Datasets:

kilian-group
/

arxiv-classifier

Dataset card Data Studio Files Files and versions Community

Subset (4)

default · ~99.6k rows (showing the first 120k)

Split (2)

train · ~72.4k rows (showing the first 93.1k)

paper_id stringlengths 9 16	version stringclasses 26 values	yymm stringclasses 311 values	created timestamp[s]	title stringlengths 6 335	secondary_subfield sequencelengths 1 8	abstract stringlengths 25 3.93k	primary_subfield stringclasses 124 values	field stringclasses 20 values	fulltext stringlengths 0 2.84M
1204.6209	2	1204	2012-07-20T16:11:12	Graphene plasmons and retardation: strong light-matter coupling	[ "cond-mat.mes-hall" ]	We study the retardation regime of doped graphene plasmons, given by the nominal crossing of the unretarded plasmon and light-cone. In addition to modifications in the plasmon dispersion relation, retardation implies strong coupling between propagating light and matter, even for homogeneous graphene, which opens up the possibility of efficient plasmonics in simple graphene devices. We exemplify this enhancement in a double-layer configuration that exhibits {\em perfect} (if lossless) light transmissions across a classically forbidden region, providing a simpler analog of the corresponding phenomenon in perforated metal sheets. We also show that (broad) Fabry-P\'erot resonances present without graphene turn into sharply peaked, quasi-discrete modes in the presence of graphene where graphene's response function is given by the typical Fano lineshape.	cond-mat.mes-hall	cond-mat	epl draft 2 1 0 2 l u J 0 2 ] l l a h - s e m . t a m - d n o c [ 2 v 9 0 2 6 . 4 0 2 1 : v i X r a Graphene plasmons and retardation: strong light-matter coupling G. G´omez-Santos and T. Stauber Departamento de F´ısica de la Materia Condensada and Instituto Nicol´as Cabrera, Universidad Aut´onoma de Madrid, E-28049 Madrid, Spain PACS 73.21.Ac -- Multilayers PACS 42.25.Bs -- Wave propagation, transmission and absorption PACS 78.67.Wj -- Optical properties of graphene Abstract -- We study the retardation regime of doped graphene plasmons, given by the nominal crossing of the unretarded plasmon and light-cone. In addition to modifications in the plasmon dispersion relation, retardation implies strong coupling between propagating light and matter, even for homogeneous graphene, which opens up the possibility of efficient plasmonics in simple graphene devices. We exemplify this enhancement in a double-layer configuration that exhibits perfect (if lossless) light transmissions across a classically forbidden region, providing a simpler analog of the corresponding phenomenon in perforated metal sheets. We also show that (broad) Fabry-P´erot resonances present without graphene turn into sharply peaked, quasi-discrete modes in the presence of graphene where graphene's response function is given by the typical Fano lineshape. Introduction. -- In addition to its amazing transport and mechanical properties [1], graphene's optical behavior is also notable. Absorption, for instance, has the exper- imentally observed [2, 3], universal value ≈ πα for light in the visible spectrum, depending on the fine structure constant α, but not on material's properties. Several pro- posals and/or realizations highlight graphene potential in optical and communication technologies, including resis- tive touchscreens of transparent and flexible displays, [4,5], opto-electronic devices such as photodetector [6], broad- band absorber [7], mode-locked laser due to current sat- uration [8], or as a highly efficient fluorescence quencher [9 -- 15]. Doped graphene has attracted much attention recently as a suitable candidate for noble metal's replacement in the active field of plasmonics [16 -- 19]. In addition to their intrinsic importance as a probe for dynamics, graphene plasmons [20 -- 22] (GPs) offer a number of advantages such as frequency tunability, long life-times, and spatial con- finement due shorter (than light) wavelengths. They have been observed via electron energy loss spectroscopy [23,24] and near-field nanoscopy. [25 -- 27] Unfortunately, GPs do not couple easily to propagat- ing electromagnetic modes over most of its range due to the large momentum mismatch, complicating their ma- nipulation. This problem could be overcome by breaking the conservation of parallel momentum either with con- fined geometries [19, 28] or artificial periodicities [29 -- 31]. However, there is a regime where even homogeneous GPs must couple strongly to (propagating) light: the retarda- tion limit. It applies for frequencies below a characteristic crossover scale, which can be taken as the crossing between the nominal, unretarded GP dispersion and the light-cone. There, phenomena associated with strong light-graphene coupling take place. Here, we demonstrate this enhanced light-graphene cou- pling and apply it to a double-layer graphene [32, 33] ar- rangement possessing extraordinary transmission. This term was originally coined to describe the enormous trans- mission experimentally observed through periodically per- forated metal sheets [34 -- 36], a situation where the naive expectation would have prescribed just the opposite. An explanation was provided in terms of the excitation of sur- face plasmons, with the result of enhanced transmission (perfect in the absence of absorption) through a nominally opaque region. In our case, the resonant coherent excita- tions of the graphene layers allow the enhanced transmis- sion of photons through the central, classically forbidden region for photons, in direct analogy with the metallic case. In the propagating region, the strongly quenched transmission turns into perfect transmission at sharply peaked resonances related to the original Fabry-P´erot res- onances, where graphene's response function is given by the typical Fano lineshape due to the nearly discrete na- p-1 G. G´omez-Santos T. Stauber ture of the spectrum in the central slab. The Letter is organized as follows. We first study the retardation limit and its effect on the GP dispersion for single-layer graphene. Then we consider the retarda- tion regime for double-layer graphene separating different dielectric media where extraordinary transmission takes place. Finally we summarize our results and an appendix details the GPs dispersion relation for the double-layer arrangement in the retardation limit. Retardation limit. Single-layer graphene. -- The standard expression for a 2d plasmon, ωp ∝ √q, assumes instantaneous Coulomb coupling between charges. [20, 21] Therefore, it cannot be correct when the nominal plas- mon dispersion meets the light-cone, ωp . cq. This is the regime where retardation effects matter and strong light- graphene interaction takes place, as we now show. Consider a graphene layer sandwiched between two di- electrics with permittivities ε1,2 = ǫ1,2 ε0 and ǫ1 > ǫ2. The longitudinal current response to an (in-plane) longi- tudinal, external vector potential is given by the standard RPA expression χl = χo l 1 − e2dlχo l , (1) where χl = Gjj is the retarded Green function for the longitudinal current, χo its non-interacting (bare) ver- l sion, and dl = GAA that of the (in-plane) longitudi- nal vector potential (in the absence of graphene). For graphene, we use the well-known results [20,21,37] for χo l , whereas for the photon field, one straightforwardly finds [15] dl = q′ 1q′ ε2q′ , with q′ 2 ω−2 1+ε1q′ 2 GPs are the poles of Eq. 1, leading to the following i = pq2 − (ω/c)2ǫi. (implicit) dispersion relation, including retardation Fig. 1: Left panel: Longitudinal plasmon exact dispersion re- lation (black continuous line), compared to the instantaneous approximation (blue dashed line) for doped graphene between medium 1 (ǫ1 = 5) and medium 2 (ǫ2 = 1), light-cones also shown. Right panel: As in left panel for the double-layer graphene showing the in-phase (upper curve, black) and out- of-phase (lower curve, magenta) plasmons for ǫ3 = ǫ1 and layer separation z = zc/2. Inset: schematic geometry and transmis- sion setup. for doping level n ∼ 1012 cm−2, νc ∼ 600 GHz for doping level n ∼ 1013 cm−2, reaching the technologically impor- tant THz regime for n ∼ 1014 cm−2. Therefore, for ω . ωc retardation always matters. Not surprisingly, this linear regime is also the region of strong graphene-light coupling. This is immediately seen by look- ing at the reflection and transmission amplitudes for the (in-plane) longitudinal vector potential upon passing from medium i to j, given by ω2 = e2 1q′ q′ 2 1 + ε1q′ 2 ε2q′ χo l . (2) rij = εiq′ εiq′ j − εjq′ j + εjq′ i + e2q′ iq′ iq′ i − e2q′ jχo jχo l ω−2 l ω−2 . (4) In the unretarded limit, c → ∞, this gives the known square-root dispersion. In contrast, the exact solution re- places this behavior with a linear dispersion which merges with the slower medium (1) light-cone below a characteris- tic crossover frequency, as illustrated in Fig. 1 (left panel), and summarized as follows: ωF (cid:19)2 (cid:18) ω = (cid:16) 4αg kF (cid:17) , ω & ωc ǫ1+ǫ2(cid:17)(cid:16) q kF (cid:17)2 v (cid:1)2(cid:16) q (cid:0) c1 ω . ωc , , (3)   where ωF , kF and v are graphene's Fermi frequency, Fermi momentum and Fermi velocity, c1 = c/√ǫ1 is medium 1 (slower) light velocity, and αg = c v α represents graphene fine structure constant. The crossover between regimes takes place for frequencies which roughly corresponds to the intersection of the unretarded GP and light-cone dis- persions. This scale is given by ωc ∼ αωF for reasonable dielectric constants. Typical frequencies are νc ∼ 200 GHz tij = 1 + rij For interband transitions at frequencies above 2 ωF , graphene terms in Eq. 4 are minute, leading to the univer- sal 2.3% weak absorption in vacuum, for instance. On the other hand, for frequencies below ωc, graphene response starts to dominate in Eq. 4 implying strong radiation- graphene coupling. For instance, in the limit ω ≪ ωc the reflection amplitude becomes r = −1 + O( ω ωc ). (5) r = −1 implies perfect reflection or, equivalently, a zero photon field at graphene position as a boundary condi- tion. Therefore, the (small) parameter ω of our strong coupling regime also measures the departure from this zero-field boundary condition. ωc ∼ ω αωF Double-layer graphene. Extraordinary transmis- sion. -- We consider two identically doped graphene p-2 sheets, separating three dielectrics with permittivities ǫ1,2,3, see Fig. 1 (right panel), and impose left-right sym- metry choosing ǫ2 < ǫ1 = ǫ3. Under these conditions, for large enough incident angles the central region does not support propagating modes: light must tunnel through this evanescent region. Doped graphene enhanced re- sponse in the retardation regime can make this tunneling perfect, as we will show. Reflection and transmission coefficients can be obtained from the corresponding amplitudes, written as r1,3 = r12 + t12r23t21e−2q′ 2z 1 − r21r23e−2q′ 2z t1,3 = t12t23e−q′ 2z 1 − r21r23e−2q′ 2z , (6) for light incoming from medium 1 and being transmitted to medium 3 (see Fig. 1 inset), with tij and rij taken from Eq. 4. In order to display the connection between light propagation and graphene behavior, we also calculate the double-layer (longitudinal) current-current response, given by the following matrix generalization of Eq. 1 (7) χl = (1 − e2χo l dl)−1χo l , 1, χo l = diag(χo 1,3), d22 = d3 (1 + ro 3,1 = d21. Here, d1 = q′ with the 2x2 matrix χo 2) representing the non-interacting (longitudinal) response of graphene layers 1 and 2, whereas the photonic matrix d = (dij ) is given by d11 = d1 (1 + ro 3,1), and d12 = 1,3 = d3 to d1 to 2ε3ω2 , and i,j(to ro i,j ) correspond to the expressions of Eq. 6 evaluated for the dielectric geometry of Fig. 1, but without graphene layers (χo l = 0 in Eq. 4). For our left-right symmetric arrangement, the in-phase and out-of-phase components of graphene currents, j± = j1 ± j2, diagonalize χl, with corresponding (complex) eigenvalues that we denote by χ++ and χ−−, respectively. 2ε1ω2 , d3 = q′ 1 3 Genuine (non radiative) GPs correspond to the poles of χl in the totally evanescent regime, that is, the region to the right of the slower light-cone in Fig. 1. In the Appendix we show that there always are two GPs, one for each component of the response. The in-phase GP (χ++) merges with the corresponding limiting light-cone, whereas the out-of-phase GP (χ−−) either merges with the light-cone or develops its own linear dispersion at slower velocity if the separation between layers is below a critical value zc = zc kF , with zc = ǫ2 c 2 v α . Both are shown in Fig. 1 (right panel) where we have used z = zc/2, and permittivities ǫ2 = 1 (vacuum) and ǫ1 = ǫ3 = 5, a representative value for common substrates. For bulk BN, we have, e.g., ǫ1,3 ∼ 5, whereas ǫ1,3 ∼ 4 for few layer BN or SiO2. Though GPs remain in the evanescent region of both media, the presence of graphene also strongly affects prop- agating light, leading to resonances with perfect transmis- sion. At first glance, this statement seems incompatible ǫ1−ǫ2 Graphene plasmons and retardation n o i s s i m s n a r T ) . u . a ( y t i s n e D l a r t c e p S 1 0.8 0.6 0.4 0.2 0 0.6 0.4 0.2 graphene present graphene absent -Im χ+ + -Im χ− − 0 0 0.1 0.2 sin θ 0.3 i 0.4 0.5 Fig. 2: Top panel: Transmission as a function of incoming angle for frequency ω = (α/2) ωF and double-layer graphene separation zkF = 20 c/(v α) for permittivities ǫ1,3 = 5, ǫ2 = 1. Dashed (black) line: same as before in the absence of graphene. Bottom panel: Spectral density of the in-phase (solid, blue line) and out-of-phase (dashed, red line) double-layer graphene response. with the (almost) perfect reflective character of a graphene layer in our strong coupling regime. Yet, the zero field boundary condition also decouples the photon field in the central slab from the propagating modes outside, leading to the standard discrete spectrum for a particle in a box. Now, the small parameter ω measuring the depar- ture from perfect reflection is also a measure of the (small) coupling between the discrete modes of the central slab and the propagating modes outside, leading to (radiative) resonances. Drastic changes are expected close to these resonances and, indeed, we will see that the otherwise ex- pected strong reflection turns into perfect transmission. ωc ∼ ω αωF These resonances are shown in Fig. 2 (top panel), where we present the transmission for layer's separation z kF = 20 c/(vα) and frequency ω = (α/2) ωF . For a doping level of n ∼ 1012cm−2, this would imply a fre- quency ν ∼ 100 GHz, and separation z ∼ 5 × 10−3m. For n ∼ 1014cm−2, we have ν ∼ 1 THz, and z ∼ 5 × 10−4m. Although the existence of perfect transmission at Fabry- P´erot-like resonances is generic for this geometry even without graphene (see dotted line in top panel of Fig.2), these resonances are sharply modified by its presence. The otherwise strongly reflective graphene confines the trans- mission to narrow resonances close to the region where the almost discrete central slab modes are excited. The narrow width of these resonances is a measure of the long lifetime of the central photonic modes, weakly decaying into the radiative modes outside. A direct evidence of the role of these (almost) discrete spectrum is provided by the spectral densities of the in-phase (χ++) and out-of-phase (χ−−) graphene response, also shown in Fig. 2 (bottom panel). Up to an overall scale, these spectral densities co- p-3 G. G´omez-Santos T. Stauber incide with the spectral densities of the photonic field and, therefore, the sharp resonances in graphene match the ex- pected sharp spectral densities of weakly decaying, almost discrete photonic modes. Notice the one-to-one correspon- dence between response and transmission peaks, empha- sizing the role of these modes in overturning the otherwise generic strong reflection of graphene in the strong coupling limit. The spectral functions of the bottom panel of Fig. 2 exhibit a marked asymmetric Fano-like profile, vanishing precisely where grapheneless transmission is one. This Fano-like profile can be expected, given the key role played by the almost discrete modes of the central slab. It can be quantitatively explained noticing that χ±± ∼ 1 l (d11 ± d12) 1 − e2χo , (8) but the photon propagator in the absence of graphene van- ishes at the corresponding grapheneless resonances as d11 ± d12 ∼ a(δs) + ib(δs)2, (9) where a and b are real constants, and δs is the depar- ture of sin(θi) from the value corresponding to the dis- crete modes upon strictly enforcing zero-field boundary conditions, given by exp(−2q′ 2z) = 1. Notice that this zero-field boundary condition for discrete modes in the presence of graphene coincides with the perfect transmis- sion condition in the absence of graphene, explaining the vicinity of both points. The small shift of the actual solu- tion being the consequence of the small leaking of dis- crete modes into the outside continuum. This leaking also accounts for the small width of the resonance, and leads to a characteristic resonance asymmetry, given by is the res- −Imχ±± ∼ onance shift from the nominal discrete modes graphene, playing the role of small parameter in the strong coupling regime, and l being (up to a factor of order one) the central slab width in units of the minimum required for the emer- gence of the first nominal discrete mode. This expression can be approximately recast in the more common Fano lineshape [38] , (δs−s∗)2+l2(δs)4 , where s∗ ∼ ω/ωF (δs)2 αl − Imχ±± ∼ (Qγ/2 + δs′)2 (δs′)2 + (γ/2)2 , (10) with the identifications δs′ = δs − s∗, the width γ ∼ ls2 ∗, and the Fano asymmetry parameter Q = 2s∗ γ with value Q ∼ 2 − 3, for fig. 2 (bottom panel). curs for sin θi > 1/√5 ∼ 0.447, that is, in the evanes- Interestingly, the rightmost resonance in Fig. In fact, we can tune the cent regime for medium 2. arrangement to make this last resonance the only one present. This is shown in Fig. 3, where we present the transmission calculated with the same frequency as before ω = (α/2)ωF , but for three shorter separations given by z1 kF = 2 c/(vα), z2 = z1/5, and z3 = z1/10 2 oc- z1kF = 2 c / (v α) z2 = z1 / 5 z3 = z1 / 10 1 0.8 0.6 i i n o s s m s n a r T 0.4 0.2 0 0 0.2 0.4 sin θ 0.6 i 0.8 1 Fig. 3: Transmission as a function of incoming angle for fre- quency ω = (α/2) ωF , permittivities ǫ1,3 = 5, ǫ2 = 1, and double-layer graphene separations: z1kF = 2 c/(v α) (black), z2 = z1/5 (red), and z3 = z1/10 (blue). The vertical dashed line marks the onset of the evanescent regime in the central slab. (ν ∼ 100 GHz and z1 ∼ 5 × 10−4m for n ∼ 1012 cm−2). One can observe a perfect transmission peak within the classically forbidden region for photons. This resonance is rather sharp close to the propagating boundary, moving deeper into the evanescent region and broadening with de- creasing double-layer separation. Notice that, for the cases shown in Fig. 3, the transparency window is virtually con- fined to the evanescent regime. Although a finite photon tunneling always exists, the fact that the transmission be- comes perfect in the evanescent region [39] is entirely due to the presence of doped graphene sheets, and is associ- ated with a resonance in their out-of-phase response (see bottom panel of Fig. 2). This perfect transmission peak within the evanescent region extends all the way down to zero frequency, as shown in the top left panel of Fig. 4. For higher frequencies, this resonance approaches the upper light-cone and merges with the lowest propagat- ing Fabry-P´erot like resonance, as seen in the bottom left panel of Fig. 4. Our aim so far has been to expose the strong light- graphene coupling of the retardation regime in a sim- ple manner. Therefore, we have ignored absorption. Graphene losses will degrade the perfect nature of trans- mission, making the possible observation of coherent ef- fects more likely in reflection rather than transmission. Although we are not aware of any intrinsic limitation in the minimization of losses attainable in doped graphene, the low frequencies of the strong-coupling regime might pose a severe handicap. Nevertheless, absorption opens new possibilities: the sharp response of graphene in the ab- sence of absorption suggests a correspondingly enhanced absorption when losses are allowed. This is the case as the upper right panel of Fig. 4 shows, where the ab- p-4 Graphene plasmons and retardation response function is given by the typical Fano lineshape. ii) We also discussed that the transmission, perfect if losses are ignored, can be confined to the classically forbidden region for photons between graphene layers, providing a direct analog of the extraordinary transmission through perforated metals in a conceptually simpler setup. iii) Finally, we showed that enhanced transmission becomes increased absorption when losses are allowed, displaying non-monotonous behavior. Graphene's doping tunability in this or similar setups should thus open up new ways to efficiently control the flow and/or absorption of radiation. ∗ ∗ ∗ This work has been supported by FCT via grant PTDC/FIS/101434/2008 and by MICINN via grant FIS2010-21883-C02-02. Appendix. -- We outline here the calculation of double-layer GPs in the retarded regime. We take ǫ3 = ǫ1 > ǫ2. Therefore, genuine GPs are the response poles located to the right of the slower (outer media) light-cone. These are given by the solutions of 1 ∓ r21e−q′ 2z = 0, (11) corresponding to the in-phase (χ++) and out-of-phase (χ−−) response components, respectively. Taking r21 from Eq. 4 deep into the retarded regime, one finds the in-phase GP as ǫ1 ω2 + → 4παg χo q′ 1, (12) ω = where we have used dimensionless magnitudes: i = c2/ǫi, q = q/kF , ω/ωF , αg = cα, c = c/v, c2 i = pq2 − ω2/c2 q′ i , with single-layer graphene response, χo = π−1, taken in the local approximation, excellent for the retarded regime. In this limit, ω+ coincides with the GP pole of an isolated graphene layer in the boundary be- tween two (semi-infinite) media 1 and 2. Therefore, one easily sees that it merges asymptotically with the outer media light-cone, as in Eq. 3: ω+ = c1 q + O(q3). (13) Analyzing the plus Eq. 11 for the out-of-phase GP, a critical layer separation emerges, zc, given by zc kF = zc = 2(ǫ1 − ǫ2) that separates the following two regimes ǫ2 c α , (14) We conclude that the out-of-phase GP develops a linear dispersion relation that either merges with the outer media light-cone (and, therefore, with ω+ from below) for z > zc, or shows a lower velocity v− < c1 for z < zc, as stated in the paper. p-5 Fig. 4: Top left panel: Transmission in the ω −q plane showing the perfect resonance within the evanescent region for double- layer separation corresponding to z2kF = 2 c/(5v α). Bottom left panel: As in top left panel for an extended region of the ω − q plane to include additional Fabry-P´erot-like resonances. Top right panel: Absorption when graphene losses are included with ωτ = 0.65 for the same double-layer separations (and colors) as in Fig. 3. Bottom right panel: Absorption for fixed double-layer separation z1kF = 2 c/(v α) and graphene losses ωτ = 2 (black), ωτ = 0.5 (red), and ωτ = 0.05 (blue). sorption for a lifetime τ ∼ 10−12s, present in suspended graphene and corresponding to ωτ = 0.65, is plotted for the same double-layer separations of Fig. 3. Indeed, one observes an overall high absorption, reaching values above 90% precisely in the evanescent region. From this perspec- tive, our arrangement is closely connected to that of Ref. [40], providing a complementary and potentially simpler alternative to absorption enhancement based on periodic patterning [29 -- 31]. It is similar, though, to the previously suggested enhanced absorption of graphene placed in a (double) Fabry-P´erot cavity. [41, 42] In the bottom right panel of Fig. 4, the absorp- tion for different intrinsic damping rates corresponding to ωτ = 2, 0.5, 0.05 is shown where the last value would assume a phenomenological relaxation rate γ of the po- larization with γ ∼ 10meV, present in graphene on a SiO2-substrate. Notice that the double-layer absorption can display a non-monotonous behavior with the intrinsic single-layer graphene losses, providing a richer degree of control. Summary. -- We have studied the retardation regime of doped graphene plasmons, given by (ω/ωF ) . α. Apart from modifying the unretarded GP dispersion behavior from √q to q, such a limit marks the onset of strong radiation-graphene coupling and we have exhibited such enhancement in the simplest double-layer arrangement. i) We first observed that the transmission is strongly quenched except at sharply peaked resonances, consistent with the nearly discrete spectrum of the strongly confined field in the central slab brought about by graphene, whose where ω− = (cid:26) c1 q + O(q3), v− q + O(q3), c2 1 1 + (1 − ǫ2 v2 − = ǫ1 z > zc z < zc , (15) . (16) )( zc z − 1) G. G´omez-Santos T. Stauber REFERENCES [1] Geim A. K., Science, 324 (2009) 1530. [2] Nair R. R., Blake P., Grigorenko A. N., Novoselov K. S., Booth T. J., Stauber T., Peres N. M. R. and Geim A. K., Science , 320 (2008) 1308. [3] Mak K. F., Sfeir M. Y., Wu Y., Lui C. H., Misewich J. A. and Heinz T. F., Phys. Rev. Lett., 101 (2008) 196405. [4] Bae S. et al., Nature Nanotechnology, 5 (2010) 574. [5] Bonaccorso F., Sun Z., Hasan T. and Ferrari A. C., Nat. Photonics, 4 (2010) 611. S., Koppens F. H. L. and Garc´ıa de Abajo F. J., ACS Nano, 6 (2012) 431. [29] Ju L., Geng B., Horng J., Girit C., Martin M., Hao Z., Bechtel H. A., Liang X., Zettl A., Shen Y. R. and Wang F., Nat. Nanotechnol., 6 (2011) 630. [30] Echtermeyer T., Britnell L., Jasnos P., Lombardo A., Gorbachev R., Grigorenko A., Geim A., Fer- rari A. and Novoselov K., Nat. Comm., 2 (2011) 458. [31] Thongrattanasiri S., Koppens F. H. L. and Garc´ıa de Abajo F. J., Phys. Rev. Lett., 108 (2012) 047401. [32] Hwang E. H. and Sarma S. D., Phys. Rev. B, 80 (2009) [6] Mueller T., Xia F. and Avouris P., Nature Photonics, 205405. 4 (2010) 297. [7] Liu M. et al., Nature, 474 (2011) 64. [8] Sun Z. et al., ACS Nano, 4 (2010) 803. [9] Treossi E., Melucci M., Liscio A., Gazzano M., Samori P. and Palermo V., J. Am. Chem. Soc., 131 (2009) 15576. [10] Kim J., Cote L. J., Kim F. and Huang J., J. Am. Chem. Soc., 132 (2010) 260. [11] Chen Z., Berciaud S., Nuckolls C., Heinz T. F. and Brus L. E., ACS Nano, 4 (2010) 2964. [33] Stauber T. and G´omez-Santos G., Phys. Rev. B, 85 (2012) 075410. [34] Ebbesen T. W., Lezec H. J., Ghaemi H. F., Thio T. and Wolff P. A., Nature, 391 (1998) 667. [35] Mart´ın-Moreno L., Garc´ıa-Vidal F. J., Lezec H. J., Pellerin K. M., Thio T., Pendry J. B. and Ebbesen T. W., Phys. Rev. Lett., 86 (2001) 1114. [36] Garcia-Vidal F. J., Martin-Moreno L., Ebbesen T. W. and Kuipers L., Rev. Mod. Phys., 82 (2010) 729. [37] Stauber T. and G´omez-Santos G., Phys. Rev. B , 82 [12] Loh K. P., Bao Q., Eda G. and Chhowalla M., Nat. (2010) 155412. Chem., 2 (2010) 1015. [13] Swathi R. S. and Sebastian K. L., J. Chem. Phys., 130 (2009) 086101. [14] Velizhanin K. A. and Efimov A., Phys. Rev. B, 84 (2011) 085401. [38] Fano U., Phys. Rev., 124 (1961) 1866. [39] Garc´ıa de Abajo F. J., G´omez-Santos G., Blanco L. A., Borisov A. G. and Shabanov S. V., Phys. Rev. Lett., 95 (2005) 067403. [40] Bludov Y. V., Vasilevskiy M. I. and Peres N. M. R., [15] G´omez-Santos G. and Stauber T., Phys. Rev. B , 84 Europhys. Lett., 92 (2010) 68001. [41] Furchi M., Urich A., Pospischil P., Lilley G., Un- terrainer K., Detz H., Klang P., Andrews A. M., Schrenk W., Strasser G. and Muller T., Nano Lett., 12 (2012) 2773. [42] Ferreira A., Peres N. M. R., Ribeiro R. M. and Stauber T., Phys. Rev. B, 85 (2012) 115438. (2011) 165438. [16] Jablan M., Buljan H. and Soljaci´c M., Phys. Rev. B, 80 (2009) 245435. [17] Vakil A. and Engheta N., Science, 332 (2011) 1291. [18] Koppens F. H. L., Chang D. E. and Garc´ıa de Abajo F. J., Nano Lett., 11 (2011) 3370. [19] Nikitin A. Y., Guinea F., Garc´ıa-Vidal F. J. and Mart´ın-Moreno L., Phys. Rev. B, 84 (2011) 161407(R). [20] Wunsch B., Stauber T., Sols F. and Guinea F., New J. Phys., 8 (2006) 318. [21] Hwang E. H. and Sarma S. D., Phys. Rev. B, 75 (2007) 205418. [22] Barlas Y., Pereg-Barnea T., Polini M., Asgari R. and MacDonald A. H., Phys. Rev. Lett., 98 (2007) 236601. [23] Liu Y., Willis R. F., Emtsev K. V. and Seyller T., Phys. Rev. B, 78 (2008) 201403. [24] Tegenkamp C., Pfnur H., Langer T., , Baringhaus J. and Schumacher H. W., J. Phys. Condens. Matter, 23 (2011) 012001. [25] Fei Z. et al., Nano Lett., 1 (2011) 470. [26] Chen J., Badioli M., Alonso-Gonz´alez P., Thon- grattanasiri S., Huth F., Osmond J., Spasenovic M., Centeno A., Pesquera A., Godignon P., Zuru- tuza A., Camara N., Garcia de Abajo J., Hillen- brand R. and Koppens F., Nature, 487 (2012) 77. [27] Fei Z., Rodin A. S., Andreev G. O., Bao W., McLeod A. S., Wagner M., Zhang L. M., Zhao Z., Dominguez G., Thiemens M., Fogler M. M., Castro-Neto A. H., Lau C. N., Keilmann F. and Basov D. N., Nature, 487 (2012) 82. [28] Christensen J., Manjavacas A., Thongrattanasiri p-6
1702.08408	2	1702	2017-04-27T19:56:20	Current Induced Damping of Nanosized Quantum Moments in the Presence of Spin-Orbit Interaction	[ "cond-mat.mes-hall" ]	Motivated by the need to understand current-induced magnetization dynamics at the nanoscale, we have developed a formalism, within the framework of Keldysh Green function approach, to study the current-induced dynamics of a ferromagnetic (FM) nanoisland overlayer on a spin-orbit-coupling (SOC) Rashba plane. In contrast to the commonly employed classical micromagnetic LLG simulations the magnetic moments of the FM are treated {\it quantum mechanically}. We obtain the density matrix of the whole system consisting of conduction electrons entangled with the local magnetic moments and calculate the effective damping rate of the FM. We investigate two opposite limiting regimes of FM dynamics: (1) The precessional regime where the magnetic anisotropy energy (MAE) and precessional frequency are smaller than the exchange interactions, and (2) The local spin-flip regime where the MAE and precessional frequency are comparable to the exchange interactions. In the former case, we show that due to the finite size of the FM domain, the \textquotedblleft Gilbert damping\textquotedblright does not diverge in the ballistic electron transport regime, in sharp contrast to Kambersky's breathing Fermi surface theory for damping in metallic FMs. In the latter case, we show that above a critical bias the excited conduction electrons can switch the local spin moments resulting in demagnetization and reversal of the magnetization. Furthermore, our calculations show that the bias-induced antidamping efficiency in the local spin-flip regime is much higher than that in the rotational excitation regime.	cond-mat.mes-hall	cond-mat	Current Induced Damping of Nanosized Quantum Moments in the Presence of Spin-Orbit Interaction Farzad Mahfouzi∗ and Nicholas Kioussis† Department of Physics and Astronomy, California State University, Northridge, CA, USA (Dated: July 17, 2018) Motivated by the need to understand current-induced magnetization dynamics at the nanoscale, we have developed a formalism, within the framework of Keldysh Green function approach, to study the current-induced dynamics of a ferromagnetic (FM) nanoisland overlayer on a spin-orbit-coupling (SOC) Rashba plane. In contrast to the commonly employed classical micromagnetic LLG simula- tions the magnetic moments of the FM are treated quantum mechanically. We obtain the density matrix of the whole system consisting of conduction electrons entangled with the local magnetic moments and calculate the effective damping rate of the FM. We investigate two opposite limiting regimes of FM dynamics: (1) The precessional regime where the magnetic anisotropy energy (MAE) and precessional frequency are smaller than the exchange interactions, and (2) The local spin-flip regime where the MAE and precessional frequency are comparable to the exchange interactions. In the former case, we show that due to the finite size of the FM domain, the "Gilbert damping"does not diverge in the ballistic electron transport regime, in sharp contrast to Kambersky's breathing Fermi surface theory for damping in metallic FMs. In the latter case, we show that above a critical bias the excited conduction electrons can switch the local spin moments resulting in demagnetization and reversal of the magnetization. Furthermore, our calculations show that the bias-induced anti- damping efficiency in the local spin-flip regime is much higher than that in the rotational excitation regime. PACS numbers: 72.25.Mk, 75.70.Tj, 85.75.-d, 72.10.Bg I. INTRODUCTION Understanding the current-induced magnetization switching (CIMS) at the nanoscale is mandatory for the scalability of non-volatile magnetic random access mem- ory (MRAM) of the next-generation miniaturized spin- tronic devices. However, the local magnetic moments of a nanoisland require quantum mechanical treatment rather than the classical treatment of magnetization commonly employed in micromagnetic simulations, which is the cen- tral theme of this work. The first approach of CIMS employs the spin transfer torque (STT)1,2 in magnetic tunnel junctions (MTJ) con- sisting of two ferromagnetic (FM) layers (i.e., a switch- able free layer and a fixed layer) separated by an insulat- ing layer, which involves spin-angular-momentum trans- fer from conduction electrons to local magnetization3,4. Although STT has proven very successful and brings the precious benefit of improved scalability, it requires high current densities (≥ 1010 A/cm2) that are uncomfort- ably high for the MTJ's involved and hence high power consumption. The second approach involves an in-plane current in a ferromagnet-heavy-metal bilayer where the magnetization switching is through the so-called spin- orbit torque (SOT) for both out-of-plane and in-plane magnetized layers.5–8 The most attractive feature of the SO-STT method is that the current does not flow through the tunnel barrier, thus offering potentially faster and more efficient magnetization switching compared to the MTJs counterparts. As in the case of STT, the SO-STT has two compo- nents: a field-like and an antidamping component. While the field-like component reorients the equilibrium direc- tion of the FM, the antidamping component provides the energy necessary for the FM dynamics by either enhanc- ing or decreasing the damping rate of the FM depending on the direction of the current relative to the magneti- zation orientation as well as the structural asymmetry of the material. For sufficiently large bias the SOT can overcome the intrinsic damping of the FM leading to ex- citation of the magnetization precession.8 The underlying mechanism of the SOT for both out-of-plane and in-plane magnetized layers remains elusive and is still under de- bate. It results from either the bulk Spin Hall Effect (SHE)9–12, or the interfacial Rashba-type spin-orbit cou- pling,13–16 or both17–19. Motivated by the necessity of scaling down the size of magnetic bits and increasing the switching speed, the objective of this work is to develop a fully quantum me- chanical formalism, based on the Keldysh Green function (GF) approach, to study the current-induced local mo- ment dynamics of a bilayer consisting of a FM overlayer on a SOC Rashba plane, shown in Fig. 1. Unlike the commonly used approaches to investigate the magnetization dynamics of quantum FMs, such as the master equation20, the scattering21 or quasi-classical22 methods, our formalism allows the study of magnetiza- tion dynamics in the presence of nonequilibrium flow of electrons. We consider two different regimes of FM dynamics: In the first case, which we refer to as the single domain dynamics, the MAE and the precession frequency are smaller than the exchange interactions, and the FM can be described by a single quantum magnetic moment, of a typically large spin, S, whose dynamics are governed 2 into account the collective excitations. Furthermore, Ed- wards points out26 the necessity of including the effect of long-range Coulomb interaction in calculating damping for large SOC. In the second case, which corresponds to an indepen- dent local moment dynamics, the FM has a large MAE and hence the rotational excitation energy is compara- ble to the local spin-flip excitation (exchange energy). We investigate the effect of bias on the damping rate of the local spin moments. We show that above a criti- cal bias voltage the flowing conduction electrons can ex- cite (switch) the local spin moments resulting in demag- netization and reversal of the magnetization. Further- more, we find that, in sharp contrast to the single do- main precessional dynamic, the current-induced damping is nonzero for in-plane and out-of-plane directions of the equilibrium magnetization. The bias-induced antidamp- ing efficiency in the local moment switching regime is much higher than that in the single domain precessional dynamics. The paper is organized as follows. In Sec. II we present the Keldysh formalism for the density matrix of the en- tagled quantum moment-conduction electron system and the effective dampin/antdamping torque. In Sec. III we present results for the current-induced damping rate in the single domain regime. In Sec. IV we present results for the current-induced damping rate in the independent local regime. We conclude in Sec. V. II. THEORETICAL FORMALISM Fig. 1 shows a schematic view of the ferromagnetic heterostructure under investigation consisting of a 2D ferromagnet-Rashba plane bilayer attached to two semi- infinite normal (N) leads whose chemical potentials are shifted by the external bias, Vbias. The magnetization of the FM precesses around the axis specified by the unit vector, (cid:126)nM , with frequency ω and cone angle θ. The FM has length, LF M , along the transport direction. The total Hamiltonian describing the coupled conduc- tion electron-localized spin moment system in the het- erostructure in Fig. 1 can be written as, x FIG. 1: (Color online) Schematic view of the FM/Rashba plane bilayer where the FM overlayer has length Lx and is infinite (finite) along the y-direction for the case of a single domain (nano-island) discussed in Sec. III (IV). The magne- tization, (cid:126)m, of the FM precesses around the direction denoted by the unit vector, (cid:126)nM , with frequency ω and cone angle, θ. The Rashba layer is attached to two normal (N) leads which are semi-infinite along the x-direction, across which an exter- nal bias voltage, V , is applied. mainly by the quantized rotational modes of the magne- tization. We show that the magnetic degrees of freedom entering the density matrix of the conduction electron- local moment entagled system simply shift the chemical potential of the Fermi-Dirac distribution function by the rotational excitations energies of the FM from its ground state. We also demonstrate that the effective damping rate is simply the net current along the the auxiliary m- direction, where m= -S, -S+1, . . ., +S, are the eigenval- ues of the total Sz of the FM. Our results for the change of the damping rate due to the presence of a bias volt- age are consistent with the anti-damping SOT of clas- sical magnetic moments,16,23, where due to the Rashba spin momentum locking, the anti-damping SOT, to low- est order in magnetic exchange coupling, is of the form, (cid:126)m × ( (cid:126)m × y), where y is an in-plane unit vector normal to the transport direction. In the adiabatic and ballistic transport regimes due to the finite S value of the nanosize ferromagnet our formal- ism yields a finite "Gilbert damping", in sharp contrast to Kambersky's breathing Fermi surface theory for damp- ing in metallic FMs.24 On