Additionally we plot Φ−(t) = 0.75πt,0.4πt,0.1πt,2πt,0 (indicated also by dot, square, diamond, triangle, upside down triangle) describing the ro- tation of B− with dotted-dashed lines, to see how the upper disk follows B− if the angular velocity of Φ−(t) is smaller or equal to 0.75 rad/s. (compare Fig. 2 (c)) where T0 = 67pN/m if B+ = B− = 5 T. Thus the equation of motion for the rotation of the upper disk Here r and l are the radius of the upper disk and its thickness, respectively. The result of the numerical solution of Eq. (7) can be found in Fig. 4 (b). As one can see if we let the tun- able magnetic field B− rotate with an angular velocity of up to 0.75 rad/s the upper plate will follow the direction of B− and start to rotate with the same angular velocity of almost one full 2π rotation per second. If B− rotates faster than 0.75 rad/s the upper disk can not follow the direction of B− and therefore it does barely rotate (triangle). To summarize, we have found a Casimir Torque between topological-insulator plates whose direction and magnitude is easily tuneable by the external bias magnetic field. Further- more, the torque is of the same order of magnitude as the ones between quartz or calcium and barium titinate half spaces. We have further shown that in the nonretarded limit this torque is dominated by SPPs which are directional and therefore it is only symmetric under a rotation of 2π of one of the plates around its normal component. This unique periodicity in con- tradistinction from the typical π-periodicity for ordinary bire- fringent media is a clear signature of nonreciprocity. We have shown how the tuneability of the torque between two InSb disks can be exploited to set one of the disks into rotation which offers new possibilities for measurements and nanome- chanical applications of the Casimir torque on small objects. We acknowledge helpful discussions with Francesco Intra- vaia and Alexey Belyanin. This work was supported by the German Research Foundation (DFG, Grants BU 1803/3-1 and GRK 2079/1). S.Y.B is grateful for support by the Freiburg Institute of Advanced Studies. [1] H. B. G. Casimir, Proc. K. Ned. Acad Wet. 51, 793 (1948). [2] I. E. Dzyaloshinskii, E. M. Lifshitz, and L. P. Pitaevskii, Adv. Phys. 10, 165 (1961). [3] S. J. Van Enk, Phys. Rev. A 52, 2569 (1995). [4] F. S. S. Rosa, D. A. R. Dalvit, and P. W. Milonni, Phys. Rev. A 821 (2014). [14] S. A. Hassani Gangaraj, M. G. Silveirinha, and G. W. Hanson, IEEE Journal on Multiscale Multiphys. Comput. Techn. 2, 3 (2017). [15] L. Lu, J. D. Joannopoulos, and M. Soljaci´c, Nat. Photonics 8, 78 (2008). [5] R. B. Rodrigues, P. A. M. Neto, A. Lambrecht, and S. Reynaud, Europhys. Lett. 76, 822 (2006). [6] Y. S. Barash, Radiophys. Quantum El. 21, 1138 (1978). [7] T. G. Philbin and U. Leonhardt, Phys. Rev. A 78, 042107 [8] J. C. Torres-Guzm´an and W. L. Moch´an, J. Phys. A-Math. Gen. [9] J. N. Munday, D. Iannuzzi, Y. Barash, and F. Capasso, Phys. [16] A. B. Khanikaev, S. Hossein Mousavi, W.-K. Tse, M. Kargar- ian, A. H. MacDonald, and G. Shvets, Nat. Mater. 12, 233 (2012). [17] W. Gao, B. Yang, M. Lawrence, F. Fang, B. Bri, and S. Zhang, Nat. Commun. 7, 12435 (2016). [18] M. G. Silveirinha, Phys. Rev. B 92, 125153 (2015). [19] M. G. Silveirinha, Phys. Rev. B 94, 205105 (2016). [20] A. R. Davoyan and N. Engheta, Phys. Rev. Lett. 111 (2013). [21] F. D. M. Haldane and S. Raghu, Phys. Rev. Lett. 100 (2008). [22] S. A. H. Gangaraj, G. W. Hanson, and M. Antezza, Phys. Rev. (2008). 39, 6791 (2006). Rev. A 71 (2005). (2011). (2008). [10] D. Iannuzzi, M. Lisanti, J. N. Munday, and F. Capasso, Solid A 95 (2017). State Commun. 135, 618 (2005). [23] M. G. Silveirinha, S. Gangaraj, G. W. Hanson, and M. Antezza, [11] X. Chen and J. C. H. Spence, Phys. Status Solidi A , 210, 1925 preprint arXiv:1711.04941 (2017). [12] M. Z. Hasan and C. L. Kane, Rev. Mod. Phys. 82, 3045 (2010). [13] S. Raghu and F. D. M. Haldane, Phys. Rev. A 78, 033834 [24] S. Gangaraj, G. W. Hanson, M. Antezza, and M. G. Silveirinha, Phys. Rev. A 97, 022509 (2018). [25] R. M. Ekeroth, P. Ben-Abdallah, J. C. Cuevas, and A. Garc´ıa- Mart´ın, ACS Photonics (2017). ●●■■◆◆▲▲▼▼05101520253004π8π12π16π20π24πt/sθ(t)(b)[26] S. Fuchs, F. Lindel, R. V. Krems, G. W. Hanson, M. Antezza, and S. Y. Buhmann, Phys. Rev. A 96, 062505 (2017). lag, Heidelberg, 2012). [27] S. Y. Buhmann, D. T. Butcher, and S. Scheel, New J. Phys. 14, Phys. A 31, 1641029 (2016). [32] S. Y. Buhmann, V. N. Marachevsky, and S. Scheel, Int. J. Mod. [33] L. Bergstrom, Adv. Colloid Interfac. 70, 125 (1997). [34] C. Henkel, K. Joulain, J.-P. Mulet, and J.-J. Greffet, Phys. Rev. 083034 (2012). [28] L. Zhu and S. Fan, Phys. Rev. Lett. 117, 134303 (2016). [29] M. G. Silveirinha, Phys. Rev. B 95 (2017). [30] E. D. Palik, R. Kaplan, R. W. Gammon, H. Kaplan, R. F. Wallis, and J. J. Quinn, Phys. Rev. B 13, 2497 (1976). [31] S. Y. Buhmann, Dispersion forces: many-body effects, excited atoms, finite temperature and quantum friction (Springer Ver- [35] T. B. Bateman, H. J. McSkimin, and J. M. Whelan, J. Appl. [36] A. G. Grushin, P. Rodriguez-Lopez, and A. Cortijo, Phys. Rev. A 69, 023808 (2004). Phys. 30, 544 (1959). B 84, 045119 (2011). 6 

1305.2451

1

1305

Long-Range Interaction of Singlet-Triplet Qubits via Ferromagnets

We propose a mechanism of a long-range coherent interaction between two singlet-triplet qubits dipolarly coupled to a dogbone-shaped ferromagnet. An effective qubit-qubit interaction Hamiltonian is derived and the coupling strength is estimated. Furthermore we derive the effective coupling between two spin-1/2 qubits that are coupled via dipolar interaction to the ferromagnet and that lie at arbitrary positions and deduce the optimal positioning. We consider hybrid systems consisting of spin-1/2 and ST qubits and derive the effective Hamiltonian for this case. We then show that operation times vary between 1MHz and 100MHz and give explicit estimates for GaAs, Silicon, and NV-center based spin qubits. Finally, we explicitly construct the required sequences to implement a CNOT gate. The resulting quantum computing architecture retains all the single qubit gates and measurement aspects of earlier approaches, but allows qubit spacing at distances of order 1$\,\mu$m for two-qubit gates, achievable with current semiconductor technology.

cond-mat.mes-hall

cond-mat

Long-Range Interaction of Singlet-Triplet Qubits via Ferromagnets Department of Physics, University of Basel, Klingelbergstrasse 82, CH-4056 Basel, Switzerland Luka Trifunovic, Fabio L. Pedrocchi, and Daniel Loss We propose a mechanism of a long-range coherent interaction between two singlet-triplet qubits dipolarly coupled to a dogbone-shaped ferromagnet. An effective qubit-qubit interaction Hamilto- nian is derived and the coupling strength is estimated. Furthermore we derive the effective coupling between two spin-1/2 qubits that are coupled via dipolar interaction to the ferromagnet and that lie at arbitrary positions and deduce the optimal positioning. We consider hybrid systems consisting of spin-1/2 and ST qubits and derive the effective Hamiltonian for this case. We then show that operation times vary between 1MHz and 100MHz and give explicit estimates for GaAs, Silicon, and NV-center based spin qubits. Finally, we explicitly construct the required sequences to implement a CNOT gate. The resulting quantum computing architecture retains all the single qubit gates and measurement aspects of earlier approaches, but allows qubit spacing at distances of order 1 µm for two-qubit gates, achievable with current semiconductor technology. PACS numbers: 73.20.Dx, 71.70.Ej, 03.67.Lx, 76.60.-k I. INTRODUCTION At the heart of quantum computation lies the ability to generate, measure, and control entanglement between qubits. This is a difficult task because the qubit state is rapidly destroyed by the coupling to the environment it is living in. Hence, a qubit system appropriate for quantum computing must be sufficiently large for it to be controllable by experimentalist but must contain as few degrees of freedom as possible that couple to the en- vironment. One of the most successful candidates for encoding a qubit is an electron spin localized in a semi- conductor quantum dot, gate-defined or self-assembled, or a singlet-triplet qubit with two electrons in a dou- ble quantum well.1,2 These natural two-level systems are very long-lived (relaxation time T1 ∼ 1s, see Ref. 3, and decoherence time T2 > 200µs, see Ref. 4), they can be controlled efficiently by both electric and mag- netic fields,5 -- 7 and, eventually, may be scaled into a large network. It has been experimentally demonstrated that qubit-qubit couplings can be generated and controlled efficiently for these systems.8 However the separation be- tween the quantum dots needs to be small (∼ 100nm) and this renders their scaling to a very large number of qubits a perplexed task. Indeed, the physical implemen- tation of quantum dot networks requires some space be- tween the qubits for the different physical auxiliary com- ponents (metallic gates, etc.). It is therefore important to find a way to couple qubits over sufficiently large dis- tances (micrometer scale) to satisfy the space constraint. Another type of promising two-level systems are silicon-based spin qubits. They are composed of nuclear (electron) spin of phosphorus atoms in a silicon nanos- tructure. It has recently been shown that very long de- coherence times T2 ≈ 60ms9 (T2 ≈ 200µs10) and high fi- delity single qubit gates and readout are experimentally achievable. However, the realization of such qubits is subject to randomness of the phosphorus atom position in silicon. Hence, their location is a priori unknown. If two randomly chosen qubits are well-separated from each other, they will hardly interact. On the contrary, when they lie close to each other, it will be difficult to turn- off the interaction since they will not be isolated from each other. In this context it is especially important to be able to couple two qubits over relatively large dis- tances by putting some coupler between them. The aim of our work is to show that this is possible by putting a micrometer-sized ferromagnet between the qubits. We point out that our analysis is general and does not de- pend on the precise nature of the qubits that need to be coupled as long as they interact dipolarly with the ferro- magnet. Hence we think that our analysis is applicable to a variety of other spin qubits systems such as N-V centers in diamond.11 There have been various other propsals over the last years in order to couple spin qubits over large distances. Among them we mention here coupling through a super- conductor,12,13 microwave cavities,14,15 two-dimensional electron gas (RKKY),16 and floating gates.17 -- 20 In this work, we propose and study a system that al- lows for coherent coupling between ST-qubits as well as between spin-1/2 qubits over distances of about one mi- crometer. The coupler is a ferromagnet composed of two disks separated by a thin quasi-1D region, see Fig. 1. The qubits are coupled to the ferromagnet via dipolar interac- tion and they are positioned in the vicinity of each disk. The relevant quantity of the coupler, describing the effec- tive coupling between the distant qubits, is its spin-spin susceptibility -- a slowly spatially decaying real part of the susceptibility is required in order to mediate interactions over long distances. Additionally, in order to have co- herent coupling, the imaginary part of the susceptibility should be sufficiently small. The spatial decay of the spin susceptibility depends strongly on the dimensionality of the ferromagnet -- it is longer ranged in lower dimensions. The dogbone shape of the coupler considered here is thus optimal: it allows for strong coupling to the ferromagnet because many spins of each disk lie close to the qubits, while the coherent interaction between them is mediated by the quasi-1D channel. Actually the statement on the 2 gate operation time to the decoherence time -- of about 10−4 for ST-qubits as well as for spin-1/2 qubits and this is good enough to implement the surface code er- ror correction in such setups. This quantum computing architecture thus retains all the single qubit gates and measurement aspects of earlier approaches, but allows qubit spacing at distances of order 1 µm for two-qubit gates, achievable with current semiconductor technology. The paper is organized as follows. In Sections II and III A we introduce the ferromagnet and ST-qubit Hamil- tonian, respectively. In Sec. III B we derive the effec- tive dipolar coupling between the ST-qubit and the fer- romagnet. In Sec. III C we make use of a perturba- tive Scrhieffer-Wolff transformation to derive the effec- tive coupling between the two ST-qubits that is mediated by the ferromagnet. We determine the optimal position of the qubits relative to the disks of the dogbone. In Sec. III D we construct the sequence to implement the CNOT (and iSWAP) gate and calculate the correspond- ing fidelity of the sequence. In Sec. IV we study the cou- pling between two spin-1/2 qubits positioned at arbitrary location with respect to the adjacent disk of the dogbone- shaped ferromagnet. We derive an effective Hamiltonian for the interaction of the two spin-1/2 qubits mediated by the ferromagnet and determine the optimal position of the qubits. In Sec. IV A we derive the sequence to imple- ment the CNOT (and iSWAP) gate. In Sec. V, we show that spin-1/2 and ST qubits can be cross-coupled lead- ing to hybrid qubits and we derive the effective Hamil- tonian for this case. In Sec. VI, we discuss the range of validity of our effective theory. The on/off switching mechanisms of the qubit-qubit coupling are discussed in Sec. VII. In Sec. VIII, we present a table with the effective coupling strengths and operation times achievable in our setup for four experimentally relevant systems, namely GaAs spin-1/2 quantum dots, GaAs singlet-triplet quan- tum dots, silicon-based qubits, and N-V centers. Finally, Sec. IX contains our final remarks and the Appendices additional details on the models and derivations. II. FERROMAGNET We denote by Sr the spins (of size S) of the ferro- magnet at site r on a cubic lattice and σi stands for the spin-1/2 qubit spins. The ferromagnet Hamiltonian we consider is of the following form (cid:88) r Sz r , (1) (cid:88) (cid:104)r,r(cid:48)(cid:105) HF = −J Sr · Sr(cid:48) + ∆F with J > 0 and ∆F = µB, where B is externally applied magnetic field (see Fig. 1) and µ is the magnetic moment of the ferromagnet spin. The above Hamiltonian is the three-dimensional (3D) Heisenberg model with the sum (cid:105). The ferromag- net is assumed to be monodomain and below the Curie temperature with the magnetization pointing along the z-direction. restricted to nearest-neighbor sites (cid:104)r, r(cid:48) FIG. 1. Model system consisting of two identical double-QDs in the xy-plane and the dogbone-shaped coupler. The dog- bone coupler consists of two ferromagnetic disks of radius R0 connected by a thin ferromagnetic wire of length L. Each double-QD can accommodate one (two) electrons, defining the spin-1/2 (ST-) qubit. Absence of tunneling between the separate double-QD is assumed. Here RL (RR) is the in-plane distance between the left (right) well and the corresponding disk center, while h is vertical distance between the QD and the gate. The red arrow on top of the ferromagnet denote the orientation of its magnetization which is assumed to be monodomain. optimal shape of the coupler is quite general if we con- sider a realistic interaction between the qubit and the coupler, i.e., a Coulomb18 or a dipolar one as herein. Likewise, we derive a Hamiltonian for the effective in- teraction between the distant qubits positioned arbitrar- ily with respect to each disk of the dogbone and deter- mine what is the optimal position for the coupling to be strongest. For ST-qubit the optimal coupling is obtain if one quantum well is positioned directly below the disk center and the other one below the edge of the disk, while for the spin-1/2 qubits it is optimal to place them at the edge of the disk. Similar conclusions about the optimal positioning of the qubits with respect to the coupler were previously obtained for the case of electrostatic interac- tion.18 In the most favorable scenarios described above the coupling strength of 10−2µeV are achieved for ST-qubits as well as for spin-1/2 qubits. The clock speed for such coupling schemes thus varies between 1MHz and 100MHz. We summarize in Tables I and II of Section VIII the coupling strengths and corresponding operation times achievable with our scheme. In order to be useful for quantum computation, one needs to be able to turn on and off the coupling between the qubits. This is efficiently achieved by putting the qubit splitting off-resonance with the internal splitting of the ferromagnet. As we argue below, a modification in the qubit splitting of about one percent of ∆F is enough to interrupt the interaction between the qubits. Finally we derive for both qubit systems the sequence to imple- ment the entangling gates CNOT (and iSWAP) that can be achieved with a gate fidelity exceeding 99.9%. The additional decoherence effects induced solely by the cou- pling to the ferromagnet are negligible for sufficiently low temperatures (T (cid:46) 0.1K).21 We then obtain error thresholds -- defined as the ratio between the two-qubit coupling the singlet and triplet blocks. The effect of SOI in ST-qubit was studied in Ref. 23 and no major influence on the qubit spectra was found. 3 The two qubit states are T0(cid:105) and, in the absence of SOI, the linear combination of the singlet states S(cid:105) = α(2, 0)S(cid:105) + β (1, 1)S(cid:105) + γ (0, 2)S(cid:105), where the coeffi- cients α, β, γ depend on the detuning ε between the two In particular, when ε = 0 we have quantum wells. S(cid:105) = (1, 1)S(cid:105). In what follows, we always consider Hamiltonians only in the qubit subspace, thus the Hamil- tonian of two ST-qubits reads (cid:88) i=1,2 Hτ = − ∆ 2 τ z i , (5) where τ x,y,z are the Pauli matrices acting in the space spanned by vectors {S(cid:105) ,T0(cid:105)} and ∆ is the ST-qubit splitting. B. Dipolar coupling to ST-qubit In this section we derive the dipolar coupling between the ferromagnet and the ST-qubit. To this end we first project the Zeeman coupling to the ST-qubit system on the two-dimensional qubit subspace ∗ HZ = g µB (BL · SL + BR · SR) , (6) where BL (BR) is the magnetic field in the left (right) † L,R,scL,R,s(cid:48) and g∗ is the ef- quantum well, Si fective Land´e factor. After projecting on the qubit space we obtain L,R = (σi)ss(cid:48)c ∗ HZ = g µB(Bz L − Bz R)τ x. (7) With this result we are ready to write down the ferromagnet/ST-qubit interaction Hamiltonian HI = ∗ g µ0µB Bz L(i) − Bz R(i) τ x i , (8) (cid:16) (cid:17) where index i enumerates ST-qubits, and the magnetic field from the ferromagnet can be express through the integral over the ferromagnet Bz L,R(i) = µ0µ 4πa3 (cid:32) dr(i)L,R 1 L,R× r(i)3 Sz r(i)L,R − 3(Sr(i)L,R · r(i)L,R)r(i)z L,R r(i)2 L,R (cid:33) (9) , where the coordinate system for r(i)L (r(i)R) is posi- tioned in left (right) quantum well of the i-th qubit. (cid:88) i=1,2 (cid:90) We would like to stress at this point that even though herein we analyze a specific model for the ferromagnet (Heisenberg model), all our conclusions rely only on the generic features of the ferromagnet susceptibility, i.e., its long-range nature. Furthermore, the gap in the magnon spectrum can originate also from anisotropy. The pres- ence of the gap is an important feature since it suppresses the fluctuations, albeit the susceptibility is cut-off after some characteristic length given by the gap and the fre- quency at which the ferromagnet is probed. III. COUPLING BETWEEN ST-QUBITS The Hamiltonian we consider is of the following form H = HF + Hτ + HI , (2) where Hτ is Hamiltonian of the two ST-qubits2,22 and HI is the dipolar coupling between the ferromagnet and the ST-qubits (see below). A. Singlet-Triplet qubit Hamiltonian A Singlet-Triplet (ST) qubit is a system that consists of two electrons confined in a double quantum well. Herein we assume that the wells are steep enough so that we can consider only one lowest orbital level of each well. Following Ref. 23, we consider also the spin space of the two electrons and write down the total of six basis states (3) 0(cid:105), (cid:17) (cid:17) (cid:16) (cid:16) † † L↑c L↓0(cid:105), † † R↓0(cid:105), R↑c † † † † 1 √2 L↑c R↓ − c L↓c R↑ c † † L↑c R↑0(cid:105), † † † † 1 √2 L↓c R↓ + c L↑c R↑ c † † L↓c R↓0(cid:105), (2, 0)S(cid:105) = c (0, 2)S(cid:105) = c (1, 1)S(cid:105) = T+(cid:105) = c T0(cid:105) = T−(cid:105) = c † † where c R) creates an electron in the Wannier L (c state ΦL (ΦR). The Wannier states are ΦL,R = 1√1−2sg+g2 (ϕ1,2 − gϕ2,1), where s = (cid:104)ϕ1ϕ2(cid:105) = exp[− (a/aB)2] is the overlap of the harmonic oscilla- tor ground state wave functions of the two wells, aB = (cid:112)/mω0 is the Bohr radius of a single quantum dot, ω0 is g = (1 − √1 − s2)/s. Using these six basis states we is the single-particle level spacing, and 2a = l is the in- terdot distance. The mixing factor of the Wannier states can represent the Hamiltonian of the ST-qubits 0(cid:105), (cid:18) HSS (cid:19) H0 = 0 0 HTT,0 . (4) C. Effective coupling between two ST-qubits In writing the above equation we have neglected the spin- orbit interaction (SOI), thus there are no matrix elements Given the total Hamitonian, Eq. (2), we can easily derive the effective qubit-qubit coupling with help of Schrieffer-Wolff transformation The longitudinal susceptibility, defined via 4 (cid:90) ∞ (cid:90) ∞ 0 (cid:90) ∞ 0 Heff = Hτ − lim ν→0+ i 2 −νt [HI (t), HI ] , dte (10) with HI (t) = eiHτ tHI e−iHτ t. We assume that the radius of the two disks is much smaller than the distance between their centers (R0 (cid:28) L). Within this assumption we can take for the suscep- tibility between two points at opposite disks the same as the 1D susceptibility. Next we take only on-resonance susceptibility and make use of the expression τ x(∆) = 1 2 (τ x + iτ y), where τ x(ω) is the Fourier transform of τ x(t) = eiHτ tτ xe−iHτ t. We define the transverse sus- ceptibility in the standard way dte(−iω−η)t[Sz 0 ri(t), Sz χ(cid:107)(ω, ri − rj) = −i lim η→0+ rj ], (12) can be neglected compared to the transverse one because the former is smaller by a factor 1/S and is proportional to the magnon occupation number, see Eq. C2. Therefore the longitudinal susceptibility vanishes at zero tempera- ture, while the is not the case for the transverse suscep- tibility. We arrive finally at the following expression Heff = Hτ + (13) 1 τ x 2 , 9 4Bχ1D⊥ (∆, L)τ x R)(A2 L − A1 L − A2 R)/16π2a6, where B = (µ0µ)2(g∗µB)2(A1 χ1D⊥ is given in Eq. (B23) and Ai L,R = dr(i)L,R r(i) − L,Rr(i)z r(i)5 L,R L,R . (14) (cid:90) χ⊥(ω, ri − rj) = −i lim η→0+ dte(−iω−η)t[S+ ri (t), S − rj ]. (11) Assuming the dogbone shape of the ferromagnet in the above integral and integration only over the adjacent disk, we obtain (cid:16) Ai L,R = 2ihd 2Ri L,RR0 a F (acsc(wi L,R), wi L,R (cid:16) 2 2 ) L,R ) − K(wi (Ri L,R − R0)2 + h2 3Ri L,R (cid:17) + ui (cid:17)(cid:113) L,RE(wi L,R 2 ) − ui L,RE(acsc(wi L,R), wi L,R 2 ) , (15) (Ri L,R + R0)2 + h2 where R0 is the disk radius, Ri L,R is the distance from the adjacent disk axis to the left or right quantum well of the i-th qubit, acsc(x) is the inverse cosecant; F (x, y), K(x) and E(x, y) are the corresponding elliptic integrals. Furthermore, we introduced the notation ui + 2 L,R = Ri L,R R2 L,R = 0 + h2 and wi 4Ri L,R+R0)2+h2 , where h is the distance in the z-direction between the ST-qubit plane and the adjacent disk bottom and d is the disk thickness, see Fig. 1. 1 − L,RR0 (Ri (cid:114) Figure 2 illustrates the dependence of the Ai L,R inte- grals on the position of the quantum wells. Since the coupling constant is given by the difference of this inte- grals for left and right quantum well, we conclude that the strongest coupling is obtained if one quantum well of the ST-qubit is positioned below the disk center and the other exactly below the edge. Furthermore, when h (cid:28) R0 the value of the integral is strongly peaked around R ∼ R0 and this can be exploited as yet an- other switching mechanism -- moving one quantum well away from the edge of the disk. D. Sequence for CNOT gate Two qubits interacting via the ferromagnet evolve ac- cording to the Hamiltonian Heff , see Eq. (13). The Hamiltonian is therefore the sum of Zeeman terms and qubit-qubit interaction. These terms do not commute, making it difficult to use the evolution to implement stan- dard entangling gates. Nevertheless, since Hτ acts only in the subspace spanned by {↑↑(cid:105) ,↓↓(cid:105)} and ∆ (cid:29) J12 = 9Bχ⊥(∆)/4, we can neglect the effect of Heff in this part of the space and approximate it by its projection in the space spanned by vectors {↑↓(cid:105) ,↓↑(cid:105)} (cid:48) eff = Hτ + J12(τ x H 1 τ x 2 + τ y 1 τ y 2 ). (16) Within this approximation, the coupling in H(cid:48) man terms now commute. UiSWAP = e−i(τ x 1 τ y plement the CNOT gate: We consider the implementation of the iSWAP gate24 2 )3π/4, which can be used to im- eff and Zee- 2 +τ y 1 τ x UiSWAP = eiHτ te −iH (cid:48) eff t, (17) where t = 3π/(4J12). When iSWAP is available, the 5 well above this value and hence, though they do not cor- respond to the same noise model, we can expect these gates to be equally suitable for fault-tolerant quantum computation. IV. COUPLING BETWEEN SPIN-1/2 QUBITS In this section we study the coupling of two spin-1/2 quantum dots via interaction with a dog-bone shaped ferromagnet. The Hamiltonian has again the form as in Eq. (2) and we allow for splittings of the spin-1/2 qubits both along x and z direction, (cid:88) (cid:88) Hσ = ∆x 2 σx i + ∆z 2 i=1,2 i=1,2 σz i , (20) where σi are the Pauli operators of the ith spin-1/2 quan- tum dot. Hamiltonian (20) is a generalized version of the Hamiltonian studied in Ref. 21 where we considered splitting along x only. We present here a detailed deriva- tion of the effective coupling between two quantum dots located at an arbitrary position with respect to the dog- bone shaped ferromagnet, i.e., contrary to Ref. 21 we do not assume that the quantum dots are positioned at a highly symmetric point but consider the most general case. This allows us to determine the optimal position- ing of the qubit in order to achieve the strongest coupling between the qubits. The dipolar coupling between the ferromagnet and the spin-1/2 qubits is given by (cid:88) i,r HI = g∗µ0µBµ 4πr3 (cid:18) (cid:19) σi · Sr − 3(σi · r)(Sr · r) r2 , (21) where µD is the magnetic moment of the spin-1/2 qubit. The explicit expressions for the time evolution of the Pauli operators in Heisenberg picture is σ+ i (t) = − − + σz i (t) = i − i 1 ∆2 (i∆ cos(∆t/2) − ∆z sin(∆t/2))2σ+ ∆2 x 2∆2 (cos(t∆) − 1)σ ∆x ∆2 (∆z − ∆z cos(t∆) − i∆ sin(t∆))σz i , ∆x 2∆2 (∆z − ∆z cos(t∆) − i∆ sin(t∆))σ+ ∆x 2∆2 (∆z − ∆z cos(t∆) + i∆ sin(t∆))σ + z + ∆2 ∆2 x cos(t∆) ∆2 σz i , + i − i (22) (cid:112) z. We where we introduced the notation ∆ = also assume that ∆ < ∆F such that the susceptibil- ity χ⊥(∆, r) is purely real -- thus the transverse noise is gapped. By replacing the above expressions in Eq. (10), x + ∆2 ∆2 FIG. 2. Plot of aAi L,R/d defined through Eq. (14) as function of Ri/R0 for different values of h. We see that the value of aAi L,R/d is bigger when the ST-qubit is closer to the disk of the dogbone as expected. Furthermore, by placing the right dot at distance R0 of the disk axis and the left dot on the disk axis, we obtain the strongest value for the effective coupling between the two ST-qubits, see Eq. (13). 4 τ z 4 τ z 4 τ x 4 τ x −i π −i π 1 ei π 2 ei π 2 UiSWAPe UCNOT = e Since H(cid:48) CNOT gate can be constructed in the standard way25 1 UiSWAPei π 4 τ z 2 . (18) eff is an approximation of Heff , the above se- quence will yield approximate CNOT, U(cid:48) CNOT, when used with the full Hamiltonian. The success of the sequences therefore depends on the fidelity of the gates, F (U(cid:48) CNOT). Ideally this would be defined using a minimization over all possible states of two qubits. However, to characterize the fidelity of an imperfect CNOT it is sufficient to con- sider the following four logical states of two qubits18,21: +, 0(cid:105) ,+, 1(cid:105) ,−, 0(cid:105) , and −, 1(cid:105). These are product states which, when acted upon by a perfect CNOT, become the four maximally entangled Bell states Φ+(cid:105) ,Ψ+(cid:105) ,Φ− (cid:105) , and Ψ− (cid:105), respectively. As such, the fidelity of an imper- fect CNOT may be defined, min F (U † CNOTU (cid:48) CNOT) = i∈{+,−},j∈{0,1}(cid:104)i, j U (cid:48) CNOT i, j(cid:105)2. (19) The choice of basis used here ensures that F (U(cid:48) CNOT) gives a good characterization of the properties of U(cid:48) CNOT in comparison to a perfect CNOT, especially for the re- quired task of generating entanglement. For realistic pa- rameters, with the Zeeman terms two order of magni- tude stronger than the qubit-qubit coupling, the above sequence yields fidelity for the CNOT gate of 99.976%. To compare these values to the thresholds found in schemes for quantum computation, we must first note that imperfect CNOT's in these cases are usually mod- elled by the perfect implementation of the gate followed by depolarizing noise at a certain probability. It is known that such noisy CNOT's can be used for quantum compu- tation in the surface code if the depolarizing probability is less than 1.1%.26 This corresponds to a fidelity, ac- cording to the definition above, of 99.17%. The fidelities that may be achieved in the schemes proposed here are 0.00.51.01.52.001234561aAiL,R/d(1)aBi/d(2)h=1(3)h=0.5(4)h=0.1(5)1aAiL,R/d(1)aCi/d(2)h=1R0(3)h=0.5R0(4)h=0.1R0(5)Ri/R0(6)0.00.51.01.52.001234560.00.51.01.52.001234560.00.51.01.52.001234561aAiL,R/d(1)aCi/d(2)h=0.1R0(3)h=0.5R0(4)h=1R0(5)Ri/R0(6)we obtain the effective qubit-qubit coupling duced the following notation for the integrals (g∗µ0µBµ)2 ∗ 1A2χ1D⊥ (∆)σz A 1(∆)σz 2 ( 9 8 (3A1C (3A2C 1 (∆)σz 2 16π2a6 ∗ ∗ 2χ1D⊥ (∆))σ+ 1 χ1D⊥ (∆) − B1A ∗ ∗ 1(∆)σ+ 2 χ1D⊥ (∆) − B2A 1χ1D⊥ (∆))σz 2 − ∗ 1 (∆)σ+ 2 χ1D⊥ (∆))σ (B1B2χ1D⊥ (∆) + 9C1C 2 − − 1 (∆)σ 2 (B1C2χ1D⊥ (∆) + B2C1χ1D⊥ (∆))σ (23) Heff = Hσ + + + + 3 16 3 16 1 32 3 32 − + h.c.) + 1 ↔ 2 , (cid:90) (cid:90) (cid:90) Ai = Ci = Bi = dri dri dri , , i r+ rz i r5 i (r+ i )2 r5 i (cid:18) 1 r3 i 2 − (cid:19) , − 3r+ i r i r2 i 6 (24) (25) (26) where we have denoted χ1D⊥ (∆) = χ1D⊥ (∆, L) and intro- with the coordinate origin for ri at the i-th qubit and the integration goes over the adjacent disk. We also defined the Fourier transforms of the time evolution of Pauli ma- trices σ(t) = eiHσtσe−iHσt as (cid:18) (cid:18) − − 1 ∆2 ∆x ∆2 ∆2 4 ∆z 2 + + σ+ i (∆) = − + (cid:19) ∆2 z 4 σ+ i ∆2 x 4∆2 σ − i , (cid:19) ∆z∆ 2 − ∆ 2 σz i − FIG. 3. Plot of aBi/d defined in Eq. (26) as function of Ri/R0 for different values of h. The value of the integral increases in general by decreasing the value of h. and σz i (∆) = (cid:18) ∆z 2 − ∆x 2∆2 ∆2 x 2∆2 σz i . + (cid:19) + ∆ 2 σ+ i + ∆x 2∆2 (cid:18) ∆z 2 − − ∆ 2 (cid:19) − i σ (27) FIG. 4. Plot of aCi/d defined in Eq. (25) as function of Ri/R0 for different values of h. The value of the integral is peaked around Ri ∼ R0 and it increases in general by decreasing the value of h. By assuming a dogbone-shaped ferromagnet and inte- grating only over the adjacent disk as above, we obtain Ai given in Eq. (15) with Ri L,R replaced by Ri since there is now only one spin-1/2 qubit below each disk of the dog- bone. The remaining integrals yield the following results 0.00.51.01.52.0-505101aAiL,R/d(1)aCi/d(2)h=1R0(3)h=0.5R0(4)h=0.1R0(5)Ri/R0(6)1aAiL,R/d(1)aBi/d(2)h=1(3)h=0.5(4)h=0.1(5)0.00.51.01.52.001234560.00.51.01.52.001234560.00.51.01.52.001234561aAiL,R/d(1)aCi/d(2)h=0.1R0(3)h=0.5R0(4)h=1R0(5)Ri/R0(6)0.00.51.01.52.0-1012341aAiL,R/d(1)aCi/d(2)h=1R0(3)h=0.5R0(4)h=0.1R0(5)Ri/R0(6)0.00.51.01.52.001234560.00.51.01.52.001234560.00.51.01.52.001234561aAiL,R/d(1)aCi/d(2)h=1(3)h=0.5(4)h=0.1(5)1aAiL,R/d(1)aCi/d(2)h=0.1R0(3)h=0.5R0(4)h=1R0(5)Ri/R0(6)i + 2(cid:0)R2 0 + h2(cid:1)(cid:1) K(cid:0)1 − w2 i 7 (cid:1) , (28) Bi = − 2d a Ci = 2d 3a 0 R4 i + 3R2 i (cid:0)h2 − R2 (cid:16) 0 + h2(cid:1)2(cid:17) (cid:1) +(cid:0)R2 − 2(cid:0)R2 (cid:1) (cid:0)(Ri − R0)2 + h2(cid:1) K(cid:0)1 − w2 (cid:112) (cid:113) ((Ri − R0)2 + h2) i (cid:1) E(cid:0)1 − w2 (cid:0)(Ri − R0)2 + h2(cid:1)(cid:0)R2 (cid:112) 0 + h2(cid:1) E(cid:0)1 − w2 − R2 i ((Ri − R0)2 + h2) (Ri + R0)2 + h2 i − R2 (cid:1) i i , (Ri + R0)2 + h2 4RiR0 1 − where wi = (Ri+R0)2+h2 , R0 is the radius of each disk, Ri is the distance of the i-th qubit to the adja- cent dog bone axis, and R0 and h are defined as in Sec. III C. In deriving Eq. (23) we took again only 'on- resonance' terms into account (i.e. we neglected χ1D⊥ (0) and χ1D⊥ (−∆)). Furthermore we assumed, as above, that the susceptibility between two points on different disks of the dogbone is well approximated by the 1D trans- verse susceptibility. In the limit where each quantum dot lies on the vertical axis going through the center of each cylinder of the dogbone, the axial symmetry leads to A1 = A2 = C1 = C2 = 0, B1 = B2 = B, and with ∆z = 0 we recover the result (g∗µ0µBµ)2 B2 32 1σx 1 σy 2 +σz 2 +σx 16 π2a6 χ1D⊥ (∆)(2σy Heff = Hσ+ 1 σz 2) (29) derived in Ref. 21. The analysis carried out herein as- sumes arbitrary positioning of the qubit and allow us to determine the optimal positioning for the strongest coupling. To this end, we analyze integrals Ai, Bi, Ci, see Figs. 2-4. It is readily observed that the coupling strength increases as the vertical distance between the qubit and coupler plane, h, decreases. Additionally, we observe that the strongest coupling strength is obtained when the qubit is positioned below the edge of the adja- cent disk. The derived coupling is valid for any dogbone-like shape of the ferromagnet, i.e., it is not crucial to assume disk shape. A. Sequence for CNOT gate The effective Hamiltonian derived in previous section, Eq. (23), can be re-expressed in the following form Heff = (g∗µ0µBµ)2 16π2a6 χ1D⊥ (∆, L)σT 1 · H · σ2 + 1 2 ∆ · (σ1 + σ2), (30) with ∆ = (∆x, 0, ∆z)T and H being the symmetric ma- trix with all entries being non-zero. The question now arises how to construct the CNOT gate sequence for such a general Hamiltonian. We tackle this problem by taking first the quantization axis to be along the total magnetic field acting on the two qubits and denote by σi Pauli matrix vector with respect to this new quantization axis. The Hamiltonian now reads (g∗µ0µBµ)2 Heff = 16π2a6 χ1D⊥ (∆, L) σT 1 · H · σ2 + 1 2 ∆(σz 1 + σz 2), (31) (cid:69) III D, pendix D. (cid:12)(cid:12)(cid:12)↓↑ (cid:12)(cid:12)(cid:12)↑↓ i.e., we project where the components of the matrix H are given in Ap- (cid:69) We proceed further along the lines presented in Sec. the rotated Hamilto- nian, Eq. (31), on the subspace spanned by vectors { }. This procedure yields the following result (cid:48) eff = J12(σx (32) ∗ J12 = (µ0g stant A12 is defined through the following expression A12. The dimensionless con- 1 σx χ1D⊥ (∆,L) 2 ) + ∆(σz 2 + σy 1 + σz 1 σy (4π)2a6 µB µ)2 2), H 32 , ∆x 2 (36A1A2 + (B1 + 3C1) (B2 + 3C2)) 16∆2 6∆x∆z (A2 (B1 − 3C1) + A1 (B2 − 3C2)) 2∆z (B1B2 + 9C1C2) (∆ + ∆z) 16∆2 A12 = + + 16∆2 . (33) The projected Hamiltonian in Eq. (32) is identical to the one already considered in Sec. III D, Eq. (16). Thus the CNOT gate sequence can be obtained in exactly same way, namely via Eqs. (17) and (18). Similar to the previously studied case of ST-qubits, the CNOT gate sequence described in this section is only ap- proximate one. For realistic parameters, with the Zee- man terms two order of magnitude stronger than the qubit-qubit coupling, this approximate sequence yields fidelity for the CNOT gate similar to the one previously found in Sec. III D. We now use Eq. (33) to determine the optimal posi- tioning of the qubits in order to obtain shortest possi- ble gate operation times. If we assume that the qubit splitting is predominantly along the x-axis (∆x (cid:29) ∆z), we obtain the behavior illustrated in Fig. 5. We con- clude that for all values of h the optimal positioning is below the edge of the adjacent disk. It is interesting to note that when h (cid:28) R0 one can obtain more than two orders of magnitude enhancement compared to the positioning previously studied in Ref. 21. In the oppo- site limit, ∆x (cid:28) ∆z, we observe behavior illustrated in Fig. 6. When also h (cid:28) R0 we recover the same optimal positioning as before -- below the edge of the disk, while when h ∼ R0, positioning the qubit anywhere below the disk yields approximately same coupling strength. (cid:18) form (cid:88) HI = g∗µ0µBµ 4πr3 (cid:16) σ · Sr − (cid:17) Bz L − Bz R τ x i , r ∗ µB + g 8 (cid:19) (35) 3(σ · r)(Sr · r) r2 FIG. 5. Plot of A12 defined in Eq. (33) as function of Ri/R0 for different values of h, assuming R1 = R2 and ∆x = 10∆z. The value of the integral is peaked around Ri ∼ R0 and it increases in general by decreasing the value of h. FIG. 6. Plot of A12 defined in Eq. (33) as function of Ri/R0 for different values of h, assuming R1 = R2 and ∆x = 0.1∆z. The value of the integral is peaked around Ri ∼ R0 only for h (cid:28) R0 and it increases in general by decreasing the value of h. V. COUPLING BETWEEN SPIN-1/2 AND ST-QUBITS In the previous sections we have considered the cou- pling of both spin-1/2 and ST qubits individually. Since each setup has its own advantages and challenges, it is interesting to show these qubits can be cross-coupled to each other and thus that hybrid spin-qubits can be formed. This opens up the possibility to take advantage of the 'best of both worlds'. The Hamiltonian of such a hybrid system reads H = HF + Hσ + Hτ + HI , (34) where the first three term on left-hand side are given by omitting the summation over i in Eqns. (1),(20) and (5), respectively. The interaction term HI has the following with BL,R being given in Eq. (9) when index i is omitted. Continuing along the lines of the previous sections, we perform the second order SW transformation and obtain the effective coupling between the qubits 3(µ0g∗µB µ)2 Heff = χ1D⊥ (∆) (Re[3A(AL − AR) 256π2a6 ∗ ∗ (AL − AR) − 3B(AL − AR) + (C + {σi(∆) → σi, τ x → τ x(∆)}, ∗ )τ xσ+(∆) + h.c(cid:1) ]τ xσz(∆) (36) where AL,R and A are calculated in Eq. (15), while B and C are given in Eq. (28). Similarly as in the previous sections, we find that the optimal coupling for the hybrid case is obtained when the spin-1/2 qubit is positioned below the edge of one of the two discs while one quantum well of the ST-qubit is positioned below the other disc center with the other well being below the disc edge. VI. VALIDITY OF THE EFFECTIVE HAMILTONIAN We discuss herein the validity of the effective Hamil- tonian derived in Sec. III C and Sec. IV. In deriving those effective Hamiltonians we assumed that the state of the ferromagnet adapts practically instantaneously to the state of the two qubits at a given moment. This is true only if the dynamics of the two qubits is slow enough compared to the time scale for the ferromagnet to reach its equilibrium. To estimate the equilibration time of the ferromagnet we use the phenomenological Landau- Lifshitz-Gilbert equation27 in standard notation ∂tm = −γm × Heff + αm × ∂tm, (37) where m(r, t) is the magnetization of the (classical) fer- romagnet at the given time t and spatial coordinate r; γ = 2µB/ is the gyromagnetic ratio and α is Gilbert damping constant. The above equation has to be supple- mented by an equation for the effective magnetic field, Heff (r, t). However, here we do not aim at studying the exact dy- namics of the ferromagnet but rather at giving a rough estimate of its equilibration time scale. We first estimate the time needed for the effective field, Heff (r, t), to reach its final orientation along which the magnetization of the ferromagnet, m(r, t), will eventually point. This time is roughly given by the ratio of the ferromagnet size to the relevant magnon velocity. The typical magnon energy is given by A, and thus the relevant magnon velocity can be 0.00.51.01.52.00204060801001aAiL,R/d(1)aCi/d(2)h=1R0(3)h=0.5R0(4)h=0.1R0(5)Ri/R0(6)0.00.51.01.52.001234560.00.51.01.52.001234560.00.51.01.52.001234561aAiL,R/d(1)aCi/d(2)h=0.1R0(3)h=0.5R0(4)h=1R0(5)Ri/R0(6)1aAiL,R/d(1)A12(2)1x/z=10(1)1⇥10(1)1⇥100(1)0.00.51.01.52.0010203040501aAiL,R/d(1)aCi/d(2)h=1R0(3)h=0.5R0(4)h=0.1R0(5)Ri/R0(6)0.00.51.01.52.001234560.00.51.01.52.001234560.00.51.01.52.001234561aAiL,R/d(1)aCi/d(2)h=0.1R0(3)h=0.5R0(4)h=1R0(5)Ri/R0(6)1aAiL,R/d(1)A12(2)1x/z=0.1(1)estimated as vM = √JSA a/, leading to times of 10ns for magnons to travel over distances of 1µm in ferromag- nets. After Heff (r, t) has reached its final value, it takes additional time for m(r, t) to align along it. This time can be estimated from Eq. (37), as Tp/α, where Tp is the precession time around the effective magnetic field which can be approximated by the externally applied magnetic field (leading to the gap ∆F ). Using the typical Gilbert damping factor of α = 0.00127 we arrive at the time scale of 10ps. Therefore, the total time for the ferromagnet considered herein to reach its equilibrium is about 10ns. This in turn leads to the bottleneck for the operation time. Special care has to be taken for the validity of the per- turbation theory employed herein, since we are working close to resonance, i.e., ∆ − ∆F has to be small but still much larger than the coupling of a qubit to an individual spin of the ferromagnet. For the perturbation theory to be valid we also require the tilt of each ferromagnet spin r (cid:105) (cid:28) 1). The tilt of the central spin of the ferromagnetic disk can be estimated by the integral over the dogbone disk D to be sufficiently small (i.e. (cid:104)S± ± r (cid:105) = (cid:104)S χ⊥(r)B⊥(r) . (38) (cid:90) (cid:90) R D Using cylindrical coordinates we then obtain ± r (cid:105) ∼ (cid:104)S µ0µ2 B 2a ρdρ 0 1 (ρ2 + h2)3/2 S Dρ , (39) 1 where S Dρ is the spatial decay of the transversal suscepti- (ρ2+h2)3/2 is the decay of the dipolar field caus- bility and ing the perturbation of the ferromagnet. Even though each spin is just slightly tilted, we obtain a sizable cou- pling due to big number of spins involved in mediating the coupling. VII. SWITCHING MECHANISMS In this section we briefly discuss possible switching on/off mechanisms. These include changing the split- ting of the qubits and moving them spatially. The for- mer mechanism is based on the dependence of the sus- ceptibility decay length on frequency,21 see Eq. (B23). It is enough to detune the qubit splitting by less then 9 a percent to switch the qubit-qubit coupling effectively off. This is particularly feasible for the ST-qubits where qubit splitting can be controlled by all electrical means. Furthermore, the ST-qubits coupling can be switched off also by rotating them such that AL = AR, see Eq. (14). The spin-1/2 qubits can be switched either by detuning its splitting off-resonance with the magnon gap ∆F or by moving them away from the dogbone disk, see Figs. 5-6. VIII. COUPLING STRENGTHS AND OPERATION TIMES In Tables I and II we present a summary of the ef- fective coupling strengths and operation times that can be obtained in the proposed setup. We assume that the qubits are separated by a distance of 1µm and we give the remaining parameters in the table captions. The column captions correspond to four experimen- tally relevant setups considered in this work (GaAs ST and spin-1/2 quantum dots, silicon-based quantum dots, and NV-centers). The row captions denote respectively the vertical distance h between the qubit and the disk of the ferromagnet, the difference between the qubit split- ting ∆ and the internal splitting ∆F of the ferromagnet (given in units of energy and in units of magnetic field), the obtained effective qubit-qubit interaction, and the corresponding operation time. The operation times obtained in Tables I and II are significantly below the relaxation and decoherence times of the corresponding qubits. Indeed, for GaAs quan- tum dots T1 = 1s (see Ref. 3), and T2 > 200µs (see Ref. 4), respectively. Here we compare to T2 instead of T ∗ 2 since spin-echo can be performed together with two- qubit gates.28 Alternatively, the T ∗ 2 of GaAs qubits can be increased without spin-echo by narrowing the state of the nuclear spins.29,30 For silicon-based qubits decoherence time up to T2 ≈ 200µs is achievable.9 Finally decoherence times of T ∗ 2 ≈ 20µs and T2 ≈ 1.8ms have been obtained for N-V centers in diamond.31 In Table VIII, we summarize the obtained coupling strengths and operation times obtained when a ST-qubit is cross coupled with a spin-1/2 qubit. We have verified that the tilting of the ferromagnet spins given in Eq. (39) remains small. The biggest tilt we obtain (for h = 5 nm) is (cid:104)S± r (cid:105) ≈ 10−7 (cid:28) 1. Thus all the result are within the range of validity of the perturbation theory. IX. CONCLUSIONS We have proposed and studied a model that allows coherent coupling of distant spin qubits. The idea is to introduce a piece of ferromagnetic material between qubits to which they couple dipolarly. A dogbone shape of the ferromagnet is the best compromise since it allows both strong coupling of the qubits to the ferromagnet and long-distance coupling because of its slowly decaying 1D spin-spin susceptibility. We have derived an effective TABLE I. The parameters used to obtain the numbers below are: Land´e factor of the ferromagnet gF = 2; disk radius R0 = 50nm; disk thickness d = 20nm ; Curie temperature T = 550K and thus exchange coupling J/kB ≈ 824K; lattice constant of the ferromagent a = 4A. We consider the case ∆x (cid:28) ∆z. 10 ∆x (cid:28) ∆z Distance h Splitting ∆F − ∆ Coupling strength (CS) GaAs ST QD GaAs ST QD GaAs spin-1/2 QD Silicon-based QD g∗ = 0.4 50 nm g∗ = 0.4 50 nm g∗ = 0.4 50 nm g∗ = 2 25 nm NV-center g∗ = 2 5 nm 1 µeV (43.2 mT ) 0.5µeV (21.6mT ) 10−2 µeV (0.4 mT ) 10−2 µeV (0.1 mT ) 10−1 µeV (0.9 mT ) 1.4 × 10−9 eV 2.4 × 10−8 eV 1.8 × 10−8 eV 2 × 10−10 eV 1.4 × 10−8 Operation time (OS) 470 ns 47ns 3.3 µs 27.4 ns 36.6 ns TABLE II. We use the same parameters as in Table I but consider the case ∆x (cid:29) ∆z. GaAs spin-1/2 QD Silicon-based QD ∆x (cid:29) ∆z Distance h g∗ = 0.4 50 nm g∗ = 2 25 nm NV-center g∗ = 2 5 nm Splitting ∆F − ∆ 10−2 µeV (0.4 mT ) 10−2 µeV (0.1 mT ) 10−1 µeV (0.9 mT ) Coupling strength 1.2 × 10−10 eV 1.8 × 10−8 eV 3.6 × 10−8 eV Operation time 5.5 µs 36.6 ns 18.3 ns Hamiltonian for the qubits in the most general case where the qubits are positioned arbitrarily with respect to the dogbone. We have calculated the optimal position for the effective qubit-qubit coupling to be strongest and es- timated it. For both the singlet-triplet (ST) and spin-1/2 qubits, interaction strengths of 10−2µeV can be achieved. Since decoherence effects induced by the coupling to the ferromagnet are negligible,21 we obtain error thresholds of about 10−4 for ST-qubits and for spin-1/2 qubits. In both cases this is good enough to implement the surface code error correction.32 Finally, for both types of qubits we have explicitly constructed the sequence to implement a CNOT gate achievable with a fidelity of more than 99.9% Our analysis is general and is not restricted to any special types of qubits as long as they couple dipolarly to the ferromagnet. Furthermore, the only relevant quan- tity of the coupler is its spin-spin susceptibility. Hence, our analysis is valid for any kind of coupler (and not just a ferromagnet) that has a sufficiently slowly decaying sus- ceptibility. This quantum computing architecture retains all the single qubit gates and measurement aspects of earlier ap- proaches, but allows qubit spacing at distances of order 1 µm for two- qubit gates, achievable with the state-of- the-art semiconductor technology. Appendix A: Holstein-Primakoff transformation For the sake of completeness we derive in this Ap- pendix explicit expressions for the different spin-spin cor- relators used in this work C αβ(ω, q) = (cid:104)Sα q (ω)Sβ−q(0)(cid:105) . (A1) For this purpose, we make use of a Holstein-Primakoff transformation i =(cid:0)S Sz i = −S + ni, S S+ (cid:1)† − i , i = √2S − (cid:114) ni 2S 1 − ai, and (A2) in the limit ni (cid:28) 2S, with ai satisfying bosonic com- † 33. The creation op- mutation relations and ni = a i ai † erators a i and annihilation operators ai satisfy bosonic (cid:80) commutation relations and the associated particles are called magnons. The corresponding Fourier transforms are straightforwardly defined as a† i e−iq·Riai. In harmonic approximation, the Heisenberg Hamiltonian HF reads q = 1√ N (cid:88) q †
qa qaq , (A3) X. ACKNOWLEDGMENT HF ≈ We would like to thank A. Yacoby, A. Morello, and C. Kloeffel for useful discussions. This work was sup- ported by the Swiss NSF, NCCR QSIT, and IARPA. where
q = ωq + ∆F = 4JS[3 − (cos(qx) + cos(qy) + cos(qz))] + ∆F is the spectrum for a cubic lattice with lattice constant a = 1 and the gap ∆F is induced by the external magnetic field or anisotropy of the ferromagnet. TABLE III. We use the same parameters as in Table I and choose the splitting ∆F − ∆ = 10−2 µeV for the ST-qubit (the splitting of the other qubit is taken from Table I) to determine the coupling strengths and operation times achieved in the hybrid case. The column caption of the table labels GaAs ST-QD, while the row captions label the three other qubit systems, considered in this work, to which it can be hybridized. The left panel corresponds to the case ∆x (cid:28) ∆z while the right panel corresponds to ∆x (cid:29) ∆z. ∆x (cid:28) ∆z GaAs ST QD ∆x (cid:29) ∆z GaAs ST QD Coupling strength Operation time Coupling strength Operation time GaAs spin-1/2 QD 1.7 × 10−9 eV 1.8 × 10−8 eV Silicon-based QD 1.6 × 10−8 eV NV-center 387 ns 36.6 ns 41.1 ns GaAs spin-1/2 QD 1.3 × 10−9 eV 1.6 × 10−8 eV Silicon-based QD 2.2 × 10−8 eV NV-center 506 ns 41.1 ns 29.9 ns ns 11 Appendix B: Transverse correlators (cid:104)S+ q (t)S −q(0)(cid:105) − which gives for ω > ∆F Let us now define the Fourier transforms in the har- monic approximation (cid:88) (cid:88) i i (cid:88) √2S √N (cid:88) √2S √N i i i ≈ − i ≈ −iq·ri S+ e eiq·riS S+ q = − −q = S 1 √N 1 √N i = √2Sa † † −iq·ria −q , e eiq·riai = √2Sa−q .(B1) × sin 2S D = We remark that C +− (ω, r) = 4S/r (cid:18)(cid:114) y 2D(eβω − 1) (cid:19) (cid:90) ∞ 0 dyδ(y + ∆F − ω)× r sin( D 1 (cid:112) (ω − ∆F )/Dr) r (B6) . eβω − 1 From this it directly follows that C +− (ω, r) = 0, ω < ∆F . (B7) C +− q (t)S (t, q) = (cid:104)S+ † −q(t)a−q(cid:105) = 2Sei
qtnq , = 2S(cid:104)a − −q(0)(cid:105) (B2) with
q ≈ Dq2 + ∆F in the long-wavelength approxima- tion. The Fourier transform is then given by We note the diverging behavior of the above correlation function for ∆F = 0 and ω → 0, namely D r(cid:1) sin(cid:0)(cid:112) ω (cid:112) lim ω→0 1 eβω − 1 r = lim ω→0 ω/Dr rβω 1 √ω 1 √Dβ → . (B8) C +− (ω, q) = dte −iωtC +− (t, q) Similarly, it is now easy to calculate the corresponding commutators and anticommutators. Let us define dtei(
q−ω)t (cid:125) 2Snq S⊥(t, q) := 1 2{S+ q (t), S − −q(0)} . It is then straightforward to show that 1 eβω − 1 . (B3) and therefore S⊥(t, q) = Sei
qt(1 + 2nq) , (cid:90) ∞ (B9) (B10) (cid:19) .(B11) (cid:90) ∞ (cid:90) ∞ (cid:123)(cid:122) −∞ 1 √2π 1 √2π = √ −∞ 2πδ(
q−ω) (cid:124) = √2π2Sδ(
q − ω) (cid:90) (cid:90) The corresponding correlator in real space becomes (q = q) C +− dqeiq·rC +− (ω, r) = (ω, q) 1 (B4) (2π)3/2 √2π (2π)3/2 2S 2S eβω − 1 4S 1 r eβω − 1 = = = (cid:90) 1 −1 eβω − 1 1 (cid:90) ∞ (cid:90) ∞ 0 dqδ(Dq2 + ∆F − ω)eiq·r dqdxq2δ(Dq2 + ∆F − ω)eiqrx dqqδ(Dq2 + ∆F − ω) sin(qr) . 0 S⊥(ω, q) = ei(
q−ω)t(1 + 2nq) −∞ S √2π = S√2πδ(
q − ω) (cid:18) 1 + 2 1 eβω − 1 Following essentially the same steps as the ones per- formed above, we obtain the 3D real space anticommu- tator for ω > ∆F S3D⊥ (ω, q) = S coth(βω/2) × (B12) (cid:90) 1 (cid:90) ∞ −1 0 × S D dxdqq2eiqrxδ(
q − ω) (cid:112) Let us now perform the following substitution = y = Dq2, (B5) coth(βω/2) sin( (ω − ∆F )/Dr) r . (B13) Let us now finally calculate the transverse susceptibility defined as χ⊥(t, q) = −iθ(t)[S+ q (t), S − −q(0)] . (B14) 12 In the frequency domain, we then have χ⊥(ω, q) = dtei(
q−ω)t−ηt (B16) (cid:90) ∞ 2iS √2π 2S √2π = − 0 1
q − ω + iη , As before, in the harmonic approximation, one finds and thus in the small q expansion χ⊥(t, q) = iθ(t)2Sei
qt . (B15) χ⊥(ω, q) = − 2S √2π 1 Dq2 + ∆F − ω + iη . (B17) In real space, for the three-dimensional case, we obtain (cid:90) 1 (cid:90) ∞ (cid:90) ∞ 0 −1 dqq 0 2S √2π 4S √2π 2π (2π)3/2 2π (2π)3/2 1 r dxdqq2 1 eiqrx Dq2 + ∆F − ω + iη 1 Dq2 + ∆F − ω + iη χ3D⊥ (ω, r) = − = − (cid:90) ∞ (cid:90) ∞ −∞ P −∞ dqq Making use of the Plemelj formula we obtain for ω > ∆F χ3D⊥ (ω, r) = − = − = − 2S √2π 2S √2π S D 2π (2π)3/2 2π (cid:112) (2π)3/2 1 r 1 r r cos(r (ω − ∆F )/D) 1 sin(qr) Dq2 + ∆F − ω + iη dq q sin(qr) + i (cid:112) Dq2 + ∆F − ω + i sin( S 2D (ω − ∆F )/Dr) r 2S √2π 2π2 (2π)3/2 1 r sin(qr) . (B18) (cid:90) ∞ −∞ dqqδ(Dq2 + ∆F − ω) sin(qr) . (B19) It is worth pointing out that the imaginary part of the susceptibility vanishes where lF is defined as above and the imaginary part van- ishes as above, i.e., (cid:48)(cid:48) χ3D⊥ (ω, r) = 0, ω < ∆F , (B20) and therefore the susceptibility is purely real and takes the form of a Yukawa potential e−r/lF , ω < ∆F , (B21) S D χ3D⊥ (ω, r) = − (cid:113) D r where lF = ∆F −ω . Note also that the imaginary part of the transverse sus- ceptibility satisfies the well-know fluctuation dissipation theorem Similarly for ω > ∆F we have (cid:48)(cid:48) χ1D⊥ (ω, r) = 0, ω < ∆F . sin (cid:16)(cid:112) (cid:112) (cid:114) D (ω − ∆F )/Dr D(ω − ∆F ) (cid:16)(cid:112) (cid:17) (B24) , (B25) (cid:17) . (B26) cos (ω − ∆F )/Dr ω − ∆F χ1D⊥ (ω, r) = S and (cid:48)(cid:48) χ1D⊥ (ω, r) = S 2D S3D⊥ (ω, r) = coth(βω/2)χ3D⊥ (ω, r) (cid:48)(cid:48) . (B22) Appendix C: Longitudinal correlators (cid:104)Sz q(t)Sz−q(0)(cid:105) In three dimensions the susceptibility decays as 1/r, where r is measured in lattice constants. For distances of order of 1µm this leads to a reduction by four orders of magnitude. For quasi one-dimensional ferromagnets such a reduc- tion is absent and the transverse susceptibility reads The longitudinal susceptibility reads (cid:88) χ(cid:107)(t, q) = −iθ(t)[Sz 1 N = −θ(t) q(cid:48),q(cid:48)(cid:48) q(t), Sz−q(0)] eit(
q(cid:48)−
q(cid:48)+q)(cid:104)[a (C1) † † q(cid:48)(cid:48)aq(cid:48)(cid:48)−q](cid:105) . q(cid:48)aq(cid:48)+q, a χ1D⊥ (ω, r) = − S D −r/lF , ω < ∆F , lF e (B23) Applying Wick's theorem and performing a Fourier transform, we obtain the susceptibility in frequency do- main χ(cid:107)(ω, q) = − 1 N (cid:88) k nk − nk+q ω −
k+q +
k + iη , (C2) where nk is the magnons occupation number, which is given by the Bose-Einstein distribution nk = 1 eβ
k − 1 , (C3) where
k is again the magnon spectrum (
k = ωk +∆F ≈ Dk2 + ∆F for small k). Note that the transverse suscep- 13 tibility dominates, since its ratio to longitudinal one is proportional to 1/S (cid:28) 1. Furthermore, the longitudi- nal susceptibility is ∝ nk (cid:28) 1 and thus vanishes at zero temperature, while this is not the case for the transverse. Since we are interested in the decoherence processes caused by the longitudinal fluctuations, we calculate the imaginary part of χ(cid:107)(ω, q) which is related to the fluctua- tions via the fluctuation-dissipation theorem. Performing a small q expansion and assuming without loss of gener- ality ω > 0, we obtain (cid:48)(cid:48) χ3D(cid:107) (ω, q) = π (2π)3 0 (cid:90) (cid:90) ∞ (cid:90) ∞ (cid:90) 1 (cid:90) 1 −1 0 dx 0 = = = 1 4π 1 4π 1 4π = 1 4π dkk2 dkk2 dx dx −1 1 1 −1 (cid:90) 1 (cid:90) 1 dk(nk − nk+q)δ(ωk − ωk+q + ω) eβ(∆F +Dk2) − 1 − eβ(∆F +Dk2) − 1 − (cid:18) (cid:18) (cid:18) (cid:12)(cid:12)(cid:12)(cid:12)(cid:18) ω − Dq2 (cid:19)2  (cid:19)2 (cid:18) ω − Dq2  β e (cid:18) 2Dqx 2Dqx 1 2Dqx 2Dqx (cid:12)(cid:12)(cid:12)(cid:12) 1 dx ∆F +D β e 1 1 eβ(ω+∆F +Dk2) − 1 eβ(ω+∆F +Dk2) − 1 (cid:16) ω−Dq2 (cid:17)2(cid:19) 1 (cid:19) (cid:19) δ − e k − (cid:18) (cid:18) δ(ω − Dq2 − 2Dkqx) ω − Dq2 2Dqx 1 2Dqx (cid:19)(cid:12)(cid:12)(cid:12)(cid:12) 1 (cid:16) ω−Dq2 2Dqx (cid:17)2(cid:19) ∆F +D 2Dqx β ω+∆F +D 1 (cid:16) ω−Dq2 2Dqx (cid:17)2(cid:19) (cid:18) β ω+∆F +D 1 (cid:16) ω−Dq2 2Dqx (cid:17)2(cid:19) − 1 − e − 1 (cid:12)(cid:12)(cid:12)(cid:12)  θ  . − 1 − 1 (cid:19) (cid:18) ω − Dq2 2Dqx (C4) Next, since we are interested in the regime where ω (cid:29) T (and thus βω (cid:29) 1), we have nk (cid:29) nk+q. Further- more, we approximate the distribution function nk = −β(∆F +ωk) e (this is valid when βωk (cid:28) 1) and arrive at 1−e−β∆F +βωk the following expression (cid:90) 1 0 dx 1 2Dqx χ3D(cid:107) (ω, q) (cid:48)(cid:48) = 1 4π (cid:19)2 (cid:18) ω − Dq2 (cid:18) 2Dqx (cid:18) −β ∆F +D e (cid:18) 1 − e−β∆F + βD (cid:17)2(cid:19) (cid:16) ω−Dq2 (cid:16) ω−Dq2 2Dqx 2Dqx e1−e −β∆F −β∆F = − 4βD2q Ei e −β∆F + 1 4 −4 − βDq2 + 2βω − (cid:17)2 βω2 Dq2 (cid:19)(cid:19) , (C5) (cid:114) (cid:90) ∞ 0 2 π 1 r where Ei(z) is the exponential integral function. We also need the the real space represenation obtained after in- verse Fourier transformation, (cid:48)(cid:48) χ3D(cid:107) (ω, r) = (cid:48)(cid:48) dqqχ3D(cid:107) (ω, q) sin(qr) . (C6) In order to perform the above integral we note that the imaginary part of the longitudinal susceptibility, given by Eq. (C5), is peaked around q = ω/D with the width (cid:112) of the peak (1/√βD) much smaller than its position in the regime we are working in (ω (cid:29) T ). For r = 0, the integration over q can be then performed approximately and yields the following expression (cid:48)(cid:48) χ3D(cid:107) (ω, r = 0) (cid:16) ee −β∆F −3β∆F /2 = √πe−e − e√π × Erfc(e (cid:112) −β∆F /2(cid:112) 2β2D3 eβ∆F − 1 −β∆F +β∆F /2 (cid:17)(cid:112) eβ∆F − 1) βω , where Erfc(z) denotes the complementary error function. It is readily observed from the above expression that the longitudinal fluctuations are exponentially suppressed by the gap. Assuming that ∆F (cid:29) T , we obtain the following simplified expression χ3D(cid:107) (ω, r = 0) (cid:48)(cid:48) = √π − eπErfc(1) 2β2D3 e −β∆F(cid:112) (C7) (cid:48)(cid:48) χ1D(cid:107) (ω, r = 0) = 1 4π case of a quasi- one-dimensional ferromagnet (∆F (cid:29) T ) and obtain 14 (cid:90) ∞ (cid:90) ∞ dq −∞ dk 1 −∞ (cid:18) (cid:19) eβ(ω+∆F +Dk2) − 1 e−βDk2 − = = (cid:90) ∞ dk −∞ γ D√βω eβ(∆F +Dk2) − 1 δ(ω − Dq2 − 2Dkq) 1 (cid:112) 1 − e−β∆F + βDk2 −β∆F , e D 1 k2 + ω/D (C9) βω . (C8) where γ is a numerical factor of order 1. We observe that, since J(ω) = χ(cid:107)(ω, r)(cid:48)(cid:48), the longitudinal noise of the ferromagnet is -- as the transverse one -- sub- ohmic34. Next we calculate the longitudinal fluctuations for the Appendix D: Rotated Hamiltonian for CNOT Hate Here we give the general for of the matrix (cid:101)H entering Eq. (31). 2 + 2i(B1C(cid:48)(cid:48) (D1) 2 + B2C(cid:48)(cid:48) 1 )))) H12 = + + H13 = + H23 = + + 1 + B1∆− z + 3C(cid:48) 2 (−6∆xA(cid:48) z (C(cid:48)(cid:48) 3 (∆z∆− 3∆xi (12A2∆∆xA∗ xC(cid:48) 1∆+ z ) + 2∆xA(cid:48)(cid:48) 32∆3 2 + ∆x (−4∆xA(cid:48) 2) + C(cid:48)(cid:48) 1 (B2∆− z − 3C(cid:48) 1 (B2 + 3C(cid:48) 1(B2 + 3C2) + 4A1B2∆x + ∆− z )(cid:1) 2∆− 2∆+ z ))) z (3C1C∗ , 1 − 12A1∆∆xA∗ 3∆xi(cid:0)12A1∆2 1 ∆− 2 + 3C∗ 2(cid:0)∆z z(B2∆x(B1 − 3C(cid:48) 1) − 3B1(2∆zA(cid:48) 2 (2A1∆x + C1∆ − C1∆z) + 9C∗ 3i(cid:0)2i(∆x(∆− 64∆3 2 + ∆xC(cid:48) 1 ∆− 1 ) + 2iB2∆xA(cid:48) z (2A2∆ − C2∆x) + (2B1∆A(cid:48)(cid:48) 2)) − 6B2∆2 64∆2 64∆3 z + 3iC1∆A∗ 2 ∆− xA(cid:48) (cid:1) + 18∆zA∗ z ) + 4(cid:0)A1B2∆2 1(C2∆x − 2A2∆z) + 3∆xC∗ z (2A2∆z + C2∆x) + 18∆xA∗ 64∆2 1) + 2iB1∆zA(cid:48) 64∆2 2(−2A1∆x − C1∆ + C1∆z) − 6∆xA∗ 2 + B2C(cid:48)(cid:48) z (B1C(cid:48)(cid:48) 9∆xC∗ 2∆− 1 1(2A2∆z + C2∆x) , x + A2B1∆z∆− z (cid:1)(cid:1) 2(2A1∆x + C1∆ − C1∆z) 2 (2A1∆x − C1∆ + C1∆z)) 64∆2 3i (6∆zA∗ 1 ∆− 3i (3C∗ z (2A2∆z + C2∆x)) , 64∆2 (D2) (D3) (D4) , (D5) (D6) 2(6B2∆x∆zA(cid:48) 18∆xA∗ 2(6B2∆xA(cid:48) H11 = + H22 = 1 − ∆− z (6B1∆xA(cid:48) 2 + B2∆z(B1 − 3C(cid:48) 1) − 3B1∆zC(cid:48) 2)) + 18∆xA∗ 2(2A1∆x + C1∆ − C1∆z) 1(2A2∆x − C2∆z) + 9∆zC∗ 1 + ∆− z (B1(B2 + 3C(cid:48) 2 (−2A1∆x − C1∆ + C1∆z) − 9C∗ 2) + 3B2C(cid:48) z (C2∆z − 2A2∆x) 2 (2A1∆x − C1∆ + C1∆z) − 18C2∆xA∗ 1)) − 9C∗ 1 ∆− 32∆2 , 1 + 9C2C∗ 1 ∆− z 32∆2 32∆ H33 = 0, and the rest of the components Hij are obtain from Hji by exchanging i ↔ j. 15 1 C. Kloeffel and D. Loss, Annual Review of Condensed Mat- (2010). ter Physics 4, 51 (2013). 2 J. Levy, Phys. Rev. Lett. 89, 147902 (2002). 3 S. 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). 4 H. Bluhm, S. Foletti, I. Neder, M. Rudner, D. Mahalu, V. Umansky, and A. Yacoby, Nat. Phys. 7, 109 (2011). 5 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). 6 F. H. L. Koppens, K. C. Nowack, and L. M. K. Vander- sypen, Phys. Rev. Lett. 100, 236802 (2008). 7 R. Brunner, Y.-S. Shin, T. Obata, M. Pioro-Ladri`ere, and T. Kubo, K. Yoshida, T. Taniyama, Y. Tokura, S. Tarucha, Phys. Rev. Lett. 107, 146801 (2011). 8 M. D. Shulman, O. E. Dial, S. P. Harvey, H. Bluhm, V. Umansky, and A. Yacoby, Science 336, 202 (2012). 9 J. J. Pla, K. Y. Tan, J. P. Dehollain, W. H. Lim, J. J. L. Morton, F. A. Zwanenburg, D. N. Jamieson, A. S. Dzurak, and A. Morello, ArXiv e-prints (2013), arXiv:1302.0047. 10 J. J. Pla, K. Y. Tan, J. P. Dehollain, W. H. Lim, J. J. L. Morton, D. N. Jamieson, A. S. Dzurak, and A. Morello, Nature 489, 541 (2012). 11 V. Dobrovitski, G. Fuchs, A. Falk, C. Santori, and D. Awschalom, Ann. Rev. Condens. Matter Phys. 4, 23 (2013). 12 M.-S. Choi, C. Bruder, and D. Loss, Phys. Rev. B 62, 13569 (2000). 13 M. Leijnse and K. Flensberg, ArXiv e-prints (2013), arXiv:1303.3507. 14 M. Trif, F. Troiani, D. Stepanenko, and D. Loss, Phys. Rev. Lett. 101, 217201 (2008). 15 A. Wallraff, D. I. Schuster, A. Blais, L. Frunzio, R.-S. and R. J. Huang, J. Majer, S. Kumar, S. M. Girvin, Schoelkopf, Nature 431, 162 (2004). 16 Y. Rikitake and H. Imamura, Phys. Rev. B 72, 033308 (2005). 18 L. Trifunovic, O. Dial, M. Trif, J. R. Wootton, R. Abebe, A. Yacoby, and D. Loss, Phys. Rev. X 2, 011006 (2012). 19 C. Flindt, A. S. Sørensen, and K. Flensberg, Phys. Rev. Lett. 97, 240501 (2006). 20 M. Trif, V. N. Golovach, and D. Loss, Phys. Rev. B 75, 085307 (2007). 21 L. Trifunovic, F. L. Pedrocchi, and D. Loss, ArXiv e-prints (2013), arXiv:1302.4017. 22 J. Klinovaja, D. Stepanenko, B. I. Halperin, and D. Loss, Phys. Rev. B 86, 085423 (2012). 23 D. Stepanenko, M. Rudner, B. I. Halperin, and D. Loss, Phys. Rev. B 85, 075416 (2012). 24 A. Imamoglu, D. D. Awschalom, G. Burkard, D. P. Di- Vincenzo, D. Loss, M. Sherwin, and A. Small, Phys. Rev. Lett. 83, 4204 (1999). 25 T. Tanamoto, K. Maruyama, Y. X. Liu, X. Hu, and F. Nori, Phys. Rev. A 78, 062313 (2008). 26 D. S. Wang, A. G. Fowler, and L. C. L. Hollenberg, Phys. Rev. A 83, 020302 (2011). 27 Y. Tserkovnyak, A. Brataas, G. E. W. Bauer, and B. I. Halperin, Rev. Mod. Phys. 77, 1375 (2005). 28 K. Khodjasteh and L. Viola, Phys. Rev. Lett. 102, 080501 (2009). 29 X. Xu, W. Yao, B. Sun, D. G. Steel, A. S. Bracker, D. Gam- mon, and L. J. Sham, Nature 459, 1105 (2009). 30 I. T. Vink, K. C. Nowack, F. H. L. Koppens, J. Danon, Y. V. Nazarov, and L. M. K. Vandersypen, Nat. Phys. 5, 764 (2009). 31 G. Balasubramanian, P. Neumann, D. Twitchen, M. Markham, R. Kolesov, N. Mizuochi, J. Isoya, J. Achard, J. Beck, J. Tissler, V. Jacques, P. R. Hemmer, F. Jelezko, and J. Wrachtrup, Nat Mater 8, 383 (2009). 32 R. Raussendorf and J. Harrington, Phys. Rev. Lett. 98, 190504 (2007). 33 W. Nolting and A. Ramakanth, Quantum Theory of Mag- netism (Springer, 2009). 34 D. P. DiVincenzo and D. Loss, Phys. Rev. B 71, 035318 17 K. Flensberg and C. M. Marcus, Phys. Rev. B 81, 195418 (2005). 

1705.10526

1

1705

Synthetic Antiferromagnetic Spintronics: Part of a collection of reviews on antiferromagnetic spintronics

Spintronic and nanomagnetic devices often derive their functionality from layers of different materials and the interfaces between them. This is especially true for synthetic antiferromagnets - two or more ferromagnetic layers that are separated by metallic spacers or tunnel barriers and which have antiparallel magnetizations. Here, we discuss the new opportunities that arise from synthetic antiferromagnets, as compared to crystal antiferromagnets or ferromagnets.

cond-mat.mes-hall

cond-mat

Synthetic Antiferromagnetic Spintronics R. A. Duine,1, 2 Kyung-Jin Lee,3, 4 Stuart S. P. Parkin,5,6,7 and M. D. Stiles8 1)Institute for Theoretical Physics, Universiteit Utrecht, Leuvenlaan 4, 3584 CE Utrecht, The Netherlands 2)Department of Applied Physics, Eindhoven University of Technology, P.O. Box 513, 5600 MB Eindhoven, The Netherlands 3)Department of Materials Science and Engineering, Korea University, Seoul 02841, Korea 4)KU-KIST Graduate School of Converging Science and Technology, Korea University, Seoul 02841, Korea 5) Max Planck Institute for Microstructure Physics, Halle (Saale) D-06120, Germany. 6) IBM Research–Almaden, San Jose, CA 95120, USA. 7)Institute of Physics, Martin Luther University Halle-Wittenberg, Halle (Saale) D-06120, Germany 8)Center for Nanoscale Science and Technology, National Institute of Standards and Technology, Gaithersburg, Maryland 20899, USA (Dated: 20 May 2017) Spintronic and nanomagnetic devices often derive their functionality from layers of different materials and the interfaces between them. This is especially true for synthetic antiferromagnets - two or more ferromagnetic layers that are separated by metallic spacers or tunnel barriers and which have antiparallel magnetizations. Here, we discuss the new opportunities that arise from synthetic antiferromagnets, as compared to crystal antiferromagnets or ferromagnets. 1 Introduction Advances in nanofabrication techniques for magnetic materials - such as Fe, Ni, Co , Cr and their alloys - have, since the late 1980's, enabled researchers to engineer stacks of thin (nanometers) layers of magnetic and nonmagnetic material. The study of such magnetic multilayers and superlattices – i.e., periodic multilayers – has led to many discoveries and potential applications. The first among these is the existence of a coupling between two magnetic layers adjacent to the same non-magnetic spacer.1–3 This interlayer exchange coupling is essentially a spin- dependent Ruderman-Kittel-Kasuya-Yosida (RKKY) coupling that is rooted in Friedel-like spatial oscillations in the spin density of the non-magnetic spacer that are caused by the adjacent ferromagnets. The oscillating spin density in turn leads to an interlayer exchange coupling constant that oscillates with the distance between the ferromagnetic layers.4–8 By changing the thickness of non-magnetic material between two magnetic layers one can therefore tune the interaction from ferromagnetic - preferring parallel alignment - to antiferromagnetic, preferring antiparallel alignment, whereas for thick spacers the interlayer exchange coupling is suppressed. Trilayers, multilayers or superlattices, in which the interaction between magnetic layers is antiferromagnetic are now commonly referred to as synthetic antiferromagnets (see Figure). The antiferromagnetic coupling was crucial for the discovery that the resistance of metallic magnetic multilayers depends on the relative orientation of the magnetization in adjacent layers.9,10 This finding - called giant magnetoresistance (GMR) or, in the case of a tunneling barrier, tunneling magnetoresistance (TMR) - kickstarted the field of nanomagnetism and spintronics. Before we discuss the physics of GMR in somewhat more detail, let us first compare synthetic with crystal anti- ferromagnets, i.e., the antiferromagnets found in nature as bulk single crystals. Most important is the difference in physical origin between the interlayer exchange coupling described above and the direct exchange or superexchange coupling in crystal antiferromagnets. As a result, the interlayer exchange coupling is much weaker and typically two or more orders of magnitude smaller than ordinary exchange coupling. This relatively weak exchange coupling implies that external fields and anisotropy can compete with the antiferromagnetic order in synthetic antiferromagnets, but in crystal antiferromagnets an external field can typically only reorient the magnetization, for example near a spin-flop transition. F o r a synthetic antiferromagnet, a transition from antiparallel to parallel arrangement of the magnetization direction in the magnetic layers can therefore be achieved by magnetic fields that can be easily reached in a laboratory. This was an essential part of the discovery of GMR. Furthermore, the repeat distance of the antiferromagnetic order in synthetic antiferromagnets is larger than in crystal antiferromagnets. While in the latter the magnetic order alternates on atomic length scales, the layer thickness in magnetic multilayers is typically several nanometers. For most situations the electron motion within 2 one magnetic layer is therefore appropriately described by a spin-dependent semi-classical model.11 GMR, for instance, is typically modelled by taking into account electron diffusion within the magnetic layers supplemented with spin-dependent resistances of the various layers and interfaces between them, as well as spin relaxation. For crystal antiferromagnets this picture breaks down as the electrons are in that case phase coherent over a region that is larger than the length scale of the antiferromagnetic order. The large tunability of synthetic antiferromagnets via layer thickness and material composition, together with the above-mentioned unique energy and spatial scales, leads to physical phenomena that are either very different from, or have no counterpart in, crystal antiferromagnets. In the remainder of this commentary we review some of these phenomena and discuss possible new directions. Statics and dynamics One of the most basic applications of the tunability of magnetic multilayers is in magnetic field sensing.12 At a rudimentary level, field sensors consist of a layer stack that exhibits GMR or TMR and that typically incorporates a free layer and a pinned layer. Interlayer exchange coupling and the remaining dipolar fields cause an offset of the response curve of the free layer - the resistance as a function of field is no longer symmetric around zero field. Large offset fields degrade the performance of sensors that operate in the linear regime by reducing the change of resistance with changes in field. To minimize dipolar fields, and thereby reduce hysteresis and increase sensitivity, the pinned is chosen to be a synthetic antiferromagnet. Recent results demonstrate that the tunability of magnetic multilayers allows for simultaneous optimization of dipolar and offset fields.13–15 The relative orientation between pinned and free layers is used to encode binary information in magnetic random access memory (MRAM).16 Two key physical properties for the design of such memories are the thermal stability, determined by the energy barrier that separates the two states of one bit, and the current or field required to switch, i.e., write the bit. For the goal of keeping thermal stability h i g h while reducing the energy required to write the bit, current-induced switching that uses spin-polarized currents is most attractive.17 It was shown that using synthetic antiferromagnets c a n reduce the critical current for switching while maintaining thermal stability.18 Moreover, it was predicted that, owing to the dynamics resulting from interlayer exchange coupling within a synthetic antiferromagnet free layer, the time for switching can be significantly reduced.19 En route to demonstrating switching, spin-current-driven auto-oscillations in devices with a synthetic antiferromagnetic layer were demonstrated.20 These are characterized by high power of emission, narrow 3 linewidth, and nontrivial field and current dependence, due to the multiple layers whose dynamics are governed by the interlayer exchange.21,22 Linewidth broadening of dynamic modes in synthetic antiferromagnetic superlattices, on the other hand, has been reported in Ref. 22 and was interpreted to result from spin pumping from the magnetic layers into the nonmagnetic spacers. Switching by spin-orbit torques has recently been demonstrated in devices containing synthetic antiferromagnetic layers.23,24 Such switching requires a symmetry breaking field. In Ref. 23 this was provided by the interlayer exchange coupling between an in-plane and out-of-plane magnetized layer, while in Ref. 25 exchange bias was used. Domain walls and solitons The motion of domain walls and solitons in synthetic antiferromagnets has been studied both theoretically and experimentally.26–32 Due to partial cancellation of dipolar fields, the domain walls in synthetic antiferromagnets tend to be narrower than in ferromagnetic wires. The narrow walls are beneficial for the design of race track memories based on driving trains of closely-spaced domain walls. Saariskoski et al.27 showed that the interlayer exchange coupling has two important novel features for domain wall motion driven by spin transfer torques. First, the interlayer exchange coupling leads to an attraction between the domain walls in the two ferromagnetic layers that make up the synthetic antiferromagnet. This coupling reduces effects of pinning as both domain walls mutually assist each other in overcoming potential barriers. Second, the attraction between domain walls in different ferromagnetic layers of the synthetic antiferromagnet leads to increased velocities of the domain walls if they are not on top of each other. While domain walls in crystal antiferromagnets are also expected to move at large – in comparison to ferromagnets – velocities, the underlying physics is different because of their much stronger antiferromagnetic exchange and the much shorter length scales. The narrow walls and fast domain motion have been seen experimentally in a reduced threshold current for domain wall motion in a synthetic ferrimagnet in Ref. 33. Indirect support was obtained via noise measurements in synthetic antiferromagnets, while Ref. 24 attributed switching behaviour in a multilayer to domain wall motion and nucleation. Most of the ongoing research on current-driven domain wall motion focuses on multilayers that involve heavy elements with strong spin-orbit coupling, such as Pt or Ta, as the nonmagnetic layers. This spin-orbit coupling has several new physical consequences. First of all, the boundary between heavy nonmagnetic and magnetic metal leads to interface-induced Dzyaloshinskii-Moriya (DM) interactions (see box). These interactions lead to chiral domain walls - domain walls in which the spins have a preferred sense of rotation. In particular, the interfacial DM 4 interactions stabilize N´eel domain walls that are efficiently driven by spin-orbit torques. These spin-orbit torques are also induced by the spin-orbit coupling in the heavy metallic layer. The interlayer exchange coupling stabilizes the N´eel structure of the walls such that they can be driven more efficiently by spin-orbit torques. On top of this, the interlayer exchange coupling leads to additional torques that efficiently drive the domain walls in both ferromagnetic layers of the synthetic antiferromagnet in the same direction. It was experimentally shown in Ref. 28 that large domain wall velocities (of up to 750 m/s) are obtained for domain walls in synthetic antiferromagnets of Co/Ni magnetic layers, separated by thin layers of Ru. A completely different type of domain wall motion was demonstrated by Lavrijsen et al.31. The domain walls considered in this work are kink defects in the antiferromagnetic order of the synthetic antiferromagnet such that two adjacent magnetic layers have magnetizations that are parallel, rather than antiparallel. In a superlattice designed with different interlayer exchange coupling and different magnetic layer thicknesses, these authors were able to demonstrate injection and propagation of kinks by external field pulses. This latter work is an attractive example of how the large tunability of synthetic antierromagnets, and magnetic multilayers in general, can be put to use to enable new functionalities. Outlook One of the areas of interest in synthetic antiferromagnets is magnetic skyrmions. While there has not been much work yet in this direction, Ref. 35 pointed out that the Magnus force that acts on a skyrmion - or more generally on two-dimensional magnetic structures that have a nonzero winding number - and pushes them sideways is counteracted and cancelled by the interlayer exchange coupling in synthetic antiferromagnets. This is beneficial for applications in which skyrmions are driven along narrow wires, as the sideways motion may cause the skyrmions to interact with the edges of the wire and disappear. Another attractive research direction is to alter the magnetic properties of the synthetic antiferromagnets in-situ by electric fields,36 rather than by engineering different systems with different properties. For example, it was theoretically proposed in Ref. 37 that the interlayer exchange coupling can be switched from ferromagnetic to antiferromagnetic either by an electric field or by making use of a ferroelectric layer. These and other examples ultimately show that synthetic antiferromagnets can in some sense be thought of as materials with properties in between those of ferromagnets and antiferromagnets. Some of their properties derive from the ferromagnetic layers that constitute them, whereas other properties derive from the coupling between these layers. Engineering and exploiting their tunability will surely lead to new physics and applications for the years to come. 5 Acknowledgments RD is supported by the Stichting voor Fundamenteel Onderzoek der Materie (FOM), the European Research Council (ERC), and is part of the D-ITP consortium, a program of the Netherlands Organization for Scientific Research (NWO) that is funded by the Dutch Ministry of Education, Culture and Science. K.-J.L. was supported by the National Research Foundation of Korea (NRF) (NRF-2015M3D1A1070465, NRF-2017R1A2B2006119). Figure: Schematic of synthetic antiferromagnets. a, bilayers with in-plane magnetization. b, bilayers with out-of- plane magnetizations. c, multilayers. The arrows within each ferromagnetic layer indicate the direction of magnetization. 6 BOX – DZYALOSHINSKII-MORIYA (DM) EXCHANGE INTERACTIONS Apart from the well-known Heisenberg-type exchange interactions between spins in ferromagnetic materials, there can exist so-called Dzyaloshinskii-Moriya (DM) interactions in magnetic systems that lack a center of inversion and exhibit spin-orbit coupling. In the situation of two magnetic atoms (spin S1 and S2) in presence of a third non-magnetic atoms with spin-orbit coupling (see figure) this interaction has the form DS1S2, with D the Dzyaloshinskii vector and R1 and R2 the respective positions of the magnetic atoms with respect to the non-magnetic one. This interaction clearly favors a certain misaligned and turning sense (chirality) of the magnetic moments. Most important for magnetic multilayers are the DM interactions induced by interfaces between magnetic metals and metals with strong spin-orbit coupling. Figure adapted from Nature Nanotech. 8, 152 (2013). REFERENCES 1Grunberg, P., Schreiber, R., Pang, Y., Brodsky, M. B. & Sowers, H. Layered magnetic structures: Evidence for antiferromagnetic coupling of Fe layers across Cr interlayers. Phys. Rev. Lett. 57, 2442–2445 (1986). 2Majkrzak, C. F., Cable, J. W., Kwo, J., Hong, M., McWhan, D. B., Yafet, Y., Waszczak, J. V. & Vettier, C . Observation of a Magnetic Antiphase Domain Structure with Long-Range Order in a Synthetic Gd-Y Superlattice. Phys. Rev. Lett. 56, 2700–2703 (1986). 3Salamon, M. B., Sinha, S., Rhyne, J. J., Cunningham, J. E., Erwin, R. W., Borchers, J. & Flynn, C. P. Long-range incommensurate magnetic order in a Dy-Y multilayer. Phys. Rev. Lett. 56, 259–262 (1986). 4Slonczewski, J. C. Conductance and exchange coupling of two ferromagnets separated by a tunneling barrier. Phys. Rev. B 39, 6995–7002 (1989). 5Parkin, S . , More, N . & Roche, K . P . Oscillations in Exchange Coupling and Magnetoresistance in Metallic Superlattice Structures: Co/Ru, Co/Cr, and Fe/Cr. Phys. Rev. Lett. 23, 130 (1990). 6Edwards, D. M., Mathon, J., Muniz, R . B . & Phan, M. S. Oscillations of the exchange in magnetic multilayers as an analog of de Haas - van Alphen effect. Phys. Rev. Lett. 67, 493–496 (1991). 7 7Bruno, P. Theory of interlayer magnetic coupling. Phys. Rev. B 52, 411 (1995). 8Faure-Vincent, J . , Bellouard, C . T . C . , Popova, E . , Hehn, M . , Montaigne, F. & Schuhl, A . Interlayer magnetic coupling interactions of two ferromagnetic layers by spin polarized tunneling. Phys. Rev. Lett. 89, 107206 (2002). 9Baibich, M . N . , Broto, J . M . , Fert, A . , Nguyen Van Dau, F . , Petroff, F., Eitenne, P . , Creuzet, G . , Friederich, A, & Chazelas, J. Giant Magnetoresistance of (001)Fe/(001)Cr Magnetic Superlattices. Phys. Rev. Lett. 61, 2472–2475 (1988). 10Binasch, G., Grunberg, P., Saurenbach, F . & Zinn, W. Enhanced magnetoresistance in layered magnetic structures. Phys. Rev. B 39, 4828–4830 (1989). 11Valet, T. & Fert, A. Theory of the perpendicular magnetoresistance in magnetic multilayers. Phys. Rev. B 48, 7099–7113 (1993). 12Parkin, S., Jiang, X., Kaiser, C., Panchula, A., Roche, K . & Samant, M. Magnetically engineered spintronic sensors and memory. Proc. IEEE 91, 661–679 (2003). 13Bandiera, S . , Sousa, R . C . , Dahmane, Y . , Ducruet, C . , Portemont, C . , Baltz, V . , Auffret, S . , Prejbeanu, I. L. & Dieny, B. Comparison of synthetic antiferromagnets and hard ferromagnets as reference layer in magnetic tunnel junctions with perpendicular magnetic anisotropy. IEEE Magn. Lett. 1, 3000204 (2010). 14Kent, A. D. &Worledge, D. C. A new spin on magnetic memories. Nat. Nanotechnol. 10, 187–191 (2015). 15Apalkov, D., Dieny, B . & Slaughter, J. M. Magnetoresistive Random Access Memory. Proc. IEEE 104, 1796– 1830 (2016). 16Engel, B . N . , Akerman, J . , Butcher, B . , Dave, R . W . , DeHerrera, M . , Durlam, M . , Grynkewich, G . , Janesky, J . , Pietambaram, S. V., Rizzo, N. D., Slaughter, J. M., Smith, K., Sun, J. J. & Tehran, S. A 4-Mb toggle MRAM based on a novel bit and switching method. IEEE Trans. Magn. 41, 132–136 (2005). 17Lee, S. W. & Lee, K . J . Current-induced magnetization switching of synthetic antiferromagnetic free layer in magnetic tunnel junctions. J. Appl. Phys. 109, 07C904 (2011). 18Hayakawa, J . , Ikeda, S . , Lee, Y . M . , Sasaki, R . , Meguro, T . , Matsukura, F . , Takahashi, H. & Ohno, H . Current- induced magnetization switching in MgO barrier based magnetic tunnel junctions with CoFeB/Ru/CoFeB synthetic ferrimagnetic free layer. Jpn. J. Appl. Phys. 45, L1057–L1060 (2006). 19Bergman, A., Skubic, B., Hellsvik, J., Nordstrom, L., Delin, A . & Eriksson, O. Ultrafast switching in a 8 synthetic antiferromagnetic magnetic random-access memory device. Phys. Rev. B 83, 224429 (2011). 20Houssameddine, D., Sierra, J. F., Gusakova, D., Delaet, B., Ebels, U., Buda-Prejbeanu, L. D., Cyrille, M. C., Dieny, B., Ocker, B., Langer, J . & Maas, W. Spin torque driven excitations in a synthetic antiferromagnet. Appl. Phys. Lett. 96, 072511 (2010). 21Krebs, J. J., Lubitz, P., Chaiken, A . & Prinz, G. A. Observation of magnetic resonance modes of Fe layers coupled via intervening Cr (invited). J. Appl. Phys. 67, 5920–5924 (1990). 22Tanaka, K., Moriyama, T., Nagata, M., Seki, T., Takanashi, K., Takahashi, S. & Ono, T. Linewidth broadening of optical precession mode in synthetic antiferromagnet. Appl. Phys. Express 7, 063010 (2014). 23Lau, Y .-C ., Betto, D., Rode, K ., Coey, J . M . D . & Stamenov, P . Spinorbit torque switching without an external field using interlayer exchange coupling. Nat. Nanotechnol. 11, 758 (2016). 24Bi, C . , Almasi, H . , Price, K . , Newhouse-illige, T . , Xu, M . , Allen, S . R . , Fan, X. & Wang, W . Spin-orbit torque switching of synthetic antiferromagnets. Phys. Rev. B 95, 104434 (2017). 25van den Brink, A., Vermijs, G., Solignac, A., Koo, J., Kohlhepp, J. T., Swagten, H . J . M . & Koopmans, B. Field-free magnetization reversal by spin-hall effect and exchange bias. Nat. Commun. 7, 10854 (2015). 26Herranz, D . , Guerrero, R ., Villar, R . , Aliev, F . G . , Swaving, A . C. , Duine, R. A ., van Haesendonck, C . & Vavra, I . Anomalous low-frequency noise in synthetic antiferromagnets: Possible evidence of current-induced domain-wall motion. Phys. Rev. B 79, 134423 (2009). 27Saarikoski, H., Kohno, H., Marrows, C. H. & Tatara, G. Current-driven dynamics of coupled domain walls in a synthetic antiferromagnet. Phys. Rev. B 90, 094411 (2014). 28Yang, S . - H . , Ryu, K.-S. & Parkin, S . Domain-wall velocities of up to 750 m/s driven by exchange-coupling torque in synthetic antiferromagnets. Nat. Nanotechnol. 10, 221–226 (2015). 29Shiino, T., Oh, S.-H., Haney, P. M., Lee, S.-W., Go, G., Park, B . G . & Lee, K.-J. Antiferromagnetic Domain Wall Motion Driven by Spin-Orbit Torques. Phys. Rev. Lett. 117, 087203 (2016). 30Komine, T. & Aono, T. Micromagnetic analysis of current-induced domain wall motion in a bilayer nanowire with synthetic antiferromagnetic coupling. AIP Adv. 6, 056409 (2016). 31Lavrijsen, R . , Lee, J . - H . , Ferna´ndez-Pacheco, A . , Petit, D . C . M . C . , Mansell, R. & Cowburn, R . P . Magnetic ratchet for three-dimensional spintronic memory and logic. Nature 493, 647–50 (2013). 32Fernandez-Pacheco, A . , Steinke, N . J . , Mahendru, D . , Welbourne, A . , Mansell, R . , Chin, S . L . , Petit, 9 D . , Lee, J. H., Dalgliesh, R., Langridge, S . & Cowburn, R. P. Magnetic State of Multilayered Synthetic Antiferromagnets during Soliton Nucleation and Propagation for Vertical Data Transfer. Adv. Mater. Interfaces 3, 1600097 (2016). 33Lepadatu, S ., Saarikoski, H., Beacham, R., Benitez, M. J., Moore, T. A., Burnell, G., Sugimoto, S., Yesudas, D., Wheeler, M . C . , Miguel, J . , Dhesi, S . S . , McGrouther, D . , McVitie, S . , Tatara, G. & Marrows, C . H . Very low critical current density for motion of coupled domain walls in synthetic ferrimagnet nanowires. arXiv:1604.07992 (2016). 34Bauer, U., Lide, Y. & Beach, G. S. D. Magneto-ionic control of interfacial magnetism. Nat. Mater. 14, 174–181 (2014). 35Zhang, X., Zhou, Y. & Ezawa, M. Magnetic bilayer-skyrmions without skyrmion Hall effect. Nat. Commun. 7, 10293 (2015). 36Tsymbal, E. Y. Spintronics: Electric toggling of magnets. Nat. Mater. 11, 12 (2012). 37Fechner, M., Zahn, P., Ostanin, S., Bibes, M . & Mertig, I. Switching magnetization by 180 degrees with an electric field. Phys. Rev. Lett. 108, 197206 (2012). 1 0 

1502.02956

2

1502

Strongly metallic electron and hole 2D transport in an ambipolar Si-vacuum field effect transistor

We report experiment and theory on an ambipolar gate-controlled Si-vacuum field effect transistor (FET) where we study electron and hole (low-temperature 2D) transport in the same device simply by changing the external gate voltage to tune the system from being a 2D electron system at positive gate voltage to a 2D hole system at negative gate voltage. The electron (hole) conductivity manifests strong (moderate) metallic temperature dependence with the conductivity decreasing by a factor of 8 (2) between 0.3 K and 4.2 K with the peak electron mobility ($\sim 18$ m$^2$/Vs) being roughly 20 times larger than the peak hole mobility (in the same sample). Our theory explains the data well using RPA screening of background Coulomb disorder, establishing that the observed metallicity is a direct consequence of the strong temperature dependence of the effective screened disorder.

cond-mat.mes-hall

cond-mat

Strongly metallic electron and hole 2D transport in an ambipolar Si-vacuum field effect transistor Binhui Hu,1, 2 M. M. Yazdanpanah,1, 2 B. E. Kane,1, 2 E. H. Hwang,2, 3, 4 and S. Das Sarma2, 3 1Laboratory for Physical Sciences, University of Maryland at College Park, College Park, MD 20740 2Joint Quantum Institute, University of Maryland, College Park, Maryland 20742, USA 3Condensed Matter Theory Center, Department of Physics, University of Maryland, College Park, Maryland 20742-4111 4SKKU Advanced Institute of Nanotechnology and Department of Physics, Sungkyunkwan University, Suwon 440-746, Korea (Dated: May 18, 2018) We report experiment and theory on an ambipolar gate-controlled Si(111)-vacuum field effect transistor (FET) where we study electron and hole (low-temperature 2D) transport in the same device simply by changing the external gate voltage to tune the system from being a 2D electron system at positive gate voltage to a 2D hole system at negative gate voltage. The electron (hole) conductivity manifests strong (moderate) metallic temperature dependence with the conductivity decreasing by a factor of 8 (2) between 0.3 K and 4.2 K with the peak electron mobility (∼ 18 m2/Vs) being roughly 20 times larger than the peak hole mobility (in the same sample). Our theory explains the data well using RPA screening of background Coulomb disorder, establishing that the observed metallicity is a direct consequence of the strong temperature dependence of the effective screened disorder. It is now well-established that, quite generically, "high mobility" and "low-density" semiconductor-based effec- tively metallic 2D systems can manifest anomalous low temperature "metallic" (i.e., dσ/dT < 0 with σ being the 2D conductivity) transport behavior, where a modest variation in temperature (T ≈ 0.1K−4K) could decrease σ by a large amount, with variations in σ(T ) by as large as a factor of ∼2 observed in Si MOSFET based 2D elec- tron systems (2DES) [1] and GaAs-based 2D hole systems (2DHS) [2] in a temperature regime (0.1K − 4K) where phonons are inactive due to the Bloch-Gruneisen (BG) suppression of phonon occupancy. This strong metallic temperature dependence (the precise quantitative defi- nition of "high-mobility" and "low-density" is materials dependent and varies from system to system [3]) in 2D semiconductor structures is in sharp contrast with 3D metals where, at low temperatures (. 10K), the conduc- tivity typically saturates to a disorder-dependent (and temperature-independent) constant (σ0) as the system enters the BG phonon scattering regime with σ(T ) ≈ σ0−O(T 4−6). By contrast, the observed anomalous σ(T ) in high-mobility and low-density 2D semiconductor sys- tems appears to follow a leading-order linear tempera- ture dependence, with σ(T ) ≈ σ0 − O(T ) over a wide temperature range (0.1K − 4K) although eventually (for T < 50 mK) σ(T ) saturates (or manifests weak local- ization behavior[4]), perhaps because of electron heating effects invariably present in semiconductors. In the current work we report three remarkable new results on the anomalous 2D metallic behavior by com- bining experiment and theory: (i) we present the first experimental results on the 2D metallic behavior in an ambipolar system where the metallic temperature de- pendence in the conductivity is separately observed for both 2DES and 2DHS in the same device simply by changing an external gate voltage (we mention that low- mobility ambipolar Si 2D devices have earlier been re- ported in the literature[5] without any observation of the temperature-dependent metallic transport, which is the focus of our study); (ii) our observed 'metallicity' (i.e., the temperature-induced fractional change in the conductivity) is an unprecedented factor of 8 (2) in the 2DES (2DHS) for T = 0.3 − 4 K range and carrier density ∼ 3 × 1011 cm−2 -- this is by far the largest temperature-induced fractional change in the metallic conductivity ever reported in any non-superconducting system in such a small temperature window -- for exam- ple, earlier-studied 2D Si 'metallic' systems in the litera- ture [6] show at most a factor of 3 change in the conduc- tivity in the same temperature window; (iii) we explain our observations qualitatively by calculating the tem- perature and density dependent RPA-Boltzmann con- ductivity using a realistic model of screened Coulomb disorder where the main difference between 2DES and 2DHS arises from the effective valley degeneracy being 6 and 1 respectively by virtue of the qualitatively differ- ent band structures in the conduction and the valence band of the Si(111) ambipolar FET structure used in our experiment -- this leads to the screened effective dis- order in the 2DES being much weaker (and much more strongly temperature-dependent) than in the 2DHS, al- though both see exactly the same bare disorder, explain- ing the remarkable difference in the mobility and the tem- perature dependence in the two cases. The ambipolar FET device (we have actually stud- ied several such devices with similar results) we study is a high-purity and high-mobility hydrogen-terminated atomically flat (and nominally undoped) Si(111) struc- (a) electron ) s V / 2 m 20 ) s V / 2 m ( µ 15 10 hole 101 T=0.3K T=4.2K electron (b) electron 100 10-1 T=0.3K T=4.2K hole 1 10 -2 density (10 cm ) 11 hole T=4.2K T=0.3K electron T=4.2K ( µ 5 T=0.3K 0 -60 -40 -20 40 60 0 20 V (V) g (c) electron 0 -8 -6 -4 -2 0 6 -2 density (10 cm ) hole electron 2 11 4 (d) T=0.3K ) S m ( σ T=4.2K 10 1 0.1 T=0.3K T=4.2K 8 8 10 hole 8 6 4 2 ) 2 - m c 1 1 0 1 ( y t i s n e d hole 25 20 15 10 5 ) S m ( σ 101 ) S m ( σ 100 10-1 0 -8 -6 -4 -2 0 4 2 11 6 -2 density (10 cm ) T=0.3K electron (e) ~n1.3 8 -8 -6 -4 -2 0 6 -2 density (10 cm ) 2 11 4 101 ) S m 100 T=4.2K electron (f) ~n2.2 ~n 1.0 hole ~n 1.0 ( σ hole ~n 1.5 10-1 ~n 2.4 10 1 -2 density (10 cm ) 11 1 10 -2 density (10 cm ) 11 FIG. 1. Experimental results for an ambipolar Si(111)- vacuum FET. (a) Electron and hole densities vs. gate voltage are shown. (b) Mobility and (c) conductivity for both elec- trons (right panel) and holes (left panel) measured at T=0.3K (black) and T=4.2K (red) are shown as a function of density in linear scale. Inset shows the same data in (b) as a log plot, where top (bottom) two lines is for electron (hole). (d) Conductivity vs. density relation is shown in semi-log scale. (e) and (f) show conductivity vs. density in log-log scale at T=0.3K and T=4.2K, respectively. The fitting exponents α in the relation of σ ∼ nα are given in the figures. ture, where the electrons (holes) are induced near the H-Si(111) surface by applying a positive (negative) volt- age through a vacuum barrier. Details of fabrication and characterization of such Si-vacuum FETs (in contrast to the usual Si-SiO2 MOSFETs) with ultrahigh mobility, for both electrons [7] and holes [8], have been described elsewhere [7 -- 9]. The new aspect of the current work is the fabrication of a Si-vacuum ambipolar FET where we can go from a 2DES to a 2DHS simply by changing the external gate voltage from positive to negative in a single device. Such ambipolar devices have earlier been stud- ied for low-mobility Si(100)-SiO2 MOS systems[5] and for GaAs/AlGaAs based undoped 2D structures [10 -- 13], but no 2D metallic behavior or temperature-dependent conductivity was reported in either case. The great ad- vantage of such an ambipolar device is that the 2DES and the 2DHS "feel" precisely the same bare disorder, and therefore a direct comparison between the conduc- tivity data between electrons and holes in the same de- vice should give us considerable insight into the intrinsic 2 aspects of the intriguing metallic 2D phase. In Figs. 1 and 2 we show the experimental results for one typical ambipolar device (along with some of our theoretical results to be described below also shown in Fig. 2(c)). The most important salient features of the experimental results (Figs. 1 and 2) are briefly sum- marized here: (i) The peak electron (hole) mobility in the device reaches 180,000 cm2/Vs (9,000 cm2/Vs) at T = 0.3K with an astonishing factor of 20 difference in the electron versus hole mobility although both are be- ing measured in the same sample at the same tempera- ture and carrier density -- by contrast, the correspond- ing high-mobility GaAs/AlGaAs 2D ambipolar devices shows electron and hole mobilities typically within a fac- tor of 2 − 3 of each other [13] which is expected just based on the electron/hole effective mass difference; (ii) the observed temperature dependence in the conductiv- ity is much stronger for the electrons than for the holes; (iii) the extrapolation of the mobility (or the conductiv- ity) to low gate voltage indicates a rough mobility gap of 3V, which is approximately the indirect band gap of Si as expected in an ambipolar device; (iv) the carrier density dependence of the conductivity, σ(n) ∼ nα where n is the electron (or hole) density and α is the density exponent of conductivity [14], gives α = 1.3 (1.0) for electrons (holes) at T = 0.3K and α = 2.2 (1.0) for electrons (holes) at T = 4.2K in the 'high-density' (> 3× 1011 cm−2) regime with α increasing at lower density most likely due to den- sity inhomogeneity effects [15] which become strong at low carrier density in the presence of random charged im- purity centers; (v) both the conductivity and the mobil- ity (for both electrons and holes) increase monotonically with increasing density with no sign of conductivity satu- ration (or mobility decrease) at our highest experimental density (∼ 1012 cm−2), indicating that surface (or inter- face) roughness scattering, which dominates 2D carrier transport in standard Si-SiO2 MOSFETs [16, 17] and in the GaAs-based gated ambipolar devices [12, 13], plays (at best) a minor role in the Si-vacuum 2D structures (similar to the corresponding situation in GaAs-based modulation-doped high-mobility 2D systems [18]) due to the atomically flat nature of our high-quality Si(111) sur- face; (vi) the main density regime of interest (> 1.5×1011 cm−2) for the study of the 2D effective metallic behav- ior has kF l ≫ 1 (Fig. 2(a)) for both electrons and holes (with kF , l being Fermi wave vector and mean free path, respectively) implying that a Boltzmann theory based transport theory should work well for both the 2DES and the 2DHS existing in our ambipolar device; (vii) the threshold carrier density (obtained by extrapolating the measured electron or hole Hall density to zero conductiv- ity in Fig. 1(a) -- (d)) is almost the same for the 2DES and the 2DHS with the hole system having only a very small amount of (∼ 8 × 109 cm−2) higher surface charge states populated by the gate, indicating the very high quality of the sample and that the two systems have almost iden- (a) g =2v g =6v ) S m ( σ 10 1 0.1 100 l F k 10 g =1v 1 k l = 1 F -8 -6 -4 -2 0 4 -2 density (10 cm ) 2 11 6 8 0.01 0 (b) 101 100 10-1 10-2 ) S m ( σ 10 ) S m ( σ 1 0.1 0 1 1 2 T (K) 3 4 0 2 T (K) 1 3 (c) 3 2 T (K) 3 4 FIG. 2. (a) Calculated kF l using experimental conductivity with different values of gv. (b) The experimentally measured conductivity as a function of temperature for several electron densities, n = 0.85, 0.91, 0.99, 1.15, 1.30, 1.46, 1.61, 1.92, 2.39, 3.17, 3.94, 4.72, 5.49, 6.12×1011 cm−2 (bottom to top). (c) Calculated conductivity in the presence of ionized channel impurities and surface roughness for electron densities n = 1.3, 1.5, 2.0, 3.0, 4.0, 4.5, 5.5, 6.0×1011 cm−2 (bottom to top). Inset in (c) shows the experiment/theory results together for carrier densities (n = 1.3, 1.46, 1.61, 1.92, 2.39, 3.17, bottom to top) demonstrating reasonable agreement. tical background disorder (thus indicating that the very large difference in the electron versus hole mobility is an intrinsic effect not arising from any extrinsic difference in the disorder in two cases). In Fig. 2(c) we show our theoretically calculated σ(n, T ) for the 2DES to be compared with the cor- responding experimental data in Fig. 2(b) whereas Fig. 2(a) shows that for density > 1011 cm−2 the Boltz- mann theory should be valid as kF l ≫ 1 applies for the experimental conductivity. The finite temperature 2D Boltzmann theory has already been described by us in details in our earlier work on Si MOSFETs [3, 19], and we only mention that the results shown in Fig. 2(c) use finite-temperature and finite-wave vector RPA screening [20] of the background disorder which is taken to be un- intentional random quenched charged impurity centers in the 2D Si layer itself as well as a small amount of surface roughness. The most important parameter determining the theoretical σ(n, T ) here is the valley degeneracy (gv) which is taken to be gv = 6 consistent with the bulk conduction band structure of six equivalent conduction band minima along the three symmetry axes of Si. Such a high (gv = 6) valley degeneracy for the 2DES on the Si(111) surface is consistent with earlier experimental re- sults on high mobility Si-vacuum FETs [21], but not with most low mobility Si-SiO2 MOSFET samples studied in the literature [16] where gv = 2 is typically found most likely because of uniaxial interface strain at the Si-SiO2 interface which lifts four of the valleys higher in energy leaving a ground state valley degeneracy of gv = 2. Our independent SdH analysis of magnetoresistance oscilla- tions (not shown, but see, e.g., Ref. [21]) in the sample confirms that the system indeed has gv = 6. We empha- size that the effective mass difference between electrons and holes in our Si(111) 2D system (only a factor of 1.67) cannot explain at all the large difference in our measured mobility. We note that although theory and experiment agree reasonably well qualitatively using gv = 6 (and even quantitatively for density above 1.3 × 1011 cm−2) in Fig. 2, we have not attempted any quantitative fitting because the precise disorder parameters are unknown in the experiment. (We mention that using gv = 2 in the theory gives results in qualitative and quantitative dis- agreement with the experimental data for the 2DES.) In Fig. 3 we show the theoretical results for two tem- peratures (T = 0.3 K and 4.2 K) for both 2DES (both gv = 2, 6 are shown for the sake of comparison in Fig. 3(a)) and 2DHS (only gv = 1 is shown since the Si valence band has no valley degeneracy). The theory reproduces all the key features of the experimental data provided gv = 6 (1) is used for the 2DES (2DHS). In particular, there is a very large (∼ a factor of 20) differ- ence in the 2DES and 2DHS mobilities although both see identical disorder. We have checked explicitly that this mobility difference arises mainly from the different val- ley degeneracies in the two cases -- for example, changing the electron or hole effective mass does not modify the results much whereas changing the valley degeneracy for either electrons or holes has a huge effect. The theory also reproduces the much stronger temperature depen- dence of the 2DES conductivity compared with the 2DHS case, again arising primarily from the valley degeneracy difference. Finally, we show in Fig. 3(b) the calculated exponent (with σ ∼ nα) for 2DES and 2DHS at T = 0.3 K and 4.2 K, finding for the 2DES (with gv = 6) the exponent α = 1.3 and 2.2 for T = 0.3 K and 4.2 K, re- spectively, and for the 2DHS (with gv = 1) α = 1.0 and 1.1 for T = 0.3 K and 4.2 K respectively. These theo- retically calculated exponents are in agreement with the experimental data shown in Fig. 1(e) and 1(f). We em- phasize that all our theoretical results assume the same bare disorder for both 2DES and 2DHS and incorporate all the realistic microscopic details.[16] We note that the 10 1 ) S m ( σ 0.1 -10 hole (a) electron g =6v g =2 v ) S m ( σ -5 5 density (10 cm ) -2 11 0 10 102 101 100 10-1 10-2 (b) α=1.3 α=2.2 α=1.0 α=1.1 1 10 density (10 cm ) -2 11 a) Calculated conductivity as a function of carrier FIG. 3. density for two different temperatures T = 0.3K (solid lines) and 4.2 K (dashed lines). For the hole system the effective mass mh = 0.5m0 and gv = 1 are used, and for the electron system me = 0.3m0 and gv = 2, 6 are used. (b) The high density exponents in the relation of σ ∼ nα are shown for electron system with gv = 6 (top two lines) and hole system with gv = 1 (bottom two lines). The solid (dashed) lines indicate the calculated conductivity at T = 0.3 K (4.2 K). The dotted lines are for guide of exponents α shown in the figure. small threshold difference of 8× 109 cm−2 surface charge density between the 2DES and the 2DHS has no quanti- tative effect on our theoretical results. Before concluding, we provide a simple intuitive un- derstanding of the theory which successfully explains the data. At first, the conductivity data appear intriguing because of the huge difference in the quantitive behav- ior of the conductivity for 2DES and 2DHS in the same sample. Basically, this difference arises from the sub- stantial difference in the effective screened disorder seen by the two kinds of carriers (electrons or holes) in the same ambipolar device because of the large difference in the conduction/valence band structure giving rise to gv = 6 (electrons)/1 (holes). The crucial dimensionless quantities [3, 14] determining both the mobility and the temperature dependence of the conductivity are qT F /kF , where qT F and kF are the 2D Thomas-Fermi and Fermi wave vectors, and T /TF , where TF (= EF /kB) is the Fermi temperature (EF is the Fermi energy). We empha- size that the dimensionless interaction strength parame- ter rs ∝ m/√n is in fact larger for the hole system than the electron system, and is not relevant in controlling the temperature dependence with the relevant control parameter being qs(= qT F /kF ) ∼ g1.5 v rs which is much larger for the 2DES compared with the 2DHS in our sys- tem. We assume that only screened Coulomb disorder (and not phonon scattering) determines the conductivity in the 2D Si system as is expected in the T = 0.3 − 4.2 K range.[16] The constraint on T /TF is simply that it should not be too small in the experimental temperature window for σ(T, n) to have strong T -dependence. It is easy to see that T (h) v = 1 and the respective electron/hole effective masses. (The hole effective mass is known to increase from 0.3 to 0.36 F ∼ 3 using g(e) v = 6, g(h) F /T (e) 4 s = (me/mh)(g(e) in the experimental carrier density range[22], but this does not affect our theory in any quantitative manner.) Thus, the fractional conductivity change, being linear in T /TF at low temperatures [20], is expected to be much larger for 2DES than for 2DHS in a given temperature range simply by virtue of the electron valley degener- acy being six times larger! But this is only a part of the explanation. The central quantity of key importance in the theory [3, 19, 20] is the dimensionless screening strength qs = qT F /kF which determines both the over- all magnitude of the mobility as well as the magnitude of the temperature dependence. It is easy to see that v /g(h) s /q(h) q(e) v )1.5 ∼ 8, which implies that the effective screening is much stronger for the 2DES than for the 2DHS, leading to the conclusion that the mobility ratio for the 2DES compared with 2DHS goes approximately as (me/mh)(q(e) )2 ∼ 35, whereas the exact numerical calculation gives more a factor of 20 dif- ference since the system is not strictly in the qs ≫ 1 and/or T /TF ≪ 1 limit that these analytical approxima- tions assume. Similarly, the simplest analytical theory predicts that the temperature-induced fractional conduc- tivity change should go as (q(e) F ) ∼ 20 as- suming that qs ≫ 1 and T /TF ≪ 1 for both 2DES and 2DHS. Since these strong-screening and low-temperature conditions are not obeyed in the experiment, the realis- tic difference in the temperature-dependent conductivity, as obtained in our numerical results, is around a factor of 4. For the theoretical details we refer to the existing literature.[3, 19, 20] F /T (e) s /q(h) s /q(h) )(T (h) s s In conclusion, we report the first experimental obser- vation of very strong metallic temperature dependence of 2D conductivity in both electrons and holes in an ambipolar Si(111) system, with the electron (hole) con- ductivity changing by a factor of 8 (2) at a density of 3 × 1011 cm−2 for a temperature change from 0.3 K to 4.2 K with the electron mobility being 20 times larger than the hole mobility. We provide a theoretical expla- nation for the data using an RPA-Boltzmann transport theory assuming background screened Coulomb disorder as the primary scattering mechanism. Our work con- clusively shows the dominant role of valley degeneracy in determining 2D transport through carrier screening of Coulomb disorder and explains the main difference be- tween the electron and the hole conductivity as arising from the factor of six difference in their valley degener- acy. In particular, we find that the dimensionless pa- rameter qT F /kF and not the so-called rs-parameter with rs ∼ m/√n controls the strength of metallicity in the anomalous 2D metallic phase of semiconductor systems. This work is supported by LPS-CMTC. Part of this work was performed at the NIST Center for Nanoscale Science and Technology, and the support of the Maryland NanoCenter through its FabLab is also acknowledged. [1] S. V. Kravchenko, G. V. Kravchenko, J. E. Furneaux, V. M. Pudalov, and M. D'Iorio, Phys. Rev. B, 50, 8039 (1994); A. Lewalle, M. Pepper, C. J. B. Ford, E. H. Hwang, S. Das Sarma, D. J. Paul, and G. Red- mond, Phys. Rev. B, 66, 075324 (2002); V. M. Pudalov, arXiv:cond-mat/0405315 (2004). [2] Y. Hanein, U. Meirav, D. Shahar, C. C. Li, D. C. Tsui, and H. Shtrikman, Phys. Rev. Lett. 80, 1288 (1998); M. J. Manfra, E. H. Hwang, S. Das Sarma, L. N. Pfeiffer, K. W. West, and A. M. Sergent, Phys. Rev. Lett. 99, 236402 (2007); A. R. Hamilton, M. Y. Simmons, M. Pepper, E. H. Linfield, and D. A. Ritchie, Phys. Rev. Lett. 87, 126802 (2001). [3] S. Das Sarma and E. H. Hwang, Phys. Rev. B 69, 195305 (2004). [4] S. Das Sarma, E. H. Hwang, K. Kechedzhi, and L. A. Tracy, Phys. Rev. B 90, 125410 (2014). [5] V. Dolgopolov, C. Mazure, A. Zrenner, and F. Koch, J. App. Phys. 55, 4280 (1984). [6] E. Abrahams, S. V. Kravchenko, and M. P. Sarachik, Rev. Mod. Phys. 73, 251 (2001); C. M. Varma, Z. Nussi- nov, and W. van Sarloos, Phys. Rep. 361, 267 (2002); S. V. Kravchenko and M. P. Sarachik, Rep. Prog. Phys. 67, 1 (2004); B. Spivak, S. V. Kravchenko, S. A. Kivelson, and X. P. A. Gao, ibid. 82, 1743 (2010); S. Das Sarma, S. Adam, E. H. Hwang, and E. Rossi, ibid. 83, 407 (2011). [7] K. Eng, R. N. McFarland, and B. E. Kane, Appl. Phys. Lett. 87, 052106 (2005); Physica E 34, 701 (2006). [8] B. Hu, T. M. Kott, R. McFarland, and B. E. Kane, Appl. Phys. Lett. 100, 252107 (2012). [9] K. Eng, R. N. McFarland, and B. E. Kane, Phys. Rev. Lett. 99, 016801 (2007). 5 [10] B. E. Kane, L. N. Pfeiffer, K. W. West, and C. K. Har- nett, Appl. Phys. Lett. 63, 2132 (1993). [11] Y. Hirayama, J. Appl. Phys. 80, 588 (1996). [12] J. C. H. Chen, D. Q. Wang, O. Klochan, A. P. Micolich, K. Das Gupta, F. Sfigakis, D. A. Ritchie, D. Reuter, A. D. Wieck, and A. R. Hamilton, Appl. Phys. Lett. 100, 052101 (2012). [13] A. F. Croxall, B. Zheng, F. Sfigakis, K. Das Gupta, I. Farrer, C. A. Nicoll, H. E. Beere, and D. A. Ritchie, Appl. Phys. Lett. 102, 082105 (2013). [14] S. Das Sarma and E. H. Hwang, Phys. Rev. B 88, 035439 (2013). [15] C. Jiang, D. C. Tsui, and G. Weimann, Appl. Phys. Lett. 53, 1533 (1988); see also S. Das Sarma, E. H. Hwang, and Q. Li, Phys. Rev. B 88, 155310 (2013) and references therein. [16] T. Ando, A. B. Fowler, and F. Stern, Rev. Mod. Phys. 54, 437 (1982). [17] S. Das Sarma and E. H. Hwang, Phys. Rev. B 89, 121413 (2014) and references therein. [18] E. H. Hwang and S. Das Sarma, Phys. Rev. B 77, 235437 (2008). [19] S. Das Sarma and E. H. Hwang, Solid State Comm. 135, 579 (2005) and references therein. [20] F. Stern, Phys. Rev. Lett.44, 1469 (1980); A. Gold and V. T. Dolgopolov, Phys. Rev. B 33, 1076 (1986); S. Das Sarma, Phys. Rev. B 33, 5401 (1986); S. Das Sarma and E. H. Hwang, Phys. Rev. Lett. 83, 164 (1999); G. Zala, B. N. Narozhny, and I. L. Aleiner, Phys. Rev. B 64, 214204 (2001). [21] R. N. McFarland, T. M. Kott, L. Sun, K. Eng, and B. E. Kane, Phys. Rev. B 80, 161310(R) (2009). [22] J. P. Kotthaus and R. Ranvaud, Phys. Rev. B 15, 5758 (1977). 

1705.09498

1

1705

Effect of one-dimensional superlattice potentials on the band gap of two-dimensional materials

Using the tight-binding approach, we analyze the effect of a one-dimensional superlattice (1DSL) potential on the electronic structure of black phosphorene and transition metal dichalcogenides. We observe that the 1DSL potential results in a decrease of the energy band gap of the two-dimensional (2D) materials. An analytical model is presented to relate the decrease in the direct-band gap to the different orbital characters between the valence band top and conduction band bottom of the 2D materials. The direct-to-indirect gap transition, which occurs under a 1DSL potential with an unequal barrier width, is also discussed.

cond-mat.mes-hall

cond-mat

a Effect of one-dimensional superlattice potentials on the band gap of two-dimensional materials Shota Ono1, a) Department of Electrical, Electronic and Computer Engineering, Gifu University, Gifu 501-1193, Japan Using the tight-binding approach, we analyze the effect of a one-dimensional superlattice (1DSL) potential on the electronic structure of black phosphorene and transition metal dichalcogenides. We observe that the 1DSL potential results in a decrease of the energy band gap of the two-dimensional (2D) materials. An analytical model is presented to relate the decrease in the direct-band gap to the different orbital characters between the valence band top and conduction band bottom of the 2D materials. The direct-to-indirect gap transition, which occurs under a 1DSL potential with an unequal barrier width, is also discussed. I. INTRODUCTION Recently, a wide variety of graphene sister materials, such as black phosphorene (BP)1 and monolayer transition metal dichalcogenides M X2 (M =Mo, W; X =S, Se, Te)2,3, have been synthesized. Furthermore, a recent search on the Materials Project database yielded more than 600 stable two-dimensional (2D) materials that can be synthesized by exfoliation.4 Since some of these 2D materials possess a finite band gap, they are expected to be used in a host of next-generation electronic and optoelectronic devices. Strain engineering has attracted a lot of attention in the context of providing a tunable band gap of 2D materials.5 -- 9 For example, the early studies, based on the density-functional theory, have shown that the strain can yield a semiconductor-metal transition in BP.5,6 An analytical study has shown that the band gap increases (decreases) by a few hundred meV when the tensile (compressive) strain lies in the 2D plane.7 In view of the tight-binding (TB) description, the strain application can be interpreted as the hopping parameter modification. How the locally modified on-site potential param- eter influences the magnitude of the band gap in 2D materials should then be determined. Thus far, monolayer graphene in a one-dimensional superlattice (1DSL) potential has been extensively investigated based on the Dirac-type10 -- 22 and TB Hamiltonian.23,24 The pres- ence of a 1DSL potential has been shown to be able to drastically change the energy band structure around the Dirac cone, yielding electron supercollimation, extra Dirac cones, and an opening of the band gap. As the graphene superlattices have been fabricated by several experiments, where the period of the superlattice structure is from a few nm up to several hundreds of nm,25 -- 28 a theoretical study for the 1DSL potential effect on the band gap of 2D materials would be important for future applications. Herein, we study the electronic structure of the BP and M X2 in 1DSL potentials. By using the TB models developed by Rudenko and Katsnelson29 and Liu et al,30 we show that the band gap of the 2D materials is lowered by the 1DSL potential. Through a simple model analysis, the decrease in the direct band gap is shown to be attributed to the different orbital characters between the valence band (VB) top and conduction band (CB) bottom. We also show the direct-to-indirect gap transition that occurs under a 1DSL potential with an unequal barrier width. The models examined below can be applied to situations in which the 1DSL potential is controlled by two (top and back) gates,25,28 or in which the 2D layer is placed on a substrate with an appropriate lattice mismatch.26,27 a)Electronic mail: shota [email protected] 2 FIG. 1. Schematic illustration of (a) BP and (b) M X2. The conventional unit cells are depicted by the shaded rectangle with the size of ax × ay. (c) Side view on the BP structure. The atoms numbered 1, 4, 5, and 8 deviate from the plane at which the atoms numbered 2, 3, 6, and 7 are located. (d) The translation vectors Ri with i = 1, 2, 3, 4, 5, and 6 for M X2. (e) Schematic illustration of the 2D square net model. (f) The potential profile along the y-direction in the 2D material superlattice. (g) The first Brillouin zone of the 2D material superlattice with a size of SQ (see text for the definitions). II. 2D MATERIAL SUPERLATTICE To study the 1DSL potential effect on the electronic band structure of BP and M X2, we first consider a conventional unit cell that consists of N atoms. The area of the unit cell is defined as S1 ∈ [0, ax)⊗ [0, ay), as shown in Fig. 1(a) and (b) for BP and M X2, respectively. For example, ax = 4.37 and ay = 3.32 for BP and ax =3.19 and ay = √3ax for MoS2 in units of A. Next, we create a 1×Q supercell, that is, SQ ∈ [0, ax) ⊗ [0, Qay) with a positive integer Q. The position of each atom can be defined as Rs = (Rs y) with s = 1, · · · , N Q. The 1DSL potential that changes periodically along the y-direction is applied to the 2D materials as x, Rs vs =   V0 −V0 for 0 ≤ Rs for y < Λy 2 ≤ Rs Λy 2 (1) y < Λy, where Λy = Qay [see Fig. 1(f)]. In the following, we set Q = 20, unless noted otherwise. The use of a larger Q (and Λy) does not change the V0-dependence of the band gap size, as demonstrated below. The first Brillouin zone (BZ) has a rectangular shape surrounded by four lines kx = ±π/ax and ky = ±π/(Qay), as shown in Fig. 1(g). For the pristine BP and M X2 (i.e., the case of V0 = 0 and Q = 1), the minima of the band gaps (i.e., both the CB bottom and VB top) are located at the Γ point and (2π/(3ax), 0) in the first BZ, respectively. Since these are located along the ky = 0 line, the application of the 1DSL potential, given by Eq. (1) with Q 6= 1, does not change the location of the band gap minimum even if zone foldings occur. Below, we will consider only the electronic band structure along the kx-direction. For the case of the 1DSL potential that changes periodically along the x-direction, the location of the band gap minimum is also determined with consideration of the zone-folding concept. Similar to the case above, the band gap minimum is still located at the Γ point for the BP superlattices. In contrast, the location of the band gap minimum moves along the kx-direction with Q for the M X2 superlattices. By using the zone-folding concept, 3 (a) (b) (c) V0=0 eV V0=0.15 eV V0=0.3 eV 3 2 1 0 -1 -2 ] V e [ y g r e n E -3 0 kxax π 0 π 0 kxax kxax π FIG. 2. The electron band structure of the BP superlattice with Q = 20 for (a) V0 = 0, (b) 0.15, and (c) 0.3 eV. 16 dispersion curves around the band edges are shown. (a) (b) (c) V0=0 [eV] V0=0.25 [eV] V0=0.5 [eV] 2.5 2.0 1.5 1.0 0.5 0.0 ] V e [ y g r e n E -0.5 0 kxax π 0 kxax π 0 π kxax FIG. 3. The electron band structure of the MoS2 superlattice with Q = 20 for (a) V0 = 0, (b) 0.25, and (c) 0.5 eV. 16 dispersion curves around the band edges are shown. The VB top and CB bottom are located at kxax = 2π/3 (dashed line). one can observe that the band gap minimum is located at the Γ point when Q = 3i and (±2π/(3Qax), 0) when Q 6= 3i with a positive integer i. The most important observation is that the direction of the 1DSL potential does not alter the main result in this work, that is, the band gap reduction arising from the 1DSL potential. It should be noted that when the period of the 1DSL potential, given by Eq. (1), is large (Q ≫ 1), the group velocity along the ky-direction is negligibly small compared with that along the kx-direction. This is true when we consider the 1DSL potential that changes periodically along the x-direction; given a large period along the x-direction, the group velocity along the kx-direction is, in turn, negligibly small compared with that along the ky-direction. For the study of the M X2 superlattices below, we assume that the 1DSL potential is independent of the orbitals dz2 , dxy, and dx2−y2, for simplicity. The application of the orbital-dependent 1DSL potential does not change the main result of this work. 4 BP MoS2 Q=2 Q=4 Q=6 Q=8 Q=10 Q=20 Q=40 Q=2 Q=4 Q=6 Q=8 Q=10 Q=20 Q=40 2.4 2.2 2.0 1.8 1.6 1.4 1.2 1.0 ] V e [ p a g y g r e n E 0.8 0.0 0.2 0.4 0.6 2V0 [eV] 0.8 1.0 FIG. 4. The 2V0-dependence of the energy gap of MoS2 (solid) and BP (dashed) superlattices for several Qs. A. Black Phosphorene First, we study the electronic structure of the BP superlattice. We use the TB model developed by Rudenko and Katsnelson.29 They showed that the BP has a direct band gap at the Γ point, where the VB top and CB bottom consist of a mixture of the s, px, and pz orbitals and have different orbital characters. The TB Hamiltonian of the BP superlattice is given by HBP = X s (ǫs + vs) c† scs + X s6=s′ tss′ c† s′ cs, (2) where the first and second terms in the right hand side are the on-site potential and kinetic energies. cs and c† s are the electron destruction and creation operators at the sth atom site, respectively. tss′ is the hopping integral between the sites s and s′. These are given by t14 = −1.220, t12 = 3.665, t18 = −0.205, t13 = −0.105, and t25 = −0.055 in units of eV [see also Fig. 1(a) and (c) for the atom positions].29 The 1DSL potential, vs defined by Eq. (1), is added to the TB Hamiltonian. Figures 2(a)-(c) show 16 dispersion curves around the band edge for V0 =0, 0.15, and 0.3 eV, respectively. The zero energy is located at the middle of the CB bottom and VB top at the Γ point by adding the on-site potential of ǫs = 0.42 eV. When V0 is increased from 0 to 0.3 eV, the band gap decreases from 1.52 to 0.97 eV. This arises from the charge redistribution within the 1DSL potential period, which will be explained in Secs. II C and III A. B. Transition metal dichalcogenides Next, we compute the band gap of M X2 superlattice. We use the three-band TB model developed by Liu et al.30 This model is constructed by considering the d-d hoppings between M -dz2 , dxy, and dx2−y2 orbitals, where dz2 (dxy, dx2−y2) is the basis of the irreducible representation of A′ 1 (E′) for the point group D3h. The pristine M X2 exhibits a direct band gap at K and K′ points. The VB top mainly consists of the dxy and dx2−y2 orbitals, while the CB bottom consists of the dz2 orbital. By using γ and γ′ to denote the three atomic orbitals, the Hamiltonian of the M X2 superlattice is given by 5 HM X2 = X + X s s6=s′ (ǫγ + vs) c† s,γcs,γ X γ X γ,γ ′ Tγ,γ ′(Rs′ − Rs)c† s′,γ ′cs,γ, (3) where cs,γ and c† s,γ are the electron destruction and creation operators at the sth atom site with the energy ǫγ, respectively. Tγ,γ ′(R) is the hopping integral between the orbital γ at the site 0 and the orbital γ′ at the site R. The summation of s and s′ in Eq. (3) is taken up to the third nearest-neighbor (NN) sites; R = Ri for the first NN sites, R = R′ i = Ri +Ri+1 for the second NN sites, and R = R′′ i = 2Ri for the third NN sites, where i = 1, 2, 3, 4, 5, and 6, with R7 = R1 [see also Fig. 1(d) for the translation vectors Ri]. The expressions of 3 × 3 matrix Tγ,γ ′(R) for 18 NN sites are provided in the Appendix A. Among M X2, we consider, as an example, the monolayer MoS2 superlattice. We use the TB parameters of MoS2 obtained by the density-functional theory calculations within the local-density approximation.30 Then, the values of ǫγ with γ = dz2 and γ = dxy, dx2−y2 are set to 0.820 eV and 1.931 eV, respectively. Figures 3(a)-(c) show the energy dispersion curves along the kx-direction of the MoS2 superlattice for various V0s. 16 dispersion curves around the band edge are shown. Similar to the case of the BP superlattice, the band gap at kxax = 2π/3 (vertical dotted line) drastically decreases with increasing V0. C. Gap variation Figure 4 shows the energy gap of the MoS2 superlattice as a function of the barrier height 2V0 for various Qs. For comparison, the 2V0-dependence of the band gap of the BP superlattice is also shown. The energy gap Eg(V0) decreases monotonically with increasing 2V0, while the decrease in Eg(V0) is moderate for small Q. For larger Q, the band gap difference is approximately expressed by the linear relation Eg(V0) − Eg(0) ≃ −2V0. (4) This reflects the fact that the charges are strongly localized to the region with vs = V0 and −V0 at the VB top and CB bottom, respectively. Figures 5(a) and 5(b) show the Ry-dependence of the charge density at the VB top and CB bottom, respectively. The contributions from dz2, dxy, and dx2−y2 orbitals are shown when Q = 20 and V0 = 0.5 eV (solid). For comparison, the case of V0 = 0 eV is also shown (dashed). The charge density at the VB top consists of dxy and dx2−y2 and is distributed around Ry/Λy ≤ 1/2, that is, the region with vs = V0 [Fig. 5(a)], while that at the CB bottom consists of dz2 dominantly and is distributed around Ry/Λy > 1/2, that is, the region with vs = −V0 [Fig. 5(b)]. In such a charge distribution, the eigenenergy would be linearly proportional to the on-site potential energy, resulting in the relation of Eq. (4). Figures 5(c) and 5(d) show the Ry-dependence of the charge density of the three orbitals at the VB top and CB bottom, respectively, for Q = 4. For such a small Q, the charge density is finite even around Ry/Λy > 1/2 (Ry/Λy ≤ 1/2) at the VB top (the CB bottom). In this case, the magnitude of the upward and downward shifts of the bands is small for the VB and CB, respectively, yielding the slight decrease in the band gap. It should be noted that the semiconductor-metal transition would be observed when Eg(0) = 2V0 is satisfied in Eq. (4) for large Qs. For example, V0 ≃ 0.9 and 0.75 eV is needed to observe the transition in MoS2 and BP, respectively. (a) VB, Q=20 0.10 0.08 0.06 0.04 0.02 y t i s n e d e g r a h C V0 = 0.5 [eV] dz2 dxy dx2 -y2 6 V0 = 0 [eV] dz2 dxy dx2 -y2 (b) CB, Q=20 0.00 0.0 0.3 0.2 0.1 y t i s n e d e g r a h C 0.0 0.0 0.5 Ry/Λy 1.0 0.0 0.5 Ry/Λy 1.0 (c) VB, Q=4 (d) CB, Q=4 0.5 Ry/Λy 1.0 0.0 0.5 Ry/Λy 1.0 FIG. 5. The Ry-dependence of the charge density in the MoS2 superlattice (a) at the VB top for Q = 20, (b) at the CB bottom for Q = 20, (c) at the VB top for Q = 4, and (d) at the CB bottom for Q = 4, with V0 = 0.5 eV (solid) and 0 eV (dashed) given by Eq. (1). The contributions from −y2 (green) orbitals are shown. The charge distributions of dxy and dz2 (red), dxy (blue), and dx2 dx2 −y2 completely overlap when V0 = 0 eV. III. DISCUSSION A. Origin of the band gap decrease In Sec. II, it has been shown that the 1DSL potential reduces the magnitude of the direct band gap. We attribute such a reduction to the different orbital characters between the VB top and CB bottom. As mentioned in Secs. II A and II B, this condition is satisfied in BP and M X2. To show how the decrease in the direct band gap is explained in terms of the orbital characters at the band edges, we consider, as a simple example, a 2D square net that consists of the atoms shown in Fig. 1(e). There is an atom in the unit cell whose size is ax × ay. Each atom has an atomic energy of ǫγ. We assume that there are two energy levels γ = a and b satisfying the relation ǫa < ǫb. The sth atom position is denoted by Rs = (Rs y) = (nax, may) with integers n and m. The different potentials are added to study the superlattice potential effect: vs = V0 (> 0) and −V0 for m = 2l and m = 2l + 1 with an integer l, respectively. The real-space TB Hamiltonian for the 2D square net is given by x, Rs HSQ = X s b X γ=a (ǫγ + vs) c† s,γcs,γ + X s6=s′ b X γ=a tγc† s′,γcs,γ, (5) where tγ is the electron hopping integrals between the energy levels ǫγ at the nearest- neighbor sites. The electron hopping between the energy levels γ = a and γ = b is assumed to be negligible. By imposing the periodic boundary condition to form the energy bands, the TB Hamiltonian in a reciprocal-space becomes a 4 × 4 matrix: 0 0 2ta cos β 2ta cos β h+ a HSQ(k) =   0 0 h− a 0 0 0 0 h+ b 2tb cos β 2tb cos β h− b where, for γ = a and b, h+ γ = ǫγ + V0 + 2tγ cos α, h− γ = ǫγ − V0 + 2tγ cos α,   , 7 (6) (7) with α = kxax and β = kyay being the wavevector k = (kx, ky). The energy eigenvalues are given by γ (k) = ǫγ + 2tγ cos α ± qV 2 E± 0 + 4t2 γ cos2 β. (8) Below, we use the indexes (γ, ±) to denote the energy band. We assume ta > 0 and tb < 0, since we study the band structure with semiconducting properties. Then, both the energy maximum of the band (a, +) and the energy minimum of the band (b, −) are located at the Γ point. The energy difference ∆E is explicitly given by ∆E = E− b (k = 0) − E+ a (k = 0) = ǫb − ǫa + 2(tb − ta) − qV 2 0 + 4t2 b − qV 2 0 + 4t2 a. (9) When ∆E > 0, the bands (a, +) and (b, −) serve as the VB and CB, respectively. From Eq. (9), it is clear that the energy gap ∆E decreased with increasing V0. This is because the charges are redistributed to obtain the energy gain. The eigenvectors of the VB top [i.e., at the Γ point in the (a, +) band] and the CB bottom [i.e., at the Γ point in the (b, −) band] are respectively given by   C+ a C − a C+ b C − b   = 2ta qw2 a,− + 4t2 a   and 1  wa,−/(2ta)  0 0 (10) (11) =   2tb b,+ + 4t2 C+  a C −  a C+ b C − b γ + pV0 with p = ± and γ = a, b. C+     −wb,+/(2tb) qw2 0 0 1 , b 0 + 4t2 where wγ,p = qV 2 γ are the proba- bility amplitude of the orbital γ at the sites m = 2l and m = 2l + 1, respectively. Since wa,−/(2ta) < 1 and −wb,+/(2tb) > 1 for finite V0, the charges at the VB top and CB bottom are mainly distributed at the sites m = 2l (vs = V0) and m = 2l + 1 (vs = −V0), respectively. As a result, the (a, +) and (b, −) bands shift to higher and lower energies, respectively, which leads to the decrease in the band gap. Thus, we determined the rela- tionship between the band gap and the orbital difference at the band edge. Although the 2D square net model above is quite simple, it captures the main physics behind the band gap reduction observed in Figs. 2 and 3. γ and C − It is possible to construct a TB model that includes more than two atoms in a unit cell, for example, vs = V0 for m = 4l and 4l + 1, and vs = −V0 for m = 4l + 2 and 4l + 3 with an integer l. In such a case, the magnitude of the band gap decreases more significantly because more charges are redistributed within the unit cell. This is also consistent with the observation in Fig. 4. 8 (b) V0=0.5 eV (c) V0=0.75 eV 2.0 1.5 (a) 1.0 V0=0.25 eV 0.5 0.0 ] V e [ y g r e n E 0 kxax π 0 kxax π 0 π kxax FIG. 6. The electron band structure of the MoS2 under a 1DSL potential given by Eq. (12) with Q = 20, calculated with (a) V0 = 0.25, (b) 0.5, and (c) 0.75 eV. B. Direct-to-indirect gap transition While we have also studied other types of 1DSL potentials, such as a cosine-type potential, where the net total potential per unit cell vanishes, the results in this work are qualitatively the same. When the net total potential per unit cell is finite, the system exhibits an indirect band gap. To show the latter, we study the MoS2 under a 1DSL potential with unequal barrier width. The 1DSL potential used is given by vs =   V0 −V0 for 0 ≤ Rs for y < Λy 10 ≤ Rs Λy 10 (12) y < Λy. Figure 6 shows the band structure of the MoS2 under a 1DSL potential given by Eq. (12) with Λy = 20ay for various V0s. As V0 increases, the band gap decreases, similar to the cases of the 1DSL potential with an equal barrier width shown in Fig. 3. Interestingly, when V0 ≥ 0.5 eV, the energy of the VB top at kxax = 0 overcomes that at kxax = 2π/3, yielding the indirect band gap. This originates from the different orbital characters of the VB top between kxax = 0 and kxax = 2π/3 in the pristine MoS2, where the former mainly consists of a Mo dz2 orbital, while the latter consists of Mo dxy and dx2−y2 orbitals.30 In addition, since the charge of the CB bottom at kxax = 2π/3 is distributed around the region with vs = −V0, as shown in Fig. 5(b), the charge of the VB top at kxax = 0 tends to be, in turn, distributed around the region with vs = +V0 (i.e., Rs y < Λy/10). This leads to an increase in the energy of the VB top at kxax = 0, compared with that at kxax = 2π/3. IV. CONCLUSIONS Through the TB calculations for the BP and M X2 superlattices, we have demonstrated that the presence of a 1DSL potential yields a decrease in the band gap of 2D materials. An analytical investigation shows that this also holds if a pristine 2D material has (i) a direct- band gap and (ii) different orbital characters between the VB top and the CB bottom. It has also been found that the band gap experiences a direct-to-indirect gap transition when a 1DSL potential with an unequal barrier width is applied. We expect that various 2D material superlattices will be created in future experiments, as the graphene superlattices are fabricated by several experiments.25 -- 28 9 ACKNOWLEDGMENTS SO would like to thank M. Aoki for fruitful discussions and G.-B. Liu and D. Xiao for providing useful information about the TB model of M X2. This study was supported by a Grant-in-Aid for Young Scientists B (No. 15K17435) from JSPS. Appendix A: Hopping integrals The matrix elements Tγ,γ ′(R) in Eq. (3) can be expressed by 17 parameters.30 By using the notation (dz2 , dxy, dx2−y2) = (1, 2, 3), the 17 hopping parameters for the first NN sites are explicitly given as: t0 = T1,1(R1), t1 = T1,2(R1), t2 = T1,3(R1), t11 = T2,2(R1), t12 = T2,3(R1), t22 = T3,3(R1), for the second NN sites: r0 = T1,1(R′ 1), r1 = T1,2(R′ 1), r2 = T1,2(R′ 4), r11 = T2,2(R′ 1), r12 = T2,3(R′ 1), and for the third NN sites: u0 = T1,1(R′′ The other matrix elements are obtained with the aid of the symmetry property. Those are explicitly written as, for the first nearest-neighbor (NN) sites Ri, 1 ), u11 = T2,2(R′′ 1 ), u12 = T2,3(R′′ 1 ), u1 = T1,2(R′′ 1 ), u2 = T1,3(R′′ 1 ), u22 = T3,3(R′′ 1 ). T (R1) =   t1 t11 t0 t2 −t1 t12 t2 −t12 t22   , T (R2) =   t0 T (R3) =  −2ct1 + 2dt2 ct11 + 3ct22 2ct1 + 2dt2  2dt1 − 2ct2 dt11 − t12 − dt22 t0 2ct1 − 2dt2 ct11 + 3ct22 −2ct1 − 2dt2 2dt1 − 2ct2 −dt11 + t12 + dt22 −2dt1 − 2ct2 3ct11 + ct22 −dt11 − t12 + dt22   , (A1) −2dt1 − 2ct2 dt11 + t12 − dt22 3ct11 + ct22   , T (R4) =   t0 −t1 t2 t1 t11 −t12 t22 t12 t2   , t0 T (R5) =  2ct1 − 2dt2  −2dt1 − 2ct2 −dt11 − t12 + dt22 −2ct1 − 2dt2 ct11 + 3ct22 T (R6) =   t0 2ct1 + 2dt2 ct11 + 3ct22 −2ct1 + 2dt2 −2dt1 − 2ct2 dt11 + t12 − dt22 2dt1 − 2ct2 3ct11 + ct22 −dt11 + t12 + dt22   , 2dt1 − 2ct2 3ct11 + ct22 dt11 − t12 − dt22   , and, for the next NN site R′ i = Ri + Ri+1 with R7 = R1: T (R′ 1) =   r1 r11 r0 r2 −er2 r12 r11 + 2er12 −er1 r12   , T (R′ 2) =   r0 0 2er1 0 r11 + r12/e 0 2er1 0 r11 − er12   , T (R′ 3) =   r0 −r1 r11 −r2 −er2 −r12 r11 + 2er12 −er1 −r12   , T (R′ 4) =   r2 r11 r0 r1 −er1 r12 r11 + 2er12 −er2 r12   , (A2) (A3) (A4) (A5) (A6) T (R′ 5) =   r0 0 2er2 0 r11 + r12/e 0 2er1 0 r11 − er12   , T (R′ 6) =  r0 −r2 r11 −r1  −er1 −r12 r11 + 2er12 −er2 −r12   , (A7) 10 with c = 1/4, d = √3/4 and e = 1/√3. The expressions for the third NN sites R′′ i = 2Ri can be obtained by replacing (t0, t1, t2, t11, t12, t22) in the expressions for the first NN sites Ri with (u0, u1, u2, u11, u12, u22), respectively. 1H. Liu, A. T. Neal, Z. Zhu, Z. Luo, X. Xu, D. Tom´anek, and Peide D. Ye, Phosphorene: An Unexplored 2D Semiconductor with a High Hole Mobility, ACS Nano, 8, 4033 (2014). 2B. Radisavljevic, A. Radenovic, J. Brivio, V. Giacometti, A. Kis, Single-layer MoS2 transistors, Nat. Nanotechnol. 6, 147 (2011). 3Q. H. Wang, K. K. Zadeh, A. Kis, J. N. Coleman, M. S. Strano, Electronics and optoelectronics of two-dimensional transition metal dichalcogenides, Nat. Nanotechnol. 7, 699 (2012). 4M. Ashton, J. Paul, S. B. Sinnott, and R. G. Hennig, Topology-Scaling Identification of Layered Solids and Stable Exfoliated 2D Materials, Phys. Rev. Lett. 118, 106101 (2017). 5X. Peng, Q. Wei, and A. Copple, Strain-engineered direct-indirect band gap transition and its mechanism in two-dimensional phosphorene, Phys. Rev. B 90, 085402 (2014). 6A. S. Rodin, A. Carvalho, and A. H. CastroNeto, Strain-Induced Gap Modification in Black Phosphorus, Phys. Rev. Lett. 112, 176801 (2014). 7J.-W. Jiang and H. Park, Analytic study of strain engineering of the electronic bandgap in single-layer black phosphorus, Phys. Rev. B 91, 235118 (2015). 8H. Rostami, R. Rold´an, E. Cappelluti, R. Asgari, and F. Guinea, Theory of strain in single-layer transition metal dichalcogenides, Phys. Rev. B 92, 195402 (2015). 9A. J. Pearce, E. Mariani, and G. Burkard, Tight-binding approach to strain and curvature in monolayer transition-metal dichalcogenides, Phys. Rev. B 94, 155416 (2016). 10C. Bai and X. Zhang, Klein paradox and resonant tunneling in a graphene superlattice, Phys. Rev. B 76, 075430 (2007). 11C. H. Park, Y. W. Son, L. Yang, M. L. Cohen, S. G. Louie, Electron beam supercollimation in graphene superlattices, Nano Lett. 8, 2920 (2008). 12C. H. Park, L. Yang, Y. W. Son, M. L. Cohen, S. G. Louie, New Generation of Massless Dirac Fermions in Graphene under External Periodic Potentials, Phys. Rev. Lett. 101, 126804 (2008). 13L. Brey and H. A. Fertig, Emerging Zero Modes for Graphene in a Periodic Potential, Phys. Rev. Lett. 103, 046809 (2009). 14M. Barbier, P. Vasilopoulos, and F. M. Peeters, Extra Dirac points in the energy spectrum for superlattices on single-layer graphene, Phys. Rev. B 81, 075438 (2010). 15L.-G. Wang and S.-Y. Zhu, Electronic band gaps and transport properties in graphene superlattices with one-dimensional periodic potentials of square barriers, Phys. Rev. B 81, 205444 (2010). 16P. Burset, A. L. Yeyati, L. Brey, and H. A. Fertig, Transport in superlattices on single-layer graphene, Phys. Rev. B 83, 195434 (2011). 17G. M. Maksimova, E. S. Azarova, A. V. Telezhnikov, and V. A. Burdov, Graphene superlattice with periodically modulated Dirac gap, Phys. Rev. B 86, 205422 (2012). 18Kh. Shakouri, M. Ramezani Masir, A. Jellal, E. B. Choubabi, and F. M. Peeters, Effect of spin-orbit couplings in graphene with and without potential modulation, Phys. Rev. B 88, 115408 (2013). 19S. Ono, M. Zhang, Y. Noda, and K. Ohno, Tunable Seebeck Coefficient in Monolayer Graphene Under Periodic Potentials, Journal of Elec. Materi. 43 1505 (2014). 20S. K. Choi, C. H. Park, and S. G. Louie, Electron Supercollimation in Graphene and Dirac Fermion Materials Using One-Dimensional Disorder Potentials, Phys. Rev. Lett. 113, 026802 (2014). 21C. H. Chen, P. Tseng, and W. J. Hsueh, Quasi-Dirac points in one-dimensional graphene superlattices, Phys. Lett. A 380, 2957 (2016). 22P. Kim and C. H. Chen, The electronic structure and intervalley coupling of artificial and genuine graphene superlattices, Nano Res. 9, 1101 (2016). 23G. Pal, W. Apel, and L. Schweitzer, Landau level splitting due to graphene superlattices, Phys. Rev. B 85, 235457 (2012). 24A. de Jamblinne de Meux, N. Leconte, J.-C. Charlier, and A. Lherbier, Velocity renormalization and Dirac cone multiplication in graphene superlattices with various barrier-edge geometries, Phys. Rev. B 91, 235139 (2015). 25S. Dubey, V. Singh, A. K. Bhat, P. Parikh, S. Grover, R. Sensarma, V. Tripathi, K. Sengupta, and M. M. Deshmukh, Tunable Superlattice in Graphene To Control the Number of Dirac Points, Nano Lett. 13, 3990 (2013). 26K.-K. Bai, Y. Zhou, H. Zheng, L. Meng, H. Peng, Z. Liu, J.-C. Nie, and L. He, Creating One-Dimensional Nanoscale Periodic Ripples in a Continuous Mosaic Graphene Monolayer, Phys. Rev. Lett. 113, 086102 (2014). 27C. Lin, X. Huang, F. Ke, C. Jin, N. Tong, X. Yin, L. Gan, X. Guo, R. Zhao, W. Yang, E. Wang, and Z. Hu, Quasi-one-dimensional graphene superlattices formed on high-index surfaces, Phys. Rev. B 89, 085416 (2014). 28M. Drienovsky, F.-X. Schrettenbrunner, A. Sandner, D. Weiss, J. Eroms, M.-H. Liu, F. Tkatschenko, and K. Richter, Towards superlattices: Lateral bipolar multibarriers in graphene, Phys. Rev. B 89, 115421 (2014). 29A. N. Rudenko and M. I. Katsnelson, Quasiparticle band structure and tight-binding model for single- and bilayer black phosphorus, Phys. Rev. B 89, 201408(R) (2014). 30G.-B. Liu, W.-Y. Shan, Y. Yao, W. Yao, and D. Xiao, Three-band tight-binding model for monolayers of group-VIB transition metal dichalcogenides, Phys. Rev. B 88, 085433 (2013). 11 

1104.4903

3

1104

Charge transfer through single molecule contacts: How reliable are rate descriptions?

The trend to fabricate electrical circuits on nanoscale dimensions has led to impressive progress in the field of molecular electronics in the last decade. A theoretical description of molecular contacts as the building blocks of future devices is challenging though as it has to combine properties of Fermi liquids in the leads with charge and phonon degrees of freedom on the molecule. Apart from ab initio schemes for specific set-ups, generic models reveal characteristics of transport processes. Particularly appealing are descriptions based on transfer rates successfully used in other contexts such as mesoscopic physics and intramolecular electron transfer. However, a detailed analysis of this scheme in comparison with numerically exact data is elusive yet. It turns out that a formulation in terms of transfer rates provides a quantitatively accurate description even in domains of parameter space where in a strict sense it is expected to fail, e.g. for lower temperatures. Typically, intramolecular phonons are distributed according to a voltage driven steady state that can only roughly be captured by a thermal distribution with an effective elevated temperature (heating). An extension of a master equation for the charge-phonon complex to include effectively the impact of off-diagonal elements of the reduced density matrix provides very accurate data even for stronger electron-phonon coupling.

cond-mat.mes-hall

cond-mat

Charge transfer through single molecule contacts: How reliable are rate descriptions? Universitat Ulm, Institut fur Theoretische Physik, Albert-Einstein-Allee 11, 89069 Ulm, Germany D. Kast,∗ L. Kecke, and J. Ankerhold (Dated: December 4, 2018) Background: The trend to fabricate electrical circuits on nanoscale dimensions has led to impressive progress in the field of molecular electronics in the last decade. A theoretical description of molecular contacts as the building blocks of future devices is challenging though as it has to combine properties of Fermi liquids in the leads with charge and phonon degrees of freedom on the molecule. Apart from ab initio schemes for specific set-ups, generic models reveal characteristics of transport processes. Particularly appealing are descriptions based on transfer rates successfully used in other contexts such as mesoscopic physics and intramolecular electron transfer. However, a detailed analysis of this scheme in comparison with numerically exact data is elusive yet. Results: It turns out that a formulation in terms of transfer rates provides a quantitatively accurate description even in domains of parameter space where in a strict sense it is expected to fail, e.g. for lower temperatures. Typically, intramolecular phonons are distributed according to a voltage driven steady state that can only roughly be captured by a thermal distribution with an effective elevated temperature (heating). An extension of a master equation for the charge-phonon complex to include effectively the impact of off-diagonal elements of the reduced density matrix provides very accurate data even for stronger electron-phonon coupling. Conclusion: Rate descriptions and master equations offer a versatile instrument to describe and understand charge transfer processes through molecular junctions. They are computationally orders of magnitudes less expensive than elaborate numerical simulations that, however, provide exact data as benchmarks. Adjustable parameters obtained e.g. from ab initio calculations allow for the treatment of various realizations. Even though not as rigorously formulated as e.g. nonequilibrium Greens function methods, they are conceptually simpler, more flexible for extensions, and from a practical point of view provide accurate results as long as strong quantum correlations do not modify properties of relevant sub-units substantially. I. INTRODUCTION interest in the last decade [1]. Electrical devices on the nanoscale have received sub- stantial Impressive progress has been achieved in contacting single molecules or molecular aggregates with normal-conducting or even superconducting metallic leads [2, 3]. The objective is to exploit nonlinear transport properties of molecular junc- tions as the elementary units for a future molecular elec- tronics. While the first experimental set-ups have been operated at room temperature, meanwhile low tempera- tures down to the millikelvin range, the typical regime for devices in mesoscopic solid state physics, are accessible (see e.g. [4 -- 6]). This allows for detailed studies of phe- nomena such as inelastic charge transfer due to molecular vibrations [7 -- 9], voltage driven conformational changes of the molecular backbone [10], Kondo physics [11], and Andreev reflections [6] to name but a few. These developments have been accompanied by efforts to advance theoretical approaches in order to obtain an understanding of generic physical processes on the one hand and to arrive at a tool to quantitatively describe and predict experimental data. For this purpose, basi- cally two strategies have been followed. One is based on ab initio schemes that have been successfully employed ∗Electronic address: [email protected] for isolated molecular structures as e.g. density func- tional theory (DFT). Combining DFT with nonequilib- rium Greens functions (NEGF) allows to capture essen- tial properties for junctions with specific molecular struc- tures and geometries [2, 3, 12, 13]. This provides insight in electronic formations along contacted molecules and gives at least qualitatively correct results for currents and differential conductances. However, a quantitative description on the level of accuracy known from conven- tional mesoscopic devices seems out of reach yet. Further, these methods are not able to capture phenomena due to strong correlations such as e.g. Kondo resonances. Thus, an alternative route, mainly inspired by solid state methodologies, starts with simplified models that are assumed to cover relevant physical features. The intention then is to reveal fundamental processes char- acteristic for molecular electronics that give a qualita- tive description of observations from realistic samples, but provide also the basis for a proper design of molec- ular junctions to exploit these processes. Information about specific molecular set-ups appears merely in form of parameters which offers a large amount of flexibility. In general, to attack the respective many body prob- lems, perturbative schemes have been applied, the most powerful of which are nonequilibrium Greens functions [14, 15]. Conceptually simpler, easier to implement, and often better revealing the physics are treatments in terms of master or rate equations. Being approximations to the NEGF frame in certain ranges of parameters space, they sometimes lack the strictness of perturbation series, but have been extensively employed for mesoscopic devices [16] and quantitatively often provide data of at least sim- ilar accuracy. Roughly speaking, these schemes apply as long as quantum correlations between relevant sub-units of the full compound are sufficiently weak [15]. Phys- ically, it places charge transfer through molecular con- tacts in the context of inelastic charge transfer through ultra-small metallic contacts (dynamical Coulomb block- ade [17]) and in the context of solvent or vibronic medi- ated intramolecular charge transfer (Marcus theory) [18 -- 20]. While rate descriptions have been developed in a va- riety of formulations before [21 -- 28], the performance of such a framework in comparison with numerically exact data has not been addressed yet. The reason for that is simple: a numerical method which provides numerically exact data in most ranges of parameters space (temper- ature, coupling strength, etc.) has been successfully im- plemented only very recently in form of a diagrammatic Monte Carlo approach [29]. Path integral Monte Carlo methods have been used previously for intramolecular charge transfer in complex aggregates [18, 19] in a vari- ety of situations including correlated [30] and externally driven transfer [31] and, of particular relevance for the present work, transfer in presence of prominent phonon modes [32]. The goal of the present work is to study a simple yet highly non-trivial set-up, namely, a molecular contact with a single molecular level coupled to a prominent vi- bronic mode (phonon) which itself may or may not be embedded in a bosonic heat bath. We develop rate de- scriptions of various complexity, place them into the con- text of NEGF, and compare them with exact data. The essence of this study is, astonishingly enough, that rate theory provides quantitatively accurate results for mean currents over very broad ranges of parameter space, even in domains where they are not expected to be reliable. II. RESULTS AND DISCUSSION In Sec. II A we define the model and the basic ingredi- ents for a perturbative treatment. A formulation which closely follows the P (E)-theory for dynamical Coulomb blockade is discussed in Sec. II B. Nonequilibrium ef- fects in the stationary phonon distribution are analyzed in Sec. II C based on a dynamical formulation of charge and phonon degrees of freedom. The presence of a sec- ondary bath is incorporated in Sec. II D together with an improved treatment of the dot-lead coupling, which is exact for vanishing electron-phonon interaction. The comparison with numerically exact data and a detailed discussion is given in Sec. II E. 2 A. Model We start with the minimal model of a molecular con- tact consisting of a single electronic level coupled to fermionic reservoirs, where a prominent internal molec- ular phonon mode interacting with the excess charge is described by a harmonic degree of freedom (cf. Fig. 1) [15, 33, 34]. Neglecting spin degrees of freedom the total compound is thus described by H = HL/R + HT + HD + HD,P h + HP h ǫk,αc† k,αck,α + Xα=L,R; k (Tk,αc† k,αd + h.c.) = Xα=L,R; k +ǫDd†d + p2 0 2m + mω2 0 2 (cid:0)x0 + l0d†d(cid:1)2 (1) and reservoir α and l0 = M0p2/ω3 Here, the Tk,α denote tunnel couplings between dot level 0m contains the cou- pling M0 between excess charge and phonon mode. An external voltage V across the contact is applied symmet- rically around the Fermi level such that ǫk,α = ǫ0(k)+ µα with the bare electronic dispersion relation ǫ0(k) and chemical potentials µL = +eV /2, µR = −eV /2. Below, this model will be further extended to include the em- bedding of the prominent mode into a large reservoir of residual molecular and/or solvent degrees of freedom act- ing as a heat bath. Qualitatively, since the dot occupa- tion d†d can only take the values q = 0, 1, the sub-unit HD + HD,P h + HP h describes a two state system coupled to a harmonic mode (spin-boson model [20]). Depending on the charge state of the dot the phonon mode is sub- ject to potentials Vq(x0) = (mω2 0/2)(x0 + l0q)2. Now, the presence of the leads acts (for finite voltages) as an ex- ternal driving force to alternately charging (q = 1) and discharging (q = 0) the dot, thus switching alternately between V0 and V1 for the phonon mode. The classical energy needed to reorganize the phonon is the so-called reorganization energy Λ = V1(0) − V0(0) = M 2 0 /ω0. Quantum mechanically, the phonon mode may also tun- nel through the energy barrier located around x0 = −l0/2 separating the minima of V0,1. It is convenient to work with dressed electronic states on the dot and thus to apply a polaron transformation generating the shift l0 in the oscillator coordinate asso- ciated with a charge transfer process, i.e., U = exp(cid:18)−i p0l0  d†d(cid:19) (2) with momentum operator p0 = ipmω0/2(b† 0 − b0) where b† 0, b0 are creation and annihilation operators of the phonon mode, respectively. We mention in pass- ing that complementary to the situation here, the the- ory of dynamical Coulomb blockade in ultra-small metal- lic contacts is based on a transformation which gener- ates a shift in momentum (charge) rather than posi- tion [17]. Now, the electron-phonon interaction is com- pletely absorbed in the tunnel part of the Hamiltonian, +eV
2 ý GL 7
2 Ñ Ω0 5
2 Ñ Ω0 3
2 Ñ Ω0 GL Εdot 1
2 Ñ Ω0 V=0 ý GR GR -eV
2 FIG. 1: Single charge transfer through a molecular contact consisting of a single electronic level coupled to a harmonic phonon mode and contacted to metallic leads. Forward (no prime) and backward (with prime) rates are the basic ingre- dients for the approximate treatment, see text for details. thus capturing the cooperative effect of charge tunneling onto the dot and photon excitation in the molecule, i.e. H = HL/R + HD + HP h + HT with HP h = ω0(cid:18)b† HT = Xα=L,R; k 0b0 + 1 2(cid:19) k,αd exp(cid:18) i Tk,αc†  p0l0(cid:19) + h.c. (3) Single charge tunneling through the device can be captured formally exactly under weak conditions (e.g. instantaneous equilibration in the leads during charge transfer) within the Meir-Wingreen formulation based on nonequilibrium Greens functions [14, 15]. For the current voltage characteristics one finds I(V ) = 4e  Z dǫ ΣLΣR ΣL + ΣR (cid:20)fβ (cid:18)ǫ − ×(cid:2)iG>(ǫ) − iG<(ǫ)(cid:3) eV 2 (cid:19) − fβ (cid:18)ǫ + eV 2 (cid:19)(cid:21) (4) with energy dependent lead self-energies Σα(ǫ) = 2πPk Tk,α2δ(ǫ − ǫk) and with the Fourier transforms of the time dependent Greens functions G<(t) = ihd†d(t)i and G>(t) = −ihd(t)d†i. Upon applying the polaron transformation (2), one has 3 the Greens functions factorize such as e.g. G<(t) → ihd†d(t)iD exp[J(t)] with the phonon correlation eJ(t) = De− i  p0l0 e i  p0(t)l0EP h (6) into expectation values with respect to the dot (D) and the phonon (Ph), respectively. Any finite tunnel coupling induces correlations that in analytical treatments can only be incorporated perturbatively. There, the proper approximative scheme depends on the range of param- eter space one considers. Generally speaking, there are four relevant energy scales ΣL/R, M0, kBT , and ω0 of the problem corresponding to three independent dimen- sionless parameters, e.g., m0 = M0 ω0 , θ = ω0β , σ = ΣL + ΣR ω0 . (7) In the sequel we are interested in the low temperature domain θ > 1 where thermal broadening of phonon lev- els is small so that discrete steps appear in the IV - characteristics. Qualitatively, seen from the dynamics of the phonon mode, two regimes can be distinguished according to the adiabaticity parameter Σ/ω0 = σ: For σ < 1 the phonon wave packet fulfills on a given surface V0 or V1 multiples of oscillations before a charge trans- fer process happens to occur. The electron carries excess energy due to a finite voltage which may be absorbed by the phonon to reorganize to the new conformation (in the classical case the reorganization energy Λ). In the language of intramolecular charge transfer this scenario corresponds to the diabatic regime with well-defined sur- faces Vq. In the opposite regime σ > 1 charge transfer is fast so that the phonon may obey multiples of switchings between the surfaces V0,1 . This is the adiabatic regime. In this latter range the impact of the adiabaticity on the diabatic ground state wave functions is weak for m0 < 1 when the distance of the diabatic surfaces is small com- pared to the widths of the ground states. For m0 > 1 in both regimes electron transfer is accompanied by phonon tunneling through energy barriers separating minima of adiabatic or diabatic surfaces. The dynamics of the total compound is then determined by voltage driven collec- tive tunneling processes. Master equation approaches to be investigated below, rely on the assumption that both sub-units, charge degree of freedom and phonon mode, basically preserve their bare physical properties even in case of finite coupling m0. Hence, since the model (1) can be solved exactly in the limits m0 = 0 and σ = 0 and following the above discussion, we expect them to cap- ture the essential physics quantitatively in the domain m0 < 1 and for all ratios σ. We note that recently the strong coupling limit including the current statistics has been addressed as well [35, 36]. G<(t) = iDd†e− i G>(t) = −iDe i  p0l0e i  p0(t)l0 d(t)E  p0(t)l0d(t)d†e− i  p0l0E , (5) B. Rate approach I where all expectation values are calculated with the full Hamiltonian (3). for Tk,α → 0, Of course, The simplest perturbative approach considers the co- operative effect of electron tunneling and phonon exci- tation in terms of Fermi's golden rule for the tunneling part HT . For this purpose one derives transition rates for sequential transfer according to Fig. 1. A straight- forward calculation for energy independent self-energies ΣL/R (wide-band limit) gives the forward rate onto the dot from the left lead ΓL(V, ǫD) = ΣL  Z dǫfβ (cid:18)ǫ − eV 2 (cid:19)P0(ǫ − ǫD) , (8) where fβ(ǫ) is the Fermi distribution. Inelastic tunneling associated with energy emission / absorption of phonons is captured by the Fourier transform of the phonon- phonon correlation exp[J(t)] leading to
Ñ
S e

V
I 0.2 0.1 0.0 -0.1 -0.2 4 m0=1
2 m0=3
2 -6 -4 -2 0 2 4 6 V
ÑΩ0
e
δ [ǫ − ω0(l − k)] (9) FIG. 2: IV -characteristics for symmetric coupling ΣL = ΣR and for varying electron-phonon coupling m0 at inverse tem- perature θ = 25 (solid) and θ = 10 (dashed). with the total rate Γtot,0 = ΓL + ΓR + Γ′ L + Γ′ R and the rate for transfer towards the dot Γd = ΓL + Γ′ R obtained according to (8). Note that for vanishing electron-phonon coupling M0 = 0 one has Γtot,0(M0 = 0) = ΣL + ΣR. The steady state distribution pdot → pdot = Γd/Γtot,0 is approached with relaxation rate Γtot,0. For a sym- metric situation ΣL = ΣR with ǫD = 0 one shows that pdot = 1/2 independent of the voltage, while asymmet- ric cases lead to voltage dependent stationary popula- tions. The steady state current is given by I(V ) = (e/2)[(ΓL − Γ′ L − ΓR)pdot] so that R)(1 − pdot) − (Γ′ I(V ) = e ΓLΓR − Γ′ Γtot,0 LΓ′ R . (12) A transparent expression is obtained for ǫD = 0, namely, I(V ) = e  ΣLΣR ΣL + ΣR × Z dǫ(cid:20)fβ (cid:18)ǫ − eV 2 (cid:19) − fβ (cid:18)ǫ + eV 2 (cid:19)(cid:21) P0(ǫ). (13) Despite its deficiencies mentioned above, the golden rule treatment provides already a qualitative insight in the transport characteristics. Typical results are shown in Fig. 2. The IV -curves display the expected steps at eV = 2nω0, n ∈ Z. Each time the voltage eV 2 exceeds multiples of ω0 new transport channels open associated with the excitation of one additional phonon. For higher temperatures the steps are smeared out by thermal fluc- tuations. The range of validity of this description fol- lows from the fact that a factorizing assumption for the phonon-electron correlation has been used and an instan- taneous equilibration of the phonon mode after a charge transfer, that means σ < 1 and m0 < 1. The latter constraint guarantees that conformational changes of the phonon distribution remain small. There are now three ways to go beyond this golden rule approximation. With respect to the phonon mode, ρk aρl e k!l! P0(ǫ) = e−ρa−ρeXk,l 0/2)(cid:2)coth(cid:0) θ 2(cid:1) ∓ 1(cid:3) denoting mean val- with ρa/e = (m2 ues for single phonon absorption (a) and emission (e). The exponentials in the prefactor contain the dimension- less reorganization energy m2 0 = Λ/ω0. Apparently, inelastic charge transfer includes the exchange of mul- tiple phonon quanta according to a Poissonian distri- bution. Further, one has the detailed balance relation P0(−ǫ) = e−βǫP0(ǫ). For vanishing phonon-electron cou- pling m0 → 0 only the elastic peak survives P0(ǫ) → δ(ǫ). We note again the close analogy to the P (E)-theory for dynamical Coulomb blockade [17]. Moreover, golden rule rates for intramolecular electron transfer between donor and acceptor sites coupled to a single phonon mode are of the same structure with the notable difference, of course, that in this case one has discrete density of states for both sites [20, 22]. The fundamental assumption under- lying the golden rule treatment is that equilibration of the phonon mode occurs much faster than charge trans- fer. In the last two situations this is typically guaranteed by the presence of a macroscopic heat bath (secondary bath) strongly coupled to the prominent phonon mode. Here, the fermionic reservoirs in the leads impose phonon relaxation only due to charge transfer. Thus, for finite voltage the steady state is always a nonequilibrium state that can only roughly be described by a thermal distribu- tion of the bare phonon system (see below). One way to remedy this problem is to introduce a phonon-secondary bath interaction as well, see below in Sec. II D. The re- maining transition rates easily follow due to symmetry ΓR(V, ǫD) = Γ′ R(V, ǫD) = ΣR ΣL ΣR ΣL ΓL(V, −ǫD) , ΓL(−V, ǫD) , Γ′ L(V, ǫD) = ΓL(−V, −ǫD) . (10) Now, summing up forward and backward events, the dot population follows from dpdot dt = −Γtot,0 pdot + Γd (11) one is to explicitly account for its nonequilibrium dy- namics, another is to introduce a direct interaction with a secondary heat bath in order to induce sufficiently fast equilibration. With respect to the dot degree of freedom one can exploit the fact that for vanishing charge-phonon coupling the model can be solved exactly. C. Master equation for nonequilibrated phonons To derive an equation of motion for the combined dy- namics of charge and phonon degrees of freedom, one starts from a Liouville-von Neumann equation for the full polaron transformed compound (3). Then, applying a Born-Markov type of approximation with respect to the tunnel coupling to the fermionic reservoirs, one arrives at a Redfield-type of equation for the reduced density ma- trix of the dot-phonon system [15]. An additional rotat- ing wave approximation (RWA) separates the dynamics of diagonal (populations) and off-diagonal (coherences) elements of the reduced density. Denoting with P n q the probability to find q charges on the dot (here, for sin- gle charge transfer q = 0, 1) and the phonon in its n-th eigenstate, one has d P n q (t) dt = − 1  Xα=L,R; k hfβ (Eq,α fn,k2Σα × kn ) P n q − fβ (−Eq,α q+νqi (14) with ν0 = 1, ν1 = −1 and energies Eq,α kn = ω0(k − n) + νq(µα + ǫD). The matrix elements of the phonon shift operator fn,k = hn exp(ip0l0/)ki read kn ) P k fn,k = e−m2 max(n,k) (n − k)! 0/2 (−m0)n−k l  1F1(cid:0) max(n, k) + 1, n − k + 1, −m2  Yl=min(n,k+1)  1 2 × 0(cid:1) , (15) where 1F1 denotes a hypergeometric function. The un- derlying assumptions of this formulation require weak dot-lead coupling σ < 1 and sufficiently elevated temper- atures σθ < 1 for a Markov approximation to be valid. We will see below when comparing low temperature re- sults with numerically exact data that this seems to be only a weak constraint though. 0 = P n The calculation of the steady state distribution P n mean phonon number hni = Pq,n nP n q = limt→∞ P n q (t) reduces to a standard matrix inversion. One can show that for a symmetric system with ǫD = 0, ΣL = ΣR one has P n 1 . A typical example for the q is depicted in Fig. 3. The curve is well approximated by a/m0 with a ≈ 0.7. Apparently, hni diverges for m0 → 0 since then P n 0 , P n 1 approach constants independent of the phonon number. Upon closer inspection one finds that excitation is more likely than absorption, i.e f (n, n+1) > f (n, n−1), for all 0 ≤ n ≤ N0(m0) where N0(m0) increases with decreasing 5 > n < 8 6 4 2 0 0.0 0.5 1.0 m0 1.5 2.0 FIG. 3: Mean phonon number in nonequilibrium for eV = 3ω0 and vs. the electron-phonon coupling m0. 0.20 0.15 0.10 0.05 0.00 1n P + 0n P = n P kBT = 0.1 Ñ Ω0 kBTeff = 5.2 Ñ Ω0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 n FIG. 4: Phonon number distribution in nonequilibrium for eV = 5ω0, m0 = 0.5 and kBT /ω0 = 0.1 (histogram). The solid line depicts a fit to a Boltzmann distribution. See text for details. m0. The opposite is true for n > N0(m0) so that in a steady state, depending on the voltage, the tendency is to have higher excited phonon states occupied for smaller couplings m0. In particular, for strong coupling transi- tions n → n + k, k ≥ 0 are basically blocked at quite small n. A convenient strategy to include nonequilibrium ef- fects in the phonon distribution, sometimes used in the interpretation of experimental data, is the introduction of an effective temperature Teff. This way, one could return to the simpler modeling of the previous section. However, the procedure to identify P n β = exp(−βeff ω0n)/[1−exp(−βeff ω0)] is not reliable as Fig. 4 reveals. While it clearly shows the general tendency of a substantial heating of the phonon degree of freedom induced by the electron transfer, the profile of a thermal distribution strongly differs from the actual steady state distribution. 1 ≈ P n 0 + P n Non-equilibrated phonons leave their signatures also in the IV -curves as compared to equilibrated ones. The net current through the contact follows from summing up the transfer rates from / onto the dot according (14), namely, I = fn,k2nhΣLfβ (cid:16)E0,L e 2 Xn,k − hΣLfβ (cid:16)E1,L nk (cid:17) − ΣRfβ (cid:16)E0,R 1 o . nk (cid:17)i P n nk (cid:17) − ΣRfβ (cid:16)E1,R nk (cid:17)i P n 0 (16) Fig. 5 shows that deviations are negligible for low voltages in the regime around the first resonant step (eV /2 < ω0), where at sufficiently low temperatures only the ground state participates so that the steady state distribution basically coincides with the thermal one. For larger voltages deviations occur with the tendency that for smaller couplings m0 the nonequilibrated current is always smaller than the equilibrated one (Inon < Ieq), while the opposite scenario (Inon > Ieq) is observed for larger m0. At sufficiently large voltages, one always has Inon < Ieq. This behavior results from the combination of two ingredients, namely, the phonon distributions P n q and the Frank-Condon overlaps fn,k2. To see this in detail, let us consider a fixed voltage. Then, on the one hand, for smaller m0 the steady state distribution is broad (cf. Fig. 3) so that due to normalization less weight is carried by lower lying states compared to a thermal distribution at low temperatures; on the other hand, for m0 < 1 the overlaps fn,k2 favor contributions from low lying states in the current (16) which is thus smaller than Ieq. For increasing electron-phonon cou- pling m0 > 1 overlaps fn,k2 tend to include broader ranges of phonon states also covered by P n q compared to those of low temperature thermal states. A voltage de- pendence arises since with increasing voltage higher lying phonon states participate in the dynamics supporting the scenario for smaller couplings. Interestingly, as already noted in [23] the overlaps fn,k2 may vanish for certain combinations of n, m depending on m0 due to interfer- ences of phonon eigenfunctions localized on different di- abatic surfaces Vq, q = 0, 1. D. Rate approach II The assumption of a thermally distributed phonon de- gree of freedom during the transport can be physically justified only if this mode interacts directly and suffi- ciently strongly with an additional heat bath (secondary bath) realized e.g. by residual molecular modes. Here we will generalize the formulation of Sec. II B to a situa- tion where the secondary bath is characterized by Gaus- sian fluctuations. Its corresponding modes can thus effec- tively be represented by a quasi-continuum of harmonic oscillators for which the phonon correlation function (6) can be calculated easily 6 m0=0.5 m0=1.0 m0=1.5 m0=2.0
Ñ
S e

V
I 0.2 0.1 0.0 -0.1 -0.2 -5 0 5 V
Ñ Ω0
e
IV -characteristics for equilibrated (solid) and FIG. 5: nonequilibrated (dotted) phonon distributions according to (13) and (16), respectively. Here the spectral density I(ω) describes now the com- bined distribution of the prominent mode and its sec- ondary bath. It is thus proportional to the imaginary part of the dynamical susceptibility of a damped har- monic oscillator [20]. For a purely ohmic distribution of bath modes, one has I(ω) = 2m2 0ω3 0 (ω2 − ω2 γω 0)2 + γ2ω2 , (18) where γ denotes the coupling between phonon mode and bath. The Fourier transform of exp(J) reads at finite temperatures Pγ(ǫ) = e−ργ δ(ǫ) + ℜ X(k,l)6=(0,0) × e−ργ π ρk γ,a k! ρl γ,e l! i ǫ + Ωk − Ω∗l (19) with frequency Ω = ω0ξ + iγ/2 and ργ,a(Ω) = 0/2ξ Ω2)[coth(βΩ/2) − 1] where ξ = p1 − γ2/4ω2 (m2 0. Further, ργ,e(Ω) = −ργ,a(−Ω∗) (* means complex conju- gation) and ργ = ℜ[ργ,a+ργ,e]/2. In the above expression contributions from the Matsubara frequencies in Equa- tion (17) have been neglected since they are only relevant in the regime γβ ≫ 2π which is not studied here. Ap- parently, the coupling to the bosonic bath effectively in- duces a broadening of the dot levels γ(k+l)/2 compared to the purely elastic case (9). In the low temperature regime, where for equilibrated phonons absorption (re- lated to k) is negligible, the widths grow proportional to l. The presence of the secondary bath drives the prominent phonon mode towards thermal equilibrium with a rate proportional to this broadening. Hence, if the time scale for thermal relaxation is sufficiently smaller than the time scale for charge transfer, i.e. 1/τl ≡ ΣL + ΣR/γ ≪ 1, the assumption of an equilibrated phonon mode is justified and the golden rule formulation (13) can be used with P0(ǫ) → Pγ(ǫ). However, this argument no longer ap- plies in the overdamped situation γ/ω0 ≫ 1, where the J(t) = Z ∞ 0 dω π I(ω) ω2 (cid:20)coth(cid:18) ωβ 2 (cid:19) [cos(ωt) − 1] − i sin(ωt)(cid:21) . (17) phonon mode exhibits a sluggish thermalization on the time scale γ/ω2 0 which may easily exceed τl. As already mentioned above, for vanishing charge- phonon coupling m0 = 0, the model (3) can be solved exactly to all orders in the lead-dot coupling [15]. In the frame of a rate description, one observes that in this limit the dot population (11) decays proportional to ΣL + ΣR. The golden rule version of the theory neglects this broadening in (13) since it is associated with higher order contributions to the current (13). Now, recalling that P0(ǫ) reduces to a δ-function for m0 → 0, this finite lifetime of the electronic dot level is included to all orders by performing the time integral in the Fourier transform with ǫ → ǫ − i(ΣL + ΣR)/2 ≡ ǫ − iΓtot(M0 = 0)/2 [see (11)]. In fact, this way one reproduces the exact solu- tion (one electronic level coupled to leads with energy independent couplings), i.e., its exact spectral function. To be specific, let us restrict ourselves in the remain- der to the symmetric situation ΣL = ΣR ≡ Σ/2, and ǫD = 0. Then, in the presence of the phonon mode (m0 6= 0) the corresponding function P Σ 0 (ǫ) follows from (9) by replacing the δ-function by i/[ǫ+ ω(k −l)+iΣ/2]. Again following the spirit of a rate treatment, an im- proved version of this result accounting for higher order electron-phonon correlations is obtained by using instead of the bare dot level width Σ/ ≡ Γtot,0(M0 = 0), the decay rate Γtot,0(M0 6= 0). Equivalently, one replaces i/[ǫ + ω(k − l) + iΣ/2] → i/[ǫ + ω(k − l) + iΓtot,0/2] to arrive at an improved P Σ 1 (ǫ). We note that within a Greens-function approach and upon approximating the corresponding equations of motion a similar result has been found in [15, 33] with the difference though that there instead of Γtot,0 an imaginary part of a phonon state dependent self-energy Σ′′ k,l appears. One can show that the Γtot,0 appearing here within a rate scheme is related to a thermally averaged Σ′′ k,l. Now, an additional secondary bath can be introduced 1 , leading eventually as above by combining (19) with P Σ to P Σ γ,1(ǫ) = e−ργ π ℜ X(k,l)≥(0,0) ρk γ,a k! ρl γ,e l! i 2  + Ωk − Ω∗l + i Γtot,0 ǫ (20) The width of the electronic dot level is thus voltage de- pendent and approaches the bare width from below for large voltages, that is limV →∞ Γtot,0(V ) = Σ/. The range of validity of this scheme is the following: it ap- plies to all ratios σ/m0 in the domain where the electron- phonon coupling is weak m0 < 1. For m0 > 1 charge transfer is strongly suppressed and the phonon dynamics still occurs on diabatic surfaces for σ/m0 ≪ 1 so that we expect the approach to cover this range as well. E. Comparison with numerically exact results A numerically exact treatment of the nonequilibrium dynamics of the model considered here is a formidable 0.25 0.20 0.15 0.10 0.05
Ñ
S e
I 0.00 0 5 10 V
S
e
7 M0=2, Ñ Ω0=10 M0=2, Ñ Ω0=5 M0=4, Ñ Ω0=5 M0=4, Ñ Ω0=3 15 20 FIG. 6: IV -characteristics according to approximate mod- els based on equilibrated phonons (solid) and nonequilibrated phonons (dashed) together with exact DQMC data (dots) for kBT /Σ = 0.2 and without coupling to a secondary bath (γ = 0). All quantities are scaled with respect to the dot-lead coupling Σ. task. The number of formulations which allow simula- tions in non-perturbative ranges of parameter space is very limited. Among them is a recently developed di- agrammatic Monte Carlo approach (diagMC) based on a numerical evaluation of the full Dyson series which in contrast to numerical renormalization group (NRG) methods [37] covers the full temperature range. For the sector of single charge transfer results have been obtained with and without the presence of a secondary bath inter- acting with the dot phonon mode. We note that com- putationally these simulations are very demanding as for each parameter set and a given voltage the stationary current for the IV curve needs to be extracted from the saturated value of the time dependent current I(t) for longer times. Typical simulation times are on the order of several days and even up to weeks depending on the parameter range. In contrast, rate treatments require minimal computational efforts and can be done within minutes. Here, we compare numerically exact findings with those gained from the various types of rate/master equations discussed above. . We start with the scenario where the coupling to a sec- ondary bath is dropped (γ = 0) to reveal the impact of nonequilibrium effects in the phonon mode. The formu- lation for an equilibrated phonon is based on (13) with P0 replaced by P Σ γ,1 in (20), while the steady state phonon distribution is obtained from the stationary solutions to (14). In the latter approach the intrinsic broadening of the dot electronic level due the lead coupling is intro- duced in the following way: One first determines via (14) a steady state distribution P n q . This result is used for an effective self-energy contribution (total decay rate) for non-equilibrated phonons, i.e., Γtot,neq(V ) = Σ  Xα=L,R; n,k,q fn,k2P n q fβ (Eα nk) , (21) where Eα nk = ω0(n − k) + µα. We note in passing that limV →∞ Γtot,neq(V ) = Σ/ ≡ Γtot(M0 = 0). Subse- quently, an improved result for the steady state phonon distribution at a given voltage is evaluated working again with (14) but using the replacement fβ (cid:18)ω0(k − l) ∓ 2 (cid:19) Z dǫ fβ (cid:18)ǫ ∓ 2π eV eV 2 (cid:19) → Γtot,neq [ǫ − ω0(k − l)]2 + 2Γ2 tot,neq/4 . 0.25 0.20 0.15 0.10 0.05
Ñ
S e
I 8 M0=0.5 M0=2 M0=4 M0=10 0.00 0 5 10 V
S
e
15 20 (22) FIG. 7: Same as in Fig. 6 but for fixed ω0/Σ = 5 and varying electron-phonon coupling. Of course, for Σ → 0 the standard Fermi distribution is regained. The corresponding steady state phonon dis- tribution eventually provides the current according to (16) using in this expression the same replacement (22). The procedure relies on weak electron-phonon coupling m0 < 1 and requires in principle also sufficiently elevated temperatures. Results are shown in Fig. 6 together with correspond- ing diagMC data for various coupling strengths m0. In- terestingly, the equilibrated model describes the exact data very accurately from weak up to moderate electron- phonon coupling m0 ≈ 1, while deviations appear for stronger couplings m0 & 1 and voltages beyond the first plateau eV > 2ω0. For m0 > 1 nonequilibrium ef- fects are stronger and the corresponding master equation (14) gives a better description of higher order resonant steps. Moreover, as already addressed above, even in this low temperature domain the approximate descrip- tion provides quantitatively reliable results. In Fig. 7 the frequency of the phonon mode is fixed and only the electron-phonon coupling is tuned over a wider range. For strong coupling (here m0 = 2) the equilibrated (nonequilibrated) model predicts a smaller (larger) cur- rent than the exact one in contrast to the situation for smaller m0. This phenomenon directly results from what has been said above in Sec. II C: for stronger coupling the Franck-Condon overlaps favor also higher lying phonon states that are suppressed by a thermal distribution. After all, the approximate models give not only qual- itatively a correct picture of the exact IV curves, but provide even quantitatively a reasonable description in this low temperature domain. In a next step the coupling to a secondary bath is turned on (γ 6= 0) enforcing equilibration of the phonon mode, see (20). The expectation is that in this case de- partures from the equilibrated model are reduced. In Fig. 8 data are shown for a ratio m0 = 4/5 where devia- tions occur at larger voltages as observed in the previous figures. Obviously, due to the damping of the phonon mode the resonant steps are smeared out with increasing γ. However, the approximate model predicts this effect to be more pronounced as compared to the exact data, particularly for stronger coupling γ/Σ > 1, while still γ/ω0 < 1. In fact, in the limit of very large coupling only the k = l = 0 contribution to (20) survives so that at zero temperature one arrives at lim γ→∞ I(V ) = I∞ 2 π arctan(cid:18) eV Γtot,0(V )(cid:19) (23) with the current at large voltages I∞ = eΣ/4 and Γtot,0(V ) ≤ Σ/ where equality is approached for V → ∞. It seems that a broadened equilibrium distribution of the phonon induced by the secondary bath according to (20) overestimates the broadening of individual levels. Since the approach is exact in the limit m0 → 0, the de- viations appearing in Fig. 8 are due to intimate electron- phonon-secondary bath correlations not captured by the rate approach. In the overdamped regime, i.e. γ/ω0 > 1, the dynamics of the phonon mode slows down and may become almost static on the time scale of the charge transfer. In this adiabatic regime an extended version of the master equation (14) is not trivial since the con- ventional eigenstate representation becomes meaningless. One should then better switch to phase-space coordinates and develop a formulation based on a Fokker-Planck or Smoluchowski equation for the phonon. This will be the subject of future research. The essence of this comparison is that, as anticipated from physical arguments already in Sec. II A, a rate de- scription does indeed provide even quantitatively accu- rate results in the regime of weak to moderate electron- phonon coupling m0 < 1 and for all σ/m0. Deviations that occur for larger values of m0 can partially be ex- plained by nonequilibrium distributions in the phonon distribution, where, however, the master equation ap- proach seems to overestimate this effect. In order to obtain some insight into the nature of this deficiency, a minimal approach consists of adding to (14) with the extension (22) a mechanism that enforces relaxation to thermal equilibrium with a single rate constant Γ0 that 0.25 0.20 0.15 0.10 0.05
Ñ
S e
I Ñ Γ=1.95 Ñ Γ=0.39 Ñ Γ=0 0.00 0 5 10 V
S
e
15 20 FIG. 8: IV -characteristics in presence of a secondary heat bath interacting with the phonon with various coupling con- stants γ. Shown are approximate results (solid) using (20) and diagMC data (dots); energies are scaled with Σ. Other parameters are kBT /Σ = 0.2, m0 = 4/5, σ = 0.2.
Ñ
S e

V
I 0.25 0.20 0.15 0.10 0.05 0.00 0 M0=2, Ñ Ω0=10 M0=2, Ñ Ω0=5 M0=4, Ñ Ω0=5 M0=4, Ñ Ω0=3 5 10 15 20 V
S
e
FIG. 9: Same as in Fig. 6, but for nonequilibrated phonons based on an extended master equation (solid) in comparison to exact diagMC data (dots). 9 Figs. 6, 7 are shown in Figs. 9, 10 in comparison with exact diagMC data. There, the same equilibration rate Γ0/Σ = 0.25 is used for all parameter sets. Astonish- ingly, this procedure provides an excellent agreement over the full voltage range. It improves results particularly in the range of moderate to stronger electron phonon cou- pling, but has only minor impact for m0 < 1. The indica- tion is thus that electron-phonon correlations neglected in the original form of the master equation have effec- tively the tendency to support faster thermalization of the phonon. Indeed, preliminary results with a gener- 0.25 0.20 0.15 0.10 0.05
Ñ
S e

V
I M0=0.5 M0=2 M0=4 M0=10 0.00 0 5 10 V
S
e
15 20 FIG. 10: Same as in Fig. 7, but for nonequilibrated phonons based on an extended master equation (solid) in comparison to exact diagMC data (dots). alized master equation where the coupling between di- agonal (populations) and off-diagonal (coherences) ele- ments of the reduced charge-phonon density matrix are retained (no RWA approximation) indicate that this cou- pling leads to an enhanced phonon-lead interaction and thus to enhanced phonon equilibration. serves as a fitting parameter. Accordingly, the respective time evolution equation for P n q (t) receives an additional term −Γ0[P n β /2] with the Boltzmann distribu- tion for the bare phonon degree of freedom P n β . Cor- responding results for the same parameter range as in q (t) − P n We thank L. Muhlbacher for fruitful discussions and for providing numerical data of Ref. [29]. Financial support was provided by the SFB569, the Baden-Wurttemberg Stiftung, and the German-Israeli Foundation (GIF). Acknowledgments [1] K. Goser, P. Glosekotter, J. Dienstuhl, Nanoelectronics and Nanosystems: From Transistors to Molecular and Quantum Devices, (Springer, Berlin, 2004). [2] G. Cuniberti, G. Fagas, K. Richter (eds.), Introducing Molecular Electronics, (Springer, Berlin, 2005). [3] J. C. Cuevas, E. Scheer, Molecular Electronics -- An In- troduction to Theory and Experiment, (World Scientific, Singapore, 2010). [4] J. Reichert, R. Ochs, D. Beckmann, H. B. Weber, M. Mayor, H. von Lohneysen, Phys. Rev. Lett. 88, 176804 (2002). [5] M. Ruben, A. Landa, E. Lortscher, H. Riel, M. Mayor, H. Gorls, H. B. Weber, A. Arnold, F. Evers, Small 4, 2229 (2008). [6] J.-D. Pillet, C. Quay, P. Morfin, C. Bena, A. Levy Yeyati, P. Joyez, Nature Physics 6, 965 (2010). 10 [7] H. Park et al., Nature 407, 57 (2000). [8] H. Song, Y. Kim, Y. H. Jang, H. Jeong, M. A. Reed, T. Lee, Nature 462, 1039 (2009). [9] D. Secker, S. Wagner, S. Ballmann, R. Hartle, M. Thoss, [22] A. Nitzan, Annu. Rev. Phys. Chem. 52, 681 (2001). [23] J. Koch, F. v. Oppen, Phys. Rev. Lett. 94, 206804 (2005); J. Koch, M. Semmelhack, F. v. Oppen, A. Nitzan, Phys Rev B 73, 155306 (2006). H. B. Weber, Phys. Rev. Lett. 106, 136807 (2011). [24] A. Donarini, M. Grifoni, K. Richter, Phys Rev Lett 97, [10] L. Venkataraman, J. E. Klare, C. Nuckolls, M. S. Hy- 166801 (2006). bertsen, M. L. Steigerwald, Nature 442, 904 (2006). [25] M. Leijnse, M. R. Wegewijs, Phys. Rev. B, 78, 235424 [11] I. Fernandez-Torrente, K. J. Franke, and J. I. Pascual, (2008). Phys. Rev. Lett. 101, 217203 (2008). [12] Y. Xue, S. Datta, M.A. Ratner, Chem. Phys. 281, 151 (2002). [13] P. Damle, A.W. Ghosh, S. Datta, Chem. Phys. 281, 171 (2002). [26] C. Timm, Phys. Rev. B 77, 195416 (2008). [27] H. Hubener, T. Brandes, Phys. Rev. B 80, 155437 (2009). [28] R. Hartle, M. Thoss, Phys. Rev. B 83, 115414 (2011). [29] L. Muhlbacher and E. Rabani, Phys. Rev. Lett. 100, 176403 (2008). [14] Y. Meir and N. S. Wingreen, Phys. Rev. Lett. 68, 2512 [30] L. Muhlbacher, J. Ankerhold, and A. Komnik, Phys. Rev. (1992). Lett. 95, 220404 (2005). [15] A. Mitra, I. Aleiner, and A.J. Millis, Phys. Rev. Lett. B [31] L. Muhlbacher and J. Ankerhold, New J. Phys. 11, 69, 245302 (2004). 035001 (2009). [16] S. A. Gurvitz, Phys Rev B 57, 6602 (1998). [17] G.-L. Ingold and Y.V. Nazarov, Charge Tunneling Rates in Ultrasmall Junctions, in: H. Grabert and M.H. De- voret (eds.), Single Charge Tunneling, (Plenum Press, New York, 1992). [18] L. Muhlbacher, J. Ankerhold, J. C. Escher, J. Chem. [32] J. C. Escher and J. Ankerhold, Phys. Rev. A 83, 032122 (2011). [33] K. Flensberg, Phys. Rev. B 68, 205323 (2003); S. Braig and K. Flensberg, ibid 68, 205324 (2003). [34] R. Egger, A. O. Gogolin, Phys Rev. B 77, 113405 (2008). [35] S. Maier, T.L. Schmidt, A. Komnik, Phys. Rev. B 83, Phys. 121, 12696 (2004). 085401 (2011) [19] L. Muhlbacher and Ankerhold, J. Chem. Phys. 122, [36] T.L. Schmidt, A. Komnik, Phys. Rev. B 80, 041307 184715 (2005). (2009) [20] U. Weiss, Quantum Dissipative systems (3rd edition, [37] R. Bulla, T. A. Costi, T. Pruschke, Rev. Mod. Phys. 80, World Scientific, Singapore, 2008). 395 (2008). [21] D. Segal, A. Nitzan,W.B. Davis, M.R.Wasielewski, M.A. Ratner, J. Phys. Chem. B 104, 3817 (2000). 

1107.5367

2

1107

Spin Polarization Phenomena and Pseudospin Quantum Hall Ferromagnetism in the HgTe Quantum Well

The parallel field of a full spin polarization of the electron gas in a \Gamma8 conduction band of the HgTe quantum well was obtained from the magnetoresistance by three different ways in a zero and quasi-classical range of perpendicular field component Bper. In the quantum Hall range of Bper the spin polarization manifests in anticrossings of magnetic levels, which were found to strongly nonmonotonously depend on Bper.

cond-mat.mes-hall

cond-mat

Spin Polarization Phenomena and Pseudospin Quantum Hall Ferromagnetism in the HgTe Quantum Well M. V. Yakunina, A. V. Suslovb, S. M. Podgornykha, S. A. Dvoretskyc and N. N. Mikhailovc aInstitute of Metal Physics, S. Kovalevskaya Str., 18, Ekaterinburg 620990, Russia bNHMFL, FSU, 1800 East Paul Dirac Drive, Tallahassee, Florida 32310, USA cInstitute of Semiconductor Physics, Lavrentyev Ave., 13, Novosibirsk 630090, Russia Abstract. The parallel field of a full spin polarization of the electron gas in a Г8 conduction band of the HgTe quantum well was obtained from the magnetoresistance by three different ways in a zero and quasi-classical range of perpendicular field component BB⊥. In the quantum Hall range of B⊥B the spin polarization manifests in anticrossings of magnetic levels, which were found to strongly nonmonotonously depend on BB⊥. Keywords: HgTe quantum well, magnetoresistance, quantum Hall, tilted magnetic fields, magnetic level coincidences, anticrossings. PACS: 73.21.Fg, 73.43.-f, 73.43.Qt, 73.43.Nq Spin polarization underlies the basic principles of spintronics devices. Full spin polarization of the electronic system is easily achievable in the HgTe conduction band due to its large value of g*m*/m0, the effective Lande g-factor multiplied by the effective to free mass ratio, thus making it promising for applications and studies of a variety of spin phenomena. In particular, rich patterns of the spin level coincidences are observed in the HgTe quantum well (QW) under tilted magnetic fields that extend into the range of high field component BB⊥ perpendicular to the layer where the quantum Hall effect (QHE) is well realized [1]. We present a study of quantum magnetotransport under tilted magnetic fields in a 20.3 nm wide HgTe symmetric QW with the electron gas density nS ≈ 1.5×1015 m-2 and mobility of 22 m2/V·s. The spin level coincidences manifest in a number of peculiarities in the longitudinal ρxx(BB⊥,BB ) (Fig. 1a, BB – field component parallel to the layer) and Hall magnetoresistivities (MR) residing on different trajectories in the (B⊥B ,BB)-plane. First, they align well on a set of straight beams going from zero at fixed tilt angles θr that satisfy a typical relation for the coincidences: g*m*/m0 = 2r⋅cosθr, r = 1,2,3…[2]. This means that the coincidences in the Г8 conduction band in our case may be well described in terms of a usual Г6 band as in a traditional semiconductor QW. Second, another system of traces along the coincidences, descending from a 0 = 15.4 T on the BB axis, may be drawn. This series is approximately single point BB described at high enough tilts in terms of the Г6-like band by the equation BB = 2(BB1 - MB⊥)/(g*m*/m0), M = 1,3,5…, where B1 B = hnS /e is the MR minimum position for the magnetic level filling factor ν = 1, in a good agreement with our experimental data. 0 = 2B1B /(g*m*/m0) fulfil the relation g*μBBB 0 = 2EF, with EF The convergence point BB – the Fermi level, μB – Bohr magneton, thus yielding the field of a full spin polarization of the electron gas under pure parallel field (Fig. 1b). (b) B=0 0 B<B 0 B EF B μB * g (c) S 6 5 4 3 2 1 ) T ( ) B ( f y c n e u q e r f T F F 0 = 15.4T B S 2EF 0.012 0.011 0.01 ) 2 e 0.009 / h ( ) B 0.008 ( x ρx 0.007 0 0 0.006 8 10 12 14 16 18 20 B, B (T) FIGURE 1. (a) The magnetoresistivity ρxx(BB⊥,BB ) at 0.32 K after intermediate illumination. (b) The evolution of spin splitting under pure parallel field until all the electrons pass into the lower spin subband acquiring the same spin polarization. (c) Three ways of probing the full spin polarization by 0 of the descending trajectories on the (B⊥B ,BB)- means of magnetoresistivity: the convergence point BB plane (the dashed vertical); the FFT frequency as a function of B: falling of the lower (S↓) line to zero and a concomitant saturation of the upper (S↑) line; a point on ρ(B=BB ) separating two functions. 2 4 6 The other way to estimate the redistribution of electrons between two spin subbands is from the Fourier analysis of oscillations in ρxx(1/BB⊥) taken along the circle trajectories in the (B⊥B ,BB)-plane for fixed values of the total field B. This results in two lines of FFT frequency f vs. B (Fig. 1c) describing the behavior of electron densities in the subbands: nsi = fi×e/h. The lines diverge from a single point with the lower line going to zero (exhausting of the upper subband) and the upper line simultaneously going to saturation. This result is compared with MR in a pure BB (Fig. 1c). It appears that just this field separates two kinds of dependences in ρ(B=B): a somewhat complicated function at lower fields from a monotonously increasing one at higher fields as has been seen in a number of studies conducted in parallel fields [3]. Thus, all three techniques indicate a full spin polarization occurring at the same field. At higher BB⊥, where the QHE is well developed, the MR features for coincidences acquire a complicated structure with local ρxx peaks splitted in couples of peaks shifted in opposite directions of B⊥ B (Fig. 1a), thus indicating the formation of anticrossings at the points of expected level crossings. The effect is significantly enhanced after IR illumination of the sample that causes a considerable improvement of oscillations indicating a narrowing of magnetic levels. Unexpectedly, it was found that the anticrossings depend nonmonotonously on BB⊥: the anticrossing at ν = 3 manifests pronouncedly stronger than the neighboring ones at ν = 2 and 4, as it is easily seen while going along the descending trace for M = 1 on the (B⊥B ,BB)-plane. The difference is dramatically enhanced after illumination and with decreased temperature. The activation gaps deduced from the temperature dependences of MR at anticrossings confirm this nonmonotonicity with the ν = 3 gap being a half an order larger than those for its neighbors. This result looks counterintuitive since the overlapping of magnetic levels with decreased B⊥B seems to monotonously destroy the reasons for appearing of anticrossings as it has been observed so far on other materials [4,5]. The conventional explanation of the anticrossings is in terms of the pseudospin anisotropy of the electronic system [6]: as the approaching magnetic levels (having different pseudospin numbers) tend to swop their order in energy relatively EF, the Hartree-Fock energy of the system may decrease and, as this decrease starts before the level crossing (due to a hybridization of the levels), this crossing does not occur, the stronger is the energy gain the larger is the anticrossing gap. The systems inclined to transitions into pseudospin ordered states under QHE conditions are called QH Ferromagnets (QHF). Estimations for an easy-axis QHF in a 2D layer according to Eq. 21 in [6] does not yield a big difference for anticrossings at ν = 2, 3, and 4. Therefore, we tentatively attribute the observed difference to the coupling of BB with the orbital degree of freedom in a QW of a finite width resulted in a substantial difference in the charge density profiles across the QW for the two pseudospin levels [5]. This coupling enhances the magnetic anisotropy energy for ν = 3 coincidence as compared to that for ν = 2 since its BB is about a factor of 1.5 stronger, causing a relative shrinkage of the wave functions. On the other hand, the coincidences at ν ≥ 4 are restored because they go outside of the QH range of BB⊥. The observed sharp changes of anticrossings with the fields and, for fixed anticrossings, with a change in a sample state due to illumination indicate the phase-transition-like character of these transformations. ACKNOWLEDGMENTS Authors are grateful to E. Palm, T. Murphy, J. H. Park, and G. Jones for help with the experiment. Supported by RFBR, projects 11-02-00427, 09-02-96518. NHMFL is supported by NSF (DMR-0654118), the State of Florida, and the US DOE. REFERENCES 1. M. V. Yakunin, S. M. Podgornykh, N. N. Mikhailov and S. A. Dvoretsky, Physica E 42, 948-951 (2010); M. V. Yakunin et al., J. Phys.: Conf. Ser. (2011), in print. 2. R. J. Nicholas, M. A. Brummell, J. C. Portal et al., Solid State Comm. 45, 911-914 (1983); R .J. Nicholas, R. J. Haug, K. von Klitzing and G. Weinmann, Phys. Rev. B 37, 1294-1302 (1988). 3. J. Yoon, C. C. Li, D. Shahar, D. C. Tsui and M. Shayegan, Phys. Rev. Lett. 84, 4421-4424 (2000); B. Nedniyom, R. J. Nicholas, M. T. Emeny, L. Buckle, A. M. Gilbertson, P. D. Buckle and T. Ashley, Phys. Rev. B 80, 125328 (2009); I. L. Drichko, I. Yu. Smirnov, A. V. Suslov, O. A. Mironov and D. R. Leadley., Phys. Rev. B 79, 205310 (2009). 4. S. Koch, R. J. Haug, K. V. Klitzing and M. Razeghi, Phys. Rev. B 47, 4048-4051 (1993); W. Desrat, F. Giazotto, V. Pellegrini, M. Governale, F. Beltram et al., Phys. Rev. B 71, 153314 (1993). 5. T. Jungwirth, S. P. Shukla, L. Smrcka, M. Shayegan and A. H. MacDonald, Phys. Rev. Lett. 81, 2328-2331 (1998). 6. T. Jungwirth and A. H. MacDonald, Phys. Rev. B 63, 035305 (2000). 

1508.05136

1

1508

Quadrupolar and anisotropy effects on dephasing in two-electron spin qubits in GaAs

Understanding the decoherence of electron spins in semiconductors due to their interaction with nuclear spins is of fundamental interest as they realize the central spin model and of practical importance for using electron spins as qubits. Interesting effects arise from the quadrupolar interaction of nuclear spins with electric field gradients, which have been shown to suppress diffusive nuclear spin dynamics. One might thus expect them to enhance electron spin coherence. Here we show experimentally that for gate-defined GaAs quantum dots, quadrupolar broadening of the nuclear Larmor precession can also reduce electron spin coherence due to faster decorrelation of transverse nuclear fields. However, this effect can be eliminated for appropriate field directions. Furthermore, we observe an additional modulation of spin coherence that can be attributed to an anisotropic electronic $g$-tensor. These results complete our understanding of dephasing in gated quantum dots and point to mitigation strategies. They may also help to unravel unexplained behaviour in related systems such as self-assembled quantum dots and III-V nanowires.

cond-mat.mes-hall

cond-mat

Quadrupolar and anisotropy effects on dephasing in two-electron spin qubits in GaAs Tim Botzem,1 Robert P. G. McNeil,1 Dieter Schuh,2 Dominique Bougeard,2 and Hendrik Bluhm1 1JARA-Institute for Quantum Information, RWTH Aachen University, D-52074 Aachen, Germany 2Institut fur Experimentelle und Angewandte Physik, Universitat Regensburg, D-93040 Regensburg, Germany (Dated: January 23, 2021) 5 1 0 2 g u A 0 2 ] l l a h - s e m . t a m - d n o c [ 1 v 6 3 1 5 0 . 8 0 5 1 : v i X r a fundamental Understanding the decoherence of electron spins in semiconductors due to their interac- tion with nuclear spins is of in- terest as they realize the central spin model1 and of practical importance for using electron spins as qubits. Interesting effects arise from the quadrupolar interaction of nuclear spins with electric field gradients, which have been shown to suppress diffusive nuclear spin dynamics2. One might thus expect them to enhance electron spin coherence3. Here we show experimentally that for gate-defined GaAs quantum dots, quadrupo- lar broadening of the nuclear Larmor precession can also reduce electron spin coherence due to faster decorrelation of transverse nuclear fields. However, this effect can be eliminated for appro- priate field directions. Furthermore, we observe an additional modulation of spin coherence that can be attributed to an anisotropic electronic g- tensor. These results complete our understand- ing of dephasing in gated quantum dots and point to mitigation strategies. They may also help to unravel unexplained behaviour in related systems such as self-assembled quantum dots4 and III-V nanowires5. Electron spin qubits in GaAs quantum dots have played a central role in demonstrating the key operations of semiconductor spin qubits6 -- 9. A prominent and often dominant dephasing mechanism in these devices as well as other semiconductor spin qubits4,10 is the interaction of the electron spin with 104 − 106 nuclear spins of the host lattice. While the fundamentals of this interaction have been studied quite extensively11 -- 14, and theory and experiments are in reasonable agreement15, theory pre- dicts a potential for much longer dephasing times16 than observed so far and it remains an open question as to what ultimately limits electron spin coherence. Remark- able progress has also been made in eliminating dephas- ing from nuclear spins by using Si-based systems17 that can be isotopically purified, but this route is not open for III-V semiconductor systems, where all isotopes carry nuclear spin. Nevertheless, the latter remain of practical interest because of their lower effective mass, single con- duction band valley and potential for optical coupling. Recently the role of quadrupolar coupling of nuclear spins to electric field gradients (EFGs) from charged impurities or strain has been investigated both experimentally and theoretically2,3,18 -- 20, but its influence on electron spin coherence is still unclear. FIG. 1. Device Layout and quadrupole boadening a, Gates used for pulsed qubit control are depicted in blue; the energy of the conduction band edge ECB is shown on the left. b, Nuclear spins 3/2 in the proximity of the quantum dot experience quadrupolar coupling to electric field gradients Vx(cid:48)x(cid:48) induced by crystal distortion due to the electric field of the triangular quantum well. c, While the center transi- tion stays unchanged, the satellite transitions, distorted by the electron's own charge, exhibit a quadrupolar shift by ωQ. d, The resulting frequency distribution F (ω) consists of two Gaussians with different variances. e, Echo amplitude for magnetic fields along the [110] axis, showing oscillations with the relative Larmor frequencies of the three nuclear spins. A semi-classical model (solid line) is used to fit the data (dots, offset for clarity). In contrast to the expected enhancement of coherence [001][110][110]100mT200mT300mT400mT500mT600mT700mT750mTEvolution time τ (µs)Echo amplitude P(Singlet)0.511.52eI = 3/2ω0ω0ω0baF(ω)ω0ωδωQdω0+ωQVij01020ω0−ωQω0cBextEzBexteµ+-+-Vx'x'EE2 FIG. 2. B-field direction dependence Echo amplitude at 300 mT as a function of separation time for different in-plane magnetic field directions θ, with 0◦ corresponding to the [110] direction. Curves are offset for clarity. At 45◦, parallel to the crystallographic [100] axis, the coherence time is enhanced as quadrupolar couplings are suppressed. When rotating the field a g-factor anisotropy leads to oscillations, associated with the three different nuclear Larmor frequencies. A semi-classical model (solid line) is used to fit the data (dots). due to quadrupolar suppression of nuclear spin flip-flops, we find that Hahn echo coherence improves when the magnetic field is rotated to minimize quadrupolar broad- ening of nuclear levels. In addition, we find a complex pattern of collapses and revivals of the echo signal unless the magnetic field is aligned with specific crystal axes, which we explain with an anisotropic g-tensor causing a coupling of the nuclear Larmor precession to the electron spin. √ The qubit studied here is a two-electron spin qubit6,21, using the mz = 0 subspace S = (↑↓(cid:105) − ↓↑(cid:105)) / √ 2 and T0 = (↑↓(cid:105) + ↓↑(cid:105)) / 2 of two electron spins. These elec- trons are confined in a GaAs double quantum dot formed by electrostatic gating (Fig. 1a) of a two-dimensional elec- tron gas (2DEG). The effects explored in this work apply equally to single electron spins. A random configuration of the nuclear spins introduces an effective magnetic field of a few mT, the Overhauser field, whose dynamics cause qubit dephasing. Hahn echo measurements that eliminate dephasing from slow fluctu- ations allow studying these dynamics, as they become the dominant dephasing mechanism. We follow the experi- mental procedure from Ref. 15 (see also Methods), im- plementing the required π-pulse to invert the state of the qubit halfway through the evolution time τ using the ex- change interaction between the two spins. Fig. 1c shows the spin echo signals as a function of separation time for magnetic fields aligned along the [110] crystal axis. (Note that we experimentally cannot distinguish between the [110] and [110] axes, but refer to the direction parallel to the dot connection line as the [110] axis throughout the paper for ease of reading.) Similar results to Refs. 15 and 22 are obtained, but with approximately a factor two shorter coherence times (see Supplementary Infor- mation). At fields below 500 mT, a second order cou- pling to the oscillating, transverse nuclear field (i.e., its component perpendicular to the external field) leads to periodic collapses and revivals of the echo amplitude13 -- 15. Revivals occur at times corresponding to the periods of the relative Larmor precession of the three species 69Ga, 71Ga and 75As. The overall envelope decay can be mod- eled by assuming a phenomenological broadening δB of the Larmor frequencies that causes fluctuations of the transverse hyperfine field of each species. While such a broadening is expected from dipolar in- teraction between nuclei, fitting the current and earlier15 data requires a value of δB = 1.4 mT and δB = 0.3 mT respectively, at least a factor three larger than the in- trinsic dipolar nuclear linewidth of 0.1 mT obtained from NMR measurements in pure GaAs23. More direct mea- surements of the nuclear dynamics based on correlation of rapid single shot measurements24 are consistent with Echo amplitude P(Singlet)00.60.811.21.41.61.8102030Evolution time τ (µs)0°15°30°45°60°75°90°3 FIG. 3. B-field magnitude dependence Echo amplitude for magnetic field magnitudes along the [100] axis. A g-factor anisotropy causing different quantization axes for electron and nuclei spins leads to oscillations with the three nuclear Larmor frequencies. For small magnetic fields the echo signal is strongly suppressed in the first hundreds of nanoseconds, but revives at later times. A semi-classical model (solid line) is used to fit the data (dots). these values. NMR experiments on GaAs samples with impuri- ties revealed similar excess line broadening which was found to depend on the field direction and explained by quadrupolar effects23,25. Electric fields from charged im- purities, strain or the triangular quantum well, used here to confine electrons (see Fig. 1a), distort the valence or- bitals and crystal lattice, thus creating EFGs at nuclear sites (see Fig. 1b). These EFGs couple to the quadrupolar momentum of the nuclei with spin I = 3/2 and modify the splitting of the Iz = ±3/2 ↔ ±1/2 satellite Larmor transitions by23 ωQ,α = eQα 2 Vx(cid:48)x(cid:48), (1) where Qα is the quadrupolar moment of nuclear species α and Vx(cid:48)x(cid:48) denotes the component of the electric field gra- dient tensor in the direction of the external field (Fig 1b). The local field gradients are given by23 Vx(cid:48)x(cid:48) = R14,αEz cos (2θ) . (2) R14,α is the species dependent response tensor compo- nent relating electric fields to electric field gradients at the nuclear site and θ is the angle between the magnetic field and the [110] axis. The variation of the local elec- tric field Ez across the electronic wave function due to the electron's own charge density introduces a broadening of the precession frequencies. The dependence of ωQ,α on θ, arising from the crystal symmetry of the host material, implies a suppression of the effect for a field along the [100] and [010] axis. The Hahn echo amplitude as a function of separation time is shown in Fig. 2 for different in-plane field direc- tions θ between the [110] and the [1¯10] axes. Indeed a factor two longer coherence is seen for θ = 45◦, parallel to the [100] (or [010]) direction. Apart from this enhance- ment, another oscillatory modulation appears, reaching a maximum at the same angle. To further investigate the origin of these oscillations we aligned Bext along the [100]-axis and varied its mag- nitude in Fig. 3. With decreasing Bext the frequency of the modulation decreases, until at 100 mT only a very fast decay of the echo amplitude followed by a revival at τ ≈ 13 µs occurs. This envelope modulation can be ex- plained by an electronic g-factor anisotropy, arising from an asymmetric confinement of the electron in the 2DEG and spin-orbit coupling26 -- 28. The main axes of the g- tensor are expected to be the [110] and [1¯10] crystal axis, consistent with the absence of a fast echo modulation with B along these directions. For other field directions, the quantization axis of the electron differs from the ex- ternal field around which the nuclear spins precess. A linear coupling to the transverse nuclear magnetic field 500 mT400 mT300 mT200 mT100 mT1015202530350.60.811.21.41.6Evolution time τ (µs)Echo amplitude P(Singlet) 50B⊥ nuc thus appears in the effective magnetic field deter- mining the electronic Zeeman splitting (see Fig. 4a and Supplementary Information): Beff = g(cid:107)Bext + g⊥B ⊥ nuc(t), (3) where g(cid:107)(g⊥) denotes the (off-)diagonal entries of the g- tensor. During the free evolution part of the spin echo, the qubit acquires a phase arising from B⊥ nuc(t). Due to the dynamics of B⊥ nuc(t) that phase is not eliminated by the echo pulse and hence leads to dephasing. But whenever the evolution time τ /2 is a multiple of all three Larmor frequencies, the net phase accumulated vanishes and the echo amplitude recovers. Partial recovery oc- curs if the evolution time only matches a multiple of the Larmor period of two or one species. To obtain a quantitative description of quadrupolar and anisotropy effects, we adapt the semiclassical model of Ref. 15, based on computing the total electronic phase accumulated due to the precessing nuclear spins and av- eraging over the initial nuclear state. The transverse hy- perfine field is modeled as the vector sum of Gaussian distributed contributions arising from the three nuclear species and the spread of quadrupolar shifts. The dis- tribution of nuclear precession frequencies F (ω) is cho- sen such that the correlation function of the transverse field is that obtained from an ensemble of independent nuclear spins 3/2 subjected to a Gaussian distribution of quadrupolar shifts (see Supplementary Information). F (ω) is taken as the weighted sum of two Gaussians centered on the Larmor frequency, reflecting the contri- butions from the unperturbed center transition and the quadrupole broadened satellite transitions as schemati- cally depicted in Fig. 1d. The rms-width of the quadrupo- lar broadened distribution is given by the variation of electric fields via equation (1) and (2). Using this model we fit the data (Fig. 1-3) with most free parameters being independent of the magnetic field (see Supplementary Information). Most relevant for this work are the quadrupolar broadenings of nuclear transi- tion and the linear coupling to transverse hyperfine fields g⊥ (both depending on field direction only) shown in Fig. 4b and c. As predicted, the quadrupolar broaden- ing approximately vanishes at θ = 45◦ and is maximal at θ = 0◦ and θ = 90◦. The maximum magnitude of δBα is consistent with the electric field variation generated by the electron in the dot (see Supplementary Information). The off-diagonal g-tensor element g⊥ shows the predicted sin (2θ) dependence, and its maximum anisotropy of 5 % is comparable with that found in quantum wells27. One of our key results is that quadrupole broaden- ing of nuclear spins can contribute to electronic de- phasing by decorrelating the transverse nuclear polariza- tion, which contributes to the electronic Zeeman split- ting to second order. While in principle another source of anisotropy with the same angular dependence could explain the observed variation of the coherence time, we are not aware of any other plausible mechanism. Anisotropic diffusion29 shows a different angular depen- 4 FIG. 4. g -factor anisotropy and fit parameters a, Due to an anisotropic g-tensor electron and nuclear spins have different quantization axes, Beff and Bext, respectively. This leads to a linear contribution of the transverse Overhauser field to the electronic Zeeman splitting, oscillating with the Larmor frequencies of the nuclear spins. b, c, Fit parameters extracted for different in-plane magnetic field directions θ. The quadrupolar contribution to nuclear broadening δBα = δωQ,α/γα, expressed in terms of an equivalent line width for the three isotopes using equation (1), vanishes at 45◦ along the [100] direction. At the same angle, the sin(2θ) dependence of the coupling g⊥ to the transverse hyperfine field reaches a maximum. dence with the longest coherence times along the [110] direction. Our interpretation is further supported by the good quantitative agreement with the model and NMR measurements23,25. This result does not contradict the reported suppression of nuclear spin diffusion20 by quadrupole effect19 as spin diffusion mostly affects elec- tron coherence via the longitudinal polarization, whereas in our case the transverse coupling is dominant. An isotropic g-factor in combination with an anisotropic hy- perfine interaction would lead to the same echo modula- tion when rotating Bext, but the anisotropy of the hyper- fine interaction is usually assumed to be negligible as the conduction band wavefunction of GaAs is predominantly s-type. While in the present sample g-factor anisotropy and quadrupolar effects cannot be eliminated simultaneously, symmetric, possibly back-gated quantum wells27 should allow the elimination of any g-factor anisotropy. The back gate could also be used to tune quadrupolar interac- tion, as it depends on the electric field, thus allowing fur- ther studies. Given that the strain-induced quadrupole broadening in self-assembled dots was found to be 3- 4 orders of magnitudes larger19, it likely also has pro- nounced effects on the coherence of this type of quantum dot, which is currently less well understood than that of gated dots. Furthermore, the echo modulation due to an anisotropic g-factor may also play an important role in III-V nanowire qubits, where strong g-factor anisotropies and short coherence times have been measured5,30. 00.020.040.060θ (°)153045607590 0θ (°)153045607590 [110][110]a b g/gδBα (mT)Beff(t)BextBnuc(t)^^θc 01269Ga71Ga75AsI. METHODS A. Qubit system. The quantum dots used in this work were fabricated on a GaAs/Al0.69Ga0.31As heterostructure with Si-δ-doping 50 nm below the surface and a spacer thickness of 40 nm, leaving the 2DEG at 90 nm depth, as shown in Fig. 1a (see Supplementary Information). 5 by measuring the resistance of a nearby sensing dot via RF-reflectometry. Such a pulse cycle with varying evo- lution times is repeated several million times and the average echo amplitude is recorded. Simultaneous his- togramming of individual measurement outcomes is used for normalization7. The fine tuning of the pulses that was necessary in Ref. 15 to avoid artifacts from shifts of the wave function has been eliminated due to improved RF-engineering. B. Echo sequence. C. Acknowledgements Following the experimental procedure for Hahn spin echo measurements from Ref. 15 we first initialize the qubit system in the spin singlet groundstate S by puls- ing both electrons into one dot. Rapidly separating the electrons into both dots lets them evolve in different Zee- man fields arising from the external magnetic field Bext and the fluctuating local Overhauser field BL(R) for a time τ . A gradient ∆Bz = BL − BR/2 in the hyperfine field of the two dots leads to coherent rotations between S and T0 and fluctuations in ∆Bz cause dephasing. An exchange splitting between the spin singlet S and triplet state T0 arises from inter-dot tunnel-coupling. This ex- change allows electric control of the qubit by varying the difference in electrostatic potential between the two dots, on the nanosecond timescale with an arbitrary waveform generator. Using this exchange interaction to perform a π-pulse by driving rotations between the eigenstates ↑↓(cid:105) and ↓↑(cid:105), we swap the two electrons halfway through the evolution time τ . Lastly, we read out the final qubit state by pulsing the electrons into one dot. Using Pauli-spin- blockade we distinguish between singlet and triplet states This work was supported by the Alfried Krupp von Bohlen und Halbach Foundation and DFG grant BL 1197/2-1 and SFB 689. D. Author contribution Molecular-beam-epitaxy growth of the sample was car- ried out by D.S. and D.B.. T.B. and R.P.G.M. set-up the experiment. T.B. fabricated the sample and conducted the experiment. T.B. and H.B. developed the theoretical model, analyzed the data and wrote the paper. E. Additional information The authors declare no competing financial interests. Online supplementary information accompanies this pa- per. Correspondence and requests for materials should be addressed to H.B. 1 Chen, G., Bergman, D. L. & Balents, L. Semiclassical dynamics and long-time asymptotics of the central-spin problem in a quantum dot. Physical Review B 76, 045312 (2007). 2 Welander, E., Chekhovich, E., Tartakovskii, A. & Burkard, G. Influence of Nuclear Quadrupole Moments on Elec- tron Spin Coherence in Semiconductor Quantum Dots. arXiv:1405.1329 [cond-mat] (2014). ArXiv: 1405.1329. 3 Sinitsyn, N. A., Li, Y., Crooker, S. A., Saxena, A. & Smith, D. L. Role of Nuclear Quadrupole Coupling on Decoher- ence and Relaxation of Central Spins in Quantum Dots. Physical Review Letters 109, 166605 (2012). 4 Press, D. et al. Ultrafast optical spin echo in a single quan- tum dot. Nature Photonics 4, 367 -- 370 (2010). 5 Nadj-Perge, S., Frolov, S. M., Bakkers, E. P. a. M. & Kouwenhoven, L. P. Spin-orbit qubit in a semiconductor nanowire. Nature 468, 1084 -- 1087 (2010). 6 Petta, J. R. et al. Coherent Manipulation of Coupled Elec- tron Spins in Semiconductor Quantum Dots. Science 309, 2180 -- 2184 (2005). 7 Barthel, C., Reilly, D. J., Marcus, C. M., Hanson, M. P. & Gossard, A. C. Rapid Single-Shot Measurement of a Singlet-Triplet Qubit. Physical Review Letters 103, 160503 (2009). 8 Foletti, S., Bluhm, H., Mahalu, D., Umansky, V. & Ya- coby, A. Universal quantum control of two-electron spin quantum bits using dynamic nuclear polarization. Nature Physics 5, 903 -- 908 (2009). 9 Shulman, M. D. et al. Demonstration of Entanglement of Electrostatically Coupled Singlet-Triplet Qubits. Science 336, 202 -- 205 (2012). 10 Greilich, A. et al. Mode Locking of Electron Spin Co- herences in Singly Charged Quantum Dots. Science 313, 341 -- 345 (2006). 11 Cywinski, L., Witzel, W. M. & Das Sarma, S. Pure quan- tum dephasing of a solid-state electron spin qubit in a large nuclear spin bath coupled by long-range hyperfine- mediated interactions. Physical Review B 79, 245314 (2009). 12 Cywinski, L., Witzel, W. M. & Das Sarma, S. Electron Spin Dephasing due to Hyperfine Interactions with a Nu- clear Spin Bath. Physical Review Letters 102, 057601 (2009). 13 Witzel, W. M. & Das Sarma, S. Quantum theory for elec- tron spin decoherence induced by nuclear spin dynamics in semiconductor quantum computer architectures: Spectral diffusion of localized electron spins in the nuclear solid- state environment. Physical Review B 74, 035322 (2006). 14 Yao, W., Liu, R.-B. & Sham, L. J. Theory of electron spin decoherence by interacting nuclear spins in a quantum dot. Physical Review B 74, 195301 (2006). 15 Bluhm, H. et al. Dephasing time of GaAs electron-spin qubits coupled to a nuclear bath exceeding 200µs. Nature Physics 7, 109 -- 113 (2011). 16 Lee, B., Witzel, W. M. & Das Sarma, S. Universal Pulse Sequence to Minimize Spin Dephasing in the Central Spin Decoherence Problem. Physical Review Letters 100, 160505 (2008). 17 Zwanenburg, F. A. et al. Silicon quantum electronics. Re- views of Modern Physics 85, 961 -- 1019 (2013). 18 Munsch, M. et al. Manipulation of the nuclear spin en- semble in a quantum dot with chirped magnetic resonance pulses. Nature Nanotechnology 9, 671 -- 675 (2014). 19 Chekhovich, E. A., Hopkinson, M., Skolnick, M. S. & Tar- takovskii, A. I. Suppression of nuclear spin bath fluctua- tions in self-assembled quantum dots induced by inhomo- geneous strain. Nature Communications 6 (2015). 20 Maletinsky, P., Badolato, A. & Imamoglu, A. Dynam- ics of Quantum Dot Nuclear Spin Polarization Controlled by a Single Electron. Physical Review Letters 99, 056804 (2007). 21 Levy, J. Universal Quantum Computation with Spin-1/2 Pairs and Heisenberg Exchange. Physical Review Letters 6 89, 147902 (2002). 22 Medford, J. et al. Scaling of Dynamical Decoupling for Spin Qubits. Physical Review Letters 108, 086802 (2012). 23 Sundfors, R. K. Exchange and Quadrupole Broadening of Nuclear Acoustic Resonance Line Shapes in the III-V Semiconductors. Physical Review 185, 458 -- 472 (1969). 24 Dickel, C., Foletti, S., Umansky, V. & Bluhm, H. Char- acterization of S − T+ Transition Dynamics via Correla- tion Measurements. arXiv:1412.4551 [cond-mat] (2014). ArXiv: 1412.4551. 25 Hester, R. K., Sher, A., Soest, J. F. & Weisz, G. Nuclear- magnetic-resonance detection of charge defects in gallium arsenide. Physical Review B 10, 4262 -- 4273 (1974). 26 Kalevich, V. K. & Korenev, V. L. Electron g-factor anisotropy in asymmetric GaAs/AlGaAs quantum well. JETP Lett 57, 571 (1993). 27 Nefyodov, Y. A., Shchepetilnikov, A. V., Kukushkin, I. V., Dietsche, W. & Schmult, S. Electron g-factor anisotropy in GaAsAl1−xGaxAs quantum wells of different symmetry. Physical Review B 84, 233302 (2011). 28 Snelling, M. J. et al. Magnetic g factor of electrons in GaAsAl1−xGaxAs quantum wells. Physical Review B 44, 11345 -- 11352 (1991). 29 Witzel, W. M. & Das Sarma, S. Wavefunction considera- tions for the central spin decoherence problem in a nuclear spin bath. Physical Review B 77, 165319 (2008). 30 Nadj-Perge, S. et al. Spectroscopy of Spin-Orbit Quantum Bits in Indium Antimonide Nanowires. Physical Review Letters 108, 166801 (2012). 