the other hand, Costa and Muniz 25 and Edwards26 demonstrated that the prob- lem of divergent Gilbert damping is removed by taking (cid:88) rr(cid:48),σσ(cid:48) T r{sd} Htot = (cid:104)(cid:16) 1s H σσ(cid:48) rr(cid:48) + δrr(cid:48)δσσ(cid:48)1sµr + δrr(cid:48)Jsd(cid:126)σσσ(cid:48) · (cid:126)sd(r) + δσσ(cid:48)δrr(cid:48)HM (cid:17) ψ∗ {s(cid:48) d}r(cid:48)σ(cid:48)ψ{sd}rσ (cid:105) . (1) Here, (cid:126)sd(r) is the local spin moment at atomic position r, the trace is over the different configurations of the lo- cal spin moments, {sd}, ψ{sd}rσ = {sd}(cid:105) ⊗ ψe rσ is the quasi-particle wave-function associated with the conduc- tion electron (ψe) entangled to the FM states ({sd}(cid:105)), Jsd is the s−d exchange interaction, 1s is the identity ma- trix in spin configuration space, and σx,y,z are the Pauli matrices. We use the convention that, except for r, bold symbols represent operators in the magnetic configura- tion space and symbols with hat represent operators in the single particle Hilbert space of the conduction elec- trons. The magnetic Hamiltonian HM is given by HM = − gµB (cid:126)Bext(r) · (cid:126)sd(r) (cid:88) (2) − (cid:88) (cid:104)r,r(cid:48)(cid:105) r J dd rr(cid:48) s2 d (cid:126)sd(r(cid:48)) · (cid:126)sd(r) −(cid:88) (cid:126)sc(r) · (cid:126)sd(r), Jsd sd r where, the first term is the Zeeman energy due to the external magnetic field, the second term is the magnetic coupling between the local moments and the third term is the energy associated with the intrinsic magnetic field acting on the local moment, (cid:126)sd(r), induced by the local spin of the conduction electrons, (cid:126)sc(r). The Rashba model of a two-dimensional electron gas with spin orbit coupling interacting with a system of localized magnetic moments has been extensively em- ployed14,27,28 to describe the effect of enhanced spin-orbit coupling solely at the interface on the current-induced torques in ultrathin ferromagnetic (FM)/heavy metal (HM) bilayers. The effects of (i) the ferromagnet induc- ing a moment in the HM and (ii) the HM with strong spin-orbit coupling inducing a large spin-orbit effect in the ferromagnet (Rashba spin-orbit coupling) lead to a thin layer where the magnetism and the spin-orbit cou- pling coexist.27 The single-electron tight-binding Hamiltonian29 for the conduction electrons of the 2D Rashba plane, H σσ(cid:48) rr(cid:48) which is finite along the transport direction x and infinite along the y direction is of the form, xx(cid:48) (kya) = [t cos(kya)δσσ(cid:48) − tso sin(kya)σx H σσ(cid:48) σσ(cid:48)]δxx(cid:48) (3) + t(δx,x(cid:48)+1 + δx+1,x(cid:48))δσσ(cid:48) + itso(δx,x(cid:48)+1 − δx+1,x(cid:48))σy σσ(cid:48). Here, x, x(cid:48) denote atomic coordinates along the trans- port direction, a is the in-plane lattice constant, and tso is the Rashba SOI strength. The values of the local effec- tive exchange interaction, Jsd= 1eV, and of the nearest- neighbor hopping matrix element, t=1 eV, represent a realistic choice for simulating the exchange interaction of 3d ferromagnetic transition metals and their alloys (Fe, Co).30–32 The Fermi energy, EF =3.1 eV, is about 1 eV below the upper band edge at 4 eV consistent with the ab initio calculations of the (111) Pt surface33. Further- more, we have used tso=0.5 eV which yields a Rashba parameter, αR = tsoa ≈1.4 eVA (a=2.77 A is the in- plane lattice constant of the (111) Pt surface) consis- tent with the experimental value of about 1-1.5 eVA34 and the ab initio value of 1 eVA28. However, because other experimental measurements for Pt/Co/Pt stacks report35 a Rashba parameter which is an order of mag- nitude smaller, in Fig.3 we show the damping rate for different values of the Rashba SOI.. For the results in Sec. IV, we assume a real space tight binding for propa- gation along y-axis. 3 (cid:18) The single particle propagator of the coupled electron- spin system is determined from the equation of motion of the retarded Green function, E − iη − µ − H − HM − Jsd 2 (cid:126)σ · (cid:126)sd Gr(E) = 1, (4) (cid:19) where, η is the broadening of the conduction electron states due to inelastic scattering from defects and/or phonons, and for simplicity we ignore writing the identity matrices 1 and 1 in the expression. The density matrix of the entire system consisting of the noninteracting elec- trons (fermionic quasi-particles) and the local magnetic spins is determined (see Appendix A for details of the derivation for a single FM domain) from the expression, ρ = Gr(E)ηf (E − µ − HM ) Ga(E). (5) (cid:90) dE π It is important to emphasize that Eq. (5) is the central result of this formalism which demonstrates that the ef- fect of the local magnetic degrees of freedom is to shift the chemical potential of the Fermi-Dirac distribution func- tion by the eigenvalues, εm, of HMm(cid:105) = εmm(cid:105), i.e., the excitation energies of the FM from its ground state. Here, m(cid:105) are the eigenstates of the Heisenberg model describing the FM. The density matrix can then be used to calculate the local spin density operator of the con- duction electrons, [(cid:126)sc(r)]mm(cid:48) ss(cid:48),rr(cid:126)σss(cid:48)/2, which along with Eqs. (2), (4), and (5) form a closed set of equations that can be solved self consistently. Since, the objective of this work is the damping/anti-damping (transitional) behavior of the FM in the presence of bias voltage, we only present results for the first iteration. =(cid:80) ss(cid:48) ρmm(cid:48) Eq. (5) shows that the underlying mechanism of the damping phenomenon is the flow of conduction electrons from states of higher chemical potential to those of lower one where the FM state relaxes to its ground state by transferring energy to the conduction electrons. There- fore, the FM dynamical properties in this formalism is completely governed by its coupling to the conduction electrons, where conservation of energy and angular mo- mentum dictates the excitations as well as the fluctua- tions of the FM sate through the Fermi distribution func- tion of the electrons coupled to the reservoirs. This is different from the conventional Boltzmann distribution function which is commonly used to investigate the ther- mal and quantum fluctuations of the magnetization. Due to the fact that the number of magnetic configura- tions (i.e. size of the FM Hilbert space) grows exponen- tially with the dimension of the system it becomes pro- hibitively expensive to consider all possible eigenstates of the HM operator. Thus, in the following sections we consider two opposite limiting cases of magnetic config- urations. In the first case we assume a single magnetic moment for the whole FM which is valid for small FMs with strong exchange coupling between local moments and small MAE. In this case the dynamics is mainly gov- erned by the FM rotational modes and local spin flips can 4 states, S, m(cid:105), of the total angular momentum Sz, with eigenvalues mω = −Sω, . . . , +Sω, where ω = Bz is the Larmor frequency. Thus, the wave function of the cou- pled electron-spin configuration system, shown schemat- ically in Fig. 2 is of the form, ψms(cid:48)r(t) = S, m(cid:105)⊗ ψs(cid:48)r(t). One can see that the magnetic degrees of freedom corre- sponding to the different eigenstates of the Sz operator, enters as an additional auxiliary dimension for the elec- tronic system where the variation of the magnetic energy, (cid:104)S, mHMS, m(cid:105) = mω, shifts the chemical potentials of the electrons along this dimension. The gradient of the chemical potential along the auxiliary direction, is the Larmor frequency (µeV ≈ GHz) which appears as an effective "electric field"in that direction. Substituting Eq (5) in Eq (A1)(b) and averaging over one precession period we find that the average rate of angular momentum loss/gain, which we refer to as the effective "damping rate"per magnetic moment, can be written as Tm = 1 2 (cid:61)(T − m − T + m ), where, T ± m = Jsd 2SNd T rel[σ∓S± m ρm,m±1]. (6) (7) m = (cid:112)S(S + 1) − m(m ± 1) is the current along the auxiliary m-direction in Fig. 2 from the m ↔ m + 1 (± sign) state of the total Sz of the FM. Here, T rel, is the trace over the conduction electron degrees of freedom, and S± are the ladder operators. It is important to note that within this formalism the damping rate is simply the net current across the mth-layer along the auxiliary direction associated with the transition rate of the FM from state m to its nearest-neighbor states (m ± 1). Fig. 3 shows the damping rate as a function of the pre- cession cone angle, θ = cos−1( m S ), for different values of bias and for an in-plane effective magnetic field (a) along and (b) normal to the transport direction, and (c) an out- of-plane magnetic field. For cases (a) and (c) the damp- ing rate is negative and relatively independent of bias for low bias values. A negative damping rate implies that the FM relaxes towards the magnetic field by losing its angu- lar momentum, similar to the Gilbert damping rate term (cid:82) dϕ in the classical LLG equation, where its average value over the azimuthal precession angle, ϕ = ωt, is of the 2π (cid:126)m× ( (cid:126)m× (cid:126)B)· (cid:126)nM , which is nonzero form, T = −αsd (zero) when the unit vector (cid:126)nM is along (perpendicular to) the effective magnetic field. The dependence of the damping rate on the bias voltage when the effective mag- netic field (cid:126)B is inplane and normal to the transport direc- tion can be understood by the spin-flip reflection mech- anism accompanied by Rashba spin-momentum locking described in Ref.16. One can see that a large enough bias can result in a sign reversal of the damping rate and hence a magnetization reversal of the FM. It's worth mention- ing that due to the zero-point quantum fluctuations of FIG. 2: (Color online) Schematic representation of the quasi- particles of the FM and conduction electron entangled states. The horizontal planes denote the eigenstates, S, m(cid:105) of the total Sz of the FM with eigenvalues m = −S,−S + 1, . . . , +S along the auxiliary m-direction. Excitation of magnetic state induces a shift of the chemical potential of the Fermi-Dirac distribution function leading to flow of quisiparticles along the m-direction which corresponds to the damping rate of the FM. The FM damping involves two processes: (1) An intra-plane process involving spin reversal of the conduction electron via the SOC; and (2) An inter-plane process involving quasiparti- cle flow of majority (minority) spin along the ascending (de- scending) m-direction due to conservation of total angular momentum, where the interlayer hopping is accompanied by a spin flip of conduction electrons. be ignored. In the second case we ignore the correlation between different local moments and employ a mean field approximation such that at each step we focus on an indi- vidual atom by considering the local moment under con- sideration as a quantum mechanical object while the rest of the moments are treated classically. We should men- tion that a more accurate modeling of the system should contain both single domain rotation of the FM as well as the local spin flipping but also the effect of nonlocal correlations between the local moments and conduction electrons, which are ignored in this work. III. SINGLE DOMAIN ROTATIONAL SWITCHING average spin operator, (cid:126)sd(r) =(cid:80) In the regime where the energy required for the excita- tion of a single local spin moment (≈ meV ) is much larger than the MAE (≈ µeV ) the low-energy excited states cor- respond to rotation of the total angular momentum of the FM acting as a single domain and the effects of local spin flips described as the second term in Eq 2, can be ignored. In this regime all of the local moments behave collectively and the local moment operators can be replaced by the r(cid:48) (cid:126)sd(r(cid:48))/Nd = sd (cid:126)S/S, where Nd is the number of local moments and (cid:126)S is the total angular momentum with amplitude S. The mag- netic energy operator is given by HM = − (cid:126)B · S, where, (cid:126)B = gµB (cid:126)Bext + Jsd(cid:126)sc. Here, for simplicity we assume (cid:126)sc to be scalar and independent of the FM state. The eigenstates of HM operator are then simply the eigen- the magnetization, at θ = 0, π (i.e. m = ±S) we have T (cid:54)=0 which is inversely proportional to the size of the magnetic moment, S. In Fig. 4(a) we present the effective damping rate ver- sus bias for different values of the Rashba SOC. The re- sults show a linear response regime with respect to the bias voltage where both the zero-bias damping rate and the slope, dT /dV increases with the Rashba SOC. This is consistent with Kambersky's mechanism of Gilbert damping due to the SOC of itinerant electrons,24 and the SOT mechanism16. Fig. 4(b) shows that in the absence of bias voltage the damping rate is proportional to t2 so and the effect of the spin current pumped into the left and right reservoirs is negligible. This result of the t2 so depen- dence of the zero-bias damping rate is in agreement with recent calculations of Costa and Muniz25 and Edwards26 which took into account the collective excitations. In the presence of an external bias, T varies linearly with the SOC, suggesting that to the lowest order it can be fitted to T = sin2(θ)tso(c1tsoω + c2eVbias), (8) where c1 and c2 are fitting parameters. The bias-induced efficiency of the anti-damping SOT, Θ ≡ ω(T (Vbias) − T (0))/eVbiasT (0), describes how effi- cient is the energy conversion between the magnetization dynamics and the conduction electrons. Accordingly, for a given bias-induced efficiency, Θ, one needs to apply an external bias equal to ω/eΘ to overcome the zero-bias damping of the FM. Fig. 5 displays the anti-damping ef- ficiency versus the position of the Fermi energy of the FM from the bottom (-4t=-4 eV) to the top (4t=4eV) of the conduction electron band for the two-dimensional square lattice. The result is independent of the bias voltage and the Larmor frequency in the linear response regime (i.e. Vbias, ω (cid:28) t). We find that the efficiency peaks when the Fermi level is in the vicinity of the bottom or top of the energy band where the transport is driven by electron- or hole-like carriers and the Gilbert damping is minimum. The sign reversal of the antidamping SOT is due to the electron- or hole-like driven transport similar to the Hall effect.36 Classical Regime of the Zero Bias Damping rate - In the following we show that in the case of classical magnetic moments (S → ∞) and the adiabatic regime (ω → 0), the formalism developed in this paper leads to the conventional expressions for the damping rate. In this limit the system becomes locally periodic and one can carry out a Fourier transformation from m ≡ Sz space to azimuthal angle of the magnetization orientation, ϕ, space. Conservation of the angular momentum suggests that the majority- (minority-) spin electrons can propa- gate only along the ascending (descending) m-direction, where the hopping between two nearest-neighbor m- layers is accompanied by a spin-flip. As shown in Fig. 2 the existence of spin-flip hopping requires the presence of intralayer SOC-induced noncollinear spin terms which rotate the spin direction of the conduction electrons as 5 FIG. 3: (Color online) Effective damping rate for a single FM domain as a function of the precession cone angle, θ, for various bias values under an effective magnetic field which is in-plane (a) along and (b) normal to the transport direction and (c) out-of-plane. The length of the FM along the x di- rection is Lx = 25a while it is assumed to be infinite in the y-direction, ω = 10µeV , the broadening parameter η = 0, kBT = 10meV and the domain magnetic moment S = 200. The results are robust with larger values of S in either the ballistic, η (cid:28) ω, or dirty, η (cid:29) ω, regimes. FIG. 4: (Color online) Damping torque versus (a) bias voltage and (b) spin-orbit coupling strength, for m = 0 corresponding to the precession cone angle of 90o. The precession axis of the FM is along the y-direction and the rest of the parameters are the same as in Fig. 3. The zero-bias damping rate versus SOC shows a t2 so dependence while the damping rate under non- zero bias exhibits nearly linear SOC dependence. they propagate in each m-layer. This is necessary for the persistent flow of electrons along the ϕ auxiliary di- rection and therefore damping of the magnetization dy- namics. Using the Drude expression of the longitudinal conductivity along the ϕ-direction for the damping rate, we find that, within the relaxation time approximation, η/ω → ∞, where the relaxation time of the excited con- duction electrons is much shorter than the time scale of 04590135180−10−505B=[B,0,0]Damping Rate ( µeV) 4590135180B=[0,B,0]Cone Angle,θ (deg)4590135180B=[0,0,B]Vbias=−5 mVVbias=0Vbias=5 mV(a)(b)(c)Student Version of MATLAB−2−1012−3−2−101Bias Voltage (mV)Damping Rate ( µeV) −0.500.5−15−10−505Spin Orbit Coupling, tso (eV) tso=0tso=0.1 eVtso=0.2 eVVbias=−5 mVVbias=0Vbias=5 mV(a)(b)Student Version of MATLAB6 FIG. 5: (Color online) Bias-induced precessional anti- damping efficiency, Θ = ω(T (Vbias)−T (0))/eVbiasT (0), ver- sus the Fermi energy of the 2D Rashba plane in Fig.1, where the energy band ranges from -4 eV to +4 eV. The magnetiza- tion precesses around the in-plane direction (y−axis) normal to the transport direction and the rest of the parameters of the system are the same as in Fig. 3. Note, for magnetization precession around the x and z axis, T (Vbias) = T (0) for all precession cone angles and hence Θ=0. Inset shows the damp- ing rate versus the Fermi energy for different bias values used to calculate precessional anti-damping efficiency. FIG. 6: (Color online) Precessional damping rate versus broadening of the states in the presence (solid lines) and ab- sence (dashed lines) of bias voltage for two values of the do- main size S = 200 and S = 300. In both ballistic, η/ω ≈ 0, and diffusive, η/ω (cid:29) 1, regimes the precessional damping rate is independent of the domain size, while in the intermediate case, the amplitude of the minimum of damping rate shows a linear dependence versus S. Note that the value of the broad- ening at which the damping rate is minimum varies inversely proportional to the domain size, S. the FM dynamics, T is given by (cid:88) (cid:90) dkxdkydϕ (2π)3 n T = − ω η (vϕ n(cid:126)k )2f(cid:48)(εn(cid:126)k(ϕ)). (9) Here, vϕ = ∂εn(cid:126)k(ϕ)/∂ϕ is the group velocity along the n(cid:126)k ϕ-direction in Fig. 2, and εn,(cid:126)k = ε0((cid:126)k) ± (cid:126)h((cid:126)k) for the 2D-Rashba plane, where ε0((cid:126)k) is the spin independent dispersion of the conduction electrons and (cid:126)h = atsoez × (cid:126)k + 1 2 Jsd (cid:126)m, is the spin texture of the electrons due to the SOC and the s − d exchange interaction. For small precession cone angle, θ, the Gilbert damping constant can be determined from α = −T /sdω sin2(θ), where the zero-temperature T is evaluated by Eq. (9). We find that α ≈ 1 η F a)2D−(EF )(cid:3) (1+cos2(γ)), F a)2D+(EF ) + (k− (cid:2)(k+ t2 so F ) are obtained from, ε0(k± (10) where D+(−)(E) is the density of states of the majority (minority) band, γ is the angle between the precession axis and the normal to the Rashba plane, and the Fermi F ) = EF ∓ wave-vectors (k± Jsd/2. Eq. (10) shows that the Gilbert damping increases as the precession axis changes from in-plane (γ = π/2) to out of plane (γ = 0),37 which can also be seen in Fig. 3. It is important to emphasize that in contrast to Eq. (9) the results shown in Fig. 4 correspond to the ballistic regime with η = 0 in the central region and the relaxation of the excited electrons occurs solely inside the metallic reservoirs. To clarify how the damping rate changes from the ballistic to the diffusive regime we present in Fig. 6 the damping rate versus the broadening, η, of states in the presence (solid line) and absence (dashed line) of bias voltage. We find that in both ballistic (η/ω ≈ 0) and diffusive (η/ω (cid:29) 1) regimes the damping rate is independent of the size of the FM domain, S. On the other hand, in the intermediate regime the FM dynamics become strongly dependent on the effective domain size where the minimum of the damping rate varies linearly with S. This can be understood by the fact that the effective chemical potential difference between the first, m = −S and last, m = S layers in Fig.3 is proportional to S and for a coherent electron transport the conduc- tance is independent of the length of the system along the transport direction. Therefore, in this case the FM motion is driven by a coherent dynamics. IV. DEMAGNETIZATION MECHANISM OF SWITCHING In Sec. III we considered the case of a single FM do- main where its low-energy excitations, involving the pre- cession of the total angular momentum, can be described by the eigenstates m(cid:105) of Sz and local spin flip processes were neglected. However, for ultrathin FM films or FM nanoclusters, where the MAE per atom (≈ meV ) is com- parable to the exchange energy between the local mo- ments (Curie temperature), the low-energy excitations involve both magnetization rotation and local moments spin-flips due to conduction electron scattering which can in turn change also S. In this case the switching is ac- −4−3−2−101234−3−2−10123Fermi Energy (eV)Antidamping Efficiency (%)−4−2024−40−200Damping Rate (µeV) Vbias=5 mVVbias=0Vbias=−5 mVStudent Version of MATLAB10−810−610−410−2100−80−60−40−20020Broadening (eV)Damping Rate ( µeV) S=200, Vbias=0S=200, Vbias=3 mVS=300, Vbias=0S=300, Vbias=3 mVStudent Version of MATLAB7 FIG. 8: (Color online) Bias dependence of the average (over all sites) damping rate of the FM island for in-plane effective magnetic field (or equilibrium magnetization) (a) along and (b) normal to the transport direction and (c) out-of-plane magnetic field for two values of B. 1) for two bias values (Vbias = ±0.4V ) and for an in- plane effective magnetic field (a) along and (b) normal to the transport direction, and (c) an out-of-plane mag- netic field. The size of the FM island is 25a× 25a, where a is the lattice constant. Negative local moment switch- ing rate (blue) denotes that, once excited, the local mo- ment relaxes to its ground state pointing along the di- rection of the effective magnetic field; however positive local damping rate (red) denotes that the local moments remain in the excited state during the bias pulse dura- tion. Therefore, the damping rate of the local moments under bias voltage can be either enhanced or reduced and even change sign depending on the sign of the bias voltage and the direction of the magnetization. We find that the bias-induced change of the damping rate is high- est when the FM magnetization is in-plane and normal to the transport directions similar to the single domain case. Furthermore, the voltage-induced damping rate is peaked close to either the left or right edge of the FM (where the reservoirs are attached) depending on the sign of the bias. Note that there is also a finite voltage-induced damping rate when the magnetization is in-plane and and along the transport direction (x) or out-of-the-plane (z). Fig. 8 shows the bias dependence of the average (over all sites) damping rate for in- (a and b) and out-of-plane (c) directions of the effective magnetic field (direction of the equilibrium magnetization) and for two values of B. This quantity describes the damping rate of the amplitude of the magnetic order parameter. For an in- plane magnetization and normal to the transport direc- tion (Fig. 8) the bias behavior of the damping rate is lin- ear and finite in contrast to the single domain [Fig. 3(a)] where the damping rate was found to have a negligible response under bias. On the other hand, the bias behav- ior of the current induced damping rate shows similar behavior to the single domain case when the equilibrium FIG. 7: (Color online) Spatial dependence of the local damp- ing rate for the spin-1/2 local moments of a FM island under different bias voltages (±0.4V ) and magnetization directions. For the parameters we chose the size of the FM island to be 25 × 25a2, the effective magnetic field, B = 20meV , the broadening, η = 0, and kBT = 10 meV. companied by the excitation of local collective modes that effectively lowers the amplitude of the magnetic ordering parameter. For simplicity we employ the mean field ap- proximation for the 2D FM nanocluster where the spin under consideration at position r is treated quantum me- chanically interacting with all remaining spins through an effective magnetic field, (cid:126)B. The spatial matrix ele- ments of the local spin operator are [(cid:126)sd,r]r1r2 = (cid:126)sd(r1)δr1r2(1 − δr1r)1s + δr1r2 δr1r(cid:126)τ , (11) 1 2 where, (cid:126)τ s are the Pauli matrices. The magnetic energy can be expressed as, HM (r) = − (cid:126)B(r) · (cid:126)τ /2, where, the effective local magnetic field is given by, (cid:126)B(r) = gµB (cid:126)Bext + 4 rr(cid:48)(cid:126)sd(r(cid:48)) + 2Jsd(cid:126)sc(r). J dd (12) (cid:88) r(cid:48) The equation of motion for the single particle propa- gator of the electronic wavefunction entangled with the local spin moment under consideration can then be ob- tained from, E − µ − HM (r) − H − Jsd 2 (cid:126)σ · (cid:126)sd,r Gr r(E) = 1. (13) (cid:18) (cid:19) The density matrix is determined from Eq. (5) which can in turn be used to calculate the spin density of the conduction electrons, (cid:126)sc(r) = T r((cid:126)σ ρrr)/2, and the di- rection and amplitude of the local magnetic moments, (cid:126)sd(r) = T r((cid:126)τ ρrr)/2. Fig. 7 shows the spatial dependence of the spin- 1 2 local moment switching rate for a FM/Rashba bilayer (Fig. −101−30−20−100102030 Damping Torque (meV)B=[B,0,0]−101Bias Voltage (V)B=[0,B,0]−101B=[0,0,B]B=1 meVB=20 meV(c)(a)(b)Student Version of MATLAB8 the FM. When the magnetization (precession axis) is in- plane and normal to the transport direction, similar to the conventional SOT for classical FMs, we show that the bias voltage can change the damping rate of the FM and for large enough voltage it can lead to a sign reversal. In the case of independent spin-1/2 local moments we show that the bias-induced damping rate of the local quantum moments can lead to demagnetization of the FM and has strong spatial dependence. Finally, in both regimes we have calculated the bias-induced damping efficiency as a function of the position of the Fermi energy of the 2D Rashba plane. Appendix A: Derivation of Electronic Density Matrix Using the Heisenberg equation of motion for the an- gular momentum operator, (cid:126)S(t), and the commutation relations for the angular momentum, we obtain the fol- lowing Landau-Lifshitz equations of motion, ∓ i − i ∂ ∂t ∂ ∂t S±(t) = hzS±(t) − h±(t)Sz(t) (cid:0)h+(t)S−(t) − h−(t)S+(t)(cid:1) Sz(t) = (A1a) (A1b) (cid:126)hmm(cid:48)(t) = 1 Jsd(cid:126)smm(cid:48) c (r) + gµBδmm(cid:48) (cid:126)B(t), (A1c) 1 2 (cid:88) r c (σ± = σx ± σy), (cid:80) σσ(cid:48) (cid:126)σσσ(cid:48)ρmm(cid:48) where, S± = Sx ± Sy lar momentum (spin) ladder operators, (cid:126)smm(cid:48) is the angu- (r) = 1 σσ(cid:48),rr is the local spin density of the con- 2 duction electrons which is an operator in magnetic con- figuration space. Here, ρ is the density matrix of the system, and the subscripts, r, m, σ refer to the atomic cite index, magnetic state and spin of the conduction electrons, respectively. In the following we assume a pre- cessing solution for Eq (A1)(a) with a fixed cone angle and Larmor frequency ω = hz. Extending the Hilbert space of the electrons to include the angular momentum degree of freedom we define ψms(cid:48)i(t) = S, m(cid:105) ⊗ ψs(cid:48)i(t). The equation of motion for the Green function (GF) is then given Jsd(k)σz(cid:17) Gr (cid:16) − E − iη − H(k) + nω − n 2S Jsd(k)σ− Gr (cid:112)S(S + 1) − n(n + 1) (cid:112)S(S + 1) − n(n − 1) 2S nm(E, k) (A2) n+1m(E, k) 2S Jsd(k)σ+ Gr − n−1m(E, k) = 1δnm where, n = (−S,−S +1, ..., S) and the gauge transforma- tion ψnσi(t) → einωtψnσi(t) has been employed to remove the time dependence. The density matrix of the system is of the form ρnm = e−i(n−m)ωt Gr np2ηfpµ Ga pm (A3) S(cid:88) (cid:90) dE p=−S 2π FIG. 9: (Color online) Bias-induced local anti-damping efficiency due to local spin-flip, Θ = B(T (Vbias) − T (0))/eVbiasT (0), versus Fermi energy for different equilib- rium magnetization orientations. For the calculation we chose Vbias = 0.2 V and the rest of the Hamiltonian parameters are the same as in Fig. 7. magnetization direction is in-plane and normal to the transport direction (Fig. 8(b)). For an out-of-plane ef- fective magnetic field [Fig. 8(c)] the damping torque has an even dependence on the voltage bias. In order to quantify the efficiency of the voltage in- duced excitations of the local moments, we calculate the relative change of the average of the damping rate in the presence of a bias voltage and present the result versus the Fermi energy for different orientations of the magne- tization in Fig 9. We find that the efficiency is maximum for an in-plane equilibrium magnetization normal to the transport direction and it exhibits an electron-hole asym- metry. The bias-induced antidamping efficiency due to spin-flip can reach a peak around 20% which is much higher than the peak efficiency of about 2% in the sin- gle domain precession mechanism in Fig. 5 for the same system parameters. Future work will be aimed in determining the switch- ing phase diagram16 by calculating the local antidamping and field-like torques self consistently for different FM configurations. V. CONCLUDING REMARKS In conclusion, we have developed a formalism to in- vestigate the current-induced damping rate of nanoscale FM/SOC 2D Rashba plane bilayer in the quantum regime within the framework of the Kyldysh Green func- tion method. We considered two different regimes of FM dynamics, namely, the single domain FM and indepen- dent local moments regimes. In the first regime we as- sume the rotation of the FM as the only degree of free- dom, while the second regime takes into account only the local spin-flip mechanism and ignores the rotation of −4−2024−30−20−100102030Fermi Energy (eV)Antidamping Efficiency (%) B=[20,0,0] meVB=[0,20,0] meVB=[0,0,20] meVStudent Version of MATLAB9 we have S (cid:29) 1, we need a recursive algorithm to be able to solve the system numerically. The surface Keldysh GFs corresponding to ascending gu,r/<, and descending gd,r/<, recursion scheme read, where, fpµ(E) = f (E − pω − µ) is the equilibrium Fermi distribution function of the electrons. Due to the fact that pω are the eigenvalues of HM = −gµB (cid:126)B · S, one can generalize this expression by transforming into a ba- sis where the magnetic energy is not diagonal which in turn leads to Eq (5) for the density matrix of the con- duction electron-local moment entagled system. Appendix B: Recursive Relation for GFs Since in this work we are interested in diagonal blocks of the GFs and in general for FMs at low temperature (cid:16) (cid:16) α Σu,< gu,r n (E, k) = gd,r n (E, k) = n (E, k) = −(cid:88) n (E, k) = −(cid:88) n(E, k) =(cid:80) Σd,< α E − ωn − iηn − H(k) − Σr n(E, k) − n 2iηn + Σr n,α(E, k) − Σa n,α(E, k) (cid:17) − 2S Jsd(k)σz − (S n−1(E, k)σ−Jsd(k) n )2 4S2 Jsd(k)σ+gu,r (S− n )2 Σu,< 4S2 Jsdσ+gu,r n−1gu,a n−1 n−1σ−Jsd fnα + E − ωn − iηn − Σr 2iηn + Σr n(E, k) − H(k) − n n,α(E, k) − Σa n,α(E, k) n )2 (cid:17) 2S Jsd(k)σz − (S+ 4S2 Jsd(k)σ−gu,r n )2 (S+ 4S2 Jsdσ−gd,r Σd,< n+1gd,a fnα + n+1 n+1σ+Jsd n+1(E, k)σ+Jsd(k) 1 1 (B1) (B2) (B3) (B4) α(E − ωn, k) corresponds to the Σr where, Σr self energy of the leads, α = L, R refers to the left and right leads in the two terminal device in Fig. 3 and S± m = α (cid:112)S(S + 1) − m(m ± 1). Using the surface GFs we can calculate the GFs as follows, Gr n,m(E, k) = = = 1 n − n E − ωn − iηn − H(k) − Σr S+ n (E, k)Jsd(k)σ− Gr n gu,r 2S S− n 2S n (E, k)Jsd(k)σ+ Gr gd,r 2S Jsd(k)σz − Σr,u n (cid:54)= m n+1,m(E, k), n−1,m(E, k), n (cid:54)= m n − Σr,d n where the ascending and descending self energies are given by, n−1(E, k)σ−Jsd(k) n+1(E, k)σ+Jsd(k) Σr,u n = Σr,d n = (S− n )2 4S2 Jsd(k)σ+gu,r (S+ n )2 4S2 Jsd(k)σ−gd,r (cid:32)(cid:88) S− n 2S T r[ k The average rate of angular momentum loss/gain can be obtained from the real part of the loss of angular momentum in one period of precession, T (cid:48) n = 1 2 (T (cid:48)− n − T (cid:48)+ n ) = (cid:61) 1 2 σ+Jsd(k) ρnn+1(k) − S+ n 2S σ−Jsd(k) ρnn−1(k)] (B10) , n = m (B5) (B6) (B7) (B8) (B9) (cid:33) which can be interpreted as the current flowing across the layer n. T (cid:48)−/+ n = (E, k) − Σd/u,a n (E, k) nn(E, k) + Σd/u,< n (E) (cid:90) dE (cid:88) 2πi k (cid:110)(cid:104) Σd/u,r n Tr (cid:105) G< (cid:104) Gr nn(E, k) − Ga nn(E, k) 10 (cid:105)(cid:111) , (B11) Acknowledgments The work at CSUN is supported by NSF-Partnership in Research and Education in Materials (PREM) Grant DMR-1205734, NSF Grant No. ERC-Translational Ap- plications of Nanoscale Multiferroic Systems (TANMS)- 1160504, and US Army of Defense Grant No. W911NF- 16-1-0487. ∗ Electronic address: [email protected] † Electronic address: [email protected] 1 J. C. Slonczewski, Current-driven excitation of magnetic multilayers, J. Magn. Magn. Mater. 159, L1-L7 (1996). 2 L. Berger, Emission of spin waves by a magnetic multilayer traversed by a current, Phys. Rev. B 54, 9353 (1996). 3 D. Ralph and M. Stiles, Spin transfer torques, J. Magn. Magn. Mater. 320, 1190 (2008); A. Brataas, A. D. Kent, and H. Ohno, Current-induced torques in magnetic mate- rials, Nature Mater. 11, 372 (2012). 4 Ioannis Theodonis, Nicholas Kioussis, Alan Kalitsov, Mair- bek Chshiev, and W. H. Butler, Anomalous Bias Depen- dence of Spin Torque in Magnetic Tunnel Junctions, Phys. Rev. Lett. 97, 237205 (2006). 5 P. Gambardella and I. M. Miron, Current-induced spinor- bit torques, Phil. Trans. R. Soc. A 369, 3175 (2011). 6 I. M. Miron et al., Nature Mater. 9, 230 (2010) I. M. Miron, K. Garello, G. Gaudin, Pierre-Jean Zermatten, M. V. Costache, S. Auffret, S. Bandiera, B. Rodmacq, A. Schuhl, and P. Gambardella, Perpendicular switching of a single ferromagnetic layer induced by in-plane current injection, Nature 476, 189 (2011); 7 L. Liu, C. F. Pai, Y. Li, H. W. Tseng, D. C. Ralph, and R. A. Buhrman, Spin-torque switching with the giant spin Hall effect of tantalum, Science 336, 555 (2012). 8 L. Liu, T. Moriyama, D. C. Ralph, and R. A. Buhrman, Spin-Torque Ferromagnetic Resonance Induced by the Spin Hall Effect, Phys. Rev. Lett. 106, 036601 (2011); Luqiao Liu, O. J. Lee, T. J. Gudmundsen, D. C. Ralph, and R. A. Buhrman, Current-Induced Switching of Perpen- dicularly Magnetized Magnetic Layers Using Spin Torque from the Spin Hall Effect, Phys. Rev. Lett. 109, 096602 (2012). 9 M. I. Dyakonov and V. I. Perel, Current-induced spin ori- entation of electrons in semiconductors, Phys. Lett. 35A, 459 (1971). 10 J. E. Hirsch, Spin Hall Effect, Phys. Rev. Lett. 83, 1834 (1999). 11 T. Jungwirth, J. Wunderlich and K. Olejn´ik, Spin Hall effect devices, Nature Materials 11, (2012). 12 J. Sinova, S. O. Valenzuela, J. Wunderlich, C. H. Back, and T. Jungwirth, Spin Hall Effects, Rev. Mod. Phys. 87, 1213 (2015). 13 E. van der Bijl and R. A. Duine, Current-induced torques in textured Rashba ferromagnets, Phys. Rev. B 86, 094406 (2012). 14 Kyoung-Whan Kim, Soo-Man Seo, Jisu Ryu, Kyung-Jin Lee, and Hyun-Woo Lee, Magnetization dynamics in- duced by in-plane currents in ultrathin magnetic nanos- tructures with Rashba spin-orbit coupling, Phys. Rev. B 85, 180404(R) (2012). 15 H. Kurebayashi, Jairo 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, An antidamping spinor- bit torque originating from the Berry curvature, Nature Nanotechnology 9, 211 (2014). 16 F. Mahfouzi, B. K. Nikoli´c, and N. Kioussis, Antidamping spin-orbit torque driven by spin-flip reflection mechanism on the surface of a topological insulator: A time-dependent nonequilibrium Green function approach, Phys. Rev. B 93, 115419 (2016). 17 Frank Freimuth, Stefan Blugel, and Yuriy Mokrousov, Spin-orbit torques in Co/Pt(111) and Mn/W(001) mag- netic bilayers from first principles, Phys. Rev. B 90, 174423 (2014). 18 J. Kim, J. Sinha, M. Hayashi, M. Yamanouchi, S. Fukami, T. Suzuki, S. Mitani and H. Ohno, Layer thickness de- pendence of the current-induced effective field vector in Ta-CoFeB-MgO, Nature Materials 12, 240245 (2013). 19 Xin Fan, H. Celik, J. Wu, C. Ni, Kyung-Jin Lee, V. O. Lorenz, and J. Q. Xiao, Quantifying interface and bulk contributions to spinorbit torque in magnetic bilayers, 5:3042 - DOI: 10.1038/ncomms4042 (2014). 20 A. Chudnovskiy, Ch. Hubner, B. Baxevanis, and D. Pfannkuche, Spin switching: From quantum to quasiclas- sical approach, Phys. Status Solidi B 251, No. 9, 1764 (2014). 21 M. Fahnle and C. Illg, Electron theory of fast and ultrafast dissipative magnetization dynamics, J. Phys.: Condens. Matter 23, 493201 (2011) 22 J. Swiebodzinski, A. Chudnovskiy, T. Dunn, and A. Kamenev, Spin torque dynamics with noise in magnetic nanosystems, Phys. Rev. B 82, 144404 (2010). 23 Hang Li, H. Gao, Liviu P. Zarbo, K. Vyborny, Xuhui Wang, Ion Garate, Fatih Dogan, A. Cejchan, Jairo Sinova, T. Jungwirth, and Aurelien Manchon, Intraband and inter- band spin-orbit torques in noncentrosymmetric ferromag- nets Phys. Rev. B 91, 134402 (2015). 24 V. Kambersk´y, On the LandauLifshitz relaxation in fer- romagnetic metals, Can. J. Phys. 48, 2906 (1970); V. Kambersk´y, Spin-orbital Gilbert damping in common magnetic metals Phys. Rev. B 76, 134416 (2007). 25 A. T. Costa and R. B. Muniz, Breakdown of the adiabatic approach for magnetization damping in metallic ferromag- nets, Phys. Rev B 92, 014419 (2015). 26 D. M. Edwards, The absence of intraband scattering in a consistent theory of Gilbert damping in pure metallic ferromagnets, Journal of Physics: Condensed Matter, 28, 8 (2016). 27 Paul M. Haney, Hyun-Woo Lee, Kyung-Jin Lee, Aurlien Manchon, and M. D. Stiles, Current induced torques and interfacial spin-orbit coupling: Semiclassical modeling Phys. Rev. B 87, 174411 (2013). 28 Jin-Hong Park, Choong H. Kim, Hyun-Woo Lee, and Jung Hoon Han, Orbital chirality and Rashba interaction in magnetic bands, Phys. Rev. B 87, 041301(R) (2013). 29 Suprio Data, Quantum Transport: Atom to Transistor, Cambridge University Press, New York, (2005). 30 S. Zhang and Z. Li, Roles of Nonequilibrium Conduction Electrons on the Magnetization Dynamics of Ferromagnets Phys. Rev. Lett. 93, 127204 (2004). 31 Maria Stamenova, Stefano Sanvito, and Tchavdar N. Todorov, Current-driven magnetic rearrangements in spin- polarized point contacts, Phys. Rev. B 72, 134407 (2005). 32 D. E. Eastman, F. J. Himpsel, and J. A. Knapp, Exper- imental Exchange-Split Energy-Band Dispersions for Fe, 11 Co, and Ni, Phys. Rev. Lett. 44, 95 (1980). 33 Anton Kokalj and Mauro Caus , Periodic density func- tional theory study of Pt(111): surface features of slabs of different thicknesses, J. Phys.: Condens. Matter 11 7463 (1999). 34 Ioan Mihai Miron, Thomas Moore, Helga Szambolics, Lil- iana Daniela Buda-Prejbeanu, Stphane Auffret, Bernard Rodmacq, Stefania Pizzini, Jan Vogel, Marlio Bonfim, Alain Schuhl and Gilles Gaudins, Fast current-induced domain-wall motion controlled by the Rashba effect, Na- ture Materials 10, 419?423 (2011). 35 P. P. J. Haazen, E. Mure, J. H. Franken, R. Lavrijsen, H. J. M. Swagten, and B. Koopmans, Domain wall depinning governed by the spin Hall effect, Nat. Mat. 12, 4, 299303 (2013). 36 Charles Kittel, Introduction to Solid State Physics, Wiley, (2004). 37 F. Mahfouzi, N. Kioussis, Ferromagnetic Damping/Anti- damping in a Periodic 2D Helical Surface; A Nonequi- librium Keldysh Green Function Approach, SPIN 06, 1640009 (2016).