1611.08395

1

1611

Robust topological insulator surface state in MBE grown (Bi_{1-x}Sb_x)_2Se_3

(Bi1-xSbx)2Se3 thin films have been prepared using molecular beam epitaxy (MBE). We demonstrate the angle-resolved photoemission spectroscopy (ARPES) and transport evidence for the existence of strong and robust topological surface states in this ternary system. Large tunability in transport properties by varying the Sb doping level has also been observed, where insulating phase could be achieved at x=0.5. Our results reveal the potential of this system for the study of tunable topological insulator and metal-insulator transition based device physics.

cond-mat.mes-hall

cond-mat

Robust topological insulator surface state in MBE grown (Bi1-xSbx)2Se3 Y. Hung Liu1,‡, , C. Wei Chong1,‡,* , W. Chuan Chen2, J. C. A. Huang1,3,4,*, C.-Maw Cheng2, K.-Ding Tsuei2, Z. Li5, H. Qiu5, V.V. Marchenkov6 1Department of Physics, National Cheng Kung University, Tainan 70101, Taiwan. 2National Synchrotron Radiation Research Center, Hsinchu 300, Taiwan 3Advanced Optoelectronic Technology Center (AOTC), National Cheng Kung University Tainan 70101, Taiwan. 4Taiwan Consortium of Emergent Crystalline Materials (TCECM), Ministry of Science and Technology, Taipei 10622, Taiwan. 5School of Electronic Science and Applied Physics, HeFei University of Technology, Hefei, Anhui 230009, China. 6M.N. Miheev Institute of Metal Physics, Ekaterinburg 620137, Russia. *Correspondence should be addressed to C. Wei Chong ([email protected]) or J. C. Andrew Huang ([email protected]) 1 KEYWORDS. Topological surface states, (Bi1-xSbx)2Se3, molecular beam epitaxy, tunability ABSTRACT. (Bi1-xSbx)2Se3 thin films have been prepared using molecular beam epitaxy (MBE). We demonstrate the angle-resolved photoemission spectroscopy (ARPES) and transport evidence for the existence of strong and robust topological surface states in this ternary system. Large tunability in transport properties by varying the Sb doping level has also been observed, where insulating phase could be achieved at x0.5. Our results reveal the potential of this system for the study of tunable topological insulator and metal-insulator transition based device physics. 2 INTRODUCTION During the past decade, three dimensional topological insulators (TIs) have been studied extensively in both the fundamental and technological aspect. Their metallic surface states originated from strong spin-orbit coupling (SOC) and band inversion that are topological protected by time reversal symmetry.1, 2, 3, 4, 5 One of the remarkable feature is the spin-momentum locked (SML) Dirac-like surface band that enables electrically controlled spin polarization in TIs channel.6, 7, 8, 9, 10 These unique surface states make them the promising candidates for future high speed/low power electronics and spintronic devices. Nevertheless, utilization of the SML for device applications remains a challenge since the surface carrier conductions are always overwhelmed by the contribution from the bulk carriers which are non-topological protected.11, 12, 13, 14 Unique method for extraction of the surface contribution is a necessary step for the development of TI-based device applications. Utilizing the band structure engineering, (Bi1-xSbx)2Te3 ternary compound has been proven a successful approach in achieving the ideal TI with truly insulating bulk.15 With reducing the bulk carrier density by over two orders of magnitude, a clear ambipolar gating effect in (BixSb1–x)2Te3 nanoplate had been demonstrated16 However, no topological phase transition could be observed in this system, where the topological surface states (TSS) are shown to exist over the entire composition range of (Bi1-xSbx)2Te3.15 Recent attention has been paid to the issues regarding the phase transition between TI and non-topological metal or band insulator (BI) owing to its exotic quantum phenomena and the 3 versatility for device fabrication.17,18,19,20 To date, detailed studies in both experimental and theoretical works mostly focused on (Bi1-xInx)2Se3 in which Brahlek et al. demonstrated the transition from topological metal to a band insulator at x0.25.18 Nevertheless, the system became non-topological metal even at low doping level as low as x 0.03-0.07.18 It is desired to search for a system where transition between ideal TI and BI can be obtained, where large changes in the surface transport properties could be expected. (Bi1-xSbx)2Se3 is another emerged candidate that exhibits TI surface states even at large x, where the critical concentration xc 0.78-0.83, as predicted theoretically.21 Bi2Se3 is a topological insulator with a bulk band gap of 0.3 eV, while Sb2Se3 is a trivial insulator with a band gap around 1.3 eV. Both the band gap and SOC strength can be modified when Bi2Se3 is doped with Sb, leading to a topological phase transition.21,22 Experimentally, Zhang et al. demonstrated the great reduction of bulk carrier density in MBE grown (Bi1-xSbx)2Se3 at x=0.3, where the Dirac-cone like SS showed similar structure and dispersion as undoped Bi2Se3 as evidenced from angle-resolved photoemission spectroscopy (ARPES).23 Via transport measurement, Zhang et al. presented the metal-insulator transition (MIT) at x0.8 that was suggested to be attributed to topological phase transition.22 Another group, instead of using MBE, Lee et al. reported the MIT in (Bi1-xSbx)2Se3 nanosheet at x0.22 that prepared by van der Waal epitaxy method.24 Owing to the importance of this system for the study of TI phase transition, systematic studies on the material growth and characterization, including ARPES and transport evidence for the existence 4 of robust TSS are highly desirable. Here, we fabricated (Bi1-xSbx)2Se3 TI films using molecular beam epitaxy. Enhanced surface states transport was observed via high field Hall effect and weak antilocalization measurement when increasing the Sb doping level. On the other hand, temperature-dependent sheet resistance, Rs demonstrated the metal-insulator transition for the sample with x=0.5, where the Rs (at 300 K) as large as 18 k about six-time larger than that of x=0.32. The transition occurred around 100 K that is believed due to the competition between the insulating (BI phase) and metallic phase (TI phase). More importantly, ARPES revealed the signature of strong TSS up to x=0.32. Our results reveal the strong potential of this ternary (Bi1-xSbx)2Se3 TI for tunable topological insulator and metal-insulator transition device applications, where wide range of transport properties could be achieved. EXPERIMENTAL METHODS A series of (Bi1-xSbx)2Se3 thin film were prepared using molecular beam epitaxy (AdNaNo made MBE-9 system). The doping level, x was controlled by fixing the Bi/Se flux ratio, while Sb flux (in Å /min) was changed by varying the source temperature. The c-plane Al2O3 (0001) substrate was used for the growth of (Bi1-xSbx)2Se3 thin films. High-purity Bi(99.99%), Se(99.999%) and Sb(99.999%) were evaporated using Knudsen cells. All the films used in this work have a thickness of ~25 quintuple-layers (QLs) with Se capping layer 1 nm that was deposited in-situ after the growth of TI layer to avoid environmental 5 contamination.14 In situ reflection high energy electron diffraction (RHEED) was used for monitoring the film quality. Structural characterization was performed by X-ray diffraction (XRD), high-resolution transmission electron microscopic (HRTEM) and Raman spectroscopy, while the surface morphology was obtained by atomic force microscopy (AFM). For the electrical transport measurement, the TI films were patterned into Hall bar geometry using photolithography, allowing the measurement of longitudinal resistance (Rxx) and Hall resistance (Rxy). The angle-resolved photoemission spectroscopy (ARPES) experiment was performed at the National Synchrotron Radiation Research Center in Hsinchu, Taiwan using the U9-CGM spectroscopy beamline. The chamber base pressure was 10-10 torr and all the measurements were performed at 65 K. The samples for the ARPES were prepared using the same method as our previous works.25 RESULTS AND DISCUSSION The composition of the (Bi1-xSbx)2Se3 thin films were determined using TEM-EDS (energy dispersive X-ray spectroscopy) as shown in Figure S1. Figure 1a-d shows the RHEED patterns for samples with various doping level. Streaky patterns were observed for all the samples, indicating the high-quality single crystalline films were formed. Figure 1e displays the HRTEM image of the (Bi0.68Sb0.32)2Se3 film, clearly demonstrating the epitaxial growth of quintuple layers (QLs) of the TI film. For the composition at x=0.5, the streak pattern becomes blur out, revealing the crystal structure has been disturbed. The 6 evolution of the RHEED pattern is consistent with the XRD data as shown in Figure 1f. All the samples exhibit single phase that crystallized in rhombohedral structure (SG R-3m, Z=3) with (00l) c-axis orientation except for the sample with x=0.5. The extra peaks were identified as (020), (130), (240), (061) and (370), corresponding to the diffraction signals from Sb2Se3 that crystallized in orthorhombic phase (Figure S2). Bi2Se3 exhibits the characteristic triangular terraces and steps as revealed by AFM shown in Figure S3a. When Sb is doped into Bi2Se3 (Figure S2b and S2c for x=0.28 and 0.32 respectively), triangular terraces on the surface of the (Bi1-xSbx)2Se3 films become smaller with higher surface roughness. Most remarkable feature is observed at the sample (Bi0.5Sb0.5)2Se3 (Figure S3d), where needle-like morphology appeared in which the surface roughness as high as 4 nm (rms value). As revealed from XRD, it was suggested that the rough morphology originated from the precipitation of Sb2Se3 phase. 7 Figure 1. (a)-(d) RHEED patterns for (Bi1-xSbx)2Se3 thin films; (e) TEM image for x=0.32 (scale bar: 5 nm); (f) X-ray diffraction data for x=0, 0.28, 0.32, 0.5. The dashed lines indicate the peaks corresponding to the (003n) family, indicating the c-axis orientation growth. The triangular symbols reveal the peaks of orthorhombic Sb2Se3. Figure 2a displays the observed Raman spectra for the samples at x=0, 0.28, 0.32 and 0.5. Bi2Se3 is a strong Raman active material that can be identified by their characteristic phonon modes of E1 g, A1 1g, E2 g and A2 1g in the low wave number region.25,26 While the accessible range of our instrument is 50-250 cm-1, the Raman spectra of the Bi2Se3 film (x=0) clearly displays A1 1g, E2 g and A2 1g Raman peaks that corresponding to an in-plane (E2 g) and two out of plane (A1 1g and A2 1g) vibrational modes of the 8 X=0.32X=0X=0.5X=0.2810203040506070(0021)(0018)(0015)(0012)(009)(006)x=0.5x=0.32Intensity (a.u.)2-theta (Bi2Se3x=0.28(003)(a)(b)(c)(d)(e)(f)(Bi0.68Sb0.32)2Se3X=0.32(-Se(1)-Bi-Se(2)-Bi-Se(1)-) lattice.25,26 Interestingly, the vibration modes of E2 g and A1 1g are not changed significantly with increasing the Sb content (Figure 2b), indicating the rhombohedral crystalline structure is largely preserved except for x=0.5. Besides, signature of the substitution of Sb into Bi lattice was observed as the A2 1g mode shifted towards high wave number side (Figure 2b, shown as the dashed line.) Due to the incorporation of Sb, some Bi-Se bonds (175 cm-1) are replaced by Sb-Se bonds (190 cm-1) that resulting in the peak shifting.22,24 For the sample with x=0.5, peaks broadening has been observed, where there are two broad peaks emerged at 175 and 190 cm-1. The result indicated the coexistence of the rhombohedral and orthorhombic phases at x=0.5, which agrees well with XRD observations. Figure 2. (a) Typical Raman spectra of (Bi1-xSbx)2Se3 with different doping levels x=0, 0.28, 0.32, 0.5; (b) Raman shift of A1 1g, A2 1g E2 g vibration modes for various samples. 9 50100150200250 x=0.5x=0.32 Bi2Se3x=0.28 Raman shift(cm-1)Intensity (a.u.)A11gE2gA21g0.00.10.20.30.40.5100150200A11gE2gRaman shift (cm-1)x in (Bi1-xSbx)2Se3A21g(a)(b)We further investigate the doping effect by measuring the electrical transport properties. Hall effect measurement, Rxy vs. B has been presented in Figure 3a in which the nonlinear Hall resistance was observed for the Sb doped samples, indicating the multiple channels contributing to the Hall effect.18 By taking dRxy/dB, the degree of nonlinearity could be revealed as shown in the inset. Very much larger nonlinearity, in comparison to the undoped case, where the Sb doped Bi2Se3 exhibits steeper slope of dRxy/dB vs. B and larger dRxy/dB. This finding demonstrates the enhancement of the surface states conduction18,24,27 when Sb is doped into Bi2Se3 up to x=032. One of the reason is the depletion/suppression of the contribution of the bulk carrier transport. To obtain the total carrier density, we extract the slope at the high field region as shown in the graph of Rxy vs. B (dashed line). As expected, Figure 3b shows that the carrier density, n2D decreases by doping the Sb into Bi2Se3. The total reduction of carrier density, n2D at doping level of x=0.32 was 2×1013 cm-2 which is almost 70% reduction in comparison with the undoped Bi2Se3. Due to the same valency of Sb as Bi atoms, no additional charges would be introduced by substitution of Bi with Sb. However, replacement of Bi by Sb could result in the reduction of unit cell volume that enhances the formation energy of Se vacancies. Thus, the decreasing of n2D is attributed to the suppression of Se vacancies in this (Bi1-xSbx)2Se3 ternary compound.23 On the other hand, sheet resistance Rs increases with increasing the Sb content (Figure 3(c); left axis), following by the degradation of mobility (right axis). For x=0.5, about six-time larger Rs (at 300 K) was observed in comparison to x=0.32 as shown in the Rs-T (Figure 3d). In contrast to the metallic behavior (presented 10 by x=0 and x=0.32), the sample (x=0.5) exhibits insulating behavior where Rs increases with decreasing temperature until 100 K. This observation is in agreement with the XRD, AFM and Raman analysis, where coexistence phases of rhombohedra/orthorhombic were found at x=0.5. Sb2Se3 is a band insulator in nature, the large enhancement of the Rs is believed attributed to the existence of Sb2Se3 that resulted in the metal-insulator transition as shown in the Rs-T. Figure 3. (a) Hall resistance Rxy vs. B for x=0, 0.25, 0.28, 0.32; (b) Carrier density n2D for x=0, 0.25, 0.28 and 0.32 that extracted using Hall effect; (c) Sheet resistance Rs (right axis) and Hall mobility for x=0, 0.25, 0.28 and 0.32. All the measurements performed at 1.9 K; (d) Rs vs. T for x=0, 0.32 and 0.5. 11 0.00.10.20.30.51.01.52.0 x in (Bi1-xSbx)2Se3Rs (k)400600800Mobility (cm2/V.s)0.00.10.20.3123n2D (1013cm-2)x in (Bi1-xSbx)2Se3-8-4048-800-4000400800x=0.32dRxy/dB (/T)B (T)x=0x=0.32x=0.28x=0.25x=0Rxy ()B(T)(a)(b)(c)50100150200250021820Rs (k)T (K)(d)X=0X=0.32X=0.5The presence of the nonlinearity in the Rxy vs. B indicating the enhancement of the surface states transport for the Sb-doped samples. To further confirm this observation, MRxx vs. B has been measured as shown in Figure 4a. The cusp-like MRxx curves were observed (at low field region), revealing weak antilocalization (WAL) effect that originated from the high SOC of the materials. To obtain quantitative analysis of those measurements, we fit the MRxx with Hikami-Larkin-Nagaoka (HLN) equation that illustrates the 2D behavior of the WAL:27 Eq. (1) Here  is expected to be -1/2 for single coherent channel,  is digamma function and B = ħ/4el 2 is characteristic field, l = (D)1/2 is the phase coherent length,  is the phase coherent time and D is diffusion constant. All the curves could be well fitted with this HLN equation indicating the presence of TSS transport in these (Bi1-xSbx)2Se3 system.27 The characteristic coefficient  has been extracted from the fitting where it reveals the number of transporting coherent channels of the 2D system. Consistent with the observation from Rxy vs. B,  increases with increasing the Sb doping level (from 0.4 to 0.6) (Figure 4c), confirming that the reduction of the carrier density in (Bi1-xSbx)2Se3 plays the role in enhancing the transport contribution of topological surface states. 12 BBBBheln212Figure 4. (a) Rxx (B)/Rxx (0) vs. B for (Bi1-xSbx)2Se3 with x=0, 0.25 and 0.32; (b) Fitting of dGs(e2/h) vs. B using HLN equation (solid line indicates fitting curve); (c) alpha extracted from (b) for various samples. All the measurements performed at 1.9 K. Finally, ARPES measurement has been performed to confirm the existence of the topological surface states in this (Bi1-xSbx)2Se3 system. As shown in Figure 5, linear energy-momentum dispersion was evidenced by ARPES in which strong TSS was observed up to x=0.32. The results reveal that the Fermi level is shifted towards bottom of conduction band (arrows is label at Dirac point for clarity) with increasing the Sb content, indicating the depletion of electron that was in agreement with the Hall measurement result. Interestingly, for the case of x=0.5, where the mixed phase of rhombohedral/orthorhombic occurred, the TSS signal still observable although becomes blur or less obvious. This finding has proven the strong and robust TSS in this (Bi1-xSbx)2Se3 ternary system. At last, 13 0.00.10.20.3-0.8-0.6-0.4alphax in (Bi1-xSbx)2Se3-8-40481.01.11.2Rxx(B)/Rxx(0)B (T)x=0.32x=0.25x=0(a)(b)(c)0.00.10.20.3-0.4-0.20.0x=0.32x=0.25x=0dGs (e2/h)B (T)we compare the above finding with the theoretical prediction on the topological phase transition in (Bi1-xSbx)2Se3. According to the first-principles calculations, the critical concentration is xc0.78-0.83 due to the decrease of spin-orbit coupling (SOC) strength.21 Bulk band gap closing will be happened before the formation of the topological trivial band insulator phase. The prediction was supported by the transport data demonstrated by Zhang et al. where xc0.8 was determined.22 However, in our case, the fact could not be concluded due to the high photon energy was used in this experiment, where the precise bulk band gap hardly determined from the ARPES spectrum. Here, we suggest that the enhancement of Rs and metal-insulator transition that observed at x=0.5 should be attributed to mixed phase of rhombohedral/orthorhombic. The evolution of the electronic properties is more appropriately explained based on the scenario, where formation of insulating Sb2Se3 phase has block the conduction electrons and leading to the MIT. Figure 5. (a)-(d) ARPES spectra of (Bi1-xSbx)2Se3 for x=0, 0.25, 0.32 and 0.5 that measured at 65 K. 14 (a)(b)(c)(d)CONCLUSION In summary, we have synthesized high-quality (Bi1-xSbx)2Se3 thin films using MBE. Large reduction of carrier density n2D, where 70% lower than the undoped Bi2Se3 was achieved with increasing Sb content. Enhancement of the TSS transport owing to the Sb doping was revealed via the high field Hall effect and weak antilocalization measurement. Large enhancement of sheet resistance and metal-insulator transition (MIT) occurred at x=0.5 that was attributed to the formation of the insulating Sb2Se3 phase. The topological surface state was further confirmed by ARPES and the linear dispersion was observed up to x=0.5. Our results indicate the robust and strong TSS in this (Bi1-xSbx)2Se3 ternary system where large tunability in transport properties can be achieved, suggesting it could be a promising candidate for exploring the physics and technology of tunable topological insulator and metal-insulator transition. ASSOCIATED CONTENT Supporting Information. EDS measurement of (Bi1-xSbx)2Se3 grown on sapphire. XRD diffraction patterns for (Bi1-xSbx)2Se3 (BSS) and Sb2Se3 films that grown using MBE. AFM images of (Bi1-xSbx)2Se3 with x=0, 0.28, 0,32 and 0.5. This material is available free of charge via the Internet at http://pubs.acs.org. AUTHOR INFORMATION 15 Corresponding Author (C. Wei Chong) [email protected]; (J. C. Andrew Huang) [email protected] Author Contributions The manuscript was written through contributions of all authors. All authors have given approval to the final version of the manuscript. ‡These authors contributed equally. ACKNOWLEDGMENTS This work was partly supported by the National Science Council (Grant No 104-2811-M-006-029 and 104-2811-M-006-031) and by the Russian Foundation for Basic Research (Grant No 14-02-92012). REFERENCES (1) He, L.; Kou, X.; Wang K. L.; Review of 3D topological insulator thin-film growth by molecular beam epitaxy and potential applications. Phys. Status Solidi RRL 2013, 7, No. 1-2, 50-63. (2) Kong, D.; Cui Y.; Opportunities in chemistry and materials science for topological insulators and their nanostructures. Nat. Chem. 2011, 3, 845-849. (3) Hasan M. Z.; Kane, C. L.; Colloquium: Topological insulators. Rev. Mod. Phys. 2010, 82 (4), 3045-3067. (4) Moore, J. E.; The birth of topological insulators. Nature 2010, 464 (7286), 194-198. 16 (5) Qi, X.-L.; Zhang, S.-C.; Topological insulators and superconductors. Rev. Mod. Phys. 2011, 83 (4), 1057-1110. (6) Li, C. H.; van 't Erve, O. M. J.; Robinson, J. T.; Liu, Y.; Li, L; Jonker, B. T. Electrical detection of charge-current-induced spin polarization due to spin-momentum locking in Bi2Se3. Nat. Nanotech. 2014, 9, 218-224. (7) Tang, J.; Chang, L. T.; Kou, X; Murata, K; Choi, E. S; Lang, M; Fan, Y; Jiang, Y; Montazeri, M; Jiang, W; Wang, Y; He, L; Wang, K. L. Electrical detection of spin-polarized surface states conduction in (Bi0.53Sb0.47)2Te3 topological insulator. Nano Lett. 2014, 14, 5423-5429. (8) Dankert, A; Geurs, J.; M. V. Kamalakar, Charpentier, S.; Dash, S. P. Room Temperature Electrical Detection of Spin Polarized Currents in Topological Insulators. Nano Lett. 2015, 15, 7976−7981. (9) Lee, J. S.; Richardella, A.; Hickey, D. R.; Mkhoyan, K. A.; Samarth, N. Mapping the chemical potential dependence of current-induced spin polarization in a topological insulator. Phys. Rev. B 2015, 92, 155312. (10) Yang, F.; Ghatak, S.; Taskin, A. A.; Segawa, K.; Ando, Y.; Shiraishi, M.; Kanai, Y.; Matsumoto, K.; Rosch, A.; Ando, Y. Switching of charge-current-induced spin polarization in the topological insulator BiSbTeSe2. Phys. Rev. B 2016, 94, 075304. 17 (11) Checkelsky, J. G.; Hor, Y. S.; Cava, R. J.; Ong, N. P. Bulk Band Gap and Surface State Conduction Observed in Voltage-Tuned Crystals of the Topological Insulator Bi2Se3. Phys. Rev. Lett. 2009, 106, 196801. (12) Butch, N. P.; Kirshenbaum, K.; Syers, P.; Sushkov, A. B.; Jenkins, G. S.; Drew, H. D.; Paglione, J. Strong surface scattering in ultrahigh-mobility Bi2Se3 topological insulator crystals. Phys. Rev. B 2010, 81, 241301(R). (13) Taskin, A. A.; Ando, Y. Quantum oscillations in a topological insulator Bi1-xSbx. Phys. Rev. B 2009, 80, 085303. (14) Liu, Y. H.; Chong, C. W.; Jheng, J. L.; Huang, S. Y.; Huang, J. C. A.; Li, Z.; Qiu, H.; Huang, S. M.; Marchenkov, V. V. Gate-tunable coherent transport in Se-capped Bi2Se3 grown on amorphous SiO2/Si. Appl. Phys. Lett. 2015, 107, 012106. (15) 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.; Wang, Y. Band structure engineering in (Bi1-xSbx)2Te3 ternary topological insulators. Nat. Commun. 2011, DOI: 10.1038/ncomms1588. (16) 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.; Cui, Y. Ambipolar field effect in the ternary topological insulator (BixSb1-x)2Te3 by composition tuning. Nat. Nanotech. 2011, 6, 705-709. 18 (17) Neupane, M.; Richardella, A.; Sa´nchez-Barriga, J.; Xu, S. Y.; Alidoust, N.; Belopolski, I.; Liu, C.; Bian, G.; Zhang, D.; Marchenko, D.; Varykhalov, A.; Rader, O.; Leandersson, M.; Balasubramanian, T.; Chang, T. R.; Jeng, H. T.; Basak, S.; Lin, H.; Bansil, A.; Samarth, N.; Hasan, M. Z. Observation of quantum-tunnelling-modulated spin texture in ultrathin topological insulator Bi2Se3 films. Nat. Commun. 2014, DOI: 10.1038/ncomms4841. (18) Brahlek, M.; Bansal, N.; Koirala, N.; Xu, S. Y.; Neupane, M.; Liu, C.; Hasan M. Z.; Oh, S. Topological-Metal to Band-Insulator Transition in (Bi1-xInx)2Se3 Thin Films. Phys. Rev. Lett. 2012, 109, 186403. (19) Liao, J.; Ou, Y.; Feng, X.; Yang, S.; Lin, C.; Yang, W.; Wu, K.; He, K.; Ma, X.; Xue, Q. K.; Li, Y. Observation of Anderson Localization in Ultrathin Films of Three-Dimensional Topological Insulators. Phys. Rev. Lett. 2015, 114, 216601. (20) Brahlek, M. J.; Koirala, N.; Liu, J.; Yusufaly, T. I.; Salehi, M.; Han, M.-G.; Zhu, Y.; Vanderbilt, D.; Oh, S. Tunable inverse topological heterostructure utilizing (Bi1-xInx)2Se3 and multichannel weak-antilocalization effect. Phys. Rev. B 2016, 93, 125416. (21) Liu, J.; Vanderbilt, D. Topological phase transitions in (Bi1-xInx)2Se3 and (Bi1-xSbx)2Se3. Phys. Rev. B 2013, 88, 224202. 19 (22) Zhang, C.; Yuan, X.; Wang, K.; Chen, Z. G.; Cao, B.; Wang, W.; Liu, Y.; Zou, J.; Xiu, F. Observations of a Metal–Insulator Transition and Strong Surface States in Bi2–xSbxSe3 Thin Films. Adv. Mater. 2014, 26, 7110-7115. (23) Zhang, Y.; Chang, C. Z.; He, K.; Wang, L. L.; Chen, X.; Jia, J. F.; Ma, X. C.; Xue, Q. K. Doping effects of Sb and Pb in epitaxial topological insulator Bi2Se3 thin films: An in situ angle-resolved photoemission spectroscopy study. Appl. Phys. Lett. 2010, 97, 194102. (24) Lee, C. H.; He, R.; Wang, Z. H.; Qiu, R. L. J.; Kumar, A.; Delaney, C.; Beck, B.; Kidd, T. E.; Chancey, C. C.; Sankaran, R. M.; Gao, X. P. A. Metal–insulator transition in variably doped(Bi1-xSbx)2Se3 nanosheets. Nanoscale 2013, 5, 4337- 4343. (25) Tung, Y.; Chiang, Y. F.; Chong, C. W.; Deng, Z. X.; Chen, Y. C.; Huang, J. C. A.; Cheng, C.-M.; Pi, T.-W.; Tsuei, K.-D.; Li, Z.; Qiu, H. Growth and characterization of molecular beam epitaxy-grown Bi2Te3-xSex topological insulator alloys. J. Appl. Phys. 2016, 119, 055303. (26) Zhang, G.; Qin, H.; Teng, J.; Guo, J.; Guo, Q.; Dai, X.; Fang, Z.; Wu, K. Quintuple-layer epitaxy of thin films of topological insulator Bi2Se3. Appl. Phys. Lett. 2009, 95, 053114. (27) Tian, J.; Chang, C. Z.; Cao, H.; He, K.; Ma, X.; Xue, Q. K.; Chen, Y. P. Quantum and Classical Magnetoresistance in Ambipolar Topological Insulator Transistors with Gate-tunable Bulk and Surface Conduction. Sci. Rep. 2014, DOI: 10.1038/srep04859. 20 Supporting Information Robust topological insulator surface state in MBE grown (Bi1-xSbx)2Se3 Y. Hung Liu1,a, C. Wei Chong1,a,b, W. Chuan Chen2, J. C. A. Huang1,3,4,b, C.-M. Cheng2, K.-D. Tsuei2, Z. Li5, H. Qiu5, V.V. Marchenkov6 1Department of Physics, National Cheng Kung University, Tainan 70101, Taiwan. 2Advanced Optoelectronic Technology Center (AOTC), National Cheng Kung University Tainan 70101, Taiwan. 3Taiwan Consortium of Emergent Crystalline Materials (TCECM), Ministry of Science and Technology, Taipei 10622, Taiwan. 4School of Electronic Science and Applied Physics, HeFei University of Technology, Hefei, Anhui 230009, China. 5M.N. Miheev Institute of Metal Physics, Ekaterinburg 620137, Russia. a Y. Hung Liu, C. Wei Chong contributed equally to this work. b C. Wei Chong or J. C. Andrew Huang: Authors to whom correspondence should be addressed. Electronic mail: [email protected]； [email protected] 21 FIG. S1. Table below shows the composition of the MBE grown (Bi1-xSbx)2Se3 thin films that determined using TEM-EDS. FIG. S2. XRD diffraction patterns for (Bi1-xSbx)2Se3 (BSS) and Sb2Se3 films that grown using MBE. The extra peaks (pointed by the yellow lines) appeared in the BSS with x=0.5, identified as the peaks corresponding to Sb2Se3 orthorhombic phase. [1] [1] Zhai et al., Adv. Mater. 4530–4533, 22 (2010). 22 102030405060x=0.32Intensity (a.u.)2degree)Sb2Se3 x=0.5 Sb flux Å/min Component Sb fraction (%) 0.13 (Bi0.75Sb0.25)2Se3 25% 0.3 (Bi0.72Sb0.28)2Se3 28% 0.45 (Bi0.68Sb0.32)2Se3 32% 0.6 (Bi0.5Sb0.5)2Se3 50% FIG. S3. (a)-(d) AFM images of (Bi1-xSbx)2Se3 with x=0, 0.28, 0,32 and 0.5. Scale bar: 500 nm. Table shows the root mean square roughness determined from AFM. 23 x in (Bi1-xSbx)2Se3Rq (nm)01.00.281.750.321.930.54.29Bi2Se3X=0.28X=0.32X=0.5(a)(b)(c)(d)

1907.08036

2

1907

Optical Absorption and Energy Loss Spectroscopy of Single-Walled Carbon Nanotubes

The recent development of efficient chirality sorting techniques has opened the way to the use of single-walled carbon nanotubes (SWCNTs) in a plethora of nanoelectronic, photovoltaic, and optoelectronic applications. However, to understand the excitation processes undergone by SWCNTs, it is necessary to have highly efficient and accurate computational methods to describe their optical and electronic properties, methods which have until now been unavailable. Here we employ linear combinations of atomic orbitals (LCAOs) to represent the Kohn-Sham (KS) wavefunctions and perform highly efficient time dependent density functional theory (TDDFT) calculations in the frequency domain using our LCAO-TDDFT-$k$-$\omega$ code to model the optical absorbance and energy loss spectra and spatial distribution of the exciton charge densities in SWCNTs. By applying the GLLB-SC derivative discontinuity correction to the KS eigenenergies, we reproduce the measured $E_{11}$ and $E_{22}$ transitions within $\sigma \lesssim 70$ meV and the optical absorbance and electron energy loss spectra semi-quantitatively for a set of fifteen semiconducting and four metallic chirality sorted SWCNTs. Furthermore, our calculated electron hole density difference $\Delta \rho(\textbf{r}, \omega)$ resolves the spatial distribution of the measured excitations in SWCNTs. These results open the path towards the computational design of optimized SWCNT nanoelectronic, photovoltaic, and optoelectronic devices $\textit{in silico}$.

cond-mat.mes-hall

cond-mat

Optical Absorption and Energy Loss Spectroscopy of Single-Walled Carbon Nanotubes Mar´ıa Rosa Preciado-Rivas, Victor Alexander Torres-S´anchez, and Duncan J. Mowbray∗ School of Physical Sciences and Nanotechnology, Yachay Tech University, Urcuqu´ı 100119, Ecuador The recent development of efficient chirality sorting techniques has opened the way to the use of single- walled carbon nanotubes (SWCNTs) in a plethora of nanoelectronic, photovoltaic, and optoelectronic applica- tions. However, to understand the excitation processes undergone by SWCNTs, it is necessary to have highly efficient and accurate computational methods to describe their optical and electronic properties, methods which have until now been unavailable. Here we employ linear combinations of atomic orbitals (LCAOs) to repre- sent the Kohn-Sham (KS) wavefunctions and perform highly efficient time dependent density functional the- ory (TDDFT) calculations in the frequency domain using our LCAO-TDDFT-k-ω code to model the optical absorbance and energy loss spectra and spatial distribution of the exciton charge densities in SWCNTs. By applying the GLLB-SC derivative discontinuity correction to the KS eigenenergies, we reproduce the measured E11 and E22 transitions within σ (cid:46) 70 meV and the optical absorbance and electron energy loss spectra semi- quantitatively for a set of fifteen semiconducting and four metallic chirality sorted SWCNTs. Furthermore, our calculated electron hole density difference ∆ρ(r, ω) resolves the spatial distribution of the measured excitations in SWCNTs. These results open the path towards the computational design of optimized SWCNT nanoelec- tronic, photovoltaic, and optoelectronic devices in silico. I. INTRODUCTION Single-walled carbon nanotubes (SWCNTs) have drawn at- tention in the field of organic electronics due to their unique physical properties, e.g., ballistic conductance, tailorable band gaps, photoluminescence, and high optical absorbance1. These nearly one dimensional (1D) structures come in vari- ous chiralities, which, depending on the way they are rolled up, change their energy band gaps yielding a plethora of dif- ferent absorption and conductive properties2. A variety of different semiconducting SWCNTs can be used to widen the range of wavelengths that can be potentially exploited in pho- tovoltaic applications3,4. SWCNTs exhibit intense absorption peaks with band gaps between 0.9 and 1.5 eV and have high thermal stability5,6. In the case of metallic nanotubes, elec- tronic transport occurs ballistically, meaning they can carry high currents without heating7,8. Furthermore, a clear advan- tage are the recently developed methods for separating SWC- NTs based on their chirality. This provides a straightforward method for tailoring the band gap of the semiconducting layer in a solar cell. For these reasons, SWCNTs have been widely used as additives in OPVs to improve their efficiency by increas- ing the charge carrier mobility of conventional polymers9,10 and dye-sensitized solar cells. In donor-acceptor hybrid cells, SWCNTs have been used to either covalently11 or non- covalently12 graft chromophore molecules, increasing inci- dent photon to current efficiency (IPCE) by about 17%. SWC- NTs can interact with polymers via π-π stacking, porphyrins electrostatically13 to achieve an IPCE of 8.4%, lipid nan- odiscs, and human DNA14. Moreover, in many other pho- tovoltaic devices, metallic carbon nanotubes are used as elec- trodes because of their ballistic conducting properties. Spectroscopy techniques are widely used to characterize SWCNTs. The advantages of optical absorbance (OA), a spe- cific type of spectroscopy, rely on the fact that it is nonde- structive, noninvasive, and simple to perform at room tem- perature and under ambient pressure. For instance, photolu- minescence, absorption, and resonance Raman spectroscopy are widely employed in bulk SWCNTs samples in both research15,16 and industrial laboratories17. This makes spec- troscopy techniques important for the development of OPVs as these methods provide insight into the properties of the ma- terials, whether they are suitable for photovoltaic devices, and how they can be improved. For example, information about the exciton generation process can be gathered through spec- troscopy techniques to make further improvements in the de- sign of OPVs. This is because, in the case of OA, light is most often absorbed when in resonance with the band gap of the material so that the observation of absorption peaks are related to electron transitions. Theoretical calculations of the photoabsorption processes in systems provide insight into not only how excitons are gen- erated, but also other properties, such as the charge distribu- tion, which can help to explain what is observed in experi- mental data. Some of the most commonly used methods are those based on density functional theory (DFT)18. DFT, based on the hypothesis that the electron density distribution com- pletely characterizes the ground state of many electron sys- tems, uses functionals of the spatially-dependent electron den- sity to model the ground state electronic structure and proper- ties at the quantum mechanical level. DFT has made impor- tant contributions in material design projects by combining theory and computational methods to replace traditional, and often expensive, experiments19,20. For instance, DFT calcu- lations have been done to unravel the characteristics of spec- troscopy for SWCNTs with linear response time-dependent density functional theory (TDDFT) used to complement the experimental work made in Ref. 21; additionally, estimates of the internal quantum efficiency of organic photovoltaic de- vices containing polymers, fullerene C60 and SWCNTs have been obtained using DFT22. Optical selection rules for SWCNTs allow light polarized parallel to the nanotube's axis to excite intense transitions be- tween the corresponding subbands in the valence and con- duction bands. For instance, v1→c1 and v2→c2, and so on, correspond to well-defined absorption transitions between van Hove singularities with energies E11 and E22. Metallic SWC- 9 1 0 2 l u J 4 2 ] l l a h - s e m . t a m - d n o c [ 2 v 6 3 0 8 0 . 7 0 9 1 : v i X r a 2 FIG. 1. SWCNT indices (m, n) of circumference vector C ≡ ma1 + na2 where a1 (red) and a2 (blue) are the primitive unit vectors with optical absorbance (red), electron energy loss (blue), and both (mauve) data from Refs. 23 and 24, respectively. Metallic tubes (m − n = 0 mod 3) are marked in grey or dark blue. NTs also have intense absorption peaks associated to transi- tions between van Hove singularities25. Recently, experimen- tal measurements of chirality sorted SWCNTs have provided both optical absorbance23 and electron energy loss spectra24 for a large variety of SWCNTs. For this reason, SWCNTs pro- vide experimentally relevant 1D periodic systems for bench- marking our LCAO-TDDFT-k-ω code21,26 -- 28. In this work, we employ linear combinations of atomic orbitals (LCAO) to represent the Kohn-Sham (KS) wave- functions within time dependent density functional theory (TDDFT) in momentum k and frequency ω space, using our LCAO-TDDFT-k-ω code21,26 -- 28, applying the derivative dis- continuity correction29 to the KS eigenenergies. This method is applied to the set of four metallic and fifteen semicon- ducting SWCNTs mapped in Figure 1, for which optical ab- sorbance and electron energy loss spectroscopy measurements of chirality sorted samples are available from Refs. 23 and 24, respectively. Employing the exciton density method im- plemented within our LCAO-TDDFT-k-ω code26,28, we are able to provide a spatially resolved description of the experi- mentally observed transitions in metallic and semiconducting SWCNTs. The paper is organized as follows. In Sec. II we be- gin by providing a brief theoretical background in Sec. II A to the derivative discontinuity correction obtained from the exchange part of the GLLB-SC functional ∆x, the LCAO- TDDFT-k-ω method in the optical limit (cid:107)q(cid:107) → 0+, and our model for the spatial distribution of the exciton charge den- sity ∆ρ = ρe + ρh, followed by a complete description of the relevant parameters employed in our DFT and LCAO- TDDFT-k-ω calculations in Sec. II B. In Sec. III we compare our results with those obtained from experiments for the nine- teen SWCNTs studied (see Figure 1), including the atomic and electronic structure in Sec. III A, optical absorbance spec- tra in Sec. III B, the E11 and E22 transitions in semiconduct- ing SWCNT in Sec. III C, and electron energy loss spec- troscopy in Sec. III D, followed by our spatially resolved description of the exciton density for the E11 transition in semi-conducting SWCNTs and the Dirac plasmon in metallic SWCNTs in Sec. III E. Finally, concluding remarks are pro- vided in Sec. IV. Atomic units ( = e = me = a0 = 1) have been employed throughout unless otherwise noted. II. METHODOLOGY A. Theoretical Background Modelling the optical absorbance or electron energy loss spectra of a material requires a proper description of its elec- tronic structure, including the electronic band gap Egap. At the Kohn-Sham (KS) level the band gap is approximated by the energy difference between the KS eigenenergies, ∆KS = εN+1 − εN, where N is the number of electrons and we have suppressed dependence on spin and k-point. However, ∆KS often underestimates the experimental band gap by an order of magnitude. Although the exchange and correlation (xc) potential can be tuned to obtain a better agreement of ∆KS with Egap, this can lead to a potential that has unphysical fea- tures, resulting in a poor description of properties other than the band gap30. While both hybrid functionals (HSE0631) 32,33) often provide a suffi- and quasiparticle methods (G0W0 ciently accurate description of the electronic structure, their intractability makes such methods unsuitable for large macro- molecules such as the SWCNTs we will study herein. The derivative discontinuity correction to the exchange functional, ∆x, has been proposed as a first-order ab initio correction to the KS band gap34, where Egap ≈ ∆KS + ∆x. 11,812,88,89,810,816,01,02,03,04,05,06,07,011,012,013,014,015,08,09,010,01,12,13,14,15,16,17,111,112,113,114,115,18,19,110,12,23,24,25,26,27,211,212,213,214,28,29,210,215,23,34,35,36,37,311,312,313,314,38,39,310,34,45,46,47,411,412,413,414,48,49,410,45,56,57,511,512,513,58,59,510,56,67,611,612,613,68,69,610,67,711,712,78,79,710,711,99,910,9a1a2(cid:17)(cid:104)ψN+1 ψ∗ Kuisma et al.29 calculated the exchange part of the deriva- tive discontinuity ∆x from the KS equations by using a mod- ified version of the Gritsenko, van Leeuwen, van Lenthe, and Baerends (GLLB) xc potential35,36. This xc potential exhibits a step structure at the lowest unoccupied orbital when it starts to be occupied. A newer version of this potential is called GLLB-SC, for solid and correlation, and has been shown to yield a better agreement with the experimental band gaps than LDA or GGA for solids37. The derivative discontinuity correction of the ex- change part of the GLLB-SC functional is given by N(cid:88) (cid:16)√ εN+1 − εn − √ √ 8 2 3π2 n=1 ∆x = nψn ρ εN − εn ψN+1(cid:105), (1) where N is the number of electrons, ψn and εn are the nth Kohn-Sham (KS) wavefunction and eigenenergy, respectively, and we have suppressed dependence on the spin and k-point. Major advantages of employing ∆x are both its ab initio nature and its efficiency. Specifically, the calculation of ∆x re- quires a single-point calculation of the electronic structure for the relaxed geometry, and the summation given in Eq. 1. This makes the derivative discontinuity correction an attractive al- ternative to hybrid functionals or quasiparticle methods for accurately describing the electronic structure of large macro- molecules. We model the optical absorption and electron energy loss spectra using the head of the dielectric function, ε(ω), from our LCAO-TDDFT-k-ω code21,26 -- 28, neglecting local crystal field effects. Adding the derivative discontinuity correction of the exchange part of the GLLB-SC functional ∆x from Eq. 1 to the eigenenergies of unoccupied KS states, the dielectric function is then21,32 (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) eq · (cid:104)ψn∇ψm(cid:105) εn − εm + ∆x (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)2 , (cid:88) nm ε(ω) = 1 − 4π Ω f (εm) − f (εn) ω − (εn − εm + ∆x) + iη (2) where we have suppressed spin and k-point dependence, f is the Fermi-Dirac function, η ≈ 25 meV is the Lorentzian broadening of the peaks, and eq is a unit vector in the direc- tion of the light's polarization q → 0+. The matrix elements in Eq. 2 are expressed using the PAW formalism as 3 i T †∇T φa(cid:48) j (cid:105). of the gradient operator within the LCAO basis. For this rea- son, obtaining the dielectric function ε(ω) within the LCAO- TDDFT-k-ω code simply involves the multiplication of matri- ces that have already been calculated, i.e., the KS coefficient matrices ca in with the expectation values of the gradient opera- tor in the PAW-corrected LCAO basis (cid:104) φa This is a very efficient method with a scaling better than O(NM2) where N is the number of KS wavefunctions and M ≥ N is the total number of basis functions used in the LCAO calculation. Moreover, the implicit summation over spin, k-point, and domain in Eq. 2 lends itself trivially to parallelization employing the facilities available within most DFT codes. This degree of parallelizability, as implemented within the LCAO-TDDFT-k-ω code26,28, proved essential for performing distributed memory calculations of SWCNTs with large unit cells (∼ 50 Å) employing the dense k-point sam- 1200 nm−1) required to converge the room tem- pling (∆k (cid:46) 1 perature (η ≈ 25 meV) optical absorbance. It should be noted that employing an LCAO representation to solve for the non-zero dielectric matrix elements, outside the optical limit q → 0+, is unfeasible. This is because the LCAO representation does not lend itself to the efficient cal- culation of Fourier transforms, unlike real-space and plane- wave methods. For this reason, the LCAO-TDDFT-k-ω code's range of applicability is restricted to the optical limit with lo- cal crystal field effects neglected. However, as we shall see, this simplification, when the GLLB-SC derivative discontinu- ity correction is employed, leads to a semi-quantitative de- scription of optical and energy loss spectra for SWCNTs. We model the exciton density as the electron hole den- sity difference, ∆ρ(r, ω) = ρh(r, ω) + ρe(r, ω), where the electron/hole densities are obtained by averaging over the hole/electron position, as implemented in our LCAO-TDDFT- k-ω code26,28. This may be calculated using38,39 η2τm→n2(cid:16)ψm(r)2 − ψn(r)2(cid:17) (ω − (εn − εm + ∆x))2 + η2 (cid:82) ∆ρ(r, ω)d3r = 0. Here τm→n2 are the calculated intensities where we have suppressed spin and k-point dependence and of the m → n transition from Im[ε(ω)] of Eq. 2, so that (5) nm , ∆ρ(r, ω) ≈(cid:88) Im[ε(ω)] = ρh(r, ω)d3r = − ρe(r, ω)d3r. (6) (cid:90) (cid:90) (cid:88) (cid:88) a,a(cid:48) i j (cid:104)ψn∇ψm(cid:105) = ca† in ca(cid:48) jm(cid:104) φa j (cid:105), i T †∇T φa(cid:48) (3) In this way we take into account the relative strength of tran- sitions and their contribution at a given frequency ω. i is the ith smooth basis function centered on atom a where φa and T is the PAW transformation operator i (cid:105) − ϕa T = 1 + (cid:0)ϕa (cid:88) (cid:88) i (cid:105)(cid:1)(cid:104) pa i , (4) a i i and ϕa where ϕa for state i on atom a and pa functions. i are the all-electron and pseudo partial waves i are their smooth PAW projector Methods for calculating Eq. 3 are already implemented within DFT to obtain the forces, i.e., the expectation value B. Computational Details In Figure 1 we show the indices (m, n) of the SWCNTs for which we have performed calculations. Those marked in red are semiconducting SWCNTs with optical absorption spec- tra, in blue are semiconducting SWCNTs with electron energy loss spectra, in mauve are those with both optical absorption and electron energy loss spectra, and those in dark blue are metallic SWCNTs with electron energy loss spectra, as taken from Refs. 23 and 24. Our density functional theory (DFT) calculations were per- formed using the gpaw code40,41, based on the projector- augmented wave (PAW) method40,42 within the atomic simu- lation environment (ASE)43,44. We have used for the SWC- NTs a revised Perdew-Burke-Ernzerhof generalized gradi- ent approximation (GGA) for solids (PBEsol)45 for the ex- change and correlation (xc) functional, and represented the Kohn-Sham (KS) wave functions using a linear combina- tion of atomic orbitals (LCAO)46 with a double-ζ-polarized (DZP) basis set. A room temperature electronic broadening of η = kBT = 25 meV was employed throughout. Both the unit cell and atomic structure for each of the nine- teen SWCNTs studied (see Figure 1) were relaxed until the maximum force was less than 0.05 eV/Å by including 10 Å of vacuum perpendicular to the SWCNT's axis. Periodic bound- ary conditions were employed only in the direction of the SWCNT axis, with the electron density and KS wave func- tions set to zero at the unit cell boundaries perpendicular to the SWCNT's axis. A grid spacing of h ≈ 0.2 Å was employed and the Brillouin-zone was sampled with a k-point density of ∆k (cid:46) 1 30 Å−1 along the SWCNT's axis. A Harris calculation was performed for each SWCNT to 1200 nm−1, fixing the increase the k-point density to ∆k (cid:46) 1 electron density throughout the self-consistency cycle. Such a dense k-point density was found to be necessary to con- verge the calculated absorbance spectra at room temperature (η = kBT = 25 meV). In order to improve the description of the electronic gap, we employed the derivative discontinuity correction to the exchange part of the GLLB-SC functional ∆x as provided in Eq. 1, by performing a single-point calculation for the relaxed structures with GLLB-SC. All calculations of the dielectric function ε(ω) and electron hole density different ∆ρ(r, ω) were performed using linear combinations of atomic orbitals (LCAOs) to represent the KS wave functions at the time-dependent density functional the- ory (TDDFT) level in the optical limit (q → 0+) in recipro- cal k-space and the frequency ω domain using our LCAO- TDDFT-k-ω code21,26 -- 28. Here we have employed a room temperature Lorentzian broadening (η = 25 meV) to the peaks and corrected the eigenenergies by the derivative discontinu- ity correction ∆x from Eq. 1 when calculating the dielectric function using Eq. 2. We model the optical absorbance spec- tra using Im[ε(ω)] and the electron energy loss spectra using − Im[ε−1(ω)]32. 4 TABLE I. Relaxed single-walled carbon nanotube (SWCNT) diam- eters d in Å, unit cell lengths L in Å, numbers of atoms Nat per unit cell, and derivative discontinuity corrections ∆x and electronic band gaps Egap in eV. SWCNT (6,4) (9,1) (8,3) (6,5) (7,3) (7,5) (10,2) (9,4) (8,4) (7,6) (10,3) (11,1) (10,8) (9,8) (11,3) (11,5) (12,3) (10,4) (7,4) d (Å) 6.97 7.62 7.84 7.60 7.09 8.31 8.84 9.17 8.40 8.97 9.36 9.16 12.38 11.66 10.13 11.22 10.89 9.91 7.70 L (Å) 18.64 40.81 42.12 40.83 38.06 44.68 23.80 49.33 11.34 48.21 50.47 49.35 33.37 63.02 54.63 20.20 6.51 8.91 13.69 Nat 152 364 388 364 316 436 248 532 112 508 556 532 488 868 652 268 84 104 124 ∆x (eV) 0.418 0.417 0.399 0.374 0.365 0.360 0.358 0.339 0.329 0.315 0.278 0.269 0.240 0.246 0.310 -- -- -- -- Egap (eV) 1.475 1.476 1.414 1.317 1.281 1.273 1.269 1.199 1.158 1.113 0.983 0.950 0.849 0.869 1.098 -- -- -- -- tween 0.8 and 1.5 eV. In Table I we also provide the derivative discontinuity cor- rection of the exchange part of the GLLB-SC functional29, ∆x, calculated using Eq. 1. These corrections have a size on av- erage of ∼ 28% of the corrected band gap (∆x ≈ 0.28Egap), and are thus proportional to both the corrected band gap en- ergy Egap and the KS band gap ∆KS. More specifically, for the SWCNTs considered herein, ∆x ranges from 0.24 to 0.42 eV, as shown in Table I. This implies ∆x will provide a qualitative correction to both the onset and intensities of the calculated spectra. As we will see in the following sections, this cor- rection is essential for providing both a semi-quantitative and qualitative description of optical absorption and electron en- ergy loss spectra in the q → 0+ limit. III. RESULTS & DISCUSSION B. Optical Absorption Spectra A. Atomic and Electronic Structure We will begin our analysis of the SWCNTs shown schemat- ically in Figure 1 by considering their atomic and electronic structure. As shown in Table I, this is a rather diverse selec- tion of SWCNTs, with diameters ranging from 6.97 to 12.38 Å, lengths from 6.51 to 63.02 Å, and from 84 to 868 atoms per unit cell. Moreover, they exhibit different electronic prop- erties, with four of them being metallic SWCNTs and the re- maining fifteen semiconducting SWCNTs with band gaps be- In Figure 2, we directly compare the optical absorption spectra calculated with our LCAO-TDDFT-k-ω code to the experimental data provided in Ref. 23. The experimental data was normalized, that is, the highest value was set to 1.5 in ar- bitrary units. Likewise, we normalized the maximum of the calculated spectra to 1 in the same arbitrary units. In each of the optical absorption spectra, the first peak cor- responds to the first excitation associated to van Hove singu- larities, i.e., the E11 transition. The second highest peak corre- sponds to the second excitation of this kind, E22. We were able 5 that the energies of the E22 transitions are uncorrelated to those of the E11 transitions. The energies of these two tran- sitions are neither separated by the same amount nor is one proportional to the other. In some cases they are closer than in others. Although the E11 and the E22 transitions are uncorre- lated, the spectra calculated with the LCAO-TDDFT-k-ω code was able to reproduce semi-quantitatively the values of both these transitions energies. The optical absorption spectra shown in Figure 2 corre- sponds to the first 12 SWCNTs listed in Table I, which vary widely in length and number of atoms per unit cell. We can observe that the spectra calculated using the LCAO-TDDFT- k-ω code reproduce the features of the experimental data of these very different SWCNTs, showing that it is surprisingly robust when calculating the optical absorbance of carbon- based 1D nanostructures. In order to calculate the optical absorption spectra of SWC- NTs with large unit cells, it was also necessary to implement both domain decomposition of the real space grids and paral- lelization with respect to k-points. This type of parallelization proved essential for allowing us to perform distributed mem- ory calculations with limited computational resources. FIG. 2. Comparison of LCAO-TDDFT-k-ω calculated (solid lines) and measured (filled regions, Ref. 23) optical absorbance Im[ε(ω)] spectra along the SWCNT axis in nm (upper axis) and eV (lower axis) for chirality sorted (6,4), (9,1), (8,3), (6,5), (7,3), (7,5), (10,2), (9,4), (8,4), (7,6), (10,3), and (11,1) SWCNTs shown in Figure 1. to resolve in some cases a peak between these two, such as in the (9,4), (9,1), (8,3), (7,5) and (9,4). We can also observe excitations higher in energy than the E22, which are typically red shifted by about 0.3 eV relative to the experimental peaks. From hereon, we shall restrict our discussion to the E11 and the E22 excitations. All spectra were calculated using the derivative discontinu- ity correction of the exchange part of the GLLB-SC functional ∆x from Eq. 1. As seen in Eq. 2, this not only shifts the KS eigenenergies, yielding a good description of the first excita- tion energy E11 or the band gap Egap, but also changes the intensities of all the transitions. For all the SWCNTs studied herein, we obtained a semi- quantitative agreement in the description of the relative inten- sities and positions of the E11 and E22 transitions. We find that the E11 transition is more intense than the E22, except in the case of the (9,1) SWCNT. The first transition being more intense than the second transition is a feature also observed in the experimental data. When comparing the spectra of all the SWCNTs, we find C. E11 and E22 Transitions In Figure 3(a) we directly compare the measured E11 tran- sition energies from optical absorption and electron loss spec- troscopy of Refs. 23 and 24, respectively, with our LCAO- TDDFT-k-ω calculated values. We also compare the values obtained with the derivative discontinuity correction of the ex- change part of the GLLB-SC functional ∆x and without this correction, that is, using only the PBEsol xc functional. The PBEsol functional yields an estimation of the band gap with an average error of
≈ −0.30 ± 0.02 eV. Although it re- produces the trend better than in the case of GLLB-SC func- tional, the band gap is always significantly underestimated. When we add the derivative discontinuity correction of the GLLB-SC functional, the average error is
≈ 0 ± 0.07 eV, that is, much smaller than when not adding the correction and well within the expected 0.1 eV accuracy of DFT calculations. Nevertheless, the standard deviation is somewhat larger. This shows that it is important to use the derivative discontinuity correction of the GLLB-SC functional to properly describe the electronic structure and have a better agreement with the experimentally measured spectra onset for SWCNTs. In Figure 3(b), we directly compare the measured E22 transition energies from optical absorption spectroscopy23 with our LCAO-TDDFT-k-ω calculated values including the derivative discontinuity correction of the GLLB-SC func- tional ∆x. Here we obtain a similar agreement to that for the E11 transition, with an average error of
≈ 26± 33 meV. This is again well within the expected 0.1 eV accuracy of DFT cal- culations. It is important to note that the E11 and E22 tran- sition energies almost completely uncorrelated, as shown in Figure 3, so that the near quantitative agreement we obtain is rather independent and systematic. Based on the results of this subsection and the previous one, 40050060070080010001400Wavelength (nm)(11,1)(10,3)(7,6)(8,4)(9,4)(10,2)(7,5)(7,3)(6,5)(8,3)(9,1)(6,4)11.522.533.5Energy (eV)(11,1)(10,3)(7,6)(8,4)(9,4)(10,2)(7,5)(7,3)(6,5)(8,3)(9,1)(6,4)6 FIG. 3. Theoretical versus experimental (a) E11 and (b) E22 transition energies in eV from LCAO-TDDFT-k-ω including (GLLB-SC29, filled squares) and neglecting (PBEsol45, open circles) the derivative discontinuity correction ∆x and from optical absorbance and electron energy loss measurements of Refs. 23 and 24, respectively, for 15 different SWCNT chiralities. The average errors for GLLB-SC (a) E11 (
≈ 0± 70 meV) and (b) E22 (
≈ 20 ± 33 meV) transitions and for PBEsol (a) E11 (
≈ −300 ± 3 meV) transitions are shown as grey regions. Red lines are provided to guide the eye. we have shown that the LCAO-TDDFT-k-ω code can repro- duce with great accuracy the uncorrelated E11 and E22 tran- sitions energies for SWCNTs. Moreover, the fact that we are obtaining such a good agreement suggests that we are consid- ering almost all the processes that are taking place during the optical absorption. This also suggests that transitions that in- clude charge transfer, which are not described by our method, do not occur in these SWCNTs. D. Electron Energy Loss Spectroscopy So far we have considered the optical absorption spectra calculated using the LCAO-TDDFT-k-ω code, defined as the imaginary part of the dielectric function, Im[ε(ω)], for semi- conducting SWCNTs. Now we will consider the electron en- ergy loss spectroscopy of both metallic and semiconducting SWCNTs, provided in Ref. 24. In so doing we are able to also assess the accuracy of the real part of the dielectric function, Re[ε(ω)]. This is because the electron energy loss spectra is the negative of the imaginary part of the inverse of the dielec- tric function, − Im[ε−1(ω)], i.e., Re[ε(ω)]2+Im[ε(ω)]2 . In this way, we are further assessing the robustness of the LCAO-TDDFT- k-ω code by considering another of its outputs. Furthermore, the comparison will be done with respect to measured spectra that correspond to different experiments than those used in the previous sections. Im[ε(ω)] In Figure 4(a) we compare the electron energy loss spec- tra of semiconducting SWCNTs measured in Ref. 24 with our calculations using the LCAO-TDDFT-k-ω code. We find for all five semiconducting SWCNTs that the first and sec- ond peaks are somewhat blue-shifted with respect to the mea- sured spectra by about 0.2 and 0.4 eV on average, respectively. These peaks are assigned the E11 and E22 interband transi- tions. Above these two peaks in energy there is a trough and one, two or three intermediate peaks before a broader and last peak in the measured spectra. The third of all the peaks, that is, the first of the intermediate peaks or the one right after the trough, is always red-shifted with respect to the measured spectra by about 0.37 eV on average. These intermediate peaks corre- spond to the E33, E44 and E55 interband transitions which can be easily identified in the spectra of the (10,8) and the (9,8) SWCNTs. The broader and higher energy peak is blue-shifted by about 0.29 eV on average, and is the well-known π plasmon of SWCNTs47 -- 49. The spectra have the same behavior in gen- eral up to an energy shift, but in the spectra of the (8,4) and the (8,3) SWCNTs the peaks are closer together and harder to recognize. As in the experimental results, the spectra present a monotonic downshift as the diameter of the SWCNT in- creases. Turning to an analysis of the metallic SWCNTs' electron energy loss spectroscopy in Figure 4(b), we observe a strong peak at around 1 eV (marked with an *) that also matches what is observed in the experimental spectra. These peaks, which are present only in the spectra of metallic SWCNTs, correspond to free charge carrier Drude plasmons ωP. In other words, an intraband excitation that causes quantized collective (8,3)(9,8)(6,5)(10,3)(7,6)(9,4)(10,2)(9,1)(8,4)(11,3)(6,4)(11,1)(7,3)(10.8)(7,5)(8,4)(11,3)(11,1)(7,3)(10,8)(7,5)(8,3)(9,8)(6,5)(6,4)(10,3)(7,6)(9,4)(10,2)(9,1)0.60.70.80.911.11.21.31.41.5Experimental E11(eV)0.60.70.80.911.11.21.31.41.5Theoretical E11(eV)GLLB-SCPBEsol(a)(6,4)(9,1)(8,3)(7,3)(10,2)(10,3)(6,5)(7,5)(9,4)(8,4)(7,6)(11,1)1.61.71.81.922.12.22.32.42.5Experimental E22(eV)1.61.71.81.922.12.22.32.42.5Theoretical E22(eV)(b)7 FIG. 5. Exciton density difference ∆ρ(r, ω) = ρe(r, ω) + ρh(r, ω) positive (red) and negative (blue) isosurfaces for the (a,b) E11 and (c,d) E22 transitions of the (a,c) (6,4) (E11 ≈ 1.53, E22 ≈ 2.20 eV) and (b,d) (8,4) (E11 ≈ 1.16, E22 ≈ 2.20 eV) semiconducting SWCNTs along the axis (left) and in the plane (right) of the nanotube. E. Exciton Density Having demonstrated the reliability of our LCAO-TDDFT- k-ω code for describing both the optical absorption and elec- tron energy loss spectra of SWCNTs in the previous sections, we may now use the electron hole density difference ∆ρ(r, ω), calculated from Eq. 5, to model the spatial distribution of the most relevant bright excitons. In so doing, we may probe the spatially resolved optical absorption and electron energy loss spectroscopy of SWCNTS, and their underlying physical makeup. In Figure 5 we show the spatially resolved electron-hole density difference ∆ρ(r, ω) of the E11 and E22 transitions for two semiconducting SWCNTs with quite different transition energies. Regions of negative charge (blue) correspond to the excited electron, whereas regions of positive charge (red) cor- respond to the hole. For both the E11 and E22 transitions we find that the positive (or hole) density is distributed in a con- Comparison of LCAO-TDDFT-k-ω calculated (solid FIG. 4. lines) and measured (filled regions, Ref. 24) electron energy loss − Im[ε−1(ω)] spectra along the SWCNT axis in eV for chirality sorted (a) semiconducting (10,8), (9,8), (11,3), (8,4), and (8,3) and (b) metallic (11,5), (12,3), (10,4) and (7,4) SWCNTs shown in Fig- ure 1, with Drude intraband plasmons ωP (*) marked. oscillations of electrons. Going higher in energy, there is a trough and three well- known peaks. The first two of these peaks correspond to M11 (E11) transitions. The splitting of the transition into two peaks is probably caused by the trigonal wrapping effect24. The third peak corresponds to the M22 transition. All of these transitions can be compared to peaks in the experimental data, although they are red-shifted by about 0.15 eV. Finally, the broader and higher energy peak is again blue-shifted by about 0.28 eV, and is the well-known π plasmon of SWCNTs47 -- 49. We also observe intense peaks in the spectra above 5 eV of the (11,5) and (12,3) SWCNTs, which could be related to splitting of the π plasmon. In summary, we obtained an accurate energy for the plas- monic transition and also a qualitative description of the two peaks related to the M11 transition and the peak related to the M22 up to a red-shift. In this way we have assessed both the real and imaginary part of the dielectric function calculated using the LCAO-TDDFT-k-ω code and found it provides a ro- bust and efficient method for modelling electron energy loss spectra. (8,3)(8,4)(11,3)(9,8)(10,8)(a) Semiconducting SWCNT EELS123456Energy (eV)(7,4)(10,4)(12,3)(11,5)(b) Metallic SWCNT EELS****(a) (6,4) SWCNT E11 = 1.53 eV(b) (8,4) SWCNT E11 = 1.16 eV(c) (6,4) SWCNT E22 = 2.20 eV(d) (8,4) SWCNT E22 = 2.09 eV8 In Figure 6 we show the spatially resolved electron-hole density difference ∆ρ(r, ω) of the Drude intraband plasmon ωP for three different metallic SWCNTs. Again, regions of negative charge (blue) correspond to the excited electron, whereas regions of positive charge (red) correspond to the hole. In contrast to the E11 and E22 transitions of the semi- conducting SWCNTs (cf. Figure 5), we find for each of the three metallic nanotubes the plasmon excitation has nega- tive (or electron) density distributed in a continuous spiral around the nanotube, whereas the positive (or hole) density follows the same pattern but is discontinuous. This suggests the electron density corresponds to bonding orbitals wrapping the SWCNT, whereas the hole density corresponds to anti- bonding orbitals localized on individual C -- C bonds. As was the case for the semiconducting E11 and E22 tran- sitions (cf. Figure 5), the plots in Figure 6 in the SWCNT's plane show the electron hole density difference isosurfaces for the Drude intraband plasmons ωP are also composed of π- orbitals, with a nodal plane on the SWCNT's surface. These results clearly demonstrate that the Drude plasmon is also composed of π → π transitions, as expected. However, the nearly continuous excited electron's density seems to be a property of this metallic plasmon. Overall, these results provide us with added insight into the physical makeup of the experimentally observed peaks in op- tical absorbance and electron energy loss spectra. This in- formation has the potential of further optimizing a SWCNT's overlap with donor molecules when designing organic photo- voltaic cells. IV. CONCLUSIONS We have performed an in-depth analysis of the optical ab- sorption and electron energy loss spectra of SWCNTs, 1D structures with properties determined by their (m, n) chiral in- dices. We have considered a variety of SWCNTs with differ- ent indices and found that our theoretical optical absorption spectra, given by the imaginary part of the dielectric function, agree semi-quantitatively with the experimental data when the derivative discontinuity correction of the GLLB-SC func- tional ∆x is employed in our LCAO-TDDFT-k-ω code. We also see that both the calculated E11 and E22 transition en- ergies have an average error much smaller than the expected accuracy of DFT calculations, with the E22 transition energies having an even better agreement than the E11. This result is rather surprising since the E11 and E22 energies are uncorre- lated. Furthermore, we assessed the real part of the dielec- tric function by comparing our calculated electron energy loss spectra, given by minus the imaginary part of the inverse di- electric function, to experimental data. We were able to re- produce the qualitative behavior of the spectra and to obtain an accurate energy for the Drude intraband plasmon peak ωP in metallic SWCNTs. Finally, we have employed the electron hole density difference ∆ρ(r, ω) to model the spatial distribu- tion of the excitons. We find, as expected, the E11 and E22 transitions in semiconducting SWCNTs and the Drude intra- band plasmon ωP in metallic SWCNTs all involve π levels FIG. 6. Exciton density difference ∆ρ(r, ω) = ρe(r, ω) + ρh(r, ω) pos- itive (red) and negative (blue) isosurfaces for the plasmon excitations ωP of the (a) (12,3) (ωP ≈ 0.84 eV), (b) (10,4) (ωP ≈ 0.89 eV), and (c) (7,4) (ωP ≈ 1.05 eV) metallic SWCNTs along the axis (left) and in the plane (right) of the nanotube. tinuous spiral around the nanotube, whereas the negative (or electron) density follows the same pattern but is discontinu- ous. This suggests the hole density corresponds to bonding orbitals wrapping the SWCNT, whereas the electron density corresponds to anti-bonding orbitals localized on individual C -- C bonds. In fact, the plots in Figure 5 in the SWCNT's plane show the electron hole density difference isosurfaces are composed of π-orbitals, with a nodal plane on the SWCNT's surface. These results clearly demonstrate that both the E11 and E22 transitions are indeed π → π transitions, as expected. It is interesting to note that the spatial distribution of the E11 transition of the (6,4) SWCNT more closely resembles that of the E22 transition of the (8,4) SWCNT, whereas the E22 tran- sition of the (6,4) SWCNT more closely resembles that of the E11 transition of the (8,4) SWCNT. This is evident from both the direction of the wrapping of the positive hole distributions around the nanotube axis and the slice in the nanotube plane. This clearly suggests the spatial distribution of the individual excitonic peaks is highly dependent on the SWCNT's chiral- ity, and not simply a function of the peak's energy. (a) (12,3) SWCNT ωP = 0.84 eV(b) (10,4) SWCNT ωP = 0.89 eV(c) (7,4) SWCNT ωP = 1.05 eVwhich wrap around the SWCNTs. Altogether, these results demonstrate the surprising reliability and efficiency of a sim- plified LCAO-based TDDFT calculation in the optical limit for describing the optical absorbance and electron energy loss spectra of carbon-based macromolecules. This work blazes the trail towards the computational design of complex carbon- 9 based macromolecular organic photovoltaic systems in silico. ACKNOWLEDGMENTS This work employed the Imbabura cluster of Yachay Tech University, which was purchased under contract No. 2017-024 (SIE-UITEY-007-2017). ∗ [email protected] 1 R. H. Baughman, A. A. Zakhidov, and W. A. de Heer, "Carbon nanotubes -- the route toward applications," Science 297, 787 -- 792 (2002). 2 M. S. Dresselhaus, G. Dresselhaus, and P. Avouris, eds., Car- bon Nanotubes: Synthesis, Structure, Properties, and Applica- tions (Springer, Berlin, 2001). 3 C. D. Spataru, S. Ismail-Beigi, L. X. Benedict, and S. G. Louie, "Excitonic effects and optical spectra of single-walled carbon nan- otubes," Phys. Rev. Lett. 92, 077402 (2004). 4 H. Kataura, Y. Kumazawa, Y. Maniwa, I. Umezu, S. Suzuki, Y. Ohtsuka, and Y. Achiba, "Optical properties of single-wall car- bon nanotubes," Syn. Metals 103, 2555 -- 2558 (1999). 5 T. Yamamoto, K. Watanabe, and E. R. Hern´andez, "Mechani- cal properties, thermal stability and heat transport in carbon nan- otubes," in Carbon Nanotubes: Advanced Topics in the Synthe- sis, Structure, Properties and Applications, edited by A. Jorio, G. Dresselhaus, and M. S. Dresselhaus (Springer Berlin Heidel- berg, Berlin, Heidelberg, 2008) pp. 165 -- 195. 6 K. M. Liew, C. H. Wong, X. Q. He, and M. J. Tan, "Thermal stability of single and multi-walled carbon nanotubes," Phys. Rev. B 71, 075424 (2005). 7 W. Liang, M. Bockrath, D. Bozovic, J. H. Hafner, M. Tinkham, and H. Park, "Fabry-Perot interference in a nanotube electron waveguide," Nature 411, 665 -- 669 (2001). 8 S. Frank, P. Poncharal, Z. Wang, and W. A. De Heer, "Carbon nanotube quantum resistors," Science 280, 1744 -- 1746 (1998). 9 E. Kymakis and G. Amaratunga, "Single-wall carbon nan- otube/conjugated polymer photovoltaic devices," Appl. Phys. Lett. 80, 112 -- 114 (2002). 10 E. Kymakis, E. Koudoumas, I. Franghiadakis, and G. Ama- ratunga, "Post-fabrication annealing effects in polymer-nanotube photovoltaic cells," J. Phys. D Appl. Phys. 39, 1058 (2006). 11 S. Campidelli, B. Ballesteros, A. Filoramo, D. D´ıaz, G. de la Torre, T. Torres, G. A. Rahman, C. Ehli, D. Kiessling, and F. Werner, "Facile decoration of functionalized single-wall carbon nanotubes with phthalocyanines via "click chemistry"," J. Am. Chem. Soc. 130, 11503 -- 11509 (2008). 12 J. Bartelmess, B. Ballesteros, G. de la Torre, D. Kiessling, S. Campidelli, M. Prato, T. Torres, and D. M. Guldi, "Phthalocyanine-pyrene conjugates: a powerful approach toward carbon nanotube solar cells," J. Am. Chem. Soc. 132, 16202 -- 16211 (2010). 13 D. M. Guldi, G. Rahman, M. Prato, N. Jux, S. Qin, and W. Ford, "Single-wall carbon nanotubes as integrative building blocks for solar-energy conversion," Angew. Chem. Int. Ed. 44, 2015 -- 2018 (2005). 14 M.-H. Ham, J. H. Choi, A. A. Boghossian, E. S. Jeng, R. A. Graff, D. A. Heller, A. C. Chang, A. Mattis, T. H. Bayburt, Y. V. Grinkova, A. S. Zeiger, K. J. V. Vliet, E. K. Hobbie, S. G. Sli- gar, C. A. Wraight, and M. S. Strano, "Photoelectrochemical complexes for solar energy conversion that chemically and au- tonomously regenerate," Nat. Chem. 2, 929 -- 936 (2010). 15 C. Zamora-Ledezma, C. Blanc, and E. Anglaret, "Orientational order of single-wall carbon nanotubes in stretch-aligned photolu- minescent composite films," Phys. Rev. B 80, 113407 (2009). 16 F. J. Torres-Canas, C. Blanc, C. Zamora-Ledezma, P. Silva, and E. Anglaret, "Dispersion and individualization of SWNT in surfactant-free suspensions and composites of hydrosoluble poly- mers," J. Phys. Chem. C 119, 703 -- 709 (2014). 17 R. B. Weisman and J. Kono, "Introduction to optical spectroscopy of single-wall carbon nanotubes," in Handbook of Carbon Nano- materials (World Scientific, 2019) pp. 1 -- 43. 18 A. Zangwill, "A half century of density functional theory," Physics Today 68, 34 -- 39 (2015). 19 J. K. Nørskov, T. Bligaard, J. Rossmeisl, and C. H. Christensen, "Towards the computational design of solid catalysts," Nat. Chem. 1, 37 -- 46 (2009). 20 A. Jain, S. P. Ong, G. Hautier, W. Chen, W. D. Richards, S. Dacek, S. Cholia, D. Gunter, D. Skinner, G. Ceder, and K. A. Pers- son, "Commentary: The materials project: A materials genome approach to accelerating materials innovation," APL Materials 1, 011002 (2013). 21 L. N. Glanzmann, D. J. Mowbray, D. G. Figueroa del Valle, F. Scotognella, G. Lanzani, and A. Rubio, "Photoinduced ab- sorption within single-walled carbon nanotube systems," J. Phys. Chem. C 120, 1926 -- 1935 (2015). 22 L. N. Glanzmann and D. J. Mowbray, "Theoretical insight into the internal quantum efficiencies of polymer/C60 and poly- mer/SWNT photovoltaic devices," J. Phys. Chem. C 120, 6336 -- 6343 (2016). 23 X. Wei, T. Tanaka, Y. Yomogida, N. Sato, R. Saito, and H. Kataura, "Experimental determination of excitonic band struc- tures of single-walled carbon nanotubes using circular dichroism spectra," Nat. Comm. 7, 12899 (2016). 24 R. Senga, T. Pichler, and K. Suenaga, "Electron spectroscopy of single quantum objects to directly correlate the local structure to their electronic transport and optical properties," Nano Lett. 16, 3661 -- 3667 (2016). 25 R. Weisman and S. Subramoney, "Carbon nanotubes," Elec- trochem. Soc. Interface 15, 42 -- 46 (2006). 26 K. Lyon, M. R. Preciado-Rivas, V. Despoja, and D. J. Mowbray, "LCAO-TDDFT-k-ω: Spectroscopy in the optical limit," (2019), unpublished. 27 M. R. Preciado-Rivas, D. J. Mowbray, K. Lyon, A. H. Larsen, and B. F. Milne, "Optical excitations of chlorophyll a and b monomers and dimers," (2019), arXiv:1907.09430. 28 The LCAO-TDDFT-k-ω code is available free of charge from gitlab.com/lcao-tddft-k-omega/lcao-tddft-k-omega. 29 M. Kuisma, J. Ojanen, J. Enkovaara, and T. T. Rantala, "Kohn- Sham potential with discontinuity for band gap materials," Phys. Rev. B 82, 115106 (2010). 30 F. Tran and P. Blaha, "Importance of the kinetic energy density for band gap calculations in solids with density functional theory," J. Phys. Chem. A 121, 3318 -- 3325 (2017). 31 J. Heyd, G. E. Scuseria, and M. Ernzerhof, "Hybrid function- als based on a screened Coulomb potential," J. Chem. Phys. 118, 8207 (2003). 32 G. Onida, L. Reining, and A. Rubio, "Electronic excita- tions: Density-functional versus many-body Green's-function ap- proaches," Rev. Mod. Phys. 74, 601 -- 659 (2002). 33 A. Migani, D. J. Mowbray, A. Iacomino, J. Zhao, H. Petek, and A. Rubio, "Level alignment of a prototypical photocatalytic sys- tem: Methanol on TiO2(110)," J. Am. Chem. Soc. 135, 11429 -- 11432 (2013). 34 F. Tran, S. Ehsan, and P. Blaha, "Assessment of the GLLB-SC potential for solid-state properties and attempts for improvement," Phys. Rev. Materials 2, 023802 (2018). 35 O. Gritsenko, R. van Leeuwen, E. van Lenthe, and E. J. Baerends, "Self-consistent approximation to the Kohn-Sham exchange po- tential," Phys. Rev. A 51, 1944 -- 1954 (1995). 36 O. V. Gritsenko, R. van Leeuwen, and E. J. Baerends, "Di- rect approximation of the long- and short-range components of the exchange-correlation Kohn-Sham potential," Int. J. Quantum Chem. 61, 231 -- 243 (1997). 37 I. E. Castelli, T. Olsen, S. Datta, D. D. Landis, S. Dahl, K. S. Thygesen, and K. W. Jacobsen, "Computational screening of per- ovskite metal oxides for optimal solar light capture," Energy Env- iron. Sci. 5, 58145819 (2012). 38 L. N. Glanzmann, D. J. Mowbray, and A. Rubio, "PFO-BPy solu- bilizers for SWNTs: Modelling polymers from oligomers," Phys. Status Solidi B 251, 2407 -- 2412 (2014). 39 D. J. Mowbray and A. Migani, "Optical absorption spectra and excitons of dye-substrate interfaces: Catechol on TiO2(110)," J. Chem. Theory Comput. 12, 2843 (2016). 40 J. J. Mortensen, L. B. Hansen, and K. W. Jacobsen, "Real-space grid implementation of the projector augmented wave method," Phys. Rev. B 71, 035109 (2005). 41 J. Enkovaara, C. Rostgaard, J. J. Mortensen, J. Chen, M. Dułak, L. Ferrighi, J. Gavnholt, C. Glinsvad, V. Haikola, H. A. Hansen, H. H. Kristoffersen, M. Kuisma, A. H. Larsen, L. Lehto- vaara, M. Ljungberg, O. Lopez-Acevedo, P. G. Moses, J. Oja- nen, T. Olsen, V. Petzold, N. A. Romero, J. Stausholm-Møller, 10 M. Strange, G. A. Tritsaris, M. Vanin, M. Walter, B. Hammer, H. Hakkinen, G. K. H. Madsen, R. M. Nieminen, J. K. Nørskov, M. Puska, T. T. Rantala, J. Schiøtz, K. S. Thygesen, and K. W. Jacobsen, "Electronic structure calculations with GPAW: A real- space implementation of the projector augmented-wave method," J. Phys.: Condens. Matter 22, 253202 (2010). 42 P. E. Blochl, "Projector augmented-wave method," Phys. Rev. B 50, 17953 -- 17979 (1994). 43 S. R. Bahn and K. W. Jacobsen, "An object-oriented scripting in- terface to a legacy electronic structure code," Comput. Sci. Eng. 4, 56 -- 66 (2002). 44 A. H. Larsen, J. J. Mortensen, J. Blomqvist, I. E. Castelli, R. Christensen, M. Dułak, J. Friis, M. N. Groves, B. Hammer, C. Hargus, E. D. Hermes, P. C. Jennings, P. B. Jensen, J. Kermode, J. R. Kitchin, E. L. Kolsbjerg, J. Kubal, K. Kaasbjerg, S. Lysgaard, J. B. Maronsson, T. Maxson, T. Olsen, L. Pastewka, A. Peterson, C. Rostgaard, J. Schiøtz, O. Schutt, M. Strange, K. S. Thygesen, T. Vegge, L. Vilhelmsen, M. Walter, Z. Zeng, and K. W. Jacobsen, "The atomic simulation environment -- a python library for work- ing with atoms," J. Phys.: Condens. Matter 29, 273002 (2017). 45 J. P. Perdew, A. Ruzsinszky, G. I. Csonka, O. A. Vydrov, G. E. Scuseria, L. A. Constantin, X. Zhou, and K. Burke, "Restoring the density-gradient expansion for exchange in solids and sur- faces," Phys. Rev. Lett. 100, 136406 (2008). 46 A. H. Larsen, M. Vanin, J. J. Mortensen, K. S. Thygesen, and K. W. Jacobsen, "Localized atomic basis set in the projector aug- mented wave method," Phys. Rev. B 80, 195112 (2009). 47 T. Pichler, M. Knupfer, M. S. Golden, J. Fink, A. Rinzler, and R. E. Smalley, "Localized and delocalized electronic states in single-wall carbon nanotubes," Phys. Rev. Lett. 80, 4729 -- 4732 (1998). 48 C. Kramberger, R. Hambach, C. Giorgetti, M. H. Rummeli, M. Knupfer, J. Fink, B. Buchner, L. Reining, E. Einarsson, S. Maruyama, F. Sottile, K. Hannewald, V. Olevano, A. G. Marinopoulos, and T. Pichler, "Linear plasmon dispersion in single-wall carbon nanotubes and the collective excitation spec- trum of graphene," Phys. Rev. Lett. 100, 196803 (2008). 49 D. J. Mowbray, S. Segui, J. Gervasoni, Z. L. Miskovi´c, and N. R. Arista, "Plasmon excitations on a single-wall carbon nanotube by external charges: Two-dimensional, two-fluid hydrodynamic model," Phys. Rev. B 82, 035405 (2010). 

1603.04214

1

1603

Time evolution of nanoscale systems by finite difference method

Using finite difference method, time evolution of a typical metal molecule metal system is studied by introducing a new method to solve general related Volterra integro differential equation (IDE). Discretization in time domain is applied for one dimentional chain tight binding model in several cases by defining a matrix integro-differential equation (MIDE). Results are compatible with their analytical counterparts and show more accuracy than other numerical methods like Runge Kutta (RK). Charge transport properties in a trans polyacetylene chain are found by studying the time evolution of charge density in it and current voltage diagram is calculated.

cond-mat.mes-hall

cond-mat

a Time-evolution of nanoscale systems by finite dif- ference method Mohammad Nakhaee1,†∗, S Ahmad Ketabi1,‡, M Taher Pakbaz1,¶,M Ali M Keshtan2,⋆, Elham Rahmati1,§ and Zahra Abdous3,∗ 1Damghan University, Damghan, Iran , 2Department of Physics, Iran University of Science and Technology, Narmak, Tehran 16844, Iran , 3Islamic Azad Universiry Tehran Center Branch, Tehran, Iran Abstract. Using finite difference method, time evolution of a typical metal-molecule- metal system is studied by introducing a new method to solve general related Volterra integro−di f f erential equation (IDE). Discretization in time domain is applied for one dimensional chain tight binding model in several cases by defining a matrixintegro− di f f erential equation (MIDE). Results are compatible with their analytical counter- parts and show more accuracy than other numerical methods like Runge Kutta (RK). Charge transport properties in a trans-polyacetylene chain are found by studying the time evolution of charge density in it and current-voltage diagram is calculated. Key words: time evolution, green function, molecular junction, finite difference method. 1 Introduction Recently, Metal-Molecule-Metal (MMM) structures (Figure 1) have attracted scientists. Their vast applications, include electrical, optical, mechanical, etc have led scientists to produce devices with new abilities and have improved efficiencies relative to their pri- mary counterparts [1 -- 5]. Well predicting behaviour of MMM systems requires investi- gating time-evolution of their transport properties. Many efforts have been done to inves- tigate time dependence of transport properties in MMM structures. Generally, Green's function formalism and density functional theory, have extensively applied to study time evolution of MMM structures [6 -- 12]. In the Green's function formalism, transport prop- erties of MMM systems can be deduced by applying Green's function in the energy rep- resentation, G(E), which can be calculated as follows: G(E) = [(E+i0+)I− H−Σ(E)]−1 (1.1) ∗Corresponding author. Email address: [email protected] (M. Nakhaee) http://www.global-sci.com/ Global Science Preprint 2 In which H, Σ(E) and E are Hamiltonian of the molecule and self-energy of the system in energy E, respectively. The Fourier transform of G(E), suggests impulse response (G(t)) as: G(t) = e −iEt ¯h G(E)dE 1 2π¯hZ ∞ −∞ which satisfies the Fourier transform of equation (1.1) as follows: (i¯h ∂ ∂t − H−Σ)G(t) = I δ(t) (1.2) (1.3) Taking the energy dependence into account, the product of Σ and G becomes a convolu- tion in time domain then equation (1.3) can be rewritten as [13]: (i¯h ∂ ∂t − H)G(t)−Z Σ(t−t′)G(t′)dt′ = I δ(t) (1.4) Equation (1.4) is a non-homogeneous Volterra IDE [14] with an intractable and time consuming general solution process. Finite difference method (FDM), is an applicable scheme to solve coupled equations [15] and as will be mentioned in section ??, discretization of a differential equation results some ones. This method has been used to study electronic transport in nanostructures, for instance Khomyakov et. al. [16] have calculated coherent transport of a nano wire by wave functions matching in the boundary zones connecting electrods and the scattering region using FDM. In this article, using FDM, a simple formalism is proposed to solve equation (1.4). In this approach, derivative and integrator operators are defined then equation (1.4) is rewritten in matrix form in the presence of adequate boundary conditions (section ??). Figure 1: (Color online) A schematic figure of a metal-molecule-metal system. 3 In section ?? this approach is applied to calculate the Green's function, G(t) for an infinite 1D chain. Results are calculated in both time-independent and time-dependent Hamiltonian cases and are compared with their analytical solutions. Beside some numer- ical comparisons are made between computational errors of our formalism and Dyson series and RK methods. Finally, this method is used to calculate the charge current in a system composed of a trans-polyacetylene molecule connected to two semi infinite 1D metal electrodes. 2 Method 2.1 Construction of the Finite Difference Scheme to solve the Volterra IDE To solve equation (1.4) generally, consider following Volterra IDE: d dt y(t) = f (t)y(t)+Z t 0 k(t−τ)y(τ)dτ (2.1) In which y(t) and f (t) stand for functions of an arbitrary real parameter t and k(t−τ) represents a convolution between τ and t. In order to perform numerical calculations it is really lucrative to make a discretization scheme for this IDE which allow us to use FDM. The grid used for discretization is a set of points {1,2,··· ,nt} where nt is an integer and shows the number of mesh points in the t domain which may be determined properly based on the fluctuations of the functions. Commonly, yt, ft and kt are nt dimensional vectors which contain all information of functions y(t), f (t) and k(t), respectively. Beside we need two operators; a first order derivative operator and an integrator one; which are represented by Dns nt×nt, respectively. ns returns to ns-point stencil of a point in the grid in FDM formalism; the point itself together with its ns−1 neighbours. Clearly ns must usually choose in such a way that be less than or equal to nt, (ns ≤ nt). Here we try to introduce these two operators properly. The first derivative of a function y(t) respect to the parameter t at a point ti is usually nt×nt and Ins approximated using a ns-point stencil as [17]: y′(ti) ≈ di,iy(ti)+∑is∈neighborsdi,is y(tis ) (2.2) The coefficients di,j of this equation, while i and j can be integer numbers in this set {1,2,··· ,nt}, are well known as Lagrange interpolation coefficients [18] and are used widely in FDM. These coefficients should be exploited to derive Dns nt×nt as a matrix whose elements are zero except those that are di,j. Up to this precision, it is straightforward to define the inverse of this derivative operator as an adequate integrator operator, namely Ins nt×nt. Uniqueness of this integrator operator imposes a boundary condition. Keeping in mind the proper initial value condition, which is R 0 0 y(t)dt = 0, we pursue common procedure in the FDM. For each boundary condition a row and a column are added to 4 integrator matrix [15] so that the final integrator operator with boundary conditions is introduced as: Ins (nt+1)×(nt+1) =(cid:18) Dns In which, the elements of the matrix Bnt×1 are defined as: Bt,1 = δt,1 nt×nt Bnt×1 B† nt×1 0 (cid:19)−1 (2.3) (2.4) In which δt,1 represents the Kronecker delta function. After inversion, we omit added row and column of integrator operator with boundary conditions, Ins (nt+1)×(nt+1), and reshape it to new one, namely Ins nt×nt which is integrator operator without boundary conditions. In addition to bringing forward these two operators, we need to shed light on the right hand side of the equation (2.1) and its meaning in our method. For the first part we propose a new operator which multiplies y(t) by f (t) and represent it with Mnt×nt. Clearly it must multiply the same elements of yt by ft. It is satisfied by Exploiting the Kronecker delta function as follows: Mt,t′ = ft′ δt,t′ (2.5) In the second part, we must integrate a function which is multiplication of the convolu- tion function k(t−τ) by y(τ). Hiring the predefined integrator operator and representing a new integrator operator with convolution factor by Jns nt×nt, we sufficiently introduce a matrix whose elements are defined as follow: Jns t,t′ = kt−t′+1 Ins t,t′ (2.6) In which t and t′ go from 1 to ns. We can rewrite equation (2.1) in matrix form using the operators Dns nt×nt, Mnt×nt and Jns nt×nt: (Dns− M− Jns ).y = 0 (2.7) Where in it subscripts are omitted for abbreviation and matrices are multiplied in usual matrix product law. 2.2 Construction of the Matrix Finite Difference Scheme to solve MIDE To extend our method over matrix domain, retaining their definitions, we replace the functions y, f and k with Yn×n, Fn×n and Kn×n, respectively where n is the dimension of these matrices. The IDE substitutes by its matrix counterpart MIDE as follows: d dt Yn×n(t) = Fn×n(t)Yn×n(t)+Z t 0 Kn×n(t−τ)Yn×n(τ)dτ (2.8) All of the matrices are time dependent. Discretization of these matrices in t domain even- tuates 3D array version of them, namely Yx,x′,t, Fx,x′,t and Kx,x′,t where x, x′ and t are 5 integer numbers. x and x′ belong to set {1,2,··· ,n} and t goes from 1 to nt as pointed before. Supporting matrix representation, we reshape Yx,x′,t to Yξ,x′, in which ξ sweeps both parameters x and t. Consequently it becomes a (nt.n)×(n) matrix. For consistency ξ is defined as: (2.9) ξ(t,x) = t+(x−1)nt By means of our previous definitions, we construct a new first order derivative and (nt.n)×(nt.n) denotes the first order derivative an integrator operator in this scope. Let Dn,ns operator whose elements are defined as: ξ(t,x),ξ′(t′,x′) = δx,x′ Dns Dn,ns t,t′ (2.10) t,t′ is the related matrix entry of Dns Where Dns tion (2.2) and δx,x′ is the Kronecker delta function. Suppose In,ns integrator operator. The elements of this operator are determined as: nt×nt which was formerly defined in equa- (nt.n)×(nt.n) stands for the In,ns ξ(t,x),ξ(t′,x′) = Kx,x′,t−t′+1 Ins t,t′ (2.11) t,t′ represents the proper entry of Ins In this equation Ins nt×nt, once was defined in equation (2.3). Pursuing our procedure we need an adequate operator to multiply Fx,x′,t′ by Yξ,x′. Let M(nt.n)×(nt.n) represents it. Exerting the Kronecker delta function its elements are assigned as: (2.12) ξ(t,x),ξ(t′,x′) = Fx,x′,t′ δt,t′ M We put equations (2.10), (2.11) and (2.12) in to the equation (2.8) to achieve its FDM counterpart as: (Dn,ns−M−In,ns )Y(nt.n)×(n) = 0(nt.n)×(n) (2.13) In which 0(nt.n)×(n) denotes a zero matrix. Generally Y(t) at t = 0 is Y(t = 0) = Y1, i.e.: the Y1 is an arbitrary n×n matrix at t = 0 with adequate conditions based on our problem. Finally, we obtained first order linear partial integro-differential equation as a system of linear equations. To digest we get A = Dn,ns−In,ns−M. (cid:18) A(nt.n)×(nt.t) B(nt.n)×n B† (nt.n)×n Xn×n (cid:19) 0n×n (cid:19) (cid:18) Y(nt.n)×n Y1n×n (cid:19) =(cid:18) 0(nt.n)×n Where B imposes boundary condition at t = 0 by following definition: Bξ(x,t),x′ = δx,x′δt,1 (2.14) (2.15) In which δx,x′ and δt,1 are the Kronecker delta function. It should be noted that X is the discarded part of the answer after solving the equation (2.14). The remainder of this paper is reserved for material to exploit the above method. 6 3 Numerical results 3.1 Studying the electronic transport of a one dimensional system We start with a simple toy model. As a primarily system consider a simple MMM con- sists of two-atom molecule connected to two semi infinite 1D leads. As a primarily system, consider an infinite one dimensional chain of atoms which some part of it may counts as center part and its two tails as two semi infinite 1D electrodes(see figure 2). In Figure 2: (Color online) Schematic illustration of a simple MMM consists of two-atom molecule connected to two semi infinite 1D leads. the tight-binding approximation and second quantization representation, Hamiltonian for the molecule can be written as follows: H(t) = ∑ x ǫ0c† xcx + ′ ∑ <x,x′> tx,x′(t)c† xcx′ (3.1) where the summations run over the lattice sites, ǫ0 is the energy of the electrons at site x, tx,x′ is the transfer energy between site x and x′, and c† x (cx′ ) is the creation (annihilation) operator of electrons at site x (x′). The prime on summation symbol omits the cases x=x′. Σ(t) should be determined to solve equation (1.4). In this method, Σ(t) accounts for the "interaction" of an open system; namely the center part; with the attached two ideal semi infinite leads. So the non-vanishing elements of the self-energy matrix (for this considered system, the first and the last elements) will be the conventional self-energy of an ideal semi infinite 1D chain which in the energy representation can be written as [19]: Σ(ǫ) =  v2 v2 2 [V−ǫ−ip4−(ǫ−V)2] 2 [V−ǫ+ (ǫ−V) ǫ−Vp(ǫ−V)2−4] ǫ−V < 2 ǫ−V > 2 (3.2) In which V is the bias voltage, v is the coupling energy between leads and the central molecule and ǫ is the energy. Σ(t) will be obtained by the Fourier transform of the Σ(ǫ): Σ(t) =−iv2 Θ(t) J1(2t) t eiVt (3.3) where Θ(t) is the Heaviside function, J1 is the Bessel function of the first kind and other parameters are similar to their Fourier transforms. Rewriting the MIDE form of the equation (1.4) for this system, it will be found: i¯h d dt t2,1(t) G(t)−(cid:18) ǫ0 −Z t 0 (cid:18) Σ(t−τ) 0 t1,2(t) ǫ0 (cid:19)G(t) Σ(t−τ) (cid:19)G(τ)dτ = I δ(t) 0 7 (3.4) Instead of the Green's function, we continue our approach with time evolution operator U(t) which has a simple relation with G(t): [G(t) = −iΘ(t)U(t)]. The boundary con- dition for U(t) at t = 0 is U(0) = 12×2. Finally, to use the FDM for equation (3.4), some substitutions by replacing Y and Y1 in equation (2.14) with U and 12×2, respectively and turning K and F in equations (2.11) and (2.12) to self-energy and Hamiltonian matrices. For numerical calculations, except for mentioned cases, parameters ns = 11 and nt = 1000 are fixed to obtain time evolution operator (TEO) in the certain domain of t; (t = [0,··· ,50] in this work) Two major cases are distinguished, a MMM system in the limit of very weak electrods coupling (isolated molecule), which means no convolution term in equation (3.4) [v = 0 in equations (3.2) and (3.3)] and a traditional MMM system [non-zero convolution term in equation (3.4)]. The simplest model may consist of two atoms as the central molecule in which the on-site energy of the electrons ignored and the hopping terms are constant [ǫ0 = 0 and tx,x′(t) = 1.0 in the equation (3.1)]. For the isolated system, the exact analytical solution of the equation (3.4) for U(t) can straightforwardly be found as: U(t) = e−iHt =(cid:18) cos(t) −isin(t) cos(t) (cid:19) −isin(t) (3.5) In this case the FDM solutions for real and imaginary parts of U(t) are shown in figures 3 (a) and (b), respectively where in them Ui,j stands for the ith and jth entry of U(t) matrix. Clearly FDM results for elements of U(t) matrix are in excellent agreement with ones which obtained from analytical solutions. The presence of a periodic time dependent perpendicular electric field, makes a time commutative time dependent Hamiltonian with dynamic on-site energies [ǫ0(t)=cos(t)]. Again, The exact analytical solution of the equation (3.4) for U(t) in this case may be calculated as: (3.6) U(t) = e−iR t 0 H(τ)dτ = e−isin(t)(cid:18) cos(t) −isin(t) cos(t) (cid:19) −isin(t) The FDM solutions for real and imaginary parts of U(t) are shown in figures 3 (c) and (d). Once more, analytical results are consistent with FDM answers for elements of U(t) matrix superbly. In the case of time dependent, non-commutative Hamiltonian in time, a simple case with vanishing on-site energies; ǫ0(t) = 0 and time dependent phase hopping tx,x′(t) = 8 r o t a r e p O n o i t l u o o v E e m T i 1 a) 0.5 0 −0.5 1 c) 0.5 0 −0.5 −1 0 10 b) d) Re U Re U Re U Re U 1,1 1,2 2,1 2,2 Re U Re U Re U Re U 1,1 1,2 2,1 2,2 40 50 10 20 30 time (Atomic Unit) Im U Im U Im U Im U 1,1 1,2 2,1 2,2 Im U Im U Im U Im U 1,1 1,2 2,1 2,2 40 20 30 time (Atomic Unit) Figure 3: Real (a and c), and imaginary (b and d), parts of TEO elements for an isolated system, in the cases of time independent (a and b) and a time commutative time dependent (c and d) Hamiltonians, as mentioned in equations (3.5) and (3.6), respectively. e−iΦ(t) are studied. This phase hopping may emerge from an external magnetic field or strain, etc [20]. Regardless of the physical source of this phase, to facilitate the calculation, a simple time dependent function Φ(t) = Ωt with Ω = 1.0 is selected. Dyson series are usual method to study this types of Hamiltonian [21] which in compact form, exploiting time order operator Tτ to respect the time order, it can be noted as [22]: U(t) = Tτ(e−iR t 0 H(τ)dτ) (3.7) Although this equation is not as tractable as its former counterparts, it can be solved analytically for this specific Hamiltonian and so its answer will be found as: U(t) =  A(t) = C(t) =− A(t) C(t) −C∗(t) A∗(t) √5)e−i √5+1 1 [(5− 10 √5t 2i √5 2 it 2 sin( ) e 2   t +(5+√5)ei √5−1 2 t] (3.8) Using FDM, numerical solution of the equation (3.7) for this specific Hamiltonian even- tuated to figures 4 (a) and (b) for real and imaginary parts of the elements of TEO U(t). r o t a r e p O n o i t l u o o v E e m T i 1 0.5 0 −0.5 a) 1 0.5 0 −0.5 c) −1 0 Re U Re U Re U Re U 1,1 1,2 2,1 2,2 b) Re U Re U Re U Re U 1,1 1,2 2,1 2,2 10 20 30 40 time (Atomic Unit) d) 50 10 20 30 time (Atomic Unit) 9 Im U Im U Im U Im U 1,1 1,2 2,1 2,2 Im U Im U Im U Im U 1,1 1,2 2,1 2,2 40 50 Figure 4: Real (a and c), and imaginary (b and d), parts of TEO elements for an isolated (a and b) and an interacting (c and d) systems, in the case of time dependent non-commutative Hamiltonian in time, as mentioned in equations (3.7) and (3.4), respectively. In all three types of systems the elements of the TEO are not only periodic in time but also non-dissipative which the later one is a natural property of an isolated system. This may be counted as a good evidence for authenticity of the procedure. For an "interacting" system which is more interesting, equation (1.4) contains the convolution term. A simple model may be constructed with vanishing on-site energies; ǫ0(t)=0 and time dependent hopping tx,x′(t) = e−iΦ(t) with Φ(t)= Ωt for central molecule and a weak coupling probability between molecule and electrodes; v = 0.2 (as described in third former isolated system). Unfortunately analytical solution for this system is in- tractable but its numerical solution using FDM led to calculate real and imaginary parts of U(t) (figures 4 (c) and (d), respectively). Dissipative behaviour of elements of the TEO may be regarded as a good physical proof for "interacting" nature of the system. To illustrate both advantages and deficiencies of the FDM, comparison of it with Dyson series method and RK family methods is appropriate. Figure 5 (a) shows the error estimates for four distinct conditions for a short period of time after its initial condition t0 = 0. In an isolated system "without any interactions with electrode"; i.e. [v = 0 in equa- tions (3.2) and (3.3)]; the IDE equation (1.4) is modified to a trivial partial differential equation(PDE) so RK family methods may be the best choice to solve it since RK family methods are faster and more accurate than other ones [dashed line in figure 5 (a)]. Turn- ing the "interactions" on, a general Volterra IDE [equation (1.4)] must be solved. Because 10 the RK method is a very rough approximation to calculate an integral, this method devi- ates rapidly [dashed dot line in figure 5 (a)]. Power series characteristic of Dyson series method in time causes it to fluctuate in time faster than FDM. Comparison of dot line which stands for Dyson series method with solid line which represents the FDM dia- grams in figure 5 (a) can clarify it. Therefore in the vicinity of initial condition in time, both Dyson series method and FDM are applicable but the former one is more accu- rate. After a long enough period of time, effect of the higher power of time in Dyson series causes larger fluctuations and more deviations in its solution and only FDM can be tractable and accurate. All of them are illustrated in figure 5 (b). 1.5 1.0 0.5 -0.5 -1.0 e t a m i t s e r o r r E -6 10 a) -1.5 0.0 MIDE by FDM MDE by RK4 MIDE by RK20-22 MDE by Dyson series MIDE by FDM MDE by RK4 MDE by Dyson series 0.5 1.0 1.5 time (Atomic Unit) b) 2.0 67.0 67.5 68.0 68.5 69.0 time (Atomic Unit) Figure 5: Error estimate of different methods related to a short period a) and a long enough period b) of time. 3.2 FDM application to study electronic transport in a trans-polyacetylene molecule Investigating charge transport in molecular junctions has been interesting for scientists [1, 23 -- 26]. Electrical transport properties of a system have a closed relation to the time evolution of the charge density in it. Consider a MMM system composed of a trans- polyacetylene molecule which contains 20 atoms as the central molecule and two 1D ideal semi infinite electrodes (figure 6), beside the assumption that the coupling energy between the polyacetylene molecule and each electrode is v = 0.5 and an electric po- tential difference is applied between leads ∆V = 2volt; the self-energy of each electrode can be calculated by equation (3.3) in the time representation. Hamiltonian of the trans- polyacetylene was written in the tight binding approximation and second quantization representation [27, 28] as below: ′ ∑ ǫ0c† xcx + H = ∑ x tx,x′ c† tx,x′ = t0−2α(−1)min(x,x′) u0 <x,x′> xcx′ (3.9) In which the summations run over the lattice sites and the on-site energy of electrons (ǫ0) is set properly. The hopping terms tx,x′ relate to nearest neighbour sites. There are three empirical parameters which clarify the magnitude of the hopping integrals. The carbon-carbon atoms hopping integral related to π orbitals equals to t0 = 2.5eV. The electron-phonon coupling constant is α=4.1eV A−1 and u0=0.04 A represents the constant displacement due to Peierls distortion because of dimerization [20, 28]. Applying this Hamiltonian and using the FDM, the equation (1.4) can be solved numerically. Suppose 11 Figure 6: (Color online) A sketch of the polyacetylene molecule between two semi infinite 1D electrodes. ψi,t = 0 > represents a state that an electron exists on ith atom (in the local atomic state i) at the time t = 0. The TEO traces this state in the next time t as follow: ψi,t >= U(t)ψi,t = 0 > (3.10) then the probability density of transition from ith atom at time t = 0 to jth atom at time t is straightforward as: Pi,j = < ψj,tψi,t = 0 >2 (3.11) Evolution of this probability density in time may interpret as the movement of an electron wave packet between different atomic states in the system. For instance figure 7 shows the evolution of transition probability density from the first atomic state, i=1 at time t=0 to all other states at any time (equation (3.11)). P P P P 1,1 1,2 1,3 1,4 P P P P 1,5 1,6 1,7 1,8 P P P P 1,9 1,10 1,11 1,12 P P P P 1,13 1,14 1,15 1,16 P P P P 1,17 1,18 1,19 1,20 1 0.8 0.6 0.4 0.2 y t i l i b a b o r P n o i t i s n a r T 0 0 2 4 6 8 12 time (Atomic Unit) 10 14 16 18 20 Figure 7: (Color online) Transition probability from the first atomic state at time t =0 to other states at time t. To study charge transport in the system one needs to define the charge density in it. Assume qi(t) denotes the charge density of the atomic state i at time t, which is related to initial charges on all states, qj(0) as: qi(t) = ∑ j qj(0) < ψi,tψj,t = 0 >2 (3.12) 12 while its discrete counterpart can be rewritten in the matrix representation as: qψ(t) = Pψ(t)qψ(0) (3.13) In which Pψ represents the matrix of the transition probabilities between atomic states and qψ is a vector containing the charge of each one. Applying the TEO on the charge density one may find the propagation of it in the system during a specific time. For instance assume, at the initial time (t = 0) one electron is arrived in first atom of the molecule from left lead (qi(0) = δi,1). The snapshots of the qψ(t) at different times are shown in figure 8. At the time t = 5.5(AtomicUnit) this electron wave packet collides to the right lead and diffracts [see the figure 8(d)] . Some parts of it reflected back and others transferred to the right lead and this process persists in time. The time evolution of the Hamiltonian eigenstates is useful. Assume ǫ > denotes the eigenstate with eigenvalue ǫ. This means that initially the equation (3.14) holds for this state so ǫ,t = 0 > is calculated by: (3.14) H(0)ǫ,t = 0 >= ǫǫ,t = 0 > then its time evolution can be calculated by: ǫ,t >= U(t)ǫ,t = 0 > (3.15) Suppose qǫ(t) denotes the charge density in energy level ǫ at time t. Rewriting equa- tions (3.12) and (3.13) in these new basis, one may found: qǫ(t) = ∑ ǫ′ qǫ′ (0) < ǫ,tψǫ′ ,t = 0 >2 and its discrete representation as: qǫ(t) = Pǫ(t)qǫ(0) (3.16) (3.17) In which Pǫ is a matrix which represents transition probabilities between energy eigen- states and qǫ denotes a vector contains the charge of these energy levels. Consider half filling situation with sorted states as initial state at the time t=0 and define the Fermi en- ergy at the middle of the energy difference between highest full state and lowest empty one, EF = 0.5eV. The snapshots of the time evolution of this initial state were shown in Figure 9 in several times. During this process, states whose their energy are near the Fermi energy, lose their charge while others are robust. These lost charges are transferred to the leads and make a charge current. To calculate the current, definition of the total charge in the center part of the system at any time t is mandatory. This can be properly defined by the summation of total charge in all atoms of central molecule at any time t, as: qT(t) = ∑ i qi(t) (3.18) 13 1 0.5 0 0.5 0 0.5 0 0.5 0 0.5 0 0.5 0 0.5 0 (a) t=0.00 au (b) t=1.37 au (c) t=2.75 au (d) t=5.50 au (e) t=8.25 au (f) t=13.50 au Local Charge Local Charge Local Charge Local Charge Local Charge Local Charge (g) t=20.00 au Local Charge 2 4 6 8 12 10 Site i 14 16 18 20 x a M q / q i Figure 8: The evolution of a charge density from left to right snapshots in the atomic state basis. Evolution of this total charge in time illustrates the meaning of the charge current in the molecule. Three different cases were shown in figure 10. If the molecule become separated from the leads, its total charge must be constant which is compatible with the black dashed line diagram in figure 10. The red (blue) line diagram refers to the situation in which the electron is arrived in first (last) atom from left (right) lead at time t = 0. Generally, amount of the charge reduces during the time evolution. At first this charge reduces rapidly and leaks to the leads but then partially increases due to contact effect. The potential difference between the leads causes the total charge magnitude for evolution from the left to the right of the molecule be grater than its reverse direction. When the maximum of the charge density collides to the next lead, these two lines are tangent to each others during the large reduction due to the charge leaking to the leads [compare the violet dot line in figure 7 with figure 10 at time t = 5.5(AtomicUnit)] and therefore intersect where direction of the charge flow changes (points c1, c2 and c3). The difference between the charge magnitude before and after charge leaking, determines net amount of charge transferred to the leads which makes a charge current. Comparison between this current interpretation and formal one which relates to the current density operator is beneficial {j(t) = ¯h mℑ[ψ†(t)∇ψ(t)]}. The charge flow is plotted in figure 11. The extremums in this diagram are equal to collisions of the charge density with leads 14 x a M q / ε q 1 0.5 0 1 0.5 0 1 0.5 0 1 0.5 0 1 0.5 0 1 0.5 0 1 0.5 (a) t=0.00 au (b) t=1.37 au (c) t=2.75 au (d) t=5.50 au (e) t=8.25 au (f) t=13.50 au (g) t=20.00 au Ef 0 −4.44 −2.69 0.5 −0.88 1.88 Energy Levels Level Charge Level Charge Level Charge Level Charge Level Charge Level Charge Level Charge 3.69 5.44 Figure 9: The evolution of a charge density from left to right snapshots in the energy basis. which is satisfiable. Applying this method for the first and the last atoms of the molecule, we calculated the current. All the charge evolutions in these two atoms emerge from three distinct sources: 1. The initial charges which have remind there yet qL(R)(t), 2. The charges which depend on other sites qL(R)→R(L)(t), 3. The interaction with the adjacent lead qL(R)→Ll(r)(t). So we can find the total transferred charge from the first atom as: qL(t) = P1,1(t)qL(0) qL→R(t) = n ∑ j=2 P1,j(t)qL(0) qL→Ll (t) = qL(0)−qL(t)−qL→R(t) and so for the last atom we have: qR(t) = Pn,n(t)qR(0) qR→L(t) = n−1 ∑ j=1 Pn,j(t)qR(0) qR→Lr (t) = qR(0)−qR(t)−qR→L(t) (3.19) (3.20) x a M q / T q 15 1.1 1 0.9 0.8 0.7 0.6 0.5 0 left to right right to left right to left and left to right without any leads initial leaking (at left lead) initial leaking (at right lead) transmission probability (left to right) transmission probability (right to left) C 1 collision to left lead collision to right lead collision to right lead collision to left lead C 2 5 10 time (Atomic Unit) C 3 collision to right lead collision to left lead 15 20 Figure 10: (Color online) The total charge in molecule. 1 0.5 x a m j / j 0 −0.5 −1 0 Flow C1 C2 C3 5 10 time (Atomic Unit) 15 20 Figure 11: The charge flow which is calculated by the energy eigenstates. then the transferred charge from the molecule at any time t is calculated as: q(t) = 1 2{[qL→R(t)+qR→Lr (t)]−[qR→L(t)+qL→Ll (t)]} (3.21) As our computations are in the independent electron approximation, total transferred charge at every time t, is sum of all transferred charges in any infinitesimal time period δτ. So we have: q(t)T = q(t)+q(t−δτ)+q(t−2δτ)+··· ≃Z t 0 q(t−τ)dτ (3.22) Therefore its time derivation will give the current [I(t) = d dt q(t)T]. After a proper time, for every constant voltage an steady current Is, will pass trough the system. Fig- ure 12 shows the current diagram at constant voltage v = 5V where the steady current is Is = 0.96mA. This tranquility time may interpret as a relaxation time, τ, which can be calculated by fitting the diagram of figure 12 to a proper function like f (t) = 1−e− t τ . The inverse of this relaxation time (τ−1) may be regarded as resistivity (ρ) of this system. Following this procedure for other voltages, we can find the voltage dependent di- Indeed the relaxation time-voltage, the resistivity- agrams of these three parameters. 16 1 0.8 ) A m 0.6 ( ) t ( I 0.4 0.2 0 0 Time depentent current Steady state current (I ) s (1−e−t/τ ) I s 5 10 time (Atomic Unit) 15 20 Figure 12: (Color online) The charge current as a function of time (t) in the system at a constant voltage voltage and the current-voltage diagrams are found as depicted in figures 13 (a), (b) and (c), respectively. Existence of steps in the current-voltage diagram, is a proper evidence for quantum confinement effect. 4 Summery In summery, we propose a new numerical method to study time evolution in physical systems by using FDM. To solve the correspondent Volterraintegro−di f f erential equation, first we introduced a first order derivative and an integrator operators and discretized them. Using this method we studied the time evolution of a 1D chain Hamiltonian in different situations and compared our results with Dyson series and Runge Kutta. Our method not only is compatible with analytical results but also is more accurate than other numerical methods. Furthermore we study the charge transport in a trans-polyacetylene chain as a central molecule of a MMM system by considering time evolution of its charge density and then calculated its current- voltage diagram. The most significant application emerges from this method that has not instantly men- tioned is that it can properly be applied for time dependent Hamiltonians regardless of the source of this time dependency. So it not only can be used for time dependent Hamil- tonian but also may be used for time dependent self-energies related to the electrodes in MMM system. References [1] A. Nitzan, and M. A. Ratner, Science 300, 1384 (2003). [2] W. Liang, M. P. Shores, M. Bockrath, J. R. Long, and H.Park, Nature 417, 725 (2002). [3] An Introduction to molecular electronics, edited by M. C. Petty, M. R. Bryce, and D. Bloor (Ox- ford University Press, New York, 1995). [4] Molecular Electronics, edited by J. Jortner and M. A. Ratner (Blackwell, Oxford, 1997). [5] N. A. Zimbovskaya, Transport Properties of Molecular Junctions, (Springer, New York, 2013). 17 a) 4 3 2 1 0 40 b) 20 0 1 c) 0.5 ) t i i n U c m o A t ( τ 1 − ) t i i n U c m o A t ( ρ ) A m ( I 0 −0.5 −1 −5 −3 −1 0 1 V (Volt) 3 5 Figure 13: The relaxation time-voltage (a) the resistivity-voltage (b) and the current- voltage (c) diagrams. [6] Molecular and Nano Electronics: Analysis, Design and Simulation, edited by J. M. Seminario (Elsevier, Amsterdam, 2007). [7] Z. G. Yu, D. L. Smith, A. Saxena, and A. R. Bishop, Phys. Rev. B 59, 16001 (1999). [8] Y. Kwok, Y. Zhang, and G. Chen, Front. Phys. 9, 698 (2014). [9] Y. Zhu, J. Maciejko, T. Ji, and H. Guo, Phys. Rev. B 71, 075317 (2005). [10] S. H. Ke, R. Liu, W. Yang, and H. U. Baranger, J. Chem. Phys. 132, 234105 (2010). [11] C. G. Sanchez, M. Stamenova, S. Sanvito, D. R. Bowler, A. Horsfield, and T. N. Todorov, J. Chem. Phys. 124, 214708 (2006). [12] N. Renaud, M. A. Ratner, and C. Joachim, J. Phys. Chem. B 115, 5582 (2011). [13] S. Datta, Quantum Transport: Atom to Transistor, (Cambridge University Press, New York, 2005). [14] P. J. Collins, Differential and Integral Equations (Oxford University Press, New York, 2006). [15] J. W. Thomas, Numerical Partial Differential Equations: Finite Difference Methods (Springer, New York, 1995). [16] P. A. Khomyakov, and G. Brocks, Phys. Rev. B 70, 195402 (2004). [17] M. Abramowitz and I. A. Stegun, Handbook of Mathematical Functions with Formulas, Graphs, and Mathematical Tables(Dover Publication, New York, 1965). [18] I. Tsukerman Computational Methods for Nanoscale Applications: Particles, Plasmons and Waves (Springer, New York, 2007). [19] D. A. Ryndyk, R. Guti´errez, B. Song, And G. Cuniberti, Green Function Techniques in the Treatment of Quantum Transport at the Molecular Scale (Springer-Verlag, Berlin, Heidelberg, 2009). [20] R. E. Peierls,Quantum Theory of Solids (Oxford University Press, Oxford, Great Britain, 1955). [21] L. I. Schiff, Quantum Mechanics (McGraw-Hill, New York, The United States of America, 1949) [22] A. L. Fetter, and J. D. Walecka,Quantum Theory of Many-Particle Systems (McGraw-Hill, New 18 York, The United States of America, 1971) [23] N. J. Tao, Nature Nanotechnology 1, 173 (2006). [24] Y. W. Chang, and B. Y. Jin, J. Chem. Phys. 141, 064111 (2014). [25] M. A. Reed, C. Zhou, C. J. Muller, T. P. Burgin, and J. M. Tour Science 278, 252 (1997). [26] M. Kilgour, and D. Segal, J. Chem. Phys. 143, 024111 (2015). [27] J. C. W. Chien, Polyacetylene: Chemistry, Physics, and Material Science (Academic Press, Or- lando, Florida ,The United States of America, 1984). [28] P. M. Grant, and I. P. Barta, Synthetic Metals. 1, 193 (1980). 

1203.2324

1

1203

A scaling-limit approach to the theory of laser transition

The conditions for the appearance of a sharp laser transition are formulated in terms of a scaling limit, involving vanishing cavity loss and light-matter coupling, $\kappa \to 0$, $g \to 0$, such that $g^2/\kappa$ stays finite. It is shown analytically that in this asymptotic parameter domain, and for pump rates above the threshold value, the photon output becomes large in a sense that is specified, and the photon statistics becomes strictly Poissonian. Numerical examples for the case of a two-level and a three-level emitter are presented and discussed in relation to the analytic result.

cond-mat.mes-hall

cond-mat

A scaling-limit approach to the theory of laser transition Paul Gartner Institute for Theoretical Physics, University of Bremen, 28334 Bremen, Germany and National Institute of Materials Physics, Bucharest-Magurele, Romania∗ (Dated: November 16, 2018) Abstract The conditions for the appearance of a sharp laser transition are formulated in terms of a scaling limit, involving vanishing cavity loss and light-matter coupling, κ → 0, g → 0, such that g2/κ stays finite. It is shown analytically that in this asymptotic parameter domain, and for pump rates above the threshold value, the photon output becomes large in a sense that is specified, and the photon statistics becomes strictly Poissonian. Numerical examples for the case of a two-level and a three-level emitter are presented and discussed in relation to the analytic result. 2 1 0 2 r a M 1 1 ] l l a h - s e m . t a m - d n o c [ 1 v 4 2 3 2 . 3 0 2 1 : v i X r a ∗ [email protected] 1 I. INTRODUCTION An early preoccupation of laser theory was the analogy between the onset of lasing and phase transitions[1 -- 3]. Within the cavity QED models the problem became increasingly accessible to accurate numerical treatments, and evidence that an abrupt change of regime takes place has accumulated. In parallel, approximate analytic considerations also argued in the same sense. A particularly illuminating case is the so-called random injection model of Scully and Lamb[4], which has the advantage of addressing directly the cavity mode statistics. The model was extensively studied [5] and is by now textbook material[6, 7]. In emitter-plus- mode models the situation is more complicated because the photonic state information has first to be extracted, by eliminating the emitter degrees of freedom. This procedure can be carried out[8] in the case of single emitters with few (usually only two) states. Then information about the photon statistics can be obtained, either from the numerics or, using simplifying approximations, analytically too[9 -- 14]. In this context the seminal paper by Rice and Carmichael, Ref. 11, has drawn the attention on the necessity of a limiting procedure for obtaining a sharp transition, with a precisely defined threshold. The analogy with the thermodynamic limit in the theory of phase transition was invoked. It was argued that for the lasing transition the limit involves the β-factor going to 0, together with the cavity loss κ so that the ratio β/κ remains finite. In this limit, and for a pump rate exceeding a threshold value, the number of photons N becomes infinite, generating "an explosion of stimulated emision"[11], and the appropriate object of study is the rescaled value βN, which stays finite. In the present paper we show that a similar scaling limit is needed for an abrupt onset of lasing, but we use in our formulation the more ubiquitous Jaynes-Cummings (JC) coupling constant g, instead of the β-factor. The latter is proportional to g2 and indeed our scaling procedure implies κ → 0, g → 0 so that g2/κ is finite. In this scaling limit and above the threshold one obtains N → ∞ but such that κN remains finite. Moreover we are able to prove not only that the transition becomes abrupt, but also that the photon statistics above the threshold turns exactly Poissonian. The proof is analytic and does not rely on approximations. Also, analytic expression for relevant data, like the threshold pump rate, level occupancies and photon output in the lasing regime, are obtained. Numerical results 2 are shown as illustration of the statements. II. THE MODEL AND STATEMENT OF THE RESULT Single emitter lasers are commonly described as embedded JC systems. In other words, two emitter states, either quantum dot or atomic configurations, interact with the cavity mode via the JC Hamiltonian HJ C = g b† 2i h1 + g b1i h2 , (1) in the presence of, possibly, other states. Here b, b† are the photonic operators and the emitter states are denoted by ii. In particular 1i and 2i specify the upper and the lower laser state, respectively. We assume that the cavity mode is resonant with the laser transition. Dissipation effects are included in the master equation for the density operator ρ (in the interaction picture and with  = 1) ∂ ∂t ρ = −i [HJ C, ρ] + L(ρ) (2) by the Lindblad terms L(ρ) = κ 2 (cid:2)2bρb† − b†bρ − ρb†b(cid:3)+X(i,j) γij 2 h2σijρσ† ij − σ† ijσijρ − ρσ† ijσiji , σij = iihj . (3) A central role here is played by the first term, describing the cavity losses at a rate κ, while the second term summarizes transition processes from states ji to ii at rates γij. Lowering σij operators and their hermitian conjugates σ† ij give rise to Lindblad terms accounting for relaxation, but raising terms are considered as well, to simulate incoherent pumping [15], and the corresponding rate will then be denoted by P . The master equation is solved in time until a steady-state solution is reached, from which data concerning level occupancies and photon statistics can be extracted, as function of the pumping rate and other parameters. With this information at hand one can address the problem of laser transition: how to identify it and what are the conditions for its appearance. The answer to the first question is that the lasing regime is defined by accumulation of a large number of photons in the cavity, the statistics of their number n obeying a Poissonian law ρn,n = Xi ρi,i n,n = λn n! 3 e−λ , for all n . (4) Equivalently, the Poisson statistics amounts to the requirement that the normal-ordered expectation values pn = (cid:10)b† nbn(cid:11) depend exponentially on n, pn = λn, or that the zero time- delay n-th order correlation functions g(n) = pn/pn 1 all become equal to 1. The problem is that, on the one hand, the large number condition is imprecise (how large is large?) and, on the other hand, no matter which of these criteria for Poisson statistics one chooses to apply, one has to check an infinite set of equalities. This is practically impossible, either experimentally or numerically. This is why it is often encountered in the literature that one limits oneself to simpler lasing criteria, like N > 1 and g(2) = 1. The aim of the present paper is to formulate and prove the conditions under which strict Poissonian statistics is generated and at the same time to specify in what sense the photon output becomes large. Essentially these conditions involve the limit κ → 0 and simultaneously g → 0, but with the JC coupling parameter scaling like √κ so that the ratio g2/κ remains finite. The necessity of a certain limiting procedure for obtaining a well-defined transition to a purely Poissonian statistics was recognized long ago[11] and is analogous to the thermodynamic limit in the theory of phase transitions. The precise formulation of our statement runs as follows: a. Rescale κ as ε κ and g as √εg. Then, in the limit ε → 0 and for the pump rate above a threshold value, the average photon number N = p1 tends to infinity in such a way that the product εN remains finite. b. In the same limit and above the threshold, the rescaled expectation values pn = εn pn remain finite for all n, with their limit values obeying an exponential law pn = pn 1 . Equiv- alently, all the correlation functions become g(n) = 1. Below the threshold the increase in photon production is not sufficient, leading to pn = εn pn → 0 for all n ≥ 1, and the correlation functions are in general different from 1. c. In this scaling limit the transition between the two regimes becomes sharp, with a well-defined threshold point. Several comments are in order. The model parameters, g and κ,that are rescaled corre- spond to the photon source and sink, respectively. They are present in all laser models and therefore the formulation of the scaling limit in these terms makes sense in various situations. The limit of small cavity losses (good cavity) pleads in favour of photon accumulation, but with the simultaneous diminishing of the production rate, proportional to g2, their overall increase is not a foregone conclusion. The statement (a) above spells out what is meant by 4 a large photon number, namely that it should scale like 1/κ to compensate for the reduction of the rate of escape from the cavity. Thus, even for a cavity quality factor Q close to infinity there is still light coming out from the device. In situations as those described by incoherent excitation, at high pump rates the phe- nomenon of self-quenching[16] might occur. Beside producing population inversion, the pump can destroy coherence between the laser levels and inhibit the transition. This plays against lasing and therefore one may encounter a double transition, one at the onset of lasing and the other when lasing becomes quenched. In such cases the lasing regime takes place in a given interval of the excitation rates, limited below by the threshold value and above by the self-quenching Pthr < P < Psq. The endpoints of this interval depend on the model parameters. Accordingly, it may occur that the interval shrinks to zero and then no transition takes place. Such situations will also be discussed below. In what follows we will first bring numerical evidence in favor of the scaling limit result. We illustrate the situation with calculations performed on a two-level and on a three-level model. In both cases the scaling limit tendencies are quite clear. Still, numerical statements do not amount to a proof, which can only rely on an analytic argument. We are able to formulate an analytic derivation of the scaling limit result in the two-level case (see Sec. III). Also, an analytic proof is available[17] for the random injection model[4, 5], which does not belong to the class defined by Eqs. (2),(3). We believe that all these results speak in favor of a wider generality of the scaling limit statement. A. Numerical results Steady-state results are obtained from the long-time limit of the master evolution. Particularly indicative of a transition is the behavior of the population inversion w = h1ih1 − 2ih2i as a function of the pump. Depending on the parameters, one clearly detects two types of behavior[12, 14] illustrated in Fig.1(a) [18]. The plots correspond to a two-level emitter in which the pumping is described by the raising 2i → 1i Lindblad operator with the rate P = γ12, while for the rate of loss to non-lasing modes we use the notation γ21 = γ. For one set of parameters the curve is strictly concave, while for the other there appears an almost perfectly linear shortcut[9, 17, 19] separating two concave regions. This clearly suggests that in the latter case an abrupt change of regime is taking place, in 5 w 1.5 1 0.5 0 −0.5 −1 (a) 1 0 −1 0.001 0.01 0.1 0 2 4 6 Pump 1 8 10 10 (b) 1.8 1.4 ) 2 ( g 1 0.6 0.001 0.01 0.1 1 10 Pump FIG. 1. (a) Population inversion w in linear and (inset) semilogarithmic plot, and (b) second order correlation function g(2) for a two-level model. The parameters are, red (solid) line: γ = 0.02, g = 0.1 and κ = 0.01, blue (dotted) line: γ = 0.01, g = 0.01 and κ = 0.02. a given P -interval. It is also obvious that the appearance of the transition is conditioned by certain parameter values. By examining Fig.1(b), we see that this linear segment (which in a semilogarithmic representation appears as an additional convex region, see inset of Fig.1a), corresponds to g(2) being very close to unity, which is characteristic for coherent light. Therefore it is natural to assume that we are in the presence of the lasing regime. Moreover, the fact that for large P -values the linear behavior disappears is consistent with the inhibition of lasing by self-quenching. Zooming in on the leftmost point of the linear interval, as in Fig.2, it is seen that the transition becomes more and more abrupt as the scaling parameter becomes smaller, in accordance with the scaling limit statement. This is seen both in the panel (a) of Fig.2, which refers to the two-level model discussed in Fig.1, and in the panel (b), which shows the result for a three-level emitter. In this latter case the pump is raising the system from the lower laser state 2i to a third state 3i (P = γ32), wherefrom it relaxes to the upper laser state 1i with the rate γ′ = γ13. As before γ21 = γ. The same tendency is seen in plots of g(2), which becomes closer and closer to the coherent light value g(2) = 1, as the scaling parameter becomes smaller, both for the two- and for the three-level case, as illustrated in Fig.3. Finally, in Fig.4 we show numerical results for the photon output. It is seen that in the lasing interval the rescaled photon numbers N = εN have practically reached their limit 6  0.1 0.05 w 0 −0.05 −0.1 (a) 2 lev ε =0.36 ε =0.64 =1ε 0 0.05 0.15 0.1 Pump 0.2 0.25 0.1 0.05 w 0 −0.05 −0.1 (b) 3 lev =0.04 ε ε =0.25 =1ε 0 0.05 0.15 0.1 Pump 0.2 0.25 FIG. 2. Scaling parameter dependence of the population inversion w for (a) the two-level emitter with parameters: γ = 0.02, g = √ε 0.1 and κ = ε 0.01 and (b) the three-level emitter with parameters: γ = 0.02, γ′ = 0.05, g = √ε 0.1 and κ = ε 0.01. Thin solid lines correspond to the analytic result Eq. (17). (a) ) 2 ( g 1.2 1.1 1 0.9 ε =0.36 ε =0.64 =1ε 2 lev 0.1 Pump 1 (b) 3 lev =0.04 ε ε =0.25 =1ε ) 2 ( g 1.8 1.6 1.4 1.2 1 0.1 1 Pump FIG. 3. Scaling parameter dependence of g(2) for the same parameters as in Fig.2. values, given by Eqs. (18) and (19), also plotted in the figure. This means that indeed, N grows like 1/ε, in accordance with the scaling statement. The agreement seen in this section between the numerical data and the scaling limit results can hardly be accidental. Values of g(2) close to unity suggest that in the interval of intermediate pump strengths the system operates in the lasing regime, and the large photon output speaks in favor of this supposition too. Nevertheless, it is not at all clear yet how the linear dependence of the population inversion on the excitation is in any way linked to lasing. The analytic arguments of the next section will prove that, indeed, the two are 7  10 (a) 2 lev 1 N ε 0.1 0.01 ε =0.36 ε =0.64 =1ε 10 1 N ε 0.1 0.01 0.001 (b) 3 lev =0.04 ε ε =0.25 =1ε 0.001 0.01 0.1 1 10 0.001 0.01 0.1 1 10 Pump Pump FIG. 4. Rescaled photon output for the two- and three-level emitter cases. The parameters are the same as in Fig.2. Thin solid lines correspond to Eq. (18) in panel (a), and to Eq. (19) in panel (b). related and appear simultaneously. III. THE TWO LEVEL LASER A special feature of Eq. (2) is the fact that it provides a system of closed equations for In this category, the elements which are a subclass of relevant density matrix elements. diagonal in the emitter states ii are also diagonal in the photon number n, ρi,i only off-diagonal elements are of the form ρ1,2 n,n+1 and their complex conjugates ρ2,1 n+1,n. This is a consequence of fact that the JC Hamiltonian conserves the excitation number 1ih1 + b†b. We consider here the case of an emitter consisting of only the two laser states 1i and 2i. The Lindblad terms describe, beside the cavity losses, the spontaneous emission into non-lasing modes and the pumping, with the rates γ = γ21 and P = γ12, respectively. The n,n, and the master equation implies an infinite set of equations of motion for expectation values. The above-mentioned limitation for the density matrix elements involved, translates into a closed system of equations of motion for a reduced number of relevant expectation values. These are cn = (cid:10)1ih1 b† nbn(cid:11) , vn = (cid:10)2ih2 b† nbn(cid:11) , tn = −ig(cid:10)2ih1 b† nbn−1(cid:11) , 8 n = 0, 1, 2 . . . , n = 0, 1, 2 . . . and (5) n = 1, 2, 3 . . . Obviously, the average population of the upper (lower) level is given by c0 (v0), which obey c0 + v0 = 1. Of a special interest for the photon statistics are the expectation values pn = (cid:10)b†nbn(cid:11) = cn+vn, and in particular the average photon number p1, which is also denoted by N. The imaginary prefactor in the definition of the multi-photon assisted polarization tn makes it a real-valued quantity. The equation of motion for the expectation value of a given operator A can be obtained from the master equation via ∂ ∂t hAi = Tr(cid:26)A ∂ ∂t ρ(cid:27) = −ih[A, HJ C]i + κ 2 (cid:10)(cid:2)b†, A(cid:3) b + b† [A, b](cid:11) ij, Ai σij + σ† +X(i,j) 2 Dhσ† γij ij [A, σij]E . (6) For those in Eq. (5) this leads to the following equations of motion and the corresponding steady-state conditions [8, 12, 17] ∂ ∂t ∂ ∂t ∂ ∂t cn = −(nκ + γ)cn + P vn − 2tn+1 = 0 , vn = γcn − (nκ + P )vn + 2tn+1 + 2ntn = 0 , tn = g2cn + g2ncn−1 − g2vn − 2 (2n − 1)κ + P + γ (7) (8) (9) tn = 0 . By adding Eqs. (7) and (8) one obtains the steady-state balance relation between the losses from the cavity and its feeding through the photon-assisted polarization κ pn = 2 tn , n ≥ 1 . (10) Using this condition and Eq. (7) with n = 0, one obtains P v0 = γ c0 + κ N, which allows to express the steady-state level occupancies in terms of the photon output N c0 = P − κ N P + γ , v0 = γ + κ N P + γ . (11) The unknowns cn and vn can be eliminated from Eqs. (7) and (8) in favor of tn and, using again the balance condition Eq. (10), one is lead[8] to a three-term recursion equation for the photonic quantities pn An pn+1 + Bn pn − Cn pn−1 = 0 , n ≥ 1 , (12) 9 with An = Bn = Cn = 2 κ n κ + P + γ n κ − P + γ n κ + P + γ n P , + n κ (n − 1) κ + P + γ + κ (2n − 1) κ + P + γ 4g2 , (13) (n − 1) κ + P + γ . By using the well-established connection between three-term recursion problems and con- tinued fractions[20], Eq. (12) allows for obtaining directly steady-state values, without re- sorting to the time evolution[17]. The convergence of the continued-fraction solution is very good in all points, except the intermediate pumping region where the transition takes place and the population inversion becomes linear. In that interval an excellent agreement with the numerical solution can be obtained by the following simple ansatz: Assume that (i) Eq. (12) is valid for n = 0 too, and (ii) the last term C0 p−1 takes in this case the value 0, so that one has A0 p1 + B0 p0 = 0 . (14) With p0 = 1 and the values A0 and B0 as in Eq. (13) one obtains for the average photon number N = − B0 A0 = P − γ 2κ − (P + γ)(P + γ − κ) 8g2 . Using this in Eq. (11) leads for the population inversion w = c0 − v0 to w = κ P + γ − κ 4g2 , showing that the linear P -dependence of w is a direct consequence of the ansatz. In the scaling limit, κ → εκ, g → √εg with ε → 0, these results become for the population inversion and w = κ P + γ 4g2 , N = P − γ 2κ − (P + γ)2 8g2 . for the rescaled photon output N = εN. 10 (15) (16) (17) (18) Similar results can be obtained for the three-level model (details of the calculations are left for a future publication) with the conclusion that in the scaling limit the population inversion behaves in as in Eq. (17) above, while the rescaled photon population obeys N = (P − γ)γ′ κ(P + 2γ′) − (P + γ)(P γ + P γ′ + γγ′) 4g2(P + 2γ′) . (19) Note that in the limit of large γ′, that is for very fast 3i → 1i relaxation, one recovers the two-level result, Eq. (18), as expected. The agreement of these expressions, Eqs. (17 -- 19), with the numerical simulations is illustrated in Figs. 2 and 4. It is clear that the ansatz makes sense only in the interval of P values for which N ≥ 0. The first term in Eq. (15) is positive if P is not too small. On the other hand, for large P values, the second, negative term becomes dominant, so that the positivity condition can hold only for a finite interval. Needless to say, the very good agreement between the ansatz and the numerical results (in the interval where the former makes sense) does not constitute a valid proof of the former. It is not immediately obvious why Eq. (12) should hold for n = 0. Neither can one use C0 = 0 as an argument to replace the last term C0 p−1 with 0[19], because it also contains the ill-defined p−1. In order to prove the result one has to show first that, indeed, the three-term recursion relation can be extended for n = 0, identifying in the process the quantity appearing in the role of C0 p−1. In a second step, one has to show that, in certain conditions, this quantity does vanish, as required by the ansatz. The first step of this program is the easier part. It relies on the Glauber-Sudarshan (GS) P-representation for the photonic density operator [21] as an integral over the complex plane of coherent states ρ = Z αiP(α) hα d2α π . (20) Since there is no preferred phase angle in the theory (the reduced photonic density operator, obtained by tracing out the emitter indices, is diagonal in the photon number basis) the P-function depends only on s = α2, P(α) = P(s). Then, the normal-ordered photonic expectation values pn turn out to be moments of the quasi-distribution defined by P pn = Z ∞ 0 snP(s) ds . (21) Note, for further reference, that the Poissonian statistics, characterized by pn = λn would correspond to a sharp peak in the GS function P(s) = δ(s − λ) with λ > 0. 11 The P-representation contains, in principle, the same information as the density operator, but here we take advantage that it allows a natural extension for the definition of expectation values. A case in point is pn which, using Eq. (21), becomes well-defined even for n taking continuous, (not just integer) positive values. Using the differential equations which translate the master equation Eq. (2) into the P-representation formalism [21], one recovers the three- term recursion formula with a continuous index, including the result for index zero[17]. The latter can also be obtained by taking the limit n → 0 in Eq (12) and using for evaluating Cn pn−1 npn−1 = Z ∞ 0 (sn)′P(s) ds = −Z ∞ 0 snP ′(s) ds n→0−−→ P(0) . (22) It is now obvious that the third term in the recursion relation Eq. (12) for n = 0 is propor- tional to P(0) and therefore the validity of the ansatz is equivalent to the requirement that P(s) vanishes for s = 0. The second task is to establish the conditions when the vanishing takes place. This is the central point not only in the justification of the ansatz, but also in the proof of the scaling limit result. The latter amounts to showing that in this limit the P-function becomes a δ-distribution concentrated on a positive value, and this obviously entails that indeed, its value at the origin becomes vanishingly small. To this end we look at the differential equation obeyed by P(s), and which translates the master equation into the language of the P-representation. We summarize here the main steps, the details can be found in [17]. To start with, the density matrix of our problem has a two-by-two block structure, corresponding to the two levels of the emitter. Correspondingly one has four P-functions placed in a matrix Pi,j(s), i, j = 1, 2 and the photonic function we are interested in is obtained by tracing out the emitter-state indices P = P1,1 + P2,2. Using the rules for mapping the master equation for ρ into a Fokker-Planck equation for P, [21] one is lead to a system of equations which is the counterpart of Eqs. (7 -- 9). After eliminating P1,1 and P2,2 in favor of P we obtain a second-order differential equation for the latter. This equation has then to be analyzed in the scaling limit. To simplify the notation we take κ itself as the scaling parameter which goes to 0, and impose the condition g → 0 with g2/κ fixed, by writing g2 = g2κ and keeping g2 constant. The rescaled expectation values pn = κpn, which are the object of the scaling statement, can be obtained as moments of a 12 rescaled P-function with t = κs and pn = Z ∞ 0 κnsnP(s) ds = Z ∞ 0 tn P(t) dt , P(t) = 1 κ P (cid:18) t κ(cid:19) . (23) (24) If, indeed, the number of photons increases to infinity in the scaling limit, then the P quasi- distribution function moves its weight to larger and larger values and its moments cease to exist. In this situation only the rescaled function remains meaningful. Intuitively, according to Eq. (24), the graph of the rescaled function P(t) is obtained from that of the original P(s) by compressing the latter by a factor of 1/κ along the abscissa and expanding it by the same factor along the ordinate. This would bring the rescaled function to δ(t), in the limit κ → 0, were it not for the opposite tendency of P(s) to move away from the origin, as discussed above. The net result of these competing trends is what one has to establish. It is easy to rewrite the differential equation obeyed by P(s) into the corresponding one for P(t), and retain in the coefficients only the dominant terms in the scaling parameter κ. The result is [17]: t2 4g2 κ2 P ′′ −(cid:20)(cid:18)3 γ + P 8g2 + 1(cid:19) t − 1 2 P(cid:21) κ P ′ + (t − ν) P = 0 , with ν an essential parameter in the discussion ν = P − γ 2 − (P + γ)2 8g2 , (25) (26) whose P dependence is important and therefore sometimes emphasized by the notation ν(P ). Note that ν is the same as the rescaled photon population κN, see Eq. (18), so that they change sign simultaneously. The appearance of the small parameter κ along with the derivatives suggests a WKB approach to the κ → 0 asymptotics of the solution. In other words one searches the solution, up to a normalization factor, in the form P(t) = exp(cid:18)− 1 κ ϕ(t)(cid:19) , (27) in which ϕ(t) is taken in the leading, zeroth order in κ. It is clear that when κ gets smaller, the value of P(t) around the minimum of ϕ(t) is greatly enhanced, in comparison with the values at other points which, in the view of normalization, become negligible. In the limit one obtains a δ-function concentrated at the minimum of ϕ(t). 13 The equation obeyed by ϕ(t) in the leading order has the form of a quadratic equation for its derivative t2 4g2 (ϕ′)2 +(cid:20)(cid:18)3 γ + P 8g2 + 1(cid:19) t − 1 2 P(cid:21) ϕ′ + [t − ν(P )] = 0 . (28) Around t = 0 one of the roots behaves like ϕ′ ∼ 2g2P/t2, i.e. ϕ ∼ −2g2P/t which in Eq. (27) leads to a strongly singular solution. The regular one comes from the other root for which ϕ′(0) = −2ν(P )/P . Two cases arise, depending on the sign of ν(P ): (i) As long as ν(P ) is negative, ϕ′(0) > 0, then t = 0 is a minimum for ϕ(t) and, according to the above discussion, P(t) tends to δ(t) (ii) When ν(P ) becomes positive, ϕ′ starts at t = 0 with negative in the scaling limit. values and crosses the abscissa at t = ν. Then ϕ(t) has a local maximum at the origin and therefore the values of P(0) become vanishingly small in the limit κ → 0. This is precisely the requirement for the ansatz to hold. The function P(t) is now concentrated at the point of minimum t = ν(P ). As a consequence, in the interval in which ν(P ) is positive one has P(t) → δ(t − ν) and all the rescaled expectation values become pn = νn, n ≥ 0. Outside this interval P(t) → δ(t) and all pn = 0, except p0 = p0, which is equal to 1 by definition. The change is abrupt and takes place at the interval endpoints defined by the quadratic equation ν(P ) = 0. The condition for this equation to have real roots is g2 ≥ 2γ, or g2 ≥ 2κγ, and then the roots are both positive P± = 2g2 − γ ± 2gpg2 − 2γ . (29) The lowest one, P− = Pthr, corresponds to the onset of lasing and the highest, P+ = Psq, to self-quenching. The condition g2 ≥ 2γ distinguishes the two behaviors illustrated in Fig.1 because, when not fulfilled, no transition takes place. With this, the proof of the scaling limit is complete. It is instructive to see the action of the scaling limit directly on the recursion relation Eq. (12). The essential point is the observation that the n-dependence of the coefficients An, Bn, Cn gradually disappears in the limit κ → 0. More precisely, the recursion for the rescaled expectation values An κ pn+1 + Bn pn − κ Cn pn−1 = 0 , n ≥ 1 , (30) 14  1.2 1.1 1 0.9 (a) 2 lev (4) g g (3) g (2) (4) g g (3) g (2) (b) 1.15 1.1 1.05 1 0.95 3 lev 0.1 Pump 1 0.1 1 Pump FIG. 5. Second, third and fourth correlation function for (a) the two-level emitter and (b) the three-level emitter. The parameters are the same as in Fig.2 with ε = 1 in (a) and ε = 0.004 in (b). reduces, in the κ → 0 limit, to pn+1 = −κ B0 A0 pn = ν pn , n ≥ 1 , (31) with the obvious solution pn = νn−1 p1. For the values of the pump where ν(P ) is negative only the trivial solution pn = 0, n ≥ 1 is possible, in order to avoid negative results for positive expectation values. On the other hand, when ν(P ) > 0, Eq. (31) holds for n = 0 too and one has p1 = ν p0 = ν. Then the solution is exponential pn = νn, n ≥ 0 in accordance with the Poisson statistics. As κ approaches 0, the product nκ in the coefficients of the recursion relation vanishes, and this is how their n-dependence is lost. In the process it is the low-index coefficients that are the first to get close to their limit values, because the limit requires the product nκ to be small. Therefore the Poissonian condition g(n) = 1 is obeyed by g(2) first, and by g(3), g(4), . . . only later. This is numerically confirmed, as seen in Fig.5, and shows that using g(2)) = 1 as a criterion of truly coherent light may be, in this sense, somewhat premature. IV. CONCLUSION By solving the master equation for a single emitter in JC interaction with a cavity mode one observes a sudden change in the behavior of the steady-state solution. This is indicative of the onset of lasing and offers the possibility of identifying the conditions for a sharp tran- 15 sition to a pure, as opposed to approximate, Poissonian statistics. We have shown that these conditions imply an asymptotic regime for the parameters controlling the generation and loss of cavity photons. Specifically, the domain of parameters for which a sharp transitions occurs is defined by both the cavity loss κ and the JC coupling g going to 0, provided that g scales like √κ. The result is supported by numerical data, as exemplified for a two-level and a three- level emitter, and is proven using analytical methods for the two-level model. In a previous paper[17], the same scaling limit was shown to give rise to a sharp transition and to reproduce the threshold value known in the literature, for the random injection model of Scully and Lamb. It should be noted that the Scully-Lamb model does not belong to the class considered here: while the latter are "embedded" JC systems the former is rather an "intermittent" JC one. This fact, together with the numerical evidence, suggests that our scaling limit result has a range of validity that is larger than the set of cases for which a full analytic proof is available now. [1] V. DeGiorgio and M. O. Scully, Phys. Rev. A 2, 1170 (1970) [2] R. Graham and H. Haken, Z. Physik 237, 31 (1970) [3] S. Grossmann and P. H. Richter, Z. Physik 242, 458 (1971) [4] M. O. Scully and W. E. Lamb, Phys. Rev. 159, 208 (1967) [5] S. Stenholm, Phys. Rep. 6, 1 (1973) [6] D. F. Walls and G. J. Milburn, Quantum Optics (Springer, Berlin, 1994) [7] M. Orszag, Quantum Optics (Springer, Berlin, Heidelberg, 2000) [8] G. S. Agarwal and S. D. Gupta, Phys. Rev. A 42, 1737 (1990) [9] T. B. Karlovich and S. Y. Kilin, Opt. Spectrosc. 91, 343 (2001) [10] A. D. Boozer, Phys. Rev. A 78, 053814 (2008) [11] P. R. Rice and H. J. Carmichael, Phys. Rev. A 50, 4318 (1994) [12] E. del Valle, F. P. Laussy, and C. Tejedor, Phys. Rev. B 79, 235326 (2009) [13] E. del Valle and F. P. Laussy, Phys. Rev. A 84, 043816 (2011) [14] A. Auff`eves, D. Gerace, J.-M. G´erard, M. F. Santos, L. C. Andreani, and J.-P. Poizat, Phys. Rev. B 81, 245219 (2010) 16 [15] Coherent excitation can be described by an additional Hamiltonian term[16], but we will not discuss this approach here. [16] Y. Mu and C. M. Savage, Phys. Rev. A 46, 5944 (1992) [17] P. Gartner, Phys. Rev. A 84, 053804 (2011) [18] The steady-state results depend only on the ratio of the parameters, therefore their units are irrelevant and not specified. [19] E. del Valle and F. P. Laussy, Phys. Rev. Lett. 105, 233601 (2010) [20] L. Lorentzen and H. Waadeland, Continued fractions with applications (North-Holland, Am- sterdam, 1992) [21] H. J. Carmichael, Statistical Methods in Quantum Optics 1, 2nd ed. (Springer, Berlin, Heidel- berg, 2002) 17 

1611.07291

1

1611

Split-gate point-contact for channelizing electron transport on MoS2/h-BN hybrid structures

Electrostatically defined nanoscale devices on two-dimensional semiconductor heterostructures are the building blocks of various quantum electrical circuits. Owing to its atomically flat interfaces and the inherent two-dimensional nature, van der Waals heterostructures hold the advantage of large-scale uniformity, flexibility and portability over the conventional bulk semiconductor heterostructures. In this letter we show the operation of a split-gate defined point contact device on a MoS2/h-BN heterostructure, a first step towards realizing electrostatically gated quantum circuits on van der Waals semiconductors. Our devices show signatures of channelized electron flow and a complete shutdown of transport similar to the conventional point contacts defined on bulk semiconductor heterostructures. We explore the role of back-gate and the drain-source voltages on the pinch-off characteristics and, we are able to tune the pinch-off characteristics by varying the back-gate voltage at temperatures ranging from 4K to 300 K.

cond-mat.mes-hall

cond-mat

Split-gate point-contact for channelizing electron transport on MoS2/h-BN hybrid structures Chithra H. Sharma and Madhu Thalakulama School of Physics, Indian Institute of Science Education and Research Thiruvananthapuram, 695016, Kerala, India Electrostatically defined nanoscale devices on two-dimensional semiconductor heterostructures are the building blocks of various quantum electrical circuits. Owing to its atomically flat interfaces and the inherent two-dimensional nature, van der Waals heterostructures hold the advantage of large-scale uniformity, flexibility and portability over the conventional bulk semiconductor heterostructures. In this letter we show the operation of a split-gate defined point contact device on a MoS2/h-BN heterostructure, a first step towards realizing electrostatically gated quantum circuits on van der Waals semiconductors. Our devices show signatures of channelized electron flow and a complete shutdown of transport similar to the conventional point contacts defined on bulk semiconductor heterostructures. We explore the role of back-gate and the drain-source voltages on the pinch-off characteristics and, we are able to tune the pinch-off characteristics by varying the back-gate voltage at temperatures ranging from 4K to 300 K. Electrostatic gating is a versatile technique to engineer the electron flow in 2D systems. Gate defined quantum point contacts (QPC)1,2 and quantum dots3 are the basic building blocks of potential devices for quantum information,4 quantum metrology5 and charge sensing applications6–8. The inherent non-uniformity in the material properties and the associated spatial variations in the operating conditions make large-scale integration of gated devices on conventional semiconductor heterostructures technically challenging. Heterostructures combining van der Waals (vW) semiconductors and insulators,9–11 with its atomically flat interfaces and the two-dimensional nature across the entire device area, could provide the required large-scale uniformity. In addition, the substrate independence of vW systems could pave way for flexible and transferable quantum circuits. Electrostatically defined quantum dots on bilayer graphene-hexagonal boron nitride (h-BN) devices12 and split-gate defined QPCs on Graphene-Al2O3 structures have been demonstrated.13 In contrast to graphene, molybdenum disulfide (MoS2)14 offers superior electrostatically tunable devices owing to the presence of a sizable band gap. Dielectric encapsulation have improved the electron mobility and on-off ratio in MoS2 transistors.15–18 Compared to conventional oxide interfaces MoS2/h-BN devices offer better electrical properties and stability owing to cleaner interfeces.19 Quantum transport phenomena such as quantum Hall effect and Shubnikov-de Haas oscillations are observed in MoS2/h-BN devices20. In addition, recently, single electron transport has been reported on MoS2 21 making it a potential candidate for hosting future quantum circuits. Dual gating on h-BN/MoS2 has offered better and controllable devices.9,19 A natural continuation in this a Electronic mail: [email protected] 1 regard would be the electrostatic shaping and control of transport on MoS2 based vW heterostructure for nano-electronic applications. In this communication we demonstrate channelized electron transport on a MoS2/h-BN heterostructure using a split-gate defined point-contact, the first step towards the realization of controllable quantum devices via electrostatic shaping. The MoS2/h-BN heterostructure is formed by aligned transfer of layers of MoS2 and h-BN on a SiO2/Si wafer using a home-built micro-positioning system. Standard electron-beam lithography is used to define the drain and the source contacts and, the split-gate defining the point-contact. We show that the electron flow in the device can be channelized and pinched-off by controlling the voltage on the split-gate defining the point-contact, in the temperature range between 4K and 300 K. We also show that the pinch-off voltage can be continuously tuned by varying the back-gate voltage. Besides these, we also explore the dependence of the pinch-off characteristics of the channel on the drain-source bias. MoS2 flakes, mechanically exfoliated from bulk crystals are transferred onto a clean 300 nm SiO2/Si substrate and, the drain and the source contacts are defined by electron-beam lithography followed by Cr/Au metallization. A thin flake of h- BN is carefully placed on top of the MoS2 flake using the micro-positioning setup. To ensure a cleaner and a residue-free device both the MoS2 and the h-BN flakes are transferred using the PDMS dry transfer technique.22 The split-gate defining the point-contact is fabricated on top of the h-BN using electron-beam lithography and Cr/Au metallization. The heavily doped underlying Si substrate is used as the global back-gate for controlling the carrier concentration in the device. All electrical measurements are performed in a 4 K - 300 K variable temperature cryostat in high vacuum (< 10-6 mbar) unlit environment. Fig. 1 (a) shows the optical image of the final device. The top inset shows the optical image of the five-layer MoS2 flake (3.5 nm in thickness) on which the device is made. The Raman spectra shown in the bottom inset exhibit the characteristic E1 2g and A1g peaks of MoS2 verifying the structural quality of the sample. A sketch of the device drawn to-scale in Fig. 1 (b) shows the stacking scheme of MoS2, the drain-source contacts, h-BN and the split-gate. We estimate the thickness of the MoS2 and the h-BN flakes from the AFM height profiles shown in the inset. We have used two flakes of h-BN with a total thickness of ~ 11 nm (5nm + 6nm) to ensure complete coverage of the underlying MoS2 flake. Fig. 1 (c) shows the scanning electron microscope (SEM) image of the device showing the drain and the source contacts and the split-gate defining the point-contact. The lithographic dimensions of the point-contact constriction are 280 nm in length and 220 nm in width. From the optical, AFM and SEM images we infer that the h-BN flake on top of the MoS2 is flat and uniform. 2 Figure 1. (a) Optical image of the device, the drain and the source contacts marked as D and S respectively. The top inset shows the optical image of the MoS2 flake and the bottom inset shows the Raman spectra of the MoS2 showing characteristic Raman peaks. (b) A to-scale cartoon of the device, with all the layers labelled. The inset shows AFM height profiles of the MoS2, h-BN (1) and h-BN (2) flakes showing thickness of 3.5 nm, 5 nm and 6 nm respectively. (c) SEM image of the device showing the point-contact with a lithographic length of 280 nm and width 220 nm. The drain and source contacts are marked as D and S respectively. (d) Conductance versus back-gate voltage at four different temperature. The inset shows I-V characteristics at 300 K, for various back gate voltages. Fig. 1 (d) shows the conductance of the MoS2 flake versus back-gate voltage (VBG) for various temperatures, from 300 K to 4 K with the split-gate grounded. The inset shows I-V characteristics of the device at 300 K for various VBG values. We infer from the I-V characteristics and the conductance versus VBG traces that the sample shows n-type behavior and a linear change in conductance with VBG after the device is turned-on. The threshold voltage estimated from the linear extrapolation for all the temperatures are around VBG = 0 V and, we do not find any significant shift in the threshold voltage as the temperature is lowered, suggesting a lower trap density in our sample.18 Now we discuss the formation of the point-contact in our device. For all the measurements discussed in this manuscript we maintain both the electrodes defining the split-gate at the same potential. Fig. 2 (a) shows the pinch-off curve: the current through the device as a function of the voltage on the split-gate (VSG), at 4 K (blue), 77 K (green) and 300 K (red). We keep an on-state current (the current through the device while the voltage on the split-gate, VSG = 0V) of 50 nA though the device and VBG of 10 V. The pinch-off characteristics of split-gated point-contacts possess two distinct regimes; (1) the gradual depletion of electrons under the gate resulting in the formation of the point-contact constriction leading to the onset of channelized transport and, (2) the pinch-off of the point-contact itself.23 We observe a similar behavior in our device as shown in Fig. 2 (a). As VSG is reduced the electrons directly under the gate get depleted (0 V to -3.75 V) resulting in the 3 formation of the constriction exhibiting a shoulder-like structure in the pinch-off characteristics. A further reduction in the VSG results in squeezing the electron flow into the channel followed by the pinch-off of the constriction. This is characterized by the sharp decrease in the conduction followed by a complete shutdown of the transport (-5.75 V and beyond). These features are evident at all the temperatures ranging from 4 K to 300 K. We note here that the features signifying a channelized transport are not observed in a through top-gated device even though the transport could be shut down.16 The inset to Fig. 2 (a) shows the pinch-off characteristics taken by sweeping VSG in the forward (blue) and reverse (red) directions with an on- state current of 20 nA for VBG = 10 V at 4 K. We do not observe any hysteresis effects in our device suggesting that the charge traps do not play much role in the transport behavior. Figure 2: (a) Pinch-off curves at 4 K (blue), 77 K (green) and 300 K (red) where the on-state current is kept as 50 nA at VBG=10 V. The inset shows pinch-off curves taken in the forward and the reverse direction showing no hysteresis (b) Pinch-off characteristics with an on-state current of 1.7 µA showing a high on-off ratio >106. The inset shows IV curves as a function of VSG at VBG=10 V. VSG is varied from -8 V (red) to 0 V (violet). Fig. 2 (b) shows the pinch-off characteristics of the point-contact at 4 K with an on-state current of 1.7 µA and VBG = 10 V showing an on-off ratio in excess of 106. We note that by optimizing the VBG and the drain-source voltage VDS (not explored in this work) one can achieve much higher on-off ratios. The inset shows I-V characteristics for VSG starting from 0 V through the pinch-off. Until the point-contact channel formation (VSG ~ -4 V), the I-V traces lie one on top of the other suggesting that the entire device area contributes to the transport. As VSG is reduced further through the pinch-off regime, the I-V traces show increasingly flatter off regions suggesting that the transport is dominated by the point-contact barrier. To get further insight into the pinch-off characteristics we extract the dependence of the pinch-off voltage (the VSG at which the transport shuts down) on VDS and VBG. Fig. 3 (a) shows a plot of the pinch-off voltage as a function of VDS at 4 K for VBG of 10 V (green) and 2 V (blue). We find that the pinch-off voltage depends linearly on VDS which is a characteristic behavior of transport across a barrier; the current through the barrier is proportional to the voltage across the barrier and the 4 transmission coefficient of the barrier.24 As VDS is increased one need to decrease VSG to shutdown the transmission by escalating the barrier. The top inset shows pinch-off curves taken for three different VDS, 0.61 V (red), 0.72 V (green) and 0.89 V (blue), for on-state currents of 10 nA, 20 nA and 50 nA respectively through the device. The bottom inset shows a magnified view of the pinch-off region showing a reduction in the pinch-off voltage as VDS is increased while VBG is maintained at 10 V. Fig 3. (b) shows a surface-plot of the pinch-off characteristics as VBG is varied from 0 V through 10 V in steps of 1 V at 4 K. For all the traces, we have kept an on-state current of 50 nA by varying VDS. For all the VBG values we are able to pinch-off the channel and shutdown the transport. As VBG is increased we observe that the pinch-off voltage and the Figure 3. (a) Dependence of pinch off voltage on the drain-source voltage for VBG = 2 V (blue) and 10 V (green) at 4 K. Upper inset: pinch-off curves at three different VDS, 0.61 V (red), 0.72 V (green) and 0.89 V (blue) with VBG = 10 V. x-axis is VSG in Volts and y-axis is the current through the device in nA. Lower inset: an enlarged view of the pinch-off region. (b) A surface plot of pinch-off curves at 4 K for a series of VBG with the on-state current kept as 50 nA for each trace. The pink arrow shows a linear shift in the position of the shoulder representing the on-set of channelized transport. (c) Variation of the pinch-off voltage with VBG. The red trace represents the pinch-off voltage extracted from (b) and the black trace from pinch-off characteristics taken with a VDS of 1 V. The inset shows a plot of the applied VDS for each VBG values to set an on-state current of 50 nA. The x-axis is VBG in volts and y-axis VDS in volts. (d) Pinch-off curves for another point-contact device at 300 K for various VBG values. The arrow indicates the linear shift in the position of the shoulder like structure with the VBG. Inset: Optical image of the device, scale bar is 20 µm. 5 shoulder-like feature moves smoothly towards lower VSG values. For a field effect device, the carrier concentration is proportional to the gate voltage. For point-contacts, the pinch-off voltage varies linearly with the carrier concentration and consequently with the back-gate voltage, which we observe in Fig. 3(c). The black trace represents the pinch-off voltage dependence on VBG while a constant VDS of 1 V is maintained across the device. The red trace represents the pinch-off voltages extracted from the traces in Fig. 3 (b). In this case we have increased VDS as VBG is reduced to keep the on-state current 50 nA for all the pinch-off curves as shown in the inset. We observe that, for the same range of VBG the black trace is steeper than the red trace and the red trace flattens at lower VBG values. Though the reduction in the VBG results in an increase in the pinch-off voltage, the corresponding increment in VDS opposes it, resulting in a reduced slope compared to the black trace and an eventual flattening of the trace. Apart from this, we observe that the shoulder-like feature in Fig. 3 (b) is more prominent at higher VBG. At higher VBG, the carrier concentration is higher and the applied VDS is lower, as a result, the depletion of electrons under the gates and the pinch-off of the channel are distinct. Also, the feature moves linearly with VBG as represented by dotted arrows in Fig. 3 (b) as a result of the increase in the carrier concentration. Fig. 3 (d) shows the room temperature pinch-off curves as a function of VBG for another point-contact device made on a 20 nm thick MoS2 flake. An optical image of the device is shown in the inset. All the pinch off curves exhibit the shoulder- like structure and other features implying a channelized transport similar to the previous device. In this communication we have demonstrated the formation of split-gated point-contacts and channelized electron flow in a MoS2/h-BN heterostructure, the first step towards the realization of substrate independent quantum circuits to drive the future technology. Our devices exhibit clear and sharp pinch-off behavior and signatures of channelized transport at all temperatures ranging from 4 K to 300 K, unlike the point contacts on conventional semiconductors, making them potential candidates for the implementation of quantum electrical metrology and other charge detection applications at higher temperatures. The devices exhibit transistor action with an on-off ratio in excess of 106. We were able to tune the pinch-off voltage continuously by varying the back-gate voltage. We have studied the effect of carrier concentration and drain-source voltage on the pinch-off characteristics. The devices show nearly Ohmic behavior at room temperature, the drain and the source contacts turned non-linear at lower temperature ranges. Engineering the contacts and the dielectric interfaces would result in better devices exhibiting quantized transport. 6  ACKNOWLEDGMENTS Authors acknowledge IISER TVM for the infrastructure and the experimental facilities and, Mandar Deshmukh for the help with the micro-positioning setup. MT acknowledges the financial support received from DST-SERB extramural program and CHS acknowledges Abin Varghese for the micromechanical exfoliation of MoS2 in device-2, and CSIR for the fellowship. REFERENCES 1 D.A. Wharam, T.J. Thornton, R. Newbury, M. Pepper, H. Ahmed, J.E.F. Frost, D.G. Hasko, D.C. Peacock, D.A. Ritchie, and G.A.C. Jones, J. Phys. C Solid State Phys. 21, L209 (1988). 2 B.J. van Wees, H. van Houten, C.W.J. Beenakker, J.G. Williamson, L.P. Kouwenhoven, D. van der Marel, and C.T. Foxon, Phys. Rev. Lett. 60, 848 (1988). 3 M.A. Kastner, Phys. Today 46, 24 (1993). 4 D. Loss and D.P. DiVincenzo, Phys. Rev. A 57, 12 (1997). 5 B.J. van Wees, H. van Houten, C.W.J. Beenakker, J.G. Williamson, L.P. Kouwenhoven, D. van der Marel, and C.T. Foxon, Phys. Rev. Lett. 60, 848 (1988). 6 J.M. Elzerman, R. Hanson, L.H. Willems van Beveren, B. Witkamp, L.M.K. Vandersypen, and L.P. Kouwenhoven, Nature 430, 431 (2004). 7 M. Field, C.G. Smith, M. Pepper, D.A. Ritchie, J.E.F. Frost, G.A.C. Jones, and D.G. Hasko, Phys. Rev. Lett. 70, 1311 (1993). 8 M. Thalakulam, C.B. Simmons, B.M. Rosemeyer, D.E. Savage, M.G. Lagally, M. Friesen, S.N. Coppersmith, and M.A. Eriksson, Appl. Phys. Lett. 96, 183104 (2010). 9 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, T. Taniguchi, C. Nuckolls, P. Kim, and J. Hone, ACS Nano 7, 7931 (2013). 10 S. Das, R. Gulotty, A. V. Sumant, and A. Roelofs, Nano Lett. 14, 2861 (2014). 11 A.K. Geim and I. V Grigorieva, Nature 499, 419 (2013). 12 A.M. Goossens, S.C.M. Driessen, T.A. Baart, K. Watanabe, T. Taniguchi, and L.M.K. Vandersypen, Nano Lett. 12, 4656 (2012). 13 S. Nakaharai, J.R. Williams, and C.M. Marcus, Phys. Rev. Lett. 107, 36602 (2011). 14 K.F. Mak, C. Lee, J. Hone, J. Shan, and T.F. Heinz, Phys. Rev. Lett. 105, 136805 (2010). 15 H. Liu and P.D. Ye, IEEE Electron Device Lett. 33, 546 (2012). 16 B. Radisavljevic, A. Radenovic, J. Brivio, V. Giacometti, and A. Kis, Nat. Nanotechnol. 6, 147 (2011). 17 B. Radisavljevic and A. Kis, Nat. Mater. 12, 815 (2013). 18 W.S. Leong, Y. Li, X. Luo, C.T. Nai, S.Y. Quek, and J.T.L. Thong, Nanoscale 7, 10823 (2015). 19 G.-H. Lee, X. Cui, Y.D. Kim, G. Arefe, X. Zhang, C.-H. Lee, F. Ye, K. Watanabe, T. Taniguchi, P. Kim, and J. Hone, ACS Nano 9, 7019 (2015). 20 X. Cui, G.-H. Lee, Y.D. Kim, G. Arefe, P.Y. Huang, C.-H. Lee, D.A. Chenet, X. Zhang, L. Wang, F. Ye, F. Pizzocchero, B.S. Jessen, K. Watanabe, T. Taniguchi, D.A. Muller, T. Low, P. Kim, and J. Hone, Nat. Nanotechnol. 10, 534 (2015). 21 K. Lee, G. Kulkarni, and Z. Zhong, Nanoscale 8, 1271 (2016). 22 A. Castellanos-Gomez, M. Buscema, R. Molenaar, V. Singh, L. Janssen, H.S.J. van der Zant, and G. a Steele, 2D Mater. 1, 11002 (2014). 23 H. van Houten, C.W.J. Beenakker, and B.J. van Wees, Semiconductors and Semimetals (Elsevier, 1992). 24 D.K. Ferry, S.M. Goodnick, and J. Bird, Transport in Nanostructures, 2nd edn (Cambridge University Press, n.d.). 7 

1804.02694

1

1804

Micro-thermocouple on nano-membrane: thermometer for nanoscale measurements

A thermocouple of Au-Ni with only 2.5-micrometers-wide electrodes on a 30-nm-thick Si3N4 membrane was fabricated by a simple low-resolution electron beam lithography and lift off procedure. The thermocouple is shown to be sensitive to heat generated by laser as well as an electron beam. Nano-thin membrane was used to reach a high spatial resolution of energy deposition and to realise a heat source of sub-1 micrometer diameter. This was achieved due to a limited generation of secondary electrons, which increase a lateral energy deposition. A low thermal capacitance of the fabricated devices is useful for the real time monitoring of small and fast temperature changes, e.g., due to convection, and can be detected through an optical and mechanical barrier of the nano-thin membrane. Temperature changes up to ~2x10^5 K/s can be measured at 10 kHz rate. A simultaneous down-sizing of both, the heat detector and heat source strongly required for creation of thermal microscopy is demonstrated. Peculiarities of Seebeck constant (thermopower) dependence on electron injection into thermocouple are discussed. Modeling of thermal flows on a nano-membrane with presence of a micro-thermocouple was carried out to compare with experimentally measured temporal response.

cond-mat.mes-hall

cond-mat

Micro-thermocouple on nano-membrane: thermometer for nanoscale measurements Armandas Balcytis1,2,+, Meguya Ryu3,+, Saulius Juodkazis1,4, and Junko Morikawa3 1Swinburne University of Technology, John st., Hawthorn, 3122 Vic, Australia 2Center for Physical Sciences and Technology, Savanoriu ave. 231, LT-02300 Vilnius, Lithuania 3Tokyo Institute of Technology, Meguro-ku, Tokyo 152-8550, Japan 4Melbourne Center for Nanofabrication, Australian National Fabrication Facility, Clayton 3168 Vic, Melbourne, Australia +these authors contributed equally to this work ABSTRACT A thermocouple of Au-Ni with only 2.5-µm-wide electrodes on a 30-nm-thick Si3N4 membrane was fabricated by a simple low-resolution electron beam lithography and lift off procedure. The thermocouple is shown to be sensitive to heat generated by laser as well as an electron beam. Nano-thin membrane was used to reach a high spatial resolution of energy deposition and to realise a heat source of sub-1 µm diameter. This was achieved due to a limited generation of secondary electrons, which increase a lateral energy deposition. A low thermal capacitance of the fabricated devices is useful for the real time monitoring of small and fast temperature changes, e.g., due to convection, and can be detected through an optical and mechanical barrier of the nano-thin membrane. Temperature changes up to ∼ 2× 105 K/s can be measured at 10 kHz rate. A simultaneous down-sizing of both, the heat detector and heat source strongly required for creation of thermal microscopy is demonstrated. Peculiarities of Seebeck constant (thermopower) dependence on electron injection into thermocouple are discussed. Modeling of thermal flows on a nano-membrane with presence of a micro-thermocouple was carried out to compare with experimentally measured temporal response. Introduction Thermal characterisation of nanoscale heat sources and heat flows around/through nano-objects is a challenging task2 due to a deep sub-wavelength nature when IR imaging is used while a direct contact measurement suffers from a large heat capacitance and, correspondingly, alters thermal distribution pattern. Moreover, direct contact methods of temperature measurements are slow when micro-thermocouples are used3. Simulation of a heat transport by atomistic methods, e.g., Monte Carlo simulations, still have challenges for modeling of the actual sizes of nanoscale devices4. During the last decade advances in measuring heat transport through the interfaces, at conditions of phase transitions with nanoscale resolution using an atomic force microscopy (AFM) probes were reported5. A real time monitoring capability is still strongly required for research of phase transitions, 8 1 0 2 r p A 8 ] l l a h - s e m . t a m - d n o c [ 1 v 4 9 6 2 0 . 4 0 8 1 : v i X r a Figure 1. Thermocouple fabrication stages and patterns of electrodes at different magnifications. Thermocouples were fabricated on glass and Si3N4 membranes of different thicknesses: 1 µm and 30 nm. (a) Metal junctions of 2.5-µm-wide metal stripes with 100× 100 µm2 primary contact pads. (b) Photo image of a laser ablated photolithography mask used for resist exposure. It defines the secondary contact pads to interface with electrical measurements. (c) Photo of the final device on glass. (d) A SEM image of the micro-thermocouple and reference electrodes. The central pair is the Au-Ni thermocouple. (a)(b)(c)(d)AuAuNiAuNiNimaskdeviceSEMAg pastesubstratesubstrate (glass, Si3N4)AuFigure 2. Characterisation of thermocouples. (a) Temperature increase induced by laser heating at different laser powers measured by the optical modulation method (wavelength λ = 830 nm; p-polarisation at slanted front-side incidence). Thermocouple Au-Ni was made on a slide glass. Sensitivity of 10.1 µV/K1 determined for similar thermocouple was used for estimation of temperature changes; Au-Au junction was used as a reference. Illumination of the substrate was carried out from the side to contacts. (b) Temperature vs. laser power for 1 µm-thick Si3N4 membrane. The Au-Au reference electrodes had a Cr adhesion layer and formed a thermocouple which was experiencing a thermal gradient due to asymmetry of the primary contact pads during laser heating (see panel (c)). (c) An optical see-through image of the 400× 400 µm2 SiN-membrane region with thermocouple whose response is plotted in (b); laser spot was ∼ 100 µm in diameter. Note a thermal asymmetry of this layout where the upper 100× 100 µm2 primary contact square pad was on the SiN membrane while the lower one on the Si substrate. crystalline phase formations and photo-thermal cancer treatments6. For an optical light harvesting, the thermal radiation and suppression of reflectivity (impedance matching7) have to be determined for the optimised performance. A recently introduced hot-tip scanning lithography with an AFM tip heated up to ∼ 800◦C temperature (Nanofrazor, SwissLitho, Ltd.) allows to write 3D nanoscale patterns with resolution down to 10 nm in molecular glass resists. With this approach, a secondary electron damage usually occurring in a high-resolution electron beam lithography (EBL) exposure during patterning of thin layers of electronic devices is avoided. Thermal protocols of 3D material growth and structuring for nanotechnology applications (a recent review8) are strongly dependent on thermal properties and conditions, which are currently not well known at the nanoscale. Management of temperature and heat flows in 2D layered materials and structures, e.g., graphene, are important for photo-detectors and light harvesting devices9. Conceptually, a thermal microscopy with a miniaturised heat source and detector are strongly required to develop next generation of transistors beyond current 10-nm-node where thermal management will be of paramount importance. We show here fabrication and characterisation of thermocouples on 30-nm-thick Si3N4 membranes. The Au-Ni thermocou- ple was made from thin evaporated metal films of ∼ 100 nm thicknesses. Small thermal capacitance of SiN nano-membrane facilitated detection of minute temperature changes due to the absorbed energy (dose), e.g., ∼ 0.1 K measured under ∼ 1 mW red laser illumination as well as heating by an electron beam exposure (this estimate was obtained for the sensitivity of 10.1 µV/K determined for a similar thermocouple1). Results 0.1 Laser beam heating Miniaturisation of a thermocouple by fabrication of a cross pattern of few-micrometer-wide stripes of dissimilar metals is an obvious step3 and was demonstrated for direct contact measurements of the temperature diffusivity in polymers10. The next improvement carried out in this study was fabrication of such pattern onto a thin Si3N4 membrane to reduce thermal capacitance and augment sensitivity as well as to reduce a response time of thermocouple. Figure 1 shows the pattern of thermocouple used in this study, a mask for definition of contacts pads, the final device, and a scanning electron microscopy (SEM) image of the Au-Ni micro-thermocouple made on a slide glass substrate. Voltage generated by the thermocouple in response to modulated laser power when laser was illuminated from the front- side (the surface on which the contacts were made) is shown in Fig. 2(a). Sensitivity increased ∼ 34.5 times for the same thermocouple made on a 1-µm-thick SiN-membrane (Fig. 2(b)). For temperature calibration, the sensitivity 10.1 µV/K of a similar thermocouple was used1. In this study we were aiming at detection of fast temperature changes induced by the laser and electron beam irradiation rather on determination of its absolute values (see, Methods Sec. for details). A junction of Au-Au 2/7 01020300255075Au Temperature, T (10-3K)Laser power (mW)Thermocouple (TC)01230.00.10.2 Temperature, T (K)Laser power (mW) TC Au(a)(b)SiSiNlaser(c)SiNmembrane 400 x 400 mm2Figure 3. Thermocouple on a 30-nm-thick SiN-membrane. (a,b) SEM images of thermocouple made on a 250× 250 µm2 SiN window. Note, secondary (large) contact pads are made from the same metal (Au or Ni) as the smaller ones to avoid formation of a secondary thermocouple. The large contacts are placed on Si substrate to remove a thermal gradient on the thermocouple (such gradient was responsible for the observed temperature change in Fig. 2(b) measured with the Au-Au contact). (c) Temperature vs. laser power. (d) Temperature vs. position of the electron beam across the central cross section (along the line in (b)). Diameter of the e-beam was ∼ 0.5 µm at the acceleration voltage of 25 keV; modulation frequency was 30 Hz, current ∼ 1.7 nA as measured by Faraday cup without the sample. Central shaded region depicts the location of membrane. E-beam was scanned across the SiN window and Si substrate without direct exposure of metal leads/contacts. used as a reference also showed some sensitivity for the laser-induced heating. This is caused by two reasons. First, an adhesion of 5-to-10-nm-thick layer of Cr was evaporated before deposition of Au. This created an additional Au-Cr thermocouple. The second reason is due to an asymmetry in placement of the square 100× 100 µm2 contact pads, which have the bi-metal Cr-Au structure (the pads are seen in upper and lower part in Fig. 2(c)). The upper pad was placed on the SiN membrane while the lower one was in contact with the Si substrate. This caused an unwanted temperature gradient upon laser heating and the temperature sensitivity. In next design the both pads were placed on Si to eliminate temperature gradient and sensitivity of the Au-Au reference contacts. In the final design, a 30-nm-thick SiN-membrane with a smaller window was used. It secured placement of the contact pads outside the membrane region, hence, at a constant temperature defined by the bulky Si substrate. Figures 3(a,b) show SEM images of the thermocouple. Response of the thermocouple to laser heating was similar as for 1 µm membrane. Saturation tendency at a larger laser power is attributable to the heat sink effect of the substrate, which was closer to the ∼ 100 µm diameter laser spot (note, the smaller membrane window). The laser irradiation at a slanted angle caused a larger elliptical projection of the laser spot onto the membrane. The Au-Au junction showed no photo-sensitivity when the contact pads were outside the membrane region in this final thermocouple design. 0.2 Electron beam irradiation Thermal sensitivity of the thermocouple to electron beam focused to the ∼ 0.5 µm-diameter spot and scanned across the membrane with single point irradiation at 30 Hz is shown in Fig. 3(d). The largest voltage response was recorded with e-beam close to the thermocouple. Here we used the same 10.1 µV/K coefficient to estimate ∆T and validity of this judgment is discussed below. When separation between the e-beam spot and thermocouple was > 5 µm with e-beam still on the SiN-membrane, the temperature readout was almost the same. When e-beam was on the thick Si substrate, thermocouple was recording a decreasingly smaller temperature as electrons were impinging at a larger distance from the thermocouple. The slope of the voltage with distance had a characteristic single exponential decay over distance xd = 130 µm. With e-beam directly irradiating the Au-Ni junction or the metal leads (Au or Ni) there was an electrical signal generated due to electron injection and was by two orders of magnitude larger (∼ 230 µV vs ∼ 2 µV for the electron and laser exposures, respectively). All the e-beam exposure locations were selected to avoid direct electron irradiation of the metal leads. 0.3 Detection of fast heat transients Next, temporal response of thermocouples with 6 µm2 area were investigated at much higher ∼kHz frequencies with more tight definition of the laser focal spot on the junction (Fig. 4). A square-wave excitation was used to test thermocouple response time. Absorbed amount of light irradiated from the contact side is very small in the case of 30-nm-thick SiN , however, the saturation level of signal is reached faster as compared with the same thermocouple on the slide glass. The fastest segment of temperature rise and decay was τ ∼ 10 µs (the same time constant was the best fit also for the rising part of the transient but is not shown in Fig. 4). Up to a 10 kHz laser repetition rate, saturated temperature values were reached within the period of illumination. The absolute temperature values were extracted from oscillograms (Fig. 4). For the 30 nm SiN-membrane, an approximately 3/7 -500-25002505000.000.050.100.150.200.25 Temperature, T (K)Position (m) TC Au01230.00.10.2 Temperature, T (K)Laser power (mV) TC Au100 µm(a)(b)(c)(d)laserNiNiNiAuAuAu30 µmE-beamNiAuFigure 4. Temporal response of thermocouple on a 30-nm-thick SiN-membrane to a square-wave optical excitation. (a) Video image of a tightly-focused laser beam onto thermocouple with ∼ 10 µm spot diameter; λ = 830 nm. (b) Temporal response of thermocouples: (i) to a 3.6 mW laser power at repetition rate f = 2 kHz with a thermocouple on a slide glass and (ii) to 1.8 mW power with thermocouple on a 30-nm-thick SiN-membrane at f = 2,5,10 kHz; note different x-axis scales in (b). The fastest switching time was τ = RC = 10 µs with ohmic resistance of thermocouple R = 500 Ω and C = 20 nF. Electronic pre-amplifier of 100× was used for a direct observation by oscilloscope. The estimated max-min ∆T span was 16 K (30.7-to-14.7 K above RT of 22◦C) for the thermocouple on glass and ∆T = 4.1 K for 30 nm SiN-membrane (31.3-to-27.2 K) at 2 kHz; at higher 5 and 10 kHz frequencies the max temperature increase was similar ∆Tmax = 30.5 K and min-max span of ∼4 K. 4-times smaller span of min-max temperatures was observed as compared with thermocouple on the slide-glass (4 vs 16 K). Also, slightly higher maximum temperatures were reached on the SiN-membrane. Considering a half of the min-max span of ∆T ≈ 2K occurring within the fastest change of τ = 10 µs, the heating (cooling) rates up to 0.2× 106 K/s are measurable. This is a promising feature for practical applications in real time monitoring of temperature. 0.4 Heat and direct electron injection contributions When the thermocouple response was measured from the front-side (where metal contacts were deposited), larger ∼ 30 µm steps were used and positions were chosen to avoid direct electron irradiation of the thermocouple (Fig. 3). Next, a back-side e-beam irradiation was carried out in small ∼ 7 µm steps with simultaneous detection of back-scattered (reflection) and transmitted (Faraday cup) electrons during measurements of the Au-Ni thermocouple response (Fig. 5). High transmission of SiN-membrane was confirmed with no reflected electrons measured with a detector at a large scattering angle (sensitive to the secondary electrons). Transmission of the membrane to electrons is also confirmed by a high-contract SEM image (Fig. 5(a)). Thermocouple signal was normalised to the transmitted Faraday cup signal and the surface of the Au-Au and Ni-Ni junctions were grounded to eliminate possible charging effects. All the device area was covered with stainless steel foil with an only 4-mm-opening for the e-beam exposure. Small voltage detected by thermocouple (Fig. 5(b)) close to Au-Ni junction is a signature of a changing thermopower since direct electron injection into the junction occur. Thermopower of the free electron gas has a negative sign, hence, a reduction of Seebeck coefficient is expected. The phase of a lock-in amplifier signal showed an expected phase delay as the e-beam was more distant from the thermocouple junction (Fig. 5(c)). The phase delay of a heat wave generated by e-beam ∼ 0.5 µm-diameter heat source at the f = 27 Hz is expected to follow ∆θ = −(cid:112)π f /ad − β , where a [m2/s] is temperature diffusivity, d is the distance between the heat source and thermocouple, and β is instrument constant11. Temperature diffusivity of a 600-nm-thick SiN-membrane was measured and a = 1.3× 10−6 m2/s value was determined12 while that of gold is ∼ 1.2× 10−4 m2/s and ∼ 0.2× 10−4 m2/s for Ni. The linear expression between phase and distance is valid at larger separation between the heat source and the temperature measurement point (see line (1) in (c)); detailed modeling of thermal transport for the used thermocouple on the membrane is presented in Supplementary material section. The fit was achieved for a ∼ 47.7 times larger temperature diffusivity a = 0.62× 10−4 m2/s than that of SiN12. This value is higher than a typical value for Ni and approximately twice lower than that of gold. The electron beam induced heating is one of the contributions to the detected signal in addition to the charge injection which has a strong impact onto an effective thermopower of the metallic segments of thermocouple. Discussion Sensitivity of micro-thermocouple to heating by light and electron beam are demonstrated with the same device. Photo- sensitivity of Ni-Ni junction was observed and was much higher than Au-Au which was used for the reference. This is caused by formation of Schottky junction and oxidation of Ni, e.g., optically transparent Ni films sputtered on Si creates a solar cell13. Future studies and calibration of Ni-NiO, Au-Cr, and rectenna14 metal-insulator-metal structures for temperature detection are strongly required. 4/7 markerDiameter:10 mm-0.4-0.20.00.20.70.80.00.51.0T (normalised)Time (ms) glass; 2kHzSiN 30 nm: 2 kHz 5 kHz 10 kHz = 10 msSiN30 nm(a)(b)Figure 5. Response of micro-thermocouple to back-side electron irradiation. (a) SEM image of thermocouple on 30 nm SiN-membrane by back-scattered (in lens) and secondary (large angle scattered) electrons. (b,c) Measured amplitude and phase response of the Au-Ni thermocouple to a diagonal scan (dashed line in the inset in (b)) with ∼ 7 µm steps measured with a lock-in amplifier. Thermocouple voltage was normalised to the transmitted electron current measured by the Faraday cup using an additional lock-in amplifier; e-beam blanking frequency was 27 Hz. The slope of line (1) in (c) corresponds to the best fit by a linear Phase ∼ Const × d dependence, where d is the distance between heat source and measurement point (only valid at large separations); the temperature diffusivity was a = 0.62× 10−4 m2/s or 47.7 times larger than that of SiN12. The results of e-beam irradiation of thermocouple from the back-side through the SiN-membrane showed anomalous behavior of a smaller amplitude (voltage) with irradiation point closer to the thermocouple with a minimum when the e-beam is focused onto the junction (Fig. 5(b)). During the measurements it was observed that a longer equilibration time (minutes) was necessary for the amplitude and phase to be stable when e-beam was irradiated on the junction. Lower amplitude (voltage) would be equivalent to the smaller temperature for the fixed value of Seebeck coefficient (thermopower). However, injection of electrons into the junction is equivalent to creation a more conductive region with a higher electron density. The higher the density, the more negative values of Seebeck coefficient are expected as for the free electron gas. The phase signal (Fig. 5(c)) at larger separation of the e-beam irradiation from thermocouple is consistent with the temperature diffusivity of gold rather than SiN-membrane as determined above (Secs. Results and Supplement material). An electron injection into micro-junctions by direct e-beam irradiation is a complex and not well controlled phenomenon which can be investigated with the miniaturised thermocouples made for this study. Separation of the pure thermal phenomenon from a dynamic change of thermopower under increased electron density revealed in this study needs further investigation. The thermal modeling of the fabricated thermocouple on a membrane (Supplement) shows a frequency response and the signal (proportional to temperature) detected by lock-in-amplifier and predicts dissipation of heat dominated by SiN membrane. In experiments (Fig. 5(b,c)) however, the faster dissipation was observed corroborating contribution of electron injection and dissipation through metallic leads of thermocouple. Conclusions Thermocouple made of micrometer-wide Au-Ni electrodes on a 30-nm-thick SiN membrane show sensitivity to optical and electron beam excitation and can be used for direct measurement of temperature. Nano-membrane decreases a secondary electron generation which excites a considerably larger sub-surface volume with 1-2 µm cross sections in bulk samples as measured by Au-Ni micro-thermocouple under e-beam exposure1. The miniaturised heat source by a tightly focused laser beam or e-beam accompanied with a miniaturised thermocouple opens a new toolbox for investigation of heat transport at micro- and nano-scales. For the absolute temperature measurements, a dedicated calibration of sensitivity is required1. In this study, a fast temporal response was demonstrated with thermocouple-on-a-membrane. The demonstrated thermocouple on a nano-membrane can also be used in optical microscopy applications where thermal registration is decoupled from the sample, e.g., cells in a buffer solution on the opposite side of the membrane. Recently, an electron-beam excited fluorescence from a cell on transparent membrane was optically mapped with resolution down to 50 nm15 and a direct measurement of the thermal conditions at the nanoscale could further enhance versatility of such technique. In synchrotron radiation experiments, SiN-membranes are common sample support platforms which could have a thermometer function embedded for in situ monitoring of the temperature of the sample. 5/7 -50-250255078910-50-2502550-1.70-1.65-1.60-1.55-1.50-1.45-1.40-1.35Amplitude (V)Distance (m)Phase (rad)Distance (m)1e-secondarye-back scattered250 mNiNiAuAu(a)(b)(c)Methods Membranes of Si3N4 with different thicknesses of 1 µm and 30 nm (Norcada Ltd.) were used as substrates for fabrication of thermocouples; thermocouples were also made on slide glass for reference. Micro-thermocouples were made by electron beam lithography (EBL) using a simple scanning electron microscope (ACE-7000/EBU, Sanyu Electron Ltd.). Lithography steps started with definition of a 2.5 µm-wide Au segment of thermocouple in ZEP520A resist. After development, a 5 nm Cr adhesion layer was evaporated followed with deposition of 50 nm of Au. Then, lift-off was performed in developer. Second step exposure of the Ni segment of the thermocouple was made in ZEP520A. Evaporation of 50 nm of Ni followed by the lift-off. The resulting Au-Ni junction had ∼ 6.3 µm2 area, which is smaller than a typical CCD pixel. Laser-scribed optical mask was made for definition of secondary contact pads. The mask was superimposed with the lithographically defined thermocouple pattern for evaporation of 10 nm of Cr followed with 90 nm of Au. Electrical bonding was made with a silver paste. Ohmic resistance of Au-Ni thermocouple was typically 500 Ω and 90 Ω for Au-Au wire junction of the same geometry. Final device is shown in Fig. 1(c) on a glass substrate. We used the established calibration constant of 10.1 µV/K for Au-Ni thermocouple of similar dimensions1. Calibration of a particular thermocouple can be made using Au-Au (or Ni-Ni) junction fabricated at the close proximity on the same substrate and by measuring ohmic resistance. This measurement is then compared with the direct resistance measurement at the known temperature1. Design of the thermocouple pattern used in this study does not have a resistance heater which is placed equidistantly from the thermocouple and reference wires (Au-Au, Ni-Ni). Hence, such calibration was not carried out and we relied on the reported sensitivity1. Calibration of thermocouple for determination of the absolute temperature are important when films of several nanometers are used due to strong differences in Seebeck coefficient (thermopower): it changes from -4 to +14 µV/K when the film of Cr is increasing in thickness from 5 to 10 nm16 (the scale is defined with Platinum having the Seebeck coefficient S = 0, S = −15 (Ni), S = +6.5 µV/K (Au)). The bulk Cr has thermopower S = 21.8 µV/K16. References 1. Chu, D., Bilir, D. T., Pease, R. F. W. & Goodson, K. E. Submicron thermocouple measurements of electron-beam resist heating. J. Vac. Sci. Technol. B 20, 3044 -- 3046 (2002). 2. Chae, J. et al. Nanophotonic atomic force microscope transducers enable chemical composition and thermal conductivity measurements at the nanoscale. Nano Lett. 17, 5587 -- 5594 (2017). 3. Morikawa, J., Orie, A., Hashimoto, T. & Juodkazis, S. Thermal and optical properties of the femtosecond-laser-structured and stress-induced birefringent regions of sapphire. Opt. Express 18, 8300 -- 8310 (2010). 4. Cahill, D. G. et al. Nanoscale thermal transport. J. Appl. Phys. 93, 793 -- 818 (2003). 5. Cahill, D. G. et al. Nanoscale thermal transport. II. 2003-2012. Appl. Phys. Rev. 1 (2014). 6. O'Neal, D. P., Hirsch, L. R., Halas, N. J., Payne, J. D. & West, J. L. Photo-thermal tumor ablation in mice using near infrared-absorbing nanoparticles. Cancer Lett. 209, 171 -- 176 (2004). 7. Mazilu, M. & Dholakia, K. Optical impedance of metallic nano-structures. Opt. Express 14, 7709 -- 7722 (2006). 8. Seniutinas, G. et al. Tipping solutions: emerging 3D nano-fabrication/-imaging technologies. Nanophotonics 6, 923 -- 941 (2017). 9. Tielrooij, K.-J. et al. Out-of-plane heat transfer in van der Waals stacks through electron -- hyperbolic phonon coupling. Nat. Nanotechnol. 12, online published doi:10.1038/s41565 -- 017 -- 0008 -- 8 (2017). 10. Morikawa, J., Orie, A., Hashimoto, T. & Juodkazis, S. Thermal diffusivity in femtosecond-laser-structured micro-volumes of polymers. Appl. Phys. A 98, 551 -- 556 (2009). 11. Morikawa, J. & Hashimoto, T. Thermal diffusivity of aromatic polyimide thin films by temperature wave analysis. J. Appl. Phys 105, 113506 (2009). 12. Zhang, X. & Grigoropoulos, C. P. Thermal conductivity and diffusivity of free-standing silicon nitride thin films. Rev. Sci. Instr. 66, 1115 (1995). 13. Juodkazyte, J. et al. Solar water splitting: efficiency discussion. Int. J. Hydrog. Energy 41, 11941 -- 11948 (2016). 14. Jayaswal, G. et al. Optical rectification through an Al2O3 based MIM passive rectenna at 28.3 THz. Mater. Today Energy 7, 1 -- 9 (2018). 15. Kawata, Y., Nawa, Y. & Inami, W. High resolution fluorescent bio-imaging with electron beam excitation. Microsc. 63, i16 (2014). 16. Moore, J. P., Williams, R. K. & Graves, R. S. Thermal conductivity, electrical resistivity, and Seebeck coefficient of high-purity chromium from 280 to 1000K. J. Appl. Phys. 48, 610 -- 617 (1977). 6/7 Acknowledgements JSPS KAKENHI Grant No.16K06768. NATO grant No. SPS-985048. AB is grateful for research visit support by Tokyo Institute of Technology and Swinburne University. Professor Juodkazis was supported by a travel grant form MEXT The Program for Promoting the Enhancement of Research Universities for his visit to Tokyo Institute of Technology. Author contributions statement JM initiated the project, AB together with MR made thermocouples and carried out their characterisation. SJ drafted the first version of the manuscript. All authors contributed to discussion of results and writing of the manuscript. Additional information No competing financial and non-financial interests. 7/7 

1506.03141

3

1506

Quantized topological magnetoelectric effect of the zero-plateau quantum anomalous Hall state

Topological magnetoelectric effect in a three-dimensional topological insulator is a novel phenomenon, where an electric field induces a magnetic field in the same direction, with a universal coefficient of proportionality quantized in units of $e^2/2h$. Here we propose that the topological magnetoelectric effect can be realized in the zero-plateau quantum anomalous Hall state of magnetic topological insulators or ferromagnet-topological insulator heterostructure. The finite-size effect is also studied numerically, where the magnetoelectric coefficient is shown to converge to a quantized value when the thickness of topological insulator film increases. We further propose a device setup to eliminate the non-topological contributions from the side surface.

cond-mat.mes-hall

cond-mat

a Quantized topological magnetoelectric effect of the zero-plateau quantum anomalous Hall state Jing Wang,1, 2 Biao Lian,1 Xiao-Liang Qi,1 and Shou-Cheng Zhang1, 2 1Department of Physics, McCullough Building, Stanford University, Stanford, California 94305-4045, USA 2Stanford Institute for Materials and Energy Sciences, SLAC National Accelerator Laboratory, Menlo Park, California 94025, USA (Dated: August 13, 2018) Topological magnetoelectric effect in a three-dimensional topological insulator is a novel phe- nomenon, where an electric field induces a magnetic field in the same direction, with a universal coefficient of proportionality quantized in units of e2/2h. Here we propose that the topological mag- netoelectric effect can be realized in the zero-plateau quantum anomalous Hall state of magnetic topological insulators or ferromagnet-topological insulator heterostructure. The finite-size effect is also studied numerically, where the magnetoelectric coefficient is shown to converge to a quantized value when the thickness of topological insulator film increases. We further propose a device setup to eliminate the non-topological contributions from the side surface. PACS numbers: 73.43.-f 73.20.-r 85.75.-d The search for topological quantum phenomena has become an important goal in condensed matter physics. Topological phenomena in the physical systems are de- termined by topological structures and are thus univer- sal and robust against perturbations, and the electro- magnetic response is usually exactly quantized [1]. Two well-known examples of topological quantum phenom- ena are the flux quantization in superconductors [2] and Hall conductance quantization in the quantum Hall effect (QHE) [3]. The remarkable observation of such topolog- ical phenomena is that the quantization is exact, which provide the precise values of fundamental physics con- stants such as Plank's constant h [4]. The recent discovery of the time-reversal (T ) invariant (TRI) topological insulator (TI) brings the opportunity to study a large family of new topological phenomena [5, 6]. The electromagnetic response of a three-dimensional (3D) insulator is described by the topological θ term [7 -- 9] of the form (cid:90) Sθ = θ 2π e2 h d3xdtE · B, (1) together with the ordinary Maxwell terms. Here E and B are the conventional electromagnetic fields inside the in- sulator, e is the charge of an electron, and θ is the dimen- sionless pseudoscalar parameter describing the insulator, which refers to the axion field in particle physics [10]. Un- der the periodic boundary condition, all physical quan- tities are invariant if θ is shifted by integer multiples of 2π. Therefore, all TRI insulators are described by either θ = 0 or θ = π (modulo 2π). TIs are defined by θ = π, which cannot be connected continuously to trivial insu- lators, defined by θ = 0, by TRI perturbations. With open boundary condition, the effective action is reduced to a (2+1)D Chern-Simons term on the surface, which describes a surface QHE with half-quantized surface Hall conductance [7]. Such a topological θ term with a uni- versal value of θ = π in TIs leads to a magnetoelectric effect with coefficient quantized in units of e2/2h, known as the topological magnetoelectric effect (TME), i.e., an electric field can induce a magnetic polarization, whereas a magnetic field can induce an electric polarization. To obtain the quantized TME in TIs, as is first suggested in Ref. [7], one must fulfil the following stringent require- ments. First, introduce a T -breaking surface gap by fer- romagnetic (FM) ordering, where the magnetization of FM points inward or outward from the surface. Second, finely tune the Fermi level into the magnetically induced surface gap and keep the bulk truly insulating. Third, the film of TI material should be thick enough to eliminate the finite-size effect, therefore the TME is exactly quan- tized. Several other theoretical proposals [11 -- 14] have been made to realize the TME; however, observing the TME in TIs experimentally is still challenging. In this paper, we propose to realize TME effect in the newly discovered quantum anomalous Hall (QAH) state [15, 16]. Recently, a new zero-plateau QAH state in a magnetic TI has been theoretically predicted [17] and experimentally realized [18, 19]. The magnetic TI studied in the QAH experiment develops robust FM at low temperature. In the magnetized states, the magnetic domains of the material are aligned to the same direction, and the system is in a QAH state with a single chiral edge state propagating along the sample boundary, where the Hall conductance σxy is quantized to be ±e2/h. The zero-plateau state, on the contrary, appears around the coercivity when the magnetic domains reverse, where σxy shows a well-defined zero-plateau over a range of mag- netic field around coercivity while longitudinal conduc- tance σxx → 0 [shown in Fig. 1a]. In such a state, the Fermi level is in the magnetization induced surface gap, fulfilling the first two conditions above, providing a good platform to observe the TME effect as we will discuss in details below. However, due to finite thickness in mag- netic TI, the TME is non-quantized. Therefore, we fur- 2 energy physics of this system consists of Dirac-type sur- face states only [17, 20]. At the coercivity, both ran- dom magnetic domains that formed in the sample, and the exchange field ∆ introduced by the FM ordering are spatially inhomogeneous. The 3D spatial average of ex- change field vanishes ((cid:104)∆(cid:105)ave = 0). However, due to an unavoidable top-bottom asymmetry, the top and bot- tom surfaces may feel an opposite nonzero exchange field ∆t = −∆b = ∆0. In this case, the zero-plateau state is described by the mean field effective model which has the generic form as H0 = kyσ1 ⊗ τ3 − kxσ2 ⊗ τ3 + ∆0σ3 ⊗ τ3 + m1 ⊗ τ1. with the basis of t ↑(cid:105), t ↓(cid:105), b ↑(cid:105) and b ↓(cid:105), where ↑ / ↓ represent the spin up/down states, respectively. σi and τi (i = 1, 2, 3) are Pauli matrices acting on spin and layer, respectively. m describes the hybridization between the top and bottom surface states. If m = 0, due to the opposite half-integer Hall conduc- tance contributions from the top and bottom surfaces xy = −σb σt xy = 0, which gives rise to a quantized TME as discussed above. However, a nonzero m will mix the circulating current Jt and Jb, therefore, the TME is no longer quantized. In reality, the exchange field depends very much on the microscopic details of the randomness in magnetic do- mains. However, we emphasize that the TME of the zero-plateau state in a magnetic TI is in general nonzero and non-quantized. xy = sgn(∆0)(e2/2h), the system has σtot xy + σb To realize the quantized TME effect, the TI film should be thick enough so that the hybridization between top and bottom surfaces is negligible. Therefore, we pro- pose that a quantized TME can be realized in the zero- plateau state of FM-TI-FM structure as shown in Fig. 1b. The FM insulators A and B have different coercivity H c 1 and H c 2, respectively. Assume that both FM A and B have an out-of-plane magnetic easy axis, and the same sign of the exchange coupling parameter to TI surface states. When A and B have antiparallel magnetization, the system is in a zero-plateau QAH state with σtot xy = 0, xy as (1/2 − 1/2)(e2/h) which is contributed by σt or (−1/2 + 1/2)(e2/h). Such a magnetization configura- tion can be easily achieved in the hysteresis loop by an external field H with H c 2, and then remove H. Experimentally, to achieve the TME in this setup, a good proximity between FM and TI is necessary. TI material can be chosen as BiySb1−yTe3, where the Dirac cone of the surface states is observed to be located in the bulk band gap [21]. The candidate FM materials are Cr2Ge2Te6 (CGT), Crx(Bi,Sb)2−xTe3 (CBST) with 0.3 < x < 0.46 and Vx(Bi,Sb)2−xTe3 (VBST). All of them are FM insulators with an out-of-plane easy axis, and have good lattice match with Bi2Te3 family mate- rials. CGT is a soft FM insulator with Tc ∼ 61 K and Hc < 100 Oe [22], and it also shows good proximity with Bi2Te3 [23]. CBST with 0.3 < x < 0.46 is a FM insulator with Tc = 40-90 K and Hc ∼ 1.0 × 103 Oe [24]. VBST with 0.1 < x < 0.3 is a FM insulator with Tc = 30-100 K 1 < H < H c FIG. 1. (color online). Zero-plateau QAH state and magnetic field dependence of σxy. (a) Magnetic TI in an external field H, and sketch of σxy as a function of H. σxy = 0 plateau oc- curs at the coercivity. The arrow indicates the magnetization direction. (b) FM-TI-FM heterostructure, and σxy vs H. The zero-plateau appears in the hysteresis loop due to different H c 1 and H c 2 of FM A and B. The blue and red arrows indicate the magnetization direction of FM A and B, respectively. ther propose to realize the quantized TME effect in the zero-plateau QAH state of the FM-TI-FM heterostruc- ture as shown in Fig. 1b, where an in-plane ac magnetic field induces an electric current in the same direction, or an in-plane electric field induces a magnetic field. The finite-size effect is also studied numerically, where the TME coefficient is shown to converge to a quantized value when the thickness of TI film increases. Finally, we pro- pose a device setup where the non-topological contribu- tion from the side surface is negligible. The TME described by the topological θ term implies that a quantized magnetic polarization is induced by an electric field, given by M = − θ 2π e2 h E. (2) Such response can be understood in terms of a surface Dirac fermion picture. With antiparallel magnetization of the two FM layers as shown Fig. 2b, an in-plane electric xyz × E field E = Ey y induces the Hall currents Jt = σt xyz × E on top (z = d/2 and denoted as and Jb = σb superscript t) and bottom (b and z = −d/2) surfaces, respectively. Since the surface massive Dirac fermion gives rise to half-integer Hall conductance σt xy = (θ/2π)(e2/h), the currents Jt = −Jb are opposite and form a circulating total current. Inside the sample, such a circulating current can be viewed equivalently as the surface bound current generated by a constant magne- tization M = −J t y = −(θ/2π)(e2/h)E. Therefore, the TME essentially originates from the half-quantized sur- face Hall conductance. xy = −σb First we examine theoretically the TME in the zero- plateau QAH state observed in experiments. The low- 10−1xyσ2[]ehH*2H−*1H−*1H*2H10−1xyσ2[]ehHc2H−c1H−c1Hc2H1122−−1122+1122+1122−−1122−1122−+00FM Insulator AFM Insulator BTIc1Hc2HMagnetic TI(a)(b)H3 FIG. 3. (color online). Finite-size effect of TME. (a) The γ and κ(z = 0) as a function of d. The inset shows γ plotted vs the inverse of thickness 1/d. (b) The current amplitude J0 scales linearly with d. Here θ/π = γ. −iJ0e−iωt with J0 = 1.11 nA, in the range accessible by transport experiments. Moreover, as shown in Fig. 3b, the current amplitude J0 scales linearly with thickness d, for θ is a linear function of 1/d with (1 − θ/π) ∝ 1/d, i.e., thicker film gives rise to larger TME. Finite-size effect Due to the finite-size confinement along z direction, the TME effect is not quantized when the TI film is thin. However, as we shall show be- low, the TME coefficient converges quickly into the quantized value as the film thickness d increases. The generic Hamiltonian of a TI thin film can be written as −d/2 dzH3D(k, z). Here k = (kx, ky), and H2D(k) = (cid:82) d/2 we impose periodic boundary conditions in both x and y directions. The magnetoelectric response of such a thin film can be directly calculated with the Kubo formula. With the 3D in-plane current density operator defined as j3D(k, z) = (e/)∂kH3D(k, z), we can write down a dc current correlation function Πxy(z, z(cid:48)) = (cid:88) f (
nk) d2k (cid:90) (cid:34)(cid:104)unk j3D 2 2πe2 ×2Im n(cid:54)=m x (k, z)umk(cid:105)(cid:104)umk j3D y (k, z(cid:48))unk(cid:105) (
nk −
mk)2 (cid:35) , where unk(cid:105) is the normalized Bloch wavefunction in the n-th electron subband satisfying H2D(k)unk(cid:105) =
nkunk(cid:105) , and f (
) is the Fermi-Dirac distribution func- tion. The Kubo formula for magnetic field By induced by a uniform external electric field Ey is then given by where κ(z) =(cid:82) d/2 e2 2h By(z) = −µ −d/2 dz1 sgn(z − z1)(cid:82) d/2 κ(z)Ey , −d/2 dz2 Πxy(z1, z2) is a dimensionless function. Here sgn(z) gives the sign of z. Similarly, the current density j3D induced by a uniform external ac magnetic field Bx of frequency ω/2π is given by x (4) x (z) = −iω j3D e2 2h η(z)Bx , (5) x (z) is defined in Eq. (5). FIG. 2. (color online). TME effect. (a) Illustration of the ac electric current induced by an ac magnetic field Bx, the current density j3D (b) Magnetic field By induced by applying an electric field Ey through a capacitor. Ey (with direction into the paper) will induce Hall currents Jt and Jb for antiparallel magnetization. (c) The functions κ(z) and η(z) for different thickness 6, 10, 20 QL. Each QL is about 1 nm thick. and Hc ∼ 1.0 × 104 Oe [25]. TME As we discussed previously, an electric field will induce a topological contribution to bulk magnetization. From the constituent equation H = B/µ − M, with H = 0 and B continuous, we have on the middle of side surface (parallel to z) B = −µ(e2/2h)Ey y. Here µ is the material-dependent magnetic permeability. Tak- ing µ ≈ µ0, Ey = 105 V/m, we get the magnitude of magnetic field 2.43 × 10−6 T, which is easily detectable by present superconducting quantum interference devices (SQUID). The stray magnetic field effect can be well sep- arated from the quantized TME by ac modulation of the electric field and phase-locking detection, where the ac frequency is quasi-static around 10-100 Hz. Moreover, a gradiometer sensor in SQUID could also screen the ho- mogeneous stray field. The TME also indicates the induction of a parallel polarization current when an ac magnetic field is ap- plied. Consider the process of applying an ac mag- netic field B = Bx x as shown in Fig. 2a. A circulat- ing electric field E parallel to side surface (parallel to B) is generated due to Faraday's law of induction, where Et = −Eb = (∂Bx/∂t)(d/2)y. Such an electric field will induce a Hall current density j2D = j2D b , where xyz × Eb. Therefore the j2D t = σt total current J = j2D(cid:96) = J x, where xyz × Et and j2D t + j2D b = σb J = θ π e2 2h ∂Bx ∂t (cid:96)d. (3) Here d and (cid:96) are the thickness and width of the TI film as shown in Fig. 2a, and θ → π when d is large enough. For an estimation, take Bx = B0e−iωt, B0 = 10 G, ω/2π = 1 GHz, d = 20 nm, θ/π ≈ 0.91 (finite-size effect taken into account as in Fig. 3), and (cid:96) = 500 µm, we have J = (a)yEyB(b)2(2)tyJehE=0-3300.5100.510-5500.510-55-10100-330462z0-550462z0-55-10100462z(c)(nm)(nm)(nm)3DxjdxByETI3Dxj()zκ()zη2(2)byJehE=−yMxyz2zd=−2zd=0510152025300.30.40.50.60.70.80.91.0d (nm)0.00.10.20.30.40.60.81.01/d (nm-1)γγ(0)zκ=0510152025300.00.20.40.60.81.01.21.41.61.8 d (nm)(nA)B0=10 G21ωπ=500µ=GHzm(a)(b)where η(z) = 2(cid:82) d/2 −d/2 dz1 z1Πxy(z, z1) is also dimension- less. z + D2(k2 z + B2(k2 The response formulas above are generic for any TI system and do not rely on a specific model. For con- creteness, we adopt the effective Hamiltonian in Ref. [26] to describe the low-energy bands of Bi2Te3 family mate- rials, H3D(k, z) = ε1⊗1+d1τ1⊗1+d2τ2⊗σ3 +d3τ3⊗1− ∆(z)τ3 ⊗ σ3 + iA1∂zτ2 ⊗ σ2. Here τj and σj (j = 1, 2, 3) are Pauli matrices, ε(k, z) = −D1∂2 x + k2 y), d1,2,3(k, z) = (A2kx, A2ky, B0 − B1∂2 x + k2 y)), and ∆(z) is the z-dependent exchange field. We then discretize it into a tight-binding model along z-axis be- tween neighboring quintuple layers (QL) from H3D, and assume ∆(z) takes the values ±∆s in the top and bottom layers, respectively, and zero elsewhere. Fig. 2c shows the numerical calculations of κ(z) and η(z) for thin films of 6, 10 and 20 QL, where we set a typical surface exchange field ∆s = 50 meV. All the other parameters are taken from Ref. [27] for (Bi0.1Sb0.9)2Te3. The bulk value of κ(z) at z = 0 as a function of d is plotted as black line in Fig. 3. As is consistent with the topological field theory, κ(z) in the bulk tends to 1 and becomes quantized as the thickness d increases, whereas η(z) is bounded within a finite penetration depth to the top and bottom surfaces. The shape of functions κ(z) and η(z) near surfaces re- main almost unchanged as the thickness d varies. To characterize the deviation from topological quan- tization of TME in TI thin films, we further define the dimensionless number (cid:90) d/2 −d/2 (cid:90) d/2 −d/2 γ = 1 d dz κ(z) = 1 d dz η(z) , (6) y (cid:82) dzj3D which is the mean value of κ(z) or η(z) (which are equal to each other). The average magnetic field in re- sponse to the external electric field Ey is then Bmean = −γµ(e2/2h)Ey, whereas the total 2D current density in- duced by external magnetic field Bx is given by j2D x = x (z) = −iωγd(e2/2h)Bx. Compared to Eq. (3), we get θ/π = γ. The value of γ as a function of d is shown as the red line in Fig. 3, where γ → 1 with d → ∞. This shows the TME effect is quantized as the system is in the thermodynamic limit. In fact, as shown in the inset of Fig. 3, the value of 1 − γ scales lin- (cid:82) d/2 early with 1/d as the thickness d → ∞, which indicates −d/2 dz(1− κ(z)) = const. when d is large enough. This is simply because the function 1 − κ(z) is nonzero only near the top and bottom surfaces, and its shape is inde- pendent of the thickness d, as shown in Fig. 2c. Since (1 − γ) ∝ 1/d, J0 ∝ θd = γπd is a linear function of d, as is shown in Fig. 3b. Discussion The TME effect in the setup Fig. 2 is not quantized when the side surface (parallel to z) is not gapped. The gapless side surface states may give rise to a non-topological contributions to the TME effect. To eliminate such non-topological contribution, one can use 4 FIG. 4. (color online). Schematic of FM-TI-FM heterostruc- ture to observe the quantized TME, where the side surface of TI is gapped by FM proximity in (a). Au is the electrode. Such geometry of TI can be made by lithography. In reality, such configuration cook in experiments may like (b), where the side surface is uneven and step-like. In this case, the side surface is gapped by both FM ordering and quantum confine- ment. the device setup as shown in Fig. 4, where the side surface is gapped either by FM order or quantum confinement. Also in this setup, if FM A and B have opposite sign of exchange coupling parameter to TI surface, only paral- lel magnetization is needed to realize TME. Recently, the surface QHE has been realized in bulk Bi2−xSbxTe3−ySey TI [28, 29], where the systems exhibit surface-dominated conduction even at temperatures close to the room tem- perature, whereas the bulk conduction is negligible. Such experimental progress on the material growth and rich material choice of TI and FM insulator make the realiza- tion of the quantized TME in TIs feasible. We are grateful to Ke He for sharing their data prior to publication, and thank Yihua Wang and Andre Broido for useful comments on the draft. This work is supported by the US Department of Energy, Office of Basic Energy Sciences, Division of Materials Sciences and Engineering, under Contract No. DE-AC02-76SF00515 and in part by the NSF under grant No. DMR-1305677. XLQ is sup- ported by the NSF through the grant No. DMR-1151786. Note added: During the preparation of our manuscript, we learned of an independent work on a similar prob- lem [30]. However, their experimental proposal to ob- serve the TME is different from our results. [1] D. J. Thouless, Topological Quantum Numbers in Non- relativistic Physics (World Scientific, Singapore, 1998). [2] N. Byers and C. N. Yang, Phys. Rev. Lett. 7, 46 (1961). [3] D. J. Thouless, M. Kohmoto, M. P. Nightingale, and M. den Nijs, Phys. Rev. Lett. 49, 405 (1982). [4] P. J. Mohr, B. N. Taylor, and D. B. Newell, Rev. Mod. Phys. 80, 633 (2008). [5] M. Z. Hasan and C. L. Kane, Rev. Mod. Phys. 82, 3045 (2010). [6] X.-L. Qi and S.-C. Zhang, Rev. Mod. Phys. 83, 1057 (2011). [7] X.-L. Qi, T. L. Hughes, and S.-C. Zhang, Phys. Rev. B 78, 195424 (2008). FM AFM BTIAuAuc1Hc2HFM Bc2HTIFM Ac1H(a)(b)[8] A. M. Essin, J. E. Moore, and D. Vanderbilt, Phys. Rev. [9] We use SI units. For CGS units e2/h → α/2π, where α Lett. 102, 146805 (2009). is the fine structure constant. [10] F. Wilczek, Phys. Rev. Lett. 58, 1799 (1987). [11] X.-L. Qi, R. Li, J. Zang, and S.-C. Zhang, Science 323, 1184 (2009). [12] J. Maciejko, X.-L. Qi, H. D. Drew, and S.-C. Zhang, Phys. Rev. Lett. 105, 166803 (2010). [13] W.-K. Tse and A. H. MacDonald, Phys. Rev. Lett. 105, 057401 (2010). [14] K. Nomura and N. Nagaosa, Phys. Rev. Lett. 106, 166802 (2011). [15] C.-Z. Chang, J. Zhang, X. Feng, J. Shen, Z. Zhang, M. Guo, K. Li, Y. Ou, P. Wei, L.-L. Wang, Z.-Q. Ji, Y. Feng, S. Ji, X. Chen, J. Jia, X. Dai, Z. Fang, S.-C. Zhang, K. He, Y. Wang, L. Lu, X.-C. Ma, and Q.-K. Xue, Science 340, 167 (2013). [16] J. Wang, B. Lian, and S.-C. Zhang, arXiv:1409.6715 (2014). [17] J. Wang, B. Lian, and S.-C. Zhang, Phys. Rev. B 89, 085106 (2014). [18] Y. Feng et al., arXiv: 1503.04569 (2015). [19] X. Kou et al., arXiv: 1503.04150 (2015). [20] J. Wang, B. Lian, and S.-C. Zhang, Phys. Rev. Lett. 115, 036805 (2015). [21] J. Zhang, C.-Z. Chang, Z. Zhang, J. Wen, X. Feng, K. Li, M. Liu, K. He, L. Wang, X. Chen, Q.-K. Xue, X. Ma, 5 and Y. Wang, Nature Commun. 2, 574 (2011). [22] H. Ji, R. A. Stokes, L. D. Alegria, E. C. Blomberg, M. A. Tanatar, A. Reijnders, L. M. Schoop, T. Liang, R. Pro- zorov, K. S. Burch, N. P. Ong, J. R. Petta, and R. J. Cava, J. Appl. Phys. 114, 114907 (2013). [23] L. D. Alegria, H. Ji, N. Yao, J. J. Clarke, R. J. Cava, and J. R. Petta, Appl. Phys. Lett. 105, 053512 (2014). [24] He Ke, private communications. With 0.3 < x < 0.46, the topological nontrivial properties in CBST disappear due to high Cr concentration. [25] C.-Z. Chang, W. Zhao, D. Y. Kim, H. Zhang, B. A. Assaf, D. Heiman, S.-C. Zhang, C. Liu, M. H. W. Chan, and J. S. Moodera, Nature Mater. 14, 473 (2015). [26] J. Wang, B. Lian, H. Zhang, Y. Xu, and S.-C. Zhang, Phys. Rev. Lett. 111, 136801 (2013). [27] H. Zhang, C.-X. Liu, X.-L. Qi, X. Dai, Z. Fang, and S.-C. Zhang, Nature Phys. 5, 438 (2009). [28] Y. Xu, I. Miotkowski, C. Liu, J. Tian, H. Nam, N. Ali- doust, J. Hu, C.-K. Shih, M. Z. Hasan, and Y. P. Chen, Nature Phys. 10, 956 (2014). [29] R. Yoshimi, A. Tsukazaki, Y. Kozuka, J. Falson, K. Taka- and hashi, J. Checkelsky, N. Nagaosa, M. Kawasaki, Y. Tokura, Nature Commun. 6, 6627 (2015). [30] T. Morimoto, A. Furusaki, and N. Nagaosa, arXiv: 1505.06285 (2015). 

1512.04289

1

1512

Spatially dispersive dynamical response of hot carriers in doped graphene

We study theoretically wave-vector and frequency dispersion of the complex dynamic conductivity tensor (DCT), $\sigma_{lm}(\mathbf{k}, \omega)$, of doped monolayer graphene under a strong dc electric field. For a general analysis, we consider the weak ac field of arbitrary configuration given by two independent vectors, the ac field polarization and the wave vector $\mathbf{k}$. The high-field transport and linear response to the ac field are described on the base of the Boltzmann kinetic equation. We show that the real part of DCT, calculated in the collisionless regime, is not zero due to dissipation of the ac wave, whose energy is absorbed by the resonant Dirac quasiparticles effectively interacting with the wave. The role of the kinematic resonance at $\omega = v_F |{\bf k}|$ ($v_{F}$ is the Fermi velocity) is studied in detail taking into account deviation from the linear energy spectrum and screening by the charge carriers. The isopower-density curves and distributions of angle between the ac current density and field vectors are presented as a map which provides clear graphic representation of the DCT anisotropy. Also, the map shows certain ac field configurations corresponding to a negative power density, thereby it indicates regions of terahertz frequency for possible electrical (drift) instability in the graphene system.

cond-mat.mes-hall

cond-mat

Spatially dispersive dynamical response of hot carriers in doped graphene S. M. Kukhtaruk,1 V. A. Kochelap,1 V. N. Sokolov,1 and K. W. Kim2 1Department of Theoretical Physics, V. Lashkaryov Institute of Semiconductor Physics, NASU, Pr. Nauki 41, Kiev 03028, Ukraine and 2Department of Electrical and Computer Engineering, North Carolina State University, Raleigh, NC 27695-7911, USA Abstract We study theoretically wave-vector and frequency dispersion of the complex dynamic conductiv- ity tensor (DCT), σlm(k, ω), of doped monolayer graphene under a strong dc electric field. For a general analysis, we consider the weak ac field of arbitrary configuration given by two independent vectors, the ac field polarization and the wave vector k. The high-field transport and linear re- sponse to the ac field are described on the base of the Boltzmann kinetic equation. We show that the real part of DCT, calculated in the collisionless regime, is not zero due to dissipation of the ac wave, whose energy is absorbed by the resonant Dirac quasiparticles effectively interacting with the wave. The role of the kinematic resonance at ω = vFk (vF is the Fermi velocity) is studied in detail taking into account deviation from the linear energy spectrum and screening by the charge carriers. The isopower-density curves and distributions of angle between the ac current density and field vectors are presented as a map which provides clear graphic representation of the DCT anisotropy. Also, the map shows certain ac field configurations corresponding to a negative power density, thereby it indicates regions of terahertz frequency for possible electrical (drift) instability in the graphene system. 5 1 0 2 c e D 4 1 ] l l a h - s e m . t a m - d n o c [ 1 v 9 8 2 4 0 . 2 1 5 1 : v i X r a 1 I. INTRODUCTION One of fundamental aspects of the graphene physics is the electronic band dispersion which is linear ε(p) = ±vF p in the vicinity of the Dirac points for a wide region of energy ε (< 1 eV);1,2 here vF ≃ 108 cm/s is the Fermi velocity in graphene, p is the electron momentum, and p = p. Hence, the electro-physical properties essentially determined by the energy band structure will be different for graphene and conventional semiconductors with the standard energy band (Ge, Si, GaAs, etc).3 A lot of attention has been paid to the investigation of the electrical conductivity in graphene both experimentally and theoretically. A special focus is given to the investigation of electron dynamic response to electrical perturbations which can be varying in both space and time.4 -- 13 Generally, spatial dispersion for arbitrary wave vectors should be included in the description of the dynamic conductivity which, in this case, due to the existence of pref- erential direction given by the wave vector k is a tensorial quantity, σlm(k, ω), the dynamic conductivity tensor (DCT) . The detailed knowledge of wave-vector dispersion of DCT is critically important for use in tunable THz plasmonic and metamaterial nanodevices based on excitation modes in different graphene structures.14,15 The examples are patterned struc- tures such as periodic micro-ribbon arrays,16,17 antidots arrays,18 grating-gated graphene- based HEMTs,19 and other semiconductor plasmonic systems.20 -- 22 Also, a scheme which avoids patterning has been proposed for coupling light into graphene plasmons by forming a tunable optical grating with electrically generated surface acoustic waves.23 Recently, the novel nanoscope techniques have been developed where electric fields (variable in time and space) can be excited at the length scale of the order of tens of nanometers, as well as the methods which allow to explore the excited fields and plasmons on these time and length scales, including ultrafast resonance of Dirac plasmons in graphene and similar systems.24 -- 31 Another important feature associated with the wave-vector dispersion in such systems is that the wave vector k and vector of the locally excited field are not parallel as it takes place, for example, in the case of electrostatic perturbation, where they are always collinear according to the Poisson equation. Since the discovery of graphene,32 the behavior of Dirac quasiparticles in external elec- tric fields is the subject of intense theoretical and experimental studies. Such studies are stimulated by both the exceptional electronic properties of graphene3 and the perspective of 2 various ultrahigh-speed applications in future integrated-circuit technology,33 high-frequency electronics including terahertz (THz) frequencies,4,5,34 and optoelectronics.35 In early re- search, most transport studies in graphene have been carried out under low-bias conditions to probe the intrinsic electronic properties, although high-field behavior of the nonequilib- rium carriers is more appropriate for practical graphene-based device operation.36 -- 41 The wave-vector k and frequency ω dispersion of the graphene DCT σlm(k, ω) was considered in the collisionless regime9 and in the relaxation-time approximation13 for a weak electric field, which did not break the equilibrium distribution of Dirac quasiparticles over energy in the bands. In the high-field limit, the charged carrier transport is described, in particular, in terms of the hot electrons and/or holes in graphene.37,42 -- 45 In a general case of arbitrary wave vector k and frequency ω, the dispersion of DCT becomes important for such k and ω which obey the inequalities kvτp & 1, ωτp & 1, where v = ∂ε/∂p and τp are the carrier band velocity and the momentum relaxation time, respec- tively; v = v and k = k. The collisionless limit corresponds to the conditions kvτp ≫ 1, ωτp ≫ 1 , (1) which imply many oscillations of the ac field on the space and time scales given by the carrier mean free path lp = vτp and the characteristic time τp, respectively. These allow to neglect a small collision integral, compared to the rest of terms in the Boltzmann kinetic equation, and to calculate the DCT, σlm(k, ω), in the straightforward manner. In this regime, the space and time dispersion results from the fact that free movement of the carriers is affected by all values of the ac field along the trajectories of movement rather than its local and instantaneous value. In this paper, we investigate theoretically the space and time dispersion of DCT of Dirac quasiparticles in the doped single-layer graphene system under a high dc electric field. For generality, we consider the perturbing ac field to be of arbitrary configuration, that is the electric field vector and the wave vector k are treated as the two independent vectors. The theoretical model is based on the semiclassical approach which is valid for k < kF and ω < εF /, where kF and εF are the Fermi wave vector and energy, respectively. The latter inequality allows us to exclude the interband transitions between the valence and conduction bands which assumes frequency ω < 2εF /. The specific behavior of real, σ′ lm(k, ω), and imaginary, σ′′ lm(k, ω), parts of the complex 3 lm(k, ω) + iσ′′ DCT σlm(k, ω) = σ′ lm(k, ω) is tightly connected to the well-known kinematic resonance in the collisionless plasma.46 At the resonance ω = k · v, the phase velocity of the wave in the direction of propagation ω/k is equal to the carrier velocity v. For graphene, due to the linear energy spectrum, the resonance is determined by a linear dependence of frequency on wave vector, ω = vF k, which is to some extent similar to the Dirac cone ε/ = vF q; here q is the Dirac quasiparticle wave vector and q = q. We will show that for high frequencies ω > vF k [fast waves, (ω/k) > vF ] the real part of DCT becomes equal to zero. In the region of low frequencies ω < vF k [slow waves, (ω/k) < vF ], the space dispersion is large and the real part of DCT takes definite nonzero values. We note that such specific dispersion of σ′ lm(k, ω) is due to the intrinsic property of the graphene energy spectrum and totaly different from standard 2D electron systems with the parabolic energy spectrum. This paper is organized as follows. We introduce our model and describe the theoretical approach in Sec. II and use them to study the real part (Sec. III) and the imaginary part (Sec. IV) of the complex DCT. Also in Sec. III, we provide a graphic representation of the DCT anisotropy in the form of a map for isopower-density curves and distributions of angle between the ac current density and electric field vectors. In Secs. V and VI, we discuss the role of the energy band nonlinearity and screening by the charge carriers, respectively. We present conclusions in Sec. VII. In the Appendix, we describe details of the evaluation of integrals in the expressions for the DCT. II. FORMULATION OF THE MODEL In our study, we consider doped graphene systems in which the electron (hole) density is controlled by an external gate voltage and/or chemical doping.3 As far as the graphene spectrum is symmetric relative to Dirac electron and hole quasiparticles, we will refer, for definiteness, to the electron-doped graphene with the electron density n0. We assume the Fermi level is well above the Dirac point for such doping which allows us to neglect the contribution of holes. For the calculation of DCT, we consider a graphene monolayer sheet lying in the XOY plane (z = 0), driven by a strong dc electric field E0 and a weak ac electric field E1(r, t) = E1ei(k·r−ωt) with E1 ≪ E0; r = (x, y) is the in-plane coordinate vector and t is the time. In the chosen coordinate system, the wave vector k = (k, 0) has only one nonzero component kx = k (Fig. 1). The complex amplitude of the ac field (and 4 Figure 1: Geometry of the problem. Graphene single layer (on a substrate) and coordinate system show in-plane basic vectors and characteristic angles appearing in the problem: electron drift velocity vd, amplitudes of ac electric field E′ 1 and current density j′ 1 with wave vector k and frequency ω. The angles α, β, γ are marked by arcs. other perturbation quantities) is frequency and wave-vector dependent, E1 = Eωk 1 . For shortness of notations, we will omit these subindices throughout the text except Sec. VI. For linearization purposes, we use the usual plane-wave (ω, k) representation. However, where it is necessary [see Eq. (9)], we will utilize the substitution ω → ω + iδ (δ → +0) which corresponds to a slow turning on the ac field at t = −∞.46 Notice that the weak ac electric field can correspond to the electric field of a propagating electromagnetic wave, the effective field of electrostatic or deformation potentials, or a weak inhomogeneity of different origin. In general, it can depend on the z-coordinate as well, E1(r, z, t). However, in the simplest approach when the graphene is modeled as an infinitesimally thin layer in the z direction (a delta-layer), the main expressions derived below contain the field value at z = 0. We designate the in-plane value of the field as E1(r, t) = E1(r, z = 0, t). To describe the linear response of nonequilibrium Dirac quasiparticles (electrons) to the ac electric field, we result from the Boltzmann equation ∂f ∂t + vF p p ∂f ∂r − eE ∂f ∂p = I{f} , (2) where E = E0 + E1(r, t) is the lateral electric field, e = e is absolute value of the electron charge, and I = IeL + Iee is the collision integral which includes different scattering mecha- nisms such as electron-phonon and impurity scattering (IeL) and electron-electron scattering 5 (Iee). The distribution function can be represented as f (r, p, t) = f0(p) + f1(r, p, t), where f0(p) is time-independent and space-uniform distribution function of the electrons in the dc field E0 and f1(r, p, t) is a small addition due to the ac field E1(r, t). Letting a harmonic dependence on the coordinate and time for all quantities [e.g., f1(r, p, t) = f1(p)ei(k·r−ωt), etc] and utilizing the usual procedure of linearization in Eq. (2), we obtain the equations for the steady-state distribution function f0(p) eE0 ∂f0 ∂p + I{f0} = 0 and the time-dependent addition f1(r, p, t) ∂f1 ∂t + vF p p ∂f1 ∂r − eE1 ∂f0 ∂p − eE0 ∂f1 ∂p = I{f0, f1} . (3) (4) For large electron densities, the e − e scattering time appears to be shorter than all other scattering times, so that the leading term in Eq. (3) is determined by the e − e collision integral. In this regime, rapid e − e collisions establish a displaced Fermi-Dirac distribution function f0(p) = (cid:20)exp(cid:18)vF p − vdp − εF kBTe (cid:19) + 1(cid:21)−1 , (5) which satisfies the equation Iee{f0} = 0 and thus can be regarded as an approximate solution to Eq. (3).47 In Eq. (5), vd is the electron drift velocity which is collinear with the dc field E0, and kB is the Boltzmann constant. The three parameters, the Fermi energy, εF , the average electron drift velocity, vd, and the electron temperature, Te, depend on E0 and describe the steady state of the nonequilibrium graphene system in the dc electric field E0. These parameters are found from the set of coupled equations which are pertinent moments of the kinetic equation (3) and represent balance equations for the carrier density, the electron momentum, and the electron energy.37,42 -- 44 Integrating Eq. (5) over dΓp = gd2p/(2π)2, where g = gsgv = 4 is the electron spin (gs = 2) and valley (gv = 2) degeneracy factor, we obtain g (2π)2 Z d2pf0(p) = n0 . (6) For strongly degenerate electrons, the integral can be performed analytically and the asymp- totic relationship can be found for the Fermi energy εF = kF vF (1 − v2 F )3/4, where kF = √πn0 is the Fermi wave vector in the absence of the dc electric field. To avoid con- siderable decrease in εF with increasing drift velocity, we exclude from our consideration d/v2 6 such values of vd = vd (the dc field strength E0) for which the relativistic factor vd/vF approaches unity. In the dc transport, the momentum and energy gained by electrons from the electric field are balanced by losses due to electron scattering on the impurities and phonons. The results of numerical calculations of the field dependencies of vd = vd(E0) and Te = Te(E0) can be found elsewhere (see, for example, Ref. 37). In this work, we utilize vd and Te as the known model parameters for the analysis of the DCT. An approximate analytical solution to Eq. (4) can be found in the collisionless regime by neglecting the collision integral I{f0, f1}. This approach is valid for the range of frequencies and wave vectors within the criteria given in (1). Also, we omit the last term on the left-hand side of Eq. (4) to obtain f1(p) = ieE1 ω − vF k · p/p ∂f0 ∂p . (7) The latter assumption imposes restrictions on the dc field strength according to the following inequalities (eE0/ω) ≪ ¯p, (eE0/k) ≪ vF ¯p. Physically, they mean that the momentum and energy acquired by an electron on the time and space scales of the order of the ac field period (∼ 1/ω) and wavelength (∼ 1/k) are much less than the average momentum ¯p and energy vF ¯p, respectively. The ac current density j1(r, t) = j1ei(k·r−ωt) is related to the ac field E1(r, t) through the DCT j1 = σE1. Substituting f1(p) from Eq. (7) into the current density we obtain the DCT as j1 = − eg (2π)2 Z d2p vF p p f1(p) , σlm = −i e2gvF (2π)2 Z d2p ω − vF k · p/p + iδ pl p ∂f0 ∂pm . (8) (9) Under the integral, an infinitesimally small imaginary iδ in the denominator corresponds to a slow turning on the ac field at t = −∞ and results in the regularization of the integral at δ → +0.46 As it can be seen from Eq. (9), there is no singularity in the integrand for frequency ω > vF k. This is the key difference of the pole ω = vF k · p/p which takes place for the graphene from a similar pole ω = k· p/m characteristic for electron systems with the standard (quadratic) energy spectrum ε = p2/2m. Thus, the equation ω = vF k · p/p has solutions only at ω < vF k, in contrast to the equation ω = k · p/m which has solutions at any ω and k. This results in that the real part of the DCT is not zero only for frequencies ω < vF k. 7 The dynamic conductivity tensor σlm is a complex tensor of the second order. It is given by the eight functions which depend on the six parameters: ω, k, and γ which are the prop- agating wave characteristics, and n0, vd, and Te which are the dc transport characteristics. With the steady-state distribution function (5), the integrals in Eq. (9) can be performed analytically. In particular, for vd = 0, the function (5) represents the so-called electron temperature approximation which is widely used in the theory of the high-field carrier trans- port.47 In this case, the off-diagonal components of tensor σ are zero and obey the evident symmetry relations σxy = σyx = 0. The same occurs for sin γ = 0, i.e., for the drift velocity vd and the wave vector k to be collinear (Fig. 1). Below we analyze the behavior of the real σ′ lm and imaginary σ′′ lm parts of σlm separately. First, we address the real part of the DCT. III. REAL PART OF DYNAMIC CONDUCTIVITY We now turn to explicit evaluation of the integrals in Eq. (9). In this and in the following Sec. IV, we use the linear energy dispersion for graphene, and later (Sec. V) we take into account deviation from the linear band structure to analyze its influence on the DCT. It is convenient to introduce the dimensionless variables Ω = ω/ω0 and K = vF k/ω0 where the normalization frequency is ω0 = e2gkBTe/(2π2vF ), and the dimensionless parameters Vd = vd/vF (Vd < 1) and EF = εF /kBTe. After substitution of the distribution function (5) into Eq. (9), the integration therein can be performed analytically utilizing the complex plane and the residue theory method.48 The calculation details are described in the Appendix where the DCT is given by Eq. (A4). Since the explicit form of the general expression for σlm in Eq. (A4) is rather complicated, it is difficult to separate analytically the real and imaginary parts of σlm from this equation. We will utilize Eq. (A4) in Sec. IV for the numerical calculation and analysis of the imaginary part of σlm. We notice that it is more simple to separate σ′ lm in Eqs. (A1), where the integration is over the polar angle θ. Indeed, in using the well-known relation (x + iδ)−1 = P(1/x)− iπδ(x) [P(1/x) denotes the principal part of the integral and δ(x) is the delta function], it is clear that the integral with the delta 8 function is easily evaluated, and the explicit analytical expressions for σ′ lm are obtained σ′ xx = BΩ √K 2 − Ω2 (cid:18)Ω − VxK A2 − σ′ yy = B(cid:18)√K 2 − Ω2 − VyK √K 2 − Ω2 (cid:18)√K 2 − Ω2 − VyK BΩ A2 A2 − − + − + A2 A2 + Ω − VxK √K 2 − Ω2 + VyK √K 2 − Ω2 + VyK A2 + Ω − VxK + (cid:19) , (cid:19) , (cid:19) , + (cid:19) ; − A2 σ′ yx = B(cid:18)Ω − VxK A2 − σ′ xy = here A± = K − VxΩ ± Vy√K 2 − Ω2, ln (eEF + 1) θ(K − Ω), vF 2 B = (10) (11) Vx = Vd cos γ, Vy = Vd sin γ, and θ(x) is the Heaviside step function. We have verified that the real part of the DCT obtained in such manner coincides with that derived from Eq. (A4) in the Appendix. It follows from Eqs. (10) that σ′ yy is nonnegative while the other components can take both signs. It is also evident that σ′ lm = 0 at Ω > K; if the frequency approaches to the resonance Ω → K, then σ′ yy approach to zero ∼ (K − Ω)1/2. It is seen that if sin γ = 0 or Vd = 0, then Vy = 0 and A+ = A− so that the real part of DCT becomes symmetrical tensor, with σ′ yx = 0. In the analysis of xy go to infinity ∼ (K − Ω)−1/2, whereas σ′ xx and σ′ yx and σ′ xy = σ′ anisotropy properties of σ′ lm, it is instructive to consider the following four particular cases: 1. vd = 0; 2. vdx 6= 0, vdy = 0; 3. vdx = vdy 6= 0; and 4. vdx = 0, vdy 6= 0. For numerical calculations, we choose the set of parameters: K = 1.5, Vd = 0.4, n0 = 2 × 1012 cm−2, and Te = 400 K (kBTe ≃ 34.5 meV).37 With these parameters, we estimate ω0 ≃ 7.3 × 1013 s−1, kF ≃ 2.5 × 106 cm−1, εF ≃ 153 meV (Vd = 0), and εF ≃ 131 meV (Vd = 0.4); k ≃ 1.1 × 106 cm−1, λ = 2π/k ≃ 57 nm, and the resonance frequency (Ω = K) ν ≃ 17 THz. 1. In the first case, Vd = 0 so that there is no drift of the electron system as a whole in the dc electric field (the electron temperature Te can be higher than the lattice temperature). The diagonal components in Eqs. (10) are expressed as xx = vF ln(eEF + 1) σ′ Ω2 yy = vF ln(eEF + 1) σ′ K 2√K 2 − Ω2 √K 2 − Ω2 K 2 θ(K − Ω), θ(K − Ω); (12) 9 xx yy s, s, s, s, xx yy F v / m , l s a (a) (b) b W Figure 2: (Color online) (a) The real part of normalized DCT, σ′ lm/vF , as a function of frequency (in units of ω0) calculated at the dimensionless wave vector K = 1.5 for Vd = 0 (solid and dashed curves) and Vd = 0.4 (γ = 0◦) (dashed-dotted and dashed-double dotted curves). Different curves correspond to different diagonal components of σ′ lm (σ′ xy = σ′ yx = 0). (b) Distributions of angle β on the plane (Ω, α) corresponding to curves from (a) for Vd = 0.4; solid curves (β = 90◦) indicate regions of possible instability. yx = 0. The frequency dispersion of σ′ σ′ xy = σ′ vector, is shown in Fig. 2(a) with the solid (σ′ lm, calculated at a fixed value of the wave xx) and dashed (σ′ yy) curves. Note that the real part of DCT is not negative in the whole region of considered frequencies. In the limit of Ω → K, the longitudinal component σ′ component σ′ xx demonstrates a divergence, and the transverse yy approaches zero. 2. In the second case, the drift velocity is finite (Vd = 0.4) and its direction is collinear 10 with the wave vector k (sin γ = 0,Vx = Vd, Vy = 0). Then we obtain Ω(Ω − VxK) (K − VxΩ)2 θ(K − Ω), √K 2 − Ω2 (K − VxΩ)2 θ(K − Ω). yy = vF ln (eEF + 1) σ′ √K 2 − Ω2 vF ln (eEF + 1) σ′ xx = (13) The calculated frequency dependencies from Eqs. (13) are shown in Fig. 2(a) with dashed- xx) and dashed-double dotted (σ′ dotted (σ′ yy) curves. The frequency behavior of diagonal components at Ω → K is similar to the previous case (Vd = 0). However, the occurrence of a finite drift (Vd = 0.4) of the electron system adds new features. Namely, if the criterion for the Cherenkov radiation effect is fulfilled vd > ω/k, i.e., the drift velocity is greater than the ac wave phase velocity, then the longitudinal component σ′ xx(Ω) [Fig. 2(a), dashed- dotted curve] changes its sign at the frequency Ω = VxK = 0.65 (ν ≃ 7.5 THz). It becomes negative in the frequency domain Ω < VxK. The point of minimum (Ω = 0.37, ν ≃ 4.3 THz) corresponds to maximal increment for the Cherenkov instability in this frequency domain. xx = σ′ The transverse component is positive and, as a function of frequency σ′ yy = σ′ yy(Ω), has a maximum [Fig. 2(a), dashed-double dotted curve]. An important physical quantity associated with the real part of DCT is the power density P = hj′ 1(r, t)i averaged over the time period T = 2π/ω. For monochromatic space and time dependence of the ac field and the current density, the power density P takes the 1(r, t) · E′ form P = 1 2 (cid:2)σ′ xxE ′2 1x + (σ′ xy + σ′ yx)E ′ 1xE ′ 1y + σ′ yyE ′2 1y(cid:3) . (14) A negative sign of P means that the hot carriers do work over the ac field. In this case, the amplitude of the field of a given wave vector and frequency can be amplified. The occurrence of a (k, ω)-range in which P takes negative values can lead to electrical instability of the considered system. It is evident from Eq. (14) that P < 0 is possible if at least one of the terms in the square brackets is negative and its module value is larger than the sum of others. Alternatively, a positive value of P means dissipation of the ac field energy and stability of the considered system. Formally, considering P as an independent parameter, the equation (14) with P = const can be treated as a family of the second order curves on the plane of variables (E ′ 1x, E ′ 1y). These isopower-density curves are ellipses or hyperbolas depending on sign of the DCT components and P . The family of isopower-density curves give a descriptive geometric representation of the DCT as well as the occurrence of possible instability in the 11 considered system. We demonstrate this by letting sin γ = 0 (σ′ xy = σ′ yx = 0). Then Eq. (14) is written in the canonical form σ′ xx 2P E ′2 1x + σ′ yy 2P E ′2 1y = 1. (15) For positive diagonal components (σ′ xx, σ′ specifies a set of ellipses of constant power density. If the signs of σ′ yy > 0), P also should be positive and Eq. (15) yy are different, xx and σ′ then it is possible to realize the power density of both signs (P ≷ 0). In this case, the corresponding second order curve is a hyperbola. In particular, P = 0 determines asymptotes of the set of hyperbolas, so that the curves corresponding to P > 0 and P < 0 are located on different sides of these asymptotes. Thus, in going from situation with a positive P to situation with an arbitrary sign of P (i.e., from stability to instability), the set of ellipses is transformed into the set of hyperbolas. It is worth noting that for a rank-2 symmetrical tensor σ′ lm the dissipative part j′ 1 of the current density is normal to the isopower-density curve. This follows from the direct comparison of the normal vector n = ∂P (E ′ 1x, E ′ 1y)/∂E′ 1, calculated at a fixed ac electric field E′ 1 with the function P (E ′ 1x, E ′ 1y) given by Eq. (14), and the vector j′ 1. Thus, each isopower- density curve plotted on the plane (E ′ which provides such P, as well as the direction of the current density j′ 1y) for a given P specifies the field E′ 1y) 1 corresponding with 1. For a general case of sin γ 6= 0 all components of the DCT are nonzero. In this 1 does not coincide with lm is not a symmetrical tensor and the direction of vector j′ the field E′ 1 = (E ′ 1x, E ′ case, σ′ 1x, E ′ the direction of normal vector n. Another useful characteristic of induced anisotropy associated with the DCT is the angle β between vectors j′ 1 and E′ 1. It is given by cos β = xx cos2 α + (σ′ σ′ xx cos α + σ′ [(σ′ xy + σ′ yx) cos α sin α + σ′ yx cos α + σ′ yy sin2 α yy sin α)2]1/2 , xy sin α)2 + (σ′ (16) where α is the angle between the wave vector k (axis OX) and the ac field E′ 1 (Fig. 1). The numerator of the fraction in Eq. (16) is proportional to the power density P . It follows that if the current density j′ 1 corresponds to positive P > 0, the angle β ∈ (0, 90◦); for P < 0 the angle β ∈ (90◦, 180◦). If the power density is zero, then β = 90◦. In the case of stability, the ac current flows along the ac field under the angle β < 90◦; in the case of instability, it flows against the ac field under the angle β > 90◦. The angle β characterizes the property of the DCT to turn the current density j′ 1 relatively to the ac field E′ 1. Figure 12 2(b) shows distributions of β on the plane (Ω, α) corresponding to curves with dots from Fig. 2(a). Inside the two regions (bounded with the black curves), which are determined by the condition β = 90◦, the angle β ∈ (90◦, 180◦]. Hence, in these regions the considered system may be electrically unstable. Such regions can exist only at frequencies Ω < VxK. It is also seen that there are regions of very small β where the tensor σ′ lm reduces to a scalar. Such representation of the DCT with plots of the angle β is very demonstrative; it can be utilized for any other second rank tensor, for example, the dielectric permittivity tensor, etc. In Fig. 3, we show the curves of a constant power density [Eq. (15)], which correspond to the curves from Fig. 2(a) at Vd = 0.4, calculated at different frequency. The curves which are more distant from the coordinate origin correspond to bigger values of P. The arrows show vector field of the ac current density [in coordinate axes (j1x, j1y)] which direction is normal to the corresponding isopower-density curve. The arrow length reflects relative value of the ac current density. Figure 3(a) corresponds to σ′ xx < 0 (Ω = 0.3); therefore, the curves are hyperbolas. A negative value of P associated with σ′ xx < 0 is responsible for the Cherenkov radiation in the region of frequencies Ω < KVd cos γ. Figure 3(b) corresponds to σ′ xx ≈ σ′ figures that the angle β can substantially be changed with changing the ac field frequency. yy > 0 (Ω = 1.22); therefore, the curves in this figure are circles. It is seen from these 3. In the third case, the drift velocity is finite (Vd = 0.4) and its direction is determined by the angle γ = 45◦ (Fig. 1). In this case Vx = Vy 6= 0 so that the off-diagonal components of σ′ lm are not a trivial zero. As it was mentioned above, a negative power density (P < 0) can be realized if the second term in the square brackets of Eq. (14) is negative and its module is larger than the rest of terms. This is possible for (σ′ xy + σ′ yx) < 0 with E1xE1y > 0, or for (σ′ xy + σ′ yx) > 0 with E1xE1y < 0. For the former, polarization of the ac field is such that both in-plane components of the field have the same sign, while for the latter they have opposite signs. In Fig. 4(a), we present the components σ′ lm, as well as the sum In Fig. 4(b), we show distributions of the angle (σ′ xy + σ′ yx), as functions of frequency. β. Similar to Fig. 2(b), there are wide regions of frequency (Ω < VxK), adjoining to the left vertical coordinate axis, where instability associated with the Cherenkov effect can take place. In addition, we revealed narrow regions at high frequencies (Ω → K), adjoining to the right vertical coordinate axis, which occurrence is due to a negative sum of the off-diagonal components σ′ lm. In Fig. 5, we show the isopower-density curves calculated at different frequencies with 13 Figure 3: (Color online) Lines of constant power density (P = const ≷ 0) corresponding to curves from Fig. 2(a) calculated at different frequency for Vd = 0.4: P > 0 (solid), P < 0 (dashed), and P = 0 (straight solid). Arrows represent vector field of the ac current density (relative units). (a) Ω = 0.3 (σ′ xx < 0, σ′ yy > 0) and (b) Ω = 1.22 (σ′ xx ≈ σ′ yy > 0). the same numerical parameters used in Fig. 4(a). The curves correspond to P > 0 (solid), P < 0 (dashed), and P = 0 (straight solid lines). The arrows represent vector field of the ac current density. The curves in Fig. 5(a) were calculated with σ′ xx < 0 and (σ′ xy + σ′ xx > 0 and (σ′ xy + σ′ yx) < 0 yx) < 0 at at Ω = 0.3. The curves in Fig. 5(b) were calculated with σ′ Ω = 1.49, i.e., at Ω → K. 4. In the forth case, the drift velocity is finite (Vd = 0.4) and its direction is perpendicular to the k vector, γ = 90◦ (Fig. 1). In this case Vy = Vd and Vx = 0 which excludes the possibility for the Cherenkov instability in the considered system. To illustrate more clearly 14 yy xx s, s, s, s, xy+s, s, yx xy yx F v / m , l s a (a) (a) (b) b W Figure 4: (Color online) (a) The real part of normalized DCT, σ′ lm/vF , as a function of frequency (in units of ω0) calculated at the dimensionless wave vector K = 1.5 (γ = 45◦) for Vd = 0.4. Different curves correspond to different components of σ′ lm. (b) Distributions of angle β on the plane (Ω, α) corresponding to curves from (a); solid curves (β = 90◦) indicate regions of possible instability. the influence of the off-diagonal term (σ′ xy + σ′ yx) on the sign of P , we compare all nonzero components σ′ be realized due to the term (σ′ lm calculated as a function of frequency in Fig. 6(a). It follows that P < 0 may yx) which is the only negative at high frequency (Ω → K); the rest of the terms in Eq. (14) are definite positives at the considered frequencies. The xy +σ′ corresponding distributions of angle β are presented in Fig. 6(b), which demonstrates at what frequency the negative power density can take place. It is seen that the regions of possible instability (P < 0) are slightly wider compared to the similar regions in Fig. 4(b). In Fig. 7, we show the isopower-density curves corresponding to curves from Fig. 6(a) calculated at different frequencies. The curves demonstrate a qualitatively different behavior. At Ω = 0.3, we get P > 0 as both σ′ xy + σ′ yx > 0 [Fig. 7(a)]. At Ω = 1.48, we can get P ≷ 0 as σ′ xx > 0 but (σ′ yx) < 0 [Fig. 7(b)]. The results show that with xx > 0 and σ′ xy + σ′ 15 Figure 5: (Color online) Lines of constant power density (P = const ≷ 0) corresponding to curves from Fig. 4(a) calculated at different frequency: P > 0 (solid), P < 0 (dashed), and P = 0 (straight solid). Arrows represent vector field of the ac current density (relative units). (a) Ω = 0.3, σ′ xx < 0, σ′ xy + σ′ yx > 0 and (b) Ω = 1.49 , σ′ xx > 0, σ′ xy + σ′ yx < 0. γ = 90◦ a negative power density is achieved in the case when the Cherenkov criterion is not fulfilled, i.e., due to the negative off-diagonal term alone (σ′ yx) < 0. We have also yx) may become negative at Ω → K in a wide region of the xy + σ′ xy + σ′ revealed that the term (σ′ angle γ ∈ (0◦, 180◦). Thus, there exists a certain set of the specific parameter values at which the negative power density (P < 0) may be realized. This comprises three different cases with σ′ xx < 0 yx) < 0 in the limit of high in the Cherenkov region of frequencies (Fig. 2); with (σ′ xy + σ′ 16 yy xx s, s, s, s, s, xy+s, yx xy yx F v / m , l s a (a) (b) b W Figure 6: (Color online) (a) The real part of normalized DCT, σ′ lm/vF , as a function of frequency (in units of ω0) calculated at the dimensionless wave vector K = 1.5 (γ = 90◦) for Vd = 0.4. Different curves correspond to different components of σ′ lm. (b) Distributions of angle β on the plane (Ω, α) corresponding to curves from (a). frequencies Ω → K (Fig. 6); and with both σ′ yx) < 0 simultaneously (Fig. 4). We also found that if γ ∈ (180◦, 360◦), then one may achieve P < 0 at the positive off-diagonal term (σ′ 1y < 0 in the limit of Ω → K. yx) > 0 due to the ac field polarization with E ′ xx < 0 and (σ′ xy + σ′ xy + σ′ 1xE ′ To conclude this section, we note that so far we considered the ac field of arbitrary configuration given by the two vectors E′ 1 was not necessarily parallel to the wave vector k. Let us briefly discuss the case when the effective ac field is of an 1 and k, where E′ electrostatic origin. Then, the Fourier components of the field and the electrostatic potential are connected by Eωk 1 = −ikϕ ωk 1 for the power density . In this case, E′ 1 is collinear with k (Fig. 1), and we get P = 1 2 xxk2ϕ ωk σ′ 1 2 , (17) which now replaces (14). From the comparison of (17) and (14), we conclude that in the 17 Figure 7: (Color online) Lines of constant power density (P = const ≷ 0) corresponding to curves from Fig. 6(a) calculated at different frequency. (a) Ω = 0.3 (P > 0) and (b) Ω = 1.48 (P ≷ 0). P > 0 (solid), P < 0 ( dashed), and P = 0 (straight solid). Arrows represent vector field of the ac current density (relative units). case of ac electrostatic field the negative power density (P < 0) can be realized only on the Cherenkov criterion Ω < KVd cos γ. IV. IMAGINARY PART OF DYNAMIC CONDUCTIVITY In this section, the frequency and wave vector behavior of the imaginary part of DCT is analyzed by separating σ′′ lm in Eq. (A4) (Appendix). In the analysis, we will use the same exemplary cases of particular parameter values considered in the previous section. In the 18 absence of the electron drift (case 1, Vd = 0) and for small space dispersion (k → 0), the DCT reduces to pure imaginary scalar σlm = iσ′′δlm with σ′′(k = 0, ω) = e2kBTe π2ω ln(eεF /kB Te + 1) . (18) Hence, the dependence of σ′′(k = 0, ω) on frequency is similar to the Drude-Lorentz conduc- tivity at high frequencies (ωτp ≫ 1). A similar result was obtained for intraband contribution into σ′′(k, ω) in Ref. 9 within a quantum approach for low electric field and at finite lattice temperature T (ω < kBT ), when the electron distribution over the energy is equilibrium one [Eq. (10) in Ref. 9]. The expression (18) extends this result to the region of hot elec- trons in high electric fields, when the electron temperature Te is higher than the lattice temperature. If space dispersion is large, then the imaginary part of DCT reads vF ln (eEF + 1) σ′′ xx = − K 2 vF ln (eEF + 1) Ω(cid:18)1 − √Ω2 − K 2 (cid:19) , Ω θ(Ω − K) √Ω2 − K 2 θ(Ω − K)(cid:17) . (19) (cid:16)Ω − σ′′ yy = K 2 From these expressions, we see that the diagonal elements σ′′ yy are linear functions of Ω at Ω < K and take nonzero values at Ω > K. In the limit Ω → ∞, we obtain the asymptotic behavior σ′′ xx(Ω) yy = vF ln (eEF + 1)/2Ω ∼ Ω−1. At Ω = K, the function σ′′ xx and σ′′ xx = σ′′ has an infinite discontinuity, whereas σ′′ yy(Ω) is a continuous function of Ω. In a general case, the corresponding integrals are evaluated by the theory of residues (see the Appendix). The results of numerical calculations of the imaginary part of DCT as a function of frequency are shown in Fig. 8. The parameter values used in the calculations are for Fig. 8(a) the same as for Fig. 2(a) [case 1 (Vd = 0) and case 2 (sin γ = 0, Vd = 0.4)]; for Fig. 8(b) the same as for Fig. 4(a) [case 3 (γ = 45◦, Vd = 0.4)]. As mentioned above, the off-diagonal elements of σkl are zero for the considered cases 1 and 2. The diagonal elements k=l, unlike the real part σ′ σ′′ asymptotic behavior of σ′′ k=l, are not zero at Ω > K. It is seen from Fig. 8(a) that the yy) at Ω → K and Ω → ∞ for a finite electron drift are similar to the case of Vd = 0. The occurrence of the electron drift leads to an increase in σ′′ k=l near the point Ω = K. The transversal diagonal component σ′′ in a frequency region Ω < K, starting from zero frequency. For comparison, in Fig. 8(b), we yy takes negative values xx (and σ′′ show the results obtained for γ = 45◦, in which case the off-diagonal components σ′′ σ′′ yx are not zero. The frequency behavior of the diagonal components σ′′ shown in Fig. 8(a) for Vd 6= 0. The component σ′′ xy and k=l is similar to that xy has an infinite discontinuity, approaching 19 (a) (b) F , , v / m s l F , , v / m s l xx xx yy s,, s,, s,, s,, yy xx yy s,, s,, s,, s,, xy yx W Figure 8: (Color online) The imaginary part of normalized DCT, σ′′ lm/vF , as a function of frequency (in units of ω0) calculated at different values of the dimensionless wave vector. (a) K = 1.5 (γ = 0◦), xy = σ′′ σ′′ yx = 0; Vd = 0 (solid and dashed curves) and Vd = 0.4 (dashed-dotted and dashed- double dotted curves). (b) K = 1.5 (γ = 45◦), Vd = 0.4. Different curves correspond to different components of σ′′ lm. −∞ when Ω → K + 0. The frequency behavior of the off-diagonal component σ′′ to that of the diagonal component σ′′ yy. yx is similar V. ROLE OF ENERGY BAND NONLINEARITY So far, our calculations were restricted to the Dirac-spectrum approximation. In the integral of Eq. (9), the integration is over all possible values of the electron momentum p. Therefore, corrections to the Dirac spectrum may be important at certain values of the model parameters. In this section, we address the question of how the frequency and wave vector behavior of the dynamic conductivity can be changed by including corrections to the Dirac 20 cone spectrum. Recently, the dynamic and optical properties in graphene have been studied taking into account the nonlinear energy dispersion.49,50 The full energy band spectrum has been obtained in the tight-binding approximation.1,3 In using such the energy spectrum in Eqs. (5) and (9), the DCT can be calculated numerically performing the integration in Eq. (9) over the entire Brillouin zone. For the purpose of this section, it is sufficient to use a quadratic correction to the linear term ε(p) = vF p which follows from Taylor expansion of the full band spectrum close to the Dirac point, ε(p) = vF p − 3Eta2 82 sin(3ϑ)p2 . (20) Here Et ≃ 2.8 eV is the nearest-neighbor hopping energy, a ≃ 1.42A is the carbon-carbon dis- tance, and ϑ is the angle in momentum space.3 Assuming the strongest nonlinearity [letting sin(3ϑ) = 1] and using the dimensionless variables P = p/p0, p0 = kBTe/vF , the spectrum (20) can be rewritten as Ep ≡ ε(p)/kBTe = P − (1/2)α0P 2, where α0 = 3kBTeEta2/4v2 is a dimensionless parameter. For example, if we take Te = 400 K, then the numerical estimation gives α0 = 3.6 × 10−3 ≪ 1. Substituting Ep in Eq. (9), we find ∂f0 ∂pm (2π)2 Z σlm = −i e2gvF 2 F d2p ω − vF k · p/p + α0vF k · p/p0 + iδ pl p . (21) As in Sec. III, the integration is convenient to be performed in the polar coordinate system. Importantly, the denominator of the integrand in Eq. (21) contains not only the polar angle but also the electron momentum, which modifies essentially the pole of the integrand compared to the Dirac cone approximation. As a consequence, the diagonal component σ′ xx does not go to infinity at the limit ω → vF k. width at the resonance frequency. The distribution function in Eq. (5) now depends on the Instead, it takes a finite hight and parameter α0 through the modified electron spectrum, f0 = f0(p, α0). Note that f0(p, α0) cannot be normalized according to Eq. (6), because it does not approach zero when the momentum goes to infinity. Nevertheless, it is possible to evaluate an asymptotic value for the integral in Eq. (21), utilizing the small parameter α0 and expanding the function f0(p, α0) into a Taylor series f0(p, α0) = f0(p) + α0[∂f0(p, α0)/∂α0]α0=0 + · · · . Substituting this distribution function in Eq. (21), we obtain the asymptotic expansion for the DCT lm + α0σ1 lm + · · · . Further, we retain only the main (zeroth order) term σ0 σlm = σ0 ignore all higher orders in α0 of this expansion. Then, the expression for σ0 lm and lm is given in Eq. (21) with the function f0(p) given in Eq. (5). The real and imaginary parts of the 21 DCT, σlm ≃ σ0 lm, can be derived in a fashion similar to Sec. III, where they have been calculated within the Dirac cone approximation. As the derived analytical formulas are rather complicated, we demonstrate only the results of numerical computation and compare them with those obtained in the Dirac cone approximation. Using the polar coordinates (P, θ), the longitudinal component σ′ xx can be written as σ′ xx = − vF 2K Z ∞ 0 dP 1 − α0P Z 2π 0 dθ(cid:18)P ∂f0 ∂P cos2 θ − cos θ sin θ ∂f0 ∂θ (cid:19) δ(s − cos θ) , (22) where s = s/(1 − α0P ). Due to the delta function δ(s − cos θ), the angle integral in (22) is not zero if only s ≤ 1. Then, we have P Ixx 1 − α0P P Ixx 1 − α0P 2K Z Pc θ(1 − s) = (23) dP dP vF , 0 vF σ′ xx = 2K Z ∞ where Pc = (1 − Ω/K)/α0, 0 Ixx = s2 − Vxs (1 − s2)1/2 [f + 0 (1 − f + 0 ) + f − 0 (1 − f − 0 )] , (24) and f ± 0 = {exp[(1 − Vxs ∓ Vy√1 − s2)P − EF ] + 1}−1. In particular, for the case of Vd = 0, Eq. (23) becomes σ′ xx = −vF Ω2 K 3 Z Pc 0 P dP (1 − α0P )2p(1 − α0P )2 − Ω2/K 2 ∂f0 ∂P . (25) Figure 9 shows the frequency dependence of the real part of DCT calculated with the quadratic correction to the linear energy spectrum. The parameter values used for Fig. 9(a) are the same as for Fig. 2(a), and for Fig. 9(b) are the same as for Fig. 4(a). As expected, the divergence at the resonance frequency (Ω → K) is removed due to the adopted quadratic correction. The integrals in Eqs. (23) and (25) approach zero in the limit Pc → 0 (Ω → K). Thereby, we observe a normal resonance curve with the finite peak and width values. In Fig. (Vd = 0.4). 9(a), the peak values are σ′ xx/vF = 28.3 at Ω = 1.47 (Vd = 0) and σ′ xx/vF = 31.0 at Ω = 1.45 xx/vF = 28.2 (maximum) at Ω = 1.46 xy/vF = −11.52 (minimum) at the same Ω [inserts in Fig. 9(a) and 8(b)]. Far from the resonance, the real part of DCT practically coincides with that obtained in the previous In Fig. 9(b), the peak values are σ′ and σ′ section but remains finite near the resonance. Note that the off-diagonal element σ′ xy changes its sign from positive (for small frequencies) to negative (near the resonance frequency). The element σ′ yx changes its sign from negative (for small frequencies) to positive and, again, to 22 (a) (b) 30 20 F v / m ¢ l s 10 0 1.3 1.4 W 1.5 F v / m , l s F v / m ¢ l s F v / m , l s 27 18 9 0 -9 1.3 W 1.4 1.5 W xx yy s, s, s, s, xx yy yy xx s, s, s, s, s, xy+s, yx xy yx Figure 9: (Color online) The real part of normalized DCT, σ′ lm/vF , calculated with the quadratic correction to the graphene Dirac spectrum as a function of frequency (in units of ω0). (a) K = 1.5 (γ = 0◦), σ′′ xy = σ′′ yx = 0; Vd = 0 (solid and dashed curves) and Vd = 0.4 (dashed-dotted and dashed-double dotted curves). (b) K = 1.5 (γ = 45◦), Vd = 0.4. Different curves correspond to different components of σ′ lm. The inserts show details of the resonance. negative (near the resonance frequency). As a consequence, the sum (σ′ yx) shows two frequency windows in which it takes negative values [dotted curve in Fig. 9(b)]. Within xy + σ′ these windows, the considered graphene system can be electrically unstable, and an external electromagnetic wave of such frequency can be amplified. VI. SCREENING EFFECT Within the employed theoretical approach, the screening effect can be considered as sity n1(r, t) in the graphene layer. follows. The weak ac electric field E1(r, t) induces a modulation of the electron den- In turn, the induced electron charge −en1(r, t) gen- erates an electrostatic potential ϕ1(z, r, t) and an additional lateral electric field F1(r, t) = 23 −∂ϕ1(z, r, t)/∂rz=0, which changes the initial field E1(r, t) and thereby gives rise to screen- ing effect. Further, the electric field F1(r, t) is included in the expression for the ac current density j1l = σlm(E1m + F1m) (summation over the repetitive indices). Since the field F1 is proportional to E1, the ac current density can be written as j1l = σs renormalized DCT σs lm takes onto account the effect of screening. lmE1m, where the The electrostatic potential is given by the Poisson equation (cid:18) ∂2 ∂z2 + ∂2 ∂r2(cid:19) ϕ1(z, r) = 4πe ε0 n1(r)δ(z) , which solution is ϕ1(z, r, t) = − 2πe ε0k nωk 1 exp(−kz) exp[i(k · r − ωt) . Then, it follows that the induced lateral electric field is along the wave vector k, F1(r, t) = ik 2πe ε0k nωk 1 exp[i(k · r − ωt) . (26) (27) (28) The local change of the electron density n1(r, t) can be found from the continuity equation ∂n1(r, t) ∂t 1 e − ∂j1(r, t) ∂r = 0 , which takes the form of an equation for the Fourier coefficient nωk 1 iωnωk 1 + i e klσlm(cid:18)Eωk 1m + 2πie ε0k kmnωk 1 (cid:19) = 0 . Then, we obtain and n1(r, t) = − klσlmE1m(r, t) e(ω + iτ −1) F1(r, t) = − 2πi ε0 k k klσlmE1m(r, t) ω + iτ −1 , (29) (30) (31) (32) where the quantity τ = ε0k/2πklσlmkm has the dimensionality of time and depends on the wave vector. Upon substituting F1 from Eq. (32) into the current density j1l = σlm(E1m + F1m) = σs lmE1m and comparing both the expressions, one finally obtains σs xx = σxx − 2πi ε0k klσlxσxmkm ω + iτ −1 σs yy = σyy − 2πi ε0k klσlyσymkm ω + iτ −1 , , 24 (33) σs xy = σxy − σs yx = σyx − 2πi ε0k 2πi ε0k klσlyσxmkm ω + iτ −1 klσlxσymkm ω + iτ −1 , . The result (33) is quite general and is valid not only for the graphene, but also for an arbitrary quantum well system. The equations (33) are simplified in the chosen coordinate system: kx = k, ky = 0 (Fig. 1). As an example, we consider the wave propagating along the dc field (γ = 0◦). Then Eqs. (33) read σs xx = σxx 1 + i(ωτ )−1 , (34) σs yy = σyy, and σs yx = 0. The similar expression is obtained for the total electric field Et = E1 + F1 which longitudinal component is Etx = E1x/[1 + i(ωτ )−1], and the transverse xy = σs component is not screened Ety = E1y. The denominator in these expressions can be treated as an effective dielectric function ε(k, ω) = 1 + i(ωτ )−1, where τ −1 = (2π/ε0)kσxx. Such renormalization of the DCT changes its frequency behavior near the resonance. considering the limit Ω → K, we obtain from Eq. (13) the asymptotic expression Indeed, ln(eEF + 1) 1 σ′ xx = vF (2K)1/2(K − Ω)1/2 , 1 − Vx xxΩ→K ∼ (K − Ω)−1/2. Separating the real and imagi- (35) which diverges at the resonance as σ′ nary parts in Eq. (34), we find (σs xx)′ = σ′ xx A2(k, ω) + B2(k, ω) , (36) where A(k, ω) = (1 − 2πkσ′′ the considered limit σ′′ xx is finite and σ′ xx/ε0ω) and B(k, ω) = 2πkσ′ xx goes to infinity, then B(k, ω) ∼ σ′ xx/ε0ω. Taking into account that at xx → ∞, we can xx). Now, utilizing Eq. (35), we obtain the write according to (36) (σs xx)′ ≃ (ε0vF /2π)2(1/σ′ asymptotic expression (σs xx)′Ω→K = vF ε2 0(1 − Vx) 4π2 ln(eEF + 1) (2K)1/2(K − Ω)1/2 , (37) in which (σs Etx(x, t) = (1/2)[Eωk similar behavior. Specifically, we find the relative amplitude Eωk xx)′ ∼ (K − Ω)1/2, i.e., it approaches zero for Ω → K. The total electric field tx cos(kx− ωt + ϕ0) is characterized by the 1x of the total tx exp(ikx− iωt) + c.c.] = Eωk tx = Eωk tx /Eωk field Etωk x = hA2(k, ω) + B2(k, ω)i−1/2 25 (38) and the phase shift cos ϕ0 = A(k, ω)/[A2(k, ω) + B2(k, ω)]1/2. From Eq. (38), we obtain the asymptotic behavior near the resonance Eωk xx)−1 ∼ (K − Ω)1/2 → 0. We may con- clude that due to self-consistent electrostatic potential the considered divergence is removed tx ∼ (σ′ leading to finite values of the DCT at the resonance. VII. CONCLUSION We have investigated the dynamic conductivity tensor, σlm(k, ω), of doped graphene subjected to a strong dc electric field. The steady-state transport has been characterized in terms of the hot Dirac quasiparticles with the electron (hole) temperature Te and the drift velocity vd. The analysis has been carried out in the most general fashion considering a weak ac field of arbitrary configuration and regarding the ac field vector and the wave vector k as independent vectors. We found that specific features of dispersion of DCT are essentially determined by the Dirac energy spectrum and by the existing resonance at ω = vF k, at which the phase velocity of the wave in the direction of propagation coincides with the carrier velocity. The real part of DCT takes nonzero values (except the points where it changes its sign) at frequencies below the resonance (ω < vF k) and diverges [(σxx(k, ω)] at ω → vF k. This means dissipation of the energy of ac wave since in relation to such carriers the ac field is effectively a steady-state one and therefore can do work, which is not zero when averaging over the wave period. We have shown that deviation from the Dirac-like spectrum at high energies and screening by the charge carriers regularizes the divergence, such that the resulting resonance curve is characterized by definite (finite) peak and width values. The imaginary part of DCT is nonzero for all considered frequencies and goes to zero for ω → ∞. The anisotropy of the carrier distribution function induced by a strong dc electric field leads to a new isolated direction (the electron drift) in addition to the wave vector k, so that tensor structure of the DCT is complicated. We have revealed certain ac field configurations for which the ac power density is negative. This has allowed us to indicate regions of terahertz frequency for possible electrical (drift) instability in the considered graphene system. We suggest that the knowledge of the DCT dispersion controlled by a strong dc electric field may be useful for further progress in graphene plasmonics and applications. Finally, we note that the theoretical formalism developed in this work is quite general and 26 may be useful for calculations of spatially dispersive dynamic response under a strong dc elec- tric field in graphene-analogous 2D semiconductor nanomaterials, such as transition-metal dichalcogenides, the silicon and germanium counterparts of graphene, etc.51 -- 53 In addition, it would also be interesting to explore, for comparison, other steady-state regimes of the high- field transport with known distribution function of the hot carriers, in particular, obtained with advanced numerical methods for the graphene54,55 and graphene-like nanomaterials56 in combination with the results obtained in this work. ACKNOWLEDGMENTS This work was partially supported by NASU (State Targeted Scientific-Technical Program "Nanotechnology and Nanomaterials" 2010-2014, No. 2.3.4.22), STCU (Targeted R&D Initiatives Program 2012-2014, No. 5716), and by a grant for young scientists of NAS of Ukraine. The work performed at North Carolina State University was supported, in part, by FAME (one of six centers of STARnet, a SRC program sponsored by MARCO and DARPA). 27 Appendix A: Evaluation of the kinematic integrals The kinematic integrals in Eq. (9) are evaluated by integrating over p in the polar co- ordinate system px = p cos θ and py = p sin θ, in which the direction of the x axis in the momentum space is chosen to be along the k vector, θ is the polar angle, and d2p = p dp dθ. With the function f0(p) of Eq. (5), the integral over p from 0 to ∞ is easily evaluated, and we get 2π dθ σxx = i vF ln (eEF + 1) 2πK σyy = i vF ln (eEF + 1) 2πK σxy = i vF ln (eEF + 1) 2πK σyx = i vF ln (eEF + 1) 2πK Z Z 0 2π 0 2π Z 0 Z 0 dθ dθ 2π dθ (A1) cos2 θ − Vx cos θ (s − cos θ)(1 − Vx cos θ − Vy sin θ)2 , sin2 θ − Vy sin θ (s − cos θ)(1 − Vx cos θ − Vy sin θ)2 , sin θ cos θ − Vy cos θ (s − cos θ)(1 − Vx cos θ − Vy sin θ)2 , sin θ cos θ − Vx sin θ (s − cos θ)(1 − Vx cos θ − Vy sin θ)2 , where Vx = Vd cos γ and Vy = Vd sin γ. Here, we use the dimensionless variables introduced in Sec. III. The complex quantity s = (Ω + iδ)/K is reduced to the dimensionless phase velocity Ω/K of the propagating wave in the limit of δ → 0. Further, we make a substitution z = exp(iθ), then cos θ = (z + z−1)/2 and sin θ = (z − z−1)/2i; the integrals over the angle θ, from 0 to 2π, become the integrals of rational functions of z in the complex plane, where the integration is over a unit circle z = 1 centered in the origin. These functions (the integrands) are given by (A2) Fxx(z) = Fyy(z) = Fxy(z) = Fyx(z) = −(z2 + 1)(z2 − 2Vxz + 1) (z2 − 1)(z2 − 2iVyz − 1) (z − z+)(z − z−)(z − z+)2(z − z−)2 , (z − z+)(z − z−)(z − z+)2(z − z−)2 , (z − z+)(z − z−)(z − z+)2(z − z−)2 , (z − z+)(z − z−)(z − z+)2(z − z−)2 , i(z2 + 1)(z2 − 2iVyz − 1) i(z2 − 1)(z2 − 2Vxz + 1) where z± = s ± √s2 − 1 and z± = (1 ±p1 − V 2 Flm(z) of the first and second order, respectively. d )/(Vx − iVy) are the poles of the functions It is a straightforward procedure to 28 evaluate the integrals by using the theorem of residues, which gives each integral of the functions (A2) equals the sum of residues at a number of isolated poles located inside the integration contour, multiplied by 2πi. The residue at an n-order pole at z = a is evaluated by48 Res(n)Flm(a) = 1 (n − 1)! lim z→a d (n−1) dz (n−1)h(z − a)nFlm(z)i. (A3) It should be noted that the expressions for the first order poles z± contain a square root of complex function, which is a double-valued function in the complex plane. If we chose a cut along the real axis in the complex plane s from −1 to +1, then only one pole z = z− is inside the integration path over the unit circle. The reason is that, at such cut, the function z− conformally maps points of the plane s into the interior of the unit circle; correspondingly, the function z+ transforms them onto the exterior of the unit circle. The poles z± do not depend on frequency, and there is only one pole z− inside the unit circle. Thus, finally, the two poles z = z− and z = z− are inside the unit circle, and we get the exact expression σlm = i 2vF ln (eEF + 1) K(Vx − iVy)2 (cid:2)Res(1)Flm(z−) + Res(2)Flm(z−)(cid:3) . (A4) Since the drift velocity Vd is in the denominator of formula (A4), the situation with the absence of the electron drift should be treated as the limit of Vd → 0. 29 1 P. R. Wallace, Phys. Rev. 71 (1947) 622. 2 S. Das Sarma, S. Adam, E. H. Hwang, and E. Rossi, Rev. Mod. Phys. 83 (2011) 407. 3 A. H. Castro Neto, F. Guinea, N. M. R. Peres, K. S. Novoselov, and A. K. Geim, Rev. Mod. Phys. 81 (2009) 109. 4 Y.-M. Lin, C. Dimitrakopoulos, K. A. Jenkins, D. B. Farmer, H.-Y. Chiu, A. Grill, and Ph. Avouris, Science 327 (2010) 662. 5 J. Lloyd-Hughes and T.-I. Jeon, J. Infrared Millim. Terahertz Waves 33 (2012) 871. 6 L. A. Falkovsky, JETP 106 (2008) 575. 7 J. Horng, C.-F. Chen, B. Geng, C. Girit, Y. Zhang, Z. Hao, H. A. Bechtel, M. Martin, A. Zettl, M. F. Crommie, Y. R. Shen, and F. Wang, Phys. Rev. B 83 (2011) 165113. 8 K. F. Mak, M. Y. Sfeir, Y. Wu, C. H. Lui, J. A. Misewich, and T. F. Heinz, Phys. Rev. Lett. 101 (2008) 196405. 9 L. A. Falkovsky and A. A. Varlamov, Eur. Phys. J. B 56 (2007) 281. 10 V. P. Gusynin, S. G. Sharapov, and J. P. Carbotte, Phys. Rev. Lett. 96 (2006) 256802; Phys. Rev. B 75 (2007) 165407. 11 Z. Q. Li, E. A. Henriksen, Z. Jiang, Z. Hao, M. C. Martin, P. Kim, H. L. Stormer, and D. N. Basov, Nat. Phys. 4 (2008) 532. 12 L. Hao, J. Gallop, S. Goniszewski, O. Shaforost, N. Klein, and R. Yakimova, Appl. Phys. Lett. 103 (2013) 123103. 13 G. Lovat, G. W. Hanson, R. Araneo, and P. Burghignoli, Phys. Rev. B 87 (2013) 115429. 14 X. Luo, T. Qiu, W. Lu, and Z. Ni, Mater. Sci. Eng. R74 (2013) 351. 15 M. Wagner, Z. Fei, A. S. Mcleod, A. S. Rodin, W. Bao,E. G. Iwinski, Z. Zhao, M. Goldflam, M. Liu, G. Domingues, M. Thiemens, M. F. Fogler, A. H. C. Neto, C. N. Lau, S. Amarie, F. keilmann, and D. N. Basov, Nano Lett. 14 (2014) 894. 16 F. Rana, Nat. Nanotechnol. 6 (2011) 611; 17 L. Ju, B. Geng, J. Horng, C. Girit, M. Martin, Z. Hao, H. A. Bechtel, X. Liang, A. Zettl, Y. R. Shen, and F. Wang, Nat. Nanotechnol. 6 (2011) 630. 18 A. Yu. Nikitin, F. Guinea, and L. Martin-Moreno, Appl. Phys. Lett. 101 (2012) 151119. 19 N. N. Esfahani, R. E. Peale, C. J. Fredricksen, J. W. Cleary, J. Hendrickson, W. R. Buchwald, 30 B. D. Dawson, and M. Ishigami, Proc. SPIE 5 (2012) 8261-14. 20 V. A. Kochelap and S. M. Kukhtaruk, J. Appl. Phys. 109 (2011) 114318. 21 V. A. Kochelap and S. M. Kukhtaruk, Ukr. J. Phys. 57 (3) (2012) 367. 22 Yu. M. Lyaschuk and V. V. Korotyeyev, Ukr. J. Phys. 59 (2014) 495. 23 J. Schiefele, J. Pedr´os, F. Sols, F. Calle, and F. Guinea, Phys. Rev. Lett. 111 (2013) 237405. 24 R. Hillenbrand, T. Taubner, and F. Keilmann, Nature 418 (2002) 159. 25 Z. Fei, A. S. Rodin, G. O. Andreev, W. Bao, A. S. McLeod, M. Wagner, L. M. Zhang, Z. Zhao, M. Thiemens, G. Dominguez, M. M. Fogler, A. H. Castro Neto, C. N. Lau, F. Keilmann, and D. N. Basov, Nature 487 (2012) 82. 26 A. Huber, N. Ocelic, D. Kazantsev, R. Hillenbrand, Appl. Phys. Lett. 87 (2005) 081103. 27 Z. Fei, G. O. Andreev, W. Bao, L. M. Zhang, A. S. McLeod, C. Wang, M. K. Stewart, Z. Zhao, G. Dominguez, M. Thiemens, M. M. Fogler, M. J. Tauber, A. H. Castro-Neto, C. N. Lau, F. Keilmann, and D. N. Basov, Nano Lett. 11 (2011) 4701. 28 J. Chen, M. Badioli, P. Alonso-Gonz´alez, S. Thongrattanasiri, F. Huth, J. Osmond, M. Spasenovi´c, A. Centeno, A. Pesquera, P. Godignon, A. Z. Elorza, N. Camara, F. J. Garcia de Abajo, R. Hillenbrand, and F. H. L. Koppens, Nature 487 (2012) 77. 29 P. Alonso-Gonz´alez, A. Y. Nikitin, F. Golmar, A. Centeno, A. Pesquera, S. V´elez, J. Chen, G. Navickaite, F. Koppens, A. Zurutuza, F. Casanova, L. E Hueso, and R. Hillenbrand, Science 344 (2014) 1369. 30 S. Gangadhariah, A. M. Farid, and E. G. Mishchenko, Phys. Rev. Lett. 100 (2008) 166802. 31 S. M. Kukhtaruk and V. A. Kochelap, Phys. Rev. B 92 (2015) 041409(R). 32 K. S. Novoselov, A. K. Geim, S. V. Morozov, D. Jiang, Y. Zhang, S. V. Dubonos, I. V. Grig- orieva, and A. A. Firsov, Science 306 (2004) 666. 33 F. Schwierz, Nat. Nanotechnol. 5 (2010) 487. 34 S. Y. Liu, N. J. M. Horing, and X. L. Lei, IEEE Sensors J. 10 (2010) 681. 35 T. Mueller, F. Xia, and Ph. Avouris, Nat. Photon. 4 (2010) 297. 36 K. Tahy, T. Fang, P. Zhao, A. Konar, C. Lian, H. G. Xing, M. Kelly, and D. Jena, Graphene Transistors, Physics and Applications of Graphene - Experiments (edited by S. Mikhailov, In- Tech, 2011), p. 473. 37 A. M. DaSilva, K. Zou, J. K. Jain, and J. Zhu, Phys. Rev. Lett. 104 (2010) 236601. 38 V. Perebeinos and Ph. Avouris, Phys. Rev. B 81 (2010) 195442. 31 39 A. D. Liao, J. Z. Wu, X. Wang, K. Tahy, D. Jena, H. Dai, and E. Pop, Phys. Rev. Lett. 106 (2011) 256801. 40 M.-H. Bae, S. Islam, V. E. Dorgan, and E. Pop, ACS Nano 5 (2011) 7936. 41 S. Islam, Z. Li, V. E. Dorgan, M.-H. Bae, and E. Pop, IEEE Electron Device Lett. 34 (2013) 166. 42 R. Bistritzer and A. H. MacDonald, Phys. Rev. B 80 (2009) 085109. 43 D. Svintsov, V. Vyurkov, S. Yurchenko, T. Otsuji, and V. Ryzhii, J. Appl. Phys. 111 (2012) 083715. 44 D. Svintsov, V. Vyurkov, V. Ryzhii, and T. Otsuji, Phys. Rev. B 88 (2013) 245444. 45 A. Y. Serov, Z.-Y. Ong, M. V. Fischetti, and E. Pop, J. Appl. Phys. 116 (2014) 034507. 46 L. D. Landau and E. M. Lifshits Course of Theoretical Physics: Physical Kinetics, 1981 Perg- amon New York 10. 47 V. F. Gantmakher and Y. B. Levinson Carrier Scattering in Metals and Semiconductors, 1987 North-Holland Amsterdam. 48 P. M. Morse and H. Feshbach, Methods of Theoretical Physics (McGraw Hill, New York, 1953) part 1, p. 147. 49 T. Stauber N. M. R. Peres, and A. K. Geim, Phys. Rev. B 78 (2008) 085432. 50 C. H. Yang, Y. Luo, Z. S. Liu, and J. J. Jiang, J. Appl. Phys. 114 (2013) 113701. 51 M. Xu, T. Liang, M. Shi, and H. Chen, Chem. Rev. 113 (2013) 3766. 52 M. Chhowalla, H. S. Shin, G. Eda, L.-J. Li, K. P. Loh, and H. Zhang, Nat. Chem. 5 (2013) 263. 53 Q. Tang and Z. Zhou, Prog. Mater. Sci. 58 (2013) 1244. 54 X. Li, E. A. Barry, J. M. Zavada, M. Buongiorno Nardelli, and K. W. Kim, Appl. Phys. Lett. 97 (2010) 082101; 97 (2010) 232105. 55 X. Li, K. M. Borysenko, M. Buongiorno Nardelli, and K. W. Kim, Phys. Rev. B 84 (2011) 195453. 56 X. Li, J. T. Mullen, Z. Jin, K. M. Borysenko, M. Buongiorno Nardelli, and K. W. Kim, Phys. Rev. B 87 (2013) 115418. 32 

1001.4024

2

1001

RKKY coupling in graphene

We study the carrier-mediated exchange interaction, the so-called RKKY coupling, between two magnetic moments in graphene using exact diagonalization on the honeycomb lattice. By using the tight-binding nearest neighbor band structure of graphene we avoid the use of a momentum cut-off which plagues results in the Dirac continuum model formulation. We extract both the short and long impurity-impurity distance behavior and show several corrections to earlier long distance results. In the bulk the RKKY coupling is proportional to $1/|{\bf R}|^3$ and displays $(1+\cos(2{\bf k}_D\cdot {\bf R})$-type oscillations. A-A sublattice coupling is always ferromagnetic whereas A-B subattice coupling is always antiferromagnetic and three times as large. We also study the effect of edges in zigzag graphene nanoribbons (ZGNRs) and find that for impurities on the edge, the RKKY coupling decays exponentially because of the localized zero energy edge states. For impurities inside a ZGNR the bulk characteristics are quickly regained.

cond-mat.mes-hall

cond-mat

RKKY coupling in graphene NORDITA, Roslagstullsbacken 23, SE-106 91 Stockholm, Sweden Annica M. Black-Schaffer (Dated: October 24, 2018) We study the carrier-mediated exchange interaction, the so-called RKKY coupling, between two magnetic impurity moments in graphene using exact diagonalization on the honeycomb lattice. By using the tight-binding nearest neighbor band structure of graphene we also avoid the use of a momentum cut-off which plagues perturbative results in the Dirac continuum model formulation. We extract both the short and long impurity-impurity distance behavior and show on a qualitative agreement with earlier perturbative results in the long distance limit but also report on a few new findings. In the bulk the RKKY coupling is proportional to 1/R3 and displays (1 + cos(2kD · R)- type oscillations. A-A sublattice coupling is always ferromagnetic whereas A-B subattice coupling is always antiferromagnetic and three times as large. We also study the effect of edges in zigzag graphene nanoribbons (ZGNRs). We find that for impurities on the edge the RKKY coupling decays exponentially because of the localized zero energy edge states and we also conclude that a non-perturbative treatment is essential for these edge impurities. For impurities inside a ZGNR the bulk characteristics are quickly regained. PACS numbers: 75.20.Hr, 75.75.-c, 73.20.-r I. INTRODUCTION Graphene is a two-dimensional honeycomb lattice of carbon and has since its isolation in 20041 generated a lot of attention (see e.g. Ref. 2 and references therein). Its two-dimensionality, linear energy dispersion, where the quasiparticles are massless Dirac fermions, and chemi- cal potential tunable by a gate voltage are all novel fea- tures in an, essentially table-top, condensed matter sys- tem. Together with a very high mobility these prop- erties have helped to raise the expectation of graphene being a post-silicon era candidate.3 -- 6 For this, function- alization of graphene using for example finite geome- tries, adatoms, hydrogen chemisorption, or intrinsic de- fects such as vacancies, has become an important goal. Adding magnetic atoms or defects have the added benefit of opening the possibility for spintronics where not only the electron charge but also its spin is actively used in devices.7,8 One of the most important property of mag- netic impurities is the effective interaction between them propagated by the conduction electrons in the bulk host, the so-called Ruderman-Kittel-Kasuya-Yoshida (RKKY) coupling.9 -- 11 This coupling is crucial for magnetic order- ing of the impurities but also offers access to the intrinsic magnetic properties of the host. Previous work on the RKKY coupling in graphene12 -- 16 have exclusively used a field-theory continuum model where the graphene band structure is approximated with a Dirac spectrum at each of the two inequivalent Bril- loiun zone corners. Using perturbation theory the RKKY coupling has then been calculated analytically from the static spin susceptibility of this model. The earliest work12,13 failed to recognize the importance of the bi- partite lattice and predicted that the RKKY coupling is always ferromagnetic. Later it was, however, shown by Saremi14 that for all bipartite lattices at half-filling the RKKY coupling is ferromagnetic (FM) only for impuri- ties on the same sublattice but antiferromagnetic (AFM) for impurities on different sublattices. A perturbative ap- proach also requires an explicit use of an ultraviolet mo- mentum cut-off scheme for the non-interacting graphene band structure and it was recently showed that a regular sharp cut-off does not produce the correct results thus demonstrating the need for a carefully chosen regulariza- tion scheme.14 This most likely explains the discrepancies in the later results for the RKKY coupling.14,15 Further- more, any results from a continuum model does not fully resolve the lattice structure which are likely to be im- portant for short impurity-impurity distances R. In fact, the later work14,15 on the RKKY coupling in graphene have only considered the long-distance limit. While this is the traditional RKKY limit, knowledge of the short distance behavior is important in nano-structures as well as when comparing with ab-initio results where the unit cell always has a finite size. To circumvent the use of perturbation theory, the ul- traviolet cut-off dependency, and to also get results for finite impurity distances we will here explicitly calculate the RKKY coupling on the honeycomb lattice using exact diagonalization in a finite system. We show that by using a large enough system it is possible to extract both short- range and long-range behavior of the RKKY coupling. We report on a qualitative agreement between our results and one of the perturbative results in the long distance limit14, but also report on a few new findings, includ- ing a phase shift for different sublattice coupling. This establishes not only that the standard perturbative ap- proach to RKKY coupling in graphene is in general valid but also finally settles the issue of the exact form of the RKKY coupling. Furthermore, we also study the RKKY coupling in zigzag graphene nanoribbons (ZGNRs) and show that the zigzag edge will significantly modify the re- sults due to the presence of the localized zero-energy edge state.17 -- 20 In contrast to the bulk, a non-perturbative treatment seems to be essential for impurities along the zigzag edge. For impurities inside a ZGNR bulk-like be- havior is however achieved even for narrow ribbons. The rest of the paper is organized as follows: In the next section we introduce the model and solution method. Then we discuss the results for on-site and pla- quette impurities, followed by the results for impurities on the edge and inside ZGNRs. We end with comparing our results with the previous perturbative results, and a discussion on how to experimentally achieve magnetic impurities in graphene as well as on the importance of electron-electron interactions. II. METHOD We are here focusing on how impurity magnetic mo- ments, or spins, interact with each other on a graphene surface. We are not concerned with the details of the in- teraction between the moments and graphene nor about how the moments are originally formed. We will there- fore model the interaction between an impurity moment and graphene with a simple Kondo coupling term. Using the nearest neighbor tight-binding Hamiltonian for the pz-orbitals in graphene, the system can be formulated as H = −t X <i,j>,σ (a† iσbjσ + H.c.) + Jk X Si · si, (1) i=imp where aiσ (biσ) annihilates an electron on sublattice A (B) in unit cell i [see Fig. 1(a)], < i, j > means nearest neighbors, and σ is the spin index. Moreover, S = ±Sz is the impurity spin and s = 1 ασαβaβ, with σαβ being the Pauli matrices, is the electron spin (for sublattice B in- terchange a → b). The constants entering are the nearest neighbor hopping in graphene t ∼ 2.7 eV and the Kondo coupling Jk which depends on the particular impurity moment. We consider here only undoped graphene and thus no chemical potential term enters in Eq. (1). 2 a† (a) Zigzag direction (b) Armchair direction A c1 B a c2 p p p R p FIG. 1: (Color online) (a) The graphene honeycomb lattice with the two sublattices A and B in dark and light colors, re- spectively. The lattice unit vectors c1 and c2, lattice constant a = 2.46 A as well as the two most common directions, the zigzag and the armchair, are displayed. (b) Unit cell setup in the AFM configuration with the distance R between impurity spins shown as well as the padding p surrounding them. In standard RKKY perturbation theory21 the leading 2 interaction between two impurity moments at sites i and j is given by HRKKY = Jij Si · Sj, (2) with the effective RKKY coupling constant Jij propor- tional to the static spin susceptibility of the imbedding bulk, Jij ∝ χ0 ij. A. Exact diagonalization Instead of using the above perturbative result for the RKKY coupling we will calculate Jij by exact diagonal- ization of Eq. (1) in a finite system with two impurity spins either aligned ferromagnetically (FM) or antifer- romagnetically (AFM). Then Jij can be expressed as the energy difference between the two configurations:22 Jij = [E(FM) − E(AFM)]/(2S2). We will for simplicity set S = 1. This solution method is non-perturbative, automatically avoids any artificial ultraviolet cut-off de- pendencies, and is also capable of generating results for any R. However, solving in a finite system creates its own problems. We apply periodic boundary conditions (PBC) to avoid the effects of edge states but then have to deal with the two impurities in one unit cell also in- teracting with the impurities in neighboring cells. By systemically increasing the "padding" p around the two impurities, see Fig. 1(b), we can determine the necessary size of the unit cell for converged results for Jij. As seen later on in Figs. 3 and 4, p = 2R is in general sufficient. We have also ensured convergence with respect to the number of k-points in reciprocal space. Apart for studying the RKKY interaction in the bulk we have also looked at ZGNRs and strips. Here the graphene lattice is terminated along the zigzag direc- tion with saturated σ-bonds whereas the pz-orbital on the edge atom is unsaturated because of the one miss- ing nearest neighbor. We have primarily studied narrow symmetric ZGNRs with width W = 8/√3a, where a is the lattice constant, see Fig. 1. As is well established, the zigzag graphene edge hosts localized states at zero energy which significantly changes many of the physical properties compared to the bulk.17 -- 20 However, the arm- chair edge does not have any such zero energy localized states and we find that that for large enough unit cells, the results for a ZGNR, where PBC are applied in the direction of the ribbon, are the same as those for a strip, where instead armchair edges terminate the strip. III. RESULTS ↑a↑ − a† ↓a↓)/2 and sz We start with displaying in Fig. 2 the spin polarization, A = (a† sz B, induced into the graphene from two impurity spins positioned on-site (a,b) and in a plaquette site (c,d), in the FM (a,c) and AFM (b,d) configurations, respectively. Below we will discuss each of these results. (a) (b) 3 Figure 3 shows Jij as a function of the impurity dis- tance R for all four different configurations of sublat- tice and zigzag/armchair directions for on-site impuri- ties. First of all we can directly verify that the RKKY ) 3 − 0 1 ( J / t J − k2 j i ) 3 − 0 1 ( J / t k2 j i J (a) (b) (c) (d) 1 10 0 10 −1 10 −2 10 −3 10 1 10 0 10 −1 10 −2 10 −3 10 0 5 10 R (a) 15 0 5 15 10 R (a) k Jij for A-A (a,b) and t/J 2 FIG. 3: (Color online) −t/J 2 k Jij for A-B (c,d) on-site positioning of the impurity spins along the zigzag (a,c) and armchair (b,d) directions as function of impurity distance R. Solid curves represent calculated results with padding p = 1R − 4R (black ×, blue +, green ⋆, red ◦), where the lines are only a guide to the eye, whereas the large- R dependence in Eqs. (3-4) with C = 1/(72√3π) is displayed with black (cid:3)s. coupling is always FM for A-A sites and AFM for A-B sites as seen in the sign difference of Jij . This has been predicted before14,15 and is a property of the bipartite lattice.14 Secondly, we conclude that padding p = 2R is enough for well converged results. Third, we extract the following functional dependence in the large R = R limit: Jij,A−A(R) = −C 1 + cos(2kD · R) Jij,A−B(R) = C 1 + cos(2kD · R + π) J 2 k t 3J 2 k t R3 R3 (3) , (4) for R measured in units of the lattice constant a. In Fig. 3 these results are plotted as black (cid:3)s using the nu- merical prefactor C = 1/(72√3π) and, as seen, there is essentially a perfect agreement at larger R. In the above equations kD is the reciprocal vector for the Dirac points, (c) (d) FIG. 2: (Color online) Two impurity spins along the zigzag direction positioned on-site on the B-B sublattice (a,b) and in the plaquette site (c,d) in the FM (a,c) and AFM (b,d) configurations. The impurity spins are marked with white or black arrows and the area of the circles on each site is propor- tional to the spin polarization, where excess spin-↑ (spin-↓) density is red/dark grey (black). A. On-site impurities An on-site positioning of the impurity spins, where the spins sit directly on top of an A or B atom of the graphene lattice, has so far been the dominating setup for RKKY studies in graphene.14 -- 16 Since on-site positioning breaks the symmetry of the lattice, it is important to distinguish between A-A and A-B positioning of the two impurity spins. We have studied both of these configurations along both the zigzag and armchair directions as well as verified our predictions for the asymptotic large-R behavior for several other, chiral, directions. Figures 2(a,b) show typ- ical spin polarization patterns of two impurity spins both on the B sublattice and along the zigzag direction. We clearly see that the spin polarization has different signs on the two sublattices close to an impurity and therefore one would expect A-A (B-B) impurities to prefer a FM coupling whereas AFM coupling should be the case for A-B (B-A) impurities. i.e. the corners of the Brillouin zone. There are six such vectors and for the A-A configuration the result is inde- pendent on the particular choice of kD since R is then a lattice translation vector, i.e. R = n1c1 + n2c2 where n1, n2 are integers. However, for the A-B configuration R is not a lattice translation vector and cos(2kD · R) can then depend on the choice of kD. Eq. (4) is only correct when kD · R is maximized, i.e. when kD is cho- sen to be as parallel as possible to R. Eqs. (3-4) are very similar to the results derived earlier by Saremi14 using perturbation theory in a continuum model except for the addition of the π-phase factor and the need for specifying kD in the A-B result, as just discussed. The π-phase factor is essential for reproducing the numeri- cal results for A-B sublattice coupling as these are 180◦ out of phase with the A-A results. These two discrepan- cies stem from the fact that in all previous work R has improperly been treated as a lattice translation vector even for A-B impurities. As seen in our results, choos- ing R to be the correct impurity-impurity distance not only changes the RKKY coupling at short distances, as one might have expected, but also produces the π-phase shift and a need for an explicit choice of kD for all R. We strongly believe that a proper handling of R in a perturbative treatment in the continuum model will also include these two additional corrections found in our nu- merical results. Our numerical prefactor C = 1/(72√3π) also agrees with the results of Saremi14 if one explicitly use a factor of t = 8/3 ≈ 2.67 eV to produce the proper energy dimension of their results. We thus conclude that our non-perturbative results agree, up to a few small, and traceable differences, with the perturbative results of Saremi and they firmly establish that the RKKY cou- pling is oscillatory for certain R-vectors, although never changes sign on the same sublattice, which is contrary to some other recent RKKY results.15 However, note that due to the impurities only appearing at lattice sites, the (1 + cos(2kD · R))-oscillation is in general undersampled. For the zigzag direction the period is 3a instead of 3a/4 whereas for the armchair direction the period is infinitely long. Also, for the cases reported in Fig. 3, it is only for the zigzag A-B configuration that the RKKY cou- pling completely disappears at certain sites. What makes the RKKY coupling in graphene unusual is this non-sign changing oscillation on the same sublattice as well as the 1/R3 decay as compared to Jij ∝ sin(2kF R)/(2kF R)2 for an ordinary 2D metal.23,24 Finally, we are not only able to extract the long-distance behavior but can also directly see the deviations from this behavior at short im- purity distances. As seen in Fig. 3, the results along the zigzag direction are well converged toward the large-R results around R ∼ 10a. The exceptions are the results for every third lattice site in the A-B configuration which are zero in Eq. 4. Along the armchair direction the con- vergence is even faster and, except for nearest neighbor impurity spins, Eqs. (3-4) give correct results for any R. We thus conclude that the large-R limit is reached for surprisingly small impurity distances R. 4 B. Plaquette impurities For magnetic atoms deposited on graphene, the on- site position might not the most energetically favorable but instead the atoms can prefer to sit in the middle of the hexagon, in the plaquette site.25,26 We will here first study the simple situation where the impurity spin cou- ples incoherently, and with the same coupling constant Jk, to all six nearest neighbors in the honeycomb lat- tice. Representative spin polarization maps in this case are shown in Figs. 2(c,d) which have plaquette impurities along the zigzag direction. For this incoherent, symmet- ric, coupling to the lattice the large-R result can in fact be directly derived from the on-site results in Eqs. (3-4) by summing the interactions between all combinations of nearest neighbors of each impurity spin. Doing so, it turns out that in this case the oscillations cancel, the AFM coupling prevails, and the asymptotic large-R re- sult is given by Jij,plaq(R) = C 36J 2 k t 1 R3 . (5) This is the same result as derived by Saremi14 despite the additional π-phase factor in Eq. (4). Figure 4 shows our exact diagonalization results together with Eq. (5). As before, we see that p = 2R is enough to reach con- verged numerical results. For plaquette impurities along the armchair direction, Fig. 4(b), there is a systematic increase in Jij,plaq for small R compared to the long- distance result but the asymptotic behavior is nonethe- less reached before R = 10a. For the zigzag direction, Fig. 4(a), there is some oscillations around the asymp- totic value for short R but convergence with respect to this value is reached already at R ∼ 4a. ) 3 − 0 1 ( J / t k2 q a l p , j i J − 2 10 1 10 0 10 −1 10 0 (a) (b) 5 R (a) 10 0 5 R (a) 10 FIG. 4: (Color online) −t/J 2 k Jij,plaq for impurities in the pla- quette site along the zigzag (a) and the armchair (b) direction. Same color coding as in Fig. 3. While the incoherent case above is straightforward to derive from the results of on-site impurities, coherent cou- pling, where of a plaquette impurity spin couples to a linear coherent combination of the nearest neighbor con- duction electrons, is physically more realistic. Saremi14 showed that the same cancelation of the oscillations that occur in Eq. (5) also makes the 1/R3 contribution disap- pear altogether for coherent coupling. Since we get the same cancellation in the incoherent case when we prop- erly treat the impurity distance, we conclude that this conclusion is still valid. Coherent plaquette impurities thus have a significantly weaker, and thus less relevant, RKKY coupling than the other configurations studied here. C. Impurities in ZGNRs Zigzag edges in graphene have proven to be an excit- ing new playground for magnetism since this termination leads to a localized, flat-band, zero energy edge state and is thus extremely prone to spin polarization.17 -- 20 This zero energy singularity in the local density of states (LDOS) has been claimed to significantly change the perturbative RKKY coupling result because of a qual- itatively modified behavior of the spin susceptibility at the edge.16 It is therefore of large interest to further study these systems and establish the consequences of zigzag edges on the RKKY behavior in an exact, non- perturbative setting. We have found that an armchair edge does not significantly effect the RKKY coupling and thus it is only necessary to study the zigzag edge in order to establish the qualitative effect of edges on the RKKY coupling. Figure 5 shows the spin polarization for two prototyp- ical situations in a narrow ZGNR: impurities inside the ZGNR (a,b) and along the edge (c,d). The RKKY cou- pling −Jij for the configurations in Fig. 5 is shown in Fig. 6(a) for the parameters t = Jk = 1. While this is a rather unphysically high value for Jk it helps display- ing the essential features in a numerically accessible R- range. We have for comparison also included the results for Jk = t/100, and, as discussed below, the same physi- cal behavior is governing both Jk-values, although all sig- nificant features get extended over a longer R-range for smaller Jk, making a complete numerical study harder. For reference in Fig. 6, the asymptotic zigzag result in the bulk is shown in black and we directly see that impuri- ties in narrow ZGNRs behave significantly different. The coupling between edge impurities (red, ◦) is larger than in the bulk for small distances and also displays some os- cillations in that regime, but then decays exponentially for larger R which is in sharp contrast to the power-law decay in the bulk. Displayed is also Jij for the RKKY coupling between opposite edges, the opposite sign be- ing a consequence of the two edges belonging to different sublattices. As seen, after an initial short distance, where the width of the ZGNR is important, same edge and op- posite edge impurities couple with equal strength. This is another difference compared to the bulk results where the AFM coupling is 3 times larger than the FM cou- pling. We thus draw the conclusion that the edges are dominating the response for edge impurities. This could (a) (b) (c) (d) 5 FIG. 5: (Color online) Two on-site impurity spins inside a narrow ZGNR (a,b) and on the edge of the same ZGNR (c,d) in the FM (a,c) and AFM (b,d) configurations, respectively. Same color coding as in Fig. 2. have been anticipated already from Figs. 5(c,d) where the spin polarization is seen to be large only in the vicinity of the impurity and only along the edge, it does not spread significantly into the bulk. The essential features of the RKKY coupling along the edge can be understood from the following argument: A magnetic impurity will always force a spin polarization of the graphene in its vicinity. For a zigzag edge impurity this spin polarization can triv- ially, and essentially without energy penalty, be achieved by polarizing the localized edge state at zero energy with- out much polarization of the surrounding bulk. Conse- quently, away from the impurity, the edge state will also quickly become unpolarized, much more quickly than the bulk can lose its spin polarization, and thus the RKKY coupling between two edge impurity spins decays much faster than in the bulk. With this argument one would expect the total amount of polarization in the B sub- lattice for the FM configuration, P sB, to be large for small R, as then the two impurities interact very strongly with each other, but then rapidly decay with R until its flattens out to a value equal to the sum of the total spin polarization of two uncoupled edge impurity spins. This behavior is confirmed in Fig. 6(b) where the asymptotic value is shown to be reached already around R = 7a for Jk = t. This should be compared to the evolution of the spin polarization for the equivalent configuration in the bulk where the total spin polarization is almost constant, as displayed by the black curve in Fig. 6(b). The exponential decay of Jij is in sharp contradiction to ) V e m ( J − j i 2 10 0 10 −2 10 −4 10 −6 10 −8 10 0 2 1 0 0 10 R (a) 20 B s Σ 25 R (a) 50 FIG. 6: (Color online) (a) −Jij for impurities along the zigzag direction on the zigzag edge of a narrow ZGNR (red, ◦), inside the same ZGNR (green, ⋆), and the asymptotic behavior in the bulk (black, ×) as function of impurity distance R. In (blue, +) is Jij for impurities on opposite edges of the ZGNR. Here t = Jk = 1 expect the gray line where t = 1, Jk = t/100. (b) The total spin polarization in sublattice B, P sB, for the same situations as in (a) in the FM impurity configuration. A large enough unit cell was used to ensure convergence of P sB in the R-range displayed. earlier calculations on the same width ZGNR by Bunder et al.16 who used an analytical approach to the tight- binding structure of graphene to calculate the spin sus- ceptibility and thus obtained perturbative results for the RKKY coupling. Both methods predict an equivalence between same and opposite edge impurity spins but Bun- der et al. report an almost linear decay with distance for small distances followed by sign oscillations in the RKKY coupling. Despite a thorough investigation of the RKKY coupling for R ≤ 130a, we were not able to detect any deviations from the exponential decay, and thus no sign changes, within the numerical accuracy which was a factor of 10−11 of the RKKY coupling at R ∼ 0. We thus conclude that when studying these edge states, a non-perturbative method appears to be essential when calculating the RKKY coupling. In the bulk Jij ∝ J 2 k and PsB ∝ Jk but for edge impurities these simple relations do not longer hold. For all Jk ≥ t/10 we have found that the asymptotic large-R induced spin polarization from edge impurities is PsB ≈ 1.1 and even for Jk = t the asymptotic value of PsB is only slightly higher due to a finite amount of induced polarization in the nearby bulk, see Fig. 6(b). Thus even in the limit of vanishing Jk, edge impurities are going to elicit a finite spin polarization response of the ZGNR. This is again due to the extreme easiness of polarizing the zero energy edge state. However, when Jk decreases the spin polarization per edge site naturally goes down so the polarization now instead have to be spread over more edge sites. This has an interesting con- sequence for the RKKY coupling: When Jk decreases, the more elongated polarization response causes the os- cillations in the RKKY coupling for small R to be spread 6 out over a longer R-range. Eventually, however, an expo- nential decay is achieved for large enough R for any Jk, as also seen in the gray curve in Fig 6(a) for Jk = t/100. But note that the exponential decay rate also becomes smaller when the spin polarization gets more elongated along the edge, thus making the RKKY coupling at large R larger for decreasing Jk. As a direct consequence, the RKKY coupling between edge impurities is going to be larger than the coupling between bulk impurities over a larger R-distance the smaller the Jk. Of course in the ex- treme large-R limit the exponential decay is always going to make the edge impurity RKKY coupling smaller than the bulk coupling. It is worth pointing out before leaving the treatment of same edge impurities that for all of the effects described above, the width of the ribbon is not essential and we thus expect the same behavior even for much wider ribbons. Physically in between zigzag edge and bulk impuri- ties we find impurities positioned inside a narrow ZGNR. The spin polarization for this situation is displayed in Fig. 5(a,b). Here the bulk is polarized in the direction parallel to the ZGNR but this bulk polarization also in- duces an edge polarization. Thus the absolute value of the polarization is here very large and it grows with the distance R since all the bulk between the two impurities is at least lightly polarized and any amount of bulk po- larization is going to elicit a large polarization of the edge state. This behavior is clearly seen in Fig. 6(b) where the total spin polarization increases linearly with R. How- ever, the edge states are also very easily unpolarized, as discussed above, so the effect on the RKKY coupling from the edges is not expected to be very large compared to the bulk contribution. This prediction is confirmed in Fig. 6(a) where we see that, while the coupling is some- what larger than in the bulk and lacks oscillations, it decays with approximately the same power-law as in the bulk. Also, Jij ∝ J 2 k and PsB ∝ Jk for these impu- rities, which further establish the bulk-like behavior of impurities inside a narrow ZGNR. We anticipate that for wider ZGNRs all bulk properties are fully recovered. IV. DISCUSSION Our results above are obtained without any approxi- mations except the finite size of the system, assuming a non-interacting picture for the electrons in graphene, and using a nearest-neighbor hopping band structure. The first approximation we have taken care to handle system- atically and as long as the padding p around the impu- rities are twice or larger than the impurity-impurity dis- tance R in each direction, the results are well converged. Also, since the asymptotic behavior is reached for rela- tively small R we have been able to extract the large R limit to make a comparison with earlier, partly disagree- ing, results12 -- 15 obtained using a perturbative approach within the continuum field-theory model. Eqs. (3)-(4) are our results for large R for bulk impurities and these are closely related to one of the results in the literature14 al- though a π-phase factor and the need for explicitly choos- ing kD for the A-B configuration are new findings. This difference is most likely due to a previous improper treat- ment of the impurity distance for impurities on differ- ent sublattices. Thus our results establish both that the usual perturbative treatment of RKKY coupling is appro- priate in the bulk and, that, while the RKKY coupling in graphene does not undergo sign changes with distance on the same sublattice, it still has a (1 + cos(2kD · R))- oscillation, a fact that has only been pointed out in one of the previous works.14 Despite these oscillations the facts that the AFM A-B coupling is 3 times larger than the FM A-A coupling and that the coupling is stronger at short distances result in a strong tendency towards AFM order for any random configuration of impurities. Thus these new results still support the conclusion of mag- netic defects in graphene creating a dilute antiferromag- net at low enough temperatures.15 The moments will here be oriented in opposite directions on the two sublattices with the total magnetic moment equal to zero. Since the RKKY coupling is always AFM (incoherent coupling) or very small (coherent coupling) for plaquette impurities a dilute AFM state could also be present for plaquette impurities. We have also studied ZGNRs where we have shown that the zero energy edge state present on the zigzag graphene edge significantly modifies the coupling between impurity spins. We show that defects along the edges couple very strongly at short distances but that the coupling finally decays exponentially with distance in contrast to the R−3 power-law decay in the bulk. The exponential decay is a consequence of the extreme eas- iness of polarizing and unpolarizing the localized zero energy state. This result disagrees with earlier perturba- tive results16 where they found the decay to be almost linear for small R followed by oscillatory sign changes. So while the standard perturbative RKKY treatment is approximately valid in the bulk we have here shown that for ZGNR edge impurities a numerical non-perturbative treatment is essential. The approximation of ignoring electron-electron inter- action in graphene is on the other hand harder to moti- vate. The non-interacting picture has been predominant in the study of RKKY coupling13 -- 16 as well as in other theoretical work on magnetic adatoms in graphene.27 -- 30 However, there seem to be growing evidence for the importance of electron-electron interactions in graphene with theoretical results pointing to graphene being close to a Mott insulator state.31 -- 39 Electron interactions could thus potentially significantly modify the magnetic prop- erties of graphene and therefore also the RKKY coupling. One such example would be if the insulating state is a N´eel phase.37 While a comprehensive treatment of elec- tron interactions in graphene is extremely hard, a density functional theory (DFT) calculation of a carefully chosen system should be able to offer insight into the impor- tance of interactions for the RKKY coupling. However, not only will the magnetic moment and its coupling to 7 graphene be more complicated in a real system than in the idealized Eq. (1) but also a large unit cell is needed, making the DFT calculation highly computationally in- tensive. Nonetheless, some DFT results exist for short distances40 and line defects,41 both pointing to a slower power-law decay than R−3. A detailed comparison of such results with our short distance RKKY coupling may offer insights in the applicability of the non-interacting approximation and is the goal of a future study. In fact, initial data points to the possibility of electron-electron interactions producing strong enough corrections to the non-interacting results that it is reasonable to then ig- nore any corrections to the nearest neighbor hopping band structure in comparison. In a ZGNR the effect of electron-electron interaction is possibly larger than in the bulk as here any electron-electron interaction will drive a spontaneous spin polarization of the edges.51 -- 56 Thus, in a real ZGNR the edge is already polarized and an im- purity spin can then not as easily polarize the edge as found in the non-interacting picture, especially since Jk is usually a small parameter. A recent DFT study57 has shown on an almost exponentially decaying RKKY cou- pling for nearby impurities inside a narrow ZGNR which points to the importance of electron-electron interactions even inside ZGNRs. This would be in contrast to the non-interacting picture where impurities inside a ZGNR behave essentially bulk-like. Detailed results on the influ- ence of electron-electron interactions for ribbons is also a subject of the future study. We have here also explicitly ignored the issue of how an external magnetic moment is created in graphene, but for any experimental verification or use of our results such considerations have to be taken into account. The sim- plest implementation is probably to deposit a magnetic adatom on top of graphene, such as Co or Fe.25,26 We also expect the long distance behavior to be qualitatively similar for substitutional magnetic atoms, of which at least Co has been shown to be magnetic in graphene.40 Single-atom vacancies and hydrogen chemisorption de- fects have also both been shown to possess a magnetic moment in graphene41 -- 46 and should behave similarly to substitutional magnetic atoms. In fact, for vacancies, Lieb's theorem47 gives the same AFM/FM state as found here. It has even been proposed that the FM state found in proton-radiated graphite48,49 is due to excess vacancy creation on one sublattice, which is in agreement with our findings.44 At very short impurity distances, however, an- nihilation of vacancies as well as direct exchange between magnetic moments can also be of importance. In aggre- gate, there exist multiple possibilities for experimentally studying the RKKY coupling in graphene. Moreover, the critical coupling for the Kondo effect is likely too high in undoped graphene26,28,30,50 and therefore the Kondo ef- fect should not compete with the RKKY coupling in any of these systems. In summary we have studied the RKKY coupling be- tween two impurity spins on graphene for any impu- rity distance R using exact diagonalization. Our results largely agree with an earlier perturbative result in the large R limit where Jij ∝ ±(1 + cos(2kD · R))/R3, with A-A sublattice arrangement FM and A-B sublattice ar- rangement AFM, although an additional π-phase shift for the A-B sublattice arrangement is a new finding. The large R limit is reached within a few lattice unit con- stants and the deviations are in general not large even for small R in the bulk. We have also studied the ef- fect of zigzag edges and found that impurities along this edge display an exponentially decaying coupling at large R due to the easiness of polarization of the zero energy edge, state in an non-interacting picture. This result for edge impurities is however in contrast with earlier pertur- 8 bative results,16 pointing to the importance of an exact, non-perturbative, treatment of the RKKY coupling in ZGNRs. Impurities away from the edge, even in narrow ZGNRs, regain most of the bulk characteristics. Acknowledgments The author thanks Sebastian Doniach, Jonas Fransson, and Biplab Sanyal for valuable discussions and Eddy Ar- donne for helpful comments on the manuscript. 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. H. Castro Neto, F. Guinea, N. M. R. Peres, K. S. Novoselov, and A. K. Geim, Rev. Mod. Phys. 81, 109 (2009). 3 C. Berger, Z. Song, X. Li, X. Wu, N. Brown, C. Naud, D. Mayou, T. Li, J. Hass, A. N. Marchenkov, et al., Science 312, 1191 (2006). 4 P. Avouris, Z. Chen, and V. Perebeinos, Nature Nanotech. 2 (2007). 5 A. K. Geim and K. S. Novoselov, Nat. Mater. 6, 183 (2007). 6 Y.-M. Lin, K. A. Jenkins, A. Valdes-Garcia, J. P. Small, D. B. Farmer, and P. Avouris, Nano Lett. 9 (2009). 7 S. A. Wolf, D. D. Awschalom, R. A. Buhrman, J. M. Daughton, S. von Molnar, M. L. Roukes, A. Y. Chtchelka- nova, and D. M. Treger, Science 294, 1488 (2001). 8 I. Zuti´c, J. Fabian, and S. Das Sarma, Rev. Mod. Phys. 76, 323 (2004). 9 M. A. Rudermann and C. Kittel, Phys. Rev. 96, 99 (1954). 10 T. Kasuya, Prog. Theor. Phys. 16, 45 (1956). 11 K. Yosida, Phys. Rev. 106, 893 (1957). 12 M. A. H. Vozmediano, M. P. L´opez-Sancho, T. Stauber, 24 M. T. B´eal-Monod, Phys. Rev. B 36, 8835 (1987). 25 Y. Mao, J. Yuan, and J. Zhong, J. Phys.: Condens. Matter 20 (2008). 26 T. O. Wehling, A. V. Balatsky, M. I. Katsnelson, A. I. Lichtenstein, and A. Rosch, ArXiv:0911.2103 (unpub- lished). 27 B. Uchoa, V. N. Kotov, N. M. R. Peres, and A. H. Cas- tro Neto, Phys. Rev. Lett. 101, 026805 (2008). 28 P. S. Cornaglia, G. Usaj, and C. A. Balseiro, Phys. Rev. Lett. 102, 046801 (2009). 29 B. Uchoa, L. Yang, S.-W. Tsai, N. M. R. Peres, and A. H. Castro Neto, Phys. Rev. Lett. 103, 206804 (2009). 30 Z.-G. Zhu, K.-H. Ding, and J. Berakdar, ArXiv:0912.0182 (unpublished). 31 J. E. Drut and T. A. Lahde, Phys. Rev. B 79, 165425 (2009). 32 J. E. Drut and T. A. Lahde, Phys. Rev. Lett. 102, 026802 (2009). 33 J. E. Drut and T. A. Lahde, Phys. Rev. B 79, 241405(R) (2009). 34 D. V. Khveshchenko, Phys. Rev. Lett. 87, 246802 (2001). 35 D. V. Khveshchenko and H. Leal, Nucl. Phys. B 687, 323 (2004). and F. Guinea, Phys. Rev. B 72, 155121 (2005). 36 E. V. Gorbar, V. P. Gusynin, V. A. Miransky, and I. A. 13 V. K. Dugaev, V. I. Litvinov, and J. Barnas, Phys. Rev. Shovkovy, Phys. Rev. B 66, 045108 (2002). B 74, 224438 (2006). 14 S. Saremi, Phys. Rev. B 76, 184430 (2007). 15 L. Brey, H. A. Fertig, and S. Das Sarma, Phys. Rev. Lett. 99, 116802 (2007). 37 I. F. Herbut, Phys. Rev. Lett. 97, 146401 (2006). 38 I. F. Herbut, V. Jurici´c, and B. Roy, Phys. Rev. B 79, 085116 (2009). 39 I. F. Herbut, V. Jurici´c, and O. Vafek, Phys. Rev. B 80, 16 J. E. Bunder and H.-H. Lin, Phys. Rev. B 80, 153414 075432 (2009). (2009). 40 E. J. G. Santos, D. S´anchez-Portal, and A. Ayuela, 17 M. Fujita, K. Wakabayashi, K. Nakada, and K. Kusakabe, arXiv:0906.5604 (unpublished). J. Phys. Soc. Jpn 65, 1920 (1996). 41 L. Pisani, B. Montanari, and N. M. Harrison, New J. Phys. 18 K. Nakada, M. Fujita, G. Dresselhaus, and M. S. Dressel- 10, 033002 (2008). haus, Phys. Rev. B 54, 17954 (1996). 19 K. Wakabayashi, M. Fujita, H. Ajiki, and M. Sigrist, Phys. Rev. B 59, 8271 (1999). 42 P. O. Lehtinen, A. S. Foster, Y. Ma, A. V. Krasheninnikov, and R. M. Nieminen, Phys. Rev. Lett. 93, 187202 (2004). 43 O. V. Yazyev and L. Helm, Phys. Rev. B 75, 125408 20 Y. Miyamoto, K. Nakada, and M. Fujita, Phys. Rev. B 59, (2007). 9858 (1999). 21 C. Kittel, in Solid State Physics, edited by F. Seitz, D. Turnbull, and H. Ehrenreich (Academic, New York, 1968), vol. 22, p. 1. 22 D. M. Deaven, D. S. Rokhsar, and M. Johnson, Phys. Rev. B 44, 5977 (1991). 23 B. Fischer and M. W. Klein, Phys. Rev. B 11, 2025 (1975). 44 O. V. Yazyev, Phys. Rev. Lett. 101, 037203 (2008). 45 J. J. Palacios, J. Fern´andez-Rossier, and L. Brey, Phys. Rev. B 77, 195428 (2008). 46 H. Kumazaki and D. Hirashima, Physica E 40, 1703 (2008). 47 E. H. Lieb, Phys. Rev. Lett. 62, 1201 (1989). 48 P. Esquinazi, D. Spemann, R. Hohne, A. Setzer, K.-H. Han, and T. Butz, Phys. Rev. Lett. 91, 227201 (2003). 49 H. Ohldag, T. Tyliszczak, R. Hohne, D. Spemann, P. Es- quinazi, M. Ungureanu, and T. Butz, Phys. Rev. Lett. 98, 187204 (2007). 50 K. Sengupta and G. Baskaran, Phys. Rev. B 77, 045417 (2008). 51 T. Hikihara, X. Hu, H.-H. Lin, and C.-Y. Mou, Phys. Rev. B 68, 035432 (2003). 53 L. Pisani, J. A. Chan, B. Montanari, and N. M. Harrison, Phys. Rev. B 75, 064418 (2007). 54 J. Fern´andez-Rossier, Phys. Rev. B 77, 075430 (2008). 55 O. V. Yazyev and M. I. Katsnelson, Phys. Rev. Lett. 100, 047209 (2008). 56 H. Feldner, A. Honecker, D. Cabra, S. Wessel, Z. Y. Meng, and F. F. Assaad, arXiv:0910.5360 (unpublished). 57 D. Soriano, F. Munoz Rojas, J. Fern´andez-Rossier, and 52 H. Lee, Y.-W. Son, N. Park, S. Han, and J. Yu, Phys. Rev. J. J. Palacios, ArXiv:1001.1263 (unpublished). B 72, 174431 (2005). 9 

1112.3859

1

1112

Probing fractional topological insulators with magnetic edge perturbations

We discuss detection strategies for fractional topological insulators (FTIs) realizing time-reversal invariant analogues of fractional quantum Hall systems in the Laughlin universality class. Focusing on transport measurements, we study the effect of magnetic perturbations on the edge modes. We find that the modes show unexpected robustness against magnetic backscattering for moderate couplings and edge interactions, allowing for various phase transitions signaling the FTI phase. We also describe protocols for extracting the universal integer m characterizing the phase and the edge interaction parameter from the conductance of setups with magnets and a quantum point contact.

cond-mat.mes-hall

cond-mat

Probing fractional topological insulators with magnetic edge perturbations TCM Group, Cavendish Laboratory, J. J. Thomson Ave., Cambridge CB3 0HE, UK B. B´eri and N. R. Cooper (Dated: December 2011) We discuss detection strategies for fractional topological insulators (FTIs) realizing time-reversal invariant analogues of fractional quantum Hall systems in the Laughlin universality class. Focusing on transport measurements, we study the effect of magnetic perturbations on the edge modes. We find that the modes show unexpected robustness against magnetic backscattering for moderate couplings and edge interactions, allowing for various phase transitions signaling the FTI phase. We also describe protocols for extracting the universal integer m characterizing the phase and the edge interaction parameter from the conductance of setups with magnets and a quantum point contact. PACS numbers: 73.43.-f,73.63.-b,72.25.-b One of the currently most intensive fields of con- densed matter research, the study of topological phases of matter[1], was born out of the discovery of quan- tum spin-Hall (QSH) insulators[2, 3]. These two dimen- sional gapped systems, realized in HgTe/CdTe quantum wells, display the time-reversal invariant (TRI) counter- part of the integer quantum Hall effect. Recent stud- ies of lattice systems with topologically nontrivial flat bands[4], including time-reversal invariant models[5, 6], point towards realizing TRI analogues of fractional quan- tum Hall (FQH) systems: fractional topological insula- tors (FTIs)[3, 5, 7, 8]. While there are yet no predictions for concrete host materials, the theory of FTIs has sev- eral universal aspects. One can thus already ask: if FTIs were to be realized, how one would detect them in exper- iments? Our goal here is to answer this question focusing on the simplest and presumably most robust FTI phases, adiabatically connected to systems where electrons of op- posite spins form opposite chirality Laughlin-like states at filling fraction 1 m . The odd integer m > 1 is the single universal parameter characterizing this phase. Note that by adiabaticity we specify only the universality class and we do not require, e.g., that the z component of the spin is conserved by the spin-orbit coupling. The experience with QSH and FQH systems shows that a powerful way to identify the underlying phase is to demonstrate the existence and the universal properties of the edge modes the system supports. For QSH sys- tems, the edge modes lead to a universally quantized two terminal conductance when the chemical potential lies in the gap[1]. In the FQHE, the zero bias (linear) tunneling conductance between the edges through a nearly pinched off quantum point contact (QPC) shows universal tem- perature dependence[9, 10] GQPC(T ) ∼ T 2m−2. (These results are valid for temperatures, voltages, etc., much smaller than the bulk gap, which we assume throughout this paper.) For both the QSH and FQH cases there are fortunate circumstances which allow such universal re- sults: the quantization in the QSH case holds because the contacts can be treated as Fermi liquid leads[11], while FIG. 1. FTI (shaded region) with a magnetic edge perturba- tion (hatched). The solid and dashed arrows indicate the pair of counterpropagating edge modes between the source (volt- age V ) and drain (grounded) terminals. Inset: QPC setup near the pinched off limit. for the 1 m FQHE, the universality of the tunneling expo- nent is rooted in the chirality of the edge mode[12]. In the FTI phase the edge supports a pair of counterpropagat- ing FQH modes and these circumstances are absent[12]. As a result, the contact details enter the two terminal conductance, and the exponent in GQPC(T ) becomes de- pendent on the intermode interactions. To extract m from such a compound dependence, one has to measure a set of well chosen quantities. For TRI topological phases, it is natural to look for the effect of TRI breaking edge perturbations, e.g., due to magnetic fields or contacts to ferromagnetic insulators. To our knowledge, this is a direction yet unexplored in the FTI context and, as we show here, it is a fruitful one: the behavior of the conduc- tance in the presence of such perturbations always allows one to extract m from measuring at most two quantities. A sketch of the proposed setup is shown in Fig. 1. Our results show that the effect of magnetic perturba- tions is much richer than for QSH systems. While in the QSH case these always gap the edge modes[1], the FTI edge is more robust: magnetic perturbations are irrele- vant [in the sense of the renormalization group (RG)] as long as the edge interactions or the perturbation itself does not reach a critical strength. (Here and henceforth we focus on repulsive edge interactions.) This allows for the possibility to tune the system through various phase transitions as the magnetic coupling or the edge inter- actions are varied, providing hallmark signatures of the FTI phase. To begin our analysis, we summarize the relevant el- ements of the FTI edge theory in the absence of per- turbations, following Ref. 5 and 7. The edge can be described in terms of two bosonic quantum fields φα. The label α =↑, ↓ if the z component of the spin is con- served; for more general spin-orbit couplings α = L, R (for left/right movers). The fields obey the Kac-Moody equal-time commutation relations [φα(x), φβ (y)] = (σ3)αβ iπ m sgn(x − y). (1) Here and henceforth σ1,2,3 denote the Pauli matrices. The density and current of the electrons (relative to the ground state) is given by ρα(x) = 1 2π ∂xφα and jα(x) = − 1 H =Z πmv X 2π ∂tφα, respectively. The Hamiltonian is Vαβρα(x)ρβ (x)dx. (2) ρα(x)ρα(x) + X αβ α The real positive definite matrix V accounts for screened two body interactions and v (with mv > 0) is the edge velocity for V = 0. Due to TRI, one has V = σ1V σ1[5], which implies V = u4112+u2σ1. The operator Ψ† qp,α(x) ∝ exp [i(σ3)ααφα(x)] (3) e,α ∝ [Ψ† creates an edge excitation of charge 1 m ; the electron op- qp,α]m. (The precise form of Eq. (3) erator is Ψ† depends on Klein and regularization factors which need not be specified for our purposes.) We emphasize that Eqs. (1)-(3) represent the most general abelian FTI edge theory with a single pair of edge fields that is compatible with TRI[5]. Let us now consider what happens if a magnetic per- turbation is present (Fig. 1), taken as a proximity fer- romagnet of length LM for definiteness. A Zeeman-like coupling of the counterpropagating electrons reads as HZ = EZ Z LM 0 n3[ρR − ρL] + n⊥[eiχΨ† eRΨeL + h.c.]dx, (4) where EZ measures the strength of the perturbation[13] and n is a unit vector related to the magnetization. (Our analysis does not depend on the precise form of this re- lation.) The n3 term can be dropped, as it can be elim- inated by a gauge transformation that leaves the total density and current invariant. This leaves us with the n⊥ term which describes backscattering. The perturbation ∝ Ψ† eRΨeL is however not the most general backscattering term that can be introduced, and 2 the presence of the ferromagnet might generate other terms as well. It is an important fact that these terms can contain only electron operators. In a 1 m FQH system, it follows from the statistical angle π m of quasiparticles that local operators can change the number of quasipar- ticles only by integer multiples of m[14]. In the systems we consider, quasiparticles come in two species related by time-reversal. They have self-statistics angle π m and trivial mutual statistics. This means that the number of quasiparticles can be changed only by multiples of m for each species. As L/R movers belong to opposite species, quasiparticle backscattering [∝ Ψ†n qpL, n 6= 0 mod m] is forbidden. In the case of the simplest spin-orbit coupling which conserves the z component of the spin this reduces to the requirement that the particle number for each spin is an integer, as it should be. qpRΨn The form of the n electron backscattering Ψ†n eL ∝ exp[nm(φR + φL)] (n ∈ N) suggests the introduction of the fields ϕ = m(φR + φL) and θ = φR − φL. They satisfy eRΨn [ϕ(x), θ(y)] = 2πi sgn(x − y) (5) and [ϕ(x), ϕ(y)] = [θ(x), θ(y)] = 0, while the Hamilto- nian becomes H =  8π Z u(mK)(∂xθ)2 + u mK (∂xϕ)2dx + ∞ Xn=1Z LM 0 an[Fnei(nϕ+χn) + h.c.]dx, (6) where Fn accounts for the Klein and regularization fac- tors. The commutator (5) and the quadratic part of H takes the form of a spinless Luttinger liquid (sLL) with interaction parameter mK and velocity u with K = r 1 + λ4 − λ2 1 + λ4 + λ2 , u = vq(1 + λ2 4) − λ2 2, (7) where λj = uj πmv . The density of sLL electrons is ρ = ∂xϕ 2π . Note that it is K that behaves as the standard Luttinger interaction parameter: K < 1 for repulsive edge interac- tions (λj > 0), while K = 1 corresponds to a noninteract- ing edge (λj = 0). (Even though we call λj = 0 "non- interacting", electrons form a correlated FTI edge fluid even in this case.) Eq. (6) shows that any K dependence will be through mK. Note also that n electron backscat- tering processes in the original theory remain n electron backscattering processes of the sLL. Despite its simplic- ity, this mapping proves convenient by allowing the use of some basic results for sLLs[15] to predict the behavior of the FTI edge - ferromagnet system. A similar mapping also appeared in the context of FQH antiwires[9, 16, 17]. The relevance of the backscattering terms with differ- ent n can be inferred from their scaling dimensions[15] ∆n = n2mK. The low energy properties are domi- nated by the process with the smallest ∆n: single elec- tron backscattering. Our observation that quasiparticle backscattering is forbidden now becomes crucial: were it not the case, it would be quasiparticle backscattering that determines the low energy behavior. Henceforth, we focus our attention to single electron backscattering and neglect the terms with n > 1. The remaining term is the original perturbation Eq. (4), thus a1 ∝ EZ and χ1 = χ. We will analyze the effect of the ferromagnet in two opposite limits, LM ≪ LT and LM ≫ LT where LT = u kB T is the thermal length. To establish experimentally whether LM ≪ LT or LM ≫ LT , one needs the value of u; this can be obtained for example from time-domain measurements[18]. For LM ≪ LT , the characteristic wavelength of the relevant excitations is much longer than LM , hence the magnet can be taken as a delta-function impurity with strength EZ LM . The behavior of the system can be in- ferred following the by now classic analysis of Ref. 15. For a weak Zeeman term, EZ LM ≪ u, the leading order RG flow of the dimensionless coupling cM = EZ LM is u dcM dl = (1 − mK)cM . (8) This immediately gives our first result, that the ferromag- net is an irrelevant perturbation, as long as the repulsive interactions are moderate, K > 1 m . This is in stark con- trast to the naive expectation that TRI breaking pertur- bation should always be relevant for a TRI topological phase. Eq. (8) also determines the temperature depen- dence of the leading correction to the linear conductance, δG(T ) ∝ T 2mK−2. (9) The correction decays as T → 0 for K > 1 m , while for K < 1 m it is divergent, indicating that the flow is to strong coupling. Eq. (9) is then valid only as long as δG(T ) G(cM =0) ≪ 1. The behavior in the strong coupling regime corresponds to the physical picture in which the backscat- tering at the ferromagnet is so strong that it effectively cuts the edge into two halves. We then have two half- infinite sLLs [with fields ϕj , θj with j = 1(2) to the left (right) of the impurity], and the most relevant process is single electron tunneling, described by[15] Htun ∝ dM ei(θ1−θ2)/2 + h.c. (10) Htun introduces a (−1)j2π kink in ϕj , implementing the desired changes in the sLL charge density. Note that a ±2π kink in ϕj amounts to a ± 2π m kink in φRj + φLj, which corresponds to the transfer of charge e m , i.e., quasi- particle tunneling in the physical system. The scaling dimension of dM is now[15] (mK)−1 leading to a con- ductance (defined without the constant contribution of the complementary edge) that decays for T → 0 as G ∝ T 2 mK −2. (11) The demonstration of the robustness of the edge modes against a ferromagnet in systems with moderate edge in- teractions would already provide a strong signature of the 3 FIG. 2. The Kosterlitz-Thouless flow in the regime LM ≫ LT with conducting (C) and insulating (I) phases (separated by the dashed line). FTI phase. An even more apparent signature would be a demonstration of the phase transition upon tuning the interaction strength across K = 1 m , e.g., by changing the confinement potential. Note that from the temperature dependence itself only the combination mK can be ex- tracted. In the K < 1 m case, the fact that the transport takes place through quasiparticle tunneling events can be used to obtain m independently through shot noise mea- surements; these are however more difficult technically than measuring the conductance. Let us now turn to the opposite limit, LM ≫ LT . The system is now governed by a sine-Gordon Hamiltonian, resulting in the RG equations[19] drZ dl dK dl = (2 − mK)rZ , = −γr2 Z K 2, (12) (13) to leading order in EZ. Here rZ = EZ is a dimension- Ec less coupling (Ec is the high energy cutoff), and γ is a nonuniversal positive constant. Equations (12),(13) describe a Kosterlitz-Thouless (KT) flow (Fig. 2) sep- arating conducting and insulating phases. For weak Zee- man coupling and K > 2 m the magnetic perturbation is again irrelevant (conducting phase). In this regime, the leading order correction to the linear conductance obeys δG = e2 ), with the short distance cutoff Lc, where we neglected the flow of the interaction parameter K as it scales only to second order in rZ . As LT defines the dephasing length[17], we expect an Ohmic behavior δG ∼ LM for LM ≫ LT . This leads to a power law decay h f (rZ ( LT )2−mK , LM LT Lc δG(T ) ∝ T 2mK−3 (14) as T → 0. As before, δG(T ) depends only on mK. In the complementary, strong coupling regime of Fig. 2 the system becomes gapped, and the conductance (of the rel- evant edge) is exponentially suppressed in LM (insulating phase). Reaching this regime does not require tuning K below 2 m ; it can also be reached through the KT transi- tion by increasing EZ . As we now show, in this regime m can be extracted independently using a setup with spa- tially varying magnetization, as suggested originally in the QSH context[20]. For large EZ , the Hamiltonian can be subjected to a saddle point analysis through the imag- inary time path integral for action S = R dxdτ L with L = [u(∂xϕ)2 +u−1(∂τ ϕ)2]+gM cos(ϕ+χ), (15) 1 8πmK 2πm e(χLM −χ0) where gM ∝ EZ . For large EZ and constant χ, the field ϕ is locked to a minimum of the cosine. Now if χ varies spatially from χ0 to χLM as one goes from 0 to LM , will be accumulated along the mag- charge netic region. To relate χ to the magnetization note that time-reversal amounts to χ → χ + π. A magnetic do- main wall with opposite polarizations thus corresponds to a 0 − π domain wall in χ. The trapped e 2m charge in such a structure can be conveniently detected through Coulomb blockade measurements[20]. So far we have seen that using ferromagnetic perturba- tions, we can measure mK through the temperature de- pendence of the conductance. We have also shown that m can be measured independently using a domain wall configuration, or noise measurements if small K or large EZ can be achieved. In closing we now show that an addi- tional conductance measurement in a QPC geometry (see Fig. 1) provides a way to access K and m separately, even if small K or large EZ is not reachable. We focus on the limit when the QPC is almost completely pinched off. In this case, transport is through the tunneling of electrons from the left to the right. Similarly to QSH QPCs[11], this problem can now be mapped to a spinful Luttinger liquid with charge and spin interactions Kρ = mK, Kσ = m 2m charge transport is dominated by single electron tunneling processes[21], with scaling dimension ∆e = m(K+K−1) . The T → 0 decay of the linear conductance is thus K . We find that for K > 1 2 GQPC(T ) ∝ T m(K+K−1)−2. (16) Our results can be used to devise protocols for mea- surements for all regimes of EZ and K. For example, starting with a LM ≫ LT setup, the conductance can either show a power law increase or an exponential sup- pression as T → 0. The former behavior is already in- dicative of an FTI with K > 2 m . The product mK can be measured through δG(T ), and a further measurement using the QPC setup provides m and K separately. In the latter case m can be directly measured using a do- main wall setup. The value of K can also be obtained from measurements in the LM ≪ LT limit. In summary, we have shown that the behavior of the edge modes under magnetic perturbations can be used to detect the FTI phase through conductance measure- ments. Searching for the predicted phase transitions and 4 the temperature dependences provides a straightforward way for identifying the FTI phase in future experiments. NRC acknowledges useful discussions with Ady Stern. This work was supported by EPSRC Grant EP/F032773/1 and BIRAX. [1] M. Z. Hasan and C. L. Kane, Rev. Mod. Phys. 82, 3045 (2010); X.-L. Qi and S.-C. Zhang, Rev. Mod. Phys. 83, 1057 (2011). [2] C. L. Kane and E. J. Mele, Phys. Rev. Lett. 95, 146802 (2005); C. L. Kane and E. J. Mele, Phys. Rev. Lett. 95, 226801 (2005); R. Roy, Phys. Rev. B 79, 195321 (2009); M. Konig et al., Science 318, 766 (2007). [3] B. A. Bernevig and S.-C. Zhang, Phys. Rev. Lett. 96, 106802 (2006). [4] E. Tang et al, Phys. Rev. Lett. 106, 236802 (2011); K. Sun et al., Phys. Rev. Lett. 106, 236803 (2011); T. Neupert et al., Phys. Rev. Lett. 106, 236804 (2011); S. A. Parameswaran, R. Roy, and S. L. Sondhi, arXiv:1106.4025 (2011); M. Goerbig, arXiv:1107.1986 (2011); G. Murthy and R. Shankar, arXiv:1108.5501 (2011); J. Venderbos, M. Daghofer, and J. v-d Brink, Phys. Rev. Lett. 107, 116401 (2011); J. Venderbos et al, arXiv:1109.5955 (2011). [5] T. Neupert, L. Santos, S. Ryu, C. Chamon, and C. Mudry, Phys. Rev. B 84, 165107 (2011). [6] C. Weeks and M. Franz, arXiv:1111.1447 (2011). [7] M. Levin and A. Stern, Phys. Rev. Lett. 103, 196803 (2009). [8] L. Santos, T. Neupert, S. Ryu, C. Chamon, and C. Mudry, Phys. Rev. B 84, 165138 (2011). [9] X.-G. Wen, Phys. Rev. B 44, 5708 (1991). [10] A. M. Chang, Rev. Mod. Phys. 75, 1449 (2003). [11] C.-Y. Hou, E.-A. Kim, and C. Chamon, Phys. Rev. Lett. 102, 076602 (2009); J. C. Y. Teo and C. L. Kane, Phys. Rev. B 79, 235321 (2009). [12] C. L. Kane, M. P. A. Fisher, and J. Polchinski, Phys. Rev. Lett. 72, 4129 (1994); C. L. Kane and M. P. A. Fisher, Phys. Rev. B 51, 13449 (1995); C. L. Kane and M. P. A. Fisher, Phys. Rev. B 52, 17393 (1995). [13] If the edge electrons have nonzero Fermi wavevector kF , EZ is the 2kF component of the perturbation. [14] D. J. Thouless and Y.-S. Wu, Phys. Rev. B 31, 1191 (1985); T. Einarsson, Phys. Rev. Lett. 64, 1995 (1990). [15] C. L. Kane and M. P. A. Fisher, Phys. Rev. Lett. 68, 1220 (1992); C. L. Kane and M. P. A. Fisher, Phys. Rev. B 46, 15233 (1992). [16] S. R. Renn and D. P. Arovas, Phys. Rev. B 51, 16832 (1995). [17] C. L. Kane and M. P. A. Fisher, Phys. Rev. B 56, 15231 (1997). [18] Ashoori, R. C. et al., Phys. Rev. B 45, 3894 (1992). [19] T. Giamarchi, Quantum physics in one dimension, vol. 121 (Oxford University Press, USA, 2004). [20] X. Qi, T. Hughes, and S. Zhang, Nature Physics 4, 273 (2008). [21] For K < 1 2m two electron processes become more rel- 2K , leading to a evant with scaling dimension ∆2e = m GQPC(T ) ∝ T m/K−2 decay. 

1205.1807

1

1205

A Transfer Hamiltonian approach in self-consistent field regime for transport in arbitrary quantum dot arrays

A transport methodology to study the electron transport between quantum dots arrays based in Transfer Hamiltonian approach is presented. The interactions between the quantum dots and between the quantum dots and the electrodes are introduced by transition rates and capacitive couplings. The effects of the local potential are computed within the self-consistent field regime. The model has been developed and expressed in a matrix form in order to make it extendable to larger systems. Transport through several quantum dot configurations have been studied in order to validate the model. Despite the simplicity of the model, well-known effects are satisfactorily reproduced and explained. The results qualitatively agree with other results obtained using more complex theoretical approaches.

cond-mat.mes-hall

cond-mat

A Transfer Hamiltonian approach in self-consistent field regime for transport in arbitrary quantum dot arrays S. Illera MIND/IN2UB Departament d'Electr`onica, Universitat de Barcelona, C/Mart´ı i Franqu`es 1, E-08028 Barcelona, Spain E-mail: [email protected] N. Garcia-Castello MIND/IN2UB Departament d'Electr`onica, Universitat de Barcelona, C/Mart´ı i Franqu`es 1, E-08028 Barcelona, Spain J. D. Prades MIND/IN2UB Departament d'Electr`onica, Universitat de Barcelona, C/Mart´ı i Franqu`es 1, E-08028 Barcelona, Spain A. Cirera MIND/IN2UB Departament d'Electr`onica, Universitat de Barcelona, C/Mart´ı i Franqu`es 1, E-08028 Barcelona, Spain Abstract. A transport methodology to study the electron transport between quantum dots arrays based in Transfer Hamiltonian approach is presented. The interactions between the quantum dots and between the quantum dots and the electrodes are introduced by transition rates and capacitive couplings. The effects of the local potential are computed within the self-consistent field regime. The model has been developed and expressed in a matrix form in order to make it extendable to larger systems. Transport through several quantum dot configurations have been studied in order to validate the model. Despite the simplicity of the model, well-known effects are satisfactorily reproduced and explained. The results qualitatively agree with other results obtained using more complex theoretical approaches. PACS numbers: 72.10.Bg, 73.63.-b, 73.63.Kv 2 1 0 2 y a M 8 ] l l a h - s e m . t a m - d n o c [ 1 v 7 0 8 1 . 5 0 2 1 : v i X r a A Transfer Hamiltonian approach in self-consistent field regime for transport in arbitrary quantum dot arrays 2 1. Introduction Confined structures have been available to the experimentalist for a very long time, the MOS (metal-oxide-semiconductor) transistor is the archetype of a confined two- dimensional system [1]. Nevertheless, the possibility to enhance this confinement by embedding low-dimensional structures in an insulating matrix has renewed the interest. These structures (quantum dots, wires or layers) can be used in single- electron device [2],new memory concepts [3] and photon or electroluminescent devices [4]. Concerning quantum dots (Qds), they are particularly attractive because they possess discrete energy levels and quantum properties similar to natural atoms or molecules. From a fundamental point of view, research has been mostly concentrated on single quantum dots. These simple systems have been studied using many-body approaches, including non-equilibrium Green's function formalism (NEGFF) [5, 6]. From a practical point of view, many novel phenomena have been discovered, such as the staircaselike current-voltage (I-V) characteristic [7], Coulomb blockade oscillation [8], negative differential capacitance [9], and the Kondo effect in Qds [10]. Researchers have recently paid much attention to electron transport through several Qds since multiple Qd provides more Feynman paths for the electron transmission [11]. However, up to now the only computation of transport in an extended arbitrary array of Qds was done by Carreras et al. [12] but no local potential due to self-charge was included. Sun et al.[13] have also studied the electron transport using NEGFF for different arrangements of Qds, from one to three Qds, without including the potential due to self-charge neither. The inclusion of the self-charge potential using this complex framework is usually impossible for large systems. In this work, we use non-coherent rate equations (NCRE) [14, 15] to study the electrical transport in Qds in an extendible, arbitrary, matrix of Qds taking into account self charge effects. In a previous work [16], we applied NCRE to obtain analytical solutions for electron transport in simple cases. Using this approach each Qd is treated as a separate system, therefore we can write a NCRE for each dot since these equations describe relationship between the charge inside the Qds and applied bias voltage. The interactions between the Qds, and between the Qds and the electrodes are introduced by transition rates and capacitive couplings. Electron transport and charge densities inside the Qds depend on the tunnel transparency of the barriers limiting each dot. In order to effectively solve the multielectron problem, the effects of the local potential are computed within the self-consistent field (SCF) regime. Moreover, we show how our approach can be easily extended to an arbitrary number of Qds and configurations using a matrix formalism. Therefore, this methodology allows to simulate realistic devices based on large scale Qds arrays. Finally, we compare this methodology with NEGFF, obtaining similar results. 2. Theoretical background Our system consist of two electrodes (L lead and R lead) coupled to a central transport region. The central region contains several quantum dots, N Qds, distributed inside of an insulator matrix. In order to find the current voltage curve I-V of the total system we use the transfer Hamiltonian formalism[17, 18]. Using this formalism we can write an expression for the current flowing across two parts of the system. Assuming no inelastic scattering and symmetry in the transmission coefficient [19] the net current A Transfer Hamiltonian approach in self-consistent field regime for transport in arbitrary quantum dot arrays 3 flux between two parts of the system is (cid:90) Iij = 4πq ¯h Tij(E)ρi(E)ρj(E)(fj(E) − fi(E))dE, (1) where Tij(E) is the transmission probability, ρi(E) and ρj(E) are the density of states while fi(E) and fj(E) are the distribution functions of the different parts of the system. In the equilibrium, the electrochemical potential of the whole system is equal and the particular distribution functions (DFs) are described by the equilibrium Fermi Dirac DF therefore the current between each part of the system is zero. If an external bias voltage (V) is applied, which will drive the system out of the equilibrium, the electrochemical potential of the leads will change by µL−µR = qV . From the definition of the total charge Ni inside the ithQd, we can write (cid:90) Ni = ρi(E)ni(E)dE, (2) where ni(E) is an unknown DF and ρi is the density of states (DOS) of the ith Qd. For sake of clarity, we only consider one single state with energy level
in each Qd. In order to take into account the coupling with the surrounding elements we assign a Lorentzian shape DOS centered in
. We can write the evolution charge in time for (cid:82) Ijidt, where the subscript i refers to ithQd and j runs over each Qd as Ni = (cid:80) the other components of the system. Thus, a set of integro-differential equations are obtained for the time charge evolution j (cid:90) TLiρLρi(fL − ni)dE + TRiρRρi(fR − ni)dE Tjiρjρi(nj − ni)dE) ∀i = 1 . . . N, (3) (cid:90) ( dNi dt = (N−1)(cid:88) 4πq ¯h (cid:90) + j(cid:54)=i where we explicitly write all the current terms: the leads current contributions (first and second term) and the neighbor contribution (the last term). We assume that the DFs in the electrodes (fL and fR) are similar to the Fermi Dirac DF using different electrochemical potentials (µL and µR). Equation (3) can be rewritten for the steady state and assuming no inelastic scattering we can obtain the DF in each Qd for each energy step as a solution of the system of equations j(cid:54)=1  −TL1ρL − TR1ρR −(cid:80)(N−1)  −TL1ρLfL − TR1ρRfR  . T1N ρ1 −TLN ρLfL − TRN ρRfR ... ... = T1jρj T1N ρN . . . . . . . . . −TLN ρL − TRN ρR −(cid:80)(N−1) ... j(cid:54)=N TN jρj   n1 ... nN  (4) The effect of the applied voltage to the external electrodes on the electrostatic potential inside each Qd must also be taken into account. The classical solution for the potential at each quantum dot (Vi) involves the Poisson equation (cid:126)∇ · (εr (cid:126)∇Vi) = − q(cid:52)Ni Ωε0 , (5) A Transfer Hamiltonian approach in self-consistent field regime for transport in arbitrary quantum dot arrays 4 q2 (6) Ui = where εr is the relative permittivity of the dielectric media, ε0 is the vacuum permittivity and Ω is the Qd volume. The general solution for the potential energy Ui = −qVi in the ith Qd is [20] Cij Ctot,i (−qVj) + (cid:88) (cid:52)Ni, j(cid:54)=i Ctot,i pling between the different components and Ctot,i =(cid:80) where the subscript j runs over all components of the system, Cij is the capacitive cou- j,j(cid:54)=i Cij is the total capacitive coupling of ithQd. The charge energy constant U0i = q2/Ctot,i is the potential increase as a consequence of the injection of one electron into the Qd and (cid:52)Ni is the change in the number of electrons, calculated respect to the number of electrons N0 initially in the ithQd. The effects of local potential on each Qd, which modify the Qd charge and the currents, should be taken into account in the Qd DOS ρi(E) → ρi(E − Ui). In (6) we observe that the local potential depends on the increasing charge density but at the same time the charge depends on the DOS that it is modified by the local potential. These considerations impose a self-consistent solution of (2) and (6). 3. Results and discussion In this section, we first show the calculated current voltage curves I-V for different arrangements and we compare their with the results obtained using NEGFF [13]. We also present the number of electrons Ni accumulated in the ithQd in each configuration. In these cases, analytical expressions for the current are presented as well. Finally, the extension of the model developed in the previous section is presented as a powerful method to study the electron transport in an arbitrary extended array of Qds. The electrochemical potentials in the two leads are set at µL = 0 and µR = −qV . Electrons flow from the left lead to the right one. For simplicity, we consider that the transmission probability is constant and the same between all the parts of the system. We do not consider direct transmission between the leads. For clarity the DOS of the leads were considered constant in all the energy range. Using this framework, transport without inelastic scattering, the position of the energy levels in the Qds plays an important role therefore the evolution of it with the applied bias voltage define the shape of the I(V) curve. As expected, the I(V) curves exhibit strong dependence with the electrostatic coupling of the different parts of the system. We present expressions for the evolution of the energy level with the applied bias voltage assuming that the elements which are coupled have equal capacitive coupling between them. We set a constant charge energy for all Qds, U0 = 0.25eV . 3.1. One single Qd We briefly review electron transport through one Qd. Using (3) and taking into account only lead contributions the current can be written as (fL − fR)dE. (7) (cid:90) TR1TL1ρLρ1ρR TL1ρL + TR1ρr I = 4πq ¯h Fig. 1 shows the numerical result of the current I(V). In the calculation we assumed symmetric coupling respect to the leads, TR1 = TL1 = 0.2 ‡. The evolution of the ‡ We use similar transmission values than Sun et al.[13] in order to make possible the qualitative comparison between models. A Transfer Hamiltonian approach in self-consistent field regime for transport in arbitrary quantum dot arrays 5 Figure 1. (a) The I-V curve for one single Qd obtained using NCRE. We also show the NEGFF results for the same system, the NEGFF data are taken from Sun et al. [13]. (b) The electron number in the Qd as a function of the applied bias V. The inset shows the connection geometry. The rectangles represent two leads and the circle represents a Qd. energy level with the applied bias voltage is
1(V ) = 1 − V /2 + U0(cid:52)N1, (8) where the second and third terms are due to the electrostatic effect. As expected the current increases with the bias when the energy of the Qd moves across the left lead; which is µL =
1(V ) → V ≈ 2. When V is hight enough, the current saturates to a constant value as
1(V ) is placed between the two electrochemical potentials of the leads. Fig. 1(b) shows the dependence of the electron number with the applied bias. 3.2. Two Qds We now study the case of two Qds. There are four different connection geometries between Qds and leads. In our calculations we assume symmetric coupling respect to the leads, TR1 = TL1 = 0.2, and the Qd coupling T12 = 0.2. 3.2.1. Parallel case The first configuration of two Qds is the case that they are in parallel. Both Qd are coupled to all elements of the system, the leads and the neighbor Qd. In this configuration the expressions for the current are (cid:90) TL1TR1(TL1ρL + T1RρR + T12(ρ1 + ρ2))ρLρRρ1 (cid:90) TL1TR1(TL1ρL + T1RρR + T12(ρ1 + ρ2))ρLρRρ2 D2 4πq ¯h I1 = ×(fL − fR)dE (9b) where D2 = (T1RρL + T1LρR)2 + T1LT12ρR(ρ1 + ρ2) + TL1T12ρL(ρ1 + ρ2). The position of the energy level of each Qd is (9a) (10a) (10b) 4πq ¯h I2 = ×(fL − fR)dE, D2
1(V ) = 1 − qV /3 − qV2/3 + U0(cid:52)N1
2(V ) = 3.5 − qV /3 − qV1/3 + U0(cid:52)N2. A Transfer Hamiltonian approach in self-consistent field regime for transport in arbitrary quantum dot arrays 6 Figure 2. (a) The total I-V curve and partial I-V curves obtained using NCRE for a parallel configuration. The NEGFF results are taken from Sun et al. [13]. (b) The electron number in the Qds as a function of the applied bias V. We show the total and partial currents Fig. 2(a). The I-V curve shows two steps when the energy levels of the Qds are placed between the electrochemical potentials of the leads. This case is equivalent to a single Qd with two energy levels. Fig. 2(b) shows the electron number ni with the applied bias voltage. The charge increases until it reach the saturation value. Figure 3. (a) The I-V curve for two Qds in a serial configuration obtained using NCRE. We also show the NEGFF results for the same system, the NEGFF data are taken from Sun et al. [13]. The inset shows the connection geometry. (b) The electron number in the Qds as a function of the applied bias V. 3.2.2. Serial case The second type of arrangement is the case of two Qds in a serial configuration. The system is shown in the inset of Fig. 3(a). Each Qd only interacts with one lead and the other Qd. In this case, the expression for the current is (cid:90) 4πq ¯h I = ×(fL − fR)dE TL1T12T2RρLρ1ρ2ρR TL1T12ρ1ρL + TL1T2RρRρL + T12T2Rρ2ρR (11) A Transfer Hamiltonian approach in self-consistent field regime for transport in arbitrary quantum dot arrays 7 and the evolution of the energy level of each Qd with the applied bias voltage is
1(V ) = 1 − qV2/2 + U0(cid:52)N1
2(V ) = 3.5 − qV /2 − qV1/2 + U0(cid:52)N2, (12a) (12b) where we assumed that the Qds are only coupled to each other and to one lead. In order to have current flowing through the system, the energy levels must lie between the electrochemical potentials of the leads and overlapping of the Qd energy levels is necessary. This means that the electrons need available states in each part of the system in order to move from Left lead to Right lead. When the energy levels are equal,
1 =
2 → V ≈ 7.5, this is a maximum overlapping between Qd DOS, the current is maximum and the system is in a resonance state therefore the channel is open. When the voltage increases further the Qd DOS overlapping decreases. Therefore a negative differential resistance appears [21]. In Fig. 3(b) we show the evolution of the charge in each Qd Ni as a function of the applied voltage V. Initially, N1 increases since the channel between the first and second Qd is closed. At the resonant condition, the channel between the Qds opens and some charge stored in the first Qd flows to the second Qd. At higher voltages the channel closes again and N1 stores all the incoming charge, while N2 loses its charge. Figure 4. (a) The I-V curve, for the configuration plotted in the inset, obtained using NCRE. We also show the NEGFF results for the same system, the NEGFF data are taken from Sun et al. [13]. (b) The electron number in the Qds as a function of the applied bias V. 3.2.3. Other two Qds configurations We first examine the case in which one Qd interacts with the two leads and it is also connected to the second Qd, while the second Qd is only connected to the first Qd. The current is (cid:90) TR1TL1ρLρ1ρR TL1ρL + TR1ρr I = 4πq ¯h (fL − fR)dE and the position of the energy levels are
1(V ) = 1 − qV /3 − qV2/3 + U0(cid:52)N1
2(V ) = 3.5 − qV1 + U0(cid:52)N2. The obtained current expression 13 is the same than the one we obtained for the single Qd case. The DF in the second Qd is the same as in the first Qd therefore the current (13) (14a) (14b) A Transfer Hamiltonian approach in self-consistent field regime for transport in arbitrary quantum dot arrays 8 Figure 5. (a) The total I-V curve and partial I-V curves obtained using NCRE for the configuration showed in the inset. The NEGFF results are taken from Sun et al. [13]. (b) The electron number in the Qds as a function of the applied bias V. between the Qds is zero. The results are presented in Fig. 4. The second arrangement of Qds is shown in the inset of Fig. 5. The expressions for the current are I1 = I2 = 4πq ¯h 4πq ¯h D (cid:90) T1Rρ1ρR(TR2ρRTL1ρL + T12ρ1TL1ρL) (cid:90) T2Rρ2ρRT12ρ1TL1ρL D (fL − fR)dE, (fL − fR)dE (15a) (15b) where D = T2RρRTL1ρL + TR1TR2ρ2 the total current is I = I1 + I2. The energy level position is R + TR2ρRT12ρ2 + T12ρ1TL1ρL + TR1ρRT12ρ1 and
1(V ) = 1 − qV /3 − qV2/3 + U0(cid:52)N1
2(V ) = 3.5 − qV1/2 + U0(cid:52)N2. (16a) (16b) In this case we show the total and partial currents. The I-V partial current shows an interesting behavior. The current through the first Qd is similar to the single one Qd configuration but the current through the second reminds the slope of a resonant state. This fact can be easily understood in the following way: if the channel between the two Qds is closed the current only flows through the first Qd. When the Qd1-Qd2 channel is opened the Qd2 also conducts. In the same case as before, when the voltage increases the overlapping decreases and the Qd2 current decreases. 3.3. Three Qds The methodology developed in the first section can be easily extended into more complicated systems. Here, we present the results for some configurations based in three Qds. The analytical expressions for the current are too large to write here but in Fig. 6 [(a),(c),(e)] we show the I-V curves and the charge in each Qd Fig. 6[(b),(d),(f)]. As we have shown before the position of the energy levels plays an important role in the I-V and N-V curves, using 6 we can write the position of the each energy level as A Transfer Hamiltonian approach in self-consistent field regime for transport in arbitrary quantum dot arrays 9 Figure 6. [(a),(c) and (e)] The I-V curves and the electron number [(b), (d) and (f)] respectively for three Qds with different configuration. The insets show the connection geometry. The NEGFF results are taken from Sun et al. [13]. a function of the applied bias voltage C1j
1(V ) = 1 −(cid:88)
2(V ) = 2 −(cid:88)
3(V ) = 3.5 −(cid:88) j j Vj + U0(cid:52)N1 Vj + U0(cid:52)N2 Ctotal1 C2j Ctotal2 C3j Ctotal3 Vj + U0(cid:52)N3, j (17a) (17b) (17c) where the subscript j runs over all connected elements of the system. The Qd-lead coupling and the interdot coupling are set equal Tij = 0.2. In the insets of the Fig. 6 we show the scheme of the system under study. 3.4. Large Qds arrangement To conclude we present the results for larger systems that they are close to the experimental measurements. The systems are formed by 100 Qds placed in a parallel configuration, serial configuration and in an array geometry (10 × 10). The total I-V curves and the geometries are presented in Fig. 7. The Qd-lead coupling and the interdot coupling are set equal Tij = 0.2. The capacitance between the linked elements are also equal. In order to represent an experimental system we considerer that the value of the energy level of each dot follows a normal distribution with mean value 1eV and deviation 0.2eV. This fact represents the usual distribution size that appears in the experiments. The relationship between the Qd radius and the energy level position is a well known effect and it is related to the quantum confinement of the electrons [22]. The I-V curves show an interesting behavior. First, in the parallel case, Fig. 7(a), the I-V curve shows a staircaselike structure and saturates to a constant value at high bias. A Transfer Hamiltonian approach in self-consistent field regime for transport in arbitrary quantum dot arrays 10 Figure 7. The I-V curves for the larger systems: (a) 100 Qds in parallel configuration, (b) 100 Qds in serial configuration and (c) 100 Qds in an array disposition 10 × 10. As we have seen before in the parallel configuration each Qd acts as an independent channel therefore the total current is the sum of all partial currents. As expected, the saturation current is 100 times the saturation current of a single dot. For the serial configuration, Fig. 7(b), we obtain a current peak as we expect due to the resonant state is necessary in order to have electron transport in this configuration. The maximum value of the peak is hard to determine because it depends on the transmission coefficient, but it also depends of the overlapping between the DOS of the Qds. Concerning to the array configuration, Fig. 7(c), the I-V curve is determined by a combination of the two previous cases. In order to have transport the resonant state condition must be fulfilled therefore a current peak appears but the total current is the sum of the partial currents of each row. 3.5. Comparison with NEGFF Finally, the results obtained using the proposed approach have been compared to the results of Sun et al. [13]. In that paper the authors have used the nonequilibrium Green's function method (NEGFF) to study the electron transport between one, two and three Qds in several configurations. Their I-V results have been plotted in our figures (NEGFF in the legend). The main results are: • The results presented in Figs. 1, 3, 2, 6(a) and 6(c) are in accordance between the two approaches. For the serial configuration Fig. 3 the differences are due to the different values of the Qd coupling, we also obtain a resonant peak when the energy levels of the Qds are placed in a resonant state. The resonant state is strongly dependent on the capacitive coupling of the Qds, as the position of the energy level with the applied bias voltage depends on the capacitive coupling of the Qd. In the parallel configuration we obtain the same staircase shape, but, in our case we also take into account the energy charge terms, therefore the steps occur at higher voltages. • The main difference appears in the case described in Fig. 4. For this configuration Sun et al. predicts an antiresonance effect. We do not recover this effect because A Transfer Hamiltonian approach in self-consistent field regime for transport in arbitrary quantum dot arrays 11 our model considers each Qd as a separate quantum systems. For this reason, our approach is known as a non-coherent model. • For the systems presented in Figs. 5 and 6(b) we obtain similar results. The position of the current peak is different because Sun et al. assume that the bias is uniformly applied throughout the whole system meanwhile we take into account all the electrostatic coupling between the different parts of the system. As we have shown the electrostatic coupling plays an important role to determine the I-V characteristic of the system. The electrostatic effect has two terms: the first term is determined by the influence of the leads and the neighbor Qd and it is described by the capacitive coupling of the Qd and its surrounding. The second term takes into account the charge stored inside the Qd, this effect is related to the electron-electron interaction and the self consistent solution of 2 and 6 is the first approach to introduce many body effects, like the Coulomb Blockade. If we create nanodevices in oder to take advantage to the quantization of the current, only small number of discrete energy levels are available for conduction, the accurate control of the energy levels with the applied bias voltage is one of the most important points that we need to take into account. Therefore a good modelization of the Qd-Qd and Qd-lead capacitances is necessary. This paper precedes future works in which realistic DOS, energy dependent transmission coefficients as well as a realistic capacitive couplings can be introduced. 4. Conclusion We propose a theoretical model to study the electron current in systems based in quantum dots (Qds). This model is based in the transfer Hamiltonian formalism and computes the I-V and N-V curves in the self consistent field regime (SCF), using non- coherent rate equations (NCRE). This approach provides a simple and transparent method to describe the electron transport. Due to the simplicity of the model, this can be easily extended to analyze arbitrary large arrays of Qds of interest in technological applications. Despite its simplicity and in contrast with other approaches the effect of self-charge has been taken into account, by solving the Poisson equation with appropriate boundary conditions for each Qd. As expected, the calculation of the local potential inside each Qd is one of the most critical points, since the I-V curves depends on the position of the energy level. In order to show the potential of this method to analyze realistic configurations, we have studied the electron transport between different Qd configurations. We have also compared the NCRE results with well established data obtained with the NEGFF approach. Such a successful comparison shows that NCRE is a powerful and intuitive method to describe the electron transport. Acknowledgments N. Garcia acknowledges the Spanish MICINN for her PhD grant in the FPU program. A. Cirera acknowledges support from ICREA academia program. The authors thankfully acknowledge the computer resources, technical expertise and assistance provided by the Barcelona Supercomputing Center - Centro Nacional de Supercomputaci´on. The research leading to these results has received funding from A Transfer Hamiltonian approach in self-consistent field regime for transport in arbitrary quantum dot arrays 12 the European Communitys Seventh Framework Programme (FP7/2007-2013) under grant agreement n: 245977. References [1] Ando T, Fowler A B and Stern F 1982 Rev. Mod. Phys. 54(2) 437 -- 672 URL http://link.aps. org/doi/10.1103/RevModPhys.54.437 [2] Meirav U and Foxman E B 1996 Semiconductor Science and Technology 11 255 URL http: //stacks.iop.org/0268-1242/11/i=3/a=003 [3] Tiwari S, Rana F, Chan K, Shi L and Hanafi H 1996 Applied Physics Letters 69 1232 -- 1234 URL http://link.aip.org/link/?APL/69/1232/1 [4] Lockwood D J Semiconductor and Semimetals, Light Emission in Silicon From Physics to Devices (Academic Press, San Diego) [5] Yeyati A L, Mart´ın Rodero A and Flores F 1993 Phys. Rev. Lett. 71(18) 2991 -- 2994 URL http://link.aps.org/doi/10.1103/PhysRevLett.71.2991 [6] Meir Y, Wingreen N S and Lee P A 1991 Phys. Rev. Lett. 66(23) 3048 -- 3051 URL http: //link.aps.org/doi/10.1103/PhysRevLett.66.3048 [7] Barner J B and Ruggiero S T 1987 Phys. Rev. Lett. 59(7) 807 -- 810 URL http://link.aps.org/ doi/10.1103/PhysRevLett.59.807 [8] Weis J, Haug R J, Klitzing K v and Ploog K 1993 Phys. Rev. Lett. 71(24) 4019 -- 4022 URL http://link.aps.org/doi/10.1103/PhysRevLett.71.4019 [9] Wang S D, Sun Z Z, Cue N, Xu H Q and Wang X R 2002 Phys. Rev. B 65(12) 125307 URL http://link.aps.org/doi/10.1103/PhysRevB.65.125307 [10] van der Wiel W G, Franceschi S D, Fujisawa T, Elzerman J M, Tarucha S and Kouwenhoven L P 2000 Science 289 2105 -- 2108 (Preprint http://www.sciencemag.org/content/289/5487/ 2105.full.pdf) URL http://www.sciencemag.org/content/289/5487/2105.abstract [11] Gong W, Zheng Y, Liu Y and L T 2008 Physica E: Low-dimensional Systems and Nanostructures 40 618 -- 626 ISSN 1386-9477 URL http://www.sciencedirect.com/science/article/pii/ S1386947707004390 [12] Carreras J, Jambois O, Lombardo S and Garrido B 2009 Nanotechnology 20 155201 URL http://stacks.iop.org/0957-4484/20/i=15/a=155201 [13] Sun Z Z, Zhang R Q, Fan W and Wang X R 2009 Journal of Applied Physics 105 043706 (pages 7) URL http://link.aip.org/link/?JAP/105/043706/1 [14] Averin D V, Korotkov A N and Likharev K K 1991 Phys. Rev. B 44(12) 6199 -- 6211 URL http://link.aps.org/doi/10.1103/PhysRevB.44.6199 [15] Gurvitz S A 1998 Phys. Rev. B 57(11) 6602 -- 6611 URL http://link.aps.org/doi/10.1103/ PhysRevB.57.6602 [16] Illera S, Prades J D, Cirera A and Cornet A 2012 EPL (Europhysics Letters) 98 17003 URL http://stacks.iop.org/0295-5075/98/i=1/a=17003 [17] Payne M C 1986 Journal of Physics C: Solid State Physics 19 1145 URL http://stacks.iop. org/0022-3719/19/i=8/a=013 [18] Passoni M and Bottani C E 2007 Phys. Rev. B 76(11) 115404 URL http://link.aps.org/doi/ 10.1103/PhysRevB.76.115404 [19] Supriyo D 1995 Electronic Transport in Mesoscopic Systems (Cambridge University Press) [20] Datta S 2004 Nanotechnology 15 S433 URL http://stacks.iop.org/0957-4484/15/i=7/a=051 [21] Borgstrom M, Bryllert T, Sass T, Gustafson B, Wernersson L E, Seifert W and Samuelson L 2001 Applied Physics Letters 78 3232 -- 3234 URL http://link.aip.org/link/?APL/78/3232/1 [22] Proot J P, Delerue C and Allan G 1992 Applied Physics Letters 61 1948 -- 1950 URL http: //link.aip.org/link/?APL/61/1948/1 

1312.3362

1

1312

First Principles Study of Bismuth Films at Transition Metal Grain Boundaries

Recent experiments suggest that Bi impurities segregate to form bilayer films on Ni and Cu grain boundaries but do not segregate in Fe. To explain these phenomena, we study the total energies of Bi films on transition metal (TM) $\Sigma$3(111) and $\Sigma$5(012) GBs using density functional theory. Our results agree with the observed stabilities. We propose a model to predict Bi bilayer stability at Ni GBs which suggests that Bi bilayer is not stable on (111) twist CSL GBs but is stable in most (100) twist CSL GBs. We investigate the interaction and bonding character between Bi and TMs to explain the differences among TMs based on localization of orbitals and magnetism.

cond-mat.mes-hall

cond-mat

Supplemental Material for: First Principles Study of Bismuth Films at Transition Metal Grain Boundaries Qin Gao and Michael Widom Department of Physics, Carnegie Mellon University, Pittsburgh, Pennsylvania 15213, USA PACS numbers: 3 1 0 2 c e D 1 1 ] l l a h - s e m . t a m - d n o c [ 1 v 2 6 3 3 . 2 1 3 1 : v i X r a 1 BI BILAYER ON FE Σ5(012) GB We calculated Bi bilayer enthalpy on Fe Σ5(012), a high energy GB which is created by cleaving the BCC bulk along the (012) plane and rotating one grain around [001] by 53.1o and rejoining the two parts. Our calculated GB energy is 98 meV/A2, close to the GB energy 104 meV/A2 of the lowest energy structure in the literature [1]. We first studied Bi monolayers on Fe (012) surfaces and then calculated bilayer films on the GB with the stable surface structure at two sides of the GB plane. We used 10 layers of Fe at each side of the Bi film. The relaxed structure is shown in Fig. S1. The lowest ∆H/A is +17 meV/A2 for bilayer films, which is large and positive indicating that the Bi bilayer is not stable even on this high energy GB. Fe Bi 10 5 0 -5 Fe Bi 5 0 -5 -10 -10 -5 0 5 10 -5 0 5 FIG. S1: (Color online) Side view (left) of relaxed Bi bilayer at Fe Σ5(012) GB and top view (right) of Bi monolayer on one side of the GB plane. The cyan cell is the Fe GB unit cell, the green cell is the Bi segregated GB unit cell. Atom size indicates depth (large below small). Length units are in A. 2 GB Σ 3(111) Σ 5(120) Co -1.6 80 Ni Cu 2.8 (2.7 [2]) 0.9 (1.4 [2]) 77 (89 [3]) 55 (59 [4]) TABLE S1: GB energies, units are meV/A2. The energy conversion factor is 1 meV/A2 =0.016 J/m2. Values from other studies are in parentheses. Note the Co Σ3(111) GB energy is negative because the T = 0 K state is HCP rather than the high-T FCC that we choose to compare with. Surface Esurf ∆HML/A Emin GB (111) 0.118 -0.090 0.056 (001) 0.137 -0.122 0.030 (120) 0.150 -0.133 0.034 TABLE S2: Calculated input quantities for the enthalpy model (Eq. 3). Predicted Emin GB values for GB with the same surface plane (i.e. a = b) at two sides. The energy units are eV/A2. The Bi monolayer structure on Ni(100) surface is the c(2 × 2) structure as observed in experiment [5]. GB EGB ∆H model/A ∆H calc/A W bare sep Wsep Reduction Σ3(111) 0.003 Σ7(111) 0.029 0.053 0.027 0.045 0.235 0.009 96.2% 0.023 0.209 0.009 95.7% Σ5(100) 0.064 -0.034 -0.037 0.208 0.007 96.6% Σ5(120) 0.077 -0.043 -0.054 0.220 0.010 95.5% (111)/(100) 0.055 -0.012 -0.004 0.207 0.004 98.0% TABLE S3: Model and calculated Ni GB energies and Bi bilayer enthalpies of formation at different Ni GBs. Work of separation for bare and Bi bilayer segregated GBs is shown on right. The energy units are eV/A2. The Σ7(111) GB is made by twisting the one side of bulk Ni by 21.8o around the [111] axis with (111) as GB plane. The resulting GB cell is [3-112] as defined in [6] on which Bi favors 3 atoms per layer. The Σ5(100) GB is made by twisting one side of bulk Ni by 36.9o around the [001] axis with (001) as GB plane. 3 Co(nonmag) Co(mag) Ni(nonmag) Ni(mag) Cu Esurf (eV/A2) ∆H (eV/A2) 0.193 0.164 0.152 0.150 0.100 -0.160 -0.118 -0.152 -0.133 -0.070 iCOHP(Bi-TM) iCOHP(TM-TM)a iCOHP(TM-TM)b -1.75 -1.38 -1.17 -1.63 -1.32 -1.16 -1.77 -1.13 -0.85 -1.75 -1.13 -0.83 -1.33 -0.43 -0.66 TABLE S4: The (012) surface energy, Bi monolayer enthalpies of formation, integrated COHP (iCOHP) energies of Bi-TM bond, TM-TM bond near to Bi(a) and TM-TM bond away from impurities(b). The energy units are eV/bond for the iCOHP energies. ] V e / d n o b / V e [ P H O C d Co -0.5 -10 0.5 0 0.5 0 Ni -0.5 -10 0.5 0 Cu -0.5 -10 bulk near Bi 5 5 5 -5 -5 -5 E-EF [eV] 0 0 0 FIG. S2: (Color online) Differential COHP of Metal-Metal interaction in the bulk and near to Bi. Negative is bonding while positive is antibonding. The Ni and Co results are the summation of two spin components. The dashed green line is the x axis. The zero in x axis is the Fermi energy. 4 [1] T. Ossowski, J. Kuriplach, E. Zhurkin, M. Hou, and A. Kiejna (Pre- sented as the 27th Max Born Symposium, Wroclaw, Poland, 2010), URL http://www.ift.uni.wroc.pl/~mborn27/download/files/SessionX_3_Ossowski.pdf. [2] V. L. E. Murr, Interfacial Phenomena in Metal and Alloys, Addison-Wesley Publishing Com- pany p. 138 (1975). [3] M. Yamaguchi, M. Shiga, and H. Kaburaki, Journal of Phys.: Condens. Matter 16, 3933 (2004). [4] M. A. Tschopp and D. L. Mcdowell, Phil. Mag. 87, 3871 (2007). [5] C. Panja, M. E. Jones, J. M. Heitzinger, S. C. Gebhard, and B. E. Koel, The Journal of Physical Chemistry B 104, 3130 (2000). [6] T. R. J. Bollmann, R. van Gastel, H. J. W. Zandvliet, and B. Poelsema, Phys. Rev. Lett. 107, 176102 (2011). 5 3 1 0 2 c e D 1 1 ] l l a h - s e m . t a m - d n o c [ 1 v 2 6 3 3 . 2 1 3 1 : v i X r a First Principles Study of Bismuth Films at Transition Metal Grain Boundaries Department of Physics, Carnegie Mellon University, Pittsburgh, Pennsylvania 15213, USA Qin Gao and Michael Widom Recent experiments suggest that Bi impurities segregate to form bilayer films on Ni and Cu grain boundaries but do not segregate in Fe. To explain these phenomena, we study the total energies of Bi films on transition metal (TM) Σ3(111) and Σ5(012) GBs using density functional theory. Our results agree with the observed stabilities. We propose a model to predict Bi bilayer stability at Ni GBs which suggests that Bi bilayer is not stable on (111) twist CSL GBs but is stable in most (100) twist CSL GBs. We investigate the interaction and bonding character between Bi and TMs to explain the differences among TMs based on localization of orbitals and magnetism. Segregation at grain boundaries (GBs) affects various properties of polycrystals such as grain growth [1, 2], liq- uid metal embrittlement (LME) [3, 4] and corrosion [5, 6]. However, the exact segregated structures, and hence the underlying mechanisms at atomic level, are far from be- ing fully revealed. As a generalization of Gibbs' defini- tion of phase, the new concept "complexion" was pro- posed to describe thermodynamically stable interfacial structures [7, 8]. A particular type of complexion, Dillon- Harmer complexions [9], attracted attention after the dis- covery of six discrete complexions and their striking con- nection with grain boundary kinetics in alumina [10], a widely used ceramic material. Recently, Dillon-Harmer complexions were discovered in metallic systems Bi-Ni [11] and Bi-Cu [12], which could possibly explain the long standing puzzle of LME. In these experiments, Bi formed bilayer films ubiquitously in Ni at general orientation GBs around the penetration tip (as far as 983 µm). In contrast, a clean low energy Ni grain boundary was found 239 µm away from the tip. Bi also formed bilayer films at Cu GBs around the penetra- tion tip. However, the bilayer films were only observed close to the tip (as far as 257 µm) indicating bilayer films were stable over a much narrower Bi chemical potential window. Similarly to Ni, Bi did not segregate at low energy Cu GBs. A study of Fe revealed no Bi films [13]. A recent theoretical study [14] of Bi at Ni and Cu(111) twist and Σ5(310) GBs found the Bi bilayer enthalpy of formation on Σ5(310) is negative, which indicates ther- modynamic stablity, while on (111) twist GBs it is posi- tive. The authors proposed that bilayers are more stable than monolayers based on interaction strength between Bi and Ni layers and an electric dipole generated in the Bi bilayer on (111) twist GB. However, neither the origin of different segregation behavior of Bi on Ni compared with Cu, nor the relative stability of bilayer and trilayer films, was discussed. Moreover, a detailed study of the film structure, registry and bonding character is needed. In this letter, we present a first-principles study of Bi films on low energy Σ3(111) and high energy Σ5(210) transition metal GBs. Our results agree with the exper- imental result that bilayer films form on Ni and Cu high energy GBs but do not form on Fe GBs (see Supplemen- tal Material [15] for Fe). To compare the trends within TMs of similar structure, we analyzed Bi on FCC Co GBs. By exploiting the weak Bi interlayer interaction, we proposed a model that can be used to predict Bi bi- layer stability on various Ni GBs with relatively simple surface calculations. Moreover, we analyze the difference between transition metals based on localization of or- bitals and magnetization, and confirm this analysis with crystal orbital Hamilton populations (COHP) [16, 17] calculations. We also confirm the antibonding between Cu atoms induced by Bi segregation which was inferred in Reference [18]. Finally, we discuss the effects of Bi bilayers on grain growth and embrittlement. Our calculation methods are similar to our study of Bi on Ni(111) [19], namely PAW potentials [20, 21] in the PBE [22] generalized gradient approximation with de- fault energy cutoffs using VASP [23, 24]. To find stable structure at GBs, we first study Bi structures on free sur- faces. For Bi on TM(111) and (120), we construct models based on four and six metal layers normal to the surface respectively with Bi films on one side. We choose the Σ3(111) twist and the Σ5(012) tilt GBs as representative low energy and high energy GBs respectively. Σ3(111) is formed by cleaving the bulk along the (111) plane, ro- tating one grain around [111] by 600 and rejoining the two parts [25]. It is the lowest energy GB among all co- incidence site lattice (CSL) types that differ from bulk by just a stacking fault. Σ5(012) is formed by cleaving the bulk along the (012) plane, rotating one grain around [100] by 53.10 and rejoining the two parts after removing overlapping atoms (see Fig. 1). For Bi on Σ3(111) GBs we stack six layers of metal with periodic boundary conditions and rotate three layers rel- ative to the other three, thus creating the GBs. Then we insert our Bi film at one GB, leaving the other bare. To reduce computational complexity, the segregated struc- tures at Σ5(012) GBs are calculated with six layers TM at each side of the Bi films and terminated by bare TM surfaces with vacuum at both sides. The Σ5(120) GB plane is shown in Fig. 1. Later on, we refer the blue solid cell as (1 × 1), green dashed cell as (3 × 1) and cyan dash- dotted cell as (1 × 4). To analyze interaction strength and bonding character, we perform COHP calculations 10 5 0 -5 10 5 0 -5 (1x1) (3x1) (1x4) -10 -5 0 -10 -10 5 -5 0 5 10 FIG. 1: (Color online) Left: Side view of our Σ5(012) GB. The black cell is our unit cell. The dashed green lines are GB planes. Right: Top view of Σ5(012) GB plane. Three layers of atoms are shown. The black solid cell is the orthorhombic unit cell we use to calculate Σ5(012) GB energies. Atom size indicates depth (large below small). Units are A. which evaluate matrix elements of the total energy be- tween pairs of atomic orbitals on neighboring atoms. The differential (dCOHP) reveals the bonding and antibond- ing orbitals while the integral up to the Fermi energy (iCOHP) measures the bond strength. Our calculated GB energies EGB are shown in Supple- mental Material [15], Table S1, and agree well with prior literature. For Bi on the Ni(111) surface, we found a 4- atom Bi monolayer on a (3×3) surface cell is stable over a wide Bi chemical potential [19], and the same holds true for Co. For Bi on the Cu(111) surface, 2-atom Bi mono- layer on a [2012] cell is stable, which agrees with exper- mental observation [26]. On TM(120) surfaces, Bi sitting on the valley sites of (1 × 1) cells are stable over a wide range of chemical potential. This stable Bi on Cu(120) surface structure was observed in experiment [27]. We then study various Bi films at GBs. To compare the stability of these films, we calculate the enthalpy of formation, which is defined as, ∆H/A = [Etot − ETM slab − EBi bulkNBi]/A, (1) where Etot is the energy of a TM slab containing GB segregated by Bi, ETM slab is the energy of a TM slab con- taining a bare GB, EBi bulk is the Bi bulk energy, A is the GB area. Fig. 2 shows our enthalpies of formation. On the low energy Σ3(111) GBs, the enthalpies of Bi film for- mation are all large and positive, which suggests that Bi does not form stable films at these GBs. This is expected since the Σ3(111) GB differ from bulk only by a low en- ergy stacking fault. It is energetically unfavorable to cut the strong bulk-like metal bonds and replace them by bonds with Bi. These results agree with the experimen- tal observation [11, 12] of bare Ni and Cu low energy GBs near the Bi penetration tip. At the high energy Σ5(120) GB, ∆H is reduced for all TM. At Co Σ5(120) GB, ∆H 2 remains positive suggesting all films are unstable. For Ni, all Bi films have negative ∆H which means Bi pene- tration is favorable for all these films. Moreover, bilayer Bi is most favorable, with lower enthalpy of formation than monolayer and trilayer. For Cu, Bi monolayer and bilayer film have negative enthalpies of formation. The bilayer preference is less pronounced than on Ni. Overall, the enthalpy of formation is less negative on Cu than on Ni, which indicates interfacial films are less favorable in the case of Cu. 100 50 0 -50 ] 2 Å V e m [ / A H ∆ -100 0 1 CoΣ3(111) NiΣ3(111) CuΣ3(111) CoΣ5(012) NiΣ5(012) CuΣ5(012) 0.1 0.05 3 3′ 4′ 4 2 0.2 0.15 Coverage [Bi/Å2] 0.25 0.3 0.35 FIG. 2: (Color online) Enthalpies of Bi films at Σ3(111) and Σ5(012) GBs. Solid lines connect Bi films in (3 × 3) cells of Σ3(111) GBs. Films of 1-3 layer thickness (labeled 1-3) have 4 Bi atoms per layer while the 4-layer films contain 4 Bi atoms per layer in layers adjacent to Ni but 3 Bi per layer in the middle two layers [19] for Ni and Co. Films for Cu have 2 Bi per layer. Dashed lines connect Bi films in (1 × 1) cells of Σ5(120) GBs. Red, green and blue colors indicate Co, Ni and Cu respectively. Square points (labeled as 3′) stand for trilayer films in (3 × 1) cells of Σ5(012) GBs with denser middle layer (4 Bi in (3 × 1) cell). Diamond points (labeled as 4′) stand for four layer films in (1 × 4) cells of Σ5(012) GBs with the in-plane density of the middle bilayer similar to bulk Bi (3 Bi in a (1 × 4) cell). To further illustrate the stability of Bi films at Σ5(120) GBs, we calculate the GB free energy. This is the Legen- dre transformation of the enthalpy of formation (Eq. 1), replacing the Bi coverage with relative chemical poten- tial. From equilibrium thermodynamics, the most stable structure at a certain Bi chemical potential minimizes the GB free energy γ [28], γ = [∆H − ∆µBiNBi]/A, (2) where ∆µBi ≡ µBi − EBi bulk is the Bi relative chemical po- tential. Note that ∆µBi = 0 corresponds to the chemical potential of bulk Bi. As shown in Fig. 3, the stable sequence at Co Σ5(120) GB goes from a bare GB plane directly to an infinite height bulk-like film at ∆µBi = 0 eV. In contrast, the stable sequence at Ni Σ5(120) GB goes from bare GB to bilayer Bi film with (1 × 1) cell at ∆µBi = −0.37 eV, and then to the infinite height bulk like films at 50 0 -50 50 0 -50 50 0 -50 ] 2 / Å V e m [ γ Co -0.4 Bare GB Ni -0.4 Cu -0.4 Bare GB Infinte layer -0.2 Bilayer Infinte layer -0.2 Bare GB Bilayer Infinte layer -0.2 ∆µ Bi [eV] 0 0 0 0.2 0.2 0.2 FIG. 3: (Color online) GB free energy of Bi films at Co, Ni, Cu Σ5(120) GBs, respectively, top to bottom. The black solid lines stand for bare GBs while the black dashed lines stand for infinite bulk-like Bi films. Other lines are for different Bi films, with stable bilayer labeled. ∆µBi = 0 eV. The stable sequence at Cu Σ5(120) GB goes from bare GB to bilayer Bi film with (1 × 1) cell at ∆µBi = −0.067 eV, and then to the infinite height bulk-like films at ∆µBi = 0 eV. Bi films are thus not stable at Co Σ5(120) GB, and the Bi bilayer film at Ni Σ5(120) GB is stable over a much wider chemical poten- tial window than at Cu Σ5(120) GB. These results are consistent with the experimental observations [11, 12] at high energy GBs of Ni and Cu. Ni Bi 5 0 -5 -10 -5 0 5 10 FIG. 4: (Color online) Relaxed monolayer and bilayer Bi films at (1×1) cell of Ni Σ5(120) GB. Only Ni atoms close to Bi are shown. Atom size indicates depth (large below small). Units are A. Studying other bilayer films with different registry and coverage, it turns out the valley site of a missing Ni atom is a strong Bi adsorption site. For structures with Bi density smaller than the (1 × 1) film, all Bi atoms relax into valley sites. The (1 × 1) film is more stable than these films due to the energy gain by putting more Bi at the remaining empty valley sites. With Bi density larger than the (1 × 1) film, Bi atoms in each layer bond with each other in the unfavorable metallic form and also leave some empty valley sites which weakens the bonds with Ni. Both these effects destabilize such films. This result 3 is also consistent with our surface study that the (1 × 1) monolayer structure is most stable. Trilayer films are unfavorable at all GBs, again be- cause of the bonding character of Bi. Bulk Bi has the common α-As group-V semimetal (strukturbericht A7, Pearson hR2) with rhombohedral space group R¯3m form- ing a bilayer structure. Each Bi atom has strong covalent bonds with three intrabilayer neighbors at the distance of 3.1 A and bonds weakly with three interbilayer neigh- bors at 3.5 A. The trilayer films contain a chemically adsorbed monolayer on each side of the GB plus a mono- layer of atoms in between that forms metallic bonds. The four layer structure has a bilayer film similar to the bulk structure between the strong adsorbed monolayer films. Thus, bilayer films and four layer films are more favor- able than trilayer films. The observed trilayer Bi film at Ni GB near the penetration tip is thus indeed predicted to be a metastable structure as inferred in Reference [11]. The Bi interlayer interaction is weak in the bilayer films at both Ni Σ3(111) and Σ5(120) GBs, with bond lengths around 3.9 A and 4.2 A respectively, which are larger than the weak Bi-Bi metallic bond. In experiment, the observed Bi layer spacing is 3.9±0.6 A. Based on these observations, we propose a model to calculate the en- thalpy of formation of Bi bilayer at Ni GBs with bare GB energies and surface adsorptions, by neglecting the interlayer interaction, ∆H/A ≈ E a surf +∆H a ML/A+E b surf +∆H b ML/A−EGB, (3) surf and E b where E a surf are the surface energies of Ni sur- faces a and b adjacent to the GB plane, ∆H a ML and ∆H b ML are the enthalpies of formation of Bi monolayers on Ni surfaces a and b. The first four terms represent the excess energy per area with bilayer intercalation. EGB is the excess energy per area without intercalation. We define Emin GB as the minimum energy of GB consisting of surfaces a and b such that formation of Bi bilayer is ener- getically favorable, i.e. for which ∆H/A ≤ 0 (see values in Tab. S2 in Supplemental Material [15]). Hence Emin GB ≈ E a surf + ∆H a ML/A + E b surf + ∆H b ML/A. (4) Results of this model are given in Table S3 in Supplemen- tal Material [15]. The enthalpies of formation from model predictions (∆H model) and direct calculations (∆H calc) are within 0.01 eV/A2, and slightly exceed the direct cal- culations because we neglect the interaction between Bi bilayers which lower the total energy. This model thus ac- curately predicts Bi bilayer enthalpies of formation, while being easier to calculate than direct Bi at Ni GBs. Approximate energies for many bare GBs can be ob- tained from empirical potential calculations [29]. Based on those values, our model predicts that Bi bilayer en- thalpies of formation are positive on all Ni(111) CSL twist GBs, but are negative on (100) CSL twist GBs for rotation angles between 10 and 50 degrees. Moreover, for structures where the optimized Bi monolayer is not com- mensurate with the GB cell, and for GBs with different adjacent surfaces (i.e. a 6= b) that are not commensurate with each other, the model prediction might be more ac- curate than affordable direct calculations. An example is the (111)/(100) GB in Table S3 in Supplemental Mate- rial [15], where a general GB is constructed with a =(111) and b =(100) planes with (3 × 3) surface cell at the a side and [2 -2 1 3] surface cell (following the notation of [30]) at the b side. Unlike the CSL GBs, the ∆H calc is greater than ∆H model due to strain of the Ni cells (around 5%). This artificial strain introduced by forcing the two weakly interacting grains to share a common small cell makes the direct calculation inaccurate. This model could easily be generalized to other polycrystal materials providing the interlayer interaction of segregated films is small. Ni GBs are severely embrittled by Bi bilayer segrega- tion. In Table S4 in Supplemental Material [15], we show the work of separation (defined as Wsep = 2Esurf − EGB, the work needed to separate the GB [31]) at several bare and Bi segregated GBs. In all these GBs, Wsep is reduced by more than 95% due to the weak interaction between Bi layers [14]. The differences of Bi interactions among these three transition metal GBs can be understood with a combi- nation of localization of TM electron orbitals and mag- netism. With increasing of atomic number from Co to Ni and to Cu, the 3d orbital becomes more localized, and thus the interactions between TMs and with Bi decreases. For example, shown in Tab. S4 in Supplemental Mate- rial [15] the (012) surface energies Esurf and Bi monolayer ∆H/A on (012) surface diminish for Co, Ni and Cu re- spectively in nonmagnetic calculation. With magnetism, Co and Ni(012) Esurf decrease further due to increas- ing surface magnetic moment. The remaining greater Co surface energy due to stronger interaction between less localized orbitals makes the Co GBs harder to separate. Values of iCOHP measure bond strength. The Bi-Co bonds at surface is weaker than Bi-Ni when magnetism is included (see Tab. S4 in Supplemental Material [15]) due to the fact that Bi is nonmagnetic and quenches the TM surface magnetic moments. All of these effects are greater at the Co surface than Ni due to Co's larger surface magnetic moments, 1.93µB/atom compared with 0.78µB/atom for Ni. This leads to a greater increase in Bi ∆H/A on Co(012) than Ni(012) surface, compared with the nonmagnetic case. This effect is also manifested from our COHP calculation that the Bi-Co bond is weakened by 0.12 eV while Bi-Ni bond is weakened only by 0.02 eV due to magnetism. Thus Bi monolayers on Co(012) surface and Bi bilayers on Co Σ5(012) GB are less favor- able to form than on Ni due to the stronger interaction between Co atoms than between Ni atoms resulting from greater localization of 3d electrons on Ni, and weaker in- teraction between Bi and Co than between Bi and Ni, due to magnetism. 4 Apart from weaker interaction of Bi with Cu than with Ni, the Bi bilayer is less favorable on Cu than on Ni, since Bi gives electrons to Cu, half filling the Cu s states. This leads to stronger antibonding among Cu atoms close to Bi as is inferred in [18] and confirmed by our COHP calcula- tion shown in Fig. S2 in Supplemental Material [15]. The Ni-Ni and Co-Co bonds close to Bi however are stronger than in the bulk, where no antibonds appear. In conclusion, we have studied Bi segregation at Co, Ni and Cu low energy Σ3(111) and high energy Σ5(120) GBs using density functional theory. Our results reproduce the experimental result that Bi does not form film at all Fe GBs but forms a bilayer film ubiquitously at Ni high energy GBs, and in a much narrower chemical potential window at Cu high energy GB. The difference between these metals can be explained by the localization of 3d orbitals and also the loss of magnetism near the GB of Co (and presumably Fe). Moreover, Bi on Cu GB also increases the strength of antibonding, as confirmed by COHP calculation. We propose a model to predict the stability of Bi bilayer at various Ni GBs. Combining with the empirical GB energies from Reference [29], the model suggests Bi bilayer is not thermodynamically stable on (111) twist CSL GBs but should be stable in most (100) twist CSL GBs. LME is a long standing puzzle that has not been well understood at an atomic level. Several studies based on monolayer adsorption propose explanations of atomic size effects [31] or electronic effects [18]. Our theoretical study confirms the formation of Bi bilayers as the origin of the LME of Ni. The Bi bilayer (instead of monolayer) film at Ni and Cu GB thus serves as a starting point for future study of LME. Segregation also affects grain growth by altering the GB energy and mobility. When GB energy drops, grain growth is reduced since the driving force for grain growth is proportional to GB energy. Indeed, some metastable structures with low but positive GB energy halt grain growth. W dopants at Ni GB stops the grain growth when the averaged GB energy drops by ∼60% [32]. We find Bi bilayer segregation at Ni Σ5(120) GB reduces the GB energy from 77 meV/A2 to 28 meV/A2, a reduction of 64%. We thus anticipate a significant change in the grain growth behavior for Bi segregated Ni polycrystal since Bi bilayer films are ubiquitous at general GBs. The authors thank Jian Luo, Jeffrey Rickman, Zhiyang Yu, Anthony Rollett, Gregory Rohrer and Martin Harmer for helpful discussion. Financial support from the ONR-MURI under the grant NO. N00014-11-1-0678 is gratefully acknowledged. [1] C. Scott, M. Kaliszewski, C. Greskovich, and L. Levin- son, J. Am. Ceram. Soc. 85, 1275 (2002). 5 [2] J. Cho, M. Harmer, H. Chan, J. Rickman, and densed Matter 97, 35 (1995). A. Thompson, J. Am. Ceram. Soc. 80, 1013 (1997). [18] G. Duscher, M. F. Chisholm, U. Alber, and M. Ruhle, [3] B. Joseph, M. Picat, and F. Barbier, The Eur. Phys. Nat. Mater. 3, 621 (2004). J.-Appl. Phys. 5, 19 (1999). [4] K. Wolski and V. Laporte, Mater. Sci. Eng.: A 495, 138 (2008). [5] R. Viswanadham, T. Sun, and J. Green, Metall. Mater. [19] Q. Gao and M. Widom, Phys. Rev. B 88, 155416 (2013). [20] P. E. Blochl, Phys. Rev. B 50, 17953 (1994). [21] G. Kresse and D. Joubert, Phys. Rev. B 59, 1758 (1999). [22] J. P. Perdew, K. Burke, and M. Ernzerhof, Phys. Rev. Trans. A 11, 85 (1980). Lett. 77, 3865 (1996). [6] S. Knight, N. Birbilis, B. Muddle, A. Trueman, and S. Lynch, Corros. Sci. 52, 4073 (2010). [23] G. Kresse and J. Hafner, Phys. Rev. B 47, RC558 (1993). [24] G. Kresse and J. Furthmuller, Phys. Rev. B 54, 11169 [7] M. Tang, W. C. Carter, and R. M. Cannon, Phys. Rev. (1996). Lett. 97, 075502 (2006). [8] M. Harmer, Science 332, 182 (2011). [9] P. R. Cantwell, M. Tang, S. J. Dillon, J. Luo, G. S. Rohrer, and M. P. Harmer, Acta Materialia 62, 1 (2014). [10] S. J. Dillon, M. Tang, W. C. Carter, and M. P. Harmer, [25] M. D. Sangid, H. Sehitoglu, H. J. Maier, and T. Niendorf, Materials Science and Engineering: A 527, 7115 (2010). [26] D. Kaminski, P. Poodt, E. Aret, N. Radenovic, and E. Vlieg, Surface Science 575, 233 (2005). [27] B. Blum and H. Ascolani, Surface Science 482-485, Part Acta Mater. 55, 6208 (2007). 2, 946 (2001). [11] J. Luo, H. Cheng, K. M. Asl, C. Kiely, and M. Harmer, [28] J. R. Kitchin, K. Reuter, and M. Scheffler, Phys. Rev. B Science 333, 1730 (2011). 77, 075437 (2007). [12] A. Kundu, K. M. Asl, J. Luo, and M. P. Harmer, Scripta [29] D. L. Olmsted, S. M. Foiles, and E. A. Holm, Acta Ma- Mater. 68, 146 (2013). terialia 57, 3694 (2009). [13] J. Luo, Private communication (2013). [14] J. Kang, G. C. Glatzmaier, and S.-H. Wei, Phys. Rev. Lett. 111, 055502 (2013). [30] T. R. J. Bollmann, R. van Gastel, H. J. W. Zandvliet, and B. Poelsema, Phys. Rev. Lett. 107, 176102 (2011). [31] R. Schweinfest, A. Paxton, and M. W. Finnis, Nature [15] See supplementary material for supplemental tables and 432, 1008 (2004). figures. (link TBD). [32] A. J. Detor and C. A. Schuh, Acta Mater. 55, 4221 [16] R. Dronskowski and P. E. Bloechl, The Journal of Phys- (2007). ical Chemistry 97, 8617 (1993). [17] O. Jepsen and O. Andersen, Zeitschrift fr Physik B Con- 

1109.6923

1

1109

Correlated magnetic states in domain and grain boundaries in graphene

Ab initio calculations indicate that while the electronic states introduced by grain boundaries in graphene are only partially confined to the defect core, a domain boundary introduces states near the Fermi level that are very strongly confined to the core of the defect, and that display a ferromagnetic ground state. The domain boundary is fully immersed within the graphene matrix, hence this magnetic state is protected from reconstruction effects that have hampered experimental detection in the case of ribbon edge states. Furthermore, our calculations suggest that charge transfer between one-dimensional extended defects and the bulk in graphene is short ranged for both grain and domain boundaries.

cond-mat.mes-hall

cond-mat

Correlated magnetic states in domain and grain boundaries in graphene Simone S. Alexandre1, A. D. L´ucio2, A. H. Castro Neto3,∗, and R. W. Nunes1 1Departamento de F´ısica, ICEx, Universidade Federal de Minas Gerais, 31270-901, Belo Horizonte, MG, Brazil 2Departamento de Ciencias Exatas, Universidade Federal de Lavras, 37200-000, Lavras, MG, Brazil 3Graphene Research Centre and Department of Physics National University of Singapore, 2 Science Drive 3, Singapore, 117542 (Dated: December 26, 2018) Ab initio calculations indicate that while the electronic states introduced by grain boundaries in graphene are only partially confined to the defect core, a domain boundary introduces states near the Fermi level that are very strongly confined to the core of the defect, and that display a ferromagnetic ground state. The domain boundary is fully immersed within the graphene matrix, hence this magnetic state is protected from reconstruction effects that have hampered experimental detection in the case of ribbon edge states. Furthermore, our calculations suggest that charge transfer between one-dimensional extended defects and the bulk in graphene is short ranged for both grain and domain boundaries. PACS numbers: 73.22.-f, 73.20.Hb, 71.55.-i Controlling electronic transport and tailoring magnetic states at the nanoscale in graphene rank among the main issues related to the prospective application of this ma- terial in nanoelectronics and spintronics. In particular, electronic and magnetic states of extended line defects in graphene have been considered as possible conducting one-dimensional (1D) electronic channels and platforms for tailored spin states for spintronic applications. [1, 2] Two recent developments highlight the focus on extended line defects: (i) the recognition that mass-scale produc- tion of graphene should inevitably lead to a polycrys- talline material, containing one-dimensional grain bound- aries (GB) [3]; and (ii) the recent theoretical predic- tion [4] and experimental realization of domain bound- aries (DB) in graphene by controlled deposition of the FIG. 1: Geometries of extended one-dimensional defects in graphene. Carbon (oxygen) atoms are shown as grey (black) circles. (a) DB558: a domain boundary. (b) DB558+O: oxi- dized form with one O atom per defect unit. (c) DB558+2O: oxidized form with two O atoms per defect unit. (d) GB: a large-angle tilt grain boundary. (e) GB+O: oxidized form of the GB. (f) Schematic view of the first quadrant of the Bril- louin zone associated to the defect supercells. High-symmetry points and Cartesian axes are indicated. material on metallic substrates [2]. In the case of GB's, a large volume of works has accumulated in the last few years [5 -- 8], while for the DB produced in Lahiri et al. ex- periment, valley-filter properties [9] and its effects on the magnetic edge states of a graphene ribbon [10] have been theoretically investigated. Further, when chemical reduc- tion of a readily available material, such as graphene ox- ide, is used as a viable route for mass-scale graphene syn- thesis [11, 12], the presence of residual functional groups should affect the material electronic and magnetic prop- erties. In this context, the nature of the electronic states in- troduced by such extended 1D defects in graphene is a topic that deserves close inspection. More specifically, whether GB's and DB's act as quasi-1D conducting chan- nels immersed in the bulk of graphene is the question we seek to address in this work. We also consider the is- sue of self-doping in graphene induced by the presence of such extended 1D defects [13], that occurs when the line defect attracts charge carriers, resulting in charged defective lines surrounded by a doped graphene matrix. Our calculations indicate that while the electronic states introduced by a model GB structure in graphene hybridize with the bulk states and are only partially con- fined to the defect core, the DB defect introduces a sharp resonance in the density of states (DOS) of graphene, that lies just above the Fermi level in the neutral sys- tem, and is associated to electronic states that are very strongly confined to the core of the defect. Our re- sults suggest that, when a graphene sample containing a DB is doped and these quasi-1D states are populated, a ferromagnetic state is realized, which is confined to zigzag chains along the defect core that are fully im- mersed within a bulk graphene matrix. The quantum confinement induced by the presence of a domain bound- ary leads to an enhancement of the Coulomb interactions and stronger electron-electron correlations. Because of their 1D nature, these correlated states do not show long range order. Instead, they present power law or algebraic correlation functions. Furthermore, given that these 1D states are immersed in a metallic environment, they are unique examples of open Luttinger liquids [14]. Moreover, we find that charge transfer between bulk graphene and the line defects is essentially local, with charge redistribution taking place within a region of ∼3-5 A from the geometric center of the 1D defects. Both this charge-transfer region and the surrounding bulk regions remain essentially neutral. The absence of a charge monopole moment means that these 1D defects should act as weak charge carrier scatterers, in agreement with recent transport experiments in graphene grown by chemical vapor deposition (CVD) [15]. In this study, we examine the nature of the electronic states introduced by the two aforementioned types of extended 1D defects in graphene: (i) large-angle tilt grain boundaries [6, 7], expected to occur commonly in mass-scale production of graphene; (ii) domain bound- aries produced by deposition of a graphene layer on a Nickel substrate [2, 4]. For both 1D defects, we also con- sider oxidized forms, with oxygen atoms bound to the atoms at the core of the defect. Figure 1(a) shows the atomistic model for the domain boundary defect, which consists of periodic units composed of one octagon and two side-sharing pentagons along the defect line, which we label DB558 in the following discussion. The oxi- dized forms with one (DB558+O) and two oxygen atoms (DB558+2O) per defect unit are shown respectively in Figs. 1(b) and (c). Figure 1(d) shows the GB model pro- posed in Ref. 6, while Fig. 1(e) shows the fully unzipped oxidized form of this defect (GB+O). Our calculations are performed in the framework of Kohn-Sham density functional theory (DFT), within the generalized-gradient approximation (GGA) [16] and norm-conserving pseudopotentials in the Kleinman- Bylander factorized form [17]. We use the LCAO method implemented in the SIESTA code [18], with a double-zeta basis set plus polarization orbitals. A Mulliken charge partition is employed for the analysis of the charge- transfer between bulk and defects. Supercells of 66 (84) atoms are employed for the DB558 (GB) calculations. We performed convergence tests to ensure that, at these cell sizes, our results are not affected by interaction be- tween the defects and their periodic images. In the case of GB's in graphene, the electronic struc- ture has been amply discussed in Ref. 5, where the oc- currence of an anisotropic Dirac cone in a Brillouin-zone (BZ) line along the GB direction was analyzed. In Ref. 5, the GB electronic states near the Fermi level (FL) were found to disperse in all directions, indicating hybridiza- tion with bulk states. Further indication of such hy- bridized nature of GB states is provided by the analysis of the full band structure and density of states (DOS) of the GB supercell. Figure 2(a) shows the band struc- ture along the BZ lines shown in Fig. 1(f). Note that 2 FIG. 2: (Color online) (a) Band-structure for the GB along the Brillouin-zone lines in Fig. 1(f). (b) DOS for the GB. (c) Band-structure for the DB558. (d) DOS for the DB558. the states near the Dirac point show sizeable dispersions in directions perpendicular to the defect (Y-M and Γ-X lines). The contributions of the orbitals centered on the atoms at the GB core to the resonances introduced in the DOS of graphene are shown in Figure 2(b). For the purpose of our analysis, we define the core of both the GB and DB558 defects as composed by the ten atoms that form the topological defects along the defect line, as indicated in Fig. 1. In the DOS plots, we refer to atoms 1 and 2, that are placed along the line at the geometric centers of both defects, as core center, while the remain- ing eight core atoms are referred to as core perimeter. In the case of the GB, the contribution of the ten core atoms to the four DOS peaks near the FL ranges from 41% to 57%. These results will be contrasted with the case of the DB558 in the following. The band structure and the DOS for the DB558 are shown in Fig. 2(c) and (d), respectively. In Fig. 2(c) we also include the band structure of a 64-atom bulk supercell, obtained by removing core atoms 1 and 2 from the geometry in Fig. 1(a). Given the periodicity of both defect and bulk cells along the DB558 direction (a zigzag direction of the bulk matrix), the Dirac point of the bulk cell folds onto the point at 2π/3ℓ, where ℓ is the period of the DB558 [19]. Note that, to a large degree, the changes in the DB558 band structure with respect to the bulk one are concentrated in a region of ±0.5eV from the FL, and that within ∼0.2 eV above (below) the FL we find bands with flat sections along the Γ−Y (L-X) direction (both parallel to the defect line). These states show little or no dispersion along the Γ-X and Y-M lines that are perpendicular to the defect line. The flat-band character of the states above the FL, near the zone center, leads to a ferromagnetic state for a bulk 1D defect in graphene consisting entirely of carbon atoms, as discussed below. Considering now the DOS for the DB558 in Fig. 2(d), 3 FIG. 3: (Color online) Isosurface of spin polarization density for domain-boundary magnetic state. we find two sharp resonances within ∼ ±0.1 eV from the FL, associated to the 1D defect. These resonances reflect the presence of extended van Hove regions in the in the bandstructure, as seen in Fig. 2(c). We observe that in one feature the DOS of the DB558 differs markedly from the GB one: the peak just above the Fermi level shows a very strong concentration on the core atoms, with ∼80% of the total DOS concentrated on the orbitals centered on the perimeter core atoms numbered 3, 4, 5, and 6, along the zigzag chains that bond to the dimers (atoms 1 and 2) at the core center in Fig. 1(b). Adding the contributions of the atoms from the same sublattice as atoms 3-6, in the zigzag chains nearest to the core on both sides of the defect line, we already account for ∼96% of the total DOS for this peak. A strongly one- dimensionally confined empty state along the defect core is thus a characteristic of the DB558. In our calculations, however, these states are empty in the neutral system, hence no spin polarization is induced. In order to investigate the formation of magnetic states in gate-doped versions of this system, we added one elec- tron per supercell, a doping concentration that raises the FL by about ∼0.07 eV. A spin-polarized calculation sta- bilizes a ferromagnetic state, with a magnetic moment of 0.52 µB per defect unit, and a formation energy that is ∼40 meV per defect unit lower in energy than the unpo- larized state. Figure 3 shows a representative isosurface of the difference between majority and minority spin den- sities. The spin density is concentrated on the sublattice of the atoms 3-6, with a negligible contribution from the dimer atoms themselves and also from the atoms near the core that belong to the other sublattice. We see here a manifestation of a magnetic state along a line defect that is fully immersed within the bulk of graphene. The fact that these states have a negligible contribution along the dimer atoms is perhaps the origin of the topological disruption in the electronic states that leads to the local- ization of the related π-orbital bands and hence to the magnetic state. We speculate that such DB magnetic states should be more easily detectable experimentally FIG. 4: (Color online) (a) Band-structure for the GB+O. (b) Band-structure for the DB558+2O. (c) Band-structure for the DB558+O. (d) Band-structure for the DB558+O, including spin polarization. Black (red) symbols show majority (minor- ity) spin bands. than those predicted to occur along the edges in graphene ribbons [19 -- 22], since the zigzag chains along the DB are protected from reconstruction. To address the effect of chemical doping on the elec- tronic states of the 1D defects, we considered the oxi- dized forms shown in Fig. 1. The corresponding band structures are shown in Fig. 4. For both the GB+O and DB55+2O, where oxygen-induced unzipping of the C-C bonds takes place, we observe the opening of small gaps of 55 meV and 190 meV, respectively. In neither of these two cases we find magnetic ground states in the neutral system. An interesting case is the DB558+O system, with one oxygen atom per defect cell, in which the oxy- gen atom bonds with the two carbon atoms at the defect core in the bridge position. In this case, as shown in Fig. 4(c), the occupied band along the Γ−Y line that is ∼0.1 eV below the FL in the pure DB558 system is now deeper in energy. More importantly, a significant change is observed in the flat band of confined 1D states near Γ that now crosses the FL. As a result, a magnetic ground state is stabilized, with a net magnetic moment of 1.1µB per unit cell, as shown in Fig. 4(d), where major- ity and minority spin bands are plotted. Note that spin- polarization is strongly localized on the 1D defect, with significant exchange splitting restricted to the quasi-1D bands crossing the FL. This ferromagnetic ground-state is 43 meV (per defect unit) lower in energy than the non-polarized state. This value is very close to that for the pure DB558 system with an extra electron per cell. Hence, chemical doping offers another way of stabilizing this ferromagnetic state. The ferromagnetic instability of these flat-band states is further confirmed by a calcu- lation we performed for the DB558+2O system with an extra charge added to the defect supercell. Again, the FL shifts into the band of quasi-1D states, and the system 4 confined to the defect core, while a domain boundary in- troduces unoccupied electronic states near the Fermi level that are very strongly confined to the core of the defect, and that, when populated by doping, display a ferromag- netic ground state in a 1D defect that is fully contained within the bulk matrix and that consists entirely of car- bon atoms. Being fully bulk-immersed, this ferromag- netic state is protected from reconstruction and should be more easily detectable experimentally than those pre- dicted to exist along the edges of graphene ribbons. Fur- thermore, our calculations indicate that charge transfer between bulk graphene and both types of extended 1D defects is strongly localized, with charge redistribution confined to ∼3-5 A from the geometric center of the 1D defect, implying that these 1D defects should act as weak charge scatterers in graphene. ∗ On leave from Boston University, USA. SSA, RWN, and ALD acknowledge support from Brazilian agencies CNPq, FAPEMIG, and Instituto do Milenio em Nanociencias-MCT. AHCN thanks the finan- cial support of the NRF-CRP award Novel 2D Materi- als with Tailored Properties: Beyond Graphene (R-144- 000-295-282), US/DOE grant DE-FG02-08ER46512, and US/ONR grant MURI N00014-09-1-1063. [1] A. H. Castro Neto et al., Rev. Mod. Phys. 81, 109 (2009), and references therein. [2] J. Lahiri et al., Nature Nanotech. 5, 326 (2010). [3] Q. Yu et al., Nature Mat. 10, 443 (2011). [4] M. U. Kahaly, S. P. Singh, and U. V. Waghmare, Small 4, 2209 (2008). [5] J. da S. Ara´ujo and R. W. Nunes, Phys. Rev. B 81, 073408 (2010). [6] P. Simonis et al., Surf. Sci. 511, 319 (2002). [7] J. Cervenka and C. F. J. Flipse, Phys. Rev. B 79, 195429 (2009). [8] O. V. Yazyev and S. G. Louie, Phys. Rev. B 81, 195420 (2010). [9] D. Gunlycke and C. T. White, Phys. Rev. Lett 106, 136806 (2011). [10] X. Lin and J. Ni, Phys. Rev. B 84, 075461 (2011). [11] C. G´omez-Navarro et al., Nano Lett. 10, 1144 (2010). [12] A. Bagriet al., Nature Chem. 2, 581 (2010). [13] N. M. R. Peres, F. Guinea, and A. H. Castro Neto, Phys. Rev. B 73, 125411 (2006) [14] A. H. Castro Neto, C. de C. Chamon and C. Nayak, Phys. Rev. Lett. 79, 4629 (1997). [15] A. Ferreira et al., EPL 94, 28003 (2011). [16] W. Kohn and L. J. Sham, Phys. Rev. 140, A1133 (1965); J. P. Perdew, K. Burke, and M. Ernzerhof, Phys. Rev. Lett. 77 3865 (1996). [17] N. Troullier and J .L. Martins, Phys. Rev. B 43, 1993 (1991); L. Kleinman and D. M. Bylander, Phys. Rev. Lett. 48, 1425 (1982). [18] J. M. Soler et al., J. Phys. Cond. Matt. 14 2745 (2002). [19] K. Nakada, M. Fujita, G .Dresselhaus, and M .S. Dres- selhaus, Phys. Rev. B 54, 17954 (1996); FIG. 5: (Color online) (a) Charge density distribution for the DB558 (blue circles) and the GB (red squares). (b) Integrated charge I(r) (see text) for the DB558 and the GB. stabilizes a ferromagnetic state. Just as in the case with- out O doping, the presence of extended van Hove regions is also clearly seen in the band structures in Fig. 4. We turn our attention now to the charge transfer be- tween the 1D defects and the surrounding graphene bulk, since this may give important qualitative information about the nature of the scattering of charge carriers by such 1D defects. In Fig. 5, we show the profile of the charge distribution around the 1D defects, and the inte- grated charge per atom, both as a function of the distance to the geometric center of the defect, for the DB558 and the GB. The charge distribution is shown as linear charge densities representing the net charge summed over atoms at the same distance from the 1D defect, divided by the period of the defect unit. The integrated charge is the integral of this linear charge density profile, computed as total net charge per atom, and defined as follows: I (R) = 1 N (R) R X ri=0 λ (ri) × ℓd ; (1) where λ (ri) is the linear charge density for the line of atoms at a distance ri from the defect, ℓd is the period of the 1D defect, and N (R) is the number of atoms summed over all lines with ri ≤ R. Figure 5 shows that the charge redistribution around the 1D defects is non-monotonic, with the linear charge distributions alternating in sign on both sides of the de- fect, as seen in Fig. 5(a). Note also that charge trans- fer between the 1D defects and the graphene matrix oc- curs in a range of ∼3-5 A from the defect line, with the graphene matrix becoming neutral beyond this range, as shown in the plots for I(R) in Fig. 5(b). Since the charge distribution region recovers neutrality within ∼3- 5 A from the defect center, we expect no long range Coulomb scattering of charge carriers from both 1D de- fects. In our calculations, oxidation of the 1D defects is found not to affect this localized character of the charge balance between the defects and the graphene bulk. Such extended 1D defects should then play a minor role as a source of carrier scattering in graphene. In summary, we find that the electronic states intro- duced by grain boundaries in graphene are only partially [20] H. Lee, Y. W. Son, N. Park, S. Han, and J. Yu, [22] S. S. Alexandre, M. S. C. Mazzoni, and H. Chacham, Phys. Rev. B. 72, 174431 (2005). Phys. Rev. Lett 100, 146801 (2008). [21] Y. W. Soon, M. L. Cohen, and S. G. Louie, Nature 444, 347 (2006). 5 

1802.06352

1

1802

Dynamic spin injection into a quantum well coupled to a spin-split bound state

We present a theoretical analysis of dynamic spin injection due to spin-dependent tunneling between a quantum well (QW) and a bound state split in spin projection due to an exchange interaction or external magnetic field. We focus on the impact of Coulomb correlations at the bound state on spin polarization and sheet density kinetics of the charge carriers in the QW. The theoretical approach is based on kinetic equations for the electron occupation numbers taking into account high order correlation functions for the bound state electrons. It is shown that the on-site Coulomb repulsion leads to an enhanced dynamic spin polarization of the electrons in the QW and a delay in the carriers tunneling into the bound state. The interplay of these two effects leads to non-trivial dependence of the spin polarization degree, which can be probed experimentally using time-resolved photoluminescence experiments. It is demonstrated that the influence of the Coulomb interactions can be controlled by adjusting the relaxation rates. These findings open a new way of studying the Hubbard-like electron interactions experimentally.

cond-mat.mes-hall

cond-mat

Dynamic spin injection into a quantum well coupled to a spin-split bound state N. S. Maslova,1 I. V. Rozhansky,2, 3 V. N. Mantsevich,1 P. I. Arseyev,4, 5 N. S. Averkiev,2 and E. Lahderanta3 1Lomonosov Moscow State University, 119991 Moscow, Russia 2Ioffe Institute, St.Petersburg, 194021 Russia 3Lappeenranta University of Technology, FI-53851 Lappeenranta, Finland 4P.N. Lebedev Physical Institute RAS, 119991 Moscow, Russia 5Russia National Research University Higher School of Economics, 119991 Moscow,Russia (Dated: September 20, 2018) We present a theoretical analysis of dynamic spin injection due to spin-dependent tunneling between a quantum well (QW) and a bound state split in spin projection due to an exchange interaction or external magnetic field. We focus on the impact of Coulomb correlations at the bound state on spin polarization and sheet density kinetics of the charge carriers in the QW. The theoretical approach is based on kinetic equations for the electron occupation numbers taking into account high order correlation functions for the bound state electrons. It is shown that the on-site Coulomb repulsion leads to an enhanced dynamic spin polarization of the electrons in the QW and a delay in the carriers tunneling into the bound state. The interplay of these two effects leads to non-trivial dependence of the spin polarization degree, which can be probed experimentally using time-resolved photoluminescence experiments. It is demonstrated that the influence of the Coulomb interactions can be controlled by adjusting the relaxation rates. These findings open a new way of studying the Hubbard-like electron interactions experimentally. I. INTRODUCTION The research field of spintronics continues to grow cov- ering various spin phenomena in solid-state physics1. The first generation or "metallic" spintronics is associated with magnetism and exchange interaction. It has suc- ceeded in suggesting practical applications, the vivid ex- ample is the giant magnetoresistance effect widely used in hard drives2. Semiconductor spintronics of the sec- ond generation is focused on the effects based on spin- orbit interaction, which locks a particle motion with its spin. This locking is key to many attracting practical applications3. In the traditional semiconductor materi- als the effective spin-orbit interaction is relatively weak, so the latest research in the field of spintronics devel- ops in two overlapping directions: new semiconductors with stronger spin-orbit interaction, including topologi- cal insulators4,5, and new spin phenomena based on the exchange interaction6. Our work contributes to the sec- ond direction, in the present paper we focus on a dynam- ical spin injection into a quantum well (QW) through a tunnel barrier with account for the exchange interaction in the leads. The spin injection into a semiconductor remains the cornerstone of modern spintronics. The conductivity mismatch prevents an efficient spin injection from a fer- romagnetic metal into a semiconductor7. The widely dis- cussed solutions of this problem (apart from those based on spin-orbit interaction) include using dilute magnetic semiconductor as a spin injector8,9, spin injection from a ferromagnet through a tunnel barrier10, superdiffusive spin transport11. In our paper we consider a spin injec- tion into a semiconductor due to spin-dependent relax- ation of initially unpolarized ensemble of charge carriers. In this sense it is similar to spin-depdendent recombina- tion phenomena12,13. In our model a non-equilibrium distribution of the non- polarized carriers is assumed created instantaneously in the QW. We analyse theoretically the subsequent kinet- ics of the spin polarization. The study of this physical model is motivated by the experimental studied reported in Refs.14,15. In these experiments an InGaAs based het- erostrcutures with a QW and remote Mn doping layer was optically pumped with non-polarized carriers; the ex- perimentally measured time-resolved photoluminescence indicated a development of non-stationary spin polar- ization of the 2D carriers in the QW on the time scale smaller than the radiative recombination time. The ori- gin of this phenomena was explained theoretically as elec- tron tunneling from the QW into a Mn dopant layer16,17. The spin splitting of the impurity bound state in the dopant layer due to exchange interaction results in the spin-dependent tunneling rate and thus leads to the spin polarization of the carriers remaining in the QW16. In this paper we generalize the theory by developing non-stationary formalism to describe the charge and spin kinetics in the QW coupled to a bound state. An impor- tant feature of our study is that in addition to the con- ventional Heisenberg exchange interaction we account for the Coulomb correlations at the bound state which can contribute to the spin splitting as it is well known from Anderson and Stoner models18. We show how the kinet- ics of the spin polarization in the QW depends on the relaxation rates and the strength of the Coulomb on-site correlation at the bound state. II. THEORETICAL MODEL Although motivated by the experiments14,15, in this paper we do not restrict ourselves to a particular semi- conductor heterostructure design. Let us consider two 8 1 0 2 b e F 8 1 ] l l a h - s e m . t a m - d n o c [ 1 v 2 5 3 6 0 . 2 0 8 1 : v i X r a (cid:88) σ (cid:88) form: H = HQW + H1 + Hint, where HQW describes the QW: (cid:88) σ,k HQW = εkc+ kσckσ, 2 (1) (2) H1 describes the bound state with the Hubbard term for on-site Coulomb repulsion: H1 = ε1 nσ 1 + U nσ 1 n−σ 1 , (3) and HT is the tunneling part describing the QW and bound state coupling: HT = tk(c+ kσc1σ + c+ 1σckσ). kσ Here index k labels continuous spectrum states in the QW, tk is the tunneling transfer amplitude between QW states and the bound state. The bound state is char- acterized by the energy level ε1, which can be split due to an exchange interaction or an external magnetic field into two spin sublevels with the energies εσ = ε1 + σ∆, where σ = ±1/2 is the electron spin projection and ∆ is the energy gap. Operators c+ kσ, ckσ are the creation and annihilation operators for the electrons in the QW, n1σ = c+ 1σc1σ is the occupation number operator for the bound state, operator c1σ destroys electron in the bound state with the spin projection σ. U is the on-site Coulomb repulsion energy for the doubly occupied bound state. Further analysis deals with the low temperature regime when the Fermi level is well defined and the tem- perature is much lower than all other energy scales in the system. III. NON-STATIONARY ELECTRONIC TRANSPORT FORMALISM Let us further consider  = 1 and e = 1 elsewhere, so the equations of motion for the electron operators prod- 1 = c+ ucts nσ ckσ can be written as: 1σckσ and nσ (cid:48) k 1σc1σ, nσ 1k = c+ = c+ (cid:48) k σ k = −(cid:88) k,σ i ∂ nσ 1 ∂t i ∂ nσ 1k ∂t tk · (nσ k1 − nσ 1k), = − (εσ 1 − εk) · nσ 1 − nσ k ) − (cid:88) 1k − U · n−σ tk 1 nσ 1k (cid:48) · nσ k + tk · (nσ (cid:48) k, ∂ nσ (cid:48) k ∂t i k = − (εk (cid:48) − εk) · nσ (cid:48) k (cid:48) · nσ 1k + tk · nσ k(cid:48)1. k (cid:48)(cid:54)=k k − tk (4) (5) (6) FIG. 1. Two possible realizations of the considered sys- tem: (a) corresponds to the design of experimentally stud- ied (Ga,In,Mn)As heterostructures with bound states formed by paramagnetic impurities, (b) is an alternative design with bound states formed by a quantum dot (QD). model systems shown in Fig. 1. Essential for both con- figurations is a QW with one size quantization subband. At initial time t = 0 it is filled with unpolarized non- equilibrium 2D electrons (for example, created by an optical pumping) with the energies εk, where k is the in-plane wavevector. A spin-split bound state with the energy ε1 is separated from the QW by a tunnel bar- rier. The barrier is characterized by the tunneling rate Γ. The radiative electron-hole recombination in the QW is characterized by the rate γk, which is spin-indepedent. There is also a spin-independent relaxation channel emp- tying the bound state with a characteristic rate γ1. in Fig.1,a The model system shown corre- sponds to the experimentally studied (Ga,In,Mn)As heterostructures14, in which the bound state is formed by Mn ions in interstitial position providing donor-like states. The delta-doping Mn layer is located at a distance of 3-10 nm from the QW. The spin splitting is due to the effective exchange field in the doping Mn layer and the relaxation from the impurity donor levels is due to a very fast non-radiative recombination with the holes in the low-temperature grown (Ga,Mn)As layer16,17. A somewhat different model system is shown in Fig.1,b. Here the bound state is formed by a quantum dot (QD) located at a small distance from the QW. This state can accept one or two electrons. The spin splitting inside the QD occurs due to an exchange interaction if QD has paramagnetic impurities, an external magnetic field if present and the dynamic Coulomb correlations in the QD. The spin-independent relaxation channel is represented by a metallic lead, so that an electron can escape from QD with a characteristic rate γ1. The re- sults presented below are equally valid for both schemes, Fig. 1,a and Fig. 1,b, moreover, we believe that the pre- sented theory can be applied to other systems of a similar character. Below we make no difference between the two variants of the model system shown in Fig. 1. The Hamiltonian of the system can be written in the Following the logic of Ref.19 we substitute the solution of Eq. (6) into Eq. (5) to obtain: i ∂ nσ 1k ∂t = − (εσ 1 − εk + iΓk)nσ (cid:88) t(cid:90) tk(cid:48)tk + i k(cid:48)(cid:54)=k 1k − U n−σ 1 nσ 1k + tk(nσ 1 − nσ k ) ei(εk(cid:48)−εk)(t−t(cid:48)) nσ k(cid:48)1dt(cid:48), (7) where Γk = πν0 (εk) t2 k and ν0 (εk) is the unperturbed density of states in the QW. Further we assume that the tunneling parameter tk has a negligibly weak dependence on k, so for 2D density of states in the QW the tunneling relaxation rate is a constant Γk ≡ Γ, which we take as a parameter. If condition ε1−εF 1 is a slowly varying quantity in comparison with nσ 1k. Conse- quently, it is reasonable to consider that: Γ (cid:29) 1 is fulfilled, nσ n±σ 1 nσ 1k ∼ n±σ 1 ∂ ∂t (8) nσ 1k. ∂ ∂t 1 and (1− nσ 1k, n−σ 1 )nσ 1 )· nσ 1 nσ 1 , nσ 1 )2 = nσ Taking into account that (nσ 1 = 0 one can find expressions for (1 − n−σ 1k and obtain the equations for the time evolution of the particle number operators nσ k for the bound state and QW, respectively. The suggested theoretical approach allows one to analyze dynamic spin injection due to the spin- dependent tunneling with account for high order correla- tion functions for the bound state electrons. Therefore, it gives possibility to analyze the effects of the on-site Coulomb repulsion. We also add explicitly the spin-independent relaxation terms describing recombination in the QW with the rate γk and relaxation at the bound state due to non-radiative recombination (Fig. 1,a) or tunnel leakage into the lead (Fig. 1,b) with the rate γ1. Thus, we account for the ef- fect of Coulomb correlations on the tunneling between QW and the bound state and also for the additional bound state broadening due to the relaxation into a lead or reservoir. This approach neglects the influence of the QW and bound state relaxation channels on each other, but allows for decoupling of the QW and bound state equations of motion. Therefore, we obtain: ∂ nσ 1 ∂t ∂ nσ k ∂t 1 − (1 − n−σ 1 ) · Φ (εσ) = −2Γ · [nσ − n−σ 1 2Γ ν0π = · Φ (εσ + U )] − γ1 · nσ 1 , · [(1 − n−σ 1 − nσ k )(cid:101)Γ k ) (εσ + U − εk)2 +(cid:101)Γ2 1 − nσ n−σ 1 (nσ 1 )(nσ ] − γk · nσ k . + (cid:101)Γ (εσ − εk)2 +(cid:101)Γ2 (9) Here we introduced the QW occupation operators Φ (εσ) and Φ (εσ + U ) as: 3 (cid:90) (cid:90) Φ (εσ) = dεk · nσ k (εk) · 1 π (cid:101)Γ (εσ + U − εk)2 +(cid:101)Γ2 where (cid:101)Γ = Γ + γ1. Note, that we introduced (cid:101)Γ in order (εσ − εk)2 +(cid:101)Γ2 (cid:101)Γ k (εk + U ) · 1 π Φ (εσ + U ) = dεk · nσ (10) , to properly account for the structure of the bound state, which is affected both by the hybridization with the QW and the separate relaxation channel with the rate γ1. One can obtain equations for the bound state occupa- tion numbers nσ 1 by averaging equations for the operators and by decoupling electron occupation numbers for the QW states from the bound state occupation numbers. Such decoupling procedure is reasonable if one consid- ers that Kondo correlations can be neglected20,21. Af- ter decoupling the QW occupation numbers operators nσ (9)-(10) have to be replaced by the dis- k in Eqs. tribution functions f σ In order to take into account k . spin-independent relaxation processes from the QW and the bound state we add the corresponding rates γk and γ1 into kinetic equations for the bound state occupation numbers and the QW electron distribution function. As- suming that equilibrium state corresponds to the empty bound state and empty QW we obtain the following equa- tions: (cid:101)Γ · Φ (εσ + U ) (εσ − εk)2 +(cid:101)Γ2 (11) (12) ∂nσ 1 ∂t ∂f σ k ∂t = −2 · Γ · I σ = 2 · Γ · J σ k − γ1 · nσ 1 , k − γk · f σ k , where 1 + J σ k = 1 − (1 − n−σ 1 ) · Φ (εσ) − n−σ k = nσ I σ 1 − f σ [(1 − n−σ 1 1 )(nσ k )(cid:101)Γ k ) ν0π (εσ + U − εk)2 +(cid:101)Γ2 1 − f σ n−σ 1 (nσ (cid:90) (cid:90) k (εk) · 1 π dεk · f σ Φ (εσ) = ], dεk · f σ k (εk + U ) · 1 π Φ (εσ + U ) = (εσ − εk)2 +(cid:101)Γ2 (cid:101)Γ (cid:101)Γ (εσ + U − εk)2 +(cid:101)Γ2 , and QW occupation functions Φ (εσ) and Φ (εσ + U ) read (13) 1 = n−σ We further solve Eqs. (11)-(12) implying the following initial conditions at t = 0: the bound state is empty, therefore nσ 1 = 0; the QW is filled by the photo- excited carriers with a non-equilibrium energy distribu- tion function characterized by chemical potential µ∗ and electron temperature T : fk(0) = 1 . e(εk−µ∗ )/kT +1 , . 4 FIG. 2. (Color online) Time evolution of spin polarization and QW sheet density for different QW relaxation rates γk. For solid lines U/Γ = 35, for dashed lines U/Γ = 0. Parameters ε↑/Γ = 2, ε↓/Γ = −2, µ∗/Γ = 0, γ1/Γ = 0.15 and Γ = 1 are the same for all the figures. QW is given by N↑ − N↓, where Nσ = (cid:82) fσ (εk) dεk it The spin polarization of the electrons remaining in the manifests itself in the circular polarization of the photolu- minescence (PL) from the QW which can be measured16. The spin polarization degree which would be measured by optical means is defined as: P = N↑ − N↓ N↑ + N↓ . (14) The polarization degree P is negative when electrons with spin projection σ = − 1 2 prevail. The considered theoretical approach is more general than in Ref.16 as Eqs. (11) cover both cases of resonant and non-resonant tunneling between the QW and the bound state. IV. RESULTS AND DISCUSSION The spin kinetics calculated according to the theory de- scribed above is shown in Figs. 2-4. In all the calculations we assume the same transparency of the tunnel barrier FIG. 3. (Color online) Time evolution of spin polarization and QW sheet density for different bound state relaxation rates γ1. Insert in the panel (a) demonstrates that for large values of γ1/Γ spin polarization exhibits the same behavior with (black curve) and without (red dashed curve) Coulomb interaction. Parameters ε↑/Γ = 2, ε↓/Γ = −2, µ∗/Γ = 0, γk/Γ = 1.5 and Γ = 1 are the same for all the figures. FIG. 4. (Color online) Time evolution of spin polarization degree. Parameters ε↑/Γ = 2, ε↓/Γ = −2, µ∗/Γ = 0, γk/Γ = 1.5 and Γ = 1 are the same for all the figures. Colors of the curves and values of parameters U/Γ and γ1/Γ in the main panel correspond to the colors and values shown in Fig. 3. Insert shows the behavior of black and blue curves at the beginning of the time evolution. N+NNN-GN+NNN-GGNN-characterized by the tunneling rate, which is taken Γ = 1. Fig. 2 shows the time evolution of the spin polarization (a) and total number of electrons (b) in the QW with account for the on-site Coulomb repulsion U for the elec- trons in the bound state. The calculation results are pre- sented for various relaxation rates γk, which describe the radiative recombination in the QW. As shown in Fig. 2,b, the total number of electrons in the QW is decreased as the initial non-equilibrium concentration of electrons re- laxes due to the tunneling and radiative recombination. In the absence of Coulomb correlations (U = 0), the to- tal relaxation rate for the electrons in the QW is simply the sum of the two γQW = γk + Γ. This conclusion holds also for the case U (cid:54)= 0 if γk > Γ. The electrons mostly relax through the recombination channel in the QW and do not have enough time to be affected by the correla- tions at the bound state. This situation is illustrated by blue and red curves in Fig. 2. However, if Γ > γk and U (cid:54)= 0, the decrease of the carriers sheet density in the QW due to the tunneling is delayed as now the tunneling of an electron requires an additional energy cost if the final state is occupied. This case corresponds to the solid and dashed black lines in Fig. 2. Since the bound state is split in spin projection, the relaxation rate through the tunneling channel is spin- dependent. Therefore, the spin polarization of the elec- trons in the QW shown in Fig. 2,a increases with time. The increase is linear at t < (γk + Γ) in agreement with Ref.16,17, later on the spin polarization decays as QW becomes empty. As can be clearly seen in Fig. 2,a, when the Coulomb correlations at the bound state become im- portant, that is U (cid:54)= 0, Γ > γk, the maximum of the spin polarization is substantially increased. For the pa- rameters used for Fig. 2 the enhancement is more than two times. The position of the maximum on time scale is also substantially shifted to larger times. Thus, the strong Coulomb correlations lead to a stronger dynamic spin injection into the QW and the delayed kinetics, con- sequently, the spin polarization in the QW is preserved for a longer time. The effect of the Coulomb correlations on the spin polarization in the QW also depends on the spin- independent relaxation rate at the bound state γ1 (as- sumed to be the same for both spin sublevels). This in- fluence is shown in Fig. 3. Obviously, if the bound state sublevels are emptied faster than the rate of the incoming tunneling electrons from the QW, the Coulomb correla- tions shouldn't play a role as the bound state would never get doubly occupied. Indeed, for γ1/Γ > 1 the evolution of the total sheet density and the spin polarization in the QW is the same for U/Γ = 30 (red curve) and for U = 0. In the latter case the magnitude of γ1 does not matter as occupation of the bound state is not accompanied with an additional energy cost. The effect of the Coulomb cor- relations becomes important as γ1 is enhanced so that the electrons are less effectively removed from the bound state. The spin injection in this case in enhanced as can be clearly seen in Fig. 3,a, blue line. The QW to- 5 tal occupation dynamics is also affected by the Coulomb interactions, which lead to a decrease in the decay rate analogously to what was seen in Fig. 3,b. However, one can note that the discrepancy between different lines in Fig. 3,a develops at times t > Γ. That is, when the bound state becomes significantly populated with the tunneling electrons so that the correlations become important. Finally, Fig. 4 shows the spin polarization degree P introduced in Eq. (14). It is this quantity that can be measured experimentally as a degree of circular polar- ization of the photoluminescence from the QW. Its time evolution in the presence of the Coulomb correlations is somewhat non-trivial. The linear growth of the spin po- larization degree at 0 < t < Γ is common for the cases with and without on-site Coulomb correlations. Start- ing from t = Γ the increase of P is suppressed by the Coulomb correlations (blue and green lines in Fig. 4. This is a net effect of the two: the spin polarization, which is the nominator in (14) is enhanced due to an effectively large spin splitting of the bound state but the total oc- cupation of QW, which is the denominator (14) remains larger as the electrons are stuck in the QW. The Coulomb correlations do not manifest themselves if γ1 (cid:29) Γ as was discussed above (red line in Fig. 4). As the total number of non-equilibrium carriers in the system decreases the Coulomb correlations effect on polarization degree van- ishes and all the curves converge at larger times in Fig. 4. The characteristic time evolution of the polarization de- gree demonstrated in Fig. 4 has been never reported be- fore. In our opinion, it gives a good opportunity to verify the role of the Coulomb correlations in the systems of the considered type experimentally. It is also clear, that the influence of the correlations can be well controlled by adjusting the system parameters. In particular, for the system design shown in Fig. 1,b the bound state relax- ation rate γ1 is directly related to the transparency of the barrier on the left, which can be tuned by changing the barrier height or its thickness. V. SUMMARY We have studied dynamic spin injection by the mech- anism of spin-dependent relaxation in a quantum well coupled to the spin-split bound state. In this work for the first time the impact of the Coulomb correlations at the bound state on the spin and sheet density kinetics in the QW were analyzed. As supported by our analysis, the effect of the Coulomb correlations is twofold. Firstly, the on-site Coulomb repulsion leads to an effectively larger spin splitting and, consequently, an enhanced spin polar- ization of the electrons remaining in the QW. Secondly, it increases the characteristic time of the carriers relax- ation in the QW since it reduces the electron tunneling into the bound state. We predict that the interplay of these two effects would lead to the non-trivial depen- dence of a circular polarization degree of photolumines- cence from the QW. This characteristic dependence will 6 (RFBR) grant No. 18-02-00668 and Scientific Program of RAS No. 9 "Terahertz optoelectronics and spintronics". V.N.M. also acknowledges the support by the RFBR grant 16 − 32 − 60024 mol − a − dk. allow probing the strength of the on-site Coulomb cor- relations experimentally. As shown by our analysis, the effect of the Coulomb correlations can be controlled by affecting the relaxation times. For example, the bound state relaxation time can be tuned by a tunnel barrier separating it from the lead. This opens a way of study- ing the Hubbard-like electron-electron interactions exper- imentally. VI. ACKNOWLEDGEMENTS This work has been carried out under the financial support of the Russian Foundation for Basic Research 1 A. Hoffmann and S. D. Bader, Phys. Rev. Applied 4, 12 E. L. Ivchenko, L. A. Bakaleinikov, and V. K. Kalevich, 047001 (2015). 2 I. Zuti´c, J. Fabian, and S. Das Sarma, Rev. Mod. Phys. 76, 323 (2004). 3 J. Sinova and I. Zuti´c, Nature Materials 11, 368 (2012). 4 C. H. Li, O. M. J. van t Erve, J. T. Robinson, Y. Liu, and L. Li, Nature Nanotechnology 9, 218 (2014). 5 D. Pesin and A. H. MacDonald, Nature Materials 11, 409 (2012). 6 S. Bader and S. Parkin, Annual Review of Condensed Mat- ter Physics 1, 71 (2010). 7 A. Fert and H. Jaffr`es, Phys. Rev. B 64, 184420 (2001). 8 M. Oltscher, M. Ciorga, M. Utz, D. Schuh, D. Bougeard, and D. Weiss, Phys. Rev. Lett. 113, 236602 (2014). 9 T. Dietl and H. Ohno, Rev. Mod. Phys. 86, 187 (2014). 10 E. I. Rashba, Phys. Rev. B 62, R16267 (2000). 11 M. Battiato and K. Held, Phys. Rev. Lett. 116, 196601 (2016). Phys. Rev. B 91, 205202 (2015). 13 E. L. Ivchenko, L. A. Bakaleinikov, M. M. Afanasiev, and V. K. Kalevich, Physics of the Solid State 58, 1539 (2016). 14 V. Korenev, I. Akimov, S. Zaitsev, V. Sapega, L. Langer, D. Yakovlev, Y. A. Danilov, and M. Bayer, Nat. Commun. 3, 959 (2012). 15 I. Akimov, V. L. Korenev, V. F. Sapega, L. Langer, S. V. Zaitsev, Y. A. Danilov, D. R. Yakovlev, and M. Bayer, physica status solidi (b) 251, 1663 (2014). 16 I. V. Rozhansky, K. S. Denisov, N. S. Averkiev, I. A. Aki- mov, and E. Lahderanta, Phys. Rev. B 92, 125428 (2015). and 17 K. S. Denisov, I. V. Rozhansky, N. S. Averkiev, E. Lahderanta, Semiconductors 51, 43 (2017). 18 A. Hewson, The Kondo Problem to Heavy Fermions (Cam- bridge University Press, 1993). 19 N. Maslova, P. Arseyev, and V. Mantsevich, Solid State Communications 248, 21 (2016). 20 J. Q. You and H.-Z. Zheng, Phys. Rev. B 60, 8727 (1999). 21 J. Q. You and H. Z. Zheng, Phys. Rev. B 60, 13314 (1999). 

1306.6478

1

1306

Charge transport through interfaces: a tight-binding toy model and its implications

With the help of a tight-binding (TB) electronic-structure toy model we investigate the matching of parameters across hetero-interfaces . We demonstrate that the virtual crystal approximation, commonly employed for this purpose, may not respect underlying symmetries of the electronic structure. As an alternative approach we propose a method which is motivated by the matching of wave functions in continuous-space quantum mechanics. We show that this method obeys the required symmetries and can be applied in simple band to band transitions. Extension to multiple interfaces and to more sophisticated TB models is discussed.

cond-mat.mes-hall

cond-mat

Noname manuscript No. (will be inserted by the editor) Charge transport through interfaces: a tight-binding toy model and its implications B. A. Stickler · W. Potz the date of receipt and acceptance should be inserted later PACS 72.25.Hg,85.75.-d,75.50.Cc Abstract With the help of a tight-binding (TB) electronic- structure toy model we investigate the matching of param- eters across hetero-interfaces . We demonstrate that the vir- tual crystal approximation, commonly employed for this pur- pose, may not respect underlying symmetries of the elec- tronic structure. As an alternative approach we propose a method which is motivated by the matching of wave func- tions in continuous-space quantum mechanics. We show that this method obeys the required symmetries and can be ap- plied in simple band to band transitions. Extension to mul- tiple interfaces and to more sophisticated TB models is dis- cussed. 1 Introduction The modeling of quantum transport in nano-structured semi- conductor devices is one of the major challenges in applied solid-state physics [1,2,3,4,5,6,7,8,9,10]. In most cases a reliable quantum mechanical treatment of charge transport based on ab-initio electronic structure calculations seems to be impossible with state-of-the-art methods. For many mate- rial classes even "ab-initio" methods require empirical input in order to reliably capture the electronic structure in equi- librium (T = 0). Furthermore, they generally cannot cap- ture the non-linear-response regime [11,12]. We remark that these shortcomings may not apply to the method suggested by Brandbyge et al. [13]. As a resort, an empirical framework for the modeling of electronic devices under a finite bias has commonly been used. The calculation is subdivided into three steps: (i) the electronic properties of (the components of) the device are B.A. Stickler · W. Potz Institute of Physics, Karl-Franzens Universitat Graz, Austria. E-mail: [email protected] identified by means of computational or experimental ef- forts, (ii) the findings of step (i) are implemented into a model Hamiltonian which in step (iii) is used in the transport calculations, usually, with the inclusion of (self-consistent) corrections to account for the specific non-equilibrium sit- uation. Typically, step (i) yields the (bulk) electronic struc- ture of individual components of the device, as well as the band offset, for example, from a super-cell calculation. In step (ii) this information is combined into a model Hamil- tonian for the entire device. Here one faces the problem of matching two or more regions with limited information re- garding the interface. In the simplest case, one may know the electronic structure of two bulk semiconductors and their relative band alignment and be faced with the task to design an effective Hamiltonian for the study of charge transport across the hetero-interface. In the case of single-band electronic transport in the para- bolic regime, an effective-mass model (EMM) may be ap- propriate. There is a plethora of analytic, as well as nu- merical, methods available to solve the associated boundary value problem. In particular, if a linear voltage drop is as- sumed, Gundlach's method may be employed to solve the problem analytically [14]. However, the eigenvalues of the problem are substantially influenced by the boundary and matching conditions imposed upon the Schrodinger equa- tion. We note that the EMM may be viewed as a special one- band limit of the envelope-function approach. The matching strategy for the latter also faces problems when the overlaps of the Bloch functions for the two bulk materials to be joined are not known [15]. For more complex band structure situations, empirical tight-binding (TB) models provide a popular modeling tool [5,4,10,1,8,7,2]. In this case the TB parameters have to be regarded as pure fitting parameters, [16,17,18,19], to be distinguished from the case where they stem from ab-initio calculations [20,21,22]. Within this model, transport calcu- 3 1 0 2 n u J 7 2 ] l l a h - s e m . t a m - d n o c [ 1 v 8 7 4 6 . 6 0 3 1 : v i X r a 2 B. A. Stickler, W. Potz lations usually are performed within the framework of non- equilibrium Green's functions (NEGFs) or the transfer ma- trix method [10,5]. Besides its numerous benefits, such as conceptual sim- plicity and computational effectiveness, the empirical TB approach also comes with caveats. Recently it has been ar- gued by Tan et al. [22] that the reliability of transport cal- culations based on empirical TB parameters is rather ques- tionable and, therefore, a direct mapping procedure based on ab-initio methods is preferable. Here, we shall address an even more serious disadvantage of commonly employed techniques in cases where the only information available consists of the bulk electronic structure and the band-offsets between the materials involved [23,24]. A common ap- proach to the modeling of the interface for binary compounds is the virtual crystal approximation (VCA) [25,26,4,1,2,3, 5]. It is based on a simple (linear) interpolation of TB hop- ping and/or on-site parameters at the interface for establish- ing a matching between materials. Within this work and utilizing a toy model we demon- strate that the commonly used VCA for constructing the interface TB elements may not respect all symmetries of the underlying TB model and we propose an alternative ap- proach which preserves these symmetries. It is important to remark that the method proposed by Brandbyge et al. [13] also employs an ad hoc approximation which is of the form of the VCA. This paper is structured as follows: In Sec. 2 we define a rather simple model Hamiltonian and briefly discuss the NEGF method, as well as the EMM, as required for our purpose. In Sec. 3 we demonstrate that the VCA is likely to lead to inconsistencies by investigating an artificial and a genuine interface. Finally, in Sec. 4 we propose an alter- native formulation of the interface matching problem and demonstrate that it respects the symmetries supplied by the input information. Conclusions are drawn in Sec. 5. 2 The model We define a bulk toy model in order to formulate the match- ing problem and to demonstrate the inconsistencies which may arise. It is an infinite one-dimensional two component tight-binding chain with nearest-neighbor hopping only. Each element of the chain contains one orbital and the grid-spacing is given by a > 0. The Hamiltonian may be written as H = (cid:229) e s l,s i hl,s ls +t12(cid:229) (l,2i hl + 1,1 + l,1i hl,2) , (1) l where l ∈ Z labels the unit cells, s ∈ {1,2} labels the atoms within the unit cells and l,s i are the basis-kets localized at lattice point (l,s ). Furthermore, e s ∈ R are the onsite ener- gies of atom s and t12 is the hopping element. Please note that in writing Eq. (1) we assume that t12 ∈ R for reasons of simplicity. Let us define an onsite matrix e and a hopping matrix t via e =(cid:18) e 1 t12 t12 e 2(cid:19) and t =(cid:18) 0 0 t12 0(cid:19) . (2) (3) The eigen-energies of the Hamiltonian (1) can be expressed as E1,2(k) = B ±qB2 − A(k), where we introduce the wavenumber k ∈(cid:2)− A(k) = det [h(k)] , p a , together with B = 1 2 Tr [h(k)] . (4) p a(cid:3) and define (5) (6) Moreover, we define the Hamiltonian matrix h(k) as the rep- resentation of the Hamiltonian H in reciprocal space, h(k) = e + t exp(ika) + t† exp(−ika). (7) Please note that we refer to the operator (1) as the Hamilto- nian while we denote the matrix h(k) as the Hamiltonian ma- trix. The band structure defined by Eq. (4) is completely de- termined by the set of TB parameters x = (e 1,e 2,t12) ∈ R3. In order to emphasize this dependence we denote En(k) ≡ En(k,x ), where n = 1,2. We shall now consider two different bulk materials each characterized by a Hamiltonian of the form (1) with TB pa- rameters x L and x R, respectively. We bring these two materi- als into contact and investigate the resulting heterostructure assuming that the band-offset is known and no further in- terface effects are taken into account. In what follows we present two common approaches to calculate the transmis- sion function T (E) through the interface for some particular energy E. Please note that we restrict our discussion to the case of zero bias for reasons of simplicity. The general argu- ments also apply to heterostructures under bias. As a first technique we shall discuss the NEGF approach with the VCA. We introduce two Hamiltonians of the form Eq. (1), however, each defined only on one half axis, i.e. l ∈ Z− and l ∈ Z+. Furthermore, we denote HL and HR the Hamiltonians defined on the semi-infinite domains of the left material L (l = −1, −2, . . .) and the right material R (l = 1,2, . . .). Diagonal elements of the TB Hamiltonians are shifted to comply with the band alignment. These two Charge transport through interfaces: a tight-binding toy model and its implications 3 Hamiltonians are then connected with the help of an inter- face Hamiltonian HI such that the total Hamiltonian H of the system is of the form H = HL + HI + HR. (8) 2 It is obvious that the particular choice of HI significantly influences the transport physics of the system, yet, in most applications its exact structure is unknown. In the particular case of two diatomic materials AB and CB which share the atomic constituent B, as for instance in the case of a GaAs / AlAs heterostructure [4], a common approach is the vir- tual crystal approximation, for instance [25,26], in which HI contains one element from the left chain, associated with A, while the onsite element of the common atom B is averaged at the interface, i.e. 2 + (1 − x)e R e I 2 = xe L where x ∈ [0,1]. Here, e L/R denote the onsite enery of atom B according to the TB parameter sets x L/R. The coupling to the left semi-infinite Hamiltonian is then described ac- cording to tL 12, while the coupling to the right semi-infinite Hamiltonian is described by the hopping between atoms B and C, i.e. tR 12. The transmission function T (E) across the interface as- sociated with Hamiltonian (8) may, for instance, be calcu- lated within the NEGF formalism, dividing the TB chain into three segments: the left semi-infinite lead consisting of the unit cells l = −2, −3, . . ., the interface region referred to as system S given by the unit cells l = −1,0,1, and the right semi-infinite lead consisting of the cells l = 2,3, . . .. T (E) is given by [10] (9) 2 SG LGA S(cid:3) , T (E) = Tr(cid:2)G RGR (10) where Tr [·] denotes the operator trace, G R/L are the coupling functions to the right (R) and the left (L) semi-infinite leads, respectively, and GR/A denote the retarded and advanced system's Green's function. The latter are given by [10] S , (11) GR S =(cid:2)(E + ih )I − HS − S R L − S R R(cid:3)−1 S =(cid:0)GR S(cid:1)† and HS is the system Hamiltonian. In Eq. and GA (11) h > 0 is a small parameter, I is the identity, and S R L/R is the retarded self-energy of the leads. The retarded self- energy of the leads is calculated from the surface Green's functions of the left and the right lead GR L/R, respectively, via [10] S R L = t†GR Lt, and S R Rt† R = tGR (12) help of layer doubling, as suggested by Sancho et al. [27]. The coupling functions appearing in Eq. (10) are given by G L/R = i(cid:16)S R L/R − S A L/R(cid:17) . (13) An entirely different approach is to calculate the trans- mission with the help of an EMM. We calculate the effective mass of the band of interest, say n, at k = 0 via m(x ) = ¯h2(cid:20) d2En(k,x ) dk2 (cid:21)−1 k=0 . In what follows, we employ the notation m =(mL mR x ≤ 0, x > 0, (14) (15) where we replace the discrete index l ∈ Z by the continu- ous variable x ∈ R. Again, we partition the domain into two different regimes: (I) with m = mL for x ≤ 0 and (II) with m = mR for x > 0 and solve the corresponding eigenvalue problem analytically. In particular, in region (I) we have y ′′(x) + 2mL ¯h2 Ey (x) = 0, with the solution y (x) = exp(ikLx) + R exp(−ikLx). (16) (17) Here, R is the reflection amplitude and the wavenumber kL is given by kL =r 2mL ¯h2 E. (18) In similar fashion we obtain for regime (II) the solution y (x) = T exp(ikLx), when no right-incident waves are considered. Here T is the transmission amplitude and the wavenumber kR reads (19) (20) ¯h2 (E + D ), kR =r 2mR with D ∈ R the bandoffset. If we consider a linear volt- age drop between two leads, one may solve the correspond- ing Schrodinger equation analytically with the help of Airy functions [14]. The numerical constants R and T are deter- mined by the continuity conditions of the wave function and the current density. These conditions read where we take into account that t12 ∈ R. The surface Green's functions, in general, can be determined recursively with the T (E) = kRmL kLmR T 2. 1 + R = T , kR mR (1 − R) = kL mL The transmission function T (E) is then calculated via T . (21a) (21b) (22) 4 B. A. Stickler, W. Potz A major benefit of the EMM is clearly its conceptual simplicity, but, in most cases the parabolic approximation may not be justified and the method of choice is the determi- nation of T (E) within a NEGF approach as discussed above. However, by writing Eq. (8) for the total Hamiltonian H we employ information which we actually do not have. Even worse, it turns out that the electronic structure is invariant under certain transformations of the TB parameters and that these symmetries are destroyed by the particular choice of HI in the VCA (9). Let us exemplify this dilemma within the next section in more detail. 3 The dilemma Let us define the operators A and B acting on vectors x ∈ R3 via A : (e 1,e 2,t12) 7→ (e 1,e 2, −t12), (23a) and B : (e 1,e 2,t12) 7→ (e 2,e 1,t12). (23b) Then we note from Eq. (4) - (7) that the electronic structure En(k,x ) obeys the invariances En (k, A x ) = En (k,x ) , (24a) and En (k, Bx ) = En (k,x ) , (24b) for all x ∈ R3. Thus, the electronic structure described by the Hamiltonian Eq. (1) is invariant under a change of the sign of the hopping parameter connecting the two atomic species, see Eq. (24a). Furthermore, the electronic struc- ture is also invariant under a relabeling of the atoms within the unit cell, as expressed in Eq. (24b). This has the conse- quence that TB parameters are not uniquely determined by the band structure while the reversed statement is entirely true. Moreover, given a set of TB parameters we can im- mediately construct three further sets of parameters which yield completely identical bands by simply employing the invariances (24a) and (24b). In this light, the VCA for TB models may become prob- lematic because one needs to unambiguously identify the associated atomic species, which according to Eq. (24b) is not possible in many cases. In particular, e 2(Bx L) = e 1(x L). On the other hand, the EMM respects the symmetries Eqs. (24a) and (24b), since the effective mass is solely based on the electronic structure, see Eq. (14). In summary, if the only information available is the elec- tronic structure En(k) then the set of TB parameters defin- ing the underlying Hamiltonian (1) is not unique. This is not problematic as long as one deduces from the Hamiltonian 4 2 0 −2 V e / E −2 0 k/a 2 Fig. 1 (Color online) Electronic structure of the toy model Eq. (1) with TB parameters x = (2, −1, 1). (1) observables which too are invariant under a change in the TB parameters according to A and B. On the other hand, if the observables are not invariant under these transformations one introduces an inconsistency because one utilizes infor- mation which is not contained in the electronic structure, as for instance, the sign of t12. In what follows we investigate this problematic in terms of the toy model in order to quantify the error. 3.1 An artificial heterostructure For a first example we consider a homogeneous system with a bulk electronic structure described by a Hamiltonian of the form (8), i.e. we model a diatomic material comparable to, for instance, GaAs. The TB parameters x = (2, −1,1) uniquely determine the electronic structure. Now an artifi- cial interface between two semi-infinite linear chains is con- structed by employing the symmetry operators A and B to the right chain. For the left semi-infinite chain we keep x L = x , while for the right semi-infinite chain x R is one of the parameter sets: x a = x L, x b = A x L, x c = Bx L and x d = A Bx L. Since the transmission obtained with an EMM is invariant under these substitutions we shall restrict the fol- lowing discussion to the NEGF formalism with the VCA. The electronic structure, as well as the resulting transmis- sion function for the lower band with x = 0.5, are illustrated in Fig. 2. The transmission function resulting from the correct treat- ment of the material (x R = x a, red solid line) agrees with the result obtained with the help of x R = x b and with that of an EMM. However, the transmission originating from an ex- change of the parameters 1 ↔ 2 according to B strongly de- viates from the original curve. The reason obviously stems from modeling an interface of the form · · · ABABBABA · · · instead of · · · ABABABAB · · ·. This example clearly illustrates a major problem in the TB treatment of interfaces with the help of the VCA. Moreover, we do not know which value Charge transport through interfaces: a tight-binding toy model and its implications 5 1 0.8 0.6 0.4 0.2 T a c d 4 2 0 −2 V e / E 0 −2 −1.5 E / eV −1 −2 0 k/a 2 Fig. 2 (Color online) The transmission function through the lower band resulting from an artificial interface as obtained by applying the symmetry operators A and B to x . Please note that we did not plot the transmission of x b because it is equivalent to the correct case x a. Fig. 4 (Color online) Electronic structure according to the TB param- eters x L = (2, −1, 1) (red solid lines) and x R = (2.6398, −0.0602, 1.5) (blue dashed lines). 1 0.8 0.6 0.4 0.2 T a x = 0.25 x = 0.375 x = 0.5 1 0.8 0.6 0.4 0.2 T a c eff. mass 0 −2 −1.5 E / eV −1 0 −2 −1.5 E / eV −1 Fig. 3 (Color online) The transmission function for the lower band resulting from an artificial interface with x c for different values of x. of x in the VCA (9) should give the correct result. We illus- trated this dependency for the case x R = x c in Fig. 3 for three different values of x. 3.2 A genuine heterostructure For the second example we investigate two truly different materials characterized by x L and x R and which have one atom in common, i.e. we are looking at an interface of the form · · · ABABCBCB · · ·. According to the above discussion it is not possible to assign TB parameters to a particular atom within the unit cell as expressed in Eq. (24b). This point is particularly crucial if the TB parameters are genuine fitting parameters stemming from the electronic structure solely, as in the case of empirical TB approaches [16,17,18]. Then it is not possible to unambiguously identify certain parameters with one atomic species in general. The two electronic structures depicted in Fig. 4, respec- tively, correspond to the TB parameters x L = (2, −1,1), for AB, and x = (2.6398, −0.0602,1.5), for material BC. Again, Fig. 5 (Color online) Transmission T (E) for the composite material with the NEGF technique with parameters x a (red solid), x c (magenta dash dotted) and for the EMM (black solid line). The curves for x b and x d have been omitted since they are equivalent to x a and x c, respec- tively. x R takes is equivalent to one of the TB parameters x a = x , x b = A x , x c = Bx and x d = A Bx . The resulting trans- mission functions for the lower band are depicted in Fig. 5. In summary, for a simple model system we have iden- tified the reasons why the VCA does not yield satisfactory results. The problem is that (i) an arbitrary mixing parameter x is introduced at the interface, and (ii) the symmetry of the electronic structure under the operators A and B acting on x , Eqs. (24a) and (24b), is not respected within the VCA. It is clear that ambiguities grow dramatically with increasing complexity of the TB model. A determination of the trans- mission function within the EMM respects this symmetry, however, the EMM itself often is too crude for an electronic structure model. In the EMM the preservation of symme- try in the transmission function is a result of the matching condition at the interface, see Sec. 2. In what follows we propose a comparable approach for TB models as a possible method for avoiding ambiguities in the interface problem. x x x x x x 6 B. A. Stickler, W. Potz 4 The interface matching problem - a possible solution Within this section we propose a discrete matching approach for TB models which is adopted from the matching proce- dure within the EMM. The main procedure consists in solv- ing the bulk problem for the two different materials sepa- rately and matching the wave functions, as well as the cur- rent between unit cells, at the interface. Such an approach has two main advantages: (i) it does not introduce arbitrary parameters at the interface and (ii) it respects the symme- tries of the electronic structure, e.g. under variation of x ac- cording to A and B, Eqs. (24a) and (24b), for the case of the TB Hamiltonian (1). However, this approach still cannot account for interfacial effects as long as no additional infor- mation is made available. It has, therefore, to be regarded as a best solution based on the information given. We shall first illustrate the procedure for the matching of two single- atomic linear TB chains in subsection 4.1 since it allows an entirely analytic exposition of the method. Subsequently, the method is generalized and then applied to the toy model in subsection 4.2. The assumption of an abrupt interface between to semi- infinite crystals inevitably represents an approximation which leads to a loss of information regarding microscopic details at the interface. In an empirical TB model parameters should become position dependent near the interface. However, the mapping of ab-initio band structure calculations onto em- pirical TB models numerically is rather intractable (due to supercell size) and one may resort to an approach based on stationary scattering theory. Considering a quasi-1D system, such as a heterostructure, a stationary solution for given en- parallel k-vector k , may be written as n,k,ki(cid:11), with band index n and ergy E with in-asymptote(cid:12)(cid:12) Ei = n,k,ki(cid:11) + (cid:229) (cid:12)(cid:12)  tn,i;m′, j′(E,k)(cid:12)(cid:12) (out) m′, j′ (out) m, j rn,i;m, j(E,k)(cid:12)(cid:12) m′,k,k j′(cid:11) m,k,k j(cid:11) for z < 0, for z > 0, (25) when the interface is positioned at "z = 0". The sum is over the out-channels for which E = Em(k,k j) = Em′(k,k j′). In- and out-channels, respectively, are identified as having group velocity z-components towards the interface and away from the interface. In general, degeneracy may imply more than two out-channels m,k j for each in-channel. Unitarity of the S matrix and possibly other symmetries, such as time- reversal invariance, reduce the number of independent ele- ments rn,i;m, j(E,k) and tn,i;m′, j′(E,k) but are not sufficient to uniquely specify them if the interface potential at z ≈ 0 is unknown. Let NL and NR denote the number of available in-channels on the left- and the right-hand side of the interface at a given energy E. Under time-reversal symmetry, we also have NL and NR out-channels on each side. Then, unitarity of the S matrix leads to (NL + NR)2 conditions for 2(NL + NR)2 un- knowns. For NL = NR = 1, i.e., one out-channel on each side, one is left with two unknowns. For this case, and this represents our suggested alternative for an ad-hoc VCA, the condition of continuity in charge and current density across the interface determine the transmission and reflection at the hetero-interface, in analogy to the effective-mass case and a finite potential step. For higher degeneracy, additional in- formation is necessary to determine all of the S matrix ele- ments. 4.1 Tight-binding interface matching for single-atomic chain Two semi-infinite single-atomic linear TB chains are to be connected at the grid point l = 0. The connection between these two chains is not established by introducing an ad- hoc hopping parameter t across the interface, as for instance suggested by Harrison [25], but rather by matching the left and right solutions at the interface. For this purpose, the left- and right-semi-infinite TB Hamiltonian both are extended to site l = 0. Specifically we propose the matching conditions y L(0)2 = y R(0)2, and (26a) (26b) +1/2. JL −1/2 = JR Here, y L/R are the wave functions at site l = 0 expressed respectively, in terms of the left (L) and right (R) TB or- bitals. The second condition (26b) guarantees stationarity of the solution, i.e. the current flowing between the grid points l = −1 and l = 0 in the left chain equals the current flowing between l = 0 and l = 1 in the right chain. While the match- ing conditions (21) can be derived from the Schrodinger equation, these two conditions (26) may be viewed as a (sta- tionary version of the) particle continuity equation in terms of TB orbitals. Condition (26a) ensures that the probability for finding a particle at the interface is the same when ap- proaching the interface from left or right: at the interface this probability can be expressed either in terms of the L or R ba- sis functions. In the parabolic regime above conditions (26) become exact (i.e., agree with the ones for the Schrodinger equation) and coincide with Eqs. (21). The isolated Hamiltonian of the L and R single-particle TB chain, respectively, is denoted by HL and HR with on- site energies e L, e R and hopping elements tL, tR, respectively. The corresponding energy bands are EL/R(k) = e L/R + 2tL/R cos(aL/Rk). (27) (cid:229) Charge transport through interfaces: a tight-binding toy model and its implications 7 For both bulk Hamiltonians, we write the eigenfunctions as linear combination of localized states(cid:12)(cid:12) Y L/R(cid:11) = (cid:229) (cid:12)(cid:12) lL/R(cid:11) , aL/R l (cid:12)(cid:12) l lL/R(cid:11) where the general form of the coefficients aL/R is well known to be of the form aL/R l (k) = AL/R exp(ikla) + BL/R exp(−ikla). l (28) for given k (29) We assume for simplicity that the lattice constants coincide aL = aR ≡ a. In analogy to Eq. (17) we choose AL = 1, BL = R, AR = T, and BR = 0. (30) For given energy E, condition (26a) reads 0 [kR(E)]2, 0[kL(E)]2 = aR aL where we define the wave numbers kL/R(E) = E −1 L/R(E) with the help of the inverse of Eq. (27). With Eq. (30) we rewrite this condition as (31) 1 + R(E)2 = T (E)2. (32) We calculate the current from lattice point −1 to lattice point 0 via J−1/2 = 0)∗tLaL −1 − i(aL −1)∗tLaL = − tL(R2 − 1)sin(kLa), i ¯h(cid:2)(aL 2 ¯h 0(cid:3) and from 0 to 1 as J+1/2 = 2 ¯h T 2tR sin(kRa). Furthermore, it is convenient to define the velocities 1 ¯h vL/R(k) = 2 ¯h and rewrite condition (26b) as EL/R(k) = − d dk atL/R sin(ka), R(E)2 + vR[kR(E)] vL[kL(E)] T (E)2 = 1. (33) (34) (35) (36) Hence, we solve the coupled equations (32) and (36). Un- der the assumption that R,T ∈ R these equations are easily solved to give T (E) = 2vL(kL) vL(kL) + vR(kR) , and R(E) = vL(kL) − vR(kR) vL(kL) + vR(kR) . Clearly, the quantities of physical interest are vL(kL) vR(kR) T (E)2 = 4vR(kR)vL(kL) [vL(kL) + vR(kR)]2 , (37a) (37b) (38a) 1 0.8 0.6 0.4 0.2 0 0 T R 1 2 E / eV 3 4 Fig. 6 (Color online) Reflection and transmission coefficient through the interface between two semi-infinite single atom TB chains as ob- tained with the matching method (solid lines) and with the EMM (dashed lines). V e / E 5 4 3 2 1 0 EL(k) ER(k) −2 0 k / a 2 Fig. 7 (Color online) Electronic structures of the two isolated semi- infinite chains (solid lines) and parabolic approximations (dashed lines). and vL(kL) + vR(kR)(cid:19)2 R(E)2 =(cid:18) vL(kL) − vR(kR) . (38b) The transmission and reflection as a function of energy E can be obtained by inverting Eq. (27), however, it has to be kept in mind that, due to our choice of Eq. (30), one has to take the branch vL(kL),vR(kR) ≥ 0. In Fig. 6 we illustrate the transmission and reflection coefficients Eq. (38) for e L = 2, e R = 1, tL = −1, tR = −0.5 in comparison with the result obtained with an EMM. The corresponding electronic struc- tures Eq. (27) together with the parabolic approximations are illustrated in Fig. 7. From Fig. 6 we observe that the reflection and the trans- mission coefficients from the left to the right TB chain are well approximated by the EMM only for energies near to the band minimum. This is in accordance with the effective mass approximation, see Fig. 7. Moreover, we note that the reflection correctly approaches 1 as the energy E approaches 2 eV, see Fig. 6. This is due to the reduced band width of the 8 B. A. Stickler, W. Potz electronic structure of the right semi-infinite chain, see Fig. 7. The transmission is maximal for low energies. In what follows we shall formulate the matching method for more general TB Hamiltonians and then apply it to the diatomic TB chain model discussed in subsection 3.2. 4.2 The genuine heterostructure revisited Let us begin with a brief review of some important proper- ties of general nearest neighbor TB models. The Hamilto- nian H is of the form H = (cid:229) e s ls i hls (39) ls + (cid:229) ll′ss ′ ss t ll′ ′ ls i(cid:10)l′s ′(cid:12)(cid:12) where l and l′ label the unit cell and s = 1, . . . ,N is an ad- ditional index. This additional index might include, for in- stance, the atoms in the unit cells as well as their orbitals and the parallel momentum kk, if the Hamiltonian (39) is derived from a three-dimensional TB model by partial Wan- nier transformation [4]. The hopping elements tss couple ll′ only up to neighboring unit cells. This is a valid assumption for any finite-range TB model since the unit cell can be cho- sen as large as necessary [28]. We expand the wavefunction as y i = (cid:229) cls ls i ≡ (cid:229) cl · li , (40) ′ ls l where in the very last step we introduce the N-component vectors cl = {cls } and a "vector ket" li = {ls i}. Bloch's theorem allows one to extract the space dependence of the vector cl at a given energy E in form of a phase factor cl(E) = c[k(E)]exp [ik(E)la] +c[−k(E)]exp[−ik(E)la] , (41) where k(E) is the inverse of the electronic structure E = En(k) for a given energy E and c(k) is the eigenvector of the Hamiltonian matrix h(k) with eigenvalue E. The Hamilto- nian matrix h(k) reads h(k) = e + exp(ika)t + exp(−ika)t†, (42) where we define an onsite matrix e with matrix elements e s d ss ll+1, respectively. Please note that the Hamiltonian matrix Eq. (7) is a special case of this general form. Furthermore, the cur- rent Jl+1/2 between unit cells l and l + 1 can be written as and a coupling matrix t with elements tss ′ + tss ll ′ ′ Jl+1/2 = l tcl+1 − c† i ¯h(cid:16)c† l+1t†cl(cid:17) . Hence, the general matching conditions for given energy E read cL[kL(E)] + cL[−kL(E)]R2 = T cR[kR(E)]2 , (44a) (43) and JL −1/2 = JR +1/2, where we use i JL −1/2 = † tLcL ¯hh(cid:0)cL −1(cid:1) 0 −(cid:0)cL 0(cid:1) † t† −1i , LcL with cL l = cL[kL(E)]exp [ikL(E)laL] +RcL[kL(E)]exp [−ikL(E)laL] , and JR +1/2 = with i † tRcR ¯hh(cid:0)cR 0(cid:1) 1 −(cid:0)cR 1(cid:1) † t† 0i , RcR cR l = T cR[kR(E)]exp [ikR(E)laR] . (44b) (44c) (44d) (44e) (44f) Here, cL/R[kL/R(E)] are the eigenvectors to the Hamiltonian matrices hL/R(k) with eigenenergies E, which are of the form Eq. (7), however, corresponding to the Hamiltonians HL and HR, respectively. Moreover, with kL/R(E) for a given energy E we define the inverse of the dispersion relations EL n (k) and ER n (k), respectively. Finally, aL/R denote the lattice constants of the two materials. Note that these matching conditions re- spect time-reversal symmetry. We solve the above equations (44) for the problem dis- cussed in Sec. 3.2. The resulting transmission and reflection functions are plotted in Fig. 8, in comparison to the result obtained with the help of Eqs. (38). It is very important to notice that the result depicted in Fig. 8 is independent of the actual choice of the parameter sets x L and x R, i.e. the transmission function is independent under the action of the symmetry operators A and B. This follows from the match- ing conditions (38), since all quantities entering these equa- tions are solely determined by the bulk Hamiltonians HL/R. In Fig. 9 we illustrate the transmission through the diatomic TB heterostructure discussed in Sec. 3.1 in comparison with the results obtained with the help of the virtual crystal ap- proximation, see Sec. 3.1. The transmission and reflection function versus energy computed in this fashion may subsequently be mapped back onto a TB model. This can be achieved by constructing a Hamiltonian H of the form H = HL +V (E) + HR, (45) where one introduces an energy dependent coupling Hamil- tonian V (E) between the two chains. Following the proce- dure outlined in Sec. 2 to obtain the transmission T (E) with the help of a Green's function approach, the elements of the hopping matrix V (E) is then determined in such a way that the transmission obtained by solving Eqs. (38) is repro- duced. Hence, V (E) will depend on the particular choice of Charge transport through interfaces: a tight-binding toy model and its implications 9 1 0.8 0.6 0.4 0.2 0 −2 T R Ts Rs −1.5 −1 E / eV −0.5 0 Fig. 8 (Color online) Transmission and reflection coefficient through the interface between two semi-infinite diatomic TB chains as obtained with the matching method (solid lines) and as obtained when employ- ing relations (38) (dashed lines). T 1 0.8 0.6 0.4 0.2 0 −2 T Tfit TVCA1 TVCA2 −1.5 −1 E / eV −0.5 0 Fig. 9 (Color online) Transmission coefficient throught the interface between two semi-infinite diatomic TB chains as obtained with the NEGF formalism where the hopping has been fitted to the result of the matching method, (solid red and dashed blue lines). Furthermore, we present the solutions obtained with the VCA in Sec. 3.2. x L and x R, however, is determined in such a way that it re- produces a parameter-independent result, such as the trans- mission T (E). Moreover, as long as the voltage drop at the interface is small, V (E) may provide a good approximation for the heterostructure under bias. An extension to multiple interfaces of this type is straight-forward. This method works as long as to every in-channel Bloch state there is only one out-channel state, respectively, to the left and to the right. In other words, as long as one deals with the problem of matching a doubly degenerate band in one material with another one in the other. The proposed match- ing conditions (44) alone are not sufficient to tackle the case of multi-channel scattering where, for given in-channel E,n,k (energy, band index, and k-parallel) the degenerate out-channel states lie in different energy bands n, or in the same band n when a degeneracy of greater than two is present. A unique solution cannot be obtained without further assumptions. This is most easily observed by inspection of a situation with one in-channel on the left hand side i and two out-channels on the right hand side. In this case the wave function to the right of the interface is given by a linear combination of the first and the second channel states with weights ti,1(E) and ti,2(E) in Eq. (25). It is clear, that the matching condi- tions Eqs. (44) may be solved only if the ratio ti,1(E)/ti,2(E) is known. An estimate for this ratio might be obtained by "blanking off" one out-channel at a time and determining the two transmission functions at a time. However, in this proce- dure, phase information is lost. It should also be mentioned that in many multichannel problems, there may be dominant out-channels (those which couple dominantly with a given in-channel). This may be used to reduce the problem in an approximation to a tractable 2 by 2 form of one out-channel on each side of the interface. In addition, of course, empiri- cal input from experiment may be helpful in the fitting pro- cedure. 5 Conclusions We have studied the electron transmission through a hetero- interface as modeled by the linking of two diatomic single- orbital TB chains. The two separate TB chains are char- acterized solely by their bulk properties while no particu- lar information about the interface, other than the relative band alignment, is available. For this example we demon- strate that the commonly employed VCA does not respect the underlying symmetries of the electronic structure in re- lation to the TB model used in the fitting procedure. Com- monly, however, the latter is the key input information avail- able. Hence, the VCA introduces an arbitrary error which is hard to estimate. As a remedy to this ambiguity we sug- gest a matching method of the wave function and current density which is motivated by the continuity relations of continuous space quantum mechanics. In particular, the ob- tained transmission functions respect the symmetries of the band structures associated with the TB model and can there- fore be regarded as a best result under available informa- tion. The transmission function determined with the help of the proposed matching method can subsequently be used to construction of a hopping Hamiltonian which reproduces this transmission function. Thus, multiple interfaces can be treated in this fashion. We stress that the proposed matching method does not give a physically complete description of the interface. In fact, this is not possible since the required information re- garding the coupling matrix at the interface which enters the total Hamiltonian H is considered to be unknown. Hence, the approach discussed here is no substitute for a full (micro- scopic) study of the hetero-interface. But, since such a study is almost always based on large supercell calculations, one is confronted with the problem of fitting numerous bands which, in most cases, practically is not feasible. 10 B. A. Stickler, W. Potz 15. W. Potz and D. K. Ferry. On the boundary conditions for envelope- function approaches for heterostructures. Superlattices and Mi- crostr., 3(57), 1987. 16. P. Vogl, Harold P. Hjalmarson, and John D. Dow. A semi-empirical tight-binding theory of the electronic structure of semiconduc- tors. Journal of Physics and Chemistry of Solids, 44(5):365 -- 378, 1983. 17. Jean-Marc Jancu, Reinhard Scholz, Fabio Beltram, and Franco Bassani. Empirical spds∗ tight-binding calculation for cubic semi- conductors: General method and material parameters. Phys. Rev. B, 57:6493 -- 6507, Mar 1998. 18. F. Starrost, S. Bornholdt, C. Solterbeck, and W. Schattke. Band- Phys. Rev. B, structure parameters by genetic algorithm. 53:12549 -- 12552, May 1996. 19. Timothy B. Boykin, Gerhard Klimeck, and Fabiano Oyafuso. Va- lence band effective-mass expressions in the sp3d5s∗ empirical tight-binding model applied to a si and ge parametrization. Phys. Rev. B, 69:115201, Mar 2004. 20. Skriver H. The LMTO method: Muffin Tin Orbitals and Electronic Structure. Springer, Berlin, 1984. 21. O. K. Andersen, C. Arcangeli, R. W. Tank, T. Saha-Dasgupta, G. Krier, O. Jepsen, and I. Dasgupta. Third-generation tb-lmto. eprint arXiv:cond-mat/9804166, April 1998. 22. Yaohua Tan, Michael Povolotskyi, Tillmann Kubis, Yu He, Zheng- ping Jiang, Gerhard Klimeck, and TimothyB. Boykin. Empiri- cal tight binding parameters for gaas and mgo with explicit ba- sis through dft mapping. Journal of Computational Electronics, 12(1):56 -- 60, 2013. 23. Su-Huai Wei and Alex Zunger. Calculated natural band offsets of all ii -- vi and iii -- v semiconductors: Chemical trends and the role of cation d orbitals. Applied Physics Letters, 72(16):2011 -- 2013, 1998. 24. Yong-Hua Li, Aron Walsh, Shiyou Chen, Wan-Jian Yin, Ji-Hui Yang, Jingbo Li, Juarez L. F. Da Silva, X. G. Gong, and Su-Huai Wei. Revised ab initio natural band offsets of all group iv, ii-vi, and iii-v semiconductors. Applied Physics Letters, 94(21):212109, 2009. 25. W.A. Harrison. Elementary Electronic Structure. World Scientific - Singapore, 1999. 26. Massimiliano Di Ventra. Eletrical Transport in Nanoscale Sys- temambridges. Cambridge University Press, Cambridge, 2008. 27. M P Lopez Sancho, J M Lopez Sancho, J M L Sancho, and J Ru- bio. Highly convergent schemes for the calculation of bulk and surface green functions. Journal of Physics F: Metal Physics, 15(4):851, 1985. 28. J. N. Schulman and Yia-Chung Chang. Reduced hamiltonian method for solving the tight-binding model of interfaces. Phys. Rev. B, 27:2346 -- 2354, Feb 1983. The matching method is applicable to the case of one out-channel on each side (per in-channel). If more in- or out- channels are present, a simple matching technique cannot yield the channel resolved transmission since the problem is underdetermined (since the Hamiltonian at the interface is unknown). However, if the transmission ratio between all available out-channels is known, the proposed method can still be employed in order to obtain the total transmission function. 6 Acknowledgments This work was supported financially by FWF project P221290- N16. References 1. Roger Lake, Gerhard Klimeck, R. Chris Bowen, and Dejan Jo- vanovic. Single and multiband modeling of quantum electron transport through layered semiconductor devices. Journal of Ap- plied Physics, 81(12):7845 -- 7869, 1997. 2. C. Strahberger and P. Vogl. Model of room-temperature resonant- tunneling current in metal/insulator and insulator/insulator het- erostructures. Phys. Rev. B, 62:7289 -- 7297, Sep 2000. 3. Timothy B. Boykin, Jan P. A. van der Wagt, and James S. Har- ris. Tight-binding model for gaas/alas resonant-tunneling diodes. Phys. Rev. B, 43:4777 -- 4784, Feb 1991. 4. J. A. Støvneng and P. Lipavsk´y. Multiband tight-binding approach to tunneling in semiconductor heterostructures: Application to g X transfer in gaas. Phys. Rev. B, 49:16494 -- 16504, Jun 1994. 5. Aldo Di Carlo. Microscopic theory of nanostructured semi- conductor devices: beyond the envelope-function approximation. Semiconductor Science and Technology, 18(1):R1, 2003. 6. Aldo Di Carlo, P. Vogl, and W. Potz. Theory of zener tunnel- ing and wannier-stark states in semiconductors. Phys. Rev. B, 50:8358 -- 8377, Sep 1994. 7. Christian Ertler and Walter Potz. Electrical control of ferromag- netism and bias anomaly in mn-doped semiconductor heterostruc- tures. Phys. Rev. B, 84:165309, Oct 2011. 8. Christian Ertler, Walter Potz, and Jaroslav Fabian. Proposal for a ferromagnetic multiwell spin oscillator. Applied Physics Letters, 97(4):042104, 2010. 9. Christian Ertler and Walter Potz. Disorder effects on resonant tun- neling transport in gaas/(ga,mn)as heterostructures. Phys. Rev. B, 86:155427, Oct 2012. 10. S. Datta. Electronic Transport in Mesoscopic Systems. Cambridge Studies in Semiconductor Physics and Microelectronic Engineer- ing, 1999. 11. O. Bengone, O. Eriksson, J. Fransson, I. Turek, J. Kudrnovsk´y, and V. Drchal. Electronic structure and transport properties of CrAsGaAsCrAs trilayersfrom first principles theory. Phys. Rev. B, 70:035302, Jul 2004. 12. Phivos Mavropoulos, Nikolaos Papanikolaou, and Peter H. Ded- erichs. Korringa-kohn-rostoker green-function formalism for bal- listic transport. Phys. Rev. B, 69:125104, Mar 2004. 13. Mads Brandbyge, Jos´e-Luis Mozos, Pablo Ordej´on, Jeremy Tay- lor, and Kurt Stokbro. Density-functional method for nonequilib- rium electron transport. Phys. Rev. B, 65:165401, Mar 2002. 14. K.H. Gundlach. Zur berechnung des tunnelstroms durch eine trapezfrmige potentialstufe. Solid-State Electronics, 9(10):949 -- 957, 1966. 

1206.3474

2

1206

Molecular adsorption study of nicotine and caffeine on the single-walled carbon nanotube from first principles

Using first-principles calculations, we investigate the electronic structures and binding properties of nicotine and caffeine adsorbed on single-walled carbon nanotubes to determine whether CNTs are appropriate for filtering or sensing nicotine and caffeine molecules. We find that caffeine adsorbs more strongly than nicotine. The different binding characteristics are discussed by analyzing the modification of the electronic structure of the molecule-adsorbed CNTs. We also calculate the quantum conductance of the CNTs in the presence of nicotine or caffeine adsorbates and demonstrate that the influence of caffeine is stronger than nicotine on the conductance of the host CNT.

cond-mat.mes-hall

cond-mat

Molecular adsorption study of nicotine and caffeine on the single-walled carbon nanotube from first principles Hyung-June Lee,1 Young-Kyun Kwon,1 and Gunn Kim2 1Department of Physics and Research Institute for Basic Sciences, Kyung Hee University, Seoul 130-701, Korea 2Department of Physics and Graphene Research Institute, Sejong University, Seoul 143-747, Korea (Dated: August 31, 2018) Using first-principles calculations, we investigate the electronic structures and binding properties of nicotine and caffeine adsorbed on single-walled carbon nanotubes to determine whether CNTs are appropriate for filtering or sensing nicotine and caffeine molecules. We find that caffeine adsorbs more strongly than nicotine. The different binding characteristics are discussed by analyzing the modification of the electronic structure of the molecule-adsorbed CNTs. We also calculate the quantum conductance of the CNTs in the presence of nicotine or caffeine adsorbates and demonstrate that the influence of caffeine is stronger than nicotine on the conductance of the host CNT. I. INTRODUCTION Carbon nanotubes (CNTs) have attracted much aca- demic and industrial interest because of their remarkable physical and chemical properties.1 -- 11 One of the most intriguing characteristics of CNTs is a very large surface area12 -- 14 comparable to that of activated carbon. This property has been exploited to develop various applica- tions including hydrogen storage, filters removing toxic compounds, and chemical sensors. In spite of the over- estimated earlier studies,15 -- 17 however, CNTs did not appear to be a good storage medium for hydrogen due to their inert surfaces, where H2 molecules adsorb very weakly via van der Waals interaction.18 -- 21 The binding properties of hazardous molecules on CNTs were also ex- amined theoretically and experimentally22 -- 26 for poten- tial filter application. One of the most promising applica- tions of CNTs is a sensor application. CNT-based sensors were developed and demonstrated to detect a wide range of molecules such as ammonia,27 nitrogen dioxide,27 oxy- gen,28 alcohol,29 and other molecules.30 -- 32 They verified that CNT-based sensors are able to detect not only the type of gas molecule, but also the concentration of the adsorbate. Nicotine and caffeine are both alkaloids and therapeu- tic compounds. They serve as stimulants for the central nervous system in humans. Through tobacco and coffee, nicotine and caffeine are addictive drugs used heavily by humans. Because of their psychoactive effects, nicotine and caffeine could have positive effects. For example, nicotine appears to enhance concentration and memory due to the increase of acetylcholine,33 and caffeine seems to enhance performance in endurance sports.34,35 How- ever, these drugs often produce side effects. Both of them may increase the heart rate that further limits the abil- ity of the body to maintain homeostasis during exercise. Overdose of these drugs causes severe problems. In par- ticular, nicotine overdose can be deadly, and excessive ingestion of caffeine over extended periods of time re- sults in a toxic condition (caffeinism), with symptoms of vomiting, elevated blood pressure, rapid breathing, heart palpitations, and insomnia. As beverages containing a lot of caffeine to provide mental or physical stimulation, en- ergy drinks have been recently produced. Because they contain more caffeine than strong coffee, some countries have restrictions on the sale and/or manufacture of en- ergy drinks. Therefore, the fabrication of sensors and filters with high sorption capability of these chemicals is desired for public health. In this paper, we report our study on the adsorption properties of nicotine and caffeine on CNTs. We ex- amined the modification of the electronic structures of CNTs due to adsorbates. We found that both nicotine and caffeine have non-covalent interaction with the CNT through π-stacking. Although both of nicotine and caf- feine contain aromatic rings and nitrogen atoms, it was found that they show very different binding geometries and binding energies. II. COMPUTATIONAL DETAILS We performed first-principles calculations based on the density functional theory36,37 to study the structural and electronic properties of the adsorbates on CNTs. We employed a plane-wave basis set and the projector aug- mented wave (PAW) implemented in the Vienna ab ini- tio Simulation Package (VASP).38,39 The cutoff energy for the kinetic energy was set to 500 eV. The generalized gradient approximation (GGA)40 was used to describe the exchange-correlation energy functional. In general, chemical bonding and electron transfer can be described well within the GGA. However, the dispersion or van der Waals (vdW) forces are not represented well by the GGA functional because quantum electronic interactions in the regions of low electron densities are not correctly expressed. In these calculations, thus we made the cor- rection of vdW interaction between the adsorbates and the CNT, using Grimme's method41 for the GGA (i.e., the GGA plus vdW). To verify the validity of our computational results, we also repeated our work with the linear combination of pseudoatomic orbitals implemented in the SIESTA code.42,43 We used 210 Ry as the mesh cutoff energy and 3 1 0 2 l u J 9 ] l l a h - s e m . t a m - d n o c [ 2 v 4 7 4 3 . 6 0 2 1 : v i X r a double-ζ basis with polarization for the basis set. All the model structures were relaxed until none of the resid- ual forces acting on any atom would exceed 0.03 eV/aB, where aB is the Bohr radius. The binding energies were corrected using the basis set superposition error correc- tion with ghost atoms. The binding energies, electronic band structures, and wavefunctions presented in this pa- per were all obtained using the pseudoatomic orbital ba- sis set implemented in the SIESTA package. The general trends in the binding energies and the electronic struc- tures for nicotine and caffeine are very similar for the two computational packages. We selected the semiconducting (8, 0) CNT as the host material for molecular adsorption. We used a tetragonal supercell with a length of 16.94 A along the CNT axis and 20 A along the lateral directions. We utilized 1 × 1 × 4 Monkhorst-Pack sampling44 for the Brillouin zone inte- grations. On the other hand, a cubic supercell with a length of 20 A along all three directions and only one k point at the Γ point were used for calculations of the iso- lated molecules (nicotine and caffeine). Charge transfer between the adsorbate and the host CNT was estimated using the Bader charge analysis. For each configuration, we obtained its binding energy (Eb) defined by Eb = E[CNT] + E[molecule] − E[CNT+molecule], where E[CNT+molecule] and E[CNT] are the total en- ergies of the CNT with and without a nicotine (or caf- feine) molecule, respectively, and E[molecule] represents the energy of an isolated nicotine (or caffeine) molecule. The quantum transport calculations were performed using the TranSIESTA code, based on density functional theory (DFT), and the self-consistent electronic structure of a nanostructure connected to the electrodes. Non- equilibrium Green's functions were used to solve an open system from the DFT Hamiltonian. In the calculations of the quantum transport, 1 × 1 × 36 k points were used in the Monkhorst-Pack scheme.44 III. RESULTS AND DISCUSSION First, we calculated the individual nicotine and caf- feine molecules in the vacuum condition to optimize the gas phase geometry. Figures 1a and 1b show the opti- mized geometries of nicotine and caffeine molecules. All the nitrogen atoms form essentially a planar structure with carbon atoms implying sp2 orbital hybridization, maintaining the aromatic characteristics. Figures 2a -- c show the density of states (DOS), the highest occupied molecular orbital (HOMO) and the low- est unoccupied molecular orbital (LUMO) of a nicotine molecule, respectively. Similarly, Figs. 2d -- f represent the same quantities of a caffeine molecule. The energy gaps between the HOMO and the LUMO levels are 3.34 eV and 3.37 eV for the nicotine and caffeine molecules, re- spectively, which are very similar to each other. However, the energy difference between the LUMO and LUMO+1 2 FIG. 1: Model structures of (a) nicotine and (b) caffeine. Left: top view. Right: side view. FIG. 2: (a) DOS, (b) HOMO and (c) LUMO orbitals of a nicotine molecule. (d) DOS, (e) HOMO and (f) LUMO or- bitals of caffeine molecule. (the state one level above the LUMO) of the nicotine molecule in Fig. 2a is much smaller that that of the caf- feine molecule in Fig. 2d. Interestingly, the HOMO of nicotine is dominant at the five-membered ring as shown in Fig. 2b, whereas the electronic density in the LUMO is high at the phenyl group (six-membered ring) as dis- played in Fig. 2c. In contrast, the HOMO and LUMO wavefunctions of caffeine appear to be distributed over the whole molecule as shown in Figs. 2e and 2f. (a) (b) Hydrogen Carbon Nitrogen Oxygen -4 -3 -2 -1 0 1 2 3 4 DOS(states/eV/cell) E-EF (eV) (a) DOS(states/eV/cell) (b) (c) (d) (e) (f) E-EF (eV) -4 -3 -2 -1 0 1 2 3 4 3 FIG. 3: Optimized structure of the (8, 0) CNT with (a) an adsorbed nicotine molecule and (b) an adsorbed caffeine molecule. The figures on the left hand side show side view, whereas the right hand side figures do top view. To avoid confusion, the carbon atoms in the CNT are shown in yellow. Figure 3 shows the optimized structures of the (8, 0) CNTs with adsorbed nicotine and caffeine molecules. As shown in Fig. 3a, the nicotine molecule prefers to contact with the CNT through its six-membered ring rather than its five-membered ring. Moreover, the most stable bind- ing geometry appears similar to Bernal ("AB") stacking of graphite with the binding energy of Eb ≈ 0.46 eV and the binding distance of db ≈ 3.29 A. On the other hand, the caffeine molecule is bound to the CNT stronger than the nicotine molecule (almost twice). Its binding energy and distance are Eb ≈ 0.88 eV and db ≈ 3.25 A, respec- tively. It is noteworthy that according to an ab initio study of Debbichi et al., the binding energy of antracene on the (10, 10) CNT is ∼ 1 eV when the van der Waals interaction is considered.45 Our charge analysis demon- strates that no appreciable charge transfer occurs be- tween the molecules and the CNT. Then, what makes the difference in the binding energy for the two types of molecules? Unlike nicotine, a hexagon and a pentagon in caffeine are both on the same plane, and thus this planar structure enhances the π stacking with the CNT surface. In contrast to AB stacking in the nicotine adsorption, the caffeine molecule prefers "AA" to AB stacking, which is only ∼ 5 meV less stable though. Figure 4 displays the electronic structures of the bare, nicotine-, and caffeine-adsorbed (8, 0) CNTs. The bare (8, 0) CNT has an energy band gap of 0.58 eV (Fig. 4b), whereas both nicotine- and caffeine-adsorbed CNTs in- crease their band gap to ∼ 0.61 eV, as seen in Figs. 4a and 4c. In Figs. 4a -- c, we did not find a significant shift in the Fermi level by the molecular adsorption implying no significant charge transfer between the adsorbate and the CNT. In each of the band structure of the molecule- adsorbed CNT, there are two flat bands originating from FIG. 4: The band structures of (a) the nicotine-adsorbed CNT, (b) the bare (8, 0) CNT and (c) the caffeine-adsorbed CNT. (d-e) The states A and B indicate wavefunctions at la- beled "A" and "B" in (a). (f-g) The states C and D indicate wavefunctions at labeled "C" and "D" in (c). The red and blue surfaces indicate the phase of the wave functions. the molecule. For nicotine (caffeine), such flat bands are located near 0.6 eV (0.6 eV) and 1.3 eV (0.9 eV) below the Fermi level. Figures 4d and 4e display the wavefunc- tions corresponding to the states labeled "A" and "B" in Fig. 4a at the Γ point. The state A in Fig. 4d shows a localized state corresponding to the HOMO of nico- tine, whereas the state B exhibits a weakly hybridized π state on the CNT together with the nicotine HOMO−1 state (the state just below the HOMO). As shown in Fig.2b, the electronic density in the HOMO of nicotine is mainly at the five-membered ring and thus it has no practical coupling with any CNT state. Similarly, we show in Figs. 4f and 4g the wavefunctions at the Γ point for the two states labeled "C" and "D" for the caffeine adsorption in Fig. 4c. Interestingly, due to the stronger interaction of caffeine with the CNT than that of nicotine via π-stacking, the state C corresponding to the HOMO state of caffeine shows interaction between an extended state of the CNT and a molecular orbital of caffeine. In 2010, Girao et al. investigated nicotine adsorption on the single-wall CNT46. Thus, we need to compare the differences between their models and ours. For the binding energy, our model has stronger adsorption than (a) (b) (a) (b) (c) 2 1 0 1 2 E-EF (eV) Γ A Γ A Γ A k State A State B State D A B C D State C Nicotine + CNT Bare CNT Caffeine + CNT Caffeine + CNT (d) (f) (e) (g) C D 4 to very narrow dips in conductance (from 4G0 to 3G0) around 1.5 eV below the Fermi level, which are related to the states B and D in Figs. 4e and 4g, respectively. As mentioned above, the state C corresponds to the HOMO of caffeine strongly coupled to the extended CNT state, whereas the states B and D originate from the HOMO−1 states of nicotine and caffeine and they are very weakly hybridized with CNT states, respectively. We note that the energy levels of conductance dips may have an off- set with those of energy eigenstates. The wavefunctions and the electronic band structures are obtained for a fi- nite CNT with an adsorbed molecule in periodic bound- ary conditions. In contrast, the quantum conductance is calculated for a finite CNT with an adsorbate molecule connected to two semi-infinite perfect CNTs (electrodes) in both sides. Therefore, the energy offset may occur between the quantum electron conductance and enegy eigenstates. Consequently, the caffeine molecule using a CNT-based device can be monitored in the current- voltage (I-V) measurement. In the case of nicotine, on the other hand, it is difficult to detect the molecule using a CNT-based electronic device. Therefore, one should use the CNTs to filter nicotine molecules. IV. CONCLUSION In conclusion, we have investigated the electronic structures and binding properties of nicotine and caffeine molecules on (8, 0) single-walled carbon nanotubes us- ing first-principles calculations. We found that both ad- sorbates show different binding characteristics on CNTs. The nicotine molecule binds to the CNT with an ad- sorption energy of 0.46 eV, whereas the caffeine molecule adsorbs with a higher binding energy of 0.88 eV. To fig- ure out the difference, we discussed the modification of the electronic structures of the nicotine- and caffeine- adsorbed CNTs. Although there is no significant charge transfer between the CNT and the adsorbate, we found that there is an noticeable interaction between them. In particular, the caffeine-adsorbed CNT exhibits a special binding feature formed by electron donation and back- donation. Our study of quantum transport demonstrates that I-V curves can distinguish nicotine from caffeine us- ing a CNT-based electronic sensor. We believe our find- ings of the adsorption properties of caffeine and nicotine on the CNT will help the future studies of intermolecular interaction between the host nanotube and other organic molecules. Acknowledgement G. K. acknowledges the Priority Research Center Pro- gram (Grant No. 2010-0020207) and the Basic Science Research Program through the National Research Foun- dation of Korea (NRF) (Grant No. 2013R1A1A2009131). Y. K. acknowledges the financial support from the Na- FIG. 5: Charge density difference and quantum conductance plots of (a) nicotine- and (b) caffeine-adsorbed CNTs. In the charge density difference plots, red and blue colors repre- sent regions of electron accumulation and depletion, respec- tively. Here, G0 (= 2e2/h) represents the conductance quan- tum, where e is called the elementary charge (≈ 1.602× 10−19 C) and h is the Planck constant (≈ 1.436 × 10−15 eV·s). their models for nicotine adsorbed on the CNT. For the band structure, their models have localized states origi- nating from the nicotine molecule in the forbidden band (just above the valence band top). In contrast, our most stable model for nicotine-adsorbed CNT has no localized state in the forbidden band (see Fig. 4). For further analysis, we calculated charge density dif- ferences shown in Fig. 5. Isodensity surfaces indicate the regions of electron accumulation (red) and depletion (blue). It is shown that the nicotine molecule creates a local polarized region at the CNT near the adsorbate as shown in Fig. 5b. We found a more interesting feature in the caffeine-adsorbed CNT, where an electron deple- tion takes place near the O atom bonded to the C atom between two N atoms in the six-membered ring of the caf- feine molecule. At the same time, an electron accumula- tion region is formed at the opposite end of the adsorbate (near the five-membered ring). This phenomenon can be interpreted as a "donation and back-donation" between the CNT and the caffeine molecule in the axial direc- tion of the CNT. It is noteworthy that the DFT calcula- tion does not describe accurately the amount of charge transfer but it guarantees the overall shape of isodensity surfaces. Finally, we present the quantum conductance plots of the (8, 0) CNTs with the adsorbed nicotine and caffeine molecules. As shown in Fig. 5, only caffeine gives rise to a broad conductance dip about 1.1 eV below the Fermi level with the reduction of conductance from 4G0 to 3G0, which is attributed to the state C in Fig. 4f. Here, G0 (= 2e2/h) represents the conductance quantum, where e is called the elementary charge (≈ 1.602×10−19 C) and h is the Planck constant (≈ 1.436×10−15 eV·s). On the other hand, both of nicotine and caffeine molecules give rise G/G0 G/G0 E-EF (eV) E-EF (eV) (a) Nicotine (b) Caffeine (8,0) CNT (8,0) CNT+Nicotine (8,0) CNT (8,0) CNT+Caffeine tional Research Foundation of Korea (Grant No. 2011- 0016188). Some portion of our computational work was done using the resources of the KISTI Supercomputing Center (KSC-2012-C2-19 and KSC-2012-C2-72). 5 1 L. Chico, V. H. Crespi, L. X. Benedict, S. G. Louie, M. L. Cohen, Phys. Rev. Lett. 76 (1996) 971. 2 S. J. Tan, A. R. M. Verschueren, C. Dekker, Nature 393 (1998) 49. 3 Y.-K. Kwon, D. Tom´anek, S. Iijima, Phys. Rev. Lett. 82 (1999) 1470. B. H. Tian, Z. P. Jia, Chem. Phys. Lett. 376 (2003) 154. 24 S. B. Fagan, A. G. Souza, J. O. G. Lima, J. Mendes, O. P. Ferreira, I. O. Mazali, O. L. Alves, M. S. Dresselhaus, Nano Lett. 4 (2004) 1285. 25 S. B. Fagan, E. C. Girao, J. Mendes, A. G. Souza, Int. J. Quan. Chem. 106 (2006) 2558. 4 Z. Yao, H. W. Ch. Postma, L. Balents, C. Dekker, Nature 26 F. Tournus, S. Latil, M. I. Heggie, J. C. Charlier, Phys. 402 (1999) 273. Rev. B 72 (2005) 075431. 5 S. Sanvito, Y.-K. Kwon, D. Tom´anek, C. J. Lambert, Phys. 27 J. Kong, N. Franklin, C. Zhou, M. Chapline, S. Peng, K. Rev. Lett. 84 (2000) 1974. Cho, H. Dai, Science 287 (2000) 622. 6 S. Berber, Y.-K. Kwon, D. Tom´anek, Phys. Rev. Lett. 84 28 P. G. Collins, K. Bradley, M. Ishigami, A. Zettl, Science (2000) 4613. 7 C. Zhou, J. Kong, E. Yenilmez, H. Dai, Science 290 (2000) 1552. 8 J. Lee, H. Kim, S.-J. Kahng, G. Kim, Y.-W. Son, J. Ihm, H. Kato, Z. W. Wang, T. Okazaki, H. Shinohara, Y. Kuk, Nature 415 (2002) 1005. 9 G. Kim, S. B. Lee, T.-S. Kim, J. Ihm, Phys. Rev. B 71 (2005) 205415. 287 (2000) 1801. 29 H. J. Song, Y. Lee, T. Jiang, A.-G. Kussow, M. Lee, S. Hong, Y.-K. Kwon, H. C. Choi, J. Phys. Chem. C 112 (2008) 629. 30 K. Bradley, J.-C. P. Gabriel, A. Star, G. Gruener, Appl. Phys. Lett. 83 (2003) 3821. 31 M.-F. Yu, B. S. Files, S. Arepalli, R. S. Ruoff, Phys. Rev. Lett. 84 (2000) 5552. 10 J.-C. Charlier, X. Blase, S. Roche, Rev. Mod. Phys. 79 32 P. Burt, N. R. Wilson, J. M. R. Weaver, P. S. Dobson, J. (2007) 677. V. Macpherson, Nano Lett. 5 (2005) 639. 11 W. I. Choi, J. Ihm, G. Kim, Appl. Phys. Lett. 92 (2008) 33 J. Rusted, L. Graupner, N. O'Connell, C. Nicholls, Psy- 193110. 12 R. Q. Long, R. T. Yang, Ind. Eng. Chem. Res. 40 (2001) 4288. 13 M. Cinke, J. Li, B. Cen, A. Cassell, L. Delzeit, J. Han, M. Meyyappan Chem. Phys. Lett.365 (2002) 69. 14 Y. F. Yin, T. Mays, B. McEnanaey, Langmuir 15 (1999) 871. 15 A. C. Dillon, K. M. Jones, T. A. Bekkedahl, C. H. Kiang, D. S. Bethune, M. J. Heben, Nature 386 (1997) 377. 16 A. Chambers, C. Park, R. T. K. Baker, N. M. Rodriguez, J. Phys. Chem. B 102 (1998) 4253. 17 P. Chen, X. Wu, J. Lin, K. L. Tan, Science 285 (1999) 91. 18 C. Liu, Y. Y. Fan, M. Liu, H. T. Cong, H. M. Cheng, M. S. Dresselhaus, Science 286 (1999) 1127. 19 C. C. Ahn, Y. Ye, B. V. Ratnakumar, C. Witham, R. C. Bowman Jr., B. Fultz, Appl. Phys. Lett. 73 (1998) 3378. 20 B. Panella, M. Hirscher, S. Roth, Carbon 43 (2005) 2209. 21 Y.-K. Kwon, J. Korean Phys. Soc. 57 (2010) 778. 22 R. Q. Long, R. T. Yang, J. Am. Chem. Soc. 123 (2001) 2058. 23 X. J. Peng, Y. H. Li, Z. K. Luan, Z. C. Di , Y. H. Wang, chopharmacology 115 (1994) 547. 34 D. Bishop, Sports Med. 40 (2010) 995. 35 S. A. Conger, G. L. Warren, M. A. Hardy, M. L. Millard- Stafford, Int. J. Sport Nutr. Exerc. Metab. 21 (2011) 71. 36 P. Hohenberg, W. Kohn, Phys. Rev. 136 (1964) B864. 37 W. Kohn, L. J. Sham, Phys. Rev. 140 (1965) A1133. 38 G. Kresse, J. Hafner, J Phys. Rev. B 47 (1993) R558. 39 G. Kresse, J. Furthmuller, Phys. Rev. B 54 (1996) 11169. 40 J. P. Perdew, K. Burke, M. Ernzerhof, Phys. Rev. Lett. 77 (1996) 3865. 41 S. Grimme, J. Comp. Chem. 27 (2006) 1787. 42 D. S´anchez-Portal, P. Ordej´on, E. Artacho, J. M. Soler Int. J. Quantum Chem. 65 (1997) 453. 43 E. Artacho, D. S´anchez-Portal, P. Ordej´on, A. Garc´ıa, J. M. Soler Phys. Status Solidi B 215i (1999) 809. 44 H. J. Monkhorst, J. D. Pack, Phys. Rev. B 13 (1976) 5188. 45 L. Debbichi, Y. J. Dappe, and M. Alouani, Phys. Rev. B 85 (2012) 045437. 46 E. C. Giriao, S. B. Fagan, I. Zanella, A. G. Souza Filho, J. Hazard. Mater. 184 (2010) 678. 