1604.06873	1	1604	2016-04-23T07:50:37	Reflectivity properties of graphene with nonzero mass-gap parameter	[ "cond-mat.mes-hall", "quant-ph" ]	The reflectivity properties of graphene with nonzero mass-gap parameter are investigated in the framework of Dirac model using the polarization tensor in (2+1)-dimensional space-time. For this purpose, a more simple explicit representation for the polarization tensor along the real frequency axis is found. The approximate analytic expressions for the polarization tensor and for the reflectivities of graphene are obtained in different frequency regions at any temperature. We show that the nonzero mass-gap parameter has a profound effect on the reflectivity of graphene. Specifically, at zero temperature the reflectivity of gapped graphene goes to zero with vanishing frequency. At nonzero temperature the same reflectivities are equal to unity at zero frequency. We also find the resonance behavior of the reflectivities of gapped graphene at both zero and nonzero temperature at the border frequency determined by the width of the gap. At nonzero temperature the reflectifities of graphene drop to zero in the vicinity of some frequency smaller than the border frequency. Our analytic results are accompanied with numerical computations performed over a wide frequency region. The developed formalism can be used in devising nanoscale optical detectors, optoelectronic switches and in other optical applications of graphene.	cond-mat.mes-hall	cond-mat	Reflectivity properties of graphene with nonzero mass-gap parameter G. L. Klimchitskaya1, 2 and V. M. Mostepanenko1, 2 1Central Astronomical Observatory at Pulkovo of the Russian Academy of Sciences, Saint Petersburg, 196140, Russia 2Institute of Physics, Nanotechnology and Telecommunications, Peter the Great Saint Petersburg Polytechnic University, St.Petersburg, 195251, Russia Abstract The reflectivity properties of graphene with nonzero mass-gap parameter are investigated in the framework of Dirac model using the polarization tensor in (2+1)-dimensional space-time. For this purpose, a more simple explicit representation for the polarization tensor along the real frequency axis is found. The approximate analytic expressions for the polarization tensor and for the reflec- tivities of graphene are obtained in different frequency regions at any temperature. We show that the nonzero mass-gap parameter has a profound effect on the reflectivity of graphene. Specifically, at zero temperature the reflectivity of gapped graphene goes to zero with vanishing frequency. At nonzero temperature the same reflectivities are equal to unity at zero frequency. We also find the resonance behavior of the reflectivities of gapped graphene at both zero and nonzero temperature at the border frequency determined by the width of the gap. At nonzero temperature the reflectifities of graphene drop to zero in the vicinity of some frequency smaller than the border frequency. Our analytic results are accompanied with numerical computations performed over a wide frequency re- gion. The developed formalism can be used in devising nanoscale optical detectors, optoelectronic switches and in other optical applications of graphene. PACS numbers: 12.20.Ds, 78.20.Ci, 78.67.Wj 6 1 0 2 r p A 3 2 ] l l a h - s e m . t a m - d n o c [ 1 v 3 7 8 6 0 . 4 0 6 1 : v i X r a 1 I. INTRODUCTION Graphene is a two-dimensional sheet of carbon atoms arranged in a hexagonal lattice, which possesses unusual mechanical, electrical and optical properties. At energies below a few eV the electronic excitations (quasiparticles) in graphene are well described by the Dirac model, i.e., display the linear dispersion relation, where the speed of light c is replaced with the Fermi velocity vF ≪ c [1, 2]. This makes graphene a unique laboratory for testing some effects of quantum electrodynamics and quantum field theory, such as the thermal Casimir force [3], the Klein paradox [4], and the creation of particles from vacuum in strong external fields [5 -- 7]. It is important also that technological applications of graphene are numerous and diverse. One could mention graphene coatings used in the optical detectors [8], solar cells [9], transparent electrodes [10], corrosium protection [11], optical biosensors [12], optoelectronic switches [13], etc. In order to make the best use of graphene in both fundamental physics and its applica- tions, it is necessary to investigate the reflectivity properties of this novel material over the wide frequency region. As was shown experimentally, in the region of visible light, where the photon energy is much larger than both thermal and Fermi energies, the transparency and conductivity of graphene are defined by the fine structure constant α = e2/(c) [14] and by the quantity e2/(4) [15], respectively. These facts were understood in theoretical studies of the conductivity of graphene using the current-current correlation function and the Kubo formula (see Ref. [16] and review in Ref. [17]). The obtained results for the in-plane (longitudinal) conductivity have been applied to in- vestigate the reflectivity properties of graphene. This was done in the framework of spatially local approximation, i.e., with no regard for dependence on the wave vector. The reflection coefficients and reflectivities of graphene with zero mass-gap parameter (quasiparticle mass) for the transverse magnetic (TM), i.e., p polarized, electromagnetic waves have been found at both low [18] and high [19] frequencies. The case of gapless graphene deposited on a substrate was considered in Ref. [20]. The most fundamental quantity allowing full investigation of the reflectivity properties of graphene over the entire frequency range is the polarization tensor in (2+1)-dimensional space-time. It was found in Refs. [21, 22] at zero and nonzero temperature, respectively, and used to investigate the Casimir effect in graphene systems in Refs. [23 -- 30] (several 2 calculations of the Casimir force were performed also using the density-density correlation functions of graphene and other methods [3, 31 -- 35]). We stress, however, that for calculation of the Casimir force using the Lifshitz theory one should use the reflection coefficients defined at the discrete imaginary Matsubara frequencies [36, 37]. It was shown [38] that the polarization tensor of Ref. [22] is valid strictly at the Matsubara frequencies and cannot be immediately continued either to the real or to the entire imaginary frequency axis. The polarization tensor of graphene valid at all complex frequencies, including the real frequency axis, was derived in Ref. [38]. At the imaginary Matsubara frequencies it coincides with that of Ref. [22], but, in contrast to Ref. [22], allows immediate analytic continuation to the whole complex plane. Using this representation for the polarization tensor, the reflectivity properties of gapless graphene were investigated at low, high and intermediate frequencies for both polarizations of the electromagnetic field TM and TE (i.e., transverse electric or s polarization). In so doing, some new properties were found which escaped notice in previous literature. The polarization tensor of Ref. [38], has been also used to find the reflectivities of material plates coated with gapless graphene [39]. In this paper, the polarization tensor of Ref. [38] is used to investigate the reflectivity properties of graphene with nonzero mass-gap parameter m. It is common knowledge that the quasiparticles of pristine (ideal) graphene are massless. However, they acquire some small mass m under the influence of defects of the structure, electron-electron interaction, and in the presence of substrates [1, 40 -- 43]. The exact value of m for some specific graphene sample usually remains unknown, but, according to some estimations [21], may achieve 0.1 eV (we measure masses and frequencies in the units of energy). Here, we derive convenient explicit expressions for the polarization tensor of graphene with nonzero mass-gap parameter at any temperature. We demonstrate that under some conditions the nonzero mass-gap parameter may have a dramatic effect on the reflectivity properties of graphene. Specifically, we find the resonance behavior of graphene reflectivities at the border frequency of incident electro- magnetic waves ω ≈ 2mc2/. This effect occurs at both zero and nonzero temperature. At nonzero temperature the reflectivities of gapped graphene drop to zero at some frequency smaller than the border one. We find the asymptotic expressions for the reflectivities of gapped graphene at both low and high frequencies. At zero temperature the TM and TE reflectivities of graphene with m 6= 0 take the zero value at zero frequency. This is different from the case of gapless graphene, where both reflectivities at zero temperature are equal 3 to nonzero constants depending on α and on the angle of incidence [38]. At nonzero tem- perature the reflectivities of gapped graphene go to unity with vanishing frequency. We also perform numerical computations of the reflectivities of gapped graphene over the wide range of frequencies at both zero and nonzero temperature and find the regions, where nonzero value of m makes a profound effect on the obtained results. The paper is organized as follows. In Sec. II we derive convenient expressions for the polarization tensor of graphene with nonzero mass-gap parameter and present the general formalism. Section III contains calculation of the reflectivities of gapped graphene at zero temperature. An investigation of the impact of temperature on the reflectivity of gapped graphene is presented in Sec. IV. In Sec. V the reader will find our conclusions and discussion. II. SPECIAL FEATURES OF THE POLARIZATION TENSOR OF GRAPHENE WITH NONZERO MASS-GAP PARAMETER The description of graphene by means of the polarization tensor in (2+1)-dimensional space-time is in fact equivalent [26] to the method using the density-density correlation func- tions, which is more often used in atomic and condensed matter physics. The formalism of the polarization tensor, however, offers some advantages over the more conventional meth- ods, especially at nonzero temperature, because it has been much studied and elaborated in the framework of thermal quantum field theory. It is conventional [44, 45] to notate the polarization tensor as Πµν, where in (2+1)- dimensional case µ, ν = 0, 1, 2, and Πtr ≡ Πµ µ. In our configuration the polarization tensor Πµν = Πµν(ω, θi), where ω is the frequency of electromagnetic wave incident on graphene and θi is the angle of incidence. We assume that the graphene sheet is situated in the plane (x, y) and the z azis is perpendicular to it. Then for real photons on a mass-shell the magnitude of the wave vector component parallel to graphene (i.e., perpendicular to the z axis) is defined as The amplitude reflection coefficients for TM and TE polarizations of the electromagnetic kk = ω c sin θi . (1) 4 field incident on a graphene sheet are given by [22 -- 30, 38] rTM(ω, θi) = cos θi Π00(ω, θi) 2i ω c sin2 θi + cos θi Π00(ω, θi) , rTE(ω, θi) = Π(ω, θi) 2i ω3 c3 sin2 θi cos θi − Π(ω, θi) , where the following notation is introduced: Π(ω, θi) = ω2 c2 (cid:2)sin2 θi Πtr(ω, θi) + cos2 θi Π00(ω, θi)(cid:3) . Note that so important quantity as the conductivity of graphene can be immediately expressed via the polarization tensor. Thus, the spatially nonlocal (i.e., depending on the wave vector) longitudinal and transverse conductivities of graphene are given by [26] σk(ω, kk) = −i ω 4πk2 k Π00(ω, kk), σ⊥(ω, kk) = i c2 4πk2 kω Π(ω, kk). (4) It is convenient to represent the quantities Π00 and Π in the form Π00(ω, θi) = Π(0) 00 (ω, θi) + ∆T Π00(ω, θi), Π(ω, θi) = Π(0)(ω, θi) + ∆T Π(ω, θi), (5) where Π(0) 00 and Π(0) are defined at zero temperature and ∆T Π00, ∆T Π are the thermal corrections which go to zero in the limiting case of vanishing temperature. The explicit expressions for the quantities Π(0) 00 and Π(0) at the imaginary frequencies in the case of arbitrary mass-gap parameter m have been found in Ref. [21]. In our case of real frequencies they can be presented in the form (2) (3) (6) (7) Π(0) 00 (ω, θi) = − Π(0)(ω, θi) = α Φ(ω, θi), α sin2 θi η2(θi) ω2 c2 sin2 θiΦ(ω, θi), where η(θi) ≡ η = [1 − v2 F sin2 θi]1/2, vF ≡ vF /c and the function Φ is given by [38] 4mc − 2ωη 4mc − 2ωη 2ω2η2i arctanh ωη 2ω2η2i(cid:16)arctanh 2mc2 c h1 + 4m2c4 c h1 + 4m2c4 ω < 2mc2 η 2(cid:17) , ω ≥ 2mc2 η ωη + i π 2mc2 , , . Φ(ω, θi) =  Note that for a gapless graphene (m = 0) only the second line in Eq. (7) is applicable at all ω. In this case the function Φ is pure imaginary and has the negative imaginary part Φ(ω, θi) = −iπ ω c η(θi). 5 (8) Substituting Eqs. (6) and (8) in Eq. (4) with account of Eq. (1), one finds the conductivities of graphene at zero temperature σ(0) k (ω, kk) = e2 4 ω qω2 − v2 F c2k2 k , σ(0) ⊥ (ω, kk) = e2 4 qω2 − v2 ω F c2k2 k , (9) which are in accordance with the previous knowledge (see, for instance, Refs. [14 -- 17, 46]). Now we consider thermal corrections to the 00 component of the polarization tensor and to its trace entering the definition (3) of the quantity Π. According to Eqs. (33) -- (36) of Ref. [38], along the real frequency axis these thermal corrections can be written in the form ∆T Π00(tr)(ω, θi) = 16α v2 F Z ∞ 0 dq q Γ(q) 1 eβΓ(q) + 1"1 + 1 2 Xλ=±1 M (λ) 00(tr)(ω, θi, q) N (λ)(ω, θi, q) # , (10) where β ≡ c/(kBT ). Here, the integration variable q has the dimension 1/cm, Γ(q) = [q2 + (mc/)2]1/2, kB is the Boltzmann constant, T is the temperature, and the following notations are introduced: M (λ) 00 (ω, θi, q) = 4Γ2(q) + M (λ) tr (ω, θi, q) = ω c Γ(q), ω2 c2 η2(θi) + 4λ F (cid:16)mc (cid:17)2 ω2 c2 η2(θi) + 4v2 The expression for N (λ) depends on the fulfilment of the following conditions: N (λ)(ω, θi, q) = signQλhQλ 2 −(cid:0)2vF −ih−Qλ 2 +(cid:0)2vF  Qλ ≡ Qλ(ω, θi, q) = − ω where F(cid:1)hΓ2(q) + λ ω c Γ(q)i . , Qλ ≥ 2vF Qλ < 2vF ω c q sin θi, ω c q sin θi, ω + 4(cid:0)1 − v2 c q sin θi(cid:1)2i1/2 c q sin θi(cid:1)2i1/2 ω2 c2 η2(θi) − 2λ , ω c Γ(q). (11) (12) (13) (14) (15) In the case of m 6= 0 under consideration here it is more convenient to represent the temperature correction (10) as an integral with respect to the new variable u = Γ(q): ∆T Π00(tr)(ω, θi) = where 16α v2 F Z ∞ mc/ du 1 eβu + 1"1 + 1 2 Xλ=±1 M (λ) 00(tr)(ω, θi, u) N (λ)(ω, θi, u) # , M (λ) 00 (ω, θi, u) = 4u2 + M (λ) tr (ω, θi, u) = u, ω c ω2 c2 η2(θi) + 4λ F (cid:16)mc (cid:17)2 ω2 c2 η2(θi) + 4v2 6 + 4(cid:0)1 − v2 F(cid:1)(cid:16)u2 + λ ω c u(cid:17) . The quantity Qλ defined in Eq. (13) takes the form Qλ ≡ Qλ(ω, θi, u) = − ω2 c2 η2(θi) − 2λ ω c u. (16) It is easily seen that for λ = 1 the quantity Qλ is always negative and satisfies the condition Q1 ≥ 2vF ω c (cid:20)u2 −(cid:16) mc (cid:17)2(cid:21)1/2 Thus, the first line in Eq. (12) is applicable leading to sin θi . (17) ω N (1)(ω, θi, u) = − The case λ = −1 is more complicated and deserves separate consideration. According to Eq. (16) in this case Q−1 can be both positive and negative. In order to explicitly rewrite F sin2 θi(cid:20)4(cid:16)mc (cid:17)2 c (cid:26)η2(θi)(cid:16)ω + 2u(cid:17)2 c2 η2(θi)(cid:21)(cid:27)1/2 + v2 (18) − ω2 c . Eq. (12) in terms of the variable u, we introduce the quantity ω2 χ(ω, θi, u) ≡ Q 2 c2 (cid:26)η2(θi)(cid:16)ω −1 − 4v2 c − 2u(cid:17)2 ω2 c2 (cid:20)u2 −(cid:16) mc (cid:17)2(cid:21) sin2 θi F sin2 θi(cid:20)4(cid:16)mc (cid:17)2 + v2 = F (19) − ω2 c2 η2(θi)(cid:21)(cid:27) . The sign of the quantity χ under different values of parameters is found by solving the equation with respect to the variable u: u(±) = If the condition χ(ω, θi, u) = 0 1 2"ω c ± vF sin θisω2 c2 − ω < 2mc2 η(θi) 4m2c2 2η2(θi)# . (20) (21) (22) is satisfied, Eq. (20) does not possess real roots and the function χ is always positive. In this case the first line in Eq. (12) is applicable and N−1(ω, θi, u) =pχ(ω, θi, u), where χ is defind in Eq. (19). Now we consider the opposite condition mc ≤ u < ∞, 2mc2 η(θi) , ω ≥ 7 (23) (24) when Eq. (20) has two different real roots u(±) defined in Eq. (21) [they coincide in the case of equality in Eq. (24)]. It is easily seen that under the condition (24) u(−) ≥ mc/. Thus, in the regions mc/ ≤ u ≤ u(−) and u ≥ u(+) the function χ is nonnegative and the quantity N (−1) is given by the first line of Eq. (12). In so doing Q−1 > 0 in the latter interval whereas it can be both negative and positive in the former. As to the region u(−) < u < u(+), here the function χ is negative and, as a consequence, N (−1) is given by the second line of Eq. (12). Combining all these facts together, for the quantity N (−1) one obtains N (−1)(ω, θi, u) =  signQ−1pχ(ω, θi, u), mc −ip−χ(ω, θi, u), pχ(ω, θi, u), ≤ u ≤ u(−), u(−) < u < u(+), u ≥ u(+). Simple analysis using Eqs. (13) and (21) shows that for u ≤ u(−) within the interval 2mc2 η(θi) ≤ ω < 2mc2 η2(θi) we have signQ−1 = +1, whereas within the interval ω > 2mc2 η2(θi) (25) (26) (27) it holds signQ−1 = −1. Equations (14) -- (16), (18), (19), (23) and (25) are convenient for both analytic and nu- merical calculations using the polarization tensor of gapped graphene. To apply them for calculations of the reflection coefficients and reflectivities, it is convenient also to present an explicit expression for the quantity ∆T Π(ω, θi). It is obtained from Eqs. (3), (14) and (15) and has the form ∆T Π(ω, θi) = 16α v2 F Z ∞ mc/ du 1 eβu + 1"ω2 c2 + 1 2 Xλ=±1 M (λ)(ω, θi, u) N (λ)(ω, θi, u)# , where the quantity M (λ) is given by M (λ)(ω, θi, u) = ω2 c2 (cid:20)η2(θi)(cid:16)2u + λ ω c(cid:17)2 + 4v2 F (cid:16)mc (cid:17)2 sin2 θi(cid:21) . (28) (29) The denominator N (λ) in Eq. (28) is presented in Eqs. (18) and (25). In the limiting case m → 0 the above expressions take a more simple form used in Ref. [38] to investigate the reflectivity properties of gapless graphene. 8 III. REFLECTIVITY OF GAPPED GRAPHENE AT ZERO TEMPERATURE At T = 0 the polarization tensor of graphene is given by Eqs. (6) and (7). We consider first the region of frequencies (22). Here, the first line of Eq. (7) is applicable which can be rewritten in the form Φ(ω, θi) = 4mc(cid:20)1 − 2mc2 ωη (cid:18)1 + 2ω2η2 4m2c4 (cid:19) arctanh ωη 2mc2(cid:21) . (30) This is real function in contrast with the case of gapless graphene [in the latter case the function Φ is pure imaginary, see Eq. (8)]. Substituting Eq. (6) with the real function Φ in Eq. (2), one obtains the reflection coef- ficients and reflectivities of gapped graphene at sufficiently low frequencies RTM(ω, θi) ≡ rTM(ω, θi)2 = α2 cos2(θi)Φ2(ω, θi) 42 ω2 c2 η4(θi) + α2 cos2(θi)Φ2(ω, θi) , RTE(ω, θi) ≡ rTE(ω, θi)2 = α2Φ2(ω, θi) 42 ω2 c2 cos2(θi) + α2Φ2(ω, θi) , (31) where Φ is defined by Eq. (30). It is seen that at the normal incidence (θi = 0) RTM(ω, 0) = RTE(ω, 0), (32) as it should be. A. Asymptotic results Now we consider the asymptotic regime of very small frequencies satisfying the condition ωη(θi) 2mc2 ≪ 1. Expanding Eq. (30) in powers of small parameter (33), one obtains Φ(ω, θi) ≈ − 42ω2η2(θi) 3mc3 . (33) (34) Substituting this in Eq. (31) with account of Eq. (33), we arrive at RTM(ω, θi) ≈ 4 9 α2 cos2(θi) 2ω2 m2c4 , RTE(ω, θi) ≈ 4α22ω2 9m2c4 cos2(θi) . (35) Note that in Eq. (35) and below we have also used η(θi) ≈ 1. 9 As can be seen in Eq. (35), RTM(ω, θi), RTE(ω, θi) → 0 when ω → 0. (36) This is different from the case of gapless graphene at zero temperature. In the latter case, substituting Eq. (8) in Eq. (6), one finds Π(0) 00 (ω, θi) = iαπ ω c sin2 θi, Π(0)(ω, θi) = −iαπ ω3 c3 sin2 θi. Substituting these equations in Eq. (2), after simple calculation, we obtain RTM(θi) = α2π2 cos2 θi (απ cos θi + 2)2 , RTE(θi) = α2π2 (2 cos θi + απ)2 . (37) (38) Note that both reflectivities in Eq. (38) do not depend on the frequency and, thus, take nonzero values at ω = 0, as opposed to Eq. (36) for a gapped graphene. We are coming now to the asymptotic region ω → 2mc2 η(θi) , (39) where the quantity on the left-hand side of Eq. (33) approaches unity from the left. It is convenient to present the frequency in the form ω = 2mc2 η(θi)(cid:0)1 − 2e−τ(cid:1) , where the limiting case (39) corresponds to τ → ∞. Substituting Eq. (40) in Eq. (30), one finds that at τ → ∞ Φ(ω, θi) ≈ 4mc(1 − τ ) ≈ −4mcτ. Then from Eq. (31) we conclude that in the limiting case (39) RTM(ω, θi) → 1, RTE(ω, θi) → 1. (40) (41) (42) The next region to consider is given by Eq. (24). Here, the second line of Eq. (7) is applicable, where the function Φ has both real and imaginary parts ωη 2c (cid:18)1 + 4m2c4 2ω2η2(cid:19) arctanh 2mc2 ωη (cid:21) , (43) ReΦ(ω, θi) = 4(cid:20)mc − ImΦ(ω, θi) = − πωη c (cid:18)1 + 4m2c4 2ω2η2(cid:19) . 10 Using Eqs. (2) and (6), we find the reflectivities in the form RTM(ω, θi) = RTE(ω, θi) = α2[Re2Φ(ω, θi) + Im2Φ(ω, θi)] c η2 − α cos θiImΦ(ω, θi)(cid:3)2 , α2 cos2(θi)[Re2Φ(ω, θi) + Im2Φ(ω, θi)] α2 cos2(θi)Re2Φ(ω, θi) +(cid:2)2 ω α2Re2Φ(ω, θi) +(cid:2)2 ω c cos θi − αImΦ(ω, θi)(cid:3)2 . Now we consider the asymptotic behavior of the reflectivities at high frequencies In this case Eq. (43) leads to ωη(θi) 2mc2 ≫ 1. ReΦ(ω, θi) ≈ − 64m3c5 32ω2η2(θi) → 0 when ω → ∞, ImΦ(ω, θi) ≈ −π ω c η(θi), (44) (45) (46) i.e., to the same result as for a gapless graphene [see Eq. (8)]. Then, the reflectrivities are again given by Eq. (38) or, neglecting by the small terms, as compared to unity, by RTM(θi) ≈ α2π2 cos2 θi (2 + απ cos θi)2 , RTE(θi) ≈ α2π2 (2 cos θi + απ)2 . (47) Note that these reflectivities do not depend on the frequency and on the mass-gap parameter. The remaining asymptotic region is given by Eq. (39) when the quantity on the left-hand side of Eq. (45) approaches unity from the right. In this case from the second line of Eq. (43) one obtains ImΦ(ω, θi) ≈ −2π ω c η(θi). (48) The asymptotic behavior of the real part of Φ in the first line of Eq. (43) can be found similar to Eqs. (40) and (41) with the same results for the reflectivity properties as in Eq. (42). B. Numerical computations We are coming to numerical computations of the reflectivity properties of gapped graphene at zero temperature over wide frequency range of incident electromagnetic waves. Computations are performed by using Eqs. (2), (6) and (7). Taking into account that due to the conditions vF ≪ 1, η ≈ 1 the angular dependences of reflectivities are similar to those for gapless graphene [38], computations are performed at the normal incidence. In 11 Fig. 1 the three lines from right to left show the computational results for the TM and TE reflectivities (32) at the normal incidence calculated as functions of frequency for the graphene sheets with mass-gap parameter m equal to 0.1, 0.05, and 0.01 eV, respectively. The double logarithmic scale is used allowing to cover the frequency range from 1 µeV to 20 eV. The linear dependence of the logarithm of reflectivities on the logarithm of frequency at low frequencies satisfying the condition (33) corresponds to the asymptotic regime where Eq. (35) is applicable. At the frequency value ω = 2mc2 , (49) which is different for different m [recall that at the normal incidence η(0) = 1], the nar- row resonances are observed in Fig. 1. Under the condition (39) these resonances satisfy Eq. (42). At high frequencies satisfying the condition (45) the asymptotic equations (47) are applicable. Note that at θi = 0 one can neglect by απ, as compared with 2 in the denominators. It would be interesting to investigate in more detail an immediate vicinity of the resonance frequency (49), where the above asymptotic expressions do not apply. As an example, in Fig. 2 we again plot the reflectivity at the normal incidence (32) as a function of frequency for the graphene sheet with m = 0.1 eV . Now the computational results are plotted over a narrow frequency region from 0.14 to 0.30 eV using the natural scale in the frequency axis. As is seen in Fig. 2, even over a so narrow range, the reflectivity of graphene varies by four orders of magnitude, whereas the width of the resonance peak remains unresolved. In fact the half-width of this resonance can be found analytically from the asymptotic representation (41). Substituting Eq. (41) in Eq. (31) taken at θi = 0, we find RTM(ω, 0) = RTM(ω, 0) ≈ α2τ 2 2ω2 4m2c4 + α2τ 2 . (50) Taking into account Eq. (49), one obtains that RTM (TE) ≈ 1/2 when α2τ 2 = 1, i.e., for τ = 1/α ≈ 137. Then, using Eq. (40), we conclude that the half-width of the resonance is equal to 4e−137 mc2 . (51) ∼ 10−59 mc2 This half-width is too small and makes the resonance under consideration unobservable. However, at zero temperature it might be possible to observe the change in the reflectivity of graphene by a factor of ten with increasing frequency of the incident light. Thus, the 12 reflectivity increases from 0.00013 to 0.00134 when ω increases from 0.16 to 0.199 eV. In a similar way, with further increase of frequency from 0.201 to 0.3 eV the reflectivity of graphene changes from 0.00141 to 0.000141. We consider now more detailly the region of frequencies from 0 to 0.4 eV, where the reflectivity varies by more than an order of magnitude for all values of m under discussion. Note that at room temperature (T = 300 K) it holds kBT ≈ 0.025 eV, i.e., the thermal energy is of the same order as mc2 for the smallest value of m. In Fig. 3 we plot the reflectivity of gapped graphene at the normal incidence as a function of frequency by the three lines from right to left for the mass-gap parameter m = 0.1, 0.05, and 0.01 eV, respectively. In this figure, both the reflectivity and frequency are plotted in their natural scales. Because of this, only the lower parts of the resonance peaks are shown. As is seen in Fig. 3, an impact of the mass-gap parameter of graphene on the reflectivity properties decreases with decreasing m and increasing frequency. For small mass-gap parameters m < 0.01 eV in the region ω > 2mc2 the reflectivity is approximately equal to that obtained for m = 0 at any temperature T ≤ 300 K [38]. The role of nonzero temperature in the case of gapped graphene is considered below. IV. INFLUENCE OF NONZERO TEMPERATURE ON THE REFLECTIVITY OF GAPPED GRAPHENE In this section, we consider the reflection coefficients (2), where the quantities Π00 and Π are given by Eq. (5), i.e., are defined at any nonzero temperature. In so doing, the zero- temperature contributions are presented in Eqs. (6) and (7) and the thermal corrections to them are given in Eqs. (14) and (28) with all necessary notations in Eqs. (15) -- (19), (23), (25) and (29). 13 ×  − + (2cu + ω)2 − v2 F ω2 sin2 θi r(2cu + ω)2 + v2 F ω2(cid:16) 4m2c4 F ω2 sin2 θi 2ω2η2 − 1(cid:17) sin2 θi     2ω2η2 − 1(cid:17) sin2 θi F ω2(cid:16) 4m2c4 (2cu − ω)2 − v2 r(2cu − ω)2 + v2 (52) . A. Asymptotic results We begin with thermal correction to the polarization tensor of graphene calculated at frequencies satisfying Eq. (22). From Eqs. (14), (15), (18), (19) and (23) one obtains ∆T Π00(ω, θi) = 16α v2 F Z ∞ mc/ du eβu + 1 1 + 1 2ωη   Now we impose a more stringent than Eq. (22) requirement, i.e., ω . mc2.Under this requirement the quantity 2cu− ω cannot become too small and one can expand both square roots in Eq. (52) up to the first order of respective small parameters with the result ∆T Π00(ω, θi) ≈ ×(cid:26)1 + 2η(cid:20)−2 + 1 16α v2 F Z ∞ mc/ F sin2 θi v2 2 du eβu + 1 (cid:18) 4m2c4 2ω2η2 + 1(cid:19)(cid:18) ω 2cu + ω − ω 2cu − ω(cid:19)(cid:21)(cid:27) . Then, using the smallness of vF , we arrive at ∆T Π00(ω, θi) ≈ −8α sin2 θiZ ∞ mc/ du eβu + 1 1 + 4 m2c4 2 + ω2 4u2c2 − ω2! . (53) (54) Note that Eq. (54) is also valid under a less stringent condition (22) if one assumes that ω ≪ kBT . This is because the main contribution to the integral in Eq. (52) is given by cu ∼ kBT and hence the quantity 2cu − ω cannot become too small. Both application regions of Eq. (54) are used below to obtain approximate expressions for the polarization tensor and reflectivity of gapped graphene at nonzero temperature. It is convenient to introduce two dimensionless parameters and new integration variable according to mc2 kBT , µ ≡ ν ≡ ω kBT , v = cu kBT ≡ βu. (55) 14 Then Eq. (54) takes the form ∆T Π00(ω, θi) ≈ −8α sin2 θi kBT c Z ∞ µ dv ev + 1(cid:18)1 + 4µ2 + ν2 4v2 − ν2(cid:19) . (56) We consider, first, Eq. (56) under its validity condition ν ≪ 1 and assume that µ ≪ 1 as well. Taking into account that the main contribution to the integral in Eq. (56) is given by v ∼ 1, one can neglect by the small second term in the round brackets. Then, after the integration, we arrive at ∆T Π00(ω, θi) ≈ −8α kBT c sin2 θi ln(cid:0)1 + e−µ(cid:1) ≈ −8α ln 2 kBT c sin2 θi. (57) The latter thermal correction does not depend on m and coincides with that obtained in Eq. (89) of Ref. [38] for a gapless graphene. Note, however, that for a gapped graphene at T = 0 K from Eqs. (6) and (34), i.e., in the case of small frequencies, one obtains Π(0) 00 (ω, θi) ≈ 4 3 α sin2 θi 2ω2 mc3 . (58) This is real quantity, as opposed to the case of m = 0, where Π(0) 00 at small frequencies is pure imaginary. As is seen using Eq. (22) and the conditions µ, ν ≪ 1, Π(0) the TM reflectivity of graphene from Eqs. (2) and (57) we find 00 ≪ ∆T Π00. As a result, for RTM(ω, θi) ≈ 16α2 ln2 2(kBT )2 cos2 θi 2ω2 + 16α2 ln2 2(kBT )2 cos2 θi . Similar approximate calculation with account of Eqs. (6) and (34) results in ∆T Π(ω, θi) ≈ 8α ln 2 ω2 c2 kBT c sin2 θi ≫ Π(0)(ω, θi). Then, substituting this equation in Eq. (2), one obtains RTE(ω, θi) ≈ 16α2 ln2 2(kBT )2 2ω2 cos2 θi + 16α2 ln2 2(kBT )2 . (59) (60) (61) Unlike the case of gapped graphene at zero temperature, Eqs. (59) and (61) result in RTM (TE)(ω, θi) → 1 when ω → 0 (62) similar to the case of gapless graphene at nonzero T [38]. Below we demonstrate, both ana- lytically and numerically, that this is general property of the reflectivity of gapped graphene at T 6= 0. 15 Now we return to Eq. (56) derived under the condition ω . mc2 and consider it under additional assumptions µ ≫ 1 and ω ≪ mc2. Under these assumptions one can neglect by unity, as compared with ev, and by ν2, as compared with µ2 and v2. As a result, Eq. (56) takes the form ∆T Π00(ω, θi) ≈ −8α sin2 θi = −8α sin2 θi kBT µ2 kBT c Z ∞ c (cid:26)e−µ + µ2(cid:20)e−µ dve−v(cid:18)1 + v2(cid:19) + Ei(−e−µ)(cid:21)(cid:27) , µ µ (63) where Ei(z) is the exponential integral. Taking into account the asymptotic expansion [47] Ei(−x) = e−x(cid:20)− 1 x + 1 x2 + O(cid:18) 1 x3(cid:19)(cid:21) , one arrives at ∆T Π00(ω, θi) ≈ −16α sin2 θi Combining this expression with Eq. (58), we find kBT c e−mc2/(kB T ). Π00(ω, θi) ≈ 4α sin2 θi ω c (cid:20) 1 3 2ω mc2 − 4 kBT ω e−mc2/(kB T )(cid:21) , (64) (65) (66) i.e., the thermal correction is exponentially small. Substituting this equation [and similar for Π(ω, θi)] in Eq. (2), we again obtain Eq. (62) in the limiting case of vanishing frequency. As is seen from Eq. (66), the thermal correction to the 00 component of the polarization tensor is negative, whereas the zero-temperature contribution to it is positive. Taking into account that they can be of the same order of magnitude, we find the value of frequency ω0 where the quantity Π00 vanishes ω2 0 = 12 mc2 2 kBT e−mc2/(kB T ). (67) At room temperature (kBT = 0.025 eV) only the maximum value of the mass-gap parameter under consideration here (m = 0.1 eV) leads to µ = 4 and can be considered as close to the application region of the approximation used. For these values of parameters we find from Eq. (67) the value ω0 = 0.0234 eV, which is more or less in agreement with the second application condition of our approximation (ω ≪ mc2). Below we compare the obtained approximate value of ω0 with the results of numerical computations and find rather good agreement. Just now we only note that vanishing of Π00(ω0, θi) leads in accordance with 16 Eq. (2) to vanishing of the reflectivities of graphene RTM (TE)(ω0, θi) at the frequency ω0 6= 0. This does not happen either for a gapless graphene at any temperature or for a gapped graphene at T = 0 K. In the next subsection we consider the effect of vanishing reflectivity of graphene in more details by means of numerical computations. Here, we continue with the approximate analytic expressions for the polarization tensor and consider the frequency region of higher frequencies defined in Eq. (24). We start with the imaginary part of the thermal correction (14) which arises only for λ = −1 and is obtained from the first line of Eq. (15) and the second line of Eq. (25): Im∆T Π00(ω, θi) = where 8α v2 F Z u(+) u(−) du eβu + 1 F ω2 sin2 θi (2cu − ω)2 − v2 F ω2 sin2 θi A(ω, θ) − (2cu − ω)2 ωηqv2 A(ω, θi) ≡ 1 − 4 m2c4 2ω2η2(θi) , (68) (69) and u(±) are defined in Eq. (21). Note that in the case of equality in Eq. (24) the imaginary part (68) vanishes. It is convenient to introduce new dimensionless integration variable Then Eq. (68) takes the form τ = 2cu − ω vF ω sin θipA(ω, θi) . Im∆T Π00(ω, θi) = −4α ω c sin2 θi η Z 1 −1 dτ eω/(2kB T )eγτ + 1 1 − A(ω, θi)τ 2 √1 − τ 2 , where γ ≡ γ(ω, θi) = vF sin θipA(ω, θi) ω 2kBT . (70) (71) (72) The integral in Eq. (71) can be calculated approximately under different relationships between ω and T . We first assume that ω ≪ kBT [and also use η(θi) ≈ 1]. In this case both exponents in the denominator of Eq. (71) can be replaced with unities and after the integration one obtains Im∆T Π00(ω, θi) ≈ −απ ω c sin2 θie− ω 2kB T (cid:18)1 + 4m2c4 2ω2 (cid:19) ≈ −απ ω c sin2 θi(cid:18)1 + 4m2c4 2ω2 (cid:19) . (73) For m = 0 this expression coincides with Eq. (83) of Ref. [38] obtained for gapless graphene. 17 Now we consider the opposite case ω ≫ kBT . In this case it is convenient to rewrite Eq. (71) in the form Im∆T Π00(ω, θi) = −4α ω c sin2 θi η e− ω 2kB T Z 1 −1 dτ eγτ + e−ω/(2kB T ) 1 − A(ω, θi)τ 2 √1 − τ 2 . (74) Now we can neglect by the second exponent in the denominator, as compared to the first, put η(θi) ≈ 1 and calculate the integral with the result 2kB T π(cid:20)I1 (γ) γ (cid:18)1 − Im∆T Π00(ω, θi) ≈ −4α 2ω2 I0 (γ)(cid:21) , 2ω2 (cid:19) + sin2 θie− ω 4m2c4 4m2c4 (75) ω c where In(z) is the modified Bessel function of the first kind. If the frequency is not too large, i.e., ω ≫ kBT but γ(ω, θi) ≪ 1, one can put I1(γ) ≈ γ/2 and I0(γ) ≈ 1. Substituting this in Eq. (75) one finds Im∆T Π00(ω, θi) = −2απ ω c sin2 θie− ω 2kB T (cid:18)1 + 4m2c4 2ω2 (cid:19) . (76) Note that for m = 0 this result agrees with the estimation in Eq. (70) of Ref. [38] made for a gapless graphene. We now turn our attention to the real part of the polarization tensor under the condition (24). It is given by Eq. (14), where all notations are contained in the first line of Eq. (15) for λ = ±1, in Eq. (18) and in the first and third lines of Eq. (25). Using the dimensionless quantities µ and ν and the integration variable v introduced in Eq. (55), the real part of the polarization tensor takes the form Re∆T Π00(ω, θi) = 16α v2 F kBT c (I1 + I2 + I3), (77) where the integrals I1,2,3 are defined as I1 ≡ Z v(−) µ + v(−) I2 ≡ Z v(+) I3 ≡ Z ∞ − v(+) dv dv ev + 1(1 − (2v − ν)2 − v2 p(2v − ν)2 − v2 ev + 1"1 − ev + 1(1 − (2v − ν)2 − v2 p(2v − ν)2 − v2 dv 18 F ν2 sin2 θi F ν2A sin2 θi 1 2νη " (2v + ν)2 − v2 p(2v + ν)2 − v2 F ν2A sin2 θi#) , F ν2 sin2 θi F ν2 sin2 θi F ν2A sin2 θi# , F ν2 sin2 θi F ν2A sin2 θi 1 1 2νη (2v + ν)2 − v2 p(2v + ν)2 − v2 2νη" (2v + ν)2 − v2 p(2v + ν)2 − v2 F ν2A sin2 θi#) . F ν2 sin2 θi (78) Here, the integration limits are v(±) ≡ βu(±) and u(±) are presented in Eq. (21), η ≡ η(θi) is defined below Eq. (6) and A ≡ A(ω, θi) is introduced in Eq. (69). We consider first the region of relatively low frequencies ω ≪ kBT (ν ≪ 1). Note that under the condition (24) assumed now the inequality ν ≪ 1 can be valid only if µ ≪ 1 is valid as well. Now we take into account that the main contributions to all integrals (78) are given by v ∼ 1 and that in our case the inequalities ν, µ ≪ 1 lead to v(±) ≪ 1. Therefore only the integral I3 contains the values v ∼ 1 in the integration region and, thus, I3 alone determines the value of Re∆T Π00 in this case. Under the integral I3 one can expand in F /c2 because for v ∼ 1 we have 2v ≫ ν and the quantity F ≡ v2 powers of a small parameter v2 2v − ν cannot vanish. As a result, one obtains Re∆T Π00(ω, θi) ≈ −8α = −8α kBT c kBT c sin2 θiZ ∞ v(+) dv ev + 1 sin2 θi ln(cid:2)1 + e−ω/(2kB T )(cid:3) ≈ −8α ln 2 (79) kBT c sin2 θi. Comparing this with Eq. (57), we conclude that under the condition µ ≪ 1 (mc2 ≪ kBT ) there is a smooth transition between the frequency regions satisfying the inequalities (22) and (24). At this point we can present the complete polarization tensor of graphene under the conditions mc2 ≪ ω ≪ kBT . In this case the zero-temperature contribution Π(0) 00 is given by Eqs. (6) and (46), whereas the imaginary and real parts of thermal correction are found in Eqs. (73) and (79). The results is Π00(ω, θ) ≈ −8α ln 2 kBT c sin2 θi. (80) It is seen that under our conditions the imaginary part of this expression is much smaller than the real one and can be neglected. Taking into account the similar result for Π(ω, θi) and using Eq. (2), for the reflectivities of graphene we return back to Eqs. (59) and (61) obtained above in the frequency region ω ≪ mc2. We will return to this fact below when discussing the results of numerical computations. Now we consider the opposite case ω ≫ kBT , i.e., ν ≫ 1. In the region of frequencies satisfying Eq. (24) this is possible for both mc2 > kBT and mc2 < kBT . Then we have v(±) ≫ 1 and, thus, the major contribution to ∆T Π00 is given by the integral I1, where the upper integration limit v(−) can be replaced with infinity. Taking into account that for v ∼ 1 19 we have ν ≫ 2v, it is easy to expand the function under the integral I1, as was done above, and arrive at Re∆T Π00(ω, θi) ≈ 32α kBT c sin2 θiZ ∞ µ dv ev + 1(cid:18) v2 ν2 + m2c4 2ω2(cid:19) . After the integration in Eq. (81) we have kBT c sin2 θi Re∆T Π00(ω, θi) ≈ 64α ×"(cid:18)mc2 ω (cid:19)2 ln(1 + e−µ) − mc2kBT ω (cid:19)2 2ω2 Li2(−e−µ) −(cid:18)kBT Li3(−e−µ)# , where Lin(z) is the polylogarithm function. (81) (82) If we additionally assume that mc2 ≪ kBT , we can neglect by the first and second terms on the right-hand side of Eq. (82), as compared to the third one. Then, by putting exp(−µ) ≈ 1, one arrive at Re∆T Π00(ω, θi) ≈ −64α (kBT )3 c(ω)2 sin2 θi Li3(−1). (83) Taking into account that Li3(−1) = −3ζ(3)/4, where ζ(z) is the Riemann zeta function, this is in agreement with Eq. (67) of Ref. [38] obtained for m = 0. Then, the complete polarization tensor and also the reflectivities of gapped graphene in this case are the same for a gapless one [we remind that under our conditions the imaginary part of the thermal correction to the 00 component of the polarization tensor is exponentially small according to Eq. (76), whereas the dominant contributions to the imaginary parts of Π(0) 00 and Π(0) do not depend on m]. Under another additional condition, kBT ≪ mc2, the main contribution in Eq. (82) is given by the first term on the right-hand side and the result is Re∆T Π00(ω, θi) ≈ 96α kBT m2c3 (ω)2 sin2 θie− mc2 kB T . (84) In this case both the real and imaginary parts of the thermal correction are exponentially small [see Eq. (76)], and we again return to the same reflectivities, as were obtained for a gapless graphene in Ref. [38]. B. Numerical computations Now we perform numerical computations of the reflectivity of graphene at the normal incidence (θi = 0) defined via the reflection coefficients by the first equalities in Eq. (31). 