1202.2635

1

1202

Unifying explanation for carrier relaxation anomaly in gapped systems

We develop a theory to describe energy relaxation of photo-excited carriers in low-temperature ordered states with band gap opening and formulate carrier relaxation time $\tau$ near and below transition temperature $T_{\mathrm{c}}$ by quantifying contributions from different carrier-phonon scatterings to the relaxation rate. The theory explains anomalous experimental observations of $\tau$ in gapped systems. Transverse acoustic (TA) phonon modes play a crucial role in carrier relaxation; their heat capacity determines $\tau$-divergence near $T_{\mathrm{c}}$. The theory is validated by fitting $\tau$ of fullerene polymers onto a theoretical curve.

cond-mat.mes-hall

cond-mat

Unifying explanation for carrier relaxation anomaly in gapped systems Shota Ono,∗ Hiroyuki Shima, and Yasunori Toda Division of Applied Physics, Faculty of Engineering, Hokkaido University, Sapporo, Hokkaido 060-8628, Japan (Dated: July 7, 2018) We develop a theory to describe energy relaxation of photo-excited carriers in low-temperature ordered states with band gap opening and formulate carrier relaxation time τ near and below transition temperature Tc by quantifying contributions from different carrier-phonon scatterings to the relaxation rate. The theory explains anomalous experimental observations of τ in gapped systems. Transverse acoustic (TA) phonon modes play a crucial role in carrier relaxation; their heat capacity determines τ -divergence near Tc. The theory is validated by fitting τ of fullerene polymers onto a theoretical curve. PACS numbers: 78.47.-p, 63.20.kk, 71.45.Lr, 74.25.Gz The nonequilibrium dynamics of photo-excited carri- ers in solids has attracted considerable research interest in the field of condensed matter physics. These dynamics are governed by multiple scatterings of carriers and en- ergy transfer to the phonon field, both of which are quan- tified by the carrier relaxation time τexp and its tempera- ture (T ) dependence. Usually anomalous T -dependence of τexp is observed in many gapped systems, for example, phase-ordered systems showing an energy gap opening in the electron band below the transition temperature Tc. It has been found that a wide variety of superconductors [1 -- 12] and density-wave compounds [13 -- 19] show diverg- ing behavior of τexp near Tc, as confirmed by femtosecond time-resolved optical spectroscopy [1, 4, 5, 7 -- 17, 19]. The divergence of τexp in these gapped systems is believed to result from recursive energy transfer between electrons and phonons. Photo-excited electrons having high energy emit a number of phonons through relaxation from above to below the energy gap. Conversely, relaxed electrons can be re-excited above the gap by absorbing phonon energy. This phonon emission-reabsorption process be- comes efficient near Tc because of the small gap energy, and thus, it suppresses the relaxation of carriers, extend- ing τexp significantly. This anomalous phonon-mediated relaxation in gapped systems is called the phonon bottle- neck effect. This bottleneck enables the reproduction of these experimental observations of τexp in various gapped systems that exhibit τexp-divergence at Tc. There exists, however, a distinct class of gapped sys- tems such as Tl-based superconductors [2, 6] and C60- related materials [3, 18] that, instead of showing τexp- divergence at Tc, show monotonic increases in τexp with cooling across Tc. In light of the bottleneck, the mono- tonic variation in τexp near Tc appears controversial, lead- ing to the questions of why the τexp-divergence vanishes in a portion of gapped systems despite well-defined en- ergy gap formation at the Fermi level and whether the bottleneck concept is completely invalid in these gapped systems. Theoretical studies have been unable to satis- factorily answer these two questions over the last decade. In this Letter, we develop a theory of photo-excited carrier relaxation dynamics with the objective of resolv- ing the abovementioned issues. We postulate that the lack of τexp-divergence even in gapped systems can be as- cribed to the presence of transverse acoustic (TA) phonon modes. Further, we state that an ensemble of TA modes in gapped systems serves as a high-capacity thermal sink, and energy release to the sink from other phonon modes facilitate efficient cooling of carriers, in other words, a significant reduction in τexp close to Tc. The proposed theory is validated by the quantitative agreement with experimental data of τexp for peanut-shaped C60 poly- mers, a typical charge density wave (CDW) material that does not show any carrier relaxation anomalies. First, we briefly review the conventional bottleneck concept for τexp-divergence. It is based on the as- sumption that photo-excited carrier relaxation in gapped systems is regulated by the anharmonic slow decay of phonons [1, 20]. By the absorption of a pump laser pho- ton, electrons in the valence band are excited far above the initial states, and they rapidly accumulate in the up- per end of the gap through carrier-carrier and carrier- phonon interactions. Then, the carriers emit high-energy phonons (HEPs) whose energies are higher than the gap width 2∆ to relax into the lower end of the gap. The HEPs produced can be reabsorbed to create new carriers above the gap [see Fig. 1(a)] or they can decay in an an- harmonic manner into low-energy phonons (LEPs) that no longer excite carriers because their energy is lower than 2∆ [see left-hand side panel in Fig. 1(b)]. If the re- absorption probability per unit time is much larger than the inverse of the anharmonic decay time (τ −1), photo- excited carriers and HEPs settle in nearly steady states that obey the Fermi and Bose distribution functions, re- spectively, with temperature T ′ that is higher than the lattice temperature T [1]. Meanwhile, LEPs obey the Bose distribution function with T , because they remain unperturbed after the laser pulse incident. Consequently, the carrier relaxation is dominated by the energy transfer from HEPs to LEPs, that is, τexp ≃ τ , which is described (a) 2 (b) HEP HEP (c) S O D n o n o h P l TA h HEP LEP HEP TA TA FIG. 1: (Color online) (a) Emission-reabsorption process of HEPs across the energy gap with width 2∆. (b) Diagram of three-phonon scattering processes through which HEPs dissi- pate. Only the scatterings that involve TA modes are relevant to the lack of divergence in τ . (c) Phonon density of states of TA modes with cutoff frequency ΩTA, and those of LA modes with Ωh. The maximum frequency of the LEP is Ωl (see text for its definition). 2 plies that TA modes serve as a thermal receiver into which HEPs can dissipate, as a result of which the τ - divergence vanishes even at Tc. Below, we prove that this holds true in certain gapped systems. To formulate the anharmonic decay time τ of HEPs, we consider the time evolution of the phonon distribu- tion function. Following the above discussion, we divide phonon excitations into two groups. One group consists of HEPs (Ωl < ωq,LA < Ωh) in equilibrium temperature at T ′ that can participate in the reabsorption process. The other group involves TA modes (ωq,TA < ΩTA) and LEPs [ωq,LA < 2∆/(≡ Ωl)] in equilibrium at T that do not participate in the reabsorption process. Here, ωq,TA and ωq,LA represent the phonon dispersion rela- tions, where q is the wavevector. Ωh, Ωl, and ΩTA are the cutoff frequencies for the HEP, LEP, and TA modes within the Debye approximation, respectively. The dis- tribution functions for the two groups are given by n(ωq,j) =(cid:20)exp(cid:18) ωq,j kBx (cid:19) − 1(cid:21)−1 with an appropriate variable x = T or T ′. , (1) The rate of change in n(ωq,j) due to phonon-phonon collisions is written as [21] by the time evolution of the two temperatures, that is, T ′ = T ′(t) and T = T (t). It should be emphasized that in the conventional bot- tleneck concept, only longitudinal acoustic (LA) phonon modes are considered. Here, we point out the unnoticed but important role of TA phonon modes in the HEP's decay [see right-hand side panel in Fig. 1(b)]. Because TA phonon modes generate no density modulation in the lattice, they cannot interact with photo-excited carriers within the deformation potential theory. This fact im- ∂n(ωq,j) ∂t = Jcol[n(ωq,j)]. (2) The collision integral Jcol[n(ωq,j)] describes phonon scattering, and it is defined by three- Jcol[n(ω0)] = 2π N 2 Xq1,q2,j1,j2 wi→f 2(cid:18) 1 2 SA + SB(cid:19) , (3) where SA = {[n(ω0) + 1]n(ω1)n(ω2) − n(ω0)[n(ω1) + 1][n(ω2) + 1]} δ(ω0 − ω1 − ω2), SB = {[n(ω0) + 1][n(ω1) + 1]n(ω2) − n(ω0)n(ω1)[n(ω2) + 1]} δ(ω2 − ω0 − ω1). (4) (5) Here, ωs(s = 0, 1, 2) is an abbreviation of ωqs,js, and wi→f is the matrix element between the initial and the final state [22]. Note that Jcol regulates the time evolution of the total energy through the relationship ∂ ∂t (ELA + ETA) = Xq(LA) ωq,LAJcol + Xq(TA) ωq,TAJcol, ωq,TA < ΩTA, respectively. The energy Ej is defined by Ej =Xq(j) ωq,jn(ωq,j) =Z y 0 ωn(ω)ρj(ω)dω, (7) where y = ΩTA(Ωl) for j =TA (LA). ρj is the density of states defined by where the first and second summations on the right-hand side run over qs satisfying 0 < ωq,LA < Ωl and 0 < is the num- where z = ΩTA(Ωh) for j =TA (LA), νj ′ ber of phonon branches participating in the three-phonon (6) ρj(ω) = αjω2θ(z − ω), αj = 3νjN z3Pj ′ νj ′! , (8) scattering, N is the number of unit cells, and θ is the Heaviside step function; the total density of states is given by ρ = ρTA + ρLA [see Fig. 1(c)], satisfying the normalization condition 1 = R ρ(ω)dω/N . The Debye approximation we have used in Eq. (8) is valid when kBT < ΩTA(< Ωh), which holds for many gapped sys- tems below Tc. We simplify n(ω) in Eq. (7) as kBT /ω [1] to obtain Ej = CjT , where Cj = kBαjy3/3 (kB is the Boltzmann constant). In addition, we focus on the fact that each sum in Eq. (6) gives a non-zero value only when one or two HEPs contribute to the three-phonon scatter- ing represented by Jcol [see Fig. 1(b) as an example]. As a result, Eq. (6) is rewritten as [23] ∂T (t) ∂t = τ = using the definitions [T ′ − T (t)], 1 τ CLA + CTA ILA + ITA , ILA = w2 ITA = w2 1γ0 (VAT + VBT ′) , 2γ1 (VC T + VDT ′) + w2 2γ2VET, where γk = 2πk2 Bα3−k LA αk TA/(3N 2). T ′ is given by (9) (10) (11) (12) kBT ′ = − , (13) ∆ ln(cid:0)ǫ + e−∆/kB T(cid:1) where ǫ is the dimensionless photoexcitation energy [1]. To derive Eqs. (11) and (12), we assumed that the matrix element is momentum independent, that is, wi→f ≡ w1 (or w2) =const. when TA modes are absent from (join in) the three-phonon scattering. VX (X = A to E) in Eqs. (11) and (12) are functions of Ωl, ΩTA, and Ωh. Equation (10) is the main finding of this study, because it clearly shows the contribution of TA phonon modes to carrier relaxation. Figure 2(a) shows the T -dependence of τ that we have formulated in Eq. (10). The parame- ter p (≡ w2/w1) is the coupling strength through which the HEPs decay into TA modes, and it is tuned from 0 to 0.5 in increments of 0.1. With an increase in p, the magnitude of τ decreases over the entire T range. When p is smaller (larger) than ∼ 0.3, τ increases (decreases) as T approaches Tc from below. The most significant phenomenon is the drastic reduction in τ at Tc with in- creasing p. In fact, τ at Tc is inversely proportional to p2, as shown in Fig. 2(b). These results indicate that the lack of τ -divergence at Tc is attributable to efficient phonon-phonon coupling between the HEP and the TA mode characterized by p. To explain the microscopic mechanism for the lack of divergence, in Fig. 2(c), we show the T -dependence of the magnitude of ILA and ITA given in Eqs. (11) and (12). Here, ILA(TA) quantifies the efficiency of the HEP's energy dissipation through the anharmonic interaction with LA (TA) phonon modes. In the limit of T → Tc (a) p=0.0 p=0.1 p=0.2 p=0.3 p=0.4 p=0.5 105 104 A T 103 102 101 0.0 0.2 0.4 0.6 0.8 1.0 T/Tc (b) A T 103 102 101 100 ) A T B k ( / N I 0.1 100 (c) 10-2 10-4 10-6 10-8 0.0 3 p 1 ILA ITA (p=0.1) ITA (p=0.5) 0.5 T/Tc 1.0 FIG. 2: (Color online) (a) Numerical result of τ (T ) based on Eq. (10). See text for detailed numerical conditions. (b) Inverse square law of τ at T = Tc with respect to p. (c) T -dependence of ILA and ITA defined by Eqs. (11) and (12). Because ITA differs from zero at T = Tc, the τ -divergence at Tc is strongly suppressed. (that is, ∆ → 0), ILA vanishes (irrespective of p) but ITA converges to a finite value as long as p 6= 0. Therefore, we obtain a non-diverging τ at Tc if p 6= 0, which readily follows from Eq. (10). When p = 0, on the other hand, ITA ≡ 0 for arbitrary T [because ITA ∝ w2 2; see Eq. (12)]. In this case, ILA = ITA = 0 at Tc so that τ → ∞ at Tc. We thus conclude that the anharmonic decay of HEPs to TA modes plays a prominent role in the efficient cooling of HEPs as well as in determining the significance of the τ -divergence at Tc. Upon reducing the temperature to zero, the energy dissipation of HEPs gradually reduces due to monotonic decreases in ILA and ITA [see Fig. 2(c)]. The decrease in ILA(TA) is attributed to the reduced phonon population, which results in a monotonic increase in τ at low T , as shown in Fig. 2(a). A similar increase has been found in various gapped systems [2 -- 5, 10 -- 12, 15, 18], and it is attributable to the inefficient cooling of HEPs. Now, we apply the proposed theory to photo-excited carrier relaxation in quasi one-dimensional C60 poly- mers. It was previously observed in experiments [24] that the C60 polymers undergo the CDW transition [25] at Tc = 60K, forming a well-defined energy gap at the Fermi level. This result implied the possibility of τexp- divergence at 60K; nevertheless, optical pump-probe in- vestigations [18] of the C60 polymers revealed a mono- tonic variation in τexp near Tc, whose origin has yet to be clarified. This problem is solved by considering that the twisting phonon modes of the C60 polymer [26] play the same role as the above-described TA modes. Fig- ure 3 shows the numerical reproduction (indicated by lines) of the experimental data [18] (indicated by circles) of the T -dependent τexp of the C60 polymers. An over-  experiment p=0.25 p=0.30 p=0.35 8 6 4 2 ] s p [ 0 20 40 80 100 60 T [K] FIG. 3: (Color online) Experimental data of τexp in the C60 polymer (after Ref. [18]) and their numerical reproduction based on Eq. (10). The data for T ≤ Tc = 60K fits the theoretical curve with p = 0.3. all agreement between the theory and the experiments is obtained by assuming p ∼ 0.3. The other parame- ters we used are ΩTA = 220cm−1, Ωh = 360cm−1, and 2∆(0) = 360K; the first two values were estimated from the phonon model for 1D C60 polymers [26], and the last value gives 2∆(0)/kBTc = 6 consistent with many CDW compounds [27]. The generality of Eq. (10) is of great importance. It is applicable to τexp in other gapped systems such as Tl-based cuprate superconductors [2, 6] and the solid fullerenes K3C60 and Rb3C60 [3]. Even these materials show a lack of τexp-divergence; however, no theoretical studies have attempted to clarify their relaxation anoma- lies. We believe that the proposed theory will serve as a unified framework for nonequilibrium carrier dynamics in gapped systems. In conclusion, we have developed a theory to describe the energy dissipation from HEPs (ωq,LA > 2∆) to LEPs (ωq,LA < 2∆) and TA modes in gapped systems below Tc. This theory enables the evaluation of the T - dependence of τ as a function of the coupling strength p ≡ w2/w1 between the HEPs and the TA modes, and it explains the crossover between the diverging and the non-diverging behaviors of photo-excited carrier relax- ation. The latter behavior can quantitatively account for the anomalous carrier relaxation time in C60 polymers, as measured by recent pump-probe laser experiments. The coupling between the HEPs and the TA mode also sug- gests the variation of τexp-divergence in typical gapped systems. We thank K. Yakubo, Y. Asano, J. Onoe, K. Ohno, Y. Noda, and T. Matsuura for the helpful discussions. This study was supported by a Grant-in-Aid for Scientific Research from the MEXT, Japan. HS acknowledges the financial support from the Inamori Foundation and the Sumitomo Foundation. 4 hailovic, Phys. Rev. B 59, 1497 (1999). ∗ Electronic address: [email protected] [1] V. V. Kabanov, J. Demsar, B. Podobnik, and D. Mi- [2] D. C. Smith, P. Gay, D. Z. Wang, J. H. Wang, Z. F. Ren, and J. F. Ryan, Physica C (Amsterdam) 341, 2219 (2000). [3] S. B. Fleischer, B. Pevzner, D. J. Dougherty, H. J. Zeiger, G. Dresselhaus, M. S. Dresselhaus, E. P. Ippen, and A. F. Hebard, Phys. Rev. B 62, 1366 (2000). [4] J. Demsar, R. Hudej, J. Karpinski, V. V. Kabanov, and D. Mihailovic, Phys. Rev. B 63, 054519 (2001). [5] V. V. Kabanov, J. Demsar, and D. Mihailovic, Phys. Rev. Lett. 95, 147002 (2005). [6] E. E. M. Chia, J.-X. Zhu, D. Talbayev, R. D. Averitt, A. J. Taylor, K.-H. Oh, I.-S. Jo, and S.-I. Lee, Phys. Rev. Lett. 99, 147008 (2007). [7] T. Naito, Y. Yamada, T. Inabe, and Y. Toda, J. Phys. Soc. Jpn. 77, 064709 (2008). [8] Y. H. Liu, Y. Toda, K. Shimatake, N. Momono, M. Oda, and M. Ido, Phys. Rev. Lett. 101, 137003 (2008). [9] T. Mertelj, P. Kusar, V. V. Kabanov, L. Stojchevska, N. D. Zhigadlo, S. Katrych, Z. Bukowski, J. Karpinski, S. Weyeneth, and D. Mihailovic, Phys. Rev. B 81, 224504 (2010). [10] E. E. M. Chia, D. Talbayev, J.-X. Zhu, H. Q. Yuan, T. Park, J. D. Thompson, C. Panagopoulos, G. F. Chen, J. L. Luo, N. L. Wang, and A. J. Taylor, Phys. Rev. Lett. 104, 027003 (2010). [11] G. Coslovich, C. Giannetti, F. Cilento, S. Dal Conte, G. Ferrini, P. Galinetto, M. Greven, H. Eisaki, M. Raichle, R. Liang, A. Damascelli, and F. Parmigiani, Phys. Rev. B 83, 064519 (2011). [12] Y. Toda, T. Mertelj, P. Kusar, T. Kurosawa, M. Oda, M. Ido, and D. Mihailovic, Phys. Rev. B 84, 174516 (2011). [13] J. Demsar, K. Biljakovi´c, and D. Mihailovic, Phys. Rev. Lett. 83, 800 (1999). [14] J. Demsar, L. Forr´o, H. Berger, and D. Mihailovic, Phys. Rev. B 66, 041101 (2002). [15] E. E. M. Chia, J.-X Zhu, H. J. Lee, N. Hur, N. O. Moreno, E. D. Bauer, T. Durakiewicz, R. D. Averitt, J. L. Sarrao, and A. J. Taylor, Phys. Rev. B 74, 140409(R) (2006). [16] K. Shimatake, Y. Toda, and S. Tanda, Phys. Rev. B 73, 153403 (2006); Phys. Rev. B 75, 115120 (2007). [17] R. V. Yusupov, T. Mertelj, J.-H. Chu, I. R. Fisher, and D. Mihailovic, Phys. Rev. Lett. 101, 246402 (2008). [18] Y. Toda, S. Ryuzaki, and J. Onoe, Appl. Phys. Lett. 92, 094102 (2008). [19] Y. Toda, R. Onozaki, M. Tsubota, K. Inagaki, and S. Tanda, Phys. Rev. B 80, 121103 (2009). [20] A. Rothwarf and B. N. Taylor, Phys. Rev. Lett. 19, 27 (1967). [21] J. M. Ziman, Electrons and Phonons (Oxford University Press, New York, 2001). [22] For example, the first term in the curly brackets in Eq. (5) implies that one phonon with ω2 decays into two phonons with ω0 and ω1. [23] In the calculation, we assume that T ′ is fixed [that is, T ′(0) ≃ T ′(∞)] in order to derive the analytical expres- sion of τ . A similar assumption has been used in Ref. [1] for the case of the absence of the TA mode. [26] S. Ono and H. Shima, J. Phys. Soc. Jpn. 80, 064704 (2011). [24] J. Onoe, A. Takashima, and Y. Toda, Appl. Phys. Lett. [27] G. Gruner, Density Waves in Solids (Addison-Wesley, 97, 241911 (2010). [25] S. Ono and H. Shima, EPL 96, 27011 (2011). MA, 1994). 5 

1902.00345

1

1902

Magnetic Properties of Epitaxially-grown $SrRuO_3$ Nanodots

We present the fabrication and exploration of arrays of nanodots of $SrRuO_3$ with dot sizes between 500 nm and 15 nm. Down to the smallest dot size explored, the samples were found to be magnetic with a maximum of the Curie temperature $T_C$ achieved by dots of 30 nm diameter. This peak in $T_C$ is associated with a dot-size-induced relief of the epitaxial strain, as evidenced by scanning transmission electron microscopy.

cond-mat.mes-hall

cond-mat

Magnetic Properties of Epitaxially-grown SrRuO3 Nanodots G. Laskin,1, a) H. Wang,1, a) H. Boschker,1 W. Braun,1 V. Srot,1 P. A. van Aken,1 and J. Mannhart1, b) Max Planck Institute for Solid State Research, Heisenbergstr. 1, 70569 Stuttgart, Germany We present the fabrication and exploration of arrays of nanodots of SrRuO3 with dot sizes between 500 nm and 15 nm. Down to the smallest dot size explored, the samples were found to be magnetic with a maximum of the Curie temperature T C achieved by dots of 30 nm diameter. This peak in T C is associated with a dot-size-induced relief of the epitaxial strain, as evidenced by scanning transmission electron microscopy. 9 1 0 2 b e F 1 ] l l a h - s e m . t a m - d n o c [ 1 v 5 4 3 0 0 . 2 0 9 1 : v i X r a I. INTRODUCTION Nanostructures of complex functional materials provide an experimental platform to investigate novel physical phe- nomena if the characteristic device size reaches intrinsic length scales, such as the inelastic scattering length or the Fermi wavelength. For mean-field systems, especially semiconductors, confinement effects have been studied for decades and have already been implemented in widespread technological applications1,2. Recently, a significant num- ber of investigations have shown that such confinement of correlated systems in 2D3, e.g., at interfaces4 -- 6, in thin films7 or superlattices8, promises a variety of interesting advantages over their mean-field counterparts. Although novel electronic properties are expected9 if correlated sys- tems are confined in fewer dimensions, i.e., to 1D and 0D, such systems have been explored only marginally. This work focuses on the limiting case of spatial confine- ment into 0D objects, i.e., into quantum dots or artificial atoms. In Refs. 10 and 11 it was shown that such dots can be fabricated at the LaAlO3/SrTiO3 interface using surface- modification with an atomic force microscope (AFM) tip. Ion-beam based fabrication of magnetic SrRuO3 nanodots with diameters of about 80 nm has been reported in Ref. 12. In Ref. 13 the magnetization reversal in similar nanostruc- tures of various sizes was studied by magnetic force mi- croscopy. We here use the top-down approach based on epitaxial growth and electron-beam lithography that allows such objects to be fabricated with high precision and re- producibility. At the same time, the form and mutual ar- rangement of these objects may be chosen and varied as desired. The main challenge for fabricating such devices is to synthesize a highly pure material with low defect den- sity that along with small sizes allows the formation of a co- herent many-body electron wave function. However, it is well known that the patterning process causes local mate- rial damage and dead layers, which can significantly reduce the mean free path of the electrons14. In order to study the material behavior upon confine- 15 as a model system. ment in nanodots, we chose SrRuO3 SrRuO3 is a correlated conducting perovskite oxide with strong itinerant ferromagnetism (bulk Curie temperature T C ≈ 160 K) and an intensively studied perovskite material. It exhibits orthorhombic symmetry at room temperature. a)These two authors contributed equally b)corresponding author, [email protected] The ferromagnetic properties of SrRuO3 depend highly on the material quality as well as on the film thickness16,17. Even slight variations of the stoichiometry or lattice pa- rameters may lead to a significant change of magnetic properties18 -- 20. Hence, SrRuO3 is a model system to study the relevance of dead-layer creation and how materials be- have upon confinement within low-dimensional objects. We therefore prepared epitaxial SrRuO3 nanodots with vari- ous diameters and explored how the magnetic properties of the material depend on the size of the structures. II. EXPERIMENTAL TECHNIQUES For the sample preparation, special care was taken to de- posit SrRuO3 films of high purity and with the desired sto- ichiometry. Prior to SrRuO3 deposition, SrTiO3 substrates were terminated in situ in the growth chamber by thermal annealing at 1300 °C for 200 s at a molecular oxygen pres- sure of 0.08 mbar21. Immediately after that, and in the same UHV chamber, SrRuO3 thin films were grown by pulsed laser deposition with a target-substrate distance of 56 mm. The substrate temperature during growth was 680 °C, and the molecular oxygen pressure in the chamber was kept at 0.08 mbar. The SrRuO3 target was ablated using a KrF ex- cimer laser with a wavelength of 248 nm and an energy den- sity on the target of 2.5 J/cm2. The nanodots were patterned using a commercial 100 keV electron beam lithography sys- tem (JEOL JBX6300) and CSAR (Allresist, AR-P 6200) as resist material. After resist development, a 20-nm layer of amor- phous Al2O3 was deposited, which was then used as a hard mask for dry etching with Ar ions at an energy of 600 eV and a pressure of 1.4× 10 −4 mbar. For a pattern design, we used rectangular arrays of dots, where the dot periodicity equalled two dot diameters. Arrays covered the entire sub- strate area, excluding edge regions. For 20-nm dots, ≈ 109 dots were fabricated on each substrate. Scanning transmission electron microscopy (STEM) specimens in cross-sectional orientation were prepared by focused ion beam (Zeiss CrossBeam XB 1540) cutting, then nanomilled at liquid nitrogen temperature (Fischione NanoMill Model 1040). STEM investigations were car- ried out using a spherical aberration-corrected microscope (JEOL JEM-ARM 200F) with a DCOR probe corrector (CEOS GmbH) at 200 kV. 2 Figure 1. (a) AFM image of SrRuO3 dot arrays grown on SrTiO3 showing both the 1-unit-cell high substrate surface steps and the dot arrays. The height of the dots is approximately 3 nm. (b) Top view SEM image of a SrRuO3 dot array with a diameter of 20 nm and height of 12 nm. Figure 2. (a) Z-contrast STEM image of a cross-sectional cut through a single dot close to its center. (b) Magnified STEM image showing the interface (red arrow) between the dot and the sub- strate. III. RESULTS Figures 1 and 2 give an overview of the dots' microscopic properties. Figure 1(b) is a scanning electron microscopy (SEM) image of a sample with a rectangular array of 20- nm SrRuO3 dots. A Z-contrast STEM image of a single-dot cross section from a different sample is shown in Fig. 2(a), which depicts a clear contrast between the SrRuO3 and the substrate material with a sharp interface. This is magnified in Fig. 2(b). During the sample preparation for TEM, the dot was cut through its center with a lamella thickness of ≈ 20 nm. Therefore, the cross section shown in Fig. 2(a) pro- vides a projection through almost the entire dot. The curved shape of the substrate near the dot and the shape of the dot itself are caused by non-uniform removal of material during dry etching with Ar ions. The magnetic properties of different samples were mea- sured using a SQUID magnetometer (Quantum Design MPMS) equipped with the reciprocating sample option (RSO) head. For all samples, the thickness of the SrRuO3 film equaled 12 nm (i.e., 32 unit cells of SrRuO3). Every sam- ple was measured twice at different steps of the fabrication process: directly after the thin-film growth and then after being patterned into nanodots. To measure the zero-field- cooled and the field-cooled behavior of the magnetization, the samples were cooled to 4 K in zero magnetic field, after which an out-of-plane magnetic field of 0.1 T was applied. (a)(b)2 m200 nm100 nmSrRuO3SrTiO3(a)(b)10 nm2 nm3 Figure 3. Temperature-dependent magnetization of (a) an un- patterned 12-nm-thick SrRuO3 film and (b) an array of 20-nm- diameter dots patterned from the same film, measured with zero- field cooling (zfc) and field-cooling (fc) in an out-of-plane ori- ented magnetic field of 0.1 T. The black and red lines correspond to zero-field-cooled and field-cooled measurements, respectively. The diamagnetic contribution of the SrTiO3 substrate has been subtracted. The samples were then warmed to room temperature and cooled again. A typical characteristic of the temperature-dependent magnetization of a SrRuO3 film is shown in Fig. 3(a). The shape of the curve and the saturation magnetization of 1.5 µB/Ru are in good agreement with the literature data22,23. The Curie temperature T C is determined by the linear ex- trapolation of the transition curve to M = 0, although a finite magnetic moment exists above this temperature, attributed to the domain structure and film reorientation24. For all films, the transition has been found at a T C = 149±1 K, which is comparable to the T C values reported in the literature for equally thick films deposited by pulsed laser deposition8,17. Fig. 3(b) shows an example of the magnetic transition curve for a patterned sample with a dot diameter of 20 nm. Con- siderably different is the behavior of the magnetic response at low temperatures, where a significant increase of the magnetization is observed. This is due to the paramag- 25, which causes difficulties for netism inherent in SrTiO3 measuring the dot magnetization because of the very small SrRuO3 volume compared to the SrTiO3 substrate. However, this contributes only to the low-temperature properties and Figure 4. (a) Measured temperature dependence of the magnetiza- tion of several samples with different dot sizes in the temperature range close to T C. All curves are normalized by the magnetization of the film. (b) Curie temperature of the samples plotted as a func- tion of dot diameter. does not affect the magnetization close to the ferromagnetic transition region, in which we are primarily interested here. After patterning, the sample remains ferromagnetic with a T C = 153 K, i.e., 4 K higher than that of the film. This in- crease of the Curie temperature is systematically observed for dots of different sizes. To clarify this effect in more de- tail, a series of samples with different dot sizes (from 500 to 15 nm) was fabricated. The initial film morphology and their magnetic properties were similar. With decreasing dot size, the distance between them was scaled proportionally to keep the total amount of SrRuO3 constant. After pat- terning, the temperature dependence of the magnetization was measured for all samples using the procedure described above. The results of the measurements for dots of differ- ent sizes are depicted in Fig. 4(a) and the extracted vari- ation of the Curie temperature in Fig. 4(b). The T C does not change after patterning for dots equal to or larger than 500 nm. For smaller dots, a gradual increase of T C with de- creasing dot size is observed. This is valid down to a dot size of approximately 30 nm, after which T C starts to decrease, presumably due to surface defects and dead layers. A de- crease of T C with shrinking feature size is well known for thin films17,26 and has been observed in nanoparticles of 0501001502000.00.51.01.5M (μB/Ru)T (K)filmdots(a)(b)0.00.30.60.91.21.5 zfc fcM (μ B/Ru)1441481521561600.000.040.080.120.160.20 film 500 nm 180 nm 90 nm 40 nm 20 nmM/M0T (K) dots Ø:(a)(b)0100200300400500film148150152154156158TC (K)D (nm)4 Figure 5. Original Z-contrast STEM images are shown for (a) a SrRuO3 unpatterned thin film, (b) a 80-nm-sized dot and (c) a 30-nm-sized dot on (100) SrTiO3. The respective out-of-plane εyy (d,e,f) and in-plane εxx (g,h,i) components of the strain are obtained from STEM- based geometric phase analysis. For the 80-nm-sized dot, the central region of the dot (red rectangle) is analyzed. The data in the graphs is extracted from the horizontally averaged regions marked by the boxes. The 0% level corresponds to the lattice constant of bulk SrTiO3. other perovskites27 and also for metals28. This reduction of T C is generally attributed to the reduced number of neigh- bors and the corresponding destabilization of the magnetic ordering28. Even for the smallest dot sizes, however, the Curie temperature exceeds that of the unstructured film. Strain is one of the mechanisms known to influence mag- netic properties of SrRuO3. Using ab-initio density func- tional theory (DFT) calculations, it was shown in Ref. 29 that uniaxial and epitaxial strain in SrRuO3 significantly alters its magnetic properties. Experimental studies of SrRuO3 grown on different substrates19 demonstrate that the mag- netic properties differ depending on the amount of strain in the film. Furthermore, SrRuO3 films released from their substrates demonstrate a 10 K higher T C of 160 K30. A dif- ferent study20 showed that a SrTiO3 capping layer grown on SrRuO3 epitaxial films deposited on DyScO3 modifies the oxygen octahedral structure in the SrRuO3 with a cor- responding increase of the Curie temperature. To explore whether strain relief is responsible for the T C increase in our dot structures, we studied the strain in the dots by STEM. We used geometric phase analysis to analyze high spatial resolution STEM images in the in-plane and out-of-plane directions31,32. The lattice constant of the SrTiO3 substrate was taken as a reference level. Fig. 5(d,g) shows the strain distribution of a SrRuO3 thin film grown on SrTiO3. The data reveals that the film is fully strained, i.e., the lattice con- stant in the in-plane direction remains unchanged across the interface (Fig. 5(g), right panel), whereas in the out-of- plane direction the unit cell of SrRuO3 is elongated by a constant value of ≈1.4±0.3 % compared to SrTiO3 (Fig. 5(d), right panel). These strain characteristics are altered by patterning the epitaxial film into nanodots, as shown by Fig. 5(e,h) for a 80-nm-diameter dot and Fig. 5(f,i) for a 30-nm-diameter dot. For both dot sizes, in the in-plane direction (Fig. 5(h,i)), starting from the interface, the lat- tice constant of SrRuO3 increases gradually, which indicates strain relaxation. For the 80-nm dot, the lattice constant has relaxed by 0.35±0.10 % at the top of the dot, whereas for the 30-nm dot the relaxation is even stronger. It reaches 0.75±0.10 %, matching fully relaxed SrRuO3. We link this lattice relaxation to the observed increase of the Curie tem- perature. In the out-of-plane direction (Fig. 5(e,f)) the be- havior for dots of both sizes is similar. Here, in the first two unit cells close to the interface, we observed the strain close to that of the film, which, as expected, then decays to- gether with the lateral strain relaxation to approach a three- dimensionally relaxed structure. A similar mechanism of patterning-induced strain relaxation has been observed be- 50-5-10nmεxx (%)(d)(g)5 nmyx-55 (%)compressivetensile strain0.00-0.500.50-0.250.25-1012350-5-10nmεyy (%)nm-55 (%)compressivetensile strainεxx (%)εyy (%)(e)(h)yxnmnm-55 (%)compressivetensile strainεxx (%)εyy (%)(f)(i)10 nmyx50-5-100.00.51.0-1012350-5-10nm5 nm50-5-10-101230.000.250.5050-5-1010 nm30 nm5 nm(a)(b)(c)unpatterned film80 nm dot30 nm dotout-of-planestrainin-planestrainfore in patterned semiconductor nanostructures, for exam- ple Si33, InGaN/GaN34 or (Ga, Mn)As35. We therefore con- clude that the strain relaxation is responsible for the in- crease of the T C in these SrRuO3 nanodots. 12D. Ruzmetov, Y. Seo, L. J. Belenky, D.-M. Kim, X. Ke, H. Sun, V. Chan- drasekhar, C.-B. Eom, M. S. Rzchowski, and X. Pan, Adv. Mater. 17, 2869 (2005). 13L. Landau, J. W. Reiner, and L. Klein, J. Appl. Phys. 111, 07B901 (2012). 14J. Katine, M. Ho, Y. Ju, and C. Rettner, Appl. Phys. Lett. 83, 401 (2003). 15G. Koster, L. Klein, W. Siemons, G. Rijnders, J. S. Dodge, C.-B. Eom, 5 D. H. A. Blank, and M. R. Beasley, Rev. Mod. Phys. 84, 253 (2012). 16K. Ishigami, K. Yoshimatsu, D. Toyota, M. Takizawa, T. Yoshida, G. Shi- bata, T. Harano, Y. Takahashi, T. Kadono, V. K. Verma, V. R. Singh, Y. Takeda, T. Okane, Y. Saitoh, H. Yamagami, T. Koide, M. Oshima, H. Ku- migashira, and A. Fujimori, Phys. Rev. B 92, 064402 (2015). 17Y. J. Chang, C. H. Kim, S.-H. Phark, Y. S. Kim, J. Yu, and T. W. Noh, Phys. Rev. Lett. 103, 057201 (2009). 18W. Siemons, G. Koster, A. Vailionis, H. Yamamoto, D. H. A. Blank, and M. R. Beasley, Phys. Rev. B 76, 075126 (2007). 19W. Lu, W. Song, P. Yang, J. Ding, G. M. Chow, and J. Chen, Sci. Rep. 5, 10245 (2015). 20S. Thomas, B. Kuiper, J. Hu, J. Smit, Z. Liao, Z. Zhong, G. Rijnders, A. Vail- ionis, R. Wu, G. Koster, and J. Xia, Phys. Rev. Lett. 119, 177203 (2017). 21M. Jäger, A. Teker, J. Mannhart, and W. Braun, Appl. Phys. Lett. 112, 116601 (2018). 22W. Tian, J. H. Haeni, D. G. Schlom, E. Hutchinson, B. L. Sheu, M. M. Rosario, P. Schiffer, Y. Liu, M. A. Zurbuchen, and X. Q. Pan, Appl. Phys. Lett. 90, 022507 (2007). 23A. Grutter, F. Wong, E. Arenholz, M. Liberati, A. Vailionis, and Y. Suzuki, Appl. Phys. Lett. 96, 082509 (2010). 24J. Xia, W. Siemons, G. Koster, M. R. Beasley, and A. Kapitulnik, Phys. Rev. B 79, 140407 (2009). 25J. M. D. Coey, M. Venkatesan, and P. Stamenov, J. Phys.: Condens. Matter 28, 485001 (2016). 26F. J. Himpsel, J. E. Ortega, G. J. Mankey, and R. F. Willis, Adv. Phys. 47, 511 (1998). 27S. Vasseur, E. Duguet, J. Portier, G. Goglio, S. Mornet, E. Hadova, K. Knizek, M. Marysko, P. Veverka, and E. Pollert, J. Magn. Magn. Mater. 302, 315 (2006). 28C. Rong, D. Li, V. Nandwana, N. Poudyal, Y. Ding, Z. L. Wang, H. Zeng, and J. P. Liu, Adv. Mater. 18, 2984 (2006). 29A. T. Zayak, X. Huang, J. B. Neaton, and K. M. Rabe, Phys. Rev. B 77, 214410 (2008). 30Q. Gan, R. Rao, C. Eom, J. Garrett, and M. Lee, Appl. Phys. Lett. 72, 978 (1998). 31M. J. Hytch, E. Snoeck, and R. Kilaas, Ultramicroscopy 74, 131 (1998). 32M. J. Hytch, J.-L. Purtaux, and J.-M. Penisson, Nature 423, 270 (2003). 33C. Himcinschi, R. Singh, I. Radu, A. P. Milenin, W. Erfurth, M. Reiche, U. Gösele, S. H. Christiansen, F. Muster, and M. Petzold, Appl. Phys. Lett. 90, 021902 (2007). 34V. Ramesh, A. Kikuchi, K. Kishino, M. Funato, and Y. Kawakami, J. Appl. Phys. 107, 114303 (2010). 35J. Wenisch, C. Gould, L. Ebel, J. Storz, K. Pappert, M. J. Schmidt, C. Kumpf, G. Schmidt, K. Brunner, and L. W. Molenkamp, Phys. Rev. Lett. 99, 077201 (2007). IV. SUMMARY In summary, using e-beam lithography of epitaxial SrRuO3 films, we have succeeded in fabricating nanosized dots of SrRuO3 ranging in size from several hundred nm down to 15 nm. SrRuO3 shows ferromagnetism even in the smallest dots. Unexpectedly, magnetometry reveals that the magnetic properties are enhanced for dot sizes below 500 nm, which manifests itself in an increase of the Curie temperature for small dots. We demonstrate that the Curie temperature depends on the size of the dots, increasing gradually from 149 K in the thin film and dots larger than 500 nm to 157 K at 30 nm dot size. This behavior is at- tributed to strain relaxation in the material caused by the removal of lateral constraint around the dot. V. ACKNOWLEDGEMENTS GL acknowledges E. Goering (MPI for Intelligent Systems, dept. Modern Magnetic Systems) and E. Bruecher (MPI for Solid State Research, Chemical Service) for the support with magnetometry measurements. We thank T. Reindl, U. Waizmann and J. Weis for the technical support with sam- ple fabrication. Discussions with J. Mydosh and M. Ternes are gratefully appreciated. HW gratefully acknowledges the China Scholarship Council (No. 201504910813) and the Max Planck Society for financial support. We thank Y. Wang, U. Salzberger, K. Hahn and P. Kopold for their support during TEM sample preparation and the TEM experiments. We acknowledge the assistance by B. Fenk for preparing FIB lamellas and by L. Jin from the Ernst Ruska-Centre Juelich for the nanomill preparation of TEM lamellas. REFERENCES 1R. Ashoori, Nature 379, 413 (1996). 2M. Kastner, Phys. Today 46, 24 (1993). 3M. Bibes, J. E. Villegas, and A. Barthelemy, Adv. Phys. 60, 5 (2011). 4J. Mannhart and D. G. Schlom, Science 327, 1607 (2010). 5H. Hwang, Y. Iwasa, M. Kawasaki, B. Keimer, N. Nagaosa, and Y. Tokura, Nat. Mater. 11, 103 (2012). 6F. Baiutti, G. Logvenov, G. Gregori, G. Cristiani, Y. Wang, W. Sigle, P. A. van Aken, and J. Maier, Nat. Commun. 6, 8586 (2015). 7K. Yoshimatsu, K. Horiba, H. Kumigashira, T. Yoshida, A. Fujimori, and M. Oshima, Science 333, 319 (2011). 8F. Bern, M. Ziese, A. Setzer, E. Pippel, D. Hesse, and I. Vrejoiu, J. Phys.: Condens. Matter 25, 496003 (2013). 9J. Mannhart, H. Boschker, T. Kopp, and R. Valenti, Rep. Prog. Phys. 79, 084508 (2016). 10C. Cen, S. Thiel, J. Mannhart, and J. Levy, Science 323, 1026 (2009). 11G. Cheng, P. F. Siles, F. Bi, C. Cen, D. F. Bogorin, C. W. Bark, C. M. Folkman, J.-W. Park, C.-B. Eom, G. Medeiros-Ribeiro, and J. Levy, Nat. Nanotech- nol. 6, 343 (2011). 

1911.03523

1

1911

Topological Nanospaser

We propose a nanospaser made of an achiral plasmonic-metal nanodisk and a two-dimensional chiral gain medium -- a monolayer transition-metal dichalcogenide (TMDC). When one valley of the TMDC is selectively pumped (e.g., by a circular-polarized radiation), the spaser generates a mode carrying a topological charge (chirality) that matches that of the gain valley. The chirally mismatched, time-reversed mode has exactly the same frequency but the opposite topological charge; it is actively suppressed by the gain saturation (population clamping) and never generates leading to a strong topological protection for the generating matched mode. This topological spaser is promising for the use in nanooptics and nanospectroscopy in the near-field especially in applications to biomolecules that are typically chiral. Another potential application is a chiral nanolabel for biomedical applications emitting in the far field an intense circularly-polarized coherent radiation.

cond-mat.mes-hall

cond-mat

Topological Nanospaser Rupesh Ghimire,∗ Jhih-Sheng Wu,† Vadym Apalkov,‡ and Mark I. Stockman§ Center for Nano-Optics (CeNO) and Department of Physics and Astronomy, Georgia State University, Atlanta, Georgia 30303 (Dated: November 12, 2019) We propose a nanospaser made of an achiral plasmonic-metal nanodisk and a two-dimensional chiral gain medium -- a monolayer transition-metal dichalcogenide (TMDC). When one valley of the TMDC is selectively pumped (e.g., by a circular-polarized radiation), the spaser generates a mode carrying a topological charge (chirality) that matches that of the gain valley. The chirally- mismatched, time-reversed mode has exactly the same frequency but the opposite topological charge; it is actively suppressed by the gain saturation (population clamping) and never generates leading to a strong topological protection for the generating matched mode. This topological spaser is promising for the use in nanooptics and nanospectroscopy in the near-field especially in applications to biomolecules that are typically chiral. Another potential application is a chiral nanolabel for biomedical applications emitting in the far field an intense circularly-polarized coherent radiation. I. INTRODUCTION Spaser (surface plasmon amplificaiton by stimulated emission of radiation) was originally introduced in 2003 [1] as a nanoscopic phenomenon and device: a genera- tor and amplifier of coherent nanolocalized optical fields. Since then, the science and technology of spasers expe- rienced a rapid progress. Theoretical developments [2 -- 6] were followed by the first experimental observations of the spaser [7, 8] and then by an avalanch of new devel- opments, designs, and applications. Currently there are spasers whose generation spans the entire optical spec- trum, from the near-infrared to the near-ultraviolet [9 -- 17]. Several types of spasers, which are synonymously called also nanolasers, have so far been well developed. Historically, the first is a nanoshell spaser [7] that con- tains a metal nanosphere as the plasmonic core that is surrounded by a dielectric shell containing gain material, typically dye molecules [7, 18]. Such spasers are small- est coherent generators produced so far, with sizes in the range of tens nanometers. Almost simultaneously, an- other type of nanolasers was demonstrated [8] that was built from a semiconductor gain nanorod situated over a surface of a plasmonic metal. It has a micrometer-scale size along the nanorod. Its modes are surface plasmon polaritons (SPPs) with nanometer-scale transverse size. Given that the spasers of this type are relatively efficient sources of far-field light, they are traditionally called nanolasers though an appropriate name would be polari- tonic spasers. Later, this type of nanolasers (polaritonic spasers) was widely developed and perfected [9, 14, 19 -- 22]. There are also spasers that are similar in design to the polaritonic nanolasers but are true nanospasers ∗Electronic address: [email protected] †Electronic address: [email protected] ‡Electronic address: [email protected] §Electronic address: [email protected] whose dimensions are all on the nanoscale. Such a spaser consists of a monocrystal nanorod of a semiconductor gain material deposited atop of a monocrystal nanofilm of a plasmonic metal [23]. These spasers possess very low thresholds and are tunable in all visible spectrum by changing the gain semiconductor composition while the geometry remains fixed [13, 24, 25]. There are also other types of demonstrated spasers. Among them we men- tion semiconductor-metal nanolasers [26] and polaritonic spasers with plasmonic cavities and quantum dot gain media [27]. A fundamentally different type of quantum generators is the lasing spaser [6, 28, 29]. A lasing spaser is a peri- odic array of individual spasers that interact in the near field and form a coherent collective mode. Such lasing spasers have been built of plasmonic crystals that incor- porate gain media. One type of the lasing spasers is a pe- riodic array of holes in a plasmonic metal film deposited on a semiconductor gain medium [12]; another type is a periodic array of metal nanoparticles surrounded by a dye molecules solution [30]. We have recently proposed a topological lasing spaser that is built of a honeycomb plasmonic crystal of silver nanoshells containing a gain medium inside [31]. The generating modes of such a spaser are chiral surface plsmons (SPs) with topological charges of m = ±1, which topologically protects them against mixing. Only one of the m = ±1 topologically- charged modes can generate at a time selected by a spon- taneous breaking of the time-reversal symmetry. The spasers are not only of a significant fundamental interest but also are promising for applications based on their nanoscale-size modes and high local fields. Among such demonstrated applications are those to sensing of minute amounts of chemical and biological agents in the environment [20, 21, 32]. Another class of the demon- strated applications of the spasers is that in cancer thera- nostics (therapeutics and diagnostics) [18]. An important perspective application of spasers is on-chip communica- tions in optoelectronic information processing [33]. It is of a great interest to explore intersections of In our the spaser technology and topological physics. 2 II. SPASER STRUCTURE AND MAIN EQUATIONS The geometry and the fundamentals of functioning of the proposed topological nanospaser is illustrated in Fig. 1. This spaser consists of a thin silver nano-spheroid placed atop of the two-dimensional (2d) gain medium (a nanodisk of a monolayer TMDC) -- see Fig. 1(a). As panel (b) illustrates, the gain medium is pumped with circularly-polarized light, which is known to selectively populate one of the K or K(cid:48) valleys depending on its helicity [42, 43]. Due to the axial symmetry, the plas- monic eigenmodes, φ(r), depend on the azimuthal angle, ϕ: φm(r) ∝ exp (imϕ), where m = const is the magnetic quantum number. Figure 1(b) illustrates that the con- duction band (CB) to valence band (VB) transitions in the TMDC couple predominantly to the SPs whose chi- rality matches that of the valley: the transitions in K- or K(cid:48)-valley excite the m = 1 or m = −1 SPs, respectively. The surface plasmon eigenmodes φn(r) are described by the quasistatic equation [44] ∇Θ(r)∇φn(r) = sn∇2φn(r), (1) where n is a set of the quantum numbers defining the eigenmode, sn is the corresponding eigenvalue, which is a real number between 0 and 1, and Θ(r) is the char- acteristic function, which is equal to 1 inside the metal and 0 outside. We assume that the metal nanoparticle is a spheroid whose eigenmodes can be found in oblate spheroidal coordinates [45] -- see Supplemental Materials (SM). They are characterized by two integer spheroidal quantum numbers: multipolarity l = 1, 2, . . . and az- imuthal or magnetic quantum number m = 0,±1, . . . . We will consider a dipolar mode, l = 1 where m = 0,±1. Note that the dipole transitions in the TMDC at the K-, K(cid:48)-points are chiral, and they couple only to the modes with m = ±1. The Hamiltonian of the SPs is HSP = ωsp a† mam, (2) (cid:88) m=±1 where ωsp is the SP frequency, and a† m and am are the SP creation and annihilation operators (we indicate only the magnetic quantum number m). The electric field operator is [1, 3] Fm(r, t) = −Asp∇φm(r)(ame−iωspt + a† meiωspt), (cid:115) Asp = 4πssp
ds(cid:48) sp , (3) (4) sp = Re[ds(ω)/dωω=ωsp ]. The monolayer TMDC where s(cid:48) is coupled to the field of the SPs via the dipole inter- action. We choose the proper thickness of the silver spheroid so that the SP energy ωsp is equal to the band gap of the TMDC gain medium. The Hamiltonian of the FIG. 1: Schematic of the spaser geometry and operation. (a) Geometry of the spaser: Silver nanospheroid on TMDC. (b) Schematic of spaser operation. Pumping with a circular- polarized light excites the valley whose chirality is matched to the light helicity. The stimulated CB→CB transitions at the corresponding K or K(cid:48) point excite SPs matched by chirality to that of the valley; the other, mismatched valley couples only weakly to these SPs. recently proposed topological lasing spaser [31], the topologically-charged eigenmodes stem from the Berry curvature [34, 35] of the plasmonic Bloch bands of a hon- eycomb plasmonic crystal of silver nanoshells. In con- trast, the gain medium inside these nanoshells is com- pletely achiral. This topological lasing spaser is predicted to generate a pair of mutually time-reversed eigenmodes carrying topological charges of ±1, which strongly com- pete with each other, so only one of them can be gener- ated at a time. In this Article we propose a topological nanospaser that also generates a pair of mutually time-reversed chi- ral SP eigenmodes with topological charges of ±1, whose fields are rotating in time in the opposite directions. In a contrast to Ref. 31, this proposed spaser is truly nanoscopic, with a radius ∼ 10 nm. The topological charges (chiralities) of its eigenmodes originate from the Berry curvature of the gain-medium Bloch bands. This gain medium is a two-dimensional honeycomb nanocrys- tal of a transition metal dichalcogenide (TMDC) [36 -- 38]. The plasmonic subsystem is an achiral nanodisk of a plasmonic metal. Note that previously the TMDCs have been used as the gain media of microlasers where the cavities were formed by microdisk resonators [39, 40] or a photonic crystal microcavity [41]. None of these lasers generated a chiral, topologically charged mode. The interaction between the monolayer TMDC and the SPs is described by an interaction Hamiltonian Hint = − (cid:88) (cid:90) (cid:88) K=K,K(cid:48) νK d2r Fm(r)dK(r) , (8) γ2K(r) = m=±1 (cid:88) (cid:12)(cid:12)(cid:12)2 (cid:12)(cid:12)(cid:12) Ωm,K(r) Following Ref. 3, the equations of motion of the SPs and the monolayer TMDC electron density matrix are am = [i(ω − ωsp) − γsp]am+ ρ∗ K(r) Ω∗ iνK d2r (cid:88) (cid:90) m,K(r) , (cid:88) S K (cid:104) (cid:105) nK(r) = − 4 Im ρK(r) Ωm,K(r)am + gK [1 − nK(r)] − γ2K(r) [1 + nK(r)] , m=1,−1 3 (13) (14) ρK(r) = [−i(ω − ∆g)−Γ12]ρK(r)+ (cid:88) inK(r) m=1,−1 Ω∗ m,Ka∗ m , (15) where S is the entire area of the TMDC, ωsp is the SP frequency, γsp is the SP relaxation rate, Γ12 is the polar- ization relaxation rate for the spasing transition 2 → 1, gK is the pumping rate in valley K, the population in- version, nK, is defined as nK ≡ ρ(c)K − ρ(v)K , (16) and the spontaneous emission rate of the SPs is [3] 2(γsp + Γ12) (ωsp + ∆g)2 + (γsp + Γ12)2 m=1,−1 (17) III. RESULTS AND DISCUSSION A. Parameters of Spaser and Chiral Coupling to Gain Medium We consider a spaser consisting of an oblate silver spheroid with semi-major axis a = 12 nm placed atop of a circular TMDC flake of the same radius. We as- sume that the system is embedded into a dielectric ma- trix with permittivity
d = 2. We choose the value of the semi-minor axis c (the height of the silver spheroid) to fit ωsp to the K-point CV→VB transition frequency in the TMDC, ωsp = ∆g. We employ the three-band tight- binding model for monolayers of group-VIB TMDCs of Ref. 49. We also set Γ12 = 10 meV. From the tight-binding model, we calculate the band including band gap ∆g and the transition structure, dipole matrix element d. Note that at the K- and K(cid:48)- points, the band gaps are the same, ∆g (K) = ∆g (K(cid:48)), while the transition dipole matrix elements are complex K(cid:48), as protected by the T -symmetry. conjugated, dK = d∗ The values used in the computations are listed in the SM. Here we give an example for MoS2: c = 1.2 nm; ∆g = 1.66 eV; dK = 17.7 e+ D, and dK(cid:48) = 17.7 e− D, √ where e± = (ex ± iey) / 2 are chiral unit vectors. TMDC near the K or K(cid:48) point can be written as HK = d2q Eα(K + q)α,K + q(cid:105)(cid:104)α,K + q, (5) (cid:90) (cid:88) α=v,c where K = K or K(cid:48), and v and c stand for the valence band and the conduction band, correspondingly. We ex- pand the Hamiltonian around the K and K(cid:48) points as HK (cid:39) νK Eα(K)α,K(cid:105)(cid:104)α,K, (6) (cid:88) α=c, v where νK is the density of electronic states in the K valley, which we adopt from experimental data [39, 46]: νK = νK(cid:48) = 7.0 × 1012 cm−2. The field of the SPs in nanoparticles is highly nonuni- form in space, which gives rise to a spatial non-uniformity of the electron population of the TMDC monolayer. To treat this, we employ a semiclassical approach where the state α,K, r(cid:105) represents a electron in the K valley at position r. The corresponding Hamiltonian in the semi- classical approximation can be written as HK = νK Eα(K) d2rα,K, r(cid:105)(cid:104)α,K, r (7) (cid:90) (cid:88) α=c, v where the dipole operator is given by dK(r) = dKei∆gtc,K, r(cid:105)(cid:104)v,K, r + h.c. , (9) where ∆g is the band gap (at the K- or K(cid:48)-point). The transition dipole element, dK, is related to the non-Abelian (interband) Berry connection A(cv)(k) as dK = eA(cv)(k) , A(cv)(k) = i uc(k) (cid:28) (cid:12)(cid:12)(cid:12)(cid:12) ∂ ∂k (cid:12)(cid:12)(cid:12)(cid:12) uv(k) (cid:29)(cid:12)(cid:12)(cid:12)(cid:12)k=K , (10) where uα(k) are the normalized lattice-periodic Bloch functions. In this Article, we consider the dynamics of the sys- tem semiclassically: we treat the SP annihilation and creation operators as complex c-numbers, am = am and a† m = a∗ m, and describe the electron dynamics quantum mechanically by density matrix ρK(r, t). Furthermore, we assume that the SP field amplitude is not too large, Ωm,K (cid:28) ∆g, where the Rabi frequency is defined by Ωm,K(r) = − 1  Asp∇φm(r)d∗ K . (11) Then we can employ the rotating wave approximation (RWA) [47, 48] where the density matrix can be written as ρK(r, t) = ρ(c)K (r, t) ρ∗ K(r, t)e−iωt ρK(r, t)eiωt ρ(v)K (r, t) . (12) (cid:32) (cid:33) 4 virtually no coupling to the TMDC transitions in the K valley. Thus, for the gain medium whose radius Rg is within the footprint of the metal spheroid, i.e., for Rg ≤ a, only one mode with chirality m = 1 will be generated. However, for a larger gain medium, Rg ≥ 13 nm, there is a circle of strong coupling of the mismatched mode, which can potentially go into the generation. B. Kinetics of Continuous-Wave Spasing Below in this Article, we provide numerical examples of the spaser kinetics. For certainty, we assume that the K-valley is selectively pumped, which can be done with the right-hand circularly polarized pump radiation. (As protected by the T -symmetry, exactly the same results are valid for the left-handed pump and K(cid:48)-valley.) Thus, we set gK = g and gK(cid:48) = 0. A continuous wave (CW) solution can be obtained by solving Eqs. (13)-(15) where the time derivatives in the left-hand sides are set to zero. The calculated dependences of the generated coherent SP population, Nm = am2 where m = ±1, on the pumping rate, g, for various TMDC's are shown in Fig. 3(a). As we can see, there is a single spasing threshold for each of the TMDCs. Significantly above the threshold, for gK > 30 ps−1, the number of SPs, Nm, grows linearly with pumping rate g. This is a common general property of all spasers: it stems from the fact that the feedback in the spasers is very strong due to the extremely small modal volume. Therefore, the stimulated emission dominates the elec- tronic transitions between the spasing levels, which is the prerequisite of the linear line Nm(g). The slope of this straight line (the so-called slope efficiency) is specific for every given TMDC. For all these spasers, the threshold condition of Eq. (18) for the generation of the matched mode is satisfied. We have verified that the mismatched mode (m = −1) does not have a finite threshold, i.e., it is not generated at any pumping rate. The reason is that the matched mode (m = 1) above its threshold clamps the inversion at a constant level [3] preventing its increase with the pumping and, thus, precluding the generation of the mismatched mode. In this case, the single chiral mode generation enjoys a strong topological protection. At the threshold, the spasing curves experience a bi- furcation behavior. This is clearly seen in the magnified plot in Fig. 3(b): there is the threshold as the bifurca- tion point and two branches of the spasing curve above it. As we see from Fig. 3(c), these two branches differ by the stationary values of population inversion nK: for the upper branch it is significantly lower than for the lower branch. To answer a question whether these two branches are stable, we slightly perturb the accurate nu- merical solutions at g = 25 ps−1 by changing the number of SPs by ∆Nm = 0.0001. The density matrix solution for the dynamics of the SP population induced by such a perturbation is shown in Fig. 3(d). As we see, the upper branch is absolutely stable but the lower branch is un- FIG. 2: Coupling amplitude between SPs and TMDC dipole m , for m = ±1 as indicated. The magnitude is transitions, I (+) color coded by the bar to the right. The radius of the metal spheroid is a = 12 nm. A fundamental question regarding any spaser is the existence of a finite spasing threshold. There are two modes with the opposite chiralities, m = ±1, and identi- cal frequencies, ωsp, which are time-reversed with respect to each other, whose wave functions are ∇φ ∝ e±iϕ. In the center of the TMDC patch, i.e., at r = 0, the point symmetry group of a metal nanospheroid on the TMDC is C3v. It contains a C3 symmetry operation, i.e., a rota- tion in the TMDC plane by an angle ϕ = ±2π/3, which brings about a chiral selection rule m = 1 for the K-point and m = −1 for the K(cid:48)-point, i.e., the chirality of the SPs matches that of the valley. For eccentric positions, which are not too far from r = 0, it is not exactly the case but still there is a preference for the chirally-matched SPs. We assume that the pumping is performed with the circularly polarized radiation, and one of the valleys, say the K valley, is predominantly populated. Consequently, the first mode that can go into generation is the m = 1 SP. To find the necessary condition that the correspond- ing threshold can be achieved, we will follow Ref. 3 and set nK = 1. Then from Eqs. (13) and (15) we obtain this condition as νK m (r) d2r ≥ 1 , (18) where ± is the chirality of the pumped TMDC valley, and the coupling amplitude is (cid:90) S (cid:12)(cid:12)(cid:12)I (±) (cid:12)(cid:12)(cid:12)2 (cid:112)γspΓ12 AmdK I (±) m (r) = ∇φm(r)e± . (19) Note that I (±)∗ m = I (∓)−m. This coupling amplitude is illustrated in Fig. 2 for I (+) m . As we see, for the chiral-matched SP with m = 1, the m ≈ coupling amplitude is approximately constant, I (+) 1 within the geometric footprint of the silver spheroid, which is seen in panel (a) as an almost uniform orange disk with radius a = 12 nm. In a sharp contrast, the chiral-mismatched mode with m = −1 [panel (b)] has 5 FIG. 3: Spaser kinetics. (a) Dependence of the number of SP quanta in the spasing mode on the pumping rate for gain medium of the matched radius, Rg = a = 12 nm. Only the chirality-matched SP with m = 1 are generated. (b) Magni- fied near-threshold portion of panel (a) for MoS2. The num- ber of the SPs, Nm, is indicated for the points shown on the graphs for the two branches. (c) Radial distribution of the in- version, nk for each of the two branches. (d) Test of stability of the two SP branches. The kinetics of the SP population, Nm, after the number of the SPs in each branch is increased by ∆Nm = 0.0001. stable, and it evolves in time towards the upper branch within less than half a picosecond. As a result of this bifurcation instability, the system actually evolves with the increase of pumping along a path indicated by ar- rows in Fig. 3(b): Below the threshold, the population of the coherent SPs Nm = 0; it jumps to the apex of the curve at the bifurcation point and then follows the upper branch. One can state that the spatial inhomogeneity of the field and the inversion cause the spasing transition to become the first order. This is in contrast to the previ- ous homogeneous case of Ref. 3 where this transition was continuous, i.e., of the second order. The chiral optical fields generated by the topological spaser are not stationary -- they evolve in time rotating clockwise for m = 1, as illustrated in Fig. 4, and coun- terclockwise for m = −1. The magnitude of the field is large even for one SP per mode, E ∼ 107 V/A, which is a general property of the nanospasers related to the the SP population, the field increases as E ∝ √ nanoscopic size of the mode. Note that with increase of Nm. FIG. 4: Temporal dynamics of the local electric field, E, in topological spaser generating in the m = 1 mode. The curved arrow indicates the rotation direction of the field (clockwise). The magnitude of the field is calculated for a single SP per mode, Nm = 1; it is color-coded by the bar to the right. The phase of the spaser oscillation is indicated at the top of the corresponding panels. FIG. 5: Number of SPs Nm as a function of time t for a spaser with MoS2 as a gain material. The pumping is performed by a radiation whose electric field rotates clockwise in the plane of system (m = 1). The solid lines denote the chiral SPs with m = −1, and the dashed lines denote the SPs with m = 1. The pumping rates are indicated in the panels. (a) Dependence of SP number Nm on time t after the beginning of the pumping for different initial SP populations (color coded as indicated) for pumping rate g = 50 ps−1. (b) Dependence of SP number Nm on time t for different pumping rates g (color coded). The initial SP number is Nm = 10. C. Stability and Topological Protection of Spaser Modes In Fig. 5(a), we test the stability and topological pro- tection of the spasing mode. Panel (a) displays the dy- namics of the SP population of the topological spaser, Nm(t), for different initial numbers of SPs, Nm(0), and for their different chiralities, m = ±1. As these data show, the left-rotating SPs (m = −1) are not amplified irrespectively of their initial numbers: the corresponding curves evolve with decaying relaxation oscillations tend- ing to N−1 = 0. In contrast, the m = 1 SPs exhibit -1Initial Plasmonst (ps)0 0.1 0.2 0.3 0.4 0.5 0.6 0 0.1 0.2 0.3 0.4 0.5 0.6 (a)Nm 3025201510591625g=60 psg=50 psg=40 psg=30 ps252015105t (ps)-1-1-1-1g=50 ps(b)Nm a stable amplification: their number increases to a level that is defined by the pumping rate, g, and does not de- pend on the initial populations. The m = 1 chirality SP-amplification stability with respect to the injection of the m = −1 quanta, which these data demonstrate is due to the topological protection: matching the phase windings of the SP mode and the electronic states in the pumped K-valley. As a complementary test, we show in Fig. 5(b) the temporal dynamics of the SP population for equal initial number of SPs but different pumping rates. The dy- namics in this case is again stable with the mismatched m = −1 SPs decaying to zero, and the matched m = 1 being amplified to the stable levels that linearly increase with the pumping rate. D. Far-Field Radiation of Spaser The spaser is a subwavelength device design to gener- ate intense, coherent nanolocalized fields. Generation of far-field radiation is not its primary purpose. However, the proposed spaser, as most existing nanospasers, gen- erates in a dipolar mode that will emit in the far field. This emission, in absolute terms, can be quite intense for a nanosource. In particular, the spaser emission was used to detect cancer cells in the blood flow model [18]; it was, actually, many orders of magnitude brighter than from any other label for biomedical detection. To describe the spaser emission, we note that the radi- ating dipole uniformly rotates with the angular velocity of ωsp. The emitted radiation will be right-hand circu- larly polarized for the pumping at the K point and left- hand circularly polarized for the K(cid:48) pumping. Note that the corresponding two radiating modes are completely uncoupled. This is equivalent to having two independent chiral spasers in one. To find the intensity, I, of the emitted radiation, we need to calculated the radiating dipole. To do so, we will follow Ref. 50. We take into account that the modal field, Em = ∇φm, inside the metal spheroid is constant. Then from Eq. (8) of the SM, we can find E2 m = ssp Vm , (20) where Vm is the spheroid's volume. The physical field squared inside the metal is found from Eqs. (3) and (4), F 2 m = 4πs2
ds(cid:48) spNm spVm . (21) From this, we find the radiating dipole squared as (cid:18) (cid:19)−1 d0p2 =  4π Re ∂
m(ωsp) ∂ωsp Re [
m(ωsp) −
d]2 VmNm . (22) 6 The dipole radiation rate (photons per second) can be found from a standard dipole-radiation formula [51] as (cid:18) ω (cid:19)3 c0 I = 4 9 (cid:18) Re ∂
m(ωsp) ∂ωsp (
d)1/2 Re [
m(ωsp) −
d]2 × (cid:19)−1 a2cNm , (23) where c0 is speed of light in vacuum. For our example of MoS2, substituting parameters that we used everywhere in our calculations [see Sec. III A], we obtain I = 2.1×1012Nm s−1; P = ωspI = 0.55Nm µW , (24) where P is the power of the emission. From these num- bers, we conclude that the emission is bright for a nano- emitter and easily detectable. This is in line with the observation of the emission from single spasers of the comparable size in Ref. 18. IV. CONCLUDING DISCUSSION In this Article we introduce a topological nanospaser that consists of a plasmonic metal spheroid as the SP resonator and a nanoflake of a semiconductor TMDC as a gain medium. This spaser has two mutually T -reversed dipole modes with identical frequencies but opposite chi- ralities (topological charges m = ±1). Only the mode whose chirality matches that of the active (pumped) val- ley, i.e., m = 1 for the K-valley and m = −1 for the K(cid:48) valley, can be generated while the conjugated mode does not go into generation at any pumping level. The topological spaser is stable with respect to even large per- turbations: the SP with mismatched topological charge injected into the system even in large numbers decay ex- ponentially within a ∼ 100 fs time. This implies a strong topological protection. This protection is not trivial because the exact val- ley selection rule matching its chirality to the of the SPs is strictly valid only on the symmetry axis of the metal spheroid (in the center of the TMDC gain medium flake). Off-axis, there is a coupling of the gain to the chirally-mismatched SPs. However, the strong topologi- cal protection appears due to the fact that the spaser is a highly-nonlinear, threshold phenomenon. In fact, it is the nonlinear saturation of the gain and the concurrent clamping of the inversion that cause the strong mode competition. The topologically-matched mode (m = 1 for the K-valley and m = −1 for the K(cid:48) valley) reaches the threshold first and saturates the gain, thus, prevent- ing the mismatched mode from the generation under any pumping or any perturbations. The proposed topological spaser is promising for the use in nanooptics and nanospectroscopy where strong rotating nano-localized fields are required. It may be especially useful in applications to biomolecules and bi- ologicals objects, which are typically chiral. It may also be used as a nano-source of a circularly polarized radia- tion in the far field, in particular, as a biomedical multi- functional agent similar to the use of spherical spasers [18]. Acknowledgments Major funding was provided by Grant No. DE-FG02- 11ER46789 from the Materials Sciences and Engineering Division of the Office of the Basic Energy Sciences, Office of Science, U.S. Department of Energy. Numerical simu- lations have been performed using support by Grant No. DE-FG02-01ER15213 from the Chemical Sciences, Bio- sciences and Geosciences Division, Office of Basic Energy Sciences, Office of Science, US Department of Energy. The work of V.A. was supported by NSF EFRI NewLAW Grant EFMA-17 41691. Support for J.W. came from a MURI Grant No. N00014-17-1-2588 from the Office of Naval Research (ONR). V. APPENDIX A. Modes of a metallic oblate spheroid For an oblate spheroid, geometry is defined by the spheroidal coordinates, ξ, η and ϕ related to the Carte- sian coordinates, x, y and z as [45]: (cid:112) (cid:112) (cid:112) (cid:112) 1 − η2 cos(ϕ), 1 − η2 sin(ϕ), x = f ξ2 + 1 (25) ξ2 + 1 y = f z = f ξη, (26) (27) 0 ≤ ϕ < 2π and with 0 ≤ ξ < ∞, −1 ≤ η ≤ 1, f > 0. The corresponding equation in the Cartesian coordinate system describes a spheroidal shape with the semi-principal axes a and c, x2 + y2 a2 + z2 c2 = 1 . The eccentricity, ε, can be written as ε = f = εa. (cid:113) (28) 1 − c2 a2 and The surface plasmon eigenmodes of the spheroid are described by the quasistatic equation [44] ∇ [θ(r)∇φm] = ssp∇2φm. (29) Here θ(r) is the characteristic function that is equal to 1 inside the metal and 0 elsewhere. For the spheroid, the multipole quantum number is l = 1, and m is the angular momentum projection. Eigenmodes are given by φm = CNP m 1 (η)eimφ 0 < ξ < ξ0, ξ0 < ξ, (30) l (x) and Qm where P m the first and second kind, respectively, and ξ0 = l (x) are the Legendre functions of 1−ε2 . ε √ (cid:40) P m 1 (iξ) 1 (iξ0) , P m Qm 1 (iξ) 1 (iξ0) , Qm 7 The constant CN is determined by the normalization con- dition of the eigenmodes, ∇φ(r)m2d3r = 1. (31) Due to the axial symmetry, the corresponding eigen- values, ssp, do not depend on m; they are given by the equation [1, 3] θ(r)∇φm(r)2d3r ∇φm, (r)2d3r . (32) ssp = All Space All Space (cid:90) All Space (cid:90) (cid:90) . (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)x=iξ0 (cid:12)(cid:12)(cid:12)ω=ωsp From this, we derive an analytical form of the eigenvalue ssp = dP m 1 (x) dx 1 (x) 1 (x) Qm 1 (x) dx − P m dP m dQm 1 (x) dx (33) To find the SP frerquency, ωsp, and the SP relaxation rate, γsp, we use relations [1, 3] ssp = Re[s(ωsp)], Im[s(ωsp)] rsp = , s(cid:48) sp sp ≡ dRe[s(ω)] s(cid:48) dω (34) , (35) where the Bergman spectral parameter is defined as s(ω) =
d
d −
m(ω) . (36)
d is the dielectric constant of the environment, and
m(ω) is the dielectric function of the metal. In our com- putations, we use silver with the dielectric function from the Ref. 52. B. TMDC Parameters The eigenfrequency, ωsp, depends on the aspect ratio, c/a. We set a = 12 nm and vary the value of c to have the surface plasmon frequency match the band gap, ωsp = ∆g. Table I shows the calculated values of c to match the band gaps of the specific TMDCs. The transition dipole matrix element of the gain me- diaum (TMDC) were calculated using a three-band tight binding model [49]. The calculated values of the tran- sition dipoles along with the band gap are indicated in Table I. The transition dipoles at the K and K(cid:48) points are purely chiral: they are proportional to e± = 2−1/2 (ex ± iey), where ex and ey are the Cartesian unit vectors. A plot of absolute value of the chiral dipole, d±, where d± = e∗ It clearly points to the chirality of transition dipoles (d+ at the K point and d− at the K point) in the TMDCs. ±d, is shown in Fig.6. 8 FIG. 6: Absolute value of the left-rotating chiral dipole com- ponent, d− = e+d, in MoS2 TMDC MoS2 MoSe2 WSe2 MoTe2 Semi-principal Dipole elements (D) Band gap axis c (nm) 1.20 1.45 0.85 1 dK 17.68e+ 19.23e+ 18.38e+ 20.08e+ dK(cid:48) 17.68e− 19.23e− 18.38e− 20.08e− (eV) 1.66 1.79 1.43 1.53 TABLE I: Parameters employed in the calculations: Semi- principal axes of the spheroids, and the dipole matrix elements and band gaps of the TMDCs. [1] D. J. Bergman and M. I. Stockman, Phys. Rev. Lett. 90, 027402 (2003). [2] K. Li, X. Li, M. I. Stockman, and D. J. Bergman, Phys. Rev. B 71, 115409 (2005). [3] M. I. Stockman, Journal of Optics 12, 024004 (2010). [4] D. Y. Fedyanin, Opt. Lett. 37, 404 (2012). [5] D. G. Baranov, A. P. Vinogradov, A. A. Lisyansky, Y. M. Strelniker, and D. J. Bergman, Opt. Lett. 38, 2002 (2013). Veldhoven, B. Barcones, B. Koopmans, R. Notzel, M. K. Smit, and M. T. Hill, Opt. Express 19, 15109 (2011). [12] F. v. Beijnum, P. J. v. Veldhoven, E. J. Geluk, M. J. A. d. Dood, G. W. t. Hooft, and M. P. v. Exter, Phys. Rev. Lett. 110, 206802 (2013). [13] Y.-J. Lu, C.-Y. Wang, J. Kim, H.-Y. Chen, M.-Y. Lu, Y.-C. Chen, W.-H. Chang, L.-J. Chen, M. I. Stockman, C.-K. Shih, et al., Nano Lett. 14, 43814388 (2014). [14] Q. Zhang, G. Li, X. Liu, F. Qian, Y. Li, T. C. Sum, C. M. [6] N. I. Zheludev, S. L. Prosvirnin, N. Papasimakis, and Lieber, and Q. Xiong, Nat. Commun. 5, 4953 (2014). V. A. Fedotov, Nat. Phot. 2, 351 (2008). [7] M. A. Noginov, G. Zhu, A. M. Belgrave, R. Bakker, V. M. Shalaev, E. E. Narimanov, S. Stout, E. Herz, T. Sutee- wong, and U. Wiesner, Nature 460, 1110 (2009). [8] R. F. Oulton, V. J. Sorger, T. Zentgraf, R.-M. Ma, C. Gladden, L. Dai, G. Bartal, and X. Zhang, Nature 461, 629 (2009). [15] B. T. Chou, Y. H. Chou, Y. M. Wu, Y. C. Chung, W. J. Hsueh, S. W. Lin, T. C. Lu, T. R. Lin, and S. D. Lin, Sci. Rep. 6, 19887 (2016). [16] C.-J. Lee, H. Yeh, F. Cheng, P.-H. Su, T.-H. Her, Y.- C. Chen, C.-Y. Wang, S. Gwo, S. R. Bank, C.-K. Shih, et al., ACS Photonics 4, 1431 (2017). [17] S. Sun, C. Zhang, K. Wang, S. Wang, S. Xiao, and [9] R.-M. Ma, R. F. Oulton, V. J. Sorger, G. Bartal, and Q. Song, ACS Photonics 4, 649656 (2017). X. Zhang, Nat. Mater. 10, 110 (2010). [10] R. A. Flynn, C. S. Kim, I. Vurgaftman, M. Kim, J. R. Meyer, A. J. Makinen, K. Bussmann, L. Cheng, F. S. Choa, and J. P. Long, Opt. Express 19, 8954 (2011). [18] E. I. Galanzha, R. Weingold, D. A. Nedosekin, M. Sa- rimollaoglu, J. Nolan, W. Harrington, A. S. Kuchyanov, R. G. Parkhomenko, F. Watanabe, Z. Nima, et al., Nat. Commun. 8, 15528 (2017). [11] M. J. H. Marell, B. Smalbrugge, E. J. Geluk, P. J. van [19] R. Ma, X. Yin, R. F. Oulton, V. J. Sorger, and X. Zhang, 9 Nano Lett. p. doi: 10.1021/nl302809a (2012). [20] R.-M. Ma, S. Ota, Y. Li, S. Yang, and X. Zhang, Nature Nanotechnology 9, 600 (2014). [21] S. Wang, B. Li, X. Y. Wang, H. Z. Chen, Y. L. Wang, X. W. Zhang, L. Dai, and R. M. Ma, ACS Photonics 4, 1355 (2017). [22] Z. Wu, J. Chen, Y. Mi, X. Sui, S. Zhang, W. Du, R. Wang, J. Shi, X. Wu, X. Qiu, et al., Advanced Optical Materials pp. 1800674 -- 1 -- 8 (2018). [23] Y.-J. Lu, J. Kim, H.-Y. Chen, C. Wu, N. Dabidian, C. E. Sanders, C.-Y. Wang, M.-Y. Lu, B.-H. Li, X. Qiu, et al., Science 337, 450 (2012). [24] C.-J. Lee, H. Yeh, F. Cheng, P.-H. Su, T.-H. Her, Y.- C. Chen, C.-Y. Wang, S. Gwo, S. R. Bank, C.-K. Shih, et al., ACS Photonics 4, 1431 (2017). [25] S. Gwo and C.-K. Shih, Rep. Prog. Phys. 79, 086501 (2016). [26] C.-Z. Ning, Advanced Photonics 1, 014002 (2019). [27] S. J. P. Kress, J. Cui, P. Rohner, D. K. Kim, F. V. Antolinez, K.-A. Zaininger, S. V. Jayanti, P. Richner, K. M. McPeak, D. Poulikakos, et al., Science Advances 3, e1700688 (2017). [28] E. Plum, V. A. Fedotov, P. Kuo, D. P. Tsai, and N. I. Zheludev, Opt. Express 17, 8548 (2009). [29] Y.-W. Huang, W. T. Chen, P. C. Wu, V. A. Fedotov, N. I. Zheludev, and D. P. Tsai, Sci. Rep. 3, 1237 (2013). [30] W. Zhou, M. Dridi, J. Y. Suh, C. H. Kim, D. T. Co, M. R. Wasielewski, G. C. Schatz, and T. W. Odom, Nature Nano 8, 506 (2013). [31] J.-S. Wu, V. Apalkov, and M. I. Stockman, Topological [36] Y. M. You, X. X. Zhang, T. C. Berkelbach, M. S. Hy- bertsen, D. R. Reichman, and T. F. Heinz, Nat. Phys. 11, 477 (2015). [37] K. S. Novoselov, A. Mishchenko, A. Carvalho, and A. H. C. Neto, Science 353, 461 (2016). [38] D. N. Basov, M. M. Fogler, and F. J. G. de Abajo, Science 354, 195 (2016). [39] O. Salehzadeh, M. Djavid, N. H. Tran, I. Shih, and Z. Mi, Nano letters 15, 5302 (2015), ISSN 1530-6984. [40] Y. Ye, Z. J. Wong, X. Lu, X. Ni, H. Zhu, X. Chen, Y. Wang, and X. Zhang, Nat. Phot. 9, 733 (2015). [41] S. Wu, S. Buckley, J. R. Schaibley, L. Feng, J. Yan, D. G. Mandrus, F. Hatami, W. Yao, J. Vukovi, A. Majumdar, et al., Nature 520, 69 (2015). [42] T. Cao, G. Wang, W. Han, H. Ye, C. Zhu, J. Shi, Q. Niu, P. Tan, E. Wang, B. Liu, et al., Nat. Commun. 3, 887 (2012). [43] Z. Ye, D. Sun, and T. F. Heinz, Nat. Phys. 13, 26 (2016). [44] M. I. Stockman, S. V. Faleev, and D. J. Bergman, Phys. Rev. Lett. 87, 167401 (2001). [45] M. Willatzen and L. C. L. Y. Voon, in Separable Bound- aryValue Problems in Physics (WileyVCH, Weinheim, Germany, 2011), pp. 155 -- 164. [46] S. Wu, S. Buckley, J. R. Schaibley, L. Feng, J. Yan, D. G. Mandrus, F. Hatami, W. Yao, J. Vuckovic, and A. Ma- jumdar, arXiv preprint arXiv:1502.01973 (2015). [47] F. Bloch and A. Siegert, Phys. Rev. 57, 522 (1940), URL https://link.aps.org/doi/10.1103/PhysRev.57.522. [48] G. S. Agarwal, Phys. Rev. A 4, 1778 (1971), URL https: //link.aps.org/doi/10.1103/PhysRevA.4.1778. spaser (2019), arXiv: 1909.11113 [cond-mat.mes-hall]. [49] G. B. Liu, W. Y. Shan, Y. G. Yao, W. Yao, and D. Xiao, [32] X. Y. Wang, Y. L. Wang, S. Wang, B. Li, X. W. Zhang, Phys. Rev. B 88, 085433 (2013). L. Dai, and R. M. Ma, Nanophotonics 6, 472 (2017). [33] M. Stockman, Spasers to speed up cmos processors (2018). [50] M. I. Stockman, Opt. Express 19, 22029 (2011). [51] L. D. Landau and E. M. Lifshitz, The Classical Theory of Fields (Pergamon Press, Oxford, New York, 1975). [34] M. V. Berry, Proc. Royal Soc. London Ser. A 392, 45 [52] P. B. Johnson and R. W. Christy, Phys. Rev. B 6, 4370 (1984). (1972). [35] D. Xiao, M.-C. Chang, and Q. Niu, Reviews of Modern Physics 82, 1959 (2010). 

1007.4910

1

1007

Band topology and quantum spin Hall effect in bilayer graphene

We consider bilayer graphene in the presence of spin orbit coupling, to assess its behavior as a topological insulator. The first Chern number $n$ for the energy bands of single and bilayer graphene is computed and compared. It is shown that for a given valley and spin, $n$ in a bilayer is doubled with respect to the monolayer. This implies that bilayer graphene will have twice as many edge states as single layer graphene, which we confirm with numerical calculations and analytically in the case of an armchair terminated surface. Bilayer graphene is a weak topological insulator, whose surface spectrum is susceptible to gap opening under spin-mixing perturbations. We also assess the stability of the associated topological bulk state of bilayer graphene under various perturbations. Finally, we consider an intermediate situation in which only one of the two layers has spin orbit coupling, and find that although individual valleys have non-trivial Chern numbers, the spectrum as a whole is not gapped, so that the system is not a topological insulator.

cond-mat.mes-hall

cond-mat

Band topology and quantum spin Hall effect in bilayer graphene 1Instituto de Ciencia de Materiales de Madrid, (CSIC), Cantoblanco, 28049 Madrid, Spain E. Prada1, P. San-Jose2 and L. Brey1 2Instituto de Estructura de la Materia (CSIC), Serrano 123, 28006 Madrid, Spain Department of Physics, Indiana University, Bloomington IN 47405 (Dated: October 24, 2018) H.A. Fertig We consider bilayer graphene in the presence of spin orbit coupling, to assess its behavior as a topological insulator. The first Chern number n for the energy bands of single and bilayer graphene is computed and compared. It is shown that for a given valley and spin, n in a bilayer is doubled with respect to the monolayer. This implies that bilayer graphene will have twice as many edge states as single layer graphene, which we confirm with numerical calculations and analytically in the case of an armchair terminated surface. Bilayer graphene is a weak topological insulator, whose surface spectrum is susceptible to gap opening under spin-mixing perturbations. We also assess the stability of the associated topological bulk state of bilayer graphene under various perturbations. Finally, we consider an intermediate situation in which only one of the two layers has spin orbit coupling, and find that although individual valleys have non-trivial Chern numbers, the spectrum as a whole is not gapped, so that the system is not a topological insulator. INTRODUCTION The study and characterization of the properties of topological insulators has sparked considerable interest [1, 2]. A topological insulator has a bulk energy gap separating the occupied electronic bands from the empty ones. However, the conduction and valence bands in the bulk are inverted with respect to their energy position in the vacuum. This necessarily results in the existence of localized surface states that cross the energy gap and are protected by time reversal symmetry. The electronic and magnetic properties of the surface states of a three- dimensional topological insulator are chiral, and are gov- erned by a two dimensional Dirac Hamiltonian. In a two dimensional topological insulator the chiral surface states become helical one dimensional edge states, and at a given edge, states with opposite spin orienta- tions transport charge in opposite spatial directions[3]. In the absence of time reversal symmetry breaking, the helicity of the edge states prevents backscattering be- tween states in the same edge and the Hall conductivity per spin becomes quantized, although with different sign for opposite spins. Note that this system does not vio- late time reversal symmetry because, when reversing the time direction, both the spin and the Hall conductivity per spin reverse. In this way the total Hall conductivity of the system vanishes. As pointed out by Thouless et al. [4], the quantization of the Hall conductivity is a conse- quence of the topology of the band structure. Thouless et al. showed that the Hall conductivity should be an in- teger in units of e2/h, i.e., a topological invariant called the first Chern number. Because the Hall conductivity per spin is quantized, two-dimensional topological insu- lators are also called quantum spin Hall systems. The quantum spin Hall effect was theoretically predicted [5] to occur, and later experimentally observed [6], in HgTe quantum wells confined by CdTe barriers. Graphene, a two-dimensional carbon crystal [7], has also been proposed to be a two-dimensional quantum spin Hall system [8 -- 10] when the intrinsic spin orbit interac- tion is taken into account. Spin orbit coupling (SOC) in graphene opens a gap at the Dirac points and the system becomes a topological insulator. However, due to the π character of the graphene bands close to the Fermi en- ergy, the opened gap is proportional to the square of the intra-atomic spin orbit coupling constant divided by the energy difference between the s and p orbitals of graphene [11 -- 13]. Since carbon is a very light atom, the spin orbit gap is very small, ∼ 10−3meV. Consequently, the partic- ular transport properties of graphene as a topological in- sulator may only be observed in extremely clean samples and at extremely low temperatures. Graphene, neverthe- less, is a pedagogical toy model for analyzing properties of two-dimensional topological insulators [1]. In a Bernal stacked graphene bilayer, the conduction and the valence bands touch at two inequivalent points of the Brillouin zone and the system is a semimetal [14]. SOC in bilayer graphene also opens a gap in the band structure [15]. The spin-orbit interaction in bilayer graphene is larger, by about one order of magnitude, than in a single layer of graphene, due to the mixing of π and σ bands by the interlayer hopping [16, 17]. Nevertheless, the spin-orbit induced gap in the bilayer is small and difficult to detect experimentally. As mentioned above, our main interest in this work is the use of graphene based structures to understand properties of topological insulators. In par- ticular, we use bilayer graphene as a model to analyze the coupling between two topological insulators. Our aim in this work is to study the topological prop- 0 1 0 2 l u J 8 2 ] l l a h - s e m . t a m - d n o c [ 1 v 0 1 9 4 . 7 0 0 1 : v i X r a erties of a graphene bilayer in the presence of SOC. In graphene and bilayer graphene, the low energy proper- ties can be described by 2×2 Hamiltonians and for each momentum (cid:126)p, the wavefunctions have a spinor form. In undoped systems, the ground state of the system can be characterized by an unit vector field, (cid:126)h((cid:126)p), that indi- cates, at each point of the reciprocal space, the expec- tation value of the orientation of the pseudospin. In the reciprocal space (cid:126)h has the form of a topological object. We find that in the case of graphene (cid:126)h has the form of a meron, and in the case of bilayer graphene it takes the form of a double vortex meron. Thus, we find that the first Chern number of bilayer graphene is twice that of the monolayer one, and consequently the number of edge states is also doubled. In bilayers, edge states are not Kramer-protected against backscattering, which give the topological insulating phase a weak character. We also analyze the stability of the insulating bulk topology in bilayer graphene with respect to bias voltage, staggered sublattice potential and trigonal warping effects. Finally, we study a bilayer graphene system in which only a sin- gle layer has SOC. We find that this system has a finite Chern number, but is a zero gap semiconductor for which no surface states are possible. GRAPHENE Carbon atoms in graphene are covalently bonded and arranged in a honeycomb lattice, which is composed of two triangular sublattices A and B. The low energy prop- erties in graphene are mainly determined by the π or- bitals. A tight-binding Hamiltonian with hopping γ0 be- tween nearest-neighbors appropriately describes its band structure. In graphene the intrinsic SOC does not break the inversion symmetry of the honeycomb lattice and the electronic bands are spin degenerate. In addition, since the intrinsic SOC in graphene is a second order effect, the z-component of the electron spin commutes with the Hamiltonian, and the bands can be indexed by sz = ±1, the up/down electron spin component perpen- dicular to the graphene layer. The corresponding tight binding Hamiltonian of graphene is H = −γ0 (cid:88) (cid:88) bj,sz + itso cj,sz . szνijc+ i,sz a+ i,sz <i,j>,sz <<i,j>>,sz (1) Here ai,sz (bi,sz ) annihilates an electron on sublattice A(B) at site i and spin sz, tso is the next-nearest neigh- bor spin orbit hopping amplitude and cj is either aj or bj, depending wether the index j labels an A or B-sublattice site, respectively. The factor νij is +1 if the next-nearest neighbor hopping path rotates counterclockwise and -1 if it rotates clockwise. In undoped graphene, and for tso = 0, the conduction and valence bands touch at two inequivalent points of 2 3 π a (1, 0) and K(cid:48) = − 4 π the Brillouin zone: K = 4 a (1, 0), 3 being a the lattice parameter of the triangular lattice. These are known as Dirac points. The main effect of the √ SOC in the electronic spectrum is the opening of a energy 3tso, at the Dirac points. Near these points the gap, 6 wave functions for each spin sz can be expressed via the k · P approximation [18] in terms of envelope functions ψ+,sz = [Asz (r), Bsz (r)] and ψ−,sz = [A(cid:48) (r)] for states near the K and K(cid:48) points respectively. These wavefunctions satisfy the Dirac equations Hτz,sz ψτz,sz = εψτz,sz , where τz = ±1 specifies the Dirac points K and K(cid:48) and (r), B(cid:48) sz sz Hτz,sz = vF (pxτzσx + pyσy) + ∆soσzτzsz. (2) √ √ 3 2 γ0a, ∆so = 3 3tso and σi are Pauli matri- Here vF = ces representing the pseudospin degree of freedom corre- sponding to the two sites per unit cell of the graphene lattice. Note that (cid:126)p denotes the distance in momentum from the K and K(cid:48) points. It is important to note that for a given sz, the Dirac mass terms in the Hamiltonian induced by SOC at K and K(cid:48) have the same magnitude but different sign. Graphene with intrinsic SOC is a topological insulator with a finite spin Hall conductivity. For a given sz and valley index τz, the spin Hall conductivity as obtained from the Kubo formula has the form (cid:88) (cid:126)k,i,j σsz,τz xy = 2 e2 Ω Im[< ψi (cid:126)k vyψj (εi (cid:126)k >< ψj (cid:126)k (cid:126)k − εj (cid:126)k )2 vxψi (cid:126)k >] , (3) where j(i) runs over occupied (empty) states, vν is the velocity operator in the ν direction and the wavefunc- tions ψi are obtained by diagonalizing (cid:126)k Eq. In undoped graphene (Fermi energy crossing the Dirac points), the Hall conductivity for each valley takes the value (2) for the corresponding τz and sz. and energies εi (cid:126)k xy = − 1 σsz,τz 2 e2 h sz, (4) where h is the Planck constant. When summing over spins the total Hall conductivity of the system is zero, as it should be for a system with time reversal symmetry [1, 2]. However, for each spin the Hall conductivity is quantized, although with opposite sign. This is the sig- nature of the quantum spin Hall effect. Note also that although for an isolated valley the Hall conductivity is a half integer in units of e2/h, the sum of the K and K(cid:48) conductivities is quantized to integer multiples of e2/h, as it should be for a filled band of noninteracting electrons [4, 19]. In an insulator, the value of the Hall conductivity in units of e2/h is related to the first Chern number [4] of its bandstructure. The Chern number corresponding to a 2×2 Hamiltonian Hτz,sz =
((cid:126)p) (cid:126)h((cid:126)p) · (cid:126)σ , (5) is related to the number of times the unit sphere is cov- ered by the unit vector (cid:126)h((cid:126)p) when (cid:126)p runs over the whole reciprocal space [1, 2, 20]. This number takes the form n = 1 4π d2p(∂px (cid:126)h × ∂py (cid:126)h) · (cid:126)h, (6) (cid:90) where (cid:126)h((cid:126)p) depends on τz and sz and ±
((cid:126)p) are the energy eigenvalues. n is a topological invariant, the Pontyagin index of the mapping h((cid:126)p) [20]. Physically, the vector (cid:126)h((cid:126)p) represents the expectation value of the orientation of the pseudospin associated with the wavefunctions of the Hamiltonian, Eq. (2). , τz (cid:104) (cid:105) The vector (cid:126)h defines the topology of the band structure and may be written in the form (cid:112)1 − [hz(p)]2 cos θ,(cid:112)1 − [hz(p)]2 sin θ, hz(p) (cid:126)h = with hz(p) = szτz∆so/(cid:112)v2 F p2 + ∆2 (7) so. Here the valley index τz = ± defines the right and left vorticity of the topological structure and θ is the azimuthal angle made by the momentum vector (cid:126)p. At asymptotically large mo- mentum, hz vanishes, while in the cortex core we have hz = τzsz. This implies that there are four flavors of topological objects which are usually referred to as merons [21, 22], since they are essentially half skyrmions. The four possible merons are illustrated in Fig. 1. The Chern number index corresponding to the field (cid:126)h takes the form n = τz 4π d2p 1 p dhz dp = −τz 1 2 hz(0) = − 1 2 sz . (8) (cid:90) ∞ 0 Thus, the topological charge corresponding to each Dirac point is ±1/2 depending on the sign of the electron spin. When summing over the contribution from both Dirac points the Chern number in graphene is ±1. This result follows from the fact that, topologically, a meron has half the winding number of a skyrmion, that is, the topolog- ical object formed by the two Dirac points together. 3 FIG. 1: (Color online) Four flavors of merons, see Ref. 21. These are vortices that are right or left handed depending on the valley index τz. The topological charge, ±1/2, is deter- mined by the sign of the spin sz. one dimensional channels appear at an armchair termi- nated edge. We consider an edge running along the y-direction and the vacuum region is defined by the x < 0 condition. The graphene armchair termination consists of a line of A-B dimers, so it is natural to have the wave function amplitude vanish on both sublattices at x = 0. To do this we must admix valleys[25] and require Asz (x = 0) + A(cid:48) Bsz (x = 0) + B(cid:48) sz sz (x = 0) = 0 and (x = 0) = 0. (9) Note that the boundary conditions should be satisfied by each spin separately. We consider solutions with momen- tum py along the surface and energy E inside the gap, E < ∆so, where px = iκ. The general solution has the (cid:19) (cid:18) sin ξ(cid:48) cos ξ(cid:48) (cid:21) eiK(cid:48) xx py ye−κx ei (10) form(cid:20) (cid:18) sin ξ cos ξ (cid:19) (cid:113) Helical Edge States α eiKxx + β The most spectacular consequence of the existence of an insulator with a topologically non trivial band structure is the appearance of gapless conducting states at interfaces where the topology of the band structure changes. In two dimensional quantum spin Hall sys- tems, helical edge states appear at the surfaces of the material. At each edge there exists a pair of one di- mensional channels with opposite spins that propagate in opposite directions. In graphene these states have been obtained numerically by diagonalizing nanoribbons terminated with different geometries, zigzag[10, 23] or armchair[24]. Here we describe analytically how helical with vF κ = so − E2 + v2 ∆2 F p2 y and tan ξ = i vF (κ − py) E − ∆sosz and tan ξ(cid:48) = −i vF (κ + py) E + ∆sosz . (11) The boundary conditions, Eq. (9), imply that tan ξ = tan ξ(cid:48). This, together with the definition of κ, gives the solutions E = szvF py and vF κ = ∆so . (12) Since the graphene flake exists in the x > 0 region, nor- malizability of the wavefunction implies that κ should be τzsz=1τz= -1τzsz=-1τz= 1τzsz=1τz= 1τzsz=-1τz= -14 . denoted by ε(1,2)± (p), are (cid:115) (cid:114) γ2 1 2 v2 F p2 + ∆2 so + v2 F p2γ2 1 + + (−1)α ε(α)± = ± can be approximated by ε(1)± = ±(cid:113) p4 γ4 1 4 (14) These energies are independent of sz and τz [26]. The eigenvalues ε(2)± describe two strong interlayer coupling + ≥ γ1 and ε(2)− ≤ −γ1. These bands with energies ε(2) bands do not touch at the Dirac points, and correspond to wavefunctions mostly localized at sites A1 and B2, which form strong dimers [27]. The eigenvalues ε(1)± de- scribe low energy bands. Performing the usual low energy approximation vF p (cid:28) γ1, the dispersion of these bands so, where m is an effective mass induced by the interlayer hopping, m = γ1/2v2 F . The corresponding low energy eigenstates are mostly localized at sites B1 and A2. All of these states are degenerate in the spin sz and valley τz indices. For a given sz and τz, the Hall conductivity can be ob- tained numerically by plugging the eigenvalues and eigen- functions of the Hamiltonian (13) into the expression (3). For charge-neutral bilayer graphene, only negative energy bands are filled, and the Hall conductivity takes the form 4m2 + ∆2 xy = − e2 σsz,τz h sz. (15) When summing over the valleys, the Hall conductivity per spin is twice that of graphene. Interestingly, this is the same result one would expect for the case of two decoupled layers, although the eigenfunctions are com- pletely different from those of the coupled bilayer. As in the case of a monolayer, time reversal symmetry dictates opposite σsz,τz upon reversing the spin, so that the total Hall conductivity of the system is zero. xy Band structure topology The value of the Hall conductivity can be understood from the topology of the band structure through the first Chern number. Because of the gap ∼ 2γ1 between the high energy dimer bands ε(2)± , these cannot change their topology by adding a weak SOC. Therefore, only the gapless ε(1)± bands acquire non-trivial topology that con- tributes to a finite Chern number n. To compute n, it is then sufficient to consider the effective 2×2 low energy Hamiltonian that describes the low energy bands of the system in the limit ε (cid:28) γ1 [14], (cid:32) Hτz,sz = ∆soτzsz − (τzpx+ipy)2 2m = ∆soτzszσz − (p2 (cid:33) 2m − (τzpx−ipy)2 −∆soτzsz x − p2 y) 2m σx − pxpy m τzσy .(16) FIG. 2: (Color online) Tight-binding band structure of an armchair terminated wide nanoribbon. Shadow regions rep- resent bulk states. positive for any spin. Therefore, quasiparticles with pos- itive spin, sz > 0, move in the positive y direction and have and energy E = vF py. On the contrary, quasipar- ticles with sz < 0 move in the negative y direction and have energy E = −vF py. On the opposite surface, for which the wavefunction exists for x < 0, normalizability implies that px = −iκ. In this case E = −szvF py, and the edge states with positive (negative) velocity in the y-direction have sz < 0 (sz > 0). In Fig. 2 we plot the band structure of a wide armchair terminated nanorib- bon with a SOC tso = 0.01γ0. GRAPHENE BILAYERS Bilayer graphene consists of two stacked graphene sheets. In bilayer graphene there are four sites per unit cell, which we label A1, B1 and A2 and B2 in the the first and second layer respectively. We consider the so- called Bernal stacking, commonly found experimentally, in which site B2 is exactly on top of the sublattice A1. Interlayer coupling is modeled by hopping amplitude γ1 between these two sites in each unit cell. The low en- ergy properties of this model are well described for each spin sz by the following Hamiltonian acting on the four- component spinor (A1, B1, A2, B2) [14], H BG τz,sz = T0 ⊗ Hτz,sz − γ1 2 (Tx ⊗ σx − Ty ⊗ σy) , (13) where Hτz,sz is the monolayer Hamiltonian of Eq. (2), σi and Ti are the Pauli matrices for the sublattice and layer degree of freedom respectively and T0 is the unit matrix in the layer subspace. The four energy bands of H BG, kya-0.2-0.10.00.10.2Energy(γ0)-0.15-0.10-0.050.000.050.100.152∆so5 FIG. 3: (Color online) Four flavors of double vortex merons. These are vortices that are right or left handed depending on the valley index τz. The topological charge ±1 is determined by the sign of the spin. FIG. 4: (Color online) Band structure of a wide armchair terminated bilayer graphene ribbon for a single spin. The (green) shaded areas represent the continuum bulk states. This Hamiltonian acts on the two-component spinor (B1, A2). This Hamiltonian describes the low energy ef- fective coupling between carbon atoms in different layers which are not directly connected by tunneling. Their coupling arises as a result of virtual transitions through the high energy dimer states that have been integrated out. From this Hamiltonian and the relation of Eq. (5) we get the following expression for the unit vector field (cid:126)h((cid:126)p): (cid:104)−(cid:112)1 − [hz]2 cos 2θ,−τz (cid:126)h = (cid:105) (cid:112)1 − [hz]2 sin 2θ, hz) (cid:16) p2 (cid:17)2 , (17) (cid:114) (cid:90) ∞ 0 ∆2 2m so + . As in the case of with hz((cid:126)p) = ∆soτzsz/ a monolayer, the valley index τz determines the vorticity of the field (cid:126)h((cid:126)p) and the angle θ is the azimuthal angle of momentum (cid:126)p. At large momentum p, the z-component of the field (cid:126)h vanishes and the field is confined to be in the x − y plane. As in the case of the monolayer, there are four flavors of these topological objects, shown in Fig. 3. The Chern number of the corresponding fields (cid:126)h((cid:126)p) is n = 2 τz 4π d2p 1 p dhz dp = −τzhz(0) = −sz. (18) Thus, the Chern number corresponding to each valley is ±1 depending on the sign of the electron spin. It is in- teresting to note that, although these objects have twice the Pontryagin charge of merons (which are often thought as half a skyrmions), they are not skyrmions. Far away from the center of a skyrmion, i.e., at p → ∞, the orien- tation of the vector field (cid:126)h((cid:126)p) is constant. This is not the case of bilayer graphene (see Fig. 3), although, as in a skyrmion, the Chern number is unity for each τz, sz. We call these objects double vortex merons. When summing over valleys the Chern number in bilayer graphene is ±2 (depending on the spin), in agreement with the results obtained for the Hall conductivity. Edge states According to the bulk-boundary correspondence rule [1], and since the Chern number of bilayer graphene is twice that of the monolayer, the number of helical sur- face states should also be double. In order to analyze the edge states, we diagonalize the tight-binding Hamil- tonian corresponding to a wide armchair terminated bi- layer graphene ribbon. In Fig. 4 we plot its band struc- ture as function of the wavevector ky along the ribbon for a given electron spin sz. As in the case of a monolayer of graphene, the Hamiltonian commutes with sz and the band structure is degenerate in spin. The bulk bands are indicated by shaded regions. We identify two pairs of parallel edge states inside the gap. The states with positive velocity are located on the opposite ribbon edge from the states with negative velocity. This can be seen by analyzing the eigenfunctions of the low-energy Hamil- tonian, Eq. (16). As a result of the helicity of the edge states, their location changes to the opposite edge when the spin of the carriers changes sign. Bilayer graphene edge states can be understood as bonding/antibonding combinations of the surface states of the two constituent graphene layers. In the absence of tunneling between the graphene layers, the surface states have the form of Eq. (10) with ξ, ξ(cid:48), E and κ determined by Eq. (11) and Eq. (12). For finite γ1, surface states from layers 1 and 2 become coupled. To leading order in γ1, the energies and wavefunctions of the graphene τzsz=1τz= -1τzsz=-1τz= 1τzsz=1τz= 1τzsz=-1τz= -1kya-0.2-0.10.00.10.2Energy(γ0)-0.15-0.10-0.050.000.050.100.156 Fig. 4. As a result, SOC perturbations will in general split their degeneracy, and open a gap in the edge state dispersion around zero energy. Therefore, this model is not, strictly speaking, a topological insulator, since there exist single-body effects that are symmetric under time reversal that can induce backscattering between edge channels, and make the edges insulating. This happens for any two-dimensional insulator with an even number NK of edge state pairs, related by time reversal symme- try (ky, sz → −ky,−sz), at any given energy inside the gap (NK = 2 for the case at hand). An odd number of pairs NK, like in monolayers or trilayers with SOC, im- plies that there exists at least one of those pairs crossing at a high symmetry point, and are therefore protected from scattering by Kramer's theorem. A Z2 topological invariant of the form ν = NKmod 2, (21) is defined [1] to differentiate strict topological insulators (dubbed 'strong'), ν = 1, from 'weak' topological insu- lators, with ν = 0 and NK (cid:54)= 0. Within this classifi- cation, bilayer graphene is a weak topological insulator. The 'weakness' is in relation to the class of time-reversal- invariant perturbations that can open a gap in the edge states, in this case general spin-mixing perturbations. FIG. 5: (Color online) Schematic diagram of the edge states in a bilayer graphene system. At the edge of the sample there are four channels: two with spin up and moving in one direc- tion, and two with spin down an moving in the opposite direc- tion. Dotted and dashed lines represent channels formed by bonding/antibonding combinations of individual layers edge channels. bilayer Hamiltonian Eq. (13), defined for x < 0, are Esz± = szvF py ∓ γ1 4 (19)  −isz 1±sz±i (cid:104) ψsz± = √ 1 2 2 xx(cid:105) eiKxx − eiK(cid:48) eikyye−∆sox/vF . Stability of the topological insulating phase (20) where ± correspond to bonding/antibonding states. These expressions are valid when the SOC ∆so is much larger than γ1/4. In other situations, bulk states should reduce the bonding/antibonding energy separation. By inspection of Eq. (20), we see that at each edge of the system there exist four one dimensional channels: two with spin up moving in one direction and the other two with spin down and moving in the opposite direc- tion, see Fig. 5. The two terminal electrical conductance in this model is 4e2/h. The injection of charge current through edge states would result in antisymmetric spin accumulation at the edges. In a four terminal geometry the spin conductances should be quantized. However, in bilayer graphene, backscattering between channels at the same edge moving in opposite directions with opposite spins is not forbidden, in contrast to the case of monolayer graphene. Indeed, Kramers's theorem ensures degeneracy of time-reversed pairs in the absence of time reversal symmetry breaking, preventing backscat- tering between energy degenerate edge states with oppo- site spins at high symmetry points of the Brillouin zone, such as the Γ-point (for which (cid:126)k = −(cid:126)k modulo a recip- rocal lattice vector) [1]. While in monolayer graphene edge states branches cross at the Γ-point ky = 0, bilayer graphene edge state branches cross away from ky = 0, see The non-trivial insulating phase created by the instrin- sic SOC may be destroyed by Rashba SOC or other per- turbations if they are strong enough. This happens not by inducing backscattering between edge states, but by band reconnection, which changes the topology of the bulk bandstructure back to that of a conventional insu- lator [9]. It is a generic possibility both in weak and strong topological insulators. We now consider the destruction of bilayer graphene's topological insulator phase by a symmetric Rashba SOC, a staggered sublattice mass term and a voltage bias be- tween the two layers. We also discuss the effect of the trigonal warping on the topological insulating phase. 1) The Rashba spin orbit coupling term arises due to an electric field perpendicular to the bilayer plane or from the interaction with a substrate, HR = λRT0 ⊗ (τzσx ⊗ sy − σy ⊗ sx) . (22) Here sx and sy are the Pauli matrices in the spin sub- space and λR is the intensity of the coupling. This term violates mirror symmetry about the planes. In the pres- ence of this term the z-component of the electron spin is not conserved, but there is a region of values of the coupling, 0 < λR < ∆so, for which the ground state is adiabatically connected[9, 10] to the topological insu- lating phase. For values of the coupling λR > ∆so the ·xx·▲▲▼▼energy gap closes and the electronic structure is that of a zero gap semiconductor with quadratically dispersive bands. 2) A staggered sublattice potential of the form Hv = λvT0 ⊗ σz (23) 7 stabilizes the band insulator phase that competes with the topological insulating phase. This term is typically zero in graphene but would be present for a similar boron nitride film. In the effective 2×2 Hamiltonian this term appears as a mass term independent of the spin and val- ley, λvσz. In the presence of this term, the form of the unit vector field (cid:126)h((cid:126)p) of Eq. (17) is unaffected, but now the z-component takes the form ∆soτzsz + λv (∆soτzsz + λv)2 + (cid:17)2 (cid:16) p2 2m (cid:114) hz(p) = . (24) The Chern number associated with this vector field (cid:126)h((cid:126)p) is n = −τz sign(∆soτzsz + λv) (25) For values λv < ∆so the Chern number is −sz (equal for both valleys) and the system is a topological insulator. For values λv > ∆so we have n = −τz and the total Chern number per spin is zero when summing over valleys. In this case the bilayer graphene becomes a band insulator. 3) A bias voltage between the layers gives a contribution to the bilayer Hamiltonian, Eq. (13), of the form Hbias = V Tz ⊗ σ0. (26) This term acts equally on both sublattices and on both spin orientations. In the effective 2×2 low energy Hamil- tonian Eq. (16), this term has exactly the same form as the staggered sublattice potential V σz. Thus, the condi- tion for the existence of a topological insulating phase in the bilayer is V < ∆so. In the last two cases, when λv > ∆so or V > ∆so, the system becomes a conventional insulator with vanishing Hall and spin Hall conductances. The two inequivalent valleys, considered independently, have nonzero topolog- ical charge with opposite sign, n = −τz, and a valley de- pendent Hall effect may occur [28]. The nonzero Chern number of each valley induces chiral modes in topological domain walls in graphene [29] and bilayer graphene [30]. 4) Trigonal warping. The inclusion in the Hamiltonian of bilayer graphene of a weak direct A2-B1 hopping term γ3 (cid:28) γ1 introduces in the 2×2 effective Hamiltonian a term of the form Hw = v3(τzpxσx − pyσy), (27) √ 2 aγ3/. This term produces a trigonal warp- with v3 = ing in the band structure, which stretches the isoenergy 3 FIG. 6: (Color online) Density plot in the reciprocal space of the integrand of Eq. (18) in the absence (a) and the presence (b) of trigonal warping. The center of each square corresponds to the original Dirac point at K. Cold colors represent regions where the integrand is bigger. The parameters used in these plots are γ1=0.1γ0, ∆so = 10−5γ0 and in (b) γ3=0.05γ0. 3 π and 4 3 π, π and 5 lines along the directions φ = 0, 2 3 π for the val- ley K and along the directions φ = 1 3 π for the valley K(cid:48)[27]. For ∆so=0, the trigonal warping pro- duces a dramatic change in the band structure. For a given valley, instead of two parabolic bands touching at the Dirac point, there are now four Dirac points, one at the center (K or K(cid:48)-point), and three others that occurs at finite momentum and in the directions φ mentioned above. The SOC opens gaps at the four Dirac points, making the system insulating. To analyze the nature of the insulating phase we compute the Chern number. The warping term breaks the symmetric form of the unit vector field (cid:126)h((cid:126)p) given in Eq. (17) and prevents an ana- lytical calculation of the Chern number. By integrating numerically Eq. (3) we obtain that the Chern number is not affected by the trigonal warping. In the absence of trigonal warping the main contribution to the integral comes from an annulus around each Dirac point where dhz/dp is maximum, see Fig. 6(a) (dark region). When the trigonal warping effects are taken into account, the main contribution to the Chern number comes from the regions around the three new Dirac points, Fig. 6(b). SPIN-ORBIT PROXIMITY EFFECT In this Section we analyze the case of a graphene bi- layer in which only one of the layers has a finite intrinsic SOC Λso, whereas the other layer has zero coupling. This could happen in a bilayer in contact with a strong SOC metal[31], or in a bilayer where heavy atoms or molecules have been deposited on top of one of the two layers[32]. When only one of the layers is affected by the SOC, the 2×2 effective Hamiltonian takes the form − (τzpx−ipy)2 (cid:32) (cid:33) Hτz,sz = ∆soτzsz − (τzpx+ipy)2 2m 2m 0 (a)(b)8 (Color online) Tight binding band structure of a FIG. 8: zigzag (a) and a armchair (b) terminated bilayer graphene nanoribbon in which there is SOC only in one of the layers. These results correspond to sz=1. Results for sz=-1 are ob- tained by changing (cid:126)k to −(cid:126)k. In the numerics γ1 = 0.1γ0 and √ 3 × 10−2γ0. The width of the armchair and zigzag √ ∆so = 3 In the arm- nanoribbons is 151a and 120 chair case all the states are extended across the ribbon. In the zigzag one the states in the center of the Brillouin zone are localized at the edges. 3a respectively. conditions[25], states coming from different valleys can be treated independently and dispersive surface localized channels appear in the spectrum, see Fig. 8(a). However, for armchair terminated ribbons where, in order to satisfy the boundary conditions, the wavefunctions have to be a linear combination of both valleys, no localized surface states appear in the band structure, see Fig. 8(b). The absence of a full energy gap and the lack of surface states in some boundary terminations are clear indicators that a graphene bilayer with SOC in only one layer is not a quantum spin Hall system. CONCLUSION In this work we reviewed the description of single layer graphene with SOC as a topological insulator in terms of the first Chern number, which arises naturally in the computation of the Hall conductivity. We showed then that the Chern number per spin in bilayer graphene is two, twice that of the monolayer. This doubling is re- flected also in the number of topological surface states, (Color online) Low energy band structure of a FIG. 7: graphene bilayer where only one of the two layers has an finite SOC ∆so. Left and right panels correspond to the bands near the K and K(cid:48) points, respectively. Bands for states with spin up sz = 1 (down sz = −1) are represented with solid (dashed) lines. The Fermi energy is indicated by an dotted horizontal line. The SOC opens a gap ∆so with a different sign for each valley and each spin orientation. = ∆so 2 τzsz(σ0 + σz) − (p2 x − p2 y) 2m σx − τzpxpy m σy(28) and the unit vector field (cid:126)h has almost the same form as in Eq. (17). Thus, the Chern number of this system is equal to that of a graphene bilayer with intrinsic SOC in both layers, n = −sz . (29) Although from the value of the Chern number the system appears to be a topological insulator, this is not the case: it turns out it is not a true insulator at all. From the diagonalization of the Hamiltonian of Eq. (28) we get the band dispersion ∆soτzsz ±(cid:113) 2 ε± = ∆2 so + p4 4m2 . (30) This band structure, shown in Fig. 7, corresponds to an indirect zero band gap semiconductor. The absence of SOC in one of the layers breaks the inversion sym- metry, making the bands non-degenerate under indepen- dent inversion of valley and spin. When both valleys and/or both spins are considered, there is no gap in the spectrum. However, for each spin and valley there is an energy gap ∆so at (cid:126)p = 0, and the system is thus an in- sulator from an optical point of view. Moreover, if the system is free from perturbations that mix the two val- leys and the two spins, it will also behave as an effec- tive electronic insulator. This has an important conse- quence on the existence of surface states in the bilayer structure. In zigzag edge terminations for which it is not necessary to admix valleys to satisfy the boundary EF↑↓ΔsoF↓↑KK'KKkya-0.2-0.10.00.10.2Energy(γ0)-0.020.000.02kx(π/a)-0.4-0.20.00.20.4Energy(γ0)-0.04-0.020.000.020.04(a)(b)as a consequence of the bulk-surface correspondence rule. By numerically computing the spectrum of finite size samples, we checked that bilayer systems have twice as many edge states as a monolayer. This was furthermore confirmed analytically in the case of surfaces with arm- chair termination. The fact that the total Chern number per spin is even means that the bilayer system is a weak topological insulator, susceptible to gap opening if the system is subject to spin-mixing perturbations (such as a weak Rashba SOC). We also assessed the general sta- bility of the bulk topological insulating state of bilayer graphene with respect to Rashba SOC, a staggered sub- lattice potential, interlayer bias, and trigonal warping. The first three perturbations compete with the intrinsic SOC and, when sufficiently large, spoil the inverted gap property that is crucial to making the system a topolog- ical insulator. Finally, we examined a bilayer graphene system in which only one layer has intrinsic SOC. Al- though in this system individual valleys have non-trivial Chern numbers, the spectrum as a whole is not gapped, so that the system is not a topological insulator. Note added: While this manuscript was in the final stage of preparation, a manuscript by Cortijo, Grushim and Vozmediano appeared[33] which studies the Chern Simons coefficients in graphene and bilayer graphene us- ing an effective action formalism. Acknowledgments Funding for the work described here was provided by MICINN-Spain via grants FIS2009-08744 (EP and LB) and FIS2008-00124 (PSJ), and by the NSF through Grant No. DMR-1005035 (HAF). [1] M. Z. Hasan and C. L. Kane, Topological insulators (2010), URL http://www.citebase.org/abstract?id= oai:arXiv.org:1002.3895. [2] X.-L. Qi and S.-C. Zhang, Physics Today 63, 33 (2010). [3] For simplifying the discussion, here we consider systems for which the z-component of the spin commutes with the Hamiltonian. [4] D. J. Thouless, M. Kohmoto, M. P. Nightingale, and M. den Nijs, Phys. Rev. Lett. 49, 405 (1982). [5] B. A. Bernevig, T. L. Hughes, and S.-C. Zhang, Science 314, 1757 (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). Castro-Neto, N.M.R.Peres, and A.K.Geim, Rev. Mod. Phys. F.Guinea, [7] A. H. K.S.Novoselov, 81, 109 (2009). 9 [8] F. D. M. Haldane, Phys. Rev. Lett. 61, 2015 (1988). [9] C. L. Kane and E. J. Mele, Phys. Rev. Lett. 95, 226801 (2005). [10] C. L. Kane and E. J. Mele, Phys. Rev. Lett. 95, 146802 (2005). [11] D. Huertas-Hernando, F. Guinea, and A. Brataas, Phys. Rev. B 74, 155426 (2006). [12] H. Min, J. E. Hill, N. A. Sinitsyn, B. R. Sahu, L. Klein- man, and A. H. MacDonald, Phys. Rev. B 74, 165310 (2006). [13] Y. Yao, F. Ye, X.-L. Qi, S.-C. Zhang, and Z. Fang, Phys. Rev. B 75, 041401 (2007). [14] E. McCann and V. I. Fal'ko, Phys. Rev. Lett. 96, 086805 (2006). [15] R. van Gelderen and C. M. Smith, Phys. Rev. B 81, 125435 (2010). [16] F. Guinea, Spin-orbit coupling in a graphene bilayer and in graphite (2010), URL http://www.citebase. org/abstract?id=oai:arXiv.org:1003.1618. [17] H.-W. Liu, X. C. Xie, feng Sun, Gi- ant intrinsic spin-orbit coupling in bilayer graphene (2010), URL http://www.citebase.org/abstract?id= oai:arXiv.org:1004.0881. and Q. [18] T.Ando, J.Phys.Soc.Jpn. 74, 777 (2005). [19] N. A. Sinitsyn, J. E. Hill, H. Min, J. Sinova, and A. H. MacDonald, Phys. Rev. Lett. 97, 106804 (2006). [20] R.Rajaraman, Solitons and Instantons (North-Holland, The Netherlands, 1982). [21] S.M.Girvin and A.H.MacDonald, Perspectives in Quan- tum Hall Effects (Wiley-Interscience, 1997). [22] L. Brey, H. A. Fertig, R. Cot´e, and A. H. MacDonald, Phys. Rev. B 54, 16888 (1996). [23] W. Li and R. Tao, Edge states in monolayer and bi- layer graphene (2010), URL http://www.citebase.org/ abstract?id=oai:arXiv.org:1001.4168. [24] M. Zarea and N. Sandler, Phys. Rev. Lett. 99, 256804 (2007). [25] L. Brey and H. Fertig, Phys. Rev. B 73, 235411 (2006). [26] If in the bilayer system, apart from the SO term, we also include a mass term coming form a bias voltage V between layers, the eigenvalues would depend on sz and τz in the form [ε(α)]2 = v2 +(−1)α so + V 2 4 F p2 + τzszV ∆so) + V 2(v2 F p2 + ∆2 so) + γ4 4 . [27] E. McCann, D. S. Abergel, and V. I. Fal'ko, Solid State F p2 + ∆2 (cid:113) γ2 1 (v2 γ2 1 2 1 Communications 143, 110 (2007). [28] D. Xiao, W. Yao, and Q. Niu, Phys. Rev. Lett. 99, 236809 (2007). [29] W. Yao, S. A. Yang, and Q. Niu, Phys. Rev. Lett. 102, 096801 (2009). [30] I. Martin, Y. M. Blanter, and A. F. Morpurgo, Phys. Rev. Lett. 100, 036804 (2008). [31] Y. S. Dedkov, M. Fonin, U. Rudiger, and C. Laubschat, Phys. Rev. Lett. 100, 107602 (2008). [32] A. H. Castro Neto and F. Guinea, Phys. Rev. Lett. 103, 026804 (2009). [33] A. Cortijo, A. G. Grushin, and M. A. H. Vozmediano (2010), URL http://www.citebase.org/abstract?id= oai:arXiv.org:1007.3704. 

1112.0552

2

1112

Spin-dependent Klein tunneling in graphene: Role of Rashba spin-orbit coupling

Within an effective Dirac theory the low-energy dispersions of monolayer graphene in the presence of Rashba spin-orbit coupling and spin-degenerate bilayer graphene are described by formally identical expressions. We explore implications of this correspondence for transport by choosing chiral tunneling through pn and pnp junctions as a concrete example. A real-space Green's function formalism based on a tight-binding model is adopted to perform the ballistic transport calculations, which cover and confirm previous theoretical results based on the Dirac theory. Chiral tunneling in monolayer graphene in the presence of Rashba coupling is shown to indeed behave like in bilayer graphene. Combined effects of a forbidden normal transmission and spin separation are observed within the single-band n to p transmission regime. The former comes from real-spin conservation, in analogy with pseudospin conservation in bilayer graphene, while the latter arises from the intrinsic spin-Hall mechanism of the Rashba coupling.

cond-mat.mes-hall

cond-mat

Spin-dependent Klein tunneling in graphene: Role of Rashba spin-orbit coupling Ming-Hao Liu (劉明豪), Jan Bundesmann, and Klaus Richter Institut fur Theoretische Physik, Universitat Regensburg, D-93040 Regensburg, Germany (Dated: October 22, 2018) Within an effective Dirac theory the low-energy dispersions of monolayer graphene in the presence of Rashba spin-orbit coupling and spin-degenerate bilayer graphene are described by formally identical expressions. We explore implications of this correspondence for transport by choosing chiral tunneling through pn and pnp junctions as a concrete example. A real-space Green's function formalism based on a tight-binding model is adopted to perform the ballistic transport calculations, which cover and confirm previous theoretical results based on the Dirac theory. Chiral tunneling in monolayer graphene in the presence of Rashba coupling is shown to indeed behave like in bilayer graphene. Combined effects of a forbidden normal transmission and spin separation are observed within the single-band n ↔ p transmission regime. The former comes from real- spin conservation, in analogy with pseudospin conservation in bilayer graphene, while the latter arises from the intrinsic spin-Hall mechanism of the Rashba coupling. PACS numbers: 72.80.Vp, 72.25. -- b, 73.23. -- b, 73.40.Gk I. INTRODUCTION After the first successful isolation of monolayer graphene (MLG) was announced,1 intriguing properties based on its low-energy excitation that mimics massless, gapless, and chi- ral Dirac fermions were intensively investigated.2,3 Spin-orbit coupling (SOC), on the other hand, is the key ingredient of semiconductor spintronics4,5 that was undergoing a rapid de- velopment before the rise of graphene.6 The question about the role of SOC effects in graphene then naturally emerged, including the proposal of graphene as a topological insulator,7 which attracted the attention of various first-principles-based studies.8 -- 10 SOC in MLG includes an intrinsic and an extrinsic term. The former reflects the inherent asymmetry of electron hop- ping between next nearest neighbors7 (i.e., a generalization of Haldane's model11). The latter is induced by the elec- tric field perpendicular to the graphene plane, which can be externally controlled, and resembles the Rashba model12,13 for the two-dimensional electron gas. Agreement has been achieved, based on first-principles calculations,9,10 that the in- trinsic SOC term opens a gap of the order of 2l I ≈ 24 m eV, while the Rashba SOC removes the spin degeneracy and cre- ates a spin-splitting 2l R at the K and K′ points that has a lin- ear dependence on an external electric field E with the slope of about 100 m eV per V/ A of E. Under a strong gate voltage, the Rashba coupling may in principle dominate the intrinsic SOC in MLG.9,10 The low-energy spectrum of MLG plus the Rashba cou- pling (MLG+R) was derived by Rashba,14 based on the Kane- Mele model7 ( i.e., an effective Dirac Hamiltonian). An ear- lier work by one of us15 started with a tight-binding model (TBM) and obtained an equivalent form of the low-energy expansion,16 EMLG+R (q) ≈ m 1 2 [q(3tR)2 + (3ta· q)2 + n (3tR)], (1) which also agrees with expressions given in Refs. 9 and 10 when l I = 0. Here m ,n = ±1 are band indices, t and tR are nearest-neighbor kinetic and Rashba hopping parameters, re- spectively, a ≈ 1.42 A is the bonding length, and q = K + d k with d k a ≪ 1. Recall for comparison the low-energy spec- trum of bilayer graphene (BLG),2,17 1), (2) 3 and g 2 (qg 2 1 + (3ta· q)2 + ng EBLG (q) ≈ m 1 where g 1 is the nearest-neighbor hopping between the two graphene layers. Note that the next nearest-neighbor inter- layer hoppings g 4 do not influence the band dispersion near K. The completely different mechanisms of (i) pseu- dospin coupling between carriers from the two graphene lay- ers of BLG through interlayer hopping g 1 and (ii) real-spin coupling between up and down spins within MLG through Rashba hopping tR happen to lead to an identical mathemati- cal form in Eqs. (1) and (2) that can be clearly mapped onto each other18,19 with g 1 ↔ 3tR as sketched in Fig. 1. This un- ambiguously implies that low-energy physics in MLG+R and BLG should behave similarly. In this paper we tackle the question of whether the trans- port in MLG+R behaves as in BLG by choosing the issue of Klein tunneling2,3,20,21 (or, in general, chiral tunneling) as a concrete example. Chiral tunneling in graphene has been shown to exhibit completely different behavior in MLG and BLG based on the Dirac theory.22 Tunneling at normal inci- dence in MLG shows a suppression of backscattering, which resembles the original Klein paradox in relativistic quantum electrodynamics23 and hence the name Klein tunneling, while K(K ′) γ1 3tR tR γ1 FIG. 1. (Color online) Schematic of the pseudospin coupling through g 1 in BLG (left panel) and real-spin coupling through tR in MLG (right), which lead to an identical low-energy dispersion near K and K′. b 2 in BLG it shows a perfect reflection, which is strictly speaking a consequence of forbidden interband transition also due to the chiral nature of graphene. The theoretical discussion of chiral tunneling so far focuses mainly on spin-independent tunnel- ing through pn and pnp junctions,21,22,24 -- 30 while SOC effects are less discussed.18,31 -- 33 In addition, the relevant theoretical understanding so far is based on Dirac theory, which is valid only for the Fermi level close to the charge neutrality point and allows only to consider certain relatively simple systems. A recent study discussing the interplay between the Aharanov- Bohm effect and Klein tunneling in graphene, started with a TBM,34 but the nanoribbon type of the leads used in that work may have edge effects included that can be very different from the bulk properties of graphene. A more transparent theoreti- cal study of chiral tunneling in graphene directly bridging the analytical Dirac theory and the numerical TBM computation is so far missing and deserves consideration. In the present work, we re-treat this issue of chiral tunneling in graphene based on the TBM and show a unified description, allowing for a broad range of geometries and complementing the existing results based on the Dirac theory. Straightfor- ward generalization to the case of MLG+R reveals a spin- dependent tunneling behavior in close analogy with that in BLG, with the role of pseudospin in BLG replaced by real spin in MLG+R. Specifically, a combined behavior of spin- Hall-based spin separation and suppression of normal trans- mission will be shown. This paper is organized as follows. In Sec. II we briefly summarize the theoretical formalism applied in the present calculation, namely, real-space Green's function formalism in noninteracting bulk graphene. In Sec. III we show our TBM results including the consistency with the Dirac theory, a di- rect comparison between BLG and MLG+R, and a deeper dis- cussion of the MLG+R case. We review also briefly the re- cent experimental progress on the Rashba spin splitting and Klein tunneling in graphene in Sec. IV, and finally conclude in Sec. V. II. FORMALISM A. Tight-binding model for "bulk" graphene We choose the TBM for describing the electronic prop- erties of graphene, which is a well established way to treat graphene numerically. For spin-degenerate MLG, the Hamil- tonian reads hopping term, usually characterized by t′ ≈ 0.1t, can be added in Eq. (3) but will not be considered in the present work due to the minor role it plays in the bulk transport properties for low-energy excitation. c† i Vis 0c† i ci + (cid:229) Spin-orbit interactions can be incorporated into the TBM by altering the spin-dependent hopping between nearest and next-nearest neighbors,7,35 modifying Eq. (3) as HMLG+R = (cid:229) i (cid:2)−ts 0 + itR(~s × di j)z(cid:3) c j. Here s 0 is the 2 × 2 identity matrix, tR is the Rashba spin- orbit hopping parameter, di j is the unit vector pointing from site j to i, and ~s = (s x,s y,s z) is the vector of (real-) spin Pauli matrices. We take into account only the extrinsic SOC and neglect the intrinsic term in order to highlight the role of the Rashba SOC. hi, ji (4) For spin-degenerate BLG, we consider HBLG = (cid:229) m=1,2 MLG − g H (m) 1(cid:229) j (cid:16)b† 2, ja1, j + H.c.(cid:17) , (5) where H (m) MLG is HMLG given by Eq. (3) of the mth graphene layer, am, j (bm, j) annihilates an electron on sublattice A (B) in layer m = 1,2 at unit cell j (that contains two sublattice sites belonging to A and B), and the interlayer coupling strength g 1 ≈ 0.4 eV corresponds to the nearest neighbor hopping be- tween the two MLG layers. Further interlayer hopping terms,2 −g 4 (cid:229) 2, jb1, j+ H.c.), are not considered in the present calculation, since they do not influence the low-energy excitation. Throughout the presen- tation of the numerical results in Sec. III, the kinetic hopping parameters will be fixed at t = 3 eV and g 1 = 0.39 eV, while the value of the Rashba hopping parameter tR depends on the context. 2, jb1, j+ H.c.) and −g 2, ja1, j + b† j(a† j(a† 3 (cid:229) For the simulation of bulk graphene, we impose the Bloch theorem along the transverse direction with periodicity W . This is equivalent to considering a nanoribbon and modify- ing the hopping between atomic sites connected through the periodic boundary conditions by a Bloch phase factor eikBW with a Bloch momentum kB,36 as schematically shown for MLG in Fig. 2. At the same time the Bloch momentum is the component of the electron's momentum perpendicu- lar to the nanoribbon, hence defining the propagation angle f = sin−1(kB/kF ), where kF is the Fermi wave vector. To be consistent with the literature related to Klein tunneling based HMLG = (cid:229) i ci − t (cid:229) Vic† hi, ji i c† i c j, (3) ··· W eik y W eik y W eik y W eik y W eik y W eik y W eik y W eik y W eik y W eik y W eik y ··· where the operator c† i (ci) creates (annihilates) an electron at site i (including both sublattices A and B). The first sum in Eq. (3) runs over all the atomic sites in the considered region with on-site potential Vi, and the second sum runs over all the pairs of neighboring atomic orbitals hi, ji with kinetic hop- ping parameter t (≈ 3 eV). The next nearest neighbor kinetic L lead S region R lead FIG. 2. Schematic of a minimum tight-binding model that simulates a bulk MLG up to nearest neighbor hoppings with W = 3a. Fur- ther nearest neighbor hoppings can be accounted for by enlarging the transverse periodicity W to at least 6a. b b b b b b b b b b b b b b b b b b b b b b on the Dirac theory, in Sec. III we will refer to the Bloch mo- mentum as ky. In the present calculations, we will apply a minimal TBM by imposing the periodic boundary conditions on a zigzag nanoribbon with chain number Nz = 2, that is, periodicity of W = 3a (as the case sketched in Fig. 2). The present model ap- plies equally well for metallic armchair ribbon (chain number Na being a multiple of 3) with periodic boundary conditions, but the minimal model would require Na = 3 (i.e., periodicity of W = 3√3a). B. Brief summary of real-space Green's function formalism We consider open systems connected to the outer world by two leads (see Fig. 2). According to the real-space Green's function formalism37 we numerically calculate the Green's functions of our system, S = [E − HS − S Gr/a r/a ± ih ]−1, r/a = S (6) r/a L + S where the self-energies of the leads (S r/a R ) re- flect the fact that our system is open. The powerful recipe constructed in Ref. 36 for graphene handles a lead as a semi- infinite repetition of unit cells and allows for incorporating any kind of lattice structure and one-body interaction such as SOCs. The transmission probability for an electron traveling from lead L to lead R is given by the Fisher-Lee relation36,37 TRL = Tr(G LGr S G RFa S ), (7) L/R − S a L/R). where the trace is done with respect to the lattice sites. The spectral matrix functions G L/R are given by the lead self- energies as G L/R = i(S r For a given Bloch momentum ky and a given Fermi energy EF [subject to a Fermi wave vector kF via Eq. (1) for MLG+R or Eq. (2) for BLG], the incoming propagation angle f of the electron wave can be defined as f = sin−1(ky/kF ). The angle- dependent transmission function T (f ) is obtained from Eq. (7), which can be generalized to a spin-resolved version.38 III. TRANSPORT RESULTS In this section we present numerical results of our tight- binding transport calculations. We first show the consistency of our tight-binding calculations with the existing effective Dirac theory in Sec. III A. A direct comparison between BLG and MLG+R will then be shown in Sec. III B. Finally, Sec. III C is devoted to MLG+R for pn junctions, in particular the role of Rashba SOC for chiral tunneling. A. Consistency with Dirac theory We first consider tunneling in graphene without SOC and confirm existing results, limited to low energy excitations, by 0 30 −30 3 −60 V0 60 90 2 1 1 −90 2 0 (b) 0 γ1 V0 30 −30 60 −60 90 2 1 1 −90 2 0 (d) top view k F φ ky ~k ~k = EF V (x) ~k EF V (x) D (a) D (c) FIG. 3. (Color online) Tunneling through a barrier for (a), (b) MLG with EF = 3takF /2 = 81.6meV and (c), (d) BLG with EF = (3takF /2)2/g 1 = 17.1meV. In (b), red (light gray) and blue (dark gray) curves correspond to V0 = 196.8meV and V0 = 280.3meV, respectively. In (d), red (light gray) and blue (dark gray) curves cor- respond to V0 = 48.7meV and V0 = 100.7meV, respectively. In both cases the barrier width is D = 100nm and the incoming Fermi wave vector is kF = 2p /50nm−1, as considered in Ref. 22. our tight-binding calculations. We pick two pioneering theo- retical works to demonstrate the consistency explicitly. Con- sistency with recent works of tunneling in graphene hetero- junctions in the presence of SOC18,31 has also been checked, but is not explicitly shown here. 1. Chiral tunneling in MLG vs BLG Tunneling in MLG and BLG behaves quite differently as mentioned in Sec. I and pointed out by Katsnelson et al.22 For a quantitative comparison we consider a barrier of width D = 100 nm and the incoming Fermi wave vector kF = 2p /50 nm−1 as in Ref. 22 for both MLG and BLG [see Figs. 3(a) and 3(c)]. Note that in order to exactly match the bar- rier width, we set the bonding length a = (4√3)−1 nm, which differs from the realistic value of about 1.42 A by only less than 2%, so that the number of hexagons used here amounts to D/(√3a) = 4× [D] nm = 400. The resulting transmission probabilities as a function of the incident angle f are depicted in Figs. 3(b) and 3(d). They reproduce the results of Fig. 2 in Ref. 22 almost perfectly, if we choose slightly different EF and V0, to which the trans- missions at finite angles are sensitive. The remaining tiny difference between our TBM results and their Dirac theory results39 simply reflects the basic difference between the two approaches: For graphene the effective Dirac theory is valid only for energies close to the Dirac point, while the TBM is suitable for the entire energy range. Note that the maximal values of the transmission functions in Fig. 3 are 2, since the valley degeneracy is automatically incorporated in the tight-binding formalism. Later when we (cid:176) (cid:176) (cid:176) (cid:176) (cid:176) (cid:176) (cid:176) (cid:176) (cid:176) (cid:176) (cid:176) (cid:176) (cid:176) (cid:176) 4 EF V (x) 2 1.5 1 0.5 ) φ ( T ~k Eq. (8) TBM ~k d V0 d = 20 nm V0 EF V (x) 2 1.5 1 0.5 ) φ ( T 0 −90 −60 −30 0 φ (deg) (a) 30 60 90 0 −90 −60 −30 30 60 90 0 φ (deg) (b) FIG. 4. (Color online) Klein tunneling in MLG through a pn junc- tion with a (a) sharp and (b) smooth interface. (a) Comparison be- tween TBM (dashed line) and Eq. (8) [solid green (gray)] showing perfect agreement (EF = 80meV). (b) Comparison between TBM (long and short dashed) and Eq. (11) [solid green (light gray) and red (dark gray)] for kF d ≈ 6.16 (EF = 200meV) and kF d ≈ 1.54 (EF = 50meV), respectively. take spin also into account, the maximum of the transmission function will be 4. The transmission probabilities calculated by the Dirac theory always have their maximum of 1 due to the normalized incoming wave, unless a proper degeneracy factor is taken into account. 2. Klein tunneling in MLG: Sharp vs smooth interface Tunneling in MLG through a pn junction exhibits probabil- ity one at normal incidence and is called Klein tunneling. In experiments, a graphene pn junction can be realized by using a backgate, which tunes the carrier density (and hence the Fermi level) globally, and a topgate that tunes locally the carrier den- sity, equivalent to the potential step V0 at the other side.40 The carrier densities on the two sides can be controlled to be of opposite signs, forming the pn junction. In between, however, the variation of the carrier density is never abrupt in reality. Cheianov and Fal'ko showed, based on the Dirac theory, that the interface of the pn junction actually matters.24 They con- sidered symmetric pn junctions (i.e., V0 = 2EF) with sharp and linearly smooth interfaces, which we briefly review and compare with our TBM results in the following. a. Sharp interface For a symmetric pn junction with a sharp interface [see the schematic in Fig. 4(a)], the transmis- sion probability as a function of f was written as24 T (f ) = cos2 f , (8) which does not depend on the potential step height. This sur- prisingly simple expression matches our TBM result always perfectly as long as V0 = 2EF, as shown in Fig. 4(a). For a step potential with arbitrary height V0 6= 2EF, the transmission probability as a function of the incident angle f and the outgoing angle q can be derived as T (f ,q ) = 2 cosf cosq 1 + cos(f + q ) , (9) which agrees with our TBM calculation equally well as the symmetric case (not shown). The two angles f and q are connected to each other due to conservation of transverse mo- mentum by sinq = s EF EF −V0 sinf , (10) c with f where s = +1 for nn′ or pp′ and −1 for np or pn. Equation (9) clearly recovers the symmetric pn junction case of Eq. (8) when choosing s = −1 and V0 = 2EF in Eq. (10). Note that in the case of EF −V0 < EF, the Fermi wave vector in the outgoing region is shorter than that in the incoming region, and an additional constraint for f has to be applied to ensure sinq ≤ 1 [i.e., f ≤ f c = sin−1(EF −V0 /EF)]. Previously it has been stated that the single-valley Dirac picture, based on which Eqs. (8) and (9) are derived, is not equivalent to the TBM.41 The difference in their work, how- ever, becomes noticeable only when the distance between one of the involved energies and the Dirac point exceeds roughly 300 meV. In our simulation, indeed the deviation for the sym- metric pn junction case with, say EF = 300 meV, is less than 0.5%. The agreement of our TBM and the Dirac theory there- fore confirms that the intervalley scattering, which is mainly responsible for the nonequivalence at high energies, is indeed negligible. b. Smooth interface For symmetric pn junctions with a linearly varying region of width d [see the schematic in Fig. 4(b)], the analytical derivation for the transmission probability within the Dirac theory yields24 T (f ) = exp(cid:18)−p kF d 2 sin2 f (cid:19) (11) for kF d ≫ 1.42 This formula, together with the validity cri- terion kF d ≫ 1, are tested by our tight-binding calculations shown in Fig. 4(b), where two sets of parameters are consid- ered. For kF d ≈ 6.16 we find very good agreement with Eq. (11), while the result for kF d ≈ 1.54 exhibits noticeable devi- ations from the analytical prediction at large angles f . The smoothing function was assumed in their work as linear but the reality might be much more complicated, which is then not accessible by the Dirac theory but again straightforward by our tight-binding calculation. Nevertheless, the exponen- tial form of Eq. (11) is still a good description regardless of the actual form of the smoothing function, as we have numer- ically checked. What really matters is only the product kF d. Unlike the sharp pn interface, a compact form of transmis- sion probability for the asymmetric case does not exist so far. B. pnp junction: BLG vs MLG+R We next show the direct correspondence between BLG and MLG+R by considering exactly the same potential barrier 1 = 0.39 eV for MLG+R here. and incident Fermi energy as in Fig. 3(d) for BLG, and set 3tR = g (A discussion with weaker, realistic tR will be continued in the next section.) The total transmission shown in Fig. 5 for MLG+R indeed resem- bles the curves in Fig. 3(d) for BLG, as expected due to the identical form of their low-energy dispersions (1) and (2). The most important feature of chiral tunneling in BLG, forbidden normal transmission, now appears also in the case of MLG+R. In BLG, T (f = 0) = 0 was understood as the consequence of pseudospin conservation. For MLG+R, T (f = 0) = 0 can be expected as the consequence of real-spin conservation. In- deed, this can be demonstrated by computing the nonequilib- rium local spin density, which can be obtained from the lesser Green's function,43 considering two cases, 0 < EF < 3tR and −3tR < EF < 0, both with ky = 0. Within this single-band transmission, the local spin densities for positive and negative EF point to opposite directions, indicating that normal inci- dence transmission between n and p regions will be forbid- den. Next we discuss the spin-resolved transmission. The quan- tization axis is chosen as the out-of-plane direction, so that the transmission of, for example, T↓↑ means the probability of an incoming +Sz electron ending up as an outgoing −Sz one. Since the incoming angle dependence f of the transmis- sion probabilities are analyzed, we define T↑ = T↑↑ +T↓↑ as the transmission ability of the +Sz electron (or ↑ spin), and vice versa. (Alternatively, one can also analyze the outgoing angle dependence and define T↑ as T↑↑ + T↑↓, not used here. Either way, the total transmission (cid:229) ′ = T↑ + T↓ = T is ensured.) The choice of quantization axis z is not necessary but facil- itates relating the present spin-dependent tunneling in MLG with the issue of intrinsic spin-Hall effect previously dis- cussed in semiconductors. The spin-resolved transmission curves shown in Fig. 5 exhibit opposite lateral preference of ′=↑,↓ Tss s ,s Total T 60 30 0 −30 −60 V0 = 48.7 meV V0 = 100.7 meV 90 2 1.6 1.2 0.8 0.4 0 0.4 0.8 1.2 1.6 2 −90 90 2 1.6 1.2 0.8 0.4 0 0.4 0.8 1.2 1.6 2 −90 Spin-resolved T↑ and T↓ V0 = 48.7 meV 60 30 0 −30 90 2 1.6 1.2 0.8 0.4 0 0.4 0.8 1.2 1.6 V0 = 100.7 meV 60 30 0 −30 −60 T↑ = T↑↑ + T↓↑ T↓ = T↓↓ + T↑↓ 2 −90 −60 T↑ = T↑↑ + T↓↑ T↓ = T↓↓ + T↑↓ FIG. 5. (Color online) (a) Angle-resolved total transmission T for tunneling through a pnp junction in MLG+R with the same barrier height V0, barrier width D, and Fermi energy EF as used in Fig. 3(d) for BLG, and a substitution 3tR = g 1 = 0.39eV. (b) and (c) show spin-resolved transmission probabilities for V0 = 48.7meV and V0 = 100.7meV, respectively. 5 EF 2 1 T 0 −240 −30 60 150 V0 (µeV) 360 (Color online) Transmission T at normal incidence (ky = FIG. 6. 0) as a function of potential step height V0 for tunneling through a pn junction in MLG+R. The leftmost solid band diagram above the main panel corresponds to the incoming n side. The five ticks on the V0 axis correspond to the above five dashed band diagrams for the outgoing side. the ↑ and ↓ electron spins, which is an intrinsic spin-Hall mechanism due to the Rashba SOC. In a semiconductor two- dimensional electron gas (i.e., a continuous system rather than discrete as in the TBM), such an intrinsic spin-Hall deflection of opposite Sz electrons can be easily explained by the con- cept of a spin-orbit force based on the Heisenberg equation of motion,44,45 Fso = m i¯h(cid:20) 1 i¯h [r, H ] , H (cid:21) = 2ma 2 R ¯h3 (p× ez)s z. (12) Here H = p2/2m + (a R/¯h)(pys x − pxs y) is the continuous two-dimensional Hamiltonian with Rashba SOC, r and p are the position and momentum operators, a R is the Rashba cou- pling parameter (rather than the hopping one, tR), and s z is the sign of the Sz spin component. The T↑ and T↓ curves shown in Fig. 5 therefore reveal a combined effect of forbidden normal transmission due to conservation of real spin and the intrinsic spin-Hall deflection that can be understood by Eq. (12). A few remarks are due before we move on. To connect BLG with MLG+R we put 3tR = g 1 = 0.39 eV, which is ap- parently far from reality. In general the Rashba splitting in- duced by electrical gating is roughly of or less than the order of 100 m eV (see Sec. IV). Fermi energy lying within this split- ting, which is also our main interest, projects to a much shorter Fermi wave vector kF , leading to a much longer d up to a few or a few tens of microns in order for kF d ≫ 1 to be valid. This implies that the influence of the interface on the tunneling in MLG+R is normally negligible, unless d is that long. In addi- tion, tunneling through a pnp junction will also require a long barrier width D for electrons subject to such a short kF; other- wise, the barrier is merely a weak perturbation to the electron due to its long Fermi wave length. Based on these remarks, we will focus in the next section only on pn junctions in MLG+R with a reasonable Rashba hopping parameter. (cid:176) (cid:176) (cid:176) (cid:176) (cid:176) (cid:176) (cid:176) (cid:176) (cid:176) (cid:176) (cid:176) (cid:176) (cid:176) (cid:176) (cid:176) (cid:176) (cid:176) (cid:176) (cid:176) (cid:176) (cid:176) 6 C. pn junction in MLG+R (EF , tR, V0) = (60, 30, 0) µeV T↓ T↑ 0 −30 30 In the following we demonstrate in detail the role of Rashba SOC in tunneling through a potential step in MLG+R. The Rashba hopping parameter will be fixed to tR = 30 m eV and the Fermi energy in most cases to EF = 2tR, which lies within the spin-orbit splitting 3tR (see Fig. 1). EF V (x) −60 L R −90 2 1.6 1.2 0.8 0.4 0 0.4 0.8 1.2 1.6 2 90 (a) T 60 T 60 1. Normal incidence (EF , tR, V0) = (60, 30, 100) µeV T↓ T↑ 0 −30 30 We begin with the case of normal incidence, ky = 0. In Sec. III B we have discussed the one-band transmission selection rule (i.e., n ↔ p transmission is forbidden). The transmission from the left side at Fermi energy 0 < EF < 3tR to the right side with potential V0 is expected to be zero whenever a single- band n → p transmission is attempted. Indeed, as shown in Fig. 6, a zero transmission gap of T as a function of V0 is found. The gap lies in the interval of EF < V0 < EF + 3tR, cor- responding to the single-band n → p transmission. Note that contrary to the valley-valve effect in zigzag nanoribbons,46 -- 48 the gap shown here arises solely due to a bulk property. 2. Angle- and spin-resolved transmission We proceed with angle- and spin-resolved transmission and consider first the trivial case with EF = 0.5 meV well above the Rashba splitting 3tR = 90 m eV, as shown in Fig. 7. In this case the maximum of T = T↑ + T↓ is 4 since two spin subbands and two valleys are involved in transport. The to- tal transmission curve resembles the expected cos2 f behavior as discussed in Sec. III A 2, showing that the Rashba effect plays only a minor role. The spin-resolved T↑ and T↓ curves differ only slightly at f = sin−1(kin F ) ≈ 56◦, where kin F and kout F are the inner and outer radius of the two concentric Fermi circles, respectively. Tunneling in BLG with EF well above g 1 in BLG no longer plays an important role in the process of chiral tunneling when the transport occurs at EF ≫ g 1), as we have numerically checked. In other words, the chiral tunneling in 1 behaves similarly (i.e., the interlayer coupling g F /kout (EF , tR, V0) = (0.5, 0.03, 1) meV 4 3 2 1 ) φ ( T 0 −90 EF V (x) L R T↑ T↓ T −60 −30 0 30 60 90 φ (deg) FIG. 7. (Color online) Angular dependence of total (T ) and spin- resolved (T↑ and T↓) transmissions for EF = 0.5meV well above the Rashba splitting 3tR = 90m eV. EF V (x) −60 L R −90 1.2 0.8 0.4 0 0.4 0.8 90 1.2 (b) FIG. 8. (Color online) Angular dependence of total and spin-resolved transmissions through a pn junction in MLG+R with (a) zero poten- tial and (b) finite potential. Parameters used are given above. the Klein tunneling behavior as in MLG. BLG with EF ≫ g 1 and in MLG+R with EF ≫ 3tR recovers Of particular interest is the nontrivial case with EF < 3tR. As a test, we first consider V0 = 0 as shown in Fig. 8(a). In the absence of the potential step, the total transmission function T reaches its maximum of 2 (one spin subband times valley degeneracy of two) for any angle f , as it should. The opposite lateral deflection tendency of the ↑ and ↓ spins is again clearly seen and can be explained based on Eq. (12) as discussed in Sec. III B. The most important case is that of Fermi energy EF ∈ (0,3tR) and potential height V0 ∈ (EF ,EF + 3tR). A specific example with V0 = 100 m eV is shown in Fig. 8(b), which ex- hibits the combined effect of the forbidden normal transmis- sion [T (f = 0) = 0] and spin-Hall deflection. The number of high transmission peaks is always two.49 Compared to the previous trivial case (EF > 3tR, Fig. 7) where T↑ and T↓ do not significantly differ, the separation of the opposite ↑ and ↓ spins is distinctly enhanced. Whether this could be a new type of intrinsic spin-Hall mechanism in graphene deserves a further investigation, and is left as a possible future direction. We summarize the discussion of angle- and spin-resolved transmission by mapping T (f ,V0) in Fig. 9. Four different transport regimes can be identified: 1. V0 < 0, single n band to single/multiple n band(s) trans- mission regime. 2. 0 < V0 < EF, single n band to single n band trans- mission regime; distinct spin-resolved T↑ and T↓, and high total T limited by a critical angle f c = sin−1(EF −V0 /EF). 3. EF < V0 < EF + 3tR, single n band to single p band transmission regime; combined effects of forbidden (cid:176) (cid:176) (cid:176) (cid:176) (cid:176) (cid:176) (cid:176) (cid:176) (cid:176) (cid:176) (cid:176) (cid:176) (cid:176) (cid:176) T 2 T↓ T↑ 7 1 = + 0 30 60 90 0 −90 −60 −30 0 30 60 90 −90 −60 −30 0 30 60 90 180 150 120 90 60 30 0 −30 ] V e µ [ 0 V −60 −90 −60 −30 φ (deg) φ (deg) φ (deg) FIG. 9. (Color online) Transmission through a pn junction in MLG+R as a function of incident angle f and potential step height V0. Four transmission regimes can be distinguished: (i) V0 < 0, (ii) 0 < V0 < EF , (iii) EF < V0 < EF + 3tR, and (iv) V0 > EF + 3tR, with EF = 60m eV and 3tR = 90m eV. normal transmission and spin-Hall deflection. 4. V0 > EF + 3tR, single n band to multiple p bands trans- mission regime. Note that a vertical scan in Fig. 9 at f = 0 corresponds to Fig. 6, and horizontal scans at V0 = 0 and V0 = 100 m eV to Figs. 8(a) and 8(b), respectively. These four regimes will be helpful in the following discussion of conductance. 3. Integrated conductance Finally, we calculate the conductance of the pn junction in MLG+R by integrating T (f ), or equivalently, T (ky), with re- spect to the transverse Bloch momentum, G = e2/h 2kF Z kF −kF T (ky)dky, (13) where the prefactor ensures the maximal value of the Landauer-Buttiker type ballistic conductance to be e2/h times 4 3 2 1 ) h / 2 e ( G 0 EF EF + 3tR 2 1 MLG (tR = 0) MLG+R (tR = 30 µeV) 0 300 0 −180 −120 −60 0 60 120 180 240 V0 (µeV) FIG. 10. (Color online) Integrated conductance of the ballistic pn junction in MLG with tR = 0 and MLG+R with tR = 30m eV. the maximal number of modes.37 We compare the conduc- tance of the pn junction in MLG (tR = 0) and in MLG+R (tR = 30 m eV) as a function of the potential step height V0, as shown in Fig. 10. Since the Fermi level is fixed to EF = 60 m eV for both cases, the transport for tR = 0 will involve two spin and two valley degeneracies, leading to the maximal G of 4e2/h, while in the case of tR = 30 m eV only one spin subband is pro- jected, leading to the maximal G of 2e2/h. The maximal G occurs always at V0 = 0 that corresponds to an ungated clean bulk graphene. Zero conductance, on the other hand, occurs at V0 = EF since no states at the outgoing region are available at this charge neutrality point. Different transmission regimes can be distinguished based on our previous discussion for Fig. 9. For V0 ∈ [0,60] m eV (n → n transmission), the rise of V0 shrinks the Fermi circle at the outgoing region and hence introduces a critical transverse momentum, outside which the transmission is suppressed due to the lack of out-going states. The critical transverse mo- mentum reduces linearly with V0 for MLG due to the linear dispersion. The conductance G, Eq. (13), therefore reduces also linearly with V0. In the presence of the Rashba SOC, the low-energy dispersion becomes quadratic, and so does the re- duction of G with V0 in MLG+R. For V0 ∈ [60,150] m eV (n → p transmission), the conduc- tance of MLG rises faster than that of MLG+R, possibly due to the help of Klein tunneling. At V0 = 150 m eV, a sudden jump (or a shoulder) occurs in the case of MLG+R since the second spin subband at the outgoing region starts to partici- pate in transport. This jump does not occur in the MLG case since both spin subbands are always degenerate. An earlier re- lated work based on Dirac theory considered both intrinsic and Rashba SOCs.18 The V0 dependence of G for the Rashba dom- inated case in that work agrees well with the MLG+R curve shown in Fig. 10, including the shoulder. 8 IV. EXPERIMENTAL ASPECTS the nn and np regimes. A. Rashba spin splitting in graphene Whereas the Rashba spin splitting in MLG induced by an applied electric field is in general in the order of no more than 100 m eV, which is beyond the present resolution of angle- resolved photoelectron spectroscopy (ARPES), direct experi- mental observation of the Rashba spin splitting at K and K′ in agreement with the first-principles calculations9,10 is so far not reported. An earlier experiment on epitaxial graphene layers on a Ni(111) surface reported a large Rashba interaction50 up to 225 meV but was soon questioned since the splitting might simply reveal a Zeeman type splitting due to the ferromagnetic nature of nickel.51 An intercalated Au monolayer between the graphene layer and the Ni(111) substrate reduced the split- ting to about 13 meV and was concluded as the Rashba effect on the p states supported by spin-resolved ARPES.52 How- ever, the low-energy band structure of MLG+R at that time was not yet clear, and a simplified picture was adopted in the explanation of the measured spin splitting. In addition, trans- port properties of graphene based on metallic substrates can be difficult to isolate since a large bulk current will interfere as background.53 Throughout the above calculations we have mostly focused on a rather weak Rashba hopping parameter tR = 30 m eV, yielding a splitting at the K and K′ points 3tR = 90 m eV, which is a realistic and rather conservative estimate for the gate- voltage-induced Rashba SOC strength. A recent proposal of impurity-induced SOC in graphene,54 however, indicated that the coupling strength can be strongly enhanced by putting heavy adatoms55 as well as by hydrogenation.54,56 B. Klein tunneling in MLG Indirect and direct experimental evidences of Klein tunnel- ing in MLG have been reported recently.57,58. For detailed reviews, we refer to Refs. 2, 3, 20, 21, and 59. A very re- cent experiment on transport through a pnp junction in MLG used an embedded local gate, which yields high quality bal- listic transport and perfectly independent control of the local carrier density, as well as the feature of Klein tunneling.60 Recall the tR = 0 curve of conductance for MLG shown in Fig. 10. Overall, the conductance for n → n transmission with V0 < 0 is always higher than that for n → p transmission with V0 > EF. Even though Klein tunneling leads to perfect trans- mission at normal incidence in the latter case, the decay of T with incident angle eventually yields a lower conductance after integration. This feature has been agreed in recent ex- periments for pn and pnp junctions in MLG.40,57,58,60 -- 64 The difference of the conductance, or equivalently the resistance, between the nn and np (or between pp and pn) in experiments is even more obvious possibly due to the smooth interface that leads to an exponentially decaying form of T ,24 as we have re- viewed and discussed in Sec. III A 2. In fact, for MLG we have numerically checked G for pn junctions with a smooth inter- face, which indeed can enhance the difference of G between Another interesting feature so far experimentally reported only in Refs. 58 and 60 is the Fabry-Perot oscillation of the conductance for pnp junctions due to the interference between the two interfaces of the central barrier. This feature requires the system to be ballistic and can be naturally revealed by our tight-binding transport calculation, which we will elaborate elsewhere in the future. V. CONCLUSION AND OUTLOOK In conclusion, we have employed tight-binding calcula- tions to show that transport properties of MLG+R behave as BLG due to their identical form of the low-energy disper- sion, choosing the chiral tunneling in pn and pnp junctions as a concrete example. Within single-band transmission, nor- mal incidence transmission through a pn junction in BLG with EF < g 1 is forbidden as a consequence of pseudospin conservation,22 while in MLG+R with EF < 3tR this forbid- den transmission also occurs but as a consequence of real-spin conservation. In mapping the angle- and spin-resolved trans- mission for the MLG+R case, a combined effect of forbidden normal transmission and intrinsic spin-Hall deflection is re- vealed [Fig. 8(b)]. Compared to the potential-free spin-Hall deflection case as shown in Fig. 8(a), where T↑ = T↓ = 1 at f = 0, the effect of the pn junction seems to force the up and down spins to separate since T↑ = T↓ = 0 at f = 0. The fea- ture revealed in Fig. 8(b) may therefore suggest a new type of intrinsic spin-Hall mechanism in MLG. Within multiband transmission, however, the Rashba SOC in MLG no longer plays an important role when EF ≫ 3tR (Fig. 7). Likewise, the interlayer hopping g 1 in BLG becomes unimportant when EF ≫ g 1. Transport in both MLG+R with EF ≫ 3tR and BLG with EF ≫ g 1 recovers to that in MLG, despite the usually very different energy scales of 3tR and g 1. In view of the distinct transmission patterns in MLG+R with EF < 3tR [Fig. 8(b)] and EF ≫ 3tR (Fig. 7), as an inter- esting conjecture for the BLG case one expects very different scattering regimes for EF < g 1. The former is well discussed in the literature and exhibits strong scattering [Fig. 3(d)] while the latter is less discussed and the scattering is expected to be strongly suppressed. 1 and EF ≫ g MLG and BLG are known to behave quite differently in general, in the sense of single-band transmission. Whereas turning MLG directly into BLG is in principle not possible, steering MLG to MLG+R can be achieved simply by gating, and therefore the effect of Rashba SOC provides a possibility to continuously change the MLG-like transport properties to BLG-like. We expect further transport properties to behave similarly in BLG and in MLG+R, such as the quantum Hall effect,65 as was also noted by Rashba.14 ACKNOWLEDGMENTS We gratefully acknowledge Alexander von Humboldt Foundation (M.H.L.) and Deutsche Forschungsgemeinschaft (within SFB689) (J.B. and K.R.) for financial support. 9 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. H. Castro Neto, F. Guinea, N. M. R. Peres, K. S. Novoselov, and A. K. Geim, Rev. Mod. Phys. 81, 109 (2009). 3 S. Das Sarma, S. Adam, E. H. Hwang, and E. Rossi, Rev. Mod. 36 M. Wimmer, Quantum transport in nanostructures: From compu- tational concepts to spintronics in graphene and magnetic tunnel junctions, Ph.D. thesis, Universitat Regensburg (2008). 37 S. Datta, Electronic Transport in Mesoscopic Systems (Cambridge University Press, Cambridge, 1995). 38 B. K. Nikoli´c and R. L. Dragomirova, Semicond. Sci. Tech. 24, Phys. 83, 407 (2011). 4 D. D. Awschalom, D. Loss, and N. Samarth, eds., Semiconductor Spintronics and Quantum Computation (Springer, Berlin, 2002). 5 I. Zuti´c, J. Fabian, and S. Das Sarma, Rev. Mod. Phys. 76, 323 (2004). 6 A. K. Geim and K. S. Novoselov, Nat. Mater. 6, 183 (2007). 7 C. L. Kane and E. J. Mele, Phys. Rev. Lett. 95, 226801 (2005). 8 H. Min, J. E. Hill, N. A. Sinitsyn, B. R. Sahu, L. Kleinman, and A. H. MacDonald, Phys. Rev. B 74, 165310 (2006). 9 M. Gmitra, S. Konschuh, C. Ertler, C. Ambrosch-Draxl, J. Fabian, Phys. Rev. B 80, 235431 (2009). 10 S. Abdelouahed, A. Ernst, J. Henk, I. V. Maznichenko, and and I. Mertig, Phys. Rev. B 82, 125424 (2010). 11 F. D. M. Haldane, Phys. Rev. Lett. 61, 2015 (1988). 12 E. I. Rashba, Sov. Phys. Solid State 2, 1109 (1960). 13 Y. A. Bychkov and E. I. Rashba, JETP Lett. 39, 78 (1984). 14 E. I. Rashba, Phys. Rev. B 79, 161409(R) (2009). 15 M.-H. Liu and C.-R. Chang, Phys. Rev. B 80, 241304(R) (2009). 16 Due to a minor difference in the definition of the Rashba cou- pling in the tight-binding Hamiltonian, the splitting 3tR here cor- responds, e.g., to l in Ref. 14 and to 2l R in Ref. 9. 17 E. McCann and V. I. Fal'ko, Phys. Rev. Lett. 96, 086805 (2006). 18 A. Yamakage, K. I. Imura, J. Cayssol, and Y. Kuramoto, EPL 87 (2009). 19 P. Rakyta, A. Korm´anyos, and J. Cserti, Phys. Rev. B 82, 113405 (2010). 20 C. W. J. Beenakker, Rev. Mod. Phys. 80, 1337 (2008). 21 P. Allain and J. Fuchs, The European Physical Journal B - Con- 064006 (2009). 39 The unity transmission peaks (except the 0◦ peaks for MLG) are shifted by less than 3◦ compared to Fig. 2 of Ref. 22. 40 J. R. Williams, L. DiCarlo, and C. M. Marcus, Science 317, 638 (2007). 41 C. Tang, Y. Zheng, G. Li, and L. Li, Solid State Communications 148, 455 (2008). 42 Note that an additional factor of 1/2 in the exponent of Eq. (11) as compared to the original formula given in Ref. 24 comes from the fact that the linear potential profile across the interface changes from −V0 to V0 in Ref. 24, but here from 0 to V0, i.e., kF reduces to kF /2. 43 B. K. Nikoli´c, L. P. Zarbo, and S. Souma, Phys. Rev. B 73, 075303 (2006). 44 J. Li, L. Hu, and S.-Q. Shen, Phys. Rev. B 71, 241305 (2005). 45 B. K. Nikoli´c, L. P. Zarbo, and S. Welack, Phys. Rev. B 72, 075335 (2005). 46 K. Wakabayashi and T. Aoki, Int. J. Mod. Phys. B 16, 4897 (2002). 47 A. Rycerz, J. Tworzydlo, and C. W. J. Beenakker, Nat. Phys. 3, 172 (2007). 48 A. Cresti, G. Grosso, and G. P. Parravicini, Phys. Rev. B 77, 233402 (2008). 49 We have numerically checked that the double-peak feature of Ttot in the single-band n ↔ p transmission regime shown in Fig. 8(b) still holds even if the intrinsic SOC is present, as long as the Rashba coupling dominates. 50 Y. S. Dedkov, M. Fonin, U. Rudiger, and C. Laubschat, Phys. densed Matter and Complex Systems 83, 301 (2011). 22 M. I. Katsnelson, K. S. Novoselov, and A. K. Geim, Nature Physics 2, 620 (2006). 23 O. Klein, Zeitschrift fur Physik 53, 157 (1929). 24 V. V. Cheianov and V. I. Fal'ko, Phys. Rev. B 74, 041403 (2006). 25 L. M. Zhang and M. M. Fogler, Phys. Rev. Lett. 100, 116804 (2008). 26 E. B. Sonin, Phys. Rev. B 79, 195438 (2009). 27 C. Bai, Y. Yang, and X. Zhang, Physica E: Low-dimensional Sys- tems and Nanostructures 42, 1431 (2010). 28 E. Rossi, J. H. Bardarson, P. W. Brouwer, and S. Das Sarma, Phys. Rev. B 81, 121408 (2010). 29 J. M. Pereira Jr, F. M. Peeters, A. Chaves, and G. A. Farias, Semicond. Sci. Tech. 25, 033002 (2010). 30 T. Tudorovskiy, K. J. A. Reijnders, and M. I. Katsnelson, (2011), arXiv:1106.3042 [cond-mat.mes-hall]. 31 D. Bercioux and A. De Martino, Phys. Rev. B 81, 165410 (2010). 32 C. Bai, J. Wang, J. Tian, and Y. Yang, Physica E: Low- dimensional Systems and Nanostructures 43, 207 (2010). 33 M. Rataj and J. Barna´s, Appl. Phys. Lett. 99, 162107 (2011). 34 J. Schelter, D. Bohr, and B. Trauzettel, Phys. Rev. B 81, 195441 (2010). 35 S. Konschuh, M. Gmitra, and J. Fabian, Phys. Rev. B 82, 245412 (2010). Rev. Lett. 100, 107602 (2008). 51 O. Rader, A. Varykhalov, J. S´anchez-Barriga, D. Marchenko, and A. M. Shikin, Phys. Rev. Lett. 102, 057602 A. Rybkin, (2009). 52 A. Varykhalov, J. Sanchez-Barriga, A. M. Shikin, C. Biswas, E. Vescovo, A. Rybkin, D. Marchenko, and O. Rader, Phys. Rev. Lett. 101, 157601 (2008). 53 K. Yaji, Y. Ohtsubo, S. Hatta, H. Okuyama, K. Miyamoto, T. Okuda, A. Kimura, H. Namatame, M. Taniguchi, and T. Aruga, Nat. Comm. 1 (2010), 10.1038/ncomms1016. 54 A. H. Castro Neto and F. Guinea, Phys. Rev. Lett. 103, 026804 (2009). 55 C. Weeks, J. Hu, J. Alicea, M. Franz, and R. Wu, Phys. Rev. X 1, 021001 (2011). 56 D. C. Elias, R. R. Nair, T. M. G. Mohiuddin, S. V. Morozov, P. Blake, M. P. Halsall, A. C. Ferrari, D. W. Boukhvalov, M. I. Katsnelson, A. K. Geim, and K. S. Novoselov, Science 323, 610 (2009). 57 N. Stander, B. Huard, and D. Goldhaber-Gordon, Phys. Rev. Lett. 102, 026807 (2009). 58 A. F. Young and P. Kim, Nat. Phys. 5, 222 (2009). 59 A. F. Young and P. Kim, Annual Review of Condensed Matter Physics, 2, 101 (2011). 60 S.-G. Nam, D.-K. Ki, J. W. Park, Y. Kim, J. S. Kim, and H.-J. 10 Lee, Nanotechnology 22, 415203 (2011). 61 B. Huard, J. A. Sulpizio, N. Stander, K. Todd, B. Yang, and D. Goldhaber-Gordon, Phys. Rev. Lett. 98, 236803 (2007). 62 B. Ozyilmaz, P. Jarillo-Herrero, D. Efetov, D. A. Abanin, L. S. Levitov, and P. Kim, Phys. Rev. Lett. 99, 166804 (2007). 63 G. Liu, J. J. Velasco, W. Bao, and C. N. Lau, 92, 203103 (2008). 64 N. M. Gabor, J. C. W. Song, Q. Ma, N. L. Nair, T. Taychatanapat, K. Watanabe, T. Taniguchi, L. S. Levitov, and P. Jarillo-Herrero, Science 334, 648 (2011). 65 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). 

1108.3858

1

1108

Magneto-Optical Faraday and Kerr Effects in Topological Insulator Films and in Other Layered Quantized Hall Systems

We present a theory of the magneto-optical Faraday and Kerr effects of topological insulator (TI) films. For film thicknesses short compared to wavelength, we find that the low-frequency Faraday effect in ideal systems is quantized at integer multiples of the fine structure constant, and that the Kerr effect exhibits a giant $\pi/2$ rotation for either normal or oblique incidence. For thick films that contain an integer number of half wavelengths, we find that the Faraday and Kerr effects are both quantized at integer multiples of the fine structure constant. For TI films with bulk parallel conduction, we obtain a criterion for the observability of surface-dominated magneto-optical effects. For thin samples supported by a substrate, we find that the universal Faraday and Kerr effects are present when the substrate is thin compared to the optical wavelength or when the frequency matches a thick-substrate cavity resonance. Our theory applies equally well to any system with two conducting layers that exhibit quantum Hall effects.

cond-mat.mes-hall

cond-mat

Magneto-Optical Faraday and Kerr Effects in Topological Insulator Films and in Other Layered Quantized Hall Systems Wang-Kong Tse and A. H. MacDonald Department of Physics, University of Texas, Austin, Texas 78712, USA We present a theory of the magneto-optical Faraday and Kerr effects of topological insulator (TI) films. For film thicknesses short compared to wavelength, we find that the low-frequency Faraday effect in ideal systems is quantized at integer multiples of the fine structure constant, and that the Kerr effect exhibits a giant π/2 rotation for either normal or oblique incidence. For thick films that contain an integer number of half wavelengths, we find that the Faraday and Kerr effects are both quantized at integer multiples of the fine structure constant. For TI films with bulk parallel conduction, we obtain a criterion for the observability of surface-dominated magneto-optical effects. For thin samples supported by a substrate, we find that the universal Faraday and Kerr effects are present when the substrate is thin compared to the optical wavelength or when the frequency matches a thick-substrate cavity resonance. Our theory applies equally well to any system with two conducting layers that exhibit quantum Hall effects. PACS numbers: 78.20.Ls,73.43.-f,75.85.+t,78.67.-n I. INTRODUCTION Because topological insulators (TIs) have gapless he- lical surface states1–4 that respond strongly to time- reversal symmetry breaking perturbations, magneto- optical studies have emerged as an important tool for their characterization5–14. This paper expands on pre- vious work10,11 in which we demonstrated that ideal TIs exhibit striking unversal features in their long-wavelength response - a universal Faraday angle equal to the fine structure constant and a giant 90 degrees Kerr rotation. The present paper details the formalism used to obtain these results and generalizes the theory to new circum- stances motivated by current experimental activity. In particular, we include the influence of bulk conduction on the magneto-optical properties, analyze the role of the substrate material, and examine the case of oblique incidence of the light source. We also extend our theory to include thick TI films in which the electromagnetic wave can excite one of the cavity resonance modes of the film. Although we focus on the case of TI thin films, our results apply generically to systems containing two conducting layers that exhibit quantum Hall effects. When bulk conduction and surface longitudinal con- duction are both negligible, the magneto-optical proper- ties of a TI thin film can be elegantly characterized by adding a magneto-electric coupling term to the electro- magnetic Lagrangian to obtain topological field theory8. This description of magneto-electric properties shows that TIs provide a solid-state realization of axion elec- trodynamics, similar to those anticipated by Wilzcek in Ref. [15]. The topological field theory formulation of magneto-electric properties can be derived by inte- grating out the electronic degrees of freedom to obtain the magneto-electric polarizability of the bulk insulator, which is expressible as a Chern-Simons 3 form8,9. By ap- pealing to bulk time-reversal invariance considerations, it is possible to conclude that the coupling constant θ of the topological field theory8,9 is 0 mod(2π) for ordi- nary insulators and π mod(2π) for TIs. These possibili- ties correspond respectively to integer quantized surface Hall conductances in the case of ordinary insulators and to half-integer quantized surface Hall conductances11 in the case of topological insulators, providing a demonstra- tion of this important TI property. The topological field theory approach has a number of limitations, however, in describing real experiments because i) it does not ac- count for the surface longitudinal conductance which is never precisely zero at finite temperatures even when the quantum Hall effect is well established, ii) the surface Hall conductivity in topological field theory is ambigu- ous up to an integer multiple of e2/h, and iii) real thin film samples often have a finite bulk conductivity that is not readily incorporated. In addition, the thin film geom- etry normally used for magneto-optical studies requires seemingly artificial spatial profiles of the θ coupling con- stant, as discussed in the following paragraph. For these reasons we prefer to model the surface Hall conductivi- ties using a microscopic two-dimensional massless Dirac model for the TI surface states that is fully detailed be- low. The main disadvantage of our approach is that it captures the precise quantization of the DC surface Hall conductivity only when the massless Dirac model's ultra- violet cutoff is set to infinity. An important lesson from our approach is that topological field theory applies only when time-reversal symmetry breaking is strong enough to overcome disorder and establish a surface quantum Hall effect, and then only in the limit of temperatures and frequencies small compared to the surface gap in- duced by time-reversal symmetry breaking. The original discussion of Wilzcek15 imagined axion electrodynamics induced by a bulk spherical medium with a non-zero θ parameter separated from vacuum with θ = 0 by a single simply-connected surface. In the case of a spherical TI sample, this formulation can correctly capture the material's surface Hall conductivity. Simi- larly, the case of an ideal semi-infinite TI slab8 can be 2 FIG. 1: (Color online) (a). Schematic illustration of the Faraday and Kerr effects. Incident linearly polarized light becomes elliptically polarized after transmission (Faraday effect) and reflection (Kerr effect), with polarization plane angle rotations θF and θK respectively. (b) Fabry-Perot-like reflection and transmission in the TI film geometry. 'T' stands for the top surface and 'B' the bottom surface. described by a topological field theory model with θ (cid:54)= 0 in the topological insulator and θ = 0 in vacuum. In a magneto-optics setting, however, a propagating elec- tromagnetic wave necessarily interacts with two nearby surfaces, since a real TI thin film sample has both top and bottom surfaces. Assuming that the mechanism that breaks time-reversal invariance at the TI surface does so in the same sense on both top and bottom, both surfaces will have the same Hall conductivity. In topological field theory, this would be captured by a model in which the discontinuity in θ in the direction of light propagation has the same sign and magnitude at both surfaces. To describe a thin film with identical half-quantized Hall conductivities on opposite surfaces it is then necessary to take θ = ±2π in one of the vacuum regions. A TI thin film model in which the surrounding material has θ = 0 everywhere would instead describe a system with oppo- site Hall conductivities on opposite surfaces. Because of the opposite contributions, the magneto-optical effects would then vanish for films thinner than a wavelength. Given the surface massless Dirac models, our ap- proach to magneto-electric properties is more conven- tional. Magneto-electric properties are completely de- termined by the conductivity of the material, including bulk and surface, longitudinal and Hall contributions. The role of the theta-term in axion electrodynamics equa- tions is completely replaced by the appearance of explicit surface Hall conductivities that influence electromagnetic wave boundary conditions. These calculations make it clear that magneto-electric properties depend essentially on the numerical values of the surface conductivities, not just on whether they are quantized or half-quantized. This is especially important when surface time-reversal invariance is broken by an external magnetic field, since both the sign and the magnitude of the surface currents will be sensitive to the position of the Fermi level within the bulk gap. In addition to this advantage, the optical response of the surface Dirac fermions can be evaluated microscopically, enabling a natural incorporation of dy- namical and many-body effects. Interesting magneto-optical effects occur in both low- frequency and higher frequency regimes. In the latter case, we have found interband absorption10 and cyclotron resonance11 features in the Faraday and Kerr rotations that are dramatically enhanced by the cavity confine- ment effect of the TI thin film. We focus mainly on the large topological magneto-optical effects at low frequen- cies which can be observed only if the following condi- tions are satisfied: (1) The TI surface has a quantized Hall effect allowed by time-reversal breaking due to either external magnetic field or exchange coupling to external electronic degrees of freedom. We know from vast experi- ence with the quantum Hall effect in an external magnetic field in graphene, which is also described by a massless Dirac equation, that the quantum Hall effect can occur at quite weak magnetic fields when the Fermi level is close to the Dirac point and disorder is very weak. Exchange- coupling to spin yields a half-quantized anomalous Hall effect in the absence of disorder16,17. Although there are no experimental examples of quantized anomalous Hall effect as yet, the requirements on time-reversal break- ing perturbation strength in the presence of disorder are likely to be similar. The magneto-electric anomalies re- quire large Hall effects at finite electromagnetic wave fre- quency. This condition requires that the frequency be much smaller than the surface gap. (2) The dimension of the TI along the direction of light travel should be shorter than the electromagnetic wavelength or a nonzero integer multiple of the half wavelength. In either case, the di- electric properties of the TI bulk medium that separates the two quantized Hall layers do not influence the trans- mitted and reflected light. When these conditions are TBθFθKa bsatisfied, the magneto-optical effects are universal and topologically protected against weak surface disorder. The outline of our paper is as follows. In the Section II, we review linear response theory for the optical con- ductivity tensor of TI surface helical quasiparticles. In Section III, we summarize the electromagnetic scatter- ing formalism appropriate for a system with two metallic surfaces surrounded by dielectrics, applicable to TI films and to other similar layered systems. We then present and discuss our results for the magneto-optical Faraday and Kerr effects, first for TI films that are thinner than a wavelength and then for the general case. In Section V we address some issues pertinent to experiments includ- ing the influence of bulk conduction, the role of oblique incidence, and the role of substrates. Finally in Section VI we discuss the closely related magneto-optical prop- erties of graphene-based layered massless Dirac systems. II. DYNAMICAL RESPONSE OF A SPIN-HELICAL DIRAC FERMIONS We first derive explicit analytic expressions for the longitudinal and Hall optical conductivities of the spin- helical quasiparticles of topological insulator surfaces with time-reversal symmetry broken in two different ways: (1) an exchange field that couples to spin, and (2) an external magnetic field that couples to both spin and orbital degrees of freedom. The former case can be realized by exchange coupling between the TI surface and an adjacent ferromagnetic insulator8,10. The response of TI surface carriers to exchange coupling is unique, and ongoing progress along this direction has been reported by several experimental groups18–21. In the presence of time-reversal symmetry breaking, the massless Dirac Hamiltonians for the top (T) and bot- tom (B) surfaces are H = (−1)L [vτ · (−i∇ + eA/c) + V /2] + ∆τz, (1) where τ is the spin Pauli matrix vector (expressed in 90◦ rotated basis from the real spins22), A is the magnetic vector potential, ∆ is the Zeeman coupling strength, V accounts for a possible potential difference between top and bottom surfaces due to doping or external gates, and L = 0, 1 for the top (0) and bottom (1) surfaces. Note that in spite of the sign difference in the kinetic energy terms for the top and bottom surfaces, the conductivi- ties are identical on the two surfaces. The massless Dirac surface states description is valid for energies below the energy cut-off of the Dirac Hamiltonian εc, which we as- sociate with the separation between the Dirac point and the closest bulk band. A. Exchange Field 3 lating ferromagnet with magnetization oriented perpen- dicularly. Magnetic proximity coupling with strength ∆ will favor alignment of the TI surface spins. The numer- ical value of ∆ would be determined by the orbital hy- bridization between the ferromagnetic material and the TI surface. A = 0 in Eq. (1) in the absence of an applied magnetic field. In this limit, the Dirac cone is gapped with conduction (µ = 1) and valance (µ = −1) band dispersions and eigenstates (cid:20) C↑kµ (cid:21) kµ(cid:105) = εkµ = µεk, where εk = (cid:112)(vk)2 + ∆2, φk is the azimuthal angle of C↓kµeiφk (2) . the crystal momentum, and C↑kµ = sgn(µ)(cid:112)εk + sgn(µ)∆/ C↓kµ = (cid:112)εk − sgn(µ)∆/ 2εk, √ √ 2εk, (3) The optical conductivity tensor of the helical quasi- particles on topological insulator surface can be obtained from the Kubo formula: σαβ (ω) = ig fkµ − fkµ(cid:48) εkµ − εkµ(cid:48) (cid:104)kµjαkµ(cid:48)(cid:105)(cid:104)kµ(cid:48)jβkµ(cid:105) ω + εkµ − εkµ(cid:48) + i/2τs , (cid:88) (cid:88) µµ(cid:48) k (4) where α, β = {x, y}, µ, µ(cid:48) = ±1 denote the band index, fkµ is the Fermi factor for band µ, 1/τs is the quasi- particle lifetime broadening of the surface states, and g is the (odd) number of Dirac cones on the TI surface, which we take for simplicity and concreteness to be 1. This expression is not quantitatively correct when dis- order is present (i.e. when τs is finite), because it does not capture the localization physics that is important in the quantum Hall regime, but it is adequate for our present interest. We evaluate the longitudinal conduc- xx + iσI tivity σxx(ω) = σR xx and the Hall conductivity xy + iσI σxy(ω) = σR xy in the topological transport regime when the Fermi level lies within the TRS breaking gap. Since all surface optical conductivities eventually appear in the outgoing electromagnetic fields in the combination σαβ/c, we shall adopt the 'natural units' for the opti- cal conductivities and express σαβ in e2/ = α c units (α = 1/137 is the vacuum fine structure constant) and set c = 1 except where specified. For the dissipative components of the optical conductivity we find that: σR xx = σI xy = 1 16πx(ω2 + Γ2)2 ×(cid:8)x(cid:2)(ω2 + Γ2)2 + 4(ω2 − Γ2)∆2(cid:3) g(x) +4Γ∆2(cid:2)ω2 + Γ2 + ωxf (x)(cid:3)(cid:9)(cid:12)(cid:12)x=εc [−2ωg(x) − Γf (x)](cid:12)(cid:12)x=εc x=∆ ∆ x=∆ 8π(ω2 + Γ2) (5) Time-reversal symmetry breaking by an exchange field can be realized by interfacing the TI surface with an insu- For the reactive, non-dissipative components of the opti- cal conductivity σI xy, which are due to off-shell xx and σR virtual transitions, we find that: σI xx = σR xy = 1 ∆ 8π(ω2 + Γ2) 32πx(ω2 + Γ2)2 +x(cid:2)(ω2 + Γ2)2 + 4∆2(ω2 − Γ2)(cid:3) f (x)(cid:9)(cid:12)(cid:12)x=εc (cid:8)8ω∆2(ω2 + Γ2) − 16xωΓ∆2g(x) [−2Γg(x) + ωf (x)](cid:12)(cid:12)x=εc (cid:12)(cid:12)(cid:12)(cid:12) (ω − 2x)2 + Γ2 (cid:18) ω − 2x (cid:19) means R(x2) − R(x1), Γ = 1/2τs, and (cid:18) ω + 2x (ω + 2x)2 + Γ2 (cid:12)(cid:12)(cid:12)(cid:12) , (cid:19) x=x1 x=∆ x=∆ (6) − tan−1 Γ . (7) Γ f (x) = ln g(x) = tan−1 where R(x)x=x2 When the disorder broadening is small such that Γ (cid:28) ∆, it is useful to obtain analytic results in the disorder-free limit Γ → 0, in which case x=∆ = −πθ(ω − 2∆), (8) g(x)(cid:12)(cid:12)x=εc 4 In the n = 0 LL spins are aligned with the perpendicular field so that ε0 = −∆, 0(cid:105) = . (12) (cid:20) 00(cid:105) (cid:21) (cid:88) nn(cid:48) In the quantum Hall regime (ΩBτs (cid:29) 1 where ΩB = v/(cid:96)B is a characteristic frequency typical of the LLs spac- ing), the conductivity tensor can be expressed in LL basis as σαβ (ω) = ig 2π(cid:96)2 B fn − fn(cid:48) εn − εn(cid:48) (cid:104)njαn(cid:48)(cid:105)(cid:104)n(cid:48)jβn(cid:105) ω + εn − εn(cid:48) + i/2τs , (13) where the current operator is j = ie [H, x] = evτ . For convenience we rewrite the LL index as n = sm, where m = 0, 1, 2,··· Nc and s = ±1 for electron-like and hole- like LLs. The current matrix element (cid:104)njαn(cid:48)(cid:105) captures the selection rule n(cid:48) − n = ±1 for LL transitions. Af- ter some algebra we find that the conductivity tensor Eq. (13) in e2/ = αc units is given by Nc−1(cid:88) (cid:88) m=0 s,s(cid:48)=±1 fsm − fs(cid:48)(m+1) εsm − εs(cid:48)(m+1) Γs,s(cid:48) αβ (m, ω), (14) and Eqs. (5)-(7) reduce to our previous results [Eqs. (2)- (3) in Ref. 10] obtained from the quantum kinetic equa- tion approach. σαβ(ω) = v2 2π(cid:96)2 B B. External Quantizing Magnetic Field Landau level (LL) quantization of the TI's sur- face Dirac cones has recently been observed by STM experiments23,24. In the presence of a quantizing field, the vector potential in Eq. (1) is given in the Landau gauge by A = (0, Bx) and the Zeeman coupling by ∆ = √ gJµBB/2, where gJ is the electron g factor. Define raising and lowering operators a = ((cid:96)B/ B] and 2)[−∂x + (x + x0)/(cid:96)2 a† = ((cid:96)B/ the magnetic length and x0 = ky(cid:96)2 coordinate, Eq. (1) can be written as B] with (cid:96)B = 1/(cid:112)eB (cid:1) a (cid:21) B the guiding center −i(cid:0)√ 2)[∂x+(x+x0)/(cid:96)2 √ n(cid:105) = εn(cid:105), (9) (cid:20) i(cid:0)√ (cid:1) a† 2v/(cid:96)B −∆ ∆ 2v/(cid:96)B where ···(cid:105) denotes an eigenspinor. The LLs are labeled by integers n and for n (cid:54)= 0 and have eigenenergies [rela- tive to the respective Dirac point energies (−1)LV /2] and eigenspinors (cid:20)−iC↑nn − 1(cid:105) (cid:21) C↓nn(cid:105) .(10) εn = sgn(n) n + ∆2, n(cid:105) = 2v2 (cid:96)2 B where n(cid:105) without an overbar denotes a Fock state (n is the absolute value of n), and (cid:113) εn + sgn(n)∆/(cid:112)2εn, εn − sgn(n)∆/(cid:112)2εn. (cid:113) C↑n = sgn(n) C↓n = (11) (cid:115) (cid:26) i (cid:27) where Γs,s(cid:48) xy}(m, ω) = − {xx ω − εsm + εs(cid:48)(m+1) + i/2τs (cid:18) 1 1 C2↑s(cid:48)(m+1)C2↓sm ± 1 ω + εsm − εs(cid:48)(m+1) + i/2τs (15) (cid:19) . B(ε2 c − ∆2)/2v2 is the largest LL In Eq.( 15), Nc (cid:39) (cid:96)2 index with an energy smaller than the ultraviolet cut-off εc, prefactor i and sign '+' inside the parenthesis apply to the Γxx expression, 1 and '−' to the Γxy expression. Eqs. (14)-(15) express σαβ as a sum over interband and intraband dipole-allowed transitions which satisfy n(cid:48) − n = ±1. s = 0 limit Eq. (14) yields correct half-quantized plateau values for the Hall conductivity. In the ω = T = τ−1 III. LIGHT PROPAGATION THROUGH A TI SLAB In this section, we formulate the problem of electro- magnetic wave scattering through a topological insula- tor slab illustrated schematically in Fig. 1(b). Here we discuss only the normal incidence case. More general re- sults for the oblique incidence case are presented in an Appendix. Consider an electromagnetic wave propagating along the z direction through two materials, labeled by i and j, with dielectric constant and magnetic permeability
i, µi and
j, µj respectively. The interface between them is at z = ai. We write the electric field component of the electromagnetic field in the form (cid:21) (cid:20) Eti x Eti y (cid:21) (cid:20) Eri x Eri y Ei = eikiz + e−ikiz , (16) where the tilde accents denote vectors E = [Ex Ey]T, the superscripts 'r' and 't' on E denote the reflected and transmitted components of the electric field, and ki = (ω/c)
iµi is the wavevector in medium i. The corresponding magnetic field is given by Faraday's law as √ (cid:26) (cid:114)
i µi (cid:21) (cid:20) −Eti y Eti x (cid:21)(cid:27) (cid:20) Eri y−Eri x H i = eikiz + e−ikiz , (17) The electric and magnetic fields at the interface z = ai satisfy the electrodynamic boundary conditions Ei = Ej and −iτy( H j − H i) = (4π/c) J i, where τy is the Pauli matrix and J i = ¯σi Ei is the surface current density at z = ai. Note that this surface current can have longitu- dinal and Hall response components in this microscopic theory and not only Hall components as assumed in topo- logical field theory. The scattering matrix that relates incoming [ Eti Erj]T and outgoing [ Eri Etj]T fields at a conducting interface can be written in the form (cid:21) (cid:20) ¯r ¯t(cid:48) ¯t ¯r(cid:48) S = , (18) where the superscripts 'r' and 't' on E denote reflected and transmitted components of the electric fields, and are 2×2 reflection and transmission tensors, (cid:48) ¯r, ¯r which are of the form: (cid:48) and ¯t, ¯t (cid:20) rxx (cid:21) ¯r = rxy −rxy ryy , ¯t = (cid:21) (cid:20) txx txy −txy tyy , (19) : = × rxy (cid:48) , ¯t ei2kiai (cid:48) and ¯t . Matching boundary conditions, (cid:48) and similarly for ¯r (cid:48) we obtain the following expressions for ¯r, ¯t ¯r −8π(cid:112)
i/µiσxy ((cid:112)
i/µi +(cid:112)
j/µj + 4πσxx)2 + (4πσxy)2 (cid:26) rxx (cid:27) (cid:26)
i/µi − ((cid:112)
j/µj + 4πσxx)2 − (4πσxy)2 (cid:27) (cid:26) txx (cid:26) 2(cid:112)
i/µi((cid:112)
i/µi +(cid:112)
j/µj + 4πσxx) ((cid:112)
i/µi +(cid:112)
j/µj + 4πσxx)2 + (4πσxy)2 −8π(cid:112)
i/µiσxy ei(ki−kj )ai , (20) (cid:27) (cid:27) (21) × txy = . For normal incidence, the two diagonal elements of the reflection and transmission matrices are identical: r((cid:48)) yy = r((cid:48)) xx and t((cid:48)) xx. ¯r(cid:48) can be obtained from ¯r by making the replacement ki → −kj and interchanging
i/µi and yy = t((cid:48))
j/µj, and ¯t(cid:48) from ¯t by interchanging
i/µi and
j/µj. In addition, ¯t and ¯t(cid:48) are related by ¯t/(cid:112)
i/µi = ¯t(cid:48)/(cid:112)
j/µj. 5 Scattering from a TI film presents an electromagnetic problem in which scattering occurs from two interfaces at which currents flow and dielectric constants are dis- continuous. The reflection and transmission tensor can be composed from the single-interface scattering matrices ¯r((cid:48)), ¯t((cid:48)) for the top and bottom surfaces to obtain ¯r = ¯rT + ¯t(cid:48) ¯t = ¯tB (cid:0)1 − ¯r(cid:48) T ¯rB (cid:1)−1 ¯tT, (cid:0)1 − ¯r(cid:48) (cid:1)−1 ¯tT. T ¯rB T ¯rB (22) (23) The presence of a dielectric substrate underneath the TI film is easily accounted for by propagating the reflection and transmission tensors Eqs. (22)-(23) through an ad- ditional layer of dielectric. Detailed expressions for the reflection and transmission tensors that allow for oblique incidence and account for a dielectric substrate are given in an Appendix. IV. MAGNETO-OPTICAL FARADAY AND KERR EFFECTS For linearly polarized incoming light, the Faraday and Kerr angles can be defined in terms of the relative rota- tions of left-handed and right-handed circularly polarized light to obtain: θF = (cid:0)arg{Et θK = (cid:0)arg{Er +} − arg{Et−}(cid:1) /2, +} − arg{Er−}(cid:1) /2, (24) (25) x ± iEr,t y where Er,t± = Er,t are the left-handed (+) and right-handed (-) circularly polarized components of the outgoing electric fields. In this section, we present our results for an ideal topo- logical insulator under normal light incidence. We then discuss non-ideal effects that may be relevant in exper- imental situations in Section V. It is important to em- phasize that the magneto-optical effects are essentially the same in this limit for time-reversal symmetry broken by exchange coupling or by a quantizing magnetic field. In the case of a quantizing field, there are many gaps in the surface spectrum because of Landau quantization. The quantized Hall conductivity in e2/h units is equal to the filling factor νT,B. The largest gap in the magnetic field case occurs occurs at νT,B = 1/2 and has the same Hall conductivity as for the Zeeman gap case. In the magnetic field case, it is possible to shift the Hall con- ductivities of either surface by integer multiples of e2/h away from e2/2h simply by shifting the position of the Dirac point relative to the chemical potential, and this shift would influence the magneto-optical effects. When the chemical potential is placed in the largest gap in the magnetic field case, the only differences between the two scenarios are in the details of the higher frequency re- sponse. A. Thin Film d (cid:28) λ and the Kerr angle First we consider the case of a TI film that is thinner than the light wavelength. In this limit it follows from Faraday's law that the electric field is spatially constant across the film so that the two interfaces can be consid- ered as one. Amp`ere's Law implies that the magnetic field changes by a value proportional to the current inte- grated across the TI film, − iτy( H T − H B) = (4π/c) (¯σT + ¯σB) E, (26) where H T,B denotes the magnetic fields in the top and bottom vacuum regions outside of the film. Eq. (26) says that, from the viewpoint of the long electromag- netic wave, the TI film behaves effectively as a single two-dimensional surface with a conductivity equal to the conductivities integrated across the film. We therefore obtain the transmitted and reflected fields Et = Er = 1 (2 + 4πσxx)2 + (4πσxy)2 1 (2 + 4πσxx)2 + (4πσxy)2 (cid:20)1 − (1 + 4πσxx)2 − (4πσxy)2 8πσxy (cid:20)4 (1 + 2πσxx) (cid:21) (cid:21) 8πσxy ,(27) , (28) for simplicity here we use σxx, σxy to denote the total longitudinal and Hall conductivities from both surfaces, respectively. An important observation is that in this limit the transmitted and reflected fields are indepen- dent of the bulk dielectric properties of the TI film. For weak disorder and frequencies much smaller than charac- teristic transition frequencies (ω (cid:28) ∆ for the exchange field case and ω (cid:28) ΩB for the magnetic field case), the optical conductivity has only a dissipationless DC Hall conductivity contribution: σR xy = (29) , νT,B 2π xy = 0. xx = σI xx = 0, σI and σR In this low-frequency regime, the magneto-optical response of the exchange field case becomes a special case of the quantizing mag- netic field case with νT = νB = 1/2 as explained above. From Eqs. (27)-(28) we find the Faraday and Kerr an- gles θF = tan−1 [(νT + νB) α] , (cid:20) (cid:21) θK = − tan−1 1 (νT + νB) α . (31) For total filling factor νT + νB values that are not too large, the Faraday angle is quantized in integer multiples of the fine structure constant θF (cid:39) (νT + νB)α, (32) 6 (33) θK = − π 2 , becomes a full quarter polarization rotation. It is also possible to understand Eq. (33) in terms of the scattering mechanism of the reflected partial wave components. This understanding is crucial to see that Eq. (33) applies over a finite frequency range, as dis- cussed in Section V B. The reflected electric field can be easily found from Eq. (22). The algebra is simplified and the physics underlying Eq. (33) more easily illustrated when spatial-inversion symmetry across the TI film is preserved, i.e. when top and bottom surfaces have the same conductivities. This happens when the exchange fields for both surfaces are the same, or in the quantizing magnetic field case, when the surface densities (and thus filling factors) are the same. Spatial-inversion symmetry then implies that ¯rB = ¯rT , and ¯tT = ¯tB . This allows the reflected electric field to be ex- pressed solely in terms of the reflection and transmission matrix elements of one (e.g., the top) of the two surfaces. The first term on the right hand side of Eq. (22) repre- sents the partial wave directly reflected from the top sur- face, which we can evaluate in the low-frequency regime as (cid:48), ¯rT = ¯rB (cid:48), ¯tB = ¯tT (cid:48) (cid:48) Er0 = (cid:16) (cid:17)2 1 +(cid:112)
/µ 1 +(cid:0)4πσR xy (cid:1)2 (cid:20)1 −
/µ −(cid:0)4πσR (cid:1)2 (cid:21) xy 8πσR xy , (34) where
, µ (cid:39) 1 are the bulk dielectric constant and mag- netic permeability of the TI, and the second term consti- tutes all the partial waves that originate from successive reflections from the bottom surface Er(cid:48) = 1 (cid:20) (cid:1)2 1 +(cid:0)4πσR (cid:17)2 (cid:16) 1 +(cid:112)
/µ xy 4πσR (cid:21) +(cid:0)4πσR xy 1 1 − xy (cid:20)2 (1 +
/µ) 8πσR xy (35) (cid:21) . (cid:1)2 (30) Er = . (36) Summing the two contributions, we find that part of the second term on the right-hand side of Eq. (35) cancels out the first term on the right-hand side of Eq. (34) com- pletely, yielding a total reflected field 1 +(cid:0)4πσR 1 xy (cid:1)2 (cid:20)−(cid:0)4πσR (cid:1)2 (cid:21) xy 4πσR xy Eq. (36) implies that the reflected partial waves that orig- inate from successive scattering off the bottom surface destructively interfere with the partial wave scattered off the top surface, resulting in a suppression by a factor ∼ (σR xy)2 of the reflected electric field component along the incident polarization direction. This leads to the gi- ant Kerr angle in Eq. (33). It is worthwhile to make clear that the large Kerr angle occurs because almost all of the reflected partial waves have a 90◦ rotated polar- ization plane; it is not true, however, that almost all of the light is reflected. B. Thick Film d (cid:38) λ In the previous section, we have focused on films with a thickness that is only a fraction of the wavelength. In this section, we shall relax this assumption and generalize our considerations to thicker films, with thickness com- parable to or greater than the wavelength inside the film. Thick films do not in general show spectacular magneto- optical effects because the Faraday and Kerr angles are suppressed by the large dielectric constant of the TI bulk. Exceptions occur when the film thickness contains an in- teger multiple of half wavelengths inside the film, i.e. when the cavity resonance condition kTId = N π is sat- isfied. Here kTI =
µ ω/c is the wave number in the TI film and N (cid:54)= 0 is an integer. This property was first identified in Ref. [13], however the discussion there as- sumed an infinitely thick dielectric substrate underneath the TI film, neglecting scattering from the inevitable substrate-vacuum interface and thereby overestimating substrate suppression of magneto-optical responses. In this section, we first consider a free-standing thick film. We will then study the influence of a substrate, with its finite thickness properly accounted for, in Section V. √ At resonance, a standing wave is established inside the film with the tangential components of the electric and magnetic fields on the interior of the top and bottom surfaces inside the film related simply by a ± sign, i.e. E(cid:107)(z = −d/2+)/E(cid:107)(z = d/2−) = (−1)N (here z = 0 is taken at the center of the film), and similarly for H. Under such circumstances, the transmitted and re- flected electric fields become independent of the film's bulk dielectric properties, and are found to be given by Eqs. (27)-(28) multiplied by a phase factor e−ik0d, where k0 = ω/c is the vacuum wave number. In contrast to the long-wavelength regime we considered earlier in which the electromagnetic field varies slowly across the TI film, at resonance the field amplitudes change rapidly inside the film and the adiabatic condition k0d (cid:28) 1 does not apply. Regardless of the film thickness, however, the adi- abaticity of TI surface electronic response can always √ be established at a sufficiently low frequency satisfying ωN (cid:28) ∆ or ΩB [ωN = N πc/(
µ d) is the cavity reso- nance frequency], such that the quantum Hall condition Eq. (29) still holds. It follows from these considerations that the phases of the left-handed and right-handed cir- cularly polarized components of the transmitted and re- flected light are given by arg(Et λ) = arg(Er λ) = −λ sin(k0d) + 2π cos(k0d)σR λ cos(k0d) + 2π sin(k0d)σR cos(k0d) + λ2π sin(k0d)σR sin(k0d) − λ2π cos(k0d)σR xy , xy xy xy , (37) (38) 7 where λ = ±1 labels the left and right-handed circularly polarized light, and σR xy contains the sum of the top and bottom surface Hall conductivities. Let us make several remarks here. If we set k0d → 0, Eqs. (37)-(38) coincide with the long-wavelength (k0d < kTId (cid:28) 1) results, from which we recover Eqs. (30)-(31) for the Faraday and Kerr angles. In general, if in addition to the requirement kTId = N π for resonance we also have k0d = M π, then Eqs. (37)-(38) would imply that θF = tan−1 [(νT + νB) α] and θK = − tan−1 [1/ (νT + νB) α] for integer M , and θF = − tan−1 [1/ (νT + νB) α] and θK = tan−1 [(νT + νB) α] for half-odd integer M . These conditions would require
µ = N/M . With real materi- als this would seem to be rather impossible, however with the advent of metamaterials it may be possible to engi- neer one with a matching dielectric constant (N/M )2, and employ it as an intervening dielectric between two single-layer graphene sheets. This point will be discussed further in Section VI. In this light, we see that the long- wavelength limit k0d < kTId → 0 is special because it automatically satisfies both conditions kTId = N π and k0d = M π with N = M = 0. √ When k0d is not equal to a multiple of integer or half- odd integer of π, which is generally the case, only the cavity resonance condition is satisfied and k0d cannot be assumed as small in Eqs. (37)-(38). Evaluating θF and θK from Eqs. (24)-(25) using Eqs. (37)-(38), we find that the Faraday and Kerr rotations at resonance have the same universal quantized value θF,K = tan−1 [(νT + νB) α] . (39) Note that Eqs. (37)-(39) do not depend on the value of N and therefore all cavity resonant modes yield the same Faraday and Kerr rotations given by Eq. (39). It is important to emphasize, at resonance, that although the Faraday angle is the same as the long-wavelength re- sult Eq. (30), the Kerr angle is not given25 by Eq. (31). Rather, both the Faraday and Kerr rotations at reso- nance are given by the same quantized response in units of α like Eq. (30). The giant Kerr effect Eq. (33), there- fore, is a unique long-wavelength low-frequency property of the thin film system only. V. DEVIATIONS FROM AN IDEAL TOPOLOGICAL INSULATOR FILM Magneto-optical measurements of Faraday and Kerr rotations produced by topological insulators5–7 subjected to an external magnetic field have been performed by several groups. The samples studied include bulk Bi2Se3 crystals5, thin Bi2Se3 films6 and strained HgTe films7. It has not yet been possible to achieve ideal samples in which the half-quantized quantum Hall effect occurs, ei- ther in DC transport or optically. In this section we con- sider non-ideal factors that often arise in experimental TI samples, and explain their consequences when they act independently. We focus on the influence of bulk carriers, light scattering at oblique incidence, and the influence of a dielectric substrate. For the Kerr effect, Eq. (43) implies a stricter condition for negligible bulk conduction: 8 A. Influence of Bulk Carrier Conduction Real TI samples are complicated by the presence of bulk free carriers which are present because of unin- tentional doping by bulk defects26–29. Recently some progress has been reported30 in reducing the density of bulk carriers in TI thin films. Bulk conduction can be described by a complex bulk dielectric function
(ω) which is related to the bulk con- ductivity Σ(ω) by
(ω) =
b + i 4π ω Σ(ω), (40) where
b is the high-frequency dielectric constant of the TI. When the quantized Hall regime is approached on the TI surfaces, the bulk Hall angle tan−1(Σxy/Σxx) is expected to be much smaller than the surface Hall angle (which becomes infinite when σxx → 0). For definite- ness the numerical results reported below assume that the longitudinal bulk conductivity dominates, and that its frequency-dependence can be described by the Drude- Lorentz form, Σ(ω) = where Ωb =(cid:112)4πNbe2/mb is the plasma frequency of the 4π (1/τb − iω) (41) , Ω2 b bulk carriers (with density Nb and effective mass mb), and 1/τb is the disorder scattering rate due to impurities present in the bulk. The influence of a finite bulk conductivity is partic- ularly simple to describe in the long-wavelength low- frequency limit of Eqs. (27)-(28). The total current in- tegrated across the TI film in Eq. (26) now picks up an extra bulk conductivity contribution given by Σd (d is the film thickness), in addition to the conductivities from the two surfaces. The change in the expressions for the transmitted and reflected electric fields Eqs. (27)-(28) is therefore altered by the replacement σxx → σxx + Σd/c. When the surface has a perfect quantum Hall effect, the modified expressions for the Faraday and Kerr angles are θF = tan−1 (cid:21) (42) , (cid:20) (νT + νB) α (cid:26) 1 + 2πΣd/c (cid:27) , (43) θK = tan−1 4 (νT + νB) α 1 − (1 + 4πΣd/c)2 − [2 (νT + νB) α]2 where Σ = Σ(0) is the bulk DC conductivity. The bulk carriers thus enter as an effective longitudinal surface conductivity Σd. Eq. (42) implies that the influence of bulk conduction is negligible on the Faraday effect when Σd (e2/h) (cid:46) α. (45) When the bulk conductivity is sufficiently small that Eq. (45) is satisfied, Eqs. (42)-(43) reduce to the universal results for the Faraday and Kerr effects, Eqs. (30)-(33). Eq. (45) can alternately can be expressed as a condition that has to be satisfied by the bulk carrier density: Nb (cid:46) αmb hτbd . (46) For a 30-nm thick Be2Se3 film and disorder broadening /τb = 1 − 10 meV, we estimate that the to observe the giant Kerr effect, the bulk carrier density must be smaller than 1014 − 1015 cm−3. To reach the regime of the quan- tized Faraday effect given by Eq. (44), the bulk carrier density is allowed to be larger by a factor 1/α2 so that Nb (cid:46) mb/(αhτbd) (cid:39) 1018 − 1019 cm−3. Fig 2 shows the low-frequency Kerr angle in the presence of bulk con- duction. The magneto-optical response is modified by the presence of bulk carriers principally in the low fre- quency regime where the bulk Drude-Lorentz conductiv- ity is peaked. Because bulk carriers originate from bulk defects, Nb and τb are related. For the purpose of studying their influence on the magneto-optical response we will nev- ertheless treat Nb and τb as independent parameters. We first illustrate the case when there are no impuri- ties in the bulk, i.e. the case of a bulk free plasma. We find that the giant Kerr angle remains but under- goes a shift to progressively higher frequencies for in- creasing bulk carrier density [Fig. 2 (a)]. Including bulk impurities broadens [Figs. 2 (b)-(c)] the giant Kerr ef- fect, but the Kerr angle remains substantial ∼ 1 rad for a bulk density Nb = 1017 cm−3 and disorder broadening /τb = 1 − 10 meV. For thick films, we note that the presence of bulk conduction the cavity resonance condition becomes non- trivial. For this reason there is no simple analytic cri- terion for neglecting the bulk conductivity. Experimen- tally, this may also present a challenge since resonance frequencies cannot be readily estimated unless the bulk conductivity is known from a separate transport mea- surement. In passing, we mention that bulk optical phonon modes of the topological insulator can also be excited at higher frequencies. Since phonon energies are specific to dif- ferent TI materials and we are mainly interested in the low-frequency regime for the magneto-optical effects, we shall neglect the bulk optical phonon contributions to the conductivity. Phonon effects can be modeled by includ- ing an additional term5,6 Σd (e2/h) (cid:28) 1/α. (44)
ph(ω) = fph 2 − ω2 − iω/2τph , ωph (47) 9 respectively. At low frequencies (ω (cid:28) ∆ or ΩB) and long wavelength compared to both the film thickness d and substrate thickness ds, the analysis is again simple. We find the following Faraday and Kerr angles under oblique incidence (Expressions for the transmission and reflection coefficients at oblique incidence are presented in the Appendix.): (cid:20) 2α(νT + νB)cosθi (cid:21) cosθi + cosθo θF = tan−1 , (48) (cid:20) θK = −tan−1 8α(νT + νB)cosθicosθo cos2θo − cos2θi + 8α2(νT + νB)2cosθicosθo (49) (cid:21) , FIG. 2: (Color online) Kerr rotation for top and bot- tom surface densities nT = nB = 3 × 1011 cm−2 and fill- ing factors νT = νB = 1/2 (corresponding to a mag- netic field of 25 T) for (a) no bulk carrier scattering (b) /τb = 1 meV, and (c) 10 meV at different values of bulk carrier densities Nb = 1015 cm−3(black solid line), 1016 cm−3(red dashed), 1017 cm−3(blue dot-dashed). The di- electric constant of Bi2Se3
b = 29, bulk carrier effective mass mb = 0.15me 29,31,32, film thickness d = 30 nm. to the dielectric constant Eq. (40), where fph is the spec- tral weight of the phonon mode with frequency ωph, and 1/2τph is the phonon damping rate. This separation be- tween the electronic and the phononic contributions to the bulk dielectric function applies only when the elec- tronic time scales (Ω−1 F b, where εF b is the bulk Fermi energy) are much smaller than the phonon time scale (ω−1 ph ), making plasmon-phonon coupling negligible. b , ε−1 B. Influence of Oblique Incidence and the Substrate In this section, we examine the effects of oblique inci- dence and of the substrate on the magneto-optical effects. This discussion is particularly germane to experiments because oblique incidence may afford an advantage over normal incidence for Kerr effect measurement as it allows for spatial separation between the incident light source and the reflected light polarizer, and enhances the re- flected light intensity. 1. Thin TI Film and Thin Substrate We now account for a dielectric substrate layer un- derneath the TI film and allow for the (semi-infinite) medium underneath the substrate to be different from vacuum, solely for generality. The incident angle on the TI film and the emerging angle from the substrate can therefore assume different values, denoted by θi and θo It is important to recognize that Eqs. (48)-(49) are inde- pendent of the bulk dielectric constants of not only the TI film, but also importantly the substrate. The transmit- ted and reflected fields are generally dependent on the dielectric constant of the ambient medium surrounding the TI film however; these dependences enter the Fara- day and Kerr angles expressions through the incident and emergent angles θi and θo. Since the measurement appa- ratus is almost inevitably located in vacuum, so one has θo = θi by Snell's law. It is easy to verify that Eqs. (48)- (49) then reduce to the universal results in Section IV A, and it follows that the long-wavelength low-energy re- sults are not influenced by the angle of incidence or the presence of a thin (ds (cid:28) λ) dielectric substrate. The giant Kerr effect survives10,11 up to a relatively large frequency which we refer to as the Kerr frequency ωK. First we derive an analytic formula for ωK at normal incidence that also allows for the presence of a substrate. For small frequencies, the reflected circularly polarized components can be decomposed into separate leading- order contributions from the top and bottom TI surfaces, the bulk TI dielectric, and the substrate dielectric: [(
− µ) d + (
s − µs) ds] − i2πσI xx Er± (cid:39) iω 2c ±i2πσR xy(0), (cid:48) (0)ω (50) where σxx, σxy contain the top and bottom TI surface conductivities,
s, µs are the dielectric constant and per- meability (= 1) of the substrate, and (cid:48) in σI xx denotes a frequency derivative. The real components of Er± are smaller by a factor ∼ α in this regime. As frequency in- creases, the dielectric contribution to the imaginary part of Er± eventually dominates so that the ± components have the same sign and θK rapidly falls to a small value. The frequency range for which giant Kerr angles occur is approximately given by ωK = 2πσR xy(0) [(
− µ) d + (
s − µs) ds] /2c − 2πσI xx . (51) (cid:48) (0) A similar analysis can be carried out for the case of oblique incidence. The Kerr frequency θK at oblique in- 00.10.2ω / εc-2-101θK (rad)00.10.2ω / εc-2-1011015 cm-31016 cm-31017 cm-300.10.2ω / εc-2-101(a)(b)(c)10 √
µ sin θ. sin θi = Eqs. (51)-(52) show that oblique incidence and the presence of a substrate reduce the frequency window over which the giant Kerr angles occur. This is illustrated nu- merically in Fig. (3) where we have calculated the Kerr angle as a function of frequency and magnetic field for different dielectric substrates and different values of inci- dence angle. 2. Thin TI film and Thick Substrate Above we considered the case when the substrate thick- ness is small compared with the wavelength. We see that as long as the substrate thickness remains smaller than the wavelength, increasing the thickness only suppresses the Kerr frequency window, but the π/2 rotation at very small frequencies survives. Experimentally, however, one may have to employ a substrate with supra-wavelength thickness for various reasons; this motivates us to con- sider the effect of a thicker substrate. When the sub- strate thickness is increased beyond one wavelength, one can expect that the magnitude of the giant Kerr rota- tion is suppressed. But that is not the end of the story. Indeed, a logic similar to that employed in Section IV B tells us that when the substrate thickness contains an integer number of half wavelength, the resulting Fara- day and Kerr rotations will again be independent of the substrate dielectric properties. For wavelength short compared with the substrate thickness, but still long compared with the TI film thick- ness (kTId (cid:28) 1) and ω (cid:28) ∆ or ΩB, we find the following phases for the left- and right-handed (λ = ±) circularly polarized transmitted and reflected light for normal inci- dence FIG. 3: (Color online) Effect of substrate and oblique in- (a)-(b) illus- cidence on the low-frequency Kerr rotation. trate the Kerr angle versus frequency for two cases: (a) normal incidence on a TI film with a dielectric substrate of thickness ds = 0.5 µm:
= 1 (a free-standing sample),
= 4 (a SiO2 substrate),
= 12 (a Si substrate), and (b) oblique incidence on a free-standing TI film at incidence an- gle θi = 0, π/12, π/6, π/4. The top and surface densities are nT = nB = 3 × 1011 cm−2 and B = 25 T. (c)-(d) show re- spectively the Kerr frequency as a function of magnetic field corresponding to cases (a) and (b) at the same surface densi- ties. The TI film thickness d = 30 nm. cidence without substrate is given by ωK = 2πσR xy(0) cos θi (
cos2 θ − µ cos2 θi) d/2c − 2πσI xx . (cid:48) (0) (52) where θi and θ are the incident angle and refracted angle inside the TI film, respectively, related by Snell's law s s λ arg (Er cos(ksds)(cid:2)λ2πσR cos(ksds)(cid:2)λ2πσR 2(cid:8)(cid:0)Z−2 s − 1 +(cid:0)4πσR arg(cid:0)Et (cid:1) = tan−1 2Z−1 2Z−1 xy +(cid:0)Z−2 λ) = − tan−1 { (cid:0)Z−2 s + 1(cid:1) [Z−2 s − 1(cid:1) [Z−2 impedance Zs =(cid:112)µs/
s for the substrate. If, in addition, where for notational simplicity we have defined the wave xy cos(k0ds) − sin(k0ds)(cid:3) + sin(ksds)(cid:2)(cid:0)1 + Z−2 xy sin(k0ds) + cos(k0ds)(cid:3) + sin(ksds)(cid:2)(cid:0)1 + Z−2 s + 1(cid:1) λ4πσR ] −(cid:0)Z−2 (cid:1)2 s − 1(cid:1)(cid:2)λ4πσR s + 1 −(cid:0)4πσR (cid:1)2 xy xy s s we impose the requirement that the substrate thickness contains an integer multiple of half wavelengths (ksds = (cid:1) cos(k0ds) + λ4πσR xy sin(k0ds)(cid:3) (cid:1) sin(k0ds) − λ4πσR xy cos(k0ds)(cid:3) , sin(2ksds)(cid:3)(cid:9) (cid:0)Z−2 s − 1(cid:1) λ8πσR xy cos(2ksds) + Z−1 s ] cos(2ksds) + Z−1 s (53) (cid:41) (54) , xy sin(2ksds) N π, N (cid:54)= 0) then Eqs. (53)-(54) greatly simplify, yielding arg(cid:0)Et (cid:1) = tan−1 λ2πσR λ arg (Er λ) = −λ tan−1 , 2πσR xy xy cos(k0ds) − sin(k0ds) λ2πσR xy sin(k0ds) + cos(k0ds) 1 ,(55) (56) -2-1.5-1-0.50θK (rad)ε = 1ε = 4ε = 1204812ωK (THz)00.010.020.030.04ω / εc-2-1.5-1-0.50θK (rad)θ = 0θ = π/12θ = π/6θ = π/40510152025B (T)04812ωK (THz)(a)(c)(b)(d)11 FIG. 4: (Color online) (a). Dependence of the low-frequency Kerr rotation of a TI thin film on the substrate thickness ds for frequency ω/εc = 10−3, and substrate dielectric constants
= 4 (SiO2),
= 12 (Si). At this frequency, λ (cid:39) 7 mm is √ always much longer than the TI film thickness, but becomes comparable to the substrate thickness when ds ∼ λ/
sµs. Cavity resonances of the giant Kerr rotation can be seen at values of ds equal to integer multiples of λ/2 (b). Close-up showing the finer features of θK additional to the cavity resonances; Fabry-Perot type oscillations are clearly seen. For thick substrates, the value of θK is strongly sup- pressed compared to the long-wavelength result away from the cavity resonance values. The values of nT, nB, and d are the same as in Fig. 3. √
sµs. from which we recover the quantized Faraday [Eq. (30)] and giant Kerr rotations [Eq. (31)]. This tells us that the Faraday and Kerr rotations can survive suppression effects from a thick substrate as long as the substrate thickness satisfies the resonance condition. Although it is remarkable that Eqs. (30)-(31) still hold in this circum- stance, it is important that the light frequency needs to be precisely tuned to the resonant frequency of the sub- strate. In contrast, if one has the liberty to use a thin substrate, the giant Kerr rotation can be observed in a relatively broad range of frequencies up to ωK [Eq. (51)]. Fig. 4 shows the Kerr angle calculated as a function of substrate thickness. We see that the Kerr angle re- mains π/2 for substrate thickness ds much smaller than the wavelength (up to ∼ 1 µm in the plot) and then be- comes suppressed for larger substrate thickness. How- ever, when ds becomes comparable to the wavelength in the substrate, a series of sharply-defined cavity resonance peaks is seen that preserves the giant π/2 value [Fig. 4 (a)]. Around the same range of ds values, Fabry-perot like oscillations of the Kerr rotation are also clearly seen in Fig. 4 (b). FIG. 5: (Color online) Experimental setup of the double-layer graphene heterostructure sandwiching a hexagonal boron ni- tride (h-BN) substrate. A quantizing magnetic field is applied perpendicularly to the layers. VI. GENERALIZATION TO OTHER SYSTEMS WITH QUANTIZED HALL CONDUCTING LAYERS Because topological insulator properties are at present still obscured by the issue of bulk conduction26–29, and because samples do not yet have the quality necessary to yield strongly developed quantum Hall effects, it is natural to ask if similar magneto-optical effects can be achieved in other materials systems. Indeed, the magneto-optical effects we have discussed are not es- sentially distinct from those of other systems with two nearby conducting layers that exhibit quantum Hall ef- fects. Eqs. (30), (31), (39) for the Faraday and Kerr rotations in Section IV apply to a wide variety of other systems when they are placed in an external magnetic field. Systems with two nearby graphene layers appear to be particularly attractive because they are also described by massless Dirac equations and, like TI surface states, quite sensitive to time-reversal symmetry breaking per- turbations. In fact, the quantum Hall effect can be real- ized in graphene sheets at magnetic field strengths that are so low33–35 that applications in optics are not out of the question. Aside from integer and fractional quantum Hall effects in external magnetic fields, monolayer and bilayer graphene can also potentially exhibit quantized anomalous Hall effects36–38 due to surface adsorption of transition metal atoms. One experimental system that has now been realized experimentally39, but not yet studied optically, consists of two graphene layers separated by a few layers of hexag- onal boron nitride. Transport experiments have demon- strated that the quantum Hall effect is already strong in this type of system at fields well below 1 Tesla. We pro- pose the experimental setup shown in Fig. 5 to observe 10-710-610-510-410-310-2-2-1.5-1-0.50θK (rad)ε = 4ε = 1210-510-410-310-2ds (m)-10-50510θK (10-3 rad)(a)(b) GrapheneGrapheneh-BNB12 the dramatic magneto-optical effects. Because of valley and approximate spin degeneracies the strongest quan- tum Hall effects occur at filling factor νT,B = ±2, rather than at νT,B = ±1/2, but this only changes some details of the magneto-optical properties. Indeed, the change in filling factors might be desirable for some potential ap- plications. Realizing systems with two (or indeed many) essentially decoupled layers separated by much less than a wavelength is feasible. The small spacing between es- sentially isolated quantum Hall layers increases the fre- quency window over which strong magneto-optical effects are anticipated. In addition bulk conduction is automat- ically eliminated. robust as long as the effective two-dimensional conduc- tivity from the bulk, in e2/h units, is smaller than the fine structure constant. The effect of a thick substrate, which may sometimes be experimentally necessary, can be nullified either by making it thinner than a light wave- length or, if it must be thick, by tuning its thickness to an integer number of half wavelength. The giant Kerr ef- fect remains unaffected by oblique incidence when a thin film with a thin substrate is used. These magneto-optical effects can also be realized, perhaps even more readily, in systems with two graphene layers separated by hexagonal boron nitride or another thin dielectric. VII. CONCLUSION We have presented a comprehensive theory for the magneto-optical Faraday and Kerr effects of topologi- cal insulator films, and more generally of layered quan- tized Hall systems. We identify a topological regime in which the light frequency is low compared to surface gaps opened up by time-reversal symmetry breaking perturba- tions and the light wavelength is either long compared to the film thickness or an integer multiple of twice the film thickness. In the topological regime, the magneto-optical effects are dramatic and universal. For thin films, the Faraday rotation angle is quantized in units of the fine structure constant, and the Kerr angle exhibits a giant 90 degrees rotation. For thick films that contain a commen- surate number of half wavelength, both the Faraday and Kerr rotations are quantized in units of the fine struc- ture constant. In the presence of bulk conduction, the dramatic Faraday and Kerr effects for thin films remain VIII. ACKNOWLEDGEMENT This work was supported by the Welch under Grant No. F1473 and by DOE under Grant No. DE-FG03- 02ER45985. We are indebted to many for their useful discussions throughout the progress of this work: N. Pe- ter Armitage, Kenneth Burch, Dennis Drew, Jim Ersk- ine, Zhong Fang, Jun Kono, Joel Moore, Xiao-Liang Qi, Gennady Shvets, Rolando Vald´es Aguilar, David Vander- bilt, and Shou-Cheng Zhang. IX. APPENDIX. TRANSMISSION AND REFLECTION COEFFICIENTS AT OBLIQUE INCIDENCE We denote the incident and emergent angles of the elec- tromagnetic wave after scattering with the interface by θi and θj. The matrix elements of the reflection and transmission tensors can be found as rxx = cos2 θi sec θj 4πσxx + + 4πσxx sec θj 4π σxx + sec θj + xx + σ2 xy σxx sec θj (cid:19) (cid:21)(cid:27)(cid:27) , (cid:19) + 4πσxx sec θj (cid:21)(cid:27)(cid:27) σxx sec θj , cos2 θi sec θj 4πσxx + sec θj (cid:18) −
i µi (cid:18) (cid:20) (cid:20) −
i µi + (cid:19) − (cid:19) sec θj µj
j µj (cid:114)
j + 16π2(cid:0)σ2 (cid:114)
j + 16π2(cid:0)σ2 (cid:20)(cid:114)
i (cid:20)(cid:114)
j + cos θi
j µj µj µi + sec θj µj (cid:18) (cid:18) − µj µi µj (cid:18)(cid:114)
j (cid:114)
i (cid:114)
j (cid:1) + 4π (cid:18)(cid:114)
j (cid:114)
i (cid:114)
j (cid:1) + 4π (cid:114)
j (cid:114)
i µj µi µj µj cos θi sec θj 4πσxx + cos θi µi cos θi sec θj 4πσxx + sec θj (cid:19)(cid:21) (cid:19)(cid:21) , , (57) (58) (cid:26) ei2kiai cos θi Dji − cos θi (cid:26) ryy = − ei2kiai cos θi Dji µi (cid:26)(cid:114)
i (cid:114)
j (cid:26)(cid:114)
i (cid:114)
j (cid:114)
i µj µj µi + cos θi 4π σxx + sec θj xx + σ2 xy σxy cos θi sec θj, 8π Dji txx = rxy = − ei2kiai cos θi µi ei(ki cos θi−kj cos θj )ai (cid:114)
i (cid:114)
i (cid:114)
i txy = − ei(ki cos θi−kj cos θj )ai ei(ki cos θi−kj cos θj )ai tyy = Dji Dji 8π µi µi 2 2 Dji cos θi sec θjσxy, µi 4πσxx + sec θj µi µi µi − (cid:26) (cid:26) r(cid:48) xx = r(cid:48) yy = µj − σxx + sec θj cos2 θi sec θj cos2 θi sec θj e−i2kj ai cos θj Dji − cos θi Dji − cos θi −4π e−i2kj ai cos θj (cid:26) (cid:114)
i (cid:20)
i (cid:114)
j (cid:26) (cid:114)
i (cid:20)
i (cid:114)
j (cid:114)
j xy = − e−i2kj ai cos θj (cid:20)(cid:114)
i r(cid:48) (cid:114)
j (cid:20)(cid:114)
j t(cid:48) (cid:114)
j xx = t(cid:48) (cid:114)
j yy = xy = − ei(ki cos θi−kj cos θj )ai t(cid:48) ei(ki cos θi−kj cos θj )ai ei(ki cos θi−kj cos θj )ai σxx + sec θj Dji Dji Dji 4π 8π 8π µj µj µj µj µj 2 2 µi µi (cid:18) (cid:18) −
j µj 4πσxx − −
j µj µj (cid:114)
j + 16π2(cid:0)σ2 (cid:114)
j + 16π2(cid:0)σ2 (cid:18) (cid:18) 4πσxx + µj + cos θi σxy cos θi sec θj, cos θi sec θjσxy, µj µj + µi µj − (cid:19) (cid:18)(cid:114)
j (cid:114)
i (cid:114)
j (cid:1) + 4π (cid:18)(cid:114)
j (cid:19) (cid:114)
i (cid:114)
j (cid:1) − 4π (cid:19)(cid:21) (cid:19)(cid:21) sec θj µj µi , µj xx + σ2 xy sec θj xx + σ2 xy (cid:114)
j (cid:114)
i µj µi + sec θj 4πσxx + cos θi , − 4πσxx sec θj (cid:21)(cid:27)(cid:27) σxx sec θj , + 4πσxx sec θj (cid:21)(cid:27)(cid:27) σxx sec θj where (cid:114)
i µi Dji = Dji (cid:26) (cid:18) (cid:114)
j µj cos2 θi sec θj 4πσxx + (cid:114)
j (cid:20)
i µj µi sec θj +
j µj (cid:18)(cid:114)
j µj (cid:19) (cid:114)
i + 16π2(cid:0)σ2 µi + xx + σ2 xy + 4πσxx sec θj (cid:1) + 4π (cid:114)
j µj (cid:19) + cos θi 4π σxx + sec θj σxx sec θj . (61) (cid:19) (cid:19) 13 (59) (60) , (cid:21)(cid:27) Note that ryy, tyy are no longer equal to rxx, txx at oblique light incidence. Eqs. (58)-(61) recover the normal incidence results Eqs. (20)-(21) when θi = θj = 0. For completeness, we also include the expressions of the total reflection and transmission tensors in the pres- ence of a layer of dielectric substrate. These can be com- posed from the expressions Eqs. (22)-(23): ¯r = ¯rT + ¯t(cid:48) ¯t = ¯tS,B (cid:0)1 − ¯r(cid:48) T ¯rS,B (cid:1)−1 ¯tT, (cid:0)1 − ¯r(cid:48) (cid:1)−1 ¯tT. T ¯rS,B T ¯rS,B (62) (63) t((cid:48)) ¯ where S,B (the subscript 'S' denotes substrate) are the reflection and transmission tensors for light prop- r((cid:48)) ¯ S,B and agation from the bottom surface to the substrate-vacuum interface ¯rS,B = ¯rB + ¯t(cid:48) ¯tS,B = ¯tS (cid:0)1 − ¯r(cid:48) B ¯rS (cid:0)1 − ¯r(cid:48) (cid:1)−1 ¯tB, B ¯rS (cid:1)−1 ¯tB, (64) (65) B ¯rS r((cid:48)) ¯ S t((cid:48)) ¯ where S are the reflection and transmission tensors for light scattering at the substrate-vacuum in- terface. and 1 L. Fu, C. L. Kane, and E. J. Mele, Phys. Rev. Lett. 98, 106803 (2007). 2 J. E. Moore and L. Balents, Phys. Rev. B 75, 121306(R) (2007). 3 R. Roy, Phys. Rev. B 79, 195321 (2009). 4 For recent reviews, see M. Z. Hasan and C. L. Kane, Rev. Mod. Phys. 82, 3045 (2010); X.-L. Qi and S.-C. Zhang, arXiv:1008.2026v1. 5 A. B. Sushkov, G. S. Jenkins, D. C. Schmadel, N. P. Butch, J. Paglione, and H. D. Drew, Phys. Rev. B 82, 125110 (2010); G. S. Jenkins, A. B. Sushkov, D. C. Schmadel, N. P. Butch, P. Syers, J. Paglione, and H. D. Drew, Phys. Rev. B 82, 125120 (2010). 6 R. Vald´es Aguilar, A. V. Stier, W. Liu, L. S. Bilbro, D. K. George, N. Bansal, J. Cerne, A. G. Markelz, S. Oh, and N. P. Armitage, arXiv:1105.0237v2. 7 J. N. Hancock, J. L. M. van Mechelen, A. B. Kuzmenko, D. van der Marel, C. Brune, E. G. Novik, G. V. Astakhov, H. Buhmann, L. Molenkamp, arXiv:1105.0884v1. 8 X.-L. Qi, T. L. Hughes, and S.-C. Zhang, Phys. Rev. B 78, 195424 (2008). 9 A. M. Essin, J. E. Moore, and D. Vanderbilt, Phys. Rev. Lett. 102, 146805 (2009). 10 W.-K. Tse and A. H. MacDonald, Phys. Rev. Lett. 105, 057401 (2010). 11 W.-K. Tse and A. H. MacDonald, Phys. Rev. B 82, 161104(R) (2010). 12 M. Z. Hasan, Physics 3, 62 (2010). 13 J. Maciejko, X.-L. Qi, H. D. Drew, and S.-C. Zhang, Phys. Rev. Lett. 105, 166803 (2010). 14 G. Tkachov and E. M. Hankiewicz, Phys. Rev. B 84, 035405 (2011). 15 F. Wilczek, Phys. Rev. Lett. 58, 1799 (1987). 16 F. D. M. Haldane, Phys. Rev. Lett. 61, 2015 (1988). 17 C. L. Kane and E. J. Mele, Phys. Rev. Lett. 95, 226801 (2005). 18 Y. Xia, L. Wray, D. Qian, D. Hsieh, A. Pal, H. Lin, A. Bansil, D. Grauer, Y. S. Hor, R. J. Cava, and M. Z. Hasan, arXiv:0812.2078v1 (2008). 19 L. A. Wray, S.-Y. Xu, Y. Xia, D. Hsieh, A. V. Fedorov, H. Lin, A. Bansil, Y. S. Hor, R. J. Cava, and M. Z. Hasan, Nature Physics 7, 32 (2011). 20 Y. L. Chen, J. H. Chu, J. G. Analytis, Z. K. Liu, K. Igarashi, H. H. Kuo, X.-L. Qi, S. K. Mo, R. G. Moore, D. H. Lu, M. Hashimoto, T. Sasagawa, S.-C. Zhang, I. R. Fisher, Z. Hussain, and Z. X. Shen, Science 329, 659 (2010). 21 J. J. Cha, J. R. Williams, D. Kong, S. Meister, H. Peng, A. J. Bestwick, P. Gallagher, D. Goldhaber-Gordon, and Y. Cui, Nano Lett. 10, 1076 (2010). 22 D. Hsieh, Y. Xia, D. Qian, L. Wray, J. H. Dil, F. Meier, J.Osterwalder, L. Patthey, J. G. Checkelsky, N. P. Ong, A. V. Fedorov, H. Lin, A. Bansil, D. Grauer, Y. S. Hor, R. J. Cava, and M. Z. Hasan, Nature 460, 1101 (2009). 23 P. Cheng, C. Song, T. Zhang, Y. Zhang, Y. Wang, J.-F. Jia, J. Wang, Y. Wang, B.-F. Zhu, X. Chen, X. Ma, K. He, L. Wang, X. Dai, Z. Fang, X. C. Xie, X.-L. Qi, C.-X. Liu, S.-C. Zhang, Q.-K. Xue, Phys. Rev. Lett. 105, 076801 (2010). 24 T. Hanaguri, K. Igarashi, M. Kawamura, H. Takagi, and T. Sasagawa, Phys. Rev. B 82, 081305 (2010). 14 25 In Ref. [13], the Kerr angle result Eq. (2) is not correct outside of the long-wavelength limit, since k0d is not an integer multiple of π when kTId is. The result applies cor- rectly in the long-wavelength limit only with an infinitely thick substrate. 26 V. A. Kulbachinskii, N. Miura, H. Nakagawa, H. Arimoto, T. Ikaida, P. Lostak, and C. Drasar, Phys. Rev. B. 59, 15733 (1999). 27 A. D. LaForge, A. Frenzel, B. C. Pursley, T. Lin, X. Liu, J. Shi, and D. N. Basov, Phys. Rev. B 81, 125120 (2010). 28 N. P. Butch, K. Kirshenbaum, P. Syers, A. B. Sushkov, G. S. Jenkins, H. D. Drew, and J. Paglione, Phys. Rev. B 81, 241301(R) (2010). 29 J. G. Analytis, J.-H. Chu, Y. Chen, F. Corredor, R. D. McDonald, Z. X. Shen, and I. R. Fisher, Phys. Rev. B 81, 205407 (2010); J. G. Analytis, R. D. McDonald, S. C. Riggs, J.-H. Chu, G. S. Boebinger, and I.R. Fisher, Nat. Phys. 6, 960 (2010). 30 N. Bansal, Y. S. Kim, M. Brahlek, E. Edrey, S. Oh, arXiv:1104.5709v1. 31 K. Eto, Z. Ren, A. A. Taskin, K. Segawa, and Yoichi Ando, Phys. Rev. B 81, 195309 (2010). 32 N. P. Butch, K. Kirshenbaum, P. Syers, A. B. Sushkov, G. S. Jenkins, H. D. Drew, and J. Paglione, Phys. Rev. B 81, 241301(R) (2010). 33 K. I. Bolotin, F. Ghahari, M. D. Shulman, H. L. Stormer, and P. Kim, Nature 462, 196 (2009). 34 F. Ghahari, Y. Zhao, P. Cadden-Zimansky, K. I. Bolotin, and P. Kim, Phys. Rev. Lett. 106, 046801 (2011). 35 A. S. Mayorov, D. C. Elias, M. Mucha-Kruczynski, R. V. Gorbachev, T. Tudorovskiy, A. Zhukov, S. V. Morozov, M. I. Katsnelson, V. I. Fal'ko, A. K. Geim, and K. S. Novoselov, arXiv:1108.1742v1 (2011). 36 W.-K. Tse, Z. Qiao, Y. Yao, A. H. MacDonald, and Q. Niu, Phys. Rev. B 83, 155447 (2011). 37 Z. Qiao, S. A. Yang, W. Feng, W.-K. Tse, J. Ding, Y. Yao, J. Wang, and Q. Niu, Phys. Rev. B 82, 161414 (2010). 38 C. Weeks, J. Hu, J. Alicea, M. Franz, and R. Wu, arXiv:1104.3282v1 (2011). 39 C. R. Dean, A. F. Young, I. Meric, C. Lee, L. Wang, S. Sor- genfrei, K. Watanabe, T. Taniguchi, P. Kim, K. L. Shep- ard, and J. Hone, Nat. Nanotech. 5, 722 (2010). 