20 Computations are performed using Eqs. (5) -- (7), (14) -- (19), (23) and (25). As noted in Sec. IIIB, under the conditions vF ≪ 1, η ≈ 1 the angular dependences of reflectivities are similar to the case of gapless graphene. They are presented in the above analytic expressions and in Figs. 3 and 5 of Ref. [38]. Taking into account the possibility of comparison with the measurement data, we consider the temperature at a laboratory, T = 300 K, and cover the maximally wide range of frequencies. In Fig. 4 we present the computational results for the reflectivities RTM = RTE as functions of frequency by the two solid lines 1 and 2 for graphene sheets with the mass-gap parameter m equal to 0.1 and 0.05 eV, respectively. The double logarithmic scale is used. The dashed lines 1 and 2 reproduce from Fig. 1 the respective results at zero temperature. As is seen in Fig. 4, the reflectivity of graphene depends heavily on the fact that temperature is not equal to zero. First of all note that the reflectivity at zero frequency is equal to unity rather than zero, as it was at zero temperature (see Figs. 1 and 3). This fact was already established analytically in the asymptotic expressions (59), (62) and (66) valid at small frequencies. Second, as was noted above, in the region of sufficiently small frequencies the thermal correction ∆Π00 in Eq. (65) is real and negative. Taking into account that at small frequen- cies Π(0) 00 is real and positive [see Eqs. (6) and (30)], it is possible that Π00 turns into zero at some frequency ω0 depending on m and T . As is noted above, the reflection coefficients and reflectivities also vanish at the frequency ω0. From Fig. 4 it is seen that RTM (TE)(ω0, θi) = 0 at ω0 ≈ 0.02159 and 0.03033 eV for graphene sheets with m = 0.1 and 0.05 eV, respectively. An approximate analytic expression for ω0 is obtained in Eq. (67). As discussed above, at room temperature it can be applied only for the largest value m = 0.1 eV and even in this case it is slightly outside the validity region of the approximation used. However, comparing the numerical (0.02159 eV) and analytic (0.0234 eV) results for ω0, one can conclude that the approximate expression (67) leads to only 8.4% error. In the region of higher frequencies satisfying Eq. (24) for the mass-gap parameters con- sidered in Fig. 4 the computational results are similar to those obtained at T = 0 K. Here, we observe the resonance peaks at the border frequencies and the asymptotic regime at high frequencies, which does not depend on ω. In this region the dashed line 1 coincides with the solid line 1 (m = 0.1 eV), and the dashed line 2 (m = 0.05 eV) only minor deviates from the solid line 2. One can conclude that in the range of low frequencies satisfying the condition 21 (22) the reflectivity properties of gapped graphene at zero and room temperature are quite different. Now we perform numerical computations in order to investigate these differences in more details. In Fig. 5, we present the reflectivity of graphene with m = 0.1 and 0.05 eV at T = 300 K at the normal incidence in an immediate vicinity of the frequency ω0 where it drops to zero (the lines 1 and 2, respectively). For this purpose the natural scale is used along the frequency axis. The dashed lines 1 and 2 show the respective results at T = 0 K. As is seen in Fig. 5, near the frequencies ω0 it should be possible to observe the change in the reflectivity of graphene by several orders of magnitude. From Fig. 4 it can be seen that with decreasing m the interval between the frequency ω0, where the reflectivity vanishes, and the border frequency 2mc2/, where it is equal to unity, becomes more narrow. To illustrate this tendency, in Fig. 6 we plot the reflectivity of graphene with m = 0.001 eV at T = 300 K as a function of frequency in the double logarithmic scale (the solid line). From Fig. 6 it is seen that zero reflectivity is achieved at the frequency ω0 which is only slightly less than 0.002 eV. Because of this, the minimum and maximum values of the reflectivity virtually coincide. For not too high frequencies satisfying the condition ω ≪ kBT and with exception of some vicinity of the border frequency 2mc2/, the solid line in Fig. 6 is in a good agreement with the asymptotic expressions (59) and (61) taken at θi = 0. In the same figure the reflectivity of graphene with m = 0.001 eV at zero temperature is shown by the dashed line. In this case the reflectivity has only the maximum value at 0.002 eV and goes to zero with vanishing frequency. Note that all the above results related to the region of frequencies (22) are valid only for graphene with nonzero mass-gap parameter. Specifically, the reflectivity of gapless graphene does not vanish at any frequency [38]. Finally, we perform numerical computations of the frequency ω0, where the reflectivity of graphene vanishes, as a function of the mass-gap parameter m. The computational results in the region of m from 0.001 to 0.1 eV at T = 300 K are shown in Fig. 7 by the solid line. As is seen in Fig. 7, with increasing m, the frequency ω0 increases, achieves the maximum value at m ≈ 0.03 eV and then gradually decreases. In the same figure, the border frequency ω = 2mc2/ is plotted by the dashed line. It is seen that the frequency interval between the dashed and solid lines becomes narrower with decreasing m (as already was concluded from Fig. 4 basing on only two values of m). The dashed line remains above the solid line 22 at all m and almost coincides with it at m . 0.015 eV. This is in accordance with Fig. 6, where for m = 0.001 eV the frequencies, where reflectivity of graphene takes the minimum and maximum values, cannot be discerned. V. CONCLUSIONS AND DISCUSSION In the foregoing, we have investigated the reflectivity properties of gapped graphene at both zero and nonzero temperature in the framework of the Dirac model. It was shown that the presence of nonzero mass-gap parameter has a profound effect on the reflectivity of graphene. To find this effect, we have employed the polarization tensor of graphene found in Ref. [38], which was further adapted and simplified for the case of gapped graphene. Specifically, it was shown that at zero temperature the reflectivities of gapped graphene go to zero when the frequency vanishes. This is not the case for a gapless graphene, where the reflectivities at zero temperature are nonzero and depend only on the fine structure constant and the angle of incidence. We have shown also that at nonzero temperature the reflectivities of gapped graphene go to unity with vanishing frequency, as it is for a gapless graphene. Another distinctive property of gapped graphene is that its reflectivities possess the narrow resonances having the maximum values equal to unity at the border frequency ω = 2mc2/[η(θi)]. This is the case at both zero and nonzero temperature and can be observed experimentally as a jump in the reflectivities of graphene when passing across the border frequency. At frequencies larger than the border frequency there is only minor influence of the mass-gap parameter on the reflectivities of graphene. The next remarkable property of gapped graphene at nonzero temperature is that in the vicinity of some frequency ω0 smaller than the border frequency its reflectivities drop to zero. This local minimum also can be observed as a jump in the reflectivities of graphene when passing across the frequency ω0. There is no such effect for a gapless graphene or for a gapped graphene at T = 0 K. It is caused by the fact that the thermal correction to the polarization tensor takes negative values. All the above effects have been derived and investigated analytically in different frequency regions. As a result, the approximate analytic expressions for the reflectivities of graphene and for the frequency ω0 have been found. Both the cases of the normal incidence and 23 an arbitrary incidence angle were considered. The analytic results have been illustrated by numerical computations over a wide range of frequencies from 1 µeV to 20 eV. From the comparison between approximate analytic and computational results, the exactness of the former was estimated. The developed theory of the reflectivity of gapped graphene at any temperature is based on the first principles of quantum electrodynamics. It can be used in many prospective applications of graphene, such as optical detectors, optoelectronic switches and other men- tioned in Sec. I. In future it would be interesting to generalize this theory for the case of nonzero chemical potential (see Ref. [48] for the polarization tensor of graphene taking the chemical potential into account) and to apply it to graphene-coated substrates. Acknowledgments The authors are grateful to M. Bordag for helpful discussions. [1] A. H. Castro Neto, F. Guinea, N. M. R. Peres, K. S. Novoselov, and A. K. Geim, Rev. Mod. Phys. 81, 109 (2009). [2] M. I. Katsnelson, Graphene: Carbon in two Dimensions (Cambridge University Press, Cam- bridge, 2012). [3] G. G´omez-Santos, Phys. Rev. B 80, 245424 (2009). [4] M. I. Katsnelson, K. S. Novoselov, and A. K. Geim, Nat. Phys. 2, 620 (2006). [5] D. Allor, T. D. Cohen, and D. A. McGady, Phys. Rev. D 78, 096009 (2008). [6] C. G. Beneventano, P. Giacconi, E. M. Santangelo, and R. Soldati, J. Phys A: Math. Theor. 42, 275401 (2009). [7] G. L. Klimchitskaya and V. M. Mostepanenko, Phys. Rev. D 87, 215011 (2013). [8] X. Jiang, Y. Cao, K. Wang, J. Wei, D. Wu, and H. Zhu, Surf. Coatings Technology 261, 327 (2015). [9] Springer Handbook of Nanomaterials, edited by R. Vajtai (Springer, Berlin, 2013). [10] S. Watcharotone, D. A. Dikin, S. Stankovich et al., Nano Lett. 7, 1888 (2007). [11] L. F. Dum´ee, L. He, Z. Wang et al., Carbon 87, 395 (2015). 24 [12] T. Kuila, S. Bose, P. Khanra, A. K. Mishra, N. H. Kim, and J. H. Lee, Biosens. Bioelectr. 29, 4637 (2011). [13] Yu. V. Bludov, M. I. Vasilevskiy, and N. M. R. Peres, Europhys. Lett. 92, 68001 (2011). [14] 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 320, 1308 (2008). [15] A. B. Kuzmenko, E. van Neumen, F. Carbone, and D. van der Marel, Phys. Rev. Lett. 100, 117401 (2008). [16] V. P. Gusynin, S. G. Sharapov, and J. P. Carbotte, Phys. Rev. Lett. 96, 256802 (2006). [17] V. P. Gusynin, S. G. Sharapov, and J. P. Carbotte, Int. J. Mod. Phys. B 21, 4611 (2007). [18] L. A. Falkovsky and S. S. Pershoguba, Phys. Rev. B 76, 153410 (2007). [19] T. Stauber, N. M. R. Peres, and A. K. Geim, Phys. Rev. B 78, 085432 (2008). [20] F. H. L. Koppens, D. E. Chang, and F. J. Garc´ıa de Abajo, Nano Lett. 11, 3370 (2011). [21] M. Bordag, I. V. Fialkovsky, D. M. Gitman, and D. V. Vassilevich, Phys. Rev. B 80, 245406 (2009). [22] I. V. Fialkovsky, V. N. Marachevsky, and D. V. Vassilevich, Phys. Rev. B 84, 035446 (2011). [23] M. Bordag, G. L. Klimchitskaya, and V. M. Mostepanenko, Phys. Rev. B 86, 165429 (2012). [24] M. Chaichian, G. L. Klimchitskaya, V. M. Mostepanenko, and A. Tureanu, Phys. Rev. A 86, 012515 (2012). [25] G. L. Klimchitskaya and V. M. Mostepanenko, Phys. Rev. B 87, 075439 (2013). [26] G. L. Klimchitskaya, V. M. Mostepanenko, and Bo E. Sernelius, Phys. Rev. B 89, 125407 (2014). [27] G. L. Klimchitskaya, U. Mohideen, and V. M. Mostepanenko, Phys. Rev B 89, 115419 (2014). [28] G. L. Klimchitskaya and V. M. Mostepanenko, Phys. Rev A 89, 052512 (2014). [29] G. L. Klimchitskaya and V. M. Mostepanenko, Phys. Rev. B 91, 045412 (2015). [30] G. L. Klimchitskaya and V. M. Mostepanenko, Phys. Rev B 91, 174501 (2015). [31] D. Drosdoff and L. M. Woods, Phys. Rev. B 82, 155459 (2010). [32] D. Drosdoff and L. M. Woods, Phys. Rev. A 84, 062501 (2011). [33] Bo E. Sernelius, Europhys. Lett. 95, 57003 (2011). [34] Bo E. Sernelius, Phys. Rev. B 85, 195427 (2012). [35] A. D. Phan, L. M. Woods, D. Drosdoff, I. V. Bondarev, and N. A. Viet, Appl. Phys. Lett. 101, 113118 (2012). 25 [36] M. Bordag, G. L. Klimchitskaya, U. Mohideen, and V. M. Mostepanenko, Advances in the Casimir Effect (Oxford University Press, Oxford, 2015). [37] G. L. Klimchitskaya, U. Mohideen, and V. M. Mostepanenko, Rev. Mod. Phys. 81, 1827 (2009). [38] M. Bordag, G. L. Klimchitskaya, V. M. Mostepanenko, and V. M. Petrov, Phys. Rev. D 91, 045037 (2015); Phys. Rev. D93, 089907(E) (2016). [39] G. L. Klimchitskaya, C. C. Korikov, and V. M. Petrov, Phys. Rev. B 92, 125419 (2015); Phys. Rev. B 93, 159906(E) (2016). [40] S. A. Jafari, J. Phys.: Condens. Matter 24, 205802 (2012). [41] P. K. Pyatkovsky, J. Phys.: Condens. Matter 21, 025506 (2009). [42] V. P. Gusynin, S. G. Sharapov, and J. P. Carbotte, New J. Phys. 11, 095013 (2009). [43] V. P. Gusynin and S. G. Sharapov, Phys. Rev. B 73, 245411 (2006). [44] A. I. Akhiezer and V. B. Berestetskii, Quantum Electrodynamics (Interscience, New York, 1965). [45] S. S. Schweber, An Introduction to Relativistic Quantum Field Theory (Dover, New York, 2005). [46] M. I. Katsnelson, Eur. Phys. J. B 51, 157 (2006). [47] I. S. Gradshtein and I. M. Ryzhik, Table of Integrals, Series and Products (Academic Press, New York, 1980). [48] M. Bordag, I. Fialkovsky, and D. Vassilevich, Phys. Rev. B 93, 075414 (2016). 26 1 0.001 10-6 ) 0 = i θ ( ) E T ( M T R 10-9 10-6 10-4 0.01 ω (eV) 1 FIG. 1: (Color online) The reflectivity of gapped graphene at T = 0 K at the normal incidence is shown as a function of frequency by the three lines from right to left for the mass-gap parameter m=0.1, 0.05, and 0.01 eV, respectively. 27 1 0.1 0.01 ) 0 = i θ ( ) E T ( 0.001 M T R 10-4 0.15 0.20 0.25 ω (eV) FIG. 2: (Color online) The reflectivity of graphene with the mass-gap parameter m=0.1 eV at T = 0 K at the normal incidence is shown as a function of frequency in the vicinity of the resonance frequency ω = 2mc2/. 28 10 8 6 4 2 ) 0 = i θ ( ) E T ( M T R 4 0 1 0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 ω (eV) FIG. 3: (Color online) The reflectivity of gapped graphene at T = 0 K at the normal incidence is shown as a function of frequency using the natural scales in the vicinity of resonance frequen- cies. The three lines from right to left are for the mass-gap parameter m=0.1, 0.05, and 0.01 eV, respectively. 29 1 0.01 10-4 10-6 ) 0 = i θ ( ) E T ( M T R 10-8 1 2 2 1 10-5 10-4 0.001 0.01 ω (eV) 0.1 1 FIG. 4: (Color online) The reflectivity of gapped graphene at T = 300 K at the normal incidence is shown by the solid lines 1 and 2 as a function of frequency for the mass-gap parameters m = 0.1 and 0.05 eV, respectively. The dashed lines 1 and 2 reproduce the respective results at T = 0 K. 30 10-6 10-8 2 1 1 2 ) 0 = i θ ( ) E T ( 10-10 M T R 10-12 0.02 0.04 0.06 ω (eV) 0.08 0.10 FIG. 5: (Color online) The reflectivity of gapped graphene at T = 300 K at the normal incidence is shown in the vicinity of frequencies where it drops to zero by the solid lines 1 and 2 as a function of frequency for the mass-gap parameters m = 0.1 and 0.05 eV, respectively. The dashed lines 1 and 2 reproduce the respective results at T = 0 K. 31 1 0.01 10-4 10-6 ) 0 = i θ ( ) E T ( M T R 10-8 10-5 10-4 0.001 0.01 0.1 ω (eV) FIG. 6: The reflectivity of gapped graphene with m = 0.001 eV at the normal incidence is shown as a function of frequency by the solid and dashed lines at T = 300 K and T = 0 K, respectively. 32 0.20 0.15 ) V e ( 0 ω 0.10 0.05 0.00 0.00 0.02 0.04 0.06 m (eV) 0.08 0.10 FIG. 7: (Color online) The frequency, where the reflectivity of graphene at T = 300 K at the normal incidence vanishes, is shown by the solid line as a function of the mass-gap parameter. The dashed line shows the border frequency 2mc2/. 33
1003.3162	1	1003	2010-03-16T15:09:19	Macroscopic quantum state in a semiconductor device	[ "cond-mat.mes-hall" ]	We show how nanostructuring of a metallic gate on a field-effect transistor (FET) can lead to a macroscopic, robust and voltage controlled quantum state in the electron channel of a FET. A chain of triple quantum dot molecules created by gate structure realizes a spin-half Heisenberg chain with spin-spin interactions alternating between ferromagnetic and anti-ferromagnetic. The quantum state is a semiconductor implementation of an integer spin-one antiferromagnetic Heisenberg chain with a unique correlated ground state and a finite energy gap, originally conjectured by Haldane.	cond-mat.mes-hall	cond-mat	Macroscopic quantum state in a semiconductor device Yun-Pil Shim,1, ∗ Anand Sharma,1, 2 Chang-Yu Hsieh,1, 2 and Pawel Hawrylak1 1Institute for Microstructural Sciences, National Research Council of Canada, Ottawa, Canada K1A 0R6 2Department of Physics, University of Ottawa, Ottawa, Canada K1N 6N5 (Dated: November 13, 2018) Abstract We show how nanostructuring of a metallic gate on a field-effect transistor (FET) can lead to a macroscopic, robust and voltage controlled quantum state in the electron channel of a FET. A chain of triple quantum dot molecules created by gate structure realizes a spin-half Heisenberg chain with spin-spin interactions alternating between ferromagnetic and anti-ferromagnetic. The quantum state is a semiconductor implementation of an integer spin-one antiferromagnetic Heisenberg chain with a unique correlated ground state and a finite energy gap, originally conjectured by Haldane. PACS numbers: 73.21.La,73.22.-f,03.67.-a 0 1 0 2 r a M 6 1 ] l l a h - s e m . t a m - d n o c [ 1 v 2 6 1 3 . 3 0 0 1 : v i X r a 1 Haldane predicted that spin-one antiferromagnetic Heisenberg chain has a unique ground state with a finite gap [1]. Affleck et al. obtained an exact analytical solution of the ground state with a finite gap in an isotropic spin-one chain with a special biquadratic interaction [2] and suggested that the ground state of the Heisenberg spin-one chain has the valence-bond-solid state character [3]. These predictions were subsequently confirmed by numerical calculations [4 -- 7] and experimental studies [8 -- 10] in quasi-one-dimensional complex compounds. In this work, we propose a way of realizing an effective spin-one chain in a system of lateral gated quantum dots (QD) in a field effect transistor (FET) semiconductor structure. Quantum dots [11] are defined by confinement potentials in all three dimensions and often called artificial atoms due to their similarity to the real atoms. Many QDs can be connected in a network to form more complex artificial structures. In a double quantum dot molecule, the ground state of two-electron system is always spin singlet without external magnetic field. Therefore, the effective spin-spin interaction between two localized electrons in respective quantum dots is always antiferromagnetic. It was shown by some of us that the spin-spin interaction can be tuned, both in magnitude and sign, by bringing another quantum dot that contains two electrons to make a triple quantum dot (TQD) molecule [12]. We will call a system of coupled QDs an artificial molecule since the constituent dots are connected by the tunnelling which determines the state of the system. The presence of the doubly-occupied dot enables us to generate a triplet ground state due to the tunnelling between QDs. When all three dots are on resonance, a triple quantum dot system in triangular geometry with four electrons has a spin triplet ground state. As we lower the energy level of one of the dots, this dot will have two electrons and the other two dots will have single electron each. If the energy level of the doubly occupied dot is well below the energy levels of the other two dots, the system is similar to a double quantum dot with spin singlet ground state. Therefore, there is a critical value of the bias that changes the ground state between spin singlet and spin triplet. Near the critical value, the effective interaction between the two localized electrons in the singly occupied dots can be tuned by the bias, both in magnitude and sign. When tuned to be in triplet ground state with total spin S=1, the TQD molecule can be used as a building block of a spin-one Heisenberg chain system. The architecture of the proposed linear chain of TQD molecules in a semiconductor device is shown in Fig. 1(a). By patterning the top metallic gate in a typical metal-oxide-semiconductor field-effect-transistor (MOSFET) device, one 2 can create local potential minima in the two-dimensional electron gas (2DEG) and each potential minimum represents a quantum dot. Progress in the experimental technique allows precise control of the potential profile of the 2DEG and the number of electrons in each QD [13 -- 17]. For individual control of each QD, many plunger gates will be needed and the gate pattern would be more complicated than shown in Fig. 1(a). Each TQD molecule consists of three quantum dots of which one contains two electrons and the other two dots contain single electron [see Fig. 1(b)]. The doubly occupied dots in neighbouring TQD molecules are in the opposite position to minimize the effects of the Coulomb interaction of those dots. The physics of a quantum dot network is well described by the Hubbard model [18, 19]. For simplicity, we will assume that the on-site Coulomb repulsion U is the same for all QDs, the Coulomb interaction between two electrons in different dots in the same molecule is V , and two electrons in different molecules interact via Coulomb interaction V ′ only when they are in neighbouring QDs. The intra-molecular tunnelling is t, and the inter-molecular tunnelling t′ is only allowed between two neighbouring dots. In an isolated TQD molecule, we can occupy one of the dots with two electrons and the other two dots with single electron by lowering the potential energy of the doubly occupied dot (or, equivalently, raising the potential energy of the singly occupied dots). In the present TQD chain system, the inter- molecular Coulomb interaction V ′ acts as if it effectively raises the energy levels of the singly occupied dots compared to the doubly occupied dots. By controlling V ′ the ground state of a TQD molecule can be changed from singlet to triplet. The only exceptions are two end dots which have no neighbouring molecules. Thus we assume that the energy levels of the two end dots are raised by V ′ through gate control. In second quantized form, the Hubbard Hamiltonian with single level per site is given by bHHubbard = X i,σ iσciσ + X εic† X iσcjσ + X tijc† U ni↑ni↓ + i6=j σ i ≡ bH0 + bHT + bHU + bHV , 1 2 X i6=j Vij ninj (1) where i, j are indices for QD sites, σ =↑, ↓ is the spin, niσ = c† iσciσ is the number of electrons with spin σ in QD i, and ni = ni↑ + ni↓ is the total number of electrons in QD i. tij = t and Vij = V for i and j in the same molecule and tij = t′ and Vij = V ′ if i and j are in different molecules but neighbouring each other. All the other Vij and tij are zero. The QD energy level is εi = ε0 for all i except for the first and last QDs where εi = ε0 + V ′. Most of the parameters are highly tunable but, in typical devices, the strong Coulomb repulsion is much 3 larger than tunnelling between dots. In those cases, the number of electrons in each dot is well defined and the tunnelling Hamiltonian bHT gives rise to effective interactions between well localized electrons. V ′=0.2, t=-0.05 and t′=-0.02 in unit of the effective Rydberg defined by Ry = m∗ In following numerical examples we use ε0 = 0, U=2.0, V =0.5, ee4/2ǫ22. e is the electron effective mass, e is the electron charge and ǫ is the dielectric constant. For m∗ GaAs, for example, Ry is about 6meV. These parameters are based on the first experimental realization of a lateral TQD molecule device [16] and the theoretical model for the charging diagram of the device [18]. The sign of the tunnelling elements t and t′ can be determined by comparing the Hubbard model and more microscopic linear-combination-of-harmonic- oscillator (LCHO) model [18, 19]. For a system with a charge configuration as shown in Fig. 2(a), we can obtain an ef- fective Hamiltonian corresponding to a Hamiltonian of a chain of localized electron spins [Fig. 2(b)] by treating the tunnelling Hamiltonian bHT as a perturbation for the parameter range t, t′ ≪ V, V ′ ≪ U [12]. The resulting spin-half chain has alternating ferromagnetic and antiferromagnetic interactions as envisaged in the works of ,e.g., Hida [20] and Hung and Gong [21]. The effective Hamiltonian is given by bHeff = E0 + E′ 0 + J NMX m=1 s3m−2 · s3m + J ′ NM −1X m=1 s3m · s3m+1 , (2) where E0 = 4NM ε0 + NM U + 5NM V + (NM + 1) V ′ , 2t2NM E′ 0 = − t2NM U − V V ′ − 4t3 V ′2 , − J = J ′ = 4t2 U − V 4t′2 U − V ′ . − t′2 (NM − 1) U − V ′ − t3NM V ′2 , (3) m is the index for TQD molecules and NM is the number of TQD molecules in the chain. The intra-molecular spin-spin interaction J consists of two terms. The first term is antiferro- magnetic (positive sign) superexchange term and the second term is ferromagnetic (negative sign) term which comes from third order processes involving the doubly occupied dot. This third order term is comparable to the second order superexchange term in magnitude, and for small enough V ′ the overall interaction J can be either ferromagnetic or antiferromag- netic depending on the values of the parameters. Near the critical value of V ′ where J = 0, 4 each molecule has one doubly occupied dot and two singly occupied dots, and the effective spin-spin interaction between the two localized spins changes between ferromagnetic and antiferromagnetic. The critical value of V ′ is c = pt(U − V ) , V ′ (4) and we need to tune V ′ to be smaller (larger) than V ′ c for (anti)ferromagnetic intra-molecular interaction. On the other hand, the inter-molecular spin-spin interaction between two neigh- bouring electrons is always antiferromagnetic due to the superexchange. This effective spin- half chain Hamiltonian gives a clearer picture of the system and describes the low energy physics as will be shown below. Another important advantage of this effective Hamiltonian is that it reduces the dimension of the Hilbert space significantly and allows us to study much longer chains by exactly solving the associated eigenvalue problem. The solution of the full Hubbard Hamiltonian is numerically demanding for chains with more than 5 molecules since the dimension of the Hilbert space rapidly exceeds 106. Figure 3 shows the lowest eigenvalues of the Hubbard model (green triangles) and the effective spin-half chain model (red circles) for a chain of up to 5 molecules using Jacobi-Davidson method implemented in PRIMME library [22]. Each eigenstate is labelled by the total spin S. The values of J = −5.83 × 10−3 and J ′ = 8.89 × 10−4 used in Eq. (2) are obtained from parameters of the microscopic model using Eq. (3). It is clear that the effective spin-half chain Hamiltonian describes the low energy levels of the Hubbard model quite well, reproducing the main features such as (i) the spin singlet (S = 0) ground state for even number of molecules and spin triplet (S = 1) ground state for odd number of molecules, (ii) the decrease of the gap between spin singlet and spin triplet states with the number of molecules, and (iii) appearance of the Haldane gap between the ground state and spin quintuplet (S = 2) excited states. In the limit- ing case of J/J ′ → −∞, i.e., if the intra-molecular ferromagnetic interaction is extremely strong compared to the inter-molecular antiferromagnetic interaction, each molecule will be in a triplet state and the system will be a chain of spin-one dimmers [Fig. 2(c)]. Thus the ground state will be separated from excited states by Haldane gap as the length of the chain increases. Even for modest values of J/J ′, the system will have a robust ground state with a finite energy gap. Figure 4 shows the formation of Haldane gap as function of increasing number NM of triple quantum dot molecules described by the Heisenberg Hamiltonian, Eq. (2). The 5 Haldane gap for NM up to 5 molecules is also shown in Fig. 3. Note that the energies in Fig. 3 and Fig. 4 are scaled with respect to the inter-molecular antiferromagnetic interaction J ′ of the spin-half chain. The results for spin S = 1/2 finite chain are compared with results for spin S = 1 chain in Fig. 4. The antiferromagnetic interaction J (1) AF in spin-one chain was chosen to be 1/4 of the inter-molecular antiferromagnetic interaction J ′ of the spin- half chain, because the interaction is only between two neighbouring spins in the spin-half chain [21]. Main panel of Fig. 4 shows the energy eigenvalues of S = 2 excited state with respect to the ground state and the inset shows the energy difference between S = 1 and S = 0 states. The triplet-singlet gap oscillates around zero since the ground state is spin singlet(triplet) for even(odd) number of molecules, and it goes to zero as the number of molecules increases. For long chains, the ground state is four-fold degenerate (S = 0, 1) and there is a finite gap between the ground state and the first excited state (S = 2). The four-fold degeneracy can be manipulated and removed by adding additional QDs with localized spin 1/2 electrons at both ends [5]. The Haldane gap is 0.41J (1) AF for infinitely long spin-one chain [5, 6], and the energy gap of infinite spin-half chain depends on both the ferromagnetic and antiferromagnetic interactions J and J ′ [20]. For J/J ′ → −∞, the effective antiferromagnetic interaction J (1) AF of the spin-one chain is exactly J ′/4. For finite J/J ′ the gap is larger than the Haldane gap and increases as the magnitude of J/J ′ decreases, reaching the maximum value at J/J ′ = 0. Around this point, the ground state is simply a chain of singlet dimers and the first excited state is "triplet wave" state [20]. With the Hubbard parameters used, we get J/J ′ = −6.56. As can be seen in Fig. 4, the energy gap of spin-half chain is larger than that of the spin-one chain due to the finite ferromagnetic intra-molecular interaction and it would reach a value about twice the Haldane gap of spin- one chain at infinite length [20]. Our estimate for material such as GaAs lead to the gap of infinite TQD chain with J/J ′ = −6.56 to be 1.07µeV , which corresponds to T = 12.4mK. The maximum gap for J/J ′ = 0 is J ′ or 62mK. In addition to changing the ratio J/J ′, one can further increase the gap and hence strengthen the robustness of the ground state through controlling the parameters that determine J ′ = 4t′2/(U − V ′) by building quantum dot networks using self-assembled quantum dots on nanotemplates [23]. For characteristic parameters for self-assembled quantum dots U=20meV, V ′=10meV, t′=10meV one can reach J ′ 40meV, exceeding room temperature. Such a robust ground state could be used as an essential part of a larger quantum circuit, extending existing semiconductor technology to 6 quantum applications. The Authors thank Ian Affleck for discussions. This work was supported by Canadian Institute for Advanced Research and QuantumWorks. ∗ Present address: Department of Physics, University of Wisconsin-Madison, WI 53706 [1] F. D. Haldane, Phys. Lett. 93A, 464 (1983). [2] I. Affleck, T. Kennedy, E. H. Lieb, and H. Tasaki, Phys. Rev. Lett. 59, 799 (1987). [3] I. Affleck, T. Kennedy, E. H. Lieb, and H. Tasaki, Commun. Math. Phys. 115, 477 (1988). [4] M. P. Nightingale and H. W. J. Blote, Phys. Rev. B 33, 659 (1986). [5] S. R. White, Phys. Rev. Lett. 69, 2863 (1992). [6] S. R. White and D. A. Huse, Phys. Rev. B 48, 3844 (1993). [7] S. R. White and I. Affleck, Phys. Rev. B 77, 134437 (2008). [8] W. J. L. Buyers, R. M. Morra, R. L. Armstrong, M. J. Hogan, P. Gerlach, and K. Hirakawa, Phys. Rev. Lett. 56, 371 (1986). [9] R. M. Morra, W. J. L. Buyers, R. L. Armstrong, and K. Hirakawa, Phys. Rev. B 38, 543 (1988). [10] E. Cizm´ar, M. Ozerov, O. Ignatchik, T. P. Papageorgiou, J. Wosnitza, S. A. Zvyagin, J. Krzys- tek, Z. Zhou, C. P. Landee, B. R. Landry, et al., New Journal of Physics 10, 033008 (2008). [11] L. Jacak, P. Hawrylak, and A. Wojs, Quantum Dots (Springer-Verlag, Berlin, 1998). [12] Y.-P. Shim and P. Hawrylak, Phys. Rev. B 78, 165317 (2008). [13] M. Ciorga, A. S. Sachrajda, P. Hawrylak, C. Gould, P. Zawadzki, S. Jullian, Y. Feng, and Z. Wasilewski, Phys. Rev. B 61, R16315 (2000). [14] 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). [15] F. H. L. Koppens, C. Buizert, K. J. Tielrooij, I. T. Vink, K. C. Nowack, T. Meunier, L. P. Kouwenhoven, and L. M. K. Vandersypen, Nature 442, 766 (2006). [16] L. Gaudreau, S. A. Studenikin, A. S. Sachrajda, P. Zawadzki, A. Kam, J. Lapointe, M. Ko- rkusinski, and P. Hawrylak, Phys. Rev. Lett. 97, 036807 (2006). [17] L. Gaudreau, A. Kam, G. Granger, S. A. Studenikin, P. Zawadzki, and A. S. Sachrajda, Appl. Phys. Lett. 95, 193101 (2009). 7 [18] M. Korkusinski, I. P. Gimenez, P. Hawrylak, L. Gaudreau, S. A. Studenikin, and A. S. Sachra- jda, Phys. Rev. B 75, 115301 (2007). [19] F. Delgado, Y.-P. Shim, M. Korkusinski, and P. Hawrylak, Phys. Rev. B 76, 115332 (2007). [20] K. Hida, Phys. Rev. B 45, 2207 (1992). [21] H. H. Hung and C. D. Gong, Phys. Rev. B 71, 054413 (2005). [22] A. Stathopoulos and J. R. McCombs, SIAM J. Sci. Comput. 29, 2162 (2007). [23] M. E. Reimer, W. R. McKinnon, J. Lapointe, D. Dalacu, P. J. Poole, G. C. Aers, D. Kim, M. Korkusinski, P. Hawrylak, and R. L. Williams, Physica E 40, 1790 (2008). 8 FIG. 1: (Color online) Schematic picture of the triple quantum dot chain on a semiconductor device. (a) The triple quantum dot chain is defined by a top gate, with a pattern of holes to create potential minima in the two-dimensional electron gas in the active region below. (b) For the realization of effective spin-one chain, each dot on the linear chain contains a single electron and the other dots on the top and bottom contain two electrons. 9 FIG. 2: (Color online) Different models for the chain of triple quantum dot molecules. (a) In Hubbard model, each dot has a single level and they are connected by the tunnelling and Coulomb interaction. (b) When the Coulomb interaction is strong and the tunnelling can be considered as a perturbation, the singly occupied dots are represented with localized spins with alternating spin-spin interaction. (c) If the intra-molecular ferromagnetic interaction J is much stronger than the inter-molecular antiferromagnetic interaction J ′, each molecule is in spin triplet state and the system can be considered as a spin-one antiferromagnetic Heisenberg chain. 10 0.8 0.6 0.4 ' J / ) S G E E - ( 0.2 0 1 1 2 1 0 2 1 0 1 2 1 1 2 0 1 1 2 0 1 1 3 NM 1 1 2 1 0 1 1 2 1 0 0 1 2 0 1 0 1 2 0 1 4 5 FIG. 3: (Color online) Comparison of Hubbard model and spin-half chain model. The effective Hamiltonian of spin-half chain with alternating spin-spin interaction describes the low energy states of the full Hubbard Hamiltonian very well both qualitatively and quantitatively. Green triangles are the lowest eigenstates of the Hubbard Hamiltonian and red circles are the eigenstates of the effective spin-half chain Hamiltonian. The numbers adjacent to the triangles and circles represent the total spin of each eigenstates. The result is shown for TQD chains with up to 5 molecules. 11 0.8 0.6 0.4 ' J / ) 0 = S E - 1 = S E ( 0.3 0.2 0.1 0 -0.1 -0.2 -0.3 0.3 0.2 0.1 0 -0.1 -0.2 -0.3 2 4 6 8 NM 10 12 14 ' J / ) S G E - 2 = S E ( 0.2 spin-half chain spin-one chain 0 2 3 4 5 6 7 9 10 11 12 13 14 8 NM FIG. 4: (Color online) Comparison between spin-half chain and spin-one chain. The main panel shows the energy gap between spin-quintuplet(S=2) and ground state, in unit of the effective antiferromagnetic interaction J ′ of the spin-half chain. The energy gap reaches a finite value at infinite chain, which is the Haldane gap. Spin-half chain has a larger energy gap than the spin-one chain. The inset shows the spin-triplet (S=1) and spin-singlet (S=0) energy gap. Both spin-half alternating chain and spin-one chain show oscillating spin singlet and spin triplet ground state and the energy gap between spin singlet and triplet states decreases as the length of the chain increases and goes to zero for infinite chain. 12
1005.3937	2	1005	2010-08-26T07:42:53	Quantifying Transition Voltage Spectroscopy of Molecular Junctions	[ "cond-mat.mes-hall" ]	Transition voltage spectroscopy (TVS) has recently been introduced as a spectroscopic tool for molecular junctions where it offers the possibility to probe molecular level energies at relatively low bias voltages. In this work we perform extensive ab-initio calculations of the non-linear current voltage relations for a broad class of single-molecule transport junctions in order to assess the applicability and limitations of TVS. We find, that in order to fully utilize TVS as a quantitative spectroscopic tool, it is important to consider asymmetries in the coupling of the molecule to the two electrodes. When this is taken properly into account, the relation between the transition voltage and the energy of the molecular orbital closest to the Fermi level closely follows the trend expected from a simple, analytical model.	cond-mat.mes-hall	cond-mat	Quantifying Transition Voltage Spectroscopy of Molecular Junctions Jingzhe Chen,1 Troels Markussen,1, 2, ∗ and Kristian S. Thygesen1 1Center for Atomic-scale Materials Design (CAMD), Department of Physics, Technical University of Denmark, DK-2800 Kgs. Lyngby, Denmark 2Danish National Research Foundations Center of Individual Nanoparticle Functionality (CINF), Department of Physics, Technical University of Denmark, DK-2800 Kgs. Lyngby, Denmark (Dated: May 22, 2018) Transition voltage spectroscopy (TVS) has recently been introduced as a spectroscopic tool for molecular junctions where it offers the possibility to probe molecular level energies at relatively low bias voltages. In this work we perform extensive ab-initio calculations of the non-linear current voltage relations for a broad class of single-molecule transport junctions in order to assess the applicability and limitations of TVS. We find, that in order to fully utilize TVS as a quantitative spectroscopic tool, it is important to consider asymmetries in the coupling of the molecule to the two electrodes. When this is taken properly into account, the relation between the transition voltage and the energy of the molecular orbital closest to the Fermi level closely follows the trend expected from a simple, analytical model. PACS numbers: 73.63.Rt,73.23.-b Molecular electronics holds the promise of continu- ing the miniaturization of electronic devices beyond the limits of standard silicon technologies by using single molecules as the active elements [1, 2]. In the design and characterization of molecular devices, the electronic structure of the molecule is naturally of vital impor- tance. In fact, when quantum interfererence effects are disregarded[3], there is a direct relation between a junc- tion's transport properties and the distance of the molec- ular energy levels to the electrode Fermi level. The latter may in principle be determined from peaks in the dI/dV curve, where I is the current and V is the bias voltage. Assuming that the molecular level closest to the Fermi level is the highest occupied molecular orbital (HOMO) it would require a voltage of V ∼ 2EF − εH to probe the HOMO position in a symmetric junction. However, in practice the molecular junction often becomes unstable and breaks down due to the large current density before the peak in the dI/dV is reached. Transition voltage spectroscopy (TVS) was introduced as a spectroscopic tool in molecular electronics by Beebe [4]. Beebe found that the Fowler-Nordheim graph et al. of a molecular junction, i.e. a plot of ln(I/V 2) against 1/V , showed a characteristic minimum at a bias volt- age Vmin which scaled linearly with the HOMO energy obtained from ultraviolet photo spectroscopy (UPS). Im- portantly, the TVS minimum is obtained at relatively low bias voltage before electrical break-down. TVS is now becoming an increasingly popular tool in molecular electronics [2, 5 -- 9]. The original interpretation of TVS introduced by Beebe [4] and applied in most later experimental work, is based on a Simmons tunnel barrier model [10]. Within this interpretation, the TVS minimum is obtained, when the tunnel barrier, due to the applied bias potential, changes from being trapezoidal to triangular. The tran- sition voltage equals the barrier height, which is inter- preted as the distance from the Fermi level to the closest molecular level. However, it has recently been pointed out by Huisman et al. that the barrier model is incon- sistent with experimental data [11] which on the other hand is more appropriately described by transport via a single electronic level. In the single-level model the transmission function was assumed to have a Lorentzian shaped, and all non-linear effects due to the finite bias, were neglected. While these assumptions may be reason- able, they are not obviously fulfilled in a realistic molec- ular junction at high electric fields. It is therefore of interest to compare the simple Lorentzian transmission model with more realistic calculations. In this Letter, we present a quantitative analysis of TVS based on extensive ab-initio calculations of the non- linear current-voltage relations for a broad class of molec- ular junctions. The ratio of the TVS minimum to the HOMO level position is found to vary between 0.8 and 2.0 depending on the junction asymmetry, i.e. quite dif- ferent from the one-to-one relation assumed so far. The large variation is due to the difference in the non-linear response of the molecular level to the bias voltage. The importance of asymmetry effects is further signified by the fact that many of the experiments using TVS were performed on asymmetric molecules in an asymmetric conductive AFM measurement setup [4 -- 6, 8]. Indeed, as we show below a larger degree of consistency between TVS measurements and UPS data is obtained when the asymmetry is taken into account. We begin our analysis by considering a simple model for transport via a single electronic level equivalent to the work in Ref. [11]. The transmission function is assumed to be a Lorentzian T (E; ε0, Γ) = f (E − ε0)2 + Γ2/4 , (1) ε0 − EF /Vmin η = 0 ε0 − EF /Vmin η = 1/2 FE + /2V /2VFE − ε + 0 ηV ε 0 Figure 1: Ratio between molecular level energy and transition voltage, ε0 − EF /Vmin, vs. molecular level energy, ε0 and broadening, Γ, for a symmetric junction, η = 0, (left), and a completely asymmetric junction, η = 1/2, (right). Note the different scales of ε0 −EF /Vmin for the symmetric and asym- metric junctions. The lower part illustrates how the molecu- lar level moves under a finite bias voltage. At zero bias the level is located at ε0, but when a bias voltage is applied, the level follows to some degree the chemical potential of the left electrode, due to a stronger coupling to this electrode. where ε0 and Γ are the molecular level energy and broad- ening, respectively. f is a constant factor to account for multiple molecules in the junction, asymmetric coupling to the electrodes, etc. Our analysis will not be depen- dent on the actual value of f . We shall assume that the molecular level is the HOMO level, i.e. ε0 < EF . This is the typical case for thiol bonded molecules (as consid- ered in this work), where electron transfer from the metal to the sulphur end group shifts the molecular levels up- ward in energy[12]. At finite bias voltage, the molecular level may be shifted relative to the zero-bias position, as shown schematically in Fig. 1 (lower part). Such non- linear effects express themselves differently in symmetric and asymmetric junctions [13, 14]. For example, if the molecule is only strongly coupled to the left electrode, the molecular levels will follow the chemical potential of the left contact. On the other hand, for a symmetric junction the molecular level will remain at the zero-bias position. We describe the degree of asymmetry by the parameter η ∈ [−1/2; 1/2] such that the current is given by I = 2e h Z ∞ −∞ T (E; ε0 + η V, Γ) [fL(V ) − fR(V )] dE, (2) where fL/R(V ) = 1/[exp (EF ± eV /2)/kBT + 1] are the Fermi-Dirac distributions for the left and right contact, respectively. The dependence of the level position on the bias voltage is the main difference between our model and the one considered in Ref. 11. Using Eq. (2) we calculate ln(I/V 2) and find the min- imum at the bias voltage Vmin. At this voltage, the slope 2 of the current depends quadratically on the bias voltage, I ∝ V 2, and d(ln(I/V 2))/dV = 0. An example of a cal- culated Fowler-Nordheim plot can be seen in the inset of Fig. 4. In Fig. 1 we plot the ratio ε0 − EF /Vmin as a function of broadening, Γ and molecular level energy, ε0, for a symmetric junction (η = 0) and a completely asym- metric junction (η = 1/2). Fig. 1 illustrates two main points: First, the ratio ε0 − EF /Vmin is nearly constant over a large range of Γ and ε0 values. For a given de- gree of asymmetry (value of η), the TVS minimum can therefore be used as a direct measure of the molecular level position, independently of Γ. Note, however, that for Γ/ε0 − EF & 1, there is no minimum in the Fowler- Nordheim plot, since the molecular level is too close to the Fermi level, and the current increases slower than ∝ V 2 at all bias values. The second conclusion from Fig. 1 is that the ratio ε0 − EF /Vmin ranges from 0.86 to 2.0 depending on the asymmetry of the molecular junction. In order to use TVS as a quantitative tool, knowledge of the asymmetry factor is therefore needed. On the other hand, if the molecular levels can be determined by other means, the TVS can by used to measure the asymmetry factor. i n m V / F E − H ε 2.2 2 1.8 1.6 1.4 1.2 1 0.8 0 Model Naph, Anth, BP, TP (flat surfaces) Ph,Naph,Anth (STM) C6−C9 BDT Naph, Anth, C6, η=1/2 More symmetric More asymmetric 0.1 0.2 0.3 η 0.4 0.5 Figure 2: Ratio between the HOMO energy (at zero bias) and the transition voltage, Vmin, vs. asymmetry parameter, η. Solid line is obtained from a Lorentzian transmission function using Eqs. (1) and (2) and symbols are results of ab-initio finite bias calculations. We now turn to our ab-initio finite bias calculations and Fig. 2, which is our main result. It shows the ratio of εH − EF /Vmin vs. η for the 17 molecular junctions listed in Fig. 3. The solid line is the result obtained from the one-level model using Eqs. (1) and (2). Al- though there are deviations between the model and the ab-initio results, both data sets clearly follow the same trend. This result further supports TVS as a spectro- scopic tool, but it also underlines that quantitative in- formation about the molecular level position can only be obtained from TVS provided some knowledge of the junc- tion asymmetry. The calculated numbers for the data points are listed in Fig. 3. The current at finite bias voltage is calculated us- ing DFT in combination with a non-equilibrium Green function (NEGF) method, following the principles of Ref. [15]. Our DFT-NEGF method is implemented in GPAW, which is a real space electronic structure code based on the projector augmented wave method [16, 17]. We use the PBE exchange-correlation functional [18], and a 4 × 4 k-point sampling in the surface plane. The elec- tronic wave functions are expanded in an atomic orbital basis [17]. In all calculations, the molecule and the closest Au layers are described by a double-zeta plus polariza- tion (dzp) basis set, while the remaining Au atoms are described by a single-zeta plus polarization (szp) basis. The upper panel of Fig. 3 shows the atomic structure of two representative junctions: a hexanethiol (left) in an STM-like configuration with the STM-tip ∼ 4 A away from the molecule, and anthracenethiol (right) between two flat Au (111) surfaces with a distance of 1.5 A be- tween the right Au electrode and the closest H atom of the molecule. For all STM-setups, we initially re- lax the molecule on a single Au(111) surface and sub- sequently add the STM electrode without further relax- ations. For the systems with two flat Au surfaces, we relax the molecule and the two closest Au layers. Ex- cept for the benzene junctions (S-Ben-S and HS-Ben- SH), the linking S atom relaxes into a bridge site of the Au(111) surface shifted slightly towards the hollow site. The structure of the symmetric benzene-dithiol (S-Ben- S) junction is taken from Ref. [19] where the molecule binds to an Au adatom together with a SCH3 unit. In the other benzene junction (HS-Ben-SH) the hydrogenated sulfur atoms bind to Au adatoms of the Au(111) surfaces following Ref. [20]. In Fig. 3 we also list for each junction the obtained TVS minimum voltage, Vmin, the position of the HOMO level relative to EF , and the calculated asymmetry pa- rameter. Figure 4 illustrates how we determine these quantities, in the case of the C6-S junction. The TVS minimum is simply found from the minimum in the Fowler-Nordheim plot, as show in the inset. To deter- mine the HOMO level, we project the density of states onto the carbon atoms. Molecular levels then appears as clear peaks in the projected density of state (PDOS) vs. energy plot as shown in Fig. 4. We note that there are situations where a peak in the PDOS does not show in the transmission function due to symmetry mismatch between the molecular orbital and the electrode wave- functions. This is indeed the situation for Ph-S, Naph- S and Anth-S in the STM setup where a peak in the PDOS is observed at E − EF ∼ −0.55eV, but only a vanishingly small transmission peak occurs at this energy. In these three cases we take the second PDOS peak at E − EF = −2.35eV as the HOMO level. At this energy, the Au tip wavefunctions also have d-character, which enables non-zero coupling to the molecular π-orbital. At finite bias voltage, the energy of the HOMO level (c.f. Fig. 1). Figure is shifted by the bias voltage. 3 Figure 3: Top: Atomic structure of two of the major groups of molecular transport junctions investigated in this work and in experiments [2, 4, 5], namely alkanes in an STM-like con- figuration (top left), and anthracenethiol (Anth) between two flat Au (111) surfaces (top right). The table in the lower part shows schematically all the considered molecular junctions to- gether with calculated values of Vmin, εH − EF , and η. The electrode configurations are indicated with gray boxes show- ing both the flat Au surface (rectangles) and STM-like geome- tries (pentagons). The super-script a indicates that the data are obtained from non-self consistent calculations by manually setting η = 1/2. The HOMO energy, εH, refers to the first peak in the Kohn-Sham (PBE) projected density of states. See text for more details. 4 shows the PDOS of the C6-alkane junction under bias voltage V = 0 V (red solid) and V = 2.0 V (blue dashed). The equilibrium HOMO energy of εH − EF = −3.6 eV shifts upward by ∆ = 0.4 eV when the bias is applied. The asymmetry factor, η, is calculated from the shift, ∆, according to η = ∆/V = 0.2. Notice that even for the very asymmetric STM config- urations, the asymmetry factor is never larger than 0.35. This is due to the relatively small distance between the molecule and the Au tip (4.0 A). Increasing this distance would increase η further, however, this is computation- ally problematic due to the finite range of the atomic or- bital basis. Instead we simulate a completely asymmetric junction (η = 0.5), by using the zero-bias transmission function and fixing the electrode chemical potentials to µL = EF and µR = EF − V . The data for the three last junctions in Fig. 3 are obtained in this way and are thus ) s t i n u . b r a ( S O D P 6 5 4 3 2 1 0 ∆ 4.0 3.5 3.0 2.5 2.0 1.5 1.0 EEF (eV) Figure 4: Projected density of states on the C-atoms in the C6-alkane junction shown in Fig. 3. The solid red line refers to Vbias = 0 V and the dashed blue line refers to Vbias = 2.0 V. The HOMO level (at Vbias = 0 V) is located at E − EF = −3.6 eV. We determine the asymmetry factor, η, from the shift, ∆, of the HOMO level with the bias volt- age. The inset shows the calculated Fowler-Nordheim plot with the characteristic TVS-minimum at Vmin = 2.4 V. non-self consistent results. The deviations of the numerical data from the ana- lytical result in Fig. 2 may be due to several reasons. (i): The ab-initio transmission functions do not have a perfect Lorentzian shape as assumed in the analytical model. (ii) The dependence of the HOMO energy on the bias voltage is not strictly linear. (iii) The bias voltage influences the shape of the molecular orbitals which in turn affects the coupling strength. (iv) It may in some cases be difficult to accurately determine εH and η, as discussed above. The TVS minimum voltage was originally interpreted as a transition from tunneling to field emission [4]. How- ever, according to our calculations, Vmin does not mark such a transition, and in line with Refs. [11, 21] we do not view Vmin as a special transition voltage. Rather the usefulness of TVS relies on the direct relation between Vmin and εH when the transmission function can be well described by a Lorentzian. In the conducting AFM measurements of Ref. 4, the ra- tio εH −EF /Vmin was found to vary between 2.1 and 2.6 for a set of molecules which included Naph-S, BP-S, and TP-S also considered in this work. This is significantly larger than the value of 1 predicted by Simmons bar- rier model. According to our model the ratio should be smaller than 2.0 which is obtained for a completely asym- metric junction (η = 0.5). To resolve this puzzle, we first note that our ab-initio calculations yield η ∼ 0.3 for the three molecules (c.f. Fig. 3) giving εH − EF /Vmin ∼ 1.6 according to our model. Next, we note that the pres- ence of the AFM tip in the transport measurements will lead to a renormalization of the HOMO energy due to an image charge effect [22] which is not present in the UPS measurements. Based on the method described in 4 Ref. 23 we estimate the image charge interaction to be 0.5 − 0.35 eV depending on the size of the molecule. Cor- recting the UPS values for εH by these values lead to ratios εH −EF /Vmin in the range 1.6−1.8 in good agree- ment with the model prediction of ∼ 1.6. Finally, we note that the well known inability of DFT to describe energy gaps and level alignment of molecules at surfaces [22] does not affect the conclusions of the present work. This is because, according to Fig. 1, the ratio (εH − EF )/Vmin is independent of the value of εH . In particular, the dependence of (εH −EF )/Vmin on asym- metry is expected to be a general result independent of the absolute position of the molecular levels. In conclusion, we have performed extensive ab initio DFT transport calculations to simulate transition voltage spectroscopy for a large number of molecular junctions. The numerical data closely follow the trend expected from an analytical model with a Lorentzian shaped trans- mission function. We have explicitly shown that in order to use TVS as a quantitative spectroscopic tool to probe the molecular levels, it is necessary to take the asymme- try of the molecular junction into account. The present analysis should therefore be considered in future applica- tions of transition voltage spectroscopy. The center for Atomic-scale Materials Design (CAMD) is funded by the Lundbeck Foundation. The authors ac- knowledge support from FTP through grant no. 274-08- 0408 and from The Danish Center for Scientific Comput- ing. ∗ [email protected] [1] K. Moth-Poulsen and T. Bjornholm, Nat. Nano. 4, 551 (2009). [2] H. Song, Y. Kim, Y. H. Jang, H. Jeong, M. A. Reed, and T. Lee, Nature 462, 1039 (2009), ISSN 0028-0836. [3] R. Baer and D. Neuhauser, Journal of the American Chemical Society 124, 4200 (2002). [4] J. M. Beebe, B. Kim, J. W. Gadzuk, C. Daniel Fris- bie, and J. G. Kushmerick, Phys. Rev. Lett. 97, 026801 (2006). [5] J. M. Beebe, B. Kim, C. D. Frisbie, and J. G. Kushmer- ick, ACS Nano 2, 827 (2008). [6] C. D. Zangmeister, J. M. Beebe, J. Naciri, J. G. Kushm- erick, and R. D. van Zee, Small 4, 1143 (2008). [7] S. Ho Choi, B. Kim, and C. D. Frisbie, Science 320, 1482 (2008). [8] A. Tan, S. Sadat, and P. Reddy, Appl. Phys. Lett. 96, 013110 (2010). [9] Y. Noguchi, T. Nagase, R. Ueda, T. Kamikado, T. Kub- ota, and S. Mashiko, Jpn. J. Appl. Phys. 46, 2683 (2007). [10] J. G. Simmons, J. Appl. Phys. 34, 1793 (1963). [11] E. H. Huisman, C. M. Gu´edon, B. J. van Wees, and S. J. van der Molen, Nano Letters 9, 3909 (2009). [12] G. Peng, M. Strange, K. S. Thygesen, and M. Mavrikakis, J. Phys. Chem. C 113, 20967 (2009). [13] S. J. van der Molen and P. Liljeroth, J. Phys. Condens. Matter. 22, 133001 (2010). Lett. 77, 3865 (1996). [14] C. Kergueris, J.-P. Bourgoin, S. Palacin, D. Esteve, C. Urbina, M. Magoga, and C. Joachim, Phys. Rev. B 59, 12505 (1999). [15] M. Brandbyge, J.-L. Mozos, P. Ordej´on, J. Taylor, and [19] M. Strange, O. Lopez-Acevedo, and H. Hakkinen, J. Phys. Chem. Lett. 1528 (2010). [20] Z. Ning, W. Ji, and H. Guo, arXiv:0907.4674v2 (2010). [21] M. Araidai and M. Tsukada, Phys. Rev. B 81, 235114 K. Stokbro, Phys. Rev. B 65, 165401 (2002). (2010). [16] J. J. Mortensen, L. B. Hansen, and K. W. Jacobsen, [22] J. M. Garcia-Lastra, C. Rostgaard, A. Rubio, and K. S. Phys. Rev. B 71, 035109 (2005). Thygesen, Phys. Rev. B 80, 245427 (2009). [17] A. H. Larsen, M. Vanin, J. J. Mortensen, K. S. Thygesen, [23] D. J. Mowbray, G. Jones, and K. S. Thygesen, J. Chem. and K. W. Jacobsen, Phys. Rev. B 80, 195112 (2009). Phys. 128, 111103 (2008). [18] J. P. Perdew, K. Burke, and M. Ernzerhof, Phys. Rev. 5
1505.02817	3	1505	2016-08-24T19:43:04	Transport evidence for Fermi-arc mediated chirality transfer in the Dirac semimetal Cd$_3$As$_2$	[ "cond-mat.mes-hall", "cond-mat.mtrl-sci", "cond-mat.str-el" ]	Dirac semi-metals show a linear electronic dispersion in three dimension described by two copies of the Weyl equation, a theoretical description of massless relativistic fermions. At the surface of a crystal, the breakdown of fermion chirality is expected to produce topological surface states without any counterparts in high-energy physics nor conventional condensed matter systems, the so-called "Fermi Arcs". Here we present Shubnikov-de Haas oscillations involving the Fermi Arc states in Focused Ion Beam prepared microstructures of Cd$_3$As$_2$. Their unusual magnetic field periodicity and dependence on sample thickness can be well explained by recent theoretical work predicting novel quantum paths weaving the Fermi Arcs together with chiral bulk states, forming "Weyl orbits". In contrast to conventional cyclotron orbits, these are governed by the chiral bulk dynamics rather than the common momentum transfer due to the Lorentz force. Our observations provide evidence for direct access to the topological properties of charge in a transport experiment, a first step towards their potential application.	cond-mat.mes-hall	cond-mat	Transport evidence for Fermi-arc mediated chirality transfer in the Dirac semimetal Cd3As2 Authors: Philip J.W. Moll1,2,, Nityan L. Nair1, Toni Helm1,2, Andrew C. Potter1, Itamar Kimchi1, Ashvin Vishwanath1, and James G. Analytis1,3, Affiliations: 1Department of Physics, University of California, Berkeley, California 94720, USA 2Max-Planck-Institute for Chemical Physics of Solids, Noethnitzer Strasse 40, D-01187 Dresden, Germany 3Materials Science Division, Lawrence Berkeley National Laboratory, Berkeley, California 94720, USA Correspondence to: [email protected] , [email protected] The dispersion of charge carriers in a metal is distinctly different from that of free electrons due to their interactions with the crystal lattice. These interactions may lead to quasiparticles mimicking the massless relativistic dynamics of high-energy particle physics1–3, and they can twist the quantum phase of electrons into topologically nontrivial knots – producing protected surface states with anomalous electromagnetic properties4–9. These effects intertwine in materials known as Weyl semimetals, and their crystal symmetry protected analogs, Dirac semimetals10. The latter show a linear electronic dispersion in three dimensions described by two copies of the Weyl equation, a theoretical description of massless relativistic fermions. At the surface of a crystal, the broken translational symmetry creates topological surface states, so-called "Fermi Arcs"11, which have no counterparts in high-energy physics nor conventional condensed matter systems. Here we present Shubnikov-de Haas oscillations in Focused Ion Beam prepared microstructures of Cd3As2 that are consistent with the theoretically predicted "Weyl orbits", a kind of cyclotron motion that weaves together Fermi arc and chiral bulk states12. In contrast to conventional cyclotron orbits, this motion is driven by the transfer of chirality from one Weyl node to another, rather than momentum transfer of the Lorentz force. Our observations provide evidence for direct access to the topological properties of charge in a transport experiment, a first step towards their potential application. The bulk electrons in topological semimetals are described by an ultra-relativistic dispersion relation, E(k)=±ħvF σk, resembling the Weyl equation for massless spin-1/2 particles. Here σ is a pseudo-spin-1/2 degree of freedom that is energetically locked parallel or anti-parallel to the momentum, k, of the electron, giving electrons definite chirality k ±σ. Applying electromagnetic fields to Weyl or Dirac semimetals induces a pumping of electric charge between Weyl nodes with opposite chirality, a phenomena known in high energy physics as the chiral anomaly13–15. At the surface of these materials, this anomalous chirality transfer is facilitated by topologically such as strong protected surface arcs, the so-called "Fermi arc" surface states, which act as a pipeline connecting opposite chirality Weyl points11,16. Recently, Na3Bi17 and Cd3As2 18 have been predicted to be three-dimensional bulk Dirac semimetals. This has sparked significant research interest and the linear dispersion in these materials was confirmed by ARPES19–24 and STM25 experiments. A number of unusual material properties linear magnetoresistance19,26 and high mobilities27 were identified that are potentially linked to the relativistic nature of the Dirac quasiparticles. Yet the prospect of studying ultra-relativistic particles and their accompanying topological surface states, as well as potential applications exploiting their unusual behavior naturally requires a more direct measurement capable of revealing both the relativistic dynamics and topological surface states. The main aim of this study is to present such evidence in four-terminal transport measurements, showing the possibility of detecting and manipulating chiral states in Dirac semimetals. One intriguing fingerprint of Weyl quasiparticles in the electronic transport properties in strong magnetic fields has been recently predicted by Potter, Kimchi and Vishwanath (PKV)12. This "Weyl orbit" (Fig.1) weaves together the chiral states in the bulk with the topological Fermi-arc states on opposite surfaces into a closed orbit. Its quantization produces a distinctive contribution to the quantum oscillation spectrum that provides an observable signature of the chiral and topological character of these materials. This closed orbit is strikingly different from typical electrons orbiting around a Fermi surface in a metal as the quasiparticle experiences zero Lorentz force on the chiral path segments traversing the bulk. The main result of this study is an additional quantum oscillation frequency observed in microstructures smaller than the mean-free-path that exhibits characteristics of both surface-like and bulk-like states, as naturally expected for Weyl orbits. The microstructures were prepared from Cd3As2 single crystals by FIB etching (Fig.1, see Methods). Down to the smallest thickness of L=150nm, the magnetoresistance at temperatures below 100K shows pronounced Shubnikov- de Haas oscillations signaling the low effective mass of the charge carriers and the high crystal quality of the devices (Fig.1b). Quantum oscillations on bulk crystals have reported one single frequency19,26, arising from an essentially spherical 3D Fermi surface in agreement with ARPES and STM experiments. This single bulk frequency (FB) is also consistently observed in all of the studied parent bulk crystals as well as all microstructures. However, when the field is applied perpendicular to the [0 1 0] surface (0°), a second frequency FS= 61.5T appears which is distinct from the higher harmonics of the bulk. We find an effective mass of 0.044me for the bulk, similar to previously reported measurements on bulk crystals26, and a similar mass of 0.050me in the additional orbit FS (See methods section). The value of the observed surface frequency is compatible with the prediction for Weyl orbits. Their trajectory combines segments of both chiral bulk and Fermi arc surface states, and the quantum oscillation frequency FS can be estimated from the time spent in each of them12: FS = EF k0 / e π vF ≈ 56T The length of the surface Fermi arc in reciprocal space may be approximated as k0≈0.8 nm-1 using the k-space separation of the Dirac points determined by ref.19. The bulk quantum oscillation frequency and effective mass provide direct access to the Fermi energy EF=192meV and the Fermi velocity vF=8.8 105 m/s. The resulting estimate of FS~56T is in good quantitative agreement with the measured FS=61.5T. The quantum oscillation frequency dependence on the angle between the magnetic field and the surface normal shows a clear signature of a surface state as shown in Figure 2, where 0° denotes fields perpendicular to the surface and 90° parallel to it. The angular dependencies of both frequencies are strikingly different: while the low frequency FB (blue) remains essentially constant and is observed at all angles, the second frequency FS (red) strongly increases as the field is tilted away from the surface. FS(θ) is well described by a cos(θ)-1 dependence, indicating that the field component perpendicular to the surface is relevant for the orbit. This angle dependence is a hallmark of two-dimensional (2D) Fermi surfaces. The bulk band structure however does not support a 2D Fermi surface and it is not observed in our bulk crystals. While the angle dependence clearly suggests a surface character of the quantum path associated with the additional frequency, the quantum oscillations also show pronounced bulk-like characteristics which are unexpected for orbits simply consisting of surface states. First of all, the additional quantum oscillation is only observed in samples where the bulk mean-free-path is longer than or comparable to the sample thickness. The low-field transverse magnetoresistance for in-plane fields quantitatively confirms our devices to be in this limit. The resistance maximum at small fields is a hallmark signature of quasi-ballistic transport in clean metals known as the "Knudsen effect" (Fig.3a, see methods). The frequency spectrum of the quantum oscillations is found to be strongly thickness dependent (Fig.3b). No trace of the surface frequency FS has been observed in devices thicker than 3µm, in agreement with its absence in our bulk quantum oscillation measurements. As the sample thickness is reduced, its relative weight compared to the bulk frequency strongly increases, and devices thinner than 500nm are dominated by the surface oscillation. The increase in the various studied samples follows an exponential behavior, with an exponent of d = 675nm. This value should be compared to the bulk mean free path estimated from transport as l = vFτ = vF m* / (ne2ρ0) = 1.0µm, where ρ0 = 55µΩcm is the zero-field resistivity in our crystals at 2K, n = 2 kF 2 / (3π2) = 2.5 1018 cm-3 the bulk carrier density estimated for the two-fold degenerate spherical Fermi surface and kF = mvF / ħ = 3.3 108m-1 the Fermi momentum extracted from the Shubnikov-de Haas oscillations. This estimate of the mean free path is in good quantitative agreement with the observation of the Knudsen flow maximum. Intriguingly, this thickness dependence matches well with the expectation for Weyl orbits. Ordinarily, any scattering phases acquired suppress quantum oscillations in which case the quantum oscillations would be expected to decay exponentially in L over the quantum mean free path lQ which is generally much shorter than the mean-free path measured in transport. However, for large fields the bulk chiral Landau level is expected to exhibit extra resilience to such dephasing and allow for quantum oscillations for thicker samples. As the closed Weyl orbit contains two path segments traversing the bulk, each at opposite chirality, the effective bulk path length is twice the device thickness L. The agreement between 2d = 1.25µm and l = 1.0µm suggests that the relevant thickness scale for the appearance of the additional quantum oscillations is comparable to the bulk transport mean-free path. Additional clue as to the origin of the Shubnikov-de Haas oscillations comes from their phase as a function of field. We observe deviations from ideal periodicity of the oscillations in inverse magnetic field, appearing as a continuously drifting phase as shown in Figure 4. While similarly subtle deviations from periodicity can arise due to g-factor band splitting in strongly spin-orbit coupled materials28, the direction of the shift is opposite to those previously observed in experiment: The peak positions Bn are shifted towards higher fields compared to the purely periodic case, while spin-splitting should shift them towards lower fields. On the other hand, this direction and magnitude of shift of the Landau levels is expected for Weyl orbits. Non-adiabatic corrections are expected to appear due to field-induced tunneling between Fermi arc states and bulk states, occurring as the orbiting quasiparticles approach the Weyl node. A single Weyl orbit will encounter four such tunneling processes, leading to a phase deviation that is in quantitative agreement with what we observe (See Fig.4 and methods). The discussion so-far considered the Dirac material Cd3As2 as two independent Weyl subsystems overlapping in k-space. However, magnetic fields can break the crystal symmetry protecting the superimposed Weyl nodes of opposite chirality and thus introduce a gap. In sufficiently strong fields, the particle approaching the node may thus tunnel into the oppositely dispersing Fermi-arc on the same surface instead into the bulk, thus forming closed orbits purely from Fermi-arc states similar to the surface states of topological insulators. While our experiments cannot rule out localized topological surface states as the origin of the oscillations, the thickness dependence of the surface state amplitudes and the non-adiabatic corrections are more suggestive of the Weyl orbit in the present field range. In both cases, the Fermi arcs constitute in the quantum orbit and our results confirm the detection of topological surface currents in the microstructures. However quantum oscillations may also arise from trivial surface states or from the defect layer introduced by the fabrication technique, without the involvement of topological states. Thus it is essential to find ways to probe if topological states participate in the observed orbit, setting the involved orbit apart from topologically trivial states. The existence of a "saturation field" above which oscillations were expected to cease was predicted to provide such test12. However, a reexamination of the equation for quantum oscillations, reproduced in Eqn. 1 below, reveals that the oscillations associated with the Weyl orbit actually persist up a much larger field of order FS=56T in the present instance (see methods), and the thickness dependence of the saturation field is highly complex. Consequently, the Weyl surface oscillations are expected to persist over the full range of fields explored in this experiment, in agreement with our observations. Additional evidence for the topological nature of the surface oscillations comes from their remarkable resilience against surface disorder. We have purposely introduced significant surface damage by almost normal incidence FIB-irradiation. Remarkably, the surface state oscillations grow upon increasing disorder, in strong contrast to the usual disorder dampening of conventional surface oscillations. The counter-intuitive increase of amplitude with increasing surface damage, however, would be a natural consequence of the protection of topological surface states (see methods). Another striking piece of evidence for the non-trivial nature of the quantum orbit comes from a distinguishing feature of the Fermi arc orbit: the quantum oscillation phase depends on the thickness of the sample, L, given by: Bn -1 = e k0 -1 (n πvF / EF – L) (1) The position of the nth resistance minimum, Bn, is predicted to follow Eq.1, where k0 describes the length of the Fermi arc surface state in k-space, vF is the Fermi velocity, and EF the Fermi energy. To test this concept, we fabricate a sample geometry that is by design sensitive to such a thickness dependence of the phase. In Figure 5 we show a sample structured with a triangular cross-section as well as a rectangular one as a reference. The quantum oscillations for fields perpendicular to the surface for each device (0° for the rectangle, 60° for the equilateral triangle) are strikingly different. While the rectangular device clearly shows the presence of a surface state, the triangular one shows only the bulk frequency without any sign of the second surface frequency. Crucially, all surfaces were fabricated under the exact same conditions from the same piece of crystal. The sample was tilted with respect to the ion beam to ensure a grazing incidence condition for every surface, both on the rectangular and the triangular device. Both devices have comparable cross-section and surface area by design, so that a trivial surface state acting as a parallel conductance channel should lead to observable surface quantum oscillations in both of them. The absence of the surface state in the triangle is unexpected for oscillations arising from a trivial surface state, yet a natural consequence of Weyl orbits. Unlike conventional surface states that are confined to a single surface, the Weyl orbit cannot be observed in triangular geometries: All paths are of different length, each contributing to the field induced density-of-state modulation at different magnetic field 1/Bn (see Eq.1, Fig.5). This results in destructive interference due to a sum of oscillations with random phases, rendering the quantum oscillations unobservable in experiment. In contrast, all quantum paths involve bulk path segments of the same length L in a rectangular geometry, and thus all contribute to a density-of-state modulation at the same field. This directly evidences that the orbit associated with the frequency FS is sensitive to the shape and size of the bulk underneath the surface, in contrast to those arising from trivial surface states. The ensemble of presented results highlights an essential aspect of the additional quantum oscillations appearing when Dirac fermions are confined into microstructures smaller than the mean free path: They share characteristic features of both surface-like and bulk-like oscillations. Such hybrid characteristics arise naturally from the idea of the mixing of chiral bulk- and Fermi- arc surface-states into a coherent orbit. Nevertheless, there remain questions that challenge our current understanding of topological matter on microscopic length scales and in strong magnetic fields. For example, understanding the exact amplitude of the Weyl oscillations may require the extension of the present theory into the quantum limit, where the number n of occupied Landau levels cannot be treated as being large. Furthermore the understanding of the mixing of chiral states in strong magnetic fields and the influence on the Weyl orbit should be included and experimentally investigated in the future. By utilizing FIB structuring, we have demonstrated an experimentally simple route towards studying strongly confined topological matter, which will both increase our understanding of the transport characteristics of Fermi-arc states as well as provide a path to investigate the potential of these materials in future electronic applications. References: 1. Katsnelson, M. I., Novoselov, K. S. & Geim, A. K. Chiral tunneling and the Klein paradox in graphene. Nat. Phys. 2, 620–625 (2006). Novoselov, K. S. et al. Two-dimensional gas of massless Dirac fermions in graphene. Nature 438, 197–200 (2005). Vafek, O. & Vishwanath, A. Dirac Fermions in Solids-from High Tc cuprates and Graphene to Topological Insulators and Weyl Semimetals. Annu. Rev. Condens. Matter Phys. 5, 83–112 (2014). Fu, L., Kane, C. L. & Mele, E. J. Topological insulators in three dimensions. Phys. Rev. Lett. 98, 106803 (2007). Fu, L. & Kane, C. Superconducting Proximity Effect and Majorana Fermions at the Surface of a Topological Insulator. Phys. Rev. Lett. 100, 096407 (2008). Roushan, P. et al. Topological surface states protected from backscattering by chiral spin texture. Nature 460, 1106–1109 (2009). Zhang, T. et al. Experimental demonstration of topological surface states protected by Time-Reversal symmetry. Phys. Rev. Lett. 103, 266803 (2009). Pan, Z.-H. et al. Electronic structure of the topological insulator Bi2Se3 using angle- resolved photoemission spectroscopy: evidence for a nearly full surface spin polarization. Phys. Rev. Lett. 106, 257004 (2011). Park, S. R. et al. Chiral orbital-angular momentum in the surface states of Bi 2Se 3. Phys. Rev. Lett. 108, 046805 (2012). 2. 3. 4. 5. 6. 7. 8. 9. 10. Young, S. M. et al. Dirac semimetal in three dimensions. Phys. Rev. Lett. 108, 140405 (2012). 11. Wan, X., Turner, A. M., Vishwanath, A. & Savrasov, S. Y. Topological semimetal and Fermi-arc surface states in the electronic structure of pyrochlore iridates. Phys. Rev. B 83, 205101 (2011). Potter, A. C., Kimchi, I. & Vishwanath, A. Quantum oscillations from surface Fermi arcs in Weyl and Dirac semimetals. Nat. Commun. 5, 5161 (2014). 12. 13. Adler, S. L. Axial-vector Vertex in spinor electrodynamics. Phys. Rev. 177, 2426–2438 (1969). 14. Bell, J. S. & Jackiw, R. A PCAC Puzzle : π0->γγ in the σ-Model. nuovo Cim. 60A, 47–61 (1969). 15. Nielsen, H. B. & Ninomiya, M. The Adler-Bell-Jackiw anomaly and Weyl fermions in a crystal. Phys. Lett. B 130, 389–396 (1983). 16. Haldane, F. D. M. Attachment of Surface 'Fermi Arcs' to the Bulk Fermi Surface: 'Fermi- Level Plumbing' in Topological Metals. Arxiv Prepr. arXiv:1401.0529v1 (2014). at <http://arxiv.org/abs/1401.0529> 17. Wang, Z. et al. Dirac semimetal and topological phase transitions in A3Bi (A=Na, K, Rb). Phys. Rev. B 85, 195320 (2012). 18. Wang, Z., Weng, H., Wu, Q., Dai, X. & Fang, Z. Three-dimensional Dirac semimetal and quantum transport in Cd3As2. Phys. Rev. B 88, 125427 (2013). 19. Liang, T. et al. Ultrahigh mobility and giant magnetoresistance in the Dirac semimetal Cd3As2. Nat. Mater. 