1110.0826

2

1110

Tunneling of Graphene Massive Dirac Fermions through a Double Barrier

We study the tunneling of Dirac fermions in graphene through a double barrier potential allowing the carriers to have an effective mass inside the barrier as generated by a lattice miss-match with the boron nitride substrate. The consequences of this gap opening on the transmission are investigated. The realization of resonant tunneling conditions is also analyzed.

cond-mat.mes-hall

cond-mat

ucd-tpg:1103.07 Tunneling of Graphene Massive Dirac Fermions through a Double Barrier Hocine Bahloulia,b, El Bouazzaoui Choubabic, Ahmed Jellala,c,d∗ and Miloud Mekkaouic aSaudi Center for Theoretical Physics, Dhahran, Saudi Arabia bPhysics Department, King Fahd University of Petroleum & Minerals, cTheoretical Physics Group, Faculty of Sciences, Chouaıb Doukkali University, Dhahran 31261, Saudi Arabia PO Box 20, 24000 El Jadida, Morocco dPhysics Department, College of Science, King Faisal University, PO Box 380, Alahsa 31982, Saudi Arabia Abstract We study the tunneling of Dirac fermions in graphene through a double barrier potential. This is allowing the carriers to have an effective mass inside the barrier as generated by a lattice miss- match with the boron nitride substrate. The consequences of this gap opening on the transmission are investigated and the realization of resonant tunneling conditions is analyzed. PACS numbers: 73.63.-b; 73.23.-b; 11.80.-m Keywords: Dirac, Graphene, Tunneling, Double Barrier Potential 6 1 0 2 y a M 5 ] l l a h - s e m . t a m - d n o c [ 2 v 6 2 8 0 . 0 1 1 1 : v i X r a ∗[email protected] -- [email protected] 1 Introduction After the experimental realization of graphene in 2005, this system became an attractive subject not only to experimentalists but also to theoretical physicists. This is partially due to the relativistic-like behavior of its massless carriers. In addition to its anomalous quantum Hall effect [1,2], graphene pro- vides an example of condensed matter systems where quantum electrodynamics tools can be applied [3]. These new developments offered a condensed matter laboratory to perform many investigations, which were solely available to high energy physicist in the past. Because of this similarity with relativistic fermions, graphene systems triggered the interest of the scientific community for studying massless Dirac fermions in two-dimensional spaces. One of the characteristics of Dirac fermions in graphene is their ability to tunnel through very high potential barriers with unit probability [4,5], contrary to the usual intuition. This so-called Klein tunneling of chiral particles has long ago been proposed in the framework of quantum electrodynam- ics [6 -- 8] and was recently observed experimentally [9,10]. However, as appealing as the Klein tunneling may sound from the fundamental research point of view, its presence in graphene is unwanted when it comes to applications of graphene because space confinements of the carriers is of great importance in nanoelectronics. One way to overcome these difficulties is through generating a gap in the energy spectrum or equivalently to generate a mass term in the associated Dirac equation. Experimentally it has been shown that one can generate a gap in the graphene spectrum if a graphene sheet is deposited on top of hexagonal boron nitride (BN) [11]. In graphene the carbon- carbon distance is 1.42 A, when deposited on a substrate of boron nitride, a band gap insulator with a boron to nitrogen distance of the order of 1.45 A [12] creates a lattice miss-match which is at the origin of gap opening in the graphene sheet. It was shown that the most stable configuration obtains when one carbon is on top of a boron and the other carbon in the unit cell is centered above a BN ring, the value of the induced gap is of the order of 53 meV. Depositing graphene on a metal surface with a BN buffer layer leads to n-doped graphene with an energy gap of 0.5 eV [13]. Theoretically, tunneling phenomena through graphene deposited on SiO2-BN substrate in zero magnetic field was discussed [14]. In that work it was assumed that it was possible to manufacture a thin layer with SiO2-BN interfaces, on top of which a graphene flake was deposited. This arrangement induces spatial regions where graphene has a vanishing gap intercalated with regions where BN will cause a finite gap. The graphene physics was considered in two different regions: the k−region, where the graphene sheet is standing on top of SiO2, and a q−region where the sheet stands over a BN sheet hence causing a mass-like term and inducing an energy gap of value 2t′. The effect of chiral electrons in graphene through these region was studied, the electronic spectrum changes from the usual linear dispersion to a hyperbolic dispersion, due to the presence of a gap. It was shown that contrary to the tunneling through a potential barrier, the transmission of electrons was, in this case, smaller than the one for normal incidence. These results were extended to the case of a magnetic field case [15] and different generalizations were obtained. Motivated by the above reasons and in particular the investigation made in [14] and [15], we would like to extend the study of tunneling phenomena through a graphene sheet deposited on SiO2-BN layer in the presence of an external magnetic field B and double barrier potentials. Note that, actually there were many theoretical investigations on tunneling of electrons in graphene in the presence of magnetic 1 field e.g. [16 -- 19]. However, they did not consider a graphene system deposited on SiO2-BN and hence their graphene system was gapless. More precisely, we consider a system made of two different regions, where the second is charac- terized by an energy gap t′, in a perpendicular magnetic field. The energy spectra are obtained in both regions in terms of two different Landau levels and t′. For concreteness, we consider a barrier in magnetic field and study the tunneling effect. Indeed, from the continuity condition we obtain dif- ferent solutions, which allowed us to explicitly determine the reflection and transmission coefficients in different regions those are then shown to obey the probability conservation law. We study the transmission behavior in terms of the energy ratio E t′ and discuss the possibility of resonant tunneling. We emphasis that the inhomogeneous magnetic field B discussed in our manuscript is applied externally. It can be created for instance by depositing a type-I superconducting film on top of the system and remove a strip −d1 < x < d1 of the superconductor and apply a perpendicular magnetic field. This patterning technique of creating the desired magnetic field profile was proposed by Matulis et al. [20]. One of the interesting features of such inhomogeneous magnetic field profile is that it can bind electrons, contrary to the usual potential step. Such a step magnetic field will indeed result in electron states that are bound to the step B-field and that move in one direction along the step. Thus there is a current along the y-direction but that is a very small effect and it is not relevant for our problem (those electrons have kx = 0). Indeed, we consider free electron states that have in general kx non zero, because otherwise they will not tunnel. A recent work studied double barriers with magnetic field for graphene, i.e. no mass term, can be found in [21]. The present paper is organized as follows. In section 2, we formulate our problem to include different part in the Hamiltonian describing the system made of graphene. The corresponding solution of the energy spectrum will be the subject of section 3. The scattering problem for Dirac fermions will be solved in section 4 where the transfer matrix will be explicitly given. In section 5, we determine the reflection and transmission coefficients using the associated current density. We conclude our work in the final section. 2 Problem formulation To deal with our task let us first describe the geometry of our system, which is made of five regions denoted by j = 1, 2,··· 5. Each region is characterized by its potential and interaction with external sources. The potential profile is summarized in Figure 1: V 2 1 U B t' 3 V 4 5 -d2 -d1 d1 d2 x 2 Figure 1: The energy diagram of the potential V (x) describing our double barrier along with its physical parameters. The double barrier potential V (x) is defined by V, U, 0, d1 < x < d2 x < d1 otherwise. (1) V (x) =  According the region labels, we denote the corresponding constant potential values by Vj. We introduce a uniform perpendicular magnetic field, along the z-direction, constrained to the well region between the two barriers and is defined by B(x, y) = B0Θ(cid:0)d2 1 − x2(cid:1) where B0 is the strength of the magnetic field within the strip located in the region x ≤ d1 and B = 0 otherwise, Θ is the Heaviside step function. Choosing the Landau gauge and imposing continuity of the vector potential at the boundary to avoid nonphysical effects, we end up with the following vector potential Ay(x) = c el2 B ×  −d1, x, d1, x < −d1 x < d1 x > d1 (3) with the magnetic length defined by lB =p1/B0 in the unit system ( = c = e = 1). Once we have defined the potential parameters relevant to all regions, we introduce the Hamiltonian for one-pseudospin component describing our system. In region j = 1, 2, 4, 5 it can be written as while in region (3) we have Hj = vF~σ · ~π + Vj H3 = vF~σ · ~π + U + t ′ σz (4) (5) where vF is the Fermi velocity, ~π = ~p + e ~A/c is the two-component momentum, t′ is the mass term, ~σ = (σx, σy) and σz are the usual Pauli matrices. To proceed further, we need to find the solutions of the corresponding Dirac equation and therefore the energy spectrum. 3 Energy spectrum We determine the eigenvalues and eigenspinors corresponding to the Hamiltonian Hj in different regions. Indeed, from (4) we can write in regions (1, 2, 4, 5) the Hamiltonian as where the parameter d is defined by pjx − ipy + i d l2 B Vj/vF   pjx + ipy − i d l2 B Vj/vF Hj = vF  d =( d1, −d1, x < −d1 x > d1. 3 (2) (6) (7) To solve the eigenvalue problem, we can separate variables and write the eigenspinors as ψj(x, y) = ϕj(x)χ(y) (8) where ϕj(x) = (ϕj+ ϕj−)t, ϕj± are the upper and lower spinor components, and χ(y) = eikyy, with ky a real parameter that stands for the wave number of the excitations along the y-axis. After rescaling energy ǫ = E/vF and potential vj = Vj/vF, the resulting reduced time independent Dirac equation is given by 0 pjx + ipy − i d l2 B   pjx − ipy + i d l2 B 0 ϕj− ! = (ǫ − vj) ϕj+ ϕj− ! . This eigenproblem can be written as two linear differential equations of the from   ϕj+ B(cid:19) ϕj− = (ǫ − vj)ϕj+ B(cid:19) ϕj+ = (ǫ − vj)ϕj− d l2 d l2 (cid:18)pjx − ipy + i (cid:18)pjx + ipy − i (9) (10) (11) where pjx is the x-component of the momentum operator in the j-th potential region. The above equation can be solved, using the unit system ( = c = e = 1), to get the energy eigenvalues ǫ − vj = sjs(kjx)2 +(cid:18)ky − d l2 B(cid:19)2 (12) where sj = sign(ǫ − vj) is the usual sign function, equal to ±1 for a positive and negative argument, respectively. It is clear that the wave vector along the x-direction can be written as kjx =s(ǫ − vj)2 −(cid:18)ky − d l2 B(cid:19)2 . (13) Consider now a particle-like scattering state (ǫ > 0), the particle is incident from the left side in the j-th potential region, with incoming wave vector kj = (kjx, ky) and position r = (x, y). Thus, the incoming spinor in the j-th potential region is given by ψj(x, y) = 1 √2 1 zj ! eikj.r l2 kjx + i(cid:16)ky − d B(cid:17) r(kjx)2 +(cid:16)ky − d B(cid:17)2 l2 = sjeiφj (14) (15) where zj is given by zj = sj and the phase is φj = arctan ky− d kjx !. The sign sj = ±1 again refers to conduction and valence l2 B bands in region j = 1, 2, 4, 5. The corresponding left and right moving spinors are defined by ψj+(x, y) = ψj−(x, y) = 1 √2 1 √2 1 zj ! ei(kjxx+kyy) −z∗j ! ei(−kjxx+kyy). 1 4 (16) (17) To complete the solutions of the energy spectrum, we consider now region 3 where x ≤ d1, πx = px, πy = py + Ay, and vector potential is A(x) = x/l2 B, then (5) becomes H3 =  ′ U/vF + t /vF −i∂x + iky + i x l2 B −i∂x − iky − i x U/vF − t /vF   . l2 B ′ To diagonalize this Hamiltonian we proceed by defining the usual bosonic operators a = lB√2(cid:18)∂x + ky + x l2 B(cid:19) , a† = lB√2(cid:18)−∂x + ky + x l2 B(cid:19) (18) (19) which satisfy the commutation relation [a, a†] = 1. Rescaling our energies µ = t we can write the Hamiltonian in terms of a and a† as follows √2 lB a H3 = vF u + µ −i a† √2 lB i u − µ ! . ′ /vF and u = U/vF, (20) The eigenvalue equation for the spinor ψ3(x, y), that is H3ψ3(x, y) = Eψ3(x, y) gives ψ3(x, y) = ϕ3(x)χ(y) with χ(y) = eikyy and ϕ3(x) = (ϕ3+ ϕ3−)t verifies √2 lB c u + µ −i √2 lB c +i a† u − µ ! ϕ3+ ϕ3− ! ϕ3− ! = ǫ ϕ3+ a (21) where again ǫ = E/vF. The eigenproblem can then be rewritten as two relations between the two spinor components √2 lB aϕ3− = ǫϕ3+ (u + µ)ϕ3+ − i √2 lB i a†ϕ3+ + (u − µ)ϕ3− = ǫϕ3− which lead to an eigenvalue equation for the operator aa† and ϕ3+ (cid:2)(ǫ − u)2 − µ2(cid:3) ϕ3+ = aa†ϕ3+. 2 l2 B It is clear that ϕ3+ is an eigenstate of the number operator N = a†a and therefore we identify ϕ3+ with the eigenstates of the harmonic oscillator n − 1i, namely (22) (23) (24) (25) (26) (27) and the corresponding energy spectrum reads ϕ3+ ∼ n − 1i The second spinor component can now be written as follows (ǫ − u) = ±ǫn = ± i√2a† ϕ3− = (ǫ − u)lB + µlB n − 1i = (ǫ − u)lB + µlB ni 1 lBp(µlB)2 + 2n. i√2n 5 where √2n =p(ǫnlB)2 − (µlB)2, which gives ϕ3− = ±ir ǫn ∓ µ ǫn ± µ ni. Where the ± signs correspond to positive and negative energy solutions, respectively. After normal- ization we obtain the eigenspinors for positive and negative energies (28) (29) (ϕ)±3n = 1 √2  ǫn q ǫn±µ ±iq ǫn∓µ n − 1i ni ǫn   . By introducing the parabolic cylinder functions Dn(x) = 2−n/2e− x barrier region x ≤ d1 can be expressed as ψ3n(x, y) = 1 √2  ǫn q ǫn±µ √2/lB √ǫn(ǫn±µ) ±i D[(ǫnlB )2−(µlB )2]/2−1h±√2(cid:16) x D[(ǫnlB )2−(µlB )2]/2h±√2(cid:16) x lB lB 2 4 Hn(cid:16) x√2(cid:17), the solution in the + kylB(cid:17)i + kylB(cid:17)i eikyy. (30)   We should note that the above ± signs stand for waves traveling to right and left, respectively, along our usual convention for traveling waves e±ikjxx. In this context the full solution in region 3 is given by a linear combination, c+ times (30) with the upper sign added to c− times (30) with the lower sign where c± are arbitrary coefficients to be fixed by the boundqry conditions as exposed in the following section. The above results (16)-(17) and (30) summarizes the solutions of the energy spectrum in different regions composing our graphene sheet. Next, we will show how to implement the results obtained so far to study the tunneling of Dirac fermions through a double barrier potential. 4 Boundary conditions It is straightforward to solve the scattering problem for Dirac fermions. We assume that the incident wave propagates at an angle φ1 with respect to the x-axis, where the superscript indicate the potential region. Let us first write explicitly the eigenvalues and the corresponding eigenspinors with reflected and transmitted parts in each region. In the incident region for x < −d2 (region 1), we have ǫ =s(k1x)2 +(cid:18)ky − d1 l2 B(cid:19)2 ψ1 = 1 √2 1 z1 ! ei(k1xx+kyy) + r 1 √2 1 −z∗1 ! ei(−k1xx+kyy) (31) (32) where r is the reflection amplitude. It is clear that the shift in ky is due to our gauge choice for the vector potential. It is convenient to parameterize the momenta as follows k1x = ǫ cos φ1, ky = ǫ sin φ1 + d1 l2 B . In the barrier region −d2 ≤ x ≤ −d1 (region 2), we have ǫ − v = s2s(k2x)2 +(cid:18)ky − B(cid:19)2 z2 ! ei(k2xx+kyy) + √2 1 √2 1 ψ2 = d1 l2 a b −z∗2 ! ei(−k2xx+kyy) 6 (33) (34) (35) where a and b are two wavefunction parameters and v = V /vF . The corresponding, suitably parame- terized, momentum is given by k2x = (ǫ − v) cos φ2, ky = (ǫ − v) sin φ2 + d1 l2 B . Similarly for the barrier region d1 ≤ x ≤ d2 (region 4) we have ǫ − v = s4s(k2x)2 +(cid:18)ky + B(cid:19)2 z4 ! ei(k4xx+kyy) + √2 1 √2 1 ψ4 = d1 l2 f e −z∗4 ! ei(−k4xx+kyy) (36) (37) (38) where e and f are the corresponding wavefunction parameters. Similarly, the momentum can be defined by (39) (40) (41) (42) k4x = (ǫ − v) cos φ4, ky = (ǫ − v) sin φ4 − d1 l2 B . Finally, in the transmission region, x > d2 (region 5), we have only transmitted waves ǫ =s(k5x)2 +(cid:18)ky + d1 l2 B(cid:19)2 z5 ! ei(k5xx+kyy) t √2 1 ψ5 = where t is the transmission amplitude, the corresponding momentum reads as k5x = ǫ cos φ5, ky = ǫ sin φ5 − d1 l2 B . The refraction angles φ2, φ4 and φ5 at the interfaces are obtained by requiring conservation of the momentum py. This leads to a simplified expression of these angles in terms of φ1 sin φ2 = sin φ4 = ǫ ǫ − v ǫ ǫ − v sin φ5 = sin φ1 + sin φ1 sin φ1 + 2d1 ǫl2 B . 2d1 (ǫ − v)l2 B (43) (44) (45) We should point out at this stage that we were unfortunately forced to adopt a somehow cumbersome notation for our wavefunction parameters in different potential regions due to the relatively large number of necessary subscripts and superscripts. Before matching the eigenspinors at the boundaries, let us define the following shorthand notation for the dimensionless parameters an± =r1 ± µ ǫn , dn± = √2 ǫnlBan± (46) 7 and the parabolic cylindric functions by η1± = D[(ǫnlB )2−(µlB )2]/2−1(cid:20)± ξ1± = D[(ǫnlB )2−(µlB )2]/2(cid:20)± √2(cid:18)−d1 √2(cid:18)−d1 + kylB(cid:19)(cid:21) + kylB(cid:19)(cid:21) . lB lB (47) (48) Now, requiring the continuity of the spinor wavefunctions at each junction interface give rise to a set of equations. Indeed, at the point z1e−ik1xd2 − rz∗1eik1xd2 = az2e−ik2xd2 − bz∗2eik2xd2 x = −d2 −→ ( e−ik1xd2 + reik1xd2 = ae−ik2xd2 + beik2xd2 x = −d1 −→ ( ae−ik2xd1 + beik2xd1 = c+an+η+ 1 + c−an−η−1 x = d1 −→ ( c+an+η2+ + c−an−η2− = eeik4xd1 + f e−ik4xd1 x = d2 −→ ( eeik4xd2 + f e−ik4xd2 = teik5xd2 ez4eik4xd2 − f z∗4e−ik4xd2 = tz5eik5xd2 c+idn+ξ2+ − c−idn−ξ2− = ez4eik4xd1 − f z∗4e−ik4xd1 az2e−ik2xd1 − bz∗2eik2xd1 = c+idn+ξ1+ − c−idn−ξ1− (49) (50) (51) (52) In the above formulae, the parameters η2± and ξ2± are defined similarly to (47)-(48) but with d1 replaced by −d1. All these equations can be written in compact form by introducing the transfer matrix M , which is given by eik2xd2 −1 z2 eik2xd1 e−ik2xd2 1 e−ik2xd1 z2e−ik2xd1 −1 z2 0 0 0 0 0 0 0 0 eik2xd2 0 0 −anη1+ 0 0 −an−η1− eik2xd1 −idn+ξ1+ idn−ξ1− −an+η2+ −an−η2− idn+ξ2+ 0 0 −idn−ξ2− −z4eip4xd1 0 0 eik4xd2 z4eik4xd2 0 0 0 0 eik4xd1 0 0 0 0 e−ik4xd1 1 z4 e−ik4xd2 − 1 0 0 0 0 0 0 −eik5xd2 e−ik4xd2 −z5eik5xd2 e−ik4xd1 z4 .   −eik1xd2 eik1xd2 1 z1 0 0 0 0 0 0   (53) Defining Φ = (r, a, b, c+, c−, e, f, t)t as a vector quantity made of all unknown wavefunction param- eters in our problem and Ξ = (e−ik1xd2, z1e−ik1xd2 , 0, 0, 0, 0, 0, 0)t , where the superscript t stands for transpose, the compact form of equation (49)-(52) can then be written as M Φ = Ξ. (54) Of course the same problem could be formulated in terms of 2 × 2 matrices if we use the concept of transfer matrix from one potential region to another. This last formulation will be much more adequate in dealing with periodic systems and applying the Bloch theorem to find the associated energy bands. 8 5 Reflection and transmission coefficients Now we are ready for the computation of the reflection R and transmission T coefficients. For this purpose, we introduce the associated current density J, which defines R and T as R = Jre Jin , T = Jtr Jin (55) where Jin, Jre and Jtr stand for the incident, reflected and transmitted components of the current density, respectively. It is easy to show that the current density J is given by which gives the following results for the incident, reflected and transmitted components J = evFψ†σxψ Jin = evF(ψ1in)†σxψ1in = evF Jre = evF(ψ1re)†σxψ1re = evFr∗r Jtr = evF(ψ5tr)†σxψ5tr = evFt∗t k1x l2 k1x r(k1x)2 +(cid:16)ky − d1 B(cid:17)2 r(k1x)2 +(cid:16)ky − d1 B(cid:17)2 r(k5x)2 +(cid:16)ky + d1 B(cid:17)2 p5x l2 . l2 Injecting these results in (55) we obtain T = k5x k1x l2 r(k1x)2 +(cid:16)ky − d1 B(cid:17)2 B(cid:17)2t2 r(k5x)2 +(cid:16)ky + d1 l2 R = r2. Now using the conservation of energy we find the constraint s(k1x)2 +(cid:18)ky − d1 l2 B(cid:19)2 =s(k5x)2 +(cid:18)ky + d1 l2 B(cid:19)2 k1x =s(k5x)2 + 4ky d1 l2 B which allows us to express the transmission probability T in the following form (56) (57) (58) (59) (60) (61) (62) (63) (64) T = k5x q(k5x)2 + 4ky t2. d1 l2 B To get an explicit expression for T , we should determine the transmission amplitude t. After some lengthy algebra, one can solve the linear system of equations (54) to obtain the transmission and reflection amplitudes in closed form i√2 ei(d1+d2)(k2x+k4x)(1 + (z1)2)(1 + (z2)2)(1 + (z4)2) t = r = lBǫ − u e−2id2k1xz1 [A(z1 − z2) − B(1 + z1z2)] eid2(k1x+k5x) [A(1 + z1z2) + B(z1 − z2)] A(1 + z1z2) + B(z1 − z2) 9 (η2+ξ2− + η2−ξ2+) (65) (66) where we have defined the following quantities A = e2i(d1k2x+d2k4x)(z4 − z5)(α + βz2z4 + γz4 − δz2) +e2id1(k2x+p4x)(1 + z4z5) (−αz4 + βz2 + γ + δz2z4) B = e2id2(k2x+k4x)(z4 − z5) (−αz2 + βz4 − γz2z4 − δ) +e2i(d2k2x+d1k4x)(1 + z4z5) (αz2z4 + β − γz2 + δz4) as well as α = β = ǫ − u + µ ǫ − u (η1−η2+ − η1+η2−) 2 lB(ǫ − u)(ǫ − u + µ) √2 (ξ1−ξ2+ − ξ1+ξ2−) γ = i δ = i lBǫ − u √2 lBǫ − u (η1−ξ2+ + η1+ξ2−) (η2−ξ1+ + η2+ξ1−) . (67) (68) (69) (70) (71) (72) The resulting reflection and transmission coefficients have been computed numerically and are plotted in Figures 3, 4, 5, 6 for several parameter values (ǫ, v, u, d1, d2, µ). For instance a typical value of the magnetic field say B0 = 4T, the magnetic length is lB = 13 nm, and ǫlB = 1 corresponds to the energy E = 44 meV [16]. Now, let us study the reflection and transmission coefficients versus the energy ǫlB. The quantity kylB = m∗ plays a very important role in the transmission of Dirac fermions via the obstacles created by the series of scattering potentials, because it associates an effective mass to the particle and hence determines the threshold for the allowed energies. However, the application of the magnetic field in the intermediate zone seems to reduce this effective mass to(cid:16)kylB − d1 it increases it to (cid:16)kylB + d1 are then determined by the greater effective mass, namely ǫlB ≥ kylB + d1 lB(cid:17) in the incidence region while lB(cid:17) in the transmission region as shown in Figure 2. The allowed energies . lB Ε lB ky lB + d1
lB ky lB - d1
lB ky lB Figure 2: The energy configuration of the double barrier potential illustrating the effect of inhomogeneous effective mass in the incidence and reflection regions. "Color figure online". If we cancel the well region by setting d1 = 0 then we reproduce the usual single barrier transmis- sion as reflected in figure 3 which shows clearly the Klein zone followed by a full reflection zone. On 10 the other hand if we keep the well region and cancel both the applied magnetic field and mass term in the well region, the series of potentials behaves like a simple double barrier with the same effective mass ky all over. Thus, in this case we reproduce exactly the transmission obtained in [22], for the massive 1D Dirac equation with m = ky. 1.0 0.8 0.6 0.4 0.2 T, R ky lB v lB 2 ky lB T R 0.0 0 2 4 6 ΕlB 8 10 12 Figure 3: Transmission and reflection coefficient as a function of energy ǫlB for a single barrier with = 0, d2lB = 1.5, vlB = 6 and kylB = 1. "Color figure online". d1 lB The result shown in Figure 3 can be interpreted as follows. The transmission as function of energy, through a single barrier, starts from the minimum required energy equal to the effective mass m∗ = kylB then oscillates reaching full transmission due to constructive interference in the (Klein zone) for energies between m∗ and V − m∗. This is then followed by a gap region between V − m∗ and V + m∗ where the transmission drops to zero. For energies above V + m∗ the transmission has the usual oscillating behavior then reaches full total transmission asymptotically. In Figure 4 we show the influence of the size of the well region and magnetic field on the trans- mission. First we consider the same conditions as above but we vary d1 to obtain Figure 4a: 1.0 0.8 0.6 0.4 0.2 T 0.0 0 2 4 d1 lB = 0.05 d1 lB = 0.1
a
6 ΕlB 11 d1 lB = 0.2 8 10 12 Figure 4a: Transmission as a function of energy ǫlB with d1 lB ulB = 4, µlB = 4, kylB = 1. "Color figure online". = {0.05, 0.1, 0.2}, d2 lB = 1.5, V lB = 6, According to Figure 4, the size of the intermediate zone affects the form of transmission in these four zones. As d1 increases, the effective mass increases and the oscillations in the Klein zone get reduced. This strong reduction in the transmission in the Klein zone seem to suggest the potential suppression of the Klein tunneling as we increase d1. On the other hand the size of the gap region increases as we increase d1. However, this is not the case as we increase µlB where there is some change amplitude as shown in Figure 4b: 1.0 0.8 0.6 0.4 0.2 T 0.0 0 2 4 Μ lB = 0.5 Μ lB = 4 Μ lB = 6 Μ lB = 8
b
6 ΕlB 8 10 12 Figure 4b: Transmission as a function of energy ǫlB with µlB = {0.5, 4, 6, 8}, d1 lB vlB = 6, ulB = 4, kylB = 1. "Color figure online". = 0.05, d2 lB = 1.5, It is worthwhile to analyze the reflection and transmission coefficients versus the potential. In doing so, we fix the energy and choose a value of d1, then we compute the transmission as shown in Figure 5a and 5b: 1.0 0.8 0.6 0.4 0.2 T d1
lB = 0 d1
lB = 0.05 d1
lB = 0.1 d1
lB = 0.2 d1
lB = 0.3 0.0 0 2 4 8 10 12 6 ΕlB 12 Figure 5a: Transmission as a function of potential vlB for d1 lB ǫlB = 6, ulB = 4, µlB = 4 and kylB = 1. "Color figure online". = {0, 0.05, 0.1, 0.2, 0.3} with d2 lB = 1.5, T 1.0 0.8 0.6 0.4 0.2 0.0 0 d1
lB = 0 d1
lB = 0.7 d1
lB = 0.1 d1
lB = 0.3 d1
lB = 0.2 d1
lB = 0.5 d1
lB = 0.6 5 10 ulB 15 20 Figure 5b: Transmission as a function of potential ulB for d1 lB = 1.5, ǫlB = 6, vlB = 4, µlB = 4 and kylB = 1. "Color figure online". = {0, 0.1, 0.2, 0.3, 0.5, 0.6, 0.7} with d2 lB It is clearly seen that for a given energy, the transmission decreases if d1 increases and then vanishes. Let us examine the transmission coefficient as function of angle incidence φ1. This is shown in Figure 6: 1.0 0.5 0.0 -0.5 -1.0 d1
lB = 0.5 d1
lB = 1.5 d1
lB = 3 d1
lB = 3.67
a
1.0 0.5 0.0 -0.5 -1.0 Ε lB = 2 Ε lB = 8 Ε lB = 4 Ε lB = 1
b
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 Figure 6a,6b: Polar plot of a curve with radius (transmission T) as a function of angle φ1 with: = {0.5, 1.5, 3, 3.67} with ǫlB = 3.7, d2 (a): d1 lB (b): ǫlB = {1, 2, 4, 8} with d1 "Color figure online". = 0.5, d2 lB lB = d1 lB lB , vlB = ulB = 0, µlB = 0, kylB = 1. = 0.6, vlB = 0.5, ulB = 0.4, µlB = 1, kylB = 1. Figure 6a reproduces exactly the result of De Martino et al. [16], this reference was the first to treat the confinement of Dirac fermions by an inhomogeneous magnetic field. This polar graph shows the 13 transmission as a function of the incidence angle, the outermost circle corresponds to full transmission, T = 1, while the origin of this plot represents zero transmission. We note that for certain incidence angles the transmission is not allowed, in fact for ǫlB ≤ kylB + d1 all waves are completely reflected. It is worth mentioning that the transmission is uniquely defined by the incidence angle i.e. each radial line representing a given incidence angle intersects the transmission curve at one point. In Figure 6a we see that the transmission vanishes for values of d1 below the critical value ǫlB = 6.7. In Figure 6b lB we fix d1 and vary the energy, we observe that the transmission get reduced as we increase the lB increase the energy. In Figure 6c we fix d2 , the transmission vanishes below the critical lB value d1 lB = 3.7 for our parameters. and vary d1 lB and d2 lB lB 1.0 0.5 0.0 -0.5 -1.0 d1
lB = 0.5 d1
lB = 1.5 d1
lB = 3.65 d1
lB = 3
c
1.0 0.5 0.0 -0.5 -1.0 d2 lB = 0.6 d2 lB = 1 d1 lB = 3 d1 lB = 2
d
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 Figure 6c,6d: Polar plot of a curve with radius (transmission T) as a function of angle φ1 with: (c): d1 lB (d): d2 lB "Color figure online". = {0.5, 1.5, 3, 3.67} with ǫlB = 2, d2 = {0.6, 0.1, 2, 3} with ǫlB = 8, d2 lB = 0.5, vlB = 0.5, ulB = 0.4, µlB = 1, KylB = 1. = 0.6, vlB = 0.5, ulB = 0.4, µlB = 1, kylB = 1. lB 1.0 0.5 0.0 Μ lB = 0.5 Μ lB = 3 Μ lB = 2 -0.5 Μ lB = 1 -1.0
e
0.0 0.2 0.4 0.6 0.8 1.0 Figure 6e: Polar plot of a curve with radius (transmission T) as a function of angle φ1 with : 14 (e): µlB = {0.5, 1, 2, 3} with ǫlB = 4, d1 lB figure online". = 0.5, d2 lB = 0.6, vlB = 0.5, ulB = 0.4, KylB = 1. "Color and vary d2 lB In Figure 6d we fix d1 and shows lB an oscillatory behavior which becomes more apparent for larger values of d2 . Finally in Figure 6e lB we show how the transmission is affected by the effective mass term reflected by µ, the transmission decreases sharply as we increase µ. , the transmission is less sensitive to the variations of d2 lB 6 Conclusion To conclude, we have studied the effect of gap opening on the transport properties of graphene Dirac fermions. The effective mass is generated by a lattice miss-match between the boron nitride substrate and the 2D graphene sheet. An additional homogeneous magnetic field localized within the well potential region causes the effective mass to change and affects asymmetrically the incidence and reflection regions. We have found that in contrast to electrostatic barriers, magnetic barriers are able confine Dirac fermions, a property of great importance from the practical point of view. In fact as can be seen in Figure 6, the transmission reduces a lot as we increase the effective mass parameter, t′ in our original model (5), or the width, d1, of the effective mass region. These results were exposed clearly on the polar graph showing the transmission as a function of the incidence angle in Figure 6. The outermost semicircle corresponds to unit transmission is not intersected by the transmission curve, even at normal incidence, if we increase the effective mass or width of the magnetic field region. We hope that the present interplay between gap opening and local magnetic field effect will give more freedom to experimentalists to develop graphene based electronic devices. Acknowledgments The generous support provided by the Saudi Center for Theoretical Physics (SCTP) is highly appreci- ated by all Authors. AJ acknowledge partial support by King Faisal University. We also acknowledge the support of KFUPM under project RG1108-1-2. References [1] K.S. Novoselov A.K. Greim, S.V. Morosov, D. Jiang, M.I. Katsnelson, V.I. Grigorieva, L. Levy, S.V. Dubonos and A.A. Firsov, Nature 438 (2005) 197. [2] Y. Zhang, Y.W. Tan, H.L. Stormer and P. Kim, Nature 438 (2005) 201. [3] M.I. Katsnelson and K.S. Novoselov, Solid State Commun. 143 (2007) 3. [4] M.I. Katsnelson, K.S. Novoselov and A.K. Geim, Nature Physics 2 (2006) 620. [5] C. Bai and X.D Zhang, Phys. Rev. B76 (2007) 75430. 15 [6] O. Klein, Z. Phys. 53 (1929) 157. [7] A. Calogeracos and N. Dombey, Contemp. Phys. 40 (1999) 313. [8] C. Itzykson and J.-B. Zuber, Quantum Field Theory (McGraw-Hill, New York, 1985). [9] N. Stander, B. Huard and D. Goldhaber-Gordon, Phys. Rev. Lett. 102 (2009) 026807. [10] A.F. Young and P. Kim, Nat. Phys. 5 (2009) 222. [11] G. Giovannetti, P.A. Khomyakov, G. Brocks, P.J. Kelly and J. van den Brink, Phys. Rev. B76 (2007) 73103. [12] J. Zupan, Phys. Rev. B6 (1972) 2477. [13] Y.H. Lu, P.M. He and Y.P. Feng, "Graphene on metal surface: gap opening and n-doping", arXiv:0712.4008. [14] J. Viana Gomes and N.M.R. Peres, J. Phys.: Cond. Matt 20 (2008) 325221. [15] E.B. Choubabi, M. El Bouziani and A. Jellal, Int. J. Geom. Meth. Mod. Phys 7 (2010) 909. [16] A. De Martino, L. DellAnna and R. Egger, Phys. Rev. Lett. 98 (2007) 066802. [17] A.V. Shytov, M.S. Rudner and L.S. Levitov, Phys. Rev. Lett. 101 (2008) 156804. [18] M. Barbier, F.M. Peeters, P. Vasilopoulos and J. Milton Pereira, Phys. Rev. B77 (2008) 115446. [19] M.R. Masir, P. Vasilopoulos and F.M. Peeters, New J. Phys. 11 (2009) 095009. [20] A. Matulis, F.M. Peeters and P. Vasilopoulos, Phys. Rev. Lett. 72 (1994) 1518. [21] M. Ramezani Masir, P. Vasilopoulos and F.M. Peeters, Phys. Rev. B82 (2010) 115417. [22] A. D. Alhaidari, H. Bahlouli and A. Jellal, "Relativistic Double Barrier Problem with Three Sub-Barrier Transmission Resonance Regions", arXiv:1004.3892. 16 

1206.3827

2

1206

Theoretical polarization dependence of the two-phonon double-resonant Raman spectra of graphene

The experimental Raman spectra of graphene exhibit a few intense two-phonon bands, which are enhanced through double-resonant scattering processes. Though there are many theoretical papers on this topic, none of them predicts the spectra within a single model. Here, we present results for the two-phonon Raman spectra of graphene calculated by means of the quantum perturbation theory. The electron and phonon dispersions, electronic lifetime, electron-photon and electron-phonon matrix elements, are all obtained within a density-functional-theory-based non-orthogonal tight-binding model. We study systematically the overtone and combination two-phonon Raman bands, and, in particular, the energy and polarization dependence of their Raman shift and intensity. We find that the ratio of the integrated intensities for parallel and cross polarized light for all two-phonon bands is between 0.33 and 0.42. Our results are in good agreement with the available experimental data.

cond-mat.mes-hall

cond-mat

Theoretical polarization dependence of the two-phonon double-resonant Raman spectra of graphene Valentin N. Popov Faculty of Physics, University of Sofia, BG-1164 Sofia, Bulgaria Philippe Lambin Research Center in Physics of Matter and Radiation, Facult´es Universitaires Notre Dame de la Paix, B-5000 Namur, Belgium (Dated: November 16, 2018) Abstract The experimental Raman spectra of graphene exhibit a few intense two-phonon bands, which are enhanced through double-resonant scattering processes. Though there are many theoretical papers on this topic, none of them predicts the spectra within a single model. Here, we present results for the two-phonon Raman spectra of graphene calculated by means of the quantum perturbation theory. The electron and phonon dispersions, electronic lifetime, electron-photon and electron- phonon matrix elements, are all obtained within a density-functional-theory-based non-orthogonal tight-binding model. We study systematically the overtone and combination two-phonon Raman bands, and, in particular, the energy and polarization dependence of their Raman shift and inten- sity. We find that the ratio of the integrated intensities for parallel and cross polarized light for all two-phonon bands is between 0.33 and 0.42. Our results are in good agreement with the available experimental data. 2 1 0 2 n u J 5 2 ] l l a h - s e m . t a m - d n o c [ 2 v 7 2 8 3 . 6 0 2 1 : v i X r a 1 I. INTRODUCTION Presently, graphene is considered as a prospective material for nanoelectronics and nanophotonics.1,2 Among the various experimental techniques, the Raman spectroscopy has proven to be an indispensable tool for investigation of this material.3 -- 6 Graphene has a single Raman-active phonon E2g observed as an intense line (the G band) in the first-order Raman spectra. The second-order spectra of graphene with low defect density has several intense bands, which originate from scattering of electrons and holes by two phonons of the same/different frequency and non-zero momentum and are called overtone/combination bands. The appearance of intense second-order bands can be explained by the double-resonant (DR) scattering mechanism.7,8 These bands contain valuable information on the phonon dispersion9 -- 12 and the electron-phonon and electron- electron matrix elements.13 The Raman spectra of graphene with defects show additional bands, which arise from DR scattering of electrons and holes by phonons and defects. The theoretical investigation of the two-phonon DR scattering in graphene has been performed using various approximations: replacement of the electron-photon and electron- phonon interactions with constants, using a constant electronic lifetime, considering only high-symmetry directions in the Brillouin zone, as well as exact DR conditions. The pre- dicted dispersive behavior of the Raman bands7 and the frequency shift of the Stokes and anti-Stokes Raman bands14 have been found in quantitative agreement with the experimental data. It has also been realized that the integration over the entire Brillouin zone of graphene is essential for predicting the Raman intensity.15,16 While most of the theoretical papers fo- cus on the most intense overtone band, in a recent study, all two-phonon Raman bands with observable intensity have been calculated using an electron-phonon matrix element de- rived within a nearest-neighbor π-band tight-binding model.17 The dominant contribution to the two-phonon bands from different parts of the Brillouin zone17,18 and from different scattering processes17 has also been discussed. The polarization dependence of the most intense overtone band has been studied experimentally and theoretically.19 The progress, made so far in the modeling of the DR bands, has been achieved either with simple π-band tight-binding models, or with more sophisticated models but relying on approximations of the electron-phonon matrix element, and in most cases concerns only a few intense bands. Our experience indicates that, although the Raman shift of the bands is not sensitive to the 2 used matrix element, their Raman intensity crucially depends on it. Here, we calculate the two-phonon DR Raman spectra of graphene using a non-orthogonal tight-binding (NTB) model, which implements parameters derived from a density functional theory (DFT) study and thus has no adjustable parameters.20 In particular, the electronic21 and phonon22 dispersion, the electron-photon and electron-phonon matrix elements,23 as well as the electronic linewidth24 are all obtained within this model. The NTB model for electrons and phonons in graphene is introduced in Sec. II. The calculated two-phonon DR Raman spectra of graphene and its polarization dependence are discussed in Sec. III. The paper ends up with conclusions (Sec. IV). II. THEORETICAL PART A. The NTB model We use a NTB model with four valence electrons per carbon atom to calculate the elec- tronic dispersion of graphene.21 This model is based on matrix elements of the Hamiltonian and overlap matrix elements derived from DFT20 and therefore it does not rely on any ad- justable parameters. It also allows one to estimate the total energy and the forces on the atoms. This feature is utilized for relaxation of the atomic structure. Up to a few electron volts away from the Fermi energy, the electronic structure of graphene has the form of conic valence and conduction bands (Dirac cones) with a common apex (the Dirac point) at two non-equivalent special points, K and K′, of the Brillouin zone. This specific form of the electronic bands plays an important role in the enhancement of the two-phonon Raman scattering through the DR mechanism. The dynamical model of graphene uses a dynamical matrix derived by a perturbative approach within the NTB model.22 The electron-photon and electron-phonon matrix ele- ments are calculated explicitly.23 The summation over the Brillouin zone in the first-order perturbation term of the dynamical matrix is performed over a 40 × 40 mesh of k points, for which the phonon frequencies converge within 1 cm−1. The calculated in-plane phonon branches of graphene, after scaling by a factor of 0.9, agree fairly well with the available experimental data22 (Fig. 1). The phonons with displacement in the graphene plane (in- plane phonons) can interact with electrons and thus can contribute to the Raman spectra. 3 Those with atomic displacement perpendicular to graphene (out-of-plane phonons) are less well reproduced but they do not contribute to the spectra. It will be shown below that only phonons, close to the high-symmetry directions ΓK, ΓM, and KM of the Brillouin zone, are of major importance for the two-phonon spectra. The phonon branches along these directions will be denoted, as usual, by two-letter acronyms describing their vibrational pattern: the letters O and A stand for "optical" and "acoustic", respectively; the letters L, T, and Z denote in-plane longitudinal, in-plane transverse, and out-of-plane atomic displacement, respectively. The acronyms for the branches along the KM direction will be primed. Alternatively, for each wavevector, the phonons with be ascribed the index ν, ν = 1, ..., 6, in order of increasing frequency. The phonons with a certain ν can belong to branches with a different vibrational pattern. For example, a phonon with ν = 6 can belong to the LO or TO branch. It will also be argued that only phonons close to the Γ and K points give a significant contribution to the two-phonon Raman spectra. Such phonons will be denoted by acronyms ending with @Γ and @K. For example, LO phonons close to the Γ point will be denoted by LO@Γ and TO phonons close to the K point along the KM direction will be denoted by TO′@K (see, Fig. 1). B. The double-resonant processes The amplitude for two-phonon DR Raman scattering processes in graphene is described by fourth-order terms in perturbation theory.25 The underlying processes include virtual scattering of electrons/holes by phonons between states of the Dirac cone at the K point or the K′ point, or between states of the Dirac cones at the K and K′ points. The momentum is conserved in each virtual process but the energy is conserved only for the entire DR process. Below, we will consider only Stokes processes. In this case, a two-phonon DR process includes an absorption of a photon with a creation of an electron-hole pair, two consecutive processes of scattering of an electron/hole with creation of a phonon, and a recombination of the electron-hole pair with an emission of a photon (Fig. 2). There are altogether eight such processes.17,26 The total two-phonon Raman intensity is given by the expression 4 I ∝ X f X c,b,a (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) Mf cMcbMbaMai (Ei − Ec − iγ) (Ei − Eb − iγ) (Ei − Ea − iγ) δ (Ei − Ef ) (1) 2 (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) Here, the inner sum is the scattering amplitude. Eu, u = i, a, b, c, f , are the energies of the initial (i), intermediate (a, b, c), and final (f ) states of the system of photons, electrons, holes, and phonons. In the initial state, only an incident photon is present and, therefore, Ei = EL, where EL is the incident photon energy. In the final state, there is a scattered photon and two created phonons. Muv are the matrix elements for virtual processes between initial, intermediate, and final states. In particular, Mai and Mf c are the matrix elements of momentum for the processes of creation and recombination of an electron-hole pair, re- spectively. Mba and Mcb are the electron/hole-phonon matrix elements. γ is the sum of the halfwidths of pairs of electronic and hole states, and will be referred to as the electronic linewidth. The electron-photon and electron-phonon matrix elements, and the electronic linewidth are calculated explicitly.23,24 The Dirac delta function ensures energy conservation for the entire process. In the calculations, it is replaced by a Lorentzian with a halfwidth of 5 cm−1. The summation over the intermediate states runs over all valence and conduction bands, and over all electron wavevectors k. The summation over the final states runs over all phonon branches and phonon wavevectors q. For both summations, convergence is reached with a 800 × 800 mesh of k and q points in the Brillouin zone. For the discussion of the polarization dependence of the Raman intensity of the two- phonon bands, it is advantageous to rewrite Eq. (1) in the form I ∝ X f eS · R · eL2 δ (Ei − Ef ) (2) Here, eL and eS are the polarization vectors of the incident and scattered laser light, re- spectively, and R is the Raman tensor. Everywhere below we consider only backscattering geometry in accord with the usual experimental Raman setup for graphene and, therefore, the polarization vectors lie in the graphene plane. 5 III. RESULTS AND DISCUSSION A. Electronic linewidth The electronic linewidth γ is due to a large extent to scattering of the electrons (holes) by phonons and other electrons (holes). The majority of the published reports assume that γ is energy-independent. Recently, it has been argued that in undoped graphene γ is dominated by electron-phonon processes and the expression γ = 9.44E + 3.40E 2 has been derived, where the energy separation between the valence and conduction bands E is in eV and γ is in meV.17 Here, the electronic linewidth was calculated by summing up the contributions of all electron/hole-phonon scattering processes for all phonons in the Brillouin zone as a function of E. The obtained energy dependence was approximated in the range [1.0, 3.5] eV with the expression γ = 12.60E + 3.45E 2 (3) For energies in this energy range, γ changes more than four times from 16 to 86 meV (Fig. 3). B. Overtone bands The calculated overtone Raman spectrum for EL = 2.0 eV and parallel light polarization along a zigzag line of carbon bonds is shown in Fig. 4 (bottom). It has two intense bands (2D and 2D ′) and three weaker ones (2D3, 2D4, and 2D ′′) and a much weaker one (2D5). There are also other bands of in-plane phonons close to the Γ and K points but they are either very weak, or are in the shoulders of intense bands, and in both cases are practically unobservable. The assignment of the overtone spectrum can be performed by analyzing the contribution of phonons with different ν and from different parts of the Brillouin zone of graphene. First, the contributions of the phonons with ν = 2, ..., 6 are given in Fig. 4 (top five graphs). The bands in these spectra originate from pairs of phonons TA@Γ (ν = 2), LA@Γ and TA@K (ν = 3), LA@K (ν = 4), TO@K (ν = 5), and TO′@K and LO@K (ν = 6) (see also Fig. 1). The contributions for ν = 5, 6 are by about three orders of magnitude larger than the remaining ones. 6 Secondly, the assignment of the Raman bands to definite phonons at the Γ and K points is supported by the analysis of the contributions to the spectra from different parts of the Brillouin zone for parallel light polarization with averaging over all orientations in the graphene plane. In Fig. 5, the regions with major contribution to the bands are given by shaded areas. In particular, the 2D3 band comes from phonons TA@Γ along the ΓK direction, the 2D4 band - from phonons LA@Γ along ΓM, the 2D ′′ band - from phonons LA@K along ΓK (not shown in Fig. 5), and the 2D ′ band - from phonons LO@Γ from all directions in the Brillouin zone. The 2D band originates from phonons TO@K and TO′@K, which contribute to the integrated Raman intensity of this band in ratio ≈ 10 : 1. It has long been accepted that the dominant contribution to the 2D band comes from phonons along the KM direction and that the contribution along the ΓK direction vanishes because of interference.16 Our graphs in Fig. 4 and 5 for ν = 5, 6 confirm the recent conclusion that the main contribution to this band comes from the ΓK direction and only a small part comes from the KM direction.17,18 We also confirm the result of Ref.17, that among the eight DR processes, those with scattering of an electron and a hole have dominant contribution to the bands. The overtone bands are dispersive, i.e., their Raman shift depends on the laser energy (Fig. 6). The shift increases (decreases) due to increase (decrease) of the phonon frequency away from the Γ and K points. It is approximately linear in EL with slopes of ≈ 0 for the 2D ′ band, 89 cm−1/eV (2D band), −193 cm−1/eV (2D ′′), 419 cm−1/eV (2D4), and 256 cm−1/eV (2D3) at EL = 2.0 eV . The slope of the 2D band corresponds to the experimental one of 88 cm−1/eV (Ref. [11]). The integrated Raman intensity of the overtone bands (Fig. 7) is quasi-linear in EL except for the 2D3 band. The ratio of the integrated intensities A (2D) /A (cid:16)2D ′(cid:17) has been discussed a lot in the literature because it is related to the electron-phonon matrix elements at the K and Γ points. Here, we find that this ratio depends on the energy: it varies from 4.7 for EL = 1.0 eV to 15.0 for EL = 3.5 eV; for EL = 2.4 eV it is equal to 8.7. By contrast, recent sophisticated DFT-GW calculations17 yield 21.5, which is several times larger than our value. The reason for this disagreement can be found in the use of GW corrected electron-phonon matrix elements for the LO phonon at the Γ point and the TO phonon at the K point. While the ratio of the squares of these matrix elements at the two points derived by DFT is M 2 K/M 2 Γ = 2.02, the DFT-GW result is 3.03 (Ref. [27]). Since the Raman 7 intensity depends on the square of this ratio, the DFT-GW gives an intensity ratio that is larger than the DFT one by a factor of 2.25. Our result, corrected by the same correction factor, is 19.6. Both theoretical results underestimate the experimental values of 26 (Ref. [28]) and 27 (Ref. [3]) by ∼ 25%. The origin of this underestimation is still unknown. C. Combination bands The combination Raman spectrum is calculated for EL = 2.0 eV and parallel light polar- ization along a zigzag line of carbon bonds (Fig. 8, bottom). The spectrum is dominated by two intense bands: D + D ′′ and a higher frequency one, which originate from the branches 5 + 4 and 6 + 5, respectively. There are also four other less intense bands: D ′ + D3, D ′ + D4, and D + D5. As above, the assignment is facilitated by considering the contributions to these bands from pairs of phonons with ν, ν ′ = 2, 3, 4, 5, 6. It is seen in Fig. 8 (top six graphs) that for each pair ν + ν ′ there is a single band. These bands are due to pairs of phonons TOTA@K (ν + ν ′ = 5 + 3), TOLA@K (5 + 4), LOTA@Γ (6 + 2), LOLA@Γ (6 + 3), TO′LO′@K (6 + 4), and TO′LA′@K (6 + 5) (see also Fig. 1). Unlike the case of the overtone bands, the intensity of the various combination bands varies only by one order of magnitude. The band D + D ′′ is mainly due to phonons TOLA@K with a small contribution of phonons TO′LO′@K. The phonons TO′LA′@K give rise to an intense band at ∼ 2700 cm−1. The large contribution of the phonons TO′LO′@K and TO′LA′@K is consistent with the large electron-phonon matrix element for the phonon TO′@K and the nonzero matrix element for the phonons LA′@K and LO′@K. We note that the latter contrubite to the overtone spectra as well but their bands are masked in the shoulder of the intense 2D band. The band at 2700 cm−1 also overlaps considerably with the much more intense 2D band and cannot be observed as a separate feature. Thus, the total Raman spectrum exhibits the intense 2D band, the two weak bands 2D ′ and D + D ′′(Fig. 9), as well as the two very weak bands D ′ + D3 and D ′ + D4. Similarly to the overtone bands, the combination bands are dispersive. The dependence of the Raman shift on the laser energy (Fig. 10) is almost linear with a slope of 74 cm−1/eV for the highest-frequency band, −50 cm−1/eV (D + D ′′), −54 cm−1/eV (D + D5), 210 cm−1/eV (D ′ + D4), and 123 cm−1/eV (D ′ + D3) at EL = 2.0 eV. Each slope is equal approximately to the sum of half of the slopes of the corresponding overtone bands. The calculated energy 8 dependence of the D + D ′′ band frequency reproduces fairly well the experimental data but the derived slope is almost three times larger than the experimental one of −18 cm−1/eV for energies between 1.92 and 2.71 eV (Ref. [11]). The calculated frequencies for the D ′ + D4 and D ′ + D3 bands underestimate the Raman data up to 80 and 30 cm−1, respectively. Similar underestimation is present in other precise calculations.17,29 On the other hand, the frequency slopes of the bands D ′ + D4 and D ′ + D3 are in fair agreement with the measured ones of 221 cm−1/eV and 140 cm−1/eV, respectively.30 We note that neither we, nor the authors of Ref. [17], have found any observable combination band for phonons TOLA@Γ though such band has been established30 by fitting a low-intensity Raman band with two Lorentzians for two overlapping bands assigned to phonons TOLA@Γ and LOLA@Γ. The calculated integrated Raman intensity is shown in Fig. 11. The curves are increasing functions of laser energy for all bands except for the band D ′ + D3 and the highest frequency one. For the most studied D + D ′′ band, we find A (2D) /A (cid:16)D + D ′′(cid:17) = 12.2. Introducing the DFT-GW correction factor for the electron-phonon matrix element of the TO phonon at the K point of 1.76 (Ref.27), for the latter ratio we obtain the value of 21.5, which agrees with the previous estimate17 of 16 and the experimental one3 of ≈ 21. The inset of Fig. 11 shows the dependence of the DFT-GW corrected ratios A (2D′) /A (2D) and A (cid:16)D + D ′′(cid:17) /A (2D) as a function of EL. Both curves agree well with the theoretical ones17 and the available experimental data.3 D. Dependence on the electronic linewidth Most of the previously reported two-phonon Raman spectra have been calculated for energy-independent γ. In order to study the effect of this approximation we performed calculation of the integrated intensity Ac for constant linewidth γc = γ(EL = 2.0 eV) = 39 meV as well. We found that, for any of the two-phonon peaks discussed above, the calculated ratio of the integrated intensities, A and Ac, for variable and constant linewidth, respectively, can be fitted very well with the expression A/Ac = (γc/γ)2. Therefore, A can be written as A = f /γ2, where f depends on EL and does not depend on γ. It is clear from Figs. 7 and 11 that the rate of change of f as a function of EL is different for the different two-phonon bands. Thus, the ratio of the integrated intensities for any pair of two-phonon bands 1 and 2, A1/A2 = f1/f2, should depend on EL in agreement with the recent conclusions.17 We 9 note that, in the latter paper, deviation from the 1/γ2 behavior has been found. The correct dependence should be established by comparison with experimental Raman data collected at more values of the laser energy. The intensity ratio for the 2D and 2D ′ overtone bands can be obtained using the simple tight-binding analytical formulas:31 A (2D) ∝ 2 (γK/γ)2 and A (cid:16)2D ′(cid:17) ∝ (γΓ/γ)2. Here, γ = γΓ + γK is the total electronic linewidth; γΓ,K ∝ ELM 2 Γ,K/ωΓ,K are the linewidths due to the LO phonon at the Γ point and the TO phonon at the K point, M 2 Γ,K, and ωΓ,K are the corresponding square of the electron-phonon matrix element and the phonon frequency, respectively. The predicted ratio of the integrated intensity of the two bands is a constant, equal to 3 (or 7.9), evaluated with electronic linewidth from DFT (or DFT-GW).27 The latter values are about 7 (or 2.7) times smaller than those of the precise derivations at EL = 2.4 eV, presented here and in Ref. [17]. The disagreement for the intensity ratio can be sought in the approximations used in the theoretical scheme of Ref. [31]. E. Polarization dependence of the bands Let us choose a coordinate system in the graphene plane with x axis along a zigzag line of carbon bonds (and therefore the y axis is along an armchair line of carbon bonds), and set eL = (cos α, sin α) and eS = (cos β, sin β). In the case of parallel and cross polarizations, the intensity is not angle-dependent because of the high-symmetry of graphene. On the contrary, for fixed polarization angle and variable analyzer angle, or vice versa, the intensity depends only on the difference α − β and does not depend on the sign of this difference. Therefore, the Raman intensity, Eq. (2), can be written as I = (cid:16)I − I⊥(cid:17) cos2 (α − β) + I⊥ (4) The intensity for parallel and cross polarized light, I and I⊥, can be obtained by fitting this expression to the calculated angular dependence of the intensity. Expression, similar to Eq. (4), holds for the integrated intensity as well. Fig. 12 shows the polarization dependence of the ratio of the calculated integrated intensities A⊥/A at EL = 2.0 eV. As could be expected, all two-phonon bands have similar angular dependence of the Raman intensity, which follows from Eq. (4). The derived values of the ratio A⊥/A are: 0.3678 (2D3), 0.4195 (2D4), 0.3321 (2D ′′), 0.3377 (2D), 0.4075 10 (2D ′), 0.3718 (D ′ + D3), 0.4073 (D ′ + D4), 0.3401 (D + D5), 0.3350 (D + D ′′). The intensity ratio for all bands is between 0.33 and 0.42. The ratio for the 2D band corresponds to the value 1/3 derived within a simple tight-binding model31 and the tight-binding estimate19 of 0.32, and is in fair agreement with the experimental value19 of 0.3. So far as we are aware, there are no reports on the polarization dependence of the other two-phonon bands. IV. CONCLUSIONS We have presented a complete theoretical treatment of the two-phonon Raman bands of graphene within a non-orthogonal tight-binding model with no adjustable parameters and no other approximations. We have calculated the laser energy and polarization dependence of the Raman shift and intensity. In particular, the ratio of the integrated Raman intensity for parallel and cross polarized light for all bands is between 0.33 and 0.42. The agreement of our results with available experimental data is very good and the predictions can be used for further comparison to experiment and for support in the assignment of the two- phonon Raman spectra. Furthermore, we have made definite conclusions on the dominant contributions to the Raman scattering amplitude. Our computations can easily be extended to include higher-order Raman processes. ACKNOWLEDGMENTS V.N.P. acknowledges financial support from Facult´es Universitaires Notre Dame de la Paix, Namur, Belgium, and Grant No.71/05.04.2012 of University of Sofia, Sofia, Bulgaria. 1 A. K. Geim and K. S. Novoselov, Nature Materials 6, 183 (2007). 2 F. Bonaccorso, Z. Sun, T. Hasan, and A. C. Ferrari, Nature Photonics 4, 611 (2010). 3 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). 4 L. M. Malard, M. A. Pimenta, G. Dresselhaus, and M. S. Dresselhaus, Phys. Rep. 473, 51 (2009). 11 5 M. S. Dresselhaus, A. Jorio, M. Hofmann, G. Dresselhaus, and R. Saito, Nano Lett. 10, 751 (2010). 6 V. Z´olyomi, J. Koltai, and J. Kurti, Phys. Stat. Sol. B 248, 2435 (2011). 7 C. Thomsen and S. Reich, Phys. Rev. Lett. 85, 5214 (2000). 8 S. Reich and C. Thomsen, Phil. Trans. R. Soc. Lond. A 362, 2271 (2004). 9 R. Saito, A. Jorio, A. G. SouzaFilho, G. Dresselhaus, M. S. Dresselhaus, and M. A. Pimenta, Phys. Rev. Lett. 88, 027401 (2002). 10 A. Gruneis, R. Saito, T. Kimura, L. G. Can¸cado, M. A. Pimenta, A. Jorio, A. G. SouzaFilho, G. Dresselhaus, and M. S. Dresselhaus, Phys. Rev. B 65, 155405 (2002). 11 D. L. Mafra, G. Samsonidze, L. M. Malard, D. C. Elias, J. C. Brant, F. Plentz, E. S. Alves, and M. A. Pimenta, Phys. Rev. B 76, 233407 (2007). 12 A. Gruneis, J. Serrano, A. Bosak, M. Lazzeri, S. L. Molodtsov, L. Wirtz, C. Attaccalite, M. Krisch, A. Rubio, F. Mauri, and T. Pichler, Phys. Rev. B 80, 085423 (2009). 13 D. M. Basko, S. Piscanec, and A. C. Ferrari, Phys. Rev. B 80, 165413 (2009). 14 L. G. Can¸cado, M. A. Pimenta, R. Saito, A. Jorio, L. O. Ladeira, A. Grueneis, A. G. SouzaFilho, G. Dresselhaus, and M. S. Dresselhaus, Phys. Rev. B 66, 035415 (2002). 15 J. Maultzsch, S. Reich, and C. Thomsen, Phys. Rev. B 70, 155403 (2004). 16 R. Narula and S. Reich, Phys. Rev. B 78, 165422 (2008). 17 P. Venezuela, M. Lazzeri, and F. Mauri, Phys. Rev. B 84, 035433 (2011). 18 R. Narula, N. Bonini, N. Marzani, and S. Reich, Phys. Stat. Sol. B 248, 2635 (2011). 19 D. Yoon, H. Moon, Y.-W. Son, G. Samsonidze, B. H. Park, J. B. Kim, Y. P. Lee, and H. Cheong, Nano Lett. 8, 4270 (2008). 20 D. Porezag, T. Frauenheim, T. Kohler, G. Seifert, and R. Kaschner, Phys. Rev. B 51, 12 947 (1995). 21 V. N. Popov and L. Henrard, Phys. Rev. B 70, 115407 (2004). 22 V. N. Popov and P. Lambin, Phys. Rev. B 73, 085407 (2006). 23 V. N. Popov, L. Henrard, and P. Lambin, Phys. Rev. B 72, 035436 (2005). 24 V. N. Popov and P. Lambin, Phys. Rev. B 74, 075415 (2006). 25 R. M. Martin and L. M. Falicov, in Light Scattering in Solids I, Vol. 8, edited by M. Cardona (Springer-Verlag, Berlin, 1983). 26 J. Kurti, V. Z´olyomi, A. Gruneis, and H. Kuzmany, Phys. Rev. B 65, 165433 (2002). 12 27 M. Lazzeri, C. Attaccalite, L. Wirtz, and F. Mauri, Phys. Rev. B 78, 081406(R) (2008). 28 F. Alzina, H. Tao, J. Moser, Y. Garc´ıa, A. Bachtold, and C. M. Sotomayor-Torres, Phys. Rev. B 82, 075422 (2010). 29 K. Sato, J. S. Park, R. Saito, C. Cong, T. Yu, C. H. Lui, T. F. Heinz, G. Dresselhaus, and M. S. Dresselhaus, Phys. Rev. B 84, 035419 (2011). 30 C. Cong, T. Yu, R. Saito, G. F. Dresselhaus, and M. S. Dresselhaus, ACS Nano 5, 1600 (2011). 31 D. M. Basko, Phys. Rev. B 78, 125418 (2008). ) 1 - m c ( y c n e u q e r f n o n o h P 2400 2200 2000 1800 1600 1400 1200 1000 800 600 400 200 0 LO x TO ZO LA x x TA KM K' x TO' xxx LO' LA' x x x x x ZA K M FIG. 1. Calculated phonon dispersion of graphene along the high-symmetry directions in the Brillouin zone.22 The six phonon branches are marked by the acronyms LO, TO, ZO, LA, TA, and ZA. The letters O and A stand for "optical" and "acoustic", respectively; L, T, and Z denote in-plane longitudinal, in-plane transverse, and out-of-plane atomic displacement, respectively. The crosses mark the phonons which play major role in the DR processes. Inset: the hexagonal Brillouin zone of graphene with the special points Γ, M, K, and K ′ . 13 G G G eh+, eh− ep hp eh− FIG. 2. Schematic representation of the DR processes in graphene. The solid lines are a cross- section of the Dirac cones at the K and K ′ points of the Brillouin zone. The dashed lines are virtual processes: "eh+" and "eh−" are electron (e) - hole (h) creation and annihilation processes, respectively; "ep" and "hp" are electron and hole scattering processes by a phonon (p). DR processes can also take place between bands at the K or K ′ point only. ) V e m ( 100 80 60 40 20 0 0 1 2 3 4 Energy (eV) FIG. 3. Calculated electronic linewidth γ(E) (solid symbols). The dashed line is a fit to the calculated data. The fitting function is given by Eq. (3). 14 g 2LO@G 2LA@G 2TA@K 2TO'@K 2TO@K 2LA@K 6 5 4 3 2 x4 ) s t i n u . b r a ( y t i s n e n t I n a m a R x1000 x5000 x5000 2TA@G 0 2D3 2D4 2D5 2D 2D'' 2D' 500 1000 1500 2000 2500 3000 3500 Raman Shift (cm-1) FIG. 4. Calculated overtone Raman spectrum for EL = 2.0 eV (bottom). The contributions from phonons with ν = 2, ..., 6 are shown in the top graphs. The notation of the Raman bands 2D3, 2D4 and 2D5 is taken from Ref. [17]. 15 n = 2 n = 3 K' M K 2TA@G n = 5 2LA@G n = 6 2TO@K 2LO@G 2TO'@K FIG. 5. The contribution to the overtone bands from phonons with ν = 2, 3, 5, 6 from different parts of the rhombic Brillouin zone of graphene. The graph for ν = 4 for phonons 2LA@K (not shown) is similar to that for ν = 5. 16 G 3500 3000 2500 2000 1500 1000 500 0 2D' 2D 2D'' 2D4 2D3 1.0 ) 1 - m c ( t f i h S n a m a R ~0 89 -193 419 256 1.5 2.0 2.5 EL (eV) 3.0 3.5 FIG. 6. Calculated dependence of the Raman shift of the overtone bands on EL (solid symbols). The numbers are the slopes of the curves for EL = 2.0 eV. The empty symbols are experimental data.11 ) s t i n u . b r a ( y t i s n e n t I n a m a R d e t a r g e t n I 105 2D 104 2D' 103 102 101 100 10-1 2D'' 2D3 2D4 1.0 1.5 2.0 2.5 EL (eV) 3.0 3.5 FIG. 7. Calculated dependence of the integrated Raman intensity of the overtone bands on EL (solid symbols). 17  TO'LA'@K TO'LO'@K LOLA@G LOTA@G TOTA@K TOLA@K D'+D3 D'+D4 D+D5 D+D" 6+5 6+4 6+3 6+2 5+4 5+3 ) s t i n u . b r a ( y t i s n e t n I n a m a R x10 x10 x10 x30 e l t i i T s x A Y 1600 1800 2200 2000 Raman Shift (cm-1) 2400 2600 2800 FIG. 8. Calculated combination Raman spectrum for EL = 2.0 eV (bottom). The contributions from pairs of phonons with ν + ν′ = 5 + 3, ..., 6 + 5 are shown in the top graphs. ) s t i n u . b r a ( y t i s n e n t I n a m a R overtone spectrum x20 D+D" combination spectrum 2D total two-phonon spectrum 2D' 2200 2400 2800 2600 3000 Raman Shift (cm-1) 3200 3400 FIG. 9. Calculated total two-phonon Raman spectrum for EL = 2.0 eV (bottom). The constituting combination and overtone spectra are also shown (middle and top). 18 2800 2600 2400 2200 2000 1800 1600 ) 1 - m c ( t f i h S n a m a R 74 -50 -54 210 123 D+D" D+D5 D'+D4 D'+D3 1.0 1.5 2.0 2.5 EL (eV) 3.0 3.5 FIG. 10. Calculated dependence of the Raman shift of the combination bands on EL (solid sym- bols). The highest-frequency band is due to TO ′ ′ LA @K phonons. The numbers are the slopes of the curves for EL = 2.0 eV. The empty symbols - squares11 and circles,30 are experimental data. 104 ) s t i n u . D+D" 103 b r a ( y t i s n e n t I n a m a R d e a r g e n t t I 102 D'+D3 101 D'+D4 100 10-1 D+D5 1.0 1.5 A(D+D")/A(2D) A(2D')/A(2D) 0.10 0.08 0.06 0.04 0.02 1.0 1.5 2.0 2.5 3.0 3.5 2.0 2.5 EL (eV) 3.0 3.5 FIG. 11. Calculated dependence of the integrated Raman intensity of the combination bands on EL (solid symbols). The highest-intensity band at low energies is due to TO Inset: The DFT-GW corrected ratios A(cid:16)2D in comparison with experimental data3 (large symbols). The scale is the same as in Fig. 7. @K phonons. ′′(cid:17) /A (2D) (small symbols) ′(cid:17) /A (2D) and A(cid:16)D + D LA ′ ′ 19 ) s t i n u . b r a ( y t i s n e n t I n a m a R d e a r g e n t t I 1.2 1.1 1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.0 (a ) (0a ) (a - 90,a ) 0 10 20 30 40 50 60 70 80 90 a (degree) FIG. 12. Calculated polarization dependence of the integrated Raman intensity of the 2D band (solid symbols). The angle α is relative the x axis along a zigzag line of carbon bonds. The polarizer and analyzer angles are given by the notation (αβ). The curves for parallel and cross polarization, (αα) and(α − 90, α), as well for fixed polarizer direction along the x axis and variable analyzer orientation (0α) are shown. The solid line for polarization (0α) is a fit of Eq. (4). The obtained ratio of the integrated intensities is A⊥/A ≈ 0.34. 20 a 

1308.5078

1

1308

Negative-U Centers as a Basis of Topological Edge Channels

We present the findings of the studies of the silicon sandwich nanostructure that represents the high mobility ultra - narrow silicon quantum well of the p - type (Si - QW), 2 nm, confined by the delta - barriers, 3 nm, heavily doped with boron on the n - type Si (100) surface. The ESR studies show that nanostructured delta - barriers confining the Si - QW consist predominantly of the dipole negative - U centers of boron, which are caused by the reconstruction of the shallow boron acceptors along the <111> crystallographic axis, 2B(0) = B(+) + B(-). The electrically ordered chains of dipole negative - U centers of boron in the delta -barriers appear to give rise to the topological edge states separated vertically, because the value of the longitudinal, Gxx = 4e2/h, and transversal, Gxy = e2/h, conductance measured at extremely low drain-source current indicates the exhibition of the Quantum Spin Hall effect. Besides, the Aharonov - Casher conductance oscillations and the 0.7(2e2/h) - feature obtained are evidence of the interplay of the spontaneous spin polarisation and the Rashba spin - orbit interaction that is attributable to the formation of the topological edge channels. We discuss the phenomenological model of the topological edge channel which can demonstrate the ballistic, Aharonov - Chasher effect or Josephson junction behaviour in dependence on the disorder in the distribution of the negative - U dipole centers in the upper and down delta - barriers.

cond-mat.mes-hall

cond-mat

Negative-U Centers as a Basis of Topological Edge Channels Nikolay Bagraeva, Eduard Danilovskiia, Wolfgang Gehlhoffb, Leonid Klyachkina, Andrey Kudryavtseva, Anna Malyarenkoa and Vladimir Mashkovc aIoffe Physical Technical Institute, Polytekhnicheskaya 26, 194021 St. Petersburg, Russia bTechnische Universitaet Berlin, D-10623, Berlin, Germany cSt. Petersburg State Polytechnical University, Polytekhnicheskaya 29, 195251 St. Petersburg, Russia Abstract. We present the findings of the studies of the silicon sandwich nanostructure that represents the high mobility ultra- narrow silicon quantum well of the p-type (Si-QW), 2 nm, confined by the δ-barriers, 3 nm, heavily doped with boron on the n- type Si (100) surface. The ESR studies show that nanostructured δ-barriers confining the Si-QW consist predominantly of the dipole negative-U centers of boron, which are caused by the reconstruction of the shallow boron acceptors along the <111> crystallographic axis, 2B0→B+ + B-. The electrically ordered chains of dipole negative-U centers of boron in the δ–barriers appear to give rise to the topological edge states separated vertically, because the value of the longitudinal, Gxx = 4e2/h, and transversal, Gxy = e2/h, conductance measured at extremely low drain -source current indicates the exhibition of the Quantum Spin Hall effect. Besides, the Aharonov-Casher conductance oscillations and the “0.7·(2e2/h)-feature” obtained are evidence of the interplay of the spontaneous spin polarisation and the Rashba spin -orbit interaction that is attributable to the formation of the topological edge channels. We discuss the phenomenological model of the topological edge channel which can demonstrate the ballistic, Aharonov-Chasher effect or Josephson junction behaviour in dependence on the disorder in the distribution of the negative-U dipole centers in the upper and down δ–barriers. Keywords: Quantum Spin Hall Effect, Double Barrier Structure, Edge Channels, Negative-U centers, Boron in Silicon. PACS: 72.25.Dc, 72.20.-i INTRODUCTION The investigation of edge spin-dependent transport is the subject of a considerable amount of research due to their applications in the modern physics direction – spintronics [1]. One of the best candidate on the role of such device that is able to demonstrate the topological helical edge channels with quantized conductance at high temperature appears to be the high mobility p -type silicon quantum well (Si-QW), 2 nm, confined by the δ-barriers heavily doped with boron [2]. Here the measurements of longitudinal and transversal voltage by top bias gating in the absence of the external magnetic field reveal the edge channels in such silicon nanosandwiches. FIGURE 1. (a) The experimental dependence of the longitudinal, Gxx, and Hall, Gxy, conductance on the top gate voltage Vtg at the stabilized drain-source current Ids = 0.25 nA demonstrate the Quantum Spin Hall effect. T = 77 K. The inset in the middle of Fig.1 depicts the scheme of the experimental device. (b) Longitudinal conductance Gxx as a function of the top gate voltage demonstrates the “0.7·(2e2/h)-feature” at Ids = 0.5 nA and the Aharonov–Casher oscillations at Ids = 5 nA which are caused by the Rashba SOI changes. RESULTS AND DISCUSSION The inset in the middle of Fig.1 represents the schematic diagram of the sandwich nanostructure device that demonstrates a perspective view of the p-type silicon quantum well (Si-QW), 2 nm, confined by the δ-barriers heavily doped with boron, NB = 5 × 1021 cm-3, on the n-type Si (100) surface. Thus, the  - barriers represent really alternating arrays of the smallest undoped microdefects and doped dots with dimensions restricted to the value of 2 nm. The value of the boron concentration determined by the SIMS method seems to indicate that each doped dot located between undoped microdefects contains two impurity atoms of boron. Since the b oron dopants form shallow acceptor centers in the silicon lattice, such high concentration has to cause a metallic-like conductivity. Nevertheless, in order to identify the edge states, the longitudinal and transversal (Hall) voltage were measured by varying the top gate bias voltage at different values of the highly-stabilized drain-source current in the silicon nanosandwiches. The longitudinal, Gxx = 4e2/h, and transversal, Gxy = e2/h, conductance, registered at extremely low value of the stabilized drain-source current, 0.25 nA, indicate the exhibition of the Quantum Spin Hall effect (Fig.1). Besides, the Aharonov-Casher conductance oscillations and the “0.7·(2e2/h)-feature” obtained at small-scale drain- source current changes from 0.25 nA to 0.5 nA are evidence of the interplay of the spontaneous spin polarisation and the Rashba spin-orbit interaction that is attributable to the formation of the topological edge channels [2,3]. This conductivity properties of the  - barriers between which the Si-QW is formed was quite surprising, when one takes into account the high level of their boron doping. To eliminate this contradiction, the ESR technique has been applied for the studies of the boron centers packed up in dots [4]. The angular dependences of the ESR spectra at different temperatures in the range 3.8÷27 K that reveal the trigonal symmetry of the boron dipole centers have been obtained with the Brucker -Physik AG ESR spectrometer at X-band (9.1-9.5 GHz) by the rotation of the magnetic field in the {110}-plane perpendicular to a {100}-interface (Bext = 0, 180 parallel to the Si-QW plane, Bext = 90 perpendicular to the Si-QW plane) (Fig. 2a, b, c and d). No ESR signals are observed, if the Si-QW confined by the  - barriers is cooled down in the external magnetic field (Bext) weaker than 0.22 T, with the persistence of the amplitude and the resonance field of the trigonal ESR spectrum as function of the crystallographic orientation and the magnetic field value during cooling down process at Bext ≥ 0.22 T (Fig. 2a, b and c). With increasing temperature, the ESR line observed changes its magnetic resonance field position and disappears at 27 K (Figure 2d). FIGURE 2. The trigonal ESR spectrum observed in field cooled ultra-shallow boron diffusion profile that seems to be evidence of the dynamic magnetic moment due to the trigonal dipole centers of boron inside the  - barriers confining the Si-QW which is persisted by varying both the temperature and magnetic field values. Bext  <110> (a),  <112> (b),  <111> (c, d). Rotation of the magnetic field in the {110}-plane perpendicular to a {100}-interface (Bext = 0o, 180o  interface, Bext = 90o  interface), ν = 9.45 GHz, T = 14 K (a, b, c) and T=21 K (d).  FIGURE 3. (а) Model for the elastic reconstruction of a shallow boron acceptor, which is accompanied by the formation of the trigonal dipole (B+ - B-) centers as a result of the negative-U reaction: 2Bo  B+ + B-. (b) A series of the dipole negative-U centers of boron located between the undoped microdefects that seem to be a basis of nanostructured  - barriers confining the Si- QW. The observation of the ESR spectrum is evidence of the fall in the electrical activity of shallow boron acceptors contrary to high level of boron doping. Therefore, the trigonal ESR spectrum observed seems to result from the dynamic magnetic moment that is induced by the exchange interaction between the small hole bipolarons whi ch are formed by the negative-U reconstruction of the shallow boron acceptors, 2B0B+ + B-, along the <111> crystallographic axis (Fig. 3a and b) [5]. These small hole bipolarons localized at the dipole boron centers, B + - B-, seem to undergo the singlet-triplet transition in the process of the exchange interaction through the holes in the Si - QW thereby leading to the trigonal ESR spectrum (Fig. 2a, b, c and d). Besides, electrostatically ordered dipole negative-U centers of boron seem to give rise to the topological superconductive edge states separated spatially in two δ–barriers. The electrical resistivity, magnetic susceptibility and specific heat measurements are actually evidence of the superconductor properties for the δ-barriers, Tc =145 K, 2Δ = 44 meV, Hc2 = 0.22 T [6]. The superconducting gap, 2Δ, that was also found by the local tunnelling spectroscopy (LTS) measurements has to reveal the THz emission due to the Josephson junctions self-assembled in the silicon sandwich nanostructure. FIGURE 4. (a) The Fiske steps revealed by the Ids-Uxx dependence. Inset - the dI/dU-U dependence. (b) The negative differential resistance controlled efficiently by varying the top gate voltage that verifies the GHz Josephson emission from the silicon sandwich nanostructure. T = 77 K. The difference in the negative and positive Ug effect is caused by the p-n junction presence in the device design. The absence of the CV characteristics symmetry appears to be due to disordering the dipole negative-U centers of boron which seem to give rise to the properties for topological insulator. The HTS properties for the δ-barriers have been shown to result from the transfer of the small hole bipolarons through the negative-U dipole centers of boron, which cause the GHz generation under applied voltage or optical pumping [6]. This generation can be enhanced by introducing the internal microcavities in the Si-QW plane by varying the dimensions of the sandwich nanostructure using the photolithography technique. The dimensions of the sandwich nanostructure used in this work correspond to the formation of the 9.3 GHz microcavities. The Fiske steps experiment was used to identify the Josephson generation revealed by measuring the CV characteristics of the device prepared (Fig. 4a and b). It is well known that the magnetic twist of the phase difference along the Josephson junction leads to the so -called Fiske steps [7]. These are nearly constant-voltage steps in the CV characteristic at voltages Vn ≈ (h/2e)csn/2L, where n = 1, 2,…and L is the junction length perpendicular to the magnetic field, cs is the Swihart velocity of electromagnetic waves that are created by the Josephson generation and are able to propagate in the junction plane [8]. The Fiske steps in the CV characteristic and in the dI/dU-U dependence are detected in the laboratorial magnetic field by the potential Uxx measurements under the conditions of the stepwise stabilization of the drain-source current, Ids (Figs. 4a and b). The Fiske steps observed are in agreement with the theory suggested by Kulik, with the Swihart velocity, cs, equal to 1.6 × 107 m/s [8]. The negative differential resistance in the longitudinal CV characteristics, Ids > 0, that results from the GHz Josephson emission appears to be controlled efficiently by varying the top gate bias voltage, Ug (Fig. 4b). It necessary to be noticed, that the effect is maximal when the top gate voltage is equal to zero. This fact seems to be evidence of the weak disorder in the chain of the dipole boron centers by varying the top gate voltage that results in the asymmetry of the Josephson longitudinal CV characteristic in the presence of the p-n junction. FIGURE 5. The models of the HTS topological channel which can demonstrate the ballistic (a), Aharonov -Chasher effect (b) or Josephson junction (c) behavior in dependence on the negative-U dipole centers disorder in the upper and down δ–barriers that is controlled by varying the longitudinal electric field value. The high sensitivity of the longitudinal conductance to the drain-source current value appears to be due to disordering the dipole negative-U centers of boron by varying the external electrical field value which seem to give rise to the properties for the topological edge states separated spatially for the opposite spin orientation in the two δ- barriers. Topological insulators are electronic materials that have a bulk band gap like an ordinary insul ator but have protected conducting states on their edge or surface. These states are theoretically possible due to the combination of spin-orbit interactions and time-reversal symmetry [1]. The topological edge channels appear to consist of the superconducting chains divided by weakly disordered dipole boron centers. Thus, the topological edge channels represent a series of parallel Josephson junctions shunted by ballistic conductors. Figs. 5 demonstrate examples of possible topological channel configurations. In dependence on external electrical field and edge channel quality it is appear to be realized the ballistic Si-QW topological channel or the ring shunted quantum dot contact resulting in the Aharonov-Casher effect registration or the ballistic shunted Josephson junction, (Figs. 5a, b and c, respectively). However, upper and down δ–barriers confining the Si-QW are differently disordered that is due to the device design, which results in the anisotropy of the spin-dependent transport of 1D holes. ACKNOWLEDGMENTS The work was supported by the programme of fundamental studies of the Presidium of the Ru ssian Academy of Sciences “Quantum Physics of Condensed Matter” (grant 9.12); programme of the Swiss National Science Foundation (grant IZ73Z0_127945/1); the Federal Targeted Programme on Research and Development in Priority Areas for the Russian Science and Technology Complex in 2007–2012 (contract no. 02.514.11.4074), the SEVENTH FRAMEWORK PROGRAMME Marie Curie Actions PIRSES-GA-2009-246784 project SPINMET. REFERENCES 1. M. Z. Hasan and C. L. Kane, Rev. Mod. Phys. 82, 3045-3067 (2010). 2. N. T. Bagraev, N. G. Galkin, W. Gehlhoff, L. E. Klyachkin, A. M. Malyarenko, J. Phys.: Condens. Matter 20, 164202- 164217 (2008). 3. K. J. Thomas, J. T. Nicholls, M. Y. Simmons, M. Pepper, D. R. Mace and D. A. Ritchie, Phys. Rev. Lett. 77, 135 (1996). 4. N.T. Bagraev, A.D. Bouravleuv, L.E. Klyachkin, A.M. Malyarenko, W. Gehlhoff, Yu.I. Romanov, S.A. Rykov, Semiconductors 39(6), 716-728 (2005). 5. W. Gehlhoff, N. T. Bagraev, L. E. Klyachkin, Materials Sci.Forum 196-201, 467-472 (1995). 6. N. T. Bagraev, V. A. Mashkov, E. Yu. Danilovsky, W. Gehlhoff, D. S. Gets, L. E. Klyachkin, A. A. Kudryavtsev, R. V. Kuzmin, A. M. Malyarenko, V. V. Romanov, Appl. Magn. Res. 39, 113-135 (2010). 7. M. D. Fiske, Rev. Mod. Phys. 36, 221-225 (1964). 8. I. O. Kulik, JETP Lett. 2, 134-138 (1965). 

1503.07353

1

1503

Photophoresis in Single Walled Carbon Nanotubes

We report specifically two things, one the phenomenon of optically induced motion in pristine single walled carbon nanotubes (SWNT) on the macro-scale and other the consequent separation of metallic and semiconducting enriched SWNT aggregates. Experiments provide direct evidence of both positive and negative photophoresis of SWNTs in solution i.e. motion away and towards light respectively. This optically induced motion was found to be dependent on frequency and intensity of light. Aggregates of pristine SWNT, moving under UV and Visible lamp were separated and characterized using absorption and Raman spectroscopy. Aggregates separated from pristine SWNT show enrichment in metallic or semiconducting SWNTs, depending on the spectral frequency of the lamp. Photophoresis in selective SWNTs show direct relation between frequency of illumination and absorption of specific nanotubes. The observed phenomenon is also verified using pre-separated, metallic and semiconducting SWNTs in solution. Metallic SWNTs show enhanced photophoresis with an UV lamp, whereas pre-separated semiconductor SWNT exhibit motion with a mercury lamp with broadband visible frequency. However, under IR lamp, both metallic and semiconducting enriched SWNT show continuous motion, as seen for un-separated pristine SWNTs, where no specificity was observed and all particles were seen moving.

cond-mat.mes-hall

cond-mat

Abstract: Photophoresis in Single Walled Carbon Nanotubes Gopannagari Madhusudana,1Vikram Bakaraju1 and H. Chaturvedi1,a) 1Dept. of Physics, Indian Institute of Science Education & Research, Pune 411008 We report specifically two things, one the phenomenon of optically induced motion in pristine single walled carbon nanotubes (SWNT) on the macro-scale and other the consequent separation of metallic and semiconducting enriched SWNT aggregates. Experiments provide direct evidence of both positive and negative photophoresis of SWNTs in solution i.e. motion away and towards light respectively. This optically induced motion was found to be dependent on frequency and intensity of light. Aggregates of pristine SWNT, moving under UV and Visible lamp were separated and characterized using absorption and Raman spectroscopy. Aggregates separated from pristine SWNT show enrichment in metallic or semiconducting SWNTs, depending on the spectral frequency of the lamp. Photophoresis in selective SWNTs show direct relation between frequency of illumination and absorption of specific nanotubes. The observed phenomenon is also verified using pre-separated, metallic and semiconducting SWNTs in solution. Metallic SWNTs show enhanced photophoresis with an UV lamp, whereas pre- separated semiconductor SWNT exhibit motion with a mercury lamp with broadband visible frequency. However, under IR lamp, both metallic and semiconducting enriched SWNT show continuous motion, as seen for un-separated pristine SWNTs, where no specificity was observed and all particles were seen moving. a)Electronic mail:[email protected]  Directed motion of particles provides an important paradigm for specific functionalization and bottom up self-assembly of nanoparticles and biomolecules into functional devices. The motion of particles has been widely reported using electric, magnetic and thermal gradients.1-3 Efficient processes of separation, selective binding in particles, advances in biological assays, and microfluidic devices are envisioned using directed motion of micro/nano particles or biomolecules.4, 5 Photophoresis, which is optically-induced motion in particles has been studied from the observations of Ehrenfratin in 19166 to the motion of micron sized aerosol, carbon or smoke particles in stratospheres under various external conditions.7, 8 The dependence of photophoresis on various parameters such as size, shape, conductivity, and refractive index, specifically in the case of metal nanoparticles has also been analytically reported.9-11 The first report of photophoresis in an aqueous solution was by Barkas12 in 1926, but since then, there has been limited experimental observations with respect to nanoparticles in soluion.13, 14 Motion in single walled carbon nanotubes (SWNT) due to electro-magnetic fields, and high-energy optical traps has been reported.15, 16 Here, we report large macro-scale motion in aggregates of pristine SWNTs by simple lamps. Aggregates of SWNTs show both positive and negative photophoresis i.e. motion away or towards the light source. Our experiments distinctly demonstrate controlled directed motion of the aggregates of pure pristine SWNTs. Under uniform UV or visible illumination, SWNTs show distinct separation of the aggregates, and motion is observed only in part of the aggregates while rest remains at the bottom of the vial. Separated particles show enrichment in metallic SWNTs, whereas enhanced motion and enrichment is observed for semiconducting SWNTs under visible illumination. Photophoresis in selective SWNT show direct relation between the spectral frequency of illumination and dielectric property of the SWNTs used, whether metallic or semiconducting. Separation of semiconducting and metallic 2  SWNT is an area of active research.17, 18 We believe optically directed motion in absorbing SWNT may lead to a non-surfactant and non-destructive process of separation of SWNT of specific diameters. Considering both photophoresis and nanoparticles has wide implications across diverse fields such as atmospheric sciences, colloidal and nanosciences19, 20; we believe optically-induced directed motion in SWNT as reported here, may be important for diverse applications such as micro-fluidics, optical sensors, actuators and future technologies requiring stability and separation of nanoparticles. FIG. 1. (a) Conceptual model showing temperature difference (ΔT) in aggregate of SWNTs in light with intensity(I) and wavelength(λ), inducing motion. (b) Controlled motion of aggregated pristine SWNT's with directed beam of light. (c) show timeline of negative photophoresis in aggregated SWNT's. (d) Timeline of optical induced motion and distinct separation of specific nanotubes under UV light. Light induced motion in particles occurs due to the temperature difference (ΔT) on the surface of particle. If the surface exposed to the light gets hotter than the surface facing away from the light, then the particle moves away from light as positive photophoresis. However, in certain cases depending on particle size and refractive index, the surface facing away from light may be 3  hotter than the surface towards light. In such cases, negative photophoresis is observed and particles are directed towards light.21Photophoresis in individual cylindrical particle is caused by the localized, non-uniform heat distribution on the particle.22 A particle absorbing light at higher temperature than the surrounding gas will experience photophoretic force depending on the accommodation coefficient Δα.23 Photophoretic velocity for a cylindrical particle subjected to an electric field, E0, is given as . Where, photophoretic asymmetry factor depends on the complex refractive index and the normalized size λ of the particle,21, 24 λ being wavelength of incident radiation and is the dimensionless electric field distribution. Hence, direction of motion significantly depends on dielectric response of the particle to the optical excitation. Depending on the asymmetry factor (J1), photophoretic velocity may be positive or negative for an individual cylindrical particle absorbing light. Numerical results for optically-induced forces in aggregates of metal nanoparticles also show dependence on plasmonic resonance frequencies.10Hence, photophoresis in nanoparticles such as either metallic or semiconducting SWNTs is expected to depend on the characteristic plasmonic or absorption frequency respectively, i.e. on complex refractive index or dielectric property of the particle. Pristine pure SWNTs, as produced are mixture of nanotubes with both metallic and semiconducting properties. Aggregates of pure SWNT have nanotubes of various diameters and dielectric properties, signifying different absorption coefficients and refractive indices. Hence, when these aggregates of pristine SWNT are subjected to resonant optical frequency, motion in absorbing particle is observed due to significant, induced photophoretic forces, as shown in Fig.1. 4  Pristine SWNTs (purity 99.8 %) were used, as purchased from Nanointegris, without any further modification or purification. In this report we have only considered pure SWNTs in solution of DMF; however we also observe similar optically induced motion of SWNT in other organic solvents like acetone and chloroform. Fig 1 b. show controlled motion of the SWNT aggregates as directed by the optical beam. As shown, aggregated floc can be moved either to the top or to the bottom of the vial, controlled by the narrow beam of light. Photophoresis in these aggregates of SWNT show a complex dependence on the intensity of the beam. Samples were illuminated using optical fiber coupled through the mercury lamp. At lower intensities (10-20 J/m2) particles moves slowly towards the light source and remains in the inclined beam flux, however for higher intensities (140-220 J/m2) same aggregates show motion away from the central flux of the beam. Particles show rotating motions there by dissipating energy, while the aggregated floc as a whole remains in the directed narrow beam. Although experimental observations confirm controlled, directed photophoretic motion, care was taken so as to minimize the theromophoretic effects in solution if any. The solvent DMF also do not absorb in the UV or visible frequencies used for illumination, thus negating any doubts of the thermal convection due to absorption of the solvent. Experiments in the specially designed glass contraption in Fig 1 (c) clearly shows negative photophoresis, a part of the aggregated floc can be seen moving over 10 cm of the channel connecting two flasks, in the direction of light in just 2 minutes of illumination. The glass contraption clearly demonstrates potential application of the phenomenon for optically directed, controlled transportation of nanoparticles functionalized by various molecules, as required by biological or microfluidic applications and separation technologies. 5  To further probe the dependence of the phenomena on the frequency of illumination, aggregated SWNTs were uniformly illuminated from top , using commercially available UV (125W, 352 nm) and broadband mercury spectral lamp (80W). Under uniform optical illumination, part of the aggregated pristine SWNT in DMF solution moves to the top of the vial, while rest remains at the bottom. As shown in the Fig.1 (d), within 50 seconds particles moves to the top of the vial under UV lamp and clearly separates out from the rest. Part of the aggregate of SWNT under visible lamp also shows similar motion and distinctly separates out as shown in Fig 2 (a). However, aggregates under IR lamp show neither any distinct separation nor directed motion. Unlike the distinct separation under UV or visible illumination, showing motion only in selective part of the aggregated flocs; rapid, continuous motion in complete aggregated floc under NIR lamp is observed across the vial within 10 seconds (check video clip as supplementary Information). SWNTs moving to the top under lamps were separated from the ones at the bottom of vial and characterized using UV-Vis-NIR absorption spectrometer and Lab Ram Raman spectrometer with 632 nm laser line. In case of SWNTs the optical absorption significantly depends on the diameter of the nanotube.25 Absorption of pristine one-dimensional carbon nanotubes differ significantly from bulk material with its characteristic van hove singularities. The transitions are relatively sharp and characteristic to specific diameter and band gap energies of semiconducting nanotubes. However, significant absorption by pristine SWNTs in UV is essentially due to the plasmonic resonance of metallic nanotubes. Hence absorption spectra of the pristine SWNT show characteristic features of both metallic M11 (350-600 nm) and semiconducting inter-band transitions S11 (900-1400 nm) and S22 (550-900 nm) as shown in Fig.2(c, d). The SWNTs collected from the top of the vial under UV lamp, shows enhanced absorption in the plasmonic 6  metallic M11 band; as also seen for SWNTs collected from bottom of the vial, under visible frequency lamp. Similarly, nanotubes separated from the top of vial under mercury lamp and bottom of vial under UV lamp, show similar and significant increase in absorption at visible NIR frequency. Since NIR frequency is directly related to the band gap absorption by semiconducting tubes, increase in NIR absorption in separated aggregates indicate enrichment of semiconducting SWNTs. FIG. 2. (a, b) Optically induced motion in selective pristine SWNTs and consequent separation under visible frequency mercury and UV lamps. (c, d) Normalized absorption spectra of pristine SWNT aggregates separated from bottom (red) and top (black) of vials under mercury and UV lamps. (e, f) The G-band of raman spectra aggregates pristine SWNTs shows motion and enrichment of metallic and semiconducting nanotubes at the top of vial under UV and mercury lamps. 7  Further characterization by Raman spectroscopy confirms this enrichment in separated SWNTs, exhibiting optically induced motion and switching depending on the frequency of illumination. Analysis of the Raman spectra of SWNT show two main components of the G band, as shown in Fig.2(c), one at 1592 cm−1 (G+) and the other broadly peaked at 1568 cm−1 (G−).26 The G+ feature is associated with vibrations of carbon atom along the nanotube axis (LO phonon mode) and the G− feature, is associated with vibrations along the circumferential direction of the SWNT (TO phonon). The G−is highly sensitive to the metallic to semiconducting ratio in pristine SWNT aggregates. Metallic enriched SWNT show Breit–Wigner–Fano line shape; whereas lorentzian line-shape is observed for semiconducting enriched SWNTs. An increase in G−/G+ ratio, and corresponding broadening suggests increase in metallic SWNTs,27 as shown in Fig.2 (e, f). Broadening and changes in relative intensity of the G+, G-band for pristine SWNT collected from the top and bottom of the vial under UV and visible lamps; verifies enrichment of specific metallic and semiconducting SWNT, as was indicated by similar changes in either the plasmonic M11 or semiconducting S11 band of the absorption spectra. Both samples, ones separated from the top of the vial under UV illumination and the other from the bottom of vial under visible lamp; show similar increase in G−/G+ band ratio. In contrast, the ones separated from the top under visible illumination show similar enrichment as ones from the bottom of the vial under UV lamp. Additionally, the radial breathing mode (RBM) of the separated nanotubes shows enrichment of specific diameter tubes due to motion under UV and mercury irradiation. Fig.3(a, b) show the RBM of pristine SWNTs separated under UV and visible illumination. The frequency of the RBM is inversely proportional to the diameter of individual SWNT.28 8  FIG. 3. (a, b) Radial Breathing Mode (RBM) showing enrichment of specific diameter of SWNTs under mercury and UV lamps. Aggregates of SWNTs separated from the top (black) under mercury lamp and bottom (red) under UV (Black) lampshow similar enrichment. (c) histogram as plotted from the RBM data showing enrichment in specific diameter of the SWNTs at the top or bottom of the vial due to motion under visible mercury lamp (green) and UV lamp (violet). The RBM of each sample was fit using multiple lorentzian functions to identify relative enrichment in individual diameter tubes in the Raman spectra of pristine SWNT. The intensity of the RBM in each separated SWNT sample was normalized with respect to similar diameter tubes fit in the un-separated pristine SWNT sample that was used as a control. The diameter of the SWNT was calculated using ωRBM = (αRBM/d) + αbundle.29 Where d is the diameter of nanotube and αbundle, αRBM are constants associated with bundling effect and scaling factor in RBM spectra of nanotubes respectively. Individual RBM of each sample was fitted and normalized using multiple lorentzian peaks, representing different diameter of SWNTs. Corresponding histogram Fig.3(c)is plotted using RBM spectra of separated SWNTs. It shows relative enrichment of specific diameter SWNT and the direction of motion with respect to un-separated pristine- SWNTs. The discernible changes in the RBM and the G band of the vibrational spectra provide further evidence enrichment of specific nanotubes; due to optically induced motion. The motion of SWNT aggregates under UV illumination conclusively show enrichment in metallic SWNT 9  with specific diameters. Similarly, under visible mercury lamp, SWNT aggregates moving to the top of vial are found enriched in specific diameter of semiconducting SWNTs. FIG. 4. Photophoresis in aggregates of pre separated (95% pure) metallic (left) and semiconducting (right) SWNTs. (a) Aggregates at bottom of vials in dark. (b) show optically induced motion in metallic SWNTs under UV lamp. Whereas (c), shows enhanced motion of semiconducting SWNTs under visible frequency mercury lamp. However as shown in (d) both metallic and semiconducting SWNTs show enhanced motion under IR lamp. Photophoretic separation and consequent enrichment of metallic and semiconducting SWNTs from the pristine SWNTs was further verified using pre-separated semiconductor, metallic enriched solutions of SWNTs. Pre-separated semiconductor and metallic SWNT were purchased from Nanointegris (with 95% Purity) and used without any further modification. As shown in Fig.4(a-d), samples of semiconductor and metallic enriched aggregates show complimentary motion under UV and broadband mercury lamp. However, in case of illumination by IR lamp, there is no specificity in direction of motion or in diameter of nanotubes. Solutions of both metallic and semiconducting SWNT show continuous motion under IR lamp, as observed in case 10  of pristine SWNTs. Photophoresis in pre-separated metallic or semiconducting SWNTs, along with comparative enrichment shown in the absorption and Raman spectra of separated pristine SWNT indicates photophoresis in SWNT aggregates depends on the respective absorption either by metallic or semiconducting SWNTs under UV and mercury lamps. Large scale, optically controlled motion and consequent separation of SWNTs opens up an interesting areas, on one hand it potentially provides solutions for issues of active concern such as sepration and controlled transportation in SWNTs on other hand it challenges conventional fundamental understanding seeking and active research. Motion in selective SWNTS based on frequency of simple lamp, proposes to be an efficient way for both large scale and specific separation of SWNTs. Since, motion in SWNT depends on dielectric response we believe the phenomenon should be useful for directed motion and efficient separation in other nanoparticles too. Photophoresis in SWNT aggregates presents an exciting avenue for optically controlled motion and induced separation in other nanoparticles such as metallic nanoparticles and quantum dots based on either their size or electro-optical properties. Along with its applications in various advanced technologies, SWNT also provide an opportunity to understand novel effects in colloidal solutions of one-dimensional nanoparticles. Clearly both theoretically and experimentally much research is needed, to fully understand the phenomenon of photophoresis in the aggregates of one dimensional particles like SWNTs, having different diameters, with distinct optical and dielectric properties. However, in this study, we experimentally demonstrate and attempt to provide possible theoretical reasons for optically induced motion especially negative photophoresis and enrichment in separated SWNTs. In conclusion, we propose this optically controlled motion and subsequent separation of SWNT in solution may play an important role for directed self-assembly and technological advances in diverse fields such as 11  molecular devices, micro-fluidics, biotechnology and industrial applications requiring separation of nanoparticles. Acknowledgments Authors are deeply indebted to Dr. Jordan C. Poler and Andrea Giordano for invaluable scientific input, Ramanujan fellowship (SR/S2/RJN-28/2009) and funding agencies DST (DST/TSG/PT/2012/66), Nanomission (SR/NM/NS-15/2012) for generous grants. REFERENCES 1C.-X. Zhang and A. Manz, Analytical Chemistry 75 (21), 5759-5766 (2003). 2S. Lee, T. A. Hatton and S. Khan, Microfluidics and Nanofluidics 11 (4), 429-438 (2011). 3J. Brzoska, F. Brochard-Wyart and F. Rondelez, Langmuir 9 (8), 2220-2224 (1993). 4P. R. Gascoyne and J. Vykoukal, Electrophoresis 23 (13), 1973 (2002). 5A. Terray, J. Oakey and D. W. M. Marr, Science 296 (5574), 1841-1844 (2002). 6P. Ehrenfest, Proceedings of the Amsterdam Academy 19, 576-597 (1916). 7N. Tong, Journal of Colloid and Interface Science 51 (1), 143-151 (1975). 8O. Jovanovic and H. Horvath, Journal of Aerosol Science 31, Supplement 1 (0), 831-832 (2000). 9A. Akhtaruzzaman and S. Lin, Journal of Colloid and Interface Science 61 (1), 170-182 (1977). 10V. P. Drachev, S. V. Perminov and S. G. Rautian, Optics express 15 (14), 8639-8648 (2007). 11S. Arnold and M. Lewittes, Journal of Applied Physics 53 (7), 5314-5319 (1982). 12W. W. Barkas, The London, Edinburgh, and Dublin Philosophical Magazine and Journal of Science 2 (11), 1019-1026 (1926). 13J.-P. Abid, M. Frigoli, R. Pansu, J. Szeftel, J. Zyss, C. Larpent and S. Brasselet, Langmuir 27 (13), 7967-7971 (2011). 14H. Lei, Y. Zhang, X. Li and B. Li, Lab on a Chip 11 (13), 2241-2246 (2011). 15S. Tan, H. A. Lopez, C. W. Cai and Y. Zhang, Nano Letters 4 (8), 1415-1419 (2004). 16S. Blatt, F. Hennrich, H. v. Löhneysen, M. M. Kappes, A. Vijayaraghavan and R. Krupke, Nano letters 7 (7), 1960-1966 (2007). 17Y. Maeda, S.-i. Kimura, M. Kanda, Y. Hirashima, T. Hasegawa, T. Wakahara, Y. Lian, T. Nakahodo, T. Tsuchiya and T. Akasaka, Journal of the American Chemical Society 127 (29), 10287-10290 (2005). 18M. S. Arnold, A. A. Green, J. F. Hulvat, S. I. Stupp and M. C. Hersam, Nat Nano 1 (1), 60-65 (2006). 19K. Nogi, M. Hosokawa, M. Naito and T. Yokoyama, Nanoparticle technology handbook. (Elsevier, 2012). 20D. W. Keith, Proceedings of the National Academy of Sciences 107 (38), 16428-16431 (2010). 21H. J. Keh and H. J. Tu, Colloids and Surfaces A: Physicochemical and Engineering Aspects 176 (2), 213-223 (2001). 12  22O. Jovanovic, Journal of Quantitative Spectroscopy and Radiative Transfer 110 (11), 889-901 (2009). 23M. Knudsen, Radiometer pressure and coefficient of accommodation. (AF Høst & søn, 1930). 24H. Fu, Y. Huang and L. Gao, Physics Letters A 377 (39), 2815-2820 (2013). 25M. J. O'connell, S. M. Bachilo, C. B. Huffman, V. C. Moore, M. S. Strano, E. H. Haroz, K. L. Rialon, P. J. Boul, W. H. Noon and C. Kittrell, Science 297 (5581), 593-596 (2002). 26A. Jorio, M. Pimenta, A. Souza Filho, R. Saito, G. Dresselhaus and M. Dresselhaus, New Journal of Physics 5 (1), 139 (2003). 27R. Krupke, F. Hennrich, H. v. Löhneysen and M. M. Kappes, Science 301 (5631), 344-347 (2003). 28M. S. Dresselhaus, G. Dresselhaus, R. Saito and A. Jorio, Physics Reports 409 (2), 47-99 (2005). 29S. M. Bachilo, M. S. Strano, C. Kittrell, R. H. Hauge, R. E. Smalley and R. B. Weisman, Science 298 (5602), 2361-2366 (2002). 13 

1209.2040

1

1209

Nonequilibrium Landau-Zener-Stuckelberg spectroscopy in a double quantum dot

We study theoretically nonequilibrium Landau-Zener-St\"uckelberg (LZS) dynamics in a driven double quantum dot (DQD) including dephasing and, importantly, energy relaxation due to environmental fluctuations. We derive effective nonequilibrium Bloch equations. These allow us to identify clear signatures for LZS oscilations observed but not recognized as such in experiments [Petersson et al., Phys. Rev. Lett. 105, 246804, 2010] and to identify the full environmental fluctuation spectra acting on a DQD given experimental data as in [Petersson et al., Phys. Rev. Lett. 105, 246804, 2010]. Herein we find that super-Ohmic fluctuations, typically due to phonons, are the main relaxation channel for a detuned DQD whereas Ohmic fluctuations dominate at zero detuning.

cond-mat.mes-hall

cond-mat

Nonequilibrium Landau-Zener-Stuckelberg spectroscopy in a double quantum dot 1I. Institut fur Theoretische Physik, Universitat Hamburg, Jungiusstrasse 9, 20355 Hamburg, Germany P. Nalbach1, J. Knorzer1, and S. Ludwig2 2 Center for NanoScience and Fakultat fur Physik, Ludwig-Maximilians-Universitat Munchen, Geschwister-Scholl-Platz 1, D-80539 Munchen, Germany (Dated: November 9, 2018) We study theoretically nonequilibrium Landau-Zener-Stuckelberg (LZS) dynamics in a driven double quantum dot (DQD) including dephasing and, importantly, energy relaxation due to en- vironmental fluctuations. We derive effective nonequilibrium Bloch equations. These allow us to identify clear signatures for LZS oscilations observed but not recognized as such in experiments [Pe- tersson et al., Phys. Rev. Lett. 105, 246804, 2010] and to identify the full environmental fluctuation spectra acting on a DQD given experimental data as in [Petersson et al., Phys. Rev. Lett. 105, 246804, 2010]. Herein we find that super-Ohmic fluctuations, typically due to phonons, are the main relaxation channel for a detuned DQD whereas Ohmic fluctuations dominate at zero detuning. PACS numbers: 03.65.Yz,85.35.Gv,73.21.La Quantum electronic devices, as qubits realized by dou- ble quantum dots (DQD), require coherence times which exceed their quantum operation time during which the DQD is typically strongly driven by external voltage pulses. Tremendous research efforts studied semiconduc- tor based devices to achieve coherent quantum control [1 -- 7]. Many fluctuation sources of the noisy solid state envi- ronment, which act on the electron in the DQD and thus destroy coherence, were revealed but a comphrehensive picture is elusive. Furthermore, driving by voltage pulses causes an intrinsic nonequilibrium situation in which re- laxation competes with driving [8 -- 10] which renders a theoretical description of the dissipative nonequilibrium dynamics highly nontrivial. Here, we theoretically study the dissipative nonequi- librium dynamics of a single electron charge qubit de- fined in a DQD embedded in a noisy solid state envi- ronment driven by voltage pulses. While DQD charge qubits have relatively short coherence times, this disad- vantage is compensated by the possibility of fast quan- tum operations. We model the DQD and its dissipa- tion as a quantum two-level system in an open quan- tum system approach [11]. We determine the dissipa- tive nonequilibrium real time dynamics (initialized by applying voltage pulses) by deriving effective nonequilib- rium Bloch equations (NBEs). These allow fast numer- ical treatment in contrast to numerical exact methods [8, 9] and thus allow a comphrehensive analysis of re- cent experiments by Petersson et al. [1] and Dovzhenko et al. [12]. In these experiments quantum control of a single electron was achieved by means of applying ultra short voltage pulses to control gates of the laterally de- fined DQD. In these time ensemble measurements the DQD was cycled (with 40 MHz repetition rate) between two different ground state configurations while its aver- age charge occupation was continuously detected via the electric current through a capacitively coupled quantum point contact (QPC). The applied voltage pulses gener- 2 = T ⋆ 2 ate Landau-Zener-Stuckelberg (LZS) dynamics [13] and we can identify so far unexplained features in the exper- imental data as signatures of coherent LZS oscillations. In the experimental ensemble measurements the de- phasing time T ⋆ 2 due to slow noise [7, 14] is much shorter than relaxation times T1 and thus dominates the decoher- −1 + (2T1)−1. Relaxation ence times T2 since T −1 and dephasing could be caused by thermal phonons, in- trinsic or externally triggered charge noise [15, 18], or detector back-action [16 -- 22]. Ref. [1, 12] neglected re- laxation and employed solely an 1/f noise model [23 -- 27] for dephasing which allowed them to describe the detun- ing dependence of the observed decoherence of the DQD. The experimentally observed steady state charge occu- pation of the DQD is, however, heavily influenced by re- laxation [10]. Therefore, in this letter, we go beyond this simple dephasing model and include various environmen- tal fluctuation spectra to describe dephasing and relax- ation. This allows us to simulate the full nonequilibrium real time dynamics experimentally studied. We find that the observed visibilty reduction is caused by relaxation. Moreover, by analyzing the measured LZS dynamics we identify the full environmental fluctuation spectrum as a sum of three processes. In addition to slow noise, already considered in ref. [1], which causes detuning dependent dephasing, super-Ohmic fluctuations, as typically origi- nating from phonons, are the main relaxation channel for the detuned DQD. Near zero detuning, however, Ohmic fluctuations dominate relaxation. The latter also limit the experimentally observed maximal decoherence time of T2 ∼ 7 ns. Modelling of pulse driven DQD We model the single electron DQD using a two-level system Hamiltonian H = 1 2 ∆σx + 1 2 ǫ(t)σz (1) where the eigenstates of σz correspond to the electron in the left / right dot, ǫ(t) is the level detuning and ∆ is the interdot tunnel splitting [28]. Experimentally, ε 0 εp εp pt tr rept vt = pt tr+2 2 3 1 1 ] 1 - s n [ 2 T / 1 = 2 γ 0.3 0.1 -40 -30 ε -20 p [µeV] -10 0 FIG. 1. (Left): Energy detuning profile for a voltage pulse with rise time tr, pulse plateau time tp, pulse duration tv = tp+2tr. If cycled the pulse periodically repeats with repetition time trep. (Right): Decoherence rate versus detuning ǫp. Full theory (black full line), Ohmic dephasing (red dashed line) and decoherence due to relaxation (blue dot-dashed line). gate voltage pulses are applied to change the level detun- ing as sketched in Fig. 1(left). Initially the detuning is ǫ0 = ǫ(0) ≫ ∆ and the DQD in the according ground state (0,1) [electron is in the right dot]. A voltage pulse then drives the system within a rise time tr to a plateau detuning ǫp close to resonance (ǫ = 0) and keeps it there for a plateau time tp. Then the DQD is driven back within tr to the initial detuning and the probability P(1,0) of occupation of the excited state (1,0) [electron in the left dot] is studied as a function of the pulse duration tv = 2tr + tp. For a quantum mechanical two-level sys- tem without dissipation we expect coherent oscillations in the voltage pulse time tv. A voltage change on a single gate [1, 12] not only af- fects the level detuning but also causes a common energy shift of both states (1, 0) and (0, 1) which is not relevant in the following and thus neglected. The sharp corners of ǫ(t) in Fig. 1(left) are smoother in reality and might even contain oscillatory features (due to a finite band- width transfer function). At voltage pulse times tv & 2tr deviations between linear voltage ramps and more accu- rate descriptions are negligible. Solving the quantum dynamics of the DQD with a sin- gle applied voltage pulse (details given below) results in the probability P(1,0)(t = tv) after the voltage pulse, plot- ted in Fig. 2 as a function of tv and plateau detuning ǫp. In order to model the experiments [1, 12] we use ∆/h = 4.5 GHz, ǫ0 = 200µeV ≃ 11∆ and tr = 35 ps which corresponds to the fastest experimental achievable rise time [29]. The sweep speed vp = ǫp − ǫ0/tr can be approximated by v0 = ǫ0/tr for ǫ0 ≫ ǫp [slope in Fig. 1(left)] which provides an indication on the overall adia- baticity of the dynamics [13]. For v0 = 200µeV/35ps ≃ 11∆2/, Fig. 2 shows coherent oscillations between states (0,1) and (1,0) with frequency E/h and eigenenergy E = q∆2 + ǫ2 p of Eq. (1). At zero detuning P(1,0)(tv) oscillates between 0 and 1. With increased ǫp the os- FIG. 2. P(1,0)(tv) for the fully coherent case (T1 = T2 = T ⋆ 2 = ∞), right after a single voltage pulse, vs. level detuning ǫp and pulse duration tv = tp + 2tr for ǫ0 = 200µeV (ǫ0 = 2000µeV) for the main figure (inset). cillation frequency grows while its amplitude gradually decreases. Fig. 2 shows a clear asymmetry: the visibility is larger for ǫp < 0 compared to ǫp > 0. Remarkably, this asymmetry was observed experimentally, too (but not explained) [1]. The asymmetry decreases with increasing sweep speed as highlighted by the inset of Fig. 2 which shows an almost symmetric P(1,0)(tv) for ǫ0 = 2000 µeV leading to v0 & 110∆2/. We conclude that the asym- metry is solely an effect of adiabaticity when the DQD is driven through the avoided crossing at zero detuning. For ǫp . 0, the quantum system accumulates, during the voltage pulse, not only phase due to the coherent oscilla- tion of the electron between the two dots but also due to the superposition state occupied between two successive Landau-Zener transitions [13]. As such, the asymmetry is a clear signature of coherent LZS oscillations. Periodically Cycled Pulses In an ensemble measure- ment with continuous charge detection, as in Ref. [1] with trep = 25 ns, it is essential to choose trep ≫ tv to ensure readout of the charge occupation after application of the pulses (of duration tv . T2). An interpretation in terms of an ensemble measurement, which simply averages over many independent shots, further requires trep ≫ T2 and, interestingly, trep ∼ T1, where T1 is the (thermal) en- ergy relaxation time which depends on the detuning: for trep ≫ T1 initialization into configuration (1,0) is guar- anteed, but the visibility of the continuous measurement is close to zero as mostly (1,0) is occupied; for trep ≪ T1 initialization into (1,0) independently of P(1,0) -- right af- ter the pulse -- is impossible. No matter of the choice of pulse sequence, continuously cycled pulses will result in a steady state which, in principle, contains the information of dephasing and energy relaxation times [10]. In order to model the experimental repeated pulse train sequence we repeat our simulation after the first cycle (up to the repetition time trep) with the final statis- tical operator of the previous cycle as initial state. This procedure is repeated until the population ¯P(1,0) [mea- ∆/h = 3.3 GHz ∆/h = 4.5 GHz ∆/h = 6.6 GHz 3 0,4 0,4 0,4 0,2 0,2 0,2 ) ) ) 0 0 0 , , , 1 1 1 ( ( ( P P P 0 0 0 0 0 0 500 500 500 1000 1000 1000 tv [ps] tv [ps] tv [ps] 1500 1500 1500 2000 2000 2000 FIG. 4. tv = tp + 2tr for various tunnel couplings. ¯P(1,0) for zero detuning versus voltage pulse time with time dependent momentary equilibrium req tanh[β 1 2 E(t)] and time dependent decay rates 2 βE(t)]u2(t)G[E(t)] γ1(t) = x (t) = (4) (5) coth[ 1 π 2 2 γ1(t) + Γ2(t) γ2(t) = 1 where Γ2(t) = 1/T ⋆ 2 (t) is the dephasing time. Herein, φ(t) = arctan[ǫ(t)/∆], u(t) = cos φ(t), and v(t) = sin φ(t). The presented nonequilibrium Bloch equations with time-dependent rates and momentary equilibrium describe adequately dissipative Landau-Zener dynamics, i.e. the regime of competition between driving and relax- ation, for weak DQD-environment coupling. We ensured this by extensive comparisson with numerical exact re- sults [8, 9]. Details will be presented elsewhere. To proceed, we need the spectra of fluctuations acting on the DQD. Estimates can be gained by the observations for fixed DQD parameters of Petersson et al. [1] of a relaxation time T1(ǫ0) = 10 ns at the initial detuning ǫ0, a decoherence time of T2 ∼ 7 ns at zero detuning and the detuning dependent dephasing times presented in Fig.4c of Ref. [1]; all at the temperature T = 80 mK. Relaxation rates Charge fluctuations and bulk phonons likewise couple to the DQD. In order to include both, we consider an Ohmic fluctuation spectrum with s = 1 (charge fluctuations) as well as a super-Ohmic one with s = 3 (phonons) [11]. The rate (4) reflects a one- boson process and accordingly only yields a substantial relaxation if the spectrum at the eigenenergy of the DQD is finite, i. e. E(t) ≪ ωc. 1 At fixed T , ∆ and ǫ0 we can determine the coupling strength α (assuming E(t) ≪ ωc) using the known energy relaxation time T1(ǫ0), and estimate T1(ǫ = 0) using Eq. (4) and u(t) = ∆/E(t): For Ohmic fluctuations T s=1 (ǫ = 0) = (∆/p∆2 + ǫ2 0) · T1(ǫ0) while for super-Ohmic fluc- tuations T s=3 0/∆) · T1(ǫ0). With the reported T1(ǫ0) = 10 ns we find T s=1 (ǫ = 0) ≃ 900 ps as- 1 suming Ohmic fluctuations compared to T s=3 (ǫ = 0) ≃ 110 ns assuming super-Ohmic fluctuations. According to Eq. (5) energy relaxation causes an upper bound of the decoherence T2 ≤ 2T1. The decoherence time of T2 = 7 ns, observed in the experiment at ǫ ≃ 0 [1], lies well in between our predictions T s=1 1 ≪ 7 ns ≪ T s=3 . (ǫ = 0) = (p∆2 + ǫ2 1 1 1 (Main): ¯P(1,0) [(Inset): P(1,0)(tv) with color scale FIG. 3. from 0 to 1] versus detuning ǫp and voltage pulse time tv = tp + 2tr for a driven DQD with ǫ0 = 200µeV includ- ing relaxation and dephasing. sured as average over a full cycle as plotted in Fig. 1(left)] changes by less than 0.005. This ¯P(1,0) approximates the experimentally observed steady state population. Driven dissipative dynamics Including dissipation into the driven dynamics of the DQD causes a compe- tition between driving and relaxation [8, 9] which ren- ders all but expansive numerical treatments inadequate. Following the standard approach within open quantum dynamics [11] we couple the driven two-level Hamiltonian (1) of the DQD to environmental fluctuations described by harmonic oscillators. This results in Htot = H(t) + σz 2 X k λk(bk + b† k) + X k ωkb† kbk (2) with bosonic annihilation/creation operators bk/b† k. The spectrum G(ω) = Pk λ2 kδ(ω − ωk) is typically a smooth function [11] at the energies of interest, i. e. G(ω) = 2αωs exp(−ω/ωc) with spectral exponent s, cut-off fre- quency ωc and the coupling strength α. Typically[1], ∆ ≫ γ2 ≫ γ1 with the relaxation rate γ1 = 1/T1 and the decoherence rate γ2 = 1/T2, which justifies a weak DQD-environment coupling approach. Due to the time dependence of H(t) the weak coupling Born- Markov approximation, however, fails. Assuming that the change of energy during the memory time in the en- vironment is small, allows an additional adiabatic rate approximation. For this, we switch to the time depen- dent basis where H(t) = 1 2 E(t)τx is diagonal with Pauli matrizes τj describing the eigenstates with energy dif- ference E(t) = p∆2 + ǫ(t)2. Employing a lowest or- der Born approximation for the memory kernel in the system-bath coupling [30, 31], one obtains for the time evolution of the components of the statistical operator, ρ(t) = 1 2 (1l + Pj rjτj), nonequilibrium Bloch equations ∂trx(t) = φ′(t)rz(t) ∂try(t) = −γ2(t)ry(t) −E(t)rz(t) ∂trz(t) = E(t)ry(t) −γ2(t)rz(t) −φ′(t)rx(t) −γ1(t)[rx(t) − req x (t)] (3) An Ohmic fluctuation spectrum alone -- consistent with T1(ǫ0) = 10 ns -- would result in much too fast decoher- ence at ǫ = 0 compared to the experimental results. Hence, super-Ohmic fluctuations, typically caused by phonons, are the main relaxation mechanism at ǫ0. Decoherence and Dephasing rates For a super-Ohmic spectrum Γ2(t) ≡ 0, and the decoherence rate γ2 [see Eq. (5)] would be solely determined by the energy relax- ation mechanism. The spectrum of charge noise, which is well known to cause additional dephasing, strongly de- pends on the sample and how it has been treated [15]. Often, charge noise can be assumed to be slow noise as done by Petersson et al. [1]. They use an 1/f dephas- ing model [25] which typically results from background charge noise [23, 24] and might be described in terms of sub-Ohmic noise [26, 27] with s = 0 and ωc ≪ ∆. Such slow noise is a major dephasing source in realistic devices and causes solely dephasing. We aim at unraveling the relaxation mechanisms present rather than the dephasing sources. Relaxation is not influenced by slow noise and thus we describe slow noise here using a simplified Ohmic slow noise model resulting in Γ2(t) = (4π/)v2(t)α2kBT characterized by a coupling strength α2 which we fit to the experiment. Our Ohmic slow noise model cap- tures two important points, namely, that it does not influence relaxation, i.e. ωc2 ≪ ∆, and that it cou- ples to the position of the charge in the DQD leading to Γ2(t) ∝ v2(t) = ǫ2(t)/(∆2 + ǫ2(t)), which coincides with the dependence used by Petersson et al.[1] in their 1/f noise model. As typically done, we, herein, neglect fluctuations in ∆. The red dashed line in Fig. 1(right) represents the con- tribution to decoherence by our slow Ohmic dephasing noise and thus reflects most of the decoherence (black full line in Fig. 1(right) which reproduce the measured data in Fig.4c of Ref. [1] remarkebly well) except at very small detunings ǫp ≪ ∆. For v(t) = ǫp = 0, however, the slow noise dephasing rate Γ2(t) ≡ 0 and γ2(t) = 1 2 γ1(t). Super-Ohmic fluctuations result in T s=3 (ǫ = 0) ≃ 110 ns, much longer than the zero detuning decoherence time T2 = 7 ns actually measured. To resolve this discrep- ancy, we add a third noise source, namely weak Ohmic fluctuations which contribute to relaxation at zero de- tuning in addition to the super-Ohmic contribution of phonons. The contribution of weak Ohmic fluctuations to energy relaxation at large detuning is very small. The physical origin of this mechanism could be related to fast local potential fluctuations (e. g. via voltage noise on the lead gates). 1 Our complete model includes three fluctuations spectra G(ω) = 2α1ω3e−ω/ωc1 + 3 X j=2 2αjωe−ω/ωcj (6) with ωc1, ωc3 ≫ ∆ and ωc2 ≪ ∆. Choosing α1∆2 = 8.09 · 10−5, α2 = 3.53 · 10−2 and α3 = 2.73 · 10−3 for 4 the coupling strengths reproduces the experimental ob- servations [1] of T1(ǫ0) = 10 ns, T2(ǫ = 0) ∼ 7 ns and the decoherence time versus detuning with measured data in Fig.4c of Ref. [1] and our predicted decoherence rate plotted as black full line in Fig. 1(right). Dynamics of the driven dissipative DQD Solving the nonequilibrium Bloch equations (3) with the fluctuation spectrum (6) numerically results in Fig. 3, the main re- sult, where we plot the steady state occupation ¯P(1,0) ver- sus voltage pulse time and plateau detuning for a tunnel coupling ∆/h = 4.5GHz. In comparison, the inset shows the instantaneous probability P(1,0)(t = tv) with the color code stretched by a factor of two, i. e. yellow = 1. The continuous current measurement in the QPC reduces the visibility for the coherent dynamics by a factor of 2. In comparison to the undamped case in Fig. 2 we observe an overall reduction in oscillation amplitude. On top, the LZS oscillations smear out with both, more negative detunings ǫp and longer voltage pulse times tv: ¯P(1,0) is flat in the upper right corner of Fig. 3. The same be- haviour was experimentally observed, but not explained, by Petersson et al. [1]. Our results indicate that the LZS oscillations smear out as a result of relaxation but the remaining asymmetry with respect to ǫp is a result of adiabaticity. Thus, the experimentally observed fea- ture is a clear signature of LZS oscillations. The overall qualitative agreement between our simulation results and the experimental data [1] is very good. The quantitative overestimation of the oscillation amplitude being a factor of 2 is likely due to uncertainties regarding experimental details [32]. Interdot tunnel coupling Petersson et al. [1] ob- serve an increased (decreased) oscillation amplitude for a smaller (larger) interdot tunnel coupling, i. e. ∆/h = 3.3 GHz (6.6 GHz). We find the same tendency. It results from the fact that the relaxation rate [eq. (4)] increases strongly with the tunnel coupling. Fig. 4 plots our pre- dictions for ¯P(1,0) as a function of tv for zero detuning for the three different tunnel couplings. Conclusions We studied theoretically the dissipative nonequilibrium dynamics of a single electron DQD driven by voltage pulses. We couple the DQD additionally to environmental fluctuations causing relaxation and de- phasing. Extending standard Born-Markov approaches to driven systems, we derive nonequilibrium Bloch equa- tions exhibiting time dependent rates and the momen- tary equilibrium. This approach allows efficient numer- ical simulations of the full nonequilibrium real time dy- namics and a comprehensive analysis of the experimental data [1]. We identify an asymmetric occupation of the left dot in respect to the detuning between the dots in ref. [1] as a clear experimental signature of LZS dynam- ics. A full analysis of the LZS dynamics furthermore allows us to specify the full environmental fluctuation spectrum acting on the DQD studied by Petersson et al. [1] as sum of three processes. Besides slow noise causing the detuning dependent strong dephasing, super-Ohmic fluctuations, as typically originating from phonons, are the main relaxation channel for a detuned DQD. At zero detuning, however, Ohmic fluctuations, which might be caused by gate voltage noise, dominate relaxation and are also the main cause for the decoherence at zero detuning. Achnowledgements We thank M. Thorwart for con- tinuous financial and intellectual support. J.K. acknowl- egdes support from the DFG via SFB-925. S.L. acknowl- edges support from the DFG via SFB-631, the Cluster of Excellence Nanosystems Initiative Munich, and a Heisen- berg fellowship. [1] K.D. Petersson, J.R. Petta, H. Lu, and A.C. Gossard, Phys. Rev. Lett. 105, 246804 (2010). [2] T. Fujisawa, T. Hayashi, and S. Sasaki, Rep. Prog. Phys. 69, 759 (2006). [3] R. Hanson, , L.P. Kouwenhoven, J.R. Petta, S. Tarucha, and L.M.K. Vandersypen, Rev. Mod. Phys. 79, 1217 (2007). [4] M. Pioro-Ladriere, T. Obata, Y. Tokura, Y.-S. Shin, T. Kubo, K. Yoshida, T. Taniyama, and S. Tarucha, Nature Phys. 4, 776 (2008). [5] G. Shinkai, T. Hayashi, T. Ota, and T. Fujisawa, Phys. Rev. Lett. 103, 056802 (2009). [6] L. Gaudreau, G. Granger, A. Kam, G.C. Aers, S.A. Stu- denikin, P. Zawadzki, M. Pioro-Ladri`ere, Z.R. Wasilewski and A.S. Sachrajda, Nature Phys. 8, 54 (2012). [7] H. Bluhm, S. Foletti, I. Neder, M. Rudner, D. Mahalu, V. Umansky, and A. Yacoby , Nature Phys. 7, 109 (2011). [8] P. Nalbach and M. Thorwart, Phys. Rev. Lett. 103, 220401 (2009). [9] P. Nalbach and M. Thorwart, Chem. Phys. 375, 234 (2010). [10] D. Harbusch, S. Manus, H. P. Tranitz, W. Wegscheider, and S. Ludwig, Phys. Rev. B 82, 195310 (2010). [11] U. Weiss, Quantum Dissipative Systems, 2nd ed., World Scientific, Singapore, 1998. [12] Y. Dovzhenko, J. Stehlik, K.D. Petersson, J.R. Petta, H. Lu, and A.C. Gossard, Phys. Rev. B 84, 161302(R) (2011). 5 [13] L.D. Landau, Phys. Z. Sowjetunion 2 (1932) 46; C. Zener, Proc. Roy. Soc. London A 137 (1932) 696; E.C.G. Stueck- elberg, Helv. Phys. Acta 5 (1932) 369; E. Majorana, Nuovo Cimento 9 (1932) 43. [14] B. Lee, W.M. Witzel, and S. Das Sarma, Phys. Rev. Lett. 100, 160505 (2008). [15] M. Pioro-Ladri`ere, J.H. Davies, A.R. Long, A.S. Sachra- jda, L. Gaudreau, P. Zawadzki, J. Lapointe, J. Gupta, Z. Wasilewski, and S. Studenikin, Phys. Rev. B 72, 115331 (2005). [16] R. Aguado and L.P. Kouwenhoven, Phys. Rev. Lett. 84, 1986 (2000). [17] C.E. Young, and A.A. Clerk, Phys. Rev. Lett. 104, 186803 (2010). [18] D. Taubert, M. Pioro-Ladri`ere, D. Schroer, D. Harbusch, A.S. Sachrajda, S. Ludwig Phys. Rev. Lett. 100, 176805 (2008). [19] U. Gasser, S. Gustavsson, B. Kung, K. Ensslin, T. Ihn, D.C. Driscoll and A.C. Gossard Phys. Rev. B 79, 035303 (2009). [20] G.J. Schinner, H.P. Tranitz, W. Wegscheider, J.P. Kot- thaus, and S. Ludwig Phys. Rev. Lett. 102, 186801 (2009). [21] D. Harbusch, D. Taubert, H.P. Tranitz, W. Wegscheider, and S. Ludwig, Phys. Rev. Lett. 104, 196801 (2010). [22] G. Granger et al., Nature Physics 8, 522 (2012). [23] J. Bergli, Y. M. Galperin, and B. L. Altshuler, New J. Phys. 11, 025002 (2009). [24] I. V. Yurkevich, J. Baldwin, I. V. Lerner, and B. L. Alt- shuler, Phys. Rev. B 81, 121305(R) (2010). [25] E. Paladino, L. Faoro, G. Falci, and Rosario Fazio, Phys. Rev. Lett. 88, 228304 (2002). [26] P. Nalbach and M. Thorwart, manuscript submitted. [27] P. Nalbach and M. Thorwart, Phys. Rev. B 81, 054308 (2010). [28] Notice that our definition ∆ includes a factor of 2 in comparisson to Ref.[1]. [29] Longer rise times up to 100ps under nonoptimal exper- imental conditions render the dynamics more adiabatic and thus would only strengthen our arguments below. [30] P. Nalbach, Phys. Rev. B 66, 134107 (2002). [31] P. Nalbach and M. Thorwart, J. Chem. Phys. 132, 194111 (2010). [32] Better quantitative agreement could be observed using ǫ0 & ∆. Then, relaxation for the DQD (detuned and sym- metric) could be described by Ohmic fluctuations alone. However, experimentally ǫ0 ≫ ∆ [1]. 

1106.5705

1

1106

Structure and Growth in the Living Tissue and in Carbon Nanotubes

The topological organisation of cells in a model of living tissue (the crypt of intestinal epithelium) is identical to the topological organisation of atoms in carbon nanotubes. The existing models of growth of these two structures contain identical elements. It is proposed that the growth of carbon nanotubes can depend also on other mechanisms postulated in the crypt model where the growth in the bottom is transmitted to the cylinder, and, in this case, the growth of nanotubes should involve the same peculiar lattice transformations as the ones found in the crypt models. The crypt models also suggest a possible initiation of growth in the nanotubes by the loss of carbon atoms. We consider the crystalline structures that can be formed from graphene and the structures of the living tissues as one distinct class of structures.

cond-mat.mes-hall

cond-mat

Structure and Growth in the Living Tissue and in Carbon Nanotubes Michael Pyshnov1 and Sergei Fedorov 2 1Toronto, Canada; 2Moscow, Russian Federation The topological organisation of cells in a model of living tissue (the crypt of intestinal epithelium) is identical to the topological organisation of atoms in carbon nanotubes. The existing models of growth of these two structures contain identical elements. It is proposed that the growth of carbon nanotubes can depend also on other mechanisms postulated in the crypt model where the growth in the bottom is transmitted to the cylinder, and, in this case, the growth of nanotubes should involve the same peculiar lattice transformations as the ones found in the crypt models. The crypt models also suggest a possible initiation of growth in the nanotubes by the loss of carbon atoms. We consider the crystalline structures that can be formed from graphene and the structures of the living tissues as one distinct class of structures. 1. Introduction The structure of tissues in the organisms emerges as a result of division of cells during the process of development. In the emerging tissues, cells are physically attached to each other in the order and at the place in which they appeared after cell division. To the every curious microscopist, tissues appear as ordered, even crystalline structures, although, with a complex symmetry. The tissue structure, however, presents a greater puzzle: in many tissues, the cells continue proliferation in the adult organism while these tissues and organs maintain their size and shape unchanged and the cells remain connected with each other. The first attempt to solve this puzzle was made with topological modeling of cell division in the crypt of intestinal epithelium [1]. The intestinal crypt is one of the tissues that consist of one layer of cells and can topologically be considered as a two-dimensional curved and folded lattice. In the crypt, the lattice forms a cylinder closed at one end (Fig. 1 and Fig. 2); such shape is often observed in other tissues, in plants and animals. In 2005, we published a computer simulation demonstrating the process of cell division in crypt [2]. It then appeared that the topological structure in the crypt model was identical with the later found structure of carbon nanotubes. The similarity was noted in the work on graphene where the structure of the crypt was called "a living curiosity" [3]. Furthermore, while the living tissues grow by proliferation of cells within the tissue (interstitial growth), carbon nanotubes as well as other graphene structures, similarly, are capable to grow by absorbing new carbon atoms (in this case, from the surrounding medium) and intercalating them between the elements of the already formed lattice. Here, we will compare topological structures of the crypt and the carbon nanotubes (CNT) and consider possible mechanisms of interstitial growth in these structures. 2. The model 2 Fig. 1. Photo of the crypt Fig. 2. The crypt model [1], side view and the bottom. Spaces of intestinal epithelium, inside the polygons are cells. The model turns into a carbon from [2]. Cell niclei are nanotube if vertices are atoms of carbon and the sides of the stained. polygons are chemical bonds. Pentagons are red. In the model, polygons of the bottom are fully connected to the polygons of the cylinder, so, the bottom represents a topological closure in which there are no unattached cell sides at one end of the crypt cylinder. In the CNT, the closure eliminates dangling chemical bonds at one end of the CNT cylinder. The closure at the bottom is achieved by introduction of six pentagons (we call them structural pentagons) into the hexagonal lattice. The positions of the six pentagons determine the configuration of the bottom and the configuration of the cylinder. This is an idealised model: due mainly to the variations in cell size, the actual cellular lattice is not as regular as it appears here. The crypt model, therefore, represents only the net outcome of the topological events which is dictated by the Euler's theorem. The lattice is much more regular in the CNT. 3. The lattice defects, the division wave and other mechanisms involved in the crypt growth In a hexagonal lattice, polygons other than hexagons are considered topological defects (dislocations). In a two-dimensional lattice, the introduction of some topological defects (a pentagon or a heptagon) will necessitate the removal or addition of large parts of the lattice and, if the elements of the lattice cannot differ much in size - will also force it to curve and fold. However, introduction of the pair pentagon-heptagon can cause only a minor buckling on the surface of the lattice. The important property of this 5-7 pair is that it can appear to be moving within the lattice when the elements of the lattice consecutively exchange their connections (a gliding dislocation). The 5-7 defect can also move in another direction within the lattice by the addition or the loss of the elements of the lattice (a climbing dislocation). The proposed mechanism of cell division in crypt [1] was based on the moving of the 5-7 pair along the row of hexagons, inducing the cells to divide and resulting in a division wave (Fig. 3). Such interstitial growth does not require cells to break up any existing connections. 3 (a) (b) Fig. 3. The division wave proceeding from left to right. (a) - the planes of cell division are indicated; (b) - cell shape and size are partially restored. Heptagons are blue; new cells are green. Heptagon is leading the wave; in the next step it will divide, directing the plane of division to the pentagon behind (making it a hexagon) and to the hexagon ahead (making it a heptagon). Note: since there is more than one hexagon ahead, the division wave, generally speaking, can make turns. The crucial finding in the model that allowed the crypt to grow in length while preserving its original diameter was that the division wave should first occur in the crypt bottom, in a row of hexagons situated between two pentagons (Fig. 4). Fig. 4. The division of hexagons between two structural pentagons in the bottom, from [1]. In (a), the pentagons are on the same row with three hexagons. In (b), the hexagons have divided; the new sister cells are shown connected with dots. The pentagons are in new positions. The cells in the Fig.4 divide locally in the bottom. However, if two division waves meet at the same pentagon in the bottom, a topological situation arises that gives birth to a new division wave that can now proceed into the crypt cylinder, Fig. 5, from [1]. In the process, the structure of the spiral rows of the hexagons in the cylinder undergoes a peculiar structural transformation described further below. Fig. 5. (c) - two division waves meet at the pentagon and initiate a new division wave (dotted lines); (d) - cells have divided; sister cells are connected with dots. 4 A different mechanism for the initiation of cell division was used in [2]. It was shown that the removal/death of one pentagonal cell in the bottom topologically results in the emergence of a single 5-7 pair and in the appearance of a new pentagon restoring the required number of structural pentagons in the bottom to six (Fig. 6); see [2] for more details. The emerged 5-7 pair initiates a new division wave that can proceed in different directions. If this division wave moves to the crypt cylinder, it also results in the structural transformation of the cylinder (see below). Let's already note here that the removal of a pentagonal cell is topologically equivalent to the loss of one C 2 molecule from a pentagon in the CNT. The removal of C2 molecule opens 4 dangling chemical bonds that form a new configuration: two pentagons and a heptagon. (a) (b) (c) (d) Fig. 6. The division wave initiated by the removal of a pentagon. In (b), a heptagon is formed and it is dividing in (c), directing the plane of division to one of the pentagons (here, to the pentagon that in (b) was on the left of the heptagon). In Fig. 7, we show the same division wave as in Fig. 6, but now meeting a pentagon on the other end. This wave is proceeding between two pentagons which, however, are not on the same row of hexagons, because, as it is now obvious, the division wave initiated by the removal of a pentagon is making a turn. This case is important because such division wave follows the path known in the CNT growth models as "twisted Endo-Kroto patch" [6]. It is "twisted" since the two pentagons are not on the same row of hexagons. Fig. 7. The division wave initiated by the removal of pentagon in the topological situation of the twisted Endo-Kroto patch. Consecutive cell divisions are shown, from left to right. 4. Structural transformations in the crypt cylinder The transformations accompanying the passing of the division wave in the crypt cylinder were described in [1]. Later, the model using the removal of a pentagon as the mechanism for initiation of cell division, with much greater choice of the outcomes, was studied with the computer animation program [2]; it showed the same structural transformations in the cylinder. Depending on the position and orientation of the arising 5-7 pair, the division waves proceeding to the cylinder result in changes of the pitch and threadedness of the spiral rows of hexagons in the cylinder (Fig. 8). The changes have cyclical character, i. e. after several consecutive division waves, the cylinder can be returned to its initial configuration. It is possible to keep the perimeter of the cylinder essentially constant (although, strictly speaking - slightly fluctuating). Other division waves having different direction and/or orientation of the division planes (see note to the Fig. 3) can increase the perimeter of the cylinder. However, there is no such division wave that will result in the reduction of the crypt perimeter. 5 (a) (b) (c) (d) (e) (f) (g) Fig. 8. Structural transformations in the crypt cylinder. The growth was generated by the removal of a pentagon followed by a division wave having the same orientation in all figures. Each figure was produced by the removal of the pentagon that was produced in the previous figure. This pentagon was changing its location in the bottom after every division wave. The other five pentagons of the bottom were not affected. In the CNT terminology, figures (a) to (e) are hiral structures; (f) is armchair; (g) has reverted to hiral of the same pitch as (e), but of the opposite handedness. The properties of the spiral structure (threadedness in two directions, as used in [2] and in the CNT literature) are as follows: (a) - (12,2); (b) - (11,3); (c) - (10,4); (d) - (9,5); (e) - (8,6); (f) - (7,7); (g) - (8,6). 5. Mechanisms of growth in carbon nanotubes In 1992, Endo and Kroto [4] proposed that insertion of two carbon atoms (C2 molecule) into a hexagon situated between two pentagons can serve as a basic mechanism of growth of the fullerenes and the CNT. The result was the "division" of the hexagon into two pentagons and the addition of one side to each original pentagon making them hexagons, i. e. the case described for the cell division in the crypt bottom and many times shown in CNT literature in various configurations and with various numbers of hexagons "dividing" between two pentagons; this being known as the Endo-Kroto patch. The authors [4] suggested that "active sites [for the absorption of C2 molecule] are in the vicinity of pentagons and that the process involves insertion-reorganisation process in which strain is equilibrated by diffusion of the pentagonal cusps". 6 The situation around the pentagons in the crypt bottom was described similarly: " the crypt cells are trying to eliminate dislocations and rectify the surface by means of cell division" [1], and it was suggested that the six structural pentagons that cannot be removed from the crypt bottom and can only change their positions as a result of cell division, remain there as permanent sources of instability initiating new cell division and growth in the crypt. Saito et all., also in 1992 [5], noted the state of electronic excitation that exists in a row of hexagons situated between two pentagons in the CNT bottom (where "the curvature of the bonds between two pentagons is larger than for the other bonds ") and also said that "those hexagons between the two pentagons can be considered as active sites for absorbing a C 2 molecule." They noted that when the bottom grows, the new hexagons at the perimeter of the bottom will become a part of the cylinder, producing its growth in length. The authors also described a number of topological situations, such as changing positions of the pentagons in the CNT bottom and the meeting of two dislocations at a hexagon. Yet, they made no suggestion that a meeting of two dislocations at a pentagon can result in the formation of a new 5-7 pair (as in Fig. 5). It is now accepted that the absorption of new carbon atoms and the growth of the CNT is possible due to the strain on the chemical bonds near the pentagonal defects. In both 1992 papers that defined the role of the pentagons in the CNT growth, the growth occurs only locally (between two pentagons in fullerenes and in the CNT bottom), as there are no pentagons in the cylinder; there is no absorption of carbon atoms in the cylinder. Other models of CNT growth describe the growth from the open end of the cylinder (strictly speaking, not an interstitial growth). These are interpreting the results of an experimental procedure that allows CNT growth on a grain of a catalyst attached to the CNT open end. The essential difference of the open end growth models is that they do not explicitly use pentagonal lattice defects as active sites for the absorption of C2 molecules; the open end itself (while being in contact with catalyst) represents the active site. Here, the topological situation allowing the growth of the CNT remains obscure. Yet, one interesting model suggests the topological arrangements that should allow the CNT cylinder to grow as a node extending from a flat sheet of graphene [7]. 6. Are there division waves in the CNT? The mechanisms of the CNT growth discussed above produce only a local growth at one of the ends of the cylinder. They explain the elongation of the cylinder and, in some cases, are able to grow the new part of the cylinder with a different diameter. They, however, cannot change the diameter of the already formed part of the cylinder. Yet, the typical experiments show not only axial growth, but also increase in the CNT diameters which however are seen as remarkably uniform through the entire length of the individual CNTs. Topologically, this would require the addition of new atoms along the entire length of the cylinder, but that does not seem to be possible within the proposed mechanisms. Now, the mechanisms of growth proposed in the crypt model where the division wave is proceeding along the row of hexagons in the cylinder, allow the cylinder to grow in length as well as to uniformly increase its diameter (depending on the direction of the wave and the orientation of the planes of division). We are proposing here that these mechanisms might operate in the CNT. The "division" waves in the CNT, i. e. sequential absorption of carbon atoms in a row of hexagons, can probably be responsible for maintaining the wonderful uniformity of the diameter of the CNT cylinder over its entire length as well as its uniform changes. Such mechanism can be responsible for maintaining the structure relatively free from defects. 7 Generally speaking, the "division" of several hexagons between two pentagons (Endo- Kroto patch with several hexagons) can be regarded as a division wave starting at one pentagon (or - at both) and self-annihilating when all hexagons have divided. The strain on the bonds is higher at the hexagons immediately adjacent to the pentagons, which makes these hexagons the likely initiation points for the absorption of C 2 molecules, in which case the process of absorption would represent a wave, not a simultaneous reaction in the hexagons. Let's now remember that the chemical bonds of the hexagons on a curved surface of the CNT cylinder are under the radial strain. This strain is generally smaller than the strain near the pentagonal defects in the bottom (that of course depends on the radius of the cylinder), but, it seems possible that the radial strain can assist the absorption of the C2 molecules in a row of hexagons in the cylinder once the wave has been initiated in the bottom. If the initiation of C2 absorption requires the adjacency to a pentagon, the further propagation of the wave might not. This is also the case in the row with several hexagons between two pentagons where the surface curvature at the hexagons not adjacent to pentagons is similar to the curvature at the cylinder. Furthermore, the strain on the bonds belonging to the different sides of the hexagons in the cylinder is unequal, it depends on the orientation of the hexagons in the cylinder. In the cylinder and parallel to the spiral directions, there are lines involving the bonds under greater radial strain and the lines where the strain on the bonds can be smaller or absent (that depends of the particular configuration). The C2 absorption should be favored along the lines where the strain on the bonds is greater. And that probably can play a role in defining the direction of the division waves and the orientation of the division planes (see note to the Fig. 3). If the above propositions are correct, there could be a cessation of the division waves in the cylinder and a much slower growth in the CNT having larger diameter. If the experiments prove that the individual CNT cylinder can, during the growth, change its configuration (armchair, chiral and zigzag), this can be explained as a result analogous to the one shown in Fig. 8. 7. A class of structures A number of structures can be produced from graphene: spheres, tubes, nodal growth, branching, etc. These structures closely resemble living tissues and they are also capable of interstitial growth. Interestingly, branching is observed also in the intestinal crypts. Whether produced from graphene by absorption of atoms from the surrounding medium or produced by the division of cells, these structures have common properties. While completely different in the operating physical forces, these structures are produced by the higher law, that of topology. We unite them in one class of structures. The main properties that we believe exist in this class of structures are: 1. The structures are formed by the topological laws operating only in two dimensions and that allows the lattice to produce a considerable variety of patterns. 2. Most importantly, the structures can combine in one lattice parts having different shape and symmetry. They in fact are multi-periodic crystals. 3. When, in this class of structures, dislocations are introduced, they might not lead to the weakening of the structure, but, instead, will lead to the local changes in the shape and symmetry of the contiguous lattice. Even the term dislocation here has a different meaning - this is not necessarily a destructive defect in the lattice because the lattice can remain fully 8 contiguous; the result can be that only a domain with a different shape and symmetry is formed. 4. The lattice in this class of structures is capable of interstitial growth, which would be very difficult to achieve in a three-dimensional lattice. 5. While a three-dimensional lattice is always open, i. e. it allows indefinite growth of the crystal on its edges, the two-dimensional lattice can close on itself to form a topological sphere. The growth in such closed structure can only be interstitial. 6. When only the interstitial growth is possible, there should exist intrinsic topological conditions specifying the patterns of growth and limiting the size of the structure and the number of various shapes that can be produced. In living organisms, normal cells divide only within the contiguous lattice. Evolution therefore has the above limitations. There should also exist conditions determining the possibilities and impossibilities of the process of regeneration of the lost parts. 7. A multicellular organism developing from an embryo to a certain adult shape and size can essentially be, at all stages, a topological sphere. Notes 1. In this paper, we are concerned with the climbing dislocations producing growth, i. e. the continual addition of carbon atoms capable of preserving the regular structure of the CNT. It is well known that structural transformations (the changes in the CNT diameter and length, change of handedness, etc.) can also be caused by the gliding dislocations, by the climbing dislocations that move by removal of the lattice elements and by other defects acting locally. On the other side, the gliding dislocations cannot be considered in the living tissue if the cell- cell contacts are permanent. 2. The important factor considered in suggesting different mechanisms of the CNT growth is which mechanism of the C 2 absorption is energetically more favored. There are indications that the division wave (which is a 5-7 climbing dislocation moving by the addition of lattice elements) should be energetically costly. However, the actual mechanism depends on the choice of the experimental procedure. The procedure, temperature in particular, can vary greatly, but, apparently, at higher temperatures, different reactions are possible at the same time, producing the regular structure as well as different topological defects. We wonder if the way to produce a regular structure could possibly be to reduce the temperature and the variety of fragments available in vapor and to supply larger fullerene fragments that can act as donors of C2 molecules by breaking up and releasing C2 upon the contact with the specific sites on the CNT. 3. Images with color in the article were generated with the 3-D computer simulation program [2]. References [1] M.B. Pyshnov, Topological Solution for Cell Proliferation in Intestinal Crypt. I. Elastic Growth Without Cell Loss, J. theor. Biol. 87 (1980) 189-200 [2] S. Fedorov and M. Pyshnov, Cell Division Program, http://www.cell-division-program.com/ (2005) [3] A. Cortijo and M.A.H. Vozmediano, Electronic Properties of Curved Graphene Sheets, http://www.mpipks-dresden.mpg.de/~coqusy06/SLIDES/vozmediano.pdf (2006) 9 [4] M. Endo and H.W. Kroto, Formation of Carbon Nanofibres, J. Phys. Chem. 96 (1992) 6941-6944. [5] R. Saito, G. Dresselhaus and M.S. Dresselhaus, Topological Defects in Large Fullerenes, Chem. Phys. Lett. 195 (1992) 537-542 [6] G. Brinkmann, D. Franceus, P.W. Fowler and J.E. Graver, Growing Fullerenes from Seed: Growth Transformations of Fullerene Polyhedra, Chem. Phys. Lett. 428 (2006) 386– 393 [7] F. Beuneu, Nucleation and Growth of Single Wall Carbon Nanotubes, Solid State Communications 136 (2005) 462–465 

1106.5091

2

1106

Detecting non-Abelian statistics of Majorana fermions in quantum nanowire networks

We propose a scheme in semiconducting quantum nanowires structure to demonstrate the non-Abelian statistics for Majorana fermions in terms of braid group. The Majorana fermions are localized at the endpoints of semiconducting wires, which are deposited on an \emph{s}-wave superconductor. The non-Abelian nature of Majorana fermion is manifested by the fact that the output of the different applied orders of two operations, constructed by the braid group elements, are different. In particular, the difference can be unambiguously imprinted on the quantum states of a superconducting flux qubit.

cond-mat.mes-hall

cond-mat

Detecting non-Abelian statistics of Majorana fermions in quantum nanowire networks Zheng-Yuan Xue Laboratory of Quantum Information Technology, and School of Physics and Telecommunication Engineering, South China Normal University, Guangzhou 510006, China (Dated: November 16, 2018) We propose a scheme in semiconducting quantum nanowires structure to demonstrate the non-Abelian statis- tics for Majorana fermions in terms of braid group. The Majorana fermions are localized at the endpoints of semiconducting wires, which are deposited on an s-wave superconductor. The non-Abelian nature of Majorana fermion is manifested by the fact that the output of the different applied orders of two operations, constructed by the braid group elements, are different. In particular, the difference can be unambiguously imprinted on the quantum states of a superconducting flux qubit. PACS numbers: 74.78.Na, 74.50.+r, 05.30.Pr Quantum statistics is a fundamental concept in physics which distinguishes fermions from bosons in three dimen- sions. For quasi-particles live in two dimensions, they may have two different classes of exotic statistics: Abelian or non- Abelian statistics. When one particle is exchanged in a coun- terclockwise manner with the nearest particle, the relation be- tween the initial wave function ψ(ri, rj) with the final wave function ψ′(ri, rj) is given by ψ′(ri, rj) = Mijψ(ri, rj), where ri is the position of the particle i. The particles are called Abelian anyons [1] with the statistics angle θ if Mij = exp(iθ) where θ is not zero or π (θ = 0 and π correspond to bosons and fermions, respectively). Furthermore, Mij can be the elements of a braid group, where two elements Mij and Mi′j ′ may even be non-commutative, and quasi-particles with such features are called non-Abelian anyons [2]. Potential host systems for the exotic non-Abelian statistics including the ν = 5/2 quantum Hall state [3, 4], chiral p-wave superconductors [5], topological insulator-superconductor [6] and semiconductor-superconductor structures [7 -- 9]. With the potential applications in topological quantum computation, Majorana fermions (MF) with non-Abelian statistics have at- tracted strong renewed interests. MF are a kind of self- conjugate quasi-particles induced from a vortex excitation in px + ipy superconductor. However, due to the instability of the p-wave superconducting states, its implementation still re- mains an experimental challenge. Therefore, setups with s- wave superconductor proximity effect [6 -- 9], which is more stable, is more preferred. In principle, they should allow robust topological superconducting phase without unrealistic experimental conditions. More recently, it is recognized that topologically protected states may be most easily engineered in 1D semiconducting wires deposited on an s-wave supercon- ductor [10 -- 12]: the endpoints of such wires support localized zero-energy MF. This setup provides the first realistic exper- imental setting for Kitaev's 1D topological superconducting state [13]: MF can be created, transported, fused and braided by applying locally tunable gates to the wire. However, de- tecting MF in such setup is a challenge task because they hold neutral charge. Recently, it has been shown that combine two MF into a single Dirac fermion allows the neutral quasi- particles to be probed with charge transport [14 -- 17]. But, they suffer from the same impediment: Abrikosov vortices are so massive objects that behave classically. Fortunately, alterna- tives using Josephson vortexes are proposed by introducing a controllable superconducting flux qubit [18 -- 20], which en- able one to detect unambiguously the states of MF in this 1D scenario. Recently, Zhu et al. [21] proposed a scheme to directly test the quantum statistics of the braid group for MF in cold atoms scenario. Confirming this aspect of MF is not only of signif- icant important in its own, but also the crucial and first step towards the realization of topological quantum computation. However, simulate exotic statistics of MF in a macroscopic material is another story. Here, we propose such an alterna- tive scheme to detect the non-Abelian statistics of the MF in terms of braid group in semiconducting quantum nanowires structure. The non-Abelian nature is manifested by the fact that the output of the different applied orders of two opera- tions are different, which is different form previous schemes based on charge transport [14 -- 18]. The operations are con- structed by the braid group elements of MF. Furthermore, the two different final states can be distinguished by measuring the states of MF by the superconducting flux qubit. The setup we consider is shown in Fig. 1, which is a spin- orbit coupled semiconducting wire deposited on an s-wave superconductor. Applying a magnetic field perpendicular to the superconductor surface, the Hamiltonian describing such a wire is [10] H = Z (cid:20)ψ† x(cid:18) − 2∂ 2 x 2m − µ − iue · σ∂x + VBσz(cid:19)ψx +(∆eiϕψ↓xψ↑x + h.c.)(cid:21)dx, (1) where ψαx corresponds to electrons with spin α, effective mass m, and chemical potential µ; the third term denotes spin- orbit coupling with u the strength, and σ = (σx, σy, σz) is the vector of Pauli matrices; the fourth term represents the energy shift due to the magnetic field; and the terms in the second line are the spin-singlet pairing from the s-wave superconductor via proximity effect. In the setup, the magnetic field is weak, and the supercon- ductor of the flux qubit is the conventional s-wave supercon- (cid:158)x FIG. 1: (Color online) Superconducting flux qubit with two Joseph- son junctions (pink) and an enclosing magnetic flux Φx. MF (black) are induced at the interface between a topologically trivial (blue) and a topologically nontrivial (red) section of an quantum nanowire. Gate electrodes (not shown) can be used to move the MF along the wire. ductor. The interplay of Zeeman effect, spin-orbit coupling, and the proximity to an s-wave superconductor drive the wire into a chiral p-wave superconducting state [10 -- 12], providing that the wire is long compared to the superconducting coher- ence length (ξ ≃ 40 nm for the superconducting substrate being Nb). Specifically, when ∆ < VB, µ lies inside of the zero-mode energy gap, the wire is in the Kitaev's topolog- ical phase [13], which supports Majorana modes; otherwise the system is an ordinary superconductor (topological triv- ial). The zero-mode excitation gap and µ dependence is [12] B − ∆2 the topological phase with end MFs emerge, or a topologi- cally trivial phase. Thus, applying a gate voltage uniformly allows one to create or remove the MF. To avoid gap closure, A "keyboard" of local tunable gate electrodes to the wire [12] is used to control whether a region of the wire is topological or not. For InAs quantum narowire, assuming VB ∼ 2∆ and u ∼ 0.1eV A, the gap for a 0.1µm wide gate is of order 1K [12]. It is note that heavy-element wires and/or narrower gates could generate even larger gap. Therefore, MF are induced at the interface between topologically trivial and nontrivial sec- tions of the quantum nanowire. E0 = (cid:12)(cid:12)(cid:12)VB −p∆2 + µ2(cid:12)(cid:12)(cid:12) . For µ < µc = pV 2 For a pair of MF, they can be combined to form a com- plex fermionic states c = γ2 + iγ1, which can be occupied 1i or unoccupied 0i, differ in fermion parity, and therefore 2-fold degenerated. A winding of one MF around another is associated with a unitary transformation in the subspace of degenerated ground states. For 2N MF, such unitary transfor- mations form a set named braid group, which is generated by elementary interchanges of neighboring MF [22]. For our purpose to verify the no-Abelian nature of MF, four MF is enough. They combine into two complex fermions c1 and c2 and the ground state has degeneracy four, but the Hamiltonian (1) conserves parity of the fermion number, and 2 the even-number subspace is decoupled from the odd-number subspace. At low temperature (significantly below the zero- mode gap), the initial state is typically a vacuum state 0i, and in the even-number subspace {0i0i,1i1i}, only two of the three generators are independent: −i 1 (cid:19) .(2) √2 (cid:18) 1 −i √2 (cid:18) 1 − i 1 + i (cid:19) , τ1 = τ3 = τ2 = 0 From these, two composite braiding operations [21] 1 0 1 1 1 τ2τ −1 A = τ1τ2τ −1 1 , B = τ −1 (3) can be constructed with the property AB = iσz, BA = −iσx, i.e., AB 6= BA, manifested the non-Abelian nature of MF, where τi and τ −1 denote cunter-clockwise and clockwise in- terchange of MF i and i+1, respectively. Specifically, with the initial state 0i0i, AB and BA yield two orthogonal output states 0i0i and 1i1i, respectively. We on purposely chose the two outputs to be orthogonal with each other so that the difference between the two can be detected unambiguously. i However, to implement braiding of MF in a 1D structure is impossible. Therefore, in order to exchange the MF for braid- ing operations, one can use a second wire to form a T-junction or more efficiently with the "railroad track" geometry [12]. The two red sections in Fig. 1 is tune into the topological phase while the blue sections are not. In this way, we induce two pairs of MF with the left pair (labeled as 1 and 2) resi- dents in the superconducting island of the flux qubit; the right pair of MF is labeled as 3 and 4. Another wire, perpendicular to the wire holds MF, is introduced to enable interchange of neighboring MF. In our setup, to demonstrate the non-Abelian nature of MF, i.e., to implement braiding operations in Eq. (3), we need to counterclockwise interchange of MF 1, 2 and 2, 3 for τ1 and τ2, respectively. Counterclockwise interchange of MF can be achieved along the line as proposed in Ref. [12]. For example, τ1 can be implemented as following: transports MF 1 downward by driving the vertical wire into a topologi- cal phase; transports MF 2 leftward in a similar fashion; and finally transports MF 1 up and to the right. Similarly, τ −1 can be implemented as following: transports MF 2 downward, transports MF 1 rightward and finally transports MF 2 up and to the left. 1 We now turn to detect the two orthogonal output states 0iL0iR and 1iL1iR . As the output states of the two pairs of MF are the same, detecting one of them can then fulfil the purpose of distinguishing the output states. The two states 0i and 1i are distinguished by the parity of the number np of particles in the island. Therefore, they can be distinguished by np. As 0iL0iR and 1iL1iR have the same parity, we need to remove one pair of MF out of the measurement circle, otherwise they will also contribute to the measurement results and make the detection impossible. Without loss of general- ity, we choose to detect the left pair of MF (MF 1 and 2) while move MF 3 and 4 along the wire out of the flux qubit circuit, as shown in Fig. 1, by local tunable gates. Note that the wire is not interrupted by the junctions providing that the junctions' thickness is much smaller than ξ. 2 1 (cid:177) (cid:169) / 0 -1 -2 -1 -0.5 0 0.5 1 1.5 2 (cid:158)/ (cid:158)0 3 mizing the tunnel splitting in the absence of vortices in the island. Meanwhile, the read-out is nondestructive, which is necessary for the realization of a two-qubit CNOT gate [18]. Moreover, the read out is insensitive to sub-gap excitations in the superconductor (they do not change the fermion parity). In summary, we have proposed a scheme in semiconductor- superconductor structure to demonstrate the non-Abelian statistics for MF in terms of braid group. The non-Abelian nature is manifested by the fact that the output of the differ- ent applied orders of two operations, constructed by the braid group elements, are different. Meanwhile, the different final states of MF can be unambiguously detected by a supercon- ducting flux qubit. FIG. 2: (Color online) Potential of the symmetric superconducting flux qubit with Φx = Φ0/2 and EJ = 0.507Φ2 0/(2L). The energy minima are connected by two tunneling paths indicated by the red arrows. This work was supported by the NSFC (No. 11004065), the NSF of Guangdong Province (No. 10451063101006312), and the Startup Foundation of SCNU (No. S53005). To measure the parity of np, we make use of the sup- pression of macroscopic quantum tunneling by the Aharonov- Casher effect [23]: a vortex encircling a superconducting is- land picks up a phase increment φ = πq/e determined by the total charge q coupled capacitively to the superconduc- tor. For our case, the two Josephson junctions have the same Josephson coupling energy EJ . The flux Φ in the qubit is related to the phase differences across the junctions, ϕ1 and ϕ2, by 2πΦ/Φ0 = ϕ1 + ϕ2; the phase θ = (ϕ1 − ϕ2)/2 is conjugate to the number of excess Cooper pairs of the su- perconducting island as [θ, n] = i. The potential energy is plotted in Fig. 2 for the external magnetic Φx = Φ0/2 and 0/(2L) with L being the inductance of the flux EJ = 0.507Φ2 qubit, which is strictly periodic in the θ direction. Therefore, neighboring minima are always separated by δθ = ±π (as in- dicated by red arrows in Fig. 2), i.e., the energy minima are connected by two tunneling paths with same amplitude, which amounts to the circulation of a Josephson vortex around the superconducting island. The interference produces an oscilla- tory tunnel splitting of the two levels of the flux qubit ∆ = ∆0 cos(cid:18) φ 2(cid:19) , (4) where ∆0 is the tunnel splitting associated with one path. Therefore, if q is an odd (even) multiple of the electron charge e, the two tunneling paths interfere destructively (construc- tive), so the tunnel splitting is minimum (maximal). As we only need to distinguish maximal from minimal tun- nel splitting, the flux qubit does not need to have a large qual- ity factor. In addition, ∆0 ≃ 100µeV ≃ 1K for parameters in typical experiments of flux qubits [23], which should be readily observable by microwave absorption. To make sure the total charge is solely comes from 1iL, one would first calibrate the charge on the gate capacitor to zero, by maxi- [1] F. Wilczek, Phys. Rev. Lett. 48, 1144 (1982). [2] G. Moore and N. Read, Nucl. Phys. B 360, 362 (1991). [3] N. Read and D. Green, Phys. Rev. B 61, 10267 (2000). [4] S. Das Sarma, M. Freedman, and C. Nayak, Phys. Rev. Lett. 94, 166802 (2005). [5] S. Das Sarma, C. Nayak, and S. Tewari, Phys. Rev. B 73, 220502(R) (2006). [6] L. Fu and C. L. Kane, Phys. Rev. Lett. 100, 096407 (2008). [7] J. D. Sau, R. M. Lutchyn, S. Tewari, and S. Das Sarma, Phys. Rev. Lett. 104, 040502 (2010). [8] J. Alicea, Phys. Rev. B. 81, 125318 (2010). [9] P. Bonderson, S. Das Sarma, M. Freedman, and C. Nayak, arXiv:1003.2856. [10] R. M. Lutchyn, J. D. Sau, and S. Das Sarma, Phys. Rev. Lett. 105, 077001 (2010). [11] Y. Oreg, G. Refael, and F. von Oppen, Phys. Rev. Lett. 105, 177002 (2010). [12] J. Alicea, Y. Oreg, G. Refael, F. von Oppen, and M. P. A. Fisher, Nature Phys. 7, 412 (2011). [13] A. Y. Kitaev, Physics-Uspekhi 44, 131 (2001). [14] L. Fu and C. L. Kane, Phys. Rev. Lett. 102, 216403 (2009). [15] A. R. Akhmerov, J. Nilsson, and C. W. J. Beenakker, Phys. Rev. Lett. 102, 216404 (2009). [16] J. D. Sau, S. Tewari, and S. Das Sarma, arXiv:1004.4702. [17] E. Grosfeld, B. Seradjeh, and S. Vishveshwara, Phys. Rev. B 83, 104513 (2011). [18] F. Hassler, A. R. Akhmerov, C.-Y. Hou, and C. W. J. Beenakker, New J. Phys. 12, 125002 (2010). [19] L. Jiang, C. L. Kane, and J. Preskill, Phys. Rev. Lett. 106, 130504 (2011). [20] P. Bonderson and R. M. Lutchyn, Phys. Rev. Lett. 106, 130505 (2011). [21] S.-L. Zhu, L.-B. Shao, Z. D. Wang, and L.-M. Duan, Phys. Rev. Lett. 106, 100404 (2011). [22] D. A. Ivanov, Phys. Rev. Lett. 86, 268 (2001). [23] J. R. Friedman and D. V. Averin, Phys. Rev. Lett. 88, 050403 (2002). 

1608.03068

1

1608

Creating arbitrary quantum vibrational states in a carbon nanotube

We theoretically study the creation of single- and multi-phonon Fock states and arbitrary superpositions of quantum phonon states in a nanomechanical carbon nanotube (CNT) resonator. In our model, a doubly clamped CNT resonator is initialized in the ground state and a single electron is trapped in a quantum dot which is formed by a electric gate potential and brought into the magnetic field of a micro-magnet. The preparation of arbitrary quantum phonon states is based on the coupling between the mechanical motion of the CNT and the electron spin which acts as a non-linearity. We assume that electrical driving pulses with different frequencies are applied on the system. The quantum information is transferred from the spin qubit to the mechanical motion by the spin-phonon coupling and the electron spin qubit can be reset by the single-electron spin resonance. We describe Wigner tomography which can be applied at the end to obtain the phase information of the prepared phonon states.

cond-mat.mes-hall

cond-mat

Creating arbitrary quantum vibrational states in a carbon nanotube Department of Physics, University of Konstanz, D-78457 Konstanz, Germany Heng Wang and Guido Burkard (Dated: August 7, 2021) We theoretically study the creation of single- and multi-phonon Fock states and arbitrary super- In positions of quantum phonon states in a nanomechanical carbon nanotube (CNT) resonator. our model, a doubly clamped CNT resonator is initialized in the ground state and a single electron is trapped in a quantum dot which is formed by a electric gate potential and brought into the magnetic field of a micro-magnet. The preparation of arbitrary quantum phonon states is based on the coupling between the mechanical motion of the CNT and the electron spin which acts as a non-linearity. We assume that electrical driving pulses with different frequencies are applied on the system. The quantum information is transferred from the spin qubit to the mechanical motion by the spin-phonon coupling and the electron spin qubit can be reset by the single-electron spin resonance. We describe Wigner tomography which can be applied at the end to obtain the phase information of the prepared phonon states. I. INTRODUCTION The peculiar feature of quantum states is that they can be in a superposition of their basis states. Preparation, manipulation and measurement of Fock states, which are quantum states with fixed numbers of quanta, and their superpositions are especially important for quan- tum computation with trapped ions1. The photon Fock states are widely used in quantum cryptography2,3. The harmonic phonon states of a single trapped ion have been used as the control qubit with the hyperfine ground state as the target qubit in an experimental realization of two-qubit controlled-NOT quantum gate4. The prepa- ration of Fock states and their arbitrary superpositions in linear resonators has been proposed by transferring the quantum information of a nonlinear quantum sys- tem which can be controlled by a classical source5 and has already been realized experimentally by coupling a trapped ion6 or a superconducting quantum circuit7,8, to a resonator. Recently, single phonon states were proposed as qubit states in optomechanical schemes9. Measurements of the quantum ground state and prepa- ration of a single phonon state of a piezoelectric res- onator coupled to a superconducting quantum bit have been achieved a few years ago10. Heralded single-phonon preparation is obtained by detecting the photon of the photon-phonon pair generated by optomechanical para- metric down-conversion11. The improvement of photon detection in the laboratory promises the precise single- photon counting allowing for single-phonon counting12. In nanomechanical or micromechanical systems, cool- ing the mechanical system13,14 to the ground state and preparing nonclassical states are required to operate me- chanical resonators in the quantum regime. Ground-state cooling of a mechanical system has been achieved with direct or active cooling in several laboratories10,15. Both electrical and mechanical properties of carbon nanotubes (CNTs) make them very interesting for quan- tum physics. Because of the additional valley degrees of freedom, semiconducting CNTs are promising candidates for valleytronics and valley-spin based technology16 -- 18. FIG. 1: Schematic view of a single electron being trapped in a quantum dot (QD) formed by gate voltages in a sus- pended carbon nanotube (CNT). The resonance frequency of the CNT can be adjusted by the voltages on the back gate. An external ac electric field is applied on the CNT by the an- tenna. The micro-magnet is deposited in the vicinity of the CNT. The single electron wavefuction can be electrostatically shifted by applying voltage on the back gates. Therefore the spin-splitting of the electron in the slanting magnetic field of the micro-magnet can be manipulated. The curvature induced spin-orbit coupling in CNTs has been predicted to be significant19 -- 21 and been observed in the laboratory22. A magnetic field leads to the lifting of the four-fold spin and valley degeneracy23 -- 28. On the other hand, suspended nanomechanical CNTs have high and widely tunable resonance frequencies and enormous quality factors29 -- 32, hence the vibrational modes of CNTs last long until they are totally damped out (Fig. 1). The two lowest energy levels of anharmonic nanomechanical CNT oscillators have been proposed as the two states of one qubit in quantum information processing33. The coupling of the electron spin and the mechanical motion of the CNT via the intrinsic spin-orbit coupling pro- vides a non-linearity34,35. Many theoretical proposals for the read-out of the vibrational frequency of a sus- pended CNT36 and the electron spin states37, for obtain- ing single- and two-qubit quantum gates38,39 and cool- ing a suspended CNT40 are based on this spin-phonon coupling. Recently a theoretical work has proposed the ground-state cooling of a suspended carbon nanotube (CNT) resonator between a normal and superconduct- 6 1 0 2 g u A 0 1 ] l l a h - s e m . t a m - d n o c [ 1 v 8 6 0 3 0 . 8 0 6 1 : v i X r a Back gateMicromagnetQuantum point contactBack gatesMicromagnetSingle electronQuantum point contactAntennaing lead by the interference of vibration-assisted Andreev reflections41. We present theoretically how to prepare the Fock states and arbitrary quantum phonon states based on the spin- phonon interaction in a suspended CNT. The basic work- ing principle is similar to the one used previously for superconducting qubits7,8 and consists of the following steps. Two-electron spin states split by a magnetic field are defined as our qubit. The qubit flip, the qubit- phonon swap, and the phase operations are applied al- ternately to obtain an arbitrary quantum vibration state of the CNT. The qubit is flipped from the ground state ↓(cid:105) into the excited state ↑(cid:105) by the electron spin reso- nance which is obtained in the presence of an external ac electric field matching the qubit frequency. The quan- tum dot is moved back and forth by the ac electric field hence the electron in the quantum dot experiences effec- tively a time-dependent magnetic field. A qubit-phonon swap converts the energy from the excited qubit state to the resonator from the ground phonon state by the spin-phonon coupling. The qubit is brought into reso- nance with the phonon to have an effective spin-phonon coupling strength by electrostatically moving the quan- tum dot in the stray field of a micro-magnet42 -- 44. A phase rotation of the spin can be applied to adjust the relative phase of the qubit. A sequence which alternates these three operations is applied until the desired quan- tum phonon state is obtained. This paper is organized as follows. The quantum me- chanical system and the effective Hamiltonian are intro- duced in Sec. II. The respective time-evolution opera- tors for the three necessary operations are presented in Sec. III. In Sec. IV, the steps to obtain Fock states and ar- bitrary quantum phonon states are explained. In Sec. V, we discuss the Wigner tomography for extracting the full information of the quantum phonon states. II. MODEL We assume that a single electron is trapped within a quantum dot (QD) formed in a suspended CNT which lies between two supports (Fig. 1). The QD is controlled by voltages on the electrodes at the ends of the CNT. The resonance frequency of the CNT ωp can be adjusted by the back gates. The strength and the frequency of the electric driving field are denoted as λ and ω. In CNTs, there exists a curvature induced spin-orbit interaction which already splits the degeneracy of spin in each val- ley without any magnetic field. With a magnetic field B applied along the CNT, the four-fold energy degener- acy of the valley and the spin is completely lifted. The electrons in the K and K(cid:48) valleys move in the directions of clockwise and anti-clockwise around the circumference of the CNT, respectively. Two spin states in the same K(cid:48) valley cross at the field B∗ = ∆SO/2µB where ∆SO is the spin-orbit interaction and µB is the spin magnetic moment. Since these two spin states are well separated 2 in momentum space from the states in the other valley, we choose them as the qubit. The energy splitting of the qubit is ωq = geµB(B−B∗) where B is the applied mag- netic field, ωq denotes the qubit frequency and ge is the electron g-factor37. The micro-magnet, which produces a slanting magnetic field can be deposited near the CNT such that the QD is located in the field. One can electro- statically move the QD and hence adjust the qubit fre- quency. The frequency difference of the energy between the phonon and the qubit is denoted as ∆ = ωp − ωq. An external ac driving electric field is applied to the system for the spin flip operation. It is the spin-phonon interaction that converts the ex- citation of the qubit into quantum vibrational motion. The spin-phonon interaction applies with both of the de- flection and the deformation phonon modes. In the fol- lowing, we only consider a single polarization of the de- flection mode of the CNT. It is possible to make general- izations to other deflection modes and to the deformation modes. We assume that the resting CNT axis is along z axis. The vibration of the CNT causes local changes in the direction of the CNT axis, and hence the tan- gent vector t(z) is dependent on the displacement u(z) of the CNT. The interaction of the spin and the deflec- tion phonon mode is induced by the spin-orbit interaction HSO = ∆SOσ · t (cid:39) ∆SOσz + ∆SO(dux(z)/dz)σx. Here (cid:126)σ = (σx, σy, σz) is the vector of Pauli matrices, ux(z) is assumed as the displacement at point z along the CNT in the x direction and ux(z) ∝ f (z) l0√ (a + a†) as a function of phonon creation and annihilation operators a† and a, where f (z) is the waveform of the QD and l0 is the zero- √ point amplitude of the phonon mode35. The spin-phonon interaction strength of the QD is g = ∆SO (cid:104)f(cid:48)(z)(cid:105) l0/2 2. The Hamiltonian for this system is 2 H = H0 + Hd + Hsp, ωq 2 σz + ωpa†a, H0 = Hd = 2λ(a + a†) cos(ωt), Hsp = g(a + a†)(σ+ + σ−), (1) where σ+ and σ− are qubit raising and lowering oper- ators, respectively. Here, H0 is the undisturbed Hamil- tonian of the phonon mode and the electron spin qubit. Hd contains the external ac electric driving term where λ is the driving strength and ω is the driving frequency, and the third part Hsp denotes the spin-phonon coupling which is induced from the spin-orbit coupling. We assume that the detuning fulfills ∆ (cid:29) g, λ. By applying a Schrieffer-Wolff transformation, an effective Hamiltonian from Eq. (1) is obtained in the interaction picture with respect to H0 38: I = −ασx + βnσz, H(cid:48) (2) III. TIME-EVOLUTION OPERATORS 3 Since we have the effective Hamiltonian for the qubit flip and the phase operations in Eq. (2), and for the spin-phonon swap in Eq. (4), we can derive their time- evolution operators with the aim of calculating the se- quence of pulses for obtaining arbitrary quantum phonon states. The interaction Hamiltonian in Eq. (2) can be writ- I = b · σ. The time-evolution operator of the ten as H(cid:48) ESR for the qubit flip operation, which is obtained by e−ib·σt = cos(bt)1 − i sin(bt)(b · σ), with the Hamilto- nian in Eq. (3) in the basis {gn(cid:105) ,en(cid:105)} with the phonon number n, is found to be I t/ Rn = e−i H(cid:48) (cid:18) cos(ϑt) + i βn where ϑ =(cid:112)α2 + β2 i α ϑ sin(ϑt) n. = ϑ sin(ϑt) i α ϑ sin(ϑt) cos(ϑt) − i βn ϑ sin(ϑt) (cid:19) (5) , We can obtain the time-evolution operator of the phase gate with α = 0 and ω = 0 in Eq. (3) and the electron spin rotates about the z axis in the magnetic field. Hence we obtain the phase operation, (cid:18) eiβnt 0 (cid:19) Pn = e−i H(cid:48) I t/ = 0 e−iβnt , (6) = where the coefficients βn are different for the phase op- erators with different phonon numbers. −i I t/ √ −i ηn n + 1 sin(gηnt) cos(gηnt) + i∆ sin(gηnt) The time-evolution operator for the qubit-phonon swap with the Hamiltonian in the Eq. (4) in the basis of {n ↑(cid:105) ,n + 1 ↓(cid:105)} with n = 0, 1 . . . is Un = e−i H(cid:48)(cid:48) √ (cid:33) (cid:32) where ηn = (cid:112)n + 1 + ∆2/g2. The swap between the (cid:19) cos(gηnt) + i∆ sin(gηnt) qubit and the phonon can be achieved best when they are on resonance. For the resonant case ∆ = 0, we have a simple time-evolution operator of the qubit-phonon swap Un = e−i H(cid:48)(cid:48) √ √ n + 1) −i sin(gt √ √ −i sin(gt cos(gt n + 1) (cid:18) cos(gt n + 1 sin(gηnt) n + 1) n + 1) I t/ , (7) ηn ηn = ηn (8) It is worth pointing out that the qubit flip and the qubit-phonon swap both depend on the phonon numbers. States with different phonon numbers have different co- efficients hence require different times for the same swap or flip operations. For example, the swap operations of ↓ 1(cid:105) → ↑ 0(cid:105) and ↓ 2(cid:105) → ↑ 1(cid:105) require different times because the phonon numbers are different. This leads to dephasing in the electronic sector in the state preparation protocol. The dephasing can be canceled in the process of preparation by applying uncompleted swap and flip operations together with phase operations. . FIG. 2: The energy-level diagram of the spin-phonon states. (a) The electron spin resonance between ↓ 0(cid:105) and ↑ 0(cid:105) (Blue dashed). The qubit is detuned from the phonon. The param- eter α is the effective strength of the spin operator σx in the effective Hamiltonian (2). (b) The qubit is brought into res- onance with the phonon in the slanting magnetic field of the micro-magnet by electrostatically moving the QD. The spin- phonon interaction strength is g (red solid). Phase operation σz with the strength β is applied to adjust the relative phase of the state (green dotted). where α = λgωp(ω2 − 2ω2 (ω2 − ω2 p)(ω2 ωq − 1 1 2 2 p + ω2 q ) p − ω2 q ) 2(2n + 1)g2 p − ω2 ω2 ωq , q βn = (3) − 1 2 ω. The eigenstates after the Schrieffer-Wolff transformation are slightly different from the original states because the higher order terms in the approximation are omitted. In the effective Hamiltonian, the term σx (σz) denotes a rotation of the spin about the x (z) axis of the qubit. We can obtain one of these two spin rotations separately by setting the coefficient of the other rotation to zero. For example, to obtain a rotation about the axis x, we set βn to zero as shown in Fig. 2. The rotations about the x axis can be used for obtaining electron spin resonance (ESR) and flipping the qubit states in the preparation of the arbitrary quantum phonon states. The rotations about the z axis can be used as a phase operation. Here, n denotes the phonon number. The spin-phonon interaction is used to exchange the information between the qubit and the phonon where the driving field is off. The Hamiltonian without the external driving Hd in the rotating wave approximation in the interaction picture with respect to H0 is I = −∆σz + g(aσ+ + a†σ−). H(cid:48)(cid:48) (4) If a large detuning ∆ is present, the effective coupling be- tween the spin and the phonon is too small to convert the energy from the qubit to the resonator. To obtain a per- fect swap of the qubit and the phonon, one can tune the frequency of the qubit to be in resonance with the phonon as shown in Fig. 2. Together with the slanting magnetic field of a micro-magnet, electrostatically tuning the elec- tron wave function of the QD serves this purpose42 -- 44. 4 ↑ n(cid:105) completely in each qubit flip operation. The qubit- phonon swap transfers the energy completely from the from ↑ n(cid:105) to excited spin state to the resonator, i.e. (cid:80) ↓ n + 1(cid:105). ψ(cid:105) = To obtain the arbitrary phonon state n cn ↓ n(cid:105), a sequence of operations with n steps is ap- plied on the initial state ↓ 0(cid:105) as ψ(cid:105) =U (τn)P (lnR)R(rn) . . . . . . P (l1U )U (τ1))P (l1R)R(r1)↓ 0(cid:105) . (9) (10) A sequence of one qubit flip R, one qubit-phonon swap operation U, and two phase rotation operations P is ap- plied in each step, except only one phase rotation is ap- plied in the last (n-th) step. The sequence is calculated backwards from the target state to the ground state ↓ 0(cid:105). Each step decreases the highest phonon number by 1. We can apply one step of the sequence of operations for ob- taining the state ↓(cid:105) (0(cid:105) + 1(cid:105)) as ↓(cid:105)(0(cid:105) + 1(cid:105)) = U (τ1)P (l1R)R(r1)↓ 0(cid:105) . (11) Here we apply a complete swap U (τ1) and an uncom- pleted qubit flip R(r1). The phase operator P (l1R) is applied to regulate the relative phase of the state. The state ↓(cid:105) (0(cid:105) + 1(cid:105)) will transfer into (↑(cid:105) + ↓(cid:105))0(cid:105) under the complete swap operation. The spins (↑(cid:105) + ↓(cid:105))0(cid:105) are flipped in the uncompleted qubit flip operation R(r1) so that only ↓ 0(cid:105) is left, while a complete qubit flip would lead to an unwanted state ↑ 0(cid:105) which causes de- phasing. We assume the parameter βn = β = 0 in the qubit flip operator R. The analytical expressions 4 +2πC1 of the operation times are obtained as r1 = , l1 = − 2i(ω2 , where Ci=1,2,3 are non-negative integers. To explain how to apply the sequence of operations, we consider an example of obtaining the state ↓(cid:105) (0(cid:105)+i2(cid:105)). As shown in Fig. 3, the sequence is calculated in the time reversed order from ↓(cid:105) (0(cid:105) + i2(cid:105)) to ↓ 0(cid:105) and we obtain (12) ↓(cid:105)(0(cid:105) + i2(cid:105)) = and τ1 = π/2+2πC3 p+ω2 q ) p−ω2 ωq(2g2−ω2 q )(iπ+2iπC2) U (τ2)P (l2R)R(r2)P (l1U )U (τ1))P (l1R)R(r1)↓ 0(cid:105) . − 3π α g (13) Fig. 3 (a) shows the frequencies of the phonon ωp, the qubit ωq and the driving ω as a function of the time. We can see that the qubit frequency is brought into resonance with the phonon frequency during the qubit- phonon swap. In the qubit flip operation, the driving is applied and a large detuning ∆ of the qubit frequency and the phonon frequency is required. Completed oper- ations of the qubit-phonon swap and the qubit flip are applied in order to decrease the highest phonon number by one in step No. 2. In qubit flip operation R(r2), the state ↑ 1(cid:105) flips completely to ↓ 1(cid:105). However, due to the spin-phonon coupling strength and the spin flip strength both depend on phonon numbers, dephasing of the states with lower phonon numbers appear in the process. Here FIG. 3: (a) Sequence of operations for obtaining arbitrary quantum phonon states. An external ac electrical field with the frequency ω (blue dashed) is applied for obtaining the qubit flip in the time intervals r1 and r2, and ωp is the fre- quency of the phonon mode (red dotted) in the CNT. The large detuning ∆ is required for the qubit flip operation R. For the qubit-phonon swap U (τ ), the qubit with the frequency ωq (black solid) is brought into resonance with the phonon that ωp = ωq by moving the QD in the slanting magnetic field of the nearby micro-magnet. The swap operation's time inter- vals are τ1 and τ2. The phase operations P (l1U ), P (l1R) and P (l2R) adjust the relative phase of the state. (b) Diagram for calculating the operation sequence in a backwards direction for obtaining the superposition of Fock states ↓(cid:105) (0(cid:105) + i2(cid:105)) from the ground state ↓ 0(cid:105). The operation U (τ2) is applied to fully transfer the state ↓ 2(cid:105) to the state ↑ 1(cid:105) and the flip operation R(r2) is applied to fully transfer the state ↑ 1(cid:105) to ↓ 1(cid:105) in step No. 2. In step No. 1, phase operations are ap- plied to adjust the relative phases to cancel some states (red crosses), e.g. the states ↓ 1(cid:105) and ↑ 0(cid:105) in the following qubit flip or qubit-phonon swap. IV. ARBITRARY QUANTUM PHONON STATES Now we explain the operation sequence of obtaining ar- bitrary phonon Fock states and superpositions of phonon Fock states. To obtain a phonon Fock state ψn(cid:105) = ↓ n(cid:105) from the ground state ↓ 0(cid:105), a sequence of operations with the qubit-phonon swaps and qubit flips for n steps is ap- plied as ψn(cid:105) = U (τn)R(rn)...U (τ1)R(r1)↓ 0(cid:105) g n and r = π/2 √ α are the timescales of oper- where τ = π/2 ations. Here we assume βn = 0 in qubit flip operators R. Each step contains a qubit-phonon swap and a qubit flip and the highest phonon number increases by 1 after each step. By applying the ac electrical field in the pres- ence of the large detuning, the qubit flips from ↓ n(cid:105) to Step 1Step 2since the qubit flip depends on phonon numbers, the state ↓ 0(cid:105) could not fully flip to the state ↑ 0(cid:105) in time r2 there- fore causes the dephasing in the electronic sector. The remaining state ↑ 0(cid:105) could be swapped to the state ↓ 1(cid:105) in the next qubit-phonon swap operation, which would be with the highest phonon number, therefore we want to cancel ↑ 0(cid:105) to avoid this. To cancel these dephasing we apply phase operations to adjust the relative phase of the state and perform uncompleted qubit flips and qubit- phonon swaps. The phase rotations are necessary when the next swap or flip operations are not applied com- pletely. When a spin up state and the Fock state with the highest phonon number need to be canceled, we ap- ply the phase operation and an uncompleted qubit flip. To cancel a spin down state with the highest Fock state, we apply a phase operation and a qubit-phonon swap operation. Therefore in step No. 1 the phase operator P (l1u) is applied to adjust the relative phase of the state and the qubit-phonon swap U (τ1) is applied partially to cancel ↓ 1(cid:105). Hence we have only ↑ 0(cid:105) and the leftover state ↓ 0(cid:105) and the rest of the operation is similar with the relevant part in the preparation of ↓(cid:105) (0(cid:105) +1(cid:105)). For obtaining other superpositions of Fock states with larger highest phonon numbers n or with more than two Fock states, one repeats the second step n times. V. WIGNER TOMOGRAPHY One can use Wigner tomography45 -- 49 to obtain the relative phase of the quantum phonon states which has been used for quantum photon states7,8. Wigner tomog- raphy is based on representing the Wigner function as a quasiprobability distribution on the complex phase space. The Wigner function can be written as the expectation value of the operator D†(−α)ΠD(−α)8, 2 π W (α) = (cid:104)ψD†(−α)ΠD(−α)ψ(cid:105) . (14) To obtain D(−α), the resonator is driven with an ac elec- tric field pulse as −α = (1/2)(cid:82) λ(t)dt, where α is the phase space amplitude of the resonator and D is the dis- placement operator D(−α) = D†(α) = exp(α∗a − αa†). FIG. 4: The Wigner tomography of the quantum phonon states 0(cid:105) + 2(cid:105), 0(cid:105) − i2(cid:105) and 0(cid:105) + i2(cid:105). The change of the relative phase of a two-state superposition of Fock states ro- tates the Wigner function W (α). 5 Γ. The fidelity F =(cid:112)(cid:104)ψ ρψ(cid:105) shows how close the obtained The fidelity F of obtaining ψ(cid:105) = 0(cid:105) + i2(cid:105) at FIG. 5: temperature T = 10 mK as a function of the damping rate state ρ is to the target state ψ(cid:105). The Wigner tomography of the obtained states at T = 10 mK with (a) damping Γ = 0, (b) Γ = 104 s−1 and (c) Γ = 3 × 104 s−1. The fidelity for the state obtained at (a) is F = 0.999, for (b) is F = 0.945 and for (c) is F = 0.859. The other parameters are λ/2π = 0.8 MHz, ∆ = 100 MHz, ωp/2π = 1.5 GHz, and g/2π = 0.56 MHz. For the parity operator Π, Fock states have eigenvalues 1 and −1 for even and odd phonon numbers, respectively. For mixed states, Eq. (14) can be written as a trace W (α) = (cid:88) Tr(D(−α)ρD(α)Π) nn(−α), (−1)nρ(cid:48) 2 π 2 π (15) n = where ρ is the density matrix ρ =(cid:80) i Pi ψi(cid:105)(cid:104)ψi of the resonator before being displaced8. For the displaced res- onator, the density matrix is ρ(cid:48) = D(−α)ρD(α). To cal- culate the Wigner function, we need to obtain the phonon 50. In principle, the numbers ρnn from the probability Pn phonon number n can be measured directly with charge detector37, but for small numbers of phonons in the CNT the accuracy is limited. After the displacement pulse, one brings the qubit on resonance with the resonator for a variable time and then performs the read-out of the qubit. The qubit can be read out by the mechanical re- sponse of the resonator to the pulsed external driving37. States with different spin states react to the external driving differently such that the excited spin states can be driven to other states with larger phonon numbers. Therefore one can tell apart the spin states by measur- ing the amplitude of the resonator via a charge detector. From the probability Pu(t) of finding the qubit in state ↑(cid:105), we can obtain the measured probability for being into Fock state n(cid:105) as Pn = cn27,8. The Wigner function ro- tates with the changes of the relative phase of a two-state superposition of Fock states as shown in Fig. 4. For su- perpositions of more than two Fock states, the shapes of Wigner functions change. -1.0-0.500.51.0-1.0-0.500.51.001234567890.70.80.90.91Γ[104s-1]F(cid:88) We can simulate a full set of measurements with proba- bility Pn for having Fock state n(cid:105) via the density matrix ρ from the set of the linear equations8 nn(α) = (cid:104)n D(−α)ρD(α)n(cid:105) = ρ(cid:48) Mnjiρji, (16) where the matrix M has the form j,i Mnji = (cid:104)j D(α)n(cid:105)∗ (cid:104)i D(α)n(cid:105) . (17) min{u,v}(cid:88) The displacement operator can be expanded in the basis of Fock states as √ u!v! (cid:104)u D(α)v(cid:105) = e−α2/2 αu−k(−α∗)(v−k) k!(u − k)!(v − k)! (18) Therefore we can obtain the Wigner function from the density matrix ρ, which is used in the following simula- tion. k=0 . VI. NONUNITRARY EVOLUTION We use a master equation for the non-unitary evolution taking the damping of the CNT and the thermal bath into account. The spontaneous qubit relaxation rate is neglected due to the small density of other phonon modes which have similar frequencies in the CNT and in the surroundings such as the substrate and the supports. The master equation for the density matrix ρ is of the form (cid:18) (cid:19) ρ = − i  [H, ρ] + (nB + 1)Γ (cid:18) {a†a, ρ} aρa† − 1 2 (cid:19) (19) + nBΓ a†ρa − 1 2 {aa†, ρ} , ωp/kBT−1) is the Bose-Einstein occupa- where nB = 1/(e tion factor and Γ (cid:28) g is the damping rate of the CNT. CNTs with high factor Q = ωq/Γ ≈ 150000 have been found in laboratories51,52. We take the following param- (cid:80)∞ eters: Γ = 104 s−1, Q = 950000, and ωp/2π = 1.5 GHz. The phonons follow the Bose-Einstein statistics in the n=0 e−nωp/kBT n(cid:105)(cid:104)n ⊗ thermal equilibrium, ρ = 1 Z 1 J. I. Cirac and P. Zoller, Phys. Rev. Lett. 74, 4091 (1995), URL http://link.aps.org/doi/10.1103/PhysRevLett. 74.4091. 2 C. H. Bennett and G. Brassard, Proceedings of IEEE In- ternational Conference on Computers, Systems and Signal Processing, pp. 175 -- 179 (1984). 3 A. Beveratos, R. Brouri, T. Gacoin, A. Villing, J.- P. Poizat, and P. Grangier, Phys. Rev. Lett. 89, 187901 (2002), URL http://link.aps.org/doi/10.1103/ PhysRevLett.89.187901. 6 ψ(cid:105)(cid:104)ψ where Z = (cid:80)∞ the fidelity F =(cid:112)(cid:104)ψ ρψ(cid:105) decreases with the damping n=0 e−nωp/kBT is the partition function. We obtain the total phonon state by the par- tial trace over the spins ρph = Trsρ. We have simulated a procedure to produce the state ψ(cid:105) = ↓(cid:105) (0(cid:105) + i2(cid:105)) at finite temperature T = 10 mK. The Fig. 5 shows how rate at finite temperature. The fidelity for the state ob- tained at Γ = 0 is F1 = 0.999, and F2 = 0.945 with the damping rate Γ = 104 s−4, and the fidelity is found to be F3 = 0.859 with the damping rate Γ = 3 × 104 s−1. VII. CONCLUSION In conclusion, single Fock states and arbitrary super- positions of the Fock states can be obtained by sequences of qubit-phonon swaps, qubit flips, and phase operations. The exchange of the spin and the phonon is obtained by the spin-phonon interaction, which is based on the coupling of the phonon and the spin due to the intrin- sic spin-orbit interaction. To obtain a large spin-phonon coupling strength it requires the resonance of the spin and the phonon. The mechanically induced ESR, which is obtained by applying a external ac electric field, is used to flip the qubit in the presence of a large detuning of the qubit and the phonon. The frequency of the qubit can be adjusted by electrostatically moving the electron wave function in the CNT in the slanting magnetic field of a nearby micro-magnet. A phase operation is applied to change the relative phase of the state to cancel unwanted Fock states in the next qubit-phonon swap or the next qubit flip. Wigner tomography can be used to obtain the phase and the amplitude information of the states. Non-unitary evolution of the system is simulated with the master equation. Our proposal introduces a way of electrically creating arbitrary quantum phonon states by interacting the CNT resonator with the electron spin in CNT. The formation of maximally entangled quantum phonon states between two modes of a mechanical res- onator can be further studied by transferring the infor- mation from two coupled electron spins in two quantum dots to the resonator or coupling one spin to two different modes. 4 C. Monroe, D. M. Meekhof, B. E. King, W. M. Itano, and D. J. Wineland, Phys. Rev. Lett. 75, 4714 (1995), URL http://link.aps.org/doi/10.1103/PhysRevLett. 75.4714. 5 C. K. Law and J. H. Eberly, Phys. Rev. Lett. 76, 1055 (1996), URL http://link.aps.org/doi/10.1103/ PhysRevLett.76.1055. 6 D. M. Meekhof, C. Monroe, B. E. King, W. M. Itano, and D. J. Wineland, Phys. Rev. Lett. 76, 1796 (1996), URL http://link.aps.org/doi/10.1103/PhysRevLett. 76.1796. 7 M. Hofheinz, E. M. Weig, M. Ansmann, R. C. Bial- czak, E. Lucero, M. Neeley, A. D. O'Connell, H. Wang, J. M. Martinis, and A. N. Cleland, Nature 454, 310 (2008), ISSN 0028-0836, URL http://dx.doi.org/10. 1038/nature07136. 8 M. Hofheinz, H. Wang, M. Ansmann, R. C. Bial- czak, E. Lucero, M. Neeley, A. D. O'Connell, D. Sank, J. Wenner, J. M. Martinis, et al., Nature 459, 546 (2009), ISSN 0028-0836, URL http://dx.doi.org/10. 1038/nature08005. 9 K. Stannigel, P. Komar, S. J. M. Habraken, S. D. Bennett, M. D. Lukin, P. Zoller, and P. Rabl, Phys. Rev. Lett. 109, 013603 (2012), URL http://link.aps.org/doi/10.1103/ PhysRevLett.109.013603. 10 A. D. O'Connell, M. Hofheinz, M. Ansmann, R. C. Bial- czak, M. Lenander, E. Lucero, M. Neeley, D. Sank, H. Wang, M. Weides, 697 (2010), ISSN 0028-0836, URL http://dx.doi.org/10. 1038/nature08967. et al., Nature 464, 11 C. Galland, N. Sangouard, N. Piro, N. Gisin, and T. J. Kip- penberg, Phys. Rev. Lett. 112, 143602 (2014), URL http: //link.aps.org/doi/10.1103/PhysRevLett.112.143602. 12 J. D. Cohen, S. M. Meenehan, G. S. MacCabe, S. Grob- lacher, A. H. Safavi-Naeini, F. Marsili, M. D. Shaw, and O. Painter, Nature 520, 522 (2015), ISSN 0028-0836, URL http://dx.doi.org/10.1038/nature14349. 13 T. J. Kippenberg and K. J. Vahala, Science 321, 1172 (2008), URL http://www.sciencemag.org/content/321/ 5893/1172.abstract. 14 F. Marquardt, J. P. Chen, A. A. Clerk, and S. M. Girvin, Phys. Rev. Lett. 99, 093902 (2007), URL http://link. aps.org/doi/10.1103/PhysRevLett.99.093902. 15 J. D. Teufel, T. Donner, D. Li, J. W. Harlow, M. S. Allman, K. Cicak, A. J. Sirois, J. D. Whittaker, K. W. Lehnert, and R. W. Simmonds, Nature 475, 359 (2011), ISSN 0028- 0836, URL http://dx.doi.org/10.1038/nature10261. 16 N. Rohling, M. Russ, and G. Burkard, Phys. Rev. Lett. 113, 176801 (2014), URL http://link.aps.org/doi/10. 1103/PhysRevLett.113.176801. 17 G. Széchenyi and A. Pályi, Phys. Rev. B 88, 235414 (2013), URL http://link.aps.org/doi/10.1103/PhysRevB.88. 235414. 18 A. Pályi and G. Burkard, Phys. Rev. B 82, 155424 (2010), URL http://link.aps.org/doi/10.1103/PhysRevB.82. 155424. 19 T. Ando, Journal of the Physical Society of Japan 69, 1757 (2000), URL http://dx.doi.org/10.1143/JPSJ.69.1757. 20 W. Izumida, K. Sato, and R. Saito, Journal of the Physical Society of Japan 78, 074707 (2009), URL http://dx.doi. org/10.1143/JPSJ.78.074707. 21 J.-S. Jeong and H.-W. Lee, Phys. Rev. B 80, 075409 (2009), URL http://link.aps.org/doi/10.1103/ PhysRevB.80.075409. 22 F. Kuemmeth, S. Ilani, D. C. Ralph, and P. L. McEuen, Nature 452, 448 (2008), ISSN 0028-0836, URL http:// dx.doi.org/10.1038/nature06822. 23 M. V. Entin and L. I. Magarill, Phys. Rev. B 64, 085330 (2001), URL http://link.aps.org/doi/10.1103/ PhysRevB.64.085330. 24 L. Chico, M. P. López-Sancho, and M. C. Muñoz, Phys. Rev. Lett. 93, 176402 (2004), URL http://link.aps.org/ doi/10.1103/PhysRevLett.93.176402. 25 H. Min, J. E. Hill, N. A. Sinitsyn, B. R. Sahu, L. Kleinman, 7 and A. H. MacDonald, Phys. Rev. B 74, 165310 (2006), URL http://link.aps.org/doi/10.1103/PhysRevB.74. 165310. 26 D. Huertas-Hernando, F. Guinea, and A. Brataas, Phys. Rev. B 74, 155426 (2006), URL http://link.aps.org/ doi/10.1103/PhysRevB.74.155426. 27 L. Chico, M. P. López-Sancho, and M. C. Muñoz, Phys. Rev. B 79, 235423 (2009), URL http://link.aps.org/ doi/10.1103/PhysRevB.79.235423. 28 F. Kuemmeth, H. O. H. Churchill, P. K. Herring, and C. M. Marcus, Mater. Today 13, 18 (2010), URL http://www.sciencedirect.com/science/article/ pii/S1369702110700304. 29 V. Sazonova, Y. Yaish, H. Ustunel, D. Roundy, T. A. Arias, and P. L. McEuen, Nature 431, 284 (2004), ISSN 0028- 0836, URL http://dx.doi.org/10.1038/nature02905. 30 A. K. Hüttel, G. A. Steele, B. Witkamp, M. Poot, L. P. Kouwenhoven, and H. S. J. van der Zant, Nano Lett. 9, 2547 (2009), URL http://pubs.acs.org/doi/abs/10. 1021/nl900612h. 31 J. Chaste, M. Sledzinska, M. Zdrojek, J. Moser, and A. Bachtold, Appl. Phys. Lett. 99, 213502 (2011), URL http://link.aip.org/link/?APL/99/213502/1. 32 E. A. Laird, F. Pei, W. Tang, G. A. Steele, and L. P. Kouwenhoven, Nano Letters 12, 193 (2012), URL http: //dx.doi.org/10.1021/nl203279v. 33 S. Rips and M. J. Hartmann, Phys. Rev. Lett. 110, 120503 (2013), URL http://link.aps.org/doi/10.1103/ PhysRevLett.110.120503. 34 M. S. Rudner and E. I. Rashba, Phys. Rev. B 81, 125426 (2010), URL http://link.aps.org/doi/10.1103/ PhysRevB.81.125426. 35 A. Pályi, P. R. Struck, M. Rudner, K. Flensberg, and G. Burkard, Phys. Rev. Lett. 108, 206811 (2012), URL http://link.aps.org/doi/10.1103/PhysRevLett. 108.206811. 36 C. Ohm, C. Stampfer, J. Splettstoesser, and M. R. Wegewijs, Appl. Phys. Lett. 100, 143103 (2012), URL http://link.aip.org/link/?APL/100/143103/1. 37 P. R. Struck, H. Wang, and G. Burkard, Phys. Rev. B 89, 045404 (2014), URL http://link.aps.org/doi/10.1103/ PhysRevB.89.045404. 38 H. Wang and G. Burkard, Phys. Rev. B 90, 035415 (2014), URL http://link.aps.org/doi/10.1103/PhysRevB.90. 035415. 39 H. Wang and G. Burkard, Phys. Rev. B 92, 195432 (2015), URL http://link.aps.org/doi/10.1103/PhysRevB.92. 195432. 40 P. Stadler, W. Belzig, and G. Rastelli, Phys. Rev. Lett. 113, 047201 (2014), URL http://link.aps.org/doi/10. 1103/PhysRevLett.113.047201. 41 P. Stadler, W. Belzig, and G. Rastelli, arXiv:1511.04858 (2015). 42 Y. Tokura, W. G. van der Wiel, T. Obata, and S. Tarucha, Phys. Rev. Lett. 96, 047202 (2006), URL http://link. aps.org/doi/10.1103/PhysRevLett.96.047202. 43 M. Pioro-Ladrière, Y. Tokura, T. Obata, T. Kubo, and S. Tarucha, Applied Physics Letters 90, 024105 (2007), URL http://scitation.aip.org/content/aip/journal/ apl/90/2/10.1063/1.2430906. 44 M. Pioro-Ladriere, T. Obata, Y. Tokura, Y.-S. Shin, T. Kubo, K. Yoshida, T. Taniyama, and S. Tarucha, Nat Phys 4, 776 (2008), ISSN 1745-2473, URL http: //dx.doi.org/10.1038/nphys1053. 45 B. G. Englert, Journal of Physics A: Mathematical and General 22, 625 (1989), URL http://stacks.iop.org/ 0305-4470/22/i=6/a=015. 46 H. Moya-Cessa and P. L. Knight, Phys. Rev. A 48, 2479 (1993), URL http://link.aps.org/doi/10.1103/ PhysRevA.48.2479. 47 K. Banaszek and K. Wódkiewicz, Phys. Rev. Lett. 76, 4344 (1996), URL http://link.aps.org/doi/10.1103/ PhysRevLett.76.4344. 48 S. Haroche and J.-M. Raimond, Exploring the Quantum- Atoms, Cavities and Photons (Oxford Univ. Press, 2006). 49 A. Eichler, M. del Álamo Ruiz, J. A. Plaza, and A. Bach- told, Phys. Rev. Lett. 109, 025503 (2012), URL http: 8 //link.aps.org/doi/10.1103/PhysRevLett.109.025503. 50 P. Lougovski, E. Solano, Z. M. Zhang, H. Walther, H. Mack, and W. P. Schleich, Phys. Rev. Lett. 91, 010401 (2003), URL http://link.aps.org/doi/10.1103/ PhysRevLett.91.010401. 51 A. K. Hüttel, 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), URL http: //dx.doi.org/10.1002/pssb.201000175. 52 M. Cirio, G. K. Brennen, and J. Twamley, Phys. Rev. Lett. 109, 147206 (2012), URL http://link.aps.org/doi/10. 1103/PhysRevLett.109.147206. 

1211.0372

1

1211

Magnetic power inverter: AC voltage generation from DC magnetic fields

We propose a method that allows power conversion from DC magnetic fields to AC electric voltages using domain wall (DW) motion in ferromagnetic nanowires. The device concept relies on spinmotive force, voltage generation due to magnetization dynamics. Sinusoidal modulation of the nanowire width introduces a periodic potential for a DW the gradient of which exerts variable pressure on the traveling DW. This results in time variation of the DW precession frequency and the associated voltage. Using a one-dimensional model we show that the frequency and amplitude of the AC outputs can be tuned by the DC magnetic fields and wire-design.

cond-mat.mes-hall

cond-mat

Magnetic power inverter: AC voltage generation from DC magnetic fields Jun'ichi Ieda1, 2 and Sadamichi Maekawa1, 2 1)Advanced Science Research Center, Japan Atomic Energy Agency, Tokai 319-1195, Japan 2)CREST, Japan Science and Technology Agency, Tokyo 102-0075, Japan (Dated: 3 November 2018) We propose a method that allows power conversion from DC magnetic fields to AC electric voltages using domain wall (DW) motion in ferromagnetic nanowires. The device concept relies on spinmotive force, voltage generation due to magnetization dynamics. Sinusoidal modulation of the nanowire width introduces a periodic potential for a DW the gradient of which exerts variable pressure on the traveling DW. This results in time variation of the DW precession frequency and the associated voltage. Using a one-dimensional model we show that the frequency and amplitude of the AC outputs can be tuned by the DC magnetic fields and wire-design. Spinmotive force is one of the emerging concepts in spintronics1,2, which allows direct conversion of magnetic energy to electric voltage in magnetic nanostructures. The embodiment of this effect was considered in a fer- romagnetic nanowire containing a magnetic domain wall (DW)3. Application of a DC magnetic field induces DW motion along the nanowire. The exchange interaction mediates energy transfer between the local moment and the conduction electron spin. As a result, a DC elec- trical voltage develops across the DW where the source of the electric power is the Zeeman energy. The rate of conversion is given by P ¯hγ/e ∼ 102 µV/T where P is the spin polarization of the ferromagnetic material, ¯h is the Planck constant divided by 2π, γ is the gyromag- netic ratio, and e is the elementary charge. Thanks to advances in microfabrication technology, the prediction has been confirmed in controlled experiments using the lock-in technique4 and the time-domain measurement5. Striking features of the spinmotive force are listed as follows: 1) In contrast to the inductive voltage where the time variation of a magnetic flux is required, solely a DC magnetic field can generate an electric voltage (see, e.g., Fig.2a in Hai et al.6). 2) The conversion rate is represented by fundamental constants apart from the material-dependent P , offering high energy conversion efficiency and precise determination of important param- eter P by measuring the output voltages as a function of applied fields. 3) Applications using this effect can oper- ate as active devices with zero stand-by power7 and such a power-conversion ability between magnetic and electric systems might open up spin-based power electronics. Recently, in addition to the DC voltage generation from the field-induced DW motion, gyration of a mag- netic vortex core has been shown to generate AC voltages from applied AC magnetic fields (microwaves) where con- tribution of the standard inductive voltage is carefully separated from the spinmotive force signal8. Moreover, the AC to DC magnetic power converter (i.e., from AC magnetic fields to DC voltages) was demonstrated in a comb-shaped ferromagnetic thin film in which spatially selective ferromagnetic resonance is excited9. However, the inversion of the latter device, i.e., a magnetic power inverter, has not been established yet. In this paper, we propose a method of AC voltage gen- eration due to application of a DC magnetic field us- ing DW motion in a periodically modulated nanowire as shown in Fig. 1(a). Since a DW has a surface energy due to the exchange interaction and magnetic anisotropy the modulation of the wire width introduces a variation of the potential energy for the DW. The geometrical con- finement arising from this effect is commonly used for pinning DWs10,11. It has been suggested that DW mo- tion can be induced solely by the shape effect and the intrinsic magnetic energy of the DW can be exploited via the spinmotive force mechanism7,12. We consider a perpendicularly magnetized thin wire, in which the thickness of the wire is constant while the width is modulated as w(x) = ¯w [1 − 2r cos(2πx/d)] , (1) with the average width ¯w = (w1+w2)/2, modulation rate r = (w1 − w2)/ ¯w, and period d as indicated in Fig. 1(a). FIG. 1. (a) The schematic illustration of a periodically modu- lated nanowire with perpendicular magnetization containing a DW. The wire-width w(x) (w1 ≥ w ≥ w2) is modulated sinusoidally with the period d. (b) The shape-effect magnetic field, Hshape calculated by Eqs. (1) and (2) as a function of the DW position q. (c) The peak-hight of the shape-effect magnetic field Hmax as a function of the modulation period with r = 5, 10, 15% where typical parameters for a Co/Ni multilayer are used. 2 A DW is prepared in the nanowire and a DC magnetic field HDC is applied perpendicular to the plane (the z direction) to drive the DW along the wire (the x direc- tion). Then the voltage induced by the DW motion is measured between the ends of the wire. First, we analytically investigate influence of the width modulation on the spinmotive force using a collective- coordinate model in which the dynamics of a DW is char- acterized by two collective coordinates, the center posi- tion q and the tilt angle ψ of the DW plane relative to the easy-plane. Importantly, in addition to the applied field, a DW in the spatially nonuniform nanowire is sub- jected to the "shape-effect field" due to the variation of the surface energy7,13: Hshape = ∓ σ 2Ms (cid:0)1 + Q−1 sin2 ψ(cid:1)1/2 d dx ln w(x) , (2) (cid:12)(cid:12)(cid:12)(cid:12)x=q √ where σ = σ0 is the surface energy density of a DW and Ms is the saturation magnetiza- tion. Here σ0 = 4 AsKu with As the exchange stiffness constant, Ku the uniaxial anisotropy constant, and Q is the ratio of Ku to the hard-axis anisotropy constant. Here and hereafter the upper (lower) sign corresponds to the up-down (down-up) DW for the present coordinate- system. Figure 1(b) illustrates Hshape as a function of the DW position for the sinusoidally modulated width (1). The peak hight Hmax is given by √ d r 1 − 4r2 2πσ Ms Hmax = (3) . In Fig. 1(c), we show the shape dependence of Eq. (3). For HDC <∼ Hmax, a DW is trapped by the potential well and the total field acting on the DW becomes zero. This leads to the threshold field for the voltage generation. When a DW propagates along the sinusoidal wire the AC voltage with the amplitude (P ¯hγ/e)µ0Hmax is expected. In the uniform limit, r → 0 or d → ∞, the shape effect disappears. In the one-dimension model, the voltage due to DW dynamics is given by3 V = ± P ¯h e ψ. (4) Note that the growing side of the magnetic domain be- comes high voltage being independent of the wall type5. To evaluate this voltage, the DW dynamics has to be cal- culated. The time evolution of (q, ψ) is described by the reduced Landau-Lifshiz-Gilbert equations including the shape-field, α (HDC + Hshape) + sin 2ψ , (5) γµ0 ψ = HDC + Hshape − α where α is the Gilbert damping constant, ∆ =(cid:112)As/Ku HK 2 1 + α2 sin 2ψ (6) , is the wall width, HK is the hard-axis anisotropy field, (cid:20) q = ± ∆γµ0 1 + α2 (cid:18) (cid:21) HK 2 (cid:19) FIG. 2. The time evolution of the output voltage signal for a uniform wire (black) and a sinusoidal wire (red) with d = 150 nm and r = 5 %. The same DC magnetic field, µ0HDC = 0.1 T, is applied for both geometries. The inset is the enlarged view showing the oscillation due to the anisotropy field. and µ0 is the magnetic constant. In a uniform nanowire the time-average DW velocity ¯vDW is well described by ¯vDW = α∆γµ0HDC/(1 + α2) for HDC > HW where HW = αHK/2 is the Walker breakdown field. In the sinusoidal nanowire, the condition for the Walker breakdown is replaced by HDC + Hshape > HW. The one-dimensional description is more accurate for per- pendicularly magnetized nanowires with high magnetic anisotropy such as Co/Ni multilayers14 than for Permal- loy nanowires where the DW structure is deformed during the propagation15. It has been shown that the rigid DW motion supported by the high magnetic anisotropy is fa- vorable for stable generation of the spinmotive force16. In the following, we numerically solve the one- dimensional model (5) and (6) for the sinusoidal nanowire defined by (2) to clarify the AC output characteristics of the DW-induced voltages. Figure 2 shows the calculated voltage signals due to DW motion in the sinusoidal, as well as the uniform (ref- erence), nanowires. We employ typical values for a Co/Ni multilayer; γ = 1.76 × 1011 s−1·T−1, α = 0.02, P = 0.6, Ms = 0.85 T, As = 1.3×10−11 J/m, Ku = 4.0×105 J/m3, and µ0HK = 50 mT. We set the modulation parameters for the sinusoidal wire as d = 150 nm, r = 5 %, and the applied DC magnetic field µ0HDC = 0.1 T. In addition to the small-amplitude oscillation associated to the Walker breakdown as shown in the inset of Fig. 2, the AC compo- nent with the amplitude ∼ ±1.7 µV and the period ∼ 85 ns appears for the sinusoidal nanowire. The base line is the DC component (P ¯hγ/e)µ0HDC = 6.93 µV. The asymmetry found in positive and negative waveforms of the AC component is caused by the difference of the DW velocity, q, for the shape-field assisting and being against the DW propagation respectively. We discuss the shape dependence of the asymmetry later. The AC voltage amplitude agrees well with the pre- dicted value, (P ¯hγ/e)µ0Hmax = 1.72 µV, using Eq. (3) 3 FIG. 4. The wire-geometry dependence of the output voltage signals. (a) The period d of the nanowire is varied from 50 -- 200 nm with µ0HDC = 0.1 T and r = 5 % being fixed. (b) The wire-modulation ratio r is varied from 5 -- 15 % for constant µ0HDC = 0.1 T and d = 100 nm. (c) The angular frequency of the AC component ωAC as a function of d and r for µ0HDC = 0.15 T. The dotted line indicates the threshold for the DW deppining. of the DW velocity due to the shape-field becomes small. Next we investigate the geometry dependence of the AC component in Co/Ni wires. In Figs. 4(a) and 4(b), we show the time-domain voltage signals for µ0HDC = 0.1 T varying d (with r = 5 % fixed) and r (with d = 150 nm fixed) of sinusoidal nanowires respectively. While the AC amplitudes are proportional to 1/d and (cid:39) r as estimated by Eq. (3), the AC frequencies diminish with increasing both d and r. These features will be useful for engineering the device characteristics, AC amplitude and frequency, independently. Figure 4 (c) summaries the shape dependence of the AC frequency for µ0HDC = 0.15 T. For the area Hmax >∼ HDC, DWs are trapped and the voltage is not generated. Finally, we remark that the periodic potential for a DW can be realized by modulating not only the width but the thickness of a nanowire, or the radius of a mag- netic nanotube. In a cylindrical nanotubes, the hard-axis anisotropy is absent, leading to fascinating features for DW dynamics17, and thus, for the associated voltages. Different materials have different AC characteristics, e.g., the faster DW motion could achieve the GHz operation. This also merits further investigation. In summary, we have presented the concept of a mag- netic power inverter: power conversion from DC mag- netic field to AC electric voltage. The device consists FIG. 3. The applied magnetic field dependence of the output voltage signal frequency. The symbols represent the second maximum Fourier component of the representative voltage signals, shown in the inset, for µ0HDC = 0.01 -- 0.09 T as in- dicated in the legend. The sinusoidal wire parameters are d = 150 nm and r = 5 %. The dashed curve shows the ana- lytical formula (7). for the present geometry, which is basically independent of HDC as shown in the inset of Fig. 3. The period T is determined by T =(cid:82) d 0 dq/ q (cid:39) d/¯vDW. To see the HDC dependence of the AC component, we perform the Fourier analysis of the calculated voltage sig- nals. We focus on the angular frequency ωAC = 2π/T of the AC component induced by the shape effect. In the frequency domain, we identify the second highest peak as the shape-induced frequency as well as the highest DC component and the small peak of the anisotropy origin (the Walker breakdown). In Fig. 3, we plot ωAC for representative magnetic fields, µ0HDC = 10, 30, 50, 70, and 90 mT. The param- eters used here are the same as in Fig. 2. The general features of the frequency can be understood by the ap- proximate expression:  2πα∆γµ0 (1 + α2)d 0 ωAC (cid:39) HDC (HDC ≥ Hmax), (HDC < Hmax), (7) as indicated by the dashed line in Fig. 3. The deviation from Eq. (7) near the threshold value can be attributed to the nonlinear DW dynamics13 where the contribution from the anisotropy field plays a decisive role. It is in- teresting to investigate such nonlinear behaviors in more detail but we leave these problems for future work. In the inset of Fig. 3, we show the time-domain sig- nals for corresponding HDC. Note that the threshold value is µ0Hmax (cid:39) 25 mT for the present geometry. For µ0HDC = 10 mT, the voltage signal becomes zero as the DW is trapped by the geometrical potential. For µ0HDC = 30 mT, the asymmetry of the AC signal is pronounced. On the other hand, for the larger HDC the monochromaticity increases since the relative variation 4 of a periodically modulated ferromagnetic nanowire with a DW. We have investigated systematically the output voltage characteristics as a function of the input DC magnetic fields and the geometrical parameters of the nanowire using the one-dimensional model for the DW. We have shown that by tuning the magnetic field and the wire geometry the variable frequency ranging from MHz to GHz can be achieved. The proposed device operates as an active element in future spin-based power electronics, or power spintronics. We are grateful to Y. Yamane for fruitful discussions on this work. This research was supported by MEXT KAKENHI Grant Number 24740247. 1S. Maekawa ed., Concepts in Spin Electronics (Oxford University Press, Oxford, 2006). 2S. Maekawa, S. O. Valenzuela, E. Saitoh, and T. Kimura eds., Spin Current (Oxford University Press, Oxford, 2012). 3S. E. Barnes and S. Maekawa, Phys. Rev. Lett. 98, 246601 (2007). 4S. A. Yang G. S. D. Beach, C. Knutson, D. Xiao, Q. Niu, M. Tsoi, and J. L. Erskine, Phys. Rev. Lett. 102, 067201 (2009). 5M. Hayashi, J. Ieda, Y. Yamane, J. Ohe, Y. K. Takahashi, S. Mitani, and S. Maekawa, Phys. Rev. Lett. 108, 147202 (2012). 6P. N. Hai, S. Ohya, M. Tanaka, S. E. Barnes, and S. Maekawa, Nature 458, 489 (2009). 7S. E. Barnes, J. Ieda, and S. Maekawa, Appl. Phys. Lett. 89, 122507 (2006). 8K. Tanabe, D. Chiba, J. Ohe, S. Kasai, H. Kohno, S. E. Barnes, S. Maekawa, K. Kobayashi, and T. Ono, Nature Commun. 3, 845 (2012). 9Y. Yamane, K. Sasage, A. Toshu, K. Harii, J. Ohe, J. Ieda, S. E. Barnes, E. Saitoh, and S. Maekawa, Phys. Rev. Lett. 107, 236602 (2011). 10S. S. P. Parkin, M. Hayashi, and L. Thomas, Science 320, 190 (2008). 11M. Klaui, J. Phys.: Condens. Matter 20, 313001 (2008). 12Y. Yamane, J. Ieda, J. Ohe, S. E. Barnes, and S. Maekawa, Appl. Phys. Exp. 4, 093003 (2011). 13J. Ieda, H. Sugishita, and S. Maekawa, J. Magn. Magn. Mat. 322, 1363 (2010). 14T. Koyama, D. Chiba, K. Ueda, H. Tanigawa, S. Fukami, T. Suzuki, N. Ohshima, N. Ishiwata, Y. Nakatani, and T. Ono, Appl. Phys. Lett. 98, 192509 (2011). 15M. Hayashi, S. Kasai, and S. Mitani, Appl. Phys. Exp. 3, 113004 (2010). 16Y. Yamane, J. Ieda, and S. Maekawa, Appl. Phys. Lett. 100, 162401 (2012). 17M. Yan, C. Andreas, A. K´akay, F. Garc´ıa-S´anchez, and R. Hertel, Appl. Phys. Lett. 99, 12505 (2011). 

1604.05546

2

1604

Transport through a quantum spin Hall antidot as a spectroscopic probe of spin textures

We investigate electron transport through an antidot embedded in a narrow strip of two-dimensional topological insulator. We focus on the most generic and experimentally relevant case with broken axial spin symmetry. Spin-non-conservation allows additional scattering processes which change the transport properties profoundly. We start from an analytical model for noninteracting transport, which we also compare with a numerical tight-binding simulation. We then extend this model by including Coulomb repulsion on the antidot, and we study the transport in the Coulomb-blockade limit. We investigate sequential tunneling and cotunneling regimes, and we find that the current-voltage characteristic allows a spectroscopic measurement of the edge-state spin textures.

cond-mat.mes-hall

cond-mat

Transport through a quantum spin Hall antidot as a spectroscopic probe of spin textures Alexia Rod,1, 2 Giacomo Dolcetto,2 Stephan Rachel,1 and Thomas L. Schmidt2 1Institute for Theoretical Physics, Technische Universitat Dresden, 01062 Dresden, Germany 2Physics and Materials Science Research Unit, University of Luxembourg, L-1511 Luxembourg We investigate electron transport through an antidot embedded in a narrow strip of two-dimensional topolog- ical insulator. We focus on the most generic and experimentally relevant case with broken axial spin symmetry. Spin-non-conservation allows additional scattering processes which change the transport properties profoundly. We start from an analytical model for noninteracting transport, which we also compare with a numerical tight- binding simulation. We then extend this model by including Coulomb repulsion on the antidot, and we study the transport in the Coulomb-blockade limit. We investigate sequential tunneling and cotunneling regimes, and we find that the current-voltage characteristic allows a spectroscopic measurement of the edge-state spin textures. PACS numbers: 71.10.Pm,72.10.Fk,03.65.Vf I. INTRODUCTION Two-dimensional topological insulators (2D TIs) behave as band insulators in the bulk but host gapless 1D edge states.1,2 Experimentally, 2D TIs and their edge states have been in- vestigated mostly in HgTe/CdTe quantum wells, as well as in InAs/GaSb heterostructures3 -- 8 and evidence for the ex- pected ballistic edge transport and the quantum spin Hall (QSH) effect has been found. In contrast to ordinary one- dimensional spin- 1 2 electron systems, such as quantum wires, the edge channels of 2D TIs consist of a single pair of counter- propagating electronic modes.9 -- 11 Time-reversal symmetry then severely impedes backscattering in the edge states, ren- dering them robust to disorder and weak interactions. The simplest models for 2D TIs predict 1D edge channels in which electrons with opposite spins propagate in opposite directions.12 As helicity (i.e., the projection of the electron's spin operator on its momentum) is then conserved on a given edge, such systems are called helical 1D systems. However, while time-reversal symmetry is expected to be essential for the protection of gapless helical edge states, spin conservation is not. A plethora of effects, such as Rashba spin-orbit cou- pling, bulk inversion asymmetry, or structural inversion asym- metry, give rise to effective edge-state Hamiltonians without conserved spin.13 -- 16 In the presence of spin-symmetry breaking, left- and right- moving eigenstates can be almost arbitrary linear combina- tions of spin-up and spin-down electrons.17 Time-reversal symmetry merely ensures that counter-propagating eigen- states with the same energy have opposite spin orientations, but it makes no statement relating eigenstates with differ- ent energies. Hence, the most generic helical system can be thought of as a helical channel in which the spin quantiza- tion axis can rotate with momentum, which makes the spin texture of a 2D TI edge state nontrivial even in the presence of time-reversal symmetry.18,19 Recently, this spin texture was calculated for a number of realized and proposed 2D topolog- ical insulators based on their effective Hamiltonians.20 Such generic helical liquids are the most general 2D TI edge states, characterized by time-reversal but no additional symmetries. A nontrivial spin texture leads to interesting effects. Firstly, while zero-energy observables are insensitive to the spin tex- ture, scattering processes at finite energies are greatly affected by the existence of right-movers and left-movers with non- orthogonal spins. This gives rise, for instance, to increased backscattering and thus a deviation from the quantized edge channel conductance at finite temperatures.17,21,22 Another consequence of spin-non-conservation is the appearance of novel umklapp scattering processes that can gap out the spec- trum even in the presence of time-reversal symmetry.23 More- over, the spin texture can in principle be tuned locally by the application of a perpendicular electric field.15 In that case, coupling edge states with different spin textures has been shown to lead to new transport effects.20,24,25 In this article, we will investigate QSH antidots, i.e., non- topological regions (such as holes) embedded in a narrow strip of a 2D TI. In this case, the helical edge states propagating around the antidot can be tunnel-coupled to the helical sys- tems propagating along the sample edges. Such a setup has a long history in the context of quantum Hall systems.26 -- 29 When embedded in 2D TIs, an antidot can be a useful tool to generate spin-polarized currents,30 and thus to find evi- dence for the helicity of their edge states, as well as to explore nonlinear spin thermoelectric effects31,32, entanglement33 or Kondo physics.34 -- 36 The presence of multiple antidot-induced bound states was also shown to affect the transport proper- ties of helical edge states, by inducing quantum percolation in the QSH bar.37 Moreover, due to the potentially small sizes of the antidots, and the strong confinement of the electrons to 1D channels along their circumference, the Coulomb charg- ing energy may be large. This provides a promising plat- form for studying the interplay between spin-orbit coupling and electron-electron interactions. Depending on the TI ma- terial at hand, antidots can in principle be realized either by lithographical patterning of the sample or by appropriate elec- trical gating. In contrast to previous publications, the focus of this article will be on antidot transport in 2D TIs with a nontrivial edge- state spin structure. Our motivation is twofold: on the one hand, 2D TIs realized in InAs/GaSb or HgTe/CdTe systems are expected to have a nontrivial spin structure as a conse- quence of effects such as broken structural inversion asym- metry. Its effect should therefore be taken into account for a realistic modeling of antidot transport. On the other hand, it remains a challenge to directly measure the spin texture of 6 1 0 2 l u J 8 1 ] l l a h - s e m . t a m - d n o c [ 2 v 6 4 5 5 0 . 4 0 6 1 : v i X r a 2 (cid:88) (cid:90) l/2 upper (lower) edge and the antidot. Specifically, one has for the upper and lower edges, dx Ψ† Hs = −isvF α α=± −l/2 sα(x)∂xΨsα(x), (2) where s = U, L ≡ +,−. Moreover, vF is the Fermi velocity, and l is the edge length. We mostly consider the limit l → ∞, thus assuming the upper and lower edges to have a continu- ous spectrum, contrary to the antidot, whose energy levels are discrete with an energy separation ≈ vF/R, R being the radius of the antidot. In the presence of axial spin symmetry, the quantum number α corresponds to the spin polarization of the edge states, so that for instance spin-up electrons propagate to the right on the upper edge and to the left on the lower one (opposite for spin-down electrons). Our aim is to investigate the more general experimental sce- nario in which axial spin symmetry is broken: in this case, the electron operators Ψ† s±(x) still correspond to chiral particles moving to the right (left) on the upper edge and to the left (right) on the lower one, but they are no longer eigenstates of the spin operator. In momentum space it is possible to relate l for α = ± to the √ l for σ =↑,↓ via a uni- (cid:33) the chiral basis csαk = (cid:82) spin basis csσk = (cid:82) tary transformation17(cid:32)cs+,k dxe−ikxΨsσ(x)/ dxe−ikxΨsα(x)/ (cid:33) √ (3) The form of the momentum-dependent SU(2) matrix Bsk is dictated by time-reversal symmetry and unitarity cs−,k sk = B† (cid:32)cs↑k (cid:32)cos(θsk) − sin(θsk) (cid:33) cs↓k . sin(θsk) cos(θsk) Bsk = , (4) where the function θsk, which is even in k because of time-reversal symmetry, measures the rotation of the spin- quantization axis20 on edge s at momentum k. For realistic models and momenta near the Dirac point, it was shown20 that one can usually use the approximation θsk ≈ (ks/k0s)2, where the parameter k0s represents the momentum scale over which the spin-quantization axis rotates and thus incorporates the in- formation about the spin structure of the helical states. Note that we allow in principle the upper and lower edge to have different spin structures with parameters θUk and θLk. The antidot Hamiltonian Hd = −ivF dr Ψ† dα(r)∂rΨdα(r) + E(n) (5) (cid:90) 2πR α 0 (cid:88) α=± is also characterized by a linear dispersion. However, due to its confinement, the charging energy contribution should be taken into account, (cid:33)2 (cid:32) plied to the island, and n = (cid:80) E(n) = Ec 2 n − eVg Ec (cid:82) 2πR where Ec is the Coulomb energy, Vg is the gate voltage ap- dα(r)Ψdα(r) is the dr Ψ† α 0 , (6) FIG. 1. Sketch of the setup. The light gray area is the TI embedding an antidot of radius R. The dark gray areas are the leads. The tun- neling processes (black dashed line) occur at x = 0, r = 0 and x = 0, r = πR. The chirality of the edge states is indicated by the colors, '+' is red, '−' is green. edge states. We will show that using antidot geometries in the Coulomb-blockade regime, a spectroscopic measurement of the edge-state spin texture is possible by means of standard transport measurements. The structure of this article is as follows: In Sec. II, we will introduce the general model for an antidot embedded in a topological insulator without axial spin symmetry and present the low-energy Hamiltonian describing transport in the sys- tem. In Sec. III, we will study transport in the absence of interactions on the antidot. In particular, we will present nu- merical results that allow us to fix the parameters of the analyt- ical model. In Sec. IV, we will take into account the charging energy of the antidot, and we will present transport calcula- tions in the sequential tunneling and cotunneling regimes. We present our conclusions in Sec. V. II. MODEL We consider an antidot geometry realized in a 2D TI, as schematically shown in Fig. 1. This setup can be realized ei- ther by lithographically etching the sample or, in the case of an InAs/GaSb heterostructure, by gating the central portion of the bulk and thus bringing it to the trivial insulator regime. In both cases, a pair of helical edge states appear around the antidot (d), in addition to the edge states present at the upper (U) and lower (L) edges of the QSH bar. If the Fermi energy is tuned to lie within the bulk energy gap and the temperature is much lower than the gap itself, transport only occurs via the edge states, whereas the 2D bulk states are fully insulat- ing. The overlap between the edge-state wave functions gives rise to a finite tunneling probability between the edges and the antidot. The total Hamiltonian in the presence of tunneling is H = HU + HL + Hd + HdU + HdL, (1) where HU(L) is the free Dirac Hamiltonian of the upper (lower) edge, Hd is the Hamiltonian of the edge states around the an- tidot, and Hd,U(L) is the tunneling Hamiltonian between the dre−i jr/RΨ† number operator. As for the edges, spin in general is not a good quantum number, so the operator Ψ† d±(r) refers to elec- trons propagating clockwise/anticlockwise but without a well- defined spin polarization. Following what was done for the translationally invariant edges, we can define the most general SU(2) transformation in angular momentum space. It relates 2πR for α = ± √ 2πR for dσ(r)/ the chiral states dα j = (cid:82) to the spin-polarized ones dσ j =(cid:82) (cid:32)d+, j (cid:32)d↑ j (cid:33) (cid:33) (cid:33) (cid:32)cos(θ j) − sin(θ j) (7) (8) . √ dα(r)/ dre−i jr/RΨ† σ =↑,↓ as It is given by = B† d−, j d↓ j B j = . j sin(θ j) cos(θ j) σσ(cid:48) Tunneling to and from the antidot occurs near the coordi- nates x = 0 and r = 0 =: rU for the upper contact and at x = 0 and r = πR =: rL for the lower one (see Fig. 1). We start with the most general tunneling Hamiltonian containing both spin-preserving and spin-flipping terms,38 Hds = σσ(cid:48)(x, r)Ψdσ(cid:48)(rs + sr) + H.c. Ψ† sσ(x)γs (cid:88) (cid:90) dxdr (cid:104) (cid:105) (9) where s = U, L ≡ +,− and σ, σ(cid:48) ∈ {↑,↓}. To limit the number of parameters, we assume the sample geometry to be symmet- ric about the x axis, see Fig. 1. Reflection symmetry about the x axis is defined as (x, y) → (x,−y) and (px, py) → (px,−py). This entails the transformation rule σz → −σz for the spin quantum number. As a consequence, the field operators trans- form as ΨUσ(x) → ΨL ¯σ(x) and Ψdσ(r) → Ψd ¯σ(πR− r). Invari- ance of HdU +HdL under this transformation leads to the four equations γU ¯σ ¯σ(cid:48), which allow us to eliminate γL, and it leaves only the four functions γσσ(cid:48) ≡ γU σσ(cid:48). In addition, we assume the tunnel Hamiltonian to respect time-reversal sym- metry, which is local in space and acts on the edge states as Ψsσ(x) → σΨs ¯σ(x) for s ∈ {U, L, d} and σ = ↑,↓ = +,−. The tunnel Hamiltonian has time-reversal symmetry if γ↑↑ = γ↓↓ and γ↑↓ = −γ↓↑. This leaves us with two functions γsc ≡ γ↑↑ and γsf = γ↑↓ denoting the amplitudes of spin-conserving and spin-flip tunneling, respectively.39,40 The Hamiltonian now reads σσ(cid:48) = γL (cid:90) (cid:88) dxdr(cid:2)Ψ† sσ(x)γsf(x, r)Ψd ¯σ(rs + sr) + H.c.(cid:3). sσ(x)γsc(x, r)Ψdσ(rs + sr) Hds = +sσΨ† σ (10) We would like to point out that it is important to fix the form of the tunneling Hamiltonian by reflection symmetry and not inversion symmetry, the latter being defined as (x, y) → (−x,−y). Indeed, as we will show further below for the Kane- Mele model,9,13 in the presence of Rashba spin-orbit coupling, the bulk system remains invariant under reflection, whereas inversion symmetry is usually lost. Next, we express the tunneling Hamiltonian in the basis of the chiral edge states. For this purpose, we Fourier-transform to momentum and angular momentum space and use the ro- tation matrices (4) and (8). Expressed in terms of the Fourier components of the tunneling amplitudes, we find Hds = (cid:88) (cid:88) (cid:88) (cid:2) 1√ 2πRl αα(cid:48) ei jrs/R γsc(k, s j)(B† j c† s,k)ασ Bσα(cid:48) + ei jrs/R γsf(k, s j)(B† s,k)ασ B ¯σα(cid:48) k, j σ j sαkdα(cid:48) j + H.c.(cid:3). sαkdα(cid:48) j sσc† 3 (11) We can further simplify this by assuming that the Fourier com- ponents of the tunneling amplitudes γsc,sf(k, j) as well as the rotation matrices vary slowly as functions of k and j. The for- mer is justified if the tunneling happens locally on the scale of the Fermi wavelength. The latter assumption holds if temper- ature and applied bias voltage are small compared to vFk0{U,L}. In this case, we can replace these functions by their values at the Fermi energy and define γT cos(θT ) = γsc(kF, jF) γT sin(θT ) = γsf(kF, jF) B = B jF Bs = Bs,kF for s ∈ {U, L} (12) where kF = µ/vF and jF = µR/vF are determined by the chemical potential µ. The angle θT set the ratio between the tunneling amplitudes for spin-conserving and spin-flip tun- neling, θT = tan−1(γsf/γsc). Then, we obtain by Fourier- transforming back to real space (cid:88) Hds = γT (cid:88) s)ασ(cid:16) αα(cid:48) (B† φsαα(cid:48) = Ψ† sα(0)φsαα(cid:48) Ψdα(cid:48)(rs) + h.c., + sσ sin θT B ¯σα(cid:48)(cid:17) cos θT Bσα(cid:48) (13) . σ In the following we study how the spin structure of the helical edge states can be explored by means of transport properties. We begin by investigating the noninteracting case, which we can compare with numerical simulations on a lattice. III. NON-INTERACTING ANTIDOT To investigate the transport properties in the absence of in- teractions we use the standard scattering matrix formalism.41 After calculating the Heisenberg equations of motion i∂tΨsα = [Ψsα,H] and i∂tΨdα = [Ψdα,H] with respect to the Hamilto- nian (1), and by imposing plane-wave solutions for the states coming from (with amplitude ai) and going to (with ampli- tude bi) the contacts i = 1, . . . , 4 (see Fig. 1), we can find the scattering matrix relating bi =(cid:80) j S i ja j as (cid:0)1 + Γ2(cid:1) sin φ + 2iΓ cos φ 1 S =  (cid:16) 1 − Γ2(cid:17) 0 sin φ −2iΓ sin Θ −2iΓ cos Θ (cid:16) 1 − Γ2(cid:17) sin φ 0 −2iΓ cos Θ 2iΓ sin Θ (cid:16) 2iΓ sin Θ −2iΓ cos Θ 1 − Γ2(cid:17) 0 (cid:16) −2iΓ cos Θ −2iΓ sin Θ sin φ 1 − Γ2(cid:17) sin φ 0  , 4 (14) with the dimensionless tunneling probability Γ = γT2/(4v2 F). The scattering matrix depends on the chemical potential through the phase factor φ = πRµ/vF and on the parameter Θ = θUkF −θLkF +2θT . The current measured at the i-th contact can be evaluated from Eq. (14) using the Landauer-Buttiker formula (cid:90) µ+eVi (cid:88) Ii = G0 e dE Ti j(E), µ+eV j j where G0 = e2/(2π) is the conductance quantum, Ti j = are the elements of the transmission matrix, and {V j} are the bias potentials applied to the four contacts. It is worth not- ing that the spin structure of the helical states on the antidot does not affect the transport properties. Nevertheless, the re- sult does depend on the spin textures of the edges through Θ. Therefore, if the latter can be tuned independently, for instance by applying an electric-field gradient, they can be di- rectly resolved via a current measurement even in the nonin- teracting case. This result is analogous to what was found for a tunnel junction between two edges.24 In contrast, in the homo- geneous case, i.e., with the same spin structure θUk = θLk on both edges, Θ = 2θT and the transmission matrix is uniquely determined by the ratio between spin-preserving and spin- flipping tunneling. If the chemical potential coincides with an eigenenergy of the antidot µ = vF j/R (a scenario which we will refer to as the resonant case), the phase factor sin φ = 0 so that the incoming electron is fully transmitted across the antidot, while away from resonance one recovers the typical Lorentz-shaped transmission for transport through a quantum (anti)dot. These results are confirmed by numerical transport simula- tions using the KWANT package.42 To investigate the effects induced by breaking the axial spin symmetry we consider the Kane-Mele (KM) lattice model9,13 on the honeycomb lattice, which is defined as HKM = −t (cid:88) (cid:104)i j(cid:105) + iλR c† i c j + iλSO (cid:88) νi jc† i (s × di j)zc j, c† (cid:104)(cid:104)i j(cid:105)(cid:105) (cid:88) (cid:104)i j(cid:105) i szc j (16) where ci = (ci↑, ci↓) is a two-component spinor, νi j = ±1 is a factor that is +1 (−1) if the next-nearest-neighbor hopping from site j to site i corresponds to a right turn (left turn) in the honeycomb lattice, s is the spin operator, and di j is the unit vector between the nearest-neighbor lattice sites i and j. The parameters of the model are the hopping amplitude t, the intrinsic spin-orbit coupling λSO, and the Rashba spin-orbit coupling (RSOC) λR, which is responsible for the axial spin symmetry breaking. In the following we limit ourselves to (15) (cid:12)(cid:12)(cid:12)S i j (cid:12)(cid:12)(cid:12)2 FIG. 2. Representative example of the electron densities (green dots) for scattering states originating from lead 1. Panel (a) shows the up-spin propagation and panel (b) shows the down-spin propagation. The lattice is 60 lattice constants long and 34 lattice constants wide. The leads, shown in red, are 9 lattice constant wide. The model parameters are set to: t = 1, λSO = 0.2t, λR = 0.1t and µ = 0.157t. uniform bulk parameters, which corresponds in the analytical model to the homogeneous case θUk = θLk. The KM Hamiltonian with RSOC does not preserve inver- sion symmetry.43 Inversion does not affect the spin as the lat- ter is a pseudovector. It does exchange the two sublattices forming the honeycomb lattice, but it leaves the phase νi j in- variant. Hence, the kinetic and the intrinsic spin-orbit terms of the Hamiltonian are invariant. However, the RSOC term gets a minus sign under inversion, which destroys the inver- sion symmetry of the total Hamiltonian. On the other hand, HKM has reflection symmetry. A reflec- tion about the x axis (y → −y) will change the signs of the x and z components of the spin. Moreover, reflection symme- try swaps the sublattices. Hence, the kinetic term is invariant. For the spin-orbit part, the z component of spin will pick up a minus sign, but νi j will change sign, too. For the RSOC term, the y component (x component) of the lattice vectors switches (does not switch) sign, but the x component (y component) of the spin also switches (does not switch) sign. Hence, a reflec- tion about the x axis leaves the Hamiltonian invariant. For the numerical simulation, we consider a finite lattice connected to four leads, as shown in Fig. 2. To avoid dangling bonds, we model the antidot as hexagonal-shaped. The nu- merical calculation provides access to the full scattering ma- trix, and without loss of generality we will discuss in the fol- 402002045 FIG. 3. Transmission coefficient from lead 1 to lead 2 on panel (a), to lead 3 on panel (b) and to lead 4 on panel (c), for several val- ues of RSOC. The lattice setup is the same as in Fig. 2. The model parameters are set to: t = 1, λSO = 0.2t. lowing the transmission of electrons injected from lead 1 to the other leads. The finite-length system provides, in addition to transport via the antidot, direct ballistic channels between the upper and the lower edge along the left and right edges of the sample. However, these are easy to distinguish from transport via the antidot. Fig. 2 shows the weight of the spin- up and spin-down wavefunctions for small RSOC (θskF (cid:28) 1) on the antidot and the leads. The most important effect of RSOC is to enable spin flips at the tunnel contacts. The amplitude for spin-flip processes can be quantified by calculating the transmission probability from lead 1 to lead i, Ti1, shown in Fig. 3 as a function of the chemical potential. On resonance with an antidot energy level, T21 drops to zero, as shown in Fig. 3(a), in agreement with the analytic result in Eq. (14) at φ = 0, showing that the in- jected electrons are fully transmitted to the opposite edge. The transmission T31 between leads 1 and 3, shown in Fig. 3(b), is only nonzero if there is spin-flip tunneling. Hence, in the absence of RSOC, all electrons are transmitted to lead 4 at resonance, as shown by the blue peaks in Fig. 3(c). Moreover, the transmissions in Fig. 3 are symmetric with respect to en- ergy only in the absence of RSOC. In contrast, in the presence of RSOC, particle-hole symmetry is broken so that in general Ti j(µ) (cid:44) Ti j(−µ). The peaks in the transmission probabilities all have Lo- rentzian shape around the resonance energies, but their widths change with energy. Using the numerical results for T31(µ) or T41(µ), we can calculate the values of θT , shown in Fig.4(a), and γT as functions of chemical potential. They turn out to vary slowly on the scale of the antidot level spacing. Hence, we are able to extract from the numerical simulations the de- pendence of the parameters of the analytic models on the tight-binding parameters as well as on chemical potential. Moreover, the Fermi velocity can be extracted from the band structure in the leads. Hence, we have access to all the quanti- ties entering our analytic model via the numerical simulation. We also performed tight-binding calculations based on the square-lattice discretization of the Bernevig-Hughes-Zhang model12 with added bulk inversion asymmetry,44 parametrized by ∆, whose effect is similar to λR in the KM model. The re- sults do not differ qualitatively from the ones presented here for the KM model. The main difference consists of the pre- served particle-hole symmetry as shown in Fig. 4(b). A. Non-local resistance To compare the analytical predictions with the numerical simulations, we need to take into account the additional bal- listic channels connecting contacts 1 and 4 and contacts 2 and 3 in Fig. 1 via the sample edges. This corresponds to replacing T14 → T14 + 1 and T23 → T23 + 1 (analogously for T41 and T32) obtained from Eq. (14), while all other co- efficients remain invariant. The nonlocal multi-terminal re- sistance is then computed by means of the Landauer-Buttiker formula (15) in the linear-response regime. For instance, the relation between the current flowing between contacts 1 and 4 and the voltage developed at these same contacts is given by 6 FIG. 4. (Color online) θT at the resonance energies for several values of (a) RSOC in the KM model and (b) bulk inversion asymmetry in the BHZ model. The white crosses indicate the points of transport resonance which were used to evaluate θT . The model parameters are set to: t = 1, λSO = 0.2t for KM model, and A = 3, B = −1, M = −2 and C = D = 0 for BHZ model.12 R14,14 = (V1− V4)/I1I4=−I1,I2=I3=0. A numerical result is shown in Fig. 5. In the homogeneous case θUk = θLk and at resonance (sin φ = 0) one finds (cid:35) G−1 0 . (17) (cid:34) R14,14 = 1 cos(2θT ) + 3 + 1 4 In the absence of spin-flip tunneling (θT = 0), we find R14,14 = (2G0)−1. On the other hand, if only spin-flip processes are allowed (θT = π/2), then R14,14 = 3/(4G0). As shown in Fig. 5, at resonance the nonlocal resistance reaches its minimum. Away from resonance, it tends towards 3/(4G0). Around µ = 0, the nonlocal resistance deviates slightly from its quantized value due to the finite length of the tunneling region in the numerical simulation, which has the tendency to open a small spectral gap.45,46 We can fit Eq. (17) to the envelope of the resonant peaks (green dashed line in Fig. 5) and thus determine the leading behavior θT (µ)−θT (0) ∼ µ2 for small µ. Let us briefly conclude the discussion of the non-interacting transport properties. In the case of an inhomogeneous RSOC, the transport properties depend explicitly on the spin texture. In the homogeneous case, the spin texture still appears im- plicitly in θT . With the help of the numerical simulation, we are able to extract the parameters of the analytic model as a function of the bulk parameters and the chemical potential. FIG. 5. Non-local resistance R14,14 in unit of G−1 0 . The lattice setup is the same as in Fig. 2. The model parameters are set to: t = 1, λSO = 0.2t and λR = 0.05t. The continuous (blue) line is the computed non- local resistance for the lattice. The (green) dashed line is the fit of our model assuming that θT (µ) ≈ αµ2 + β. IV. INTERACTING ANTIDOT In confined low-dimensional systems, electron interactions are known to play an important role. Therefore, to complete the study of the transport properties through the QSH anti- dot, we need to investigate how interactions affect the trans- port mechanisms. Computing the transport properties in the presence of electron interactions is generally a difficult task that cannot be solved exactly for arbitrary tunneling strength. Therefore, we focus on the lowest-order contributions to the tunneling current through the antidot, which are (A) sequen- tial tunneling and (B) cotunneling.47 0.40.30.20.10.00.10.20.30.4µ/t0.020.040.060.080.100.120.14λR/t(a) KM model0.80.60.40.20.00.20.40.60.8µ/B0.10.20.30.40.50.60.70.8∆/B(b) BHZ modelθT0.00.10.20.30.40.50.60.70.8A. Sequential tunneling If the dwell time τ of electrons on the antidot is large such that 1/τ (cid:28) {eVi, kBT}, the dominant transport processes are single-electron transfers between the edges and the antidot. The transport properties can then be evaluated within first or- der perturbation theory in γT2. We assume that the antidot contains N electrons in the ground state, and we assume N to be even without loss of generality. The initial state of the leads contains one electron at a certain momentum k in one of the leads. The initial state of the full system is thus a direct product of the initial state in the leads and in the antidot: isα(N, k)(cid:105) = N(cid:105) ⊗ c† sαkvac(cid:105). (18) We compute the transition rate for adding another electron on the antidot. The final state due to tunneling of one electron from the edge s with initial momentum k and chirality α to the antidot, with final chirality α(cid:48) and angular momentum j, reads f sαα(cid:48) (N + 1, k, j)(cid:105) = d† α(cid:48) jcsαkisα(N, k)(cid:105). (19) According to Fermi's golden rule, the rate for transitions from the initial state to the final state is47 N+1,N(k, j) = 2π(cid:104) fHd,si(cid:105)2Fiδ(E f − Ei), Γsαα(cid:48) (20) where E f − Ei = E(N + 1) − E(N) + εdα(cid:48)( j) − εsα(k) is the energy difference between final and initial states and Fi is a Fermi function denoting the probability of finding the system in the initial state i(cid:105). εdα(cid:48)( j) is the eigenenergy of the antidot with angular momentum j and chirality α(cid:48), and εsα(k) is the eigenenergy of the s edge with momentum k and chirality α. The total transition rate is then obtained by summing over all possible initial and final states, Γsαα(cid:48) N+1,N = Γsαα(cid:48) N+1,N(k, j). (21) (cid:88) j,k In the sequential tunneling regime, the tunneling current is evaluated using a rate-equation approach. In the dc limit, and by considering only two antidot states with either N or N + 1 electrons, which is valid as long as the charging energy is large enough to forbid other occupation numbers, one has with ΓN+1,N = (cid:80) ΓN+1,NP(N) = ΓN,N+1P(N + 1) s,α,α(cid:48) Γsαα(cid:48) (22) N+1,N and similarly for ΓN,N+1. Com- bined with the conservation of probability constraint, P(N) + 2P(N + 1) = 1, it is then possible to compute the occupa- tion probabilities in terms of the transition rates. Transitions between the state with N electrons and the one with N +1 elec- trons in the antidot are enabled close to the resonance condi- tion ∆E(N + 1) = E(N + 1)− E(N) = 0. Therefore, transitions between N and N + 1 states are allowed for ng ≈ N + 1 2, where ng = eVg/Ec is determined by the gate voltage and the charg- ing energy. The recursive equation (22) and the probability conservation yield the expression of the probabilities, which are necessary to compute the total current, I = −e P(N)ΓU N+1,N − P(N + 1)ΓU N,N+1 . (23) (cid:105) (cid:104) 7 The expression of the current is still complicated and depends on how many levels can be reached in the bias window. If the bias window is sufficiently small, tunneling is only possible via one energy level εd( j) situated near the Fermi energy µ. In this case, I = −e 2 +(cid:2)nF(εd( j) − µU) + nF(εd( j) − µL)(cid:3) , nF(εd( j) − µU) − nF(εd( j) − µL) 2ΓT πR (24) where ΓT = γT2/(2vF). We choose the chemical potentials as µU = µ + eV/2 and µL = µ− eV/2, where V is the bias voltage between upper and lower edges. At T = 0 and finite voltage, we get IT =0 = −2eΓT 3πR , (25) for εd( j) ∈ [µ − eV/2, µ + eV/2]. On the other hand, if the temperature is finite and eV (cid:28) kBT, the current to lowest order in the applied voltage becomes e2V 1 IT (cid:44)0 = − 1 4kBT ΓT πR , 1 + nF(εd) (26) where εd = εd( j) − µ. As is well known in the sequential tunneling regime, the limits eV → 0 and kBT → 0 do not commute.47 cosh2(εd/(2kBT)) One of the central assumptions of this rate equation ap- proach is that the electrons on the dot relax to the ground state between tunneling events, i.e., there is a separation of time scales between the fast relaxation and the slow tunnel- ing. However, in our case, either the initial state N(cid:105) or the final state N + 1(cid:105) is twofold-degenerate due to time-reversal symmetry. Since the rate-equation approach does not properly account for the fact that the chirality of the electrons on the antidot is conserved, it is not possible to calculate chirality- resolved currents within this approach. Hence, we only pre- sented results for the total current I. However, since the total current does not contain information about the spin texture, we continue by exploring higher-order coherent processes. B. Cotunneling If sequential tunneling is inhibited due to energy conserva- tion, transport between the upper and lower edge is still possi- ble via cotunneling, where electrons tunnel between the upper and lower edge via virtual states on the antidot. A sketch of a possible process is shown in Fig. 6. In the following we con- sider elastic cotunneling which, due to the discrete nature of the mesoscopic antidot, becomes relevant for transport. The cotunneling rate from the initial state isα(N, k)(cid:105), which con- tains N electrons on the antidot and a single electron with momentum k and chirality α on the edge s, to the final one f s(cid:48)α(cid:48) (N, k(cid:48))(cid:105), which is defined analogously, reads Γi→ f = 2πδ(E f − Ei)Fi × (cid:104) fHd,L Hd,U + Hd,U Hd,Li(cid:105)2, (27) 1 1 Ei − H0 Ei − H0 8 with the parameter θT evaluated at the chemical potential µ. We recover here an implicit dependence on the spin texture through θT . The Coulomb repulsion will just shift the energy in the denominator, depending on the value of the gate volt- age. Moreover, it is possible to invert the expression for the current in order to extract the value of θT as a function of the current, the voltage, and the charging energy. This makes it possible in principle to compare the tunneling processes in the noninteracting and in the interacting limits. The case in which the antidot level hosts an odd number of electrons, i.e., if the energy level at the chemical potential has only one electron, is quite different. Due to the degeneracy of the antidot energy level, the initial and final states should include the initial (final) chirality of the antidot β((cid:48)), becom- (N, k(cid:48))(cid:105).48 By applying Eq. (27), we ing isαβ(N, k)(cid:105) and f s(cid:48)α(cid:48)β(cid:48) (cid:88) (cid:104) modify the current expression in (28) to (cid:105) Iαα(cid:48) = (−e) U→L − Γα(cid:48)αββ(cid:48) Γαα(cid:48)ββ(cid:48) L→U pβ. (31) β,β(cid:48) The rates are found in Appendix A. The probabilities pβ of the highest level of the antidot being occupied by an electron with chirality β are determined by the conservation of probabilities p+ + p− = 1 and by the rate equation, = −Γβ¯β pβ + Γ¯ββ p¯β (32) = 0, where Γβ¯β = (cid:80) dpβ dt αα(cid:48)(Γαα(cid:48)β¯β U→L + Γαα(cid:48)β¯β L→U ). By evaluating the rates at the chosen chemical potentials, we are able to compute the chirality-resolved currents. As an example, we set a difference of potential between the upper edge and the lower edge, such that µU+ = µU− = µ + eV 2 , leading to N odd ≈ −2Γ2 TG0V Iαα(cid:48) v2 F (cid:34) 2 and µL+ = µL− = µ − eV 1 1 ξ2(N) αα(cid:48) cos + + (cid:16) ξ2(N + 1) 4θkF − 4θ j0 ξ(N)ξ(N + 1) (cid:17) − 1  , (33) where ξ(N) = 2πR[εd( j0) + ∆E(N) − µ]/vF. In this result, we obtain an explicit dependence of the current on the anti- dot and edge state spin textures, θ j0 and θkF , even in the case of position-independent RSOC. Hence, a measurement of the cotunneling current allows one to measure the difference be- tween the external edge spin rotation θkF and the antidot spin rotation θ j. Since the system occupies a virtual intermediate state, θkF will generally differ from θ j0. In particular, the dif- ference of currents between the two lower terminals leads to (cid:16) (cid:33)2 cos (cid:32)2ΓT vF 4θkF − 4θ j0 ξ(N)ξ(N + 1) (cid:17) N odd − I++ I+− N odd ≈ G0V (34) This measurement would enable us to extract directly the in- formation about the spin texture. Another possibility to probe the spin texture would be to apply a different voltage setting: only one lead is biased, such as for example µU+ = µ + eV and µU− = µL+ = µL− = µ. Again, we can divide the tunneling current in two contribu- tions I++ N odd that we compute using Eq. 31, leading to N odd and I+− FIG. 6. Sketch of a possible cotunneling process between lead 1 and lead 4. The energy level on the antidot is fully occupied. One of the two electrons escape the antidot towards lead 4, creating a virtual state with an additional hole. The final state is reached when the electron in lead 1 tunnels to the antidot. where H0 = HU + HL + Hd, and Ei, f are the energies of the initial and final state. Fi is again the Fermi distribution specifying the probability of finding the system in the initial state isα(N, k)(cid:105). Note that the antidot contains N electrons in the ground state in both initial and final states. The total chirality-resolved cotunneling rates Γαα(cid:48) U→L and Γαα(cid:48) L→U are then obtained by summing over all the possible initial- and final-state momenta, and over all angular momenta and chiralities in the intermediate state. From the tunneling rates it is then possible to compute the tunneling current, de- fined as flowing from the upper to the lower edge, as (cid:104) (cid:105) Iαα(cid:48) = (−e) U→L − Γα(cid:48)α Γαα(cid:48) L→U . (28) To connect to the setup shown in Fig. 1, I++ is the current flowing from lead 1 to lead 4, I+− is the current flowing from lead 1 to lead 3, I−+ is the current flowing from lead 2 to lead 4, and I−− the current flowing from lead 2 to lead 3. The results for these currents depend strongly on the parity of the antidot occupation. In the case of an even number N of electrons on the anti- dot, i.e., if all levels up to the chemical potential are doubly occupied, we find = − eΓ2 T 2π3R2 dε[nF(ε − µUα) − nF(ε − µLα(cid:48))] Iαα(cid:48) N even (cid:90) (cid:88) (cid:2)ε − εd( j) − ∆E(cid:3)2 j × cos2(2θT )δα(cid:48)α + sin2(2θT )δ ¯α(cid:48)α , (29) where εd( j) + ∆E is the energy of the intermediate state. De- pending on the gate voltage, ∆E = ∆E(N) if an additional hole occupies the j-th level of the dot or ∆E = ∆E(N +1) for an ad- ditional electron. At low bias it is possible to expand the Fermi functions around µ and, by assuming that only the j0th antidot energy level (the closest from the chemical potential) is con- tributing to the cotunneling current and ∆E + εd( j0) (cid:29) µU/Lα, one obtains N even ≈ − e2(Vα − Vα(cid:48))Γ2 Iαα(cid:48) cos2(2θT )δα(cid:48)α + sin2(2θT )δ ¯α(cid:48)α (cid:2)µ − εd( j0) − ∆E(cid:3)2 2π3R2 T (30) N odd ≈ −2Γ2 TG0V I+− v2 F ξ(N + 1)ξ(N) − [ξ(N + 1) − ξ(N)] cos (cid:32) 1 (cid:32) (cid:16) 2θkF − 2θ j0 (cid:33)2(cid:110) (cid:33)2(cid:110) ξ2(N + 1) + ξ2(N) − ξ(N + 1)ξ(N) + 2θT ξ(N + 1) cos N odd ≈ −2Γ2 TG0V I++ v2 F ξ(N + 1)ξ(N) 1 (cid:16) + [ξ(N + 1) − ξ(N)] cos 2θkF − 2θ j0 + 2θT ξ(N + 1) cos ξ2(N + 1) + ξ2(N) − ξ(N + 1)ξ(N) (cid:17)(cid:104) (cid:17)(cid:104) (cid:16) (cid:104) (cid:16) 1 + cos 2θkF − 2θ j0 − 2θT (cid:104) (cid:16) 1 − cos 2θkF − 2θ j0 − 2θT (cid:16) 4θkF − 4θ j0 (cid:17)(cid:105) (cid:17) − ξ(N) cos (cid:17)(cid:105) (cid:17) 4θkF − 4θ j0 + ξ(N) cos 9 (35) (cid:17)(cid:105)(cid:111) (cid:16) 2θkF + 2θ j0 + 2θT (cid:16) 2θkF − 2θ j0 (cid:17)(cid:105)(cid:111) + 2θT . (36) To conclude the discussion of the electron transport in the interacting case, we observe in the cotunneling regime that the occupation of the antidot plays an important role. In the case of an even occupation, we recover the implicit dependence of the current on the spin texture. In contrast, for odd occupa- tion, we find an explicit dependence on the spin texture. By tuning the chemical potentials, it is thus possible to detect the interplay between implicit (θT ) and explicit (θkF and θ j) de- pendence of the spin texture. V. CONCLUSION We have presented a detailed analysis of the electron trans- port between the edges of a two-dimensional topological in- sulator via an antidot. In particular, we investigated the effects of a nontrivial spin structure of the edge states and a charging energy due to Coulomb repulsion on the antidot. We first presented a solution of the corresponding scattering problem in the absence of interactions. We showed that, on the one hand, spin-non-conservation modifies the spin texture of the edge states, but on the other hand, it also makes spin- flip tunneling between the edges and the antidot possible. We also performed numerical calculations based on tight-binding models, and we confirmed the predictions of the scattering ap- proach. We found that the effects of spin-non-conservation be- come most important for inhomogeneous samples where the spin structures of different edge states may differ. To include the effects of charging energy, we investigated the Coulomb blockade regime, where we presented a calcu- lation of the sequential tunneling and cotunneling currents. Here, we showed in particular that, since the cotunneling oc- curs via an intermediate, virtual state on the antidot, it allows a spectroscopic measurement of the antidot and edge-state spin structure. ACKNOWLEDGMENTS This work was supported through the German DFG priority program SPP 1666 on topological insulators. AR, GD, and TLS acknowledge support from the National Research Fund, Luxembourg (ATTRACT 7556175). SR thanks the DFG for financial support through SFB 1143. FIG. 7. Sketch of the two contributions to the tunneling current I++ N odd and I+− N odd in the case only lead 1 is biased, that is µU+ = µ + eV and µU− = µL+ = µL− = µ. Panel (a) corresponds to weak Rashba inter- actions, so that spin is almost conserved: therefore the contribution I+− N odd due to spin-flip is strongly suppressed. In the opposite sce- nario of strong Rashba interactions, spin-flip contributions can even become dominant compared to spin-preserving one, leading to the scenario depicted in panel (b). = θ j0 We observe this time a more sophisticated dependence on the different parameters. However, from Eq. (35) one can see that the current flowing to lead 3 vanishes if θkF = θT = 0, that is, in the absence of RSOC; indeed, in this case spin is conserved and electrons, injected fully spin-up polarized from lead 1, can only flow to lead 4 preserving their spin, as schematically shown in Fig. 7(a). However, in the case of strong RSOC, spin-flip tunneling can become important, eventually dominating over the spin-preserving contribution. In this case, the current mostly flows from lead 1 to lead 3, as schematically shown in Fig. 7(b). Appendix A: Cotunneling rates for an odd number of electrons 10 When we apply Eq. (27), with the additional degree of freedom due to the chirality, we are able to compute the 16 transition ξ(N) + µ − ε (cid:34)cos(2θT ) + αβ cos(2θk − 2θ j) (cid:34)sin(2θT ) + αβ sin(2θk − 2θ j) (cid:34)sin(2θT ) − αβ sin(2θk − 2θ j) (cid:34)cos(2θT ) − αβ cos(2θk − 2θ j) ξ(N) + µ − ε ξ(N) + µ − ε ξ(N) + µ − ε ξ(N + 1) + µ − ε −sin(2θT ) + αβ sin(2θk − 2θ j) (cid:35)2 −cos(2θT ) − αβ cos(2θk − 2θ j) (cid:35)2 (cid:35)2 (cid:35)2 −sin(2θT ) + αβ sin(2θk − 2θ j) −cos(2θT ) − αβ cos(2θk − 2θ j) ξ(N + 1) + µ − ε ξ(N + 1) + µ − ε ξ(N + 1) + µ − ε , , , (A1) (A2) (A3) . (A4) rates appearing in Eq. (31). j (cid:90) (cid:88) (cid:90) (cid:88) (cid:90) (cid:88) (cid:90) (cid:88) j j j Γααββ U→L = Γ2 T 2πv2 F Γααβ¯β U→L = Γα ¯αββ U→L = Γ2 T 2πv2 F Γ2 T 2πv2 F Γα ¯αβ¯β U→L = Γ2 T 2πv2 F dεnF(ε − µUα)[1 − nF(ε − µLα)] dεnF(ε − µUα)[1 − nF(ε − µLα)] dεnF(ε − µUα)[1 − nF(ε − µL ¯α)] dεnF(ε − µUα)[1 − nF(ε − µL ¯α)] The remaining rates for the transition Γα(cid:48)αββ(cid:48) L→U are obtained by detailed balance. 1 M. Z. Hasan and C. L. Kane, Rev. Mod. Phys. 82, 3045 (2010). 2 X.-L. Qi and S.-C. Zhang, Rev. Mod. Phys. 83, 1057 (2011). 3 M. Konig, S. Wiedmann, C. Brune, A. Roth, H. Buhmann, L. W. Molenkamp, X.-L. Qi, and S.-C. Zhang, Science 318, 766 (2007). 4 A. Roth, C. Brune, H. Buhmann, L. W. Molenkamp, J. Maciejko, X. Qi, and S. Zhang, Science 325, 294 (2009). Phys. Status Solidi RRL 7, 1059 (2013). 19 G. Dolcetto, F. Cavaliere, and M. Sassetti, Phys. Rev. B 89, 20 A. Rod, T. L. Schmidt, and S. Rachel, Phys. Rev. B 91, 245112 125419 (2014). (2015). 21 N. Kainaris, I. V. Gornyi, S. T. Carr, and A. D. Mirlin, Phys. Rev. 5 I. Knez, R.-R. Du, and G. Sullivan, Phys. Rev. Lett. 107, 136603 B 90, 075118 (2014). (2011). 6 I. Knez, C. T. Rettner, S.-H. Yang, S. S. P. Parkin, L. Du, R.-R. Du, and G. Sullivan, Phys. Rev. Lett. 112, 026602 (2014). 7 K. C. Nowack, E. M. Spanton, M. Baenninger, M. Konig, J. R. Kirtley, B. Kalisky, C. Ames, P. Leubner, C. Brune, H. Buhmann, L. W. Molenkamp, D. Goldhaber-Gordon, and K. A. Moler, Na- ture Materials 12, 787 (2013). 8 E. M. Spanton, K. C. Nowack, L. Du, G. Sullivan, R.-R. Du, and K. A. Moler, Phys. Rev. Lett. 113, 026804 (2014). 9 C. L. Kane and E. J. Mele, Phys. Rev. Lett. 95, 146802 (2005). 10 C. Wu, B. A. Bernevig, and S.-C. Zhang, Phys. Rev. Lett. 96, 106401 (2006). (2006). 11 C. Xu and J. E. Moore, Phys. Rev. B 73, 045322 (2006). 12 B. A. Bernevig, T. L. Hughes, and S.-C. Zhang, Science 314, 1757 13 C. L. Kane and E. J. Mele, Phys. Rev. Lett. 95, 226801 (2005). 14 C. Liu, T. L. Hughes, X.-L. Qi, K. Wang, and S.-C. Zhang, Phys. Rev. Lett. 100, 236601 (2008). 15 D. G. Rothe, R. W. Reinthaler, C.-X. Liu, L. W. Molenkamp, S.-C. Zhang, and E. M. Hankiewicz, New J. Phys. 12, 065012 (2010). 16 C.-C. Liu, W. Feng, and Y. Yao, Phys. Rev. Lett. 107, 076802 (2011). 22 G. Dolcetto, M. Sassetti, and T. L. Schmidt, Rivista del Nuovo Cimento 39, 113 (2016). 91, 081406 (2015). 23 C. P. Orth, R. P. Tiwari, T. Meng, and T. L. Schmidt, Phys. Rev. B 24 C. P. Orth, G. Strubi, and T. L. Schmidt, Phys. Rev. B 88, 165315 25 J. S. Van Dyke and D. K. Morr, Phys. Rev. B 93, 081401 (2016). 26 V. J. Goldman and B. Su, Science 267, 1010 (1995). 27 V. J. Goldman, J. Liu, and A. Zaslavsky, Phys. Rev. B 77, 115328 28 M. R. Geller and D. Loss, Phys. Rev. B 56, 9692 (1997). 29 Y. Komijani, P. Simon, and I. Affleck, Phys. Rev. B 92, 075301 (2013). (2008). (2015). 30 G. Dolcetto, F. Cavaliere, D. Ferraro, and M. Sassetti, Phys. Rev. 31 S.-Y. Hwang, R. L´opez, M. Lee, and D. S´anchez, Phys. Rev. B 90, 32 R. L´opez, S.-Y. Hwang, and D. S´anchez, J. Phys.: Conf. Ser. 568, B 87, 085425 (2013). 115301 (2014). 052016 (2014). 33 G. Dolcetto and T. L. Schmidt, arXiv:1604.05967. 34 T. Posske, C.-X. Liu, J. C. Budich, and B. Trauzettel, Phys. Rev. Lett. 110, 016602 (2013). 17 T. L. Schmidt, S. Rachel, F. von Oppen, and L. I. Glazman, Phys. Rev. Lett. 108, 156402 (2012). 18 G. Dolcetto, N. T. Ziani, M. Biggio, F. Cavaliere, and M. Sassetti, 35 T. Posske and B. Trauzettel, Phys. Rev. B 89, 075108 (2014). 36 B. Rizzo, A. Camjayi, and L. Arrachea, arXiv:1605.06875. 37 R.-L. Chu, J. Lu, and S.-Q. Shen, Europhys. Lett. 100, 17013 (2012). 38 V. Krueckl and K. Richter, Phys. Rev. Lett. 107, 086803 (2011). 39 F. Dolcini, Phys. Rev. B 83, 165304 (2011). 40 F. Dolcini, Phys. Rev. B 92, 155421 (2015). 41 S. Datta, Electronic transport in mesoscopic systems (Cambridge University Press, Cambridge, 1997). 42 C. W. Groth, M. Wimmer, A. R. Akhmerov, and X. Waintal, New Journal of Physics 16, 063065 (2014). 43 L. Fu and C. L. Kane, Phys. Rev. B 76, 045302 (2007). 44 M. Koenig, H. Buhmann, L. W. Molenkamp, T. L. Hughes, C.- 11 X. Liu, X.-L. Qi, and S.-C. Zhang, J. Phys. Soc. Jpn. 77, 031007 (2008). 45 P. Sternativo and F. Dolcini, Phys. Rev. B 89, 035415 (2014). 46 R.-L. Chu, J. Li, J. K. Jain, and S.-Q. Shen, Phys. Rev. B 80, 081102 (2009). 47 H. Bruus and K. Flensberg, Many-body quantum theory in con- densed matter physics (Oxford University Press, Oxford, 2002). 48 J. Lehmann and D. Loss, Phys. Rev. B 73, 045328 (2006). 