4, 3–7 (2014). 20. Neupane, M. et al. Observation of a three-dimensional topological Dirac semimetal phase in high-mobility Cd3As2. Nat. Commun. 5, 3786 (2014). 21. Liu, Z. K. et al. A stable three-dimensional topological Dirac semimetal Cd3As2. Nat. Mater. 13, 677–81 (2014). 22. Yi, H. et al. Evidence of Topological Surface State in Three-Dimensional Dirac Semimetal Cd3As2. Sci. Rep. 4, 6106 (2014). 23. Liu, Z. K. et al. Topological Dirac Semimetal, Na3Bi. Science 343, 864–867 (2014). 24. Borisenko, S. et al. Experimental realization of a three-dimensional dirac semimetal. Phys. Rev. Lett. 113, 027603 (2014). Jeon, S. et al. Landau quantization and quasiparticle interference in the three-dimensional Dirac semimetal Cd3As2. Nat. Mater. 13, 851–856 (2014). 25. 26. He, L. P. et al. Quantum Transport Evidence for the Three-Dimensional Dirac Semimetal Phase in Cd3As2. Phys. Rev. Lett. 113, 246402 (2014). 27. Rosenberg, A. J. & Harman, T. C. Cd3As2-A Noncubic Semiconductor with Unusually High Electron Mobility. J. Appl. Phys. 30, 1621 (1959). 28. Analytis, J. G. et al. Two-dimensional surface state in the quantum limit of a topological insulator. Nat. Phys. 6, 960–964 (2010). Acknowledgments: The Focused Ion Beam work was supported by the SCOPE-M center for electron microscopy at ETH Zurich, Switzerland. We thank Philippe Gasser, Joakim Reuteler and Bertram Batlogg for FIB support, and Maja Bachmann for performing magnetoresistance measurements. A.C.P. was supported by the Gordon and Betty Moore Foundation's EPiQS Initiative through Grant GBMF4307. We also thank S. Teat and K. Gagnon for their help in conducting X-ray diffraction measurements at the Advanced Light Source (ALS) beam line 11.3.1 and N. Tamura for micro-diffraction on beam line 12.3.2. N.T. and the ALS are supported by the Director, Office of Science, Office of Basic Energy Sciences, Materials Sciences Division, of the U.S. Department of Energy under Contract No. DE-AC02-05CH11231 at Lawrence Berkeley National Laboratory and University of California, Berkeley. Transport experiments, material synthesis and Focused Ion Beam microstructuring were supported by the Gordon and Betty Moore Foundation's EPiQS Initiative through Grant GBMF4374. Single-crystal x-ray refinements and theoretical support were funded by the Quantum Materials FWP, U.S. Department of Energy, Office of Basic Energy Sciences, Materials Sciences and Engineering Division, under Contract No. DE-AC02- 05CH11231. Author contributions: P.J.W.M. microstructured the crystals, performed the measurements and data analysis. N.L.N. synthesized and characterized the single crystals. T.H. analyzed the crystal structure. I.K., A.C.P. and A.V. contributed the theoretical treatment. P.J.W.M and J.G.A. designed the experiment. All authors were involved in writing the manuscript. Figures a 450nm 350nm 150nm 10mm ℓ -1 B + Chirality transfer B - 0 = FL - k x k z + F =36.5T B F =61.5T S 90° 0° b y L d ) . u . a ( e d u t i l p m A Cd As 150nm 2K 3 2 0° 90° c ) m c W m ( y t i v i t s s e R i 125 100 75 50 25 0 0 35.5T bulk FS 2FB 2FS 2 4 6 8 10 12 14 0 Field (T) 50 100 Frequency (T) 150 Fig 1. Surface oscillations in microstructures a) SEM micrograph of a typical sample prepared by FIB cutting. The active devices are 4×10µm2 free- standing sheets (purple) of varying thickness connected to contact pads (yellow). The crystallographic direction perpendicular to the polished surface is [0 1 0] and parallel to it [1 0 0], which define the plane of rotation. This entire sample consists of one contiguous slice of a Cd3As2 crystal, ensuring that all devices are made from the exact same starting material and have the same orientation during the experiments. b) Sketch of the Weyl orbit in a thin slab of thickness L in a magnetic field B. The orbit involves both the Fermi arc surface states connecting the Weyl nodes of opposite chirality, and the bulk states of fixed chirality (blue and red). In strong magnetic fields, quasiparticles may tunnel through the energy barrier separating the bulk states from the surface over a distance associated with the magnetic length lB (orange arrows). Note that as the Weyl orbit mixes processes in real and reciprocal space, the (x,y) coordinates of this sketch are in reciprocal and the z coordinate is in real space. c) Magnetoresistance and (d) its Fourier transform measured on the thinnest device (150nm) at 2K for both fields parallel (90°) and perpendicular (0°) to the surface. The main finding of this study is directly evident in the raw data: While parallel fields lead to a single frequency (incl. spin-splitting at higher fields), an additional higher frequency component FS associated with the surface oscillations appears for perpendicular fields. This high frequency is clearly distinguishable from higher harmonics of the low frequency FB. a ) . u . a ( e d u t i l p m A b ) T ( y c n e u q e r F FB FS 50 100 150 Frequency (T) 60 80 60 40 20 0 90 120 150 FS 180 FB 210 1/cos( )q 60° 50° 40° 30° 20° 10° 0° 200 30 0 330 240 300 270 Fig. 2. Angle dependent oscillations a) Angle dependence of the quantum oscillation spectrum measured in the 150nm device. As the field is rotated away from the perpendicular configuration (0°), the surface frequency FS increases and shrinks in amplitude, while the bulk frequency is unaffected. b) Polar plot of the angle dependence for the bulk FB and the surface frequency FS. FB is almost isotropic while FS increases as the field is rotated towards a parallel configuration. It follows a surface-like cos(θ)-1- dependence resembled by a straight, off-center line in a polar plot. a ) t e s f f o ( ) m c W m ( y t i v i t s s e R i b ) B F / S F ( e d u t i l p m a e v i t a e R l 100 80 60 40 20 0 0 4 3 2 1 0 0 1000nm 500nm 150nm 3 Knudsen enhancement 1 2 Field (T) e d u t i l p m A 4 2 0 Thickness in nm 150 350 550 500 1000 2000 Normalized to bulk frequency 60 40 Frequency (T) 80 ~ e-L / 675nm 500 1000 1500 2000 Thickness L (nm) Fig. 3. Thickness dependence a) A magnetoresistance maximum was observed in all studied samples for fields applied parallel to the surface (90°). This peak arises from the semi-classical Knudsen effect observed in quasi-ballistic transport when the cyclotron radius becomes comparable to the sample dimensions. b) Relative amplitude of the surface oscillations compared to the bulk oscillations, for fields perpendicular to the surface at 2K. The surface oscillations are unobservable for devices thicker than 3µm, and their relative weight grows with thinning of the samples. a t n e m e r u s a e M l n o i t a u c a C l no correction =1.25 2K,30°, 150nm 70 Fig. 4. Non-adiabatic corrections a) Comparison of a measured oscillation spectrum to simulations, with and without the non-adiabatic correction term. Without non-adiabatic corrections, the expected spectra are well-defined peaks of similar broadening. The non-adiabatic corrections lead to an asymmetric reduction of the effective amplitude in high magnetic field, and shift spectral weight from the main surface peak to lower frequencies (shaded area). b) Difference between the observed resistance maxima corresponding to the n-th landau level, 1/Bn, and the positions 1/B0 extrapolated from low fields using the usual 1/B periodicity. The difference can be well explained by including the non-adiabatic correction term using α=1.25 (blue dots).12 10 b x10-3 5 r 50 30 Frequency (T) without correction ( =1.25) a Measured B n ) T / 1 ( 0 B / 1 - n B / 1 0 -5 -10 -15 -20 -25 Calculated B with non-adiabatic corr. ( =1.25) a n 10 25 Landau Level Index 15 20 30 0 5 a b c R ecta n gle 50 mm B Tria n gle FB FS Triangle B Rectangle B ) . u . a ( e d u t i l p m A 20 40 80 60 Frequency (T) 100 120 Fig. 5. Absence of surface oscillations in triangular samples a) SEM image of triangular and rectangular devices. The rectangular sample is L = 0.8µm wide, 3.2µm tall and 5µm long. The other device features an equilateral triangular cross-section with a base of a = 2.7µm. Both devices have a similar cross-sectional area (rectangle: 0.8 × 3.2 = 2.6µm2, triangle: a=3.1µm2) and surface length (rectangle: 2 × (3.2+0.8) = 8.0µm, triangle: 8.1µm). The crystallographic direction perpendicular to the surface of the rectangular device is [1 0 2] and [0 1 0] parallel to the surface, which defines the plane of rotation. b) Sketch of the Weyl orbits for rectangular and triangular cross-sections. c) Frequency spectrum of the triangular and rectangular samples, for field orientations perpendicular to each of the surfaces (0° for the rectangle and 60° for the triangle). Methods Focused Ion Beam sample preparation - The Focused Ion Beam (FIB) has proven to be a powerful tool to fabricate crystalline microstructures of high quality, e.g. ref. 29–33, to study meso- and microscopic transport phenomena such as the Weyl-orbit quantum oscillations presented in this study. In this process, the microstructures are carved out of macroscopic crystals using the FIB. Starting from millimeter-sized flux-grown single crystals presented in Extended Data Figure 1, we use a 30kV Ga2+ ion beam to cut the crystal. Depending on the ion flux required, the irradiation spot can be as small as a few nm, and thus structures can be fabricated with high precision. The beam is guided across a negative image of the desired structure, which defines the parts of the crystal to be removed. Electric contacts to the device can also be made in situ in the FIB. A platinum precursor gas can be introduced into the chamber, and upon interaction of the ion beam with the gas adsorbed to the surface a conductive, Pt-rich film can be grown connecting the microstructure electrically to external leads. The fabrication occurs in two steps: In the first step, a cross-sectional slice is cut from the crystal. Typical dimensions of this slice are 100µm × 15µm × 3-4µm. Cd3As2 is found to be sputtered very effectively by the ion beam and even large volumes can be sputter etched with low ion currents at a reasonable rate. This renders Cd3As2 an ideal material for FIB microstructuring and the low currents contribute to the high crystal quality of the fabricated devices. For the coarse patterning in this step, currents of 2.5nA were used. This slice is then removed ex-situ under an optical microscope and transferred to a Si/SiO2 substrate with predefined gold leads connecting the center to bonding pads. The slice is glued in the center using a small drop of epoxy and cured for 1h at 100°C. Good results were obtained with both Araldite Rapid and Stycast 1266 epoxies. Cd3As2 is found to react adversely with the platinum precursor gas leading to an insulating interface with non-ohmic behavior between the platinum contacts and the crystal slice. To avoid this barrier formation, the slice was sputter-coated in 100nm gold, which was later removed by FIB etching from the top-side of the studied microdevices. A more detailed description of the process was previously published in ref34. In the second step, the mounted slice is again introduced into the FIB for final patterning into the thin bar shape shown in the main text. Again the high sputter rate allowed us to use currents as low as 40pA for the structuring process. First, the coarse outline of the final structure was patterned at 300pA, and it was checked that no stray connection due to redeposited material remained. Then the final thickness was adjusted by 40pA milling. The scan strategy for the final approach was a line-by-line, single pass mill from both sides of the device to ensure symmetric, highly polished surfaces. The strong bulk-quantum oscillations indicate the high crystallinity of our microstructures, which was additionally confirmed via X-Ray microdiffraction at the Advanced Light Source (Lawrence Berkeley National Laboratory). Nature of the surface layer - FIB microstructuring is a very gentle process inducing very little bulk damage as evidenced by the strong bulk quantum oscillations present in our Cd3As2 devices. On the other hand, the impacting ions are known to damage the crystal surface and create a completely amorphous surface layer. Therefore particular care must be taken to investigate if a highly conductive layer is formed that exhibits conventional surface Shubnikov – de Haas oscillations. As discussed in the main text, the experimental evidence points against a trivial conductive surface layer as the source of the observed surface-like oscillations: (a) These heavily disordered damage layers are unlikely to show quantum oscillations due to their short mean-free-path, (b) it would not show adiabatic corrections to the quantum oscillation frequency and (c) the triangular device shows no sign of surface oscillations despite its comparable surface area and crystal orientation. Yet it remains essential to understand the surface layer, and therefore we analyze its expected thickness and nature by Monte Carlo simulations of ion interaction with Cd3As2, which provide an estimate of the expected surface layer thickness around 20nm. The interaction of ions with matter can be well captured using Monte-Carlo simulations. We have performed a damage analysis including a calculation of the full cascade damage using the SRIM package (Stopping Range of Ions in Matter)35. Under realistic fabrication conditions of grazing incidence polishing (88° to surface normal), we found a longitudinal stopping range of 7.6nm in Cd3As2. This parameter describes the mean implantation depth of Ga, with a standard deviation of 5.4nm. At grazing incidence, the Ga-ion backscattering probability is strongly enhanced and thus the Ga-implantation is strongly suppressed compared to normal incidence. The main nature of the defects in the surface layers are cascade damages, resulting in a displacement of Cd and As atoms from their positions in the crystal lattice. The first 20nm of material surface are thus expected to be amorphous, with an exponential reduction of defects and damage deeper in the bulk. This is in agreement with the observation of quantum oscillations corresponding to the bulk frequency in our samples. In the following we probe more directly if indeed topological surface states are involved in the surface-like orbit: One consequence of the topological character of the Fermi arc states involved in the Weyl orbit is their protection from surface disorder. This can be directly tested by purposely damaging the FIB-polished surfaces of Cd3As2. By irradiating one side of a device at close-to-normal beam incidence, the surface layer can be severely damaged and disordered. Upon increasing surface damage, the surface state quantum oscillations of a trivial surface state are expected to be dampened due to the decreased quasiparticle lifetime following the Lifshitz-Kosevich mechanism36. On the other hand, the topological surface state existing at the boundary between the periodic crystal bulk and the amorphous surface layer would remain unaffected, as it is simply pushed deeper into the bulk. To inflict maximal surface damage, exceptionally long beam dwell times at 30° angle between the beam and the surface normal were used to increase heating and cause deeper defect penetration (1ms, 80pA, 30kV). Under these conditions, a well visible pattern of shallow dimples appears on the damaged surface (Extended Data Figure (EDF) 2 b). The structure is of the same design as discussed in the main manuscript with a thickness of 810nm from the same Cd3As2 crystal, yielding a ratio of quantum oscillation amplitudes of bulk and surface states of 1.2 as expected at this thickness (Main figure 3). The same device was fabricated with polished surfaces and first well characterized in its pristine state. Then the same device was measured again after the irradiation, thus allowing to directly investigate the changes due to the irradiation. The resistivity of the device increased by 55%, indicating the effective introduction of surface disorder due to the irradiation (EDF 2c). Remarkably, however, the amplitude of the Shubnikov-de Haas oscillations of both the surface and the bulk states have increased due to the irradiation (EDF 2d,e). This is in stark contrast to the expected strong reduction of amplitude due to disorder of a trivial surface state. However, this unusual result is a natural consequence of a topologically protected surface state. The current in such structures is expected to flow in three parallel paths: the crystal bulk, the topological surface state, and the disordered surface layer. By irradiating the surface at low incidence angle, the outermost damage layer becomes more disordered, as evidenced by the increase of the total device resistance. Therefore, the relative weight of the current flowing through the topological states and the bulk increases, leading to the unusual phenomenon of increasing surface state quantum oscillations upon increasing surface damage. This is further supported by the almost unchanged surface to bulk amplitude ratio due to irradiation (EDF 2e). The robustness of the observed surface state quantum oscillations to artificial extensive surface damage strongly suggests a mechanism protecting the surface state from excess scattering at play, which would arise naturally from the topological properties of the Fermi arc states. This effect of topological protection from non-magnetic scattering 20,22 as well as in topological insulators6,7,37. Clearly, such robust quantum coherent has been shown in Cd3As2 signatures are contrary to the expectations for trivial surface states. Quantum oscillation analysis - Magnetic fields quantize the electronic degrees of freedom orthogonal to the field direction. The level spacing ΔE = ħωc = B ħe/m grows linearly with increasing field strength B and the Landau levels are successively emptied as they are pushed above the Fermi level. This leads to a well-known modulation of the density-of-states at the Fermi level (DOS), and conversely to an oscillatory behavior of physical quantities depending on the DOS in high fields. A good overview of such quantum oscillatory effects is given in ref36. The effective mass m* is obtained from the temperature dependent dampening of the quantum oscillations due to thermal broadening of the Landau levels. The thermal dampening is described by the Lifshitz-Kosevich formalism and it can be shown that the oscillation amplitude follows the functional form A(T)/A0 = X/sinh(X) where X = 2π2kBT / (ħωc). Here kB denotes the Boltzmann constant and ωc = eB/m* is the cyclotron frequency. From fitting this temperature dependence to the measured amplitudes, as shown in EDF 3a, the effective mass is obtained as a fitting parameter as 0.044me for the bulk. Using Onsager's relation, the frequency F of the quantum oscillations can be directly related to the extremal cross-sectional areas Sk of the Fermi surface, as F = 2πe Sk / ħ. As the Fermi surface of Cd3As2 is to 2, the Fermi wavevector can be estimated as kF = (4πeF/ħ)1/2 ≈ 3.33 107m-1. good approximation spherical, Sk = πkF The separation between the nodes of 0.8nm-1 as reported in ref.19 well exceeds the experimentally determined kF. The Fermi velocity for our microstructure follows as vF = ħkF/m* ≈ 8.76 105m/s. The Dirac equation now leads to an estimate of the Fermi energy above the Dirac point as EF = ħ vF kF ≈ 192meV. Additional complications for the analysis of the Shubnikov-de Haas effect arise from the quasi-ballistic transport in our clean microstructures that are thinner than the bulk mean free path. At very low fields, a peak in the magnetoresistance with a broad maximum followed by a strong negative magnetoresistance is observed before eventually the bulk magnetoresistance is recovered at high fields (Main Fig. 3a). The quantum oscillations emerge ontop of this background. We note that this field configuration is transverse and thus no negative magnetoresistance associated with the chiral anomaly is expected. Instead this enhancement of scattering in ultra-pure systems is a semi-classical effect arising from the diffuse scattering of otherwise ballistic electrons at the boundaries of strongly confined microstructures38. It is well-studied in ultra-pure hydrodynamic systems, such as the viscous flow of 3He through capillaries (Knudsen-effect) or in high quality, geometrically confined conductors (Gurzhi-flow) such as semiconductor-heterostructures and clean metal whiskers 39. The resistance maximum at the Knudsen peak occurs at maximal boundary scattering of the bent electron trajectories, i.e. at 2rc = L up to a small numerical factor40, where rc = ħkF / (eB) is the cyclotron radius. As a result, the position of the maximum is expected to shift to higher fields as the sample thickness is decreased, as is observed in the Cd3As2 microstructures (See main Fig3a). The Knudsen effect is a direct consequence of ballistic electron motion between the opposite surfaces and thus directly evidences that the thickness of the studied microstructures is indeed comparable to the bulk mean free path. Non-adiabatic corrections - The positions of the Landau levels associated with FS systematically deviate from the usual 1/B periodicity. Fig. 4 of the main text shows the deviation of each maximum of the resistance oscillation, 1/Bn, from its expected position 1/B0 for a usual 1/B periodicity (blue dashed line). At low fields, the oscillations indeed are found to be periodic in 1/B, yet slightly but consistently to deviate at higher fields. The deviations from periodicity are shown in Figure 4b. We find quantitative agreement between the observed deviation and the expectations for non-adiabatic corrections of the Weyl orbit (Fig. 4b). Yet deviations of similar magnitude from periodicity could also occur from spin-splitting of the surface state alone, without the presence of Weyl orbits. A comparison of the presented data with the well- known predictions for spin-splitting, however, shows that the observed peak positions are qualitatively different from expectations in the spin-splitting scenario, further highlighting the unusual character of the observed oscillations (EDF 4). In the presence of strong magnetic fields, the Fermi surface volume for spin-up and spin-down electrons changes due to the Zeeman energy ΔE = ½µBgσzHz, where the z-axis is pointing along the magnetic field. This leads to an effective splitting of an initially spin-degenerate Fermi-surface at zero field (in the absence of spin- orbit coupling). In favorable cases, spin splitting can be directly observed as the appearance of two, well-separated peaks positioned symmetrically around the expected position in the absence of spin-splitting41. We do not observe split peaks in the surface state related oscillations. This could either be an indication for weak spin-splitting, or for a thermal or impurity broadened situation where direct splitting of the peaks is not observable despite strong spin- splitting. In the latter case, spin-splitting is known to modify the amplitude of the oscillations only, without affecting the phase of the oscillations36. Therefore spin-splitting of a trivial surface state alone is unlikely to explain the present data. Another possibility may be the formation of a surface state akin to those in a topological insulator (TI). In this case, the spin-momentum locking on the Dirac cone changes the spin-splitting behavior and oscillation phase changes without any peak splitting are possible42. The reason for this is the opening of a gap in the Dirac spectrum due to the time reversal symmetry breaking of the applied magnetic field. This gap due to the Zeeman energy modifies the Landau level spectrum for electrons (at positive band filling): EN = vF ( 2neħH + (gµBH / (2vF) )2 )1/2 In the absence of spin-splitting (g=0), this reduces into the conventional square-root-dependence of Dirac systems. At low fields, i.e. large Landau level index n, this correction is negligible, yet becomes important at higher fields closer to the quantum limit (n=0). The resulting fields Bn, where a Landau level equals the Fermi Energy EF can be easily calculated using the material parameters self-consistently obtained from the quantum oscillation analysis (blue dots) and are contrasted to the measured peak positions (red dots), shown in EDF 4a. Spin-splitting of a TI-like surface state is at odds with our observations for two reasons: (a) the expected deviation goes in the wrong direction and (b) the observed magnitude is incompatible with measured g-factors in Cd3As2. Spin-splitting would suggest that the fields Bn occur at lower fields compared to the purely periodic case, they are on the contrary observed at higher fields. The shift to lower fields in the case of spin-splitting can be easily understood from the surplus Zeeman energy adding to the field dependence of the Landau levels in the absence of spin splitting (see figure below). This shifts the levels up in energy, and thus they intersect the constant chemical potential at lower fields, hence the upwards bending of the fields Bn as shown above. In addition, a g-factor of 300 is required to obtain a shift of similar magnitude as observed by our experiment, which is an order of magnitude larger than the experimentally measured values43. On the other hand, such a downwards deviation as well as its magnitude is expected for Weyl orbits arising from non-adiabatic corrections: At small fields, the quasiparticle evolving on one of the surface states may only enter the bulk state at the Dirac point, where the gap between the surface and bulk states vanishes. In strong magnetic fields, however, the quasiparticle may tunnel through this band-gap into the bulk state before reaching the Dirac point over a distance given by the magnetic length lB = (ħ / (eB) )1/2. This shortens the effective path length on the surface (red arrows in Fig.1b) and thus leads to a non-adiabatic correction to the low-field surface frequency FS,0 in strong fields given by FS(B) = FS,0 – 4 α FS,0 lB / k0 (2) frequency, as the oscillation period Δ is independent of it: Δ = Bn -1 - Bn-1 -1 = e k0 , where α represents a material dependent parameter encoding the tunneling barrier between the surface and the bulk bands, and FS,0 the field-independent surface frequency in the absence of strong non-adiabatic effects observed at low fields. The four tunneling processes involved in each Weyl orbit thus lower the effective quantum oscillation frequency FS in strong magnetic fields. Including this non-adiabatic correction term well explains the observed deviation (Fig.4b, red points) from the pure 1/B periodicity in high magnetic fields. A fit to the observed data (blue points) yields a coupling parameter α~1.25, in good agreement with the theoretical predicted value of order unity12. This non-adiabatic correction to the FS(B) is also directly evident in the Fourier spectra: While the signature of the bulk frequency is a sharp, symmetrical peak centered at FB, the surface frequency FS is asymmetric with significant spectral weight shifted to lower frequencies. This again is a result of the effective lower frequency in high magnetic fields, and can be well reproduced by a simple calculation inserting the field-dependent frequency given by Eq.2 into the Lifshitz-Kosevich formalism and Fourier-transform the resulting wave-form (Fig.4a of the main text). Unlike the spin-splitting driven phase modification, the non-adiabatic correction mechanism leads to a slowing-down of the oscillations at higher fields due to the effective reduction of flux enclosed in the quantum path that is cut short by the tunneling process. This naturally leads to the down-bending observed in our experiment. Thickness dependence of the oscillation phase - Equation 1 of the main manuscript describes the expected width dependence of the oscillations arising from the Weyl orbit. Crucially, the thickness L does not affect the oscillation -1 π vF / EF. In agreement with this expectation, no thickness-dependence of the frequency is observed. Instead it is expected to only affect the phase offset of the peaks vs 1/B and indeed the data shows no thickness dependence to the peak positions. However, experimentally detecting this phase shift is a challenge for four main reasons: 1) the surface- and bulk- frequencies are close to each other leading to significant beating which complicates the identification of the peak positions corresponding to the surface frequency, 2) the peak positions show an additional field dependence due to the non- adiabatic corrections, 3) the oscillatory signal is on top of a large non-linear and thickness-dependent background magneto-resistance due to finite size effects which must be subtracted to find the peak positions, and 4) the relative amplitude of the surface oscillations compared to the bulk oscillations by itself is strongly thickness dependent, which causes a thickness dependence of the beating pattern. Combined, these uncertainties in fitting the peak positions are larger than the separation between peaks, and therefore from the present data it is impossible to make a quantitative claim about the systematic dependence of the peaks. These difficulties are naturally avoided in the triangular geometry and thickness dependence experiments. By design, the triangular sample averages over parts of different effective width, leading to a cancellation of the surface signal as expected from Eq.1. Upper field limit for surface state oscillations - Reference12 predicted that quantum oscillations associated with the Weyl orbit persist only up to a maximum field that depends on sample thickness, Bsat ~ k0 / (eL). By repeating this analysis, we show that the thickness dependence is actually more involved. Instead, the surface arc oscillations persist up to a field that depends in a complicated non-monotonic fashion on film thickness, but which is at least as large as the quantum limit associated with the k-space area enclosed by the Fermi arcs. To see this, note that from Equation (1) above, we can see that the smallest integer, n=N, for which (1) has a solution is given by: != !!!! −! , where kF = EF/v, and ! indicates the closest integer to x that is larger than x. N represents the index of is then: !sat= !!!!! ! !! !!!!!! . The first factor in Bsat is precisely the 1/B frequency of the surface state, FS =56T. limit of the surface state, and is largely independent of sample thickness. In the present measurements, Bsat is The second factor depends strongly on L in a complicated oscillatory fashion, but is typically of order 1 (though possibly much larger when (kF L / π) - γ is accidentally close to an integer value). Hence, we see that, in contrast to the claims of ref.12, the upper limit field for observing Weyl surface arc oscillations is comparable to the quantum the last quantum level associated with the Weyl orbits that can be pushed across the Fermi-energy. The field at which level N crosses the Fermi-energy, i.e. the upper limit field beyond which Weyl surface-arc oscillations cease, expected to be at least ~60T, and is hence unobservable in any non-pulsed field experiment. Raw oscillation data: 150nm device - As quantum oscillations are phenomena most easily understood in Fourier space, we limit the discussion in the main manuscript on the analysis of FFT spectra. Fourier transforms, however, can be misleading and thus it is very important that the main results are clearly apparent in the raw data themselves. This is clearly the case: EDF 5 and 6 shows the oscillatory magnetoresistance after subtracting a 2nd order polynomial to remove the non-oscillatory components. The same notation of angle is used as in the main manuscript. Scans were performed in 5° steps around the surface state orientations (0°,180°) and in 10° steps elsewhere for the 150nm device, and in 5° steps for the triangle. No averaging or symmetrization of the data between field sweeps of opposite polarity was performed, i.e. between sweeps with 180° angle difference. From the dataset EDF 5, the main Figure 2b showing the surface character of the oscillations was constructed. The main result of the manuscript can be seen as follows: For in-plane fields, i.e. 90°, 270°, a single frequency corresponding to the bulk frequency is observed. As discussed in the main text, both the observation of only one single frequency as well as its amplitude is in agreement with all other published data of quantum oscillations on Cd3As2. For fields perpendicular to the surface, i.e. at 0°, 180°, the spectrum is profoundly different. An additional oscillation of higher frequency is mixed with the bulk frequency, leading to a beating pattern. The second frequency gradually disappears as the field is rotated away from 0°, and is identified with the surface state oscillation from the FFT spectrum as described in the main text. Main Figure 5 showing the differences between the rectangular and triangular devices was constructed from the raw dataset shown in EDF 6. Also here the main points can be well seen without resorting to Fourier transforms: Despite its identical surface and cross-sectional area, the triangle does not show any sign of a second frequency at any angle, as can be seen from the angle-independent positions of the quantum oscillation peaks. In the rectangular case, however, strong beating appears when the field is perpendicular to the surface, signaling the second frequency. Also the characteristic angle dependence of surface states as presented in main Figure 2 can be clearly seen when one traces the maxima from the surface to higher angles. Additional References 29. Moll, P. J. W. et al. Transition from slow Abrikosov to fast moving Josephson vortices in iron pnictide superconductors. Nat. Mater. 12, 134–8 (2013). 30. Moll, P. J. W., Zhu, X., Cheng, P., Wen, H.-H. & Batlogg, B. Intrinsic Josephson junctions in the iron-based multi-band superconductor (V2Sr4O6)Fe2As2. Nat. Phys. 10, 644–647 (2014). 31. Moll, P. J. W. et al. Field induced density wave in the heavy fermion compound CeRhIn5. Nat. Commun. 6, 6663 (2015). Jaroszynski, J. et al. Upper critical fields and thermally-activated transport of Nd(O0.7F0.3)FeAs single crystal. Phys. Rev. B 78, 174523 (2008). Ooi, S., Mochiku, T. & Hirata, K. Periodic Oscillations of Josephson-Vortex Flow Resistance in Bi2Sr2CaCu2O8+y. Phys. Rev. Lett. 89, 247002 (2002). 32. 33. 38. 39. 40. 41. 34. Moll, P. J. W. et al. High magnetic-field scales and critical currents in SmFeAs(O, F) crystals. Nat. Mater. 9, 628–633 (2010). Ziegler, J. F. SRIM-2003. Nucl. Instr. Meth. Phys. Res. B 220, 1027–1036 (2004). Shoenberg, D. Magnetic oscillations in metals. (Cambridge University Press, 1984). 35. 36. 37. Wray, L. A. et al. How robust the topological properties of Bi2Se3 surface are : A topological insulator surface under strong Coulomb, magnetic and disorder perturbations. Nat. Phys. 7, 32–37 (2011). Beenakker, C. W. J. & van Houten, H. Quantum Transport in Semiconductor Nanostructures. Solid State Phys. 44, 1–228 (1991). de Jong, M. & Molenkamp, L. Hydrodynamic electron flow in high-mobility wires. Phys. Rev. B 51, 13389– 13402 (1995). Thornton, T. J., Roukes, M. L., Scherer, A. & Van De Gaag, B. P. Boundary scattering in quantum wires. Phys. Rev. Lett. 63, 2128–2131 (1989). Tokumoto, M. et al. Direct observation of spin-splitting of the Shubnikov-de Haas Oscillations in a Quasi- Two-Dimensional Organic conductor (BEDT-TTF)2KHg(SCN)4. J. Phys. Soc. Japan 59, 2324–2327 (1990). 42. Analytis, J. G. et al. Two-dimensional surface state in the quantum limit of a topological insulator. Nat. Phys. 6, 960–964 (2010). 43. Wallace, P. R. Electronic g-Factor in Cd3As2. Phys. Stat. Sol. 92, 49–55 (1979). 44. Ali, M. N. et al. The crystal and electronic structures of Cd3As2, the three-dimensional electronic analogue of graphene. Inorg. Chem. 53, 4062–4067 (2014). Extended Data Figures a b ) . u . a ( y t i s n e t n I 20 40 2 q 60 (deg) 80 100 Extended Data Figure 1: Crystal synthesis & characterization Cd3As2 single-crystals were produced using the flux growth technique with a 5:1 ratio of Cd flux to Cd3As2. Elemental Cd and As were placed into an aluminum oxide crucible with a quartz wool plug and sealed into a quartz ampule under vacuum. The ampule was heated to 825°C and was held for two days to ensure a fully homogenized mixture. It was then cooled at a rate of 6°C per hour to 425°C, where it was centrifuged to remove excess Cd flux. Bulk crystals were found trapped in the quartz wool following this procedure, and were confirmed to be Cd3As2 by both power X-ray diffraction (PXRD) and single-crystal X-ray diffraction. The PXRD data is explained by the 44 and no parasitic phases were observed. Panel (a) shows a typical facetted I41/acd low-temperature phase of Cd3As2 crystal and (b) the PXRD spectrum. after damage before damage a before damage after damage amorphous layer topological surface state bulk d b c 4mm e after damage before damage after damage before damage Extended Data Figure 2: Surface quantum oscillations are resistant to intentional surface damage a) Sketch of the device cross-section before and after the irradiation. Three possible current paths exist in these devices: The crystal bulk (purple), the topological surface states (orange), and the amorphous FIB-induced damage shell (green). b) Device after the heavy irradiation damage. The dimples due to the beam center impact are well visible across the whole device. The polished backside remained undisturbed by this procedure. c) Comparison of the same device with pristine surfaces and after damaging the surface. The resistance increases after irradiation, as expected for the increased scattering. d) Magnetoresistance of the Cd3As2 microstructure before and after irradiation; and e) the Fourier components of the quantum oscillations. Extended Data Figure 3: Effective mass analysis of bulk and surface oscillations The temperature dependence of the amplitudes yields an effective electron mass in the bulk of 0.044me, in good agreement to previously reported values 26. The surface state appears slightly heavier (0.05me). a b Extended Data Figure 4: Spin splitting: (a) Landau level positions calculated for spin-splitting (blue) and experimentally observed (red). The observed deviation is opposite to the expectation of spin-splitting, yet qualitatively and quantitatively consistent with the non-adiabatic corrections of the Weyl orbit process. (b) Landau fan diagram for Dirac systems. ) t e s f f o ( e d u t i l p m a H d S y r o a t l l i c s O -90° 0° (surf) 90° 180° (surf) 260° 0.1 0.2 0.3 0.4 0.5 -1 Field (T ) -1 Extended Data Figure 5: Raw oscillatory signal of the 150nm device: Shubnikov-de Haas oscillations of the 150nm wide rectangular device, the smallest microstructure fabricated in this study. The appearance of the surface frequency can be well seen in the raw data as strong beating appears around 0°. Also the characteristic cos(θ)-1 angle dependence of a surface frequency can be easily seen by following the peak positions to higher angles. 0° 30° (surf) 90° (surf) 0° (surf) l r a u g n a i r T l r a u g n a t c e R 0.1 0.2 -1 -1 Field (T ) 0.3 90° Extended Data Figure 6: Raw oscillatory signal of the triangle/rectangle device: Shubnikov-de Haas oscillations of the rectangular and triangular devices. While the rectangular device shows the characteristic beating, the peak positions in the triangle remain at the same fields.