1206.0308

2

1206

Coulomb drag in graphene - boron nitride heterostructures: the effect of virtual phonon exchange

For a system of two spatially separated monoatomic graphene layers encapsulated in hexagonal boron nitride, we consider the drag effect between charge carriers in the Fermi liquid regime. Commonly, the phenomenon is described in terms of an interlayer Coulomb interaction. We show that if an additional electron - electron interaction via exchange of virtual substrate phonons is included in the model, the predicted drag resistivity is modified considerably at temperatures above 150 K. The anisotropic crystal structure of boron nitride, with strong intralayer and comparatively weak interlayer bonds, is found to play an important role in this effect.

cond-mat.mes-hall

cond-mat

Coulomb drag in graphene -- boron nitride heterostructures: the effect of virtual phonon exchange Bruno Amorim,1 Jurgen Schiefele,2 Fernando Sols,2 and Francisco Guinea1 1Instituto de Ciencia de Materiales de Madrid, CSIC, Cantoblanco, E-28 049, Madrid, Spain 2Departamento de F´ısica de Materiales, Universidad Complutense de Madrid, E-28 040, Madrid, Spain (Dated: September 28, 2012) For a system of two spatially separated monoatomic graphene layers encapsulated in hexagonal boron nitride, we consider the drag effect between charge carriers in the Fermi liquid regime. Com- monly, the phenomenon is described in terms of an interlayer Coulomb interaction. We show that if an additional electron -- electron interaction via exchange of virtual substrate phonons is included in the model, the predicted drag resistivity is modified considerably at temperatures above 150 K. The anisotropic crystal structure of boron nitride, with strong intralayer and comparatively weak interlayer bonds, is found to play an important role in this effect. I. INTRODUCTION If two systems containing mobile charge carriers are spatially separated such that direct charge transfer is not possible, but close enough to allow interaction be- tween the carriers in different layers, the resulting mo- mentum transfer will equalize the drift velocities in both systems. This frictional effect was experimentally ob- served between (quasi) two-dimensional electron gases in double quantum well structures1,2. In most of the theo- retical work the interlayer interaction was attributed to Coulomb scattering, hence the effect now bears the name 'Coulomb drag' (see Refs. 3 -- 5). Interest in the subject has been revived recently by the experimental progress which made it possible to pre- pare two-dimensional electron systems based on mono- layer graphene. A considerable number of theoretical works6 -- 16 studied Coulomb drag between massless Dirac fermions, which effectively describe the charge carriers in graphene17. However, a quantitatively correct explana- tion of the experimental data is still lacking6,18 -- 20. In the typical experiment, Coulomb drag is studied by driving a constant current I2 through one of the layers (the active one, labeled by the index λ = 2 in Fig. 1). If no current is allowed to flow in the other (passive, in- dex 1) layer, a potential difference V1 builds up there. In terms of these two quantities, the drag resistivity ρD ≡ (W/L)V1/I2 serves as a measure of the momen- tum transfer between the two layers, where W and L are, respectively, the width and the length of the layer. A theoretical expression for ρD in second order in the inter- layer interaction can be derived either using Boltzmann's kinetic equation4,10,11,21 or the Kubo formula5,9,11. In the present work, we focus on the interlayer interac- tion responsible for the drag effect in heterostructures composed of two graphene monolayers and hexagonal boron nitride (hBN), see Fig. 1. The large bandgap in- sulator hBN has a layered structure composed of stacked hexagonal crystal planes. Recently the material received much attention as it allows the construction of graphene -- hBN devices with, in comparison to the much used SiO2 substrates, favorable high carrier mobilities22 -- 25. In FIG. 1. A sketch of the double layer system under consid- eration. The two monoatomic graphene layers (yellow) with charge carrier concentration n1, n2 are placed at z = 0 and z = d and labelled by the layer index λ = 1, 2, respectively. The surrounding space (regions I, II, and III) is filled with the insulating material boron nitride with hexagonal struc- ture (hBN). particular, the Manchester group reported the fabrica- tion of devices where a few layer thin hBN crystal, ob- tained by exfoliation, is sandwiched between two mono- layers of graphene26 -- 28. If such a structure is used for a Coulomb drag experiment, the Dirac fermions in the ac- tive and passive layer can exchange momentum not only via Coulomb interaction but also by phonon exchange through the spacer medium. The effect of a combined Coulomb-phonon coupling on the drag resistivity has pre- viously only been studied for quasi two dimensional elec- tron gases in semiconductor systems29 -- 34. In the following, we first investigate the effects of the anisotropy of hBN, where the bonds in between the graphene-like planes are much weaker than the in-plane bonds, on the electron -- electron interaction via phonon exchange. We then show that the inclusion of phonon exchange into the description of Coulomb drag can sig- nificantly alter the temperature, density and distance de- pendence of the predicted value for ρD at temperatures above 150 K. IIIIIIΛ(cid:61)1Λ(cid:61)2z(cid:61)0z(cid:61)dII. INTERLAYER INTERACTION A. Combined Coulomb -- phonon mediated interaction To obtain the combined Coulomb-phonon interaction λλ(cid:48) in the anisotropic medium, we solve Poisson's equa- U (0) tion − ∇ · (
· ∇φ) = ρfree/
vac 2 In a two-layer system as shown in Fig. 1, where the re- gions I, II and III are filled with a homogeneous isotropic dielectric medium, the Fourier transform of the bare (un- screened) Coulomb potential between electrons in layers λ and λ(cid:48) has the form V (0) λλ(cid:48) (q) = 1
∞ e2 2
vacq e−qd(1−δλλ(cid:48) ) , (1) where q = (qx, qy),
vac denotes the dielectric constant of vacuum and
∞ accounts for the high frequency screen- ing properties of the medium. Apart from this Coulomb interaction, the charge carriers in each graphene layer in- teract via a substrate phonon mediated interaction. The charge carriers from each layer couple to the long range electric fields generated by optically active phonon modes in the surrounding material via Frohlich coupling35 -- 37. This remote interaction between carriers in graphene and optical phonon modes in a substrate medium was found to influence the electrical conductivity of graphene on a dielectric substrate24,38,39. In Appendix A, we show that in an isotropic medium, the combined interaction between electrons in layers λ and λ(cid:48) via the effects of a static Coulomb potential and virtual substrate phonon exchange is of the form of Eqn. (1), with
∞ replaced by the frequency depen- dent dielectric function
(ω) of the substrate material (see Eqn. (A7)). In the following, we specialize to the anisotropic spacer material hBN. From its three acoustic and nine optical phonon bands, only those that (via dipole oscillations) create long range electric fields couple to the graphene electrons40. Given the layered uniaxial crystal struc- ture of hBN, these (infrared active) optical modes are described by a dielectric tensor of the form41
(ω) = diag(cid:2)
⊥(ω),
⊥(ω),
(cid:107)(ω)(cid:3) . (2) The resonance frequencies ω retarded42 dielectric functions TO of the two
⊥,(cid:107)(ω) =
⊥,(cid:107) ∞ + f⊥,(cid:107) , (3) (cid:107) TO and ω⊥ (cid:1)2 (cid:0)ω (cid:1)2 − ω2 − iωγ⊥,(cid:107) ⊥,(cid:107) TO (cid:0)ω ⊥,(cid:107) TO are the phonon frequencies at the Γ point for trans- verse intraplane shear modes with displacements parallel and perpendicular to the c-axis of the crystal (aligned with the z direction in Fig. 1), respectively. We make the usual approximation of dispersionless optical phonon bands.29,35,43 The values for the high frequency dielec- tric constants
∞, the oscillator strengths f (related to the static,
0, and high frequency dielectric constants, f =
0 −
∞), ωTO and the damping factors γ taken from Ref. 44 are listed in Table I. with ρfree being the free charge density of a point charge −e at the origin. With Eqn. (2), Poisson's equation be- comes − ∂ ∂z
(cid:107) ∂ ∂z φ(q, z) + q2
⊥φ(q, z) = − e
vac δ(z) , (cid:18) (cid:19) 11 = U (0) 22 = −eφ(q, 0) we and as U (0) get 12 = −eφ(q, d) and U (0) (cid:115) U (0) λλ(cid:48)(q, ω) = 2
vac
(cid:107)(ω)q (cid:20) e2
(cid:107)(ω)
⊥(ω) −qd(1 − δλλ(cid:48)) (cid:115) × exp (4) (cid:21) .
⊥(ω)
(cid:107)(ω) A generalization of this result to structures where the regions I,II, and III (see Fig. 1) are filled with different insulating materials (or air) is straightforward; U (0) 11 then involves different dielectric functions than U (0) 22 . B. RPA screened interlayer interaction To take into account the screening properties of the conduction electrons in the graphene layers themselves, we employ the standard procedure of solving the Dyson equation for the two-layer system within the random phase approximation (RPA) (see Ref. 5). This finally yields the dressed interlayer interaction U12(q, ω) = U (0) 12 (q, ω)
RPA(q, ω) . (5) 12 U (0) 22 χ2) − U (0) 11 χ1)(1 − U (0) The total screening function for the coupled electron- phonon system given by (see Ref. 31 and Appendix B)
RPA = (1 − U (0) 21 χ1χ2 , (6) where χ1,2 denotes the (frequency and momentum de- pendent) polarizability of the graphene layers45. Fig. 2 shows a density plot of
RPA(q, ω), using dimen- sionless units x = q/kF and y = ω/(vF kF ), where kF is the Fermi momentum. The horizontal dashed green lines mark the transverse and longitudinal frequencies of the infrared active modes in hBN, connected by the TO =
0/
∞. For Lyddane-Sachs-Teller relation46 ω2 small damping γ (cid:28) ωTO, the real parts of
⊥,(cid:107)(ω) are ⊥,(cid:107) close to a pole at ω LO , re- spectively. Near these frequencies, the absolute value of the total screening function
RPA likewise shows an abrupt change from high values (light colors) to almost zero (dark colors). In regions where
RPA is small, the red lines Re
RPA = 0 show the coupled plasmon-phonon dispersion relation of the two-layer system. ⊥,(cid:107) TO and close to zero at ω LO/ω2 3 FIG. 4. Drag resistivity versus carrier density for various temperatures, d = 8 nm. Colors as in Fig. 3. FIG. 2. Absolute value of the total screening function
RPA Eqn. (6), with n1 = n2 = 0.02 nm−2 and d = 8 nm. Vertical (cid:107) LO, green lines show the optical resonance frequencies ω TO, and ω⊥ ω⊥ LO of hBN (bottom to top). Red curves mark the zeros of Re
RPA. The dashed black lines show the line y = x and mark the region where Imχ = 0. The hybridization between phonon and plasmon modes is clear. (cid:107) TO, ω FIG. 5. Drag resistivity versus layer separation for T = 300 K, n = 0.02 nm−2. Colors as in Fig. 3. The low-temperature asymptote ρlow T (dashed black curve, Eqn. (B2)), con- verges for large layer separation to ρlarge d (dotted green line, Eqn. (10)). Note that at this temperature and density ρlow T already differs from the full static calculation, ρCD. CD CD CD FIG. 3. Drag resistivity versus temperature for various in- terlayer distances, n = 0.02 nm−2. The blue curves show ρD (Eqn. (7)) including interaction via phonon exchange and Coulomb interaction, the dashed red curves show ρCD (Eqn. (9)) with Coulomb interaction only, and dashed black lines the low-temperature asymptote ρlow T CD (Eqn. (B2)). The lowest pair of curves (d = 8 nm) is also plotted on a linear scale in Fig. 7. III. RESULTS FOR THE DRAG RESISTIVITY In the following, we assume for the sake of simplic- ity the same positive carrier density n (corresponding to electron doping) in both layers, such that EF (cid:29) kBT . In particular, we do not address the recently reported drag at charge neutrality point20, which was attributed either to contributions from higher order perturbation theory15 or to correlated density inhomogenities in the graphene layers16,20. FIG. 6. The integral kernel K of Eqn. (8) with x = q/kF = 1, d = 8 nm, n = 0.02 nm−2 as a function of y = ω/vF kF for the temperatures 200, 100, 70 K (full curves, from top to bottom). The curves have been aligned on the left side by dividing with the y → 0 temperature dependence (T /TF)2. At T = (cid:107) 100 K (green curve) a peak near the resonance frequency ω TO appears, at T = 200 K (blue curve) there is and additional second peak near ω⊥ TO (see the vertical green lines in Fig. 2). Dashed curves show the integral kernel for ρCD, where these peaks are absent. 20501002003000.0010.010.11T
K

ΡD

(cid:87)
d(cid:61)8nmd(cid:61)6nmd(cid:61)4nmd(cid:61)2nm0.0100.0150.0200.0250.030(cid:45)0.7(cid:45)0.6(cid:45)0.5(cid:45)0.4(cid:45)0.3(cid:45)0.2(cid:45)0.10.0n
nm(cid:45)2
ΡD
(cid:87)
T(cid:61)300KT(cid:61)200KT(cid:61)100K0.10.51.05.010.050.00.0010.010.1110d
nm

ΡD

(cid:87)
0.00.20.40.60.81.001234y(cid:61)Ω

vFkF
(cid:75)
T
TF
(cid:45)2(cid:137)104T(cid:61)200KT(cid:61)100KT(cid:61)70KThe drag resistivity then assumes a negative value18, and the first non-vanishing contribution to ρD obtained in perturbation theory is of second order in the dressed interlayer interaction4,5. In terms of the variables carrier density, layer separation, and temperature, and under the assumptions that both layers are with high electron doping and T (cid:28) TF 47, it reads (refer to Refs. 6 and 10 for details) √ ρD = −  where αg = e2/(4π
vacvF ) denotes the effective fine structure constant in graphene and the integral kernel dy K(T, d, n) , (cid:90) ∞ (cid:90) ∞ α2 g 8 vF kBT (7) πn dx e2 0 0 K = vac F
2 k2 e4 U12(x, y)2 sinh2(cid:0)y TF (cid:1) x7 Φ2(x, y) x2 − y2 2T . (8) (cid:12)(cid:12)U (0) The function Φ, defined in Eqn. (B1), is related to the nonlinear susceptibility of graphene, and restricts the in- tegration range in the x, y-plane to the region ω < vF q. In order to estimate the contribution of phonon ex- change to the drag effect, we note that the drag resistiv- ity ρCD resulting from Coulomb interaction only (which is usually taken as a measure for Coulomb drag) is obtained by substituting the static value of the electron-electron interaction into the integral kernel Eqn. (8): 0 e2 (10) (cid:107) 0
⊥ ρCD = ρD λλ(cid:48) (q,ω=0) . ζ(3) π28α2 g √ CD = −  ρlarge d (kBT )2 (vF )2n3d4 . (9) For low temperatures EF (cid:29) kBT , the resistivity ρCD can CD ∝ T 2 of Eqn. (B2) (see Ref. 6 be approximated by ρlow T for a detailed derivation), under the additional condition kF d, kF d/
(cid:107) (cid:29) 1 (large layer spacing), this can be further (cid:1)3 (cid:0)
approximated to yield6 d → d(cid:112)
⊥/
(cid:107) and αg → αg/ (Note that in the static limit, one only needs to rescale
⊥
(cid:107) to take into account the anisotropy of hBN.) The full blue curves in Figures 3-5 show the absolute value of ρD Eqn. (7) for different parameters T , n, and d, while ρCD is shown by dashed red curves, ρlow T CD by the dashed black lines in Figures 3 and 5, and the dotted green line in Fig. 5 shows ρlarge d . As Figures 3 and 4 show, the contribution of phonon mediated interaction to the drag resistivity is vanish- ingly small at low temperatures, but becomes notice- able for T > 150 K, the effect being more pronounced the larger the layer separation. This temperature de- −2[yTF /(2T )] in the inte- pendence is due the factor sinh gration kernel Eqn. (8), which suppresses the integrand for values of y > T /TF . Thus at low temperatures, the main contribution to the y-integration in Eqn. (7) comes from a frequency range where the dielectric func- tions in the integrand are still close to their static val- ues. However, the phonon contribution becomes notice- able at lower temperatures than one would expect, tak- ing into account that the energy of the lowest phonon CD 4 FIG. 7. Effect of the anisotropy of hBN on the behavior of drag with temperature, with n = 0.02 nm−2 and d = 8 nm. The curves isotropy,
(cid:107) and isotropy,
⊥ were computed as- suming that the graphene layers are immersed in an isotropic dielectric medium, with dielectric functions given by
(cid:107) and
⊥, respectively (see Eqn. (2)). The curve anisotropy was computed taking into account the anisotropy of hBN as in Eqn. (4). Solid curves show ρD, dashed ones ρCD. (cid:107) TO/kB ≈ 1100 K. It is also interesting to notice mode ω that in the range from 100 to 250 K, the drag resistivity ρD, including the effect of phonons, is closer to the T 2 behaviour ρlow T than the purely Coulomb drag result, ρCD. The plot of K as a function of y in Fig. 6 shows the origin of the phonon contribution to the integral ρD: With rising temperature, peaks near the resonance fre- TO appear in the integrand, which quencies ω enhance the magnitude of ρD. (cid:107) TO and ω⊥ CD Fig. 4 shows that the relative effect of phonon exchange on ρD is larger for high densities. For high values of n, the argument of the dielectric functions
(ω) =
(yvF πn) in Eqn. (8) reaches the resonance frequency already at lower values of y. While ρCD decreases rapidly with n due to increased screening of the Coulomb interaction, the modification of the screening function
RPA by phonon interaction is seen to counteract this decrease at high temperatures. √ Finally, Fig. 7 illustrates the effect of the anisotropy in hBN that enters ρD through the electron-electron in- teraction Eqn. (4). We compare the drag resistivity in hBN with that in an isotropic medium with dielectric functions
(cid:107) and
⊥, respectively (see Eqn. (2)). The dif- ference in magnitude between ρD and ρCD is seen to be greatest in the anisotropic case, where both in-plane and out-of-plane phonon modes contribute to the interlayer interaction. IV. SUMMARY AND DISCUSSION We showed that including the electron -- electron inter- action via phonon exchange into the theory of Coulomb drag significantly changes the magnitude of the predicted drag resistivity in graphene-hBN heterostructures. For large layer separations, the deviations become noticeable at temperatures higher than 150 K. 501001502002503000.000.050.100.150.20T
K

ΡD

(cid:87)
isotropy,Ε(cid:166)isotropy,Ε(cid:254)anisotropyTABLE I. Parameters for the dielectric function of hBN (see Eqn. (3)) taken from Ref. 44a.
∞ f γ ωTO
⊥ 4.95 1.868 3.61 meV 170 meV
(cid:107) 4.10 0.532 0.995 meV 97.4 meV a The experimental data in Ref. 44 exhibits two resonances, a strong and a weaker one, for each direction of the polarization of incident light. The weaker ones are attributed to missorientation of the polycrystalline samples. As the lowest phonon resonance frequency in the spacer material hBN corresponds to a temperature of approxi- mately 1100 K, our result at first sight seems to be at odds with the notion that phonon effects should be pro- portional to the thermal population factor of the relevant modes. This is indeed the case for other transport phe- nomena, like the substrate limited electron mobility in graphene, where real momentum transfer from an elec- tronic state (in graphene) to a phonon mode (in a di- electric substrate material) plays a role24,38. The decay rate of the electronic state is then overall proportional to the thermal population of the phonon mode. Our sce- nario however involves the exchange of virtual phonons in a process that is of second order in the interlayer interaction5, and no decay processes into real phonon states are relevant for ρD. We note that in Ref. 7, the effect of substrate phonons on Coulomb drag was consid- ered for the case where a material described by a uniform dielectric function fills what is our region III of Fig. 1, and a deviation from the low-temperature T 2 behavior of ρD was predicted for temperatures roughly an order of mag- nitude lower than the phononic resonance frequency of the substrate material. Up to date, there remains considerable discrepancy between experimental data on Coulomb drag between 18,19 and hBN20 graphene layers embedded in SiO2/Al2O3 and the existing theoretical work. For hBN, the reported drag resistivities in the Fermi liquid regime are roughly a factor of three larger than predicted, and the results of the present paper do not change this situation. The ex- perimentally reported T 2 dependence of ρD for d = 6 nm and n = 0.018 nm−2 up to temperatures of 240 K48 does not disagree with our results presented in Fig. 3. Ac- tually, it appears that in the temperature range of 100 to 250 K the inclusion of phonon mediated interaction brings the behaviour of drag closer to the low temper- ature T 2 behaviour than with static Coulomb interac- tion only. Nevertheless, an extension of the experimental data shown in Ref. 20 up to room temperature would be needed to distinguish clearly between ρD and ρCD. The use of other substrate materials, such as SiO2, should not qualitatively alter the results of this paper. We think that future experiments with devices as considered in the 5 present work will be able to check our predictions. ACKNOWLEDGMENTS The authors would like to thank N.M.R. Peres for useful discussions. Financial support from Funda¸cao para a Ciencia e a Tecnologia (Portugal) through Grant No. SFRH/BD/78987/2011 (B.A.), the Marie Curie ITN NanoCTM (J.S.) and from MICINN (Spain) through Grant No. FIS2010-21372 (F.S.) and FIS2008-00124 (F.G.) is acknowleged. Appendix A: Frohlich electron -- phonon coupling and phonon mediated electron -- electron interaction Throughout the present work, we assume the dielec- tric properties of hBN layers forming heterostructures as shown in Fig. 1 to be the same as for bulk hBN. The Frohlich Hamiltonian describing the coupling of electrons to a bulk polar longitudinal phonon mode (in an isotropic homogeneous dielectric material) is given by35 -- 37 (cid:88) M (Q)eiQ·r(cid:16) (cid:17) aQ − a † −Q , He−ph = d3rρ(r) 1√ V Q (cid:90) † where ρ(r) denotes the electron density operator, a Q (aQ) the creation (annihilation) phonon operator with momentum Q = (qx, qy, qz) and the matrix element reads (cid:115) (cid:18) 1 (cid:19) M (Q) = i e2ωLO 2
vacQ2 − 1
0
∞ , (A1) with the longitudinal optical phonon frequency ωLO. The phonon mediated interaction between electrons is given by with the bare phonon propagator ψ(Q, ω) = M (Q)M (Q)∗DLO(Q, ω), DLO(Q, ω) = 2ωLO/(cid:0)ω2 − ω2 (cid:1) . LO (A2) (A3) We employ the usual approximation of dispersionless op- tical phonons29,33,35,43. The bare Coulomb interaction is given by VC = e2/(cid:0)
vac
∞Q2(cid:1) , (A4) where
∞ takes into account the high frequency screening properties of the medium. With Eqns. (A2)-(A4) and the Lyddane-Sachs-Teller TO =
0/
∞, we arrive at the combined relation46 ω2 Coulomb and phonon mediated interaction LO/ω2 U (Q, ω) = VC(Q) + ψ(Q, ω) = e2
vac
(ω)Q2 , (A5) with
(ω) the dielectric function of the medium, see Eqn. 3. Since the Frohlich coupling is derived in a phe- nomenological approach based on the dielectric proper- ties of the material, the combined Coulomb and phonon mediated interaction simply reduces to the Coulomb in- teraction screened by
(ω), as it should. where the dashed and wiggled lines denote the bare Coulomb and phonon interaction, respectively, and the full curves electron propagators (see Refs. 5 and 31). The function Φ(x, y) appearing in Eqn. (8) reads6,10 6 In a two-layer system as shown in Fig. 1, the Frohlich coupling coupling between bulk phonons and 2D elec- trons of layer λ is given by43 (cid:115) (cid:18) 1 (cid:19) Mλ(q, qz) = i e2ωLO 2
vac(q2 + q2 z )
∞ − 1
0 eiqzd(1−δλ1) , (A6) Φ(x, y) = Φ+(x, y) Θ(y − x + 2)Θ(x − y) + Φ−(x, y) Θ(1 − y − 1 − x) , (B1) where q = (qx, qy) is a two-dimensional momentum vec- tor. In analogy to the above, we now get for the combined Coulomb -- phonon interaction in a homogeneous isotropic medium where λλ(cid:48)(q, ω) ≡ Vλλ(cid:48)(q) + ψλλ(cid:48)(q, ω) U iso = 1 e2
(ω) 2
vacq e−qd(1−δλλ(cid:48) ) . (A7) Although it is possible to generalize the Frohlich electron -- phonon coupling for the case of anisotropic materials41,49 and inhomogeneous layered materials50, the easiest way to obtain the effective electron -- electron interaction, taking into account the phonon mediated in- teraction, is by solving Poisson's equation for the elec- tric potential created by a point charge in the dielectric medium taking into account the frequency dependence of its dielectric tensor. Appendix B: Mathematical details The dressed interlayer interaction Eqn. (5) is the solu- tion of the coupled set of Dyson equations U12(q, ω) = = + 2(cid:88) 2(cid:88) λ=1 λ=1 + + (cid:18) 2 ± x (cid:19) y ∓ 2 ± x y (cid:115)(cid:18) 2 ± x y (cid:19)2 − 1 . Φ± = ± cosh −1 For the low-temperature approximation of ρD, the fac- −2[yTF /(2T )] in the integration kernel Eqn. (8), tor sinh which suppresses the integrand for values of y > T /TF , allows one to expand the remaining integrand to the low- est order of y. The y integration can then be performed, yielding ρlow T e2 2πα2 eff (kBT )2 3n(vF )2 CD = −  (cid:2)(x + 4αeff )2 − 16α2 (cid:26) (cid:90) 2 (cid:113) eff exp(cid:0)−2dx dx x3(4 − x2) 0 e−2dx (cid:113) πn
⊥ 0 /
(cid:107) 0 (cid:1)(cid:3)2 πn
⊥ (cid:107) 0 /
0 (cid:27) (B2) , (cid:113) where αeff ≡ αg/
⊥ 0
(cid:107) 0 (see Ref. 6 for details). 1 T. J. Gramila, J. P. Eisenstein, A. H. MacDonald, L. N. Pfeiffer, and K. W. West, Phys. Rev. Lett. 66, 1216 (1991). 2 U. Sivan, P. M. Solomon, and H. Shtrikman, Phys. Rev. Lett. 68, 1196 (1992). 3 L. Zheng and A. H. MacDonald, Phys. Rev. B 48, 8203 (1993). Phys. Rev. B 52, 14761 (1995). 5 A. Kamenev and Y. Oreg, Phys. Rev. B 52, 7516 (1995). 6 B. Amorim and N. M. R. Peres, Journal of Physics: Con- densed Matter 24, 335602 (2012). 7 M. Carrega, T. Tudorovskiy, A. Principi, M. I. Katsnelson, and M. Polini, New Journal of Physics 14, 063033 (2012). 4 K. Flensberg, B. Y.-K. Hu, A.-P. Jauho, and J. M. Kinaret, 8 M. I. Katsnelson, Phys. Rev. B 84, 041407 (2011). 12q,ω12q12λλ12q,ω12λλ7 9 B. N. Narozhny, M. Titov, I. V. Gornyi, and P. M. Ostro- 31 C. Zhang and Y. Takahashi, Journal of Physics: Con- vsky, Phys. Rev. B 85, 195421 (2012). 10 N. M. R. Peres, J. M. B. L. dos Santos, and A. H. C. Neto, EPL (Europhysics Letters) 95, 18001 (2011). 11 E. H. Hwang, R. Sensarma, and S. Das Sarma, Phys. Rev. B 84, 245441 (2011). densed Matter 5, 5009 (1993). 32 T. J. Gramila, J. P. Eisenstein, A. H. MacDonald, L. N. Pfeiffer, and K. W. West, Phys. Rev. B 47, 12957 (1993). 33 K. Guven and B. Tanatar, Phys. Rev. B 56, 7535 (1997). 34 M. C. Bønsager, K. Flensberg, B. Yu-Kuang Hu, and A. H. 12 W.-K. Tse, B. Y.-K. Hu, and S. Das Sarma, Phys. Rev. B MacDonald, Phys. Rev. B 57, 7085 (1998). 76, 081401 (2007). 13 S. M. Badalyan and F. M. Peeters, ArXiv e-prints (2012), 35 H. Frohlich, Advances in Physics 3, 325 (1954). 36 G. D. Mahan, Many-particle physics (Plenum Press, New 1204.4598. York, 1981). 14 B. Scharf and A. Matos-Abiague, Phys. Rev. B 86, 115425 37 M. P. Marder, Condensed Matter Physics (John Wiley & (2012). 15 M. Schutt, P. M. Ostrovsky, M. Titov, I. V. Gornyi, B. N. Narozhny, and A. D. Mirlin, ArXiv e-prints (2012), 1205.5018. 16 J. C. W. Song and L. S. Levitov, ArXiv e-prints (2012), 1205.5257. 17 A. H. Castro Neto, F. Guinea, N. M. R. Peres, K. S. Novoselov, and A. K. Geim, Rev. Mod. Phys. 81, 109 (2009). 18 S. Kim, I. Jo, J. Nah, Z. Yao, S. K. Banerjee, and E. Tutuc, Phys. Rev. B 83, 161401 (2011). 19 S. Kim and E. Tutuc, Solid State Communications 152, 1283 (2012). 20 R. V. Gorbachev, A. K. Geim, M. I. Katsnelson, K. S. Novoselov, T. Tudorovskiy, I. V. Grigorieva, A. H. Mac- Donald, K. Watanabe, T. Taniguchi, and L. A. Pono- marenko, ArXiv e-prints (2012), 1206.6626. 21 A.-P. Jauho and H. Smith, Phys. Rev. B 47, 4420 (1993). 22 R. C. Dean, A. F. Young, I. Meric, C. Lee, L. Wang, S. Sor- genfrei, K. Watanabe, T. Taniguchi, P. Kim, K. L. Shep- ard, et al., Nat Nano 5, 722 (2010). 23 A. S. Mayorov, R. V. Gorbachev, S. V. Morozov, L. Brit- nell, R. Jalil, L. A. Ponomarenko, P. Blake, K. S. Novoselov, K. Watanabe, T. Taniguchi, et al., Nano Let- ters 11, 2396 (2011). 24 J. Schiefele, F. Sols, and F. Guinea, Phys. Rev. B 85, 195420 (2012). 25 J. M. Garcia, U. Wurstbauer, A. Levy, L. N. Pfeiffer, A. Pinczuk, A. S. Plaut, L. Wang, C. R. Dean, R. Buizza, A. V. D. Zande, et al., Solid State Communications 152, 975 (2012). 26 L. A. Ponomarenko, A. K. Geim, A. A. Zhukov, R. Jalil, S. V. Morozov, K. S. Novoselov, I. V. Grigorieva, E. H. Hill, V. V. Cheianov, V. I. Fal/'ko, et al., Nat Phys 7, 958 (2011). 27 L. Britnell, R. V. Gorbachev, R. Jalil, B. D. Belle, F. Schedin, A. Mishchenko, T. Georgiou, M. I. Katsnelson, L. Eaves, S. V. Morozov, et al., Science 335, 947 (2012). 28 L. Britnell, R. V. Gorbachev, R. Jalil, B. D. Belle, F. Schedin, M. I. Katsnelson, L. Eaves, S. V. Morozov, A. S. Mayorov, N. M. R. Peres, et al., Nano Letters 12, 1707 (2012). 29 R. Jalabert and S. Das Sarma, Phys. Rev. B 40, 9723 (1989). Sons, Inc., New Jersey, 2010), 2nd ed. 38 S. Fratini and F. Guinea, Phys. Rev. B 77, 195415 (2008). 39 J.-H. Chen, C. Jang, S. Xiao, M. Ishigami, and M. S. Fuhrer, Nature Nanotechnology 3, 206 (2008). 40 See Refs. 44, 51, and 52 for details on the phonon dis- persions of hBN, and the classification of the vibrational modes into Raman active, infrared active and optically silent. Figure 3 and eqns. (21) and (24) of Ref. 51 show how the long range Coulomb potential associated with the infrared active modes leads to the splitting of transverse and longitudinal optical frequencies at the Γ point. 41 R. Loudon, Advances in Physics 13, 423 (1964). 42 We are here using the retarded expression (defined as being analytic in the upper half of the complex ω plane) in order to be consistent with the likewise retarded polarizability of graphene taken from Ref. 53. Not keeping this consistency yields significantly different results. 43 S. Sarma and B. Mason, Annals of Physics 163, 78 (1985). 44 R. Geick, C. H. Perry, and G. Rupprecht, Phys. Rev. 146, 543 (1966). 45 In the numerical calculations, we use for simplicity the zero temperature expression for χ as calculated in Refs. 53 and 54, which is a good approximation for T (cid:28) TF , with TF the Fermi temperature. 46 R. H. Lyddane, R. G. Sachs, and E. Teller, Phys. Rev. 59, 673 (1941). 47 We here use a simplified form of the nonlinear susceptibility of graphene, which is valid for electron doping high enough such that the existence of the valence band can be ignored. The condition T (cid:28) TF is important as we use the zero temperature expressions for the polarizability of graphene. See Refs. 6 and 10 for a discussion of both approximations. 48 See Fig. 2a in Ref. 20. 49 M. A. Stroscio and M. Dutta, Phonons in Nanostructures (Cambridge University Press, Cambridge, 2003). 50 N. Mori and T. Ando, Phys. Rev. B 40, 6175 (1989). 51 K. H. Michel and B. Verberck, Phys. Rev. B 83, 115328 (2011). 52 J. Serrano, A. Bosak, R. Arenal, M. Krisch, K. Watanabe, T. Taniguchi, H. Kanda, A. Rubio, and L. Wirtz, Phys. Rev. Lett. 98, 095503 (2007). 53 B. Wunsch, T. Stauber, F. Sols, and F. Guinea, New Jour- nal of Physics 8, 318 (2006). 54 E. H. Hwang and S. Das Sarma, Phys. Rev. B 75, 205418 30 H. C. Tso, P. Vasilopoulos, and F. M. Peeters, Phys. Rev. (2007). Lett. 68, 2516 (1992). 

1305.6414

1

1305

Generation of hyper-entangled photon pairs in coupled microcavities

We propose and theoretically analyze a new scheme for generating hyper-entangled photon pairs in a system of polaritons in coupled planar microcavities. Starting from a microscopic model, we evaluate the relevant parametric scattering processes and numerically simulate the phonon-induced noise background under continuous-wave excitation. Our results show that, compared to other polariton entanglement proposals, our scheme enables the generation of photon pairs that are entangled in both path and polarization degrees of freedom, and simultaneously leads to a strong reduction of the photoluminesence noise background. This can significantly improve the fidelity of the entangled photon pairs under realistic experimental conditions.

cond-mat.mes-hall

cond-mat

Generation of hyper-entangled photon pairs in coupled microcavities S. Portolan1, L. Einkemmer2, Z. Voros2, G. Weihs2, P. Rabl1 1Institute of Atomic and Subatomic Physics, TU Wien, Stadionalle 2, 1020 Wien, Austria and 2Institut fur Experimentalphysik, Universitat Innsbruck, Technikerstr. 25, 6020 Innsbruck, Austria We propose and theoretically analyze a new scheme for generating hyper-entangled photon pairs in a system of polaritons in coupled planar microcavities. Starting from a microscopic model, we evaluate the relevant parametric scattering processes and numerically simulate the phonon- induced noise background under continuous-wave excitation. Our results show that, compared to other polariton entanglement proposals, our scheme enables the generation of photon pairs that are entangled in both path and polarization degrees of freedom, and simultaneously leads to a strong reduction of the photoluminesence noise background. This can significantly improve the fidelity of the entangled photon pairs under realistic experimental conditions. 3 1 0 2 y a M 8 2 ] l l a h - s e m . t a m - d n o c [ 1 v 4 1 4 6 . 5 0 3 1 : v i X r a Entanglement is considered the primary resource for quantum information processing schemes and in the op- tical domain the practicability of implementing large- scale photonic quantum computers or long distance quan- tum communication protocols relies crucially on efficient sources for entangled photon pairs (EPPs). Convention- ally, EPPs are produced by parametric down conversion in nonlinear crystals [1] or four-wave mixing in photonic crystal fibers [2]. Solid state systems like exciton po- laritons in microcavities [3] offer an intriguing alterna- tive with the prospect of building highly efficient EPP sources on a miniaturized scale [4, 5]. Being half exciton, half photon, polaritons benefit from strong Coulomb in- teractions, while they can easily be converted into propa- gating optical qubits for long-distance entanglement dis- tribution. Over the past years polariton-polariton in- teractions in microcavities have been the subject of in- tensive research [6 -- 10]. However, the predicted quan- tum properties of generated polariton pairs [11 -- 13] have remained quite elusive and no direct experimental evi- dence for entanglement detection has been demonstrated so far, mainly due to background noise caused by phonon- induced photoluminescence [14]. In this work we investigate the potential of coupled microcavity structures as an efficient platform for semi- conductor quantum technologies and apply this concept to the design of bright sources of entangled and hyper- entangled photon pairs. In particular, we consider a sys- tem of three planar microcavities [8], which are coupled via two shared Bragg mirrors as illustrated in Fig. 1 a). In this setup the splitting of the upper and lower polariton dispersion curves into three well-separated sub-branches provides an additional flexibility for engineering paramet- ric inter-branch scattering processes [12], which leads to qualitative new features and an improved performance of the EPP creation. By pumping two polariton modes in the first and third branch, the phase matching condi- tion for parametric scattering is fulfilled on an energeti- cally degenerate circle of momentum states in the middle branch as shown in Fig. 1 c). By choosing different pump- ing configurations, photon pairs with polarization entan- glement or pairs with entanglement in both the path and polarization degrees of freedom can be generated. Such hyper-entangled photons [15 -- 17] provide a valuable re- source for super-dense coding protocols [18, 19] or quan- tum key distribution [20]. Moreover, the energy separa- tion between the different branches creates a bottleneck for phonon-induced polariton scattering and under re- alistic conditions suppresses the background photolumi- nesce by several orders of magnitude. Our analysis of the resulting entanglement fidelity shows that the proposed multi-cavity setting offers a competitive option for future solid state based EPP sources and integrated quantum technologies. Model. -- In the setup shown in Fig. 1 a) the photons in each cavity are coupled to electronic interband excita- tions (excitons) of a semiconductor quantum well (QW) and form new quasi-particles -- so-called polaritons -- un- der strong coupling conditions. We assume that both FIG. 1: (Color online). a) Sketch of the three coupled cavi- ties, which are separated by two distributed Bragg reflectors (DBR) and in each cavity spacer a quantum well (QW) is inserted [8]. b) Energy dispersion curves of the 6 polariton branches (solid lines); the dashed lines show the energies of the bare exciton and the lowest photonic mode. c) Actual lower polariton branches for the sample under consideration. (see text). The red circles show the chosen pump wavevec- tors (kp = −k(cid:48) p,kp = 1.8 µm−1) which, by mixed paramet- ric scattering, generate a circle of energy-degenerate hyper- entangled signal-idler pairs (dashed blue line). a)b)c)DBRcavityQWphotons and excitons are restricted to a single mode func- tion along the confined direction (z-axis) and we denote by Ec nk) the energy dispersion of photons (excitons) with transverse momentum k and cavity index n. The Hamiltonian for the whole system is nk (Ex H = † nkank + Ex nka Ec nk † nk B Bnk + Vn † a nk Bnk + H.c. p where k = (i, k) labels the polariton branch and the 2D wavevector. Pkp and Pk(cid:48) denote the classical amplitudes of the pumped modes and the interaction strength g de- pends on all the details of the four involved states [26]. Eq. (3) describes the physical process of two coherent pump polaritons being scattered into a signal-idler po- lariton pair, which satisfies the phase matching condi- tions (cid:17) (cid:16) (cid:17) (cid:26) kp + k(cid:48) p = ks + ki Ekp + Ek(cid:48) p = Eks + Eki. (4) (cid:88) −(cid:88) n,k n,k (cid:16) 2 Jn,n+1 † a nkan+1k + H.c. + HC, (1) (cid:88) Bnk and B † where ank and a nk are bosonic annihilation and creation operators for photons in cavity n and transverse momen- † tum k. nk are the corresponding operators for excitons, which can be treated as effective bosons un- der low excitation conditions [21]. The so-called Rabi splitting Vn is the dipole coupling strength between pho- tons and excitons within cavity n and Jn,n+1 is the opti- cal mode coupling between neighboring cavities. Finally, HC accounts for the Coulomb interaction between exci- tons, which will give rise to the effective χ(3) polariton- polariton non-linearity discussed below. In the following we write H = H0 + Hint, where apart from the Coulomb term, Hint accounts for additional effective interactions due to the non-bosonic character of the excitons [9, 22]. The linear part, H0, can be diagonalized by a generalized Hopfield transformation Pik [24], and expressed in terms of a H0 = (cid:80) ik Eik P † ik set of polaritonic quasi-particle operators Pik = X n ik Bnk + C n ikank, (2) n ik and C n where X n ik are the exciton and photon compo- nents of the i-th polariton branch. For a single cavity the index i denotes the familiar lower and upper polari- ton branches, which, for zero detuning, are split by ∼ 2V around k = 0 [25]. In coupled structures, the optical mode coupling induces a further splitting of ∼ J. Typi- cal dispersion curves for three coupled cavities are shown in Fig. 1 b), where identical Rabi splittings V = 6 meV and mirror couplings J = 4 meV have been assumed. Photon pair creation. -- The nonlinear optical response of our system is governed by the strong Coulomb inter- action between excitons, Hint ∼ B† B† B B, which results in an equivalent χ(3)-type non-linearity between polari- tons. In the following we are focusing on the lower set of branches i = 1, 2, 3, which have a smaller exciton-exciton dephasing [10, 13]. We assume that two polariton modes with wavevectors kp and k(cid:48) p and energies Ekp and Ek(cid:48) , p respectively, are strongly pumped by external lasers and we linearize the interaction around the classical mean value of these two pumped modes. We obtain a para- metric scattering process analogous to four-wave-mixing in nonlinear optical crystals, (cid:17) (cid:88) (cid:16) ks,ki The resulting shape of the available states depends on the energy dispersion curves and on the positions of the pump beams. In single planar cavity setups with one [14] or two pumps [13, 27] the available phase-space reduces to curves where at most two of the final states can have the same energy. Multi-mode settings provide a much larger flexibility [12]. For the present planar device and the pump configuration shown in Fig. 1 c), the phase- matching conditions (4) are fulfilled on a whole circle of energy-degenerate states in the middle polariton branch. This will allow simultaneously the generation of hyper- entangled states and the reduction of the detrimental phonon-induced noise background as shown below. Polarization and path entanglement. -- For pump fields with a definite circular polarization (σ = ±1), this po- larization will be inherited by the polariton modes and due to spin-preserving Coulomb interactions only one of the four polarization configurations +, +(cid:105), −,−(cid:105), +,−(cid:105) or −, +(cid:105) is created. For two linearly co-polarized pump beams all the four polarization states are activated and for the generation of polarization entanglement the only useful configuration occurs for linearly cross-polarized pump fields [13]. In this case the counter-circular chan- nel (due to bound biexciton and two-exciton scattering states of opposite spin) is suppressed owing to destruc- tive interference [21]. As a result, the generated pho- ton pairs are produced in an entangled state of the form (+, +(cid:105) + −,−(cid:105))/ √ 2. By generalizing Eq. (3) to account for polarization se- lection rules and restricted to cross-polarized driving, the photon pair creation process is described by an effective Hamiltonian Heff = G † ks+ † P ki+ + P † ks− P † ki− + H.c., (5) kski p and ki = kp + k(cid:48) where G = gPkpPk(cid:48) p − ks is assumed. For the configuration shown in Fig. 1 c) ki = ks and by selecting specific paths (say k1 and k2, −k1 and −k2) the outgoing photon pair is generated in a hyper-entangled state ψ(cid:105) = (cid:16)k1,−k1(cid:105)+k2,−k2(cid:105)(cid:17)⊗(cid:16)+, +(cid:105)+−,−(cid:105)(cid:17) , (6) 1 2 (cid:88) (cid:16) P (cid:17) Hχ(3) = gPkpPk(cid:48) p P † ks † P ki + H.c. δks+ki,kp+k(cid:48) p , (3) which exhibits entanglement in both the momentum and the polarization degrees of freedom. Being degenerate in energy the photons in modes k1 and k2 can be interfered, and the momentum can be used as an independent degree of freedom in photonic entanglement experiments [16]. Photoluminescence. -- In polariton systems the fidelity of the EPPs is affected by phonon-induced scattering pro- cesses and Rayleigh scattering from the pump beams. The effect of Rayleigh scattering is strongly suppressed as soon as the pump and the signal-idler photons are non energy-degenerate. On the contrary, at temperatures of a few Kelvin, polaritons can scatter incoherently by emis- sion or absorption of acoustic phonons, and redistribute along the dispersion curve. This pump-induced photolu- minescence (PL) noise background competes with para- metric coherent photoemission and lowers the degree of non-classical correlations [27, 28]. The multi-cavity set- ting can reduce its impact by using the mode splitting J to create a large energy separation between the pump and the signal-idler photons. To quantify the amount of entanglement in the present coupled cavity setup, we study the competition between parametric coherent scattering and the incoherent PL background. Starting from Eq. (5) , the dynamics of the signal and idler mode can be evaluated within a two- mode description [29] (cid:18) (cid:19) (cid:18) ∂t Pks = −i † ∂t P ki = i ωki + i ωks − i Γks 2 (cid:19) Γki 2 + Fks, † ki Pks − iG P † + iG∗ Pks + F† P ki ki (7) , (8) where the background PL is treated separately from the parametric interaction and enters through the time- dependent Langevin noise operators Fks and Fki. Under continuous driving conditions the effect of noise is fully determined by the stationary correlators (cid:104) F† k(t) Fk(t(cid:48))(cid:105) = k(t(cid:48))(cid:105) = Γk(N PL k + 1)δ(t− t(cid:48)), ΓkN PL where the total polariton decay rates Γk and the station- ary occupations N PL k δ(t− t(cid:48)) and (cid:104) Fk(t) F† are evaluated in the following. The dominant incoherent processes for optically gen- erated polaritons in III-V systems are (acoustic) phonon- induced scattering and radiative losses. In strainfree het- erostructures, the exciton-phonon interaction is well de- scribed by a 3D bulk-like model, k H DF exc-ph = Ξ(q, qz) B † n,k+q Bn;k eiqzλnbq,qz + H.c. k,q,qz,n (9) where bq,qz is the bosonic operator for phonons with transverse momentum q and momentum qz along the confinement axis. In Eq. (9) λn = n zqw accounts for the position of the wells numbered as n = −1, 0, 1 and the coefficient Ξ(q, qz) contains the exciton overlap inte- grals in terms of the Fourier transform over the phonon wavevectors [29, 30]. Using Fermi's golden rule, the linear PL dynamics for k , for each σ, can be described mi- the populations N PL croscopically by a Boltzmann equation (cid:88) (cid:16) (cid:17) ∂tN PL k = Ik − ΓkN PL k + W (ph) k,k(cid:48) N PL k(cid:48) , (10) (cid:88) k(cid:48) 3 FIG. 2: Stationary polariton populations in a logarithmic scale and superposed on the respective energy dispersion curve. In this line cut of the 2D k-space, the two coherent pump modes are resonant at the white circles, while the blue cross refer to the intersection with the circle of degenerate signal/idler pairs of Fig. 1 c). Phonon-induced scattering is quasi-elastic and favors quasi-resonant processes. k k + γ(rad) where the total linewidth Γk = Γ(ph) includes phonon-induced and radiative losses. Under continuous wave (CW) excitation with two lasers resonant with Ekp and Ek(cid:48) , we model the pump term Ik as two Gaus- sian profiles centered around kp and k(cid:48) p [21]. The total phonon-induced polariton scattering rate from state k to state k(cid:48) is W (ph) k(cid:48),k, where the rates from phonon emission (+) and absorption (−) are given by k(cid:48),k + W − k(cid:48),k = W + p k(cid:48) − k2 + (q0 z )2 Ξ(k − k(cid:48), qz)2 W ± k(cid:48),k = 1 ρuS (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:88) p × uq0 z (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)2(cid:18) eiq0 z λ(p)X p k(cid:48)X p k nB(Eph q ) + 1 2 ± 1 2 (cid:19) (11) . , q Here nB(E) is the Bose distribution, ρ is the density and u is the sound velocity of the material and S is the quan- tization surface. In Eq. (11) Eph is the phonon energy and the 3D phonon wave vector is (cid:126)q = (q = k − k(cid:48), q0 z ), where q0 z is calculated from the condition of energy con- servation, ωk(cid:48) − ωk ± Eph q = 0. The expression for Ξ(k − k(cid:48), qz) is explicitly derived in [21]. The sum over all final scattering states gives the total phonon-induced k(cid:48) W (ph) loss rate Γ(ph) k(cid:48),k . The radiative linewidth is γ(rad) c , and in the following we take a typical cavity loss rate of γn = (cid:80) = (cid:80) k2γn n X n c = 0.35 meV. Figure 2 shows the stationary polariton occupations N PL derived from Eq. (10) and for other parameters specified above and in [14]. For simplicity, but with- out loss of generality, a radial symmetry is assumed [21] k k k 4 s plot the signal to noise ratio (SN) defined as the total generated signal Ns over the background PL N P L at the detected signal and idler wavevectors. The red line shows the results obtained from the steady state solution of Eqs. (7) and (8) and the PL noise level evaluated with Eq. (10). The black line indicates the corresponding re- sults for a single-cavity setup [13], but with otherwise identical parameters. We see that for a single cavity, the emitted photons are dominated by the PL background, whereas in the three-cavity setting the coherent signal is clearly above the PL noise level and dominated by the parametric (pair) emission. Pi Ps(cid:105)/((cid:104) P † Ps(cid:105)(cid:104) P s † i s † P i sσ(t1) P † iσ(cid:48)(t2) Pi¯σ(t2) Ps¯σ(cid:48)(t1)(cid:105). Under the same conditions we evaluate in Fig. 3 b) and c) the equal time two-photon correlation function g(2)(0) = (cid:104) P † Pi(cid:105)) and the entan- glement of formation (EOF) [32] as two measures of signal-idler correlations and of the degree of polariza- tion entanglement [33, 34] between two simultaneously emitted photons, respectively. The EOF is evaluated from the reconstructed two-polariton subsector of the full density operator [35, 36], which can be obtained ex- perimentally from measurements of all the four-operator expectation values (cid:104) P † In contrast to the total signal, the quantities plotted in Fig. 3 b) and c) are sensitive to detected photon pairs only and consistent with previous findings [13] we ob- serve bunching, g(2)(0) > 1, and polarization entangle- ment EOF > 0 for both configurations. However, the bunching of g(2)(0) ∼ 4 achievable with the single-cavity set-up is at the border of the acquisition capabilities of current experiments [33], while the triple cavity struc- ture shows a ten-fold enhanced effect, g(2)(0) > 30, in relation with a very high SN ratio. Moreover, there is a pronounced difference in the fidelity of the entangled photons pairs, and the EOF in the three-cavity setting can be close to the value of EOF = 1 expected for a pure polarization entangled Bell state. Conclusions. -- In this work we showed that coupled microcavity structures can become an innovative design for protecting quantum coherences from solid-state back- ground noise and we applied this concept to devise inno- vative bright sources of entangled and hyper-entangled photon pairs. Our analysis predicts a suppression of the phonon-induced noise of more than 3 orders of magni- tude, which thereby eliminates one of the main noise sources in current experiments. Acknowledgments S.P. thanks S. Savasta and F. Rossi for many stim- ulating discussions. This work was supported by the EU project SIQS and the Austrian Science Fund (FWF) through SFB FOQUS and the START grant Y 591-N16. FIG. 3: (color online) Single (black) - triple (red) cavity com- parison in steady-state. a) Signal to noise ratio, defined as total generated signal Ns over background PL N P L at k = 1.5 µm−1. b) Equal-spin autocorrelation at zero delay g(2)(0). c) Theoretically achievable EOF. In the simulation, a s = 10−5 has been pump independent background noise of N 0 added to account for other noise sources in the system and the reference pump intensity I0 corresponds to an injected polariton density of about 0.5 µm−2 [14]. s and the population distribution is plotted as a function of k. This is consistent with radial-symmetric steady-state populations observed in experiments [14, 31]. Phonon- induced polariton scattering favors quasi-elastic events exchanging small energies [30]. If more branches are present the intra- and inter-branch scattering rates will be of the same order and the polariton population will spread across the branches as well. The pumped polari- ton modes in the first and third branch scatter domi- nantly with low energy exchange into the second branch, which is separated from the other branches by more than k (cid:29) Γ(ph) 2 meV. Since under the relevant conditions γ(rad) multi-phonon processes are highly suppressed, this cre- ates a PL window for the signal and idler wave vectors around k = (1.5, 0) µm−1, where the PL population is reduced by more than 3 orders of magnitude with respect to the intra-branch PL on the third branch. k Entanglement quantification. -- To provide a direct comparison between the present multi-mode setup and entanglement generation schemes using only a single mi- crocavity, we evaluate in Fig. 3 three different quantities that are of importance for the experimental verification of entanglement in these systems. In Fig. 3 a) we first b)a)c)5 (cid:17) 1kp1 3kp2 X p + 2VxxX p + X p , with the saturation density nsat and the Coulomb-induced con- tribution including its correction beyond mean-field Vxx. [27] S. Savasta, O. Di Stefano, V. Savona, and W. Langbein, X p X p X p 2kp1 2kp2 2ks 2ki Phys. Rev. Lett. 94, 246401 (2005). [28] J. W. Pan, C. Simon, Caslav Brukner, and A. Zeilinger, Nature 410, 1067-1070 (2001). [29] S. Portolan, O. Di Stefano, S. Savasta, F. Rossi, and R. Girlanda, Phys. Rev. B 77, 035433 (2008). [30] F. Tassone, C. Piermarocchi, V. Savona, A. Quattropani, and P. Schwendimann, Phys. Rev. B 56, 7554 (1997). [31] S. Savasta, O. Di Stefano, and S. Portolan, Physica Sta- tus Solidi (c) 5, 334, (2008). [32] S. Hill, and W. K. Wootters Phys. Rev. Lett. 78, 5022 (1997); W. K. Wootters Phys. Rev. Lett. 80, 2245 (1998). [33] K. Edamatsu, G. Oohata, R. Shimizu, and T. Itoh, Na- ture 431, 167-170 (2004); G. Oohata, E. Shimizu, and K. Edamatsu, Phs. Rev. Lett. 98, 140503 (2007). [34] A. Dousse, J. Suffczy´nski, A. Beveratos, O. Krebs, A. Lamaıtre, I. Sagnes, J. Bloch, P. Voisin, and P. Senellart, Nature 466, 217 (2010). [35] A. G. White, D. F. V. James, P. H. Eberhard, and P. G. Kwiat Phys. Rev. Lett. 83, 3103 (1999); D. F. V. James, P. G. Kwiat, W. J. Munro, and A. G. White Phys. Rev. A 64, 052312 (2001). [36] S, Portolan, S. Savasta "Quantum Optics with interact- ing polaritons", in Optical Generation and Control of Quantum Coherence in Semiconductor Nanostructures edited by G. Slavcheva and P. Roussignol (Springer 2010). [1] L. Mandel, Rev. Mod. Phys. 71, S274 (1999). [2] O. Cohen, J. S. Lundeen, B. J. Smith, G. Puentes, P. J. Mosley, and I. A. Walmsley, Phys. Rev. Lett. 102, 123603 (2009). [3] C. Piermarocchi, A. Quattropani, P. Schwendimann, F. Tassone and V. Savona, Phase Transitions 6 8, 169-279 ¯ (1999); A. Kavokin, & G. Malpuech, Cavity Polaritons. Elsevier, Amsterdam (2003); G. Khitrova, H. M. Gibbs, F. Jahnke, M. Kira, and S. W. Koch, Rev. Mod. Phys. 71, 1591 (1999). [4] P. G. Savvidis, J. J. Baumberg, R. M. Stevenson, M. S. Skolnick, D. M. Whittaker, and J. S. Roberts, Phys. Rev. Lett. 84, 1547 (2000). [5] M. Saba, C. Ciuti, J. Bloch, V. Thierry-Mieg, R. Andre, Le Si Dang, S. Kundermann, A. Mura, G. Bongiovanni, J. L. Staehli and B. Deveaud, Nature 414, 731 (2001). [6] R. M. Stevenson, V. N. Astratov, M. S. Skolnick, D. M. Whittaker, M. Emam-Ismail, A. I. Tartakovskii, P. G. Savvidis, J. J. Baumberg, and J. S. Roberts Phys. Rev. Lett. 85, 3680 (2000). [7] J. Erland, V. Mizeikis, W. Langbein, J. R. Jensen, and J. M. Hvam Phys. Rev. Lett. 86, 5791 (2001). [8] C. Diederichs, J. Tignon, G. Dasbach, C. Ciuti, A. Lemaıtre, J. Bloch, P. Roussignol and C. Delalande, Na- ture 440, 904 (2006). [9] C. Ciuti, P. Schwendimann, A. Quattropani Semicond. Sci. Technol 18, S279 (2003). [10] S. Savasta, O. Di Stefano, and R. Girlanda Phys. Rev. B 64, 073306 (2001); Phys. Rev. Lett. 90, 096403 (2003). [11] S. Savasta, G. Martino, R. Girlanda, Solid State Com- mun. 111, 495 (1999). [12] C. Ciuti. Phys. Rev. B, 69, 245304 (2004). [13] S. Portolan, O. Di Stefano, S. Savasta, and V. Savona, Europhys. Lett. 88, 20003 (2009). [14] W. Langbein, Phys. Rev. B 70, 205301(2004). [15] P. G. Kwiat, J. Mod. Opt. 44, 2173 (1997). [16] M. Barbieri, C. Cinelli, P. Mataloni, and F. De Martini, Phys. Rev. A 72, 052110 (2005). [17] J. T. Barreiro, N. K. Langford, N. A. Peters, and P. G. Kwiat, Phys. Rev. Lett. 95, 260501 (2005). [18] S. P. Walborn, S. P´adua, and C. H. Monken, Phys. Rev. A 68, 042313 (2003). [19] J. T. Barreiro, T.-C. Wei, and P. G. Kwiat, Nat. Phys. 4, 282 (2008). [20] C. Wang, Li Xiao, W. Wang, G. Zhang, G. L. Long, J. Opt. Soc. Am. B 26, 2072 (2009); H. Xu, K. Du, C. Qiao, J. Mod. Opt. 59, 611 (2012). [21] See Supplemental Material at [URL] for details. [22] S. Portolan, O. Di Stefano, S. Savasta, F. Rossi, and R. Girlanda, Phys. Rev. B 77, 195305 (2008). [23] L. Einkemmer, Z. Voros, G. Weihs, and S. Portolan, arXiv:1305.1469 [cond-mat.str-el]. [24] J.J. Hopfield, Phys. Rev. 112, 1555 (1958); F. Bassani, F. Ruggero, A. Quattropani, Il Nuovo Cimento 7, 700 (1986). [25] V. Savona et al. Solid State Comm. 93, 733 (1995); S. Pau et al. Phys. Rev. B 51, 14437 (1995). [26] In particular, for signal-idler wavevector k, ki, the pump- mixed process in Fig. 1 has a nonlinear coefficient of the form g = 1 2 p (V /nsat) X p X p X p X p 2ks 3kp1 1kp2 2ki (cid:80) (cid:16) Generation of hyper-entangled photon pairs in coupled microcavities Supplementary Material for: S. Portolan1, L. Einkemmer2, Z. Voros2, G. Weihs2, P. Rabl1 1 Institute of Atomic and Subatomic Physics, TU Wien, Stadionalle 2, 1020 Wien, Austria 2 Institut fur Experimentalphysik, Universitat Innsbruck, Technikerstr. 25, 6020 Innsbruck, Austria 6 In this supplemental material we provide a more detailed derivation of the different results presented in the main part of the paper. To be consistent with our notation, we start in Sec. I with a brief review of the standard exciton-polariton model. In Sec. II we detail the derivation of the effective χ(3) nonlinear Hamiltonian given in Eq. (3) in the main text, taking into account both the Coulomb interaction between excitons as well as the effective nonlinearity arising from the non-bosonic nature of the excitons. In Sec. III we discuss the generation of polarization entangled photons by using cross-polarized pumping beams and in Sec. IV we present the explicit form of the acoustic phonon interaction Hamiltonian for the three-cavity setup. Finally, Sec. V contains additional details on the numerical simulation of the photoluminescence signal presented in Fig. 2 of the main text. I. MODEL Polaritons are mixed quasiparticles resulting from the strong coupling between light and electronic excitations (excitons) in semiconductor crystals. For the quasi-2D interacting electron system we adopt the usual semiconductor model Hamiltonian [1]. We shall consider III-V direct-gap materials, for which a two-band semiconductor model well reproduces the optical response near the band edge. It reads: He = H eh 0 + HC . (12) The first term describes the single-particle Hamiltonian terms for electrons in conduction band and holes in valence band H eh 0 = † Ec,kc kck + Eh,k † d k dk, (13) (cid:88) (cid:88) k k † † k ( d where c k) creates an electron (hole) in the conduction (valence) band with a 2D quasi momentum k. The second term in Eq. (12) is the Coulomb interaction, HC = (cid:88) (cid:88) (cid:88) −(cid:88) q(cid:54)=0 k,k(cid:48) 1 2 q(cid:54)=0 k,k(cid:48) (cid:88) (cid:88) q(cid:54)=0 k,k(cid:48) † † k(cid:48)−qck(cid:48) ck + k+qc Vqc 1 2 † Vqc k+q † d k(cid:48)−q dk(cid:48)ck , † d k+q † d k(cid:48)−q dk(cid:48) dk − Vq (14) where the three contributions represent the repulsive electron-electron (e-e) and hole-hole (h-h) terms and the attrac- tive (e-h) interaction, respectively. Hamiltonian (12) can be rewritten as He = H eh 0 + HC = (cid:88) EN α EN α(cid:105)(cid:104)EN α , (15) N α where the eigenstates of He, with energies EN α = ωN α, have been labeled according to the number N of eh pairs. The state EN =0(cid:105) is the electronic ground state, the N = 1 subspace is the one exciton subspace with the collective quantum number α denoting the exciton energy level ν, the in-plane wave vector k and the spin index σ. Throughout this work we are interested in studying polaritonic effects, where the optical response involves mainly excitons belonging to the 1S band with wave vectors close to normal incidence, k (cid:28) π , ax being the exciton Bohr radius, typically around 10 nm. In the following we will omit the internal quantum number ν. ax Eigenstates of the model Hamiltonian with N = 1 (called excitons) can be created from the ground state by applying the exciton creation operator, (cid:12)(cid:12)N = 1, σ, k(cid:11) = B (cid:12)(cid:12)N = 0(cid:11) , † σ,k (cid:88) k(cid:48) † σ,k = B † σ,k(cid:48) c σ,k(cid:48)+ηek Φk † d σ,−k(cid:48)+ηhk , (16) 7 σk(cid:48) is the 1S exciton wave function of total and relative wave vectors k and k(cid:48), respectively. The coordinate where Φk transformation from elecron-hole to center-of-mass and relative momenta read k = ke − kh, k(cid:48) = ηhke + ηekh [2, 22], where η(e,h) = m(e,h)/(m(e,h) + m(h,e)) and (ke, kh) are the electron and hole wave vectors.) For the multicavity setup with ncav coupled microcavities, we label excitons in different quantum well by an additional index n = 1, . . . , ncav, i.e. B † n,σ,k and whenever needed we adopt the following shorthand notation k ≡ (n, σ, k). † σ,k → B Restricted to a single transverse cavity mode, the free Hamiltonian for photons in the coupled cavity array is given ncav(cid:88) (cid:88) n=1 k nkank − ncav−1(cid:88) † ωnka (cid:88) n=1 k Hc = Jn,n+1 (cid:16) (cid:17) † a nkan+1k + H.c. . (17) n)2 + (c/nref )2k2 ≈ ω0 n +(c/nref )2k2/2 [4] is the quadratic dispersion of the individual cavities with refractive index nref and Jn,n+1 is the mode coupling between neighboring cavities, which depends on the transmission of the Bragg mirrors. The discrete modes of the outermost cavities are coupled to the external continuum of modes, which can be described within a quasi-mode theory [3] by an effective Hamiltonian by Here ωnk =(cid:112)(ω0 Hp = i (cid:88) (cid:88) tj( E(−) † k − E(+) j,k a j,k ak) , (18) where tj determines the fraction of the transmitted field amplitude at mirror j, and E(−) (negative) frequency part of the external field. j,k ( E(+) j,k ) is the positive j=1,ncav k The dipole coupling of the electron system to the cavity modes is given within the usual rotating wave approximation by HI = (cid:88) † Vnka nk Bnk + H.c. , (19) where Vnk is the photon-exciton coupling coefficient enhanced by the presence of the cavity [4]. Vnk is proportional to the overlap between the exciton wave function Φk nσk(cid:48) and the transverse optical modefunction in cavity n, which can be optimized by an appropriate design of the structure. nk II. POLARITON χ(3)-NONLINEARITY Many descriptions of polariton parametric processes make use of the picture of polaritons as interacting bosons [5 -- † 8]. In these models bosonic commutation relations are assigned to the exciton operators Bnk and B nk and an effective boson-boson coupling is added to the Hamiltonian to account for the non-bosonic exciton character ("phase-space filling"). Combined with the Coulomb interaction this coupling is then treated within the mean-field approximation. At low polariton densities this approach provides a good description of the resulting nonlinear third order response and is often used due to its simplicity and its direct analogy to nonlinear optics. However, in particular at higher densities the absolute strength of the nonlinear terms differs from those obtained from more rigorous microscopic theories [3, 9]. For all the result presented in the main part of this paper these microscopic corrections -- as described in the following -- have been taken into account. A. Heisenberg equations of motion In the case of a single cavity and a single pump field, the Heisenberg equations of motion for a polariton system up to the third-order nonlinearity have been fully worked out in Ref. [3, 10]. Taking these references as our starting point, we adapt here this analysis for the case of multiple coupled cavities. For concreteness we address directly the triple cavity case (ncav = 3), but the following line of argument is completely general. We assume that the first cavity is driven by two coherent pump fields, which resonantly excite the polariton modes with wave vectors kp and kp(cid:48), respectively. Following the same notation as in the paper, we introduce a vector Bk with components Bk = ( B1k, a1k, B2k, a2k, B3k, a3k)T and write the resulting equations of motion in a compact form as Bk = −iΩxc k Bk + Ein k − iRN L k . (20) 8 (21)   , −J2,3 V3 ωc 2k Here the matrix ωx 1k V1 k ≡ Ωxc V1 ωc 1k ωx −J1,2 V2 2k −J1,2 V2 ωc 2k ωx −J2,3 V3 2k described the linear dynamics of the coupled modes and the vector Ein p e−iωptδk,kp + Ein Ein 2k , 0, RN L pump configuration specified above(cid:0)Ein k = ( RN L (cid:16) (cid:1) k accounts for the external driving field. For the p(cid:48) e−iωp(cid:48) tδk,kp(cid:48) i = 0 otherwise. The last term in Eq. (20), where RN L 3k , 0)T , describes additional nonlinear contributions, which we address in the following. Note that in Eq. (20) we have for simplicity omitted the photon loss terms, which can be included by adding photon loss rates for the cavity operators ank and the corresponding Langevin noise operators. 2 = t1 1k , 0, RN L k k Coulomb and photon-exciton interactions will be effective only between exciton and photon mode belonging to the same cavity. As a consequence the relevant nonlinear source terms, which couple waves with different in-plane wave vector k, can be treated very similarly to the single cavity case. We write the nonlinear terms as RN L nk). The first term originates from the phase-space filling of the exciton transition and couples excitons with the same spin, nk = ( Rsat nk + Rxx (cid:17) and(cid:0)Ein (cid:1) (cid:88) k(cid:48)k(cid:48)(cid:48) Rsat nk = V nsat Bnk(cid:48) ank(cid:48)(cid:48) B † n¯k , (22) x) is the exciton saturation density [3, 12] and ¯k = k(cid:48) + k(cid:48)(cid:48) − k. The second term comes from where nsat = 7/(16πa2 the Coulomb interaction among electrons and holes and contains two contributions. In the co-circular channel, only particles with the same spin are involved and the resulting interaction is always repulsive. In contrary, the counter- circular term describes the scattering of excitons of opposite spin and it can include a bound biexciton intermediate state. As a consequence its strength and its sign can vary considerably around the biexciton binding energy [11]. Altogether, and using the same notation as in Ref. [11], we obtain Rxx nk(t) = dt(cid:48) T ++(t − t(cid:48)) Bnσk(cid:48)(t(cid:48)) Bnσk(cid:48)(cid:48)(t(cid:48)) (cid:90) t −∞ (cid:88) − (cid:88) k(cid:48),k(cid:48)(cid:48) B † nσ¯k(t) (cid:88) B † n−σ(cid:48) ¯k(t) k(cid:48),k(cid:48)(cid:48) σ(cid:48)=± (cid:90) t −∞ dt(cid:48) T +−(t − t(cid:48)) Bnσ(cid:48)k(cid:48)(t(cid:48)) Bn−σ(cid:48)k(cid:48)(cid:48)(t(cid:48)) . (23) The transition T-matrix includes the instantaneous mean-field exciton-exciton interaction contribution and a non- instantaneous term originating from four-particle correlations [3, 11, 12]. The memory-less equal-time limit of Eq. (23) corresponds to an energy independent χ(3) nonlinearity for excitons, which is usually assumed in effective mod- els, where excitons are treated as interacting bosons. As we show below, within our microscopic theory and in a quasi stationary regime the resulting χ(3)-interaction between polaritons can be derived more rigorously by using a Weisskopf-Wigner approximation for the integral kernel. The transition T-matrix includes the instantaneous mean- field exciton-exciton interaction contribution and a non-instantaneous term originating from four-particle correlations. The same notation as in Ref. [11] has been used. B. Strong coupling and polariton-polariton interactions When the coupling rate V exceeds the decay rate of the exciton coherence and of the cavity field, the system enters the strong coupling regime. In this regime, cavity-polaritons arise as the two-dimensional eigenstates of Ωxc k . In order to obtain the dynamics for the polariton system we diagonalise the linear subproblem of Eq. (20) by a (unitary) Hopfield transformation [13] Pk = UkBk, where (Pk)i = Pik, and express excitons and photons in terms of a set of polaritonic quasi-particle operators Pik = X n ik Bnk + C n ikank, (24) where X n By applying this transformation to Eq. (20) we obtain ik and C n ik can be seen as the exciton and photon components of the n-th cavity on the i-th polariton branch. ∂tPk,σ = −iωkPk,σ + Ein k,σ − i RN L k,σ , (25) n (cid:88) 9 (26) (27) where T s(Ω) =(cid:82) ∞ where the ωk are the polariton eigenfrequencies, RN L σk = UkRN L σk and Ein σk = UkEin σk . Let us consider a situation where the energies of the exciting pulses are all close to the corresponding polariton resonance values ωik and the broadenings are small compared to the splitting between the polariton branches. By adopting a Weisskopf-Wigner approximation [12] we can simplify the memory integral in Eq. (23) in the case of continuous wave (CW) excitation and express the result in terms of polariton operators Pik. The nonlinear interaction terms we obtain are given by ( RN L ( Rsat (cid:16) (cid:17) with nσk)i + ( Rxx nσk)i n=1 † P mσ¯k(t) σk )i =(cid:80)3 (cid:20) mσ¯k(t)(cid:2)X n (cid:88) m,−σ(cid:48) ¯k(t)(cid:2)X n ikX n m¯k ikX n X n P P † † (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) k(cid:48),k(cid:48)(cid:48) l,m,r k(cid:48),k(cid:48)(cid:48) l,m,r + k(cid:48),k(cid:48)(cid:48) σ(cid:48)=± l,m,r ( Rsat nσk)i = ( Rxx nσk)i = (cid:21) V nsat X n lk(cid:48)C n rk(cid:48)(cid:48) m¯kT ++(ωk(cid:48) + ωk(cid:48)(cid:48) )X n lk(cid:48)X n ikX n m¯kT +−(ωk(cid:48) + ωk(cid:48)(cid:48) )X n lk(cid:48)X n rk(cid:48)(cid:48)(cid:3) Plσk(cid:48)(t) Prσk(cid:48)(cid:48)(t) + rk(cid:48)(cid:48)(cid:3) Plσ(cid:48)k(cid:48)(t) Pr,−σ(cid:48)k(cid:48)(cid:48) (t), −∞ T s(τ )e−iΩτ dτ is the Fourier transform of the time-dependent kernel in Eq. (23) for s = ++, +−. This approach, which is valid under quasi-stationary conditions, fully accounts for the energy dependence of the exciton-exciton scattering by including in the above equations the frequency dependence of the T -matrix. It can be shown [12] that the co-circular channel can be written as the sum of a constant mean-field term Vxx plus a genuine four-particle contribution, T ++(ω) = Vxx + F (ω). In the present work we are interested CW driving and in the scattering of polaritons with a pre-specified energy. In this case the frequency dependence of T ++(ω) simply leads to a renormalization of the coherent scattering amplitude. In order to maintain a connection with the literature, in the following we will use loosely Vxx as the Coulomb-induced co-circular interaction including its correction beyond mean-field. In the last step we now assume that the two pumped polariton modes are strongly driven and we replace the corresponding operators by their classical expectations values, Pkp → Pkp = (cid:104) Pkp(cid:105) and Pkp(cid:48) → Pkp(cid:48) = (cid:104) Pkp(cid:48)(cid:105), where again the convention kp = (ip, kp, σp) is assumed. As discussed below, we are mainly interested in cross-polarized pump beams to get rid of the spin non-conserving scattering channel by destructive interference. However, for the moment we will leave the polarization of the pump fields unspecified. Then, by retaining only the most relevant contribution ∼ PkpPkp(cid:48) in the expressions given in Eqs. (26) and (27) the equation of motion for the generic polariton operator reads Piσk = −iωk Piσk − i d dt P † m¯σ¯k gσk PipσpkpPip(cid:48) σp(cid:48) kp(cid:48) , (28) (cid:16) (cid:88) σp,σp(cid:48) ,¯σ where gσk is a short notation for the coupling strength, which depends on all the σ's and k's in Eq. (28) and is detailed below. This equation of motion can also be obtained in the Heisenberg picture by the effective Hamiltonian Hχ(3) (Eq. (3) in the paper). The branch index m and wave vector ¯k are defined by the wave vector and energy conservation rules, (cid:26) kp + kp(cid:48) = k + ¯k Eip,kp(cid:48) + Eip(cid:48) ,kp(cid:48) = Ei,k + Em,¯k. (cid:17) (cid:17) (cid:20) V (cid:16) nsat m¯k2 T +−X n (cid:32) (cid:32) ncav(cid:88) ncav(cid:88) n=1 g++ σk = g+− σk = The nonlinear coupling strength can be written as the sum of the co-circular and counter-circular terms, gσk = σk + g+− g++ σk , and it includes all the spin selection rules. They read δ¯σ,σδσp,σp(cid:48) δσp,σX n ikX n m¯k X n ipkp ip(cid:48) kp(cid:48) + X n C n ip(cid:48) kp(cid:48) C n ipkp (cid:33) + 2 T ++X n ipkp X n ip(cid:48) kp(cid:48) δ¯σ,−σδσp,−σp(cid:48) X n ikX n X n ip(cid:48) kp(cid:48) . (30) ipkp n=1 III. PERPENDICULAR POLARIZATION: SPIN-CONSERVING CHANNELS As the calculation of Ref. [11] shows, the +- kernel in the four-particle nonlinear response does not become negligible even in a very negatively detuned region, but it retains a value of approximately one third of the ++ channel. However, (29) (cid:21)(cid:33) with two linearly cross-polarized pumps, we can show that the counter-circular channel (due to bound biexciton and two-exciton scattering states of opposite spin) is suppressed owing to destructive interference. We choose two pump fields with polarization vectors ep = θ and ep(cid:48) = θ + π/2 (e.g. θ = 0 means the first pump polarized along x, while the second one along y). From the equation for the pumped polariton mode Pσplpkp = −i(ωkp − iΓp/2)Pipσpkp + tcC n ipkp E(+) kp (ep · σ) , d dt and ep = cos θx(cid:105) + sin θy(cid:105) = 2−1/2(ei θ−(cid:105) + e−i θ+(cid:105)), we can see that any linearly polarized pump excites two circularly polarized polariton modes, 10 P (+) 1 (t) = Pkp (t) e−iθ√ 2 P (−) 1 (t) = Pkp (t) e+iθ√ 2 , , P (+) 2 P (−) 2 By working out explicitly the spin sum in Eq. (28) we obtain Piσk = −iωk Piσk − igσk P d dt † m¯σ¯k (cid:34) = −iωk Piσk − i P † m¯σ¯k g++ σk e−iπ/2, e+iπ/2. 2 2 (t) = Pkp(cid:48) (t) e−iθ√ (t) = Pkp(cid:48) (t) e+iθ√ (cid:1) 2 + P (−) (t)δσ,+ + P (−) (cid:1)(cid:0)P (+) 2 2 1 (31) (cid:17) + (cid:35) (t)P (−) 2 (t)δσ,− (cid:17) (t)P (+) (t) + P (+) 1 (t)P (−) 2 (t) δσ,−¯σ . (32) 1 1 (t)P (+) 1 + P (−) (cid:0)P (+) (cid:16)P (+) (cid:16)P (−) (cid:17) (t) + iPkp (t) (t)P (−) e−iθ√ g+− σk 2 1 2 2 The counter-circular term cancels out by destructive interference (cid:16)P (−) (cid:18) 1 2 (t)P (+) −iPkp (t) 1 (t) + P (+) e+iθ√ Pkp(cid:48) (t) 2 δσ,−¯σ = e−iθ√ 2 (cid:19) Pkp(cid:48) (t) e+iθ√ 2 = 0 , (33) = while the co-circular term reads 1 (t)P (+) = P (+) = e−iπ/2Pkp (t)Pkp(cid:48) (t) (t)δσ,+ + P (−) e−i2θ 2 2 1 2 (t)δσ,− = (t)P (−) δσ,+ + eiπ/2Pkp (t)Pkp(cid:48) (t) e−iπ/2(cid:104)Pkp (t)Pkp(cid:48) (t)δσ,+ − Pkp (t)Pkp(cid:48) (t)e+i4θδσ,− e+i2θ 2 (cid:105) e−i2θ 2 = . δσ,− = (34) Choosing the first pump such that θ = π/4 and neglecting an overall phase, Eq. (32) becomes (35) with G = gPkp (t)Pkp(cid:48) (t), and g = gσk/2. This equation of motion can be obtained by the effective Hamiltonian Eq. (5) of the paper and ideally generates a pure entangled triplet state of the form Piσk = −iωk Piσk − iG P † m¯σ¯k (δσ,+ + δσ,−) d dt Ψ(cid:105) = 1√ 2 (+, +(cid:105) + −,−(cid:105)) . (36) IV. ACOUSTIC PHONON INTERACTION Long-wavelength acoustic phonon interaction Hamiltonian is obtained within the deformation potential coupling method. In general, moving from bulk systems to heterostructures, the partial loss of spatial symmetry leads to modifications not only in the electronic degrees of freedom but in the phonon subsystem as well. However, detailed analysis of phonons in strainless 2D systems in III-V materials [14 -- 16] showed that exciton-phonon decoherence can be well described when assuming phonons as bulk-like. The three-dimensional electron- (hole-) phonon interaction Hamiltonian for the deformation potential coupling has the form (cid:19)1/2(cid:16) (cid:18) q (cid:88) 2ρuV (cid:126)k(cid:126)q Hint = † Dcc (cid:126)k+(cid:126)q c(cid:126)k + Dv † d (cid:126)k+(cid:126)q d(cid:126)k (cid:17)(cid:16)b(cid:126)q + b † −(cid:126)q (cid:17) , (37) 11 where V is the quantization volume, u is the sound velocity, ρ is the material density, (cid:126)k is the 3D electron (hole) momentum, while Dc and Dv are deformation potential constants for the conduction and valence band, respectively. † The operators b(cid:126)q and b (cid:126)q are the bosonic annihilation and creation operators for phonons of 3D momentum (cid:126)q, with modulus q. We want to derive the phonon-interaction Hamiltonian for quasi-two-dimensional 1S excitons in the case of GaAs/AlGaAs QW structures. This can be accomplished by projecting the full Hamiltonian in the exciton sec- tor [17]. In the case of a single QW, when considering only the lowest subbands in the confined direction, the exciton wave function depends only on the two-dimensional wave vector K related to the center-of-mass motion and reads (cid:88) 1S; K(cid:105) = = a3 0√ S (cid:88) (cid:126)k,(cid:126)k(cid:48) † e−iK·RW (ρ)χe(ze)χh(zh)c c,(cid:126)re cc,(cid:126)rh 0(cid:105) (cid:126)re,(cid:126)rh † f (K; (cid:126)k, (cid:126)k(cid:48))δk−k(cid:48),Kc (cid:126)k † −(cid:126)k(cid:48) 0(cid:105). d (38) The notation takes into account explicitly the separation between confined and free directions. Here a "two- dimensional" exciton is written in terms of a three-dimensional envelope convolution (in direct or reciprocal space), (cid:126)k = (k, kz), (cid:126)r = (x, z), ρ = xe − xh, R = (mexe + mhxh)/M , V = SL, where L is the confined direction quantization length and S is the quantization surface in the free directions, capital letters are for the wave vectors of the center- of-mass of motion while the small k's refer to the in-plane component of the three-dimensional wave vectors. The envelope function in reciprocal space reads (αe = me/M, αh = mh/M, M = me + mh) f (K; (cid:126)k, (cid:126)k(cid:48)) = 1 L2 d2ρ dze dzhW (ρ)χe(ze)χh(zh)e−i(αeK−k)·ρe−ikzze eik(cid:48) zzh . The reduction to the 1S exciton sector is obtained by an explicit projection: H DF exc-ph = (cid:104)1S; K(cid:48) Hint 1S; K(cid:105) 1S; K(cid:105)(cid:104)1S; K(cid:48) . (39) (40) (cid:90) (cid:90) (cid:90) (cid:88) KK(cid:48) The procedure is fully reported in [17]. The final results in the 1S exciton sector reads (we dropped the capital letter for brevity) DcI⊥ e (qz)I(cid:107) e (q) + DvI⊥ h (qz)I (cid:107) h(q) (cid:17) k + q(cid:105)(cid:104)k (cid:16)bq,qz + b (cid:17) † −q,−qz . (41) (cid:88) (cid:88) (cid:18) q kq qz 2ρuV (cid:19)1/2(cid:16) H DF exc-ph = All the wave vectors in this Hamiltonian are now related to the center of mass of the exciton. The two overlap integrals are given by (cid:90) (cid:90) I⊥ e(h)(qz) = (cid:107) e(h)(q) = I dzχe(h)(z)2eiqzz, d2ρW (ρ)2eiαh(e)q·ρ = (cid:18) 1 + (cid:16) mh(e) 2M (cid:17)2(cid:19)−3/2 . qax (42) (43) − ρ e ax . (cid:113) 2 πax The second integral in the plane is analytic owing to the form of the 1S exciton wavefucntion, W (ρ) = A. Three quantum wells According to Eq. (38), the generic field operators for a 1S exciton can be expanded as (Φ(1S) (cid:88) (cid:126)k,(cid:126)k(cid:48) Ψ†((cid:126)r) = Φ(1S) k † ((cid:126)r)c (cid:126)ke † d −(cid:126)kh , k ((cid:126)r) = (cid:104)(cid:126)r 1S, k(cid:105)) (44) where Φk(x) is a shorthand notation for the f above. We are interested in the case of three quantum wells. We follow the spirit of Eq. (44) and write (Φ(1S) i,k ((cid:126)r) = (cid:104)(cid:126)r 1S; i, k(cid:105)) (cid:88) Ψ†((cid:126)r) = n=1,2,3,(cid:126)ke,(cid:126)kh Φ(1S) † n,k ((cid:126)r)c n,(cid:126)ke † d n,−(cid:126)kh . (45) 12 (47) (48) The non-diagonal matrix elements in Eq. (40) involve states belonging to different, space-separated, wells and the phonon interaction Hamiltonian will be zero because the overlap integrals over direct space equivalent to Eq. (42) are zero. Taking as the origin in z the well number 2, the matrix element (cid:104)2; k(cid:48) Hint 2; k(cid:105), (46) has already been calculated as in the single well case, whereas the corresponding matrix element for QW 1 and 3 are basically identical except for the overlap integral in the z direction which has to be evaluated over the corresponding well (centered around ±zqw and of the same width as the other one). Eventually, by introducing (cid:18) q (cid:19)1/2(cid:16) Ξ(q, qz) = 2ρuV DcI⊥ e (qz)I(cid:107) e (q) + DvI⊥ h (qz)I (cid:107) h(q) (cid:17) , the Hamiltonian for the triple cavity structure becomes (cid:88) Ξ(q, qz) n; k + q(cid:105)(cid:104)n; k (cid:16) † eiqzλnbq,qz + e−iqzλnb −q,−qz (cid:17) , H DF exc-ph = k,q,qz,n=−1,0,1 where here we have numbered the wells as n = −1, 0, 1, and λn = nzqw. From this Hamiltonian we calculate the scattering rates Wk,k(cid:48) for the Boltzmann equation as outlined in Ref [10]. The numerical values for the deformation potential constants are taken from [18]. V. NUMERICAL SIMULATION OF THE BOLTZMANN EQUATION Using Fermi's golden rule, the linear PL dynamics for the populations N PL can be described microscopically by a Boltzmann equation ∂tN PL ik = −ΓikN PL ik + Iik + W (ph) ik,lk(cid:48)N PL lk(cid:48) , (49) (cid:88) lk(cid:48) where the total linewidth Γik = Γ(ph) of spin, and the scattering rates W (ph) specific k we take a Gaussian pulse centered around a wave vector kp, ik + γ(rad) includes phonon-induced and radiative losses, which are independent ik,lk(cid:48) are defined in the main text. For simulating the external driving field at a ik Iik = I 0 2D e − k−kp2 2σ2 2πσ2 (cid:12)(cid:12)coh = −ΓikN PL ik (cid:12)(cid:12)coh + Iik. ∂tN PL ik The resulting coherent part of the population is then given by (cid:12)(cid:12)coh. In general, the background photoluminescence can be calculated numerically by solving Eq. (49) and Eq. (51) In the paper we are mainly interested in the (incoherent) background photoluminescence defined as N PL N PL ik on a 2D grid. ik = N PL (51) ik − . (50) A. Equivalent symmetric problem In experimental and numerical studies it is found that the incoherent steady-state polariton population N PL ik has radial symmetry even under excitation with a specific wave vector [8, 19]. This property can be seen as peculiar of the quasi-elastic nature of the phonon-induced scattering W (ph) in this system, which is able to redistribute in a very short time the peaked coherent population in a symmetric pattern and then the steady-state incoherent population depends only on the total flux of injected particles balanced by loss and no more on the specific form of N PL ik earlier times. (cid:12)(cid:12)coh at 13 In our numerical studies this allows us to approximate the full 2D problem by an equivalent one with radial symmetry from the outset. The population calculated in this way should well reproduce the full 2D incoherent population once the steady-state is reached. Moreover, the symmetry of this equivalent system reduces the computation to the sole radial distribution for the population N PL ik , where k = k. To do so we consider a radially symmetric pump ¯Iik = I 0 e − (k−kp)2 2σ2 2πσ2 , (52) which is chosen such that the flux of injected particles into the system is the same as in the case of a single pumped wave vector, We change to polar coordinates,(cid:88) (cid:39) S (2π)2 kx,ky (cid:90) kx,ky dkxdky = S (2π)2 (cid:90) kdkdθ (cid:39)(cid:88) k,θ (cid:18) S (2π)2 ∆k∆θ (cid:19) , (53) (54) (cid:88) (cid:88) kx,ky Iik = ¯Iik. and by making use of the fact that the distribution of populations is radially symmetric, N PL manipulate the last term in Eq. (49) to obtain ikθ ≡ N PL ik , we can ∂tN PL ik = −ΓikN PL ik + ¯Iik + W(ik),(lk(cid:48))N PL lk(cid:48) , (55) (cid:88) lk(cid:48) where we introduced Wk,k(cid:48) =(cid:80) equation and plot the stationary values of the incoherent part N PL ik . θ(cid:48) W(kθ),(k(cid:48)θ(cid:48)). In Fig. (2) of the main text we solve the stationary solution of this [1] V. M. Axt and T. Kuhn, Rep. Progr. Phys. 67, 433 (2004). [2] Th. Ostreich and K. Schonhammer, and L. J. Sham, Phys. Rev. B 58, 12920 (1998). [3] S. Portolan, O. Di Stefano, S. Savasta, F. Rossi, and R. Girlanda, Phys. Rev. B 77, 195305 (2008). [4] V. Savona, L. C. Andreani, P. Schwendimann, A. Quattropani, Solid State Communication, 93, 733 (1995). [5] C. Ciuti, P. Schwendimann, and A. Quattropani, Phys. Rev. B 63, 041303 (2001). [6] C. Ciuti, P. Schwendimann, B. Deveaud, A. Quattropani, Phys. Rev. B 62, R4825 (2000). [7] S. Kundermann, M. Saba, C. Ciuti, T. Guillet, U. Oesterle, J. L. Staehli, and B. Deveaud, Phys. Rev. Lett. 91, 107402 (2003). [8] W. Langbein, Phys. Rev. B 70, 205301 (2004). [9] L. Einkemmer, Z. Voros, G. Weihs, and S. Portolan, arXiv:1305.1469 [cond-mat.str-el]. [10] S. Portolan, O. Di Stefano, S. Savasta, F. Rossi, and R. Girlanda, Phys. Rev. B 77, 035433 (2008). [11] S. Schumacher, N. H. Kwong, and R. Binder Phys. Rev. B 76, 245324 (2007); [12] S. Savasta, O. Di Stefano, R. Girlanda Semicond. Sci. Technlo. 18, S294 (2003). [13] J.J. Hopfield, Phys. Rev. 112, 1555 (1958); F. Bassani, F. Ruggero, A. Quattropani, Il Nuovo Cimento 7, 700 (1986). [14] F. Rossi, T. Meier, P. Thomas, S.W. Koch, P.E. Selbmann, and E. Molinari, Phzs. Rev. B 51, 16943 (1995). [15] L. Rota, F. Rossi, P. Lugli, E. Molinari Phys. Rev. B 52, 5183 (1995). [16] H. Rucker, E. Molinari, and P. Lugli, Phys. Rev. B 4 ¯ [17] T. Takagahara, Phys. Rev. B, 31, 6552 (1985). [18] C. Piermarocchi, F. Tassone, V. Savona, and A. Quattropani, P. Schwendimann Phys. Rev. B 53, 15834 (1996). [19] S. Savasta, O. Di Stefano, and S. Portolan, Physica Status Solidi (c) 5, 334, (2008). 5, 6747 (1992). 

1309.4433

3

1309

Folded MoS2 layers with reduced interlayer coupling

We study molybdenum disulfide (MoS2) structures generated by folding single- and bilayer MoS2 flakes. We find that this modified layer stacking leads to a decrease in the interlayer coupling and an enhancement of the photoluminescence emission yield. We additionally find that folded single-layer MoS2 structures show a contribution to photoluminescence spectra of both neutral and charged excitons, which is a characteristic feature of single-layer MoS2 that has not been observed in multilayer MoS2. The results presented here open the door to fabrication of multilayered MoS2 samples with high optical absorption while maintaining the advantageous enhanced photoluminescence emission of single-layer MoS2 by controllably twisting the MoS2 layers.

cond-mat.mes-hall

cond-mat

This is the post-peer reviewed version of the following article: A.Castellanos-Gomez et al. “Folded MoS2 layers with reduced interlayer coupling”. Nano Research, 2014, 7(4): 572–578 DOI 10.1007/s12274-014-0425-z Published in its final form at: http://link.springer.com/article/10.1007%2Fs12274-014-0425-z Folded MoS2 layers with reduced interlayer coupling Andres Castellanos-Gomez*, Herre S. J. van der Zant, and Gary A. Steele Kavli Institute of Nanoscience, Delft University of Technology, 2628 CJ Delft, The Netherlands. [email protected] We study molybdenum disulfide (MoS2) structures generated by folding single- and bilayer MoS2 flakes. We find that this modified layer stacking leads to a decrease in the interlayer coupling and an enhancement of the photoluminescence emission yield. We additionally find that folded single-layer MoS2 structures show a contribution to photoluminescence spectra of both neutral and charged excitons, which is a characteristic feature of single-layer MoS2 that has not been observed in multilayer MoS2. The results presented here open the door to fabrication of multilayered MoS2 samples with high optical absorption while maintaining the advantageous enhanced photoluminescence emission of single-layer MoS2 by controllably twisting the MoS2 layers. Introduction Single-layer molybdenum disulfide (MoS2) has emerged as a prospective complementary material to graphene because, unlike graphene, it has a large intrinsic direct bandgap [1-3]. This makes monolayer MoS2 an attractive alternative for applications such as: logic circuits [4-6], photodetectors [7-14], light emitters [15-17] and solar cells [18, 19]. Recent studies on the optoelectronic properties of single-layer MoS2 have also shown interesting phenomena including the photogeneration of charged excitons [20] and valley-selective circular dichroism [21-23]. Nevertheless, the case of multilayer MoS2 is different as it is an indirect bandgap semiconductor [24-28], making it less attractive for optoelectronic applications. The direct bandgap nature of single-layer MoS2 is very fragile as the interlayer coupling between just two MoS2 layers is strong enough to turn the bilayer MoS2 into an indirect bandgap semiconductor [24, 25]. This thickness dependent direct-to-indirect bandgap transition has strong consequences in the photoluminescence emission of MoS2: while single-layer MoS2 presents a large photoluminescence yield, the multilayer MoS2 photoluminescence emission is quenched [24, 25].  This is the post-peer reviewed version of the following article: A.Castellanos-Gomez et al. “Folded MoS2 layers with reduced interlayer coupling”. Nano Research, 2014, 7(4): 572–578 DOI 10.1007/s12274-014-0425-z Published in its final form at: http://link.springer.com/article/10.1007%2Fs12274-014-0425-z The aim of this work is to generate artificial MoS2 layered structures with modified interlayer coupling and bandstructure. In this communication we study MoS2 structures where the layers are twisted leading to a decrease in the interlayer coupling and thus an enhancement of the photoluminescence emission yield. Furthermore, we observe that in MoS2 structures formed by two or three twisted MoS2 monolayers, both neutral and charged excitons contribute to the photoluminescence. This feature, typically observed in single layer MoS2, has not been observed in multilayer MoS2 with perfect stacking. Our results demonstrate the potential of twisted MoS2 layers to fabricate optically active materials (similar to single layer MoS2) with high optical absorption (overcoming the main limitation of monolayer MoS2, the low absorption due to its reduced thickness). Experimental Sample fabrication Twisted MoS2 structures can be fabricated by folding MoS2 layers as they present excellent mechanical properties that allows one to deform the MoS2 layers in extreme ways without breaking them [29-31]. In order to generate folds in the MoS2 layers we transfer MoS2 flakes by mechanical exfoliation onto an elastic substrate (Gelfilm® by Gelpak) previously stretched by about 100%. Subsequently the strain is suddenly released thereby generating wrinkles in the MoS2 layers through buckling-induced delamination [32, 33]. Interestingly, for single- and bilayer MoS2 these wrinkles are not stable due to their reduced bending rigidity and they tend to collapse forming bifolds (see the sketch in Figure 1a and the optical image in Figure 1c). We also found that the edges of single-layer flakes tend to fold after the sample fabrication process (see Figure 1a and Figure 1d). We consider this approach to induce cleaner folded structures than other techniques previously used to generate folded graphene such as water flushing the fabricated samples to induce the folds [34, 35]. In that case the water flushing may lead to adsorbated layers between the folded layers, hampering the interpretation of the data. Moreover these techniques, that rely on partially folding the edge of the flakes, can only fabricate simply folded structures. Our technique also offers the possibility of fabricating bifolded structures by collapse of the wrinkles induced during the fabrication process. In the folded regions, the MoS2 layers are not stacked following the normal AB stacking, i.e., the MoS2 layers are twisted (see Figure 1b and the Electronic Supporting Material). Using this fabrication process, we have fabricated folded single layers (hereafter denoted by 1L+1L), bifolded single-layers (1L+1L+1L) and bifolded bilayers (2L+2L+2L). For our experiments, we use selected folds and bifolds wider than 2 µm to ensure that the laser spot (400 nm in diameter) only probes the twisted regions, and also avoid laser spot positions close to the edges of the folds where the mechanical strain can alter the optoelectronic properties of MoS2 [32, 36-42].  This is the post-peer reviewed version of the following article: A.Castellanos-Gomez et al. “Folded MoS2 layers with reduced interlayer coupling”. Nano Research, 2014, 7(4): 572–578 DOI 10.1007/s12274-014-0425-z Published in its final form at: http://link.springer.com/article/10.1007%2Fs12274-014-0425-z Figure 1 (a) Sketches of the MoS2 structures fabricated by folding twice an MoS2 layer (bifold) and by folding one edge (fold). (b) Cartoon illustrating how the folding process typically results in a MoS2 structure composed of twisted layers that are not perfectly AB stacked. (c), (d) Optical microscopy images of a bifolded MoS2 bilayer and a folded single-layer MoS2 respectively. Optical microscopy MoS2 sheets have been identified under an optical microscope (Olympus BX 51 equipped with a Canon EOS 600D camera) [43]. As the employed elastic substrates are transparent, one can employ transmission mode optical microscopy to determine the number of MoS2 layers. In a previous work we found that the transmittance (T) of MoS2 flakes up to ten layers thick depends on the number of layers (N) as T = 100 – 5.5N (see the Electronic Supplementary Material of Ref. [32]). We found that the transmittance measured for folded (1L+1L) and bifolded (1L+1L+1L and 2L+2L+2L) samples is in good agreement with the transmittance measured for pristine (perfectly AB stacked) 2L, 3L and 6L respectively (see Figure S4 in the ESM). Raman spectroscopy and photoluminescence A Renishaw in via system was employed to perform simultaneous micro-Raman spectroscopy and photoluminescence measurements on the fabricated twisted MoS2 structures. The system was used in a backscattering configuration excited with visible laser light (λ = 514 nm). The Raman and photoluminescence spectra were simultaneously collected through a 100× objective (NA = 0.95) and recorded with 1800 lines/mm grating providing the spectral resolution of ~ 1 cm-1. To avoid laser-induced modification or ablation of the samples, all spectra were recorded at P ~ 500 µW which is found to be low enough power to prevent laser-induced damage [44-46]. Note that  This is the post-peer reviewed version of the following article: A.Castellanos-Gomez et al. “Folded MoS2 layers with reduced interlayer coupling”. Nano Research, 2014, 7(4): 572–578 DOI 10.1007/s12274-014-0425-z Published in its final form at: http://link.springer.com/article/10.1007%2Fs12274-014-0425-z the use of Gelfilm® substrate provides an enhanced photoluminescence signal with respect to conventionally used SiO2/Si substrates [47] which is advantageous for photoluminescence studies. Moreover, the photoluminescence spectra of samples fabricated onto Gelfilm® presents reduced doping level in comparison to samples prepared on SiO2. This reduced doping level yields photoluminescence spectra very similar to that of MoS2 samples on boron nitride or freely suspended (See Figure S2 of the ESM). Results and discussion Raman spectroscopy has proven to be a powerful tool to characterize atomically thin materials [48] and it has been previously used to characterize atomically thin MoS2 layers [49]. In the Raman spectrum of MoS2 there are two prominent peaks corresponding to the E1 2g and the A1g vibrational modes. In the E1 2g mode, the Mo and S atoms oscillate in anti-phase and parallel to the surface plane (see scheme in the inset in Figure 2a). In the A1g mode, the S atoms oscillate in anti-phase perpendicularly to the surface plane while the Mo atoms remain static. It has been shown that the frequency difference between these two Raman modes increases monotonically with the number of layers [49, 50]. For single-layer MoS2, which is not subjected to any interlayer interaction, the frequency difference between the 2g and the A1g peaks is around 19 cm-1. For perfectly AB stacked bilayer MoS2 this value is about 21 cm-1. For bulk E1 MoS2, the layers feel the interaction of many other MoS2 layers yielding a frequency difference of 25.5 cm-1. Therefore, the frequency difference between the E1 2g and the A1g peaks can also be used to estimate the strength of the interlayer coupling. Figure 2 Raman spectra measured for folded single-layer (1L+1L, in red) MoS2 layers. The Raman spectra measured for pristine 1L and 2Lhave been included for comparison. The frequency difference between the E1 modes is indicated with dashed horizontal lines. The thin light lines are Lorentzian fits to the experimental data. 2g and A1g Raman In Figure 2, the Raman spectrum obtained for folded monolayer MoS2 (1L+1L) is compared with results acquired for pristine single- and bilayer MoS2 because they can be considered as two limiting cases: a system without interlayer coupling and one with the interlayer coupling due to the perfect AB stacking of the two MoS2 layers. The folded  This is the post-peer reviewed version of the following article: A.Castellanos-Gomez et al. “Folded MoS2 layers with reduced interlayer coupling”. Nano Research, 2014, 7(4): 572–578 DOI 10.1007/s12274-014-0425-z Published in its final form at: http://link.springer.com/article/10.1007%2Fs12274-014-0425-z single-layer MoS2 shows a frequency difference between the E1 2g and the A1g that lies in between the one measured for 1L and 2L MoS2. This indicates that the 1L+1L layer has a reduced interlayer coupling with respect to the perfectly stacked 2L MoS2. We also systematically observed blueshift of the E1 2g that may be due to a change in the strain of the layer[51, 52]. Similar results have been found for bifolded single-layer (1L+1L+1L) and bifolded bilayer (2L+2L+2L) MoS2 (see Electronic Supplementary Material). For all the studied cases (1L+1L, 1L+1L+1L and 2L+2L+2L), the frequency difference between the Raman modes lies between the limiting cases. This indicates that the twisted MoS2 layers exhibit a reduced interlayer coupling in comparison to the case of perfectly stacked layers. The reduction of the interlayer coupling, however, is not enough to consider them as completely independent layers. Figure 3 summarizes the measured frequency difference between the E1 2g and the A1g Raman modes for pristine layers (perfect stacking) and for 1L+1L, 1L+1L+1L and 2L+2L+2L MoS 2 layers. The frequency difference is systematically lower for the folded layers than for their perfectly stacked counterparts. Nonetheless, we have found dispersion on the values measured for different folded and bifolded layers which we attribute to difference in the twisting angle which may lead to different interlayer coupling. Figure 3 Frequency difference between the E1 Raman modes as a function of the number of layers 2g and A1g measured for: pristine layers (gray circles), folded and bifolded single-layer (1L+1L and 1L+1L+1L, dark and light red squares respectively) and bifolded bilayer MoS2 (2L+2L+2L, orange squares). Photoluminescence measurements have been used to study the bandstructure of atomically thin MoS2 [24-26, 53]. The photoluminescence spectrum of MoS2 presents prominent peaks around 630 nm and 670 nm which correspond to the B and A excitons. For single-layer MoS2, the A exciton peak is composed of two peaks corresponding to the contribution of both charged (labeled as A-, occurring at ~ 655 nm) and neutral excitons (labeled as A, occurring at ~ 670 nm) to the photoluminescence emission. On the other hand, multilayered MoS2 does not show a contribution due to charged excitons yielding a single A exciton peak. Furthermore, multilayer MoS2 presents an extra peak at lower  This is the post-peer reviewed version of the following article: A.Castellanos-Gomez et al. “Folded MoS2 layers with reduced interlayer coupling”. Nano Research, 2014, 7(4): 572–578 DOI 10.1007/s12274-014-0425-z Published in its final form at: http://link.springer.com/article/10.1007%2Fs12274-014-0425-z energy (between 790 nm and 1000 nm, depending on the number of layers) corresponding to the indirect bandgap transition (labeled as I). Figure 4 Photoluminescence spectra measured for single-layer (1L), folded single-layer (1L+1L) and bilayer (2L) MoS2. The peaks corresponding to the: Raman emission of the MoS2 and substrate, the B exciton, the A and A- excitons and the I exciton have been highlighted. Figure 4 shows the spectrum measured for folded single-layer MoS2 compared to the limiting cases (1L and 2L MoS2). Similar measurements for bifolded single- and bilayer MoS2 can be found in the Electronic Supplementary Material. The photoluminescence spectra have been normalized to the E1 2g Raman peak intensity to account for the fact that thicker MoS2 layers absorb more excitation light as the photoluminescence yield is expected to be proportional to the absorbed excitation (See Figure S3 of the ESM). The folded MoS2 monolayer shows an enhanced photoluminescence emission of the A exciton in comparison to pristine 2L MoS2 (1.7 to 4.1 times larger) [54]. Moreover, the A exciton peak is composed of two Lorentzian peaks, indicating that both neutral and charged excitons contribute to the photoluminescence emission, similarly to the case of pristine 1L MoS2 [20] (See the Electronic Supporting Information for more detailed analysis of the photoluminescence spectra). These two observations (enhanced photoluminescence and a contribution of charged excitons to the photoluminescence yield) indicate that a twisted MoS2 bilayer presents a reduced interlayer coupling with respect to perfectly AB stacked bilayer MoS2, a conclusion consistent with the one obtained from the Raman studies. Moreover, the folded single-layer MoS2 exhibits an indirect bandgap transition with higher energy than that of pristine bilayer MoS2, as indicated by the blueshift (20 nm) of the I excitonic peak.  This is the post-peer reviewed version of the following article: A.Castellanos-Gomez et al. “Folded MoS2 layers with reduced interlayer coupling”. Nano Research, 2014, 7(4): 572–578 DOI 10.1007/s12274-014-0425-z Published in its final form at: http://link.springer.com/article/10.1007%2Fs12274-014-0425-z Figure 5 Normalized photoluminescence intensity of the A exciton peak (at 670 nm) as a function of the number of layers, measured for pristine layers (gray circles) and folded single-layer (dark red squares), bifolded single-layer (red squares) and bifolded bilayer (orange squares). We have found that for all studied samples (four 1L+1L, three 1L+1L+1L and five 2L+2L+2L samples), the photoluminescence intensity always show an enhancement of up to a factor 5 in comparison to their perfectly stacked counterparts (see Figure 5). This result further proof that folded and bifolded MoS2 layers exhibit a reduced interlayer coupling, in agreement with the conclusions obtained from the Raman spectroscopy and the analysis of the excitonic peaks of the photoluminescence spectra of folded single-layer MoS2. Conclusions In summary, MoS2 structures with twisted layers have been fabricated by folding pristine layers via an all dry process involving pre-stressed elastic substrates. Raman spectroscopy measurements indicate that the interlayer coupling strength is reduced due to the twisting. The photoluminescence spectra of twisted 1L (both folded and bifolded 1L) shows the contribution of neutral and charged excitons: the presence of charged excitons has only been observed in single-layer MoS2. Moreover, the photoluminescence spectra of twisted layers presents a peak corresponding to an indirect bandgap transition with smaller energy than that expected for perfectly stacked layers. In addition, photoluminescence emission yield is larger than that of perfectly stacked layers (up to a factor of 5). These observations all point at a modification of the interlayer coupling, and thus the bandstructure, by altering the natural stacking in MoS2.  This is the post-peer reviewed version of the following article: A.Castellanos-Gomez et al. “Folded MoS2 layers with reduced interlayer coupling”. Nano Research, 2014, 7(4): 572–578 DOI 10.1007/s12274-014-0425-z Published in its final form at: http://link.springer.com/article/10.1007%2Fs12274-014-0425-z More specifically, the interlayer coupling is reduced with respect to layers with optimal coupling and in many aspects twisted layers behave more like single-layer MoS2. The results shown here open the door to design artificial 3D MoS2 materials whose optoelectronic properties are engineered by the twisting angle between the MoS2 layers. Such 3D structures may have applications in photodetection and photovoltaics as one could fabricate multilayered MoS2 devices with high optical absorption (potentially much higher than that of monolayer MoS2) while maintaining some of the advantageous optoelectronic properties of single-layer MoS2. Acknowledgements The authors like to acknowledge fruitful discussions with J. Fernández-Rossier (INL, Portugal), A.C. Ferrari, R. S. Sundaram (Cambridge University, UK), R. Roldán, P. San-Jose (ICMM-CSIC, Spain) and E. Prada (Universidad Autonoma de Madrid). This work was supported by the European Union (FP7) through the program RODIN. A.C.-G. acknowledges financial support through the FP7-Marie Curie Project PIEF-GA-2011-300802 (‘STRENGTHNANO’). References [1] [2] [3] Yu, W. J.;Li, Z.;Zhou, H.;Chen, Y.;Wang, Y.;Huang, Y.; Duan, X. Vertically stacked multi- heterostructures of layered materials for logic transistors and complementary inverters. Nature Mater. 2012, 12, 246-252. Radisavljevic, B.;Radenovic, A.;Brivio, J.;Giacometti, V.; Kis, A. Single-layer MoS2 transistors. Nature Nanotech. 2011, 6, 147-150. Butler, S. Z.;Hollen, S. M.;Cao, L.;Cui, Y.;Gupta, J. A.;Gutierrez, H. R.;Heinz, T. F.;Hong, S. S.;Huang, J.; Ismach, A. F. Progress, Challenges, and Opportunities in Two-Dimensional Materials Beyond Graphene. ACS nano 2013, 7, 2898-2926. [4] Wang, H.;Yu, L.;Lee, Y.-H.;Shi, Y.;Hsu, A.;Chin, M. L.;Li, L.-J.;Dubey, M.;Kong, J.; Palacios, T. [5] [6] [7] [8] [9] [10] Integrated circuits based on bilayer MoS2 transistors. Nano letters 2012, 12, 4674-4680. Radisavljevic, B.;Whitwick, M. B.; Kis, A. Integrated Circuits and Logic Operations Based on Single-Layer MoS2. ACS nano 2011, 5, 9934-9938. Liu, J.;Zeng, Z.;Cao, X.;Lu, G.;Wang, L. H.;Fan, Q. L.;Huang, W.; Zhang, H. Preparation of MoS2‐ Polyvinylpyrrolidone Nanocomposites for Flexible Nonvolatile Rewritable Memory Devices with Reduced Graphene Oxide Electrodes. Small 2012, 8, 3517-3522. Yin, Z.;Li, H.;Li, H.;Jiang, L.;Shi, Y.;Sun, Y.;Lu, G.;Zhang, Q.;Chen, X.; Zhang, H. Single-layer MoS2 phototransistors. ACS Nano 2012, 6, 74-80. Lee, H. S.;Min, S.-W.;Chang, Y.-G.;Park, M. K.;Nam, T.;Kim, H.;Kim, J. H.;Ryu, S.; Im, S. MoS2 nanosheet phototransistors with thickness-modulated optical energy gap. Nano Lett. 2012, 12, 3695-3700. Lopez-Sanchez, O.;Lembke, D.;Kayci, M.;Radenovic, A.; Kis, A. Ultrasensitive photodetectors based on monolayer MoS2. Nature Nanotech. 2013. Buscema, M.;Barkelid, M.;Zwiller, V.;van der Zant, H. S.;Steele, G. A.; Castellanos-Gomez, A. Large and Tunable Photothermoelectric Effect in Single-Layer MoS2. Nano Lett. 2013, 13, 358- 363. [11] Wu, C.-C.;Jariwala, D.;Sangwan, V. K.;Marks, T. J.;Hersam, M. C.; Lauhon, L. J. Elucidating the photoresponse of ultrathin MoS2 field-effect transistors by scanning photocurrent microscopy. The Journal of Physical Chemistry Letters 2013.  This is the post-peer reviewed version of the following article: A.Castellanos-Gomez et al. “Folded MoS2 layers with reduced interlayer coupling”. Nano Research, 2014, 7(4): 572–578 DOI 10.1007/s12274-014-0425-z Published in its final form at: http://link.springer.com/article/10.1007%2Fs12274-014-0425-z [12] [13] [14] [15] Britnell, L.;Ribeiro, R.;Eckmann, A.;Jalil, R.;Belle, B.;Mishchenko, A.;Kim, Y.-J.;Gorbachev, R.;Georgiou, T.; Morozov, S. Strong light-matter interactions in heterostructures of atomically thin films. Science 2013, 340, 1311-1314. Zhang, W.;Huang, J. K.;Chen, C. H.;Chang, Y. H.;Cheng, Y. J.; Li, L. J. High‐ Gain Phototransistors Based on a CVD MoS2 Monolayer. Adv. Mater. 2013. Lin, J.;Li, H.;Zhang, H.; Chen, W. Plasmonic enhancement of photocurrent in MoS 2 field-effect- transistor. Appl. Phys. Lett. 2013, 102, 203109-203109-203103. Sundaram, R.;Engel, M.;Lombardo, A.;Krupke, R.;Ferrari, A.;Avouris, P.; Steiner, M. Electroluminescence in Single Layer MoS2. Nano Lett. 2013, 13, 1416-1421. [16] Ye, Y.;Ye, Z.;Gharghi, M.;Zhu, H.;Zhao, M.;Yin, X.; Zhang, X. Exciton-related [17] [18] electroluminescence from monolayer MoS2. arXiv preprint arXiv:1305.4235 2013. Ross, J. S.;Klement, P.;Jones, A. M.;Ghimire, N. J.;Yan, J.;Mandrus, D.;Taniguchi, T.;Watanabe, K.;Kitamura, K.; Yao, W. Electrically Tunable Excitonic Light Emitting Diodes based on Monolayer WSe2 pn Junctions. arXiv preprint arXiv:1312.1435 2013. Fontana, M.;Deppe, T.;Boyd, A. K.;Rinzan, M.;Liu, A. Y.;Paranjape, M.; Barbara, P. Electron-hole transport and photovoltaic effect in gated MoS2 Schottky junctions. Scientific reports 2013, 3. [19] Gu, X.;Cui, W.;Li, H.;Wu, Z.;Zeng, Z.;Lee, S. T.;Zhang, H.; Sun, B. A Solution‐ Processed Hole Extraction Layer Made from Ultrathin MoS2 Nanosheets for Efficient Organic Solar Cells. Advanced Energy Materials 2013, 3, 1262-1268. [20] Mak, K. F.;He, K.;Lee, C.;Lee, G. H.;Hone, J.;Heinz, T. F.; Shan, J. Tightly bound trions in monolayer MoS2. Nature Mater. 2012, 12, 207-211. [21] Mak, K. F.;He, K.;Shan, J.; Heinz, T. F. Control of valley polarization in monolayer MoS2 by [22] optical helicity. Nature Nanotech. 2012, 7, 494-498. Zeng, H.;Dai, J.;Yao, W.;Xiao, D.; Cui, X. Valley polarization in MoS2 monolayers by optical pumping. Nature Nanotech. 2012, 7, 490-493. [23] Wu, S.;Ross, J. S.;Liu, G.-B.;Aivazian, G.;Jones, A.;Fei, Z.;Zhu, W.;Xiao, D.;Yao, W.; Cobden, D. Electrical tuning of valley magnetic moment through symmetry control in bilayer MoS2. Nature Physics 2013. [24] Mak, K. F.;Lee, C.;Hone, J.;Shan, J.; Heinz, T. F. Atomically Thin MoS_ {2}: A New Direct-Gap [25] Semiconductor. Physical Review Letters 2010, 105, 136805. Splendiani, A.;Sun, L.;Zhang, Y.;Li, T.;Kim, J.;Chim, C.-Y.;Galli, G.; Wang, F. Emerging photoluminescence in monolayer MoS2. Nano Lett. 2010, 10, 1271-1275. [26] Korn, T.;Heydrich, S.;Hirmer, M.;Schmutzler, J.; Schuller, C. Low-temperature photocarrier dynamics in monolayer MoS< inf> 2</inf>. Appl. Phys. Lett. 2011, 99, 102109-102109-102103. [27] Huang, X.;Zeng, Z.; Zhang, H. Metal dichalcogenide nanosheets: preparation, properties and [28] [29] [30] [31] [32] applications. Chem. Soc. Rev. 2013, 42, 1934-1946. Chhowalla, M.;Shin, H. S.;Eda, G.;Li, L.-J.;Loh, K. P.; Zhang, H. The chemistry of two- dimensional layered transition metal dichalcogenide nanosheets. Nature chemistry 2013, 5, 263- 275. Bertolazzi, S.;Brivio, J.; Kis, A. Stretching and Breaking of Ultrathin MoS2. ACS Nano 2011, 5, 9703-9709. Castellanos-Gomez, A.;Poot, M.;Steele, G. A.;van der Zant, H. S.;Agraït, N.; Rubio-Bollinger, G. Elastic properties of freely suspended MoS2 nanosheets. Adv. Mater. 2012, 24, 772-775. Cooper, R. C.;Lee, C.;Marianetti, C. A.;Wei, X.;Hone, J.; Kysar, J. W. Nonlinear elastic behavior of two-dimensional molybdenum disulfide. Phys. Rev. B 2013, 87, 035423. Castellanos-Gomez, A.;Roldán, R.;Cappelluti, E.;Buscema, M.;Guinea, F.;van der Zant, H. S.; Steele, G. A. Local strain engineering in atomically thin MoS2. 2013, 13, 5361–5366.  This is the post-peer reviewed version of the following article: A.Castellanos-Gomez et al. “Folded MoS2 layers with reduced interlayer coupling”. Nano Research, 2014, 7(4): 572–578 DOI 10.1007/s12274-014-0425-z Published in its final form at: http://link.springer.com/article/10.1007%2Fs12274-014-0425-z [33] Vella, D.;Bico, J.;Boudaoud, A.;Roman, B.; Reis, P. M. The macroscopic delamination of thin films from elastic substrates. Proc. Natl. Acad. Sci. USA 2009, 106, 10901-10906. [34] Ni, Z.;Liu, L.;Wang, Y.;Zheng, Z.;Li, L.-J.;Yu, T.; Shen, Z. G-band Raman double resonance in twisted bilayer graphene: Evidence of band splitting and folding. Phys. Rev. B 2009, 80, 125404. [36] [35] Hao, Y.;Wang, Y.;Wang, L.;Ni, Z.;Wang, Z.;Wang, R.;Koo, C. K.;Shen, Z.; Thong, J. T. Probing Layer Number and Stacking Order of Few‐ Layer Graphene by Raman Spectroscopy. Small 2010, 6, 195-200. Scalise, E.;Houssa, M.;Pourtois, G.;Afanas’ ev, V.; Stesmans, A. Strain-induced semiconductor to metal transition in the two-dimensional honeycomb structure of MoS 2. Nano Res. 2012, 5, 43-48. Feng, J.;Qian, X.;Huang, C.-W.; Li, J. Strain-engineered artificial atom as a broad-spectrum solar energy funnel. Nature Photon. 2012, 6, 866-872. Conley, H. J.;Wang, B.;Ziegler, J. I.;Haglund, R. F.;Pantelides, S. T.; Bolotin, K. I. Bandgap Engineering of Strained Monolayer and Bilayer MoS2. Nano Lett. 2013, 13, 3626-3630. [37] [38] [39] Ghorbani-Asl, M.;Borini, S.;Kuc, A.; Heine, T. Strain-dependent modulation of conductivity in single layer transition-metal dichalcogenides. arXiv preprint arXiv:1301.3469 2013. [40] He, K.;Poole, C.;Mak, K. F.; Shan, J. Experimental demonstration of continuous electronic structure tuning via strain in atomically thin MoS2. Nano Lett. 2013, 13, 2921-2936. [42] [41] Hui, Y. Y.;Liu, X.;Jie, W.;Chan, N. Y.;Hao, J.;Hsu, Y.-T.;Li, L.-J.;Guo, W.; Lau, S. P. Exceptional Tunability of Band Energy in a Compressively Strained Trilayer MoS2 Sheet. ACS nano 2013, 7, 7126-7131. Sengupta, A.;Ghosh, R. K.; Mahapatra, S. Performance Analysis of Strained Monolayer MoS2 MOSFET. Electron Devices, IEEE Transactions on 2013, 60, 1782-2787. Castellanos-Gomez, A.;Agraït, N.; Rubio-Bollinger, G. Optical identification of atomically thin dichalcogenide crystals. Appl. Phys. Lett. 2010, 96, 213116. Castellanos-Gomez, A.;Barkelid, M.;Goossens, A.;Calado, V. E.;van der Zant, H. S.; Steele, G. A. Laser-thinning of MoS2: on demand generation of a single-layer semiconductor. Nano Lett. 2012, 12, 3187-3192. [43] [44] [45] Najmaei, S.;Liu, Z.;Ajayan, P.; Lou, J. Thermal effects on the characteristic Raman spectrum of molybdenum disulfide (MoS< inf> 2</inf>) of varying thicknesses. Appl. Phys. Lett. 2012, 100, 013106. [46] Yan, R.;Bertolazzi, S.;Brivio, J.;Fang, T.;Konar, A.;Birdwell, A. G.;Nguyen, N.;Kis, A.;Jena, D.; Xing, H. G. Raman and Photoluminescence Study of Dielectric and Thermal Effects on Atomically Thin MoS2. arXiv preprint arXiv:1211.4136 2012. Buscema, M.;Steele, G. A.;van der Zant, H. S.; Castellanos-Gomez, arXiv:1311.3869. 2013. Ferrari, A. C.; Basko, D. M. Raman spectroscopy as a versatile tool for studying the properties of graphene. Nature Nanotech. 2013, 8, 235-246. Lee, C.;Yan, H.;Brus, L. E.;Heinz, T. F.;Hone, J.; Ryu, S. Anomalous lattice vibrations of single- and few-layer MoS2. ACS Nano 2010, 4, 2695-2700. [47] [48] [49] [50] Molina-Sánchez, A.; Wirtz, L. Phonons in single-layer and few-layer MoS_ {2} and WS_ {2}. [51] Physical Review B 2011, 84, 155413. Rice, C.;Young, R.;Zan, R.;Bangert, U.;Wolverson, D.;Georgiou, T.;Jalil, R.; Novoselov, K. Raman-scattering measurements and first-principles calculations of strain-induced phonon shifts in monolayer MoS_ {2}. Phys. Rev. B 2013, 87, 081307. [52] Wang, Y.;Cong, C.;Qiu, C.; Yu, T. Raman Spectroscopy Study of Lattice Vibration and [53] Crystallographic Orientation of Monolayer MoS2 under Uniaxial Strain. Small 2013. Eda, G.;Yamaguchi, H.;Voiry, D.;Fujita, T.;Chen, M.; Chhowalla, M. Photoluminescence from chemically exfoliated MoS2. Nano Lett. 2011, 11, 5111-5116.  This is the post-peer reviewed version of the following article: A.Castellanos-Gomez et al. “Folded MoS2 layers with reduced interlayer coupling”. Nano Research, 2014, 7(4): 572–578 DOI 10.1007/s12274-014-0425-z Published in its final form at: http://link.springer.com/article/10.1007%2Fs12274-014-0425-z [54] A folded chemical vapour deposited MoS2 monolayer was also studied in [55], but in that work the pristine material showed an unexplained redshift in the PL spectra and the fold was made using a differnt method making a direct comparison with our results less straightforward. Crowne, F.J., et al., Blue shifting of the A exciton peak in folded monolayer 1H-MoS2. Phys. Rev. B 2013, 88, 235302. [55]  This is the post-peer reviewed version of the following article: A.Castellanos-Gomez et al. “Folded MoS2 layers with reduced interlayer coupling”. Nano Research, 2014, 7(4): 572–578 DOI 10.1007/s12274-014-0425-z Published in its final form at: http://link.springer.com/article/10.1007%2Fs12274-014-0425-z Electronic Supplementary Material Folded MoS2 layers with reduced interlayer coupling Andres Castellanos-Gomez*, Herre S. J. van der Zant, and Gary A. Steele Kavli Institute of Nanoscience, Delft University of Technology, 2628 CJ Delft, The Netherlands. [email protected] Table of contents Transfer of the folded and bi-folded MoS2 samples to other substrates Doping of MoS2 samples on Gelfilm and SiO2 substrates Normalization of the photoluminescence spectra Sample characterization Raman and PL spectra measured for different folded MoS2 structures Analysis of the photoluminescence spectra measured for different folded MoS2 structures ———————————— Address correspondence to: Andres Castellanos-Gomez, [email protected]  This is the post-peer reviewed version of the following article: A.Castellanos-Gomez et al. “Folded MoS2 layers with reduced interlayer coupling”. Nano Research, 2014, 7(4): 572–578 DOI 10.1007/s12274-014-0425-z Published in its final form at: http://link.springer.com/article/10.1007%2Fs12274-014-0425-z Transfer of the folded and bi-folded MoS2 samples to other substrates After fabrication of the folded and bi-folded MoS2 structures onto a Gelfim substrate, the samples can be transferred to a different substrate. The surface of the Gelfilm substrate is pressed against the new substrate and peeled off very slowly, transferring some of the flakes. We found that the slower the peeling off the higher transfer yield. Figure S1 shows an example of a large area 2 to 8 layers thick flake with many bifolds which has been partially transferred onto a SiO2/Si substrate. Figure S1 Optical images of a bifolded MoS2 sample, as fabricated on a Gelfilm substrate (a) and after transfer part of the flake onto a SiO2/Si substrate. Doping of MoS2 samples on Gelfilm and SiO2 substrates The photoluminescence experiments shown in the main text have been carried out on samples fabricated onto Gelfilm because of their lower doping level. Indeed, we recently found that the photoluminescence spectra of samples fabricated onto Gelfilm presents reduced doping level in comparison to samples prepared on SiO2. This reduced doping level yields photoluminescence spectra very similar to that of freely suspended MoS2. Figure S2 shows a comparison of the photoluminescence spectra measured in single-layer MoS2 flakes deposited onto Gelfilm (panel a) and SiO2 (panel b). The contribution of the different excitons to the photoluminescence can be determined by performing a multi-Gaussian fit to the data. The different excitons have been highlighted on the plot, being very clear the low intensity of the neutral A exciton in the sample prepared on SiO2, which indicates a high doping level on the flake. The samples prepared on Gelfilm, on the other hand, present an almost equal ratio of charged and neutral excitons, very similar to what is observed on freely suspende MoS2. Normalization of the photoluminescence spectra In order to compare the intrinsic photoluminescence yield of MoS2 samples with different thicknesses one has to account for the fact that thicker samples will absorb more excitation light and their photoluminescence intensity will be larger. In order to discount for this effect (to determine the intrinsic photoluminescence yield) one can normalize the photoluminescence spectra with a normalization value  This is the post-peer reviewed version of the following article: A.Castellanos-Gomez et al. “Folded MoS2 layers with reduced interlayer coupling”. Nano Research, 2014, 7(4): 572–578 DOI 10.1007/s12274-014-0425-z Published in its final form at: http://link.springer.com/article/10.1007%2Fs12274-014-0425-z that depends on the absoption. In the literature the intensity of some of the Raman peaks is typically employed to account this effect. Figure S3a shows several Raman spectra measured for MoS2 samples with different thicknesses under similar conditions of incident power, acquisition time and focusing. The intensity of the Raman peaks depends linearly on the number of layers within experimental uncertainty (Figure S3b) and therefore it can be used to normalize the photoluminescence spectra. Figure S2 Photoluminescence spectra of single-layer MoS2 samples fabricated on Gelfilm and SiO2/Si substrates. A multiple-Gaussian fit has been used to determine the intensities of the B, A, and A-(trion) excitons. The neutral A exciton intensity is very small for samples fabricated on SiO2, due to high doping level. On the other hand, samples fabricated onto Gelfilm present a PL spectrum very similar to freely suspended MoS2 with high intensity of the A neutral exciton.  This is the post-peer reviewed version of the following article: A.Castellanos-Gomez et al. “Folded MoS2 layers with reduced interlayer coupling”. Nano Research, 2014, 7(4): 572–578 DOI 10.1007/s12274-014-0425-z Published in its final form at: http://link.springer.com/article/10.1007%2Fs12274-014-0425-z Figure S3 (a) Raman spectra measured for MoS2 samples with different number of layers fabricated onto Gelfilm substrates, measured under same intensity, focus and acquisition time conditions. (b) Dependence of the intensity of the Raman peaks with the number of layers, showing that the intensity is proportional to the number of layers. Sample characterization Apart from the Raman spectroscopy and photoluminescence characterization, the fabricated samples were studied by quantitative optical microscopy and atomic force microscopy. As the samples were fabricated onto Gelfilm substrates, which are transparent, they can be studied in transmission mode optical microscopy. Figure S4a shows a false color map of the transmittance of a MoS2 sample with a folded monolayer area. In order to obtain the transmittance, an optical image is acquired using a digital camera attached to the trinocular of the microscope. The values of the red, green and blue channels are summed to make a matrix of intensities per pixel. Especial attention is taken to avoid using images with some channel saturated or underexposed. Then the transmittance is obtained by dividing the matrix by the average intensity on the bare substrate. The resulting matrix can be plotted in a false color image as Figure S4a. Figure S4b presents the transmittance values measured for both pristine and folded/bifolded samples with different thicknesses. The transmittance of the folded/bifolded samples is in good agreement with the  This is the post-peer reviewed version of the following article: A.Castellanos-Gomez et al. “Folded MoS2 layers with reduced interlayer coupling”. Nano Research, 2014, 7(4): 572–578 DOI 10.1007/s12274-014-0425-z Published in its final form at: http://link.springer.com/article/10.1007%2Fs12274-014-0425-z values obtained for pristine samples. Figure S4 (a) Optical transmittance map of a folded MoS2 sample. (b) Dependence of the transmittance with the number of layers, included datapoints measured on folded and bifolded layers. Atomic force microscopy has been used to study the topography of the fabricated samples. However, this characterization is very challenging for the samples fabricated on Gelfilm substrates as the substrate easily deforms during the scanning leading to setpoint-dependent images. Moreover, the large tip-Gelfilm interaction makes the feedback loop more unstable on the regions uncovered by the MoS2. Therefore, studying the details of the folded structures results non-trivial. To overcome this issue one can transfer the folded/bifolded structures onto a flat SiO2/Si surface before characterizing with the AFM. Figure S5 shows two AFM images of bifolded bilayer MoS2 structures as fabricated on Gelfilm and after transfer to a SiO2 surface. From the image in Figure S5b one can see that the bifolded structure is composed by twisted bilayers of MoS2 stacked on top of each other. Figure S5 Atomic force microscopy images of bifolded bilayer MoS2 on gelfilm (a) and SiO2/Si (b) substrates. Topographic line profiles, measured along lines, are included below the images. the solid white  This is the post-peer reviewed version of the following article: A.Castellanos-Gomez et al. “Folded MoS2 layers with reduced interlayer coupling”. Nano Research, 2014, 7(4): 572–578 DOI 10.1007/s12274-014-0425-z Published in its final form at: http://link.springer.com/article/10.1007%2Fs12274-014-0425-z Raman and PL spectra measured for different folded MoS2 structures gure S6 Raman spectra measured for folded (1L+1L) and bifolded (1L+1L+1L) monolayer MoS 2 and bifolded bilayer MoS2 (2L+2L+2L). Measurements on pristine 1L, 2L, 3L and 6L MoS2 have been also included to facilitate the comparison as they can be considered as limiting cases: negligible and optimal interlayer coupling. The thin light lines are Lorentzian fits to the experimental data.  This is the post-peer reviewed version of the following article: A.Castellanos-Gomez et al. “Folded MoS2 layers with reduced interlayer coupling”. Nano Research, 2014, 7(4): 572–578 DOI 10.1007/s12274-014-0425-z Published in its final form at: http://link.springer.com/article/10.1007%2Fs12274-014-0425-z Figure S7 Photoluminescence spectra measured for folded (1L+1L) and bifolded (1L+1L+1L) monolayer MoS2 and bifolded bilayer MoS2 (2L+2L+2L). Measurements on pristine 1L, 2L, 3L and 6L MoS2 have been also included to facilitate the comparison as they can be considered as limiting cases: negligible and optimal interlayer coupling. Analysis of the photoluminescence spectra measured for different folded MoS2 structures Similarly to the results presented in the main text for the A exciton, the B exciton photoluminescence emission is also enhanced for folded and bifolded samples in comparison with their pristine counterparts. Figure S8 shows a plot with the comparison of the B exciton intensity (normalized to the E12g Raman peak) as a function of the number of layers. The folded and bifolded structures present B exciton intensities larger than the pristine layers (up to a factor of 5), indicating a reduced interlayer coupling due to the folded structure. Table S1 summarizes the normalized intensities of the different peaks composing the A exciton (the neutral A exciton and the charged A- exciton) determined by performing multi-Gaussian fits as those shown in Figure S2. For pristine monolayer the neutral A exciton occurs at ~655 nm and the charged occurs at ~670 nm. For multilayers, however, the emission is composed exclusively by one peak around ~670 nm associated to the emission of neutral excitions in multilayer MoS2 (in agreement with the lack of observation of trions in multilayer MoS2). Folded and bifolded single-layer MoS2 behave very differently from their pristine counterparts as their PL emission shows again two peaks exactly at the wavelengths associated to the emission of neutral and charged excitons in single-layer MoS2. This indicates than these structures  This is the post-peer reviewed version of the following article: A.Castellanos-Gomez et al. “Folded MoS2 layers with reduced interlayer coupling”. Nano Research, 2014, 7(4): 572–578 DOI 10.1007/s12274-014-0425-z Published in its final form at: http://link.springer.com/article/10.1007%2Fs12274-014-0425-z formed by folded and bifolded single-layers behave similarly to pristine single-layer MoS2. For bifolded bilayer, the PL spectra only presents one peak around ~670 nm. Figure S8 Normalized intensity of the B exciton peak as a function of the number of layers. Datapoints measured for folded and bifolded structures have been also included. Similarly to what is showed for A exciton (Figure 4 of the main text) the B exciton emission yield is enhanced for folded and bifolded structures. Table S1 Summary of the normalized intensity of the peaks appearing around 655 nm and 670 nm for pristine and folded/bifolded structures with different thicknesses. Pristine structures # of layers IPeak at ~655nm IPeak at ~670nm Folded or bifolded structures # of layers IPeak at ~655nm IPeak at ~670nm 1L 2L 3L 4L 5L 6L 6L 7L 6.81 0.0 0.0 0.0 0.0 0.0 0.0 0.0 9.83 1.24 0.60 0.38 0.30 0.28 0.22 0.21 1L + 1L 1L + 1L 1L + 1L 1L + 1L 1L + 1L +1L 1L + 1L +1L 1L + 1L +1L 2L + 2L + 2L 2L + 2L + 2L 2L + 2L + 2L 2L + 2L + 2L 2L + 2L + 2L 2.00 0.82 0.95 0.23 0.57 0.65 0.44 0.0 0.0 0.0 0.0 0.0 2.82 2.40 1.86 1.73 1.64 1.83 1.90 0.66 0.69 0.80 0.75 0.45 

1801.00971

1

1801

Strong spin-orbit interaction and magnetotransport in semiconductor Bi$_2$O$_2$Se nanoplates

Semiconductor Bi$_2$O$_2$Se nanolayers of high crystal quality have been realized via epitaxial growth. These two-dimensional (2D) materials possess excellent electron transport properties with potential application in nanoelectronics. It is also strongly expected that the 2D Bi$_2$O$_2$Se nanolayers could be of an excellent material platform for developing spintronic and topological quantum devices, if the presence of strong spin-orbit interaction in the 2D materials can be experimentally demonstrated. Here, we report on experimental determination of the strength of spin-orbit interaction in Bi$_2$O$_2$Se nanoplates through magnetotransport measurements. The nanoplates are epitaxially grown by chemical vapor deposition and the magnetotransport measurements are performed at low temperatures. The measured magnetoconductance exhibits a crossover behavior from weak antilocalization to weak localization at low magnetic fields with increasing temperature or decreasing back gate voltage. We have analyzed this transition behavior of the magnetoconductance based on an interference theory which describes the quantum correction to the magnetoconductance of a 2D system in the presence of spin-orbit interaction. Dephasing length and spin relaxation length are extracted from the magnetoconductance measurements. Comparing to other semiconductor nanostructures, the extracted relatively short spin relaxation length of ~150 nm indicates the existence of strong spin-orbit interaction in Bi$_2$O$_2$Se nanolayers.

cond-mat.mes-hall

cond-mat

Strong spin-orbit interaction and magnetotransport in semiconductor Bi2O2Se nanoplates Mengmeng Meng, 1 Shaoyun Huang, 1, * Congwei Tan, 2 Jinxiong Wu, 2 Yumei Jing, 1 Hailin Peng2,* and H. Q. Xu1, 3,* 1Beijing Key Laboratory of Quantum Devices, Key Laboratory for the Physics and Chemistry of Nanodevices, and Department of Electronics, Peking University, Beijing 100871, China 2Center for Nanochemistry, Beijing National Laboratory for Molecular Sciences (BNLMS), College of Chemistry and Molecular Engineering, Peking University, Beijing 100871, China. 3Division of Solid State Physics, Lund University, Box 118, S-221 00 Lund, Sweden Semiconductor Bi2O2Se nanolayers of high crystal quality have been realized via epitaxial growth. These two-dimensional (2D) materials possess excellent electron transport properties with potential application in nanoelectronics. It is also strongly expected that the 2D Bi2O2Se nanolayers could be of an excellent material platform for developing spintronic and topological quantum devices, if the presence of strong spin-orbit interaction in the 2D materials can be experimentally demonstrated. Here, we report on experimental determination of the strength of spin-orbit interaction in Bi2O2Se nanoplates through magnetotransport measurements. The nanoplates are epitaxially grown by chemical vapor deposition and the magnetotransport measurements are performed at low temperatures. The measured magnetoconductance exhibits a crossover behavior from weak antilocalization to weak localization at low magnetic fields with increasing temperature or decreasing back gate voltage. We have analyzed this transition behavior of the magnetoconductance based on an interference theory which describes the quantum correction to the magnetoconductance of a 2D system in the presence of spin-orbit interaction. Dephasing length and spin relaxation length are extracted from the magnetoconductance measurements. Comparing to other semiconductor nanostructures, the extracted relatively short spin relaxation length of ~150 nm indicates the existence of strong spin-orbit interaction in Bi2O2Se nanolayers. Keywords: Bi2O2Se nanoplate, weak antilocalization, spin-orbit interaction *Corresponding authors: Professor H. Q. Xu ([email protected]), Dr. Shaoyun Huang ([email protected]), and Professor Hailin Peng ([email protected]) 1 The discovery of graphene has inspired extensive investigation of electrical, mechanical and optical properties of two-dimensional (2D) materials1-5. As one of emerging 2D semiconducting materials, high-quality Bi2O2Se nanoplates has recently been successfully grown by chemical vapor deposition (CVD) on mica substrates6. The grown 2D Bi2O2Se nanoplates not only possess a band energy gap, which is desired for applications in planar nanoelectronic logic devices and circuits, but also are stable against oxidation and moisture in the air6, 7. The electron field-effect mobility of the nanoplates is found to reach ∼450 cm2/V⋅s at room temperature. At the low temperature of 2 K, the high electron Hall mobility of ∼20000 cm2/V⋅s is detected and Shubnikov-de Haas (SdH) quantum coherent oscillations are observed at moderate magnetic fields in the Bi2O2Se nanoplates6, implying that the materials have excellent transport properties and could be potentially used as a new host material system for exploration of novel quantum transport phenomena. Considering the conduction electron states and the valence hole states around the band gap are dominantly built from the p-orbitals of Bi and Se elements6, 2D Bi2O2Se nanoplates are expected to possess strong spin-orbit interaction. Such layered semiconducting, strong-spin interacting electron systems are desired for constructing planar topological superconducting systems, in which Majorana bound states8-13 can be created and manipulated, with potential applications in topological quantum computing. Moreover, since electron spins can be manipulated by an electric field in an electron system with strong spin-orbit interaction, 2D Bi2O2Se nanoplates could be utilized for developing spintronic field-effect transistors and spin-based quantum information technology14-16. Thus, the main purpose of this work is to experimentally determine the strength of spin-orbit interaction in 2D Bi2O2Se nanoplates through magnetotransport measurements. Magnetotransport is a key method to determine characteristic length scales in mesoscopic systems17, including spin relaxation length (𝐿𝐿𝑠𝑠𝑠𝑠 ), which is important to understand and control transport and spin manipulation process. In a scattered, time reversal electron system where electrons move in a diffusive instead of ballistic way, quantum interference occurs on closed paths, making a correction to the classical conductivity. With applying a small magnetic field, due to the suppression of the constructive interference between time-reversed closed trajectories, magnetoconductance increases with increasing magnetic field and displays the weak localization (WL) characteristics. In the presence of strong spin-orbit interaction, quantum correction to the zero-magnetic field classical conductivity becomes positive, leading to decrease in magnetoconductance with increasing 2 magnetic field and thus the weak antilocalization (WAL) characteristics18. Here, we report on a detailed magnetotransport study of a Bi2O2Se nanoplate field- effect device. The low-field magnetoconductance has been measured as a function of temperature and voltage applied to the back gate, and the results are analyzed by means of the Hikami–Larkin–Nagaoka (HLN) interference model. The electron phase coherence length and spin relaxation length in the measured Bi2O2Se nanoplate are extracted. Our experiment demonstrates the presence of a strong spin-orbit interaction in the Bi2O2Se nanoplate. A crossover between the WAL to WL characteristics with tuning temperature or back gate voltage has also been observed in the Bi2O2Se nanoplate. The analysis shows that the crossover is governed by the relative scales of the electron phase coherence length and spin relaxation length. This work demonstrates for the first time an experimental observation of strong spin-orbit interaction in a 2D Bi2O2Se nanoplate and will stimulate experimental and theoretical studies of the materials and their potential applications. The Bi2O2Se nanoplates used in this work are grown in a homemade low-pressure CVD system with Bi2O3 powder and Bi2Se3 bulk as sources. The Bi2O3 powder (Alfa Aesar, 99.995%) is placed at the center of a horizontally arranged quartz tube and the Bi2Se3 bulk (Alfa Aesar, 99.995%) at ~6 cm upstream19. Freshly cleaved fluorophlogopite mica is employed as growth substrate, which possesses atomically flat surface and is an ideal substrate to synthesize ultra-thin 2D materials20, 21. Argon is used as the carrier gas to transfer the precursor to the growth region. For the further details regarding the materials growth, we refer to Refs. 6 and 20. Figure 1 shows the structural characterization of the as-grown Bi2O2Se nanoplates. Bi2O2Se is a typical bismuth-based oxychalcogenide, possessing a tetragonal structure in the I4/mmm space group (a = 3.88 Å, c = 12.16 Å and Z= 2)22. Bi2O2Se is also an ionic layered 2n+ layers are sandwiched by negatively charged material, where positively charged [Bi2O2]n 2n- layers through relatively weak electrostatic interactions (Figure 1a). Cooperated with [Se]n appropriate thermodynamics, the strong interlayer/intra-layer bonding anisotropy of Bi2O2Se crystal promises to form a 2D crystal on a mica substrate. Figure 1b shows an optical image of some as-grown Bi2O2Se nanoplates on a growth mica substrate. As shown in the figure, the nanoplates exhibit a large domain size6 (about 50×50 μm2 or larger). Square shape is arising from the tetragonal structure of Bi2O2Se. Figure 1c displays a representative atomic force microscope (AFM) image of a domain of a single Bi2O2Se nanoplate, showing that as-grown nanoplates have an atomically flat surface. It has been observed that the 3 atomically flat surfaces of Bi2O2Se nanoplates remains almost the same after exposed to the air for months, demonstrating the stability of the oxide semiconductor layers in the ambient environment. The thickness of the nanoplates ranges from several to tens of nanometer, and Figure 1c shows a nanoplate with a thickness of ~6.2 nm (10 layers). The crystalline structure of the as-grown Bi2O2Se nanoplates is examined by high-resolution transmission electron microscope (HRTEM) measurements. Figure 1d displays a HRTEM image of an as-grown Bi2O2Se nanoplate, where a well-defined arrangement of Bi, O, and Se atomic lattices can be identified, see the area marked by a colored square in the figure. Our HRTEM measurements shows that the CVD grown 2D Bi2O2Se nanoplates are of high-quality single crystals. As-grown Bi2O2Se nanoplates are transferred onto a 300-nm-thick SiO2/Si substrate (which is to be used as the back gate) with predefined positioning markers for device fabrication through a PMMA (polymethyl methacrylate)-mediated method.23 After removal of the mediated PMMA used during the transfer process, the Bi2O2Se nanoplates on the device substrate are again covered with PMMA by spin coating. Contact electrodes are then fabricated on selected Bi2O2Se nanoplates by electron-beam lithography, electron-beam evaporation of a Ti/Au (5/90 nm) metal layer, and lift-off process. Here, we note that before the metal evaporation, soft Ar plasma is used to remove the residual PMMA resist in the Bi2O2Se contact areas to ensure obtaining clean metal-Bi2O2Se interfaces. Figure 2a shows an optical image of a device fabricated from a Bi2O2Se nanoplate of ~8 nm in thickness. Four parallel metal electrodes with a width of 800 nm and a pitch of 2.6 μm are made on the Bi2O2Se nanoplate. Figure 2b shows the schematic structure of the fabricated device and the circuit setup for the magnetotransport measurements presented in this work. The magnetotransport measurements are carried out in a physical property measurement system (PPMS) cryostat equipped with a uniaxial magnet. As shown in Figure 2b, the magnetotransport measurements are performed in a four-probe configuration using a standard lock-in technique, in which a 17-Hz AC current is supplied between the two outer electrodes and the voltage between the two inner contact electrodes is detected. To suppress the influence of Joule heating but in the same time also conductance fluctuations, low-field magnetoresistance measurements at each temperature are performed at several excitation currents. After evaluation of the test measurements, an optimal current excitation of 100 nA is found and is set in the following magnetotransport measurements20, 24, 25. Figure 2c shows the measured transfer characteristics of the fabricated Bi2O2Se 4 nanoplate device shown in Figure 2a, where the conductance, G=I/V, measured using the circuit setup depicted in the inset of Figure 2c at zero magnetic field is plotted as a function of voltage Vbg applied to the back gate at different temperatures. It is seen that the device is a typical n-type transistor, and the conduction channel is in an open state at zero back gate voltage and can be switched off by applying a negative gate voltage. At 2 K, the threshold voltage at which the device is switched off is about −7.2 V, see Supplementary Information. The threshold voltage moves toward to more negative back gate voltage with increasing temperature. This is a common feature of semiconductor field-effect devices, due to the presence of a high carrier density at a high temperature. At a large positive back gate voltage where the device is at an on-state, the conductance decreases with increasing temperature. This is due to increase in phonon scattering with increasing temperature, as seen in other field–effect devices made from semiconductor nanoplates26. Taking into account the device geometry, the field-effect mobility and carrier density can be extracted from the measured transfer characteristics27, see Supplementary Information. At 2 K, the extracted field-effect mobility and carrier density are 1509 cm2/V∙s and 2.7×1012 cm-2 at Vbg = 30 V, giving a carrier mean-free path of 𝑙𝑙𝑒𝑒~35 nm in the Bi2O2Se nanoplate. This mean-free path value is much smaller than the distance between the two inner probes and thus the carrier transport in our Bi2O2Se nanoplate device is in the diffusive regime even at the low temperature of 2 K (see Supplementary Information for the measurements of the carrier density, mobility, and mean free path in the Bi2O2Se nanoplate at elevated temperatures). To determine the electron phase coherence length and spin relaxation length in the Bi2O2Se nanoplate, low-field magnetotransport measurements are performed in the fabricated Bi2O2Se nanoplate device at low temperatures. Figure 3a shows the measured magnetoconductance 𝛥𝛥𝛥𝛥=𝛥𝛥(𝐵𝐵)−𝛥𝛥(𝐵𝐵=0) at 2 K and at different back gate voltages. Here, the measured curves are successively offset vertically for clarity. It is seen that the measured magnetoconductance shows a WAL-WL crossover with decreasing back gate voltage. At positive back gate voltages, the magnetoconductance near zero magnetic field shows the WAL characteristics, suggesting the existence of strong spin-orbit interaction in the Bi2O2Se nanoplate. As the back gate voltage decreases towards negative values, the WAL characteristics are gradually suppressed and the magnetoconductance shows dominantly the WL characteristics. Such a gate-voltage tunable crossover between WAL and WL has also been observed in systems with strong spin-orbit interaction, such as AlxGa1-xN/GaN heterostructures28, InAs nanowires29, 30 and InSb nanowires31. 5 In a 2D disorder system, the low-field magnetoconductance is well described by the HLN quantum interference model.32, 33 Assuming that the transport in the Bi2O2Se nanoplate is in the 2D disorder regime, as we will show below this assumption is appropriate, the quantum correction to the low-field magnetoconductance of the device is given by32, 33 ⁄ at different back gate voltages to Eq. (1). Figures 3b and 3c (black solid lines) show the ∆𝛥𝛥(𝐵𝐵)=−𝑒𝑒2𝜋𝜋ℎ�12Ψ�𝐵𝐵𝜙𝜙𝐵𝐵+12�+Ψ�𝐵𝐵𝑠𝑠𝑠𝑠+𝐵𝐵𝑒𝑒𝐵𝐵 +12�−32Ψ�(43⁄ )𝐵𝐵𝑠𝑠𝑠𝑠+𝐵𝐵𝜙𝜙 +12�−12ln�𝐵𝐵𝜙𝜙𝐵𝐵�− 𝐵𝐵 �+ 32ln�(43⁄ )𝐵𝐵𝑠𝑠𝑠𝑠+𝐵𝐵𝜙𝜙 ��. (1) ln�𝐵𝐵𝑠𝑠𝑠𝑠+𝐵𝐵𝑒𝑒𝐵𝐵 𝐵𝐵 Here, Ψ(x) is the digamma function. Three subscripts are used to denote different scattering processes: 𝜙𝜙 for inelastic dephasing process, 𝑠𝑠𝑠𝑠 for spin-orbit scattering, and e for elastic scattering. 𝐵𝐵𝜙𝜙,𝑠𝑠𝑠𝑠,𝑒𝑒 is the characteristic field for each corresponding scattering mechanism ) and is given by 𝐵𝐵𝜙𝜙,𝑠𝑠𝑠𝑠,𝑒𝑒=ℏ(4𝑒𝑒𝐿𝐿𝜙𝜙,𝑠𝑠𝑠𝑠,𝑒𝑒 . 𝐿𝐿𝜙𝜙 is the dephasing length, 𝐿𝐿𝑠𝑠𝑠𝑠 the spin 2 relaxation length, and 𝐿𝐿𝑒𝑒 the mean free path. We fit our measured magnetoconductance data results of fitting for the measured low-field magnetoconductance at 𝑉𝑉𝑏𝑏𝑏𝑏=30 V and 𝑉𝑉𝑏𝑏𝑏𝑏=−14 V. Here, the fitting range is limited within ±0.2 T to satisfy the small field Figure 3d shows the extracted dephasing length 𝐿𝐿𝜙𝜙, spin relaxation length 𝐿𝐿𝑠𝑠𝑠𝑠 and mean free path 𝐿𝐿𝑒𝑒 as a function of back gate voltage. The main feature shown in Figure 3d is that 𝐿𝐿𝜙𝜙 is strongly dependent on back gate voltage 𝑉𝑉𝑏𝑏𝑏𝑏 , while 𝐿𝐿𝑠𝑠𝑠𝑠 shows a weak dependence on 𝑉𝑉𝑏𝑏𝑏𝑏. The extracted 𝐿𝐿𝜙𝜙 decreases from 300 nm to 100 nm as 𝑉𝑉𝑏𝑏𝑏𝑏 sweeps region. In addition, 𝐿𝐿𝜙𝜙 is much larger than the thickness of the nanoplate at all back gate precondition. It is seen that the measured magnetoresistances can be well described by the HLN formula. from 30 to -14 V, indicating that the dephasing process is stronger in the low carrier density voltages, which is fully consistent with the 2D nature of transport in the Bi2O2Se nanoplate we have assumed above. At the low temperature, the main source of dephasing is electron- electron interaction in two mechanisms. One is the Nyquist scattering representing the small energy transferred interactions of electrons with electromagnetic field fluctuations generated by the movement of neighboring electrons34. The other is the direct Coulomb interaction among the electrons. The observed enhanced dephasing process in our 2D Bi2O2Se nanoplate system with decreasing electron density can be attributed to an enhancement in the Nyquist scattering. This is because the direct Coulomb interaction among the electrons is insensitive to the electron density in a 2D system. However, as the electron density decreases, the screening of cruising electrons reduces and thus the Nyquist scattering becomes stronger. 6 As seen in Figure 3d, the phase coherence length 𝐿𝐿𝜙𝜙 and the spin-orbit coupling length 𝐿𝐿𝑠𝑠𝑠𝑠 𝐿𝐿𝜙𝜙 and 𝐿𝐿𝑠𝑠𝑠𝑠. The extracted 𝐿𝐿𝑠𝑠𝑠𝑠 in the Bi2O2Se nanoplate is about 150 nm, which is shorter exhibit a crossover with decreasing electron density. Thus, the crossover between WAL and WL characteristics seen in Figure 3a can be attributed to the change in the relative values of than that in AlxGa1-xN/GaN 2DEG (~290 nm)28 and InSb nanowires (~250 nm)31. Thus, as we expected, the Bi2O2Se nanoplate exhibits a strong spin-orbit interaction and is an excellent candidate for realizing a helical electron system and a topological superconducting device for topological quantum computation. Here, we would like to note that due to the inversion symmetry presented in the material system, the SOI observed in the nanoplate is most likely of the Rashba type and should be able to be tuned by applying an electric field across the nanoplate. The weak tunability of 𝐿𝐿𝑠𝑠𝑠𝑠 by the back gate voltage observed in Figure 3(d) arises from the fact that due to the ultra-thin Bi2O2Se conduction channel and a single back gate voltage in the device structure, the voltage applied to the back gate could only efficiently tune the carrier density in the nanoplate, but not an electric field across it. A way to achieve a largely tunable SOI Bi2O2Se device is to employ a dual-gate structure by, e.g., adding a top gate to the present device. Currently, such a dual-gate Bi2O2Se device technology is under development. in this figure. At a low The WAL and WL characteristics of the magnetotransport in the Bi2O2Se nanoplate device are also studied at different temperatures. Figure 4a displays the measured low-field magnetoconductance of the device at 𝑉𝑉𝑏𝑏𝑏𝑏=30 𝑉𝑉 at temperatures ranging from 2 to 13 K. For clarity, the measured data at different temperatures are again vertically offset successively temperature, a sharp WAL characteristic magnetoconductance peak is seen in the vicinity of zero field. As temperature increases, the WAL peak is gradually suppressed and the low-field magnetoconductance evolves to show an overall WL dip, i.e., again a WAL-WL crossover behavior. The measured data at different temperatures are again fitted to the HLN formula of Eq. (1) and the results are presented by black solid lines in Figure 4a. Figure 4b shows the extracted characteristic transport lengths 𝐿𝐿𝜙𝜙 and 𝐿𝐿𝑠𝑠𝑠𝑠 from the fittings. It is shown that approximately 𝐿𝐿𝑠𝑠𝑠𝑠 does not change with increasing temperature. However, 𝐿𝐿𝜙𝜙 is strongly temperature dependent; it decreases with increasing temperature following a power law of 𝐿𝐿𝜙𝜙~𝑇𝑇−0.57, which is in a good agreement with our above assumption and analysis that the transport in the Bi2O2Se nanoplate is of the 2D nature and the dephasing dominantly arises from the electron-electron scattering processes with small energy transfers34. In correspondence to the temperature-driven WAL- 7 WL crossover characteristics of the magnetoconductance, the relative values of 𝐿𝐿𝜙𝜙 and 𝐿𝐿𝑠𝑠𝑠𝑠 also show a crossover with increasing temperature. At low temperatures, the quantum correction to the low-field magnetotransport in Bi2O2Se nanoplate is dominantly governed by the WAL characteristics originating from strong spin-orbit interaction (i.e., short spin- orbit interaction length). However, at high temperatures, the quantum correction to the low- field magnetotransport is dominantly governed by the WL characteristics due to thermally enhanced dephasing processes (i.e., short coherence length). In conclusion, we have detected the presence of a strong spin-orbit interaction in an ultra-thin Bi2O2Se nanoplate through the magnetotransport measurements at low temperatures. The low-field magnetoconductance of the nanoplate is measured at different carrier densities. The quantum correction to the magnetoconductance shows dominantly the WAL characteristics at high carrier densities and the WL characteristics at low carrier densities. The measurements are analyzed based on the HLN 2D diffusive transport theory and characteristic transport lengths, such as the phase coherence length 𝐿𝐿𝜙𝜙 and spin- relaxation length 𝐿𝐿𝑠𝑠𝑠𝑠 , in the Bi2O2Se nanoplate are extracted. It is shown that 𝐿𝐿𝑠𝑠𝑠𝑠 is relatively short (~150 nm) in the nanoplate, when compared with other semiconductor materials, representing a sign of strong spin-orbit interaction, and is roughly independent of the carrier density of the nanoplate. The magnetotransport measurements have also been performed for the Bi2O2Se nanoplate at variable temperatures. The quantum correction to the low-field magnetoconductance again shows a WAL-WL crossover with increasing temperature. The extracted spin relaxation length is also rather independent of temperature. However, the extracted phase coherence length decreases with increasing temperature following the power law of 𝐿𝐿𝜙𝜙~𝑇𝑇−𝛼𝛼 with 𝛼𝛼≈0.5 , implying that the dephasing dominantly originates from the small energy-transferred electron-electron scattering processes. Our experimental study lays out a solid foundation for the further exploration of layered semiconducting Bi2O2Se materials for possible applications in spintronics and topological quantum devices. References 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, 2004, 306, 666-669. 2. 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, 2005, 438, 197-200. 8 3. H. Peng, K. Lai, D. Kong, S. Meister, Y. Chen, X. L. Qi, S. C. Zhang, Z. X. Shen and Y. Cui, Nat Mater, 2010, 9, 225-229. 4. B. Radisavljevic, A. Radenovic, J. Brivio, V. Giacometti and A. Kis, Nature Nanotechnology, 2011, 6, 147-150. 5. L. Li, Y. Yu, G. J. Ye, Q. Ge, X. Ou, H. Wu, D. Feng, X. H. Chen and Y. Zhang, Nat Nanotechnol, 2014, 9, 372-377. 6. J. Wu, H. Yuan, M. Meng, C. Chen, Y. Sun, Z. Chen, W. Dang, C. Tan, Y. Liu, J. Yin, Y. Zhou, S. Huang, H. Q. Xu, Y. Cui, H. Y. Hwang, Z. Liu, Y. Chen, B. Yan and H. Peng, Nat Nanotechnol, 2017, 12, 530- 534. 7. J. Wu, Y. Liu, Z. Tan, C. Tan, J. Yin, T. Li, T. Tu and H. Peng, Adv. Mater., 2017, 29, 1704060. 8. R. M. Lutchyn, J. D. Sau and S. Das Sarma, Phys. Rev. Lett., 2010, 105, 077001. 9. Y. Oreg, G. Refael and F. von Oppen, Phys. Rev. Lett., 2010, 105, 177002. 10. M. T. Deng, C. L. Yu, G. Y. Huang, M. Larsson, P. Caroff and H. Q. Xu, Nano Lett., 2012, 12, 6414- 6419. 11. M. T. Deng, C. L. Yu, G. Y. Huang, M. Larsson, P. Caroff and H. Q. Xu, Scientific Reports, 2014, 4, 7261. 12. V. Mourik, K. Zuo, S. M. Frolov, S. R. Plissard, E. P. A. M. Bakkers and L. P. Kouwenhoven, Science, 2012, 336, 1003-1007. 13. S. M. Albrecht, A. P. Higginbotham, M. Madsen, F. Kuemmeth, T. S. Jespersen, J. Nygard, P. Krogstrup and C. M. Marcus, Nature, 2016, 531, 206-209. 14. S. Das Sarma, J. Fabian, X. D. Hu and I. Zutic, Solid State Commun., 2001, 119, 207-215. 15. I. Zutic, J. Fabian and S. Das Sarma, Rev. Mod. Phys., 2004, 76, 323-410. 16. H. Bluhm, S. Foletti, I. Neder, M. Rudner, D. Mahalu, V. Umansky and A. Yacoby, Nature Physics, 2010, 7, 109-113. 17. J. J. Lin and J. P. Bird, Journal of Physics-Condensed Matter, 2002, 14, R501-R596. 18. P. A. Lee and T. V. Ramakrishnan, Rev. Mod. Phys., 1985, 57, 287-337. 19. J. Wu, C. Tan, Z. Tan, Y. Liu, J. Yin, W. Dang, M. Wang and H. Peng, Nano Lett., 2017, 17, 3021-3026. 20. Y. Jing, S. Huang, K. Zhang, J. Wu, Y. Guo, H. Peng, Z. Liu and H. Q. Xu, Nanoscale, 2016, 8, 1879- 1885. 21. Y. Liu, M. Tang, M. Meng, M. Wang, J. Wu, J. Yin, Y. Zhou, Y. Guo, C. Tan, W. Dang, S. Huang, H. Q. Xu, Y. Wang and H. Peng, Small, 2017, 13, 1603572. 22. H. Boller, Monatsh. Chem., 1973, 104, 916-919. 23. L. Jiao, B. Fan, X. Xian, Z. Wu, J. Zhang and Z. Liu, J. Am. Chem. Soc., 2008, 130, 12612-12613. 24. H. Linke, P. Omling, H. Q. Xu and P. E. Lindelof, Superlattices Microstruct., 1996, 20, 441-445. 25. H. Linke, P. Omling, H. Q. Xu and P. E. Lindelof, Physical Review B, 1997, 55, 4061-4064. 26. D. Pan, D. X. Fan, N. Kang, J. H. Zhi, X. Z. Yu, H. Q. Xu and J. H. Zhao, Nano Lett., 2016, 16, 834- 841. 27. Q. Li, S. Huang, D. Pan, J. Wang, J. Zhao and H. Q. Xu, Appl. Phys. Lett., 2014, 105, 113106. 9 28. N. Thillosen, S. Cabañas, N. Kaluza, V. A. Guzenko, H. Hardtdegen and T. Schäpers, Physical Review B, 2006, 73, 241311. 29. L. B. Wang, J. K. Guo, N. Kang, D. Pan, S. Li, D. Fan, J. Zhao and H. Q. Xu, Appl. Phys. Lett., 2015, 106, 173105. 30. S. Dhara, H. S. Solanki, V. Singh, A. Narayanan, P. Chaudhari, M. Gokhale, A. Bhattacharya and M. M. Deshmukh, Physical Review B, 2009, 79, 121311. 31. I. van Weperen, B. Tarasinski, D. Eeltink, V. S. Pribiag, S. R. Plissard, E. P. A. M. Bakkers, L. P. Kouwenhoven and M. Wimmer, Physical Review B, 2015, 91, 201413. 32. S. Hikami, A. I. Larkin and Y. Nagaoka, Prog. Theor. Phys., 1980, 63, 707-710. 33. B. A. Assaf, T. Cardinal, P. Wei, F. Katmis, J. S. Moodera and D. Heiman, Appl. Phys. Lett., 2013, 102, 012102. 34. B. L. Altshuler, A. G. Aronov and D. E. Khmelnitsky, Journal of Physics C-Solid State Physics, 1982, 15, 7367-7386. Acknowledgements This work is supported by the Ministry of Science and Technology of China through the National Key Research and Development Program of China (Grant Nos. 2017YFA0303304, 2016YFA0300601, 2017YFA0204901, and 2016YFA0300802), the National Natural Science Foundation of China (Grant Nos. 91221202, 91421303, and 11274021). H.Q.X. also acknowledges financial support from the Swedish Research Council (VR). Author contributions H.Q. X. conceived and supervised the project. M.M. and S.H. fabricated the devices, carried out the electrical and magnetotransport measurements. C.T., J.W. and H.P. grew the materials. Y.J. participated in the device fabrication and measurements. M.M., S.H. and H.Q.X. analyzed the data and wrote the manuscript with contributions from all authors. All authors contributed to the discussion of the results and the interpretation of the experimental data acquired. Additional Information Supplementary Information accompanies this paper. Competing Interesting: The authors declare no competing financial interests. Reprint and permission information is available online. 10  Figure 1. (a) Crystal structure of layered Bi2O2Se. (b) Representative optical image of a few as-grown Bi2O2Se nanoplates on a growth substrate mica. (c) AFM image of an as-grown Bi2O2Se nanoplate on the growth substrate. The nanoplate has an atomically flat surface and a thickness of 6 nm (see the white solid line for the height profile measured by AFM along the dashed line across an edge of the nanoplate). (d) HRTEM image of an as-grown Bi2O2Se nanoplate. A well-defined alternative arrangement of dark and bright lattices (with a spacing of ~0.19 nm) is observed. Marked in a square dark background region are atomic positions of atoms Bi (violet dots), O (red dots), and Se (yellow dots). 11  Figure 2. (a) Optical image of the fabricated Bi2O2Se nanoplate device measured for this work. The device is made on a SiO2/Si substrate. The nanoplate has a length of ~10 µm, a width of ~3 µm and a thickness of ~8 nm. The four finger Ti/Au contacts made on the nanoplate have a width of ~800 nm and a pitch of ~2.6 µm. The scale bar is 10 μm. (b) Titled schematic view of the device and circuit setup for the magnetotransport measurements using lock-in technique. Here, the contacts are labeled as 1 to 4 and the Si substrate is used as the back gage. (c) Conductance of the device measured in the setup as shown in the inset as a function of back gate voltage Vbg at different temperatures. Here, the conductance G=Ids /V23 is plotted as a function of Vbg, where Ids is the current passing through the Bi2O2Se nanoplate channel and V23 is the voltage between contacts 2 and 3, measured with a constant source- drain bias voltage of Vds = V14 = 5 mV applied between contacts 1 and 4. 12  Figure 3. (a) Low-field magnetoconductance measured for the Bi2O2Se nanoplate device at various back gate voltage and a temperature of 2 K. Here, the curve measured at Vbg = 0 V is marked by a blue arrow and all other curves are successively vertically offset for clarity. The measurements show that a crossover from WAL to WL occurs when Vbg is swept from 30 to -14 V. (b) and (c) The same low-field magnetoconductance measurement data as in (a) for Vbg = 30 V and Vbg = -14 V. The black solid lines are theoretical fits to the experimental data. (d) Dephasing length 𝐿𝐿𝜙𝜙, spin relaxation length 𝐿𝐿𝑠𝑠𝑠𝑠, and mean free path 𝐿𝐿𝑒𝑒 extracted for the device as a function of back gate voltage Vbg from the measured data shown in (a). 13  14 Figure 4. (a) Low-field magnetoconductance of the Bi2O2Se nanoplate device measured at a fixed back gate voltage of Vbg = 30 V but different temperatures. The lowest curve is for the measurements at 2 K and all other curves are successively vertically offset for clarity. The black solid lines are theoretical fits to the experimental data. Here, a WAL–WL crossover is seen to occur with increasing temperature. (b) Dephasing length 𝐿𝐿𝜙𝜙, spin relaxation length 𝐿𝐿𝑠𝑠𝑠𝑠, and mean free path 𝐿𝐿𝑒𝑒 extracted for the device as a function of temperature T from the dephasing length 𝐿𝐿𝜙𝜙, showing 𝐿𝐿𝜙𝜙 ~ 𝑇𝑇−0.57. measured data shown in (a). The orange solid line is the power-law fit to the extracted Supplementary Materials for Strong spin-orbit interaction and magnetotransport in semiconductor Bi2O2Se nanoplates Mengmeng Meng, 1 Shaoyun Huang, 1, * Congwei Tan, 2 Jinxiong Wu, 2 Yumei Jing, 1 Hailin Peng2,* and H. Q. Xu1, 3,* 1Beijing Key Laboratory of Quantum Devices, Key Laboratory for the Physics and Chemistry of Nanodevices, and Department of Electronics, Peking University, Beijing 100871, China 2Center for Nanochemistry, Beijing National Laboratory for Molecular Sciences (BNLMS), College of Chemistry and Molecular Engineering, Peking University, Beijing 100871, 3Division of Solid State Physics, Lund University, Box 118, S-221 00 Lund, Sweden China. *Corresponding authors: Professor H. Q. Xu ([email protected]), Dr. Shaoyun Huang ([email protected]), and Professor Hailin Peng ([email protected]) Contents S1. Extraction of the carrier density n2D and the field-effect mobility μ S2. Extraction of the mean free path Le from carrier density n2D and mobility μ S3. Transfer characteristics of the Bi2O2Se nanoplate field-effect device at 2 K S4. Carrier density n2D, mobility µ, and mean free path Le extracted for the same Bi2O2Se nanoplate discussed in the main article at different temperatures 1 charge (Q) presented in the nanoplate is given by where the nanoplate is then obtained from S1. Extraction of the carrier density n2D and the field-effect mobility µ To extract the carrier density in a Bi2O2Se nanoplate, we employ the parallel plate capacitor model, which is a good approximation when the size of the nanoplate is much larger than the thickness of the dielectric. Based on this parallel plate capacitor model, the capacitance (C) can be evaluated from with Vbg being a voltage applied to the back gate and Vth the threshold pinch-off back gate voltage at which no conduction carriers are present in the nanoplate at zero temperature (or 𝐶𝐶=𝜀𝜀0𝜀𝜀𝑟𝑟𝐴𝐴𝑑𝑑 , (1) where 𝜀𝜀0 is the electric permittivity of vacuum, 𝜀𝜀𝑟𝑟 is the relative permittivity of SiO2 (𝜀𝜀𝑟𝑟= 4), A is the area of the nanoplate, and d is the thickness of the dielectric SiO2. The electric ∙∆𝑉𝑉, (2) 𝑄𝑄=𝐶𝐶∙∆𝑉𝑉= 𝜀𝜀0𝜀𝜀𝑟𝑟𝐴𝐴𝑑𝑑 ∆𝑉𝑉=𝑉𝑉𝑏𝑏𝑏𝑏−𝑉𝑉𝑡𝑡ℎ , (3) after neglecting thermal excited carriers at a finite temperature). The carrier density 𝑛𝑛2𝐷𝐷 in 𝑛𝑛2𝐷𝐷= 𝑄𝑄𝑒𝑒𝐴𝐴=𝜀𝜀0𝜀𝜀𝑟𝑟𝑒𝑒𝑑𝑑∙�𝑉𝑉𝑏𝑏𝑏𝑏−𝑉𝑉𝑡𝑡ℎ�=𝐶𝐶𝑔𝑔𝑔𝑔𝑒𝑒 ∙(𝑉𝑉𝑏𝑏𝑏𝑏−𝑉𝑉𝑡𝑡ℎ) , (4) where e is the elementary charge and 𝐶𝐶𝑏𝑏𝑔𝑔=𝐶𝐶/𝐴𝐴 is the capacitance per unit area. 𝜇𝜇= 1𝐶𝐶𝑔𝑔𝑔𝑔∙𝐿𝐿𝑊𝑊∙ 1𝑉𝑉𝑑𝑑𝑔𝑔∙𝑑𝑑𝐼𝐼𝑑𝑑𝑔𝑔𝑑𝑑𝑉𝑉𝑏𝑏𝑔𝑔 , (5) applied, and 𝑑𝑑𝐼𝐼𝑑𝑑𝑔𝑔𝑑𝑑𝑉𝑉𝑏𝑏𝑔𝑔 is the transconductance taken from the linear region of the transfer where L and W are the channel length (i.e., the distance between the source and drain contacts) and the channel width (i.e., the width of the nanoplate), Vds is a small source-drain voltage The field-effect mobility is extracted from the linear region of the transfer characteristic curve of the Bi2O2Se nanoplate device by the following equation, characteristics of the device. 2 S2. Extraction of the mean free path Le from carrier density n2D and mobility μ In the effective mass approximation, the Fermi velocity is given by In a conduction channel, the mean free path Le describes the mean distance that electrons travel between two collisions with scattering centers and phonons, and is obtained from 𝐿𝐿𝑒𝑒=𝑣𝑣𝐹𝐹𝜏𝜏 , (6) where τ is the momentum relaxation time, i.e., the time electrons travel between two collisions with scattering centers and phonons, and 𝑣𝑣𝐹𝐹 is the Fermi velocity of the carriers. 𝑣𝑣𝐹𝐹=ℏ𝑘𝑘𝐹𝐹𝑚𝑚∗= ℏ𝑚𝑚∗�2𝜋𝜋𝑛𝑛2𝐷𝐷 , (7) where ℏ is the reduced Planck constant, kF is the Fermi wave vector, and m* is the effective mass of the carriers. The momentum relaxation time 𝜏𝜏 can be evaluated from the mobility 𝜇𝜇=𝑒𝑒𝑒𝑒𝑚𝑚∗ . (8) 𝐿𝐿𝑒𝑒=ℏ𝜇𝜇𝑒𝑒�2𝜋𝜋𝑛𝑛2𝐷𝐷 . (9) through the relation Thus, the mean free path can be obtained from 3 S3. Transfer characteristics of the Bi2O2Se nanoplate field-effect device at 2 K In this supplementary note, we present in Figure S1 the transfer characteristics of the Bi2O2Se nanoplate field-effect device measured at 2K and describe, as an example, the procedure in which the carrier density, the field effect mobility, and the mean free path in the device are extracted. Figure S1. Source-drain current Ids vs. back-gate voltage Vbg (transfer characteristics) measured for the Bi2O2Se nanoplate device discussed in the main article at a source-drain bias voltage of Vds= 5 mV and a temperature of T=2 K, see the inset of Figure 2c in the main article for the measurement setup. The green curve is the plot for Ids in linear scale, while the red curve is the plot for Ids in logarithmic scale. The channel is of n-type. The black solid line is the linear fit to the transfer characteristics. The fitting line intersects with the horizontal axis, giving a channel pinch-off threshold voltage of Vth = -7.2 V. With this threshold voltage the nanoplate can be determined for Vbg in the region where the transfer characteristics show a good linear dependence on Vbg. The slope of the fitting line, together with the channel and the estimated capacitance per unit area Cgs =1.18×10−4 F/m2, the carrier density in length L=7.8 µm, the channel width W=3 µm, and Cgs= 1.18×10−4 F/m2, gives the field- effect mobility of µ= 1509 cm2/V·s in the Bi2O2Se nanoplate channel at 2 K. Using this mobility value and the carrier density extracted as a function of Vbg, the mean free path in the Bi2O2Se nanoplate channel at 2K can be estimated. The same procedure has been employed to extract the mobility, the carrier density and the mean free path for the device from the transfer characteristics measured at other temperatures. 4 S4. Carrier density n2D, mobility µ, and mean free path Le extracted for the Bi2O2Se nanoplate discussed in the main article at different temperatures In this supplementary note, the carrier density n2D, mobility µ, and mean-free path Le extracted for the Bi2O2Se nanoplate studied in the main article as a function of back-gate voltage Vbg at different temperatures, using the same procedure as described in the caption to Figure S1, are presented. Figure S2. (a) Sheet carrier density n2D as a function of back gate voltage Vbg at different temperatures, (b) carrier mobility µ as a function of temperature, and (c) carrier mean free path Le as a function of Vbg at different temperatures in the same Bi2O2Se nanoplate as studied in the main article, extracted using the same procedure as described in the caption to Figure S1. 5 

1703.04241

1

1703

Landau quantization of Dirac fermions in graphene and its multilayer

When electrons are confined in a two dimensional (2D) system, typical quantum mechanical phenomena such as Landau quantization can be detected. Graphene systems, including the single atomic layer and few-layer stacked crystals, are ideal 2D materials for studying a variety of quantum mechanical problems. In this article, we review the experimental progress in the unusual Landau quantized behaviors of Dirac fermions in monolayer and multilayer graphene by using scanning tunneling microscopy(STM) and scanning tunneling spectroscopy(STS). Through STS measurement of the strong magnetic fields, distinct Landau-level spectra and rich level splitting phenomena are observed in different graphene layers. These unique properties provide an effective method for identifying the number of layers, as well as the stacking orders, and investigating the fundamentally physical phenomena of graphene. Moreover, in the presence of a strain and charged defects, the Landau quantization of graphene can be significantly modified, leading to unusual spectroscopic and electronic properties.

cond-mat.mes-hall

cond-mat

Front. Phys. 12(4), 127208 (2017) DOI 10.1007/s11467-016-0655-5 REVIEW ARTICLE Landau quantization of Dirac fermions in graphene and its multilayers Long-jing Yin (殷隆晶), Ke-ke Bai (白珂珂), Wen-xiao Wang (王文晓), Si-Yu Li (李思宇), Yu Zhang (张钰), Lin He (何林)ǂ The Center for Advanced Quantum Studies, Department of Physics, Beijing Normal University, Beijing 100875, China Corresponding author. E-mail: ǂ[email protected] Received December 28, 2016; accepted January 26, 2017 When electrons are confined in a two-dimensional (2D) system, typical quantum–mechanical phenomena such as Landau quantization can be detected. Graphene systems, including the single atomic layer and few-layer stacked crystals, are ideal 2D materials for studying a variety of quantum–mechanical problems. In this article, we review the experimental progress in the unusual Landau quantized behaviors of Dirac fermions in monolayer and multilayer graphene by using scanning tunneling microscopy (STM) and scanning tunneling spectroscopy (STS). Through STS measurement of the strong magnetic fields, distinct Landau-level spectra and rich level-splitting phenomena are observed in different graphene layers. These unique properties provide an effective method for identifying the number of layers, as well as the stacking orders, and investigating the fundamentally physical phenomena of graphene. Moreover, in the presence of a strain and charged defects, the Landau quantization of graphene can be significantly modified, leading to unusual spectroscopic and electronic properties. Keywords Landau quantization, graphene, STM/STS, stacking order, strain and defect PACS numbers 1 2 Introduction ....................................................................................................................... 2 Landau quantization in graphene monolayer, Bernal bilayer, and Bernal trilayer ............ 5 Contents 2.1 Graphene monolayer ...................................................................................................... 5 2.2 Bernal-stacked bilayer .................................................................................................... 7 2.3 Bernal-stacked trilayer ................................................................................................... 9 3 Inherent 2D Dirac quantum properties ............................................................................ 11 3.1 Two-component Dirac-LLs .......................................................................................... 11 3.1.1 Two-component spinors ........................................................................................... 11 3.1.2 Localized Dirac-LLs ................................................................................................ 12 3.2 LL bending ................................................................................................................... 15 4 Stacking-dependent LL spectrum for multilayer graphene ............................................. 18 4.1 Stacking domain wall system ....................................................................................... 18 4.1.1 AB-BA bilayer soliton ............................................................................................. 19 4.1.2 ABA-ABC trilayer soliton ....................................................................................... 22 4.2 Twisted graphene system ............................................................................................. 26 4.2.1 Twisted bilayers ....................................................................................................... 26 4.2.2 Twisted trilayers ....................................................................................................... 29 5 Landau quantization in strained and defective graphene ................................................. 32 5.1 Electron–hole asymmetry............................................................................................. 33 5.2 Valley-polarized LLs ................................................................................................... 36 6 Conclusions and perspectives .......................................................................................... 39 Acknowledgements ................................................................................................................. 40 References .................................................................................................. 错误!未定义书签。 1 Introduction Graphene-a one-atom-thick film-consists of a honeycomb-like hexagonal carbon lattice that exhibits a truly two-dimensional (2D) nature [1–3]. Two equivalent carbon atoms-A and B-exist in each unit cell of the hexagonal lattice in graphene [Fig. 1(a)]. These atoms are strongly connected by in-plane covalent bonds, i.e., the σ bond hybridized between one s orbital and two p orbitals [4, 5]. The remaining unhybridized p orbital of each atom, which is perpendicular to the planar surface, can support the fourth valence electron of carbon, leading to the formation of a π-electronic band. The electrons filling the π bands can move freely in the graphene plane, as if they are relativistic particles [6, 7]. These π-electronic states are responsible for the electronic properties of graphene at low energies, whereas the σ-electronic states form energy bands far from the Fermi energy. In the low-energy range, the freely moving π electrons in graphene are described by the Dirac equation [8, 9], rather than the usual Schrödinger equation, because of the two-sublattice crystal structure. The electronic hopping between the neighboring sublattices leads to the formation of a conical energy spectrum, with the valence band and the conduction band touching at a point-like Fermi surface called the Dirac point, which yields two inequivalent points K and K′ at the corners of the hexagonal Brillouin zone (BZ) [Figs. 1(b) and 1(c)]. As a result, the quasiparticles in graphene exhibit the linear dispersion relationship E = ћkνF in the vicinity of the Dirac point [Fig. 1(d)], where the carriers behave as massless fermions that mimic the physics of quantum electrodynamics with a constant Fermi velocity νF ≈ 106 m/s [10, 11]. Fig. 1 Lattice and low-energy band structures of graphene. (a) Honeycomb lattice consisting of A and B sublattices. (b) BZ. (c) Energy dispersion. The Dirac cones are located at the points K and K′. (d) The linear band structure near the Dirac point. Reproduced from Refs. [4] and [5]. In addition to the special linear spectrum, graphene has other essential and unique electronic properties that are absent from conventional 2D electron gas systems (2DEGs). For instance, because of the bipartite honeycomb lattice, the wave function for a unit cell of graphene can be expressed with two components as follows: Ψ = aΨA + bΨB, where ΨA and ΨB are the wave functions at the A and B sublattices, respectively. The two-component representation for graphene is very similar to that of the spinor wave functions in quantum electrodynamics. The "spin" index, which is defined by the vector (a, b), indicates sublattices rather than the real spin of electrons and is usually referred to as pseudospin. Normally, electrons and holes are not connected and are described by separate Schrödinger equations in condensed-matter physics. However, owing to the two- component wave functions of the quasiparticles in graphene, the negatively and positively charged states are interconnected, exhibiting an extra electron–hole symmetry. Additionally, the electron with energy -E and the hole with energy E originating from the band branch of the same sublattice propagate in opposite directions. Consequently, the pseudospin is parallel to the momentum for electrons and antiparallel for holes in the same branch. This introduces the chirality of the Dirac fermions in graphene [6, 12], which is positive and negative for electrons and holes, respectively. In quantum electrodynamics, the chirality is defined as the projection of spin in the direction of momentum [13]. For graphene, this definition can be used, but the true spin is replaced by the sublattice pseudospin. The 2D chiral massless-Dirac-fermions give rise to the most interesting aspects of graphene, offering a perfect platform for exploring its amazing physical phenomena, such as chiral Klein tunneling [6, 12, 14, 15], exceptional ballistic transport [16–18], and the self-similar Hofstadter butterfly spectrum [19–24]. Another interesting feature of Dirac fermions is their unusual behaviors under magnetic fields [5, 25], which lead to various novel quantum phenomena, including the anomalous integral and fractional quantum Hall effects (IQHE and FQHE, respectively) [26–31]. In the presence of a magnetic field (B), the low-energy band structure of a 2D electron system develops into discretely dispersionless Landau levels (LLs) [32–34], which is responsible for the IQHE [25] and is also the theoretical basis for understanding the involved FQHE [35–38]. The cyclotron energies of Dirac fermions in graphene are scaled as , in contrast to the linear behavior for particles in semiconductor 2DEGs. This leads to large cyclotron energies, which, together with the small scattering [39], allows the IQHE to be observed at room temperature in graphene [40]. Furthermore, there is an additional level at zero energy (i.e., at the Dirac point) including both electron and hole states [41], and accordingly, the IQHE shows an unusual half-integer characteristic. The fully exposed electronic states provide unprecedented opportunities to directly probe 2D quantum phenomena, such as the Landau quantization, on the sample surface, not only in monilayer graphene but also in multilayer graphene [42]. In multilayer graphene, the crystal sheets are stacked on top of each other and are connected by weak van der Waals forces with different layer-stacking orders [4, 43]. The 2D nature of the quasiparticles even exists in four-layer graphene [44]. More importantly, the quasiparticles in graphene multilayers exhibit a strong layer number [45] and stacking-order dependency [46], leading to the distinct Landau quantized behaviors and rich quantum Hall physics [47–51]. This makes it possible to identify the number of layers and the stacking configurations of graphene by using the LL spectrum. One of the most powerful techniques for probing quantum phenomena such as LLs is scanning tunneling microscopy (STM) [52–54]. In scanning tunneling spectroscopy (STS) measurements, the LLs appear as a sequence of peaks in differential-conductance spectra [i.e., spectra of dI/dV, which is proportional to the local density of states (DOS)] and have been observed in many systems, including graphene [55–58], conventional 2DEGs [59, 60], and topological insulators [61–64]. This method is a local and harmless way to observe the abundant microscopic physics in graphene systems. In this article, we review the recent experimental investigations of the Landau quantization of Dirac fermions in monolyer and multilayer graphene using STM and STS. This review is organized as follows. In Section 2, we introduce the distinct LL spectra for monolayer graphene, Bernal-stacked bilayers, and Bernal- stacked trilayers. Section 3 focuses on the experimental imaging of the two-component Dirac-LLs and the LL bending in a gapped graphene monolayer and bilayer, respectively. In Section 4, the stacking order-dependent Landau quantization in multilayer graphene is described in detail. Section 5 discusses the unconventional Landau quantized behaviors in strained and defective graphene. Finally, in Section 6, we summarize and conclude the article and present prospects for fully understanding the nature of novel quantum Hall phases and detecting new physics in graphene. B2 Landau quantization in graphene monolayer, Bernal bilayer, and Bernal trilayer The truly 2D nature of quasiparticles not only exists in monolayer graphene but also extends to stacked layers of graphene sheets. Because of the exrta degree of freedom of the layer, graphene and its multilayers exhibit complex phenomena and unusual properties [65]. The most energy-stable multilayer graphene is Bernal-stacked (or AB-stacked) graphene, wherein one set of the sublattice atoms of the top layer is immediately above the atoms of the bottom layer and the other set is at the centers of the hexagonal voids in the bottom layer [66]. In graphene monolayers, the interaction of the two equivalent sublattices results in a linear band structure and gives the quasiparticles a massless Dirac fermion property near the charge neutrality with a chiral degree of l = 1 and a Berry phase of π. For a Bernal (AB-stacked) bilayer, the low-energy electronic states retain the chiral Dirac characteristic but have a quadratic dispersion (i.e., massive), and the quasiparticles have a chiral degree of l = 2 and a Berry phase of 2π [66–69]. Interestingly, Bernal trilayer (ABA) graphene exhibits the coexistence of massless and massive Dirac fermions. The various types of quasiparticles give rise to disparate Landau quantized behaviors in graphene monolayers, Bernal bilayers, and Bernal trilayers, which have been frequently observed on different substrates, especially on graphite surfaces [45, 58, 70, 71]. Graphite-a three-dimensional (3D) allotrope of carbon-consists of stacks of graphene layers with Bernal layer-stacking that are weakly coupled by van der Waals forces [43]. The surface monolayer/few-layer graphene may electronically decouple from graphite when the separation between the topmost graphene layers and the graphite is larger than the equilibrium distance of ~0.34 nm [45, 57, 72] or when the sheets have a large rotation angle with respect to the substrate after exfoliation [73]. Hence, it is possible to detect the Landau quantization of 2D Dirac fermions for graphene monolayers and multilayers on the surfaces of such 3D systems [74]. Here, we introduce the distinct LL spectra for decoupled graphene monolayers, Bernal bilayers, and Bernal trilayers on graphite and Rh substrates, which were obtained by STS under high magnetic fields. 2.1 Graphene monolayer Figure 2(a) shows a series of differential-conductance spectra for a graphene monolayer on a graphite surface, under various magnetic fields ranging from 0 to 8 T [45]. The corresponding atomic-resolution STM image shows the typical honeycomb lattice structure [Fig. 2(e)]. The LLs developed as the magnetic-field increased, and the LL peaks were well-resolved up to n = 5 in both electron and hole sectors (here, n is the LL index). Moreover, a pronounced DOS peak appeared in the vicinity of the Fermi energy under various fields, corresponding to the landmark zero-energy LL of graphene. For massless Dirac fermions in a graphene monolayer, the LL energies En depend on the square root of both the level index n and the magnetic field B, including a field-independent E0 for the zero-energy state [55–57]: where e is the electron charge, is Planck's constant, vF is the Fermi velocity, and E0 is the energy of the Dirac point. The unusual appearance of a zero-energy level (n = 0) is a direct consequence of quasiparticle chirality. Each LL in graphene is fourfold degenerate, including the zero-energy state, owing to the usual spin (1) 20sgn()2, 0, 1, 2...nFEnenBEndegeneracy and the two inequivalent corners of the BZ K and K′, which gives rise to valley degeneracy. Theoretically, each filled single-degenerate LL contributes one conductance quantum e2/h towards the Hall conductivity observed in the quantum Hall effect (QHE) [32]. In monolayer graphene, the Hall conductivity is described by σxy = (4N + 2) e2/h with the absence of a zero-σxy plateau [26, 27], in contrast to the case of conventional 2DEGs. The analysis of the data shown in Figs. 2(b) and 2(c) demonstrates that the sequence of observed LLs is described quite well by Eq. (1). The linear fit of the experimental results of the LL energies to Eq. (1), which shows electron–hole symmetry, yields a Fermi velocity of vF = (1.207 ± 0.002) × 106 m/s [45]. Similar LL spectra of graphene monolayers were observed on SiC substrates [55, 56] and on Rh foil [75, 76], with the Fermi velocity of vF ≈ 1.1 × 106 m/s in both cases. Fig. 2 Landau quantization in monolayer graphene. (a) LL spectra of a graphene monolayer on a graphite surface under various fields B. The LL indices are marked. All the spectra are shifted to make the n = 0 LL remain at the same bias. (b) LL peak energies for 1–8 T versus sgn(n)(nB)1/2. The solid line is a linear fit of the data with Eq. (1). Inset: energies of the n = 0 LL at different B. (c) Energies of the LL peaks for different levels n show the square-root dependence on the field B. The solid lines are the fits with Eq. (1). (d) Schematic of LLs in monolayer graphene. (e) Atomic-resolution STM image with the honeycomb lattice structure of single-layer graphene. Reproduced from Ref. [45]. In the presence of strong magnetic fields, the electrons are more spatially localized, and the electron– electron interaction is expected to be enhanced [5]. The enhanced interaction lifts the LL degeneracies and generates gaps in graphene [56, 77], which can be directly probed in the LL spectra. In the monolayer graphene grown on Rh foil [78], we find that the energy splitting of the LL0 is ~5.5 meV under a 5-T field and increases to ~8.8 meV under a 7-T field. Similar energy splits are observed in LL-1 and LL1 under a 7-T field, as shown in Fig. 3. The energy splitting of LLs with higher orbital indices is not observed in our experiment because the line width of the LLs increases with the energy. This behavior is related to the quasiparticle lifetimes, which decrease with the increasing energy difference from the Fermi level [56, 57]. Fitting the splitting energies of the LL0 to a Zeeman-like dependence, E = gBB, yields a g-factor of g ≈ 21. We attribute this energy splitting to the lifting of the valley degeneracies because the effective g-factor of the valley splitting in graphene is measured to be ~18.4 [56]. The observation of interaction-driven gaps, which increase with the magnetic field, clearly indicates that the electron–electron interaction in graphene is enhanced as the field increases. Fig. 3 Tunneling spectra of monolayer graphene on Rh foil under different magnetic fields. The LL peak indices are labeled. The slight splitting of LL0 and LL±1 is observed under a 7-T field. Reproduced from Ref. [78]. 2.2 Bernal-stacked bilayer Usually, in Bernal (AB-stacked) bilayer graphene, the A/B-atom asymmetry generated by the two adjacent AB-stacked layers leads to a triangular lattice, which is observed via microscopy. Therefore, Bernal bilayers (as well as multilayers) exhibit triangular contrast rather than the honeycomb lattice, as shown in the STM image [45, 72]. The bright spots of the triangular lattice, as shown in the inset of Fig. 5(a), are the sites on the top layer where one sublattice lies above the center of the hexagons in the second layer. In neutral Bernal bilayers, the LL spectrum of the massive Dirac fermions takes the form (2) where ωc = eB/m* is the cyclotron frequency, and m* is the effective mass of the quasiparticles. This LL sequence is linear in B, similar to the standard 2DEGs, but has an extra zero-energy LL, which is independent of B. For orbital index n ＞ 1, the LLs are fourfold degenerate, similar to those in monolayer graphene, while the n = 0 and n = 1 LLs are further degenerate, resulting in an eightfold degenerate zero-energy state (Fig. 4). Therefore, the Hall conductivity in bilayer graphene exhibits plateaus at integer values of 4e2/h and has a double 8e2/h step between the hole and electron states across zero density [47, 48, 79]. However, in the supported bilayers, the substrate easily induces an interlayer bias, breaking the inversion symmetry of the two adjacent layers [77, 80]. Consequently, the degeneracy of the zero-energy LL is partially lifted, and a bandgap is opened in the c(-1), 0,1,2...nEnnnFig. 4 Schematic of Landau quantization in bilayer graphene with and without an energy gap. LLs are indexed by the orbital and valley indices: n and ξ. The eightfold-degenerate zero-energy LL splits into two valley- polarized quartets under an interlayer bias. Reproduced from Ref. [77]. low-energy bands [81–84]. Then, there are two DOS peaks located at the edges of the energy gap (Eg) in the dI/dV spectra, even under zero field, as shown in Fig. 5. With an increasing magnetic field, the DOS peaks at the gap edges become two valley-polarized quartets, i.e., LL(0,1,+), and LL(0,1,-) [85], which are almost independent of B, and the spectra develop into a sequence of well-defined LL peaks showing a linear field dependence. These features are the fingerprints of the quantized behavior of massive Dirac fermions in gapped graphene bilayers. Fig. 5 Landau quantization in bilayer graphene. (a) LL spectra of a graphene bilayer on a graphite surface under various fields B. The LL indices are labeled. Inset: atomic STM image of a bilayer showing a triangular lattice. (b) LL peak energies plotted against ±(n(n-1))1/2 B. The solid lines are the fits of the data with Eq. (3). Inset: schematic of the LLs in bilayer graphene with a finite gap. (c) Fan diagram of the LLs energies as a function of the magnetic field. The solid lines represent fits of the data with Eq. (3). Reproduced from Ref. [45]. For a gapped graphene bilayer, the LL sequences can be described as follows [86, 87]:  (3) where EC is the energy of the charge-neutrality point (CNP), U is the interlayer bias, and ξ = ± are valley indices. Normally, z = << 1 for B ≤ 8 T and U ≈ Eg when U < . According to the fit between the experimental data and Eq. (3) shown in Figs. 5(b) and 5(c), we can determine Eg and m*. For the bilayer shown in Fig. 5, Eg ≈ 40 meV and m* = (0.039 ± 0.002)me (me is the free-electron mass). We observed various bilayer graphene samples on the graphite substrate, with Eg ranging from 10 to 100 meV and m* of 0.03–0.05me. Both the effective mass of the massive Dirac fermions and the bandgap agree well with the range of values reported previously for Bernal graphene bilayers on different substrates [67, 68, 77, 88]. In AB-stacked bilayers, the wave functions for one valley (+) of the lowest LL are mainly localized on the B sites of the top layer, and the wave functions for the other valley (-) of the zero-energy LL are on the A sites of the bottom layer. Therefore, the signal of LL(0,1,+) (mainly localized on the first layer) is far stronger than that of LL(0,1,-) (localized on the second layer) in the spectra, as the STS predominantly probed the DOS of the electrons on the top layer [72, 77], as shown in Fig. 5(a). This feature can be considered as another signature of gapped bilayer graphene and can be used to directly identify the bilayer region on the nanoscale. 2.3 Bernal-stacked trilayer According to tight-binding calculations that include only nearest-neighbor intralayer and nearest-layer coupling, the low-energy spectrum for Bernal trilayer can be treated as the combination of those for a graphene monolayer and a Bernal bilayer [89, 90]. Consequently, the Landau quantization of an ABA trilayer is a superposition of two sequences of massless and massive Dirac fermions [91, 92]. The dI/dV spectra measured in trilayer graphene under various magnetic fields exhibit a sequence of LL peaks of both massless and massive Dirac fermions, as shown in Fig. 6. In Fig. 6(a), some of the LL peaks depend on the square root of the magnetic field (massless-type), as shown in Fig. 6(b). The other LLs exhibit a linear-field dependence (massive-type), as shown in Fig. 6(c). In the trilayer sample of Fig. 6, the field dependence of the LLs corresponding to the massive Dirac fermions in both electron and hole sectors is extrapolated to the same zero-field value, suggesting that the eightfold degeneracy of the zero-energy LL in the bilayer-like sub-bands is not lifted. For convenience, in this case, the LL spectrum of the massive Dirac fermions can be expressed in the form [70] (4) A good fit of the bilayer-like LL energies in Fig. 6(a) with Eq. (4) is shown in Fig. 6(c). Furthermore, the ABA trilayers, having parabolic sub-bands containing a finite bandgap of ~10 meV and two valley-polarized quartets under B, are observed in our STM measurements. However, the linear sub-bands are always gapless. As a result, Bernal trilayer graphene cannot open a bandgap under low-energy excitation [93–95]. Theoretically, the CNPs of the monolayer graphene-like and Bernal bilayer-like sub-bands in the ABA trilayers are at the same energy, according to the simplest approximation [Figs. 6(d) and 6(e)]. In the experiments, there is a slight energy difference of approximately 10–30 meV between the zero-energy states of massless and massive Dirac ),1)(2(,2...4,3,2,4)2/()]1([)(C1C022Cz/UξEE/ξUEEn/ξzUUnnEEcn2/cttc0sgn()(1), 0,1,2....nEnnnEnfermions [45]. Such a difference is very reasonable when the other non-nearest neighbor hopping parameters affect the band structure of the Bernal trilayer. Fig. 6 Landau quantization in Bernal trilayer graphene. (a) LL spectra of a graphene trilayer on a graphite surface under various fields B. The LL indices of massless and massive Dirac fermions are indicated by blue and black numbers, respectively. (b, c) LL peak energies of massless and massive Dirac fermions versus sgn(n)(nB)1/2 and sgn(n)(n(n+1))1/2B, respectively. Schematic low-energy band structure of the graphene trilayer around the point K under B = 0 (d) and B ≠ 0 (e). Reproduced from Ref. [45]. The Fermi velocity of massless Dirac fermions and the effective mass of massive Dirac fermions in ABA trilayers can be obtained by fitting the LL peaks to corresponding theoretical formulas, as shown in Figs. 6(b) and 6(c). Interestingly, the value of vF in different Bernal trilayers can differ by more than 30%, similar to that observed in different graphene monolayers, which ranges from 0.79 × 106 to 1.21 × 106 m/s [55–57]. Additionally, a strong correlation between vF and m* is observed: vF generally increases as m* decreases, as shown in Fig. 7(a). Thus, there is a common origin that simultaneously affects vF and m* in the ABA trilayers. This behavior can be explored by calculating the band structure of the Bernal trilayer with different hopping parameters, as vF and m* depend sensitively on the hopping parameters. For example, the increase in the nearest-neighbor intralayer coupling in the ABA graphene can lead to opposite variations in vF and m*; i.e., m* decreases while vF increases, as shown in Fig. 7(b). This effect can quantitatively explain the experimental results. Fig. 7 (a) The vF of massless Dirac fermions and the m* of massive Dirac fermions obtained in different Bernal trilayers. (b) Low-energy band structure of a Bernal graphene trilayer with different nearest-neighbor intralayer hopping strengths. Reproduced from Ref. [45]. 3 Inherent 2D Dirac quantum properties The 2D chiral Dirac fermion system of graphene provides a perfect platform for theoretically [96, 97] and experimentally [98–101] identifying and examining extraordinary and fundamental physical problems, which are unachievable in 3D structures. The fully surface-exposed electronic states allow the internal quantum properties of the quasiparticles in graphene to be directly imaged by microscopic methods, such as scanning- probe techniques [53, 54]. We show that by using STM and STS with high space- and energy-resolution simultaneously, the two-component characteristic of Dirac fermions and the LL bending are spatially visualized in gapped graphene monolayers and bilayers, respectively. 3.1 Two-component Dirac-LLs Because of the bipartite honeycomb lattice in graphene [4–6], which has two distinct sublattices (A and B), the wave functions describing the low-energy excitations near the Dirac points in monolayer graphene are two-component spinors. This two-component nature can be manifest in localized LLs, whose degeneracy is lifted by the Coulomb potential or interactions. Therefore, with the high energy and spatial resolution of STM and STS, it is possible to image the two components of the spinors in a uniform graphene system. 3.1.1 Two-component spinors The two-component spinors of the two Dirac cones (K and K′) in graphene have the following form: , . (5) Here is defined as the angle of the wave vector in the momentum space. The two-component representation, which mathematically resembles that of a spin [102], corresponds to the projection of the electron wave function on the A and B sublattices. A site energy difference of 2 between the sublattices can break the inversion symmetry of graphene and lift the energy degeneracy of the A and B sublattices [96, 103], as shown in Fig. 8(a). This generates a gap of E = 2 at the Dirac points, as shown in Fig. 8(b), which was observed for a graphene monolayer on top of SiC [88], graphite [30, 57, 104], and hexagonal boron nitride [19, 105]. The gap, which usually ranges from 10 meV to several tens of millielectron volts, can result in valley-contrasting Hall transport in graphene monolayers [96, 105]. In the quantum Hall regime, the broken symmetry of the graphene sublattices shifts the energies of the n = 0 LL in the K and K′ valleys in opposite directions and thus splits the n = 0 LL into the 0+ KAKB112τiKieK'B-K'A112τiK'iearctanτ,yττ,xqθqyτ,xτ,τq,qqand 0- LLs (here, = + and - denote the K′ and K valleys, respectively) [57, 104], as schematically shown in Fig. 8(c). Generally, the wave functions of the LLs in graphene are given by [5, 106, 107] , , (6) where , and , with vF being the Fermi velocity and B being the magnetic field (here, n is the usual LL wave function). For n = 0, and ; hence, only the second components of the spinors are nonzero, and we have and . This indicates that we can detect the 0- LL, i.e., the spinor , only on the B sites and detect the 0+ LL, i.e., the spinor , only on the A sites. Fig. 8 Electronic band structure and Landau quantization for a gapped graphene monolayer. (a) Schematic diagram of a graphene monolayer with a staggered sublattice potential breaking the inversion symmetry. The A- and B-sites are denoted by blue and red balls, respectively. (b) Energy spectrum of a graphene monolayer with broken inversion symmetry. (c) Schematic LLs and DOS of a gapped graphene monolayer in the quantum Hall regime. The peaks in the DOS correspond to the LLs n. Reproduced from Ref. [108]. 3.1.2 Localized Dirac-LLs Figures 9(a) and 9(d) show representative STM images of the decoupled graphene layer on SiC and graphite substrates [108]. The spectra of the graphene sheets, which were recorded under a magnetic field of 8 T [Figs. 9(c) and 9(f)], on both the SiC and graphite substrates exhibit the Landau quantization of Dirac fermions, as expected for a gapped graphene monolayer [45, 55, 57]. The Fermi velocities for the graphene sheets on the SiC and graphite substrates are estimated to be vF = (0.79 ± 0.03) × 106 and (0.84 ± 0.03) × 106 m/s, respectively. A notable feature of the tunneling spectra is the splitting of the n = 0 peak and its sensitive dependency on the recorded positions depicted in Figs. 9(c) and 9(f). The splitting of the n = 0 peak, ~20 mV, nλ-1sincosnnnAnnnBαψn==αψ------1cossin++++nnnBnnnAαψn==αψ222tan()ΔΔ()0- ±BBα=hωnnhωn+nnsgnsgn22/BFevB00-α=0+α=200000--A-Bψ==ψ=00000BAψ==ψ=++0K0'Kis attributed to a gap caused by the substrate potential breaking the inversion symmetry [55, 57]. The spectra recorded at different positions, as shown in Figs. 9(c) and 9(f), indicate that the 0- LL is significant only on the B sites and that the 0+ LL is pronounced only on the A sites. At the center of the hexagons of the graphene sheets, for example, at the green dots in Figs. 9(d) and 9(e), the observed intensities of the 0+ and the 0- LLs are almost identical, as shown in Figs. 9(c) and 9(f). This feature is related to characteristics of the internal structure of the two-component spinors of the 0- and 0+ LLs. The following results demonstrate that the splitting of the n = 0 LL is a direct consequence of this two-component nature. Fig. 9 STM images and STS spectra of gapped graphene sheets. (a, d) STM images of graphene on a SiC (000- 1) terrace and on HOPG, respectively. (b, e) Magnified atomic-resolution topographies for the black frames in (a) and (d), respectively. Here, the bright and dark spots represent the A- and B-site atoms, respectively. (c, f) dI/dV spectra obtained at different positions, as indicated by the different colors, in (b) and (e), respectively. The black arrows in both panels denote the position of the CNP of the topmost graphene sheets under zero magnetic field. The LL indices are marked. Reproduced from Ref. [108]. Figures 10(a) and 10(b) show differential-conductance maps for graphene on a SiC substrate under an 8-T field at the bias voltages of the 0+ and 0- LLs, respectively. The maps reflect the spatial distribution of the local DOS at the bias voltages. Both the maps exhibit triangular contrasting, as indicated the pronounced asymmetry of the 0+ and 0− LLs on the sublattices. However, there is a very important difference between the two maps. The bright spots in the conductance map of the 0- LL correspond to the dark spots of the triangular lattice, i.e., the B sites, in the STM image, whereas the bright spots in the map of the 0+ LL correspond to the bright spots of the triangular lattice, i.e., the A sites, in the STM image. At a fixed energy, the local DOS at the position r is determined by the wave functions, according to . Therefore, the maps shown in Figs. 10(a) and 10(b) reflect atomic-resolution images of the two-component Dirac-LLs. Theoretically, the spinor of the 0+ (0-) LL only has a non-zero component on the A (B) sites, which is qualitatively consistent with the observed large asymmetry of the 0+ and 0- LLs on the sublattices. 2() ()rr Fig. 10 Conductance maps of the gapped graphene monolayer on SiC at different energies. (a–e) Conductance map recorded at the bias voltages of the 0+ LL (Vb = 65.5 mV), 0- LL (Vb= 45 mV), +1 LL (Vb = 108 mV), -1 LL (Vb = -22 mV), and +3 LL (Vb = 170 mV), respectively. The honeycomb structure of graphene and the atomic-resolution STM image are overlaid onto the maps. (f) Vertical line-cuts of the conductance maps of the 0+, 0-, -1, +1, and +3 LLs along the A and B atoms. The curves are offset vertically for clarity, and the zero lines for these curves are denoted by dashed lines. Reproduced from Ref. [108]. Theoretically, sin2(an/2)/cos2(an/2) which reflects the asymmetry between the amplitudes of the A-site and B-site components of the spinors. Figures 10(c) and 10(d) show conductance maps for graphene on a SiC substrate under an 8-T field at the bias voltages of the n = +1 and n = -1 LLs, respectively. Both the maps exhibit triangular contrasting, and the amplitude of the n = +1 (n = -1) LL on the A (B) sites is clearly stronger than that for the B (A) sites. However, the asymmetry between the amplitudes of the A-site and B-site components of the spinors for the n = +1 and n = -1 LLs is far weaker than that for the 0+ and 0- LLs, as shown in Figs. 10(a)–10(d). This n asymmetry decreases as n increases, according to Eq. (6). Therefore, Fig. 10(e) shows almost honeycomb contrasting in the conductance map recorded at a bias voltage of 170 mV (n at 170 mV is estimated to be 3). The vertical line-cuts of the conductance maps of the 0+, 0-, -1, +1, and +3 LLs along the A and B atoms in Fig. 10(f) also depict the asymmetry between the amplitudes of the A-site and B-site components.  3.2 LL bending Under a strong magnetic field B, the low-energy band structure of graphene is divided into dispersionless LLs and causes insulating behavior in the bulk material (see Fig. 11), while the confining potential at the edges [104, 109, 110] of the system bends the discrete LLs to form dispersive edge states that carry charge carriers in the QHE [111–113]. Hence, the LL bending is a fundamental effect and is significant for fully understanding the nature of the graphene edge states in novel quantum Hall phases such as the symmetry-protected quantum spin Hall state [114, 115] and exotic ferromagnetic phases [116, 117]. Because the electrons reside at the graphene surface, in contrast to the case of semiconductor-based 2DEGs, it should be possible to directly probe the level bending at the edges and perform systematic studies of the edge states for testing theoretical ideas of quantum Hall edge physics [118]. Numerous works have addressed this subject, mainly using optical and transport measurements [119–121]. Direct experimental observation of the LL bending can be achieved by STM and STS at different edge terminations of graphene on the graphite surface [72]. Fig. 11 LL structures of the 2DEGs in a finite-size sample. The LLs are dispersionless in the bulk material but dispersive near the sample edges. There are two possible (perfect) edge terminations-zigzag and armchair (see Fig. 12)-in graphene [122, 123], and the edge orientations strongly affect the electronic structures of graphene sheets [124–126]. The zigzag edge is predicted to host surface states with penetration into the bulk [127, 128] and has attracted much attention [129–132] because such surface states are believed to be closely related to the bandgap opening [128, 133, 134], magnetic order [135], and exceptional ballistic transport [16]. In the quantum Hall regime, both the zigzag and armchair edges can bend LLs to generate dispersive edge states, even in bilayer graphene [136– 138]. Graphene layers such as bilayers with different edge terminations (Fig. 12) can be identified by preforming STM and STS under magnetic fields, as discussed in Section 2. Figure 12(b) shows typical and well-defined LL spectra of gapped bilayer graphene on a graphite surface under various fields B [72]. The high-quality sample with atomically sharped edges and the ultralow random potential fluctuations due to substrate imperfections on the graphite allow us to directly probe the LL bending near the edges.  Fig. 12 Bilayer graphene with different edge terminations. (a) STM topographic image of a bilayer graphene near the sample edge on a graphite surface. Inset (upper): atomic-resolution image showing the triangular contrast. Inset (lower): height profile along the black line across the edge. (b) STS spectra of the graphene bilayers recorded away from the edges under various fields B. The LL peak indices are labeled. STM atomic images of (c) a zigzag bilayer edge and (d) an armchair bilayer edge. The insets show 2D Fourier transforms of the images. (e) Schematic of the AB-stacked bilayer graphene with zigzag and armchair edges. The green dots represent a set of sublattices imaged via STM topography. Reproduced from Ref. [72]. Figure 13 summarizes the STM results measured near two bilayer graphene edges in the quantum Hall regime. The expected LL bending at both the zigzag and armchair edges are clearly observed. Away from the edges, the LL spectra follow the sequence of massive Dirac fermions in gapped bilayers [Figs. 13(a)–13(d)]. Approaching the edges, the DOS peaks for the LLs become weak, and the LLs are shifted away from the CNP, as shown in Figs. 13(e) and 13(f). The local DOS measured at position r with a fixed energy is determined by the wave functions according to , while the wave functions of the LLs have a spatial extent of ~ ( is the distance from the edge) [61, 104]. Therefore, there is an important contribution from the bulk states, even for the LL spectra measured near the edges. Moreover, the wave functions of LLs with higher indices have greater spatial extents, as shown in the inset of Fig. 13(f). Consequently, the amplitude of the high-index LL peaks decreases more slowly than that of low-index LL peaks [Fig. 13(e)], and the bending of the low-index LLs seems stronger than that of the high-index LLs [Fig. 13(f)] (because of the greater contribution from the bulk states to the higher LLs). We also observed the LL bending under different magnetic fields, as shown in Fig. 14. 2() ()rr2BNl/BleB Fig. 13 LL bending at the bilayer graphene edges. (a, c) Spatial variation of the LL spectra near the zigzag (a) and armchair edges (c) of bilayer graphene under a 7-T field. The blue and red arrows indicate the spatial evolution of the LL(0,1,−) and LL(0,1,+) peaks. (b, d) LL spectra maps measured near the zigzag and armchair edges, respectively, under a field of 7 T. The quasi-localized surface states of the zigzag edge are indicated by green arrows in (a) and (b). (e) Evolution of the LL peaks under a 7-T field near the armchair edge on the conduction-band side. Inset: LL peak heights extracted from (e) as a function of the distance from the edge. (f) LL bending as a function of the distance around the armchair edge under an 8-T field. Inset: calculated probability densities for the wave functions of LL(0,1,+), LL2, and LL3 under an 8-T field. (g) Shift lengths of LL(0,1,+) and LL2 under different fields B. The solid dots (open dots) correspond to the data for the armchair edges (zigzag edges). The dashed lines show the average values of ∼1.4lB and ∼2.0lB for LL(0,1,+) and LL2, respectively. Reproduced from Ref. [72]. The shift length of the LL bending around the edges is theoretically predicted to be of the magnetic length [112, 136]. Figure 13(g) summarizes the measured shift length under different fields B around both the zigzag and armchair edges. Here, we observe that the shift length depends on neither the magnetic fields nor the edge types and it is of the magnetic length. However, the shift length appears to be dependent on the orbital index: the estimated shift lengths for LL(0,1,+) and LL2 are approximately 1.4lB and 2.0lB, respectively. We obtained the average value of the bending energy for the lowest LL(0,1,+): ~65 meV (see Fig. 14). This energy scale is approximately identical to the depth of the confining potential well at the sample edge [137, 139]. Fig. 14 LL bending for the armchair edge measured under various fields B. The dashed lines are the energy positions of CNPs in the bilayers. Reproduced from Ref. [72]. 4 Stacking-dependent LL spectrum for multilayer graphene Although the most common multilayer graphene is AB-stacked, other natural stable layer configurations exist in the 2D crystals [66]. Structurally, to form other configurations, one graphene sheet of the multilayers can be shifted on the atomic scale along the lattice orientation (usually in the armchair direction) or twisted at a finite angle with respect to the adjacent layers. In the former case, degenerate stacking orders related to the AB-register are generated, such as BA-stacking in bilayers and rhombohedral (ABC) stacking in multilayers [140, 141]. For the latter, it introduces a stacking-misorientation structure, that is, twisted graphene, forming Moiré superlattices [142]. In both cases, the electronic properties of the graphene layers can be modified dramatically and depend sensitively on the stacking orders [46, 143–145]. In this section, we present the characteristic features of the stacking-dependent LL spectrum in graphene bilayers and trilayers. 4.1 Stacking domain wall system When multilayer graphene partly transforms from one stacking order to another configuration, one- dimensional (1D) strain soliton-like domain walls grow in the transition boundary [146]. For example, the region of the transition between AB- and BA-stacked bilayer graphene domains forms an AB-BA domain wall [147–149], and similarly, graphene trilayers have ABA-ABC stacking solitons [150]. To realize such domain walls, one graphene layer should be shifted one C-C bond length along the armchair direction with respective to the adjacent layers (see Fig. 15). If the displacement vector is parallel to the domain wall, a shear-type soliton is produced [Fig. 15(a)], whereas if it is perpendicular to the domain wall, a tensile-type one is produced [Fig. 15(b)]. Owing to the varied interlayer stacking around the domain walls, the AB-BA and ABA-ABC stacking solitons can be clearly visualized by STM [73, 150], as shown in Figs. 15(c) and 15(d), by STEM [147] or by infrared nanoscopy [151]. The fascinating electronic and optical properties in these 1D solitons have been demonstrated theoretically [152–154] and experimentally [73, 151, 155, 156], including the topologically protected valley Hall edge states in AB-BA bilayer graphene domain walls [73, 155]. Fig. 15 Shear and tensile stacking domain wall solitons. (a, b) Schematics of the shear- and tensile-type strain solitons, respectively. (c, d) Atomic STM images of the shear and tensile domain walls, respectively. The arrows indicate the displacement vectors. Reproduced from Refs. [150] and [151]. 4.1.1 AB-BA bilayer soliton An AB-BA bilayer domain wall ~8 nm in width on graphite surface were studied [Fig. 16(a)]. In STM measurements, the electronically decoupled bilayer graphene on graphite exhibited small-period Moiré patterns because of the large rotation angle with respect to the substrate [73]. The ultra-low random potential fluctuations resulting from substrate imperfections allows us to obtain high-quality atomic-resolution STM images of the 1D domain wall, as shown in Fig. 16. Figure 16(b) shows a representative atomic STM image of the AB-BA stacking soliton. From the left to the right of the domain wall, the atomic image transforms from a triangular lattice (in the AB region) to a hexangular lattice (in the center of the domain wall) and then to a triangular lattice (in the BA region) [157, 158], completing an interlayer transition from AB to BA stacking. Moreover, the interatomic distances in the domain wall are ~1.5 % smaller than those in the surrounding Bernal regions, according to the 2D Fourier transform of the atomic STM images. The compressed interatomic distances of the 1D structure, together with the hexagonal lattice at its center and the surrounding AB and BA domains, imply that the structure is an AB-BA strain soliton. As shown below, to identify the AB-BA stacking domain wall, LL spectroscopic measurements are necessary. Fig. 16 AB-BA bilayer domain wall imaged by STM. (a) STM topographic image of an AB-BA domain wall (DW) region on a graphite surface. Insets: atomic-resolution STM images in the AB, DW, and BA regions. (b) Lattice transition from a triangular lattice (in the AB region) to a hexangular lattice (in the center of the domain wall) and then to a triangular lattice (in the BA region). Reproduced from Ref. [73]. The STS spectra recorded in both the AB- and BA-stacked regions exhibit characteristics that are expected for gapped graphene bilayers (see Fig. 17). The graphite substrate induces an interlayer bias and breaks the inverse symmetry of the topmost adjacent bilayers, generating a finite gap of ~80 meV in the parabolic bands of both the AB and BA bilayers. At the level of low-energy effective theory, the AB-stacked bilayer is equivalent to the BA-stacked bilayer subjected to the opposite gate polarity [159–161]. Thus, under a uniform interlayer bias, the sign of the energy gap (as well as the effective mass) changes across the domain wall from the AB- to BA-stacked regions, and symmetry-protected gapless modes are expected to emerge in the stacking soliton [152, 153]. Under finite magnetic fields, the positions of the two lowest LLs-LL(0,1,+) and LL(0,1,-), which are a couple of layer-polarized quartets-in the energy band depend on the sign of the gate polarity (or the sign of the energy gap) of the gapped Bernal bilayers [45, 72, 77]. Therefore, they are reversed in the AB- and BA-stacked bilayers. In our experiment, the LL spectra recorded in the AB- and BA-stacked bilayer domains exhibited strong Landau quantization of massive Dirac fermions of gapped bilayers, as shown in Figs. 17(a) and 17(e). We obtained the same effective mass-m* ≈ 0.045 me-in the two domains. The most remarkable feature is that the top layer-located quartet LL(0,1,+) lies on the valence-band side for the AB-stacked bilayer region but on the conduction-band side for the BA-stacked region. This result directly demonstrates that the band structures of the AB and BA bilayers are reversed, leading to the change in the sign of the energy gap. This feature in the LL spectra can be used to unambiguously identify the AB-BA domain-wall region in bilayer graphene. Next, we discuss the Landau quantization in the AB-BA stacking soliton region. The bilayer stacking domain walls are predicted to generate topological conducting edge states in the gated system, which was proven by STM [73] and transport measurements [155]. Under strong magnetic fields, we can also detect a series of DOS peaks in the stacking domain wall, as shown in Figs. 18 and 19. These peaks mimic the sequences of the LLs of both the AB and BA domains. This feature results from the spatial extension of the LL wave functions of the surrounding Bernal bilayers (the extent, ~ , is on the order of 10 nm for B ≤ 2BNl8 and comparable to the soliton width), as described in Section 3, which leads to the appearance of the bilayer- like LLs in the domain-wall region. The "splitting" of the LLs recorded in the domain wall, as shown in Figs. 18(a) and 18(b), arises from the shift of the CNPs of the adjacent AB and BA domains (~15 meV in the experiment). The calculated LLs in the domain wall under a 8-T field imitate the sequences of those in the AB and BA regions, which agrees well with the experimental results [see Fig. 18(c)]. Our results indicate that the layer-stacking domain walls, as well as the stacking orders, can be unambiguously identified by LL spectroscopic measurements. Fig. 17 Landau quantization of gapped AB and BA bilayer graphene. (a, e) LL spectra recorded in the AB- and BA-stacked regions under various fields B, respectively. (b, c) LL peak energies of the AB bilayer plotted versus ±[n(n-1)]1/2B and B, respectively. The solid curves are the fits of the data with the theoretical equation. (d, h) Schematics of the layer-polarized states and the LLs in gapped AB and BA bilayer graphene, respectively. (f, g) LL peak energies of the BA bilayer plotted against ±[n(n-1)]1/2B and B, respectively. Reproduced from Ref. [73]. Fig. 18 STS spectra of the bilayer domain wall under various fields B. (a) STS spectra of the domain-wall region for different fields B, showing level peaks. (b) LL spectra of the AB, BA, and domain-wall regions under an 8-T field. (c) Calculated DOS of the AB, BA, and domain-wall regions under an 8-T field. Reproduced from Ref. [73]. Fig. 19 Spatially resolved tunneling spectra maps across the AB-BA domain wall. The maps are measured at 8 T (a), 7.5 T (b), and 7 T (c). The LLs shown in the domain-wall region almost follow the sequences of those recorded in the gapped AB/BA bilayer region. The dashed lines indicate the edges of the domain wall. Reproduced from Ref. [73]. 4.1.2 ABA-ABC trilayer soliton In trilayer graphene, there are two lowest-energy stacking orders: ABA, i.e., Bernal stacking, and ABC, i.e., rhombohedral stacking [Figs. 20(a) and 20(b)]. The chiral quasiparticles differ significantly between the two allotropes because of the distinction in the stacking order [162–164]. In the ABA trilayer, both massless (l = 1) and massive (l = 2) Dirac fermions coexist [89, 95]; whereas the low-energy excitations in the ABC trilayer are l = 3 chiral quasiparticles with a cubic dispersion (the corresponding Berry phase of the quasiparticles is 3) [165–167]. Moreover, stacking domain-wall solitons separating the ABA and ABC registered regions can be formed in trilayer graphene. By using similar STM measurements as described for bilayers, ABA- and ABC-stacked trilayers, together with the ABA-ABC trilayer domain walls, were observed in our experiment [150]. Figures 20(c)–20(f) show several STM images taken around the stacking solitons of trilayer graphene. The ABA region can be directly discriminated from the ABC region in the STM images recorded at low bias voltages owing to its distinct low-energy electronic structures and properties, as shown in Figs. 20(a) and 20(b). The bright lines separating the adjacent ABA and ABC regions in the topography images are the trilayer stacking domain-wall solitons, whose heights depend strongly on the bias voltage used for imaging. Furthermore, the topographical structures of the trilayer domain walls exhibit different patterns, as shown in Figs. 20(c)–20(f). These rich patterns provide an ideal platform for exploring stacking solitons with different atomic configurations. The representative atomic-resolved STM image around a trilayer domain wall shows hexagonal lattices in the center of the solitons, but triangular lattices are observed in both the adjacent ABA and ABC regions [see the lower inset of Fig. 20(f)]. The types of the trilayer domain walls are determined according to the lattice transition across the solitons. In the experiment, we observed shear-type domain walls more frequently than in the case of the tensile-type solitons in the trilayers. Similar results were reported for the stacking domain walls of bilayer graphene [147, 151], which may arise from the fact that the energy of the shear stacking solitons is slightly lower than that of the tensile stacking solitons in multilayer graphene. Fig. 20 Schematics and band dispersions of ABA (a) and ABC (b) trilayer graphene. (c–f) STM topographic images of different ABA-ABC trilayer domain-wall structures observed on the graphite surface. Insets in (f): atomic lattices in the ABA, ABC, and domain-wall regions. (g) Typical zero-field tunneling spectra of the ABA and ABC trilayers. Reproduced from Ref. [150]. Figure 20(e) shows the STS spectra recorded in the ABA and the ABC regions away from the stacking solitons, under zero magnetic field. For the ABA trilayer, we observed a typical V-shaped spectrum at 0 T, as previously reported. For the ABC trilayer, the zero-field spectrum exhibited low-energy pronounced peaks, which were generated by the flat bands around the CNP of the ABC trilayer graphene [168]. The energy spacing ~10 meV of the two peaks around the CNP corresponds to the energy gap of the ABC trilayer. The substrate induces an effective interlayer bias, breaking the inversion symmetry of the rhombohedral trilayer [169] and thus introducing a gap, which is similar to that in bilayer graphene. The finite-field spectra (Fig. 21) exhibit quite different Landau quantization in ABA-stacked and ABC- stacked graphene trilayers. In the STS spectra of the ABA trilayer, we observed a sequence of LL peaks of both massless Dirac fermions (l = 1) and massive Dirac fermions (l = 2), as discussed in Section 2. For the ABC-stacked trilayer, the tunneling spectra show a new LL sequence of the unique Landau quantization for the l = 3 chiral quasiparticles [Fig. 21(d)]. In standard theory, the l = 3 chiral fermions of ABC trilayers are quantized under a perpendicular magnetic field with the energy of the nth level, which has a B3/2 dependency [91, 170, 171]: where EC is the energy of the CNP, ± describes electrons and holes, and t⊥ is the nearest-neighbor interlayer hopping strength. By separately fitting the energies of the LLs in the electron and hole branches to Eq. (7), as shown in Figs. 21(e) and 21(f), we obtained a bandgap of Eg ≈ 10 meV for the ABC trilayer, which is congruent with that observed under zero field. We also determined the interlayer coupling to be t⊥≈ 0.48 eV in different ABC samples. It was theoretically predicted that t⊥ ≈ 0.5 eV in the ABC trilayer graphene [165], and an almost identical value of t⊥ was extracted from transport measurements in the ABC trilayers [162]. Fig. 21 (a) STS spectra of the ABA trilayer measured under various fields B. The monolayer and bilayer LL orbital indices are indicated by orange and blue numbers, respectively. The dashed lines label the gap edges in the parabolic bands. LL peak energies obtained from (a), showing (b) B1/2 and (c) B dependency, respectively. (d) Tunneling spectra of the ABC trilayer under 0–8-T magnetic fields. The LL indices are labeled. (e) LL peak energies obtained from (d) versus ±[n(n-1)(n-2)B3]1/2. (f) LL peak energies versus B. The blue dots are the data from (g), and the black curves are results of fitting with Eq. (7). Reproduced from Ref. [150]. The LL is also fourfold degenerate in ABC-stacked trilayer graphene, with two valley degeneracies and two spin degeneracies for each orbital quantum number, n. Without the roles of an external electric field and 3/2222(1)(2),3,4,5...,(7)FnCveBEEnnnntelectron–electron interaction, the n = 0, n = 1, and n = 2 LLs are further degenerate and consequently, there is a 12-fold degenerate state at the CNP of the ABC trilayers. The degeneracy of the lowest LL can easily be partially lifted. For example, in the presence of an interlayer potential, the valley degeneracy is lifted and a finite energy gap is generated in the band structure of ABC trilayer graphene [172]. Moreover, the wave functions of the lowest LL are mainly localized on the A1 sites of the first layer for one valley (+) and on the B3 sites of the third layer for the other valley (-) [173]. Hence, we observed valley-polarized Landau quantization of the l = 3 chiral quasiparticles: the magnitude of the valley-polarized LL-LL(0,1,2,+)-was far higher than that of the LL(0,1,2,-) because the STM predominantly probes the DOS of the top graphene layer, as shown in Fig. 21(d). The layer polarization of the two lowest LLs, LL(0,1,2,+) and LL(0,1,2,-), depends on the sign of electric polarity (or the sign of the energy gap) of the ABC trilayer, and we observed both positive and negative layer-polarized LL(0,1,2,+) and LL(0,1,2,-) in different trilayers [150]. Theoretically, topological edge modes are expected to be observed in such trilayer domain walls [154], which separates opposite gated ABC- stacked domains, as demonstrated in the AB-BA domain walls of bilayer graphene [73]. Now, we discuss the Landau quantization of the quasiparticles across the stacking solitons between the ABA and ABC trilayer domains. Figure 22 shows representative measured LL spectra with respect to the STM tip position, scanned from the ABA region to the ABC region. Away from the trilayer domain wall, we observed LLs of the l = 3 chiral fermions in the ABC trilayer and detected LLs of both the l = 1 and l = 2 chiral quasiparticles in the ABA trilayer. Across the stacking soliton, we observed the evolution of quasiparticles between the chiral degree l = 1&2 in the ABA trilayer and l = 3 in the ABC trilayer, accompanying the transition of the stacking orders of the trilayers. In the stacking domain-wall region, the LLs of both the l = 1&2 and l = 3 chiral fermions are detected because of the spatial extension of the wave functions for the quasiparticles in the adjacent domains [72]. Fig. 22 Spatial evolution of the LL spectra recorded under an 8-T field across an ABA-ABC shear soliton. The colored dashed lines indicate the positions of the LLs of the massless and massive fermions in the ABA trilayer. The black bars label the LLs of the quasiparticles in the ABC trilayer. Reproduced from Ref. [150]. 4.2 Twisted graphene system In addition to the matched layer-configurations, the introduction of a stacking misorientation drastically expands the allotropes of the graphene multilayers and, more importantly, the resultant structures show strong twist-dependent electronic spectra and properties [105, 174–177]. For example, two low-energy van Hove singularities (VHSs), which originate from the two saddle points in the band structure, were observed in the twisted bilayer graphene as two pronounced peaks in the DOS [178–181]. Among such especial dispersion of the twisted bilayers, the most striking results are the angle- and coupling-dependent renormalization of the Fermi velocity [182, 183] and the appearance of almost dispersionless bands (flat bands) at a very small angle [184]. This suggests that non-Abelian gauge potentials emerge in the smallest twisted bilayers (≤1°) [185] and that electrons in a graphene bilayer can be changed from ballistic to localized by simply varying the rotation angle [186–188]. Here, we discuss the angle-dependent Landau quantization in twisted graphene bilayers and trilayers. 4.2.1 Twisted bilayers In the continuum approximation, a rotation between two graphene layers leads to a shift ΔK [Fig. 23(c)] between the corresponding Dirac points of the two sheets in the momentum space [189, 190]. The zero-energy states no longer occur at k = 0 but rather occur at k = −ΔK/2 for layer one and k = ΔK/2 for layer two [191, 192]. At a large rotation angle (θ ＞ 6°), the twisted bilayers behave as two decoupled single-layer graphene sheets with a Fermi velocity of vF ≈ 1.1 × 106 m/s [181, 182]. At a small rotation angle (θ ＜ 6°), the electronic states near the Dirac cones of two layers couple with a reduced interlayer hopping amplitude of t≈t⊥ (t⊥is the interlayer hopping for AB-stacked layers) [179, 182, 193]. As a result, the two Dirac cones intersect and hybridize at the energy of ±ћvFΔK, generating two saddle points in the energy bands [Fig. 23(d)] and two symmetric low-energy VHSs in the DOS. The energy difference between two the VHSs exhibits a strong angle- dependent variation: EVHS ≈ ћνFK - 2t [Fig. 23(e)]. Below the energy of the VHSs, the linear dispersion is preserved but with a renormalized Fermi velocity. Fig. 23 Small-angle twisted bilayer graphene. (a) Schematic of the twisted bilayer. (b) STM topographic image of a twisted bilayer with θ = 3.1°, showing the Moiré superlattice structure. (c) Shifting of the Dirac cones in twisted bilayers. (d) Low-energy band dispersion of the twisted bilayer with θ = 2.2° and tθ = 156 meV. (e) STS spectra of twisted graphene bilayers with various twisted angles. The arrows mark the positions of the VHSs. (f) Interlayer hopping and energy differences of the VHSs (the inset) with respect to the rotation angles in twisted graphene bilayers. Reproduced from Refs. [182] and [196]. The Fermi velocity of a twisted bilayer can be obtained by LL spectroscopic measurements [182]. In the twisted bilayers, two series of LL sequences are generated below the VHSs under strong magnetic fields, which arise from the two shifted Dirac cones of the two layers. Each of the LL sequences shows the same square-root dependency of both the level index n and field B as in single-layer graphene, i.e., En = ED + sgn(n) , n = 0, ±1, ±2,… (ED is the energy of the Dirac point). Figure 24 shows several representative Landau quantized spectra and their analysis for different twisted bilayer samples. The sequence of LLs shown in Figs. 24(a)– 24(c) is unique to that of massless Dirac fermions and is expected to be observed in twisted bilayer graphene. In the LL spectroscopic measurements, we only measured one LL sequence, as the quasiparticles of the topmost graphene layer mainly come from one of the two Dirac cones in the twisted bilayers and the STS predominantly probes the signal of the top layer. The Fermi velocities of the twisted bilayers with various rotation angles can then be obtained directly by a reasonable linear fit of the measured LL energies to the above equation, as shown in Figs. 24(d)–24(f). Our experimental results indicate that both the twisted angles and the interlayer-coupling strengths drastically affect the Fermi velocity of the twisted bilayers. In the twisted graphene bilayers, the stacking fault, grain boundary, defects, and roughness of the substrate may alter the interlayer distance and stabilize it at various equilibrium values, leading to large variations of the interlayer interaction [182]. For a constant t, the Fermi velocity decreases as the twisted angle decreases. This tendency can be unambiguously displayed in the band structures of different twisted bilayers with identical interlayer coupling, as shown in Fig. 25(a). The BnveF22slope of the energy dispersion near the Dirac point, which is proportional to the Fermi velocity, decreases as the rotation angle decreases. Normally, the interlayer-coupling strength t ≈ 110 meV in the small twisted bilayer graphene, which can be obtained from the relationship between t andEVHS [179, 181, 182]. In this case, the renormalization of the Fermi velocity in twisted bilayers can be described as follows [189]: . (8) Hence, the pronounced Fermi velocity renormalization is observed in small twisted bilayers, as shown in Fig. 24(b), where the measured Fermi velocity is vF ≈ 0.81 × 106 m/s for a twisted bilayer with θ ≈ 3.6° and t ≈ 130 meV. However, for relatively weak interlayer hopping, the renormalization of the Fermi velocity is negligible, and we cannot detect the significant reduction of the Fermi velocity, even for small-angle twisted bilayers. For example, in the θ ≈ 2.8° twisted bilayer with interlayer hopping, t ≈ 50 meV. The weak interlayer hopping may arise from the tilt grain boundary underlying the twisted bilayers. The Fermi velocity is measured to be vF ≈ 1.03 × 106 m/s, showing a negligible renormalization. Fig. 24 Landau quantization in twisted bilayer graphene. STS spectra taken under various field B in twisted bilayers with (a) θ = 2.8°, (b) θ = 3.6° (b), and (c) θ = 6.0°. The LL indices are marked. (d–f) LL peak energies obtained in (a)–(c), respectively, plotted against sgn(n)(nB)1/2, as expected for massless Dirac fermions. The solid lines are linear fits of the data. The slopes yielding the Fermi velocities of vF = (1.03 ± 0.02) × 106 m/s (d), vF = (0.811 ± 0.004) × 106 m/s (e), and vF = (1.140 ± 0.003) × 106 m/s (f), respectively. Reproduced from Ref. [182]. The effects of both the interlayer-coupling strength and the twisted angle on the Fermi velocity of the twisted bilayers are shown in Fig. 25(b). The observed Fermi velocity does not decrease monotonously as the twisted angle decreases, indicating that it is not determined only by the twisted angle but also depends on the interlayer interaction. By using the discrete interlayer-coupling strength, the Fermi velocity shows the 219FFFtvvvKpredicted angle dependence of the renormalization for different twisted angles, as described by Eq. (8). Fig. 25 (a) Energy dispersions for three twisted graphene bilayers with θ = 6.0°, θ = 3.9°, and θ = 2.9°, respectively. (b) The Fermi velocity as a function of the twisted angle for different interlayer hopping strengths. The circles and squares indicate the data obtained for HOPG and SiC, respectively. The Fermi velocity is normalized with respect to vF = 1.1 × 106 m/s. Reproduced from Ref. [182]. 4.2.2 Twisted trilayers We discuss twisted trilayers consisting of two Bernal-stacked graphene bilayers and one slightly rotated layer [Fig. 26(a)]. The low-energy band structures of a twisted trilayer are formed by a parabolic band resulting from the AB-stacked bilayer intersecting with a Dirac cone band originating from the rotated layer [Fig. 26(b)] [194, 195]. Similar to the case in twisted bilayers, the two band branches of the twisted trilayers hybridize in a finite interlayer coupling between the rotated layer and the Bernal bilayer. Hence, two saddle points are generated in the energy dispersion, and two VHSs are generated in the DOS [Figs. 26(c) and 26(d)]. However, with an identical twisted angle and identical interlayer coupling, the energy difference of the saddle points in the electronic band structures of the twisted trilayers, i.e., EVHS, is far smaller than that of the twisted bilayers [196, 197]. Fig. 26 Configuration and band structures of twisted trilayer graphene. (a) Schematic of the twisted trilayer graphene made out of two Bernal bilayers and one rotated layer. (b) 3D low-energy band structure of a twisted trilayer with θ = 2.2° and tθ = 156 meV. (c) Band structures along kx = 0 for the twisted trilayer (dashed curve) and bilayer (solid curve) with the same θ. (d) Calculated DOS of the twisted trilayer (dashed curve) and bilayer (solid curve) with the same θ, both showing two VHSs. Reproduced from Ref. [196]. Figure 27 shows the coexistence of the twisted graphene trilayer and the twisted bilayer regions, which are separated by a tilt grain boundary and both have a twisted angle of 2.8° [197]. The tilt grain boundary is located at the second graphene layer, and the top continuous layer is rotated with respect to the second layer, resulting in the twisted bilayer in region I, as schematically shown in Fig. 27(d). In region II, the top graphene layer and second layer are AB-stacked, and the Moiré pattern arises from the stacking misorientation between the second layer and third layer [196]. Consequently, the contrast of the Moiré pattern for the AB-twisted trilayer (ABT, region II) is much lower than that for the twisted bilayer (TB, region I), as shown in the STM images of Figs. 27(a)–27(c). Figure 28(a) shows the spatial evolution of the zero-field tunneling spectra along a line across the tilt grain boundary from TB to ABT. In both the TB and ABT regions, the spectra exhibit two low-energy VHSs. However, the energy difference of the VHSs-EVHS-abruptly decreases from ~0.54 eV in TB to ~0.44 eV in ABT, as shown in Fig. 28(c). This decrease in EVHS is a direct result of introducing a third layer on top of a twisted bilayer, as previously discussed. In the ABT, the substrate induces an effective external electric field on the topmost AB-stacked bilayer, which consequently generates a finite gap ~100 meV in the parabolic band branch of the ABT. The spectra recorded in the ABT region at positions far from the boundary (＞2.4 nm) display two DOS peaks at the gap edges within the two VHSs.  Fig. 27 Coexistence of twisted graphene trilayer and bilayer. (a) STM images of twisted graphene trilayer (region I) and bilayer (region II) on a graphite surface. The two sets of Moiré superlattices are separated by a tilt grain boundary (black dashed line). The periods of both the Moiré patterns ~5.0 nm. Inset (bottom): a profile line across the boundary. Inset (top): Fourier transforms of the STM image. STM images of a twisted (b) bilayer and (c) trilayer. Insets: height profiles along the white dashed lines. (d) Schematic of the structure in (a). (e) High-resolution current image of the area indicated by the white frame in (a). The topmost graphene layer is continuous. It shows a clear hexagonal lattice in region I and a triangular lattice in region II. (f) Fourier transforms of (e). The outer six bright spots represent the reciprocal lattice of the topmost graphene layer. Reproduced from Ref. [197]. Under strong magnetic fields, we observed distinct Landau quantization in the TB and ABT. In the TB region, the STS spectra exhibit the expected one LL sequence of the massless Dirac Fermion for the twisted graphene bilayer [197]. In the ABT region, the observed LL spectra [Fig. 28(d)] are distinct from those in the TB region and follow those of a massive Dirac fermion [72]. Such a result is reasonable because the massive Dirac fermions of the ABT are localized at the topmost AB-stacked bilayer. For the massive Dirac fermions, two DOS peaks at the gap edges are expected to develop into two valley-polarized fourfold-degenerate LLs in finite fields [77]. This is demonstrated explicitly in our experiment, as shown in Fig. 28(d). The positions of the two lowest LLs are almost independent of the magnetic field, meaning that the gap size in the parabolic band of the twisted graphene trilayer does not vary with the field B. We expect that if the rotated sheet is the topmost one in the twisted graphene trilayer, the measured LL sequence will follow that of massless Dirac fermion in the STS probing.  Fig. 28 (a) STS spectroscopy as a function of the tip position measured along the black arrow in (b). The black dotted lines indicate the positions of VHSs in the TB and ABT regions. Two peaks (labeled by black and red arrows) in the spectra of the ABT region mark the positions of the DOS peaks at the gap edges. (b) An STM image obtained around the boundary of the TB and ABT regions. (c) EVHS as a function of the position around the boundary. The black and red dashed lines show the average values for the TB (~0.54 eV) and ABT (~0.44 eV) regions, respectively. (d) dI/dV spectra measured in the ABT region for different fields B. LL indices n are marked. The zero-energy LL splits into two valley-polarized quartets and a bandgap of ~100 meV is opened, as indicated by the dashed lines. The LL(0,1,+) and LL(0,1,-) are projected on the top and bottom graphene layers, respectively. (e) LL peak energies shown in (d) plotted against ±[n(n-1)]1/2B. The red lines are fits of the data. The inset shows a schematic of the low-energy dispersion of ABT with quantized LLs. Reproduced from Ref. [197]. 5 Landau quantization in strained and defective graphene Being only one layer thick, the structure of graphene is vulnerable to the underlying substrate and surrounding environment [4]. Consequently, there are many distortions, such as corrugations and charge defects, in supported graphene, especially on metal substrates [198–200]. Around these distortions, the properties of quasiparticles are drastically modified, leading to unusual spectroscopic characteristics [201, 202]. For example, both the strained structures and charge defects will introduce an electron–hole asymmetry in graphene, and a spatially varying lattice distortion in strained graphene can create pseudo-magnetic fields, resulting in LL-like quantization [203, 204]. In this section, giant electron–hole asymmetry and unconventional valley-polarized LL splitting are investigated by revealing the Landau quantization in strained and defective graphene under external magnetic fields. 5.1 Electron–hole asymmetry Charge carriers in a graphene monolayer exhibit light-like dispersion and, usually, the electron and hole are symmetric [4, 5]. The electron–hole symmetry plays a crucial role in the chirality and chiral tunneling of massless Dirac fermions in graphene monolayers [6]. Previous researches demonstrate that strain and charged defects can induce large electron–hole asymmetry in graphene [205–208]. According to the Landau quantization under a magnetic field, the Fermi velocities of electron- and hole-like Dirac fermions in a graphene monolayer can be determined separately. This provides a unique opportunity for quantitatively studying the electron–hole asymmetry. Next, via LL spectroscopy, we show the large electron–hole asymmetry generated by strain and charged-defect scattering in graphene monolayers [76]. Fig. 29 Electron–hole asymmetry in strained graphene. (a) An STM topographic image showing quasi-periodic graphene ripples on Rh foil. (b) STM image of a graphene ripple in the white frame in (a). The solid blue line is the height profile across the rippled graphene region. Inset: fast Fourier transform showing the reciprocal lattice of the graphene lattice. The scale bar represents 20 Gm−1. (c) Electronic dispersion for a graphene monolayer with different t'. (d) STS spectra taken on the ripple in (b) with different fields B. (e) LL peak energies plotted against sgn(n)nB1/2. The red and blue solid lines are linear fits of the data with Eq. (1) for electrons and holes, respectively. Inset: the schematic LLs structure of graphene with (solid curves) and without (dashed curves) strain. Reproduced from Ref. [76]. The Hamiltonian of the graphene monolayer in the tight-binding model is [5] （9） Here, the operators ( ) create (annihilate) an electron with spin at site i, t ∼ 3 eV is the nearest-neighbor †††(H.C.)H.C.£ijijiji,ji,jH=-tab+t'aa+bb+†iaiahopping integral, and t′ is the next-nearest-neighbor hopping energy. According to Eq. (9), the simplest method to break the electron–hole symmetry in a graphene monolayer is to introduce a nonzero t′, which is easily done in strained graphene [199, 205]. Figures 29(a) and 29(b) show representative STM images of a rippled graphene region on Rh foil due to the mismatch of thermal-expansion coefficients between the graphene and the Rh substrate. STS spectra recorded in the ripples under magnetic fields exhibit the Landau quantization of massless Dirac fermions, as shown in Fig. 29(d). According to Eq. (1), we determined ve F and vh F separately. The Fermi velocities of electrons, ve F, and holes, vh F, differ significantly, as shown in Fig. 29(e), and are measured to be (1.21 ± 0.03) × 106 and (1.02 ± 0.03) × 106 m/s, respectively. Additionally, the measured ve F and vh F are almost independent of the positions in the ripple. This large electron–hole asymmetry is attributed to the enhanced next-nearest neighbor hopping caused by the lattice deformation and curvature in the rippled region. By introducing a nonzero t′ in graphene, the Fermi velocity of the filled state (electrons) increases, and that of the empty state (holes) decreases, as shown in Figs. 29(c) and 29(e). A finite value of t′ = 0.16t well describes the observed electron–hole asymmetry in the ripple shown in Fig. 29(b). Here, the observed electron– hole asymmetry differs for different graphene ripples, and the estimated t′ ranges from ~0.02t to ~0.2t in the rippled regions in our experiment. Theoretically, the calculated t′ varies widely depending on the tight-binding parameterization [4]. Recently, the t′ in high-quality graphene was measured to be ~0.4 eV in a polarization- resolved magneto-Raman-scattering experiment [209] and was determined to be ~0.3 eV by quantum capacitance measurements [210]. In our experiment, t′ is simply a fitting parameter in the tight-binding model to account for the electron–hole asymmetry observed in the graphene ripples. The observed values, when compared with those estimated in theory and those measured in previous experiments, are reasonable. We will demonstrate that in strained graphene, ve F and vh F in the graphene monolayer can be quite different around the charged defect. Figure 30(a) shows a representative STM image of a graphene sheet with high- density atomic-scale defects. A R30° interference pattern, as shown in Fig. 30(b), can be observed around the defect [200]. The STS spectra, as shown in Fig. 30(c), show the existence of a resonance peak above the Dirac point associated with the defect. Figure 30(d) shows typical spectra recorded under different magnetic fields around the defects. Obviously, the observed Landau quantization exhibits large asymmetry between the empty state above the Dirac point and the filled state below the Dirac point, as shown in Fig. 30(e). According to Eq. (1), we determine ve F and vh F to be (1.21 ± 0.03) × 106 and (1.02 ± 0.03) × 106 m/s, respectively. Around the defects (within 10–15 nm), the electron–hole asymmetry is almost independent of the recorded positions, as shown in Fig. 31(a). Comparing the structure of the graphene sheet in Fig. 30(a) with that of the pristine graphene monolayer, the main differences are the existence of nanoscale rippling and the atomic defects. These nanoscale ripples, which usually occur with h2/(la) < 0.0254, cannot induce such a large electron–hole asymmetry (here, h is the amplitude, l is the width of the ripple, and a = 0.142 nm), and more importantly, the nanoscales ripple with larger h2/(la)-as shown in Fig. 29(b)-exhibit opposite electron–hole asymmetry to that observed around the defects (Fig. 30). Therefore, we mainly attribute the large electron– hole asymmetry shown in Fig. 30 to the charged-defect scattering. The electron-like and hole-like Dirac fermions in the graphene monolayer are expected to respond differently to a Coulomb potential: they are scattered more strongly when they are attracted to the charged defect than when they are repelled from it, as shown in Fig. 31(b) [211–213]. That is, charged Dirac fermions around an attractive potential center spend 33more time there and are more significantly deflected than those around a repulsive potential center. The attractive force and enhanced scattering induced by the charged defect can affect the transport properties of the quasiparticles and reduce the Fermi velocity of the quasiparticles. According to the observed electron–hole asymmetry, these defects shown in Fig. 30(a) are determined to have a positive charge, which is consistent with the fact that the resonance peak of the charged defect is located above the charge neutrality. Fig. 30 STM images and STS spectra of a graphene monolayer with a charged defect. (a) An STM topographic image of a graphene monolayer with three atomic defects (marked by dashed ellipses). (b) Magnified STM image of a defect in the red ellipse in (a). (c) Normalized STS spectra, (dI/dV)/(I/V)-V, measured on graphene at different distances from the center of the defect in (b). (d) dI/dV spectra taken under different fields B at the blue dot marked in (a). The blue arrows indicate the resonance peak of the defect. (e) LL peak energies of (d) plotted against sgn(n)nB1/2. The red and blue solid lines show the linear fits of the data with Eq. (1) for holes and electrons, respectively. Inset: the schematic LL structure of the graphene sheet around (solid curves) and without (dotted curves) the charged defect. Reproduced from Ref. [76]. Fig. 31 (a) Summary of the Fermi velocities of electrons and holes obtained at different positions around the charged defects. (b) Transport cross-section C as a function of the impurity charge αε. αε > 0 indicates an attractive interaction between the impurity and the carriers, and αε < 0 indicates a repulsive interaction between them. Reproduced from Ref. [76]. 5.2 Valley-polarized LLs In strained graphene, lattice deformation can create pseudo-magnetic fields affecting the behavior of massless Dirac fermions, resulting in zero-field LL-like quantization [199, 203–205, 214]. The primary difference between the strain-induced pseudo-magnetic field BS and the external magnetic field B is that BS preserves time-reversal symmetry and has opposite signs in the two low-energy valleys of graphene: K and K′ [215–221]. Therefore, the combination of the pseudo-magnetic field and the magnetic field can lift the valley degeneracy of the LLs and lead to unconventional valley-polarized Landau quantization in graphene monolayers [214]. In the experimental STM spectroscopic measurements, we directly observed valley- polarized LLs in strained graphene grown on Rh foil, which are induced by the coexistence of the pseudo- magnetic fields and external magnetic fields [214]. Strained structures can easily be observed for graphene grown on metallic substrates because of the mismatch of the thermal-expansion coefficients between graphene and the supporting substrates [198, 199, 204, 205, 222]. Figures 32(a) and 32(b) show representative STM images of a strained graphene structure on Rh foil. The STS spectra [Fig. 32(c)] recorded in the strained region under magnetic fields exhibit the Landau quantization of massless Dirac fermions in the graphene monolayer, with its characteristic non-equally spaced energy-level spectrum of LLs and the hallmark zero-energy state [45, 55, 57]. The linear fit of the experimental data to Eq. (1), as shown in Fig. 1(d), yields = (1.257 ± 0.009) × 106 m/s and = (0.930 ± 0.014) × 106 m/s. The large electron–hole asymmetry is mainly attributed to the enhanced next-nearest-neighbor hopping t′ ≈ 0.3t caused by the lattice deformation and curvature in the strained graphene [4, 5, 204], as previously discussed. eFvhFvFig. 32 STM images and STS spectra of the strained graphene monolayer on Rh foil. (a) STM image of a strained graphene region showing 1D quasi-periodic ripples on Rh foil. (b) Enlarged STM image of the graphene ripples in the blue frame in (a). (c) STS spectra taken at the position marked with the red solid circle in (b) under different B. The spectra are shifted to maintain the n = 0 LL at the same bias. (d) LL peak energies obtained from (c) show a linear dependence on sgn(n)(nB)1/2, as expected for the graphene monolayer. Inset: schematic of the LLs in monolayer graphene, considering the next-nearest-neighbor hopping energy t′. Reproduced from Ref. [214]. In addition to the electron–hole asymmetry, we observe two other notable features in the STS spectra of some strained graphene regions, as shown in Fig. 33. One feature is the emergence of several peaks: a pronounced peak at the CNP and several weak peaks are observed at a relatively high bias, even in the spectra measured under zero magnetic field [Fig. 33(a)]. The other feature of the spectra is the splitting of n = -1 and n = -2 LLs recorded under different magnetic fields, as shown in Fig. 33(b). For the spectra recorded at the same position, the energy difference of the two peaks of n = -1 LL, -1, decreases with increasing magnetic fields [Fig. 33(b)].  Fig. 33 Pseudo-LLs and valley-polarized Landau quantization. (a) Two representative zero-field STS spectra recorded in graphene ripples. (b) STS spectra taken in the area marked with the black frame in Fig. 1(b) under different magnetic fields. The LL indices of massless Dirac fermions are marked. In the presence of an external B, the n = -1 LL splits into two peaks: -1- and -1+ (here, the subscripts - and + denote K and K′ valleys, respectively). (c) LL peak energies obtained from (a), showing a linear dependence on sgn(n)(nB)1/2. For clarity, the data for the split n = -1 LL are not plotted in this figure. The Fermi velocities for electrons and holes are measured to be (1.250 ± 0.007) × 106 and (0.927 ± 0.007) × 106 m/s, respectively. Reproduced from Ref. [214]. Our experimental observations can be understood within a theoretical framework that incorporates the effects of both strain-induced pseudo-magnetic fields and external magnetic fields on the Landau quantization of massless Dirac fermions in a graphene monolayer. In the case of B = 0 T, the pronounced peak at the CNP, as shown in Fig. 33(a), is attributed to the strain-induced partially flat bands (n = 0 LL) at zero energy [205, 223, 224], and the weak peaks at a high bias are attributed to higher pseudo-LLs. The signal of the higher pseudo-LLs is rather weak, which agrees with earlier STM measurements in strained graphene ripples [205]. This result is reasonable because the description of the hopping modulation in strained graphene as an effective pseudo-magnetic field is exactly valid only at the CNP, and the higher pseudo-LLs are less defined [223, 224]. From the energy spacing between the pseudo-LLs measured at B = 0 T, the pseudo-magnetic field is measured to be from 0.45 to 0.85 T in the studied ripples, and the average value is estimated to be (0.60 ± 0.05) T according to Eq. (1) (in calculating BS, we use = (1.250 ± 0.007) × 106 m/s, which was measured in our experiment). Theoretically, a 2D strain field uij(x,y) in graphene can induce a gauge field and generate a pseudo-magnetic field (here, a is on the order of the C-C bond length, and 2 <  = < 3.) [145, 216, 225]. Obviously, not all the strained graphene structures result in a non-zero pseudo-magnetic field [226, 227], and it is difficult to generate a uniform pseudo- magnetic field in large-area graphene [216, 225]. Therefore, the observed pseudo-magnetic field varies spatially in our experiment, and only a part of the graphene ripples exhibit the pseudo-LLs. The pseudo- magnetic field arises from the spatial variation of the nearest-neighbor hopping t in graphene [216, 225]. The results shown in Fig. 33 indicate that the local strain in the region where the spectra of Fig. 33 are recorded 2xxyyxyuuAua2=-()xxyyxySuuuBAaxyln/lntaeFvnot only enhances the t′ but also results in the spatial variation of t. A local strain of 1% in graphene is predicted to generate a pseudo-magnetic field of 10 T [224]. Therefore, the observed BS ≈ 0.6 T is reasonable considering the width and height of the studied nanoscale ripples. In a previous study [204], the pseudo-magnetic field observed in graphene nanobubbles was as large as 300 T, indicating a local strain far larger than 10%. Such a large strain is attributed to the smaller width and larger height of the nanobubbles. The h2/(la) for the ripples studied in this work is far smaller than that of the nanobubbles. In the presence of both the pseudo-magnetic field and the external magnetic field, the total effective magnetic field in one of the valleys, for example, the K valley, is B – BS, and that of the other valley, for example, the K′ valley, is B + BS, as schematically shown in Fig. 34(a). Then, the valley degeneracy of the n ≠ 0 LLs is expected to be lifted [8, 11] and the energy spacing of the two valleys for the nth LL is (10) Considering that the picture of the effective pseudo-magnetic field is less defined at a high energy, the valley- polarized LLs should be observed only for small Landau indices n, for example, the n = ± 1 LLs. In our experiment, the valley-polarized LL is clearly observed for n = -1, as shown in Fig. 33(b). Although the absence of splitting for the n = 1 LL measured at B ≠ 0 T remains to be understood, it is likely that this observation is closely linked to the electron–hole asymmetry observed in our experiment. Additionally, the tunneling peak for the n = 1 LL is rather weak, which almost removes the possibility of detecting the valley splitting. Figure 34(b) summarizes the values of -1 with respect to the external magnetic fields recorded at the same position of the strained graphene. The fit of the experimental result to Eq. (2) yields BS = (0.59 ± 0.03) T, which is well consistent with that estimated according to the energy spacing of pseudo-LLs measured under zero magnetic field. The magnetic-field dependent of the splitting [Fig. 34(b)] provides convincing evidence for the valley- polarized LL induced by the external magnetic field and the pseudo-magnetic field. Fig. 34 Valley-polarized LL and pseudo-magnetic fields in the strained graphene. (a) Schematic image showing Landau quantization in the graphene monolayer in the presence of both the pseudo-B and the external B. The valley degeneracy of the n = -1 LL is lifted, and the energy spacing of the valley-polarized LLs is described by Eq. (10). (b) This figure summarizes the -1 measured at the same position as a function of the external magnetic field. The solid curve is a fit to Eq. (10), with BS = (0.59 ± 0.03) T as the only fitting parameter. Reproduced from Ref. [214]. 6 Conclusions and perspectives We reviewed recent experimental progress on Landau quantization in graphene and its multilayers using STM and STS. As we showed, by performing LL spectroscopic measurements of graphene sheets, the number of layers and the interlayer stacking configurations can be unambiguously verified. Compared with other conventional 2D electron systems, graphene materials offer a strictly 2D platform with fully exposed electronic 222+)2) , ...2, 1, 1, 2... nFSFSenBBenBBn（（states, which guarantees direct observations of long-desired quantum phenomena such as the two-component nature of Dirac quasiparticles and the LL bending via high-energy and spatial-resolution LL spectroscopy, as discussed in this review. Moreover, unusual Landau quantized behaviors in strained and defective graphene have been described, including the giant electron–hole asymmetry generated by strain and charged-defect scattering and the valley-polarized splitting of LLs in the presence of both a pseudo-magnetic field and a real field. As documented in this review, the aforementioned fundamentals, as well as other insights, are important for understanding new quantum Hall phases in graphene and may help in designing future experiments to detect other physical phenomena. For example, i) it is expected that the magnetic fields can generate a variety of new interaction driven states, with natures that are rather different from those found in the conventional 2DEGs. Over the past decades, multicomponent FQHE have been experimentally observed in graphene. However, the sequence of FQHE fractions and their origin are poorly understood. To solve these problems, high-resolution tunneling spectroscopic measurements at a super-low temperature are an alternative method. ii) Owing to the extra degree of freedom provided by the layer, an abundance of exotic quantum Hall edge states is believed to emerge in different stacked graphene multilayers. Through using LL spectra with both an atomically spatial resolution and a high-energy resolution, the quantum Hall edge physics can be tested systematically at the well-arranged graphene edges under super-low potential fluctuations. iii) The recent studies on layer-stacking domain-wall solitons are very exciting and vibrant. The emergence of the topologically protected conducting channels under zero magnetic field makes such strain-like solitons promising candidates for ultralow power-consumption nanodevice applications. We consider that with in-situ identification of the stacking domain walls by STM, deeper insights and the complete physics of these confined electronic systems will be revealed. iv) As we demonstrated, the combination of the external real magnetic fields and the pseudo-magnetic fields induced by the spatially varying strain in a graphene monolayer leads to unconventional splitting of LLs not present in graphene under either field alone. We envision probing more exotic phenomena if the real magnetic fields and the pseudo-magnetic fields are combined with other properties of the graphene monolayer and the additional degrees of freedom found in graphene bilayers and multilayers. Acknowledgements This work was supported by the National Natural Science Foundation of China (Grant Nos. 11674029, 11422430, 11374035), the National Basic Research Program of China (Grants Nos. 2014CB920903, 2013CBA01603), and the program for New Century Excellent Talents in University of the Ministry of Education of China (Grant No. NCET-13-0054). L.H. also acknowledges support from the National Program for Support of Top-notch Young Professionals. References 1. K. S. Novoselov, A. K. Geim, S. V. Morozov, D. Jiang, Y. Zhang, S. V. Dubonos, I. V. Grigorieva, and A. A. Firsov, Electric Field Effect in Atomically Thin Carbon Films, Science 306(5696), 666 (2004) doi:10.1126/science.1102896 2. K. S. Novoselov, A. K. Geim, S. V. Morozov, D. Jiang, M. I. Katsnelson, I. V. Grigorieva, S. V. Dubonos, and A. A. Firsov, Two- dimensional gas of massless Dirac fermions in graphene, Nature 438(7065), 197 (2005) doi:10.1038/nature04233 3. K. S. Novoselov, Nobel Lecture: Graphene: Materials in the Flatland, Rev. Mod. Phys. 83(3), 837 (2011) doi:10.1103/RevModPhys.83.837 4. A. H. Castro Neto, F. Guinea, N. M. R. Peres, K. S. Novoselov, and A. K. Geim, The electronic properties of graphene, Rev. Mod. Phys. 81(1), 109 (2009) doi:10.1103/RevModPhys.81.109 5. M. O. Goerbig, Electronic properties of graphene in a strong magnetic field, Rev. Mod. Phys. 83(4), 1193 (2011) doi:10.1103/RevModPhys.83.1193 6. M. I. Katsnelson, K. S. Novoselov, and A. K. Geim, Chiral tunnelling and the Klein paradox in graphene, Nat. Phys. 2(9), 620 (2006) doi:10.1038/nphys384 7. A. Bostwick, T. Ohta, T. Seyller, K. Horn, and E. Rotenberg, Quasiparticle dynamics in graphene, Nat. Phys. 3(1), 36 (2007) doi:10.1038/nphys477 8. A. K. Geim, and K. S. Novoselov, The rise of graphene, Nat. Mater. 6(3), 183 (2007) doi:10.1038/nmat1849 9. A. K. Geim, Graphene: Status and Prospects, Science 324(5934), 1530 (2009) doi:10.1126/science.1158877 10. S. Das Sarma, S. Adam, E. H. Hwang, and E. Rossi, Electronic transport in two-dimensional graphene, Rev. Mod. Phys. 83(2), 407 (2011) doi:10.1103/RevModPhys.83.407 11. D. N. Basov, M. M. Fogler, A. Lanzara, F. Wang, and Y. Zhang, Colloquium : Graphene spectroscopy, Rev. Mod. Phys. 86(3), 959 (2014) doi:10.1103/RevModPhys.86.959 12. W. Y. He, Z. D. Chu, and L. He, Chiral Tunneling in a Twisted Graphene Bilayer, Phys. Rev. Lett. 111(6), 066803 (2013) doi:10.1103/PhysRevLett.111.066803 13. F. D. Haldane, Model for a Quantum Hall Effect without Landau Levels: Condensed-Matter Realization of the "Parity Anomaly", Phys. Rev. Lett. 61(18), 2015 (1988) doi:10.1103/PhysRevLett.61.2015 14. C. Beenakker, Colloquium : Andreev reflection and Klein tunneling in graphene, Rev. Mod. Phys. 80(4), 1337 (2008) doi:10.1103/RevModPhys.80.1337 15. A. F. Young, and P. Kim, Quantum interference and Klein tunnelling in graphene heterojunctions, Nat. Phys. 5(3), 222 (2009) doi:10.1038/nphys1198 16. J. Baringhaus, M. Ruan, F. Edler, A. Tejeda, M. Sicot, A. Taleb-Ibrahimi, A. P. Li, Z. Jiang, E. H. Conrad, C. Berger, C. Tegenkamp, and W. A. de Heer, Exceptional ballistic transport in epitaxial graphene nanoribbons, Nature 506(7488), 349 (2014) doi:10.1038/nature12952 17. Z. Chu, and L. He, Origin of room-temperature single-channel ballistic transport in zigzag graphene nanoribbons, Science China Materials 58(9), 677 (2015) doi:10.1007/s40843-015-0081-y 18. J. J. Palacios, Graphene nanoribbons: Electrons go ballistic, Nat. Phys. 10(3), 182 (2014) doi:10.1038/nphys2909 19. B. Hunt, J. D. Sanchez-Yamagishi, A. F. Young, M. Yankowitz, B. J. LeRoy, K. Watanabe, T. Taniguchi, P. Moon, M. Koshino, P. Jarillo-Herrero, and R. C. Ashoori, Massive Dirac Fermions and Hofstadter Butterfly in a van der Waals Heterostructure, Science 340(6139), 1427 (2013) doi:10.1126/science.1237240 20. C. R. Dean, L. Wang, P. Maher, C. Forsythe, F. Ghahari, Y. Gao, J. Katoch, M. Ishigami, P. Moon, M. Koshino, T. Taniguchi, K. Watanabe, K. L. Shepard, J. Hone, and P. Kim, Hofstadter's butterfly and the fractal quantum Hall effect in moiré superlattices, Nature 497(7451), 598 (2013) doi:10.1038/nature12186 21. L. A. Ponomarenko, R. V. Gorbachev, G. L. Yu, D. C. Elias, R. Jalil, A. A. Patel, A. Mishchenko, A. S. Mayorov, C. R. Woods, J. R. Wallbank, M. Mucha-Kruczynski, B. A. Piot, M. Potemski, I. V. Grigorieva, K. S. Novoselov, F. Guinea, V. I. Fal'ko, and A. K. Geim, Cloning of Dirac fermions in graphene superlattices, Nature 497(7451), 594 (2013) doi:10.1038/nature12187 22. G. L. Yu, R. V. Gorbachev, J. S. Tu, A. V. Kretinin, Y. Cao, R. Jalil, F. Withers, L. A. Ponomarenko, B. A. Piot, M. Potemski, D. C. Elias, X. Chen, K. Watanabe, T. Taniguchi, I. V. Grigorieva, K. S. Novoselov, V. I. Fal'ko, A. K. Geim, and A. Mishchenko, Hierarchy of Hofstadter states and replica quantum Hall ferromagnetism in graphene superlattices, Nat. Phys. 10(7), 525 (2014) doi:10.1038/nphys2979 23. H. Schmidt, J. C. Rode, D. Smirnov, and R. J. Haug, Superlattice structures in twisted bilayers of folded graphene, Nat. Commun. 5, 5742 (2014) doi:10.1038/ncomms6742 24. A. M. DaSilva, J. Jung, and A. H. MacDonald, Fractional Hofstadter States in Graphene on Hexagonal Boron Nitride, Phys. Rev. Lett. 117(3), 036802 (2016) doi:10.1103/PhysRevLett.117.036802 25. Y. Barlas, K. Yang, and A. H. MacDonald, Quantum Hall effects in graphene-based two-dimensional electron systems, Nanotechnology 23(5), 052001 (2012) doi:10.1088/0957-4484/23/5/052001 26. V. Gusynin, and S. Sharapov, Unconventional Integer Quantum Hall Effect in Graphene, Phys. Rev. Lett. 95(14), 146801 (2005) doi:10.1103/PhysRevLett.95.146801 27. Y. Zhang, Y. W. Tan, H. L. Stormer, and P. Kim, Experimental observation of the quantum Hall effect and Berry's phase in graphene, Nature 438(7065), 201 (2005) doi:10.1038/nature04235 28. Z. Jiang, Y. Zhang, H. L. Stormer, and P. Kim, Quantum Hall States near the Charge-Neutral Dirac Point in Graphene, Phys. Rev. Lett. 99(10), 106802 (2007) doi:10.1103/PhysRevLett.99.106802 29. K. I. Bolotin, F. Ghahari, M. D. Shulman, H. L. Stormer, and P. Kim, Observation of the fractional quantum Hall effect in graphene, Nature 462(7270), 196 (2009) doi:10.1038/nature08582 30. X. Du, I. Skachko, F. Duerr, A. Luican, and E. Y. Andrei, Fractional quantum Hall effect and insulating phase of Dirac electrons in graphene, Nature 462(7270), 192 (2009) doi:10.1038/nature08522 31. C. R. Dean, A. F. Young, P. Cadden-Zimansky, L. Wang, H. Ren, K. Watanabe, T. Taniguchi, P. Kim, J. Hone, and K. L. Shepard, Multicomponent fractional quantum Hall effect in graphene, Nat. Phys. 7(9), 693 (2011) doi:10.1038/nphys2007 32. R. B. Laughlin, Quantized Hall conductivity in two dimensions, Phys. Rev. B 23(10), 5632 (1981) doi:10.1103/PhysRevB.23.5632 33. T. Matsui, H. Kambara, Y. Niimi, K. Tagami, M. Tsukada, and H. Fukuyama, STS Observations of Landau Levels at Graphite Surfaces, Phys. Rev. Lett. 94(22), 226403 (2005) doi:10.1103/PhysRevLett.94.226403 34. Y. Zhang, Z. Jiang, J. P. Small, M. S. Purewal, Y. W. Tan, M. Fazlollahi, J. D. Chudow, J. A. Jaszczak, H. L. Stormer, and P. Kim, Landau-Level Splitting in Graphene in High Magnetic Fields, Phys. Rev. Lett. 96(13), 136806 (2006) doi:10.1103/PhysRevLett.96.136806 35. A. Kou, B. E. Feldman, A. J. Levin, B. I. Halperin, K. Watanabe, T. Taniguchi, and A. Yacoby, Electron-hole asymmetric integer and fractional quantum Hall effect in bilayer graphene, Science 345(6192), 55 (2014) doi:10.1126/science.1250270 36. P. Maher, L. Wang, Y. Gao, C. Forsythe, T. Taniguchi, K. Watanabe, D. Abanin, Z. Papić, P. Cadden-Zimansky, J. Hone, P. Kim, and C. R. Dean, Tunable fractional quantum Hall phases in bilayer graphene, Science 345(6192), 61 (2014) doi:10.1126/science.1252875 37. Y. Shi, Y. Lee, S. Che, Z. Pi, T. Espiritu, P. Stepanov, D. Smirnov, C. N. Lau, and F. Zhang, Energy Gaps and Layer Polarization of Integer and Fractional Quantum Hall States in Bilayer Graphene, Phys. Rev. Lett. 116(5), 056601 (2016) doi:10.1103/PhysRevLett.116.056601 38. B. E. Feldman, B. Krauss, J. H. Smet, and A. Yacoby, Unconventional Sequence of Fractional Quantum Hall States in Suspended Graphene, Science 337(6099), 1196 (2012) doi:10.1126/science.1224784 39. G. M. Rutter, J. N. Crain, N. P. Guisinger, T. Li, P. N. First, and J. A. Stroscio, Scattering and Interference in Epitaxial Graphene, Science 317(5835), 219 (2007) doi:10.1126/science.1142882 40. K. S. Novoselov, Z. Jiang, Y. Zhang, S. Morozov, H. Stormer, U. Zeitler, J. Maan, G. Boebinger, P. Kim, and A. Geim, Room- Temperature Quantum Hall Effect in Graphene, Science 315(5817), 1379 (2007) doi:10.1126/science.1137201 41. J. G. Checkelsky, L. Li, and N. P. Ong, Zero-Energy State in Graphene in a High Magnetic Field, Phys. Rev. Lett. 100(20), 206801 (2008) doi:10.1103/PhysRevLett.100.206801 [42] editorial, Nat. Photonics, 2016, 10(4): 201 43. P. Wallace, The Band Theory of Graphite, Phys. Rev. 71(9), 622 (1947) doi:10.1103/PhysRev.71.622 44. A. L. Grushina, D. K. Ki, M. Koshino, A. A. Nicolet, C. Faugeras, E. McCann, M. Potemski, and A. F. Morpurgo, Insulating state in tetralayers reveals an even–odd interaction effect in multilayer graphene, Nat. Commun. 6, 6419 (2015) doi:10.1038/ncomms7419 45. L. J. Yin, S. Y. Li, J. B. Qiao, J. C. Nie, and L. He, Landau quantization in graphene monolayer, Bernal bilayer, and Bernal trilayer on graphite surface, Phys. Rev. B 91(11), 115405 (2015) doi:10.1103/PhysRevB.91.115405 46. E. J. Mele, Interlayer coupling in rotationally faulted multilayer graphenes, J. Phys. D Appl. Phys. 45(15), 154004 (2012) doi:10.1088/0022-3727/45/15/154004 47. E. McCann, and V. Fal'ko, Landau-Level Degeneracy and Quantum Hall Effect in a Graphite Bilayer, Phys. Rev. Lett. 96(8), 086805 (2006) doi:10.1103/PhysRevLett.96.086805 48. K. S. Novoselov, E. McCann, S. V. Morozov, V. I. Fal'ko, M. I. Katsnelson, U. Zeitler, D. Jiang, F. Schedin, and A. K. Geim, Unconventional quantum Hall effect and Berry's phase of 2 in bilayer graphene, Nat. Phys. 2(3), 177 (2006) doi:10.1038/nphys245 49. M. L. Sadowski, G. Martinez, M. Potemski, C. Berger, and W. A. de Heer, Landau Level Spectroscopy of Ultrathin Graphite Layers, Phys. Rev. Lett. 97(26), 266405 (2006) doi:10.1103/PhysRevLett.97.266405 50. D. S. Lee, C. Riedl, T. Beringer, A. H. Castro Neto, K. von Klitzing, U. Starke, and J. H. Smet, Quantum Hall Effect in Twisted Bilayer Graphene, Phys. Rev. Lett. 107(21), 216602 (2011) doi:10.1103/PhysRevLett.107.216602 51. J. D. Sanchez-Yamagishi, T. Taychatanapat, K. Watanabe, T. Taniguchi, A. Yacoby, and P. Jarillo-Herrero, Quantum Hall Effect, Screening, and Layer-Polarized Insulating States in Twisted Bilayer Graphene, Phys. Rev. Lett. 108(7), 076601 (2012) doi:10.1103/PhysRevLett.108.076601 [52] M. Morgenstern, physica status solidi (b), 2011, 248(11): 2423 53. E. Y. Andrei, G. Li, and X. Du, Electronic properties of graphene: a perspective from scanning tunneling microscopy and magnetotransport, Rep. Prog. Phys. 75(5), 056501 (2012) doi:10.1088/0034-4885/75/5/056501 54. A. Deshpande, and B. J. LeRoy, Scanning probe microscopy of graphene, Physica E 44(4), 743 (2012) doi:10.1016/j.physe.2011.11.024 55. D. L. Miller, K. D. Kubista, G. M. Rutter, M. Ruan, W. A. de Heer, P. N. First, and J. A. Stroscio, Observing the Quantization of Zero Mass Carriers in Graphene, Science 324(5929), 924 (2009) doi:10.1126/science.1171810 56. Y. J. Song, A. F. Otte, Y. Kuk, Y. Hu, D. B. Torrance, P. N. First, W. A. de Heer, H. Min, S. Adam, M. D. Stiles, A. H. MacDonald, and J. A. Stroscio, High-resolution tunnelling spectroscopy of a graphene quartet, Nature 467(7312), 185 (2010) doi:10.1038/nature09330 57. G. Li, A. Luican, and E. Y. Andrei, Scanning Tunneling Spectroscopy of Graphene on Graphite, Phys. Rev. Lett. 102(17), 176804 (2009) doi:10.1103/PhysRevLett.102.176804 58. Y. Niimi, H. Kambara, T. Matsui, D. Yoshioka, and H. Fukuyama, Real-Space Imaging of Alternate Localization and Extension of Quasi-Two-Dimensional Electronic States at Graphite Surfaces in Magnetic Fields, Phys. Rev. Lett. 97(23), 236804 (2006) doi:10.1103/PhysRevLett.97.236804 59. M. Morgenstern, J. Klijn, C. Meyer, and R. Wiesendanger, Real-Space Observation of Drift States in a Two-Dimensional Electron System at High Magnetic Fields, Phys. Rev. Lett. 90(5), 056804 (2003) doi:10.1103/PhysRevLett.90.056804 60. K. Hashimoto, C. Sohrmann, J. Wiebe, T. Inaoka, F. Meier, Y. Hirayama, R. A. Römer, R. Wiesendanger, and M. Morgenstern, Quantum Hall Transition in Real Space: From Localized to Extended States, Phys. Rev. Lett. 101(25), 256802 (2008) doi:10.1103/PhysRevLett.101.256802 61. Y. Okada, W. Zhou, C. Dhital, D. Walkup, Y. Ran, Z. Wang, S. D. Wilson, and V. Madhavan, Visualizing Landau Levels of Dirac Electrons in a One-Dimensional Potential, Phys. Rev. Lett. 109(16), 166407 (2012) doi:10.1103/PhysRevLett.109.166407 62. P. Cheng, C. Song, T. Zhang, Y. Zhang, Y. Wang, J. F. Jia, J. Wang, Y. Wang, B. F. Zhu, X. Chen, X. Ma, K. He, L. Wang, X. Dai, Z. Fang, X. Xie, X. L. Qi, C. X. Liu, S. C. Zhang, and Q. K. Xue, Landau Quantization of Topological Surface States in Bi 2 Se 3, Phys. Rev. Lett. 105(7), 076801 (2010) doi:10.1103/PhysRevLett.105.076801 63. Y. Jiang, Y. Wang, M. Chen, Z. Li, C. Song, K. He, L. Wang, X. Chen, X. Ma, and Q. K. Xue, Landau Quantization and the Thickness Limit of Topological Insulator Thin Films of Sb 2 Te 3, Phys. Rev. Lett. 108(1), 016401 (2012) doi:10.1103/PhysRevLett.108.016401 64. T. Hanaguri, K. Igarashi, M. Kawamura, H. Takagi, and T. Sasagawa, Momentum-resolved Landau-level spectroscopy of Dirac surface state in Bi 2 Se 3, Phys. Rev. B 82(8), 081305 (2010) (R) doi:10.1103/PhysRevB.82.081305 65. W. Bao, Z. Zhao, H. Zhang, G. Liu, P. Kratz, L. Jing, J. Velasco, D. Smirnov, and C. N. Lau, Magnetoconductance Oscillations and Evidence for Fractional Quantum Hall States in Suspended Bilayer and Trilayer Graphene, Phys. Rev. Lett. 105(24), 246601 (2010) doi:10.1103/PhysRevLett.105.246601 66. E. McCann, and M. Koshino, The electronic properties of bilayer graphene, Rep. Prog. Phys. 76(5), 056503 (2013) doi:10.1088/0034- 4885/76/5/056503 67. T. Ohta, A. Bostwick, T. Seyller, K. Horn, and E. Rotenberg, Controlling the Electronic Structure of Bilayer Graphene, Science 313(5789), 951 (2006) doi:10.1126/science.1130681 68. Y. Zhang, T. T. Tang, C. Girit, Z. Hao, M. C. Martin, A. Zettl, M. F. Crommie, Y. R. Shen, and F. Wang, Direct observation of a widely tunable bandgap in bilayer graphene, Nature 459(7248), 820 (2009) doi:10.1038/nature08105 69. E. Castro, K. Novoselov, S. Morozov, N. Peres, J. dos Santos, J. Nilsson, F. Guinea, A. Geim, and A. Neto, Biased Bilayer Graphene: Semiconductor with a Gap Tunable by the Electric Field Effect, Phys. Rev. Lett. 99(21), 216802 (2007) doi:10.1103/PhysRevLett.99.216802 70. G. Li, and E. Y. Andrei, Observation of Landau levels of Dirac fermions in graphite, Nat. Phys. 3(9), 623 (2007) doi:10.1038/nphys653 71. Y. Niimi, H. Kambara, and H. Fukuyama, Localized Distributions of Quasi-Two-Dimensional Electronic States near Defects Artificially Created at Graphite Surfaces in Magnetic Fields, Phys. Rev. Lett. 102(2), 026803 (2009) doi:10.1103/PhysRevLett.102.026803 72. L. J. Yin, Y. Zhang, J. B. Qiao, S. Y. Li, and L. He, Experimental observation of surface states and Landau levels bending in bilayer graphene, Phys. Rev. B 93(12), 125422 (2016) doi:10.1103/PhysRevB.93.125422 73. L.J. Yin, H. Jiang, J.B. Qiao, and L. He, Direct imaging of topological edge states at a bilayer graphene domain wall, Nat. Commun. 7, 11760 (2016) doi:10.1038/ncomms11760 74. M. Orlita, C. Faugeras, J. Schneider, G. Martinez, D. Maude, and M. Potemski, Graphite from the Viewpoint of Landau Level Spectroscopy: An Effective Graphene Bilayer and Monolayer, Phys. Rev. Lett. 102(16), 166401 (2009) doi:10.1103/PhysRevLett.102.166401 75. W. Yan, S. Y. Li, L. J. Yin, J. B. Qiao, J. C. Nie, and L. He, Spatially resolving unconventional interface Landau quantization in a graphene monolayer-bilayer planar junction, Phys. Rev. B 93(19), 195408 (2016) doi:10.1103/PhysRevB.93.195408 76. K. K. Bai, Y. C. Wei, J. B. Qiao, S. Y. Li, L. J. Yin, W. Yan, J. C. Nie, and L. He, Detecting giant electron-hole asymmetry in a graphene monolayer generated by strain and charged-defect scattering via Landau level spectroscopy, Phys. Rev. B 92(12), 121405 (2015) (R) doi:10.1103/PhysRevB.92.121405 77. G. M. Rutter, S. Jung, N. N. Klimov, D. B. Newell, N. B. Zhitenev, and J. A. Stroscio, Microscopic polarization in bilayer graphene, Nat. Phys. 7(8), 649 (2011) doi:10.1038/nphys1988 78. Y. Zhang, S.Y. Li, H. Huang, W.T. Li, J.B. Qiao, W.X. Wang, L.J. Yin, W. H. Duan, and L. He, Scanning tunneling microscopy ofπ magnetism of a single atomic vacancy in graphene. Phys. Rev. Lett. 117(16), 166801 (2016) DOI: 10.1103/PhysRevLett.117.166801 79. E. McCann, Asymmetry gap in the electronic band structure of bilayer graphene, Phys. Rev. B 74(16), 161403 (2006) (R) doi:10.1103/PhysRevB.74.161403 80. K. S. Kim, A. L. Walter, L. Moreschini, T. Seyller, K. Horn, E. Rotenberg, and A. Bostwick, Coexisting massive and massless Dirac fermions in symmetry-broken bilayer graphene, Nat. Mater. 12(10), 887 (2013) doi:10.1038/nmat3717 81. J. B. Oostinga, H. B. Heersche, X. Liu, A. F. Morpurgo, and L. M. Vandersypen, Gate-induced insulating state in bilayer graphene devices, Nat. Mater. 7(2), 151 (2008) doi:10.1038/nmat2082 82. K. F. Mak, C. H. Lui, J. Shan, and T. F. Heinz, Observation of an Electric-Field-Induced Band Gap in Bilayer Graphene by Infrared Spectroscopy, Phys. Rev. Lett. 102(25), 256405 (2009) doi:10.1103/PhysRevLett.102.256405 83. M. P. Lima, A. J. R. da Silva, and A. Fazzio, Splitting of the zero-energy edge states in bilayer graphene, Phys. Rev. B 81(4), 045430 (2010) doi:10.1103/PhysRevB.81.045430 84. Y. Zhao, P. Cadden-Zimansky, Z. Jiang, and P. Kim, Symmetry Breaking in the Zero-Energy Landau Level in Bilayer Graphene, Phys. Rev. Lett. 104(6), 066801 (2010) doi:10.1103/PhysRevLett.104.066801 85. M. Nakamura, E. V. Castro, and B. Dóra, Valley Symmetry Breaking in Bilayer Graphene: A Test of the Minimal Model, Phys. Rev. Lett. 103(26), 266804 (2009) doi:10.1103/PhysRevLett.103.266804 86. K. Shizuya, Pseudo-zero-mode Landau levels and collective excitations in bilayer graphene, Phys. Rev. B 79(16), 165402 (2009) doi:10.1103/PhysRevB.79.165402 87. T. Misumi, and K. Shizuya, Electromagnetic response and pseudo-zero-mode Landau levels of bilayer graphene in a magnetic field, Phys. Rev. B 77(19), 195423 (2008) doi:10.1103/PhysRevB.77.195423 88. S. Y. Zhou, G. H. Gweon, A. V. Fedorov, P. N. First, W. A. de Heer, D. H. Lee, F. Guinea, A. H. Castro Neto, and A. Lanzara, Substrate-induced bandgap opening in epitaxial graphene, Nat. Mater. 6(10), 770 (2007) doi:10.1038/nmat2003 89. E. A. Henriksen, D. Nandi, and J. P. Eisenstein, Quantum Hall Effect and Semimetallic Behavior of Dual-Gated ABA-Stacked Trilayer Graphene, Phys. Rev. X 2(1), 011004 (2012) doi:10.1103/PhysRevX.2.011004 90. M. Koshino, and E. McCann, Landau level spectra and the quantum Hall effect of multilayer graphene, Phys. Rev. B 83(16), 165443 (2011) doi:10.1103/PhysRevB.83.165443 91. W. Bao, L. Jing, J. Velasco, Y. Lee, G. Liu, D. Tran, B. Standley, M. Aykol, S. B. Cronin, D. Smirnov, M. Koshino, E. McCann, M. Bockrath, and C. N. Lau, Stacking-dependent band gap and quantum transport in trilayer graphene, Nat. Phys. 7(12), 948 (2011) doi:10.1038/nphys2103 92. S. H. Jhang, M. F. Craciun, S. Schmidmeier, S. Tokumitsu, S. Russo, M. Yamamoto, Y. Skourski, J. Wosnitza, S. Tarucha, J. Eroms, and C. Strunk, Stacking-order dependent transport properties of trilayer graphene, Phys. Rev. B 84(16), 161408 (2011) (R) doi:10.1103/PhysRevB.84.161408 93. M. F. Craciun, S. Russo, M. Yamamoto, J. B. Oostinga, A. F. Morpurgo, and S. Tarucha, Trilayer graphene is a semimetal with a gate- tunable band overlap, Nat. Nanotechnol. 4(6), 383 (2009) doi:10.1038/nnano.2009.89 94. C. H. Lui, Z. Li, K. F. Mak, E. Cappelluti, and T. F. Heinz, Observation of an electrically tunable band gap in trilayer graphene, Nat. Phys. 7(12), 944 (2011) doi:10.1038/nphys2102 95. T. Taychatanapat, K. Watanabe, T. Taniguchi, and P. Jarillo-Herrero, Quantum Hall effect and Landau-level crossing of Dirac fermions in trilayer graphene, Nat. Phys. 7(8), 621 (2011) doi:10.1038/nphys2008 96. D. Xiao, W. Yao, and Q. Niu, Valley-Contrasting Physics in Graphene: Magnetic Moment and Topological Transport, Phys. Rev. Lett. 99(23), 236809 (2007) doi:10.1103/PhysRevLett.99.236809 97. S. Takei, A. Yacoby, B. I. Halperin, and Y. Tserkovnyak, Spin Superfluidity in the  = 0 Quantum Hall State of Graphene, Phys. Rev. Lett. 116(21), 216801 (2016) doi:10.1103/PhysRevLett.116.216801 98. N. Tombros, C. Jozsa, M. Popinciuc, H. T. Jonkman, and B. J. van Wees, Electronic spin transport and spin precession in single graphene layers at room temperature, Nature 448(7153), 571 (2007) doi:10.1038/nature06037 99. S. Morozov, K. Novoselov, M. Katsnelson, F. Schedin, D. Elias, J. Jaszczak, and A. Geim, Giant Intrinsic Carrier Mobilities in Graphene and Its Bilayer, Phys. Rev. Lett. 100(1), 016602 (2008) doi:10.1103/PhysRevLett.100.016602 100. A. F. Young, C. R. Dean, L. Wang, H. Ren, P. Cadden-Zimansky, K. Watanabe, T. Taniguchi, J. Hone, K. L. Shepard, and P. Kim, Spin and valley quantum Hall ferromagnetism in graphene, Nat. Phys. 8(7), 550 (2012) doi:10.1038/nphys2307 101. J. Mao, Y. Jiang, D. Moldovan, G. Li, K. Watanabe, T. Taniguchi, M. R. Masir, F. M. Peeters, and E. Y. Andrei, Realization of a tunable artificial atom at a supercritically charged vacancy in graphene, Nat. Phys. 12(6), 545 (2016) doi:10.1038/nphys3665 102. Y. S. Fu, M. Kawamura, K. Igarashi, H. Takagi, T. Hanaguri, and T. Sasagawa, Imaging the two-component nature of Dirac–Landau levels in the topological surface state of Bi2Se3, Nat. Phys. 10(11), 815 (2014) doi:10.1038/nphys3084 103. H. Min, J. E. Hill, N. A. Sinitsyn, B. R. Sahu, L. Kleinman, and A. H. MacDonald, Intrinsic and Rashba spin-orbit interactions in graphene sheets, Phys. Rev. B 74(16), 165310 (2006) doi:10.1103/PhysRevB.74.165310 104. G. Li, A. Luican-Mayer, D. Abanin, L. Levitov, and E. Y. Andrei, Evolution of Landau levels into edge states in graphene, Nat. Commun. 4, 1744 (2013) doi:10.1038/ncomms2767 105. R. V. Gorbachev, J. C. W. Song, G. L. Yu, A. V. Kretinin, F. Withers, Y. Cao, A. Mishchenko, I. V. Grigorieva, K. S. Novoselov, L. S. Levitov, and A. K. Geim, Detecting topological currents in graphene superlattices, Science 346(6208), 448 (2014) doi:10.1126/science.1254966 106. M. Koshino, and T. Ando, Anomalous orbital magnetism in Dirac-electron systems: Role of pseudospin paramagnetism, Phys. Rev. B 81(19), 195431 (2010) doi:10.1103/PhysRevB.81.195431 107. J. L. Lado, J. W. González, and J. Fernández-Rossier, Quantum Hall effect in gapped graphene heterojunctions, Phys. Rev. B 88(3), 035448 (2013) doi:10.1103/PhysRevB.88.035448 108. W. X. Wang, L. J. Yin, J. B. Qiao, T. Cai, S. Y. Li, R. F. Dou, J. C. Nie, X. Wu, and L. He, Atomic resolution imaging of the two- component Dirac-Landau levels in a gapped graphene monolayer, Phys. Rev. B 92(16), 165420 (2015) doi:10.1103/PhysRevB.92.165420 109. V. P. Gusynin, V. A. Miransky, S. G. Sharapov, and I. A. Shovkovy, Edge states in quantum Hall effect in graphene (Review Article), Low Temp. Phys. 34(10), 778 (2008) doi:10.1063/1.2981387 110. M. T. Allen, O. Shtanko, I. C. Fulga, A. R. Akhmerov, K. Watanabe, T. Taniguchi, P. Jarillo-Herrero, L. S. Levitov, and A. Yacoby, Nat. Phys. 2(12), 128 (2016) 111. D. A. Abanin, K. S. Novoselov, U. Zeitler, P. A. Lee, A. K. Geim, and L. S. Levitov, Dissipative Quantum Hall Effect in Graphene near the Dirac Point, Phys. Rev. Lett. 98(19), 196806 (2007) doi:10.1103/PhysRevLett.98.196806 112. D. A. Abanin, P. A. Lee, and L. S. Levitov, Spin-Filtered Edge States and Quantum Hall Effect in Graphene, Phys. Rev. Lett. 96(17), 176803 (2006) doi:10.1103/PhysRevLett.96.176803 113. H. A. Fertig, and L. Brey, Luttinger Liquid at the Edge of Undoped Graphene in a Strong Magnetic Field, Phys. Rev. Lett. 97(11), 116805 (2006) doi:10.1103/PhysRevLett.97.116805 114. A. F. Young, J. D. Sanchez-Yamagishi, B. Hunt, S. H. Choi, K. Watanabe, T. Taniguchi, R. C. Ashoori, and P. Jarillo-Herrero, Tunable symmetry breaking and helical edge transport in a graphene quantum spin Hall state, Nature 505(7484), 528 (2014) doi:10.1038/nature12800 115. P. Maher, C. R. Dean, A. F. Young, T. Taniguchi, K. Watanabe, K. L. Shepard, J. Hone, and P. Kim, Evidence for a spin phase transition at charge neutrality in bilayer graphene, Nat. Phys. 9(3), 154 (2013) doi:10.1038/nphys2528 116. K. Lee, B. Fallahazad, J. Xue, D. C. Dillen, K. Kim, T. Taniguchi, K. Watanabe, and E. Tutuc, Chemical potential and quantum Hall ferromagnetism in bilayer graphene, Science 345(6192), 58 (2014) doi:10.1126/science.1251003 117. M. Kharitonov, Canted Antiferromagnetic Phase of the  = 0 Quantum Hall State in Bilayer Graphene, Phys. Rev. Lett. 109(4), 046803 (2012) doi:10.1103/PhysRevLett.109.046803 118. K. Nomura, and A. H. MacDonald, Quantum Hall Ferromagnetism in Graphene, Phys. Rev. Lett. 96(25), 256602 (2006) doi:10.1103/PhysRevLett.96.256602 119. H. Ito, K. Furuya, Y. Shibata, S. Kashiwaya, M. Yamaguchi, T. Akazaki, H. Tamura, Y. Ootuka, and S. Nomura, Near-Field Optical Mapping of Quantum Hall Edge States, Phys. Rev. Lett. 107(25), 256803 (2011) doi:10.1103/PhysRevLett.107.256803 120. K. Lai, W. Kundhikanjana, M. A. Kelly, Z. X. Shen, J. Shabani, and M. Shayegan, Imaging of Coulomb-Driven Quantum Hall Edge States, Phys. Rev. Lett. 107(17), 176809 (2011) doi:10.1103/PhysRevLett.107.176809 121. J. Tian, Y. Jiang, I. Childres, H. Cao, J. Hu, and Y. P. Chen, Quantum Hall effect in monolayer-bilayer graphene planar junctions, Phys. Rev. B 88(12), 125410 (2013) doi:10.1103/PhysRevB.88.125410 122. Y. Kobayashi, K. i. Fukui, T. Enoki, and K. Kusakabe, Edge state on hydrogen-terminated graphite edges investigated by scanning tunneling microscopy, Phys. Rev. B 73(12), 125415 (2006) doi:10.1103/PhysRevB.73.125415 123. Y. Kobayashi, K. i. Fukui, T. Enoki, K. Kusakabe, and Y. Kaburagi, Observation of zigzag and armchair edges of graphite using scanning tunneling microscopy and spectroscopy, Phys. Rev. B 71(19), 193406 (2005) doi:10.1103/PhysRevB.71.193406 124. K. A. Ritter, and J. W. Lyding, The influence of edge structure on the electronic properties of graphene quantum dots and nanoribbons, Nat. Mater. 8(3), 235 (2009) doi:10.1038/nmat2378 125. C. Tao, L. Jiao, O. V. Yazyev, Y. C. Chen, J. Feng, X. Zhang, R. B. Capaz, J. M. Tour, A. Zettl, S. G. Louie, H. Dai, and M. F. Crommie, Spatially resolving edge states of chiral graphene nanoribbons, Nat. Phys. 7(8), 616 (2011) doi:10.1038/nphys1991 126. J. Xue, J. Sanchez-Yamagishi, K. Watanabe, T. Taniguchi, P. Jarillo-Herrero, and B. J. LeRoy, Long-Wavelength Local Density of States Oscillations Near Graphene Step Edges, Phys. Rev. Lett. 108(1), 016801 (2012) doi:10.1103/PhysRevLett.108.016801 127. Y. Niimi, T. Matsui, H. Kambara, K. Tagami, M. Tsukada, and H. Fukuyama, Scanning tunneling microscopy and spectroscopy of the electronic local density of states of graphite surfaces near monoatomic step edges, Phys. Rev. B 73(8), 085421 (2006) doi:10.1103/PhysRevB.73.085421 128. Y. Y. Li, M. X. Chen, M. Weinert, and L. Li, Nat. Commun. 5, 4311 (2014) 129. E. Castro, N. Peres, J. Lopes dos Santos, A. Neto, and F. Guinea, Localized States at Zigzag Edges of Bilayer Graphene, Phys. Rev. Lett. 100(2), 026802 (2008) doi:10.1103/PhysRevLett.100.026802 130. E. V. Castro, N. M. R. Peres, and J. M. B. Lopes dos Santos, Localized states at zigzag edges of multilayer graphene and graphite steps, Europhys. Lett. 84(1), 17001 (2008) doi:10.1209/0295-5075/84/17001 131. T. Wassmann, A. Seitsonen, A. Saitta, M. Lazzeri, and F. Mauri, Structure, Stability, Edge States, and Aromaticity of Graphene Ribbons, Phys. Rev. Lett. 101(9), 096402 (2008) doi:10.1103/PhysRevLett.101.096402 132. E. V. Castro, M. P. López-Sancho, and M. A. H. Vozmediano, New Type of Vacancy-Induced Localized States in Multilayer Graphene, Phys. Rev. Lett. 104(3), 036802 (2010) doi:10.1103/PhysRevLett.104.036802 133. M. Han, B. Özyilmaz, Y. Zhang, and P. Kim, Energy Band-Gap Engineering of Graphene Nanoribbons, Phys. Rev. Lett. 98(20), 206805 (2007) doi:10.1103/PhysRevLett.98.206805 134. Y. W. Son, M. L. Cohen, and S. G. Louie, Half-metallic graphene nanoribbons, Nature 444(7117), 347 (2006) doi:10.1038/nature05180 135. G. Z. Magda, X. Jin, I. Hagymasi, P. Vancso, Z. Osvath, P. Nemes-Incze, C. Hwang, L. P. Biro, and L. Tapaszto, Room-temperature magnetic order on zigzag edges of narrow graphene nanoribbons, Nature 514(7524), 608 (2014) doi:10.1038/nature13831 136. D. A. Abanin, P. A. Lee, and L. S. Levitov, Charge and spin transport at the quantum Hall edge of graphene, Solid State Commun. 143(1-2), 77 (2007) doi:10.1016/j.ssc.2007.04.024 137. L. Brey, and H. A. Fertig, Edge states and the quantized Hall effect in graphene, Phys. Rev. B 73(19), 195408 (2006) doi:10.1103/PhysRevB.73.195408 138. V. Mazo, E. Shimshoni, and H. A. Fertig, Edge states of bilayer graphene in the quantum Hall regime, Phys. Rev. B 84(4), 045405 (2011) doi:10.1103/PhysRevB.84.045405 139. D. B. Chklovskii, B. I. Shklovskii, and L. I. Glazman, Electrostatics of edge channels, Phys. Rev. B 46(7), 4026 (1992) doi:10.1103/PhysRevB.46.4026 140. C. Cong, T. Yu, K. Sato, J. Shang, R. Saito, G. F. Dresselhaus, and M. S. Dresselhaus, Raman Characterization of ABA- and ABC- Stacked Trilayer Graphene, ACS Nano 5(11), 8760 (2011) doi: 10.1021/nn203472f 141. C. H. Lui, Z. Li, Z. Chen, P. V. Klimov, L. E. Brus, and T. F. Heinz, Imaging Stacking Order in Few-Layer Graphene, Nano Lett. 11(1), 164 (2011) doi:10.1021/nl1032827 142. L.J. Yin, J.B. Qiao, and L. He, Structures and Electronic Properties of Twisted Bilayer Graphene, Progress in Physics 36(3), 65 (2016) (in Chinese) doi: 10.13725/j.cnki.pip.2016.03.001 143. C. Park, J. Ryou, S. Hong, B. G. Sumpter, G. Kim, and M. Yoon, Electronic Properties of Bilayer Graphene Strongly Coupled to Interlayer Stacking and an External Electric Field, Phys. Rev. Lett. 115(1), 015502 (2015) doi:10.1103/PhysRevLett.115.015502 144. J. Yin, H. Wang, H. Peng, Z. Tan, L. Liao, L. Lin, X. Sun, A. L. Koh, Y. Chen, H. Peng, and Z. Liu, Selectively enhanced photocurrent generation in twisted bilayer graphene with van Hove singularity, Nat. Commun. 7, 10699 (2016) doi:10.1038/ncomms10699 145. W. Y. He, Y. Su, M. Yang, and L. He, Creating in-plane pseudomagnetic fields in excess of 1000 T by misoriented stacking in a graphene bilayer, Phys. Rev. B 89(12), 125418 (2014) doi:10.1103/PhysRevB.89.125418 146. B. Butz, C. Dolle, F. Niekiel, K. Weber, D. Waldmann, H. B. Weber, B. Meyer, and E. Spiecker, Dislocations in bilayer graphene, Nature 505(7484), 533 (2014) doi:10.1038/nature12780 147. J. S. Alden, A. W. Tsen, P. Y. Huang, R. Hovden, L. Brown, J. Park, D. A. Muller, and P. L. McEuen, Strain solitons and topological defects in bilayer graphene, Proc. Natl. Acad. Sci. U.S.A. 110(28), 11256 (2013) doi:10.1073/pnas.1309394110 148. J. Lin, W. Fang, W. Zhou, A. R. Lupini, J. C. Idrobo, J. Kong, S. J. Pennycook, and S. T. Pantelides, AC/AB Stacking Boundaries in Bilayer Graphene, Nano Lett. 13(7), 3262 (2013) doi:10.1021/nl4013979 149. M. Koshino, Electronic transmission through A B - B A domain boundary in bilayer graphene, Phys. Rev. B 88(11), 115409 (2013) doi:10.1103/PhysRevB.88.115409 150. L.J. Yin, W.X. Wang, Y. Zhang, Y.Y. Ou, H.T. Zhang, C.Y. Shen, and L. He, Observation of chirality transition of quasiparticles at stacking solitons in trilayer graphene, Phys. Rev. B 95(08), 081402 (2017) (R) doi: 10.1103/PhysRevB.95.081402 151. L. Jiang, Z. Shi, B. Zeng, S. Wang, J. H. Kang, T. Joshi, C. Jin, L. Ju, J. Kim, T. Lyu, Y. R. Shen, M. Crommie, H. J. Gao, and F. Wang, Soliton-dependent plasmon reflection at bilayer graphene domain walls, Nat. Mater. 15(8), 840 (2016) doi:10.1038/nmat4653 152. F. Zhang, A. H. MacDonald, and E. J. Mele, Valley Chern numbers and boundary modes in gapped bilayer graphene, Proc. Natl. Acad. Sci. U.S.A. 110(26), 10546 (2013) doi:10.1073/pnas.1308853110 153. A. Vaezi, Y. Liang, D. H. Ngai, L. Yang, and E. A. Kim, Topological Edge States at a Tilt Boundary in Gated Multilayer Graphene, Phys. Rev. X 3(2), 021018 (2013) doi:10.1103/PhysRevX.3.021018 154. J. Jung, F. Zhang, Z. Qiao, and A. H. MacDonald, Valley-Hall kink and edge states in multilayer graphene, Phys. Rev. B 84(7), 075418 (2011) doi:10.1103/PhysRevB.84.075418 155. L. Ju, Z. Shi, N. Nair, Y. Lv, C. Jin, C. Velasco, H. A. Ojeda-Aristizabal, M. C. Bechtel, A. Martin, J. Zettl, Analytis, and F. Wang, Topological valley transport at bilayer graphene domain walls, Nature 520(7549), 650 (2015) doi:10.1038/nature14364 156. P. San-Jose, R. V. Gorbachev, A. K. Geim, K. S. Novoselov, and F. Guinea, Stacking Boundaries and Transport in Bilayer Graphene, Nano Lett. 14(4), 2052 (2014) doi:10.1021/nl500230a 157. D. Pierucci, H. Sediri, M. Hajlaoui, J.C. Girard, T. Brumme, M. Calandra, E. Velez-Fort, G. Patriarche, M. G. Silly, G. Ferro, V. Soulière, M. Marangolo, F. Sirotti, F. Mauri, and A. Ouerghi, Evidence for Flat Bands near the Fermi Level in Epitaxial Rhombohedral Multilayer Graphene., ACS Nano 9(5), 5432 (2015) doi:10.1021/acsnano.5b01239 158. P. Xu, Y. Yang, D. Qi, S. D. Barber, M. L. Ackerman, J. K. Schoelz, T. B. Bothwell, S. Barraza-Lopez, L. Bellaiche, and P. M. Thibado, A pathway between Bernal and rhombohedral stacked graphene layers with scanning tunneling microscopy, Appl. Phys. Lett. 100(20), 201601 (2012) doi:10.1063/1.4716475 159. I. Martin, Y. Blanter, and A. Morpurgo, Topological Confinement in Bilayer Graphene, Phys. Rev. Lett. 100(3), 036804 (2008) doi:10.1103/PhysRevLett.100.036804 160. M. Zarenia, J. M. Pereira, G. A. Farias, and F. M. Peeters, Chiral states in bilayer graphene: Magnetic field dependence and gap opening, Phys. Rev. B 84(12), 125451 (2011) doi:10.1103/PhysRevB.84.125451 161. W. Yao, S. Yang, and Q. Niu, Edge States in Graphene: From Gapped Flat-Band to Gapless Chiral Modes, Phys. Rev. Lett. 102(9), 096801 (2009) doi:10.1103/PhysRevLett.102.096801 162. L. Zhang, Y. Zhang, J. Camacho, M. Khodas, and I. Zaliznyak, The experimental observation of quantum Hall effect of l=3 chiral quasiparticles in trilayer graphene, Nat. Phys. 7(12), 953 (2011) doi:10.1038/nphys2104 163. S. Yuan, R. Roldán, and M. I. Katsnelson, Landau level spectrum of ABA - and ABC -stacked trilayer graphene, Phys. Rev. B 84(12), 125455 (2011) doi:10.1103/PhysRevB.84.125455 164. M. G. Menezes, R. B. Capaz, and S. G. Louie, Ab initio quasiparticle band structure of ABA and ABC-stacked graphene trilayers, Phys. Rev. B 89(3), 035431 (2014) doi:10.1103/PhysRevB.89.035431 165. F. Zhang, B. Sahu, H. Min, and A. H. MacDonald, Band structure of A B C -stacked graphene trilayers, Phys. Rev. B 82(3), 035409 (2010) doi:10.1103/PhysRevB.82.035409 166. Y. Barlas, R. Cote, and M. Rondeau, Quantum Hall to Charge-Density-Wave Phase Transitions in A B C -Trilayer Graphene, Phys. Rev. Lett. 109(12), 126804 (2012) doi:10.1103/PhysRevLett.109.126804 167. R. Côté, M. Rondeau, A. M. Gagnon, and Y. Barlas, Phase diagram of insulating crystal and quantum Hall states in ABC-stacked trilayer graphene, Phys. Rev. B 86(12), 125422 (2012) doi:10.1103/PhysRevB.86.125422 168. R. Xu, L. J. Yin, J. B. Qiao, K. K. Bai, J. C. Nie, and L. He, Direct probing of the stacking order and electronic spectrum of rhombohedral trilayer graphene with scanning tunneling microscopy, Phys. Rev. B 91(3), 035410 (2015) doi:10.1103/PhysRevB.91.035410 169. S. H. R. Sena, J. M. Pereira, F. M. Peeters, and G. A. Farias, Landau levels in asymmetric graphene trilayers, Phys. Rev. B 84(20), 205448 (2011) doi:10.1103/PhysRevB.84.205448 170. M. Koshino, and E. McCann, Trigonal warping and Berry's phase N  in ABC-stacked multilayer graphene, Phys. Rev. B 80(16), 165409 (2009) doi:10.1103/PhysRevB.80.165409 171. F. Guinea, A. Castro Neto, and N. Peres, Electronic states and Landau levels in graphene stacks, Phys. Rev. B 73(24), 245426 (2006) doi:10.1103/PhysRevB.73.245426 172. J. Jung, and A. H. MacDonald, Gapped broken symmetry states in ABC-stacked trilayer graphene, Phys. Rev. B 88(7), 075408 (2013) doi:10.1103/PhysRevB.88.075408 173. F. Zhang, D. Tilahun, and A. H. MacDonald, Hund's rules for the N = 0 Landau levels of trilayer graphene, Phys. Rev. B 85(16), 165139 (2012) doi:10.1103/PhysRevB.85.165139 174. R. Bistritzer, and A. H. MacDonald, Moiré butterflies in twisted bilayer graphene, Phys. Rev. B 84(3), 035440 (2011) doi:10.1103/PhysRevB.84.035440 175. E. J. Mele, Commensuration and interlayer coherence in twisted bilayer graphene, Phys. Rev. B 81(16), 161405 (2010) (R) doi:10.1103/PhysRevB.81.161405 176. P. Moon, and M. Koshino, Energy spectrum and quantum Hall effect in twisted bilayer graphene, Phys. Rev. B 85(19), 195458 (2012) doi:10.1103/PhysRevB.85.195458 177. L. Wang, Y. Gao, B. Wen, Z. Han, T. Taniguchi, K. Watanabe, M. Koshino, J. Hone, and C. R. Dean, Evidence for a fractional fractal quantum Hall effect in graphene superlattices, Science 350(6265), 1231 (2015) doi:10.1126/science.aad2102 178. G. Li, A. Luican, J. M. B. Lopes dos Santos, A. H. Castro Neto, A. Reina, J. Kong, and E. Y. Andrei, Observation of Van Hove singularities in twisted graphene layers, Nat. Phys. 6(2), 109 (2009) doi:10.1038/nphys1463 179. W. Yan, M. Liu, R. F. Dou, L. Meng, L. Feng, Z. D. Chu, Y. Zhang, Z. Liu, J. C. Nie, and L. He, Angle-Dependent van Hove Singularities in a Slightly Twisted Graphene Bilayer, Phys. Rev. Lett. 109(12), 126801 (2012) doi:10.1103/PhysRevLett.109.126801 180. I. Brihuega, P. Mallet, H. González-Herrero, G. Trambly de Laissardière, M. M. Ugeda, L. Magaud, J. M. Gómez-Rodríguez, F. Ynduráin, and J. Y. Veuillen, Unraveling the Intrinsic and Robust Nature of van Hove Singularities in Twisted Bilayer Graphene by Scanning Tunneling Microscopy and Theoretical Analysis, Phys. Rev. Lett. 109(19), 196802 (2012) doi:10.1103/PhysRevLett.109.196802 181. W. Yan, L. Meng, M. Liu, J. B. Qiao, Z. D. Chu, R. F. Dou, Z. Liu, J. C. Nie, D. G. Naugle, and L. He, Angle-dependent van Hove singularities and their breakdown in twisted graphene bilayers, Phys. Rev. B 90(7), 115402 (2014) doi:10.1103/PhysRevB.90.115402 182. L. J. Yin, J. B. Qiao, W. X. Wang, W. J. Zuo, W. Yan, R. Xu, R. F. Dou, J. C. Nie, and L. He, Landau quantization and Fermi velocity renormalization in twisted graphene bilayers, Phys. Rev. B 92(20), 201408 (2015) (R) doi:10.1103/PhysRevB.92.201408 183. A. Luican, G. Li, A. Reina, J. Kong, R. R. Nair, K. S. Novoselov, A. K. Geim, and E. Y. Andrei, Single-Layer Behavior and Its Breakdown in Twisted Graphene Layers, Phys. Rev. Lett. 106(12), 126802 (2011) doi:10.1103/PhysRevLett.106.126802 184. L. J. Yin, J. B. Qiao, W. J. Zuo, W. T. Li, and L. He, Experimental evidence for non-Abelian gauge potentials in twisted graphene bilayers, Phys. Rev. B 92(8), 081406 (2015) (R) doi:10.1103/PhysRevB.92.081406 185. P. San-Jose, J. González, and F. Guinea, Non-Abelian Gauge Potentials in Graphene Bilayers, Phys. Rev. Lett. 108(21), 216802 (2012) doi:10.1103/PhysRevLett.108.216802 186. E. Suárez Morell, J. D. Correa, P. Vargas, M. Pacheco, and Z. Barticevic, Flat bands in slightly twisted bilayer graphene: Tight- binding calculations, Phys. Rev. B 82(12), 121407 (2010) doi:10.1103/PhysRevB.82.121407 187. G. Trambly de Laissardière, D. Mayou, and L. Magaud, Localization of Dirac Electrons in Rotated Graphene Bilayers, Nano Lett. 10(3), 804 (2010) doi:10.1021/nl902948m 188. R. Bistritzer, and A. H. MacDonald, Moire bands in twisted double-layer graphene, Proc. Natl. Acad. Sci. U.S.A. 108(30), 12233 (2011) doi:10.1073/pnas.1108174108 189. J. M. B. Lopes dos Santos, N. M. R. Peres, and A. H. Castro Neto, Graphene Bilayer with a Twist: Electronic Structure, Phys. Rev. Lett. 99(25), 256802 (2007) doi:10.1103/PhysRevLett.99.256802 190. J. M. B. Lopes dos Santos, N. M. R. Peres, and A. H. Castro Neto, Continuum model of the twisted graphene bilayer, Phys. Rev. B 86(15), 155449 (2012) doi:10.1103/PhysRevB.86.155449 191. A. O. Sboychakov, A. L. Rakhmanov, A. V. Rozhkov, and F. Nori, Electronic spectrum of twisted bilayer graphene, Phys. Rev. B 92(7), 075402 (2015) doi:10.1103/PhysRevB.92.075402 192. Z. D. Chu, W. Y. He, and L. He, Coexistence of van Hove singularities and superlattice Dirac points in a slightly twisted graphene bilayer, Phys. Rev. B 87(15), 155419 (2013) doi:10.1103/PhysRevB.87.155419 193. D. Wong, Y. Wang, J. Jung, S. Pezzini, A. M. DaSilva, H. Z. Tsai, H. S. Jung, R. Khajeh, Y. Kim, J. Lee, S. Kahn, S. Tollabimazraehno, H. Rasool, K. Watanabe, T. Taniguchi, A. Zettl, S. Adam, A. H. MacDonald, and M. F. Crommie, Local spectroscopy of moiré-induced electronic structure in gate-tunable twisted bilayer graphene, Phys. Rev. B 92(15), 155409 (2015) doi:10.1103/PhysRevB.92.155409 194. B. Cheng, Y. Wu, P. Wang, C. Pan, T. Taniguchi, K. Watanabe, and M. Bockrath, Gate-Tunable Landau Level Filling and Spectroscopy in Coupled Massive and Massless Electron Systems, Phys. Rev. Lett. 117(2), 026601 (2016) doi:10.1103/PhysRevLett.117.026601 195. J. B. Qiao, and L. He, In-plane chiral tunneling and out-of-plane valley-polarized quantum tunneling in twisted graphene trilayer, Phys. Rev. B 90(7), 075410 (2014) doi:10.1103/PhysRevB.90.075410 196. L. J. Yin, J. B. Qiao, W. X. Wang, Z. D. Chu, K. F. Zhang, R. F. Dou, C. L. Gao, J. F. Jia, J. C. Nie, and L. He, Tuning structures and electronic spectra of graphene layers with tilt grain boundaries, Phys. Rev. B 89(20), 205410 (2014) doi:10.1103/PhysRevB.89.205410 197. L.J. Yin, J.B. Qiao, W. Yan, R. Xu, R.F. Dou, J.C. Nie, and L. He, arXiv:1410.1621, 2014 198. K. K. Bai, Y. Zhou, H. Zheng, L. Meng, H. Peng, Z. Liu, J. C. Nie, and L. He, Creating One-Dimensional Nanoscale Periodic Ripples in a Continuous Mosaic Graphene Monolayer, Phys. Rev. Lett. 113(8), 086102 (2014) doi:10.1103/PhysRevLett.113.086102 199. W. Yan, W. Y. He, Z. D. Chu, M. Liu, L. Meng, R. F. Dou, Y. Zhang, Z. Liu, J. C. Nie, and L. He, Nat. Commun. 4, 2159 (2013) 200. H. Yan, C. C. Liu, K. K. Bai, X. Wang, M. Liu, W. Yan, L. Meng, Y. Zhang, Z. Liu, R. Dou, J. C. Nie, Y. Yao, and L. He, Electronic structures of graphene layers on a metal foil: The effect of atomic-scale defects, Appl. Phys. Lett. 103(14), 143120 (2013) doi:10.1063/1.4824206 201. S. Jung, G. M. Rutter, N. N. Klimov, D. B. Newell, I. Calizo, A. R. Hight-Walker, N. B. Zhitenev, and J. A. Stroscio, Evolution of microscopic localization in graphene in a magnetic field from scattering resonances to quantum dots, Nat. Phys. 7(3), 245 (2011) doi:10.1038/nphys1866 202. V. W. B. Yuanbo Zhang, Caglar Girit, Alex Zettl and Michael F. Crommie, Nat. Phys. 5, 722 (2009) 203. H. Yan, Y. Sun, L. He, J. C. Nie, and M. H. W. Chan, Observation of Landau-level-like quantization at 77 K along a strained-induced graphene ridge, Phys. Rev. B 85(3), 035422 (2012) doi:10.1103/PhysRevB.85.035422 204. N. Levy, S. A. Burke, K. L. Meaker, M. Panlasigui, A. Zettl, F. Guinea, A. H. Castro Neto, and M. F. Crommie, Strain-Induced Pseudo-Magnetic Fields Greater Than 300 Tesla in Graphene Nanobubbles, Science 329(5991), 544 (2010) doi:10.1126/science.1191700 205. L. Meng, W. Y. He, H. Zheng, M. Liu, H. Yan, W. Yan, Z. D. Chu, K. Bai, R. F. Dou, Y. Zhang, Z. Liu, J. C. Nie, and L. He, Strain- induced one-dimensional Landau level quantization in corrugated graphene, Phys. Rev. B 87(20), 205405 (2013) doi:10.1103/PhysRevB.87.205405 206. J. H. Chen, C. Jang, S. Adam, M. S. Fuhrer, E. D. Williams, and M. Ishigami, Charged-impurity scattering in graphene, Nat. Phys. 4(5), 377 (2008) doi:10.1038/nphys935 207. J. R. Williams, T. Low, M. S. Lundstrom, and C. M. Marcus, Gate-controlled guiding of electrons in graphene, Nat. Nanotechnol. 6(4), 222 (2011) doi:10.1038/nnano.2011.3 208. Y. Wang, V. W. Brar, A. V. Shytov, Q. Wu, W. Regan, H. Z. Tsai, A. Zettl, L. S. Levitov, and M. F. Crommie, Mapping Dirac quasiparticles near a single Coulomb impurity on graphene, Nat. Phys. 8(9), 653 (2012) doi:10.1038/nphys2379 209. M. Kühne, C. Faugeras, P. Kossacki, A. A. L. Nicolet, M. Orlita, Y. I. Latyshev, and M. Potemski, Polarization-resolved magneto- Raman scattering of graphenelike domains on natural graphite, Phys. Rev. B 85(19), 195406 (2012) doi:10.1103/PhysRevB.85.195406 210. A. Kretinin, G. L. Yu, R. Jalil, Y. Cao, F. Withers, A. Mishchenko, M. I. Katsnelson, K. S. Novoselov, A. K. Geim, and F. Guinea, Quantum capacitance measurements of electron-hole asymmetry and next-nearest-neighbor hopping in graphene, Phys. Rev. B 88(16), 165427 (2013) doi:10.1103/PhysRevB.88.165427 211. V. M. Pereira, J. Nilsson, and A. H. Castro Neto, Coulomb Impurity Problem in Graphene, Phys. Rev. Lett. 99(16), 166802 (2007) doi:10.1103/PhysRevLett.99.166802 212. D. S. Novikov, Elastic scattering theory and transport in graphene, Phys. Rev. B 76(24), 245435 (2007) doi:10.1103/PhysRevB.76.245435 213. D. S. Novikov, Numbers of donors and acceptors from transport measurements in graphene, Appl. Phys. Lett. 91(10), 102102 (2007) doi:10.1063/1.2779107 214. S. Y. Li, K. K. Bai, L. J. Yin, J. B. Qiao, W. X. Wang, and L. He, Observation of unconventional splitting of Landau levels in strained graphene, Phys. Rev. B 92(24), 245302 (2015) doi:10.1103/PhysRevB.92.245302 215. F. de Juan, A. Cortijo, M. A. H. Vozmediano, and A. Cano, Aharonov–Bohm interferences from local deformations in graphene, Nat. Phys. 7(10), 810 (2011) doi:10.1038/nphys2034 216. M. A. H. Vozmediano, M. I. Katsnelson, and F. Guinea, Physics Reports-Review Section, Phys. Lett. 496(4-5), 109 (2010) 217. B. Uchoa, and Y. Barlas, Superconducting States in Pseudo-Landau-Levels of Strained Graphene, Phys. Rev. Lett. 111(4), 046604 (2013) doi:10.1103/PhysRevLett.111.046604 218. B. Roy, Z. X. Hu, and K. Yang, Theory of unconventional quantum Hall effect in strained graphene, Phys. Rev. B 87(12), 121408 (2013) (R) doi:10.1103/PhysRevB.87.121408 219. D. A. Abanin, and D. A. Pesin, Interaction-Induced Topological Insulator States in Strained Graphene, Phys. Rev. Lett. 109(6), 066802 (2012) doi:10.1103/PhysRevLett.109.066802 220. B. Roy, Odd integer quantum Hall effect in graphene, Phys. Rev. B 84(3), 035458 (2011) doi:10.1103/PhysRevB.84.035458 221. D. B. Zhang, G. Seifert, and K. Chang, Strain-Induced Pseudomagnetic Fields in Twisted Graphene Nanoribbons, Phys. Rev. Lett. 112(9), 096805 (2014) doi:10.1103/PhysRevLett.112.096805 222. L. Tapasztó, T. Dumitrică, S. J. Kim, P. Nemes-Incze, C. Hwang, and L. P. Biró, Breakdown of continuum mechanics for nanometre- wavelength rippling of graphene, Nat. Phys. 8(10), 739 (2012) doi:10.1038/nphys2389 223. F. Guinea, M. Katsnelson, and M. Vozmediano, Midgap states and charge inhomogeneities in corrugated graphene, Phys. Rev. B 77(7), 075422 (2008) doi:10.1103/PhysRevB.77.075422 224. T. O. Wehling, A. V. Balatsky, A. M. Tsvelik, M. I. Katsnelson, and A. I. Lichtenstein, Midgap states in corrugated graphene: Ab initio calculations and effective field theory, EPL (Euro physics Letters) 84(1), 17003 (2008) doi:10.1209/0295-5075/84/17003 225. F. Guinea, M. I. Katsnelson, and A. K. Geim, Energy gaps and a zero-field quantum Hall effect in graphene by strain engineering, Nat. Phys. 6(1), 30 (2010) doi:10.1038/nphys1420 226. Y. Jiang, T. Low, K. Chang, M. I. Katsnelson, and F. Guinea, Generation of pure bulk valley current in graphene, Phys. Rev. Lett. 110(4), 046601 (2013) 227. Z. Wu, F. Zhai, F. M. Peeters, H. Q. Xu, and K. Chang, Valley-dependent Brewster angles and Goos-Hanchen effect in strained graphene, Phys. Rev. Lett. 106(17), 176802 (2011) 

1101.5636

2

1101

Observation of a warped helical spin-texture in Bi$_2$Se$_3$ from circular dichroism angle-resolved photoemission spectroscopy

A differential coupling of topological surface states to left- versus right-circularly polarized light is the basis of many opto-spintronics applications of topological insulators. Here we report direct evidence of circular dichroism from the surface states of Bi$_2$Se$_3$ using a laser-based time-of-flight angle-resolved photoemission spectroscopy. By employing a novel sample rotational analysis, we resolve unusual modulations in the circular dichroism photoemission pattern as a function of both energy and momentum, which perfectly mimic the predicted but hitherto un-observed three-dimensional warped spin-texture of the surface states. By developing a microscopic theory of photoemission from topological surface states, we show that this correlation is a natural consequence of spin-orbit coupling. These results suggest that our technique may be a powerful probe of the spin-texture of spin-orbit coupled materials in general.

cond-mat.mes-hall

cond-mat

Observation of a warped helical spin-texture in Bi2Se3 from circular dichroism angle-resolved photoemission spectroscopy Y. H. Wang1,2, D. Hsieh1, D. Pilon1, L. Fu2, D. R. Gardner1, Y. S. Lee1, and N. Gedik1 1Department of Physics, Massachusetts Institute of Technology, Cambridge MA 02139, USA and 2Department of Physics, Harvard University, Cambridge MA 02138, USA A differential coupling of topological surface states to left- versus right-circularly polarized light is the basis of many opto-spintronics applications of topological insulators. Here we report direct evi- dence of circular dichroism from the surface states of Bi2Se3 using a laser-based time-of-flight angle- resolved photoemission spectroscopy. By employing a novel sample rotational analysis, we resolve unusual modulations in the circular dichroism photoemission pattern as a function of both energy and momentum, which perfectly mimic the predicted but hitherto un-observed three-dimensional warped spin-texture of the surface states. By developing a microscopic theory of photoemission from topological surface states, we show that this correlation is a natural consequence of spin-orbit coupling. These results suggest that our technique may be a powerful probe of the spin-texture of spin-orbit coupled materials in general. Three-dimensional topological insulators [1 -- 3] are an intensely researched phase of matter owing to their unique spin-helical metallic surfaces, where the spin di- rection is locked to the wavevector and winds by 2π around the Fermi surface [4 -- 6]. Spin-dependent absorp- tion of circularly polarized light has recently been pre- dicted to generate highly spin-polarized surface electrical currents whose direction can be switched by the light he- licity [7 -- 9]. However, a demonstration of spin-dependent differential absorption of left- versus right-circularly po- larized light [circular dichroism (CD)] from the surface states (SS) has evaded conventional probes such as trans- port and optics because of the combined need for energy- momentum resolution and surface sensitivity. In this Letter, we map the variation of the CD in the photoemission intensity [10 -- 15] over the full surface band structure of the topological insulator Bi2Se3 [16, 17] us- ing a laser-based time-of-flight angle-resolved photoemis- sion spectroscopy (TOF-ARPES) technique. Our results show a strong CD from the SS in agreement with a recent work on CuxBi2Se3 [18]. However, owing to our ultra- high polarization purity and our novel sample rotational analysis, we uniquely resolve fine modulations in the CD pattern as a function of energy and momentum that per- fectly follow the predicted [19, 20], but so far un-detected [6], three-dimensional spin-texture of Bi2Se3. By devel- oping a microscopic theory of photoemission from topo- logical SS, we show that such modulations naturally arise from the coupling of the circularly polarized light to the surface spin-texture via spin-orbit interactions. Excellent agreement between our results and theory suggests that our technique is an efficient and sensitive measure of all three components of the surface spin-texture. The experimental setup is shown in Fig. 1(a). Single crystals of Bi2Se3 are cleaved along their (111) surface at room temperature under ultra-high-vacuum (< 6 × 10−11 Torr). Femtosecond laser pulses from a Ti:Sapph ampli- fier are frequency quadrupled to 6.2eV and then circu- larly polarized to > 99% purity with a quarter wave- plate. Strain induced birefringence from the vacuum windows and dichroism from all optical elements follow- ing the quarter wave-plate, which can ruin the purity of circular polarization, were carefully compensated in our experiment [21]. Conventional hemispherical ARPES an- alyzers rely on the deflection of electron paths along one spatial dimension to measure electronic dispersion along one direction in momentum space. In contrast, our TOF- ARPES analyzer is developed for pulsed photon sources and resolves electron energies through their flight time from sample to detector, which enables measuring dis- persion spectra I(E, kx, ky) simultaneously over two di- mensional momentum space. Complete spectra of Bi2Se3 (surface Dirac cone plus bulk conduction band) obtained with TOF-ARPES is shown in Fig.1(b). This ability to measure the entire bandstructure simultaneously elimi- nates experimental geometry induced matrix element ef- fects [22] that may obscure the intrinsic CD pattern and allows for direct comparison of CD-ARPES intensities across different phase space points. To investigate whether CD is exhibited by Bi2Se3 and if it has bulk and/or surface state origin, we measure spectra like in Fig.1(b) using left- and right-circularly polarized light and then take their difference to obtain the CD spectrum ∆I(E, kx, ky). As shown in Figs. 1(c)- (d), there is clear CD from the surface Dirac cone of Bi2Se3 and not from the bulk conduction band. The CD pattern exhibits perfectly odd reflection symmetry about the photon scattering plane when it is parallel to the ¯Γ ¯M direction [Fig. 1(e)], which is expected from crystal mirror symmetry. However this odd reflection symmetry is lifted when the scattering plane is along a non mirror-symmetric direction such as ¯Γ ¯K [Fig. 1(f)], which is evidence that the CD pattern is intrinsic to the sample. Having shown that CD is intrinsic to the sample and present for the SS, we first focus on its fine features in the low energy region [±0.1eV relative to the Dirac point (DP)] where the Fermi surface is circular. Fig. 1(d) (b) I E( ,k ,k ) y x Bi Se2 3 (a) 80fs l/4 waveplate quadrupler Ti:Sapph laser flight-time Electron paths 2D-PSD ΔI (a.u.) f=0 k y M +1 (f) 0 -1 30 q k x 0.1 eV 0 3eV. (h) 60 90 hv R L E ky kx (e) 0.1 hn ) 1 - Å ( k 0.0 -0.1 ) 1 - Å ( k 0.1 0.0 -0.1 (g) (c) ) y k , x k , E ( I D (d) ) V e ( E 0.3 0.2 0.1 0.0 -0.1 -0.2 -0.1 0.0 0.1 k (Å )-1 -0.1 0.0 k (Å )-1 0.1 -0.1 0.0 k (Å )-1 0.1 FIG. 1. (a) Schematic of a time-of-flight based angle-resolved photoemission spectrometer. PSD: position-sensitive detec- tor. (b) A typical iso-intensity surface in (E, kx, ky) space from Bi2Se3 collected simultaneously using linearly polarized photons. (c) Difference of TOF-ARPES data measured using right- and left-circularly polarized light. hν denotes the inci- dent photon direction. (d) E − k cut through the CD data volume in (c) along the angle denoted by the green dash in (e). (e-h) Constant energy slices through (c) at the Fermi level and at 0.1eV above the DP (insets) for sample rotation angles of φ = 0◦, 30◦, 60◦ and 90◦ respectively. Green hexagons repre- sent the Brillouin zone of Bi2Se3(111), red lines are the mirror planes and the arrow denotes the photon incident direction. shows that at a constant energy the sign of CD is re- versed for states of opposite momentum and is reversed again across the DP. Furthermore, the CD pattern is in- variant under rotation (φ) of the crystal mirror plane relative to the scattering plane [Figs. 1(e)-(h) insets]. These behaviors match the known in-plane spin texture in the low energy region, which are tangential to the ro- tationally isotropic constant energy contours [4, 5] and reverse chirality across the DP. The CD patterns change drastically away from the low energy region (E > 0.1eV), where the Fermi sur- face evolves from being circular to hexagonal in shape [19, 23]. Here we observe new modulations in the magni- tude and sign of CD as a function of the angle θ around the constant energy contours [Figs.1(e)-(h)]. In partic- ular, extrema along the ky axis (at θ = 90◦ and 270◦) at low energy [Figs. 1(e)-(h) insets] become nodes in the 2 high energy φ = 0◦ spectrum [Fig. 1(e)]. Furthermore, in stark contrast to the low energy region, CD patterns at high energy undergo dramatic changes as φ is varied. For instance, the nodes along ky in the Fig. 1(e) spec- trum become extrema again in the φ = 60◦ spectrum [Fig. 1(g)] and only repeats itself under φ = 120◦ rota- tions, consistent with the symmetry of the underlying lat- tice. This is reminiscent of the predicted θ dependence of the out-of-plane component of spin hSzi at high energies [19, 20], which goes from minimum to maximum under φ = 60◦ rotation. Such behavior has been observed us- ing Mott polarimetry only in Bi2Te3 [5] but not in Bi2Se3 [6] owing to the much larger value of hSzi in the former. Given that the overall CD patterns observed closely fol- low the predicted spin-texture of the SS, we investigate their possible connection by developing a standard model of photoemission from topological SS. The microscopic Hamiltonian for a system with spin- orbit coupling is given by: H = ~P 2 2m + V (~r) +  4m2c2 ( ~P × ~∇V ) · ~s (1) where ~P is momentum operator, V (~r) is the crystal po- tential, and ~s is the electron spin. Coupling to an elec- tromagnetic field is obtained via ~P → ~P − e ~A, where ~A is the photon vector potential, such that to first order in ~A: H( ~A) = H − ~P · ~A (2) 4m2c2 (~∇V ×~s). The photoemission ma- where ~P ≡ e m trix element between the initial and final states is given by ~P − e M (~k, f ) = hf~k ~P · ~A~ki (3) +i + v~kφi where ~A ≡ R dt ~A(t)eiωt is the Fourier transform of ~A and f~ki is the bulk final state. The initial state ~ki = u~kφi −i is a linear combination of two- fold degenerate pseudospin states φi ±i at the DP that are eigenstates of total angular momentum (orbital plus spin), which is widely used in standard k · p descrip- tions of topological SS [2, 17, 19, 24]. Because of strong spin-orbit coupling, only pseudospin is a good quan- tum number [19]. The coefficients u~k and v~k deter- mine the expectation value of three pseudospin compo- nents: hSxi~k = (u∗ v~k + v∗ v~k), ~k ~k hSzi~k = (u~k2 − v~k2). Importantly, because spin is directly proportional to pseudospin [19], we refer to the two interchangeably. For circularly polarized light incident onto the surface with wavevector in the xz plane, ~A(t) = (Ax sin ωt, Ay cos ωt, Az sin ωt) and ~A = (−iAx, Ay, −iAz). Straight-forward application of time- reversal and crystal symmetries [21] yield the following expression for the photoemission transition rate: I(~k) = a2 (cid:0)Ax2 + Ay2(cid:1) + b2Az2 + a2Im (cid:0)AxA∗ u~k), hSyi~k = (v∗ ~k u~k − iu∗ ~k + 2abIm[A∗ xAzhSyi~k − A∗ yAzhSxi~k] y(cid:1) hSzi~k (4) (a) Δ I ( f < >Sx =0 + ( Δ ) I (b) Δ I ( f f =60 ) < >Sy Δ I =30 + ) ) V e ( E ) V e ( E 0.3 0.2 0.1 0.0 -0.1 -0.2 0.4 0.3 0.2 0.1 0.0 -0.1 -0.2 -180 -90 0 -180 -90 0 θ(°) (c) Δ I ( f ( f =90 ) < >Sz =0 ( Δ ) - I f =60 ) (a.u.) +1 0 -1 90 180 90 -90 0 θ(°) 90 -180 -90 90 180 0 θ(°) (d) 1 . 0 0 . 0 ) 1 - Å ( k 1 . 0 - ) 1 - Å ( k 0.1 0.0 -0.1 E= .0 3eV (e) (f) G Μ -0.1 Κ 0.0 k (Å )-1 0.1 -0.1 0.0 k (Å )-1 0.1 -0.1 0.0 k (Å )-1 0.1 -0.1 0.1 0.0 k (Å )-1 FIG. 2. (a) x, (b) y and (c) z components of the spin-texture over the complete surface states obtained by summing or sub- tracting ∆I data volumes at different φ (see text). Color maps were created by integrating the data radially in k-space over a ±0.015A−1 window about the surface state contours at each energy. Constant energy cuts at the Fermi level for each spin component are shown directly below in (d) through (f). where a and b are bandstructure dependent complex con- stants and Im refers to the imaginary part. Circular dichroism is obtained by taking the difference of Eq.(4) with opposite photon helicity (Ay → −Ay): ∆I = a2Im(AxA∗ y)hSzi~k − 4abIm(AzA∗ y)hSxi~k (5) 3 0.1 eV -0.1 eV (d) Helical electrons (E>0) 0.05 ) ) 1 1 1 1 - - - - Å Å 0.0 ( ( k k -0.05 (e) 0.05 0.1 eV -0.1 eV 0.05 eV -0.05 eV spin G M Helical holes (E<0) 1 - ) Å ( k 0.0 -0.05 (a) 1 > S < / > x S < 0 -1 (b) 1 0 > S < / > y S < -1 (c) ) . u . a ( > z S < 1 0 -1 -180 -90 0 θ(°) 90 180 -0.05 0.0 k (Å )-1-1 0.05 FIG. 3. The evolution of the (a) Sx, (b) Sy and (c) Sz compo- nents around constant energy contours in the vicinity of the Dirac point, obtained by taking cuts through the data in Figs. 2(b)-(d). Curves are fits to sin(θ) and cos(θ) varying functions for Sx and Sy respectively. (d) and (e) show the spin vectors (each normalized by its magnitude S) of all positive and neg- ative energy states within ±0.1eV respectively projected onto the kx − ky plane. Although the CD is a linear combination of hSxi and hSzi, these can be disentangled by applying symmetry properties of Bi2Se3(111) as follows. Time-reversal sym- metry and three-fold rotational symmetry together dic- tate that hSzi flips sign upon a ∆φ = 60◦ rotation while hSxi stays unchanged. Taking the difference (sum) of CD patterns ∆φ = 60◦ apart isolates hSzi (hSxi). The hSyi component is trivially obtained by performing the procedure for hSxi under a 90◦ sample rotation. Fig. 2 shows all three spin components measured using this method. Because calculating a and b is beyond the scope of our work, we only plot the relative and not absolute magnitude of the spins. We immediately notice that all three components reverse sign across the DP as expected and that hSxi and hSyi are modulated with a similar pe- riodicity. On the other hand hSzi exhibits a different periodicity at high energies, which vanishes in the low energy region. To quantitatively test the validity of our microscopic theory, we first compare our extracted spin components with the ideal spin texture in the low energy region that is experimentally well-known . Constant low energy slices through the spin maps [Figs. 2(a)-(c)] show that hSzi ex- hibits negligible modulation whereas hSxi and hSyi fol- low a clear sin(θ) dependence that reverses sign across the DP [Figs. 3(a)-(c)]. By projecting the planar spin vectors of each state in the low energy region of the Dirac cone onto the x−y plane, we obtain a vector field that ex- actly matches an ideal helical spin-texture with opposite chirality for electrons and holes [Figs. 3(d) and (e)]. We note that while the relative orientation of spins is directly measured in our experiment, the absolute sense of chiral- ity was set by matching our data to Mott polarimetry results at a single energy-momentum point [6], although it can also be independently obtained by calculating a and b. The excellent agreement between our measured low energy CD maps and theoretical [19] and Mott po- larimetry results [4 -- 6] strongly suggests that CD-ARPES is a sensitive measure of the topological SS spin-texture. Next we compare our CD spectra to our model in the high energy region where the spin-texture is predicted to depart from ideal planar helical form but is so far un- observed in Bi2Se3 using conventional Mott polarimetry based ARPES [6]. At high energies where the Fermi sur- face acquires a hexagonal shape, k · p theory predicts that the spin-texture should develop a finite out-of-plane component hSzi that is modulated with a sin(3θ) period- icity, with maximally upward and downward spin cant- ing along ¯Γ ¯K directions [19]. Fig. 4(a) shows a con- stant energy slice at 0.3eV through the spin map of Fig. 2(c), where there is a clear sin(3θ) dependence of hSzi that is maximal along ¯Γ ¯K and zero along ¯Γ ¯M , in per- (a) (b) d=0 d=0 ) . u . a ( > z S < ) . u . a ( > x S < 1 0 -1 1 0 -1 10 Dirac regime (c) 5 0 ) . u . a ( > 0 z S < -5 FS spin d k 60 40 20 0 d ( ) -20 -40 -180 -90 0 θ(°) 90 180 0.4 0.2 (d) K M K M -0.2 E( 0.0 )eV z G Deformed Dirac spin-texture FIG. 4. Evolution of (a) Sz and (b) Sx around the Fermi surface. Red line in (a) is a fit to k · p theory. The blue and pink lines in (b) are fits to phenomenological models [21] that respectively do and do not account for an in-plane angular deviation (δ) away from ~S ⊥ ~k locking behavior (inset). (c) The energy dependence of the amplitude of Sz modulation (green symbols), its fit (magenta line) and δ (red symbols). Inset shows the spin-integrated Fermi surface together with a fit to k ·p theory. (d) Spin-orientation around the hexagonally deformed Fermi surface constructed from CD-ARPES data. 3D view shows spins plotted around Fermi contour and top- down view shows spins projected onto a segment of the unit circle in momentum space for ease of visualizing δ. fect accord with both theory [19] and experiment [5] on Bi2Te3. To further test whether our CD-ARPES data is indeed sensitive to spin, we compare our measured en- ergy dependence of out-of-plane spin component along ¯Γ ¯K (hSzi0(E)) with its unique functional form predicted by k · p theory [19]: hSzi0(E) = 1/p1 + [k(E)β]−4 (6) where β is the only free parameter. Eq.(6) provides an excellent fit for our data [Fig. 4(c)] and the fitted value of β, which parameterizes the degree of hexagonal distor- tion of the Fermi surface, is quantitatively consistent with the parameters extracted from fitting the Fermi surface contour in Fig. 4(c) inset [21]. The accurate extraction of all three spin components over the entire Dirac cone from our CD-ARPES data [Fig. 4(d)] is a testament to the validity and sensitivity of our method. While our data produce all features predicted by k · p theory [19], the in-plane spin component at high en- ergies also exhibits unpredicted kinks in its θ depen- dence [Fig. 4(b)]. One possible origin of these kinks is that the in-plane spin magnitude is modulated purely by the out-of-plane spin canting. However a fit to the data using this assumption [Fig. 4(b) pink line] yields kinks that are too weak and fails to reproduce the nodes at θ = 0◦ even when allowing for unrealistically large (∼ 45◦) canting angles [21]. An alternative explanation 4 is that the in-plane spin direction deviates from perpen- dicular spin-momentum locking [inset Fig. 4(b)] period- ically in θ. A fit assuming periodic angular deviations of amplitude δ [Fig. 4(b) blue line] better reproduces the data. The in-plane component constructed from the data alone [Fig. 4(d)] is consistent with the latter in- terpretation and shows that the deviation is zero along ¯Γ ¯M and ¯Γ ¯K and is maximum in between the two. A plot of the amplitude of δ versus energy [Fig. 4(c)] shows that it only develops beyond ∼ 0.3eV from the DP. The spike exactly at the DP is consistent with spin degener- acy at the DP. This in-plane spin canting behavior has been shown by first-principles calculations as well as k · p expansions up to fifth order on Bi2Te3 [24]. However, although such effects have been observed in Rashba spin split metals [25], they have never been observed in any topological insulators. In conclusion, we have directly observed circular dichroism from the surface states of Bi2Se3. By com- bining a TOF-ARPES technique with ultra-pure circular photon polarization and a novel sample rotation analysis, we resolve unusual modulations in the circular dichro- ism photoemission pattern as a function of both energy and momentum for the first time, which closely follow the predicted three-dimensional spin-texture. A direct connection between CD-ARPES and spin-orbit induced spin-textures is established through our microscopic the- ory of photoemission. Our results open the possibility to generate highly-polarized spin currents with circularly polarized light, which may be detected through transport or optical means [26]. The efficiency and high spin sen- sitivity of our technique suggest that CD-ARPES may be used as a vectorial spin mapping tool to detect small deviations from a π Berry's phase in magnetically doped topological surface states, or to study spin-orbit coupled materials in general. Towards completion of this manuscript, we became aware of three related concurrent and independent works [18, 27]. None, however, resolve the fine modulations in the CD-ARPES patterns that are integral to understand- ing the connection to spin, which is the main focus of this work. This research is supported by Department of Energy award number DE-FG02-08ER46521, Army Research Of- fice (ARO-DURIP) award number W911NF-09-1-0170 (ARTOF spectrometer) and in part by the MRSEC Pro- gram of the National Science Foundation under award number DMR - 0819762 (partial support for YHW). Cor- respondence and requests for materials should be ad- dressed to N.G. ([email protected]). [1] J. E. Moore, Nature 464, 194 (2010) [2] M. Z. Hasan, C. L. Kane, Rev. Mod. Phys. 82, 3045 5 (2010) [3] X.-L. Qi, S.-C. Zhang, Physics Today 63, 33 (2010) [4] D. Hsieh, et al., Science 323, 919 (2009); D. Hsieh, et al,. Nature 460, 1101 (2009) [5] S. Souma, et al., Phys. Rev. Lett. 106, 216803 (2011); S.-Y. Xu, et al., arXiv:1101.3985 (2011) [6] Z.-H. Pan, et al., Phys. Rev. Lett. 106, 257004 (2011). [7] S. Raghu, S. B. Chung, X.-L. Qi, S.-C. Zhang, Phys. Rev. Lett. 104, 116401 (2010). [15] D. V. Vyalikh, et al., Phys. Rev. Lett. 100, 056402 (2008) [16] Y. Xia, et al., Nature Phys. 5, 398 (2009). [17] H. Zhang, et al., Nature Phys. 5, 438 (2009). [18] Y. Ishida, et al., Phys. Rev. Lett. 107, 077601 (2011). [19] L. Fu, Phys. Rev. Lett. 103, 266801 (2009). [20] W. Zhang, R. Yu, H.-J. Zhang, D. Xi, Z. Fang, New J. Phys. 12, 065013 (2010). [21] See Supplemental Material. [22] S. Hufner, Photoelectron Spectroscopy (Springer, [8] H.-Z. Lu, W.-Y. Shan, W. Yao, Q. Niu, S.-Q. Shen, Phys. Berlin,2003). Rev. B 81, 115407 (2010). [23] Y.L. Chen, et al., Science (2009); K. Kuroda, et al., Phys. [9] P. Hosur, Phys. Rev. B 83, 035309 (2011). Rev. Lett. 105, 076802 (2010). [10] C. Westphal, J. Bansmann, M. Getzlaff, G. Schonhense, Phys. Rev. Lett. 63, 151 (1989); [11] C. M. Schneider, J. Kirschner, Critical Reviews in Solid State and Materials Sciences. 20, 179 (1995); [24] S. Basak, et al., arxiv: 1103.4675 (2011) [25] F. Meier, et al., Phys. Rev. B 77, 165431 (2008) [26] D. Hsieh, et al., Phys. Rev. Lett. 107, 077401 (2011) [27] Park et al., arXiv:1107.3285 (2011); Scholz et al., [12] S. V. Halilov, et al., J. Phys: Condens. Mat. 5, 3851 arXiv:1108.1053 (2011). (1993); [13] F. Frentzen, et al., Z. Phys. B 100, 575 (1996); [14] V. B. Zabolotnyy, et al., Phys. Rev. B 76, 024502 (2007); 

1503.04077

1

1503

Test of the Atiyah-Singer Index Theorem for Fullerene with a Superconducting Microwave Resonator

Experiments have been performed using a spherical superconducting microwave resonator that simulates the geometric structure of the C60 fullerene molecule. The objective was to study with very high resolution the exceptional spectral properties emerging from the symmetries of the icosahedral structure of the carbon lattice. In particular, the number of zero modes has been determined to test the predictions of the Atiyah-Singer index theorem, which relates it to the topology of the curved carbon lattice. This is, to the best of our knowledge, the first experimental verification of the index theorem.

cond-mat.mes-hall

cond-mat

a Test of the Atiyah-Singer Index Theorem for Fullerene with a Superconducting Microwave Resonator B. Dietz,1, ∗ T. Klaus,1 M. Miski-Oglu,1 A. Richter,1 M. Bischoff,2 L. von Smekal,2, 3 and J. Wambach2 1Institut fur Kernphysik, Technische Universitat Darmstadt, D-64289 Darmstadt, Germany 2Theoriezentrum, Institut fur Kernphysik, Technische Universitat Darmstadt, D-64289 Darmstadt, Germany 3Institut fur Theoretische Physik, Justus-Liebig-Universitat Giessen, D-35392 Giessen, Germany (Dated: October 29, 2018) Experiments have been performed using a spherical superconducting microwave resonator that simulates the geometric structure of the C60 fullerene molecule. The objective was to study with very high resolution the exceptional spectral properties emerging from the symmetries of the icosahedral structure of the carbon lattice. In particular, the number of zero modes has been determined to test the predictions of the Atiyah-Singer index theorem, which relates it to the topology of the curved carbon lattice. This is, to the best of our knowledge, the first experimental verification of the index theorem. PACS numbers: 05.45.Mt,41.20.Jb,71.20.-b,71.20.Tx,73.22.-f Introduction. -- The spectrum of graphene, a mono- layer of carbon (C) atoms arranged on a hexagonal lat- tice, has been the focus of extensive theoretical [1, 2] and experimental studies [3]. Its universal properties were often also investigated experimentally in analog systems, so-called 'artificial graphene' [4], e.g., in our group in pho- tonic crystals [5 -- 10]. Moreover, theoretically much atten- tion has been devoted to curved graphene structures like fullerene molecules [11 -- 16] and the connection between their spatial symmetries and electronic properties. Here, the most famous example is the C60 molecule. It con- sists of 60 carbon atoms at the vertices of a truncated icosahedron and has the shape of a soccer ball. Con- cerning the spectral properties of fullerenes the number of near-zero modes, i.e., of electronic states with excita- tion energies close to zero, have been of particular in- terest since they determine the electrical conductivity. In [17, 18] an index theorem has been derived that allows the computation of the number of such near-zero modes from the topology of the surface. It was deduced from the renowned Atiyah-Singer index theorem [19 -- 22] which states that the analytic index of an elliptic differential op- erator on a compact manifold equals the topological one, in other words, that there is a connection between the number of zero modes of the operator and the topology of the manifold on which it is defined. The aim of the high-resolution experiments presented in this letter was to test these predictions in experiments with a supercon- ducting microwave resonator of the same topology as the C60 molecule. First we briefly review the salient features of graphene and fullerenes and outline the derivation of the index theorem from [17, 18] for deformed graphene sheets. We then describe the experimental setup and compare the results of the measurements to the predictions from the index theorem and to tight-binding model (TBM) calcu- lations. These allow us to study the approach to the ther- modynamic limit of an infinite number of carbon atoms. Graphene, fullerenes and the Atiyah-Singer theorem. -- The honeycomb structure of graphene is formed by two interpenetrating triangular sublattices. As a con- sequence, at half filling the Fermi surface in graphene reduces to two independent points in the first Brillouin zone, the so-called 'Dirac points', denoted by K+ and K− [1, 2] that are conical intersections of the valence and the conduction band. Low energy excitations within the cone regions around K± have a linear dispersion with a slope given by the Fermi velocity vF . On an infinte graphene sheet they are therefore described by a Dirac Hamiltonian for massless spin-1/2 quasiparticles consist- ing of partner Hamiltonians H± = ±vF σαqα (1) which describe excitations with momentum q = (qx, qy) in each of the two Dirac cones around K±. The Pauli matrices σα with α = x, y act on the two sublattice com- ponents of the excitations, combined in two-dimensional spinors and hence referred to as quasi-spin. Both cones together then yield a four-component Dirac equation [1]. Fullerene molecules can be constructed by introduc- ing positive curvature into an initially flat graphene sheet [23]. The bending is realized by replacing hexagons by pentagons, ensuring at the same time that the lattice is not stretched and each C atom keeps three neighbors. To determine the number of pentagons n5 necessary to generate a spherical fullerene molecule with n6 hexagons one uses the Euler formula [24] which relates the number of vertices V , of edges E, faces F and open ends Nopen of an arbitrary two-dimensional lattice to the genus g of the surface formed by it, via the Euler characteristic χ = V − E + F = 2(1 − g) − Nopen . For a lattice of pentagons and hexagons, V = (5n5 + 6n6)/3, E = (5n5 + 6n6)/2 and F = n5 + n6, this gives χ = n5/6. Without open ends χ must be an even integer on a closed orientable surface due to the Gauss-Bonnet theorem [24]. Hence, for a flat graphene sheet with peri- odic boundary conditions one has g = 1 for the torus and n5 = 0, while a sphere with g = 0 needs n5 = 12 pen- tagons to avoid open ends. Consequently, fullerenes are grown from the C60 molecule by increasing the number of hexagons, i.e., always have twelve pentagons at the same relative positions. This also applies to the ther- modynamic limit, and one expects that their low-energy electronic excitations are described by a Dirac equation on a sphere. To introduce a pentagon into the honycomb lattice, a π/3 sector is cut out and then the edges are glued to- gether [13, 14, 17, 18] as illustrated in Fig. 1. Thereby (Color online) Left panel: A two-dimensional FIG. 1: graphene sheet. The red and the blue dots mark the two independent triangular sublattices. In order to form a curved sheet which contains one pentagon, a π/3 segment is cut out from the sheet. Right panel: The conically deformed graphene sheet with the pentagon at the apex. a pentagon is created at the apex of the emerging cone. Along the seam, two C atoms from the same triangular sublattice, e.g., the red ones in Fig. 1, are connected. This results in a coupling of the Dirac operators as- sociated with the K± points. Indeed, when the four- dimensional spinor associated with the Dirac equation of the flat graphene sheet is transported around the apex by an angle 2π, it is forced to jump at the seam from a red site to another red one instead of to a blue one. It thus acquires a non-trivial phase, which can be ac- counted for by introducing a non-Abelian gauge field Aµ in the Hamiltonian which yields a flux of (π/2) τ y when integrated along a closed loop around the apex. The Pauli matrix τ y thereby couples the K+ and K− spinor components [1]. This description entails the existence of a ficitious magnetic monopole inside the surface. In the case of fullerenes it is located at the center of the spherical molecule, yielding a flux of 1/8 through each of the twelve pentagons. Thus the total magnetic monopole charge inside the sphere equals 3/2. In addition to that, analogous to the daily rotation of Foucault's pendulum, a deficit angle of π/3 arises when moving a frame along a loop around the apex. It is here described by a quasi- 2 spin connection Qµ with circulation −(π/6) σz around the apex. The coupling of the K± spinor components in the re- sulting four-dimensional Dirac equation can be removed by a rotation, which leads to two independent two- dimensional Dirac equations denoted by l = 1, 2 [18], l /D ψl = vF σαeµ α(cid:0)qµ − iQµ − iAl µ(cid:1) ψl = Eψl , (2) where eµ α is the Zweibein in the tangent plane of the surface, and Al µ are the components of Aµ in the rotated basis with circulation ±π/2 for l = 1 and 2, respectively, where ~Al is now an Abelian gauge field. The four-dimensional Dirac equation obtained from Eq. (2) provides a good description of the low-energy excitations of the C molecules [13 -- 15]. It yields the long- wavelength excitations of the deformed graphene sheet in the vicinity of the Dirac points, and thus also the zero modes that we are interested in. For the fullerenes the Dirac operators /D are elliptic and defined on a compact surface. Hence the Atiyah-Singer index theorem [19 -- 22] applies. Ten years after its first formulation a new proof was provided based on the heat equation [25] which was later employed for the derivation of an index theorem for graphene sheets deformed by pentagons and heptagons [17, 18] as briefly reviewed in the following. l l Each Dirac operator in Eq. (2) can be written in terms of off-diagonal partner operators P and P † [26], so ( /D )2 contains only diagonal operators P P † and P †P that have the same number of zero modes as P † and P , respectively. Furthermore, the non-zero eigenvalues of P P † and P †P are identical. The analytic index of /D is given by the difference of the numbers of zero modes of P and P † denoted by ν±, respectively, i.e., index( /D ) = ν+ − ν− [17, 18]. More importantly, however, this index is related to the total flux of the effective gauge field via the Atiyah- Singer index theorem [17], l l index( /D l ) = 1 2π ZZΩ F ldΩ . (3) The integral is taken over the compact surface Ω and F l = ∂ ∧ Al are the field strengths associated with the now Abelian gauge potentials Al. Stokes' theorem then implies from the closed loops around each apex that 1 2π ZZΩ F ldΩ = 1 2π n5I Al µ dsµ ≡ ± 3 2 χ . (4) The Euler formula thus leads to the Atiyah-Singer index theorem for fullerenes in the form, index( /D ) = ±3(1−g). In two dimensions either ν+ or ν− vanish. Hence, the index theorem provides the number of zero modes [17, 18]. l The total number of zero modes is the sum of those of the subsystems corresponding to l = 1, 2. Consequently, according to the index theorem, the zero modes of the four-dimensional Dirac operator for spherical fullerenes correspond to two triplets. The same result has been obtained in a continuum model for the low-energy elec- tronic states of icosahedral fullerenes [13 -- 16]. We em- phasize, however, that the eigenvalues of the near-zero modes tend to zero, i.e., coincide with the energy at the Dirac points, only in the thermodynamic limit of an in- finity number of C atoms. In a sufficiently large but finite fullerene molecule they are expected to lie much closer to the Dirac energy than all the other ones. Experimental setup and resonance spectra. -- Hitherto, experiments have been performed with flat, supercon- ducting microwave resonators, so-called 'microwave bil- liards' [27] to address problems from the fields of quan- tum chaos [28, 29] and compound nucleus reactions [30]. In this context, the equivalence of the Helmholtz equa- tion and the non-relativistic Schrodinger equation of the corresponding quantum billiard is exploited which holds below a maximum microwave frequency fmax = c/(2d) with c the velocity of light and d the height of the bil- liard. Consequently, the eigenvalues of a quantum bil- liard can be obtained experimentally from the eigenfre- qencies of the microwave billiard of corresponding shape. Recently, we realized experiments with superconducting microwave Dirac billiards and studied universal spectral properties of graphene sheets [31] with unprecedented ac- curacy [10, 32, 33]. The aim of the experiments presented here was the investigation of the universal spectral properties of the fullerene C60 molecule attributed to its lattice structure and to determine the number of zero modes which, ac- cording to the Atiyah Singer index theorem solely de- pends on the number of pentagons. For this we use a sys- tem exhibiting the same topological properties, namely a quantum fullerene billiard on a sphere, consisting of a network of 60 circular billiards at the positions of the C atoms connected by three of the altogether 90 straight leads with three adjacent ones. We studied them exper- imentally by using instead of a planar, superconducting microwave (Dirac) billiard a cavity, which is imprinted on a sphere. The microwave fullerene billiard displayed in Fig. 2 was constructed by milling a total of 60 circu- lar cavities (vertices) and 90 rectangular channels (edges) out of a brass sphere and then closing them with small triangular brass plates of 5 mm thickness and 3 mm thick rectangular ones, respectively. Before the parts were screwed together, they were covered with lead, which is superconducting below Tc = 7.2 K. The diameter of the sphere of 160 mm was limited by the size of the liquid Helium cryostat in which the resonator was cooled down to 4.2 K in order to attain superconductivity. The ra- dius of the circular cavities was 12 mm, the widths of the waveguides 14 mm, before lead coating them. Thus the cutoff frequency for the first propagating mode in the latter is f 1 c & 10.714 GHz. In total, 8 antennas were at- tached to the triangular plates. Two, covered with red 3 FIG. 2: (Color online) Lead plated fullerene billiard used in the experiments. In the left part the small plates that cover the circular cavities and the rectangular channels were re- moved. The red caps protect the antenna ports. The billiard is superconducting below Tc=7.2 K. caps, are visible in Fig. 1. The height of the resonator was 3 mm corresponding to fmax = 50 GHz. For the measurement of the transmission spectrum shown in the upper panel of Fig. 3 microwave power was coupled into the microwave billiard via antenna a and the output signal was received at antenna b, with a and b denoting two of the 8 antennas. A vectorial network analyzer determined the relative phase and amplitude of the output and input signals, thus yielding the scattering matrix element Sba. The smallest resonance frequency equals f = 8.254 GHz, so we show the spectrum from 8 - 40 GHz. Due to the high-quality factor Q > 105 of the resonator, all resonances could be resolved in that frequency range. We concentrate our discussion here on the region between 8.254 and 18.801 GHz. The spec- trum exhibits three distinct bands, containing 60, 210 and 90 resonances, respectively, in the frequency intervals [8.254, 8.779] GHz, [11.492, 16.657] GHz and [18.312, 18.801] GHz. They are located around the eigenfrequen- cies f1 ≃ 8.4 GHz, f2 ≃ 13.5 GHz and f3 ≃ 18.5 GHz of the first three quasibound states in an open circular billiard of the same size as the cavities in the resonator with openings at the positions of the waveguides. The modes excited inside the circular cavities resemble within a given band the corresponding mode in the open circular billiard, and are described by J0, J1 and J2 Bessel func- tions in the first, second and third band, respectively. In the latter two cases they are twofold degenerate due to the mirror symmetry. Note, that the circular cavities exhibit no threefold symmetry, because each of them is part of one pentagon and two hexagons and the internal angles differ. The first band is located well below the cutoff fre- quency of the waveguide. Consequently, the electric field modes excited inside the circular cavities are only weakly coupled to those in the neighboring ones. The resonance 4 lead coating. The influence of the former turned out to be negligible for sufficiently short antennas. The effect of the latter on the size of the splittings of the nearly degen- erate resonance frequencies was tested by smoothing the surface of the resonator which indeed induced a reduction of the splittings in each group. The pair of triplets visible in the middle panel of Fig. 3 and shown in a further mag- nification of the spectrum in the lowest panel, is closest to the Dirac frequency at fD = 8.504 GHz which was determined as described below. The zoom demonstrates the high resolution necessary to resolve the 6 resonances. These are the 6 modes conjectured by the Atiyah-Singer index theorem that we were looking for. As is discussed next they are corroborated also by TBM calculations. Tight-binding model description of the spectra. -- The eigenvalues of the C60 molecule have been computed pre- viously using the TBM [12], however, a stringent test of its applicability was missing. Given the experimental results on the eigenfrequencies in the first band of the fullerene resonator we are now in a position to check the validity of the TBM in detail. As stated above, the modes excited in the 60 cavities are weakly coupled, which is an essential prerequisit for the applicability of the TBM. Detailed calculations showed a quantitative agreement between the computed and the measured fre- quencies only when including next-nearest, and second and third-nearest neighbor couplings with strengths t1, t2 and t3, respectively. This yielded for the frequency of the isolated cavities f0 = 8.515 GHz and the coupling parameters t1 = −0.0929 GHz, t2 = 0.0035 GHz and t3 = 0.0005 GHz. The eigenvalues deduced from the TBM appear as 15 groups of degenerate eigenvalues with the same multiplicities as the resonances in the spectrum shown in the middle panel of Fig. 3. An even better agreement was achieved by taking into account the fact that, due to the inhomogeneities in the lead coating, the radii of the cavities are slightly different. In order to esti- mate the deviations thus induced in f0, we used the fact, that f0 is given by the first zero of the J0 Bessel func- tion, J0(kR) = 0, for a circular cavity of radius R. We inserted for each of the 60 cavities the measured radius and replaced f0 by the individual values. Thereby, the degeneracies were removed. In panel a) of Fig. 4 we com- pare the resonance density ρ(f ) = Pi δ(f − fi) obtained for the frequencies fi in the first band (black full line) with the TBM result (red dashed line). For display pur- poses we have replaced the δ functions by Lorentzians of finite width of Γ = 2 MHz. The good agreement reassures the applicability of the TBM and thus justifies its use for further numerical studies with larger fullerene molecules with the parameters determined from the experiment. Panel b) of Fig. 4 shows a comparison between the reso- nance densities of the C60, C240, C540 and C720 molecules in ascending order. Here, we used Lorentzians of width Γ = 5 MHz for all cases. As stated above, all molecules contain the same number of pentagons whereas the num- FIG. 3: Transmission spectrum of the fullerene billiard (upper panel) up to 40 GHz. The first band ranges from 8.254 to 8.779 GHz. It contains 60 resonances, that are separated into 15 groups with the number of resonances indicated in the middle panel, which shows a zoom into it. The Dirac frequency fD = 8.504 GHz is marked by an arrow. The two triplets of interest (the zero modes) are clearly resolved and shown in the lower panel. frequencies in the second band are above the cutoff fre- quency. Accordingly, the modes in the cavities are cou- pled via the modes inside the waveguides, and thus mim- ick a situation where the C atoms are coupled to the neighboring ones via an extra atom, thus explaining the number of resonances in this band. The third band is still below the frequency f 2 c & 20.143 GHz of the sec- ond propagating mode in the waveguides. As a result, the number of possible mode configurations is restricted due to the symmetry properties of the modes excited in- side the cavities, that prefereably couple to the second excited mode inside the waveguide. Above 20.232 GHz, i.e., beyond f 2 c , several bands are intertwined. In summary, only the first band can be used to model the situation in the fullerene C60 molecule. The middle panel of Fig. 3 shows a magnification of it. Fifteen groups of nearly degenerate resonances are clearly visible. The number of resonances identified in each of them is indi- cated and coincides with the degrees of degeneracy pre- dicted on the basis of group theoretical considerations for the eigenfrequencies because of its truncated icosahedral structure [12, 34]. In the group with degeneracy degree 9, in fact, the energy values of 5 and 4 degenerate eigen- frequencies, respectively, are accidentally the same. We emphasize, that we were only able to identify all 60 res- onance frequencies because the degeneracies were lifted. The reason is that the symmetries of the resonator struc- ture were slightly perturbed due to the presence of the an- tennas and unavoidable marginal inhomogeneities in the 5 line) and by a circle for the one further away are displayed in Fig. 5 as a function of the number n of C atoms. As is clearly visible, the distance between the triplets decreases with increasing size of the fullerene molecule and both approach the Dirac frequency. This behavior is well fitted by a function f (n) = fD + a/nb yielding the parameter values given in the caption of Fig. 5, and for the Dirac frequency finally a value of fD = 8.504 GHz. Note that this is essentially the only way to determine the Dirac frequency of a C60 molecule. FIG. 4: (Color online) Panel a): Resonance density deter- mined from the resonance frequencies of the first band (mid- dle panel in Fig. 3) of the fullerene billiard (black solid line) compared to the TBM calculations (red dashed line). Panel b): Resonance densities computed within the TBM for C60, C240, C540 and C720 (lowermost to upmost curve). Panel c): Experimentally determined resonance density of a rectangular Dirac billiard [10, 35]. ber of hexagons increases. The resonance densities should thus resemble more and more that of a graphene sheet. They exhibit a minimum bounded by two increasingly sharp peaks, that evolve into van Hove singularities in the limit of an infinite number of atoms [35, 36]. A com- parison of the resonance densities with that of a rectan- gular graphene sheet with periodic boundary conditions (panel c) of Fig. 4), which has been obtained in mea- surements with a microwave Dirac billiard [10], shows that they resemble for large fullerenes. In contradistinc- tion to the latter, however, the resonance densities of the fullerenes all exhibit a peak of similar size located at the minimum which is due to the two triplets of zero modes. This remains true in the limit of an infinite number of atoms [13 -- 16] and is thus a distinct feature of the spa- tial curvature and topology. According to the Atiyah- Singer index theorem a plane graphene sheet with pe- riodic boundary conditions should not exhibit any zero modes. This is observed in panel c). Zero modes are ex- pected only, if part of the graphene sheet is terminated with zigzag edges [31]. The associated states are called edge states, because their wave functions vanish every- where except at these edges [6, 10, 32]. We have also computed the wave functions of the fullerene molecules under consideration using the TBM and found, that those of the 6 zero modes are localized at the pentagons, which may be considered to be equivalent to zigzag edges within the hexagon network. The central frequencies of the two triplets, marked by squares for the one close to the Dirac frequency (dotted FIG. 5: (Color online) Calculated frequencies of the zero modes of the fullerenes C60, C180, C240, C540 and C720 vs. the number n of C atoms. All pairs of triplets (squares and circles) are located slightly above the Dirac frequency (dot- ted line). The experimental values are marked by a red filled circle and square. The dashed lines correspond to fits of the function f (n) = fD + a/nb to the data points yielding for the Dirac frequency fD = 8.504 GHz, a = 1.188 GHz, b = 0.661 for the lower zero modes and a = 1.211 GHz, b = 0.869 for the upper ones. Conclusions. -- The lowest 60 eigenvalues of a C60 fullerene were determined in high-precision experiments using a superconducting microwave billiard of corre- sponding shape. They appear in 15 groups of nearly de- generate ones, where the multiplicity coincides with that determined based on the group theory of the truncated icosahedral structure of C60. We have demonstrated in TBM calculations for spherical fullerene molecules of in- creasing size that the two triplets of resonances detected close to the Dirac frequency correspond to the triplets of zero modes predicted by the Atiyah-Singer index theo- rem, and thus provided to the best of our knowledge the first experimental test of it. The exact value of the Dirac frequency was obtained as the asymptotic value attained by the frequencies of the triplets in the limit of an infinite number of atoms. This work was supported by the DFG within the Col- laborative Research Center 634. ∗ Electronic address: [email protected] [1] C. W. J. Beenakker, Rev. Mod. Phys. 80, 1337 (2008). [2] A. H. Castro Neto, F. Guinea, N. M. R. Peres, K. S. Novoselov, and A. K. Geim, Rev. Mod. Phys. 81, 109 (2009). [3] L. A. Ponomarenko, F. Schedin, M. I. Katsnelson, R. Yang, E. W. Hill, K. S. Novoselov, and A. K. Geim, Science 320, 5874 (2008). [4] M. Polini, F. Guinea, M. Lewenstein, H. C. Manoharan, and V. Pellegrini, Nature Nanotech. 8, 625 (2013). [5] S. Bittner, B. Dietz, M. Miski-Oglu, P. Oria Iriarte, A. Richter, and F. Schafer, Phys. Rev. B 82, 014301 (2010). [6] U. Kuhl, S. Barkhofen, T. Tudorovskiy, H.-J. Stockmann, T. Hossain, L. de Forges de Parny, and F. Mortessagne, Phys. Rev. B 82, 094308 (2010). [7] M. Bellec, U. Kuhl, G. Montambaux, and F. Mortes- sagne, Phys. Rev. Lett. 110, 033902 (2013). [8] M. C. Rechtsman, J. M. Zeuner, A. Tunnermann, S. Nolte, M. Segev, and A. Szameit, Nat. Photonics 7, 153 (2013). [9] A. B. Khanikaev, S. H. Mousavi, W.-K. Tse, M. Kargar- ian, A. H. MacDonald, and G. Shvets, Nat. Mater. 12, 233 (2013). [10] B. Dietz, T. Klaus, M. Miski-Oglu, and A. Richter, Phys. Rev. B 91, 035411 (2015). [11] H. Kroto, J. Heath, S. O'Brien, R. Curl, and R. Smalley, Nature 318, 162 (1985). [12] E. Manousakis, Phys. Rev. B 44, 10991 (1991). [13] J. Gonz´alez, F. Guinea, and M. A. H. Vozmediano, Phys. Rev. Lett. 69, 172 (1992). [14] J. Gonz´alez, F. Guinea, and M. Vozmediano, Nucl. Phys. B 406, 771 (1993). [15] D. V. Kolesnikov and V. A. Osipov, Eur. Phys. J. B 49, 465 (2006). [16] M. Vozmediano, M. Katsnelson, and F. Guinea, Phys. Rep. 496, 109 (2010). [17] J. K. Pachos, A. Hatzinikitas, and M. Stone, Eur. Phys. J. 148, 127 (2007). 6 [18] J. K. Pachos and M. Stone, Int. J. Mod. Phys. B 21, 5113 (2007). [19] M. F. Atiyah and I. M. Singer, Bull. Amer. Math. Soc. 69, 422 (1963). [20] M. F. Atiyah and I. M. Singer, Ann. Math. 87, 484 (1968). [21] M. F. Atiyah and G. B. Segal, Ann. Math. 87, 531 (1968). [22] M. F. Atiyah and I. M. Singer, Ann. Math. 87, 546 (1968). [23] P. E. Lammert and V. H. Crespi, Phys. Rev. B 69, 035406 (2004). [24] R. S. Millman and G. D. Parker, Elements of differential geometry (Prentice-Hall: Englewood Cliffs, NJ, 1977). [25] M. Atiyah, R. Bott, and V. Patodi, Invent. Math. 19, 279 (1973). [26] M. Stone, Ann. Phys. 155, 56 (1984). [27] A. Richter, in Emerging Applications of Number Theory, The IMA Volumes in Mathematics and its Applications, edited by D. A. Hejhal, J. Friedmann, M. C. Gutzwiller, and A. M. Odlyzko (Springer, New York, 1999), vol. 109, p. 479. [28] F. Haake, Quantum Signatures of Chaos (Springer- Verlag, Heidelberg, 2001). [29] H.-J. Stockmann, Quantum Chaos: An Introduction (Cambridge University Press, Cambridge, 2000). [30] G. E. Mitchell, A. Richter, and H. A. Weidenmuller, Rev. Mod. Phys. 82, 2845 (2010). [31] J. Wurm, K. Richter, and I. Adagideli, Phys. Rev. B 84, 075468 (2011). [32] S. Bittner, B. Dietz, M. Miski-Oglu, and A. Richter, Phys. Rev. B 85, 064301 (2012). [33] B. Dietz, T. Klaus, M. Miski-Oglu, A. Richter, and M. Wunderle, in preparation. [34] M. Dresselhaus, G. Dresselhaus, and P. Eklund, Science of Fullerenes and Carbon Nanotubes (Academic Press, San Diego, 1996). [35] B. Dietz, F. Iachello, M. Miski-Oglu, N. Pietralla, A. Richter, L. von Smekal, and J. Wambach, Phys. Rev. B 88, 104101 (2013). [36] L. Van Hove, Phys. Rev. 89, 1189 (1953). 

1011.2650

1

1011

Nonmonotonic Evolution of the Blocking Temperature in Dispersions of Superparamagnetic Nanoparticles

We use a Monte Carlo approach to simulate the influence of the dipolar interaction on assemblies of monodisperse superparamagnetic ${\gamma}-Fe_{2}O_{3}$ nanoparticles. We have identified a critical concentration c*, that marks the transition between two different regimes in the evolution of the blocking temperature ($T_{B}$) with interparticle interactions. At low concentrations (c < c*) magnetic particles behave as an ideal non-interacting system with a constant $T_{B}$. At concentrations c > c* the dipolar energy enhances the anisotropic energy barrier and $T_{B}$ increases with increasing c, so that a larger temperature is required to reach the superparamagnetic state. The fitting of our results with classical particle models and experiments supports the existence of two differentiated regimes. Our data could help to understand apparently contradictory results from the literature.

cond-mat.mes-hall

cond-mat

Non-monotonic Evolution of the Blocking Temperature in Dispersions of Superparamagnetic Nanoparticles D. Serantes, D. Baldomir, and M. Pereiro Instituto de Investigacións Tecnolóxicas and Departamento de Física Aplicada, Universidade de Santiago de Compostela, 15782 Santiago de Compostela, Galiza (Spain) C. E. Hoppe Div.Polímeros, INTEMA (UNMdP-CONICET) Universidad Nacional de Mar del Plata B7608FDQ, Mar del Plata, Argentina F. Rivadulla Departamento de Química Física, Universidade de Santiago de Compostela, 15782 Santiago de Compostela, Spain J. Rivas Departamento de Física Aplicada, Universidade de Santiago de Compostela, 15782 Santiago de Compostela, Spain Abstract We use a Monte Carlo approach to simulate the influence of the dipolar interaction on assemblies of monodisperse superparamagnetic γ-Fe2O3 nanoparticles. We have identified a critical concentration c*, that marks the transition between two different regimes in the evolution of the blocking temperature (TB) with interparticle interactions. At low concentrations (c < c*) magnetic particles behave as an ideal non- interacting system with a constant TB. At concentrations c > c* the dipolar energy enhances the anisotropic energy barrier and TB increases with increasing c, so that a larger temperature is required to reach the superparamagnetic state. The fitting of our results with classical particle models and experiments supports the existence of two differentiated regimes. Our data could help to understand apparently contradictory results from the literature. PACS: 75.75.+a, 75.20.-g, 75.40.Mg, 75.50.Tt 1 I. INTRODUCTION Envisaging new methods for the synthesis of monodisperse magnetic nanoparticles (NPs) constitutes one of the most active and challenging research fields in materials science due to the almost infinite uses foreseen for these systems in biomedicine,1 magnetic recording,2 energy production, etc.3 Many of these applications rely on the possibility of obtaining well dispersed assemblies of NPs in a non-magnetic matrix (normally diamagnetic solids like polymers, SiO2, etc). This step can be really tough due to the high tendency of NPs to aggregate producing heterogeneous and uncontrolled structures. Aggregation is typically induced by interaction forces (van der Waals, dipolar) that depend on different variables of the system, like the nature of the matrix and the molecules that coat the surface of the particles. In this sense, determining the magnetic properties of these dispersions of NPs, particularly the magnetization under zero-field-cooling and field-cooling conditions (ZFC-FC) and the blocking temperature (TB), is becoming an increasingly popular analytical technique, due to its high sensitivity to the distribution of the particles in the matrix, its surface oxidation state, homogeneity, etc. However, it is usual to find in the literature dissimilar (sometimes contradictory) results from, a priori, similar samples. This is most probably due to the high sensitivity of the magnetic properties to interparticle interaction, which can make very easy to confuse extrinsic effects with intrinsic ones. In fact, understanding the role played by dipolar interactions into the magnetic behavior of the system remains a challenge despite the intense investigation and discussion devoted to it.4,5,6 The interest of solving it is important for practical applications since it is a key-parameter for driving the magnetic response of nanotechnological devices and applications using magnetic NPs. Here we describe the effect of the magnetic dipolar interaction in the evolution of the ZFC-FC curves and the TB of a monodisperse system of maghemite-like superparamagnetic (SPM) NPs. The anisotropy energy (responsible for the existence of TB) is for these particles rather low, and so the system is very sensitive to variations of the dipolar energy. We will compare our results from Monte Carlo (MC) simulations with our own experiments and other ones available in the literature. Our results demonstrate the existence of two different regimes in the evolution of TB with the concentration of NPs in homogeneous dispersions. II. COMPUTATIONAL DETAILS 2 We have used a Monte Carlo (MC) technique7,8,9 to study the magnetic response of an assembly of single domain magnetic nanoparticles as a function of the magnetic dipolar interactions between the particles. We have chosen for the system ideal characteristics accounting to eliminate deviations from the intrinsic behaviour, with the purpose to set an appropriate frame to understand the basic magnetic properties of the system. First, the particles are spatially distributed into an ideal liquid-like structure that resembles a magnetic ferrofluid without aggregations. To acquire such a distribution we have used a Lennard-Jones pair potential with periodic boundary conditions, in the same way as done in Ref. 7. The positions of the particles are kept fixed from now on, assuming the same condition of particles fixed in a non magnetic matrix or a frozen ferrofluid. The second idealization we make is to consider a wholly monodisperse system: the particles, characterized by its volume, anisotropy and magnetization, are assumed to be all equal in their characteristics, so that no effects on the magnetic response of the system may be attributed to polydispersity of any type. The energies that we have taken into account to govern the magnetic behaviour of the system are anisotropy (EA), Zeeman (EZ), and dipolar (ED). The anisotropy of the particles is assumed uniaxial for the  2    n  i i i sake of simplicity, so that for the i-particle , where K is the uniaxial anisotropy KV  E   i A constant and in indicates the direction of the anisotropy easy axis. In the ideal superparamagnetic frame it is considered single-domain NPs with the inner atomic moments rigidly coupled, what results the total magnetic moment for the i-particle to be i  VM S , where MS is the saturation magnetization and V is the volume of the particle. The Zeeman energy is treated in the usual way E   i H    H   , and the dipolar i interaction energy between two particles located at ir ,           r     ij i j i j  ji , D 5 r ij 3 r ij  r ij E   3 jr respectively, is given by   , with ijr the vector connecting the particles. To evaluate the long-range interactions we applied periodic boundary conditions by means of Ewald’s summation. The total energy of the system is the summation of the different terms extended to all the particles. The motion of the individual magnetic moments of the particles as a function of the temperature (T) was driven by means of the Metropolis algorithm: in every MC step, we select a particle i at random and generate a new orientation of its magnetic moment. Then, we accept the new orientation with probability 3 min[1,exp(-ΔE/kBT)], where ΔE is the energy difference between the attempted and present orientations, and kB is the Boltzmann constant. In every MC step N attempts are made, where N is the number of particles used in the simulation (N=125 in this study). The results are obtained averaging over 1000 different configurations, and extend those reported by us in Ref. 9 with a larger precision, in order to obtain more reliable conclusions. We have studied the influence of the magnetic dipolar interaction on the superparamagnetic properties of the system by analyzing its influence on its characteristic blocking temperature (TB), roughly evaluated as the temperature at which the ZFC curve exhibits a maximum. The strength of the dipolar interaction is introduced as proportional to the sample concentration, c. For a monodisperse particle assembly the equation relating the dipolar energy (ED) and c is given by E   ji , D  g  i i j  e e   i j 3 a ij      e  i   3   ea ij  j 5 a ij   a ij      (5) characterizes the strength of the dipolar interaction. The dimensionless sample g  KV N where c c 0 concentration c is the ratio between the total volume  i iV occupied by the particles (NV for the monodisperse sample) and the volume L3 of the sample, c  3LNV . The value c  0 K 2 2 SM is a dimensionless constant that characterizes the material.8 The unit vector ie stands for the direction of the magnetic moment μ of the particle i, and the reduced distance ija is defined as   a ij  Lr ij , the distance between the particles i and j divided by the size of the cubic box that contains the sample. The results obtained from our simulations are presented in reduced units directly related to the real ones, i.e., the reduced sample concentration is c/c0, and t=kBT/2KV is the reduced temperature. The reduced applied magnetization is defined as magnetic field is h=H/HA, where HA=2K/MS is the anisotropy field of the particles, and the reduced S  the i-particle and the direction of the applied magnetic field. , with θi the angle between the magnetic moment of MM=m cos i N i 4 We studied the influence of the dipolar interaction on TB by simulating ZFC processes at different sample concentrations, ranging from the non-interacting diluted limit (c/c0 = 0.000) to very dense samples (c/c0 ≤ 0.320). For every simulation, we have first demagnetized the samples at very high temperature and then cooled them down in zero applied magnetic field; once the sample has reached a very low temperature, a small reduced field h = 0.1 was applied and the magnetization was measured while the samples were heated up at constant rate until well above the reduced blocking temperature, tB. The heating/cooling rate was Δt=0.001225KV/kB every 200 MC steps. The reduced susceptibility is defined as χ=m/h. We systematically vary c to evaluate tB as a function of the dipolar interaction energy, as it is summarized in Fig. 2.10 III. RESULTS AND DISCUSION Some representative simulated ZFC susceptibility curves of the dispersions of magnetic nanoparticles at different sample concentrations are shown in Fig. 1, where also the FC curves of two different interacting conditions (c/c0 → 0.000 and c/c0 = 0.112) are included to illustrate the reliability of the code. The applied magnetic field was H = 0.1HA. This small value of the magnetic field was selected to not disturbing the intrinsic SPM behavior of the NPs. Large fields could mask this effect. 4.0 c/c0 0.000 0.048 0.112 0.176 0.240 0.288 C F Z  3.5 3.0 2.5 2.0 1.5 1.0 0.5 0.0 0.0 0.1 0.2 0.3 kBT/2KV 0.4 0.5 FIG. 1. ZFC curves (empty symbols) of some representative samples. Two FC curves (full symbols) are shown, corresponding to the non-interacting case (c/c0 → 0.000) and to c/c0 = 0.112. Numerical simulations of the ZFC/FC processes show a good agreement with the general trend of experimental results on SPM particles.11,12 The ZFC curves exhibit a maximum at the reduced blocking 5 temperature (tB) that marks the transition to the SPM regime. Above tB it is observed the superposition of the FC with the ZFC curves that shows the reversible character of the SPM behavior. Below tB, in the irreversible range, the FC curve separates from the ZFC. We also see from Fig. 1 an increase of tB as the concentration increases, in agreement with the results reported in experimental12 and theoretical works.7 The detailed dependence of tB vs. c/c0 extracted from a complete set of measurements is summarized in Fig. 2. 0 .16 0 .15 0 .14 B 0 .13 t 0 .12 0 .11 0 .10 c* 0 .00 0 .05 0 .10 0 .20 0 .25 0 .30 0 .15 c/c0 FIG. 2. Plot of tB as a function of c/c0. The solid line is a fitting to the modified single particle approach (see subsection D). The arrow indicates the concentration c*. From Fig. 2, we observe a non-monotonic dependence of tB with concentration: there is a clear change of the slope from an independent tB at low concentrations to a rapid increase at high concentrations. These different features suggest the presence of two different physical behaviors, an essentially non-interacting regime at low concentrations and an interacting regime at high values of c/c0. The crossover between both regimes is marked by a particular concentration, estimated to be c*/c0 ≈ 0.05. In the next subsections we show some tests we have done to analyze the characteristics of the observed two-regime feature. A. Time dependence. As the SPM tB is highly time-dependent,13 we have simulated the different processes in Fig. 1 at different time intervals (different MC steps), in order to rule out the existence of the two regimes to be a time dependent effect. We varied the MC simulation intervals and maintained the same cooling/heating temperature step. The simulated intervals correspond to 20 and 50 MC steps. The results are shown in Fig. 3, together with the 200 MC steps case of Fig. 2 for the comparison. 6  MC steps 20 50 200 0.22 0.20 0.18 B t 0.16 0.14 0.12 0.10 0.0 0.1 0.2 0.3 c/c0 FIG. 3. Plot of tB vs. c/c0 for different simulation times (MC steps). The dotted lines are guides to the eye. Seemingly, the overall tendency is the same for the different measuring times: at low concentrations tB remains basically constant, suggesting the particles to behave independently of each other as a non- interacting system; at higher concentrations tB increases continuously with the concentration. Hence, we have confirmed that the evolution of tB with c/c0 is robust for different time intervals. B. Maxima at TB. To test the existence of the two different regimes we have also analyzed the relative values of the susceptibility at the maximum of the ZFC curves, χ(tB), as a function of concentration. This study constitutes a more precise approach than the analysis of tB because of the higher accuracy found on its determination (see Fig. 1). As the overall observed trend results equal for the different MC step intervals we have focused our study on the 200 MC step case because the simulated ZFC processes show a better definition after more MC relaxation steps. The obtained results are shown in Fig. 4. 7 3.0 2.5 2.0 1.5 1.0 0.5 0.0 ) B t (  c* ) B t ( '  0 -5 -10 -15 c* 0.0 0.1 0.3 0.2 c/c0 0.00 0.05 0.10 0.20 0.25 0.30 0.15 c/c0 FIG 4. Plot of the reduced susceptibility at the maximum of the ZFC curves for the different interacting conditions considered. Inset shows the first derivative at the maximum of the ZFC curves, χ’(tB), as a function of c/c0. The curve χ(tB) vs. c/c0 exhibits an inflexion at low concentrations, as observed in its first derivative (see inset of Fig. 4): an unambiguous minimum appears in χ’(tB) at the same reduced sample concentration value c*/c0 ≈ 0.05 observed in Fig. 2. This inflexion on the χ(tB) vs. c/c0 curve is related to a change in the magnetic behavior, and supports the existence of two intrinsic regimes of different magnetic behavior in a system of SPM-NPs being influenced by the dipolar interaction. C. Comparison with the experiment. We report now experimental results extracted from the literature for similar systems but different particle sizes in order to analyze the size-dependence of the trend reported. A similar low-concentration behavior as that shown in Fig. 2 has been observed for highly diluted samples of a frozen ferrofluid of maghemite NPs of ~7 nm diameter.14 The existence of a defined interparticle spacing separating two different regimes on the evolution of TB has been reported for iron oxide NPs of ~5.4 nm diameter.15 With the purpose of probing the generality of the observed behavior, we reproduce in Fig. 5 our results9 on the evolution of TB with the sample concentration for a system of magnetic NPs of smaller diameter (~3.5 nm) than those reported in Refs. 14, 15. 8  FIG. 5. TB vs. sample concentration for γ-Fe2O3 NPs with different aggregation level: a well dispersed sample (open squares) and a highly aggregated one (full circles). The lines serve only to guide the eye. TEM micrographs show the dispersion of the particles for the well dispersed sample (bottom) and for the aggregated one (top) for the same concentration of particles, 1.3 wt% of γ-Fe2O3. In Fig. 5, we show two cases of different aggregation level of the same NP system: a well dispersed sample (full circles, left bottom TEM micrograph), and a highly aggregated one (empty squares, left top TEM micrograph) (see Ref. 9 for further details). The presence of a non-interacting regime at low concentrations can be observed in these systems only in the well-dispersed sample (full circles). Presence of clustering (empty squares) clearly affects the shape of the curve and hampers the observation of this regime.9 These results show that the existence of two different interacting regimes is an intrinsic property of maghemite-like NPs without aggregation, and that is independent of the NP size. D. Fitting to classical models. Different models have been developed with the purpose to take into account the effect of interparticle interactions on the behavior of the magnetic nanoparticle assemblies. The first attempts were based on modifications of the superparamagnetic single-particle’s model by Nèel.16 In such treatment the interactions between the particles are introduced as changes of the height of the energy barrier, where the different energetic terms add to the anisotropy one. It results in an increment of the thermal activation energy necessary to reach the superparamagnetic state.17 Recently, W. C. Nunes et al.11 have proposed a modification of the Random Anisotropy Model (RAM) that takes into account the concentration and size of the nanoparticles, as well as the field dependence of the correlation length. We have used both models to 9 fit our results, with the purpose of having one more test to check our arguments. The fitting to both approaches is shown in Fig. 6. A 0.16 0.15 0.14 B t 0.13 R2=0.938 R2=0.972 0.12 0.11 B R2=0.968 R2=0.982 0.16 0.15 0.14 0.13 0.12 0.11 0.0 0.1 0.2 c/c0 0.3 0.0 0.1 0.3 0.2 c/c0 FIG. 6. Fitting of the tB vs. c/c0 curve to the modified single particle approach (dashed green lines) and the modified RAM approach (red dotted lines). In Fig. 6A the whole range of values is fitted, while in Fig. 6B only the data corresponding to the interacting regime is included in the fitting. In Fig. 6A the fitting of the whole-range data (the two regimes) is included, while in Fig. 6B only the interacting regime (c  c*) data is fitted. The (green) dashed lines correspond to the modified single particle approach, and the (red) dotted lines correspond to the modification of the RAM. The square of the correlation coefficient (R2) is shown in the two fittings for both models. It is clearly observed that the fitting gives very satisfactory results in the interacting range and deviates completely from the expectations at low concentration, since it improves the R2 value for both approaches. This result in fact gives an additional support to our arguments of the existence of two different regimes of behavior of the blocking temperature as a function of the dipolar interaction. E. Additional Monte Carlo simulations. In order to assure the independency of the reported results on the system size, we have simulated the evolution of tB with sample concentration using a much larger system, of 1000 particles. Due to computational constraints with this large system size, the temperature variation ratio had to be enlarged to Δt=0.005000KV/kB every 500 MC steps and the results were averaged over 300 different configurations. We have concentrated on the values around c*, aiming to focus on the two-regime threshold feature. The values of tB, χ(tB), and χ’(tB) were evaluated, and the results are shown in Fig. 7. 10 B t 0.18 0.17 0.16 0.15 0.14 0.13 A c* 0.00 0.05 0.15 0.20 0.10 c/c0 2.5 2.0 1.5 1.0 0 -3 -6 -9 -12 ) B t (  ) B t ( '  B C c* 0.00 0.05 0.10 0.15 0.20 c/c0 FIG. 7. The evolution of tB as a function of c/c0 for the N=1000 sample is shown in Fig. 7A, and the corresponding reduced susceptibility and first derivative are plotted in Fig. 7B and Fig. 7C, respectively. Vertical dotted lines indicates the concentration c* extracted from Fig. 2 and Fig. 4. The results plotted in Fig. 7 indicate that the existence of the two-regime feature discussed in the previous subsections is also observed with the 1000 particles’ system and the different temperature interval variation. This demonstrates that the two-regime feature discussed on the previous subsections is independent of the system size used for the simulation. The number of particles considered initially (N=125) is appropriate to study the influence of dipolar interaction on such nanoparticles samples. IV. SUMMARY Monte Carlo simulations of dispersions of maghemite-like NPs demonstrate a discontinuous evolution of the blocking temperature as a function of the sample concentration that stands for two different interacting regimes. The crossover between these two regimes is determined by a critical concentration c*. At low concentrations (c < c*) TB remains basically constant (non-interacting regime), while at high concentrations (c > c*) continually increases (interacting regime). This feature has been intensively discussed and analyzed, finding that: i) it is robust for different time intervals; ii) the maxima at TB also shows an inflexion at c*, what stands for two different behaviors; iii) experimental results show the same tendency and corroborate that its shape is independent on the particle size; iv) our results are in good agreement with the classical interparticle models only in the interacting regime. Moreover, we have also assessed the independency of the reported results on the system size, performing a simulation of the evolution of tB with sample concentration for a very large system consisting of 1000 particles. On the basis 11 of the results presented here many of the data of the evolution of TB with concentration in the literature should be revised, considering the possibility of clustering of NPs. ACKNOWLEDGEMENTS We acknowledge the Xunta de Galicia for Project. No. INCITE 08PXIB236052PR, and for the financial support of D. S. and M. P. (Maria Barbeito and Isabel Barreto programs, respectively). We also acknowledge the Spanish Ministry of Education and Science (Projects No. NAN2004-09203-C04-04, NAN2004-09195-C04-01 and Consolider-Ingenio 2010), and C. E. Hoppe thanks the 2006-IIF Marie Curie Grant (Contract Nº: MIF2-CT-2006-021689, AnaPhaSeS). We thank the CESGA for computational facilities. REFERENCES (1) M. K. Yu, Y. Y. Jeong, J. Park, S. Park, J. W. Kim, J. J. Min, K. Kim, and S. Jon, Angew. Chem. Int. Ed. 47, 5362 (2008). (2) S. H. Sun, C. B. Murray, D. Weller, L. Folks, and A. Moser, Science 287, 1989 (2000). (3) B. Poudel, Q. Hao, Y. Ma, Y. Lan, A. Minnich, B. Yu, X. Yan, D. Wang, A. Muto, D. Vashaee, X. Chen, J. Liu, M. S. Dresselhaus, G. Chen, and Z. Ren, Science 320, 634 (2008). (4) X. Batlle and A. Labarta, J. Phys. D: Appl. Phys. 35, R15 (2002). (5) J. L. Dormann, L. Besáis, and D. Fiorani, J. Phys. C 21, 2015 (1988). (6) S. Morup and E. Tronc, Phys. Rev. Lett. 72, 3278 (1994). (7) J. García-Otero, M. Porto, J. Rivas, and A. Bunde, Phys. Rev. Lett. 84, 167 (2000). (8) D. Serantes, D. Baldomir, M. Pereiro, J. Botana, V. M. Prida, B. Hernando, J. E. Arias, and J. Rivas, accepted in J. Nanoscience and Nanotechnology (2009). (9) C. E. Hoppe, F. Rivadulla, M. A. López-Quintela, M. C. Buján, J. Rivas, D. Serantes, and D. Baldomir, J. Phys. Chem. C 112, 13099 (2008). (10) To give an estimation in real units, for example for the 7 nm diameter maghemite nanoparticles reported by Jonsson et al. [Phys. Rev. Lett. 75, 4138 (1995)], the characteristic values are K = 1.9 *105 erg/cm3 and Ms = 420 emu/cm3, what for our simulations results to be an effective magnetic field H = 90 Oe, a temperature variation ration of 0.3K every 200 MC steps, and the critical concentration to be c* ≈ 0.10. 12 (11) W. C. Nunes, L. M. Socolovsky, J. C. Denardin, F. Cebollada, A. L. Brandl, and M. Knobel, Phys. Rev. B 72, 212413 (2005). (12) J. M. Vargas, W. C. Nunes, L. M. Socolovsky, M. Knobel, and D. Zanchet, Phys. Rev. B 72, 184428 (2005). (13) M. F. Hansen, C. B. Koch, and S. Morup, Phys. Rev. B 62, 1124 (2000). (14) T. Jonsson, J. Mattsson, C. Djurberg, F. A. Khan, P. Norblad, and P. Svedlindh, Phys. Rev. Lett. 75, 4138 (1995). (15) B. L. Frankamp, A. K. Boal, M. T. Tuominen, and V. M. Rotello, J. Am. Chem. Soc. 127, 9731 (2005). (16) L. Néel, Ann. Geophys. 5, 99 (1949). (17) J. L. Dormann, D. Fiorani, and E. Tronc, J. Magn. Magn. Mater. 202, 251 (1999). 13 

1706.02907

2

1706

Multi-Phase-Field Model for Surface/Phase-Boundary Diffusion

The multi-phase-field approach is generalized to treat capillarity-driven diffusion parallel to the surfaces and phase-boundaries, i.e. the boundaries between a condensed phase and its vapor and the boundaries between two or multiple condensed phases. The effect of capillarity is modeled via curvature-dependence of the chemical potential whose gradient gives rise to diffusion. The model is used to study thermal grooving on the surface of a polycrystalline body. Decaying oscillations of the surface profile during thermal grooving, postulated by Hillert long ago but reported only in few studies so far, are observed and discussed. Furthermore, annealing of multi-nano-clusters on a deformable free surface is investigated using the proposed model. Results of these simulations suggest that the characteristic crater-like structure with an elevated perimeter, observed in recent experiments, is a transient non-equilibrium state during the annealing process.

cond-mat.mes-hall

cond-mat

Multi-Phase-Field Model for Surface/Phase-Boundary Diffusion Raphael Schiedung,∗ Reza Darvishi Kamachali, Ingo Steinbach, and Fathollah Varnik† Interdisciplinary Centre for Advanced Materials Simulation (ICAMS) Ruhr-Universitat Bochum Universitatsstr. 150, 44801 Bochum, Germany (Dated: October 20, 2018) The multi-phase-field approach is generalized to treat capillarity-driven diffusion parallel to the surfaces and phase-boundaries, i.e. the boundaries between a condensed phase and its vapor and the boundaries between two or multiple condensed phases. The effect of capillarity is modeled via curvature-dependence of the chemical potential whose gradient gives rise to diffusion. The model is used to study thermal grooving on the surface of a polycrystalline body. Decaying oscillations of the surface profile during thermal grooving, postulated by Hillert long ago but reported only in few studies so far, are observed and discussed. Furthermore, annealing of multi-nano-clusters on a deformable free surface is investigated using the proposed model. Results of these simulations suggest that the characteristic crater-like structure with an elevated perimeter, observed in recent experiments, is a transient non-equilibrium state during the annealing process. I. INTRODUCTION Diffusion in materials results from random motion of their constituents (atoms/molecules/voids). Aside from self-diffusion of atoms or molecules within a homoge- neous body in thermal equilibrium, diffusion is one of the most frequent transport mechanisms driven by weak deviations from the equilibrium state, such as a gradient in concentration or, more generally, chemical potential. This latter aspect becomes particularly relevant at sur- faces/interfaces with spatially varying curvature, which, due to Gibbs-Thompson relation, exhibit a gradient in chemical potential along the surface/phase boundary. It is noteworthy that, while the capillarity driven diffusion at free surfaces is rather well investigated (see e.g. [1 -- 3]), much less attention has been payed to diffusion at the boundaries between two condensed phases. This work fo- cuses on this issue using the phase-field method. In order to distinguish this type of diffusive transport from grain- boundary diffusion, in which boundaries are faster tracks for diffusion of solute atoms, the term 'phase-boundary diffusion' is introduced here to express the capillarity- driven diffusion in boundaries. A first investigation of a conserved phase-field model has been carried out by Cagi- nalp in [4, 5]. Later, a conserved phase-field method has been used to simulate the evolution of elastically stressed films [6, 7]. These models, however, are limited to surface diffusion between two phases. A comparison of the avail- able phase-field models for surface diffusion is presented in [3]. In multiphase multi-grain systems, diffusion in phase-boundaries requires a treatment of multiple con- served phases at triple and higher-order junctions. In this work, we present a conserved multi-phase- field method which is able to describe the capillarity- driven surface/phase-boundary diffusion in such complex ∗ [email protected] † [email protected] topologies, where multiple junctions may be present. The current method is based on the works reviewed by Gu- genberger et al. [3] on the two-phase surface diffusion models and the non-conserved multi-phase-field model [8, 9]. For the sake of simplicity, here we consider immis- cible phases without bulk diffusion and phase transforma- tion, and solely focus on the effects of surface and phase- boundary diffusion. The present approach can be used to investigate complex systems for which it is not possible to obtain an analytic solution or which are simply too large to be investigated by atomistic methods. One of these possible scenarios is the solid-phase sintering of many particles where we can reach beyond the time and length scales of atomistic simulations. As an example, we apply our model to study surface and phase-boundary diffusion of nano-clusters during annealing on a free surface. This is similar to the experimental work presented in [10]. Fur- thermore, the construction of the current model allows the combination of surface/boundary diffusion with dis- sipative interface diffusion [11, 12], elasticity [13, 14] and chemo-mechanical coupling [15, 16] effect in the same way that they are applied in non-conserved multi-phase-field methods. The paper is organized as follows. In Sec. I A, a short introduction into the thermodynamics of phase-boundary diffusion is given. The equilibrium solution at triple junctions is discussed in Secs. I B and I C. The phase- boundary diffusion model is presented in Sec. II. Sec- tion III compiles the obtained simulation results. After a brief discussion of the equilibrium interface profile in Sec. III A, thermal grooving is addressed in Sec. III B. The late time profiles obtained from these simulations are then used in Sec. III C to check whether the method correctly recovers the von Neumann's triangle relation. Section III D highlights that the equilibrium behavior at the junctions is strongly dependent on the conservation constraints. The evolution of multiple nano-clusters on a deformable surface is discussed in Sec. III E. A summary compiles the main findings in Sec. IV. 7 1 0 2 n u J 2 1 ] l l a h - s e m . t a m - d n o c [ 2 v 7 0 9 2 0 . 6 0 7 1 : v i X r a A. Thermodynamics of Surface Diffusion 2 We follow an approach similar to Mullins [2] and con- sider a body of single species surrounded by its vapor phase at constant temperature. The surface is assumed to be in equilibrium with a finite curvature, κ, and the interface energy, σ. The curvature is here considered as the sum of the principal curvatures, κ1 and κ2, with κ = κ1 + κ2. The vapor pressure, pV , is given by (assum- ing ideal gas and incompressible solid), (cid:18) pV (cid:19) p0 ln = σVm kBT κ , (1) where p0 is the vapor pressure for the planar surface, kB is the Boltzmann constant, Vm is the molecular volume of the solid phase and T is the temperature. The chemical potential, µ, of an ideal gas can be approximated as µ ≈ −kBT ln . (2) In this relation, λ is the thermal de Broglie wave length. Inserting the vapor pressure, pV , from Eq. (1) into Eq. (2) results in an expression for the dependency of the chem- ical potential on curvature, (cid:18) kBT (cid:19) p0λ3 µ = −kBT ln (cid:18) kBT (cid:19) pV λ3 (a) (b) FIG. 1. The figure (a) shows the three coexisting phases, α, β and γ that are in contact to each other. θα, θβ and θγ are the so called contact angles. σαβ, σβγ and σαγ are the interface energies acting on the triple point x0 alongside their interfaces. The von Neumann's triangle (b) visualizes the equilibrium condition, where the length of the triangle sides is given by the value of the interface energies. where M is a mobility coefficient (see [2]). Usually, con- served surface diffusion and non-conserved phase transi- tion processes are studied separately. In principle, how- ever, one can think of situations in which a combination of both effects determines the surface normal velocity, (cid:16) (cid:17) + σVmκ . (3) vn = σ M∇2 sκ − M κ , (8) Considering a body of arbitrary curvature, this leads to a flux of particles, j, from regions of high curvature to regions of low curvature, j = −µc∇sµ = −µσcVm∇sκ , (4) where µ is the mobility of the particles and c the con- centration of particles per unit area. Here we treat the surface, s, as an isosurface of the concentration field with a constant value of c. In a more general concept of multi- ple species, the interface energy may depend on the com- position and thus the concentrations may vary alongside the surface as well. Equation (4) describes the flux of particles parallel to the surface. The local increase of concentration of particles per unit area, c, is obtained by taking the surface divergence of j so that c = DsσcVm kBT ∇2 sκ , (5) where Ds = µkBT is the surface diffusion coefficient. By multiplying c with the molecular volume, the surface normal velocity can be obtained, DsσcV 2 m kBT ∇2 vn = sκ . (6) Equation (6) is the basic equation which characterizes the dynamics of surface diffusion. In non-conserved pro- cesses, e.g. phase transition, evaporation and condensa- tion, the surface velocity is often described by a linear frictional ansatz vn = −σM κ , (7) in which a second mobility coefficient has been intro- duced, M . Equation (8) states that the significance of the two contributions is determined by the coefficients M and M , as well as the length scale of the process. For smaller scales, i.e. larger curvatures, the effect of surface diffusion becomes increasingly more important. It is to be noted that the dynamics of surface diffusion, Eq. (6), for the solid/vapor interface is derived under the assump- tion of an ideal gas. This is a good approximation for a capillary driven diffusion at a solid/vapor interface at el- evated temperatures. B. Not Deformable Surface A common example of three phases in contact is a liq- uid droplet, α, on a flat non-deformable solid surface, β, surrounded by a gas phase, γ. By minimizing the surface energy, one can obtain Young's law for the contact angle, cos θ = σβγ − σβα σγα . (9) One can see that Eq. (9) does not have a solution for all possible values of the interface energies. The droplet can either detach form the surface, σβγ−σβα ≤ −1, or can completely wet the surface, σβγ−σβα ≥ 1. More frequently, however, one has to deal with the case of partial wetting, described by the intermediate values, σγα σγα −1 < σβγ−σβα σγα < 1. σβγσαγσαβx0γβαθγθαθβπ−θγπ−θαπ−θβσβγσαβσαγC. Deformable Surface The above consideration becomes more complex for three deformable phases in contact. At a triple point in equilibrium, the sum of all forces acting on the point is zero and the force vectors form a triangle (see Fig. 1b). Then the law of sines can be applied to express the rela- tion between the magnitude of the forces and their angles, θα, θβ and θγ, to each other, σβγ sin θα = σαγ sin θβ = σαβ sin θγ . (10) This construction for the interface energies is referred to as the von Neumann's triangle relation (see e.g. [17 -- 19]). A comprehensive derivation of Eq. (10) can be found in section 8.2 of the seminal textbook on molecular theory of capillarity by J.S. Rowlinson and B. Widom [20]. In the case of a junction between more than three phases, the situation becomes more complex. The general approach, however, is the same and is based on the idea of interface energy minimization. II. THE PHASE-BOUNDARY DIFFUSION MODEL In the following, we present a model for phase- boundary diffusion which is driven by the minimization of the local interface energy. The present approach com- bines an existing non-conserved multi-phase-field model with the conserved phase-field models for surface diffu- sion at two-phase-boundaries. The description of the referred multi-phase-field model is detailed in two arti- cles [8] and [9]. Furthermore, a review of surface diffusion models is given in [3]. In comparison to existing phase- field models for surface diffusion, the present model can describe the capillary driven diffusion of multiple phases. Since the phase-boundary diffusion is the only process considered here, the volume of each individual phase is conserved. A. The Interface Free Energy Density Functional It is convenient to start with a suitable description of the free energy, F . The free energy depends on the spacial configuration of N phase-fields φα (x, t) with α ∈ [1, N ], the spatial position x and the time t. Each phase- field φα indicates the location of a thermodynamic phase α, in a way that φα = 1 means phase α is present and φα = 0 that it is not. Additionally, a sum constrained α=1 φα = 1. In the case of phase-boundary diffusion, the phase-field value may be interpreted as an indicator function of a thermodynamic phase with a fixed composition and density. for the phase-fields is used,(cid:80)N Following the multi-phase-field approach in [8, 9], the total free energy can be written as a functional F ({φα}). 3 FIG. 2. The potential function φαφβ from Eq. (12) which reduces in the case of two phase-fields to φ (1 − φ) with φ = φα = 1 − φβ. The dashed curve corresponds to φ (1 − φ). It has no minimum, but tends to −∞ as φ → ±∞. This free energy functional is defined through the free energy densities, fαβ, of the interface between the phases α and β. The sum of all interface energy densities is integrated over the volume of interest, Ω, F ({φ}) = fαβ (φα, φβ) dV . (11) N(cid:88) N(cid:88) (cid:90) α=1 β=α+1 Ω One may add other energy densities into the concept, in order to include other physical effects. The energy densi- ties of the interfaces are proposed in a pair-wise manner between existing phase-fields, fαβ = 4σαβ η η2 π2∇φα · ∇φβ + φαφβ − . (12) (cid:19) (cid:18) Here, η defines the width of the transition region be- tween two phase-fields, e.g., φα and φβ. φαφβ is called double obstacle potential and restricts the value of the phase-field to the interval [0, 1]. Figure 2 shows a plot of the double obstacle potential which can be simplified to φ (1 − φ) with φ = φα = 1 − φβ in the case of only two phase-fields. As it can be seen from Fig. 2, the use of absolute value is necessary to ensure that φ = 0 and φ = 1 correspond to two equilibrium solutions. B. The Dynamics of Phase-Boundary Diffusion The dynamic equations for phase-boundary diffusion can now be derived by using the variational derivative of the free energy functional with respect to the phase-field, (cid:18) δ (cid:19) ∇ · M αβ∇ δφα − δ δφβ F (13) N(cid:88) N(cid:88) β=1 β=1 φα = − = − 1 N 1 N ∇ · M αβ∇ψαβ . −0.200.20.40.60.811.2−0.200.20.4PhasefieldΦφ(1−φ)φ(1−φ)Here, M αβ = M βα is a symmetric mobility tensor of the interface between the phases α and β. It is of fourth-rank with Mjiαβ = Mijαβ, where i and j indi- cate Cartesian coordinates, i, j ∈ {x, y, z}. Equation (13) ensures that each phase-field, φα, is conserved. More- φα = 0, it follows that the sum rule, α=1 φα(t) = 1, is fulfilled at all times, t, provided that In Eq. (13), ψαβ is a generalized it is valid at t = 0. diffusion potential, defined via α=1 over, since (cid:80)N (cid:80)N (cid:18) δ (cid:19) ψαβ = δφα − δ δφβ F N(cid:88) = Iαβ − Iβα + γ=1,γ(cid:54)=α,γ(cid:54)=β (14) (Iαγ − Iβγ) . The function Iαβ is a shorthand notation for the func- tional derivative of the free interface density fαβ, Iαβ = = ∂fαβ ∂φα − ∇ 4σαβ (cid:18) η2 . ∂fαβ ∂∇φα π2∇2φβ + η (cid:19) (15a) . (15b) ∂ φαφβ ∂φα It should be remarked here that Iαβ (cid:54)= Iβα. In general, the interface energy, σαβ (nαβ), is a function of the in- terface orientation which can be determined by using the interface normal vector, nαβ = φα∇φβ − φβ∇φα (cid:107)φα∇φβ − φβ∇φα(cid:107) . (16) This leads to additional terms in Eq. (15) such as the Herring torque [1]. For the sake of simplicity, however, this study is restricted to the case of a constant interface energy. A comment is necessary here regarding the use of a three dimensional ∇ operator in Eq. (13) instead of the surface gradient ∇s (see Eq. (5)). This is a consequence of the fact that, in diffuse interface approaches such as the present multi-phase-field model, a strictly two dimen- sional interface does not exist. Rather, it is replaced by an interface layer with a finite thickness, η. In the limit of a sharp interface (η → 0), however, Eq. (13) approaches Eq. (5). This is ensured by an appropriate choice of the mobility tensor, M αβ, in such a way as to effectively re- strict the diffusion flux to the directions tangential to the interface. Diffusion along the normal direction is only al- lowed in order to keep the phase-field profile stable and vanishes in the limit of a sharp interface. An asymptotic analysis of multiple surface diffusion models is performed in [3]. Here we set, Mαβ = Mαβ (1 − aαβnαβnαβ) . (17) In this way, the tensorial mobility, M αβ, is restricted to the tangential plane of the interface by using the interface normal nαβ. In Eq. (17), Mαβ is the scalar magnitude of M αβ and 1 is the unit tensor. Additionally, the func- tion aαβ interpolates between a purely isotropic mobility 4 (cid:19) (cid:18) η2 π2 ∇φα · ∇φβ φαφβ (cid:40) 1 x > 1 x else. (aαβ = 0) and a pure tangential one (aαβ = 1). Thus, a flux across the interface is generated only if the interface is not in its equilibrium shape, aαβ = limit1 . (18) The operator limit1 restricts the interpolation function aαβ to the interval [0, 1] with limit1 (x) = (19) If the sharp interface limit is not of major interest, it is tempting to use the simpler isotropic mobility tensor, M αβ = Mαβ1 . (20) However, it is shown in [3] that an isotropic mobility ten- sor is less suitable to model the dynamics of surface diffu- sion. Therefore, with the exception of a single test shown in Fig. 3, all the simulations reported here are performed using the non-isotropic mobility tensor, Eq. (17). C. Simulation Details It is noteworthy that the derivatives of the obstacle po- tential, ∂ φα /∂φα and ∂ φβ /∂φβ, are not continuous at φ = 0 and φ = 1. Therefore, a so called bent-cable model [21] is used in order to piecewise interpolate be- tween the linear derivative of the double obstacle and the non-linear interpolation at the minima. The model is implemented inside the open source soft- ware project OpenPhase (www.OpenPhase.de). A finite difference scheme with a 27-point stencil [22] is used for the discretization of the Laplacian in Eq. (15). This high number of stencil points is used in order to avoid numer- ical errors. We also tested the method with a 7-point stencil and the results showed no significant difference here. This observation is corroborated by similar studies with a focus on curvature evaluation [23]. The divergence and the gradient in Eq. (13) are both discretized with a 3-point first order central finite difference scheme. A first order explicit Euler method is used for the discretization of the time derivative. Because phase-boundary diffusion becomes more significant on small length scales, a spacial discretization of ∆x = 10−9m and temporal discretiza- tion of ∆t = 10−13s is considered. The surface/interface energies used in this study are of the order of σ ∼ 1 J/m2, which is typical for metallic systems. The mobility coeffi- cient of Eq. (13) is set to Mαβ = 10−25m2/Js. This value of the interface mobility does not correspond to a real physical system but is necessary to keep the algorithm stable. Results obtained in this work are thus of generic rather than material specific nature. To better reflect this feature, all lengths and times are reported below in units of ∆x and ∆t. If not otherwise stated, the inter- face width is set to η = 10∆x. In order to calculate the right hand side of Eq. (14), three additional boundary grid cells for φ are used around the computation domain. One way to determine these boundary cells is to use pe- riodic boundary conditions. The volume of the phases is in this case conserved up to the machine's roundoff error. III. RESULTS AND DISCUSSION The phase-boundary diffusion model is first tested with regard to the equilibrium interface profile for a spheri- cally symmetric phase-field in Sec. III A. A detailed study of thermal grooving is presented in Sec. III B. Assum- ing that the late time profile obtained from these sim- ulations is a good approximation for equilibrium shape of a three phase boundary, it is examined in Sec. III C whether these simulations are consistent with the von Neumann's triangle relation. As an example for the influ- ence of periodic boundaries, we simulate a 3D tessellation with octahedra in Sec. III D. In Sec. III E, the annealing of nano-clusters on a deformable, initially flat, surface is investigated. The results obtained from these simulations are discussed in the context of the available literature. A. Interface Profile As a very first test, we consider relaxation of the in- terface towards its equilibrium profile, similar to the test performed in [3]. In the case of only two phases and a planar interface, the analytical solution of the interface profile is known at equilibrium (see App. A). We show numerically that the phase-field profile obtained from the planar interface is also a good approximation for the pro- file of a spherically symmetric phase-field. For two phase- fields with φα = 1 − φβ = φ, 1 φeq = 2 sin 1 2 − 1 0 (cid:16) π(r−R) (cid:17) η 2 if r < R − η if R − η else, 2 ≤ r ≤ R + η 2 (21) in which, r is the radial coordinate and R the radius of the sphere. Equation (21) is used as initialization for the phase- field, but with a wider interface width (ηinitial = 2η). Figure 3 shows the value of φ along the center line of the sphere for different time steps. In the current set-up, the phase-field at the initial time, t = 0, is already in a spher- ical shape, but the interface is wider than the equilibrium solution. Thus, diffusion is expected to occur only in the direction normal to the interface. A survey of Fig. 3 re- veals that isotropic and non-isotropic tensorial mobilities result in different relaxation rates towards the equilib- rium profile. This is closely related to the fact that, in the case of a non-isotropic mobility tensor, the matrix- elements are chosen such that the transport is mainly 5 a0 a1 a2 a4 a6 a8 a10 a12 a14 −7.803 10.000 −2.886 8.130 −2.004 3.625 −4.969 5.339 −4.655 ×10−1 ×10−1 ×10−1 ×10−3 ×10−4 ×10−6 ×10−8 ×10−10 ×10−12 TABLE I. Polynomial coefficients of Eq. (22). allowed along the tangential direction, while it is rather restricted along the normal direction. In the present set- up, where the driving force acts only along the normal direction, this leads to a slower diffusion as compared to an isotropic mobility tensor, where the diffusion process is not restricted along any direction. B. An Example of Thermal Grooving Mullins has investigated the mechanisms of surface dif- fusion and evaporation-condensation which lead to the formation of surface grooves in a polycrystal that is heated up to elevated temperatures [2]. An initially flat surface with a grain boundary perpendicular to the sur- face is considered. For surface diffusion, a first order approximation of the surface profile evolution, (cid:88) n=0 (cid:20) x (cid:21)n (Bt) 1 4 + O (cid:0)m2(cid:1) , (22) yM (x, t) = m(Bt) 1 4 an has been obtained for small slopes m and with the pa- rameter B = µσcV 2 m (see [2]). The values for an are given in Tab. I. The profile of Eq. (22) is shape invariant and its amplitude, Bt, can not be changed without stretch- ing it along the x-axis. This solution suggests that the groove profile increases its amplitude over time without changing its shape. In order to examine the analytically obtained results of Mullins, we consider a quasi-two dimensional set-up, where three phase-fields are initialized so that they form a flat surface with an interface perpendicular to the sur- face (Fig. 4a,b). The evolution of the surface profile is shown in Fig. 4c. Two of the three phase-fields repre- sent the solid grains. The third phase-field stands for the surrounding gas phase, φgas. The surface profile is taken as the contour of the gas phase with φgas = 0.5, which is reconstructed with a fourth-order polynomial and a bisection method. This is necessary to obtain the equi- librium angles with a suitable accuracy for the following analysis. Between the initial and equilibrium configurations, the triple junction is moving downwards, causing the inter- faces to bend upwards. Additionally, one can see that the maximum of this bended interface travels outwards. In 6 (a) (b) FIG. 3. Relaxation of the interface profile, across the center of sphere, towards its equilibrium shape with (a) a non-isotropic and (b) an isotropic mobility tensor. For this study, a two dimensional disk with a radius of R = η and an initial interface width of ηinitial = 2η has been initialized with η = 10∆x and ∆x = 10−9m (see Eq. (21)). a private communication to Mullins [2], Hillert pointed out that since the flow of matter immediately beyond the point of inflection is toward the origin, and since the curve has a fixed shape with enlarging size, there must be oscillations of the surface profile about the x- axis. These oscillations are not predicted in the original paper by Mullins [2] who did not exclude this possibility but suspected that it would be difficult to observe these oscillations, due to the decreasing amplitude of the sur- face profile [2]. The results obtained in the seminal work of Mullins, who assumes the shape invariance of the sur- face profile, has been confirmed by numerical solution of the underlying partial differential equations [22, 24]. Recently, the above mentioned non-monotonic surface profile has been reported in experiments in the case of a tungsten polycrystal (see [25] and references therein). As seen in Fig. 4, the results of the current multi-phase- field model for surface diffusion provide an independent numerical evidence for the oscillatory propagation of the surface profile during thermal grooving. These results are also in qualitative agreement with two dimensional calculations based on a variational approach [26]. A comparison of Eq. (22) with the obtained surface profile (see Fig. 4) shows that Eq. (22) is a good ap- proximation for early times of the evolution and near the triple junction. However, our simulation reveals that, the more the wave front propagates, the flatter its profile be- comes in comparison to Eq. (22). A profile similar to the one obtained in this study and the profile obtained by Mullins albeit without oscillations has been experimen- tally observed in [27]. Since the oscillation pattern increases in size with time, it is more probable that they can be observed in the later stages of thermal grooving. One can however, also not fully exclude the possible role of thermal fluctuations in damping the oscillations. In order to investigate this as- pect for the time evolution of the phase-field variable, Eq. (13) shall be extended to a Langevin-type equation including the effect of thermal noise. In addition, on the surface of a polycrystalline body which maintains multi grooving junctions, oscillations interfere with one another. Thus, spacing between the junctions may also play an important role. This feature is not considered in the present simulations as well. A detailed investiga- tion of the effects arising from thermal noise and multiple grooving junctions is left for future work. C. Von Neumann's triangle relation In this section, we benchmark our results versus Young's law at triple junctions. For this purpose, we take the example of thermal grooving and investigate the dependency of the equilibrium groove angle on the grain boundary energy. Here, the equilibrium state is defined as the steady and quasi time independent profile, which establishes during late stages of thermal grooving. In a sharp interface picture, the groove angle is defined as the angle between the left and right tangent lines of the surface profile at the triple junction. In close simi- larity to this, we use the right and left tangents of the corresponding dual phase contour lines near the triple junction. In order to improve numerical accuracy, a high order polynomial interpolation is used to calculate the tangent lines. Since Eq. (10) is derived with the assumption of straight/planar interfaces, the accuracy will depend on the boundary conditions in the simulation box. In or- der to account for this fact, the simulation of thermal grooving has been repeated for different lengths of the computation domain along the x-direction. We find that it is possible to obtain a relation be- −2−10120.000.200.400.600.801.00transversecoordinaterηPhasefieldvalueφt=0t=1.0×104∆tt=3.5×104∆tφeq−2−10120.000.200.400.600.801.00transversecoordinaterηPhasefieldvalueφt=0t=0.5×104∆tt=1.5×104∆tφeq7 (a) (b) (c) FIG. 4. Evolution of a groove on a surface at a grain boundary by surface diffusion at (a) t = 0 and (b) t = 3 × 104∆t. Simulation results are visualized by the isosurface of the phase-field for the gas phase (φgas = 0.5) and the two isosurfaces of the phase-fields representing the two solid grains (φα = 0.5 and φβ = 0.5). The surface energies of both grains, σsurface, and the grain boundary energy, σGB, are identical with σsurface = σGB = 1 J/m2. (c) The upscaled surface profile (φgas = 0.5) with the grain boundary at x = 0. The surface profile is a reconstructed phase-field contour of φ = 0.5, obtained via a fourth-order polynomial and a bisection method. For comparison, the solid lines show the analytic solution of Mullins, Eq. (22), with m = 0.057454 and B = 0.00588. The computation domain is discretized with (a,b) 128 × 64 × 32 and (c) 512 × 64 × 3 lattice nodes. Periodic boundary condition is applied in the x and z-directions. The value of the phase-field is fixed at the boundary normal to the y-direction. Note that there is no z-dependence in this set-up, so that the obtained results essentially correspond to a two dimensional problem. tween the size of the computation domain and the equi- librium angle (see Fig. 5a). With this knowledge, one can minimize numerical errors and estimate the effect of the boundary. If one assumes that the surface energy, σsurface, is the same for both grains, then the groove an- gle, θ, can be calculated using Eq. (10), (cid:32)(cid:112) θ = π − arctan 4σ2 Surface − σ2 σGB (cid:33) GB . (23) The angles calculated by the simulations with the present method, and those predicted by Eq. (23), are listed in Tab. II. The same data is plotted in Fig. 5b, highlighting the close agreement between numerically obtained results and Eq. (23). D. Quadruple Junctions Although the above simulations of thermal grooving are performed in three dimensions, they are essentially equivalent to a two dimensional situation due to the cylindrical symmetry of the set-up (no z-dependence). For multi-grain (multi-phase) systems in 3D, however, vertex points (quadruple and higher order junctions) may also exist in which more than three grains meet. In order 0102030405060708090100110120130140150−0.200.000.200.40Positionx[∆x]Positiony[∆x]t=3.0×104∆tt=6.0×104∆tt=1.2×105∆tt=2.4×105∆tt=4.8×105∆tt=9.6×105∆tyM8 (a) (b) FIG. 5. The dependency of the contact angle, between the surface and the grain-boundary at a triple junction, on the size of the computation domain is shown in (a). The equilibrium contact angle for different grain boundary energies σGB can be seen in (b). The surface energies are constant with σsurface = 1J/m2. The error bars in (b) are estimated on the basis of (a). σGB (J/m2) 0.25 0.50 0.75 1.00 1.25 1.50 θvon Neumann (◦) 97.2 104.5 112.0 120.0 128.7 138.6 θsimulation (◦) 96.4 103.4 109.6 118.0 125.3 134.7 TABLE II. The dependency of the contact angle between the surface of two grains on the grain boundary energy, σGB. The surface energy is kept constant at σsurface = 1 J/m2. Simula- tion results are listed together with the prediction of the von Neumann's relation, Eq. (23). to investigate this type of scenario, a multi-grain set-up is constructed by dividing the simulation domain into eight cubes such that only triple lines and quadruple points form between the grains (see Fig. 6a). By surface mini- mization, one can see that the grains arrange themselves into truncated octahedra (see Fig. 6c). The equilibrium angle between the triple lines at a quadruple junction is either 90◦, when the lines are connected by a square plane, or 120◦, when the lines are connected by a hexago- nal plane. We find that the dihedral angle at a triple line between two hexagonal planes of truncated octahedron is roughly 109◦ and the angle between a hexagonal and a square plane is ≈ 125◦. These are well in line with the values of 109◦28 51" expected from geo- metrical consideration. However, from a generalization of the von Neumann's triangle relation to quadruple junc- tions, one would expect that the system does not form truncated octahedra and that all dihedral angles have the same value. The difference between the simulation re- sults and the prediction of a generalized von Neumann's triangle relation for quadruple junctions is probably due to the conservation constraint in the periodic set-up of 16" and 125◦15 (cid:48) (cid:48) simulations. It is quite simple, however, to show that the sum of forces which act alongside the triple lines on a quadruple point is zero (see App. B). E. Droplets on a Deformable Surface An important application of the phase-boundary dif- fusion model proposed in this work is the simulation of nano-clusters on free surfaces. In particular, the model allows to study the entrenching of nano-clusters, reported in recent experiments [10]. In order to study the influence of the interface en- ergy on the equilibrium contact angles between the nano- clusters and the free surface, we have placed a spherical droplet on a surface and observed its wetting dynamics. The equilibrium configurations for different interface en- ergies are shown in Fig. 7. A complete wetting of the surface, beyond Eq. (10), is visible in Fig. 7c. The equilibrium configurations shown in Fig. 7 do not resemble the geometry of the nano-clusters observed in [10] where elevated perimeters have been observed which surround the 'crater-like' entrenchments of the nano-clusters. This perimeter can be explained as in- termediate state of the entrenching process, in a sense that the entire system is not in equilibrium. This in- terpretation is confirmed by our simulations, shown in Fig. 8, where four initially connected droplets on surface are simulated with identical interface energies. One can clearly see that the droplets separate from each other and entrench into the surface. The perimeter is visualized in Fig. 8b. Thus, the present model is not only able to re- produce experimentally observed results but also allows to uncover the transient character of the elevated perime- ter in the entrenching process. A full investigation of the multi-nano-clusters on free surfaces is in progress. 0100200300400500600110115120SizeofcomputationdomainLxContactangleθ[deg]θnumericθanalytic00.20.40.60.811.21.41.61.82100120140160180GrainboundaryenergyσGBContactangleθ[deg]θnumericθanalytic9 (a) (b) (c) FIG. 6. The relaxation into a tessellation with truncated octahedra at (a) t = 5×103∆t, (b) t = 5×104∆t, and (c) t = 9×105∆t. The computation domain is divided into eight cubic phase-fields with the same volume. Periodic boundary condition is used along all the coordinate directions x, y and z. The cubic phases are shifted in a way that there are only triple lines and quadruple junctions. The dual interfaces are shown with transparent greyish planes and the triple lines with black lines. Additionally, one phase is highlighted with less transparent planes. The domain is discretized into 64 × 64 × 64 lattice nodes. (a) (b) (c) FIG. 7. The configuration of multiple solid droplets on a deformable surface after t = 106∆t is shown for different interface energies with (a) σαγ = 0.5 J/m2, σβγ = 0.75 J/m2, and σαβ = 1 J/m2; (b) σαγ = 0.5 J/m2, σβγ = 1 J/m2, and σαβ = 0.75 J/m2; and (c) σαγ = 0.5 J/m2, σβγ = 10 J/m2, and σαβ = 0.5 J/m2. The computation domain is discretized by 64 × 64 lattice nodes, periodic boundary condition in x-direction and fixed boundary conditions in y-direction are applied. IV. SUMMARY A conserved multi-phase-field model is proposed to de- scribe the physical effect of surface/phase-boundary dif- fusion. Starting from the non-conserved multi-phase-field model [8], a pairwise continuity equation is proposed for temporal evolution of the surface/phase-boundaries by curvature-driven diffusion. The model is applied to investigate the dynamics of the surface profile during thermal grooving. Simula- tion results indicate that the shape invariant solution of Mullins [2] is a good approximation to the early stages of the thermal grooving process. Moreover, new evidence is provided for the oscillations of the surface profile, first proposed by Hillert. Additional simulations of thermal grooving at variable interface energies are then performed in order to gain further confidence on the reliability of the present approach. This is achieved by a check of the late time profiles obtained from these simulations against the von Neumann's triangle relation for equilibrium angles at a triple junction. As a forecast on future applications of the proposed model, the annealing dynamics of nano-clusters on ini- tially flat surfaces is investigated. The reported crater- like structure with an elevated perimeter [10] is found to be a transient non-equilibrium state during nano-cluster annealing, thus shedding light onto this complex process from a dynamic perspective. The proposed multi-phase- field method is generic and can be used to study any type of surface and phase-boundary phenomena which contains conserved fields. 1.21.622.42.81.03.0Interfacesβγα1.21.622.42.81.03.0Interfacesβγα1.21.622.42.81.03.0Interfacesβγα10 (a) (b) FIG. 8. The entrenching of multiple nano-clusters α into a deformable surface β by surface/phase-boundary diffusion with (a) t = 0, (b) t = 5 × 103 ∆t, and (c) t = 1.5 × 105 ∆t. The interface energies are set to σαβ = 1.4 J/m2, σαγ = 1.4 J/m2 and σβγ = 2 J/m2 (γ is the phase surrounding the phases α and β). One of the four clusters has been removed in the plot in order to show the base area which is colored according to its curvature. The simulation domain is discretized into 128× 128× 64 grid cells. (c) Appendix A: Static Equilibrium Solution In the case of two phase-fields with φα = 1 − φβ = φ, the generalized diffusion potential introduced in Eq. (14) simplifies to (cid:20) η2 (cid:21) π2∇2φ + 2φ − 1 ψ = 8σαβ η , (A1) where we assumed φ (1 − φ) ∈ [0, 1] so that the partial derivate of the obstacle potential (2nd term in Eq. (15b)) reduces to ∂ φ (1 − φ) /∂φ = 2φ − 1. Using this information, one finds that the equilibrium solution of Eq. (13) for a planar interface, situated at x = 0 between two phases, is identical to Eq. (21) with the difference that the radial distance r shall be replaced by the Cartesian coordinate along the direction normal to the interface, say x. Appendix B: Forces at a quadruple junction of a tessellation with truncated octahedra If one considers a quadruple in a tessellation with trun- cated octahedra as in Fig. 6c, one sees that four triple lines are connected to that quadruple junction. The con- figuration of the phase-fields at each of these triple lines is the same. So it is reasonable to assume that the mag- nitude, m, of the forces alongside the triple line is the same and only their orientation differs. These forces can be written as follows: F3 = F4 = m √2 m √2 (0, 1,−1)T (0, 1, 1)T . 11 (B1c) (B1d) One can easily check that the sum of F 1, F 2, F 3 and F 4 is zero, and the angles between the forces are either 90◦ or 120◦. ACKNOWLEDGMENTS F1 = F2 = m √2 m √2 (−1,−1, 0)T (1,−1, 0)T (B1a) (B1b) Financial support by the German Research Foundation DFG within Priority Program SPP1713 under the grants DA 1655/1-1, STE 1116/20-1 and the DFG-project VA 205/17-1 is gratefully acknowledged. [1] C. Herring, in Fundamental Contributions to the Con- tinuum Theory of Evolving Phase Interfaces in Solids (Springer, 1999) pp. 33 -- 69. [2] W. W. Mullins, Journal of Applied Physics 28, 333 (1957). [3] C. Gugenberger, R. Spatschek, and K. Kassner, Physical Review E 78, 016703 (2008). [4] G. Caginalp, Physical Review B 38, 789 (1988). [5] G. Caginalp, IMA Journal of Applied Mathematics 44, 77 (1990). [6] A. Ratz, A. Ribalta, and A. Voigt, Journal of Compu- tational Physics 214, 187 (2006). British Columbia, Canada (2002). [22] W. F. Spotz and G. F. Carey, in Proceedings of the Thrid International Conference on Spectral and High Or- der Methods (Houston Journal of Mathematics, Univer- sity of Houston, Houston, Texas, USA, 1995) pp. 397 -- 408. [23] S. Vakili, I. Steinbach, F. Varnik, "On the numerical eval- uation of local curvature for diffuse interface models of microstructure evolution", Procedia Computer Science (in press, 2017). [24] H. Zhang and H. Wong, Acta Materialia 50, 1983 (2002). [25] P. Sachenko, J. Schneibel, and W. Zhang, Scripta Ma- [7] D.-H. Yeon, P.-R. Cha, and M. Grant, Acta Materialia terialia 50, 1253 (2004). 54, 1623 (2006). [26] K. Hackl, F. D. Fischer, K. Klevakina, J. Renner, and [8] I. Steinbach, Modelling and Simulation in Materials Sci- J. Svoboda, Acta Materialia 61, 1581 (2013). [27] T. Gladstone, J. Moore, A. Wilkinson, and C. Grovenor, IEEE Transactions on Appiled Superconductivity 11, 2923 (2001). ence and Engineering 17, 073001 (2009). [9] I. Steinbach, Annual Review of Materials Research 43, 89 (2013). [10] U. Kohler, M. Kroll, T. Lober, A. Birkner, V. Schott, and C. Woll, Physica Status Solidi (b) 250, 1222 (2013). [11] I. Steinbach, L. Zhang, and M. Plapp, Acta Materialia 60, 2689 (2012). [12] L. Zhang and I. Steinbach, Acta Materialia 60, 2702 (2012). [13] I. Steinbach and M. Apel, Physica D: Nonlinear Phenom- ena 217, 153 (2006). [14] J. Mosler, O. Shchyglo, and H. Montazer Hojjat, Journal of the Mechanics and Physics of Solids 68, 251 (2014). [15] R. D. Kamachali, E. Borukhovich, N. Hatcher, and I. Steinbach, Modelling and Simulation in Materials Sci- ence and Engineering 22, 034003 (2014). [16] R. Darvishi Kamachali and C. Schwarze, Computational Materials Science 130, 292 (2017). [17] F. E. Neumann, Vol. 7 (BG Teubner, 1894). [18] G. R. Lester, Journal of Colloid Science 16, 315 (1961). [19] M. E. R. Shanahan, Journal of Physics D: Applied Physics 20, 945 (1987). [20] J. S. Rowlinson and B. Widom, Molecular Theory of Capillarity, The International Series of Monographs on Chemistry (Clarendon Press, Oxford, 1982). [21] G. S. Chiu, Bent-cable regression for assessing abrupt- ness of change, PhD disseration, Simon Fraser University, 

1302.5716

1

1302

Design principles for HgTe based topological insulator devices

The topological insulator properties of CdTe/HgTe/CdTe quantum wells are theoretically studied. The CdTe/HgTe/CdTe quantum well behaves as a topological insulator beyond a critical well width dimension. It is shown that if the barrier(CdTe) and well-region(HgTe) are altered by replacing them with the alloy Cd$_{x}% $Hg$_{1-x}$Te of various stoichiometries, the critical width can be changed.The critical quantum well width is shown to depend on temperature, applied stress, growth directions and external electric fields. Based on these results, a novel device concept is proposed that allows to switch between a normal semiconducting and topological insulator state through application of moderate external electric fields.

cond-mat.mes-hall

cond-mat

Design principles for HgTe based Topological Insulator Devices Parijat Sengupta,1, ∗ Tillmann Kubis,1 Yaohua Tan,1 Michael Povolotskyi,1 and Gerhard Klimeck1 1Dept of Electrical and Computer Engineering, Purdue University, West Lafayette, IN, 47907 The topological insulator properties of CdTe/HgTe/CdTe quantum wells are theoretically studied. The CdTe/HgTe/CdTe quantum well behaves as a topological insulator beyond a critical well width dimension. It is shown that if the barrier(CdTe) and well-region(HgTe) are altered by replacing them with the alloy CdxHg1−xTe of various stoichiometries, the critical width can be changed.The critical quantum well width is shown to depend on temperature, applied stress, growth directions and external electric fields. Based on these results, a novel device concept is proposed that allows to switch between a normal semiconducting and topological insulator state through application of moderate external electric fields. PACS numbers: Keywords: I. Introduction An insulator is conventionally defined as a material that does not conduct electricity. In most insulators the lack of electric conduction is explained using its band- structure properties.The band theory predicts that an insulator has an energy gap separating the conduction and valence bands. As a result of this finite band-gap there are no electronic states to support the flow of cur- rent. Recently, materials that have an energy gap in bulk but possess gapless states bound to the sample sur- face or edge have been theoretically predicted and exper- imentally observed. [1] These states, in a time reversal invariant system are protected against perturbation and nonmagnetic disorder. [2–5] Materials that support such states are known as topological insulators (TI). Examples of materials with such properties include Bi2Te3, Bi2Se3, BixSb1−x alloys, and CdTe/HgTe/CdTe quantum wells. Bi2Te3,Bi2Se3, and BixSb1−x belong to the class of 3- D topological insulators (3D-TI) and host bound states on their surface. [6–8] CdTe-HgTe-CdTe quantum wells, which were the first predicted TIs are 2-D topological in- sulators (2-D TI). Unlike their 3D counterpart, they pos- sess bound states at the edge of the quantum well. [9–11] These conducting surface and edge states develop at the boundary between two insulators, where one is normal (NI) and the other of inverted band ordering. The sur- face states, which are subject to the details of the band properties of each involved material can mutually influ- ence each other. It is therefore essential to theoretically study the surface states under various conditions. This work proposes ways that can efficiently invert the band profile of a CdTe/HgTe/CdTe heterostructure and create bound edge states. Specifically, the transition from an NI to a TI through external adiabatic parameters, ad- justable lattice constants, or modulation of the electron- hole band coupling is the underlying theme. The pa- per is organized as follows: In Section II, details of the CdTe/HgTe/CdTe structure and method to compute the ∗Electronic address: [email protected] energy dispersion are described. Section III discusses var- ious inverted band structures, device and material condi- tions that can lead to topologically protected conducting surface states. The concept of creating a switch from a topological insulator is developed here. The key results discussed in the paper are summarized in Section IV. II. Materials and Methods An HgTe quantum well flanked by CdTe barriers has been shown to have edge states with topological insu- lator properties. [12] TI behaviour is possible because CdTe is a normal insulator and is placed in contact with an inverted insulator HgTe. A representative sketch of the device is shown in Fig.1. CdTe is a wide band gap semiconductor (Eg = 1.606 eV) with positive energy gap (NI) and a small lattice mismatch of 0.5% with HgTe. CdTe, because of similar lattice constants is chosen as the barrier for the HgTe well region though in principle any normal ordered material would suffice. The normal valence and conduction band are reversed in their ener- getic order in HgTe as indicated in Fig. 1 and explained in the next paragraph. FIG. 1: Sketch of a CdTe/HgTe/CdTe quantum well het- erostructure. The lowest conduction band (CB) state is la- beled with E1 and the highest valence band (VB) state with H1. The inversion of bands for the CdTe/HgTe/CdTe het- erostructure is achieved through the HgTe component. Both CdTe and HgTe belong to the zinc blende (ZB) structure with Td point group symmetry. The highest valence and lowest conduction band is made up of p and s orbitals respectively. A normal band order at Γ has low- est conduction band (j = 1/2) with Γ6 symmetry above the top of the valence bands (j = 3/2) with Γ8 symme- try. The Γ6 state has s-type symmetry and the Γ8 state has p-type symmetry. In a normal ordered material Γ6 state is energetically higher than the Γ8 state. This or- der is reversed in bulk HgTe at the Γ point due to the high spin-orbit coupling and a significant Darwin term contribution. [13] The strong spin orbit coupling pushes the valence bands upwards while the Darwin term shifts the s-type conduction band down. The Darwin term can only influence the s-type bands. [14] The combined ef- fect of spin orbit coupling and Darwin term yields an inverted band order at the Γpoint which flips the order of the high-symmetry Γ6 and Γ8 points for HgTe. [15]The energy gap at Γ which is defined as Eg = E(Γ6) − E(Γ8), (1) therefore turns out to be negative for HgTe. The normal and inverted band structures of CdTe and HgTe are illustrated in Fig. 2(a) and Fig. 2(b) respectively. FIG. 2: Bulk band structure of CdTe (a) and HgTe(b). The ordering of the conduction and valence bands near the band gap at the Γ point in HgTe (Fig. 2b) is opposite to the one in CdTe (Fig. 2a). In HgTe, the hole state Γ8 is above the electron state Γ6. In this work, electronic properties of the h001i CdTe/HgTe/CdTe heterostructure are calculated within an 8-band k.p framework that includes a linear coupling between conduction and valence bands. [16, 17] In the calculations presented, the z-axis is normal to the het- erostructure and is also the confinement direction. The valence band edge Ev, Luttinger parameters, and other related material properties are collected in Table I. [18] 2 The boundary conditions are imposed by setting the wave function to zero at the edge of the device. Strain is added to the electronic Hamiltonian using deformation poten- tials defined in the Bir-Pikus method. [19, 20] TABLE I: 8-band k.p parameters for CdTe and HgTe. Ev, Eg, Pcv, and Vso are in units of eV. The remaining Luttinger parameters are dimensionless constants and the effective mass is in units of the free electron mass. Material Ev γ2 γ3 γ1 Pcv Vso -0.27 5.372 1.671 1.981 0.11 1.606 18.8 0.91 0.0 -16.08 -10.6 -8.8 -0.031 -0.303 18.8 1.08 m∗ Eg CdTe HgTe III. Results and Discussion that report critical widthExperiments Comparison with experiment: band gap a and CdTe/HgTe/CdTe quantum well heterostructure with a well width under 6.3 nm exhibits a normal band order with positive Eg. [21, 22]. The calculation of the present work confirms that the conduction states at Γ are indeed located above the valence states and the energy gap is positive (Fig. 4(a)). All band structure parameters used to reproduce the experimental observation were valid at 0 K. When the well width is exactly 6.3 nm, a Dirac system [23] is formed in the volume of the device (Fig. 3). FIG. 3: Band structure of HgTe quantum well of thickness 6.3 nm. At this width, the lowest conduction band (E1) and highest valence band (H1) at the Γ point are equal. Beyond this critical well width of 6.3 nm, the het- erostructure has its bands fully inverted. The band pro- file has a reverse ordering of the s-type and p-type or- bitals (Fig. 4(c)) and Eg < 0. Accordingly, a nano-ribbon of width 100.0 nm formed by quantizing the quantum well in its in-plane direction has a positive band gap (as shown in Fig. 4b). Similarly, a nanoribbon of width 100.0 nm constructed out of an inverted quantum well possesses gap-less TI edge states. The band structure of this situation is illustrated in Fig. 4(d). 3 FIG. 4: Bandstructure of a HgTe quantum well of thickness 5.5 nm(a). A HgTe nano-ribbon formed out of this quantum well of thickness 5.5 nm and height of 100 nm shows a posi- tive band gap. Fig. 4c shows the bandstructure of an inverted quantum well of thickness 10.0 nm. The corresponding quan- tum wire has a linearly dispersing (Dirac-cone) edge states (d). The corresponding absolute value of the squared edge- state wave functions is plotted in Fig. 5. The absolute value of the wave functions for the two edge states is maximum at the edge and gradually decay in to the bulk. This establishes that they belong exclusively to the edge states. In conclusion, the band-gap closing Dirac cone shown in Fig. 3 marks the transition from a positive band-gap to a negative one. Band nature at finite momenta: The inversion of bands in the volume of the well is necessary for edge states with topological insulator behavior. It is impor- tant to note however, that the process of inversion hap- pens only at the Γ point. In the inverted dispersion plot (Fig. 4c), for momenta different from the Γ point, the band labeled with "H1" progresses from p to s-type. Similarly the band labeled with "E1" changes charac- ter from s to p. Both the bands, at a finite momentum acquire atomic orbital characteristics associated with a normally ordered set of bands. TI behavior is therefore restricted to a special set of momentum points where the band structure is inverted. These set of points are collectively called the time-reversal-invariant-momentum (TRIM) points. [24] Well thickness continuously tunes the TI prop- erties: With increasing well width, the band gap de- creases continuously until the HgTe well thickness reaches 6.3 nm (see Fig. 6). This is due to the diminishing confinement of the well's s and p-type bands. For well thicker than 6.3 nm, the confinement is small enough such FIG. 5: Absolute value of the wave functions ψ2 of the two edge-states of Fig. 4d. that the inverted band ordering of HgTe is restored and the absolute value of the negative band gap is increasing (see Fig. 6). At a HgTe well width of 8.2 nm, the s-type band drops even below the second confined p-type state. This re- ordering of bands with well thickness is summarized in Table II and illustrated in Fig. 6. TABLE II: Orbital character of the top most valence band and lowest conduction band in CdTe-HgTe-CdTe heterostructure depending on the well width dQW . The critical well width dc is the equal to 6.3 nm. HgTe Well thickness Highest Val.Band Lowest Cond.Band dQW < dC 8.2 nm > dQW > dC dQW > 8.2nm p-type s-type p-type s-type p-type p-type A. Stoichiometric and Temperature Control of Critical Width In the previous sections, it has been shown that the effective band gap of the CdTe/HgTe/CdTe quantum well depends on the confinement and consequently on the band gap difference of the well and barrier materials. Both, alloying and temperature are known to influence the effective band gap. The band gap of CdxHg1−xTe as a function of temperature [25] T and stoichiometry x is given by Eg = −304 + 0.63T 2 11 + T (1 − 2x) + 1858x + 54x2. (2) A plot for the band-gap variation for the CdxHg1−xTe alloy is given in Fig. 7. When the quantum well material of the orig- inal CdTe/HgTe/CdTe structure is substituted by CdxHg1−xTe alloy, the critical width becomes temper- ature and x dependent. This is shown in Fig. 5 (a). Remarkably, all critical widths are equal or larger than the intrinsic critical width of 6.3 nm. Higher concentra- tion of CdTe in the quantum well reduces the Darwin 4 FIG. 8: Critical widths inverted band struc- tures of CdTe/Cd1−xHgxTe/CdTe quantum wells (a) and CdxHg1−xTe/HgTe/CdxHg1−xTe quantum wells (b) as a function of temperature and stoichiometry x. to get Absolute value of 6: a FIG. CdTe/HgTe/CdTe quantum well as a function of the well width. Well widths larger than 6.3 nm produce inverted band structures and can be exploited for topological insulator devices. the band gap of FIG. 7: Calculated band gap of bulk CdxHg1−xTe as a func- tion of stoichiometry and temperature. At x =0, the bulk band gap of HgTe (−0.303 eV) is reproduced. contribution from HgTe. Therefore, the band inversion requires a wider HgTe region. Alternatively, replacing the barrier material with CdxHg1−xTe also allows tuning the confinement and con- sequently the critical width. It is shown in Fig. 7(b) that this replacement yields critical widths smaller then the intrinsic 6.3 nm if the temperature is allowed to at- tain values below 100 K. For a Cd molar concentration of x = 0.68 and T = 0 K, the critical width dropped to 4.4 nm. This is due to the enhanced Darwin contribution to the electronic properties with increased Hg content. 5 TABLE III: The optimal tensile stress and growth conditions for CdTe/HgTe/CdTe quantum wells to achieve the least (L), highest (H) and intermediate (I) critical width, respectively. Growth Axis h001i h110i h111i Tensile uniaxial stress h001i h110i h111i L L H H H L I I I B. Critical widths under different growth conditions Apart from alloy stoichiometry, the confinement also depends on the well and barrier masses. A way to tune these effective confinement masses is by growing the quantum well in different directions. The different masses then give different effective well confinement and accordingly different critical widths. This dependence is illustrated in Fig. 9. It shows the critical width of the CdTe/HgTe/CdTe quantum well in a sequence of growth directions. The critical widths of the h111i and h110i growth directions are 5.52 nm and 5.72 nm respectively. Both these values are smaller than the h001i grown quantum well critical width of 6.3 nm. FIG. 9: Critical widths of CdTe/HgTe/CdTe heterostruc- tures grown along hN 11i direction as a function of N (a). The bandgap closing for h100i, h110i, and h111i grown CdTe/HgTe/CdTe at different well widths is shown in (b). Band gap closing at different well dimensions give the corre- sponding critical width. Alternatively, uniaxial stress can also tune the ef- fective confinement masses. As representative cases, CdTe/HgTe/CdTe quantum wells were grown along h001i, h110i, and h111i directions. Each quantum well was then subjected to uniaxial stress along h001i, h110i, and h111i directions. Uniaxial stress along h001i, h110i, and h111i was employed on three sets of CdTe/HgTe/CdTe quantum wells grown along h001i, h110i, and h111i. The behaviour of the critical width for each case is shown in Fig. 10. The ideal stress orientation for each growth direction is summarized in Tables III and IV. C. Application of an external electric field The application of an external electric field changes the confinement and the band properties of the well states. In particular, the Rashba (structural inversion asymmetry) FIG. 10: Critical widths of CdTe/HgTe/CdTe heterostruc- tures grown along h111i (a), h110i (b), and h001i (c) direc- tion with uniaxial stress applied along h111i (solid), h110i (dashed) and h001i (dash-dotted) direction. Key observa- tions are summarized in Table III and table IV. effect gets enhanced by electric fields in growth direc- tion. [26] Figure 11 shows the critical width as function of the external electric field applied in the growth di- rection. The critical width decays with increasing field for all considered temperatures. The Rashba effect that supports the band inversion of HgTe gets increased by the electric field. Consequently, smaller well widths are required to invert the CdTe/HgTe/CdTe quantum well band structure when external electric fields are present. Since external electric fields can tune the critical width, the concept of a TI-switch is obvious: A CdTe/HgTe/CdTe quantum well with a well width that is close, but below the critical width can be switched TABLE IV: The same list of conditions as in Table III but under compressive stress. Compressive uniaxial stress Growth Axis h001i h110i h111i h001i h110i h111i H H L L L H I I I 6 circuit element in a fast digital environment. When the bandgap is closed and TI properties are turned on, a high Fermi velocity for the carriers, (which is an essential attribute of TIs) on surface is able to transmit an electric signal faster than a conventional inter-connect. A seamless transition from a topological insulator to normal insulator using an external electric field as demonstrated above and shown in Fig. 12 enables it to forbid an easy passage of charge/electric signal. A normal insulator with a finite band gap will behave as an open circuit element. FIG. 11: Critical width for CdTe/HgTe/CdTe quantum wells with varying strength of external electric fields in growth di- rection. by electric fields between normal and inverted band or- der. Such a switching band structure is expected to yield significant changes in the surface conductance, due to the unique transport properties of topological insulator states. A first prototype of such a switch can be ob- served in Fig. 12 which shows effective band gaps of CdTe/HgTe/CdTe quantum wells for various well thick- nesses under externally applied electric fields. Within the plotted range of electric field magnitude, the CdTe/HgTe/CdTe quantum well with a width of 6.0 nm switches between normal and inverted band ordering. It is worth mentioning that this switching behavior can be observed in CdTe/HgTe/CdTe quantum wells grown in h001i and h111i direction. Such topological insulator based devices under an external electric field can be employed to act as a FIG. 12: Effective band gap of CdTe/HgTe/CdTe quantum wells of different well thicknesses as a function of applied elec- tric field in growth direction. The dashed line depicts the delimiter between normal and inverted band structures. IV. Conclusion The present work investigates the conditions under which band inversion can occur in a CdTe/HgTe/CdTe quantum well heterostructure. It is shown that this band inversion is essential for topological insulator properties. In agreement with experimental results, it is found that the HgTe quantum well has to be thicker than 6.3 nm to exhibit topological insulator properties. It is exam- ined in detail how the critical width depends on vari- ous device parameters such as the growth direction, alloy stoichiometry, temperature, uniaxial stress, and external electric fields. In particular the external fields allow to switch the topological insulator properties of h001i grown CdTe/HgTe/CdTe quantum wells. This result proposes a new class of switching devices. [1] M. Z. Hasan and C. L. Kane, Rev. Mod. Phys. 82, 3045 [4] C. L. Kane and E. J. Mele, Phys. Rev. Lett. 95, 146802 (2010). (2005). [2] L. Fu and C. L. Kane, Phys. Rev. B 76, 045302 (2007). [3] R. Roy, Phys. Rev. B 79, 195322 (2009). [5] S. Murakami, New Journal of Physics 9, 356 (2007). [6] H. Zhang, C. Liu, X. Qi, X. Dai, Z. Fang, and S. Zhang, 7 Nature Physics 5, 438 (2009). [20] G. Wu and T. McGill, Applied Physics Letters 47, 634 [7] 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). [8] Y. Chen, J. Analytis, J. Chu, Z. Liu, S. Mo, X. Qi, H. Zhang, D. Lu, X. Dai, Z. Fang, et al., Science 325, 178 (2009). [9] M. Konig, S. Wiedmann, C. Brune, A. Roth, H. Buh- mann, L. Molenkamp, X. Qi, and S. Zhang, Science 318, 766 (2007). (1985). [21] J. Lu, W. Shan, H. Lu, and S. Shen, New Journal of Physics 13, 103016 (2011). [22] C. Brune, A. Roth, H. Buhmann, E. Hankiewicz, L. Molenkamp, J. Maciejko, X. Qi, and S. Zhang, Arxiv preprint arXiv:1107.0585 (2011). [23] M. Konig, H. Buhmann, L. Molenkamp, T. Hughes, and S. Zhang, Arxiv preprint C. Liu, X. Qi, arXiv:0801.0901 (2008). [10] B. Bernevig, T. Hughes, and S. Zhang, Science 314, 1757 [24] J. Teo, L. Fu, and C. Kane, Physical Review B 78, 045426 (2006). (2008). [11] A. Roth, C. Brune, H. Buhmann, L. Molenkamp, J. Ma- ciejko, X. Qi, and S. Zhang, Science 325, 294 (2009). [12] M. Buttiker, Science 325, 278 (2009). [13] N. Cade and P. Lee, Solid state communications 56, 637 (1985). [14] I. Tsidil'kovskii, Band Structure of Semiconductors (Pergamon Press, 1982), ISBN 9780080216577. [15] L. Molenkamp, (Private Communication). [16] J. Maciejko, T. Hughes, and S. Zhang, Annu. Rev. Con- dens. Matter Phys. 2, 31 (2011). [17] P. Sengupta, S. Lee, S. Steiger, H. Ryu, and G. Klimeck, in MRS Proceedings (Cambridge Univ Press, 2011), vol. 1370. [18] E. G. Novik, A. Pfeuffer-Jeschke, T. Jungwirth, V. La- tussek, C. R. Becker, G. Landwehr, H. Buhmann, and L. W. Molenkamp, Phys. Rev. B 72, 035321 (2005). [19] J. Schulman and Y. Chang, Physical Review B 33, 2594 (1986). [25] S. Krishnamurthy, A. Chen, A. Sher, and M. Van Schilf- gaarde, Journal of electronic materials 24, 1121 (1995). [26] X. Zhang, A. Pfeuffer-Jeschke, K. Ortner, V. Hock, H. Buhmann, C. Becker, and G. Landwehr, Physical Re- view B 63, 245305 (2001). Acknowledgments Computational resources from nanoHUB.org and sup- port by National Science Foundation (NSF) (Grant Nos. EEC-0228390, OCI-0749140) are acknowledged. This work was also supported by the Semiconductor Research Corporation's (SRC) Nanoelectronics Research Initia- tive and National Institute of Standards & Technology through the Midwest Institute for Nanoelectronics Dis- covery (MIND), SRC Task 2141, and Intel Corporation. 

1502.02248

1

1502

Quasiperiodic AlGaAs superlattices for neuromorphic networks and nonlinear control systems

The application of quasiperiodic AlGaAs superlattices as a nonlinear element of the FitzHugh-Nagumo neuromorphic network is proposed and theoretically investigated on the example of Fibonacci and figurate superlattices. The sequences of symbols for the figurate superlattices were produced by decomposition of the Fibonacci superlattices' symbolic sequences. A length of each segment of the decomposition was equal to the corresponding figurate number. It is shown that a nonlinear network based upon Fibonacci and figurate superlattices provides better parallel filtration of a half-tone picture than a network based upon traditional diodes which have cubic voltage-current characteristics. It was found that the figurate superlattice F011(1) as a nonlinear network's element provides the filtration error almost twice less than the conventional "cubic" diode. These advantages are explained by a wavelike shape of the decreasing part of the quasiperiodic superlattice's voltage-current characteristic, which leads to multistability of the network's cell. This multistability promises new interesting nonlinear dynamical phenomena. A variety of wavy forms of voltage-current characteristics opens up new interesting possibilities for quasiperiodic superlattices and especially for figurate superlattices in many areas - from nervous system modeling to nonlinear control systems development

cond-mat.mes-hall

cond-mat

Quasiperiodic AlGaAs superlattices for neuromorphic networks and nonlinear control systems K.V.Malysheva) Electronics and Laser Technology department, Bauman Moscow State Technical University, Moscow, 105005, Russia [email protected] The application of quasiperiodic AlGaAs superlattices as a nonlinear element of the FitzHugh– Nagumo neuromorphic network is proposed and theoretically investigated on the example of Fibonacci and figurate superlattices. The sequences of symbols for the figurate superlattices were produced by decomposition of the Fibonacci superlattices’ symbolic sequences. A length of each segment of the decomposition was equal to the corresponding figurate number. It is shown that a nonlinear network based upon Fibonacci and figurate superlattices provides better parallel filtration of a half-tone picture then a network based upon traditional diodes which have cubic voltage-current characteristics. It was found that the figurate superlattice F0 11(1) as a nonlinear network’s element provides the filtration error almost twice less than the conventional “cubic” diode. These advantages are explained by a wavelike shape of the decreasing part of the quasiperiodic superlattice’s voltage-current characteristic, which leads to multistability of the network’s cell. This multistability promises new interesting nonlinear dynamical phenomena. A variety of wavy forms of voltage-current characteristics opens up new interesting possibilities for quasiperiodic superlattices and especially for figurate superlattices in many areas - from nervous system modeling to nonlinear control systems development. I. INTRODUCTION In recent years quasiperiodic superlattices (SLs) of finite length were attracted increasing attention due to their interesting and unexpected fundamental physical properties as well as their applications in nanoelectronics [1]. For example, dielectric quasiperiodic SLs are promising for  nanophotonics (see, e.g., review [2]). Piezoelectric quasiperiodic SLs are promising for radiofrequency and microwave transmitters [3], and magnetic quasiperiodic SLs - for storage devices [4]. Semiconductor quasiperiodic SLs also have interesting properties. For example, AlGaAs quasiperiodic SLs are promising as an active media of a multicolor terahertz laser [5]. Now semiconductor quasiperiodic SLs based upon the AlGaAs are the most convenient for micro- and nanoelectronic devices due to a well-developed GaAs technology of layered heterostructures’ growth. AlGaAs SLs’ quasiperiodicity appeared already in the first implementation of the quantum cascade laser (QCL) [6], where special layered nanostructures (injector and collector) were added inside of each SL’s period to strengthen the coupling of adjacent cells. Quasiperiodic configurations of various spatial symmetry composed of such components as quantum dots (QD), are considered as promising for future nanoelectronics [7]. Such electronic configurations in the QD’s cluster will be analogs of various p-d-hybridization’s forms in the macromolecules of biological enzyme catalysts. Instrumental application of quasiperiodicity in the time domain also have a long history. Since the classical work [8] for the linear frequency modulation, linear and nonlinear radio signal’s transformations in the frequency, time, angular and polarization fields are developing [9]. Famous characteristic features of semiconductor quasiperiodic SLs are highly dissected shape of an electronic states’ spectrum and its self-similarity [1]. In contrast to spectra of periodic SLs and of traditional double-barrier resonant tunneling diodes (RTDs), spectra of quasiperiodic SL depend nonmonotonically upon external electric field. If we gradually increase the applied voltage, new resonant states of conduction electrons can appear suddenly. These are the states localized in 2-3 adjacent potential wells of conduction band bottom’s profile across quasiperiodic SL layers those provide several close frequencies of transitions with an energy of 2  about 10 meV in the multicolor terahertz laser in [5]. These weakly localized states can be useful for all applications that require separate closely spaced resonant peaks with a width of about 1-10 meV, and not continuous energy minibands having a width of about 100 meV. If some measures are taken against formation of strong electric field’s domains in the semiconductor SL, then each such resonant peak in density of states can lead to waviness of the quasiperiodic SL’s voltage- current characteristic (VCC). Therefore, quasiperiodic semiconductor superlattices can be useful in all electronic devices that are sensitive to shape of the nonlinear element’s VCC. Such devices include neuromorphic networks, in which every microelectronic neuron contains a nonlinear element such as a diode having the VCC with a negative slope. Behavior of the coupled neurons are often described by the FitzHugh–Nagumo model (see, e.g. [10], p.171). Some simplification reduces this model to the model of cellular nonlinear network (CNN), each cell of which contains just a diode, a capacitor and a few resistors - one resistor for each link with neighboring cells [11]. RTD is considered as a promising nonlinear element for such networks. By varying layers’ thickness and composition in a double-barrier heterostructure, we can change shape of an initial part of the RTD’s VCC. This shape is important for RF and microwave mixers, detectors and amplifiers (see for ex. [12]. p.12). For neuron operation not the initial, but a falling VCC’s part is especially important. The transition from the RTD to some quasiperiodic SL can make this falling part wavy, and it will change behavior of the neuron. In particular, VCC’s waviness may be increased so that new small falling parts will appear on the VCC. This will lead to formation of new equilibrium states in the network’s phase space. Therefore, using of quasiperiodic superlattices as nonlinear elements in neuromorphic networks may be promising for a substantial transformation of their phase portraits. This is interesting for many areas - from a nervous system’s modeling to the development of 3  information- measurement and control systems. Neuromorphic networks are performing well in parallel transformation of images. Therefore, it would be useful to clarify possible advantages of quasiperiodic SLs as nonlinear elements in some network for the parallel image conversion. Nonlinear elements such as RTDs with VCC’s various shapes are considered as promising for applications in digital circuits [13], as well as in analog signal processing [14] as modulators and amplifiers. For example, the efficiency of the power amplifier is significantly increased, if nonlinearly to distort the signal (by sharpening the upper part of the sine wave) before to amplify it. In the following cascades this nonlinear distortion can be removed by the power amplifier operating in the nonlinear saturation regime with occupying the horizontal part of its amplitude characteristic. Because of the large amplitude this method very effectively converts the energy of the DC power supply to the energy of the AC signal (see for ex. [14]. p.158). II. The construction of quasiperiodic superlattices for nonlinear networks Basic principles of semiconductor quasiperiodic SL’s building are described in [1], p.154. In this work Fibonacci SLs were taken as typical representatives of the quasiperiodic superlattices. In addition to them figurate SLs were constructed by decomposition of Fibonacci numbers in the sum of figurate numbers [16] and were investigated as promising network’s nonlinear elements. Fibonacci number SN of rank N is formed by adding SN = SN–1 + SN–2 the numbers of the two previous ranks SN–1 and SN–2 starting with S1 = 1 and S2 = 1. Similarly symbol sequence for N-th rank’s Fibonacci superlattice SN (denoted by the same letter as the corresponding Fibonacci number) is formed by the serial connection (concatenation) SN = SN–1 + SN–2 of the symbolic sequences for previous two ranks’ superlattices SN–1 and SN–2 , starting with S1= А and S2= В. For example, SL S5= S4 + S3= ВАВВА, then SL S6= ВАВВА + ВАВ= ВАВВАВАВ, etc. 4  Like Fibonacci numbers SN, figurate numbers FM L(N) can be calculated by recurrence formulas starting from given boundary values. M-th order’s figurate number FM L(N) is expressed through zero order’s figurate numbers F0 L(N) and F0 L(N-1) by the formula FM L(N)= F0 L(N) +М F0 L(N-1). In turn, figurate number F0 L(N) is expressed recursively F0 L(N)= F0 L–1(N) + F0 L(N-1) (Pascal's triangle), which finally leads to numbers F0 L(0)= 1 и F0 0(N)= 1 at boundaries of the Pascal's triangle. Therefore, to construct M-th order’s figurate SLs FM L(N) it is sufficient to construct zero-order’s SLs F0 L(N), and then use the recursion formula FM L(N)= F0 L(N) +М F0 L(N-1). Here, under the multiplication SL F0 L(N-1) by number M we mean repetition of M copies of this SL. To build the zero-order’s SLs F0 L(N), we reduce them to already obtained Fibonacci SLs. To do this, we apply the formula (1) for expansion of a Fibonacci number SN by figurate numbers F0 L(n) (see, eg., [17]). (1) Here the expression [N/2] denotes an integer part of the number N/2. We write down the symbolic sequence for SL SN. Then we assign to each number F0 L(n) in the sum (1) a segment of this symbolic sequence such that its length is equal to the number F0 L(n). For example, according to the formula (1), S1= F0 0(0) and S2= F0 1(0). But SL S1= А and SL S2= В, therefore we obtain SL F0 0(0)= А and SL F0 1(0)= В. Similarly we act to obtain all remaining SLs F0 L(N). Thus from the decomposition S3= F0 2(0) + F0 0(1) (in numbers it looks like 2=1+1) we obtain S3= ВА= F0 2(0) + F0 0(1) =В+ А. Hence SL F0 2(0)= В and SL F0 0(1)= А. Further, for example, from the decomposition S8= F0 7(0) + F0 5(1) + F0 3(2) + F0 1(3) (in numbers it looks like 21= 1+ 6+ 10+ 4) we obtain ВАВВАВАВВАВВАВАВВАВАВ = В + АВВАВА+ ВВАВВАВАВВ+ АВАВ. Hence SL F0 7(0) = В, SL F0 5(1) = АВВАВА, SL F0 3(2)= ВВАВВАВАВВ, and SL F0 1(3)= АВАВ. The advantage of this method for constructing of zero-order’s SLs F0 L(N) is that they inherit from Fibonacci SLs their stochastic properties. 5 )(]2/[0021nFSNnnNN In this work for the blocks A and B we take different layered AlGaAs heterostructures with thicknesses of a few GaAs -monolayers (ML), each by 0.565 nm (layered structure on the left in Figure 1) so that total SL length does not exceed the characteristic electron mean free path of about 100 nm. Figure.1. Three neighboring network’s cells based on the Fibonacci superlattice S7. At this thickness, a probability of the appearance of strong electric field’s domains is small. Therefore, states of conduction electrons are considered coherent throughout all SL’s layers in a classical transfer matrix method. This assumption about the coherence of the conduction electron’s wave function is not necessary to calculate a VCC using the Tsu-Esaki formula. In this formula we need to know only the electronic transparency of the whole structure, i.e. the probability of electron’s transition through all barriers and wells. The same results about the transparency peaks, electron localization and resonant tunneling in the quasiperiodic semiconductor superlattices are obtained by the strong coupling method [15], that requires only knowledge of the probability of an electron jump to the next site. 6  Thickness and composition of this layers in blocks A and B were chosen so that SL’s VCC had an elongated falling part (preferably in a wavy form) in moderate electric fields of about 10 kV/cm (Figure 2). Figure 2. The characteristics of cell’s nonlinear elements based upon the Fibonacci superlattice S7. Bottom left – an element’s voltage- current characteristic J(F). The vertical line indicates the falling part’s beginning. Left top - the dependence of resonance energies E on the electric field strength F. Middle - the band diagram and squared moduli of electron’s wave function depending on a coordinate x across the SL’s layers. Right - the dependence of the tunneling transparency T(E) on the electron’s energy E in the electric field corresponding to the beginning of the current-voltage characteristic’s falling part. III. Voltage-current characteristics of quasiperiodic superlattices for nonlinear networks To calculate VCCs the formula Tsu-Esaki was used in combination with the traditional transfer matrix method (see, e.g. [18], p.58, and [19]). Correctness of computational procedures was checked by reproduction of the VCC (not shown here) one of the famous traditional double- barrier AlGaAs heterostructure [20]. 7  Briefly, the method of calculation is as follows. For a given voltage U calculation of the current density J across the layers of the heterostructure is reduced to the numerical integration in the Tsu - Esaki formula (*). (*) This formula contains two functions F(E) and T(E) of an electron energy Е. The constant C is determined as С≡em/2π2ћ3≈2∙109 (А/сm2)/eV2 and sets the maximum theoretical current density. Here, e and m - charge and effective mass of the electron, ћ - Dirac's constant. The function F(E) is known - it decreases slowly with increasing energy Е and increases monotonically with increasing voltage U according to the formula F(E)=(1/β)ln[(1+exp(β(EF- E))/(1+exp(β(EF-E-eU))]. Here β=1/kBT0, where kB - Boltzmann constant, T0 - temperature. The Fermi energy EF increases monotonically with increasing dopant concentration Nd in the degenerate near-contact regions according to the formula EF =(ћ2/2m)(3π2 Nd)2/3. We find the transparency T(E) by the numerical calculations using transfer matrix method. To do this, the x- axis across the heterostructure’s layers we split into small parts such that within each n-th part the electron’s potential energy V(x) can be considered as constant Vn. Solution of the Schrödinger equation for this case is known as a sum of two plane waves y=Anexp(Knx)+ Bnexp(–Knx). Here, the wave number Kn is calculated by the formula Kn = [2m(E- Vn)]1/2/ћ. To find the unknown coefficients An and Bn the conditions of continuity of the function y(x) and of its derivative dy/dx at the boundaries of adjacent small parts are used. So we obtain a system of coupled equations for neighboring An and Bn. By involving the boundary conditions for the wave function far away from the heterostructure we obtain the amplitude of the wave passing across all layers. The required transparency T(E) is the ratio of the transmitted flux of electrons to the incident one and is proportional to the square of the A’s modulus. 8 EFETdECUJ0 In this work each quasiperiodic SL’s block B(A) was consisted of a barrier layer having a thickness of 2 ML, followed by a potential well with a thickness of 16(32) ML (see the band diagram in the middle of Figure 2). A height V of the potential barriers (in eV) and an effective mass M (in units of a free electron mass) were calculated from the traditional expressions V= 1.11x - 0.93x2 + 0.85x3 and M= 0.067 + 0.083x, where x - aluminum’s proportion in the AlxGa1- xAs- layer. This proportion was x =0.15, which gave a potential barriers’ height 0.15 эВ for conduction electrons. At near-electrode n+GaAs- layers a Fermi energy was taken equal to 0.069 eV, and an effective mass M= 0.067. A contact potential difference of 0.1 eV between the n + GaAs-layers and middle undoped i-AlGaAs-layers has been added to the potential profile. A temperature was assumed equal to 300 K. Figure 3 shows VCCs of network’s nonlinear elements based on the figurate SLs F0 11(1)= ABBABABBABBA and F8 11(1)= ABBABABBABBABBBBBBBB in comparison with the reference cubic VCC of a conventional nonlinear element like RTD. Figure 3 shows several of the same current-voltage characteristics as in the tab of the Figure 2, just shifted and scaled for the convenience of analysis of the nonlinear element’s action in parallel image conversion using nonlinear network (see next section). 9  Figure 3. The voltage-current characteristics of network’s nonlinear elements: 1- reference «cubic» diode, 2- figurate superlattice F0 11(1), 3- figurate superlattice F8 11(1). The reference cubic voltage-current characteristic has a symmetrical falling part with identical positive and negative regions and serves to compare of all other VCCs. VCCs are normalized so that they have the same maximum current I = 10 mA, and the most left (right) growing branch intersects with zero at the voltage V =0 (1) Volt respectively. At a microelectronic implementation such shift of VCC’s zero may be achieved by adding appropriate permanent current and voltage sources to all network’s cells at once. IV. Parallel image transformation using nonlinear networks We considered a nonlinear network in the form of flat lattice of the Fitzhugh-Nagumo neurons [10]. This lattice simulates some kind of a two-dimensional excitable media. A propagation of an excitation in such media is described by a system of kinetic equations, which reduces to the single reaction-diffusion equation after an assumption of the absence of a slow inhibitor [21] (2) Here V- variable describing the excited state of the medium at space point r at time moment t, DV – diffusion coefficient describing spreading of the excitation, ∆ – Laplacian, FR(V) – rate of the excitation’s growth caused by an autocatalytic reaction. To carry out a parallel signal conversion using such a medium, the signal is set as a spatial distribution V(r,t0) of the excitation at some initial moment of time t0. Next, a medium evolves according to the equation (2) and after some time, we obtain the converted signal as a new spatial distribution V(r,t). If desired transformation is to clean up the distribution V(r,t0) from a high-frequency spatial noise, it is doing good by the diffusion term FD = DV∆V in (2). However, 10 VFVDttrVRV, this term reduces not only harmful noise’s amplitude, but also amplitude of useful low-frequency signal’s components. This is the reaction term FR in (2) that must counteract this (Figure.4). Figure.4. Opposing actions of the diffusion FD and reaction FR terms in signal V(x) processing by the diffusion-reaction medium. Here x - coordinate in the medium. The diffusion term FD reduces a height of hills and a depth of wells in the relief of useful low-frequency signal’s components. For preventing this phenomenon the reaction term FR should shift down wells, and shift up hills. This explains why the dependence of FR(V) must be S-shaped. In a microelectronic design such two-dimensional diffusion-reaction medium can be approximately realized by means of a planar nonlinear network [22]. In nodes of this network there are cells that contain resistors, capacitors and nonlinear resistances (рис.1). Under this implementation the equation (2) takes the form (3) (3) Here, Vn,k - voltage in (n,k)-th network’s node, τ = RC - time constant of identical RC- chains connecting neighboring nodes. Each chain consisted of the resistance R = 10 Ohms and the capacitance C = 10 pF, which gave the characteristic time of image transformation T = 100 ps. In solving the system (3) at each step of time evolution the value of current I(V) through a 11 CVIVVVVVVdtdVknknknknknknknkn,1,,1,,1,,1,22 nonlinear element is calculated by cubic approximating of the VCC comprising 1,000 points. A role of the diffusion FD term of equation (2) is played by the first two terms in equation (3), and a role of the reaction FR term - by the last term in equation (3), i.e. by the nonlinear element’s VCC with a minus sign. Thus, to prevent smoothing of the useful signal, the nonlinear network element’s VCC must be N-shaped. In a practical realization of such a network in the matrix photoconverter an image is divided into Nx pixels horizontally and Ny pixels vertically [23]. Under every pixel the network node is located. We assume that an initial voltage Vn,k in the node cell (n,k) is equal to an intensity of light striking to this pixel, and can range from 0 (black) to 1 (white) Volt. After the initial illumination a network is disconnected from the photodetector and the voltage Vn,k at each node cell (n,k) begins to vary in time t due to presence of a capacitance and a nonlinear element within the network’s cell, and also due to the resistive connection of each cell to its neighbors. After some characteristic time, determined by the network parameters, each network’s node can be connected to the corresponding node of a light-emitting matrix to obtain a transformed image. Details of image conversion using such networks are described in [22]. In this work we assumed Nx = 64 points horizontally and Ny = 64 points vertically. A method of network’s research was used the same as in [22]. To find the dependence Vn,k(t) we have solved a system of N= Nx*Ny = 4056 ordinary differential equations (3) using a specialized MatLab package’s algorithm «ode15s». For boundary cells a condition of image’s smoothness was assumed, i.e. a condition of continuity of a spatial derivative of the voltage in each border’s cell. For example, it was assumed that V0k= 2V1k – V2k for the first row’s cells in (3). Cases of indexes 0 and 1+Ny (Nx) at the first and last rows (the first and last column) were considered similarly. 12  Gaussian noise in the form of a random sample of N numbers having a normal distribution with zero mean and a standard deviation of 0.1 was added to the original reference image. In this work the purpose of image converting via network was the obtaining an image as close to the reference image as possible, i.e., obtaining an image purified from the added Gaussian noise. The relative standard deviation D was assumed as a measure of the difference between the transformed image and the reference one. It was calculated as the ratio V-W / W of vectors’ modules. Here V - vector of the converted image, W - vector of the reference image, . denotes a modulus of the vector. Components of the images’ vector V are the voltages Vnk at network nodes. Images were compared at 10 consecutive time moments in the interval 0 - 2 in units of τ= RC. Usually in this time interval of the order of 100 ps the conversion error D first decreases and then increases for the studied VCC’s shapes. V. Action of nonlinear networks based on quasiperiodic superlattices The existence of many maxima in the quasiperiodic SL’s VCC can lead to a presence of several stable equilibrium states of the network’s cell. This multistability is important for all processes occurring in nonlinear networks. In particular, benefits of VCC with many maxima for parallel image processing using the cellular nonlinear network are described in [23]. There VCC with many maxima was obtained using several operational amplifiers and analog multipliers. Here we see that a VCC with several well distinguishable maxima can be obtained using only one quasiperiodic SL. This opens the possibility of parallel image processing in time scale of the order 10 ps. Figure 5 shows the process of a noisy image’s transformation (Fig.5b) using typical nonlinear network based on a reference diode with the cubic VCC shown in Figure 3. 13  Figure 5. Converting an image using network based upon a reference diode having the cubic VCC: а) - reference image, b) - noisy image before conversion, с)…l) – converted image at the times t/τ=0.1(c), 0.2(d), 0.3(e), 0.4(f), 0.5(g), 0.6(h), 0.8(i), 1.1(j), 1.5(k), 2.0(l). Deviation from the reference image (fig.5a) initially decreases (fig.5b-d) because of noise’s smoothing, reaches a minimum (fig.5e) at the time point t/τ = 0.3 and then increases (fig.5b-d) due to contrast increasing. Figure 6 compares time dependences of the mean-square-deviation D for images obtained using network based on the “cubic” diode, figurate superlattice F0 11(1) and figurate superlattice F8 11(1). 14  Figure 6. Time dependences of the image’s deviation from the reference image during conversion using network based on the 1- diode having a cubic VCC, 2- figurate superlattice F0 11(1), 3- figurate superlattice F8 11(1). For all structures the smallest deviations were achieved within a time comparable with a characteristic time τ. For SL F0 11(1) the smallest deviation of transformed image from the reference one is almost twice less than for the other structures. On Figure 7 converted images having the smallest deviations from the reference image are compared with each other. Figure 7. Converted images having the smallest deviations from the reference one: а) - reference image, b)…d) – best images, obtained using the network based on the: b) – diode having a cubic VCC, с) – figurate superlattice F8 11(1), d) – figurate superlattice F0 11(1). It can be seen that an improvement of image transformation in this series (Fig.7b-d) is associated with the improvement of transformation of an image’s background. Network based on the reference “qubic” diode leaves background grainy (Fig.7b). Network based on the figurate SL F8 11(1) significantly reduces a granularity of the background (Fig.7c). At last, network based on the figurate SL F0 11(1) almost entirely removes the background’s granularity (Fig.7d). The falling part of a nonlinear element’s VCC in network’s cell is responsible for the transformation of an image background. Cubic VCC of the conventional nonlinear element has one positive and one negative part in the unit voltage range (Figure 3). During transformation the VCC’s positive part shifts a dark gray point of the image toward black and the VCC’s negative 15  part shifts light gray towards white. Background’s points having a gray color turn out in an unstable state during this process. To improve transformation it is desirable that gray points were moving to extreme states of black and white very slowly, or that they were moving to some steady states which have approximately the same gray color. Middle parts of the VCCs of nonlinear elements based on figurate SLs just satisfy these requirements (Figure 3). Negative part of the VCC of the figurate SL F8 11(1) is wavy, and the positive one is non-monotonic. Therefore, the conversion of a gray background using SL F8 11(1) is better than using a reference “qubic” diode. Nonmonotonicity of the VCC of the SL F0 11(1) is so large that several positive and negative VCC’s branches appears. Rising portions of these branches are crossed with horizontal at points that are stable states of gray in an image. Therefore, the conversion of a gray background using SL F0 11(1) turns out better than using SL F8 11(1) and much better than using a reference “qubic” diode. Other quasiperiodic SLs as a part of a nonlinear network’s element behave analogously to considered figurate SLs during image filtration. Image transformation depends on the waviness of a VCC’s falling part and on the amplitude of this waviness. VCC of the Fibonacci SL S7= BABBABABBABBA (fig.2) и VCC of the figurate SL F1 1(6)= ABBABABBBABAB (not shown) have weak waviness of the falling part, so behave analogously to the figurate SL F8 11(1). The waviness of the VCC of the Fibonacci SL S8= BABBABABBABBABABBABAB (not shown) increases but does not achieve the waviness of the VCC of the SL F0 11(1). So the SL S8 in sense of image filtering error turns out between the SL F8 11(1) and the SL F0 11(1). It should be noted that advantages of quasiperiodic SL associated with multistability of cells, with a high probability will manifest itself not only in neuromorphic networks, but also in all other nonlinear dynamical systems. This follows from equivalence between above discussed nonlinear network’s cell and an overdamped oscillator located in a potential, the shape of which 16  depends on the shape of the nonlinear element’s VCC [23]. So the transition from conventional heterostructures having N-shaped VCC like RTD to quasiperiodic SLs is equivalent to transition from a dynamical system within a double-well potential to the system within a multiwell potential. Therefore, in the field of nonlinear systems, use of quasiperiodic SLs may lead to discovery of new interesting phenomena. For example, it will be interesting to investigate propagation of solitons and spiral waves in the reaction-diffusion media based on quasiperiodic SLs, as well as the phenomena of self-organization, including a self-organized criticality and chaotic oscillations. The possibly easiest way for the practical realization of the benefits of the discussed superlattice diodes - it is just to insert them in a classical digital [13] and analog [14] schemes instead of standard TD’s and RTD’s. VI. Conclusion On the example of Fibonacci and figurate AlGaAs superlattices it is shown that quasiperiodic semiconductor superlattices are promising as nonlinear elements in cells of the FitzHugh–Nagumo neuromorphic networks. A waviness of a falling branch of a current-voltage characteristic of the quasiperiodic superlattice may lead to a new equilibrium states in the network’s phase space. Arising in this way multistability has a positive effect on a process of parallel conversion of an analog signal using the neuromorphic network. In particular, figurate superlattice F0 11(1) as a part of a nonlinear network’s element provides the conversion error almost twice less than do traditional diodes with a cubic current-voltage characteristic. Therefore, quasiperiodic superlattices are promising for nonlinear information-measuring and control systems, as well as for modeling of the nervous system. The multistability of a network’s 17  cell based on the quasiperiodic semiconductor superlattices promises new interesting nonlinear dynamical phenomena. 1. E. Macia. Aperiodic structures in condensed matter: fundamentals and applications (N.Y., CRC Press, 2009). 2. E. Macia. Rep. Progr. Phys., 75, 036502 (2012). 3. X. Zhang, Y. Lu, Y. Zhu, Y. Chen and S. Zhu. Appl. Phys. Lett., 85, 3531 (2004). 4. L.D. Machado, C.G. Bezerra, M.A. Correa, C. Chesman, J.E. Pearson and A. Hoffmann. Phys. Rev. B, 85, 224416 (2012). 5. K.V. Malyshev. Quantum Electronics, 43, 503 (2013). 6. J. Faist, F. Capasso, D.L. Sivco, C. Sirtori, A.L. Hutchinson, A.Y. Cho. Science, 264, 553 (1994). 7. R. Tsu. Nanoscale Research Letters, 6, 127 (2011). 8. J.R. Klauder, A.C. Price, S. Darlington and W.J. Albersheim. Bell System Tech. J., 39, 745 (1960). 9. F. Gini, A. De Maio, L.Patton, editors. Waveform design and diversity for advanced radar systems (London, The Institution of Engineering and Technology, 2012). 10. A. Fuchs. Nonlinear Dynamics in Complex Systems: Theory and Applications for the Life-, Neuro-and Natural Sciences (N.Y., Springer, 2013). 11. P. Julian, R. Dogaru, M. Itoh, M. Hanggi and L.O. Chua. IEEE Trans. Circ. Sys. I, 50(4), 500 (2003). 12. H. K. Khalil. Nonlinear Systems. 3rd Edition (Upper Saddle River, NJ, Prentice Hall, 2002). 18  13. P. Mazumder, S. Kulkarni, M. Bhattacharya, J. P. Sun, G.I. Haddad. Proceedings of the IEEE. 86(4), 664 (1998). 14. S. Marshall, G.L. Sicuranza. Advances in nonlinear signal and image processing (N.Y., Hindawi Publishing Corporation, 2006). 15. W.L. Ma, S.S. Li. J. Appl. Phys., 112, 014314 (2012). 16. K.V. Malyshev, S.L. Chernyshev. Measurement Techniques, 54, 496 (2011). 17. T. Koshy. Fibonacci and Lucas Numbers with Applications (N.Y., John Wiley & Sons, 2001). 18. R. Tsu. Superlattice to Nanoelectronics (Amsterdam, Elsevier, 2005). 19. E. Cassan. J. Appl. Phys., 87(11), 7931 (2000). 20. E. Wolak, E. Ozbay, B.G. Park, S. K. Diamond, D.M. Bloom and J.S. Harris Jr.. J. Appl. Phys., 69(5), 3345 (1991). 21. M. Cai, J. Pan and H. Zhang. Phys. Rev. E, 86, 016208 (2012). 22. S. Morfu, P. Marquie, B. Nofiele and D. Ginhac. Adv. Imag. Electr. Phys., 152, 79 (2008). 23. S. Morfu, B. Nofiele, P. Marquie. Phys. Lett. A. 367(3), 192 (2007). 19 

1407.7706

2

1407

Modulation of electronic and mechanical properties of phosphorene through strain

We report a first-principles study on the elastic, vibrational, and electronic properties of recently synthesized phosphorene. By calculating Gr\"uneisen parameters, we evaluate the frequency shift of Raman/infrared active modes via symmetric biaxial strain. We also study an inducing semiconductor-metal transition, the gap size, and effective mass of carriers in various strain configurations. Furthermore, we unfold the emergence of a peculiar Dirac-shaped dispersion for specific strain conditions, including the zigzag-oriented tensile strain. The observed linear energy spectrum has distinct velocities corresponding to each of its linear branches and is limited to the $\Gamma-X$ direction in the first Brillouin zone.

cond-mat.mes-hall

cond-mat

Modulation of electronic and mechanical properties of phosphorene through strain Mohammad Elahi, Kaveh Khaliji, and Seyed Mohammad Tabatabaei School of Electrical and Computer Engineering, University of Tehran, Tehran 14395-515, Iran School of Electrical and Computer Engineering, University of Tehran, Tehran 14395-515, Iran Institute for Microelectronics, Technische Universitat Wien, Gusshausstrasse 27–29/E360, A-1040 Wien, Austria Mahdi Pourfath∗ Reza Asgari† School of Physics, Institute for Research in Fundamental Sciences (IPM), Tehran 19395-5531, Iran (Dated: March 26, 2015) We report a first-principles study on the elastic, vibrational, and electronic properties of the recently synthesized phosphorene. By calculating the Gruneisen parameters, we evaluate the fre- quency shift of the Raman/infrared active modes via symmetric biaxial strain. We also study a strain-induced semiconductor-metal transition, the gap size, and the effective mass of carriers in various strain configurations. Furthermore, we unfold the emergence of a peculiar Dirac-shaped dispersion for specific strain conditions including the zigzag-oriented tensile strain. The observed linear energy spectrum has distinct velocities corresponding to each of its linear branches and is limited to the Γ − X direction in the first Brillouin zone. PACS numbers: 71.15.Mb, 71.20.Mq, 63.22.Np I. INTRODUCTION The discovery of graphene in 2004 has triggered an unprecedented leap in the research on ultrathin two- dimensional (2D) crystals [1, 2]. Such crystals are mostly exfoliated into individual thin layers from their lay- ered counterparts. Famous examples include graphene, hexagonal boron nitride [3], and molybdenum disulfide, the latter being the most well-known member of the fam- ily of 2D transition metal dichalcogenides [4]. Due to the wealth of exquisite physical phenomena that arise when charge, spin and heat transport are restricted within a 2D plane, these materials have been among the most in- teresting subjects in condensed matter physics [5]. The intriguing prospect of the potential nano- electronic applications which may take advantage of the impact of quantum confinement and dimensionality re- duction in 2D materials has enticed the scientific commu- nity to actively explore possibilities of similar materials with outstanding characteristics. In this regard, phos- phorene, an atomically thin layer of the element phos- phorus which has a natural band gap, has been synthe- sized recently through mechanically cleaving bulk black phosphorus (BBP) followed by a plasma-assisted thin- ning process [6]. In phosphorene, the atoms are arranged in a rectangular lattice with the surface being slightly puckered (see Fig. 1(a)), giving rise to novel correlated electronic phenomena ranging from semiconducting to superconducting behaviors. Moreover, the monolayer is still planar enough to confine electrons so that charge ∗ [email protected]; [email protected] † [email protected] flows quickly, leading to a relatively high mobility that is promised by the electronic, optical, mechanical, chem- ical, and thermal properties [7–11]. In particular, being a semiconducting 2D material, phosphorene now renders to be an appealing candidate for nano-electronic applica- tions as asserted in field effect transistors based on multi- layers of this material [9, 12–17]. The controlled introduction of strain into semiconduc- tors offers an important degree of flexibility in both sci- entific and engineering applications. To gain insight on how phosphorene can be fruitful in the realization of high performing devices, fundamental studies on the strain- induced variation of mechanical and electronic properties of this material are essential. This can be readily evi- denced by referring to both the ubiquity of mechanical perturbations and the numerous previous investigations regarding the possibility of amending the electronic prop- erties of 2D materials through strain engineering [18–23]. In this context, it has been recently proposed that via perpendicular compression, the electronic band structure of phosphorene undergoes a semiconductor-semimetal- metal transition [24]. Peng et al. has also demonstrated a strong modulation of both band gap and effective mass of carriers in response to axial in-plane deformations [25]. Moreover, a unique anisotropic conductance is reported which can be controlled and even rotated by 90 degrees under specific uniaxial and symmetric biaxial strains [26]. In this paper, we carry out first-principles simula- tions to investigate the elastic, vibrational, and elec- tronic properties of phosphorene. Our numerical results show a negative Poisson’s ratio in the out-of-plane di- rection under uniaxial deformations oriented along the zigzag direction. To obtain the Gruneisen parameters, the frequency shift of the Raman/infrared active modes through symmetric biaxial strain are evaluated. We demonstrate the feasibility of in-plane deformations in inducing semiconductor-metal transition and manipulat- ing the gap size and effective mass of carriers in various strain distributions. With the application of specific bi- axial strain distributions, we further report on the forma- tion of a peculiar Dirac-like energy spectrum. The ob- tained electronic dispersion is queer as it is linear along the Γ − X direction, while being parabolic along the or- thogonal path. Sec. This paper is organized as follows. II de- scribes the methodology employed for the electronic and phononic calculations. In Sec. III, the relevant mechan- ical constants of phosphorene, the variation of the Ra- man/infrared active modes with strain, along with the strain-induced modulation of the electronic dispersion are investigated. Finally, a short summary and conclud- ing remarks are presented in Sec. IV. II. THEORY AND METHOD We carry out first-principles simulations based on the density-functional theory (DFT) as implemented in the SIESTA code [27] to perceive the relevant mechanical, vibrational and electronic properties of a single layer phosphorene. The VASP package [28] is also used in some instances throughout this paper to provide in- creased precision for critical results. Apart from the package used, calculations begin with the determination of the optimized geometry [29], i.e., the configuration in which the residual Hellmann-Feynman forces acting on atoms are smaller than 0.01 eV/A. In the SIESTA code, this can be achieved by employing a double-ζ- double-polarized (DZDP) basis set along with the con- jugate gradient method within the generalized gradi- ent approximation (GGA) formalism and taking advan- tage from the norm-conserving Troullier-Martins pseudo- potentials [30, 31]. In the VASP package, the projec- tor augmented wave method along with the Perdew- Burke-Ernzerhof (PBE) form of the exchange correla- tion functional are adopted for the calculation of the exchange-correlation energy [30, 32]. A cutoff energy equal to 180 Ry (500 eV) is used for calculations using SIESTA (VASP) so that assure a total energy conver- gence better than 0.01 meV/unit cell in obtaining the self-consistent charge density. A vacuum separation of 15 A, which is sufficient to hinder interactions between adjacent layers, is adopted. Sampling of the reciprocal space Brillouin zone is done by a Monkhorst-Pack grid of 12 × 12 × 1 k-points. The phonon dispersion curves and the Raman/infrared active modes are calculated by di- agonalizing the dynamical matrix obtained by the small- displacement method (SDM) with forces calculated in a 4 × 4 supercell [33]. In contrast to the flatness of graphene, phosphorene is a puckered honeycomb structure with each phosphorus atom covalently bonded with three neighboring atoms 2 within a rectangular unit cell (see Fig. 1(a)). The crys- tal structure is spanned by lattice vectors (cid:126)a1 = a1(cid:98)x and (cid:126)a2 = a2(cid:98)y along armchair and zigzag directions, respec- tively. The distinct armchair ridges in the side view of phosphorene in Fig. 1(a) are characterized by the lattice buckling constant, i.e., ∆z. We first calculate the structural parameters of BBP and a monolayer phosphorene and compare the results for BBP with those results obtained in experiment [34]. We use, by treating van der Waals (vdW) interactions between adjacent layers in BBP, the Grimme correction to the PBE functional in SIESTA [35] and thus the lat- tice parameters are calculated and summarized in Ta- ble I. The DZDP basis set along with the PAW pseudo- potentials are employed to calculate the same set of struc- tural parameters. It is noted that the DZDP method, ex- cluding the vdW treatment, provides adequate accuracy in terms of its compliance with the reported experimental values for BBP. Moreover, the tiny discrepancies between VASP and SIESTA results can be ascribed to the differ- ent parameterizations of the functionals used and to the different basis sets employed in each package (plane waves versus numerical atomic orbitals). Therefore, the DZDP basis set without the vdW correction is adopted as the main tool for evaluating the results. In some instances, the PAW method is also invoked where critical results have been encountered. It should be mentioned that our calculated structural parameters for both BBP and phos- phorene are in excellent agreement with reported before calculations [7, 9, 25, 36, 37]. Method a2 az ∆z Ecoh d - Material Bulk Monolayer Exp. DZDP a1 4.37 3.31 10.47 2.16 3.07 4.40 3.34 10.67 2.17 25.66 3.16 DZDP+vdW 4.36 3.34 10.38 2.17 26.49 3.02 4.54 3.31 11.17 2.12 21.43 3.46 4.44 3.32 DZDP+vdW 4.43 3.32 4.62 3.30 2.15 25.27 2.15 25.78 2.10 21.40 PAW DZDP PAW - - - - - - TABLE I. The equilibrium lattice constants, a1, a2, and az (in units of A), buckling ∆z, cohesive energy Ecoh (in units of eV), and interlayer distance d (in units of A) for bulk and monolayer black phosphorus. Experimental data is reported in Ref. [34] III. NUMERICAL RESULTS AND DISCUSSION In this section, we present our main numerical results based on first-principles simulations. Our aim is to ex- plore the impact of strain on vibrational, mechanical and electronic properties of phosphorene. All the first- principles calculations are performed at room tempera- ture. A. Elastic and vibrational properties of strained phosphorene In-plane lattice constants are either stretched or com- pressed by εx and εy in a 4 × 4 supercell (see the inset of Fig. 1(b)) in order to obtain the strained structure. The consequent structure is then relaxed with keeping the deformed lattice vectors unchanged. We obtain the variation in strain energy, ES, by sub- tracting the total energy of the deformed structure from the equilibrium total energy, as the strain varies from 0% to 35% in the uniform expansion regime (εx = εy ≥ 0). From Fig. 1(b), the harmonic region can be assumed within the strain range of 0-0.02 and afterwards the an- harmonic region occurs and is basically followed by a plastic region (see the shaded area in Fig. 1(b)) where irreversible changes occur in the structure of the sys- tem. The corresponding yielding strain is found to be 27%, which is similar to that reported for graphene and molybdenum disulphide (MoS2), revealing the promise of phosphorene for stretchable electronic devices [38, 39]. x + bε2 We further calculate Poisson’s ratio ν, the ratio of the transverse strain to the axial strain, along with the in- plane stiffness parameters C, to assess the mechanical response of phosphorene. Figure 1(c) shows the mesh plot of strain (εx, εy) and the corresponding strain en- ergies. The strain-energy relation is then obtained as ES = aε2 y + cεxεy, where a, b, and c are fitted pa- rameters obtained as 14.88, 50.06, and 23.94 eV, respec- tively. We then calculate stress along the x (y)-direction, denoted by σx(y) through σx(y) = V −1 0 ∂ES/∂εx(y), where V0 is the equilibrium volume. The dashed lines denoted by σx= 0 and σy= 0 shown in Fig. 1(c), correspond to uniaxial deformations along the x- and y-directions, re- spectively. The associated Poisson’s ratios are evaluated as νy=c/2a=0.81 and νx=c/2b=0.24, in consistency with those obtained via VASP package in Ref. [25] (νy=0.7 and νx=0.2). In comparison with the isotropic Poisson’s ratios reported for graphene, boron nitride (BN), and MoS2 (i.e., 0.16, 0.21, and 0.25, respectively), phospho- rene has larger Poisson’s ratios along both the armchair and zigzag directions [19, 22]. 0 (2b − c2 0 (2a − c2 With A0 as the equilibrium area of the system, the in-plane stiffness along the x (y)-direction is defined as Cx(y) = A−1 0 ∂2ES/∂2εx(y). The corresponding Pois- son’s ratio results in Cx=A−1 2b )=26.16J/m2 and Cy=A−1 2a )=88.02J/m2. These values are smaller than those values reported for graphene, BN, and MoS2 (i.e., 335, 267, and 123 J/m2, respectively), implying that phosphorene is more flexible along both armchair and zigzag directions [19, 22]. It should be mentioned that the calculated parameters are in excellent agreement with those reported in Ref. [36] (24.42 and 92.13J/m2, ob- tained by converting the given data using a thickness of 5.55A). It is worth mentioning that BBP exhibits a negative Poisson’s ratio along its armchair direction in response to 3 FIG. 1. (a) Schematic representation of the atomic struc- ture of mono-layer phosphorene from the top and side views. (b) The per unit cell strain energy as a function of strain in uniform deformation regime. The shaded region indicates the plastic range. The inset shows the 4× 4 rectangular supercell used in the calculations. (c) The surface plot of (εx, εy) and the corresponding per unit cell strain energies. The points denote actual data and the background is the fitted formula. (d) The mesh plot of ∆z at the same data points as in (c). perpendicular uniaxial strains [40]. In order to probe the existence of a similar behavior in phosphorene, Fig. 1(d) depicts the monolayer’s thickness for the same set of strain components as in Fig. 1(c). Under uniaxial strain along the y-axis, the thickness is reduced (increased) as the sheet is compressed (stretched). This manifests the existence of a negative out-of-plane Poisson’s ratio in response to y-oriented uniaxial deformations. The cal- culated out-of-plane Poisson’s ratios are 0.21 and -0.09, for uniaxial strains along the armchair and zigzag direc- tions, respectively. Employing SIESTA with a double-ζ basis set, a previous study [41] has reported the out-of- plane Poisson’s ratios to be 0.046 and −0.043, respec- tively. Despite the inconsistency, the presence of a neg- ative Poisson’s ratio is revealed in both studies. Such discrepancy can be attributed to diverse methodologies and fitting procedures adopted for obtaining the Pois- son’s ratios. Another point which might have negatively impacted the accuracy of the calculated Poisson’s ratios in Ref. [41] is that the obtained equilibrium lattice con- stants are 10% larger than those previously reported in the literature [9, 25, 36]. Phonon dispersions for an undeformed monolayer phosphorene are depicted in Fig. 2(a). To obtain the sound velocities, we calculate the slopes of in-plane acoustic branches in the vicinity of the Γ point. The sound velocities in the Γ − Y direction are derived as 7.59 km/s and 4.48 km/s for longitudinal and transverse atomic motions, respectively. Along the Γ − X axis, on a2a1yx(a)TopViewzSideViewElasticRegion4×a 24×a1(b)x[%]051510202530350.50121.52.53ES [eV](c)x[%]-1-2021-1-20210510201525353040y[%]ES[meV]y[%]x[%](d)-1-2021-1-20212.1402.1352.1452.1602.1552.165Δz[Å]x= 0y= 0x= 0y= 02.150PlasticRegion4 g, A2 g, A3 g , and B2 u, and B2 study the electronic and vibrational properties in mate- rials. The Raman spectrum is directly linked to the lat- tice dynamics of materials including phonon dispersion curves, phonon density of states, and infrared/Raman active modes. In accordance with the C2h point group symmetry of phosphorene, the modes A1 g, A4 g, B1 g are characterized to be Raman active while the other three modes, namely, Au, B1 u, are in- frared active. Figure 2(b) shows the frequency shifts of these optical phonon modes as a function of symmetric biaxial strain calculated with Γ-point only simulations. The Gruneisen parameter, the variational frequencies of the individual atoms in phosphorene lattice varied with volume, for a vibrational mode X (γX ), is then calculated as γX = −(2ω0 X is the frequency of mode X in the absence of strain. The extracted av- erage slope and the Gruneisen parameter for all Raman and infrared active modes are presented in Tables II and III. As different modes exhibit qualitatively different be- haviors in response to the applied strain, both negative and positive values for the slope and the Gruneisen pa- rameters are detected. X )−1∂ωX /∂εx, where ω0 Figure 2(c) shows the contour plots of Raman/infrared frequencies with strain. The schematic representation of the atomic motions in each optical mode is also shown on top of each panel. As different modes demonstrate distinct trends under similar strain conditions, the fre- quency shifts of Raman/infrared active modes may serve as fingerprints of certain strain conditions, rendering them viable tools for mapping strain information from spectroscopy measurements. Raman Active ωX (cm−1) Gruneisen Parameters Modes (X) DZDP PAW Exp. ∂ωX /∂εx A1 g A2 g A3 g A4 g B1 g B2 g 221.52 221.65 — 365.09 341.88 363 432.78 424.01 — 452.15 448.19 471.3 181.43 192.67 — 413.65 424.52 440 -149.6 -465.7 -688.5 -460.1 810.1 373.9 γX 0.338 0.638 0.821 0.509 -2.233 -0.432 TABLE II. Phonon frequencies of the relevant Raman mode symmetry representations of phosphorene along with the cor- responding Gruneisen parameters. Our numerical results are compared with those results measured in experiment [6]. ωX (cm−1) IR Active Modes(X) DZDP PAW ∂ωX /∂εx Au ( (cid:126)E(cid:107) (cid:126)a2) 419.30 416.53 u ( (cid:126)E(cid:107) (cid:126)az) 129.02 138.01 B1 u ( (cid:126)E(cid:107) (cid:126)a1) 465.32 457.49 B2 -761.4 739.8 -1062.4 Gruneisen Parameters γX 0.920 -2.867 1.142 TABLE III. Phonon frequencies of the relevant infrared mode symmetry representations of phosphorene along with the cor- responding Gruneisen Parameters. (cid:126)E is the polarization of the incident light. FIG. 2. (a) The phonon dispersion curve of the undeformed phosphorene. ZA marks the out-of-plane acoustic branch and LA (TA) denotes in-plane longitudinal (transverse) acous- tic vibrations. (b) Frequencies of the Raman/infrared active modes at the Γ point of phosphorene under strain. (c) Con- tour plots of Raman/infrared frequencies with strain. The eigenvector of the corresponding vibrational mode is depicted at the top of each panel. the other hand, the sound velocities are obtained as 5.69 km/s and 5.27 km/s for longitudinal and transverse vi- brations, respectively. A previous study reports the re- spective maximum sound velocities along the Γ − X and Γ − Y paths [42] as 3.8km/s and 7.8km/s, which is in good agreement with our results only along the Γ − Y direction. We attribute the difference in the x-directed velocity to the instabilities observed in the out-of-plane acoustic phonon branch presented in Ref. [42], implying that the symmetry restrictions might be neglected during the phononic calculations. At 1% uniform expansion, the respective sound velocities for LA and TA branches are equal to 7.6 km/s and 3.31 km/s along the y-direction and 3.77 km/s and 6.04 km/s along the x-axis. Raman/infrared spectroscopy, as a versatile tool for structural characterization, has been widely used to Bg1x[%]4-402-24-402-2y[%]190186182194Bg2x[%]4-402-24-402-2y[%]430410390450Aux[%]4-402-24-402-2y[%]420400380440(c)Ag2x[%]4-402-24-402-2y[%]370360350380Ag3x[%]4-402-24-402-2y[%]430410450Bu1x[%]4-402-24-402-2y[%]14010060180Ag4x[%]4-402-24-402-2y[%]450440430460470Ag1x[%]4-402-24-402-2y[%]218212224228Bu2x[%]4-402-24-402-2y[%]4704504304900100200300400500Frequency [cm-1]ΓΓYSXLATAZA(a)x[%]Ag3Ag4Ag1Ag2Bg1Bg2AuBu2Bu10100200300400500Frequency [cm-1]00.20.81(b)YS1st BZXΓB. Electronic Properties of strained phosphorene In order to assess how two aspects of mechanical and electronic properties can be beneficially merged in the context of tunable electronic features, the electronic properties of monolayer phosphorene under various strain distributions are studied in this section. Figure 3(a) com- pares the electronic band structures of deformed phos- phorene with its undeformed counterpart, along particu- lar straight lines in k-space, according to which the sub- stantial influences of strain on both band spacing and curvature are evident. Figures 3(b) and (c) illustrate the strain dependence of the size and nature of the band gap, respectively. For the undeformed phosphorene, the band gap is cal- culated to be 0.95 eV (0.91 eV)- as obtained using the SIESTA (VASP) package- in excellent agreement with previous studies [9, 24]. Inspecting the nature of the band gap, our calculations based on both packages provide identical trends for the first conduction band in the vicin- ity of Γ-point, which is the exact position of the conduc- tion band minimum (CBM). For the first valance band, however, the actual placement of valance band maximum (VBM) slightly differs from SIESTA to VASP. While the SIESTA band structure predicts the VBM to be located precisely at Γ-point, the VASP package suggests an indi- rect band gap with its actual valance maximum occurring along the Γ− Y high-symmetry line, 0.0285×2π/a2 away from the Γ-point. The discrepancy can be attributed to the different calculation methods employed, i.e. the pseudopotential scheme combined with atomic orbitals in SIESTA versus projector augmented wave formalism with plane waves in VASP. Based on our calculations, a very marginal change of the overlap between atomic or- bitals would transform the nature of the band gap from direct to indirect and vice versa. In fact, a recent symme- try analysis on undeformed phosphorene has provided a criterion based on which the direct/indirect nature of the band gap can be determined [43]. The authors, however, mentioning the marginal discrepancy between the two cases, and further by referring to shortcomings ascribed to DFT-based calculations, did not provide a determined conclusion regarding the exact position of VBM. Thus, whether phosphorene is truly a direct or nearly direct semiconductor (as dubbed in Ref. [24]), we believe it should be left to experimental studies. For deformed structures, the maximum attainable di- rect (indirect) band gap is evaluated to be 1.34 eV (1.37 eV) which occurs at εx = 6%, εy = 3% (εx = 6%, εy = 4%). For the anti-symmetric case (εx = −εy) in the strain range under study (−9% (cid:54) εx, εy (cid:54) 9%), no semiconductor-metal transition can be triggered and a direct-indirect-direct-indirect transition is observed in the band gap. For symmetric deformations (εx = εy), the band gap experiences an indirect-direct-indirect-direct transition with a semiconductor-metal transition through the application of compressive strains larger than 6%. Figures 3(d) and (e) show the details of the variations in 5 FIG. 3. (a) Modification of the electronic band structure under various strain configurations. (b) The surface plot of (εx,εy) and the corresponding band gaps. The two dashed di- agonal lines denote the symmetric (Sym.) and anti-symmetric (A-Sym.) strain distributions. (c) Nature of band gaps for the same set of data presented in (b). Squares (circles) represent indirect (direct) band gaps. (d) and (e) show the VBM and CBM in symmetric and anti-symmetric strain distributions, respectively. Light (dark) gray regions correspond to direct (indirect) band gaps. Note that in (a) the Fermi energy is set to zero. In (d) and (e), the energies are referenced to vacuum level to further illustrate the modification of band offsets in strained structure. the location of the band gap for anti-symmetric and sym- metric strain distributions, respectively. Accordingly, in the symmetric case, both CBM and VBM undergo tran- sitions between Γ and Γ − X, giving rise to four dis- tinct strain zones with boundaries at -2% , 2%, and 4%. Inspecting the anti-symmetric case, while CBM expe- riences a transition similar to symmetric deformations, VBM moves between Γ, Γ − X, and Y − Γ, resulting in four strain zones with boundaries located at -6%, -3%, and 1%. Figure 4(a) shows the variation of the band structure under anti-symmetric strain. As seen, the band gap has almost vanished at εy=−εx= 11% and a linear dispersion emerges at the D point. Figure 4(b) shows the energy dispersion along both the Γ − D path and the direction -10-50105CbB-10-505100.20.01.00.80.40.61.2-10-50105CcB-10-50510Sym.A-Sym.Sym.A-Sym.x[a]x[a]y[a]y[a]CdB-10-50510x[a]-4-3-2-5Energyo[eV]Energyo[eV]CeB-10-50510x[a]Undeformedx==o5a-5a,yx==o5a5a,yEnergyo[eV]SYX-2-1021CaBΓΓEg[eV]metalCBM@ΓCBM@Γ-XVBM@ΓVBM@Γ-XVBM@Y-ΓA-Sym.metalSym.VBM@Y-ΓCBM@ΓCBM@Y-ΓVBM@ΓVBM@Γ-X-4-3-26 and conduction band crystal wave-functions around D over the constituent atomic orbitals s, px, py, pz, and d, are depicted in Figs. 4(c) and (d) for the selected strain of εy = −εx= 11%. Symmetry of the system mandates equal contributions to the crystal wave-function from all four atoms in the unit cell of phosphorene. For each atom, despite the prevalence of px, the contributions from other orbitals, especially those of s and d, should also be taken into account to properly describe the linear energy branches. To unravel the formation possibility of similar Dirac- shaped dispersions via strain, we performed a thorough inspection of band gaps, employing a dense network of data points (εx,εy) shown in Fig. 4(e). In this figure, the bright cyan (yellow) area denotes Dirac-shaped dis- persions with band gaps smaller than 5 (20) meV. Our calculations show the attainability of Dirac-like spectrum via invoking uniaxial deformations parallel to the zigzag axis. To illustrate this, the dashed line pertaining to σx = 0 is superimposed on the mesh. The cross marks on Fig. 4(e) denote selected strained lattice vectors for which the existence of a Dirac-like dispersion is further authenticated by VASP. The maximum discrepancy be- tween band gaps obtained from VASP and SIESTA in the selected geometries is 12.62 meV. We therefore conclude that, as far as DFT based simulations are concerned, our prediction regarding the existence of a Dirac-like disper- sion is valid. It should be mentioned that the asymmetric strain of εy = −εx = 11%, shown in Fig. 4(b), is provided as a sample exterior to the cyan region of Fig. 4(e), which still clearly manifests the Dirac-like feature. According to Fig. 4(e), although the band gap opens up beyond the cyan region, our calculations shows that the anisotropic Dirac-liked energy spectrum remains intact for band gaps of up to 55meV. For the emergence of Dirac-like spectrum deformations as large as 11% might be needed. There are now various practical schemes on how to incorporate strain into a 2D material. It has been reported that flex- ible substrates can be used to apply tensile axial strains of up to 30% to a graphene sheet [44]. Moreover, as the corresponding tensile yielding strains are calculated to be 27% and 30% along the zigzag and armchair axes, respec- tively, one can conclude that for strain magnitudes of up to 11%, phosphorene sheet will experience no detrimental plastic deformation and thus will preserve its structural integrity [25, 36]. Hence, the Dirac-like feature can defi- nitely lend itself to experimental verifications and prac- tical applications. Figure 4(f) illustrates the linear velocity, vk, and y−directed effective mass of Dirac-shaped dispersions as a function of εx. The calculated effective mass of carri- ers for both conduction and valence bands illustrates the parabolic nature of the energy spectrum along D−Dy for all the considered strain magnitudes. The variation in the effective mass along D − Dy is in the range of 0.10− 0.12 and 0.06 − 0.17 (in units of m0) for the conduction and valence bands, respectively. vk along Γ−D spans a range of 7.3 − 8.6 × 105 and 4.6 − 6.6 × 105 for the conduction (a) The band structures of phosphorene un- FIG. 4. der various A-Sym. strain values and the emergence of a Dirac-shaped dispersion. Blue, cyan, green, and red denote εy= −εx= 0%, 6%, 8%, and 11%, respectively. (b) Zoom of the band structure at εy= −εx= 11% along the selected paths of the first Brillouin zone. Inset shows the equi-energy con- tours for the conduction (top) and the valence (down) bands centered at D. Both rectangles span a length of 0.02 × π/ax (0.02× π/ay) along the kx (ky) direction of the first Brillouin zone. (c) and (d) denote the orbital composition of the crys- tal wave-functions close to the Dirac point for the topmost valence and the lowest conduction bands, respectively. (e) The mesh plot of the band gap at data points (εx,εy). Bright cyan (yellow) region denotes Dirac-shaped dispersions with band gap smaller than 5 (20) meV. The cross marks denote selected strain distributions for which the existence of a Dirac- like dispersion is further validated by VASP. (f) Fermi velocity and y−directed effective mass for Dirac-shaped dispersions as functions of εx. Crosses and filled circles denote the results for conduction and valence energy bands, respectively. perpendicular to it in the vicinity of D point. For both conduction and valence bands, despite the linearity along Γ − D, the bands are parabolic along the D − Dy path. In addition, the associated slope (curvatures) of the con- duction and valence bands along the Γ− D (D − Dy) di- rection are remarkably different. This can be further ap- proved by referring to the energy contours for both con- duction and valence bands (see the insets of Fig. 4(b)). Although the D point by itself is no longer a high sym- metry point, it still lies along lines of fairly high symme- try in the Brilloun zone. Decompositions of the valence -2-1012DΓyDEnergy[eV]-2-1012Γ(a)XDy=0%,6%,8%,10%y=0%,6%,8%,11%(b)DDΓDDyy0.20.100.10.2-6-14m* /m0yElectronHole-10x[%](f)81318-15-9-12-6y[%]x[%]200550100200400600Eg[meV](e)x= 0976458velocity [×10 5 m/s]-6-14-10Energy[eV]Occupancy00.20.40.60.81CBspxpypzd00.20.40.60.81Occupancy(c)(d)VBDD-D+DD-D+D-D+spxpypzdand valence bands, respectively. For the linear branches along Γ−D, it can be seen that the associated velocity of the conduction band is at least ×1.3 larger than that of the valence band, irrespective of the strain value. Noting that the Fermi velocities calculated for graphene, silicene and germanene, are 6.3×105, 5.1×105, and 3.8×105 m/s, respectively, it can be concluded that the Dirac-shaped dispersion of phosphorene is absolutely competitive with that of previously studied materials [45]. It is worthwhile to mention that a similar Dirac like dispersion has also been reported for 6,6, 12-graphyne, which has a rectangular crystal lattice [46, 47]. Of the two anisotropic Dirac cones in the first Brillouin zone of this material, the first one shows linear dispersion with the Fermi velocities of vkx=4.9× 105 m/s and vky=5.8× 105 m/s, while the second one is parabolic near the center of the cone. Moreover, the maximum attainable Fermi velocity of deformed 6,6, 12-graphyne is nearly 6.6 × 105 m/s when uniaxially strained about 7% along the x-axis. In comparison, phosphorene is different as it has distinct velocities pertaining to each of the two linear branches crossing at the D point, with both being more adjustable via in-plane strain engineering. IV. CONCLUSION In conclusion, a highly anisotropic mechanical response of phosphorene is revealed through the calculation of in and out-of-plane elastic constants. In particular, a nega- tive out-of-plane Poisson’s ratio is observed for uniaxial deformations along the zigzag direction. Compared to graphene and two-dimensional molybdenum disulphide, phosphorene is shown to possess a smaller (larger) in- plane stiffness (Poisson’s ratio) along both armchair and zigzag axes while offering comparable yielding strength. The vibrational frequencies of phosphorene are calcu- lated and the corresponding shifts are obtained in re- sponse to various biaxial strain distributions. With the ability of detecting Raman/infrared frequency shifts via high resolution Raman/infrared spectroscopies, our re- sults are of paramount importance for the characteriza- tion and mapping of strain distributions in phosphorene samples. By inspecting various strain distributions, it is shown that in-plane deformations strongly affect the size and nature of the band gap. In addition, strain is shown to significantly modulate the effective mass of both electrons and holes in phosphorene. Furthermore, we found that for specific deforma- tions, including the y-oriented uniaxial tension, a lin- ear energy spectrum with linear velocities comparable to those of other 2D semi-metal materials can be at- tained. The Dirac-like dispersion of deformed phospho- rene, however, is distinct from those previously reported for graphene, silicene, and germanene, as in phospho- rene the anisotropic dispersion allows carriers to behave as either massless Dirac fermions or massive charges, 7 FIG. 5. (a) and (b) ((c) and (d)) depict the effective mass of electrons (holes) along the x- and y-directions, respectively. The corresponding locations of the VBM and CBM are de- noted by circles (at Γ-point), crosses (along the Γ − X high- symmetry line), and squares (along the Γ− Y high-symmetry line). depending on the transport direction along the arm- chair or zigzag axes, respectively. Such an anisotropy in the linear velocity may trigger a corresponding direc- tion dependence in resistance, rendering phosphorene as a promising candidate for future nano-electronic device applications. It is highly desirable that we are able to manipulate the electronic structure of phosphorene via strain engineering as it increases a number of potential applications in nano-electromechanical as well as nano- optomechanical systems. Note added– During the last stage of preparing this manuscript, Ref. [48], where authors reported a substan- tial shift of Raman peaks via strain engineering in phos- phorene, appeared in arXiv. Appendix A: Variation of mass with strain Here, for the sake of completeness, we report our re- sults for the variation of the effective masses of both elec- trons and holes in all the strain configurations. The mesh plot of the effective masses at data points (εx, εy) are shown in Fig. 5. In addition, in this figure we denote the exact locations of the corresponding CBM (VBM) in which the effective mass of electron (hole) is calculated. As shown, the applied strain can widely tune the effec- tive mass of carriers. The discontinuities in the values of effective masses are also found to be at strain values in which direct-indirect band gap transitions take place. 00.020.040.060.080.10.120.140.160.180.211.5m*/m0x-10-50105(a)-10-50510Sym.A-Sym.x[%]y[%]metalmetal0.040.060.080.10.120.140.160.180.20.220.24m*/m0x-10-50105(c)-10-50510Sym.A-Sym.x[%]y[%]metalmetal00.10.20.30.40.511.523510m*/m0y-10-50105(d)-10-50510Sym.A-Sym.x[%]y[%]metalmetal00.10.150.31.261.271.31.351.4m*/m0y-10-50105(b)-10-50510Sym.A-Sym.x[%]y[%]metalmetal8 [1] K. S. Novoselov, A. K. Geim, S. Morozov, D. Jiang, Y. Zhang, S. Dubonos, I. Grigorieva, and A. Firsov, “Electric field effect in atomically thin carbon films,” sci- ence 306, 666 (2004). [2] S. Z. Butler, S. M. Hollen, L. Cao, Y. Cui, J. A. Gupta, H. R. Gutierrez, T. F. Heinz, S. S. Hong, J. Huang, A. F. Ismach, et al., “Progress, challenges, and opportunities in two-dimensional materials beyond graphene,” ACS nano 7, 2898 (2013). [3] A. K. Geim and I. V. Grigorieva, Nature(London) 499, 419 (2013). [4] Q. H. Wang, K. Kalantar-Zadeh, A. Kis, J. N. Coleman, and M. S. Strano, Nature Nanotech. 7, 699 (2012). [5] M. Xu, T. Liang, M. Shi, and H. Chen, Chemical reviews 113, 3766 (2013). [6] W. Lu, H. Nan, J. Hong, Y. Chen, C. Zhu, Z. Liang, X. Ma, Z. Ni, C. Jin, and Z. Zhang, “Plasma- assisted fabrication of monolayer phosphorene and its raman characterization,” arXiv preprint arXiv:1404.0742 (2014). [7] J. Qiao, X. Kong, Z.-X. Hu, F. Yang, and W. Ji, “High- mobility transport anisotropy and linear dichroism in few-layer black phosphorus,” Nat Commun 5 (2014), ar- ticle. [8] S. Das, W. Zhang, M. Demarteau, A. Hoffmann, M. Dubey, and A. Roelofs, “Tunable transport gap in phosphorene,” Nano letters 14, 5733 (2014). [9] H. Liu, A. T. Neal, Z. Zhu, Z. Luo, X. Xu, D. Tom´anek, and P. D. Ye, “Phosphorene: An unexplored 2d semicon- ductor with a high hole mobility,” ACS nano (2014). [10] V. Tran, R. Soklaski, Y. Liang, and L. Yang, “Layer- controlled band gap and anisotropic excitons in few-layer black phosphorus,” Phys. Rev. B 89, 235319 (2014). [11] T. Low, A. S. Rodin, A. Carvalho, Y. Jiang, H. Wang, F. Xia, and A. H. Castro Neto, “Tunable optical prop- erties of multilayer black phosphorus thin films,” Phys. Rev. B 90, 075434 (2014). [12] S. P. Koenig, R. A. Doganov, H. Schmidt, A. C. Neto, and B. Oezyilmaz, “Electric field effect in ultrathin black phosphorus,” Applied Physics Letters 104, 103106 (2014). [13] M. Buscema, D. J. Groenendijk, S. I. Blanter, G. A. and A. Castellanos- Steele, H. S. van der Zant, Gomez, “Fast and broadband photoresponse of few-layer black phosphorus field-effect transistors,” arXiv preprint arXiv:1403.0565 (2014). [14] H. Liu, A. T. Neal, M. Si, Y. Du, and P. D. Ye, IEEE (2014). [15] M. V. Kamalakar, B. N. Madhushankar, A. Dankert, and S. P. Dash, arXiv:1406.4476 (2014). [16] Y. Deng, Z. Luo, N. J. Conrad, H. Liu, Y. Gong, S. Na- jmaei, P. M. Ajayan, J. Lou, X. Xu, and P. D. Ye, ACS Nano 8, 8292 (2014). [17] S. Zhang, J. Yang, R. Xu, F. Wang, W. Li, M. Ghufran, Y. Zhang, Z. Yu, G. Zhang, Q. Qin, and Y. Lu, ACS Nano (2014). [18] G. Gui, J. Li, and J. Zhong, “Band structure engineer- ing of graphene by strain: First-principles calculations,” Physical Review B 78, 075435 (2008). [19] M. Topsakal, S. Cahangirov, and S. Ciraci, “The response of mechanical and electronic properties of graphane to the elastic strain,” Applied Physics Letters 96, 091912 (2010). [20] B. Wang, J. Wu, X. Gu, H. Yin, Y. Wei, R. Yang, and M. Dresselhaus, “Stable planar single-layer hexagonal sil- icene under tensile strain and its anomalous poisson’s ra- tio,” Applied Physics Letters 104, 081902 (2014). [21] M. A. Bissett, M. Tsuji, and H. Ago, Phys. Chem. Chem. Phys. 16, 11124 (2014). [22] Q. Yue, J. Kang, Z. Shao, X. Zhang, S. Chang, G. Wang, S. Qin, and J. Li, “Mechanical and electronic properties of monolayer mos2 under elastic strain,” Physics Letters A 376, 1166 (2012). [23] S. M. Tabatabaei, M. Noei, K. Khaliji, M. Pourfath, and M. Fathipour, “A first-principles study on the effect of biaxial strain on the ultimate performance of monolayer mos2-based double gate field effect transistor,” Journal of Applied Physics 113, 163708 (2013). [24] A. Rodin, A. Carvalho, and A. Neto, “Strain-induced gap modification in black phosphorus,” arXiv preprint arXiv:1401.1801 (2014). [25] X. Peng, A. Copple, and Q. Wei, “Strain engineered direct-indirect band gap transition and its mechanism in 2d phosphorene,” arXiv preprint arXiv:1403.3771 (2014). [26] R. Fei and L. Yang, “Strain-engineering the anisotropic electrical conductance of few-layer black phosphorus,” Nano letters (2014). [27] J. M. Soler, E. Artacho, J. D. Gale, A. Garc´ıa, J. Jun- quera, P. Ordej´on, and D. S´anchez-Portal, “The siesta method for ab initio order-n materials simulation,” Jour- nal of Physics: Condensed Matter 14, 2745 (2002). [28] G. Kresse and J. Furthmuller, Phys. Rev. B 54, 11169. [29] M. R. Hestenes and E. Stiefel, J. Res. Nat. Bur. Stand. 49, 409 (1952). [30] J. P. Perdew, K. Burke, and M. Ernzerhof, “Generalized gradient approximation made simple,” Physical review letters 77, 3865 (1996). [31] N. Troullier and J. L. Martins, Phys. Rev. B 43, 1993 (1991). [32] P. E. Blochl, Phys. Rev. B 50, 17953. [33] D. Alf`e, Compu. Phys. Communi. 180, 2622 (2009). [34] L. Cartz, S. R. Srinivasa, R. J. Riedner, J. D. Jorgensen, and T. G. Worlton, J. Chem. Phys. 71, 1718 (1979). [35] S. Grimme, J. Compu. Chem. 27, 1787 (2006). [36] Q. Wei and X. Peng, arXiv:1403.7882 (2014). [37] S. Appalakondaiah, G. Vaitheeswaran, S. Lebe e, N. E. Christensen, and A. Svane, Phys. Rev. B 86, 035105 (2012). [38] C. Lee, X. Wei, J. W. Kysar, and J. Hone, “Measure- ment of the elastic properties and intrinsic strength of monolayer graphene,” science 321, 385 (2008). [39] A. Castellanos-Gomez, M. Poot, G. A. Steele, H. S. van der Zant, N. Agraıt, and G. Rubio-Bollinger, “Me- chanical properties of freely suspended semiconducting graphene-like layers based on mos2,” Nanoscale research letters 7, 1 (2012). [40] G. Qin, Z. Qin, S. Yue, H. Cui, Q. Zheng, Q. Yan, and G. Su, arXiv:1406.0261 (2014). [41] J. Jiang and H. S. Park, arXiv:1403.4326 (2014). [42] Z. Zhu and D. Tom´anek, Phys. Rev. Lett. 112, 176802 (2014). [43] P. Li and I. Appelbaum, Phys. Rev. B 90, 115439 (2014). [44] J.-H. Ahn and J. H. Je, “Stretchable electronics: mate- rials, architectures and integrations,” Journal of Physics D: Applied Physics 45, 103001 (2012). [45] L. L. Y. Voon, E. Sandberg, R. Aga, and A. Farajian, “Hydrogen compounds of group-iv nanosheets,” Applied Physics Letters 97, 163114 (2010). [46] D. Malko, C. Neiss, F. Vin, es, and A. Gorling, Phys. Rev. lett. 108, 086804 (2012). [47] G. Wang, M. Si, A. Kumar, and R. Pandey, Appl. Phys. Lett. 104, 213107 (2014). [48] R. Fei and L. Yang, “Lattice vibrational modes and ra- man scattering spectra of strained phosphorene,” Ap- plied Physics Letters 105, 083120 (2014). 9 