1002.1688	1	1002	2010-02-08T19:07:08	Simplified model for the energy levels of quantum rings in single layer and bilayer graphene	[ "cond-mat.mes-hall" ]	Within a minimal model, we present analytical expressions for the eigenstates and eigenvalues of carriers confined in quantum rings in monolayer and bilayer graphene. The calculations were performed in the context of the continuum model, by solving the Dirac equation for a zero width ring geometry, i.e. by freezing out the carrier radial motion. We include the effect of an external magnetic field and show the appearance of Aharonov-Bohm oscillations and of a non-zero gap in the spectrum. Our minimal model gives insight in the energy spectrum of graphene-based quantum rings and models different aspects of finite width rings.	cond-mat.mes-hall	cond-mat	Simplified model for the energy levels of quantum rings in single layer and bilayer graphene M. Zarenia,1 J. Milton Pereira,2 A. Chaves,2 F. M. Peeters,1, 2, ∗ and G. A. Farias2 1Departement Fysica, Universiteit Antwerpen, Groenenborgerlaan 171, B-2020 Antwerpen, Belgium 2Departamento de Física, Universidade Federal do Ceará, Fortaleza, Ceará, 60455-760, Brazil Within a minimal model, we present analytical expressions for the eigenstates and eigenvalues of carriers confined in quantum rings in monolayer and bilayer graphene. The calculations were performed in the context of the continuum model, by solving the Dirac equation for a zero width ring geometry, i.e. by freezing out the carrier radial motion. We include the effect of an external magnetic field and show the appearance of Aharonov-Bohm oscillations and of a non-zero gap in the spectrum. Our minimal model gives insight in the energy spectrum of graphene-based quantum rings and models different aspects of finite width rings. PACS numbers: 71.10.Pm, 73.21.-b, 81.05.Uw I. INTRODUCTION The investigation of low-dimensional solid state devices has allowed the direct observation of quantum behav- ior in electron systems. These effects arise due to the confinement of carriers in structures that constrain their movement along one or more directions, such as quantum wells, quantum wires and quantum dots. One important class of such low-dimensional systems are quantum rings, in which a particular type of confinement together with phase coherence of the electron wavefunction allows the observation of effects such as the Aharonov-Bohm1 and Aharonov-Casher2 effects. Quantum rings have been ex- tensively studied in semiconductor systems, both experi- mentally and theoretically3 and are expected to find ap- plication in microelectronics, as well as in future quantum information devices. In this paper we present analytical results for the eigen- states and energy levels of ideal quantum rings created with graphene and bilayer graphene. Graphene is an atomic layer of crystal carbon which has been the tar- get of intense scrutiny since it has been experimentally produced4–7. Part of this interest stems from the unusual properties of carriers in graphene, caused by the gap- less and approximately linear carrier spectrum, together with possible technological applications, such as transis- tors, gas sensors and transparent conducting materials in e.g. photovoltaics. Additionally, it has been found that two coupled graphene sheets, also known as bilayer graphene displays properties that are distinct from sin- gle layer graphene as well as from graphite. The car- rier spectrum of bilayer graphene is gapless and approx- imately parabolic at the vicinity of two points in the Brillouin zone8,9. In particular, the spectrum is strongly influenced by an external electric field perpendicular to the bilayer, leading to the appearance of a gap10. The high quality of the single layer and bilayer graphene sam- ples that have been obtained, together with the large mean free path of carriers suggests that phase coher- ence effects may be observable in these systems. Re- cently, graphene-based quantum rings produced by litho- graphic techniques have been investigated on single-layer graphene11,12. These systems have been studied theoret- ically by means of a tight-binding model, which does not provide straightforward analytical solutions for the eigen- states and eigenvalues13–17. For bilayer graphene also it was recently shown18 that is possible to electrostatically confine quantum rings with a finite width. The energy spectrum was obtained by solving the Dirac equation nu- merically. In this paper we present a toy model that allows for analytical expressions for the energy levels of quantum rings in single layer and bilayer graphene. This model permits the description of several aspects of the physics of graphene quantum rings without the additional com- plications of edge effects and finite width of the quantum ring. We are able to obtain analytical expressions for the energy spectrum and the corresponding wavefunc- tions, the persistent current, and the orbital momentum as function of ring radius, total momentum and magnetic field, which can be related to the numerical results ob- tained by other methods. The paper is organized as follows: in section II we present the theoretical model and numerical results for quantum rings in single layer graphene. Similar results for bilayer graphene are given in Sect. III. Section IV contains a summary of the main results and the conclu- sions. II. SINGLE LAYER GRAPHENE A. Model The dynamics of carriers in the honeycomb lattice of covalent-bond carbon atoms of single layer graphene can be described by the Dirac Hamiltonian (valid for E < 0.8 eV). In the presence of a uniform magnetic field B = B0ez perpendicular to the plane and finite mass term ∆, which might be caused by an interaction with the underlaying substrate10,19 The Hamiltonian in the valley isotropic form is given by13: H = vF (p + eA).σ + τ ∆σz (1) where τ = +1 corresponds to the K point and τ = −1 to the K ′ point. p is the in-plane momentum operator, A is the vector potential and vF ≈ 1.0 × 106 m/s is the Fermi velocity, and σ = (σx, σy, σz) is the pseudospin operator with components given by Pauli matrices. The eigenstates of Eq. (1) are two-component spinors which, in polar coordinates is given by Ψ(ρ, φ) =(cid:18) φA(ρ)eimφ iφB(ρ)ei(m+1)φ (cid:19) . (2) where m is the angular momentum label. We follow the earlier very successful approach20,21 of ideal one- dimensional (1D) quantum rings in semiconductors with spin-orbit interaction where the Schrödinger equation was simplified by discarding the radial variation of the electron wave function. Thus, in the case of an ideal ring with radius R, the momentum of the carriers in the ra- dial direction is zero. We treat the radial parts of the spinors in Eq. (2) as a constant 2 ǫ is real for any value of β. In the region − 1 2 < δ < 1 2 the energy is real, except for m− < m < m+, when the energy is complex. For the gapless case, i.e. δ = 0, we have m+ = −β and m− = −β − 1 and the energy is real when m < −β − 1 or when m > −β and imaginary otherwise. The wavefunctions are eigenfunctions of the total an- gular momentum operator given by the sum of orbital angular momentum Lz and a term describing the pseudo- spin Sz Jz = Lz + Sz, (8) where Sz = (1/2)σz, with σz being one of the Pauli matrices and the eigenvalues of Jz operator become [m + (1/2)]. The current is obtained using jx,y = vF [ψ†σx,yψ]. The total angular current j = vF [ψ†σφψ] can be calculated using the fact that σφ = ξ(φ)σy, where ξ(φ) =(cid:18)e−iφ 0 0 eiφ(cid:19) . (9) The current for the electrons in the K-valley becomes jK = vF (φ∗ AφB + φ∗ BφA). (10) (3) The total current is then given by j = jK + jK ′. The radial part of the two spinor components are Note that the energy spectrum for an ideal single layer quantum ring for both K and K ′ points are the same. For δ > 1/2 we have that m+ = m∗ − is complex and with β = Φ/Φ0. Ψ(R, φ) =(cid:18) φA(R)eimφ iφB(R)ei(m+1)φ (cid:19) . Because the radial motion is frozen in our model there will be no radial current and the persistent cur- rent will be purely in the angular direction. By solving HΨ(R, φ) = EΨ(R, φ) and using the symmetric gauge A = (0, B0ρ/2, 0), we obtain (m + 1 + β)φB(R) = (ǫ − τ δ)φA(R), (m + β)φA(R) = (ǫ + τ δ)φB(R), (4) where the energy and mass terms are written in dimen- sionless units as ǫ = E/E0, δ = ∆/E0 with E0 = vF /R. The parameter β = (eB0/2)R2 can be expressed as β = Φ/Φ0 with Φ = πR2B0 being the magnetic flux threading the ring and Φ0 = h/e the quantum of mag- netic flux. The homogeneous set of equations (4) is solv- able for the energies This energy can also be written as ǫ = ±p(m + β + 1)(m + β) + δ2. ǫ = ±p(m − m−)(m − m+) where m± = −(β + 1 2 ) ±r 1 4 − δ2. (5) (6) (7) φA(R) = 1, φB(R) = m + β ǫ + τ δ . (11) Notice that the radial current can be calculated using jr = vF [ψ†ξ(φ)σxψ] which leads to the the following re- lation, jr = ivF (φ∗ AφB − φ∗ BφA), (12) where, in the case of our ideal ring we have jr = 0. From Eqs. (10) and (11), one can find the following expression for the total angular current of a single layer quantum ring 4vF ǫ(m + β) j = . (13) ǫ2 − δ2 One can rewrite Eq. (13) in the following form j vF ∂ǫ ∂β(cid:1)K +(cid:0) ∂ǫ ∂β(cid:1)K ′ + =(cid:0) 2(m + β)(ǫ2 + δ2) − (ǫ2 − δ2) (14) ǫ(ǫ2 − δ2) . Since for the ground state energy pδ2 − 1/4 ≤ ǫ ≤ δ and −1/2 ≤ m + β ≤ 0 the last term in Eq. (13) is much smaller than the derivatives of the energy with respect to the flux (Φ) and oscillates around zero. Note that in 1D semiconductor rings the current is exactly given by ∂E/∂Φ, which is thus different from graphene where we have approximately j vF ≃(cid:0) ∂ǫ ∂β(cid:1)K +(cid:0) ∂ǫ ∂β(cid:1)K ′. (15) B. Results The energies as function of ring radius R are shown in Fig. 1, for ∆ = 50 meV, with −10 ≤ m ≤ −1 (magenta curves), 1 ≤ m ≤ 10 (blue curves), and m = 0 (green In the absence of an external magnetic field, curves). the energy is given by E = ±pm(m + 1)(vF /R)2 + ∆2 and the energy branches have a 1/R dependence and approach E → ±∆ for very large radii. Note that for m = 0 and m = −1 the energy E = ±∆ is indepen- dent of R and all branches are two-fold degenerate. For non-zero magnetic field (B = 3T), the right panel shows that the branches have an approximately linear depen- dence on the ring radius for large R, in particular we have E ≃ ±p(αR)2 + ∆2, with α = vF eB0/2. For small radii, E ≃ ±pm(m + 1)/R and all branches diverge as 1/R, except for m = 0 and m = −1. In those cases when R → 0 we have for m = 0 the result E = ±√∆2 + αvF , while m = −1 gives E = ±√∆2 − αvF . Figure 2 presents results for the energy as function of total angular momentum index m, for ∆ = 50 meV, R = 50 nm and for three different values of magnetic field, namely B0 = −5 T (diamonds), B0 = 0 T (circles) and B0 = 5 T (triangles). Notice that for a given B0 the electron energy obtains a minimum for a particular m, i.e. for B0 = 0 (5 T, −5 T) it is m = 0 (9, −10). In fact it is given by m = −(Φ/Φ0 + 1/2) and is independent of ∆. The energy levels as function of the external magnetic field are shown in Fig. 3, for a quantum ring with (a) δ = 1/2, (b) δ = 3/8, (c) δ = 1/4 and (d) δ = 0 with R = 50 nm for −10 ≤ m ≤ −1 (red curves), 1 ≤ m ≤ 10 (blue curves), and m = 0 (green curves). The magnetic field dependence of the spectrum becomes evident if one (5) as ǫ2 − [(m + Φ/Φ0) + 1/2]2 = δ2 − rewrites Eq. 1/4. Thus, for the special case of δ = ∆/E0 = 1/2 the gap is zero and the energy levels are straight lines given by ǫ = ±(m + 1/2 + Φ/Φ0). The energy spectra for δ > 1/2 resemble those found earlier by Recher et al.13 in the case of a finite width graphene ring with infinite mass boundary conditions. An enlargement of Fig. 3 around E = 0 is shown in Fig. 4. The spectrum has an interesting magnetic field dependence with decreasing δ. For δ = 0 the double degeneracy is restored at E = 0. This behavior can be easily illustrated by considering m = 0. The energy in this case is ǫ = ±pβ(β + 1) + δ2 which for δ = 1/2 becomes ǫ = ±(β +1/2) while for δ = 0 it is ǫ = ±pβ(β + 1) and thus ǫ ≃ ±√β for β ≃ 0. In Fig. 5(a) the energy spectrum is plotted vs mag- netic field for δ = 2. Where, the energy has a hyperbolic dependence on the applied magnetic field with minima at Φ/Φ0 = −m − 1/2 and a gap of ∆ǫ = 2pδ2 − 1/4. The exact location of the transitions (orange dots) and the location of the minima points (yellow dots) in the energy spectrum is clarified in Fig. 5(b). The dependence of the energy levels on the gap parameter ∆ is shown in Fig. 6, for zero magnetic field (left panels) and B0 = 1 T (right 3 FIG. 1: (Color online) Energy levels with m = −10, ..., 10 of single layer graphene quantum ring as function of ring radius R for B0 = 0 T (left panel) and B0 = 3 T (right panel) when the mass term is ∆ = 50 meV. panels). When m ≥ 0 (upper panels) and m < 0 (lower panels). For B0 = 0 T the energy levels are two fold de- generate where E(m) = −E(−m− 1). When a magnetic field is applied an energy gap is opened (see right-bottom panel in Fig. 6). Notice also that the m = −2 level only exists for ∆ ≥ E0/2, i. e. for ∆ < E0/2 there is no real energy solution when m = −2. The corresponding ground state expectation values for the operators in Eq. (8) are plotted as function of the magnetic field in Fig. 7(b) for both K (black dashed curve) and K ′ valley (black dash-dotted curve). Notice that for the K-valley < Lz >≃ m and < Sz >≃ /2 whereas in the K ′-valley < Lz >≃ (m+1) and < Sz >≃ −/2. Thus for both the K valley and the K ′ valley < Jz >≃ [m + (1/2)] which is approximately quantized and on the average its value decreases linearly with the applied magnetic field. The angular current density for a single layer graphene quantum ring is shown in Fig. 8(c). Note that the con- tribution from the K-valley jK (Fig. 8(a)) and the K ′- valley jK ′ (Fig. 8(b)) are not the same, they have oppo- site sign and oscillate in phase around a nonzero average value −τ vF /4. The reason is that if for given energy we have electrons in the K-valley, the corresponding parti- cles in the K ′-valley will behave as holes. The persistent current is a sawtooth shaped oscillating function of the magnetic field with period ∆B0 = Φ0/πR2. This be- havior is quantitatively very similar to those found for the standard Aharanov-Bohm oscillations in metallic and semiconductor quantum rings. B0 = 5 T B0 = 0 T B0 = −5 T ) V e m ( E 200 150 100 50 0 −50 −100 −150 −200 −30 −20 −10 10 20 30 0 m FIG. 2: (Color online) Energy levels of a single layer graphene quantum ring as function of the quantum number m for B0 = −5, 0, 5 T with ∆ = 50 meV and R = 50 nm. III. BILAYER GRAPHENE A. Model In the case of bilayer graphene the Hamiltonian in the vicinity of the K point, is given by10 4 δ = 1/2 10, ..., 1 = m m = 0 m = −1, ..., −10 (a) δ = 3/8 (b) δ = 1/4 (c) δ = 0 (d) ) V e m ( E ) V e m ( E ) V e m ( E ) V e m ( E 40 20 0 20 40 40 20 0 20 40 40 20 0 20 40 40 20 0 20 40 5 4 3 2 1 0 1 2 3 4 5 B0 (T ) FIG. 3: (Color online) Electron and hole energy levels of a single layer graphene quantum ring as function of external magnetic field B0 for (a) δ = 1/2, (b) δ = 3/8, (c) δ = 1/4 and (d) δ = 0 with R = 50 nm, and total angular quantum number −10 ≤ m ≤ −1 (red curves), 1 ≤ m ≤ 10 (blue curves) and m = 0 (green curves). τ U1 π† t 0 π τ U1 0 0 t 0 0 0 τ U2 π† τ U2 π H =    where τ = ±1 distinguishes the two K and K ′ val- leys. t ≃ 400 meV is the interlayer coupling term, π = vF [(px + eAx) + i(py + eAy)], U1 and U2 are the potentials, respectively, at the two graphene layers. Here we do not include any mass term because the gate potential across the bilayer is able to open an energy gap in the spectrum.10 The eigenstates of the Hamiltonian (16), are four-component spinors Ψ(r, φ) = [φA(ρ)eimφ , iφB(ρ)ei(m−1)φ , φC (ρ)eimφ , iφD(ρ)ei(m+1)φ]T (see Ref. ideal ring with radius R, the wave function becomes: [22]). Following our earlier approach for an (16) δ = 1/2 δ = 3/8 δ = 1/4 δ = 0 ) V e m ( E 10 8 6 4 2 0 −2 −4 −6 −8 −10 m = 1 m = 0 m = −1 −1 −0.5 0 0.5 B0 (T ) φA(R)eimφ iφB(R)ei(m−1)φ φC (R)eimφ iφD(R)ei(m+1)φ Ψ =    . (17) FIG. 4: (Color online) The same as Fig. 3, for m = −1 (red curves), m = 1 (blue curves) and m = 0 (green curves) and different values of dimensionless mass term δ. We use the symmetric gauge and obtain the following set of coupled algebraic equations −(ǫ − τ u1)φA(R) − (m + β − 1)φB(R) + t′φC (R) = 0, (m + β)φA(R) + (ǫ − τ u1)φB(R) = 0, t′φA(R) − (ǫ − τ u2)φC (R) + (m + β + 1)φD(R) = 0, (18) (m + β)φC (R) − (ǫ − τ u2)φD(R) = 0. where, t′ = t/E0 and u1,2 = U1,2/E0 are in dimension- less units. After some straightforward algebra we obtain δ = 2 10, ..., 1 = m m = 0 m = −1, .. − 10 60 40 20 0 −20 −40 −60 ) V e m ( E −8 −6 −4 −2 0 2 4 6 B0 (T ) ) V e m ( E 27 26.5 26 25.5 25 −5 E = ∆ E = p∆2 − (1/4)E 2 0 −4 −3 −2 −1 0 1 2 3 4 B0 (T ) (a) 8 (b) FIG. 5: (Color online) (a) Electron and hole energy levels of a single layer graphene quantum ring as function of external magnetic field B0 for δ = 2 and R = 50 nm. (b) An Enlarge- ment of the region which is shown in (a) by a rectangle. the following polynomial equation that determines the energy spectrum (ǫ − τ u1)2(cid:2)(ǫ − τ u2)2 − (m + β)(m + β + 1)(cid:3) −(m + β)(m + β − 1)(cid:2)(ǫ − τ u2)2 − (m + β)(m + β + 1)(cid:3) −(ǫ − τ u1)(ǫ − τ u2)t′2 = 0. After introducing the average potential u = (u1 + u2)/2 and half the potential difference δ = (u1 − u2)/2 we can rewrite this quartic algebraic equation in a more compre- hensive form: (19) s4 − 2s2[(m + β)2 + δ2 + (t′)2/2] +4sτ δ(m + β) + (m + β)2[(m + β)2 − 1] −2δ2[(m + β)2 − (t′)2/2] + δ4 = 0, (20) where s = ǫ − τ u is the energy shifted by the average potential. In the next section we report the results for the case of U1 = −U2 = Ub where, the average potential u is zero. In the limit δ → 0, the quartic equation is reduced to a quadratic equation in s2 and we obtain the real solutions ± = (m + β)2 + (t′)2/2 s2 ±p(t′)4/4 + (m + β)2(1 + t′)2, which results in four solutions for the energy. These are real when m+β ≥ 1. In the opposite case of m+β < 1 (or equivalently −1+β < m < 1−β) we have s2 − < 0 and consequently the corresponding energies are imaginary. (21) B0 = 0 T B0 = 1 T 5 6 5 4 3 2 1 −7 −6 −5 −4 −3 −2 4 3 2 1 m = 0 m = 0 10 5 0 −5 −10 −3 −3 −9 −8 −7 −6 −5 −4 −1 −5 0 5 m = −2 m = −1 −50 0 50 ∆ (meV ) −50 0 50 ∆ (meV ) ) V e m ( E ) V e m ( E 100 50 0 −50 −100 100 50 0 −50 −100 FIG. 6: (Color online) Lowest energy levels of a single layer graphene quantum ring as function of the mass term ∆ with B0 = 0 T (left panels) and B0 = 1 T (right panels) for m ≥ 0 (upper panels) and m < 0 (lower panels) with R = 50 nm. In the limit of t′ >> m+β we obtain s2 − = (m+β)2[(m+ β)2 − 1]/(t′)2 and thus the low energy solutions are given by 1 s ≃ ± t′p(m + β)2((m + β)2 − 1). For bilayer graphene, the wavefunctions are eigenfunc- tions of the following operator: Jz = Lz + τz + Sz, where now (22) (23) (24) 0 I(cid:19) , Sz = 1 0 2(cid:18)σz 0 −σz(cid:19) , 1 τz = 2(cid:18)−I 0 are 4 × 4 matrices. sity are given by In bilayer graphene the components of the current den- jx = vF (cid:20)ψ†(cid:18)σx 0 0 σx(cid:19) ψ(cid:21) , jy = vF (cid:20)ψ†(cid:18)−σy 0 σy(cid:19) ψ(cid:21) . 0 (25) The angular current can be calculated from the following relation, j = vF (cid:20)ψ†(cid:18)σ∗ 0 y ξ(φ) 0 ξ(φ)σy(cid:19) ψ(cid:21) . (26) Where, ξ(φ) is given by Eq. angular current in the K-valley, (9). We obtain for the jK = vF (φ∗ C φD + φ∗ DφC − φ∗ AφB − φ∗ BφA), (27) ) V e m ( E > ¯h / z L < 50 40 30 20 6 4 2 0 −2 −4 −6 ′ K ) ¯h / > z S < ( −0.44 −0.46 −0.48 −0.5 −3 (a) K ) ¯h / > z S < ( 0.5 0.48 0.46 0.44 −2 0 B0 (T ) 2 −2 −2 0 B0 (T ) −1 2 0 1 2 (b) 3 B0 (T ) F v / K j F v / ′ K j 0 −0.1 −0.2 −0.3 −0.4 −0.5 0.5 0.4 0.3 0.2 0.1 0 0.5 F v / j 0.25 0 −0.25 −0.5 −8 −6 −4 −2 0 2 4 6 8 Φ/Φ0 −4 −3 −2 −1 0 1 2 3 4 B0 (T ) 6 (a) (b) (c) FIG. 7: (Color online) (a) Electron energy levels of a graphene single layer quantum ring as function of external magnetic field B0 for the same parameters as used in Fig. 5. Black curve shows the ground state energy (b) Ground state expec- tation value of Lz/ as function of magnetic field for both K (black dashed curve) and K ′ valley (black dash-dotted curve). Expectation value of Sz/ versus magnetic field is plotted in upper inset for K-valley and in lower inset for K ′-valley. Blue solid curve shows the expectation value < Jz > which is the same for both valleyes. and the total angular current is given by j = jK + jK ′. Where, the four spinor components are: φA(R) = 1, φB(R) = − φC (R) = φD(R) = , m + β ǫ − τ u1 (ǫ − τ u1)2 − (m + β)(m + β − 1) (m + β)[(ǫ − τ u1)2 − (m + β)(m + β − 1)] t′(ǫ − τ u1) , . t′(ǫ − τ u1)(ǫ − τ u2) (28) Note that the radial current can be calculated through jr = vF (cid:20)ψ†(cid:18)σxξ(φ) 0 AφB − φ∗ 0 ξ(φ)σx(cid:19) ψ(cid:21) , C φD − φ∗ where, jr = ivF (φ∗ DφC ) = 0 for the case of an ideal ring. Using Eq. (27), the total current density becomes, BφA + φ∗ (29) j = Xτ =±1 ǫ − τ u1 2vF (m + β) "1 + (cid:2)(ǫ − τ u1)2 − (m + β)(m + β − 1)(cid:3)2 t′2(ǫ − τ u1)(ǫ − τ u2) # . FIG. 8: The angular current density in the (a) K-valley, (b) K ′-valley and (c) the total current density of a monolayer graphene quantum ring as function of external magnetic field B0 for the ground state energy shown by the black curve in Fig. 7(a). ) V e m ( E ) V e m ( E 500 250 0 250 500 500 250 0 250 500 B0 = 0 T m = 1, 0 m = −1 m = 1 m = 0, −1 100 50 0 50 100 0 2 4 6 8 (a) B0 = 5 T 100 m = 1 m = 0 50 0 50 100 m = −1 m = 1 m = 0 m = −1 4 2 0 6 8 (b) 0 20 40 60 80 100 R (nm) FIG. 9: (Color online) Lowest energy levels of a bilayer graphene quantum ring as function of ring radius R with (a) B0 = 0 T and (b) B0 = 5 T for Ub = 100 meV and to- tal angular quantum number −10 ≤ m ≤ −1 (red curves), 1 ≤ m ≤ 10 (blue curves) and m = 0 (green curves). The insets are an enlargement of the small energy and small R region. 7 ) V e m ( E 400 300 200 100 0 −100 −200 −300 −400 B0 = 5 T B0 = 0 T B0 = −5 T −40 −30 −20 −10 10 20 30 40 0 m FIG. 10: (Color online) Lowest energy levels of a graphene bilayer quantum ring as function of total angular momentum label m for B0 = −5, 0, 5 T with Ub = 100 meV and R = 50 nm. ) V e m ( E ) V e m ( E 110 105 100 95 90 85 80 −80 −85 −90 −95 −100 −105 −110 −15 10, ..., 1 = m m = 0 m = −1, .. − 10 −10 −5 0 5 10 15 B0 (T ) FIG. 11: (Color online) Electron and hole energy states of a graphene bilayer quantum ring as function of external mag- netic field B0 for Ub = 100 meV and R = 50 nm. The energy levels are shown for the quantum numbers −10 ≤ m ≤ −1 (red curves), 1 ≤ m ≤ 10 (blue curves) and m = 0 (green curves). (30) B. Results The dependence of the spectrum on the ring radius, for B0 = 0 T (upper panel) and B0 = 5 T (lower panel) is shown in Fig. 9, for a gate potential Ub = 100 meV, which m = −1, ..., −10 m = 0 10, ..., 1 = m ) V e m ( E 50 40 30 20 10 0 −10 −20 −30 −40 −50 −10 −5 0 5 10 B0 (T ) FIG. 12: (Color online) The same as Fig. 11, but for Ub = 0 meV. for B = 0 T opens up a gap in the energy spectrum. As compared to the single layer quantum ring results of Fig. 1, we find two main differences: i) for R → 0 there are two states inside the gap, and ii) we have a second set of levels that for large R are displaced in energy by t. In the limit R → 0 the most important term in the dispersion relation is (m + β)2(cid:2)(m + β)2 − 1(cid:3). For m = −1, 0, 1 the behavior of the spectrum is different and the corresponding energy levels don't diverge when R → 0. The same behavior was found for the single layer results, but only for m = 0,−1. Previously, we found that for rings with finite width18 the spectrum exhibits anti-crossing points which arise due to the overlap of gate-confined and magnetically-confined states. In the present model the carriers motion along the radial direction is neglected and consequently we have level crossings instead of anti-crossing points in the spec- trum. The dependence of the energy eigenstates on the angular momentum index m is displayed in Fig. 10 for Ub = 100 meV, R = 50 nm, with B0 = −5 T (di- amonds), 0 T (circles) and 5 T (triangles). Due to the finite bias in this case, the fourth-order character of the dispersion Eq. (19) causes the curves to exhibit a Mex- ican hat shape. The energy minima for B0 = −5, 0, 5 T are respectively given by m = −1,−10,−20. In Fig. 11 the energy levels are plotted as function of magnetic field, for a quantum ring with Ub = 100 meV, R = 50 nm, and for −10 ≤ m ≤ −1 (red curves), 1 ≤ m ≤ 10 (blue curves), and m = 0 (green curves). These results are very similar to those found for a finite width ring and exhibit two local minima that are separated by a saddle point. In the case of finite width quantum rings there are additional energy levels correspond with states that are partly localized outside the ring. Figs. 11(a,b) show the asymmetry between the electron and hole states, caused by the bias. It is seen that the electron and hole energies are related by Eh(m, B0) = −Ee(−m,−B0), where the indices h(e) refer to holes (electrons). In the absence of bias, the electron-hole symmetry is restored, as shown in 8 (a) ¯h / > z S < −0.425 −0.43 −0.435 −0.44 −2 −1 0 1 2 B0 (T ) m −0.32 −0.34 −2 −1 0 B0 (T ) 1 2 (b) m ¯h / > z S < 0.44 0.435 0.43 0.425 ¯h / > z τ < −2 −1 0 1 2 B0 (T ) ) V e m ( E 87.75 87.7 87.65 87.6 −6 −8 −10 −12 −14 14 12 10 8 K ) ¯h / > z L < ( ′ K ) ¯h / > z L < ( ¯h / > z τ < 0.34 0.32 −2 −1 0 B0 (T ) 1 2 6 −2 −1.5 −1 −0.5 0 0.5 1 1.5 B0 (T ) (c) 2 (Color online) Lowest energy levels of a bilayer FIG. 13: graphene quantum ring as function of the gate potential Ub when B0 = 0 T (left panels) and B0 = 1 T (right panels) for m ≥ 0 (upper panels) and m < 0 (Lower panels) with R = 50 nm. Fig. 12, for a ring with R = 50 nm and the parabolic energy spectrum is recovered with zero energy gap. In Fig. 13, the energy branches are plotted as func- tion of the bias, for both the zero field case (left panels) and for B0 = 1 T (right panels), with m ≥ 0 (upper panels) and m < 0 (lower panels). Notice that the fig- ures are quantitatively similar to those found previously for a quantum ring made of a single layer of graphene where the gate potential Ub has a similar effect as the mass term ∆. The differences are that for B0 = 0 T the degeneracies are now: i) E(0) = E(1) = E(−1) and ii) E(m) = E(−m) for m > 1. In the presence of the mag- netic field a gap is opened even for Ub = 0 meV, which is more clearly illustrated in the inset of the right-bottom panel of Fig. 13. Notice that here we found that for m = −1 and m = −2 no real energy solution is found for Ub below some critical value. Figs. 14(b,c) shows the ground state expectation value of the angular momentum versus the magnetic field to- gether with the quantum number m (blue solid curve) which is an eigenvalue of the total momentum operator Jz. Notice that the expectation value of Jz, i.e. m, is dif- ferent in the K and K ′ valley which was not the case for monolayer graphene. The energy levels for the K (solid red curves) and K ′ (dashed blue curves) valleys are de- picted in Fig. 14(a). Black curve (gray curve) shows the ground state energy for the K-valley (K ′-valley). Notice that in the considered case we find that the difference FIG. 14: (Color online) (a) Electron energy levels of a bilayer graphene quantum ring as function of external magnetic field B0 for a quantum ring of radius R = 50 nm and with Ub = 100 meV for both the K-valley (solid curves) and the K ′- valley (dashed curves). Black curve shows the ground state energy of the energy spectrum in K-valley whereas the gray curve the corresponding ground state energy of the K ′-valley (b) Ground state expectation values of Lz/, Sz/, τz/ as function of magnetic field in the K-valley. Blue solid curve shows the expectation value of Jz/ operator. (c) The same as (b) but, for K ′-valley. between < Jz >= m and < Lz > is about (0.7 − 0.8) for both K and K ′ valleys. The ground state angular current of a bilayer graphene as function of magnetic field B0 in the K-valley jK, the K ′-valley jK ′ and the total angular current j is shown respectively in Fig 15(a), (b) and (c). In the case of a bilayer graphene quantum ring, the energy levels in the vicinity of the K and K ′ points are different because of the valley splitting and consequently the total angular current versus magnetic field is a more complicated saw- tooth function. Notice that the angular current for the K or K ′ valley is not zero at B0 = 0 which is due to the valley polarization whereas the total current is zero at B0 = 0. IV. SUMMARY AND CONCLUSIONS In summary we considered the behavior of carriers in single and bilayer graphene quantum rings within a toy model. Our approach leads to analytic expressions for the energy spectrum. In our simple model we are not faced with the disadvantages of the nature of edge effects which appears in quantum rings created by cutting the −3 −2 −1 Φ/Φ0 0 1 2 3 layer of graphene (or lithography defined quantum rings). 9 We found an interesting new behavior in the presence of a perpendicular magnetic field, which has no analogue in semiconductor-based quantum rings. In single layer graphene quantum rings only for ∆ > vF /2R we found the opening of a gap in the energy spectrum between the electron and hole states. For both single layer and bilayer graphene quantum rings the eigenvalues are not invariant under a B0 → −B0 transformation and in the case of bilayer the spectra for a fixed total angular momentum index m, their field dependence is not parabolic, but exhibit two minima separated by a saddle point. The persistent current exhibits oscillations as function of the magnetic field with period Φ0/πR2 which are the well-known Aharonov-Bohm oscillations. Because of the valley splitting in the energy spectrum of bilayer graphene the total current density versus magnetic field is a more complicated sawtooth function. (a) (b) (c) −1.5 −1 −0.5 0 0.5 1 1.5 2 B0 (T ) 0.02 0 −0.02 −0.04 F v / K j 0.04 0.02 0 F v / ′ K j F v / j −0.02 0.04 0.02 0 −0.02 −0.04 −2 FIG. 15: The ground state angular current density in the (a) K-valley, (b) K ′-valley and (c) the total current density of a bilayer graphene quantum ring as function of external magnetic field B0 with Ub = 100 meV and R = 50 nm. Acknowledgements This work was supported by the Flemish Science Foundation (FWO-Vl), the Belgian Sci- ence Policy (IAP), the Bilateral program between Flan- ders and Brazil, and the Brazilian Council for Research (CNPq). ∗ Electronic address: [email protected] 1 Y. Aharonov and D. Bohm, Phys. Rev. 115, 485 (1959). 2 Y. Aharonov and A. Casher, Phys. Rev. Lett. 53, 319 (1984). 3 A. Fuhrer, S. Lüscher, T. Ihn, T. Heinzel, K. Ensslin, W. Wegscheider, and M. Bichier, Nature (London) 413, 822 (2001). 4 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). 5 Y. Zhang, Y. W. Tan, H. L. Stormer, and P. Kim, Nature (London) 438, 201 (2005). Phys. Rev. B 77, 085413 (2008). 12 F. Molitor, M. Huefner, A. Jacobsen, A. Pioda, C. Stampfer, K. Ensslin, and T. Ihn, arXiv:0904.1364v1. 13 P. Recher, B. Trauzettel, A. Rycerz, Ya. M. Blanter, C. W. Beenakker, and A. F. Morpurgo, Phys. Rev. B 76, 235404 (2007). 14 J. Wurm, M. Wimmer, H.U. Baranger, and K.Richter, arXiv:0904.3182v1 15 T. Luo, A.P. Iyengar, H.A. Fertig, and L. Brey, arXiv:0907.3150v1 16 A. Rycerz, Act. Phys. Pol. A 115, 322 (2009). 17 D. A. Bahamon, A. L. C. Pereira, and P. A. Schulz, Phys. 6 V. P. Gusynin and S. G. Sharapov, Phys. Rev. Lett. 95, Rev. B 79, 125414 (2009). 146801 (2005). 7 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 (London) 438, 197 (2005). 8 B. Partoens and F. M. Peeters, Phys. Rev. B 74, 075404 (2006). 18 M. Zarenia, J. M. Pereira Jr., F. M. Peeters, and G. A. Farias, Nano Lett. 9, 4088, (2009). 19 P. Recher, J. Nilsson, G. Burkard, and B. Trauzettel, Phys. Rev. B 79, 085407 (2009). 20 F. E. Meijer, A. F. Morpurgo, and T. M. Klapwijk, Phys. Rev. B 66, 033107 (2002). 9 T. Ohta, A. Bostwick, T. Seyller, K. Horn, and E. Roten- 21 B. Molnár, F. M. Peeters, and P. Vasilopoulos, Phys. Rev. berg, Science 313, 951 (2006). B 69, 155335 (2004). 10 Edward McCann and Vladimir I. Fal'ko, Phys. Rev. Lett. 22 J. M. Pereira Jr., P. Vasilopoulos, and F. M. Peeters, Nano 96, 086805 (2006). 11 S. Russo, J. B. Oostinga, D. Wehenkel, H. B. Heersche, S. S. Sobhani, L. M. K. Vandersypen, and A. F. Morpurgo, Lett. 7, 946, (2007).
1809.10880	1	1809	2018-09-28T06:49:32	Hydrogen adsorption-induced nanomagnetism at the Si(111)-(7$\times$7) surface	[ "cond-mat.mes-hall" ]	The creation of magnetism on non-magnetic semiconductor surfaces is of importance for the realization of spintronics devices. Especially, the coupling of electron spins within quantum nanostructures can be utilized for nanomagnetism applications. Here, we demonstrate, based on first-principles density-functional theory calculations, that the adsorption of H atoms on the Si(111)-(7$\times$7) surface induces the spin polarization of surrounding Si dangling bonds (DBs) and their spin orderings. It is revealed that the H adsorption on a rest-atom site exhibits a Jahn-Teller-like distortion that accompanies a charge transfer from the rest atom to the nearest neighboring adatoms. This charge transfer increases the local density of states of such three adatoms at the Fermi level, thereby inducing a Stoner-type instability to produce a ferrimagnetic order of adatom DBs around the adsorbed H atom. Meanwhile, the H adsorption on an adatom site cannot induce spin polarization, but, as adsorbed H atoms increase, the ferrimagnetic order of rest-atom DBs emerges through the charge transfer from rest atoms to adatoms. Our findings provide a microscopic mechanism of the H-induced spin orderings of Si DBs at the atomic scale, which paves a novel way to the design of nanoscale magnetism in the representative semiconductor surface.	cond-mat.mes-hall	cond-mat	a Hydrogen adsorption-induced nanomagnetism at the Si(111)-(7×7) surface Xiao-Yan Ren1, Chun-Yao Niu1, Seho Yi2, Shunfang Li1, and Jun-Hyung Cho2,∗ 1 International Laboratory for Quantum Functional Materials of Henan, and School of Physics and Engineering, Zhengzhou University, Zhengzhou 450001, China 2 Department of Physics, Research Institute for Natural Science, and HYU-HPSTAR-CIS High Pressure Research Center, Hanyang University, 222 Wangsimni-ro, Seongdong-Ku, Seoul 04763, Korea (Dated: May 31, 2021) The creation of magnetism on non-magnetic semiconductor surfaces is of importance for the realization of spintronics devices. Especially, the coupling of electron spins within quantum nanostructures can be utilized for nanomagnetism applications. Here, we demonstrate, based on first-principles density-functional theory calcula- tions, that the adsorption of H atoms on the Si(111)-(7×7) surface induces the spin polarization of surrounding Si dangling bonds (DBs) and their spin orderings. It is revealed that the H adsorption on a rest-atom site exhibits a Jahn-Teller-like distortion that accompanies a charge transfer from the rest atom to the nearest neighboring adatoms. This charge transfer increases the local density of states of such three adatoms at the Fermi level, thereby inducing a Stoner-type instability to produce a ferrimagnetic order of adatom DBs around the adsorbed H atom. Meanwhile, the H adsorption on an adatom site cannot induce spin polarization, but, as adsorbed H atoms increase, the ferrimagnetic order of rest-atom DBs emerges through the charge transfer from rest atoms to adatoms. Our findings provide a microscopic mechanism of the H-induced spin orderings of Si DBs at the atomic scale, which paves a novel way to the design of nanoscale magnetism in the representative semiconductor surface. PACS numbers: I. INTRODUCTION Nanoelectronic devices have been developed by the iden- tification of various types of nanostructures such as quantum dots, quantum wires, and quantum wells1,2. These artificially fabricated quantum structures often exhibit a number of ex- otic phenomena different from their bulk counterparts due to confined electrons in the low dimensions3,4. Especially, the coupling of electron spins within such nanostructures can be utilized for the design of nanoscale spintronics devices where the functionality of individual atomic spins is geared towards storage capacity, computing speed, and energy saving5 -- 7. In the present study, we demonstrate that the extensively investi- gated Si(111)-(7×7) surface exhibits the spin orderings of Si DBs (which can be treated as quantum dots)8 at the atomic scale around adsorbed H atoms, thereby providing a play- ground for the atomic engineering of future spintronics and quantum information devices. Similarly, a recent scanning tunneling microscopy (STM) experiment9 observed that the adsorption of a single H atom on graphene induces a magnetic moment by removing one pz orbital from the delocalized π-bonding network, which has al- ready been predicted by purely theoretical works10. Because of the bipartite sublattices A and B of graphene, the magnetic moment is essentially localized on the carbon sublattice op- posite to the one where the H atom is chemisorbed. Such an H-induced magnetic moment exhibits a large spatial ex- tension over several next-nearest neighbors away from the H atom9. Meanwhile, the presently predicted H-induced mag- netic moment on Si(111)-(7×7) extends only up to the nearest neighboring adatoms or rest atoms, as discussed below. Thus, the present H-induced magnetism on Si(111)-(7×7) shows a strongly localized character which would be useful for nano- magnetism applications. Typeset by REVTEX Due to its complex structural and electronic properties, the Si(111)-(7×7) surface has been the most extensively studied semiconductor surface over several decades. According to the Takayanagi's model11, this surface has 19 DBs per 7×7 unit cell, where 12 DBs belong to the adatoms, 6 DBs to the rest atoms, and 1 DB to the Si atom at the bottom of corner hole [see Fig. 1(a)]. It has been well established that there is an electronic charge transfer from the adatoms to rest atoms12. Consequently, the DB states arising from the rest atoms and corner hole atoms are completely filled, while those from the adatoms are partially occupied to leave a band that crosses the Fermi level EF [see Fig. 1(b)]. Photoemission and elec- tron energy-loss spectroscopies measured an intriguing den- sity of states (DOS) around EF, while STM observed differ- ent local DOS (LDOS) depending on the surface atoms [see Fig. 1(c)]13. By employing adsorbates on Si(111)-(7×7), this LDOS can be tuned to influence its structural and elec- tronic properties. For instance, when H atoms adsorb on the rest-atom sites of the Si(111)-(7×7) surface, an earlier STM experiment13 revealed that there is a reverse charge transfer from the rest atoms back to the surrounding adatoms. Al- though such an H-adsorbed Si(111)-(7×7) surface system has so far been much studied experimentally and theoretically, most of them were related with the adsorption sites and dif- fusion path or the structural phase transition induced by H adsorption14 -- 20. Interestingly, using density-functional the- ory (DFT) calculations, Okada et al21. predicted the mag- netic ordering of several DB networks on an otherwise H- covered Si(111)-(1×1) surface, whereas Erwin and Himpsel22 also predicted the localized spin arrays of DB electrons along the Si step edges in the Au-induced vicinal Si(111) surfaces. However, the investigation of magnetism on a realistic H- adsorbed Si(111) surface with the 7×7 structure is still lack- ing. F 3 5 1 2 3 1 2 6 4 x x (b) K M (cid:299) KM (cid:299) (cid:299) 0 0.02 0.04 y z 0.75 0.5 0.25 0 ) V e ( y g r e n E -0.25 -0.5 -0.75 (cid:299) (a) UF A R C (c) R1 A1 A2 ) V e / s e a t s ( t S O D 0.6 0.3 0 -0.3 -0.6 0.3 0.2 0.1 0 -0.1 -0.2 -0.3 M K (cid:299) -2 -1 0 1 2 Energy (eV) FIG. 1: (Color online) (a) Top and side views of the optimized struc- ture of the Si(111)-(7×7) surface. The colored large, medium, and small circles represent the adatoms (A), rest atoms (R), and corner- hole atoms (C), respectively. F and UF represent the faulted and the unfaulted half cells, respectively. The adatoms and rest atoms in the F half cell are numbered. The calculated band structure of the Si(111)- (7×7) surface is given in (b). Here, the bands originating from A, R, and C atoms are distinguished with the same colors as used in (a) and the energy zero represents the Fermi level. The surface Brillouin zone of the 7×7 unit cell is drawn in the inset of (b). The LDOS obtained at R1, A1, and A2 is given in (c). In this paper, we take the Si(111)-(7×7) surface to investi- gate the effect of H adsorption on possible spin orderings of Si DBs using first-principles DFT calculations. We reveal that under H adsorption the charge redistribution of DB electrons takes place differently depending on the adsorption sites. We find that the H adsorption on a rest-atom site causes a charge transfer back to its nearest neighboring adatoms, thereby lead- ing to an increased LDOS of such three adatoms at EF. The re- sulting high LDOS induces the Stoner-type instability to yield the ferrimagnetic order of adatom DBs around the adsorbed H atom. On the other hand, when an H atom adsorbs on an adatom site, a charge abstraction takes place from its sur- rounding adatoms to the adatom, forming an H−Si bond with- out spin polarization. However, as the number of adsorbed H atoms within the 7×7 unit cell increases, some rest atoms be- gin to participate in charge donation to the H−Si bond forma- tions, therefore inducing the ferrimagnetic order of the rest- atom DBs. The present results not only elucidate the underly- ing mechanism of the H-induced spin orderings of DBs on the Si(111)-(7×7) surface, but also open a new research area of tailoring nanomagnetism on the representative semiconductor surface by using adsorbates. 