1706.00670

1

1706

Topological antiferromagnetic spintronics: Part of a collection of reviews on antiferromagnetic spintronics

The recent demonstrations of electrical manipulation and detection of antiferromagnetic spins have opened up a chapter in the spintronics story. In this article, we review the emerging research field that is exploring synergies between antiferromagnetic spintronics and topological structures in real and momentum space. Active topics include proposals to realize Majorana fermions in an antiferromagnetic topological superconductors, to control topological protection of Dirac points by manipulating antiferromagnetic order parameters, and to exploit the anomalous and topological Hall effects of zero net moment antiferromagnets. We explain the basic physics concepts behind these proposals, and discuss potential applications of topological antiferromagnetic spintronics.

cond-mat.mes-hall

cond-mat

Topological antiferromagnetic spintronics: Part of a collection of reviews on antiferromagnetic spintronics Libor Smejkal,1, 2, 3 Yuriy Mokrousov,4 Binghai Yan,5 and Allan H. MacDonald6 1)Institute of Physics, Academy of Sciences of the Czech Republic, Cukrovarnicka 10, 162 00 Praha 6 Czech Republic 2)Institut fur Physik, Johannes Gutenberg Universitat Mainz, D-55099 Mainz, Germany 3)Faculty of Mathematics and Physics, Charles University in Prague, Ke Karlovu 3, 121 16 Prague 2, Czech Republic 4)Peter Grunberg Institut and Institute for Advanced Simulation, Forschungszentrum Julich and JARA, 52425 Julich, Germany 5)Department of Condensed Matter Physics, Weizmann Institute of Science, 7610001 Rehovot, Israel 6)Department of Physics, University of Texas at Austin, Austin, Texas 78712-0264, USA (Dated: 5 June 2017) The recent demonstrations of electrical manipulation and detection of antiferromagnetic spins have opened up a chapter in the spintronics story. In this article, we review the emerging research field that is exploring syn- ergies between antiferromagnetic spintronics and topological structures in real and momentum space. Active topics include proposals to realize Majorana fermions in an antiferromagnetic topological superconductors, to control topological protection of Dirac points by manipulating antiferromagnetic order parameters, and to exploit the anomalous and topological Hall effects of zero net moment antiferromagnets. We explain the basic physics concepts behind these proposals, and discuss potential applications of topological antiferromagnetic spintronics. 7 1 0 2 y a M 0 3 ] l l a h - s e m . t a m - d n o c [ 1 v 0 7 6 0 0 . 6 0 7 1 : v i X r a 1 Topologically protected states of matter are unusually robust because they cannot be destroyed by small changes in system parameters. This feature of topological states has suggested an appealing strategy to achieve useful quan- tum computation1,2. In spintronics, topological states provide for strong spin-momentum locking3, high charge- current to spin-current conversion efficiency4–6, large electron mobilities and long spin diffusion lengths7,8, strong magnetoresistance8, and efficient spin-filtering9. Materials exhibiting topologically protected Dirac or Weyl quasipar- ticles in their momentum-space bands, and those exhibiting topologically non-trivial real-space spin textures6,10, have both inspired new energy-efficient spintronics concepts4,7,10–13. In a topological insulator (TI), time reversal symmetry enforces Dirac quasiparticle surface states with spin- momentum locking (see Fig.1(a)) and protection against backscattering3. The much higher efficiency of magnetization switching by a current induced spin-orbit torque (SOT) in a TI/magnetically doped TI (MTI) heterostructure4,14 than in a heavy-metal/ferromagnet (FM) bilayer is thought to be associated with spin-momentum locking, and is a paradig- matic example of the potential seen for topological materials in spintronics. (A full microscopic understanding of the underlying current-spin conversion mechanism is however still absent15.) Progress in understanding and exploiting topological insulators in spintronics has so far been limited by unintentional bulk doping in TIs, and by the decreased stability of TI surface states at elevated temperatures15. The practical utility of the topologically enhanced SOT has also been limited by the cryogenic temperatures at which known MTIs order4,1416. A substantial rise in the critical temperature of a magnetic topological insulator (by a factor of 3 to 90 K) due to proximity coupling to adjacent antiferromagnet (AF) has recently been demonstrated in a heterostructure consisting of the metallic antiferromagnet CrSb sandwiched between two MTIs17. Increased SOT efficiency at heterojunctions between TI's and ferrimagnetic CoTb alloys containing antiferromagnetically coupled Co and Tb sublattices18 has also been reported. The later effect persists to room temperature, but with decreased efficiency enhancement15 at higher temperatures. Research on using antiferromagnetism to achieve a role for topological materials in spintronics is however still at an early stage and many ideas have so far only been addressed theoretically. The practical advantages of TIs over heavy-metal systems for spin-orbit torques, for example, are not yet established15. The forms of magnetism so far incorporated in magnetic TIs remain fragile because they are of interfacial17,19 or dilute-moment character14. Other new ideas, beyond simply making topological insulators magnetic, are emerging at a rapid pace. In this article we review topological antiferromagnetic spintronics, the emerging field that is exploring the interplay between transport, topological properties in either momentum space or real space, and antiferromagnetic order. I. TOPOLOGICAL INSULATORS IN ANTIFERROMAGNETS AND MAJORANA FERMIONS The roots of topological antiferromagnetic spintronics can be traced to studies of layered AFs of the SrMnBi2 type, which were thought to feature quasi 2D massive Dirac quasiparticles around the Fermi level20,21. These were associated with the observation of enhanced mobilities, similarly to those in graphene. Masuda et al.22 have demonstrated manipulation of the Dirac quasiparticle current and the quantum Hall effect in a EuMnBi2 AF by an applied strong magnetic field, with the effect of the field mediated by Eu sublattices. As pointed out by Mong et al.23, TI phases are possible in antiferromagnets even though time-reversal symmetry T is broken and are protected instead by T 1 is a half magnetic unit cell translation operation, as we illustrate in Fig.1(a). The proposed low-temperature AF candidate, GdPtBi, has not yet been confirmed as a TI by angle resolved photoemission spectroscopy (ARPES), presumably due imperfect crystal momentum resolution of the measurement24. A path of research related to topological superconductivity has demonstrated signatures of the coexistence of a 2D TI, i.e. the quantum spin Hall effect (QSHE), and a superconducting state in hole-doped and electron-doped antiferromagnetic monolayers of FeSe25. FeSe is the metallic building block of the iron-based high-TC superconductors. Surprisingly, the combined effect of substrate strain, spin-orbit coupling, and electronic correlations was shown to induce band inversion and QSHE edge states25. These can in turn lead to the realization of Majorana zero modes, superconducting quasiparticle states with real wavefunctions that prevents decoherence, providing one route for the realization of quantum computing2,26 with topological qubits. The antiferromagnetic TI in combination with superconductivity can allow for an alternative realization of the Fu-Kane Majorana fermion proposal27, possibly at higher temperatures25. Separately, QSHE states in an AF have also been predicted in honeycomb lattice systems28. T where T 1 2 2 II. TOPOLOGICAL SEMIMETAL ANTIFERROMAGNETS Topological semimetal states arise when conduction and valence bands touch at discrete points, lines, or planes in a bulk Brillouin zone at energies close to the Fermi level. The low energy physics of topological semimetals is governed by effective Dirac or Weyl equations11,13,29. 3D Dirac and Weyl quasiparticles in non-magnetic bulk systems have attracted attention because of reports of suppressed backscattering, measurements of exotic topological surface states, 2 FIG. 1. (BOX:) Overview of the materials palate of topological antiferromagnetic spintronics: All panels show energy vs. crystal momentum band diagrams and magnetic structures. (a) Schematics of the spin-momentum locked surface- state Dirac quasiparticles in a TI. In the inset, we show the magnetic Gd sublattice of GdPtBi, an AF TI candidate. The T symmetry that protects a TI state in an AF. (b) Schematics of the bulk Dirac quasiparticles of magenta arrow marks the T 1 2 an AF Dirac semimetal (DSM) that must be located along a high symmetry line in the BZ, and the magnetic structure of the AF DSM candidate CuMnAs. The PT symmetry connects the magnetic Mn sublattices in pairs, and together with additional crystalline symmetry protects the DSM state as we explain in Fig.2(b). (c) Massive Dirac quasiparticles and QAHE edge states (red dispersion). A quantized Hall conductivity can be produced by the spin-chirality, χijk = Si ·(cid:16) Sj × Sk , of a non-coplanar (cid:17) spin texture. The quantized topological Hall effect (QTHE) in a non-coplanar insulating AF gives rise to quantized edge states as in the QAHE, as we explain in the text. For the sake of simplicity, the non-coplanar spins of the antiferromagnetic QTHE candidate K0.5RhO2 are translated to a common origin. (d) Weyl points act as sources and drains of Berry curvature (blue and red arrows). In a centrosymmetric WSM, the topological charges are distributed antisymmetrically around the Γ point in the BZ. The noncollinear magnetic sublattice of antiferromagnetic Mn3Ge forms kagome planes and has inversion symmetry P. The pseudorelativistic linear band crossings in the left panels are found in the edge/surface states, whereas those in the right panels are realized in the bulk. The TI and DSM states ( the two upper panels) cannot be realized in systems with ferromagnetic order, but are possible in antiferromagnets. In contrast, the QAHE/QTHE and WSM states (the two lower panels) are expected in both FMs and AFs, and can be formally obtained by breaking symmetries of the phases depicted in the two upper panels. and interest in unique topological responses such as dissipationless axial currents.8,11,30. These properties are thought to be responsible for experimental observations of chiral magnetotransport12,31, and strong magnetoresistance32,33, although the topological origin of these phenomena is not yet firmly established8. For instance, the strong magnetore- sistance in WTe2 semimetals was originally explained on the basis on the carrier compensation in the tiny electron-hole pockets at the Fermi level34, and only later linked to the presence of Weyl fermions33. A. Topological metal-insulator transitions in 3D Dirac semimetal antiferromagnets In a system with time reversal T and spatial inversion P symmetries, the electronic bands are doubly degenerate resulting in a low energy Dirac Hamiltonian, HD(k). In its simplest form11,13,29: (cid:18)vF k · σ m HD(k) = (cid:19) m −vF k · σ where σ is the vector of Pauli matrices, vF is the Fermi velocity, k = q − q0 is the crystal momentum measured from the Dirac point at q0, and m is the mass (in units of energy). The corresponding energy dispersion reads Ek = ±vF )2. The mass can vanish only when turned-off by a crystalline symmetry, and in (cid:113) k2 x + k2 y + k2 z + ( mvF 3 , (1) this case HD(k) describes the four-fold degenerate band touching11,13 of a 3D Dirac semimetal (DSM) illustrated schematically in Fig.1(b). In a 3D DSM, the topological invariants and nontrivial surface states can be linked to the crystalline symmetry protecting the degeneracy35,36. The 3D DSM state is not possible in FMs because T -symmetry breaking prevents double band degeneracy. On the other hand, a topological crystalline 3D Dirac semimetal was predicted in an AF, namely in the orthorhombic phase of CuMnAs37,38. Here P and T symmetries are absent separately, but the combined effective PT symmetry ensures double band degeneracy over the whole Brillouin zone. The DSM is in this case protected by PT symmetry together with an additional crystalline non-symmorphic symmetry, as we explain in Fig. 2. The orthorhombic CuMnAs AF provides an attractive hydrogen atom for magnetic DSMs induced by the band inversion, since only a single pair of Dirac points appears near the Fermi level of the ab initio band structure. Electron filling enforced semimetals with a single Dirac cone are also a possibility as indicated theoretically in two dimensional model AFs39. FIG. 2. (BOX:) Topological metal-insulator transition in the antiferromagnetic DSM CuMnAs. (a) The crystal and magnetic structure of the orthorhombic CuMnAs breaks time reversal and spatial inversion symmetry but preserves the combination PT which connects the A − B and C − D Mn sublattices (Cu and As atoms are omitted for brevity). In the presence of an electrical current, owing to the PT symmetry, a non-equilibrium staggered spin-polarization, δs, is generated that can reorient the antiferromagnetic moments. (b) For a N´eel vector along the [001] axis, the crystal has a screw rotation symmetry Sz which transforms the atom A into the atom C by a π-rotation along the z-axis followed by a ( 1 2 )-unit cell translation. Sz prevents hybridization of doubly degenerate bands that have opposite eigenvalues of Sz distinguished by " + " and " − " labels, and protects the DSM phase, as seen in the ab initio band structure. (c) For the [100] orientation of the N´eel vector, the Sz symmetry is broken and the Dirac fermions acquire a mass giving rise to a semiconducting phase. This N´eel vector reorientation controlled transition is referred to as the topological metal-insulator transition and can lead to extremely large anisotropic magnetoresistance. 2 , 0, 1 Novel effects have been predicted in topological DSM AFs that are based on the possibility of controlling topological states by controlling only N´eel vector orientation, not the presence or absence of antiferromagnetic order, and this can be accomplished using current-induced spin-orbit torques. The latter effect, discussed in detail by Zelezny et in this focused issue, has been experimentally demonstrated in CuMnAs40. The coexistence of Dirac fermions al. and spin-orbit torques in CuMnAs implies a new phase transition mechanism, referred to as the topological metal- insulator transition (MIT)38. The origin of the effect is in Fermi surface topology that is sensitive to the changes in the magnetic symmetry upon reorienting the N´eel vector, as explained in Fig. 2. The transport counterpart of the topological MIT is topological anisotropic magnetoresistance (AMR), which in principle can reach extremely large values38. The topological AMR can be understood as a limiting case of the crystalline AMR. The effect is different in origin and presumably more favorable for spintronics than the MIT observed in the pyrochlore iridate family which is driven by combined correlation and external field effects41, or the extreme magnetoresistance observed in the AF topological metal candidate NdSb42. An antiferromagnetic Dirac nodal line semimetal has also been proposed38. Here the band touching lines are protected by off-centered mirror plane symmetries and lead to distinct physical properties such as drum head surface states. Since the nodal-lines were observed several eV deep in the ab initio Fermi sea of tetragonal CuMnAs, the search is still on for more favorable candidate AF materials featuring nodal lines closer to the Fermi level. B. Weyl fermions in antiferromagnets When P or T symmetry, or both, is broken and the double band degeneracy is lifted, the touching points of two non-degenerate bands can form a 3D Weyl semimetal (WSM), see Fig.1(d). Fermi states in a WSM are described by 4 the Weyl Hamiltonian11,13: H(k) = ±vFk · σ. Weyl points act as monopoles sources of Berry curvature flux and generate a topological charge defined by: (cid:90) C = 1 2π δS d2k n · (cid:104)∂kuk × ∂kuk(cid:105) = ±1 . (2) (3) Here δS is a small sphere surrounding the Weyl point with the surface normal vector n and Ωk = (cid:104)∂kuk × ∂kuk(cid:105) is the momentum space Berry curvature, which can be viewed as a fictitious magnetic field43. In the vicinity of the Weyl point the Berry curvature takes the monopole form, Ωk = ±k/(2k3). Weyl points always come in pairs with opposite topological charges and do not rely on any specific symmetry protection. The only way to remove them is to annihilate two Weyl points with opposite topological charges. This is in contrast to the DSM case in which gaps can open also due to the hybridization among the degenerate band partners when symmetries are weakly broken, as we explained in the previous section. The 3D nature of the Weyl point is crucial here since the corresponding Weyl equation uses all three Pauli matrices. Consequently, any small perturbation, that is in general expressed as a linear combination of Pauli matrices that form the basis of the 2 × 2 Hilbert space, just shifts but does not gap the Weyl point. (For example, for a perturbation of a form mσz, the dispersion is renormalized as Ek = ±vF (cid:113) )2.) k2 x + k2 y + (kz + mvF The initial prediction of the WSM by Wan et al44 was for pyrochlore AFs. Subsequently, a Weyl metal state was predicted in the chiral non-collinear centrosymmetric AF Mn3Ge (see Fig.1(c)), a metallic system in which trivial Fermi surface pockets are also present45. Figs.3(a,b) illustrates the density of states weighted by the surface contribution which exhibits the typical Fermi arc features (see also Fig.1(d)), open surface state constant energy surfaces that connect the bulk projections of Weyl points, as found in ab initio calculations. We note that Weyl semimetal states can be also be realized in the paramagnetic and AF phase of GdPtBi12,31 by applying a magnetic field and, in contrast to DSMs, WSMs can in principle exist also in FMs46. The presence of Weyl points in the pyrochlore AF Eu2Ir2O7 47 was inferred from optical experiments. However, a direct spectroscopic identification (e.g. by ARPES) of these Weyl AFs has yet to be made29,48. The YbMnBi2 AF was originally proposed to be WSM based on the ARPES identification of the bandcrossings consistent with the ab initio bandstructure calculated for the canted AF moments49, later supported by an optical study50. However, recent magnetotransport51 and optical conductivity52 measurements in the YbMnBi2 AF were shown to be rather consistent with the Dirac qusiparticles29. The Mn3Ge AF was shown to host a large anomalous Hall effect (AHE) whose origin is discussed in the next section. III. HALL EFFECTS IN NONCOLLINEAR TOPOLOGICAL ANTIFERROMAGNETS Until recently the AHE was viewed as a combined consequence of the time reversal symmetry breaking in a T symmetry or PT symmetry forces the ferromagnet and spin-orbit coupling55. In the case of collinear AFs, either T 1 Hall conductivity to vanish. Recent ab initio calculations56,57 inspired by earlier theoretical work58,5960, have however shown that time-reversal symmetry breaking by AF order can yield a finite Hall response in some non-collinear AFs, even those with zero net magnetization and even in the absence of spin-orbit coupling. The time reversal symmetry breaking is manifested by a nonzero Berry curvature, as we show in Fig.3(d)). The intrinsic contribution to the Hall conductivity depends only on the band structure of the perfect crystal and can be calculated from linear response theory55: 2 (cid:90) (cid:88) n BZ σxz = e2  d3k (2π)3 f (k)Ωn y (k), (4) where Ωn index. y (k) is the y-component of the fictitious magnetic field, or the Berry curvature (cf. Eq.(3)), and n is the band A. The Anomalous Hall effect in antiferromagnets In the simplest toy model of a WSM (with a two Weyl points in the vicinity of the Fermi level), the AHE conductivity can be calculated by integrating the quantized two-dimensional Hall conductivities of momentum-planes that are 5 (a) The density of states weighted by the surface FIG. 3. Weyl fermions and Hall effects in noncollinear AFs. contribution calculation reveals the Fermi surface with visible Fermi arcs. The Berry curvature vector field around the two Weyl points with opposite charges is plotted in two lower insets. (b) Weyl points in Mn3Ge close to the Fermi level are connected by surface state Fermi arcs as calculated ab initio. (c) The largest contribution to the intrinsic AHE originates from avoided crossing lines where the fictitious magnetic field, i.e. the Berry curvature, takes the largest value. (d) The concept of magnetic memory bits in a non-collinear AF Mn3Ge. The measured large anomalous Hall effect at the room temperature originates in breaking the effective time reversal symmetry combining time reversal and mirror plane symmetries by the non- collinear magnetic order, as shown in the inset on the magnetic structure of Mn3Ge. The magnetic information can be stored in the orientation of the non-collinear AF order since a reorientation by, e.g., a weak magnetic field B[010] can flip the sign of the anomalous Hall voltage VH (and the corresponding conductivity σxz) for the applied current along the [001] direction. Temperature dependence of (e) the AHE conductivity in Mn3Ge. The Figs.(a-b, and e) were adapted from45,54. perpendicular to the line connecting Weyl points to obtain, σxy = e2 h ∆kW 2π , (5) where ∆kW is the distance between Weyl points61. The AHE was recently observed in the hexagonal noncollinear AFs Mn3Sn and Mn3Ge54,62,63, which have Weyl points close to the Fermi level. However, ab initio calculations of the intrinsic AHE in Mn3Ge, which predict a magnitude consistent with experiment (see Fig.3(e)), reveal that the domi- nant contribution to the AHE comes instead from localized avoided crossings in the band structure53. This property is illustrated in Fig.3(c) by calculating crystal momentum projected contributions of the Berry curvature/conductivity dkzΩy(kx, ky, kz). We also note that a large AHE was achieved in the collinear AF GdPtBi, by canting the ∼ (cid:82) kz staggered order64. The discovery of a large AHE Mn3Ge, which is a metal but has a relatively small density-of-states at the Fermi level, inspires a search for the quantized and dissipationless limits of the anomalous transport in topological semicon- ducting/insulating AFs. A quantum anomalous Hall effect (QAHE) with quantized charged edge state channels was proposed for antiferromagnetic insulators65,66. In a two-band 2D Chern insulator with Hamiltonian Hk = d(k) · σ, the 2D variant of Eq.(4) reads dkxdky d × ∂ky ∂kx d , (6) (cid:90) (cid:88) n BZ σxy = e2 h 1 4π d ·(cid:16) (cid:17) where d(k) = d(k)/d(k). The integral is quantized and relates directly to the Chern number (c.f. Eq.(3)). While the AHE arises from the Berry curvature in the momentum space, other important spintronic phenomena can be associated with Berry curvatures in different parameter spaces. For instance, the spin-orbit torkance tensor is defined by the linear response relation T = τ E, where T = dm is the SOT exerted on the magnetization m in a dt magnet subject to an applied electric field E. The intrinsic part of the SOT can be rewritten in terms of a mixed Berry curvature, Ω mk (cid:11), where ei denotes the ith Cartesian unit vector and m is a ij = ei · 2Im(cid:80) (cid:10)∂ mukn∂kj ukn n 6 unit vector in the direction of magnetization67. A large SOT in a topologically nontrivial insulating FM has been associated with the existence of the monopoles of the mixed Berry curvature. These are termed mixed Weyl points as they correspond formally to a Weyl Hamiltonian H(k, m) = vF (kxσx + kyσy) + vθθσz in the mixed momentum- magnetization space67. (Here θ is the azimuthal angle of the magnetization.) The recent discovery of the SOT and the prediction of the DSM in antiferromagnetic CuMnAs motivates a search for analogous dissipationless (pronounced) SOTs in insulating AFs. Finally, we note that spintronics based on insulating AFs may utilize the concept of Weyl magnons (a boson analog of Weyl fermions), that has been proposed in pyrochlore AFs68. B. Topological Hall effect in antiferromagnets spin texture (see inset in Fig.1(c)), χijk = Si ·(cid:16) Sj × Sk Real-space order parameter textures can be induced in AFs, and their presence can be detected by the so-called topological Hall effect. In this phenomenon, the role of the spin-orbit coupling is substituted by the chirality of the . (Note that the spin-chirality vanishes in coplanar AFs like Mn3Ge.) The effect of the corresponding fictitious magnetic field, m· (∂x m × ∂y m), on the Bloch electrons generates a Hall response. The topological Hall effect can be experimentally distinguished from the AHE by, e.g., analyzing the disorder dependence69. However, the distinction at surfaces might be difficult as was pointed out in studies of monolayer Fe deposited on an Ir(001) surface70. (cid:17) The Hall effect associated with the spin-chirality was initially reported in antiferromagnetic pyrochlore iridates71 and later in MnSi chiral antiferromagnetic alloys72,73. We note that the term "topological" used to label the effect does not imply in this case a correspondence to a topological invariant. This is in contrast to skyrmions discussed in more detail in the following section. A skyrmion spin texture carries an integer topological charge which is accompanied with a topological Hall effect. In this case, the term topological refers to the association of the Hall response to a topological invariant. Another example of such a correspondence is the quantum topological Hall effect which was 74, as we show schematically in Fig.1(d). Unlike skyrmions, here the proposed for the non-coplanar AF K0.5RhO2 topological charge occurs in the momentum space, i.e., is obtained from Eq.(4) giving an integer value. Finally, topological systems were predicted also as promising generators of the spin Hall effect (SHE). The WSM TaAs was predicted to host a large spin Hall angle arising mainly from the nodal line anticrossing features75. Similarly, a topological SHE was predicted for skyrmions76 where the spin Hall response occurs even in the absence of the spin- orbit coupling, in analogy with the above topological (charge) Hall effect. Recently, it was theoretically proposed that the SHE can occur also due to the breaking of the spin rotational symmetry in non-collinear AFs without the need for either the spin-orbit coupling or the spin-chirality77. In noncollinear AFs, other "topological" phenomena might arise, such as the topological orbital magnetism in Fe/Ir(001), Mn/Cu(111) or in the γ-phase of FeMn alloy70,78,79. The effect is rooted in the spin-chirality of non- coplanar spin textures, exists also without the spin-orbit coupling, and does not have to be associated with an integer topological invariant79. Novel functionalities can be envisaged by controlling the topological orbital magnetism via spin-torque manipulation of the spin textures or via the spin-orbit coupling79. IV. ANTIFERROMAGNETIC SKYRMIONS As noted above, position-dependent magnetization textures can also be topologically non-trivial. Skyrmions are noncollinear magnetization textures in which the spin quantization axis changes continuously over length scales that vary from a few nm to a few µm. For two-dimensional systems, the winding number, (cid:90) Q(j) = 1 4π dxdy m(j) ·(cid:16) ∂x m(j) × ∂y m(j)(cid:17) , (7) of a magnetization texture measures the number of times the sphere of magnetization directions is covered upon integrating over space and must take on integer values10,82. Here m = m(x, y, z) is the normalised magnetization field in the real space and m· (∂x m × ∂y m) is the fictitious emergent magnetic field. The antiferromagnetic skyrmion can be visualized as two interpenetrating ferromagnetic skyrmions, where the index (j) = (1, 2) labels the two antiferromagnetic sublattices, as shown in Fig.4(a). Microscopically, the skyrmionic magnetization modulation is caused by the Dzyaloshinskii-Moriya interaction (DMI) of non-centrosymmetric crystals. Whereas ferromagnetic skyrmions are often generated by interfacial DMIs, antiferromagnetic skyrmions are expected to be more abundant in crystals with bulk DMI83. By comparing to Eq.(3), and (6) we see that the winding number Q topologically protects skyrmion textures in real space, just as Weyl points and the QAHE state are protected in momentum space. The observed energy barrier 7 FIG. 4. Antiferromagnetic skyrmions. (a) An antiferromagnetic skyrmion can be viewed as consisting of two antiferro- magnetically coupled ferromagnetic skyrmions. For the sake of clarity, we draw the two opposite magnetic moments in the antiferromagnetic unit cell as coinciding and their moments as perfectly compensated. Note that the structure of the skyrmion is analogous to the momentum-space Berry curvature shown in the inset of Fig.3(a). (b) A synthetic antiferromagnetic skyrmion in a Fe-Cu-Fe trilayer. (c) Micromagnetic simulation of ferromagnetic (upper panel) and antiferromagnetic (lower panel) skyrmion motion driven by a SOT. The ferromagnetic skyrmion is deflected by the Magnus force, while the antiferromagnetic skyrmion can move in a straight line due to mutual compensation between Magnus forces from the two magnetic sublattices, as schematically shown in panel (a). Panels (b)-(c) are adapted from Refs. 80 and 81. for skyrmion annihilation in discrete magnetic skyrmions is of the order of ∼0.1 eV84. Because this barrier is finite, skyrmion stability in experimental systems relies in part on other physical limitations, (e.g. a combined effect of spin rotation and skyrmion diameter shrinking) rather than from topological protection itself84. Skyrmions in non-centrosymmetric AFs were considered already some time ago85 in various contexts, including in connection with high-temperature superconductivity86. However, spintronics aspects of antiferromagnetic skyrmions, namely their manipulation by an electrical current, were discussed only recently81,87–89. Micromagnetic simulations show that antiferromagnetic skyrmions move faster than ferromagnetic skyrmions, can be driven with lower current densities and, most importantly, move in straight lines, as illustrated in Fig. 4(c)81,87,88. This important difference arises because the Magnus force which deflects ferromagnetic skyrmions has opposite sign for the two magnetic sublattices of an antiferromagnetic skyrmion, owing to the opposite topological numbers illustrated in Fig.4(a). AF in a FeCuFe trilayer) in which skyrmions in skyrmions were recently studied in detail also in synthetic AFs (e.g. the two ferromagnetic layers are coupled antiferromagnetically80,90, as shown in Fig. 4(b). The topological SHE was suggested as a probe to monitor the AF skyrmions, as well as to generate a spin current80. V. PERSPECTIVES The past year has seen important progress in coupling topological phases of matter with AF order. The fortunate lattice constant match between the Cr-doped TI (Bi,Sb)2Te3, and the high temperature AF CrSb (see Fig.5(a)) has been exploited to grow epitaxial interfaces between these materials. The resulting heterostructures exhibit strength- ened order in the MTI17, enhancing topological effects in its electronic properties. CrSb/(Bi,Sb)2Te3 /CrSb trilayers exhibit cusps in the magnetoresistance, that presumably correspond to a topological phase transition of Dirac quasi- particles at the interfaces92. Separately, in studies of Bi2Se3/CoTb heterostructures15, it was also shown that a SOT from a TI can manipulate the AF-coupled sublattices in an adjacent ferrimagnet. Further progress in which perfectly compensated antiferromagnetic materials are employed can enable the advantages of AF spintronics, dis- cussed throughout this issue, to be realized more fully. AF FeSe monolayers will likely be tested for the realization of Majorana-based topological quantum qubits. Making a p-n junction in a single layer of FeSe by gating can generate two regions - one superconducting and one with a QSHE26. Coupling this system to two FM electrodes from both sides would lead to localization of Majorana modes at the interface as illustrated in Fig.5(b). This two level state can function as a quantum bit which can nonlocally encode information and is robust against decoherence due to the real wavefunction of the Majorana modes2. In this brief review, we have focused on direct synergies between antiferromagnetic and topological properties in crystal momentum and real spaces. Novel magnetic topological phases of matter were predicted only very recently and in many cases remain experimentally elusive, e.g. AF TIs, DSMs, and WSMs, QAHE states, and skyrmions. When realized, these topological antiferromagnet states can lead not only to more stable topological nanospintronics devices, but also to unprecedented functionalities relying on the unique AF symmetries and the possibility of tuning them by 8 FIG. 5. Topological spintronics based on antiferromagnets. (a) Example of a possible building block for topological AF spintronics in heterostructures: an epitaxially matched interface between AF CrSb and a Cr-doped MTI. (b) A Majorana qubits in an AF FeSe monolayer. (c) A concept of a topological transistor explained in the text. (d) The skyrmion racetrack memory concept. Magnetic information is stored in a topological charge Q instead of the magnetization of the magnetic domain. "1" encodes the skyrmion with Q=1, while "0" corresponds to a uniform domain with Q=0. Spin transfer torque shifts the register. Bits read by the topological Hall effect and written by nucleating or deleting the skyrmion. Panels (a), and (d) are adapted from Refs. 17 and 91. external means by coupling to the AF order. Fast topological memories, in which states are written by the topological SOT in an AF TI, or an AF DSM38,67,93, and read out via the large magnetoresistance effects associated with band gap tuning38,94,95, are among anticipated applications. Another possibility is opened by exploiting topological phase transitions, as we have explained for the CuMnAs Dirac AF in the Fig.2. Here one can foresee the possibility of a topological transistor operating at high frequencies and low current densities (see Fig.5(c))96. In contrast to the related proposal in crystalline topological insulators97, one can achieve highly mobile bulk Dirac quasiparticle current (present in the "ON" state) controllable by gating or by ultrafast SOT. We note that many of the novel effects we have discussed follow directly from antiferromagnetic symmetries and cannot be realized in FMs, for instance (i) magnetism combined with the QSHE, and superconductivity, and (ii) magnetism combined with Dirac semimetal phase. The conditions for a good Dirac quasiparticle in AF spintronics were discussed recently13. The sign and magnitude of the AHE in Mn3Ge depends on the noncollinear spin texture orientation. This together with the demonstration of the possibility of manipulating the noncollinear spin texture by a spin-torque98 can allow for memory devices in noncollinear AFs, with the electrical readout via the AHE, as illustrated in Fig.3(d). Moreover, optical counterparts of the dc AHE should be present in non-collinear AFs99, opening the prospect for optical detection of topological effects and of antiferromagnetic opto-spintronic devices. The nontrivial topologies of magnetization texture might also find applications. The skyrmion might represent the smallest micromagnetic configuration for storing information before going into truly quantum mechanical single spin qubits6. For instance, in the skyrmionic racetrack memory, as shown in Fig. 5(d), the magnetic information is stored in skyrmions instead of magnetic domains separated by domains walls10. Skyrmions can be driven by the SOT at low current densities and have advantages over domain walls especially in the curved parts of the race track thanks to their stability. After almost a half century of research we have discovered many manifestations of topology playing a role in materials physics, from spin liquids, to Quantum Hall effects, to the theory of dislocations100 and beyond. In this article, we have highlighted a newly emerging field, topological antiferromagnetic spintronics. Beyond providing an interesting new context in which to identify and understand the physical consequences of topological properties in momentum-space bands or real-space textures, topological antiferromagnetic spintronics suggests the tantalizing possibility of converting important fundamental advances to truly valuable new applications of quantum materials. Acknowledgement LS acknowledges support from the Grant Agency of the Charles University, no. 280815. Access to computing and storage facilities owned by parties and projects contributing to the National Grid Infrastructure 9 MetaCentrum provided under the program Projects of Large Research, Development, and Innovations Infrastructures (CESNET LM2015042) is greatly appreciated. YM acknowledges funding from the German Research Foundation (Deutsche Forschungsgemeinschaft, Grant No. MO 1731/5-1). B.Y. acknowledges support of the Ruth and Herman Albert Scholars Program for New Scientists in Weizmann Institute of Science, Israel. AHM was supported by SHINES, an Energy Frontier Research Center funded by the U.S. Department of Energy, Office of Science, Basic Energy Sciences under Award #SC0012670, and by Welch Foundation grant TBF1473. REFERENCES 1S. D. Sarma, M. Freedman, and C. Nayak, "Majorana Zero Modes and Topological Quantum Computation," Nat. Publ. Gr. (2015), 10.1038/npjqi.2015.1, arXiv:1501.02813. 2C. W. J. Beenakker and L. Kouwenhoven, "A road to reality with topological superconductors," Nat. Phys. 12, 618–621 (2016), arXiv:1606.09439v1. 3M. Z. Hasan and C. Kane, "Colloquium: Topological insulators," Rev. Mod. Phys. 82, 3045–3067 (2010). 4Y. Fan and K. L. Wang, "Spintronics Based on Topological Insulators," SPIN 06, 1640001 (2016). 5H. Wang, J. Kally, J. S. Lee, T. Liu, H. Chang, D. R. Hickey, K. A. Mkhoyan, M. Wu, A. Richardella, and N. Samarth, "Surface-State-Dominated Spin-Charge Current Conversion in Topological-InsulatorFerromagnetic- Insulator Heterostructures," Phys. Rev. Lett. 117, 076601 (2016). 6A. Soumyanarayanan, N. Reyren, A. Fert, and C. Panagopoulos, "Spin-Orbit Coupling Induced Emergent Phe- nomena at Surfaces and Interfaces," Nature 539, 509–517 (2016), arXiv:1611.09521. 7D. A. Pesin and A. H. MacDonald, "Spintronics and pseudospintronics in graphene and topological insulators." Nat. Mater. 11, 409–416 (2012). 8T. Liang, Q. Gibson, M. N. Ali, M. Liu, R. J. Cava, and N. P. Ong, "Ultrahigh mobility and giant magnetoresistance in the Dirac semimetal Cd3As2," Nat. Mater. 14, 280–284 (2014). 9J. Wu, J. Liu, and X. J. Liu, "Topological spin texture in a quantum anomalous hall insulator," Phys. Rev. Lett. 113, 136403 (2014), arXiv:1401.0415. 10A. Fert, V. Cros, and J. Sampaio, "Skyrmions on the track," Nat. Nanotechnol. 8, 152–156 (2013). 11A. A. Burkov, "Topological semimetals," Nat. Mater. 15, 1145–1148 (2016), arXiv:1610.07866. 12C. Felser and B. Yan, "Weyl semimetals: Magnetically induced," Nat. Mater. 15, 1149–1150 (2016). 13L. Smejkal, T. Jungwirth, and J. Sinova, "Route Towards Dirac and Weyl Antiferromagnetic Spintronics," Phys. Stat. Sol. (2017), 10.1002/pssr.201700044, arXiv:1702.07788. 14Y. Fan, P. Upadhyaya, X. Kou, M. Lang, S. Takei, Z. Wang, J. Tang, L. He, L.-T. Chang, M. Montazeri, G. Yu, W. Jiang, T. Nie, R. N. Schwartz, Y. Tserkovnyak, and K. L. Wang, "Magnetization switching through giant spin-orbit torque in a magnetically doped topological insulator heterostructure." Nat. Mater. 13, 699–704 (2014). 15J. Han, A. Richardella, S. Siddiqui, J. Finley, N. Samarth, and L. Liu, "Room temperature spin-orbit torque switching induced by a topological insulator," (2017), arXiv:1703.07470. 16A recent report19 of interfacial ferromagnetism persisting to room temperature at a insulating ferromagnet (EuS) /TI heterostructure is promising in this respect. 17Q. L. He, X. Kou, A. J. Grutter, G. Yin, L. Pan, X. Che, Y. Liu, T. Nie, B. Zhang, S. M. Disseler, B. J. Kirby, W. Ratcliff II, Q. Shao, K. Murata, X. Zhu, G. Yu, Y. Fan, M. Montazeri, X. Han, J. A. Borchers, and K. L. Wang, "Tailoring exchange couplings in magnetic topological-insulator/antiferromagnet heterostructures," Nat. Mater. 16, 94–100 (2016). 18J. Finley and L. Liu, "Spin-Orbit-Torque Efficiency in Compensated Ferrimagnetic Cobalt-Terbium Alloys," Phys. Rev. Appl. 6, 054001 (2016). 19F. Katmis, V. Lauter, F. S. Nogueira, B. A. Assaf, M. E. Jamer, P. Wei, B. Satpati, J. W. Freeland, I. Eremin, D. Heiman, P. Jarillo-Herrero, and J. S. Moodera, "A high-temperature ferromagnetic topological insulating phase by proximity coupling," Nature 533, 513–516 (2016). 20J. 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, "Anisotropic Dirac Fermions in a Bi Square Net of SrMnBi2," Phys. Rev. Lett. 107, 126402 (2011), arXiv:1104.5138. 21K. Wang, D. Graf, H. Lei, S. W. Tozer, and C. Petrovic, "Quantum transport of two-dimensional Dirac fermions in SrMnBi2," Phys. Rev. B - Condens. Matter Mater. Phys. 84, 220401(R) (2011), arXiv:1204.1049v1. 22H. Masuda, H. Sakai, M. Tokunaga, Y. Yamasaki, A. Miyake, J. Shiogai, S. Nakamura, S. Awaji, A. Tsukazaki, H. Nakao, Y. Murakami, T.-h. Arima, Y. Tokura, and S. Ishiwata, "Quantum Hall effect in a bulk antiferromagnet EuMnBi2 with magnetically confined two-dimensional Dirac fermions," Sci. Adv. 2, e1501117 (2016). 23R. S. K. Mong, A. M. Essin, and J. E. Moore, "Antiferromagnetic topological insulators," Phys. Rev. B 81, 245209 (2010), arXiv:1004.1403. 10 24C. Liu, Y. Lee, T. Kondo, E. D. Mun, M. Caudle, B. N. Harmon, S. L. Bud, P. C. Canfield, and A. Kaminski, "Metallic surface electronic state in half-Heusler compounds RPtBi (R = Lu , Dy , Gd)," 83, 205133 (2011) (2011). 25Z. F. Wang, H. Zhang, D. Liu, C. Liu, C. Tang, C. Song, Y. Zhong, J. Peng, F. Li, C. Nie, L. Wang, X. J. Zhou, X. Ma, Q. K. Xue, and F. Liu, "Topological edge states in a high-temperature superconductor FeSe/SrTiO3(001) film," Nat. Mater. 15, 968 (2016). 26W.-F. Tsai and H. Lin, "Topological insulators and superconductivity: The integrity of two sides," Nat. Mater. (2016), 10.1038/nmat4700. 27L. Fu and C. L. Kane, "Superconducting Proximity Effect and Majorana Fermions at the Surface of a Topological Insulator," Phys. Rev. Lett. 100, 096407 (2008). 28C. Niu, J. P. Hanke, P. M. Buhl, G. Bihlmayer, D. Wortmann, S. Blugel, and Y. Mokrousov, "Quantum spin Hall effect and topological phase transitions in honeycomb antiferromagnets," , 1–5 (2017), arXiv:1705.07035. 29N. P. Armitage, E. J. Mele, and A. Vishwanath, "Weyl and Dirac Semimetals in Three Dimensional Solids ," (2017), arXiv:1705.01111v1. 30S. Jia, S.-Y. Xu, and M. Z. Hasan, "Weyl semimetals, Fermi arcs and chiral anomalies," Nat. Mater. 15, 1140–1144 (2016). 31M. Hirschberger, S. Kushwaha, Z. Wang, Q. Gibson, S. Liang, C. A. Belvin, B. A. Bernevig, R. J. Cava, and N. P. Ong, "The chiral anomaly and thermopower of Weyl fermions in the half-Heusler GdPtBi," Nat. Mater. 15, 1161–1165 (2016), arXiv:1602.07219. 32M. N. Ali, J. Xiong, S. Flynn, J. Tao, Q. D. Gibson, L. M. Schoop, T. Liang, N. Haldolaarachchige, M. Hirschberger, N. P. Ong, and R. J. Cava, "Large, non-saturating magnetoresistance in WTe2." Nature 514, 205–208 (2014), arXiv:1405.0973. 33A. A. Soluyanov, D. Gresch, Z. Wang, Q. Wu, M. Troyer, X. Dai, and B. A. Bernevig, "Type-II Weyl semimetals," Nature 527, 495–498 (2015). 34I. Pletikosic, M. N. Ali, A. V. Fedorov, R. J. Cava, and T. Valla, "Electronic structure basis for the extraordinary magnetoresistance in WTe2," Phys. Rev. Lett. 113, 216601 (2014), arXiv:1407.3576v1. 35B.-j. Yang and N. Nagaosa, "Classification of stable three-dimensional Dirac semimetals with nontrivial topology," Nat. Commun. 5, 4898 (2014), arXiv:1404.0754v1. 36M. Kargarian, M. Randeria, and Y.-M. Lu, "Are the surface Fermi arcs in Dirac semimetals topologically pro- tected?" Proc. Natl. Acad. Sci. 113, 8648–8652 (2016). 37P. Tang, Q. Zhou, G. Xu, and S.-C. Zhang, "Dirac fermions in an antiferromagnetic semimetal," Nat. Phys. 12, 1100–1104 (2016). 38L. Smejkal, J. Zelezn´y, J. Sinova, and T. Jungwirth, "Electric Control of Dirac Quasiparticles by Spin-Orbit Torque in an Antiferromagnet," Phys. Rev. Lett. 118, 106402 (2017). 39S. M. Young and B. J. Wieder, "Filling-enforced Magnetic Dirac Semimetals in Two Dimensions," Phys. Rev. Lett. 118, 186401 (2017), arXiv:1609.06738. 40P. Wadley, B. Howells, J. Zelezny, C. Andrews, V. Hills, R. P. Campion, V. Novak, K. Olejn´ık, F. Maccherozzi, S. S. Dhesi, S. Y. Martin, T. Wagner, J. Wunderlich, F. Freimuth, Y. Mokrousov, J. Kunes, J. S. Chauhan, M. J. Grzybowski, A. W. Rushforth, K. W. Edmonds, B. L. Gallagher, and T. Jungwirth, "Electrical switching of an antiferromagnet," Science 351, 587–590 (2016). 41Z. Tian, Y. Kohama, T. Tomita, H. Ishizuka, T. H. Hsieh, J. J. Ishikawa, K. Kindo, L. Balents, and S. Nakatsuji, "Field-induced quantum metalinsulator transition in the pyrochlore iridate Nd2Ir2O7," Nat. Phys. 12, 134 (2015). 42N. Wakeham, E. D. Bauer, M. Neupane, and F. Ronning, "Large magnetoresistance in the antiferromagnetic semimetal NdSb," Phys. Rev. B - Condens. Matter Mater. Phys. 93, 205152 (2016), arXiv:1606.03724. 43D. Xiao, M.-C. Chang, and Q. Niu, "Berry phase effects on electronic properties," Rev. Mod. Phys. 82, 1959–2007 (2010). 44X. 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). 45H. Yang, Y. Sun, Y. Zhang, W.-J. Shi, S. S. P. Parkin, and B. Yan, "Topological Weyl semimetals in the chiral antiferromagnetic materials Mn3Ge and Mn3Sn," arXiv:1608.03404 (2016). 46Z. Wang, M. G. Vergniory, S. Kushwaha, M. Hirschberger, E. V. Chulkov, A. Ernst, N. P. Ong, R. J. Cava, and B. A. Bernevig, "Time-Reversal-Breaking Weyl Fermions in Magnetic Heusler Alloys," Phys. Rev. Lett. 117, 236401 (2016), arXiv:1603.00479. 47A. B. Sushkov, J. B. Hofmann, G. S. Jenkins, J. Ishikawa, S. Nakatsuji, S. Das Sarma, and H. D. Drew, "Optical evidence for a Weyl semimetal state in pyrochlore Eu2Ir2O7," Phys. Rev. B - Condens. Matter Mater. Phys. 92, 241108(R) (2015), arXiv:1507.01038. 48B. Yan and C. Felser, "Topological Materials: Weyl Semimetals," Annu. Rev. Condens. Matter Phys. 8, 337–354 (2017), arXiv:1611.04182. 11 49S. Borisenko, D. Evtushinsky, Q. Gibson, A. Yaresko, T. Kim, M. N. Ali, B. Buechner, M. Hoesch, (2015), and R. J. Cava, "Time-Reversal Symmetry Breaking Type-II Weyl State in YbMnBi2," arXiv: 10.1017/CBO9781107415324.004, arXiv:1507.04847. 50M. Chinotti, A. Pal, W. J. Ren, C. Petrovic, and L. Degiorgi, "Electrodynamic response of the type-II Weyl semimetal YbMnBi2," Phys. Rev. B 94, 245101 (2016). 51A. Wang, I. Zaliznyak, W. Ren, L. Wu, D. Graf, and V. O. Garlea, "Magnetotransport study of Dirac fermions in YbMnBi2 antiferromagnet," 94, 165161 (2016). 52D. Chaudhuri, B. Cheng, A. Yaresko, Q. D. Gibson, R. J. Cava, and N. P. Armitage, "An optical investigation of the strong spin-orbital coupled magnetic semimetal YbMnBi2," arXiv:1701.08693v1. 53Y. Zhang, Y. Sun, H. Yang, J. Zelezn´y, S. P. P. Parkin, C. Felser, and B. Yan, "Strong anisotropic anomalous Hall effect and spin Hall effect in the chiral antiferromagnetic compounds Mn3X (X= Ge, Sn, Ga, Ir, Rh and Pt)," Phys. Rev. B 95, 075128 (2017). 54N. Kiyohara, T. Tomita, and S. Nakatsuji, "Giant Anomalous Hall Effect in the Chiral Antiferromagnet Mn3Ge," Phys. Rev. Appl. 5, 064009 (2016), arXiv:1511.04619. 55J. Sinova, S. O. Valenzuela, J. Wunderlich, C. H. Back, and T. Jungwirth, "Spin Hall effects," Rev. Mod. Phys. 87, 1213–1260 (2015), arXiv:1411.3249. 56H. Chen, Q. Niu, and A. H. MacDonald, "Anomalous Hall Effect Arising from Noncollinear Antiferromagnetism," Phys. Rev. Lett. 112, 017205 (2014). 57J. Kubler and C. Felser, "Non-collinear antiferromagnets and the anomalous Hall effect," Europhys. Lett. 108, 67001 (2014), arXiv:1410.5985. 58F. D. M. Haldane, "Model for a Quantum Hall Effect without Landau Levels: Condensed-Matter Realization of the {'Parity Anomaly'}," Phys. Rev. Lett. 61, 2015 (1988). 59R. Shindou and N. Nagaosa, "Orbital Ferromagnetism and Anomalous Hall Effect in Antiferromagnets on the Distorted fcc Lattice," Phys. Rev. Lett. 87, 116801 (2001). 60T. Tomizawa and H. Kontani, "Anomalous Hall effect in the t2g orbital kagome lattice due to noncollinearity: Significance of the orbital Aharonov-Bohm effect," Phys. Rev. B 80, 100401(R) (2009); "Anomalous Hall effect due to noncollinearity in pyrochlore compounds: Role of orbital Aharonov-Bohm effect," Phys. Rev. B 82, 104412 (2010). 61A. A. Burkov and L. Balents, "Weyl Semimetal in a Topological Insulator Multilayer," Phys. Rev. B 107, 127205 (2011); K. Y. Yang, Y. M. Lu, and Y. Ran, "Quantum Hall effects in a Weyl semimetal: Possible application in pyrochlore iridates," Phys. Rev. B 84, 075129 (2011), arXiv:1105.2353. 62S. Nakatsuji, N. Kiyohara, and T. Higo, "Large anomalous Hall effect in a non-collinear antiferromagnet at room temperature," Nature 527, 212–216 (2015). 63A. K. Nayak, J. E. Fischer, Y. Sun, B. Yan, J. Karel, A. C. Komarek, C. Shekhar, N. Kumar, W. Schnelle, J. Kubler, C. Felser, S. S. P. Parkin, J. Ku bler, C. Felser, and S. S. P. Parkin, "Large anomalous Hall effect driven by a nonvanishing Berry curvature in the noncolinear antiferromagnet Mn3Ge," Sci. Adv. 2, e1501870–e1501870 (2016), arXiv:1511.03128. 64T. Suzuki, R. Chisnell, A. Devarakonda, Y.-T. Liu, W. Feng, D. Xiao, J. W. Lynn, and J. G. Checkelsky, "Large anomalous Hall effect in a half-Heusler antiferromagnet," Nat. Phys. 12, 1119 (2016). 65P. Zhou, C. Q. Sun, and L. Z. Sun, "Two Dimensional Antiferromagnetic Chern Insulator: NiRuCl6," Nano Lett. 16, 6325–6330 (2016), arXiv:1601.07705. 66X.-Y. Dong, S. Kanungo, B. Yan, and C.-X. Liu, "Time-reversal-breaking topological phases in antiferromagnetic Sr2FeOsO6 films," Phys. Rev. B 94, 245135 (2016). 67J.-P. Hanke, F. Freimuth, C. Niu, S. Blugel, and Y. Mokrousov, "Mixed Weyl semimetals and dissipationless magnetization control in insulators," arxiv.org/1701.08050 (2017), arXiv:1701.08050. 68F.-Y. Li, Y.-D. Li, Y. B. Kim, L. Balents, Y. Yu, and G. Chen, "Weyl magnons in breathing pyrochlore antifer- romagnets," Nat. Commun. 7, 12691 (2016), arXiv:1602.04288. 69N. Kanazawa, Y. Onose, T. Arima, D. Okuyama, K. Ohoyama, S. Wakimoto, K. Kakurai, S. Ishiwata, and Y. Tokura, "Large Topological Hall Effect in a Short-Period Helimagnet MnGe," Phys. Rev. Lett. 106, 156603 (2011). 70M. Hoffmann, J. Weischenberg, B. Dup´e, F. Freimuth, P. Ferriani, Y. Mokrousov, and S. Heinze, "Topological orbital magnetization and emergent Hall effect of an atomic-scale spin lattice at a surface," Phys. Rev. B 92, 020401 (2015), arXiv:1503.01885v2. 71Y. Machida, S. Nakatsuji, S. Onoda, T. Tayama, and T. Sakakibara, "Time-reversal symmetry breaking and spontaneous Hall effect without magnetic dipole order." Nature 463, 210–213 (2010). 72C. Surgers, G. Fischer, P. Winkel, and H. V. Lohneysen, "Large topological Hall effect in the non-collinear phase of an antiferromagnet." Nat. Commun. 5, 3400 (2014). 73C. Surgers, W. Kittler, T. Wolf, and H. v. Lohneysen, "Anomalous Hall effect in the noncollinear antiferromagnet 12 Mn5Si3," AIP Adv. 6, 055604 (2016), arXiv:1601.01840. 74J. Zhou, Q. F. Liang, H. Weng, Y. B. Chen, S. H. Yao, Y. F. Chen, J. Dong, and G. Y. Guo, "Predicted Quantum Topological Hall Effect and Noncoplanar Antiferromagnetism in K0.5RhO2," Phys. Rev. Lett. 116, 256601 (2016), arXiv:1602.08553. 75Y. Sun, Y. Zhang, C. Felser, and B. Yan, "Strong Intrinsic Spin Hall Effect in the TaAs Family of Weyl Semimetals," Phys. Rev. Lett. 117, 146403 (2016), arXiv:1604.07167. 76G. Yin, Y. Liu, Y. Barlas, J. Zang, and R. K. Lake, "Topological spin Hall effect resulting from magnetic skyrmions," Phys. Rev. B 92, 024411 (2015), arXiv:1503.00242. 77Y. Zhang, J. Zelezny, Y. Sun, J. van den Brink, and B. Yan, "Spin Hall effect emerging from a chiral magnetic lattice without spin-orbit coupling," Arxiv Prepr. (2017), arXiv:1704.03917. 78J. P. Hanke, F. Freimuth, A. K. Nandy, H. Zhang, S. Bl\"{u}gel, and Y. Mokrousov, "Role of Berry phase theory for describing orbital magnetism: From magnetic heterostructures to topological orbital ferromagnets," Phys. Rev. B 94, 121114(R) (2016), arXiv:1603.07683. 79J.-P. Hanke, F. Freimuth, S. Blugel, and Y. Mokrousov, "Prototypical topological orbital ferromagnet γ-FeMn," Sci. Rep. 7, 41078 (2017). 80P. M. Buhl, F. Freimuth, S. Blugel, and Y. Mokrousov, "Topological spin Hall effect in antiferromagnetic skyrmions," Phys. status solidi - Rapid Res. Lett. 11, 1700007 (2017). 81C. Jin, C. Song, J. Wang, and Q. Liu, "Dynamics of antiferromagnetic skyrmion driven by the spin Hall effect," Appl. Phys. Lett. 109, 182404 (2016). 82G. Finocchio, F. Buttner, R. Tomasello, M. Carpentieri, and M. Klaui, "Magnetic skyrmions: from fundamental to applications," J. Phys. D. Appl. Phys. 49, 423001 (2016). 83Z. Liu and H. Ian, "Numerical studies on antiferromagnetic skyrmions in nanodisks by means of a new quantum simulation approach," Chem. Phys. Lett. 649, 135–140 (2016), arXiv:1601.05170v1. 84S. Rohart, J. Miltat, and A. Thiaville, "Path to collapse for an isolated N´eel skyrmion," Phys. Rev. B 93, 214412 (2016). 85A. N. Bogdanov, U. K. Roessler, M. Wolf, and K. H. Muller, "Magnetic structures and reorientation transitions in noncentrosymmetric uniaxial antiferromagnets," Phys. Rev. B 66, 214410 (2002), arXiv:0206291 [cond-mat]. 86T. Morinari, "Half-Skyrmion Theory for High-Temperature Superconductivity," in The Multifaceted Skyrmion (World Scientific, 2010) pp. 311–331. 87X. Zhang, Y. Zhou, and M. Ezawa, "Antiferromagnetic Skyrmion: Stability, Creation and Manipulation," Sci. Rep. 6, 24795 (2016), arXiv:1504.01198. 88J. Barker and O. A. Tretiakov, "Static and Dynamical Properties of Antiferromagnetic Skyrmions in the Presence of Applied Current and Temperature," Phys. Rev. Lett. 116, 147203 (2016). 89H. Velkov, O. Gomonay, M. Beens, G. Schwiete, A. Brataas, J. Sinova, and R. A. Duine, "Phenomenology of current-induced skyrmion motion in antiferromagnets," New J. Phys. 18, 075016 (2016), arXiv:1604.05712. 90X. Zhang, Y. Zhou, and M. Ezawa, "Magnetic bilayer-skyrmions without skyrmion Hall effect," Arxiv Prepr. 7, 1504.02252 (2015), arXiv:1504.02252. 91S. Zhang, A. A. Baker, S. Komineas, and T. Hesjedal, "Topological computation based on direct magnetic logic communication," Sci. Rep. 5, 15773 (2015). 92Q. L. He, G. Yin, L. Yu, A. J. Grutter, L. Pan, X. Kou, X. Che, G. Yu, T. Nie, B. Zhang, Q. Shao, K. Murata, X. Zhu, Y. Fan, X. Han, B. J. Kirby, and K. L. Wang, "Topological transitions induced by antiferromagnetism in a thin-film topological insulator," Arxiv Prepr. (2016), arXiv:1612.01661. 93S. Ghosh and A. Manchon, "Spin-orbit torque in two-dimensional antiferromagnetic topological insulators," Phys. Rev. B 95, 035422 (2017). 94A. Kandala, A. Richardella, S. Kempinger, C.-X. Liu, and N. Samarth, "Giant anisotropic magnetoresistance in a quantum anomalous Hall insulator," Nat. Commun. 6, 7434 (2015), arXiv:1503.0355. 95C. Carbone, P. Moras, P. M. Sheverdyaeva, D. Pacile, M. Papagno, L. Ferrari, D. Topwal, E. Vescovo, G. Bihlmayer, F. Freimuth, Y. Mokrousov, and S. Blugel, "Asymmetric band gaps in a Rashba film system," Phys. Rev. B - Condens. Matter Mater. Phys. 93, 125409 (2016). 96Q.-K. Xue, "Nanoelectronics: A topological twist for transistors," Nat. Nanotechnol. 6, 197–198 (2011). 97J. Liu, T. H. Hsieh, P. Wei, W. Duan, J. Moodera, and L. Fu, "Spin-filtered Edge States with an Electri- cally Tunable Gap in a Two-Dimensional Topological Crystalline Insulator," Nat. Mater. 13, 178–183 (2013), arXiv:1310.1044. 98H. Fujita, "Field-free, spin-current control of magnetization in non-collinear chiral antiferromagnets," Phys. status solidi - Rapid Res. Lett. (2016), 10.1002/pssr.201600360, arXiv:1610.07615. 99W. Feng, G.-Y. Guo, J. Zhou, Y. Yao, and Q. Niu, "Large magneto-optical Kerr effect in noncollinear antiferro- magnets Mn 3 X ( X = Rh , Ir , Pt )," Phys. Rev. B 92, 144426 (2015). 100D. Thouless, Topological Quantum Numbers in Nonrelativistic Physics (World Scientific, 1998). 13 

1512.00255

1

1512

Energetics and carrier transport in doped Si/SiO2 quantum dots

In the present theoretical work we have considered impurities, either boron or phosphorous, located at different substitutional sites in silicon quantum dots (Si-QDs) with diameters around 1.5\,nm, embedded in a SiO2 matrix. Formation energy calculations reveal that the most energetically-favored doping sites are inside the QD and at the Si/SiO2 interface for P and B impurities, respectively. Furthermore, electron and hole transport calculations show in all the cases a strong reduction of the minimum voltage threshold, and a corresponding increase of the total current in the low-voltage regime. At higher voltage, our findings indicate a significant increase of transport only for P-doped Si-QDs, while the electrical response of B-doped ones does not stray from the undoped case. These findings are of support for the employment of doped Si-QDs in a wide range of applications, such as Si-based photonics or photovoltaic solar cells.

cond-mat.mes-hall

cond-mat

Energetics and carrier transport in doped Si/SiO2 quantum dots Nuria Garcia-Castello,1, ∗ Sergio Illera,1 Joan Daniel Prades,1 Stefano Ossicini,2 Albert Cirera,1 and Roberto Guerra3, † 1MIND-IN2UB, Department of Electronics, Universitat de Barcelona, C/Mart´ı i Franqu`es 1, E-08028 Barcelona, Spain. 2Dipartimento di Scienze e Metodi dell’Ingegneria and Centro Interdipartimentale En&Tech, Universit`a di Modena e Reggio Emilia, via Amendola 2 Pad. Morselli, I-42122 Reggio Emilia, Italy. 3International School for Advanced Studies (SISSA), Via Bonomea 265, I-34136 Trieste, Italy; CNR-IOM Democritos National Simulation Center,Via Bonomea 265, I-34136 Trieste, Italy. In the present theoretical work we have considered impurities, either boron or phosphorous, lo- cated at different substitutional sites in silicon quantum dots (Si-QDs) with diameters around 1.5 nm, embedded in a SiO2 matrix. Formation energy calculations reveal that the most energetically-favored doping sites are inside the QD and at the Si/SiO2 interface for P and B impurities, respectively. Furthermore, electron and hole transport calculations show in all the cases a strong reduction of the minimum voltage threshold, and a corresponding increase of the total current in the low-voltage regime. At higher voltage, our findings indicate a significant increase of transport only for P-doped Si-QDs, while the electrical response of B-doped ones does not stray from the undoped case. These findings are of support for the employment of doped Si-QDs in a wide range of applications, such as Si-based photonics or photovoltaic solar cells. [ Electronic Supplementary Information (ESI) available. See DOI: 10.1039/c5nr02616d ] PACS numbers: 73.63.Kv, 71.15.Mb, 71.20.Mq I. INTRODUCTION Semiconductor quantum dots (QDs) are promising structures due to their tunable band gap with QD diam- eter. Silicon QDs (Si-QDs) are, among all, the ideal can- didates for mass-scale devices production, because of the abundancy of silicon and its non-toxic, bio-compatible, and ecologic nature. Embedding Si-QDs in a dielec- tric matrix is one way to obtain an efficient quantum- confinement effect.1 Besides, the possibility of introduc- ing dopant atoms has been suggested to improve the the achievable macroscopic currents in QD-based devices. Doping of Si-QDs embedded in silica has been already investigated by several experimental works.2–15 In par- ticular, it has been shown that electrically-activated im- purity atoms located in substitutional sites tend to en- hance the conductivity.6–10 Theoretically, several works have studied the formation and ionization energies, and the opto-electronic properties of freestanding doped Si- QDs.16–26 Instead, only recently theoretical works deal- ing with structural properties of doped embedded Si-QDs have appeared in literature.27 In any case, all the above works show that the final properties of these systems are strongly sensitive to the concentration and position of the impurities. This fact makes necessary the accurate control of the impurities at the nanoscale in order to ensure repeatability. Thanks to the recent advances, it is nowadays possi- ble to dope Si-QDs with few28 or even only one29 dopant atoms, and to experimentally obtain the density of states of the single QD.30 With these premises, a comprehen- sive understanding of structural, electrical and trans- port properties of doped Si-QDs is hopefully going to be achieved soon. The aim of the present work is to shed light in this direction. Theoretical simulations can provide a strong support in understanding the role of impurities in nanostructures, thanks to the possibility of manipulating the samples at the atomic level, and to the recent advance in the computing capabilities. Here we report a theoretical study of electron and hole transport induced by B or P substitutional doping in a crystalline Si-QD embedded in SiO2, for three different QDs. The structures with the lowest formation energies are identified, and the I-V characteristic is obtained by a novel approach (See Method). To our knowledge, this is the first study reporting theoretical I-V characteristics of doped embedded Si-QDs. II. STRUCTURES AND METHOD Despite the tremendous progress in the computational power made with the advent of supercomputers, a com- plete theoretical description of transport in large nanos- tructures is still far to be achieved. Approximations must be adopted in order to limit the computational ef- fort, like using a reduced system size, or employing a simplified approach. Most of the available studies on single- and two-QDs have been performed by using non- equilibrium Green functions formalism (NEGFF) with constant transition rates between QDs and one energy level per QD31–36, and by using tunneling transmission coefficients with planar Si/SiO2 values for the barrier height, and bulk-Si band gap.37–40 Here we make use of a different approach,41–43 based on the transfer Hamiltonian formalism and non-coherent rate equations to describe the current, that takes into account the local potential due to the QD charge, com- 2 and a full ab-initio approach is required in that case.46 The transmission probabilities are calculated using WKB approximation of Fowler-Nordheim and direct tun- nel mechanisms, which are the two more relevant tun- neling mechanisms in QDs inside dielectric matrices.49 We set the distance between the Si-QD and each lead to 1.1 nm for all the systems, the relative dielectric constant of the oxide to 3.9, and the oxide effective mass of elec- trons and holes to 0.40 me and 0.32 me, respectively,37 me being the free-electron mass. The effect of charge inside the QD induced by the applied V is taken into account. Thus, we solve self- consistently the equations for the local potential in- side the QD, and the QD non-equilibrium distribu- tion function. The details of this method are reported elsewhere.41–43 For the present study we assume the same capacitive coupling between the QD and the leads, yield- ing specular current trends for negative V .42 Thus, to avoid redundancy we report only currents for positive applied V . Assuming ballistic transport we have independent con- duction channels for electrons and holes. The current for each carrier type can be calculated from Eq. 1 using the corresponding transmission coefficient and density of states. The total current is given by the sum of electron and hole currents. The density of states of the Si-QD ρQD has been com- puted within density functional theory (DFT) using the SIESTA code50,51, and gaussian broadening of 0.05 eV. It corresponds to the PDOS of the Si atoms of the QD to- gether with the interface O atoms, in order to include the interface states.52 The embedded Si-QDs of 32, 35, and 47 Si atoms (i.e. Si32, Si35, and Si47) were obtained from a 3×3×3 β-cristobalite supercell, Si216O432 of side of 21.48 A, by removing all the O atoms inside a cut-off spheres of given radius. In this way, no dangling bonds or defects are present, and all the O atoms at the inter- face are single-bonded to the Si atoms of the QD. The so-obtained embedded systems are formed by a total of about 600 atoms. Spin polarized calculations were per- formed using norm-conserving Troullier-Martins53 pseu- dopotentials with nonlinear core corrections within the local density approximation (LDA), with a Ceperley- Alder54 exchange-correlation potential, as parameterized by Perdew-Zunger.55 A cut-off of 250 Ry on the elec- tron density and no additional external pressure or stress were applied. All the calculations were performed at the Γ-point of reciprocal space, with an electronic tempera- ture of 300 K, and standard double-ζ basis set for all the atoms. Atomic positions and cell parameters were left totally free to move, with a force threshold of 0.01 eV/A. Thanks to the metastable nature of β-cristobalite, af- ter relaxation all the SiO2 in the supercell gets amor- phized due to the presence of the QD. Structural and op- tical properties of the embedding SiO2 match well those of a “true” amorphous SiO2 glass (a-SiO2), formed by annealing.52 As reported in a previous study44 the presence of quan- FIG. 1. Sticks-and-balls representation of a Si35 QD embed- ded in SiO2 (∼600 atoms), doped at the interface, and en- closed between two semi-infinite metallic leads with an ap- plied voltage. puted in a self-consistently field regime with the non- equilibrium distribution function of the QD, and able to use more than one energy state per QD. In particular, we can use the density of states computed by ab initio calculations, a difficult issue to treat with NEGFF, al- lowing us to include implicitly the effect of dopant atoms in the transport properties. With the same approach we investigated, in a previous work, the influence of QD size and amorphization level on the transport properties of undoped Si-QDs.44 We compute the I-V characteristic between two metal- lic semi-infinite electrodes coupled to an elastic scattering region (corresponding to the doped Si-QD embedded in the silica matrix) when an external bias voltage V is ap- plied (see Fig. 1). The expression of the current under the transfer Hamiltonian formalism is47,48 (cid:90) TLTRρLρQDρR TLρL + TRρR I = 4πq  (fL − fR) dE , (1) where TL,R(E) are the transmission probabilities be- tween the left lead and the QD, and between the QD and the right lead, respectively; ρL,R,QD(E) are the density of states of each region of the system, and fL,R(E) are the Fermi-Dirac distribution functions of the electrodes. All the calculations are done at room temperature (kBT = 0.026 eV), and ρL/R are assumed constant in energy. In principle, the presence of nanostructured con- tact could be described in our model making use of cal- culated ρL(E) and ρR(E) from atomistic leads, like e.g. gold tips. However, as indicated by previous works,45 in the latter case we expect no major change of the here- presented I-V curves, but rather a reduction of the cur- rent magnitude depending on the tips DOS. Clearly, for very small (molecular-like) electrodes+QD systems, cur- rents become more sensitive to the geometrical conditions Vtum confinement makes valence band offset (VBO) and conduction band offset (CBO) between Si-QDs and SiO2 significantly different than in bulk or planar systems. In order to evaluate the band offset between SiO2 and QD, we have aligned the DOS of an a-SiO2 sample with that of the embedded Si-QD by matching the strong deep- valence peak of a-SiO2, which is well observable in all the considered structures. Thus, we have obtained the VBO as the difference between QD and SiO2 HOMOs (highest occupied molecular orbitals), and the CBO as the difference between SiO2 and QD LUMOs (lowest un- occupied molecular orbitals). We also have defined the hole barrier (HB) as the difference between the Fermi en- ergy (EF ) and the HOMO of the embedding a-SiO2, and the electron barrier (EB) as the difference between the LUMO of the embedding a-SiO2 and EF . Since EF is ap- proximately located at mid-Eg, it is HB = VBO+Eg/2, and EB = CBO+Eg/2. The computed DFT HOMO-LUMO gap Eg for a-SiO2 and bulk-Si are 7.0 eV and 0.6 eV, respectively, in agree- ment with other calculations,56 and smaller than the ex- perimental values of 9.0 eV and 1.1 eV, respectively. It is well known that Kohn-Sham eigenvalues give an under- estimated Eg due to the use of approximated exchange- correlation functionals. A correction to the fundamen- tal band gap can be obtained by many-body calcula- tions accounting for quasiparticle energies and excitonic corrections.52 However, while the total correction to Eg is noticeable in bulk materials, in strongly confined systems the enhanced excitonic interaction is known to compen- sate the self-energy of about the same amount.23,52,57,58 As a consequence, in the case of small embedded QD, one deals with “correct” Eg values (determined by QD states), but “uncorrect” band offsets due to the system- atic error in the SiO2-related energy values. In the case of a Si/SiO2 slab calculation in the bulk limit, we have obtained VBO and CBO of 2.6 eV and 3.9 eV, respectively, to be compared with the experi- mental values of 4.6 eV (VBO) and 3.1 eV (CBO).40,59 Therefore, to compensate such deviations, we have ap- plied a correction of 2.0 eV to VBO and HB values, and of -0.8 eV to CBO and EB values, while leaving Eg un- changed. Since our QD size range is small, we apply the same correction for all the samples. Note that our uncorrected band offset match that of other works inves- tigating charge-carrier transport in Si-QDs by hopping mechanisms.60,61 We have positioned the impurity atom in three differ- ent substitutional sites in the embedded system: at the QD center (in the following “dot”), at the QD/SiO2 in- terface, and in the SiO2 far away from the QD (in the fol- lowing “silica”). The Si atoms at the interface can form three possible suboxide types, Si1+, Si2+, Si3+, depend- ing on the number of bonded O atoms (in the following “int-1”, “int-2”, and “int-3”). While Si47 QD presents all the three suboxide types, Si32 presents only Si1+ and Si3+ types, while Si35 presents only Si1+ and Si2+ types. 3 III. RESULTS AND DISCUSSION A. Structural properties It is known that substitutional impurities produce a local distortion, that must be taken into account for a re- alistic description of doping. In the case of free-standing hydrogenated Si-QDs, impurities located close to the QD surface are energetically more favorable than others, thanks to a larger atomic mobility that allows a reduction of the stress around the dopant atom.16,18–21 In this case, doping the nanostructure core region could result very difficult, even for materials commonly doped in their bulk phase.19 Beside strain effects, other chemistry-governed factors, occurring at shorter scales, can determine the energetically favored site of the impurity. For example, in the case of OH-terminated or SiO2-embedded QDs, the strong electronegativity of O makes P strongly re- pelled from the interfacial sites, while conversely attract- ing B.10,24,25,27 This behavior has been observed in free- standing Si-QDs experiments,62,63 and suggested as the mechanism for IR absorption in B-doped free-standing Si-QDs, not observed in P-doped ones.16 In Figure 2 we report the formation energy Ωf of all the considered structures and doping sites, calculated fol- lowing Ref.18: Ωf = Edoped − Eundoped + ESi − Edopant, (2) where Eundoped and Edoped are the total energies of the undoped and doped systems, respectively, Edopant is the total energy per atom in a bulk configuration of the dopant atom,64 and ESi is the total energy per atom of bulk silicon. Consistently with the above discussion, we note in Fig- ure 2 that P and B impurities prefer to site inside QD FIG. 2. Formation energy Ωf as a function of QD size, for different positions of B (left panel) and P (right panel) dopant atoms. Zero energy corresponds to the undoped systems. Filled symbols highlight the most stable doped configuration. The value of Ωf for the impurity in bulk silicon (dashed line) and bulk silica (dotted line) is reported for comparison. and at the interface, respectively. Moreover, while the formation energy in P-doped systems decreases with the suboxide number, it conversely increases for B-doping. Interestingly, we also note that for the largest considered QD, Si47, the placement of P in the QD center is ener- getically similar to the int-1 case. This is because the QD core-region cannot easily accomodate the impurity- induced stress.20 Therefore, a more energetically stable site should be found close to the interface (in order to ac- comodate the stress more easily), but still inside the QD (to take advantage of the P-Si bond over P-O). The latter argument is supported by XPS measurements showing a clear B-O bond signal,10 while P-Si or P-P bonds seem preferentially formed rather than P-O bonds for Si-QDs with diameters smaller than 3.5 nm.11,12 Moreover, also PL experiments suggest B-doped Si-QDs with an intrin- sic core and heavily B-doped shells,13 while B-P codoped colloidal Si nanocrystals show an outer B-rich shell and an inner P-rich shell, arising due to the large difference in the segregation coefficient of B and P.65,66 In Fig. 2 we also report the Ωf of doped bulk-Si (dashed line) and a-SiO2 (dotted line). Clearly, the for- mation energy for doped a-SiO2 is much higher than for doped bulk-Si, especially for P-doping, on line with re- cent experiments observing P-atoms segregating toward the Si-rich regions.11 Also, secondary ion mass spec- troscopy (SIMS) experiments confirmed a negligible B or P diffusion from Si-QD layers to adjacent SiO2 layers.8,9 B. Electronic properties The inclusion of impurity atoms in the pristine sys- tem leads to a reduction of Eg due to the appearance of mid-gap states, whose energy and localization can vary in a very complex way.19,26 In our systems, doping with single group-III or group-V impurities results in an odd number of electrons, for which spin-polarized calculations must be employed. For small Si-QDs, a correlation be- tween the energy difference of spin-down and spin-up im- purity levels and the magnitude of the Stokes shift of undoped Si-QDs has been reported, signaling structural deformation.23 In Figure 3 we report the eigenvalues of all the sys- tems, with energies aligned using the embedding-SiO2 states (in order to get the band offset, see Method; see Supplementary Information†for numerical data). For the doped systems we also report the PDOS of the dopant atom. The expected acceptor behavior of B impurities – low- ering of the Fermi energy toward the valence band – which is clearly observed in hydrogenated Si-QDs,20 is only present in some of our embedded systems. Instead, in most of our structures the impurity generates deep lev- els, not a suitable condition for enhanced current trans- port. Besides, the dramatic reduction of Eg occurring in many cases, is an important requirement to foster the conductivity at low V . In the case of P impurities we ob- 4 FIG. 3. Spin-up and spin-down eigenvalue spectrum of all the considered systems: Si32 (top), Si35 (center), and Si47 (bottom). Energies have been aligned using the states of the embedding SiO2 (see Method). For each case, the PDOS of the dopant atom is also reported (black solid line). Black and grey dots mark the HOMO and LUMO states, respectively, while EF is reported by dashed line. Horizontal lines mark the uncorrected (orange) and corrected (green) band-edge of SiO2 (see Method). serve a clear donor behavior, as occurring in free-standing n-doped Si-QDs.19 It is worth to stress that the variability of the observed HOMO LUMO spin−up spin−down Fermi−level 0 1 2 3 4 5 6 7 8 9energy (eV)B-dotB-int-1B-int-3B-silicaP-dotP-int-1P-int-3P-silicaundop.Si32 0 1 2 3 4 5 6 7 8 9energy (eV)B-dotB-int-1B-int-2B-silicaP-dotP-int-1P-int-2P-silicaundop.Si35 0 1 2 3 4 5 6 7 8 9energy (eV)B-dotB-int-1B-int-2B-int-3B-silicaP-dotP-int-1P-int-2P-int-3P-silicaundop.Si47response with doping conditions, among the three con- sidered QDs, is expected due to the large QD surface-to- volume ratio.67 Nevertheless, it is possible to recognize some trends in our data. First, for QDs B-doped at the interface (the most energetically favored site for B) EF increases with the suboxide number, with correspond- ingly increasing HB and decreasing EB. Conversely, the interfacial P-doping produces an n-type effect, with EF slightly decreasing with the suboxide number. The doping at SiO2 sites, far from QD, produces for B impurity a minimal decrease of Eg (and of EF ), that should lead to a conductivity similar to the undoped case. The same behaviour is found for B-doping at the center of the QD. In the case of P impurity, Eg is dramatically reduced in all the cases, while no clear trend for EF can be deduced. Unfortunately, as discussed above, any po- tential advantage of P-doping at SiO2 sites is limited by its strongly unfavored energetics. However, Eg is reduced also in the case of P located at the QD center or at inter- facial sites with low suboxide number (the most favored sites), in which case we also observe HOMO and LUMO states populated by the PDOS of the impurity atom. In this case we expect a significant enhancement of electron current, also at low V . C. Transport properties In Figure 4 we show the computed I-V characteris- tic of each system, with total current obtained by sum- ming electron and hole currents (See Supplementary In- formation†for separate electron/hole I-V curves). The results are compared with the corresponding undoped case44 (the separate electron and hole contributions to the total current, as a function of the applied voltage, are reported in the Supplementary Information†). In the results of Fig. 4 are reflected all the above-discussed ef- fects of doping over the electronic properties: as EF ap- proaches the conduction (valence) band, electron (hole) barrier EB (HB) becomes smaller, and the electron (hole) current is enhanced with respect to the undoped sys- tem. Instead, the threshold V for triggering transport is related to Eg – typically reduced by doping – that de- termines the ionization energy required to generate free carriers. The latter aspect is well observed in B-doped struc- tures, especially for interfacial doping (the one with the lowest formation energy) showing, with respect to the undoped case, a significant enhancement of the total cur- rent at low V (due to Eg decrease), while at large V we observe no significant variation of the current. Doping at SiO2 sites seems practically irrelevant in B- doped structures, while it produces dramatic enhance- ments of the current, up to two orders of magnitude, for two of the P-doped structures, having although the largest formation energy. Nevertheless, the more energetically-favored P-doping inside QD yields still an increase of the I-V response in all the considered V - 5 FIG. 4. Calculated total current (electron + hole) as a func- tion of the applied voltage, for the considered doped config- urations (symbols) along with the undoped case (blue solid curve), for Si32 (top), Si35 (center), and Si47 (bottom) QDs. Filled symbols (in red) highlight the most stable doped con- figuration (See also Supplementary Information†). range, up to one order of magnitude, also at high V for all the QDs. Si32dotint−1int−3silica10−1010−910−810−710−610−5 0 1 2 3 4 5total current (A)voltage (V)B10−1010−910−810−710−610−5 0 1 2 3 4 5total current (A)voltage (V)PSi35dotint−1int−2silica10−1010−910−810−710−610−5 0 1 2 3 4 5total current (A)voltage (V)B10−1010−910−810−710−610−5 0 1 2 3 4 5total current (A)voltage (V)PSi47dotint−1int−2int−3silica10−1010−910−810−710−610−5 0 1 2 3 4 5total current (A)voltage (V)B10−1010−910−810−710−610−5 0 1 2 3 4 5total current (A)voltage (V)PIV. CONCLUSIONS We have studied Si-substitutional doping of Si-QDs embedded in SiO2, with either B or P impurities. Cal- culations reveal that B impurities tend to site at the QD/SiO2 interface, especially where a large number of bonding oxygens is present. Conversely, doping inside the QD or the SiO2 is unfavored, with similarly large formation energies. For P impurities, we have observed a clear trend in which the formation energy increases with the number of bonding oxygens, hence favoring the QD internal, while severely hampering interfacial and SiO2 sites. Besides, given the large Si/SiO2 interfacial energy, P-doping at interface Si1+ sites may be favored over QD- core regions, especially in large QDs, thanks to an eas- ier relaxation of the doping-induced stress at the inter- face. Therefore, we indicate sub-interfacial QD sites as the most energetically-favored ones for P impurities. In any case, the presence of impurities reduces the band-gap Eg of the material – except for B-doping in the QD or in SiO2 (the two least probable sites) – lead- 6 ing to enhanced I-V characteristic at low V . At high V , for the most favored impurity positions we observe a significant variation of the current, with respect to the undoped systems, only for P-doping. Thus, with either B or P impurities one can foster hole-current at low V or electron-current at low+high V , respectively. Such asymmetry of the response with the dopant type can be advantageous from a technolog- ical point of view, permitting the tuning of the device response in the given V range. For example, possible ap- plications can extend from Si-based photonics,65 to next- generation photovoltaic solar cells,40 among others. ACKNOWLEDGEMENTS R. G. acknowledges support from the ERC Advanced Grant no. 320796-MODPHYSFRICT. S. I. acknowledges support from the FI program of the Generalitat de Catalunya. A. C. acknowledges sup- port from the ICREA academia program. J. D. P. acknowledges support from the Serra Hunter programme. The authors thankfully acknowledge the Barcelona Supercomputing Center - Centro Na- cional de Supercomputaci´on, and the CINECA-ISCRA initiative. ∗ [email protected] † [email protected] 1 J. P. Proot, C. Delerue, and G. Allan, Appl. Phys. Lett., 1992, 61, 1948. McMurray, and P. C. Taylor, Phys. Status Solidi C, 2012, 9, 1908. 16 X. Pi, J. Nanomater., 2012, 2012, 912903. 17 D. V. Melnikov and J. R. Chelikowsky, Phys. Rev. Lett., 2 G. Conibeer, I. Perez-Wurfl, X. Hao, D. Di, and D. Lin, 2004, 92, 046802. Nanoscale Res. Lett., 2012, 7, 193. 3 A. Mimura, M. Fujii, S. Hayashi, and K. Yamamoto, Solid State Commun., 1999, 109, 561. 4 M. Fujii, A. Mimura, S. Hayashi, Y. Yamamoto, and K. Murakami, Phys. Rev. Lett., 2002, 89, 206805. 5 A. R. Stegner, R. N. Pereira, K. Klein, R. Lechner, R. Dietmueller, M. S. Brandt, M. Stutzmann, and H. Wiggers, Phys. Rev. Lett., 2008, 100, 026803. 6 D. Di, H. Xu, I. Perez-Wurfl, M. A. Green, and G. Conibeer, Prog. Photovolt: Res. Appl., 2011, 21, 569. 7 S. Huang and G. Conibeer, J. Phys. D: Appl. Phys., 2013, 46, 024003. 8 X. J. Hao, E.-C. Cho, C. Flynn, Y. S. Shen, S. C. Park, G. Conibeer, and M. A. Green, Sol. Energ. Mat. Sol. Cells, 2009, 93, 273. 9 X. J. Hao, E.-C. Cho, G. Scardera, Y. S. Shen, E. Bellet- Amalric, D. Bellet, G. Conibeer, and M. A. Green, Sol. Energ. Mat. Sol. Cells, 2009, 93, 1524. 18 S. Ossicini, F. Iori, E. Degoli, E. Luppi, R. Magri, R. Poli, G. Cantele, F. Trani, and D. Ninno, IEEE J. Sel. Topics Quantum Electron., 2006, 12, 1585. 19 M. Mavros, D. A. Micha, and D. S. Kilin, J. Phys. Chem. C, 2011, 115, 19529. 20 J.-H. Eom, T.-L. Chan, and J. R. Chelikowsky, Solid State Commun., 2010, 150, 130. 21 J. Ma, S.-H. Wei, N. R. Neale, and A. J. Nozik, Appl. Phys. Lett., 2011, 98, 173103. 22 Z. Zhou, M. L. Steigerwald, R. A. Friesner, and L. Brus, Phys. Rev. B, 2005, 71, 245308. 23 L. E. Ramos, E. Degoli, G. Cantele, S. Ossicini, D. Ninno, J. Furthmuller, and F. Bechstedt, J. Phys.: Condens. Mat- ter, 2007, 19, 466211. 24 A. Carvalho, S. Oberg, M. Barroso, M. J. Rayson, and P. Briddon, Phys. Status Solidi A, 2012, 209, 1847. 25 A. Carvalho, S. Oberg, M. Barroso, M. J. Rayson, and P. Briddon, Phys. Status Solidi B, 2013, 250, 1799. 10 M. Xie, D. Li, L. Chen, F. Wang, X. Zhu, D. Yang, Appl. 26 X. Chen, X. Pi, and D. Yang, J. Phys. Chem. C, 2011, Phys. Lett., 2013, 102, 123108. 115, 661. 11 M. Perego, C. Bonafos, and M. Fanciulli, Nanotechnology, 27 R. Guerra and S. Ossicini, J. Am. Chem. Soc. , 2014, 136, 2010, 21, 025602. 4404. 12 E.-C. Cho, S. Park, X. Hao, D. Song, G. Conibeer, S.-C. Park, and M. A. Green, Nanotechnology, 2008, 19, 245201. 13 H. Sugimoto, M. Fujii, M. Fukuda, K. Imakita, and S. Hayashi, J. Appl. Phys., 2011, 110, 063528. 14 S. Park, E. Cho, D. Song, G. Conibeer, and M. A. Green, Sol. Energy Mater. Sol. Cells, 2009, 93, 684. 28 B. Weber, S. Mahapatra, T. F. Watson, and M. Y. Sim- mons, Nano Lett., 2012, 12, 4001. 29 M. Fuechsle, J. A. Miwa, S. Mahapatra, H. Ryu, S. Lee, O. Warschkow, L. C. L. Hollenberg, G. Klimeck, and M. Y. Simmons, Nat. Nanotechnol., 2012, 7, 242. 30 O. Wolf, M. Dasog, Z. Yang, I. Balberg, J. G. C. Veinot, 15 B. J. Simonds, I. Perez-Wurfl, Y.-H. So, A. S. Wan, S. and O. Millo, Nano Lett., 2013, 13, 2516. 7 31 S. D. Wang, Z. Z. Sun, N. Cue, H. Q. Xu, and X. R. Wang, Phys. Rev. B, 2002, 65, 125307. P. Ordej´on, and D. S´anchez-Portal, J. Phys. : Condens. Matter, 2002, 14, 2745. 32 Z. Z. Sun, R. Q. Zhang, W. Fan, and X. R. Wang, J. Appl. 51 P. Ordej´on, E. Artacho, and J. M. Soler, Phys. Rev. B, Phys., 2009, 105, 043706. 1996, 53, R10441. 33 A. Levy Yeyati, A. Martin-Rodero, and F. Flores, Phys. Rev. Lett., 1993, 71, 2991. 34 Y. Meir, N. S. Wingreen, and P. A. Lee, Phys. Rev. Lett., 52 R. Guerra, I. Marri, R. Magri, L. Martin-Samos, O. Pulci, O. Degoli, and S. Ossicini, Phys. Rev. B, 2009, 79, 155320. 53 N. Troullier and J. L. Martins, Phys. Rev. B, 1991, 43, 1991, 66, 3048. 8861. 35 Y. Han, W. Gong, H. Wu, and G. Wei, Physica B, 2009, 54 D. M. Ceperley and B. J. Alder, Phys. Rev. Lett., 1980, 404, 2001. 36 E. Taranko, M. Wiertel, and R. Taranko, J. Appl. Phys., 2012, 111, 023711. 45, 566. 55 J. P. Perdew and A. Zunger, Phys. Rev. B, 1981, 23, 5048. 56 M. Ribeiro, L. R. C. Fonseca, and L. G. Ferreira, Phys. 37 J. Carreras, O. Jambois, S. Lombardo, and B. Garrido, Rev. B, 2009, 79, 241312. Nanotechnology, 2009, 20, 155201. 38 G. Conibeer, M. Green, E-C. Cho, D. Konig, Y-H. Cho, T. Fangsuwannarak, G. Scardera, E. Pink, Y. Huang, T. Puzzer, S. Huang, D. Song, C. Flynn, S. Park, X. Hao, D. Mansfield, Thin Solid Films, 2008, 516, 6748. 39 C. Flynn, D. Konig, I. Perez-Wurfl, M. A. Green, and G. Conibeer, Semicond. Sci. Technol., 2010, 25, 045011. 40 G. Conibeer, M. A. Green, D. Konig, I. Perez-Wurfl, S. Huang, X. Hao, D. Di, L. Shi, S. Shrestha, B. Puthen- Veetil, Y. So, B. Zhang, Z. Wan, Prog. Photovolt: Res. Appl., 2011, 19, 813. 41 S. Illera, J. D. Prades, A. Cirera, and A. Cornet, The Scientific World Journal, 2015, 2015, 426541. doi:10.1155/2015/426541 42 S. Illera, N. Garcia-Castello, J. D. Prades, and A. Cirera, J. Appl. Phys., 2012, 112, 093701. 43 S. Illera, J. D. Prades, A. Cirera, and A. Cornet, Europhys. Lett., 2012, 98, 17003. 44 N. Garcia-Castello, S. Illera, R. Guerra, J. D. Prades, S. Ossicini, and A. Cirera, Phys. Rev. B, 2013, 88, 075322. 45 M. Di Ventra, S. T. Pantelides, and N. D. Lang, Phys. Rev. Lett., 2000, 84, 979. 46 Z. X. Dai, X. H. Zheng, X. Q. Shi, and Z. Zeng, Phys. Rev. B, 2005, 72, 205408. 47 M. C. Payne, J. Phys. C: Solid State Phys., 1986, 19, 1145. 48 M. Passoni and C. E. Bottani, Phys. Rev. B, 2007, 76, 115404. 49 D. J. DiMaria, D. W. Dong, C. Falcony, T. N. Theis, J. R. Kirtley, J. C. Tsang, D. R. Young, and F. L. Pesavento, J. Appl. Phys., 1983, 54, 5801. 50 J. M. Soler, E. Artacho, J. D. Gale, A. Garc´ıa, J. Junquera, 57 M. Bruno, M Palummo, A. Marini, R. Del Sole, S. Ossicini, Phys. Rev. Lett., 2007, 98, 036807. 58 E. Luppi, F. Iori, R. Magri, O. Pulci, S. Ossicini, E. Degoli, V. Olevano, Phys. Rev. B, 2007, 75, 033303. 59 G. Seguini, S. Schamm-Chardon, P. Pellegrino, and M. Perego, Appl. Phys. Lett., 2011, 99, 082107. 60 K. Seino and F. Bechstedt, and P. Kroll, Phys. Rev. B, 2010, 82, 085320. 61 K. Seino, F. Bechstedt, and P. Kroll, Phys. Rev. B, 2012, 86, 075312. 62 X. D. Pi, R. Gresback, R. W. Liptak, S. A. Campbell, and U. Kortshagen, Appl. Phys. Lett., 2008, 92, 123102. 63 A. R. Stegner, R. N. Pereira, R. Lechner, K. Klein, H. Wiggers, M. Stutzmann, and M. S. Brandt, Phys. Rev. B, 2009, 80, 165326. system, 64 Bulk systems of dopant atoms: for B atom we used the B50 an alpha-tetragonal phase (doi:10.1103/PhysRevLett.90.026103) with IT number 134, space group P42/nnm, a = b = 8.75 A, c = 5.06 A, and angles of 90◦; for P atom we used the black P system, an orthorombic phase (doi:10.1103/PhysRevB.66.161202), with IT number 64, space group Cmca, a = 3.314 A, b = 10.478 A, c = 4.376 A, and angles of 90◦. 65 M. Fukuda, M. Fujii, H. Sugimoto, K. Imakita, and S. Hayashi, Opt. Lett., 2011, 36, 4026. 66 H. Sugimoto, M. Fujii, K. Imakita, S. Hayashi, and K. Akamatsu, J. Phys. Chem. C, 116, 17969. 67 R. Guerra, E. Degoli, and S. Ossicini, Phys. Rev. B, 2009, 80, 155332. 

1912.07078

1

1912

Frictional drag between superconducting LaAlO$_3$/SrTiO$_3$ nanowires

We report frictional drag measurements between two superconducting LaAlO$_3$/SrTiO$_3$ nanowires. In these experiments, current passing through one nanowire induces a voltage across a nearby electrically isolated nanowire. The frictional drag signal contains both symmetric and antisymmetric components. The antisymmetric component arises from the rectification of quantum shot noise in the drive nanowire by the broken symmetry in the drag nanowire. The symmetric component in the drag resistance is ascribed to rectification of thermal noise in the drive nanowire during superconducting-normal transition. The suppression of the symmetric component is observed when a normal nanowire is used as either a drag or drive nanowire with the other nanowire superconducting. The absence of symmetric drag resistance between a normal drag nanowire and a superconducting drive nanowire suggests a higher electron-hole asymmetry in the superconducting LaAlO$_3$/SrTiO$_3$ nanowire arising from the 1D nature of superconductivity at LaAlO$_3$/SrTiO$_3$ interface.

cond-mat.mes-hall

cond-mat

Frictional drag between superconducting LaAlO3/SrTiO3 nanowires Yuhe Tang,1, 2 Jung-Woo Lee,3 Anthony Tylan-Tyler,1, 2 Hyungwoo Lee,3 Michelle Tomczyk,1, 2 Mengchen Huang,1, 2 Chang-Beom Eom,3 Patrick Irvin,1, 2 and Jeremy Levy1, 2 1Department of Physics and Astronomy, University of Pittsburgh, Pittsburgh, Pennsylvania 15260, USA 2Pittsburgh Quantum Institute, Pittsburgh, Pennsylvania 15260, USA 3Department of Materials Science and Engineering, University of Wisconsin-Madison, Madison, Wisconsin 53706, USA Abstract We report frictional drag measurements between two superconducting LaAlO3/SrTiO3 nanowires. In these experiments, current passing through one nanowire induces a voltage across a nearby electrically isolated nanowire. The frictional drag signal contains both symmetric and an- tisymmetric components. The antisymmetric component arises from the rectification of quantum shot noise in the drive nanowire by the broken symmetry in the drag nanowire. The symmetric component in the drag resistance is ascribed to rectification of thermal noise in the drive nanowire during superconducting-normal transition. The suppression of the symmetric component is ob- served when a normal nanowire is used as either a drag or drive nanowire with the other nanowire superconducting. The absence of symmetric drag resistance between a normal drag nanowire and a superconducting drive nanowire suggests a higher electron-hole asymmetry in the superconduct- ing LaAlO3/SrTiO3 nanowire arising from the 1D nature of superconductivity at LaAlO3/SrTiO3 interface. 9 1 0 2 c e D 5 1 ] l l a h - s e m . t a m - d n o c [ 1 v 8 7 0 7 0 . 2 1 9 1 : v i X r a 1 SrTiO3 (STO) has long attracted interest as a superconducting semiconductor [1 -- 3]. Re- cently, interest in the superconducting properties of STO was revived by the development of STO-based heterostructures and nanostructures and with the LaAlO3/SrTiO3 (LAO/STO) system [4] in particular. The LAO/STO two-dimensional interface supports superconduc- tivity, which is electrostatically gateable, and various transport techniques have been used to study the superconductivity at the interface [5]. The superconducting transition tem- perature (TC) has a dome shape as a function of carrier density, which is controllable via a backgate [6]. With the use of conductive-atomic force microscope (c-AFM) lithography, nanoscale control over the conductance of the LAO/STO interface is possible. This tech- nique relies on AFM tip-controlled protonation or deprotonation of the LAO surface, which enables the creation of a wide variety of quantum-confined structures, including supercon- ducting nanowires [7], ballistic 1D electron waveguides [8], and single-electron transistors [9, 10]. These mesoscopic devices, drawn from a well-established toolset of quantum trans- port, often exhibit surprising new properties due to the unique physics of the STO interface such as electron pairing without forming superconductivity [10]. Recently by studying the superconductivity in LAO/STO nanowires of different widths and numbers, it is discovered that superconductivity exists at the boundary of nanowires and is absent within the inte- rior region of nanowires, which indicates the 1D nature of superconductivity at LAO/STO interface [3]. Coulomb drag [11], or more generally frictional drag, first proposed by Pogrebinskii [12], has proven to be a powerful technique to study electron transport and electron correlations. When two electrical conductors are placed in close proximity, current driven through one ("drive") conductor may induce a voltage (or current) in the second ("drag") conductor. Frictional drag measurements have mostly been carried out between normal state conductors in coupled 2D semiconductor systems [13 -- 17], graphene systems [18, 19], 1D semiconductor systems [20 -- 22], 1D complex oxide systems [23], and quantum dot systems [24]. Frictional drag in the superconducting regime has been carried out in normal metal-superconductor systems [25, 26] and the phenomenon is explained by the local fluctuating electric field induced by mobile vortices in the superconducting layer [27] or Coulomb coupling between two conductors. [28, 29]. There are, to our knowledge, no prior reports of frictional drag between two quasi-1D superconductors. Previously-reported frictional drag experiments at the LAO/STO-based nanowires have 2 shown surprising results, particularly in the high magnetic field regime [23]. The drag resistance is anti-symmetric, indicating that the drag resistance arises via rectification of quantum shot noise in the drive nanowire due to the broken inversion symmetry of the drag nanowire [30]. Remarkably, the drag resistance shows little to no dependence on the separation between nanowires (up to ∼ µm scales). This unusual scaling strongly indicates that non-Coulombic interactions dominate the coupling between these nanowires. Here we report frictional drag experiments between two LAO/STO superconducting nanowires. The drag resistance contains a mixture of symmetric and anti-symmetric com- ponents and the symmetric component disappears whenever one nanowire is normal and the other is superconducting. The antisymmetric component arises for the same reasons as in the high B regime. The symmetric component is ascribed to the rectification of thermal noise in the drive nanowire during the superconducting-normal transition. Suppression of the symmetric drag component, when a normal nanowire is used as the drag nanowire, suggests the existence of a higher electron-hole asymmetry [31] in the superconducting LAO/STO nanowires arises from the 1D nature of superconductivity at LAO/STO interface. Nanowire devices are "sketched" on LAO/STO heterostructures using c-AFM lithography [32] (Fig. 1(a)). LAO/STO heterostructures with an LAO thickness of 3.4 unit cells are grown by pulsed laser deposition (PLD). Further details of the sample growth and the device fabrication process are described elsewhere [33]. The width of the nanowires used for these experiments is approximately w = 10 nm, as quantified by erasure experiments [32]. Other device parameters include the separation between nanowires d and the nanowire length L. Here we focus on two sets of parameters: d = 40 nm and L = 400 nm (device 2B, Fig. 2) and d = 40 nm and L = 300 nm (device 2J, Fig. 4). To investigate frictional drag at the LAO/STO interface in the superconducting regime, the magnitude of B is kept below 0.3 T and and the temperature less than 100 mK (except for temperature-dependent measurements that explicitly go above T = 100 mK). In a frictional drag experiment, a voltage Vi in nanowire i is induced by a current Ij in nanowire j (Fig. 1(b)). All nanowires are connected to the same ground during the measurement. The current Ij is produced by applying a voltage VSj = VDC + VAC cos ωt to one end of nanowire j; the resulting current Ij(ω) and induced voltage Vj(ω) at frequency ω are measured using a lock-in amplifier. The resistance may then be expressed as a matrix Rij = dVi/dIj = Vi(ω)/Ij(ω), which is generally a function of the DC drive current Ij (as well as other parameters such as temperature T 3 FIG. 1. Experimental setup. (a) Side-view of the nanowire fabrication process. A nanowire is created at the LAO/STO interface between two Ti/Au electrical contacts with c-AFM lithography. Protons (+) patterned on the surface by the AFM tip attract electrons (−) to the interface forming a nanowire (green area). (b) Top-view schematic of the double nanowire device with length L, width w, and nanowire separation d. The setup measures the induced drag voltage V1 across nanowire 1 created by current I2, which is induced by application of a voltage VS2 across nanowire 2. and applied magnetic field (cid:126)B). The off-diagonal terms then define the drag resistance Rij characterize the mutual friction between electrons in the drive and drag nanowires. In order to ensure that the drag resistances Rij are not influenced by current leakage between the two nanowires, all measurements are performed well below the inter-wire breakdown voltage (∼10 V) measured for each device. Typical frictional drag resistance measurements in the superconducting regime are shown in Fig. 2. Both nanowires in device 2B show signatures of superconductivity [7]. As shown in the bottom panels of Fig. 2(a) and (b), nanowire 2 displays three superconducting- normal transitions with critical current Ic defined as the location of the peaks in R2T,2 [7]. The first is at ±20 nA, the second at ±110 nA, and the third at ±140 nA. Non-vanishing resistances in superconducting nanowires are common and are attributed to normal hotspots below Ic [34] or quantum phase slips [25]. The superconducting-normal transition at ±20 nA arises from the nanowire since it shows up both in R2T, 2 and four-terminal resistance R22 and the transition at ±110 nA and ±140 nA arises from wires connecting the nanowire and electrodes since it only shows up in R2T,2 (Fig. S1). The drag resistance R12 is greatly enhanced in the superconducting regime, as can be seen by examining both the temperature- dependence (Fig. 2(c)) and the magnetic-field dependence (Fig. 2(d)). The nature of R12 in the superconducting regime is qualitatively different from the high magnetic field regime (where the nanowires are not superconducting). In the high magnetic field regime, the 4 -----V2Ti/AuLaAlO3SrTiO3(a)(b)I2V112wdLVs2+++++FIG. 2. Temperature and magnetic-field dependence of the drag resistance and two-terminal resis- tance of the drive nanowire. (a) Temperature dependence and line profiles of the drive nanowire's two-terminal resistance R2T,2 from nanowire 2. Top panel, temperature dependence of drag resis- tance R12 from nanowire 1. Bottom panel, line profiles of R12 at 50 mK and 400 mK. (b) Magnetic field dependence and line profiles of the drive nanowire's two-terminal resistance R2T,2. (c) Tem- perature dependence and line profiles of drag resistance R12 from nanowire 1. (d) Magnetic field dependence and line profiles of drag resistance R12. drag resistance Rij is antisymmetric [23] with respect to the sourcing current, while the superconducting response is asymmetric with drive current. The superconducting Rij is mostly symmetric between I2 = ±40 nA with two tiny dips at ±10 nA. As the magnitude of I2 increases, an anti-symmetric component starts showing up in Rij and Rij becomes 5 500400300200100T (mK)3020R2T, 2(kΩ)-0.30-0.20-0.100.00B (T)5-5R12(Ω)500400300200100T (mK)5-5R12(Ω)-0.30-0.20-0.100.00B (T)3020R2T, 2(kΩ)20100-10R12(Ω)150-150I2(nA)50mK400mK403020R2T, 2(kΩ)150-150I2 (nA)50mK400mK3020R2T(kΩ)150-150I2(nA)0T-0.3T100-10R12(Ω)150-150I2(nA)0T-0.3T(a)(b)(c)(d)FIG. 3. Symmetric and anti-symmetric components of drag resistance. (a) Typical symmetric and anti-symmetric components of drag resistance from device 2B. Top panel: Two-terminal resistance R2T,2 of drive nanowire. Middle panel: Symmetric component of drag resistance R12. Bottom panel: Anti-symmetric component of drag resistance R12. (b) d dependence of symmetric and anti-symmetric components. d ranges from 40nm to 1.5µm. Top panel: Symmetric component RS ij as a function of d. Bottom panel: Anti-symmetric component RA ij as a function of d. asymmetric. The appearance of asymmetric R12 (Fig. 2(c) and (d)) in the superconducting regime is correlated with the superconductivity in the drive nanowire 2 (Fig. 2(a) and (b)). To further symmetric components by RS ij and RA RS ij(I) = (Rij(I)+Rij(−I))/2 and RA R2T, 2, RS pinpoint locally strongest drag resistance in RS understand the frictional drag in the superconducting regime, we extract symmetric and anti- ij(I) = (Rij(I)−Rij(−I))/2. ij are shown in top, middle an bottom panels of Fig. 3(a). Dashed lines 12. As shown in Fig. 3(a), locally strongest 12 show up around the superconducting-normal transition represented by peaks in R2T, 2 in 12. The nature of coupling between ij is still unknown. But according to devices with d ranging from ij and RA ij persist over large separations and are nearly independent nanowires for RS ij and RA 40nm to 1.5µm , both RS the drive nanowire 2 accompanied by locally strongest RA 6 -10010R12(Ω)-10010R12(Ω)100-100I2(nA)30252015R2T, 2(kΩ)110Rij(Ω)110Rij(Ω)1000200d (nm)SASA(a)(b)FIG. 4. Frictional drag in the superconducting regime with one normal-state nanowire. (a) Schematic of the device with normal-state nanowire. Black sections in nanowire 2 are normal; green sections in nanowire 2 and 1 are superconducting. (b) Left: Two terminal resistance of nanowire 2 measured between B and C. Superconductivity arises from the green portions as shown in panel (a). Right: Two terminal resistance of nanowire 2 between A and D when the whole nanowire is in normal state. (c) Measurement configurations when nanowire 2 is used as the drive nanowire. (d) Drag resistance R12 and its symmetric component RS 12. Left and right panels correspond to the measurement configuration in (c). (e) Measurement configurations when nanowire 2 is used as the drag nanowire. (f) Drag resistance R21 and its symmetric component RS 12. Left and right panels correspond to the measurement configurations in (e). of d (Fig. 3(b)). Since the e−4kF d behavior is not observed in both RS ij, where kF ∼(10nm)−1 is the Fermi wave vector, the Coulomb coupling can be ruled out as the dominating effect [35]. ij and RA To corroborate that the symmetric component of drag resistance is related to the superconducting-normal transition in the drive nanowire, we examine the drag resistance from devices with one superconducting nanowire and one normal nanowire. The supercon- ducting properties of LAO/STO is known to be gate-tunable both in 2D geometries [36] and in 1D [3, 7]. There are known inhomogeneities in electron density which most likely arise from the underlying ferroelastic domain structure[37]. While we cannot independently control the carrier density of one nanowire while keeping the second fixed, we can select 7 I1V2(a)(d)2ABCD(d)12ABCDV12015105R2T, 2(kΩ)100-100I2(nA)100-100I2(nA)100-100I2(nA)-202R12(Ω)100-100I2(nA)-202R12(Ω)-2-1012R21(Ω)50-50I1(nA)-2-1012R21(Ω)50-50I1(nA)wire 2, BCwire 2, AD(d)ABCDI21V1I212(d)ABCD12I1V2(d)ABCD12(b)(c)(d)(e)(f)SSdevices in which one nanowire shows superconducting behavior and the other does not. Fig. 4 shows the typical data from Device 2J. As illustrated in Fig. 4(a), green-colored nanowires are superconducting, while black nanowires are in the normal-state. The informa- tion about the state of the nanowires is inferred from two-terminal resistance measurements (Fig. 4(b)). We then can compare the frictional drag as sensed by nanowire 1 due to two configurations -- one in which one device contains a superconducting section and one in which the other does not. First we consider the configuration where superconducting nanowire 1 is the drag nanowire and examine the influence of drive nanowire's state on drag resistance, as shown in Fig. 4(c). When both the drive and drag nanowires are superconducting, the drag resistance R12 is asymmetric (Fig. 4(d) left top panel) with a large symmetric component (Fig. 4(d) left bottom panel). However, when the drive nanowire is normal, the drag resistance is mostly anti-symmetric (Fig. 4(d) right top panel) with a negligible symmetric component (Fig. 4(d) right bottom panel). The symmetric component of drag resistance showing up around the superconducting- normal transition in the drive nanowire can be explained by the rectification of the thermal noise in the drive nanowire [30] which requires electron-hole asymmetry in both drive and drag nanowires. When a superconducting nanowire undergoes superconducting-normal tran- sition, the nanowire's resistance increases. This process generates thermal energy, which in turn gives rise to a large thermal noise and a greatly enhanced symmetric component of drag resistance. For the normal nanowire, therefore there is no significant enhancement of the thermal noise, and the symmetric component of drag resistance remains small at all biases across the drive nanowire. The rectification of thermal noise in the drive also explains the strong correlation be- tween RA 12. RA 12 and RS 12 comes from the rectification of the shot noise in the drive nanowire [23]. Shot noise is a non-equilibrium phenomenon depending on the voltage bias across the drive nanowire [30]. During the superconducting-normal transition in the drive nanowire, the change of drive nanowire's resistance changes the bias across different portions of the nanowire, thus inducing quantum shot noise and anti-symmetric Rij is observed simultane- ously with symmetric drag resistance. The symmetric component in drag resistance is also strongly suppressed when the drag nanowire is in the normal state. As shown in the left panel of Fig. 4(e), when the drag 8 resistance is measured between B and C of nanowire 2, the drag resistance R21 is asymmetric with a large symmetric component (Fig. 4(f) left). However when the drag resistance is measured between A and D, the drag resistance is anti-symmetric with a negligible symmet- ric component. Since the drive nanowire 1 is superconducting in both configurations, the absence of symmetric drag resistance component with a normal drag nanowire cannot be ascribed to the absence of thermal noise in the drive nanowire. The magnitude of symmetric drag resistance depends on the electron-hole asymmetry in the drag nanowire [11]. The fact that the symmetric drag resistance measured from a superconducting drag nanowire is larger may be explained by the electron-hole asymmetry is stronger in superconducting nanowire than normal nanowire. Electron-hole symmetry is more easily broken in low dimensional devices [30, 31]. It is reported that the superconductivity at LAO/STO interface is 1D in nature situated at the boundary of the nanowire and is absent within the interior region of the nanowire [3]. Thus the overall dimension of the nanowire is reduced as it becomes superconducting compared to a normal nanowire due to the formation of 1D superconduct- ing boundary. This reduced dimension of the nanowire gives rise to a stronger electron-hole asymmetry. Therefore the symmetric component of drag resistance is stronger measured from a superconducting drag nanowire. In summary, frictional drag between superconducting LAO/STO nanowires exhibits a strong and highly symmetric component in drag resistance, which is distinct from the anti- symmetric drag resistance between LAO/STO nanowires in the normal state. The symmetric component arises from rectification of thermal noise in the drive superconducting nanowire based on the fact that it shows up at the vicinity of superconducting-normal transition in the drive nanowire and disappears when the drive nanowire is normal. The symmetric component in drag resistance also disappears when the drag nanowire is normal, which can be attributed to the 1D nature of superconductivity in LAO/STO systems. Work at the University of Pittsburgh was supported by funding from the DOE Office of Basic Energy Sciences under award number DOE de-sc0014417. The work at the UW- Madison (synthesis and characterizations of thin films heterostructures) was supported by the US Department of Energy (DOE), Office of Science, Office of Basic Energy Sciences (BES), under award number DE-FG02-06ER46327. 9 [1] J. Schooley, W. Hosler, and M. L. Cohen, Physical Review Letters 12, 474 (1964). [2] X. Lin, Z. Zhu, B. Fauqu´e, and K. Behnia, Physical Review X 3, 021002 (2013). [3] Y.-Y. Pai, H. Lee, J.-W. Lee, A. Annadi, G. Cheng, S. Lu, M. Tomczyk, M. Huang, C.-B. Eom, P. Irvin, et al., Physical review letters 120, 147001 (2018). [4] A. Ohtomo and H. Hwang, Nature 427, 423 (2004). [5] C. Richter, H. Boschker, W. Dietsche, E. Fillis-Tsirakis, R. Jany, F. Loder, L. Kourkoutis, D. Muller, J. Kirtley, C. Schneider, et al., Nature 502, 528 (2013). [6] A. Caviglia, S. Gariglio, N. Reyren, D. Jaccard, T. Schneider, M. Gabay, S. Thiel, G. Hammerl, J. Mannhart, and J.-M. Triscone, Nature 456, 624 (2008). [7] J. P. Veazey, G. Cheng, P. Irvin, C. Cen, D. F. Bogorin, F. Bi, M. Huang, C.-W. Bark, S. Ryu, K.-H. Cho, et al., Nanotechnology 24, 375201 (2013). [8] A. Annadi, G. Cheng, H. Lee, J.-W. Lee, S. Lu, A. Tylan-Tyler, M. Briggeman, M. Tomczyk, M. Huang, D. Pekker, et al., Nano letters 18, 4473 (2018). [9] G. Cheng, P. F. Siles, F. Bi, C. Cen, D. F. Bogorin, C. W. Bark, C. M. Folkman, J.-W. Park, C.-B. Eom, G. Medeiros-Ribeiro, et al., Nature Nanotechnology 6, 343 (2011). [10] G. Cheng, M. Tomczyk, S. Lu, J. P. Veazey, M. Huang, P. Irvin, S. Ryu, H. Lee, C.-B. Eom, C. S. Hellberg, and J. Levy, Nature 521, 196 (2015). [11] B. Narozhny and A. Levchenko, Reviews of Modern Physics 88, 025003 (2016). [12] M. Pogrebinskii, Soviet Physics-Semiconductors 11, 372 (1977). [13] T. Gramila, J. Eisenstein, A. MacDonald, L. Pfeiffer, and K. West, Physical review letters 66, 1216 (1991). [14] T. Gramila, J. Eisenstein, A. MacDonald, L. Pfeiffer, and K. West, Surface science 263, 446 (1992). [15] T. Gramila, J. Eisenstein, A. MacDonald, L. Pfeiffer, and K. West, Physica B: Condensed Matter 197, 442 (1994). [16] P. Solomon and B. Laikhtman, Superlattices and Microstructures 10, 89 (1991). [17] J. Eisenstein, Superlattices and microstructures 12, 107 (1992). [18] J. Li, T. Taniguchi, K. Watanabe, J. Hone, A. Levchenko, and C. Dean, Physical review letters 117, 046802 (2016). 10 [19] K. Lee, J. Xue, D. C. Dillen, K. Watanabe, T. Taniguchi, and E. Tutuc, Physical review letters 117, 046803 (2016). [20] P. Debray, V. Zverev, O. Raichev, R. Klesse, P. Vasilopoulos, and R. Newrock, Journal of Physics: Condensed Matter 13, 3389 (2001). [21] M. Yamamoto, M. Stopa, Y. Tokura, Y. Hirayama, and S. Tarucha, Science 313, 204 (2006). [22] D. Laroche, G. Gervais, M. Lilly, and J. Reno, Science 343, 631 (2014). [23] Y. Tang, A. Tylan-Tyler, H. Lee, J.-W. Lee, M. Tomczyk, M. Huang, C.-B. Eom, P. Irvin, and J. Levy, Advanced Materials Interfaces , 1900301 (2019). [24] A. Keller, J.-S. Lim, D. S´anchez, R. L´opez, S. Amasha, J. Katine, H. Shtrikman, and D. Goldhaber-Gordon, Physical review letters 117, 066602 (2016). [25] N. Giordano and J. Monnier, Physical Review B 50, 9363 (1994). [26] X. Huang, G. Baz`an, and G. H. Bernstein, Physical review letters 74, 4051 (1995). [27] E. Shimshoni, Physical Review B 51, 9415 (1995). [28] A. Kamenev and Y. Oreg, Physical Review B 52, 7516 (1995). [29] J.-M. Duan and S. Yip, Physical review letters 70, 3647 (1993). [30] A. Levchenko and A. Kamenev, Physical review letters 101, 216806 (2008). [31] B. Narozhny and I. Aleiner, Physical review letters 84, 5383 (2000). [32] C. Cen, S. Thiel, G. Hammerl, C. Schneider, K. Andersen, C. Hellberg, J. Mannhart, and J. Levy, Nature materials 7, 298 (2008). [33] K. A. Brown, S. He, D. J. Eichelsdoerfer, M. Huang, I. Levy, H. Lee, S. Ryu, P. Irvin, J. Mendez-Arroyo, C.-B. Eom, M. C. A., and L. Jeremy, Nature communications 7, 10681 (2016). [34] M. Tinkham, J. Free, C. Lau, and N. Markovic, Physical Review B 68, 134515 (2003). [35] O. Raichev and P. Vasilopoulos, Physical Review B 61, 7511 (2000). [36] N. Reyren, S. Thiel, A. Caviglia, L. F. Kourkoutis, G. Hammerl, C. Richter, C. Schnei- der, T. Kopp, A.-S. Ruetschi, D. Jaccard, M. Gabay, D. A. Muller, J.-M. Triscone, and J. Mannhart, Science 317, 1196 (2007). [37] A. Nethwewala, H. Lee, M. Briggeman, Y. Tang, J. Li, J. Lee, C.-B. Eom, P. R. Irvin, and J. Levy, Nanoscale Horizons 4, 1194 (2019). 11 Supplemental Material In the supplement materials, Fig. S1 shows the typical two-terminal (R2T, 2) and four- terminal resistance (R22) from a nanowire. The superconducting-normal transition at small bias showing up in both two-terminal and four-terminal resistance comes from the nanowire. The extra superconducting-normal transitions at larger biases in R2T, 2 come from wires connecting the nanowire and the electrode. FIG. S1. Typical two-terminal resistance and four-terminal resistance from a nanowire. Top panel: Four-terminal resistance R22 and superconducting-normal transition from the nanowire only shows up at small bias from ±20 nA. Bottom panel: Two-terminal resistance R2T, 2. Besides the superconducting-normal transition at small bias extra superconducting-normal transitions show up at larger bias ±110 nA and ±150 nA. Superconducting-normal transitions at larger bias come from wires connecting the nanowire and electrodes. 12 4.03.83.63.4R22(kΩ)2624222018R2T, 2(kΩ)-1000100I2(nA)

1309.4971

2

1309

Ultrahigh Q-Frequency product for optomechanical disk resonators with a mechanical shield

We report on optomechanical GaAs disk resonators with ultrahigh quality factor - frequency product Qf. Disks standing on a simple pedestal exhibit GHz breathing modes attaining a Qf of 10^13 measured under vacuum at cryogenic temperature. Clamping losses are found to be the dominant source of dissipation in this configuration. A new type of disk resonator integrating a shield within the pedestal is then proposed and its working principles and performances investigated by numerical simulations. For dimensions compatible with fabrication constraints, the clamping-loss-limited Q reaches 10^7-10^9 corresponding to Qf of 10^16-10^18. This shielded pedestal approach applies to any heterostructure presenting an acoustic mismatch.

cond-mat.mes-hall

cond-mat

Ultrahigh Q-Frequency product for optomechanical disk resonators with a mechanical shield D. T. Nguyen,1, a) C. Baker,1 W. Hease,1 S. Sejil,1 P. Senellart,2 A. Lemaıtre,2 S. Ducci,1 G. Leo,1 and I. Favero1 1)Laboratoire Mat´eriaux et Ph´enom`enes Quantiques, Universit´e Paris Diderot -- CNRS, 10 rue Alice Domon et L´eonie Duquet, 75013 Paris, France 2)Laboratoire de Photonique et de Nanostructures, route de Nozay, 91460 Marcoussis, France (Dated: 18 June 2021) We report on optomechanical GaAs disk resonators with ultrahigh quality factor - frequency product Q · f . Disks standing on a simple pedestal exhibit GHz breathing modes attaining a Q · f of 1013 measured under vacuum at cryogenic temperature. Clamping losses are found to be the dominant source of dissipation in this configuration. A new type of disk resonator integrating a shield within the pedestal is then proposed and its working principles and performances investigated by numerical simulations. For dimensions compatible with fabrication constraints, the clamping-loss-limited Q reaches 107 − 109 corresponding to Q · f of 1016 − 1018. This shielded pedestal approach applies to any heterostructure presenting an acoustic mismatch. Keywords: Optomechanics, mechanical resonator, GaAs disk resonator, mechanical quality factor, clamping. The interaction of light with mechanical motion - optomechanics1 -- 3 and its related concepts - is now in- vestigated in a wide variety of experimental settings. Optomechanical resonators of various size and geome- try continue to be developed and optimized for appli- cations like weak force sensing4,5 or optical cooling of mesoscopic mechanical systems down to the quantum regime6,7. For most of these applications, high mechan- ical frequency f, strong optomechanical coupling g0 and low optical/mechanical dissipation are desirable. Among various systems, Gallium Arsenide (GaAs) optomechan- ical disk resonators bring together all of these features with a relatively simple geometry8 and the possibility of complete on-chip optical integration9. The sub-micron optical and mechanical confinement leads to GHz me- chanical frequencies and g0 in the MHz10. Optical qual- ity factors reach today several 105 and optical dissipation sources are progressively unraveled to approach the ra- diative limit. On the mechanical dissipation side, the un- derstanding of loss mechanisms in GaAs disks is largely incomplete, despite recent efforts to model their fluid damping for air or liquid operation11. Mechanical resonators are often compared on the ba- sis of their mechanical Q · f factor (quality factor times frequency) because of the paramount importance of this figure of merit for the performances of MEMS devices12. But Q · f also turns out to play a key role in the quan- tum realm, where it indicates the number of independent operations N that can be performed on a quantum me- chanical system subject to thermal decoherence induced by an environment at temperature T 13. More specifi- cally, the number of coherent oscillations in presence of a thermal environment is given by Q · f × h/(kBT ), which indicates that a Q · f higher than 6 · 1012 is necessary to attain one coherent oscillation at room temperature. a)Electronic mail: [email protected] Two independent works have demonstrated record values for Q· f in the 1015 − 1016 range for quartz resonators at ultra-low temperature14,15 and very recent developments on Silicon optomechanical crystals allowed reaching a Q·f of 10147. Apart from these three works, current state of the art systems are evolving in the 1010 − 1013 window (see Ref.13 and3 for more comprehensive reviews on this topic) and are based on Quartz, Silicon, and Polysilicon, Silicon Nitride or Diamond, and with very few reports on III-V semiconductors. First studies on GaAs disks re- ported Q · f products between 1011 and 1012 in ambient conditions8,10, at the forefront of GaAs based mechanical systems16 -- 18. In this letter we focus on mechanical dissipation and the Q·f factor in GaAs disks, with an emphasis on clamp- ing losses. By measuring and modeling the mechanical Q of disks of varying pedestal radius, we find that clamp- ing loss is the dominant loss mechanism when these res- onators sit on a simple central pedestal and are operated in vacuum at low temperature. As compared to our pre- vious work, the improved control of the pedestal fabri- cation allows miniature GaAs disk resonators reaching a Q · f factor of 1013. Building on the obtained numerical understanding of clamping losses, we propose a mechan- ical shield geometry that allows boosting the clamping limited Q from some thousands in the simple-pedestal geometry to the 107 − 109 range with the shield, corre- sponding to a clamping-limited Q·f factor of 1016−1018. We present an optimization of this shield geometry, no- tably under fabrication constraints, and give a physical discussion of the decoupling of the disk from its support in this geometry. The disk samples are fabricated from a GaAs (320 nm)/Al0.8Ga0.2As (1800 nm) bilayer grown by molecu- lar beam epitaxy (MBE) on a GaAs substrate. Relevant mechanical properties of GaAs and Al0.8Ga0.2As seen as isotropic elastic materials are summarized in Table I. Disks of radius 1 µm are positioned in the vicinity of tapered suspended GaAs waveguides to allow evanescent 3 1 0 2 p e S 0 2 ] l l a h - s e m . t a m - d n o c [ 2 v 1 7 9 4 . 9 0 3 1 : v i X r a Parameter Unit GaAs Al0.8Ga0.2As Young's modulus (E) GPa 85.9 Density (ρ) kg/m3 5317 Poisson's ratio (ν) - 0.310 83.9 4072 0.318 TABLE I. Mechanical Al0.8Ga0.2As20,21. properties of GaAs19 and optical coupling of light into the disks. They are pat- terned in a resist mask by electron beam lithography and then dry-etched by non-selective Inductively Cou- pled Plasma Reactive Ion Etching (ICP-RIE) using a mixture of SiCl4 and Ar plasmas. Pedestals are formed by hydrofluoric acid (HF) selective underetching of the AlGaAs sacrificial layer. A diluted HF:H2O solution (1.22 % in volume) at 4◦C is combined with a slow agita- tion in the solution to allow fabricating disks with a con- trolled pedestal radius as small as 60 nm. Protecting the AlGaAs parts from air oxidation by putting the sample in acetone right after ICP-RIE proved crucial to obtain the degree of control needed in the present work. Fig. 1a shows a finished GaAs disk and its coupling waveguide with smooth sidewalls. We first experimentally measure the mechanical spec- trum of several GaAs disk resonators. Optical probing with a laser blue-detuned to an optical whispering-gallery resonance gives access to the mechanical spectrum. By virtue of the optomechanical coupling, the disk mechani- cal fluctuations are imprinted in the optical transmission noise and analyzed at the photodetector output using a spectrum analyzer. Details of the experiment setup can be found in our previous work8 -- 10. In the low laser power limit, the obtained spectra provide the intrinsic frequen- cies and quality factors of the disk mechanical modes. We focus here on the mechanical breathing mode of the disks, which appears around 1.37 GHz for the considered dimensions. Experiments are run both at room tempera- ture and at 8 K. Measurements reveal that a reduction of pedestal radius from 500 to 100 nm barely lowers the me- chanical frequency (less than 5%, not shown here). The measured Q factors are presented in Fig. 1b and show a considerable 15-fold increase between 300 and 70 nm of pedestal radius at room temperature in air (red squares). Under vacuum operation at low temperature (8 K), a Q of 6500 ± 1000 is measured for the smallest investigated pedestal radius (blue circles). This corresponds to the highest Q · f value reported in GaAs disk resonators, at- taining the 1013 limit. In air, Q is limited at about 1700, probably due to air damping11. Experimental data are compared to finite-element method (FEM) simulations using the Comsol software. Fig. 1c depicts a GaAs disk in a 2D axisymmetric model. To reflect the morphology resulting from fabrication, the disk is modeled by a cylinder and the pedestal by a cylin- der followed by an isotropic etch profile. The disk and its support are standing on a substrate surrounded by per- 2 FIG. 1. GaAs disks with a simple pedestal. (a): SEM side- view of a GaAs disk and its coupling optical waveguide in the background (in purple false color). The waveguide is 200 nm wide and has the same thickness of the disk. Roughness on the ground is due to HF wet etching of AlGaAs (in gray). (b) Experimental data (open red squares corresponding to val- ues measured in air at 300 K and blue circles corresponding to values measured in vacuum at 8 K) and numerical simu- lations (black solid line) for the mechanical Q of disk radial breathing mode as a function of pedestal radius. (c) Mechani- cal modeling of a GaAs disk resonator. The driving force used to simulate the disk's spectral response is a uniform pressure acting on the disk's sidewall (white arrow). (d) Simulated instantaneous displacement field (in log color scale). Red ar- rows show the propagation of the deformation wave. fectly matched layers (PML) that emulate the attenua- tion of the deformation wave. The dimensions of the sub- strate and PML parts are optimized to correctly simulate wave propagation within a manageable computing time. Fig. 1d shows the first radial breathing mode (RBM) of a disk with a frequency around 1.37 GHz. This is the only mechanical mode detected in this frequency range in the experiments. Fig. 1d shows clearly the propagation of the deformation wave from the disk through the pedestal before dissipating in the substrate. The simulated me- chanical Q versus the pedestal radius is shown as a solid line in Fig. 1b, which captures qualitatively and quan- titatively the increase of Q for smaller pedestal radius. The residual differences between experimental data and simulations can be ascribed to geometry imperfections and other dissipation channels such as surface-state loss which depend on the bath temperature. For example, we have observed a reduction of Q to about 3000 (data not shown here) for the smallest radius disks as the tempera- ture is increased from 8 to 300 K. Our experimental and numerical results indicate that support loss is the main dissipation channel of these GaAs disk optomechanical resonators at 8 K. As a consequence, a simple way to boost Q is to reduce the impact of the anchoring points. A first natural route is to further reduce the pedestal ra- 0100200300400500101102103104Mechanical Q-factorPedestal radius (nm)DiskPedestalPML 2PML 3PML 1AlGaAsGaAs1 µm(a)(c)1 µm(d)-35-40-50-45Symmetryaxis(b)dius, but this comes with two major drawbacks: Firstly, the system would become extremely fragile and secondly, thermal effects like thermo-optical instabilities would be exacerbated, given that the pedestal is also the main thermal dissipation channel. Here we explore a second route inspired by phononic Bragg mirrors in order to prevent the disk deformation wave from escaping to the substrate and confine it in the resonator22. To form a periodic multilayered acous- tic Bragg mirror, one needs two or more materials with different acoustic impedances, which are naturally GaAs and Al0.8Ga0.2As in our case. The phononic mirror could in principle be integrated within the substrate, just under the disk pedestal, or within the pedestal itself. However, Fig. 1d reveals that the deformation wave becomes quasi- spherical as it exits the pedestal and propagates through the substrate. Hence a conventional planar Bragg mirror under the pedestal would not block the wave efficiently, as confirmed by our simulations (not shown). Therefore the Bragg structure must be integrated directly into the pedestal in order to prevent the wave from reaching the substrate. A standard Bragg mirror consists of quarter- wavelength (λ/4) layers. For a phonon mode at frequency of 1.37 GHz considered here, the "acoustic" wavelength is λ = 3.4 µm in GaAs and 3.9 µm in Al0.8Ga0.2As. This implies that each layer has a thickness of about 1 µm. Be- sides, the first AlGaAs layer under the GaAs disk should be thick enough to minimize light coupling from the disk to the substrate in the final device, and thin enough to avoid growth and etching difficulties. This leads us to chose an optimal thickness of 2 µm. Finally, because of fabrication limitations, we avoid vertical etch depths of more than 10 µm and therefore focus on structures with a small number of layers. These constraints lead us to the structure shown in Fig. 2a. It consists of a 320 nm thick GaAs disk on a first AlGaAs pedestal of 2 µm in length, that stands on a GaAs "shield", the latter topping a second (lower) AlGaAs pedestal to isolate the shield from the GaAs sub- strate. This structure can be fabricated using the same techniques as for the disks with simple pedestal. As a result, the disk and the shield will have the same ra- dius of 1 µm and the upper and lower pedestals share a common radius. The structure's mechanical properties are computed numerically as in the simple-pedestal case above (i.e. the disk, the pedestals and the shield are mod- eled by cylinders in a 2D axisymmetric approach). The adjustable dimensions for optimization are the pedestal radius r, the shield thickness t and the height h of the lower pedestal. Compared to a simple disk presented in Fig. 1c, the higher color contrast between the disk and the substrate in Fig. 2a illustrates the good efficiency of the shield at confining the mechanical RBM mode within the disk. Fig. 3b shows the computed mechanical modes of the shielded resonator in the frequency range of the radial breathing mode considered above (1.30-1.45 GHz). As the shield thickness varies from 200 to 1600 nm, three 3 FIG. 2. Shielded disk resonator. (a) Geometry of the shielded disk. Disk and shield are both 1 µm in radius. Color log scale represents the displacement field for a breathing mode at 1.37 GHz. Notice the high color contrast between the disk and sub- strate showing excellent isolation of mechanical vibration. (b) Mechanical mode dispersion as a function of the shield thick- ness. Blue dotted lines represent mechanical modes. The bold red line points towards modes with the highest disk radial dis- placement amplitude. Pictures show the displacement profile of selected modes. distinct mechanical modes appear, each of them repre- sented by a blue dotted dispersion line. The pedestal ra- dius and the lower pedestal height are respectively fixed to 180 nm and 550 nm in these simulations (h = 550 nm is optimal as discussed below). As apparent in the displacement profiles shown in the figure, normal modes of the structure result mainly from a coupling between eigenmodes of the disk and the shield. Each mode shows a disk radial breathing nature in a given shield thickness range highlighted by the red line. The radial breathing mode of the disk is hybridized with eigenmodes of the shield and the details of this hybridization will be impor- tant to minimize support losses. To determine the optimal geometry with minimal sup- port losses and highest Q we proceed as follows: 1) Study the dependence of Q on the shield thickness t and pedestal radius r by fixing h at a certain value. 2) Ana- lyze the dependence of Q on h for several values of (t; r). This step reveals that the optimal h = 550 nm is in- dependent of the selected (t; r) in our investigation. 3) Finally repeat the first step with the optimal value of h. This strategy leads to the results shown in Fig. 3. The clamping-limited mechanical Q of the disk GHz breathing mode can reach extremely large values. In Fig. 3a, a first noticeable region of the parameter space around (t; r) = (1425; 185) nm provides for example Q = 107−109. This configuration is however not the best technological compromise because of the related narrow tolerances on t and r and the rapid drop of Q if t be- comes too large. To account for some 1% imprecision on the epitaxial thickness t and for the realistic pedestal etching conditions - obtaining a 10 nm precision on r in the region r ∼ 150 nm remains challenging - we estimate that a good compromise is reached in another region of the parameter space (with 1350 ≤ t ≤ 1400 nm and 400800120016001.301.351.401.45Frequency (GHz)Shield thicknesst(nm)ShieldSubstrate1 µm(a)Lower pedestal-40-50-60Pedestal(b)Diskthr ≤ 200 nm) that also provides ultrahigh Q from 3 · 105 to 109, but with larger tolerances on the fabrication. FIG. 3. Q optimization of the geometry presented in Fig. 2a. (a) Contour color map in log scale of the calculated Q-factor of a shielded disk resonator as a function of shield thickness and pedestal radius for h = 550 nm. (b) Radial breathing mode frequency and Q-factor as a function of the lower pedestal's height h, with (t; r) = (1000; 180) nm. Frequency scale is offset by f0 = 1.38 GHz for clarity. Even though our shielded disk geometry is originally its dimensions and inspired by planar Bragg mirrors, working principles are somewhat different due to the small number of employed layers and their finite lateral size. The boost of Q in shielded resonators can be un- derstood in terms of interference between deformation waves emitted by the disk and the shield. In case of Q enhancement, the interference is destructive in the lower pedestal, which results in a better isolation from the sub- strate. In this situation the disk and shield oscillate in anti-phase: when the disk expands, the shield contracts. The disk's expansion pulls the rest of the structure to- wards the disk while the shield contraction pushes away. These two actions add up constructively in the top parts of the structure (above the shield) but cancel out at the lower pedestal. For specific values of the shield thickness t this cancellation is quasi-total, resulting in ultrahigh Q. This interpretation as an interference between different modes is corroborated by the last results shown in Fig. 3. Fig. 3b indicates an anti-crossing as the longitudinal ex- tensional mode of the lower pedestal approaches the disk breathing mode in frequency. In the vicinity of this anti- crossing a drop in Q is observed23,24. A more complete sinusoidal dependence of Q on h is shown in Fig. 3c and indicates the crucial role of the standing wave formation in the lower pedestal. A high (low) Q value corresponds to the case where the anchoring point to the substrate is located at a node (an antinode) of this wave. In summary, we have reported what is to our knowl- edge the highest Q · f value for a GaAs mechanical res- onator. Numerical simulations show that clamping losses are the dominant source of mechanical dissipation in this generation of resonators standing on a simple pedestal. In order to quench this source of loss, we propose a shielded disk geometry whose first radial breathing mode 4 at ∼ 1.38 GHz is expected to attain a clamping-limited Q as high as 109 corresponding to Q · f = 1018. A fur- ther advantage of the proposed shielded geometry is its simplicity, which makes it applicable to any resonator built from a hetero-structured wafer with some acoustic mismatch. III-V semiconductors naturally lend them- selves to these ideas and could allow to connect mechan- ical modes of ultimately low dissipation with active pho- tonic elements like a single quantum dot or quantum well, opening the exploration of hybrid cavity-QED and op- tomechanics scenarios in semiconductors25. This work is supported by the French ANR through the NOMADE project and by the ERC through the GANOMS project. The authors thank E. Gil-Santos for technical hints. 1F. Marquardt and S. M. Girvin Physics, vol. 2, p. 40, 2009. 2I. Favero and K. Karrai Nat Photonics, vol. 3, pp. 201 -- 205, 2009. 3M. Aspelmeyer, T. J. Kippenberg, and F. Marquardt arXiv e- print 1303.0733, 2013. 4S. Forstner, S. Prams, J. Knittel, E. D. van Ooijen, J. D. Swaim, G. I. Harris, A. Szorkovszky, W. P. Bowen, and H. Rubinsztein- Dunlop Phys. Rev. Lett., vol. 108, p. 120801, 2012. 5H. Miao, K. Srinivasan, and V. Aksyuk New J. Phys., vol. 14, p. 075015, 2012. 6J. D. Teufel, T. Donner, D. Li, J. W. Harlow, M. S. Allman, K. Cicak, A. J. Sirois, J. D. Whittaker, K. W. Lehnert, and R. W. Simmonds Nature, vol. 475, pp. 359 -- 363, 2011. 7J. Chan, T. P. M. Alegre, A. H. Safavi-Naeini, J. T. Hill, A. Krause, S. Grblacher, M. Aspelmeyer, and O. Painter Na- ture, vol. 478, pp. 89 -- 92, 2011. 8L. Ding, C. Baker, P. Senellart, A. Lemaitre, S. Ducci, G. Leo, and I. Favero App. Phys. Lett., vol. 98, no. 11, p. 113108, 2011. 9C. Baker, C. Belacel, A. Andronico, P. Senellart, A. Lemaitre, E. Galopin, S. Ducci, G. Leo, and I. Favero App. Phys. Lett., vol. 99, p. 151117, 2011. 10L. Ding, C. Baker, P. Senellart, A. Lemaitre, S. Ducci, G. Leo, and I. Favero Phys. Rev. Lett., vol. 105, p. 263903, 2010. 11D. Parrain, C. Baker, T. Verdier, P. Senellart, A. Lemaitre, S. Ducci, G. Leo, and I. Favero App. Phys. Lett., vol. 100, p. 242105, June 2012. 12C. Nguyen IEEE Transactions on Ultrasonics, Ferroelectrics and Frequency Control, vol. 54, pp. 251 -- 270, 2007. 13M. Devoret Lectures at the Collge de France, 2012. 14M. Goryachev, D. L. Creedon, E. N. Ivanov, S. Galliou, R. Bourquin, and M. E. Tobar App. Phys. Lett., vol. 100, no. 24, p. 243504, 2012. 15A. G. Smagin Pribory Tekhnika Eksperimenta 6, 143, 1974. 16J. Liu, K. Usami, A. Naesby, T. Bagci, E. S. Polzik, P. Lodahl, and S. Stobbe App. Phys. Lett, vol. 99, pp. 243102 -- 243102 -- 3, 2011. 17I. Mahboob and H. Yamaguchi Nature Nano, vol. 3, pp. 275 -- 279, 2008. 18G. D. Cole, S. Groblacher, K. Gugler, S. Gigan, and M. As- pelmeyer App. Phys. Lett., vol. 92, pp. 261108 -- 261108 -- 3, 2008. 19Comsol 4.3 Library 20S. Adachi Jour. App. Phys., vol. 58, no. 3, pp. R1 -- R29, 1985. 21S. Adachi, Properties of Aluminium Gallium Arsenide. IET, 1993. 22M. Trigo, A. Bruchhausen, A. Fainstein, B. Jusserand, and V. Thierry-Mieg Phys. Rev. Lett., vol. 89, p. 227402, 2002. 23G. Anetsberger, R. Rivire, A. Schliesser, O. Arcizet, and T. J. Kippenberg Nature Photonics, no. 10, pp. 627 -- 633, 2008. 24X. Sun, X. Zhang, and H. X. Tang App. Phys. Lett., vol. 100, no. 17, p. 173116, 2012. 25J. Restrepo, C. Ciuti, and I. Favero arXiv e-print 1307.4282, 2013. Pedestal Radius r (nm)Shield thickness t (nm)2201000200180160140120100120014001600(a)Q-factor012050100-0.50.00.5Q (x103)f-f0(MHz)h (µm)(b)(c)

1711.02982

3

1711

Third-order Optical Nonlinearity in Two-dimensional Transition Metal Dichalcogenides

We present a detailed calculation of the linear and nonlinear optical response of four types of monolayer Two-Dimensional (2D) Transition-Metal Dichalcogenides (TMDCs), having the formula $\textrm{MX}_2$ with M=Mo,W and X=S,Se. The calculations are based on 6-band tight-binding model of TMDCs, and then performing a semiclassical perturbation analysis of response functions. We numerically calculate the linear $\chi_{\mu\nu}^{(1)}(-\omega;\omega)$ and nonlinear surface susceptibility tensors $\chi_{\mu\nu\zeta\eta}^{(3)} (-\omega_\Sigma;\omega_r,\omega_s,\omega_t)$ with $\omega_\Sigma=\omega_r+\omega_s+\omega_t$. Both non-degenerate and degenerate cases are studied for third-harmonic generation and nonlinear refractive index, respectively. Computational results obtained \textit{with no external fitting parameters} are discussed regarding two recent reported experiments on ${\rm MoS}_2$, and thus we can confirm the extraordinarily strong optical nonlinearity of TMDCs. As a possible application, we demonstrate generation of a $\frac{\pi}{4}-$rotated squeezed state by means of nonlinear response of TMDCs, in a silica micro-disk resonator covered with the 2D material. Our proposed method will enable accurate calculations of nonlinear optical response, such as four-wave mixing and high-harmonic generation in 2D materials and their heterostructures, thus enabling study of novel functionalities of 2D photonic integrated circuits.

cond-mat.mes-hall

cond-mat

a Third-order Optical Nonlinearity in Two-dimensional Transition Metal Dichalcogenides School of Electrical Engineering, Sharif University of Technology, Tehran, Iran ´Ecole Polytechnique F´ed´erale de Lausanne, CH-1015, Lausanne, Switzerland∗ Sina Khorasani µν (−ω; ω) and nonlinear surface susceptibility tensors χ(3) We present a detailed calculation of the linear and nonlinear optical response of four types of mono- layer two-dimensional (2D) transition-metal dichalcogenides (TMDCs), having the formula MX2 with M=Mo,W and X=S,Se. The calculations are based on 6-band tight-binding model of TMDCs, and then performing a semi-classical perturbation analysis of response functions. We numerically µνζη(−ωΣ; ωr, ωs, ωt) calculate the linear χ(1) with ωΣ = ωr + ωs + ωt. Both non-degenerate and degenerate cases are studied for third-harmonic generation and nonlinear refractive index, respectively. Computational results obtained with no ex- ternal fitting parameters are discussed regarding two recent reported experiments on MoS2, and thus we can confirm the extraordinarily strong optical nonlinearity of TMDCs. As a possible application, 4 −rotated squeezed state by means of nonlinear response of TMDCs, we demonstrate generation of a π in a silica micro-disk resonator covered with the 2D material. Our proposed method will enable accurate calculations of nonlinear optical response, such as four-wave mixing and high-harmonic generation in 2D materials and their heterostructures, thus enabling study of novel functionalities of 2D photonic integrated circuits. PACS numbers: 42.65.-k; 71.20.Mq; 71.15.-m; 42.50.-p; 42.50.Dv Keywords: 2D materials; Nonlinear optics; Transition metal dichalcogenides; Quantum optics I. INTRODUCTION Following the discovery of the celebrated material, graphene in 2004, the field of two-dimensional (2D) ma- terials has been rapidly expanding over the recent years [1 -- 9]. Despite being only a monolayer thick, 2D materi- als exhibit extraordinary electronic, mechanical, thermal, spintronic, and in particular, optical properties. These offer novel and unprecedented applications which were not foreseen earlier such as new nonlinear capacitors [10] for using in cryogenic parametric amplifiers, circulators, and mixers, as well as a new type of capacitive qubit referred to as cubit [11] in quantum nonlinear supercon- ducting circuits. Of primary importance, is the nonlinear optics of the 2D materials and structures made out of them, which is a matter of current deep investigations. A number of researches discuss the second-order non- linear susceptibility as well as second-harmonic genera- tion in various 2D materials [12 -- 16]. Similarly, the third- order nonlinear optical properties of graphene has been studied in many works [17 -- 25] and the two-dimensional nonlinear sheet susceptibility χ(3)2D or the conductivity σ(3)2D tensor of graphene have been calculated. How- ever, recent experiments [26] still disagree with the ex- pected numerical estimates from theory within an order of magnitude. The nonlinearity of graphene is remark- ably large, however, it needs to be gated to adjust its Fermi level to the resonance conditions. That implies ∗ [email protected]; Present address: Vienna Center for Quantum Science and Technology, University of Vienna, 1090 Vienna, Austria for ungated structures, there is still a tendency to ex- amine other 2D materials as well. Similar discrepancies between measured sheet optical nonlinearity of alterna- tive 2D materials and theoretical estimates is noticed by other researchers as well. In this work, we present a detailed study of the non- linear optical properties of 2D transition metal dichalco- genides (TMDCs) [27], with the general formula MX2 with M being a transition metal, here being either Molyb- denum Mo or Tungsten W, and X being a chalcogen such as Sulphur S or Selenium Se. Based on a six-band Tight- Binding Hamiltonian, we obtain the nonlinear sheet sus- ceptibility χ(3)2D of the TMDCs through semi-classical perturbation approach. The semi-classical method retains its validity for low illumination intensities, where Zener tunneling and semimetal transitions could be well ignored and dismissed from the carrier dynamics [28 -- 30], leaving only multi- photon processes as important. This approximation also is limited by the finite span of optical wavelength and non-zero extent of atomic bonds, that is, the medium has to be effectively treated as continuous and its micro- scopic granular structure shall be ignored. For ultrashort wavelength optical excitations beyond the deep ultravio- let spectrum, this treatment is no longer valid. Such nonlinear interactions [31] are of primary impor- tance in study of valley-spin dynamics [32] and Kerr spec- troscopy [33] of TMDCs. In all these last works, the TMDC monolayer has been sandwiched in protecting 2D Boron Nitride shields, which has greatly contributed to the visibility and sharpness of emission spectra [31 -- 34]. The reported data on χ(3)2D of MX2 are very scat- tered up to four orders of magnitude, as it is shown to be strongly dependent on the growth method and the post-treatment process. This makes a conclusive evalua- tion out of reach at present. However, we notice agree- ment roughly within an order of magnitude between the computed numerical figure and the stronger one of the experimental values. We study the nonlinear optical properties of an ultralow-loss amorphous silica microdisk covered by WSe2, and define an effective nonlinear susceptibility χ(3) eff for the structure. We show through detailed COMSOL calculations that the effective nonlinearity marked by χ(3) eff is improved up to two orders of magnitude by placing the TMDC on top of the silica microdisk. This platform has been shown to be useful for experimental study of optical emission and excitonic properties of TMDCs [35]. As an application example, we consider the squeez- ing property of light [36] through four-wave mixing un- der unpumped and pumped configurations and discuss these scenarios. While calculations reveal the dominance of two-photon absorption over nonlinear Kerr self-phase modulation in 2D TMDCs, we show through detailed the- ory that an unorthodox π is produced with an elongated elliptical Wigner distribu- tion and theoretically unlimited squeezing. Without con- sideration of this π to be desqueezed. However, unlimited squeezing is pos- sible only along one π thogonal quadrature will always be strongly desqueezed. 4−rotation, both quadratures appear 4−rotated quadrature and the or- 4−rotated squeezed state of light 2 cient for our purpose. Moreover, the 6-band TB method, considers the effects of valley-dependent spin-orbit inter- action of 2D TMDCs [53], or better known as trigonal warping [41, 42]. This method employs an orthonormal basis, and therefore does need inversion of the overlap matrix [37, 39] which eases out coding and efficiency at the same time. Since the details of this scheme is rather comprehensive, the reader is referred to the existing lit- erature for further information [47 -- 50]. However, in Ap- pendix A we present the necessary ingredients briefly. The orthonormal bases here are  d3z2−r2 dx2−y2 dxy p+ x p+ y p− z  , Ψ(cid:105) = (1) l = (pt l + ηpb √ l )/ where pη 2 with t and b referring to the top and bottom chalcogen atoms and η = ±. For the sake of convenience, we rewrite the basis as Ψ(cid:105) = {ψn(cid:105)}, n = 1 . . . 6. Obviously, the basis kets ψn(cid:105) are dependent on the 2D k−vector and spin σ, correspond- ing to the eigenvalues Eσ n (k) which form the energy band structure. Hence, we have the relationship m (k)(cid:105) = δnmδσσ(cid:48). n(k)ψσ(cid:48) (cid:104)ψσ (2) II. THEORY & RESULTS A. Band Structure In order to compute the linear and nonlinear suscep- tibility from first principles, we would need to have ac- curate knowledge of the electronic transitions, valleys, and spin-orbit interactions. The correct way to tackle this problem is to have an efficient code to derive the electronic band structure of the material, and since this information is going to be called upon quite frequently inside integration and summation loops, the computation has to be both accurate and very efficient. For this pur- pose, tight-binding (TB) scheme is very appealing since with correct implementation it could meet both criteria. The method TB is quite popular for low-dimensional carbon structures such as graphene and carbon nan- otubes [37, 38], and six-band TB has been used for study- ing the band structure of hydrogenated graphane [39, 40]. In recent years, two-band [41 -- 43] expansion based on k · p perturbation and Lowdin partitioning of Hamilto- nian [44], three-band [45], four-band [46], six-band [47 -- 50], seven-band [42], and eventually eleven-band [51] TB models have been developed to calculate the band struc- ture of TMDCs. However, six-band TB based [50] on Slater-Koster expressions [52] is ultimately chosen since it is a fairly accurate low-energy approximation to the ex- tensive density functional theory (DFT) calculations, and thus expected to be both sufficiently accurate and effi- In general for honeycomb lattice with C3v symmetry such as graphene [38] and in absence of chirality, it is nor- mally sufficient to compute the band structure over the irreducible Brillouin zone, which is only 1/12 of the first Brillouin zone. However, monolayer TMDCs are non- centrosymmetric and satisfy a different spatial group de- noted as D1 3h [54], which causes spin-valley coupling. As a result, the irreducible zone is only 1/6 of the first Bril- louin zone. The calculated band structures are shown in Fig. 1 for the four basic types of TMDCs MX2 with M=Mo,W and X=S,Se. Here, red and blue curves correspond to respec- tively down and up spin polarizations. Interestingly, the calculations preserve valley-dependent spin-orbit interac- tion or the trigonal warping of TMDCs, which is strongly present in the valence bands at K and K(cid:48). To illustrate this, we calculate the entire first Brillouin zone of WSe2, in which spin-orbit interaction is highly different at K and K(cid:48). As a result of this symmetry breaking, the band structure shows a C3v point-group symmetry, instead of the expected C6v, and by changing the direction of spin the band structure rotates by π/3, thereby interchanging the roles of K and K(cid:48). The conduction and valence bands of WSe2 are illustrated in Fig. 2. B. Linear Susceptibility The concept of an interface occupied by a dielectric having finite surface electric and magnetic susceptibili- ties has been originally investigated in a pair of papers 3 FIG. 2. Illustration of valley-spin coupling in WSe2 as an example, based on the 6-band TB model [50]. The effect of trigonal warping is seen to be strongly pronounced for the valence band, where K and K(cid:48) behave very differently. with the dimension of meter, or equivalently, surface con- ductivity σs = j
0ωχ2D with the dimension of Ω−1 should be known. Here, we are able to compute the surface susceptibil- ity tensor elements χ2D(ω) = χ(1)2D(−ω; ω) of TMDCs using the expression χ(1)2D µν (−ω; ω) = 1 × (3) (cid:88) mnσ (cid:90)(cid:90)
0ω2AUC 1 ABZ BZ n (k) − f σ f σ m(k) mn(k) − ω Eσ m(k) − Eσ jµσ mn(k)jνσ nm(k)d2k, n (k), f σ mn(k) = Eσ m(k) = {1 + where Eσ m(k)−EF/kBT ]}−1 is the Fermi-Dirac distribution exp[Eσ with EF being the Fermi energy (here equal to zero for the intrinsic semiconductor), and kBT being the thermal energy. The matrix elements jµσ mn(k) are related to the current operator jσ µ (k) as mn(k) = (cid:104)ψσ jµσ µ (k) = − q jσ  µ (k)ψσ n(k)(cid:105) , m(k)jσ ∂H(k; σ) , ∂kµ (4) FIG. 1. Band structures of four basic TMDCs, based on the 6-band TB model [50]. Red and blue lines correspond to the spin down and up bands. published in 1994 [55, 56]. Independently and still a few years before celebrated discovery of the first 2D material, graphene, the optical properties, possible modulation and switching applications of such media, referred to as the conducting interfaces were explored by a series of papers by the author [57 -- 62]. The transfer matrix method has been reformulated so far independently by many authors to tackle the problem of optical wave refraction from lay- ered structures containing 2D materials [56, 58, 63 -- 65] and even more recently by Morano [66, 67] for measure- ment of the surface conductivity of graphene. Based on the theory of conducting interfaces, an ul- trathin 2D material could be treated by a discontinuity in tangential electromagnetic fields across the interface. However, the corresponding surface susceptibility or χ2D ΓKMΓ-12-8-6-4-202Energy(eV)MoS2ΓKMΓ-12-8-6-4-202Energy(eV)MoSe2ΓKMΓ-50Energy(eV)WS2ΓKMΓ-10-50Energy(eV)WSe2-3-2-10123-3-2-10123ConductionBand-3-2-10123-3-2-10123ValenceBandin which kµ are elements of the vector k. Moreover, q is the electronic charge and H(k; σ) is the spin-dependent Hamiltonian in k−space, which in our TB formalism ap- pears as a 6 × 6 matrix. More on the details of this operators could be found in the literature [68]. In (3),
0 is the permittivity of vacuum, AUC is the area of unit cell, and ABZ = (2π)2/AUC is the area of first Brillouin zone [69 -- 71]. That implies the expression NAAUC is the area occupied by one mole of the 2D material, where NA is Avogadro's constant [69 -- 75]. One can easily verify 3a2/2 where a is the 2D crystal's lattice that AUC = constant, being equal to the distance between the two nearest metal atoms. √ In the third-order nonlinear processes, the location of Fermi-energy has very little observable effect on the Kerr coefficient, and much less on the third-harmonic genera- tion coefficient. This is a fact observed through extensive numerical experiments. Only if the Fermi level is deeply into the conduction or valence bands, that is, the semi- conductor is made degenerate, the result would become different. Normally, since chemical doping is not in prac- tice for TMDCs, any shift in the Fermi level would be triggered by electrostatic gating. This method of charge depletion or accumulation, however, is typically unprac- tical for making a 2D TMDC degenerate. The author believes that there is no practical reason to be concerned about the effect of extrinsic Fermi level as opposed to the intrinsic case with EF = 0. µν When the TMDC is not under strain, then linear (−ω; ω) = χ(1)2D(ω)δµν is a simple susceptibility χ(1)2D scalar and not a tensor quantity. This fact together with the spin-valley coupling of carriers can be used to sim- plify (3) a bit as χ(1)2D(−ω; ω) = 3 (cid:88) mnστ 1 ABZ (cid:90)(cid:90)
0ω2AUC IRBZ × n (k) − f στ f στ m (k) mn(k) − ω Eστ (5) jxστ mn (k)2d2k, Here, the integration on the reciprocal plane is taken only over the irreducible Brillouin zone, and summation on valley index τ = ± is for selection between K and K(cid:48) valleys. Since valley contributions from opposite spins are simply equal, one may take advantage of symmetry considerations and drop the summation on τ , just mul- tiplying the whole expression on the right by a factor of 2, thus arriving at χ(1)2D(−ω; ω) = × (6) 6
0ω2AUC n (k) − f σ f σ m(k) mn(k) − ω Eσ IRBZ 1 ABZ jxσ mn(k)2d2k. (cid:90)(cid:90) (cid:88) mnσ Results of computations for the linear surface suscepti- bilities of the four basic TMDCs in the wavelength range 760-790nm is illustrated in Fig. 3. Although the absolute values are here shown, imaginary parts of χ(1)2D(−ω; ω) in the desired range is effectively zero. 4 FIG. 3. Typical values of first-order linear susceptibility χ(1)(−ω; ω) for different TMDCs. Contribution to the ab- solute value is entirely from the real part. C. Nonlinear Susceptibility Folllowing the standard perturbation scheme [76 -- 88], we have developed a code in Mathematica to compute the non-degenerate four-wave mixing (FWM) third-order mnpq(−ωΣ; ωr, ωs, ωt) where nonlinear susceptibility χ(3) ωΣ = ωr + ωs + ωt for any of the widely used TMDCs. When studying 3rd harmonic generation, mnpq(−3ω; ω, ω, ω) must be calculated, while for degen- χ(3) mnpq(−ω; ω,−ω, ω) should erate nonlinear propagation, χ(3) be found. Expressions are given in the Appendix B. It is worth mentioning that the method of equations of mo- tions [68, 89, 90] could also be used to tackle the dynamics of nonlinear optics and four-wave mixing in semiconduct- ing materials, however, the formalism does not directly give in an expression for the nonlinear susceptibility. Figure 4 represents typical calculated values from semi- classical perturbation theory of nonlinear interactions. Similarly, Fig. 6 presents the typical nonlinear coefficient in the telecommunication window around the wavelength 1550nm. The uniform convergence of the code has been demonstrated in Fig. 7 versus computational grid. These set of figures present absolute value of the tensor com- yyyy(−3ω; ω, ω, ω) corresponding to the third- ponent χ(3) harmonic generation. Calculation of other non-zero ten- sor components has also been done, but nor presented for the sake of brevity. Further discussions on the relation- ship among non-zero tensor components is given in the following. Resonances in χ(3) mnpq(−3ω; ω, ω, ω) as shown in Fig. 2, correlate well with the resonances in linear surface susceptibility χ(1)2D(− 1 n ω) with n = 1, 2, 3, as well as the possible such direct transitions across the band structure. For the case of degenerate nonlinear coeffi- n ω; 1 MoS2MoSe2765770775780785790λ(nm)2.65×10-112.7×10-112.75×10-112.8×10-112.85×10-112.9×10-112.95×10-11χyy(1)[m]SurfaceLinearSuceptibilityWSe2WS2765770775780785790λ(nm)3.05×10-113.1×10-113.15×10-113.2×10-113.25×10-11χyy(1)[m]SurfaceLinearSuceptibility5 FIG. 7. Typical convergence of the code at the wavelength of 1560nm versus resolution of the computational grid, corre- sponding to Fig. 6. in the wavelength range of interest, that is, it is the two- photon absorption which is the dominant Kerr nonlinear effect. Furthermore, this coefficient turns out to be a scalar like χ(3)(−ω; ω,−ω, ω), as is discussed later be- low. Typical values for different types of TMDCs are illustrated in Fig. 5. Comparing to Fig. 4, much less oscillations are seen, and that is because of the much less number of available resonances with transitions among band edges. For the case of MoS2, in particular, many resonances occur as shown in Fig. 5, which can be at- tributed to the bandgap of this material being the small- est of the four studied TMDCs. As a result, for a given photon energy above the energy gap, much more possi- ble triple transitions required for a third-order nonlinear process become available. Evidently, selection of a pho- ton energy sufficiently small, that is the case in Fig. 6, no such behavior is seen, simply because of the absence in availability of corresponding possible transitions. Furthermore, the magnitude of Kerr nonlinear index is significantly larger than the third-harmonic generation. It is thus to be noticed well that these two parameters of Kerr nonlinearity and third-harmonic generation, al- though related, but are essentially much different in both their physics and behavior. Actually, these two values could be quite far apart in phase and many orders in magnitude. We here point out again that the Kerr nonlinear sus- ceptibility χ(3)(−ω; ω,−ω, ω) and Third-harmonic gener- ation susceptibility χ(3)(−3ω; ω, ω, ω) are not the same. While both are having the third-order, these correspond to entirely different processes. This difference can be un- derstood as follows. The Kerr nonlinear susceptibility only describes a pro- cess in which the local permittivity of the medium is linearly dependent on the local intensity of light, such as
(E) = n2
0 + ¯E2χ(3), where ¯E is the phasor of the electromagnetic field. Hence, this effect can for in- stance cause local increase or decrease of the effective refractive index depending on the sign of (cid:60)[χ(3)], which correspondingly lead to self-focusing or self-defocusing phenomena. While higher harmonics inevitably may weakly arise under such conditions, because of the in- herent nonlinearity in the propagation, it is only the first harmonic which undergoes self-field propagation is- FIG. 4. Typical values of third-order nonlinear susceptibility χ(3)(−3ω; ω, ω, ω) for different TMDCs. Contribution to the absolute value is entirely from the real part. FIG. 5. Typical values of Kerr nonlinear susceptibility χ(3)(−ω; ω,−ω, ω) for different TMDCs. Contribution to the absolute value is entirely from the imaginary part. mnpq(−ω; ω,−ω, ω), resonances are possible only cient χ(3) at n = 1, 2, and were found to be much less pronounced. The code is able to compute individual tensor compo- nents as arbitrated such as χ(3)2D . ijkl Computations on the Kerr nonlinear susceptibility mnpq(−ω; ω,−ω, ω) reveals that it is entirely imaginary χ(3) FIG. 6. The calculated third-order nonlinear susceptibility χ(3)(−3ω; ω, ω, ω) for MoS2 in the infrared telecommunication window. Contribution to the absolute value is entirely from the real part. MoS2MoSe2765770775780785790λ(nm)02.×10-264.×10-266.×10-268.×10-261.×10-25χyyyy(3)[m3V-2]SurfaceNonlinearSuceptibilityWSe2WS2765770775780785790λ(nm)01.×10-262.×10-263.×10-264.×10-26χyyyy(3)[m3V-2]SurfaceNonlinearSuceptibilityWSe2WS2MoS2MoSe2765770775780785790λ(nm)05.×10-241.×10-231.5×10-232.×10-23χyyyy(3)[m3V-2]SurfaceKerrSuceptibility150015201540156015801600λ(nm)1.45×10-251.5×10-251.55×10-251.6×10-251.65×10-251.7×10-25χyyyy(3)[m3V-2]MoS2051015202530Grid1.4×10-251.45×10-251.5×10-251.55×10-25χyyyy(3)atλ=1560nmConvergencePlotsues. More accurately, ¯E(3)(r)eiωt = [¯E(1)(r)eiωt : χ(3) : ¯E(1)∗(r)e−iωt]¯E(1)(r)eiωt. However, the third-order nonlinear susceptibility di- rectly is connected to the local amplitude of the har- monic, as a result of locally present electromagnetic field. Hence, one could imagine that the third-harmonic be re- lated to the first harmonic as ¯E(3)(r)ei3ωt = [¯E(1)(r)eiωt : χ(3) : ¯E(1)(r)eiωt]¯E(1)(r)eiωt. Since the physical phenomena behind these two mech- anism are not identical, one would naturally expect that χ(3)(−ω; ω,−ω, ω) (cid:54)= χ(3)(−3ω; ω, ω, ω). Sometimes, this fact is not sufficiently made clear in the literature. 1. Critical Field c = χ(1)/χ(3) Defining a critical electric field as E2 xxxx as a measure of required lights electrical field to reach the onset of third-order nonlinearity, we observe that for the TMDCs this value falls within the typical range of 7-8kV/m, regardless of the choice of the par- For all the studied cases in the ticular material. wavelength range of 750-790nm, the approximations mnpq](−3ω; ω, ω, ω) = 0, as well as mn] = 0, (cid:61)[χ(3) (cid:61)[χ(1) mnpq](−ω; ω,−ω, ω) = 0 apply well to the linear and (cid:60)[χ(3) nonlinear surface polarizabilities. We use the conducting interface formulation [57, 58] to carry out any electro- magnetic analysis of monolayer 2D materials. 2. Analogy with Bulk Values In this wavelength range of interest and for the linear response, the typical strength of linear susceptibility χ(1) is within the range of 3.5 − 4.4 × 10−8m, so that having an effective monolayer thickness of t2D we may assign an effective refractive index of n = (cid:112)1 + [χ(1)/t2D] to the Referring to Fig. ultrathin TMDC layer. Since, t2D is only of the order of typically 7 − 10A, then effective refractive index is expected to be about only n ≈ 6 − 8 in the wavelength range presented here. However, we notice that at longer infrared wavelengths, this value dramatically increases to higher values. 5, TMDCs have an extremely large surface nonlinearity χ(3)2D in the range of 5 × 10−24m3V−2 to 2 × 10−23m3V−2 [76, 91], which when divided by the typical thickness of a monolayer of the order of 0.5nm, give rise to an effective nonlinearity of the order of 10−14m2V−2 to 10−13m2V−2. Given that this value is resulting from pure electronic polarization, this is quite remarkably large when put into perspective of expected values of the order of 10−22m2V−2 [77]. This places the expected nonlinearity of 2D TMDCs, to many orders of magnitude stronger the range of III- V semiconductors such as GaAs 1.4 × 10−22m2V−2, and than that of Diamond 2.5 × 10−21m2V−2, fused silica 2.5 × 10−22m2V−2, and even GaP at 577nm 2.93 × 6 10−18m2V−2 [80] which is known to have an extremely large nonlinear index of refraction. We first define a bulk-equivalent susceptibility as χ(3) eq = χ(3)2D/t2D where t2D is the physical thick- ness of the monolayer. For instance then MoS2 at 577nm and 1560nm respectively exhibits an expected bulk-equivalent, absolute nonlinear Kerr susceptibility eq (−ω; ω;−ω; ω) of 2.11 × 10−13m2V−2 and 1.17 × χ(3) 10−12m2V−2. In a similar manner, the corresponding ex- pected bulk-equivalent, absolute third-harmonic suscep- eq (−3ω; ω; ω; ω) are 2.96 × 10−17m2V−2 and tibilities χ(3) 2.11 × 10−15m2V−2 at 577nm and 1560nm respectively. These numbers could reasonably well explain the re- cently observed ultrastrong high harmonic generation in 2D TMDCs [76, 91, 92]. There are certain classes of ma- terials or media, which could offer significantly stronger nonlinearity, however, either their slow response times (such as polymers) or complexity of formation (such as cold atomic gases), render them of limited use at optical frequencies. The unusually large nonlinearity of 2D materials com- pared to bulk 3D structures is hard to explain. However, we believe that it is a matter of geometrical confinement dimensions that sets this strength. At least it has been rigorously established that the third order nonlinear op- tical response of spin density wave insulators is much stronger in 2D than 3D and 1D structures [93]. 3. Analogy with Experimental Values The validity of numerical results could be furthermore verified against three recent experimental data [76, 91, 94] on 3rd harmonic generation from MoS2 at the wave- length of 1560nm. While our developed code estimates mnpq(−3ω; ω, ω, ω), as a value of 1.6 × 10−25m3V−2 for χ(3) shown in Fig. 7, the reported experimental data are 3.9× 10−15m2V−2×0.75nm = 2.93×10−24m3V−2 [76], as well as the very different values of 1.7 × 10−28m3V−2 [91], as well as 1.2× 10−19m2V−2 × 0.65nm = 7.8× 10−29m3V−2 as the pre-treatment and 1.95×10−28m3V−2 as the post- treatment values [94], and also 6.5 × 10−29m3V−2 [95]. This shows that while experimental results [76, 91, 94] vary within four orders of magnitude, our numerical esti- mate is much closer to the first measurement [76] report- ing the larger value. The difference between theoretical results with experimental ones, therefore, should not be surprising considering the remarkably scattered numeri- cal values among experimental observations. While the nature of such differences is not exactly known yet, it could be due to material growth and transfer issues, de- fect concentrations, substrate effects, as well as the opti- cal method of measurement. Given the large sensitivity of TMDCs to the fabrication process and even resilience under ambient conditions, there remains a question of how to unify experiments alike. It could be speculated that for a variety of reasons, the experimental reports ac- tually either underestimate or overestimate the nonlinear susceptibility over the true theoretical value. • Squeezed light in atomic ensembles, and • Squeezed light in semiconductor lasers. 7 4. Behavior of Tensor Components With regard to the individual tensor components rec- ognized by the set of indices µνζη, there should be 24 = 16 elements. However, half of them are zero and the only non-zero tensor elements are yyyy, xxxx, xyxy, yxyx, xyyx, yxxy, xxyy, and yyxx. Even though, all ten- sor elements are not quite independent and the following identities hold for all four TMDCs due to the crystal symmetries yyxx + χ(3) yxxy + χ(3) yxyx, χ(3) yyyy= χ(3) χ(3) yyxx= χ(3) χ(3) yxxy= χ(3) χ(3) yxyx= χ(3) xxxx = χ(3) xxyy, xyyx, xyxy. (7) This limits the maximum number of independent tensor elements to three. For the case of Kerr nonlinearity where the param- µνζη(−ω; ω,−ω, ω) is concerned, one may verify eter χ(3) that the elements yyxx and yxyx also identically van- yxxy = χ(3) ish. Hence, we have χ(3) xyyx and get a fairly simple scalar form for the nonlinear Kerr susceptibility as xxxx = χ(3) yyyy = χ(3) µνζη(−ω; ω,−ω, ω) = χ(3)(ω)δµηδνζ. χ(3) (8) Similarly, we could write for the third-harmonic gen- eration the following µνζη(−3ω; ω, ω, ω) = χ(3) χ(3) 2 (ω)δµηδνζ + χ(3) +χ(3) 3 (ω)δµζδνη, 1 (ω)δµνδζη (9) where χ(3) l (ω) with l = 1, 2, 3 are independent scalars. Out of the above four methods, the first three are mostly implemented in a non-monolithic experiments and normally require large optical setups. The second one is based on the χ(3) effect of fused silica in optical fibers, and thus requires long propagation paths over fibers to achieve squeezing, but squeezing as large as 9dB has been achieved using this scheme. The third one requires so- phisticated techniques of atom vapor trapping and con- densation at ultralow temperatures, and the largest ob- served squeezing of 13dB has been so far achieved this way. The fourth method [103] is relatively easy to achieve and based on monolithic integrated photonics, but is lim- ited for various practical reasons to only 3-4dB of reduc- tion in shot noise and thus squeezing. Recent advances in optomechanics, has brought the possibility of optomechanical squeezing of light into per- spective as well [104 -- 109]. Optomechanical squeezing of light in homodyne detection within a small amount also occurs, and can be observed using a novel quan- tum feedback control scheme which has been recently reported [110]. Monolithic approaches to squeezing by optical parametric oscillators [111] have been shown to be feasible as well. As an alternative route, the possibility of using Si3N4 micro-ring resonators for squeezing has been demon- strated in a multi-mode optical parametric oscillator [112] with 1.7dB squeezing. This has also apparently been verified experimentally at room temperature by pumping a continuous laser [113], and 0.5dB squeezing below shot noise was observed in a self-homodyne setup. So, it could be safely claimed that using silica microdisks covered with 2D TMDCs, certainly measurable squeez- ing could be obtained. It has been furthermore recently shown that high-quality silica micro-ring resonators could provide a versatile platform for study of emission prop- erties of 2D materials [35]. III. NON-CLASSICAL STATE OF LIGHT B. χ(3) Squeezing on Micro-Resonators In this section, we discuss production of a non-classical state of light with elliptical Wigner distribution, due to two-photon absorption. Since the method of analysis is based on the theory of squeezing, we need to initially present an overview of light squeezing schemes. It is known that the 3rd order nonlinearity could be employed to generate squeeze light. Depending on the optical setup, that whether the squeezed state is pro- duced through unitary transformation in vacuum, or an interferometric setup, it leads to either of the Hamiltoni- ans [114, 115] as A. Methods of Squeezing Squeezing of light is usually done through either of the following general methods [96 -- 102]: H = ω H = ω (cid:19) (cid:19) (cid:18) (cid:18) a†a + a†a + 1 2 1 2 + ω(cid:0)ξa†2 − ξ∗a2(cid:1) , + ωξa†2a2. (10) (11) • Squeezed light by parametric down conversion, • Squeezed light in optical fibers, The first squeezed state (10) being referred to as the quadrature squeezing, has an elliptical Wigner distribu- tion, while the second one (11) being referred to as the χ(1) are scalar quantities. This greatly simplifies (15,16) as (cid:40)(cid:90)(cid:90)(cid:90) ξ = + 1 = + 3ωF 8
0 (cid:90)(cid:90) (cid:90)(cid:90)(cid:90) (cid:90)(cid:90) 2D 2D χ(3)(r)u(r)4d3r Disk , (cid:41) (cid:12)(cid:12)(cid:12)z=0 (cid:105)u(r)2d3r (cid:12)(cid:12)(cid:12)z=0 . (cid:104) χ(3)2D(r)u(r)4d2r 1 + χ(1)(r) Disk χ(1)2D(r)u(r)2d2r 8 (17) (18) (19) photon-number squeezing results in a kidney- or crescent- shaped squeezing and is thus quite different and non- Hermitian when (cid:61)[ξ] (cid:54)= 0. Here, ω is the frequency of light and ξ is a dimensionless parameter defined as [115] (cid:90)(cid:90)(cid:90) (cid:88) mnpq ξ = 3ωF 8
0 mnpq(r)um(r)un(r)u∗ χ(3) p(r)u∗ q(r)d3r, (12) where u(r) is the vector mode profile of a single photon in the cavity with components um(r), and a factor 3 2 is already included for standing wave, as it is going to be the case under study. Furthermore, since the optical wave is strongly confined in the micro-disk and undergoes a much larger effective propagation path, an extra dimensionless finesse factor F is also included. The finesse F is a direct measure of how many times the light pulse circulates the ring [116, 117]. This is while the single photon mode u(r) having the physical dimension m−3/2 must be normalized as (cid:90)(cid:90)(cid:90) (cid:90)(cid:90)(cid:90) (cid:104) (cid:88) (cid:88) mn mn 1 = =
r,mn(r)um(r)u∗ n(r)d3r (cid:105) δmn + χ(1) mn(r) um(r)u∗ n(r)d3r. Here, χ(1)(r) =
r(r) − 1 is the relative susceptibility of the micro-disk resonator, including the cladding and substrate. Hence, (13) is effectively taken on the mode volume of the cavity. Obviously, (12) could be written for an unnormalized mode as 3ωF 8
0 mnpq ξ= (cid:80) (cid:16)(cid:80) × (cid:82)(cid:82)(cid:82) χ(3) (cid:82)(cid:82)(cid:82)(cid:104) mn mnpq(r)um(r)un(r)u∗ δmn + χ(1) mn(r) um(r)u∗ p(r)u∗ n(r)d3r q(r)d3r (cid:17)2 . (cid:105) In presence of a 2D material, and using a conducting interface approximation [57], these two latter expressions should be changed slightly as follows × (cid:88) mnpq 3ωF 8
0 (cid:40)(cid:90)(cid:90)(cid:90) (cid:90)(cid:90) (cid:88) (cid:88) mn 2D ξ = + 1 = + mn 2D p(r)u∗ q(r)d3r, (cid:41) , (cid:12)(cid:12)(cid:12)z=0 q(r)d2r Disk mnpq(r)um(r)un(r)u∗ χ(3) (cid:104) mnpq(r)um(r)un(r)u∗ p(r)u∗ (cid:90)(cid:90)(cid:90) (cid:105) χ(3)2D (cid:90)(cid:90) δmn + χ(1) mn(r) Disk mn (r)um(r)u∗ χ(1)2D n(r)d2r (cid:12)(cid:12)(cid:12)z=0 . um(r)u∗ n(r)d3r, (16) (15) where it is supposed that the 2D TMDC is placed on the z = 0 plane. Following (8) for Kerr nonlinearity and two- photon absorption in unstrained TMDCs, both χ(3) and 1. Effective Nonlinearity & Mode Volume Alternatively, (17) can be written as (13) ξ = 3ωF 8
0V χ(3) eff , where V is mode volume [118] and χ(3) nonlinear index defined as eff is the effective (cid:21)−1 u(r)4d3r , (20) V = χ(3) eff = V (cid:20)(cid:90)(cid:90)(cid:90) (cid:40)(cid:90)(cid:90)(cid:90) (cid:90)(cid:90) Disk 2D (14) + χ(3)2D(r)u(r)4d2r χ(3)(r)u(r)4d3r Disk (cid:41) . (cid:12)(cid:12)(cid:12)z=0 It should be mentioned here that ξ and therefore χ(3) eff by definition cannot be independent of the micro-disk radius r as well as wavelength λ. Another way to look into this is to view the circulating optical power in disk resonator as an optical field going through a long straight path [119, 120], thus experiencing an overall phase retardation. This is the basis of light squeezing in optical fibers [121 -- 124], and this point of view gives a more clear and straightforward measure for the squeezing parameter s as (cid:90)(cid:90)(cid:90) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) Disk 6πr
0F nλ(Nω) (cid:90)(cid:90) s = + χ(3)2D(r)E(r)4d2r 2D (cid:12)(cid:12)(cid:12)z=0 (cid:12)(cid:12)(cid:12)(cid:12)(cid:12), χ(3)(r)E(r)4d3r (21) where r and n are disk's radius and index of refraction, respectively, λ is the optical wavelength, and E(r) is the electric field inside the disk. Moreover, N is the number of photons inside the disk, and thus Nω is the total optical energy confined in the cavity. Now, plugging in (17) and further simplification gives the final expression as (cid:90)(cid:90)(cid:90) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)ξ Disk 8
2 0rV nc(Nω) (cid:90)(cid:90) s = + ¯χ(3)2D(r)E(r)4d2r 2D (cid:12)(cid:12)(cid:12)z=0 (cid:12)(cid:12)(cid:12)(cid:12)(cid:12). ¯χ(3)(r)E(r)4d3r (22) eff and ¯χ(3) = ¯χ(3)/χ(3) Here, ¯χ(3)2D = ¯χ(3)2D/χ(3) eff are It should be noted that in dimensionless quantities. the above relations, Nω is in denominator, but since E2 ∝ Nω, the overall squeeze factor is proportional to the optical energy inside the cavity. More specifically, we have E =(cid:112)Nω/2
0u [97], which gives the simple form s = ξN. 4πr λ (23) 2. Noise Squeezing/Desqueezing As it was discussed in the above, χ(3) is purely imagi- nary in the wavelength range of interest between 760nm to 790nm, and thus it is the two-photon absorption which wins over the Kerr nonlinear index. Despite this fact, this is unimportant for our particular application of produc- ing a non-classical state of light with non-circular Wigner distribution. Some studies have only considered the absolute value of χ(3) to the squeeze parameter s = ζ [121], where ζ is the complex squeeze parameter given by the squeeze operator as [97, 121, 125 -- 127] (cid:18) 1 (cid:19) S(ζ) = exp ζ∗a2 − 1 2 ζa†2 2 . (24) Ignoring the real part of ζ and noting that its imaginary part is positive, we may rewrite the above as (cid:20)−is 2 (cid:0)a2 + a†2(cid:1)(cid:21) S(s) = exp . (25) This is actually directly corresponding to the interaction Hamiltonian (10) in the above, noting that s ∝ ξ. As it will be discussed later below, the fact that χ(3) is imagi- nary does not disallow production of non-classical states. It should be mentioned that actual ratio of noise squeezing for the case of (10), here being denoted respec- tively by
and ϕ for the two orthogonal quadratures is not the same as s, since ζ is not real valued. As we can actually show in Appendix C, when (cid:60)[ζ] = 0 the correct relationship is
= cosh(s) 1 + e−2s sinh2(s), ϕ = sinh2(2s) 4e2s + sech2(s) 1 + (cid:20) (26) (cid:21)2 . sinh2(2s) 4e2s (cid:113) (cid:115) 9 √ These expressions behave as
≈ 2ϕ ≈ ( 5/4)es in the limit of large s, implying asymmetric noise increase at high intensities, while approaches unity for small s im- plying no non-classical squeezing effect at low intensities. As it has been demonstrated in the Appendix C, the measure of non-classicality of the resulting Wigner dis- tribution is limited to 6.02dB at high intensities or high χ(3) eff . C. Experimental Issues One therefore may produce a non-classical light using a micro-disk resonator covered with a 2D TMDC, such as WSe2, or MoS2. We would like to investigate this possibility by taking advantage of the very large χ(3) co- efficient of 2D TMDCs. Combined with the very strong confinement of light in low loss silica disk resonators, we would expect a significant anharmonicity ξ as defined in (15). This has yet to be calculated. If ξ is hopefully found to be reasonably large, and leading to an observ- able value of desqueezing asymmetry in light, we would move ahead with experiments to characterize the prop- erties of output emission and shot noise. The enhanced effective nonlinear susceptibility χ(3) eff should in principle allow production of non-classical elongated state at much lower intensities. 1. Ultrashort Solitons A possible advantage of this scheme in case of suc- cessful design and experiments, would be relatively ease of fabrication and operation, as well as compatibility with monolithic integrated photonics. Usage of ultra- short solitons [128, 129] together with strong confinement in micro-disks increases the overall nonlinear interactions and thereby squeezing as well. This has been already demonstrated in squeezing via optical fibers [96]. In that case, no extra pumping is needed, and the pulse under- goes squeezing by itself through self-pumping. Evidently, the squeezed output is pulsed. Since the quality factors of cavities is not infinite, a small dissipation at high optical power densities is unavoidable [118, 130 -- 133], which can be treated anyhow as either a Lugiato-Lefever equation [134] or using a perturbative expansion [135]. 2. Separate Pumping It is in practice beneficial if the optical power re- quired for excitation of nonlinearity is maintained by a pump held at a slightly different frequency such as ωp = ω ± FSR, where FSR is the free spectral range of the micro-disk resonator. The susceptibility then should be calculated from the non-resonant near-degenerate pumped value χ(3)(−ω; ωp,−ωp, ω). Since FSR << ω, 10 basic types of TMDCs discussed in this paper support both of the dark and bright excitons, which is further complicated by the trignoal warping property of these materials and presence of charged excitons (also known as trions) and biexcitons [35, 138, 139]. In general, exciton binding energies in 2D materials cannot be explained by a simple hydrogenic model, be- cause of the very different radial distribution of the wave- function, which could easily extend to a few nm in radius. It only can be investigated by DFT GW-BSE calcula- tions, which exhibits numerous lines between 1s and 2p states. While the exciton binding energies in 2D mate- rials can be large, these are directly dependent on the substrate screening effects. When there are insulating encapsulation or separation, for instance, using bilayer graphene (or BN in other works), the exciton binding energy may drastically reduce by a factor of 3 to 6. At room temperature, exciton emission peaks are sig- nificantly broadened, because of large electron-phonon coupling rates in typical TMDCs. Biexcitons in 2D ma- terials are unstable at room temperature and dissociate, since their binding energies are on the order of only a few meV. The exciton-exciton annihilation (EEA) which is a four-particle process can occur under high exciton population (demanding high illumination intensities) in principle even at the room temperature, but it can be expected that at such high rates, non-radiative recom- binations through defect states and impurities would be dominant over the entire EEA process. Dark excitons in most 2D materials can actually emit light. Since they are optically forbidden because of vanishing dipole. How- ever, the true dipole should be complex valued instead of a real one, so that they can emit non-linearly polarized light (not necessarily exactly circular). The fact that the energy different between dark and bright excitons is on the same order of exciton binding energy complicates the correct interpretation of light emission. While the existence of biexcitons at room tempera- ture is unlikely because of the small binding energy, there are strong reasons to believe that emission from charged excitons could be easily mistaken with defect emissions because of interface effects [32, 140, 141]. Quite possi- bly, the existence of higher-wavelength emission peaks in photo-luminescence spectra of TMDCs could be due completely due to interface defects, and in that case tri- ons/biexciton emissions can have no influence at room- temperature as they normally appear only at cryogenic conditions [142]. If ture, then the observed trion emission peaks could be actually a defect-related emission, and for this reason is entirely absent in boron-nitride (BN) en- capsulated monolayer TMDCs [32, 142]. Similarly, there exists experimental evidence for brightened dark exci- tons/trions in WSe2, where measurements are all done at low temperatures [143]. Another recent study [144] is citing the fact that the two emission peaks for MoS2 on thermally grown SiO2, correspond to A-exciton and interface Defect. The defect state disappears at room temperature, which is somehow FIG. 8. Computation of near-dengenerate susceptibil- ity χ(3)(−ω; ωp,−ωp, ω) versus pump wavelength difference ∆λ = λp − λ at fixed λ = 775nm for various TMDCs. then χ(3)(−ω; ωp,−ωp, ω) stays within the same order of magnitude of χ(3)(−ω; ω,−ω, ω), thus leaving the pre- sented analysis and discussions basically intact. A closely related scheme with two pumps with frequencies ω± p = ω ± ∆Ω where ∆Ω is a non-zero integer multiple of FSR has been proposed in fiber [136, 137] and recently used in integrated [130] ring resonators for squeezing and ran- dom number generation, respectively. It is worthwhile to mention that the reverse of this scheme where two iden- tical pump photons decompose into two different signal and idler photons in higher and lower frequency neigh- boring sidebands has been shown [112] to be useful for on-chip optical squeezing, too. near-degenerate evaluation χ(3)(−ω; ωp,−ωp, ω), as shown in Fig. 8 for an FSR of 3nm at the wavelength of 775nm, for instance, in case of WSe2 reveals that χ(3) is actually reduced from the degenerate value of χ(3) = 1.2 × 10−23m3/V2 roughly by 49.3%, to the new non-degenerate pumped value of χ(3) = 6.05 × 10−24m3/V2 at a pump wave- length of 778nm, which is again purely imaginary. This drop could be easily accounted for by a proportional increase in pump power. Similarly, if pumping is done at 772nm, then χ(3) = 5.95 × 10−24m3/V2, which is a 50.2% reduction again. Therefore, when ∆λ = λp − λ with λp being the pump wavelength is sufficiently small, then the approximation numerical A of χ(3)(−ω; ωp,−ωp, ω) ≈ 1 2 χ(3)(−ω; ω,−ω, ω), (27) holds. This type of behavior is more or less the same for all other sorts of TMDCs, as shown in Fig. 8. This scheme, as opposed to the ultrashort solitons, en- ables continuous pumping and signal feeding, and there- fore a continuous squeezed output may be expected as well. IV. EXCITONIC EFFECTS Without doubt, the prominent role of excitons in light emission from TMDCs could be considered as a subject of deep study. This is due to the fact that all of the four MoS2WSe2WS2MoSe2-10-5510Δλ(nm)2.×10-244.×10-246.×10-248.×10-241.×10-231.2×10-23χyyyy(3)[m3V-2]connected to observations reported in [32]. Furthermore, they propose an elegant way to identify a defect state. The emission of a defect is almost always unpolarized regardless of the pump incidence angle, intensity, and po- larization, and that is why it cannot form polaritons in cavity quantum electrodynamics (CQED) experiments. A careful polarization measurement on the emission spec- tra could unanimously reveal whether the lower energy peak is a defect or excitonic emission. In summary, the origin and nature of peaks in the emis- sion spectra of TMDCs has been a matter of unresolved debate [32, 140 -- 157] and still remains with no conclusive agreement so far. V. CONCLUSIONS We presented a detailed theoretical analysis of lin- ear and third-order nonlinear optical response of two- dimensional (2D) monolayer transition metal dichalgo- nides (TMDCs). Based on a rigorous six-band tight- binding model, we calculated first order and third or- der susceptibility tensors, and observed reasonable con- sistency with experimental data. We predict an elon- gated non-classical light using high quality silica micro- disk resonators covered with the monolayer TMDCs. ACKNOWLEDGMENTS This work has been supported in part by Laboratory of Photonics and Quantum Measurements (LPQM) at EPFL and European Graphene Flagship under grant 696656, as well as Research Deputy of Sharif University of Technology. The author would like to thank Dr. Habib Rostami at Istituto Italiano di Tecnologia (IIT) as well as Dr. Rafael Rold´an and Dr. Pablo San-Jose at Instituto de Cien- cia de Materiales de Madrid (CSIC) for discussions and assistance on the Tight-Binding method. Enlightening discussions with Prof. Steven Louie from University of California at Berkeley, Dr. Bernhard Urbaszek at Labo- ratoire de Physique et Chimie des Nano-objets, Dr. Igor Aharonovich at University of Technology Sydney, as well as Dr. Anshuman Kumar and Cl´ement Javerzac-Galy at ´Ecole Polytechnique F´ed´erale de Lausanne (EPFL) on excitonic emission effects is highly appreciated. The author is highly indebted to Hiwa Mahmoudi at Institute of Electrodynamics, Microwave and Circuit En- gineering in Technische Universitat Wien, and in particu- lar, the Laboratory for Quantum Foundations and Quan- tum Information on the Nano- and Microscale in Vienna Center for Quantum Science and Technology (VCQ) at Universitat Wien for their warm and receptive hospitality 11 during the preparation of this article. This paper is dedicated to the celebrated artist, Anas- tasia Huppmann. APPENDICES Appendix A: Six-Band Tight-Binding The 6 × 6 Hamiltonian can be decomposed as [47 -- 50] (cid:20)HMM(k; σ) HMX(k) H† MX(k) HXX(k; σ) (cid:21) H(k; σ) = , (A1) 3(cid:88) 3(cid:88) j=1 TMM j cos(k · aj), TXX j cos(k · aj), HMM(k; σ) = EM(σ) + 2 HXX(k; σ) = EX(σ) + 2 3(cid:88) j=1 TMX j exp(−ik · δj). HMX(k) = j=1 The vectors aj form the lattice basis vectors and ex- tend from every metal to the nearest neighbors. They are all equal in length, given as the lattice constant a = aj. The vectors δj have equal length, given by √ δj = δ = a/ 3, but make right angles with the ba- √ sis vectors. A desirable choice for these vectors are 2 a(−1, a1 = 1 3), and a2 = a(1, 0). 3,−1), and Then we have δ1 = 1 δ2 = δ(0, 1). With these conditions, the coordinates of high-symmetry reciprocal lattice points are K = 4π 3a (1, 0), √ M = π 3, 1), and Γ = (0, 0). In Table I, the lengths 3 3 of these vectors for various TMDCs are introduced. 2 a(−1,−√ √ 3), a3 = 1 3,−1), δ3 = 1 2 δ(−√ 2 δ( √ ( TABLE I. Lattice constants a = √ 3δ of TMDCs [50]. MoS2 MoSe2 WS2 WSe2 3.160A 3.288A 3.153A 3.260A The 3 × 3 submatrices in the above are given as 0 0  , ∆0 ∆p + Vppπ − i 0 ∆2 −iλMσ 0 iλMσ ∆2 2 λXσ i 2 λXσ ∆p + Vppπ 0 0 EM(σ) = EX(σ) = (A2)  . 0 0 ∆z − Vppσ The rest of the matrices are introduced as follows. Starting with TMM j we have TABLE II. Slater-Koster parameters for TMDCs. All param- eters are in units of electron-volts (adapted from [50] with minor corrections). SOC Crystal Fields M − X M − M X − X 0.271 0.057 0.089 0.256 MoS2 MoSe2 WS2 WSe2 λM 0.086 0.251 λX 0.052 0.439 ∆0 −1.094 −1.144 −1.155 −0.935 ∆1 −0.050 −0.250 −0.650 −1.250 ∆2 −1.511 −1.488 −2.279 −2.321 ∆p −3.559 −4.931 −3.864 −5.629 ∆z −6.886 −7.503 −7.327 −6.759 Vpdσ 3.689 5.083 Vpdπ −1.241 −1.222 −1.220 −1.081 Vddσ −0.895 −0.823 −1.328 −1.129 0.094 Vddπ 0.252 0.317 Vpdδ 0.228 Vppσ 1.225 1.530 Vppπ −0.467 −0.205 −0.273 −0.123 0.215 0.192 1.256 0.121 0.442 1.178 3.728 4.911  3Vddδ + Vddσ √ 2 (−Vddδ + Vddσ) 2 (Vddδ − Vddσ) − 3 3 TMM 1 = 1 4 TMM 3 = 1 4 For TXX j we have The submatrices TMX j TXX 1 = 0 1 4 √ 3(Vppπ − Vppσ) TXX 2 =  3Vppπ + Vppσ Vppσ  3Vppπ + Vppσ −9Vpdπ + √ 5 −Vpdπ − 3 3Vpdπ + 3Vpdσ 3(Vppπ − Vppσ) −√ √ √ 1 4 0 0 0 TXX 3 = TMX 1 = √ √ 2 7 7 given in [50], with minor corrections take the form 3Vdpπ − Vpdσ −12Vpdπ − √ √ 9Vpdπ − √ 3Vpdσ 3 √ √ 3Vpdπ + 3Vpdσ −6Vpdπ + 3 3Vpdσ 5 3Vpdσ 3Vpdπ − 3Vpdσ 3Vpdσ √ 2 3Vpdσ  , √ 2 (−Vddδ + Vddσ) 3 1 4 (Vddδ + 12Vddπ + 3Vddσ) √ 4 (Vddδ − 4Vddπ + 3Vddσ) 3 − 3 2 (Vddδ − Vddσ) √ 4 (Vddδ − 4Vddπ + 3Vddσ) 1 4 (3Vddδ + 4Vddπ + 9Vddσ) 3  3Vddδ + Vddσ 3(Vddδ − Vddσ) √ 0 √ 3(Vddδ − Vddσ) 3Vddδ + Vddσ 0 0 0 4Vddπ 3 √ 2 (−Vddδ + Vddσ) 4 (Vddδ + 12Vddπ + 3Vddσ) −√ 4 (Vddδ − 4Vddπ + 3Vddσ) 3 2 (Vddδ − Vddσ) 3 4 (Vddδ − 4Vddπ + 3Vddσ) 1 4 (3Vddδ + 4Vddπ + 9Vddσ) TMM 2 = 1 4  3Vddδ + Vddσ √ 2 (−Vddδ + Vddσ) 3 2 (Vddδ − Vddσ) −√ 1 3 3  ,  .  ,  , 0 0 4Vppπ √ 3(Vppπ − Vppσ) Vppπ + 3Vppσ 0 0 0 0 Vppπ 0 Vppπ  , −√ 3(Vppπ − Vppσ) Vppπ + 3Vppσ 0 0 0 4Vppπ  . 12 (A3) (A4) (A5) 0 7 √ √ 2 7  0  9Vpdπ − √ 3Vpdπ + 2Vpdσ −12Vpdπ − √ √ −6 √ −6Vpdπ − 4 0 √ 3Vpdπ − Vpdσ −12Vpdπ − √ √ 9Vpdπ − √ √ 3Vpdσ 3 −5 3Vpdπ − 3Vpdσ √ √ √ 2 3Vpdσ 6Vpdπ − 3 3Vpdπ − 3Vpdσ 3Vpdσ −5 −Vpdπ − 3 3Vpdσ 3Vpdπ + 6Vpdσ √ 3Vpdσ −4 14Vpdπ 0  , 3Vpdσ 3Vpdπ − 3Vpdσ 3Vpdσ 13  . TMX 2 = TMX 3 = √ √ 2 7 7 Table II presents the Slater-Koster parameters needed for this analysis. Data are compiled from literature [50] with minor corrections, and numerical results from the model presented in this Appendix are basically identical to the ones reported therein. Comparison to experimental values are already done in many of works, however, where the interested reader is referred to. Appendix B: Expressions for χ(3) Following the standard perturbation method to calculate the nonlinear susceptibility tensor [76 -- 88], the χ(3) is composed of two paramagnetic Π and diamagnetic ∆ parts, as µνζη(−ωΣ; ωr, ωs, ωt) = (cid:88) χ(3) (cid:90)(cid:90) 1 (cid:2)Πσ ABZ σ IRBZ 6 AUC
0(ωΣωrωsωt) µνζη(k; ωr, ωs, ωt) + ∆σ × µνζη(k; ωr, ωs, ωt)(cid:3) d2k, (B1) where ωΣ = ωr + ωs + ωt. For the case of third-harmonic generation, we have ω = ωr = ωs = ωt and ωΣ = 3ω. For the Kerr nonlinearity we have ω = ωr = −ωs = ωt = ωΣ. For the paramagnetic contribution, we have Πσ µνζη(k; ωr, ωs, ωt) = P (cid:88) mnpq mn(k)jνσ jµσ np (k)jζσ ωΣ + Eσ pq (k)jησ mq(k) qm(k) Ξσ mnpq(k; ωr, ωs, ωt), (B2) (B3) Ξσ mnpq(k; ωr, ωs, ωt) = mnp(k; ωr + ωs, ωt, ωt) − W σ W σ npq(k; ωt + ωs, ωt, ωr). The operator P represents all possible intrinsic permutations among frequencies ωr, ωs, ωt. That implies no permu- tation when all three frequencies are equal, three permutations when only one frequency is different as (ω1, ω2, ω2), (ω2, ω1, ω2), and (ω2, ω2, ω1), and six permutations otherwise as (ω1, ω2, ω3), (ω1, ω3, ω2), (ω2, ω1, ω3), (ω2, ω3, ω1), (ω1, ω2, ω3), and (ω1, ω3, ω2). The functions W σ mnp(k; w1, w2, w3) and U σ pq(k; w) are given as For the diamagnetic contribution, we have W σ mnp(k; w1, w2, w3) = U σ pq(k; w) = µνζη(k; ωr, ωs, ωt)= P (cid:88) ∆σ np(k; w3) , mp(k) mn(k; w2) − U σ U σ w1 + Eσ q (k) pq(k) p (k) − f σ f σ w + Eσ . mnpq (k; ωr, ωs, ωt) − [Aµνζησ mnpq mnpq (k; ωr, ωs, ωt) − C µνζησ Bµνζησ mn (k)gζησ nm (k), mnpq (k; ωr, ωs, ωt)], Aµνζησ Bµνζησ mnpq (k; ωr, ωs, ωt)= U σ mnpq (k; ωr, ωs, ωt)= jµσ C µνζησ mnpq (k; ωr, ωs, ωt)= jησ nm(k)gζησ mn(k; ωr + ωs)gµνσ pm(k)jνσ mp (k) mnp(k; ωr + ωs, ωt, ωt), nm(k)jησ pm (k) mnp(k; ωΣ, ωr + ωs, ωt). np (k)gµνσ W σ W σ (B4) (B5) (B6) (B7) (B8) The matrix element gµνσ pm (k) is obtained as n(k)(cid:105) , µν(k)ψσ m(k)gσ mn (k) = (cid:104)ψσ gµνσ µν(k) = − q2 gσ 2 ∂2H(k; σ) ∂kµ∂kν . in which x is in the units of zero-point fluctuations xzp, x0 = (cid:60)[α], p0 = (cid:61)[α], and (B9) g = coshζ + (cid:60)[ζ] ζ sinhζ, h = (cid:61)[ζ] sinhζ C =(cid:112)g(1 + 2ih). 2ζeζ , It is straightforward to verify via Fourier transformation that the momentum representation would be given by (cid:104)pα, ζ(cid:105) = π 1 4 eix0(p−p0) (cid:113) 1 g − 2ihC exp − (p − p0)2 (cid:16) 1 (cid:17) gC2 − 2ih 2  . Since we have assumed that ζ = is, we have 14 (C2) (C3) (C4) (cid:27) (cid:27) , In all the above relationships, a small positive imagi- nary part is normally added to all three frequencies, in order to preserve causality as well as to avoid numerical overflow at resonances. Needless to say, coding these rela- tions are not straightforward and needs extra care. The double integration over the reciprocal lattice could be done by evaluation of the integrand over a discrete trian- gular grid and multiplying by element sizes. Employing adaptive numerical integration methods is not practical because of the extremely large numerical burden. Appendix C: Imaginary Squeezing If the third-order nonlinear susceptibility χ(3) is imag- inary, then the two-photon absorption is dominant over the Kerr effect. While the Kerr effect could in princi- ple produce unlimited squeezing, two-photon absorption may cause a limited squeezing of shot noise. This effect has been noticed by a number of authors in the past [158 -- 164]. However, in all these works the temporal evolution of squeeze parameter is considered, in which squeezing generally increases with propagation, reaching an ulti- mate value in the limit of infinite propagation in a two- photon absorbing medium. That type of analysis is useful for nonlinearly lossy long fibers. Here, for the case of a ring resonator with constant propagation length, it is the intensity, or equivalently, the cavity photon number N , of the input beam which could be varied. In what follows, we show that squeezing generally increases with the input power, first within the approximation of negligible loss over propagation, that is constant ζ. 1. Non−rotated Squeezing The wavefunction corresponding to the squeezed co- herent state α, ζ(cid:105) with α being the complex coherent state number, generated by an interaction of the type (10) or equivalently produced from a coherent state such as α(cid:105) by application of the operator S(ζ) (24), is given by [165 -- 167] (cid:104)xα, ζ(cid:105) = ei(x−x0)p0 π 1 4 C exp (cid:20) − (cid:18) 1 2gC 2 − ih (cid:19) (cid:21) (x − x0)2 , g = cosh s, sinh s 2es . h = (cid:26) (cid:20) 1 (cid:21) gC 2 Therefore, the probablity distribution in position is given by (cid:104)xα, ζ(cid:105)2 = 1√ πC2 exp −(cid:60) (x − x0)2 , (C5) where we have noticed that h is real-valued. Similarly, for the probablity distribution in the momentum repre- sentation we obtain (cid:26) −(cid:60) (cid:20)(cid:16) 1 (cid:17)−1(cid:21) gC2 − 2ih √ π 1 g − 2ihC (p − p0)2 (cid:104)pα, ζ(cid:105)2 = exp (C6) Comparing to the Gaussian distribution of a simple co- herent state [97], we may deduce the noise squeezing ra- tio
in position and ϕ in momentum, also known as the Fano factor [164], respectively by solving the equations (cid:20) 1 (cid:21) (cid:60)(cid:104) 1 (cid:105) 1 gC2 − 2ih2 gC 2 gC2 =
−2 = (cid:60) ϕ−2 = 1 g2(1 + 4h2) , (C7) = g2 (1 + 4h2g2)2 + 4h2g4 . √ Interestingly, for a purely real ζ with (cid:61)[ζ] = 0, we have g, h = 0, and this expression takes the simple C = solution
= exp(ζ) and ϕ = exp(−ζ). This corresponds to a minimum uncertainty squeezed packet ∆ϕ∆
= (C8) with ∆ϕ = ϕ/ 2. But for the present case where (cid:60)[ζ] = 0, after some alegbraic manipulations it gives the solutions 2 and ∆
=
/
= cosh(s) 1 + e−2s sinh2(s), √ (cid:113) (cid:115) (C9) (cid:21)2 . sinh2(2s) 4e2s (C1) ϕ = sinh2(2s) 4e2s + sech2(s) 1 + 1 , √ 2 (cid:20) √ It is easy to observe that for large values of s, we get
≈ 2ϕ ≈ ( 5/4)es, which represents unlimited desqueezing and shot noise increase. A plot of
and ϕ versus s is illustrated in Fig. 9. The momentum quadrature exhibits a minimum of ϕmin = −2.73dB at smin = 0.797 15 FIG. 9. Variation of squeezing amplitudes versus absolute value of purely imaginary squeeze parameter. Both quadra- tures desqueeze for large s. FIG. 10. Wigner distribution of the squeezed coherent state with α = 1 and ζ = i. Both non-rotated orthogonal quadra- tures (x, p) appear to be desqueezed almost to the same amount. 2. π 4 −rotated Squeezing (cid:90)(cid:90) (cid:20) While non-rotated quadratures both appear to be strongly desqueezed, the fact is that the true squeezing actually happens along the π illustrate this, we need to rigorously calculate the Wigner function of the squeezed vacuum first. This is given by 4−rotated quadratures. To W S(ζ)0(cid:105)(θ) = 1 π2 exp(θβ∗ − θ∗β) (C10) (cid:21) exp − 1 2 βµ + β∗ν2 d2β, where θ = x + ip, β = a + ib, µ = coshζ, and ν = exp(i∠ζ) sinhζ. Using the fact that ζ = is, and some straightforward but significant algebra, we get the Wigner distribution of purely-imaginary squeezed vac- uum as W S(ζ)0(cid:105)(x, p) = 2 π exp exp (cid:104)xα, ζ(cid:105)2 = (cid:104)pα, ζ(cid:105)2 = × (cid:21) − 2p2 cosh(2s) (cid:20) (cid:110)−2 cosh(2s) [x + tanh(2s)p]2(cid:111) (cid:90) (cid:90) W D(α) S(ζ)0(cid:105)(x, p)dp, W D(α) S(ζ)0(cid:105)(x, p)dx, (C11) . (C12) The probablity distributions obtained as [97] from the above Wigner function both appear to be desqueezed almost equally, as discussed in the above and shown in Fig. 9 for large s. Here, the displacement oper- ator D(α) which produces a coherent state simply trans- forms x → x − (cid:60)[α] and p → p − (cid:61)[α]. In the the time- dependent case, we only need to replace α = A exp(iωt) where A is the amplitude of the coherent source and ω FIG. 11. Variation of squeezing amplitudes in the π system versus absolute value of purely imaginary squeeze pa- rameter. One rotated quadrature is squeezed and the other is strongly desqueezed. 4 −rotated is the angular frequency of light. The typical resulting Wigner function for the squeezed coherent state is illus- trated in Fig. 10. Now, we can define the π 4−rotated coordinates as X + P√ 2 X − P√ , . x = p = 2 (C13) In this new system of coordinates, we obtain the Wigner distribution of the squeezed state α, ζ(cid:105) = D(α) S(is)0(cid:105) as exp(cid:2)−2e2s(X − (cid:60)[α])2(cid:3) × exp(cid:2)−2e−2s(P − (cid:61)[α])2(cid:3) , 2 π Wα,ζ(cid:105)(X, P ) = (C14) (C15) with the respective squeeze ratios √ √ 2∆
= e−s, 2∆ϕ = es,
X = ϕP = plotted in Fig. 11, and giving the uncertainty product ∆
∆ϕ = 1 2 , (C16) ϱϑ0.050.100.501s-2246810ϱ,ϑ(dB)-3-2-10123-3-2-10123xpϑPϱX0.050.100.501s-20-101020ϱX,ϑP(dB)which refers to the minimum-uncertainty packet. Appendix D: Units of Surface Quantities While the linear bulk susceptibility χ(1) should be di- mensionless, the surface linear susceptibility χ(1)2D is not, actually having the dimension of length. But it is 16 preferable to write the dimension as m(cid:3) or A(cid:3) where the redundant dimensionless square notation (cid:3) emphasizes a 2D surface quantity. Similarly, the dimension of nonlin- ear susceptibility χ(3)2D would be preferably m3V−2(cid:3) instead of m3V−2. It follows then, that the preferable di- mension of linear and nonlinear sheet conductivity σ(3)2D must be respectively Ω−1(cid:3) and Ω−1m2V−2(cid:3). [1] A. V. Kolobov and J. Tominaga, Two-Dimensional Transition-Metal Dichalcogenides (Springer, Switzer- land, 2016). [2] P. Avouris, T. F. Heinz, and T. Low, eds., 2D Ma- terials: Properties and Devices (Cambridge University Press, Cambridge, 2017). [3] J. R. Schaibley, H. Yu, G. Clark, P. Rivera, J. S. Ross, K. L. Seyler, W. Yao, and X. Xu, Nature Rev. Mat. 1, 16055 (2016). [25] G. Soavi, G. Wang, H. Rostami, D. G. Purdie, D. D. Fazio, T. Ma, B. Luo, J. Wang, A. K. Ott, D. Yoon, S. A. Bourelle, J. E. Muench, I. Goykhman, S. D. Conte, M. Celebrano, A. Tomadin, M. Polini, G. Cerullo, and A. C. Ferrari, Nature Nanotech. (2018), doi:10.1038/s41565-018-0145-8. [26] K. Alexander, N. A. Savostianova, S. A. Mikhailov, B. Kuyken, and D. V. Thourhout, ACS Photonics 4, 3039 (2017). [4] M. Chhowalla, D. Jena, and H. Zhang, Nature Rev. [27] S. Manzeli, D. Ovchinnikov, D. Pasquier, O. V. Yazyev, Mat. 1, 16052 (2016). [5] Q. H. Wang, K. Kalantar-Zadeh, A. Kis, J. N. Coleman, and M. S. Strano, Nature Nanotech. 7, 699 (2012). [6] B. Radisavljevic, A. Radenovic, J. Brivio, V. Gia- cometti, and A. Kis, Nature Nanotech. 6, 147 (2011). [7] S. A. Jafari, J. Phys.: Condens. Matter 24, 205802 and A. Kis, Nature Rev. Mater. 2, 17033 (2017). [28] T. Tamya, A. Ishikawa, T. Ogawa, Phys. Rev. Lett. 116, 016601 (2016). [29] T. Tamya, A. Ishikawa, T. Ogawa, Phys. Rev. B 94, 241107(R) (2016). and K. Tanaka, and K. Tanaka, [30] N. Yoshikawa, T. Tamaya, and K. Tanaka, Science 356, (2012). 736 (2017). [8] A. Pospischil and T. Mueller, Appl. Sci. 6, 78 (2016). [9] M. Koperski, M. R. Molas, A. Arora, K. N. A. O. Slo- bodeniuk, C. Faugeras, and M. Potemski, Nanophoton. 6, 1289 (2017). [10] S. Khorasani and A. Koottandavida, npj 2D Mat. Appl. 1, 7. [11] S. Khorasani, (2018), preprints:2018050201. [12] M. L. Trolle, G. Seifert, and T. G. Pedersen, Phys. Rev. B 89, 235410 (2014). [13] K. L. Seyler, J. R. Schaibley, P. Gong, P. Rivera, A. M. Jones, S. Wu, J. Yan, D. G. Mandrus, W. Yao, and X. Xu, Nat. Nanotech. 10, 407 (2015). [14] T. K. Fryett, K. L. Seyler, J. Zheng, C.-H. Liu, X. Xu, and A. Majumdar, 2D Mater. 4, 015031 (2016). [15] T. K. Fryett, A. Zhan, and A. Majumdar, Opt. Lett. 42, 3586 (2017). [16] T. K. Fryett, A. Zhan, and A. Majumdar, Nanophoton. 7, 355 (2017). [17] J. L. Cheng, N. Vermeulen, and J. E. Sipe, New J. Phys. 16, 053014 (2014). [18] S. A. Mikhailov, Europhys. Lett. 79, 27002 (2007). [19] S. A. Mikhailov and K. Ziegler, J. Phys.: Cond. Mat. 20, 384204 (2008). [20] J. L. Cheng, N. Vermeulen, and J. E. Sipe, Phys. Rev. B 92, 235307 (2015). [21] S. A. Mikhailov, Phys. Rev. B 93, 085403 (2016). [22] B. Semnani, A. H. Majedi, and S. Safavi-Naeini, J. Optics 18, 035402 (2016). [31] M. Manca, M. M. Glazov, C. Robert, F. Cadiz, T. Taniguchi, K. Watanabe, E. Courtade, T. Amand, P. Renucci, X. Marie, G. Wang, and B. Urbaszek, Nat. Commun. 8, 14927 (2017). [32] F. Cadiz, E. Courtade, C. Robert, G. Wang, Y. Shen, H. Cai, T. Taniguchi, K. Watanabe, H. Carrere, D. La- garde, M. Manca, T. Amand, P. Renucci, S. Tongay, X. Marie, and B. Urbaszek, Phys. Rev. X 7, 021026 (2017). [33] P. Dey, L. Yang, C. Robert, G. Wang, B. Urbaszek, X. Marie, and S. A. Crooker, Phys. Rev. Lett. 119, 137401 (2017). [34] G. Wang, C. Robert, M. M. Glazov, F. Cadiz, E. Cour- tade, T. Amand, D. Lagarde, T. Taniguchi, K. Watan- abe, B. Urbaszek, and X. Marie, Phys. Rev. Lett. 119, 047401 (2017). [35] C. Javerzac-Galy, A. Kumar, R. D. Schilling, N. Piro, S. Khorasani, M. Barbone, I. Goykhman, J. B. Khurgin, A. C. Ferrari, and T. J. Kippenberg, Nano Lett. 18, 3138 (2018). [36] L. Hilico, J. M. Courty, C. Fabre, E. Giacobino, and J. L. Oudar, Appl. Phys. B 55, 202 I. Abram, (1992). [37] R. Saito, G. Dresselhaus, and M. S. Dresselhaus, Phys- ical Properties of Carbon Nanotubes (Imperial College Press, London, 1998). [38] B. Gharekhanlou, M. Alavi, and S. Khorasani, Semi- cond. Sci. Tech. 23, 075026 (2008). [23] N. Yoshikawa, T. Tamaya, and K. Tanaka, Science 356, [39] B. Gharekhanlou and S. Khorasani, IEEE Trans. Elec- 736 (2017). [24] N. A. Savostianova and S. A. Mikhailov, Opt. Express 25, 3268 (2017). tron Dev. 57, 209 (2010). [40] B. Gharekhanlou and S. Khorasani, in Graphene: Prop- erties, Synthesis and Applications, edited by Z. Xu (Nova Science Publishers, New York, 2011) pp. 1 -- 36. 17 [41] D. Xiao, G.-B. Liu, W. Feng, X. Xu, and W. Yao, Phys. [71] S. Wang, Fundamentals of Semiconductor Theory and Rev. Lett. 108, 196802 (2012). [42] H. Rostami, A. G. Moghaddam, and R. Asgari, Phys. Rev. B 88, 085440 (2013). [43] H. Rostami, F. Guinea, M. Polini, and R. Rold´an, npj 2D Mat. Appl. 2, 15 (2018). [44] R. Winkler, Spin-Orbit Coupling Effects in Two- Dimensional Electron and Hole Systems (Springer, Berlin, 2003). [45] G.-B. Liu, W.-Y. Shan, Y. Yao, W. Yao, and D. Xiao, Phys. Rev. B 88, 085433 (2013). [46] A. Kormanyos, V. Zolyomi, N. D. Drummond, P. Rakyta, G. Burkard, and V. I. Falko, Phys. Rev. B 88, 045416 (2013). [47] R. Rold´an, M. P. L´opez-Sancho, F. Guinea, E. Cappel- luti, J. A. Silva-Guill´en, and P. Ordej´on, 2D Mater. 1, 034003 (2014). [48] R. Rold´an, J. A. Silva-Guill´en, M. P. L´opez-Sancho, F. Guinea, E. Cappelluti, and P. Ordej´on, Ann. Phys. (Berlin) 526, 347 (2014). [49] H. Rostami, R. Roldan, E. Cappelluti, R. Asgari, and F. Guinea, Phys. Rev. B 92, 195402 (2015). Device Physics (Prentice-Hall, 1989). [72] A. Avogadro, J. Physique 73, 58 (1811). [73] R. Fox and T. Hill, American Scientist 95, 104 (2007). [74] B. Andreas, Y. Azuma, G. Bartl, P. Becker, H. Bet- tin, M. Borys, I. Busch, M. Gray, P. Fuchs, K. Fu- jii, H. Fujimoto, E. Kessler, M. Krumrey, U. Kuet- gens, N. Kuramoto, G. Mana, P. Manson, E. Massa, S. Mizushima, A. Nicolaus, A. Picard, A. Pramann, O. Rienitz, D. Schiel, S. Valkiers, and A. Waseda, Phys. Rev. Lett. 106, 030801 (2011). [75] G. Mana, E. Massa, and C. P. Sasso, J. Phys. Chem. Ref. Data 44, 031209 (2015). [76] A. Saynatjoki, L. Karvonen, H. Rostami, A. Autere, S. Mehravar, A. Lombardo, R. A. Norwood, T. Hasan, N. Peyghambarian, H. Lipsanen, K. Kieu, A. C. Ferrari, M. Polini, and Z. Sun, Nat. Commun. 8, 893 (2017). [77] R. W. Boyd, Nonlinear Optics, 3rd ed. (Academic Press, New York, 2008). [78] N. Peyghambarian, S. W. Koch, and A. Mysyrowicz, Introduction to Semiconductor Optics (Prentice-Hall, Englewood Cliffs, 1993). [50] J. A. Silva-Guill´en, P. San-Jose, and R. Rold´an, Appl. [79] M. Kira and S. W. Koch, Semiconductor Quantum Op- Sci. 6, 284 (2016). [51] E. Cappelluti, R. Rold´an, J. A. Silva-Guill´en, P. Or- dej´on, and F. Guinea, Phys. Rev. B 88, 075409 (2013). [52] J. C. Slater and G. F. Koster, Phys. Rev. 94, 1498 (1954). [53] X. Xu, W. Yao, D. Xiao, and T. F. Heinz, Nature Phys. 10, 343 (2014). tics (Cambridge University Press, Cambridge, 2012). [80] R. L. Sutherland, Handbook of Nonlinear Optics (Mar- cel Dekker, New York, 1996). [81] P. D. Drummond and M. Hillery, The Quantum The- ory of Nonlinear Optics (Cambridge University Press, Cambridge, 2014). [82] Y. R. Shen, The Principles of Nonlinear Optics (Wiley [54] Y. Li, Y. Rao, K. f. Mak, Y. You, S. Wang, C. R. Dean, Interscience, Hoboken, 2003). and T. F. Heinz, Nano Lett. 13, 3329 (2013). [55] M. A. Dupertuis, M. Proctor, and B. Acklin, J. Opt. Soc. Am. A 11, 1159 (1994). [56] M. A. Dupertuis, B. Acklin, and M. Proctor, J. Opt. Soc. Am. A 11, 1167 (1994). [83] P. N. Prasad and D. J. Williams, Introduction ot Non- linear Optical Effects in Molecules and Polymers (John Wiley & Sons, New York, 1991). [84] J. F. Ward, Rev. Mod. Phys. 37, 1 (1965). [85] S. J. Lalama and A. F. Garito, Phys. Rev. A 20, 1179 [57] S. Khorasani and B. Rashidian, J. Opt. A: Pure Appl. (1979). Opt. 3, 380 (2001). [58] S. Khorasani and B. Rashidian, J. Opt. A: Pure Appl. Opt. 4, 251 (2002). [59] S. Khorasani, A. Nojeh, and B. Rashidian, Fiber Int. Opt. 21, 173 (2002). [86] A. F. Garito and K. Y. Wong, Polymer J. 19, 51 (1987). [87] J. A. Armstrong, N. Bloembergen, J. Ducuing, and P. S. Pershan, Phys. Rev. 127, 1918 (1962). [88] R. Braunstein, Phys. Rev. 125, 475 (1962). [89] M. Lindberg, Y. Z. Hu, R. Binder, and S. W. Koch, [60] E. Darabi, S. Khorasani, and B. Rashidian, Semicond. Phys. Rev. B 50, 18060 (1994). Sci. Tech. 18, 60 (2003). [61] S. Khorasani and B. Rashidian, Scientia Iranica 10, 426 (2003). [62] F. Karimi and S. Khorasani, IEEE J. Quantum Elect. 50, 607 (2013). [90] V. M. Axt and A. Stahl, Z. Phys. B 93, 195 (1994). [91] R. I. Woodward, R. T. Murray, C. F. Phelan, R. E. P. de Oliveira, T. H. Runcorn, E. J. R. Kelleher, S. Li, E. C. de Oliveira, G. J. M. Fechine, G. Eda, and C. J. S. de Matos, 2D Mater. 4, 011006 (2017). [63] T. Zhan, X. Shi, Y. Dai, X. Liu, and J. Zi, J. Phys.: [92] H. Liu, Y. Li, Y. S. You, S. Ghimire, T. F. Heinz, and Condens. Matter 25, 215301 (2013). D. A. Reis, Nature Phys. 13, 262 (2017). [64] X.-H. Deng, J.-T. Liu, J.-R. Yuan, Q.-H. Liao, and [93] S. A. Jafari, T. Tohyama, and S. Maekawa, J. Phys. N.-H. Liu, Europhys. Lett. 109, 27002 (2015). Soc. Jpn. 75, 054703 (2006). [65] G. Sz´echenyi, M. Vigh, A. Korm´anyos, and J. Cserti, J. Phys.: Condens. Matter 28, 375802 (2016). [66] M. Merano, Opt. Express 23, 31602 (2015). [67] M. Merano, Phys. Rev. A 93, 013832 (2016). [68] E. Malic and A. Knorr, Graphene and Carbon Nan- otubes: Ultrafast Optics and Relaxation Dynamics (Wiley-VCH, Weinheim, 2013). [94] L. Karvonen, A. Saynatjoki, M. J. Huttunen, A. Autere, B. Amirsolaimani, S. Li, R. A. Norwood, N. Peygham- barian, H. Lipsanen, G. Eda, K. Kieu, and Z. Sun, Nature Commun. 8, 15714 (2017). [95] R. Wang, H.-C. Chien, J. Kumar, N. Kumar, H.-Y. Chiu, and H. Zhao, ACS Appl. Mater. Interfaces 6, 314 (2013). [69] N. W. Ashcroft and N. D. Mermin, Solid State Physics [96] U. L. Andersen, T. Gehring, C. Marquardt, and (Brooks Cole, South Melbourne, 1976). G. Leuchs, Phys. Scr. 91, 053001 (2016). [70] C. Kittel, Introduction to Solid State Physics (Wiley, [97] W. Schleich, Quantum Optics in Phase Space (Wiley- 2004). VCH, New York, 2001). [98] R. Schnabel, Phys. Rep. 684, 1 (2017). 18 [99] H. Vahlbruch, M. Mehmet, K. Danzmann, and R. Schn- [124] R. M. Shelby, M. D. Levenson, D. F. Walls, A. Aspect, abel, Phys. Rev. Lett. 117, 110801 (2016). and G. J. Milburn, Phys. Rev. A 33, 4008 (1986). [100] E. Oelker, G. Mansell, M. Tse, J. Miller, F. Matichard, L. Barsotti, P. Fritschel, D. E. McClelland, M. Evans, and N. Mavalvala, Optica 3, 682 (2016). [101] M. C. Teich and B. E. A. Saleh, Quantum Opt. 1, 153 [125] C. M. Caves, Phys. Rev. D 23, 1693 (1981). [126] D. Stoler, Phys. Rev. D 1, 3217 (1970). [127] D. Stoler, Phys. Rev. D 4, 1925 (1971). [128] R. M. Shelby, P. D. Drummond, and S. J. Carter, Phys. (1989). [102] D. Mart´ın-Cano, H. R. Haakh, and M. Agio, in Quan- tum Plasmonics, edited by S. I. Bozhevolnyi, L. Martin- Moreno, and F. Garcia-Vidal (Springer, Switzerland, 2017) pp. 25 -- 46. [103] W. H. Richardson and R. M. Shelby, Phys. Rev. Lett. 64, 400 (1990). [104] A. H. Safavi-Naeini, S. Groblacher, J. T. Hill, J. Chan, and O. Painter, Nature 500, 185 M. Aspelmeyer, (2013). Rev. A 42, 2966 (1990). [129] M. Rosenbluh and R. M. Shelby, Phys. Rev. Lett. 66, 153 (1991). [130] Y. Okawachi, M. Yu, K. Luke, D. O. Carvalho, M. Lip- son, and A. L. Gaeta, Opt. Lett. 41, 4194 (2016). [131] P. Grelu and N. Akhmediev, Nat. Photon. 6, 84 (2012). [132] H. Guo, M. Karpov, E. Lucas, A. Kordts, M. H. P. Pfeiffer, V. Brasch, G. Lihachev, V. E. Lobanov, M. L. Gorodetsky, and T. J. Kippenberg, Nat. Phys. 13, 94 (2017). [105] T. P. Purdy, P. L. Yu, R. W. Peterson, N. S. Kampel, [133] E. Lucas, M. Karpov, H. Guo, M. Gorodetsky, and and C. A. Regal, Phys. Rev. X 3, 031012 (2013). T. Kippenberg, Nat. Commun. 8, 736 (2017). [106] K. Qu and G. S. Agarwal, Phys. Rev. A 91, 063815 [134] L. A. Lugiato and R. Lefever, Phys. Rev. Lett. 58, 2209 (2015). (1987). [107] C. Fabre, M. Pinard, S. Bourzeix, A. Heidmann, E. Gia- cobino, and S. Reynaud, Phys. Rev. A 49, 1337 (1994). [108] S. Mancini and P. Tombesi, Phys. Rev. A 49, 4055 (1994). [109] T. Corbitt, Y. Chen, F. Khalili, D. Ottaway, S. Vy- atchanin, S. Whitcomb, and N. Mavalvala, Phys. Rev. A 73, 023801 (2006). [110] V. Sudhir, D. Wilson, R. Schilling, H. Schutz, S. Fe- dorov, A. Ghadimi, A. Nunnenkamp, and T. Kippen- berg, Phys. Rev. X 7, 011001 (2017). [111] H. Yonezawa, K. Nagashima, and A. Furusawa, Opt. Express 18, 20143 (2010). [112] A. Dutt, K. Luke, S. Manipatruni, A. L. Gaeta, P. Nussenzveig, and M. Lipson, Phys. Rev. Appl. 3, 044005 (2015). [113] T. S. Iskhakov, U. B. Hoff, T. Gehring, and U. L. Andersen, in Presented at 605 WE-Heraeus-Seminar (Physikzentrum Bad Honnef, Germany, 17-22 January 2016). [114] M. Kitagawa and Y. Yamamoto, Phys. Rev. A 34, 3974 (1986). [115] P. D. Drummond and D. F. Walls, J. Phys. A 13, 725 (1980). [116] J. Heebner, R. Grover, and T. Ibrahim, Optical Mi- croresonators: Theory, Fabrication and Applications (Springer, London, 2008). [117] W. Bogaerts, P. DeHeyn, T. V. Vaerenbergh, K. De- Vos, S. K. Selvaraja, T. Claes, P. Dumon, P. Bienstman, D. VanThourhout, and R. Baets, Laser Photonics Rev. 6, 47 (2012). [118] T. Herr, M. L. Gorodetsky, and T. J. Kippenberg, in Nonlinear optical cavity dynamics: From Microres- onators to Fiber Lasers, edited by P. Grelu (Wiley-VCH Weinheim, New York, 2016) pp. 129 -- 162. [119] M. D. Levenson, R. M. Shelby, A. Aspect, M. Reid, and D. F. Walls, Phys. Rev. A 32, 1550 (1985). [120] R. M. Shelby, M. D. Levenson, S. H. Perlmutter, R. G. DeVoe, and D. F. Walls, Phys. Rev. A 57, 691 (1986). [121] R. Loudon and P. L. Knight, J. Mod. Opt. 34, 709 (1987). [122] C. H. Caves and B. L. Schumaker, Phys. Rev. A 31, 3068 (1985). [123] M. D. Levenson, R. M. Shelby, and S. H. Perlmutter, Opt. Lett. 10, 514 (1985). [135] F. Dini, M. M. Emamzadeh, S. Khorasani, J. L. Bobin, R. Amrollahi, M. Sodagar, and M. Khoshnegar, Phys. Scr. 77, 025504 (2008). [136] R. M. Shelby, M. D. Levenson, D. F. Walls, A. Aspect, and G. J. Milburn, Phys. Rev. A 33, 4008 (1986). [137] B. L. Schumaker, S. H. Perlmutter, R. M. Shelby, and M. D. Levenson, Phys. Rev. Lett. 58, 357 (1987). [138] J. Xiao, M. Zhao, Y. Wang, and X. Zhang, Nanopho- ton. 6, 1309 (2017). [139] L. Dobusch, S. Schuler, V. Perebeinos, and T. Mueller, Adv. Mater. 29, 1701304 (2017). [140] S. Tongay, J. Suh, C. Ataca, W. Fan, A. Luce, J. S. Kang, J. Liu, C. Ko, R. Raghunathanan, J. Zhou, F. Ogletree, J. Li, J. C. Grossman, and J. Wu, Sci. Rep. 3, 2657 (2013). [141] L. Sun, X. Zhang, F. Liu, Y. Shen, X. Fan, S. Zheng, J. T. L. Thong, Z. Liu, S. A. Yang, and H. Y. Yang, Sci. Rep. 7, 16714 (2017). [142] T. T. Tran, S. Choi, J. A. Scott, Z.-Q. Xu, C. Zheng, G. Seniutinas, A. Bendavid, M. S. Fuhrer, M. Toth, and I. Aharonovich, Adv. Opt. Mater. 5, 1600939 (2017). [143] 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, and T. F. Heinz, Nat. Nanotech. 12 (2017). [144] Y.-J. Chen, J. D. Cain, T. K. Stanev, V. P. Dravid, and N. P. Stern, Nat. Photon. 11, 431 (2017). [145] Z. Ye, T. Cao, K. O'Brien, H. Zhu, X. Yin, Y. Wang, S. G. Louie, and X. Zhang, Nature 513, 214 (2014). [146] D. Y. Qiu, F. H. da Jornada, and S. G. Louie, Phys. Rev. Lett. 111, 216805 (2013). [147] D. Y. Qiu, F. H. da Jornada, and S. G. Louie, Phys. Rev. Lett. 115, 119901 (2015). [148] D. Y. Qiu, F. H. da Jornada, and S. G. Louie, Nano Lett. 17, 4706 (2017). [149] M. Baranowski, A. Surrente, D. K. Maude, M. Ballot- tin, A. A. Mitioglu, P. C. M. Christianen, Y. C. Kung, D. Dumcenco, A. Kis, and P. Plochocka, 2D Mater. 4, 025016 (2017). [150] Z. Sun, J. Gu, A. Ghazaryan, Z. Shotan, C. R. Con- sidine, M. Dollar, B. Chakraborty, X. Liu, P. Ghaemi, S. K´ena-Cohen, and V. M. Menon, Nat. Photon. 11, 491 (2017). [151] S. Dufferwiel, T. P. Lyons, D. D. Solnyshkov, A. A. P. Trichet, F. Withers, S. Schwarz, G. Malpuech, J. M. 19 Smith, K. S. Novoselov, M. S. Skolnick, D. N. Krizhanovskii, and A. I. Tartakovskii, Nat. Photon. 11, 497 (2017). bauer, A. W. Holleitner, M. Kaniber, K. Muller, and J. J. Finley, Sci. Rep. 7, 12383 (2017). [158] L. Gilles and P. L. Knight, Phys. Rev. A 48, 1582 [152] M. Onga, Y. Zhang, T. Ideue, and Y. Iwasa, Nat. Mat. (1993). 16, 1193 (2017). [159] E. Ginossar, Y. Levinson, and S. Levit, Phys. Rev. B [153] O. L. Berman and R. Y. Kezerashvili, Phys. Rev. B 96, 77, 035307 (2008). 094502 (2017). [160] P. Garcia-Fern´andez, L. S. D. L. Terreros, and F. J. [154] K.-D. Park, T. Jiang, G. Clark, X. Xu, and M. B. Bermejo-Barrera, Opt. Acta 33, 945 (1986). Raschke, Nat. Nanotech. 13, 59 (2018). [155] A. Arora, M. Druppel, R. Schmidt, T. Deilmann, R. Schneider, M. R. Molas, P. Marauhn, S. M. de Vas- concellos, M. Potemski, M. Rohlfing, and R. Brats- chitsch, Nat. Commun. 639, 8 (2017). [156] A. Steinhoff, M. Florian, M. Rosner, G. Schonhoff, T. O. Wehling, and F. Jahnke, Nat. Commun. 1166, 8 (2017). [157] J. Wierzbowski, J. Klein, F. Sigger, C. Straubinger, M. Kremser, T. Taniguchi, K. Watanabe, U. Wurst- [161] H. Ezaki, J. Phys. Soc. Jpn. 68, 1562 (1999). [162] R. Loudon, Opt. Commun. 49, 67 (1984). [163] G. S. Agarwal, Opt. Commun. 62, 190 (1987). [164] R. G. Ispasoi, Y. Jin, J. Lee, F. Papadimitrakopoulos, and R. Goodson, Nano Lett. 2, 127 (2002). [165] M. M. Nieto, Quantum Semiclass. Opt. 8, 1061 (1996). [166] M. M. Nieto and D. R. Truax, Fortsch. Phys. 45, 145 (1997). [167] M. Miri and S. Khorasani, Int. J. Opt. Photon. 4, 9 (2010).