2 II. COMPUTATIONAL METHODS The present DFT calculations were performed using the Vienna ab initio simulation package with the projector- augmented wave (PAW) method23 -- 25. For the exchange- correlation energy, we employed the generalized-gradient ap- proximation functional of Perdew-Burke-Ernzerhof (PBE)26. The Si(111)-(7×7) surface (with the optimized lattice con- stant 5.431 A) was modeled by a periodic slab geometry con- sisting of the seven-layer slab and ∼15 A of vacuum in be- tween the slabs. Here, each Si atom in the bottom layer was passivated by one H atom. A plane-wave basis was employed with a kinetic energy cutoff of 300 eV, and the k-space in- tegration was done with 2×2 uniform meshes in the surface Brillouin zone. All atoms except the bottom two Si layers were allowed to relax along the calculated forces until all the residual force components were less than 0.01 eV/ A. III. RESULTS AND DISCUSSION We begin by optimizing the atomic structure of a clean Si(111)-(7×7) surface to examine its electronic band struc- ture. Figure 1(a) shows the optimized structure of the dimer- adatom-stacking fault (DAS) model of Si(111)-(7×7)11. The DAS model features the presence of 12 adatoms, 6 rest atoms, and 1 corner-hole atom, each of which offers one DB elec- tron. The calculated band structure of Si(111)-(7×7) is given in Fig. 1(b). We find that the surface states originating from 19 DB electrons of the adatoms, rest atoms and corner-hole atom are located at −0.18∼0.32 eV, −0.68∼−0.55 eV, and −0.48 eV around EF, respectively, in good agreement with previous DFT calculations and STM measurements27,28. Therefore, the 7 surface bands arising from the rest atoms and core-hole atom are completely filled by 14 DB electrons, while the remain- ing 5 DB electrons occupy two and half surface bands arising from the adatoms. This occupation of surface bands indicates a charge transfer from the adatoms to the rest and corner-hole atoms. In Fig. 1(a), there are two symmetrically different types of adatoms, i.e., coner adatoms (A1, A3, A5) and center adatoms (A2, A4, A6) in the faulted (F) half cell. The LDOS obtained at the A1 and A2 adatoms exhibit similar patterns [see Fig. 1(c)], each of which is also nearly identical to that at the counterpart adatom in the unfaulted (UF) half cell (see Fig. S1 of the Supplemental Material). Therefore, each adatom is likely to have ∼5/12 DB electrons. It is noticeable that the peaks of the LDOS of the adatoms A1 and A2 are located just above EF, whereas those of the rest atoms (R1, R2, R3) are identically located below EF [see Fig. 1(c)]. By employing H adsorption on the Si(111)-(7×7) surface, the LDOS peak positions of adatoms or rest atoms can be shifted toward EF, as discussed below. The resulting increased LDOS at EF may in turn give rise to the local spin polarization derived from the Stoner instability29. In this sense, tuning the LDOS at EF is very challenging for the realization of the spin orderings of DBs around adsorbed H atoms on the Si(111)-(7×7) surface. To explore how the LDOS changes by H adsorption on the Si(111)-(7×7) surface, we first consider the adsorption of a single H atom on the rest atom R1 within the F half cell, which was well known as the most stable adsorption site13 -- 15. The optimized structure using the spin-unpolarized DFT calcula- tion shows that the height of R1 (A1, A2, and A6) is lowered (raised), relative to that (those) of the clean Si(111)-(7×7) sur- face, by 0.41 A (0.10, 0.09, and 0.09 A). Such a Jahn-Teller- like distortion30 is accompanied by a charge transfer from R1 to the three neighboring adatoms. This charge transfer can be confirmed by examining the LDOS of each atom. Unlike the case of the clean Si(111)-(7×7) surface [see Fig. 1(c)], the LDOS peak of R1 disappears upon the H adsorption [see Fig. 2(a)], but its LDOS sum below EF keeps invariant. Fur- ther, the LDOS peaks of A1 and A2 (A6) become closer to EF [see Fig. 2(a)], whereas those of other adatoms remain nearly unchanged compared to the clean Si(111)-(7×7) sur- face [see Fig. S2 of the Supplemental Material]. These results indicate that the H adsorption on the R1 atom induces a lo- cal charge redistribution mostly up to the nearest neighboring adatoms. Noting that the sum of the LDOS of A1 or A2 (A6) increases because of their shifts to higher binding energy, we can say that one excess electron obtained by H adsorption is transferred from the R1 atom to its neighboring adatoms A1, A2, and A6. Figure 2(b) shows the schematic diagram for the interaction of an adsorbed H atom with the rest atom. Due to the fully occupied DB electrons at the rest atom, forma- tion of the Si−H bond leads one excess electron to occupy the antibonding state. However, since such an antibonding state is higher in energy than the surface state located at EF, it is natural that a local charge transfer from the R1 atom to its surrounding adatoms A1, A2, and A6 takes place through the above-mentioned Jahn-Teller-like distortion, lowering the electronic energy of the H/Si(111) system. Due to the increased LDOS of A1 and A2 (A6) at EF [see Fig. 2(a)], the H adsorption on the R1 atom may be expected to induce the local spin polarization around the adsorbed H atom. Indeed, our spin-polarized DFT calculation for the ad- sorption of a single H atom on the R1 atom obtains the spin po- larization of the surrounding A1, A2, and A6 adatoms. Among several spin-polarized configurations (see Fig. S3 of the Sup- plemental Material), the most stable configuration is a ferri- magnetic spin ordering of A1, A2, and A6 as shown in Fig. 2(c), where the spin directions of the two center adatoms A2 and A6 are the same with each other, but they are opposite to that of the corner adatom A1. This ferrimangetic configu- ration is energetically favored over the spin-unpolarized one by 11 meV per 7×7 unit cell. By integrating the spin density inside the PAW sphere with a radius of 1.574 A for Si, we obtain an identical spin moment of m = 0.22 µB for A2 and A6, while m = −0.15 µB for A1. The relatively larger magni- tude of spin moment for A2 and A6 may be associated with their higher LDOS at EF [see Fig. 2(a)], which gives rise to a larger spin splitting. Interestingly, as shown in Fig. 2(c), the spin density is largely delocalized up to the fifth deeper atomic layers. Note that (i) the total spin moment mt obtained (a) R1 A1 A2 ) V e / s e t a t s ( S O D 0.3 0.2 0.1 0 -0.1 -0.2 -0.3 0.3 0.2 0.1 0 -0.1 -0.2 -0.3 -2 -1 0 1 2 Energy (eV) (c) -0.15 0.22 0.22 A R ) V e / s e t a t s ( S O D (b) A 3 H (d) E ex A1 A2 0.3 0.2 0.1 0 -0.1 -0.2 -0.3 -2 -1 0 1 2 Energy (eV) FIG. 2: (Color online) (a) calculated LDOS of the R1, A1, and A2 atoms for the adsorption of a single H atom on the rest atom R1, obtained using the spin-unpolarized DFT calculation. In (b), the schematic diagram for the interaction of an adsorbed H atom with the rest atom is displayed. The filled (open) circles represent the full (partial) occupation of spin-up or -down electron, while the arrow indicates charge transfer. The top view of spin density is displayed in (c), together with the side view taken in the cross section along the dashed line. The majority (minority) spin density is displayed in bright (dark) color with an isosurface of 0.014 (−0.014) e/ A3. The numbers in (c) represent the spin moments (in µB) for the A1, A2, and A6 atoms. The spin-polarized LDOS of A1 and A2 is given in (d). Eex represents the exchange splitting of majority and minority bands. (b) Asur H A R (a) R1 ) V e / s e a t t s ( S O D 0.6 0.3 0 -0.3 -0.6 0.3 0.2 0.1 0 -0.1 -0.2 -0.3 A1 A2 -2 -1 0 1 2 Energy (eV) FIG. 3: (Color online) (a) Calculated LDOS of the R1, A1, and A2 atoms for the adsorption of a single H atom on the adatom A2, obtained using the spin-unpolarized DFT calculation. In (b), the schematic diagram for the interaction of an adsorbed H atom with the adatom is displayed. The filled (open) circles represent the full (partial) occupation of spin-up or -down electron, while the arrow indicates charge transfer. Asur represents the surrounding adatoms around A2. including the interstitial region outside the PAW sphere is 0.66 µB per 7×7 unit cell and (ii) the sum of the spin moments of subsurface Si atoms is slightly larger than that of the A1, A2, and A6 atoms. This large spin delocalization enables the spin coupling of the three neighboring adatoms (far-separated by ∼7.7 A) through the subsurface layers. In this sense, it is most likely that the H adsorption on R1 induces the local spin po- larizations at the nearest neighboring adatoms due to the intra- atomic exchange of localized DB electrons, and such spin mo- ments are in turn coupled with each other via the interatomic exchange interaction. It is noteworthy that the local spin po- larizations at A1, A2, and A6 can be derived from Stoner's criterion D(EF)I > 129, where D(EF) is the LDOS at EF and the Stoner parameter I can be estimated dividing the exchange splitting Eex by the corresponding magnetic moment [see Fig. 2(d)]31,32. Here, the calculated values of D(EF) and I are 0.203 and 6.2 (0.252 and 5.5) at A1 (A2), respectively, thereby satisfying D(EF)I > 1. Interestingly, the present ferrimagnetic order of adatoms has two types of spin-spin interactions: i.e., the ferromagnetic (FM) coupling between A2 and A6, while the antiferromagnetic (AFM) coupling between A1 and A2 (or A6). The FM coupling reduces the potential energy of repul- sive electron-electron interactions due to the Pauli exclusion principle, whereas the AFM coupling lowers the kinetic en- ergy of electrons. Therefore, the ferrimagnetic order of A1, A2, and A6 is likely to involve the synergetic effects of the potential and kinetic energies in the three localized spins at the triangular sites. Although the most stable H-adsorption site on Si(111)- (7×7) is the rest-atom site, STM experiments identified the adatom site as a precursor or metastable intermediate state13,15. Using the spin-unpolarized DFT calculation, we op- timize the structure of a single H atom adsorbed on the adatom site A2, which is almost energetically degenerate (less than ∼3 meV per 7×7 unit cell) with the adatom site A1. We find that the H adsorption on the A2 site is less stable than on the R1 site by 0.187 eV per 7×7 unit cell, in good agreement with the value (∼0.2 eV) of a previous DFT calculation14. Fur- ther, based on the nudged elastic-band method33, we calcu- late the energy profile along the H-diffusion pathway on go- ing from the A1 to the R1 site, and obtain an energy barrier of 1.38 eV, consistent with that (1.3 eV) of a previous DFT calculation14. As shown in Fig. 3(a), when H adsorbs on the A2 site, the LDOS peak of A2 disappears around EF. Mean- while, the LDOS peak of A1, located just above EF, slightly shifts toward lower binding energy, implying a charge transfer from A1 to A2. Note that the LDOS obtained at the rest atom R1 remains intact compared to the case of a clean Si(111)- (7×7) surface [see Figs. 3(a) and 1(c)]. Figure 3(b) shows the schematic diagram for the interaction of an adsorbed H atom with the A2 atom. In order to make an Si−H bond formation, the A2 atom needs 7/12 more electrons because it is partially occupied by 5/12 DB electrons before H adsorption. There- fore, unlike the above-discussed H adsorption on the R1 site, H adsorption on the A2 site needs to abstract electrons from surrounding adatoms. This charge abstraction decreases the LDOS of surrounding adatoms below EF [see the LDOS of A1 in Fig. 3(a)]. It is thus unlikely to show spin polarization ac- cording to Stoner's criterion. Indeed, our spin-polarized DFT calculation for the H adsorption on the adatom site A1 or A2 does not show any spin polarization. It is interesting to note that, as the number of H atoms ad- sorbed on the adatom sites increases, the DB electrons ab- stracted from unreacted surrounding adatoms become insuf- ficient to form the Si−H bonds. Hence, some neighboring rest atoms begin to participate in charge donation to the H- 4 adsorbed adatoms, leading to the shift of the LDOS peaks of the rest atoms toward EF. This de-charge transfer may in- duce spin polarization at the rest atoms. Our spin-unpolarized DFT calculation shows that, when the number (denoted as nA) of adsorbed H atoms on the adatom sites increases to 5, the LDOS peak of R1 is located near EF [see Fig. 4(a)], thereby possibly inducing spin polarization due to the Stoner-type in- stability. Indeed, our spin-polarized DFT calculation demon- strates that the case of nA = 5 exhibits a spin splitting in the LDOS of R1 [see Fig. 4(b)], resulting in the spin polarization at R1 [see Fig. 4(c)]. Figure 4(c) also displays the energeti- cally most stable spin configurations for nA = 6 (saturating the adatom sites in the F half cell) and 12 (saturating the adatoms sites in the F and UF half cells), respectively. Here, for nA = 6 we obtain the AFM order of two rest atoms R1 and R3 with m = ±0.57 µB, while for nA = 12 the ferrimagnetic order of six rest atoms in the F and UF unit cells [see each atomic spin moment in Fig. 4(c)]. It is thus likely that the number of adsorbed H atoms on the adatom sites can vary the pat- tern of spin polarization at the rest-atom sites. Similarly, we find that, when the numbers (denoted as nR) of adsorbed H atoms on the rest-atom sites are 2, 3, and 6, the ferrimagnetic orders of adatoms become most stabilized with different spin moments, as shown in Fig. 4(d). (a) (b) ) V e / s e t a t s ( S O D ) V e / s e t a t s ( S O D R1 0.6 0.3 0 -0.3 -0.6 -2 -1 0 1 2 Energy (eV) (c) nA = 5 0.57 nA = 6 -0.57 0.57 nA = 12 0.57 -0.57 -0.57 0.57 -0.27 0.57 -0.57 R1 0.6 0.3 0 -0.3 -0.6 -2 -1 0 1 2 Energy (eV) (d) = 2 nR -0.18 -0.18 0.21 0.21 0.27 -0.18 -0.02 0.03 = 3 nR -0.24 0.03 -0.24 0.31 0.31 -0.03 0.31 -0.03 0.03 -0.03 -0.24 0.03 = 6 nR -0.27 0.25 0.34 0.34 -0.31 0.34 -0.31 0.25 -0.31 -0.27 0.25 FIG. 4: (Color online) (a) Spin-unpolarized and (b) spin-polarized LDOS of the R1 atom for nA = 5. Spin densities for nA = 5, nA = 6, and nA = 12 are displayed in (c), while those for nR = 2, nR = 3, and nR = 6 in (d). Here, the majority (minority) spin density is drawn with an isosurface of 0.014 (−0.014) e/ A3. In (c) and (d), the numbers represent the spin moments (in µB) for rest atoms and adatoms, respectively. 5 IV. SUMMARY Based on first-principles DFT calculations, we presented the spin orderings of Si DBs at the atomic scale, induced by H adsorption on the Si(111)-(7×7) surface. We revealed that the adsorption of a single H atom on a rest-atom site causes a charge transfer back to the nearest neighboring adatoms, therefore giving rise to an increase in the LDOS of such three adatoms. The resulting high DOS at EF is attributed to the Stoner-type instability to yield a ferrimagnetic order of adatom DBs around the adsorbed H atom. Meanwhile, when an H atom adsorbs on an adatom site, a charge abstraction from its surrounding adatoms takes place, forming an H-Si bond without spin polarization. However, as the number of adsorbed H atoms on the adatom sites within the 7×7 unit cell increases, some neighboring rest atoms begin to partici- pate in charge donation to the adatoms, therefore shifting the LDOS peak of such rest atoms toward EF to induce a fer- rimagnetic order. In the present study, we propose the un- derlying mechanism of the H-induced spin orderings of DBs on the Si(111)-(7×7) surface in terms of the intra-atomic and interatomic exchanges of localized DB electrons through the subsurface layers. We anticipate that the present way to the design of nanoscale magnetism on the Si(111)-(7×7) surface is rather generic and hence, it should be more broadly appli- cable to tailor the spin orderings of Si DBs by using other adsorbates. Acknowledgement. We are grateful for discussions with Y. Jia. This work is supported by the NSFC of China (Grant Nos. 11804306, 11674289, and 11504332) and National Research Foundation of Korea (NRF) grant funded by the Korean Government (Grant Nos. 2015M3D1A1070639 and 2016K1A4A3914691). The calculations were performed at the High Performance Computational Center of Zhengzhou University. J.H.C acknowledges the strategic support pro- gram (KSC-2017-C3-0080) for supercomputing application research from the KISTI supercomputing center. X.-Y. R. and C.-Y. N. contributed equally to this work. ∗ Corresponding author: [email protected] 1 S. Folsch, J. Martinez-Blanco, J. Yang, K. Kanisawa, and S. C. Erwin, Nat. Nano, 9, 505-508 (2014). 2 S. Bohoul, Q. Shi, R. A. Wolkow, and H. Guo, Nano Lett., 17, 322-327 (2017). 3 M. B. Haider, J. L. Pitters, G. A. DiLabio, L. Livadaru, J. Y. Mu- 18 A. Yurtsever, Y. Sugimoto, H. Tanaka, M. Abe, S. Morita, M. Ondr´acek, P. Pou, R. P´erez, and P. Jel´ınek, Phys. Rev. B 87, 155403 (2013). 19 M. Durr and U. Hofer, Prog. Surf. Sci. 88, 61-101 (2013). 20 T. Arai, R. Inamura, D. Kura, and M. Tomitori, Phys. Rev. B 97, tus, and R. A. Wolkow, Phys. Rev. Lett. 102, 046805 (2009). 115428 (2018). 4 M. Kepenekian, R. Robles, C. Joachim, and N. Lorente, Nano 21 S. Okada, K. Shiraishi, and A. Oshiyama, Phys. Rev. Lett. 90, Lett. 13, 1192-1195 (2013). 5 O. V. Yazyev, Rep. Prog. Phys. 73, 056501 (2010). 6 M. Yengui, E. Duverger, P. Sonnet, and D. Riedel, Nat. Commun. 8, 2211 (2017). 7 B. E. Kane, Nature, 393, 133-137 (1998). 8 S. R. Schofield, P. Studer, C. F. Hirjibehedin, N. J. Curson, G. Aeppli, and D. R. Bowler, Nat. Commun. 4, 1649 (2013). 9 H. Gonz´alez-Herrero, J. G´omez-Rodriguez, P. Mallet, M. Moaied, J. J. Palacios, C. Salgado, M. M. Ugeda, J.-Y. Veuillen, F. Yndurain, and I. Brihuega, Science 352, 437-441 (2016). 10 O. V. Yazyev and L. Helm, Phys. Rev. B 75, 125408 (2007). 11 K. Takayanagi, Y. Tanishiro, S. Takahashi, and M. Takahashi, Surf. Sci. 164, 367-392 (1985). 12 J. E. Northrup, Phys. Rev. Lett. 57, 154 (1986). 13 I. C. Razado, H. M. Zhang, R. I. G. Uhrberg, G. V. and Hansson, Phys. Rev. B 71, 235411 (2005). 14 A. Vittadini and A. Selloni, Phys. Rev. Lett. 75, 4756 (1995). 15 R.-L. Lo, I.-S. Hwang, M.-S. Ho, and T. T. Tsong, Phys. Rev. 026803 (2003). 22 S. C. Erwin, and F. J. Himpsel, Nat. Comm. 1, 58 (2010). 23 G. Kresse and J. Hafner, Phys. Rev. B 48, 13115 (1993). 24 G. Kresse and J. Furthmuller, Comput. Mater. Sci. 6, 15 (1996). 25 P. E. Blochl, Phys. Rev. B 50, 17953 (1994). 26 J. P. Perdew, K. Burke, and M. Ernzerhof, Phys. Rev. Lett. 77, 3865 (1996); 78, 1396 (1997). 27 G. Lee, C. G. Hwang, N. D. Kim, J. Chung, J. S. Kim, and S. Lee, Phys. Rev. B 76, 245409 (2007). 28 Hamers, R. J.; Tromp, R. M.; Demuth, J. E. Phys. Rev. Lett. 1986, 56,1972. 29 E. C. Stoner, Proc. R. Soc. A 165, 372 (1938). 30 T. Hitosugi, S. Heike, T. Onogi, T. Hashizume, S. Watanabe, Z. Q. Li, K. Ohno, Y. Kawazoe, T. Hasegawa, and K. Kitazawa, Phys. Rev. Lett. 82, 4034 (1999). 31 J. D. Burton and E. Y. Tsymbal, Phys. Rev. Lett. 107, 166601 (2011). 32 D. M. Edwards and M. I. Katsnelson, J. Phys.: Condens. Matter Lett. 80, 5584 (1998). 18, 7209 (2006). 16 T. Fukuda, J. Nakatani, and H. Nakayama, Surf. Sci. 600, 2443- 33 G. Henkelman, B. P. Uberuaga, and H. J´onsson, J. Chem. Phys. 2448 (2006). 17 K. Mortensen, D. M. Chen, P. J. Bedrossian, J. A. Golovchenko, and F. Besenbacher, Phys. Rev. B 43, 1816 (1991). 113, 9901 (2000).
1012.3128	2	1012	2011-04-01T12:41:46	Nonlocal correlations of the local density of states in disordered quantum Hall systems	[ "cond-mat.mes-hall" ]	Motivated by recent high-resolution scanning tunneling microscopy (STM) experiments in the quantum Hall regime both on massive two-dimensional electron gas and on graphene, we consider theoretically the disorder averaged nonlocal correlations of the local density of states (LDoS) for electrons moving in a smooth disordered potential in the presence of a high magnetic field. The intersection of two quantum cyclotron rings around the two different positions of the STM tip, correlated by the local disorder, provides peaks in the spatial dispersion of the LDoS-LDoS correlations when the intertip distance matches the sum of the two quantum Larmor radii. The energy dependence displays also complex behavior: for the local LDoS-LDoS average (i.e., at coinciding tip positions), sharp positive correlations are obtained for tip voltages near Landau level, and weak anticorrelations otherwise.	cond-mat.mes-hall	cond-mat	Nonlocal correlations of the local density of states in disordered quantum Hall systems Thierry Champel Laboratoire de Physique et Mod´elisation des Milieux Condens´es, CNRS and Universit´e Joseph Fourier-Grenoble 1, B.P. 166, 25 Avenue des Martyrs, 38042 Grenoble Cedex 9, France Institut N´eel, CNRS and Universit´e Joseph Fourier, B.P. 166, 25 Avenue des Martyrs, 38042 Grenoble Cedex 9, France Serge Florens Department of Physics, University of Utah, Salt Lake City, Utah 84112, USA (Dated: November 15, 2018) M. E. Raikh Motivated by recent high-resolution scanning tunneling microscopy (STM) experiments in the quantum Hall regime both on massive two-dimensional electron gas and on graphene, we consider theoretically the disorder averaged nonlocal correlations of the local density of states (LDoS) for electrons moving in a smooth disordered potential in the presence of a high magnetic field. The intersection of two quantum cyclotron rings around the two different positions of the STM tip, correlated by the local disorder, provides peaks in the spatial dispersion of the LDoS-LDoS corre- lations when the intertip distance matches the sum of the two quantum Larmor radii. The energy dependence displays also complex behavior: for the local LDoS-LDoS average (i.e. at coinciding tip positions), sharp positive correlations are obtained for tip voltages near Landau level, and weak anticorrelations otherwise. I. INTRODUCTION Quantum Hall systems offer a surprising dichotomy be- tween very universal macroscopic properties, such as the near perfect quantization of the Hall conductance, and sample-dependent physics dominated by local imperfec- tions, as recently observed in several local scanning tun- neling spectroscopy (STS) experiments both on massive two-dimensional electron gas1 and on graphene2,3. It is, however, known that some degree of universality can be recovered by performing sample (or disorder) average of local quantities, and indeed theoretical predictions for the averaged STS local density of states (LDoS) lead to Gaussian behavior near the Landau levels, with an en- ergy width and a lineshape that depend on the width and correlation length of the disorder distribution, re- spectively4,5. Because the information extracted from transport experiments is limited, correlations of current (i.e., noise measurements) have been previously exam- ined6, successfully demonstrating the existence of frac- tionally charged quasiparticles for the fractional quan- tum Hall effect. The question we wish to raise here is the nature of the disorder averaged nonlocal correlations of local physical quantities (such as the LDoS) in the quantum Hall regime. Such study can, in principle, be experimentally achieved by sampling large spatial areas of the sample surface using the displacements of the STM tip and correlating the measured LDoS at two different tip positions (and possibly two different tip voltages). The possibility to probe the LDoS at different spatial lo- cations in STM experiments offers new perspectives in comparison with previous experimental studies of fluc- tuations of the LDoS at a fixed position, using resonant tunneling through a localized impurity state7 -- 10. More explicitly, from the LDoS ρ(r, ω), which depends on tip position r and voltage ω (although we keep the electron charge e = −e and Planck's constant in what follows, we assume that voltage, energy and frequency are loosely identified with each other), we define the nonlocal disorder averaged LDoS-LDoS correlations χ(r, ω1, ω2) ≡ hρ(r1, ω1)ρ(r2, ω2)i − hρ(r1, ω1)i hρ(r2, ω2)i (1) as the centered two-point correlation function of the LDoS (here r = r1 − r2). Clearly this is a complicated object that depends on two tip voltages, but only on the distance between the two positions of the STM tip be- cause of translation invariance and spatial isotropy after averaging. Before turning to detailed calculations, we wish to give some general physical interpretation of this physical quantity. The basic idea is that important correlations are obtained whenever the two quantum cyclotron rings (associated to circular wave functions in a perpendicular magnetic field B) have a large spatial overlap (disorder plays, however, a crucial role in correlating locally the states, as we will see later on). Focusing first on the spa- tial dependence of χ(r, ω1, ω2) for equal tip voltages, one readily understands that the intersection of the quan- tum rings (with width of the order of magnetic length lB = pc/eB) around the two tip positions provides an increased area of intersection [from l2 B (RL)1/2] when the distance between the points is close to the sum of the Larmor radii RL (see Fig 1 and caption for de- tails). This already suggests that a peak should occur in the LDoS-LDoS correlations for this particular distance. B to l3/2 r1 r2 a) c) b) d) r1 r2 model is a single-particle Hamiltonian for an electron confined in two dimensions in the presence of both a per- pendicular magnetic field B and an arbitrary potential energy V (r), 2 lB RL H = 1 2m∗ (cid:16)−i∇r − e c A(r)(cid:17)2 + V (r), (2) r1 r2 r1 r2 FIG. 1: Geometric interpretation of correlations of the LDoS in terms of overlap of quantum cyclotron rings of width lB. Panel a) taken for close intertip distance r1 − r2 ≪ 2RL presents a nearly maximal overlap area of the order lBRL (with RL the Larmor radius), hence maximal correlations. Panel b) taken in the intermediate situation r1 − r2 ≃ RL has a small overlap area l2 B, thus relatively weak correlations. Panel c) taken at tangent cyclotron orbits r1− r2 ≃ 2RL has a small increased overlap area of the order l3/2 B (RL)1/2 (for RL > lB), so that a moderate peak as a function of intertip distance occurs. Finally panel d) presents the situation of non-overlaping cyclotron orbits at r1 − r2 ≫ 2RL, hence exponentially suppressed LDoS-LDoS correlations. We note that this effect has no classical analog, as quan- tum cyclotron rings collapse into circular cyclotron orbits at vanishing lB (in fact, we will see that a kink instead of a peak occcurs in the classical limit). The energy depen- dence (given by the two tip voltages) can be also easily inferred, for instance, at coinciding tip positions. When both tip energies precisely match the Landau levels, max- imal overlap of the whole quantum cyclotron rings occur, leading to sharp and large positive correlations. However, detuning the two tip energies will probe correlations be- tween different circular wave functions, and destroy the correlations. In that case, the square averaged term in Eq. (1) will dominate the averaged square one, leading then to weak and negative correlations. The paper is organized as follows. Section II recalls the form of the local density of states (for fixed but smooth disorder) in quantum Hall systems11. We consider, in turn, the disorder averaged LDoS in Sec. III, allowing us to present the technique on a simple and more usual quantity, and to test the approximation scheme. Then Sec. IV presents the derivation of the disorder averaged LDoS-LDoS correlations, which are finally discussed in great detail in Sec. V. II. LOCAL DENSITY OF STATES AT HIGH MAGNETIC FIELD We now present our theoretical analysis of this prob- lem, which can be carried out fully analytically for smooth disorder in the high magnetic field regime, and which confirms the above argumentation. The basic with the vector potential A such that ∇× A = B = Bz, and m∗ the electron effective mass [here r = (x, y) is the position of the electron in the plane]. We do not consider the case of graphene here, which can be easily extended following our previous results5, although this system is quite relevant experimentally for the considerations of the present work. The starting point is the realistic assumption of large cyclotron frequency (compared to local amplitude fluctu- ations of the disordered potential) at large magnetic field, so that Landau level mixing can be disregarded and the guiding center coordinate R follows a quantum motion along weakly curved equipotential lines. In that case the guiding center Green's function obeys a single pole struc- ture for each Landau level n as found in Refs. 11 and 5 [this generalizes the results of Ref. 12 to arbitrary Lan- dau levels]: Gn(R) = 1 ω − En − Vn(R) + i0+ (3) where En = ωc(n + 1/2) are the Landau level energies and ωc = eB/(m∗c) is the cyclotron frequency. One important quantity above is the effective potential Vn(R) that results from averaging the bare disorder potential V (R) along the quantum cyclotron motion Vn(R) =Z d2 η Fn(R − η)V (η). (4) The kernel Fn(R) here is given by the following expression11: with As = (1 − s)/(1 + s), and can also be written in an equivalent form13 Fn(R) = 1 πn!l2 B ∂n ∂sn Fn(R) = (−1)n πl2 B Ln(cid:18) 2R2 l2 e−AsR2/l2 B 1 + s (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)s=0 B (cid:19) e−R2/l2 B , (5) (6) where Ln(z) is the Laguerre polynomial of degree n. This kernel is also known as the form factor in the literature on quantum Hall effects14. Expression (5) turns out to be very useful for the study of the first several Landau levels, while expression (6) is more suited for the con- sideration of high Landau levels (n ≫ 1). Apart from the case n = 0, we note that the kernel Fn(R) is not a positive-definite function, and cannot be interpreted as a wave function probability density. Instead, it rather cor- responds to a Wigner distribution because the physical space of the guiding center coordinates R = (X, Y ) is, in fact, associated to a pair of conjugate variables owing to the commutation relation [ X, Y ] = il2 B in the operato- rial language. However, integration of the kernel Fn(R) over an arbitrary line provides a translationally invari- ant Landau states probability density. In the classical limit n → ∞, while keeping the (Larmor) cyclotron ra- n = √2nlB finite (hence for lB → 0), one gets15 dius RL Fn(R) ≃ 1 n ), so that the effective poten- tial Eq. (4) corresponds to an average over the classical cyclotron orbit, as previously shown in Ref. 16. For a nonzero lB, Fn(R) is an oscillating function that shows a sharp peak of width lB centered around R ≃ RL n . This quantity will be a crucial ingredient later on for the math- ematical identification of the quantum cyclotron rings. δ(R − RL 2πRL n Finally, the LDoS is readily connected5,11 to the guid- ing center Green's function given in Eq. (3) ρ(r, ω) =Z d2R 2πl2 B +∞ Xn=0 Fn(R − r)−1 π Im Gn(R, ω). (7) Here one did not include the overall spin degeneracy. In practice, the signal measured by local scanning tunnel- ing spectroscopy is directly related to the local density of states through an energy convolution with the deriva- tive of the Fermi-Dirac distribution and the experimental resolution window. Apart from the condition of high cy- clotron energy, the present calculation is valid for smooth potentials at the scale of lB. Such a restriction on the spa- tial variations of the potential is, in fact, not very drastic because the above results are exact for arbitrary one- dimensional (1D) potentials11 (even very rough ones). For the realistic two-dimensional (2D) potential the va- lidity of the calculation is controlled by the small energy scale associated typically to potential curvature11 Ecurvature = l2 B 2 p∂XX Vn∂Y Y Vn − (∂XY Vn)2. A more precise evaluation of the degree of reliability of our approximation scheme for rough potentials will be given at the end of the next section. Before considering the correlations of the LDoS, we compute now in some detail the disorder averaged LDoS itself. III. DISORDER AVERAGED LDOS AT HIGH MAGNETIC FIELD We first examine the disorder averaged LDoS, obtained experimentally by spatially sampling the STS current over a single STM tip position. Theoretically, the aver- aging procedure of expression (7) will be carried through an isotropic distribution function C(q) in Fourier space (here q = q) that describes the spatial correlations of disorder hV (R1)V (R2)i = C(R1 − R2) (9) = Z d2q (2π)2 C(q) e−iq·(R1−R2). 3 Typically one can take C(R) = v2e−R2/ξ2 , defining the correlation length ξ and the root mean square value v =phV (R)2i of the bare disorder distribution. In that case C(q) = πv2ξ2e−ξ2q2/4. Averaging of the LDoS (7) then simply follows from exponentiating the single pole in Eq. (3) by going in the time domain, and performing Gaussian integration over all possible disorder realiza- tions hρ(r, ω)i = Z dt 2π Z d2R 2πl2 B +∞ Xn=0 (10) Fn(R − r)ZDV 2Z d2q 1 × exp(i[ω − En − Vn(R)]t − C(q) ) . (2π)2 V (q)2 The effective potential Eq. (4) is given in Fourier space by Vn(q) = Fn(q) V (q), where Fn(q) is the Fourier transform of kernel (5), which is easily shown to obey Fn(q) = Z d2r eiq.rFn(r) = 1 n! ∂n ∂sn = (−1)nπl2 B Fn(l2 B q/2). e−l2 Bq2/(4As) (1 + s)As (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)s=0 (11) One can then readily perform the functional integral over the disorder realizations in Eq. (10): hρ(r, ω)i = Z dt +∞ 2πl2 B 2π Z d2R × exp(cid:26)− 1 4 Xn=0 ei(ω−En)tFn(R − r) n(cid:27) t2 Γ2 (12) where the energy width Γn is given by the relation (8) Γ2 n = 2Z d2q (2π)2 2 . (13) C(q)(cid:12)(cid:12)(cid:12) Fn(q)(cid:12)(cid:12)(cid:12) This result was obtained initially in Ref. 5 for the case of graphene. Expression (12) for the averaged density of states (DoS) is obviously r independent, so that the R inte- gral can be carried using the normalization condition the final result R d2R Fn(R) = 1. The remaining time integral gives Γn (cid:19)2) , (14) hρ(r, ω)i = exp(−(cid:18) ω − En 1 2πl2 B √πΓn +∞ 1 Xn=0 which takes the expected Gaussian lineshape. The renormalized disorder width Γn given in Eq. (13) can be analyzed15 in the classical limit n → +∞, keeping n = √2nlB fixed and lB → 0. In the cyclotron radius RL this regime, we recover results first derived in Ref.16 using a completely unrelated method Γn =sZ qdq 2π 2 C(q)[J0(RL n q)]2, (15) 1.5 1 0.5 i ) ω , r ( ρ h 0 0 ξ/lB = 5 ξ/lB = 1 1 2 ω/ωc 3 4 FIG. 2: (color online) Disorder averaged density of states [in units of (2πl2 Bv)−1] as a function of energy for two values of the correlation length of the disorder distribution (ξ/lB = 5 and ξ/lB = 1) spanning the first four Landau levels. with J0(z) the zeroth order Bessel function. For very large classical orbits such that RL n ≫ ξ, the large- argument asymptotics of the zeroth order Bessel function J0(z) ≃p2/(πz) cos(z − π/4) can be used, so that (16) Γn ≃sR dq 2 C(q) π2RL n 1 n1/4 , ∝ showing a decrease of the Landau level energy width with increasing index n. We note that expression (13) is more general than the classical result (15) because it incorporates wave function spreads on the scale lB, a purely quantum lengthscale which has completely disap- peared in the classical limit. In all cases (classical or quantum), the general trend is that the cyclotron mo- tion averages out the local potential at increasing radius RL n , so that the energy width of the sample averaged DoS decreases with n. This effect is clearly seen1 -- 3 from the experimentally measured spatial dispersion of the LDoS, which shows a rapid narrowing for higher Landau lev- els. However, in the opposite limit of very smooth dis- order ξ ≫ RL n , this averaging by the cyclotron orbits becomes less efficient, and the energy width Γn depends very weakly on the Landau level index n. Both regimes are presented in Fig. 2 showing the energy-dependent dis- order averaged density of states for two values of ξ/lB. We end up by commenting on the reliability of our cal- culations for realistic disorders, and possibly rough ones at the scale of lB. From the construction of the quan- tum guiding center theory11,12 as a systematic gradient expansion, the lowest order calculations used here be- come exact for smooth disorder, namely ξ ≫ lB. In the opposite limit ξ ≪ lB, the approximation scheme does not fully break down per se, thanks to the wave function averaging on the scale lB performed within the effective potential (4). This implies that the effective po- tential varies on the scale pξ2 + l2 B and remains smooth In this even for very rough bare potential (ξ ≪ lB). regime, the theory simply misses then the small param- 1 0.75 0.5 0.25 i ) ω , r ( ρ h 0 −4 4 Approximate Exact −2 0 2 4 (ω − E0)/ωc FIG. 3: (color online) Comparison of the disorder averaged density of states [in units of (2πl2 BΓ0)−1] in the Landau level n = 0 between Wegner's exact formula and our leading order gradient approximation for the extreme case of δ-correlated disorder (ξ/lB = 0). Rapid convergence of the approximation scheme is expected as soon as ξ & lB. B/(ξ2 + l2 eter l2 B) of the case ξ ≫ lB, and its adequation becomes purely a quantitative matter. Interestingly, we are able to assess its validity thanks to Wegner's exact solution4 for the disorder averaged density of states in the Landau level n = 0, calculated for δ-correlated disorder hV (R)V (0)i = w δ(2)(R) hρ(r, ω)iexact = 1 2πl2 B 23/2 π3/2Γ0 exp[2(ω − E0)2/Γ2 0] 1 + {Erfi[√2(ω − E0)/Γ0)]}2 (17) 0 du eu2 0 = w/(πl2 In the previous formula, Erfi(x) = 2π−1/2R x is the so-called complex error function and Γ2 B). In our calculation, the δ-correlated potential is obtained from the ξ → 0 limit in Eq. (9), which corresponds to the most stringent limit to test our approximation scheme. The comparison (derived for the same microscopic disor- der parameters) between the Gaussian expression (14) in the lowest Landau level n = 0, using the linewidth (13), with Wegner's exact formula4, written in Eq. (17), is given in Fig. 3. Although neither the peak value nor the tails (not shown) are exactly reproduced within our approximation, the result is close to the exact answer, and this is surprising because we are far from the naive domain of validity of the theory. This comparison gives some evidence that both the dis- order averaged LDoS and its nonlocal correlations (to be calculated now), will be quantitatively obtained by a gra- dient expansion in the disorder for realistic cases where ξ & lB (granted by the high magnetic field limit). More- over, we stress that Wegner's method does not apply to obtain two-particle averages, and cannot be used to cal- culate the LDoS-LDoS correlations, showing the greater generality of our approach. IV. DISORDER AVERAGED LDOS-LDOS CORRELATIONS AT HIGH MAGNETIC FIELD We are now ready to calculate the two-point correla- tions of the LDoS, taken at two different tip positions r1, r2 and for two different energies ω1, ω2. The starting point is the expression +∞ +∞ 2πZ d2R1 2πZ dt2 hρ(r1, ω1)ρ(r2, ω2)i =Z dt1 Xn1=0 Xn2=0 ×Fn1(R1 − r1)Fn2 (R2 − r2)ei(ω1−En1 )t1+i(ω2−En2 )t2 ×A(R1, n1, t1; R2, n2, t2) where the following disorder average must be performed: BZ d2R2 2πl2 B 2πl2 (18) 1 A(R1, n1, t2; R2, n2, t2) =ZDV e−i[Vn1 (R1)t1+Vn2 (R2)t2] C(q) ) (2π)2 V (q)2 2  × exp(− C(q)(cid:12)(cid:12)(cid:12) Xi=1,2 n2 + 2t1t2Sn1,n2 (R1 − R2)(cid:3)(cid:27) . 2Z d2q ti Fni (q)eiq.Ri(cid:12)(cid:12)(cid:12) = exp 2Z d2q −  = exp(cid:26)− 4(cid:2)t2 n1 + t2 (2π)2 1Γ2 2Γ2 1 1 The energy width Γn was already defined in Eq. (13), and a spatially dependent disorder correlator now appears Sn1,n2(R) = 2Z d2q (2π)2 C(q) Fn1 (q) Fn2 (q) cos(q · R). (19) Using expressions (5) and (11), we can perform exactly the integrals in Eq. (19) and find a result which can be written under the form Sn1,n2(R) = 2v2 0 n1!n2! ∂n1+n2 ∂sn1 1 ∂sn2 ξ2 ξ2 + 2Bs1,s2 l2 B × exp(cid:18)− 2 "(1 − s1 − s2 + s1s2)−1 B(cid:19)#s1=s2=0 ξ2 + 2Bs1,s2 l2 R2 ,(20) where we have introduced the short-hand notation Bs1,s2 = 1 − s1s2 1 − s1 − s2 + s1s2 . (21) Note that the disorder correlator Sn1,n2(R) is isotropic (i.e., only depends on the distance R = R). In addition, its diagonal elements taken at R = 0 are related to the energy width Γn via the relation Sn,n(0) = Γ2 n. For a smooth disordered potential such that ξ ≫ lB, we can easily check from Eq. (20) that the functions Sn1,n2(R) are peaked at R = 0 and decay in a Gaussian way with a characteristic length scale given by ξ for the first few Landau levels. For higher Landau levels, the disorder correlator Sn1,n2(R) spreads over a bigger characteristic 5 n )2 + ξ2. distance. For instance, for n1 = n2 = n ≫ 1, it has an extent of the order of p(RL Shifting the space integrals in Eq. (18), we see that the quantity hρ(r1, ω1)ρ(r2, ω2)i is clearly described by a function of r = r1 − r2 only. In the following we will consider the centered two-point correlator of the LDoS defined in Eq. (1). Integration over the time variables in Eq. (18) can be performed analytically and yields the expression for the LDoS-LDoS correlator χ(r, ω1, ω2) =Z d2R1 BZ d2R2 2πl2 B 2πl2 +∞ +∞ Xn1=0 Xn2=0 Fn1 (R1 − r1) ×Fn2(R2 − r2) Θn1,n2 (R1 − R2 , ω1, ω2) (22) with Θn1,n2 (R, ω1, ω2) = 1 π n(cid:2)Γ2 n1Γ2 n2 − [Sn1,n2(R)]2(cid:3)−1/2 −2(ω1−En1 )(ω2 −En2 )Sn1,n2 (R) (23) o . (ω1−En1 )2 Γ2 n2 − ×e Γ2 +(ω2−En2 )2 Γ2 n1 −[Sn1,n2 (R)]2 n2 (ω1 −En1 )2 n1 Γ2 − − [Γn1Γn2 ]−1 e Γ2 n1 (ω2−En2 )2 Γ2 n2 − e Although the above expressions are still too complicated to make precise statements on the nature of the LDoS- LDoS correlations, we can already infer some of the early predictions formulated in the Introduction. Clearly the two kernels Fn(R − r) in Eq. (22) centered on the two po- sitions of the tip r1 and r2 impose a constraint on the two guiding center coordinates R1 and R2, which must live predominantly within cyclotron rings of extent RL n1 and RL n2 with width lB (we note that the kernels have rather n oscillations within a given radius RL n , so that there are in total n different rings for a given quantum state). This formula thus confirms our initial expectation that the area of overlap between two cyclotron rings dictates the behavior of the LDoS-LDoS correlations. A second interesting aspect is that it is the additional disorder- dependent kernel Θn1,n2 which permits nontrivial corre- lations. Should this complicated function of energy and space be negligible, one would end up with vanishing cor- relations. Yet, to get more quantitative understanding of the complete integral in Eq. (22), one needs to further sim- plify the above expressions. This can be achieved by in- troducing the change in variables R′ = (R1 + R2)/2 and R = R2 − R1. After integration over the center-of-mass position R′, we get χ(r, ω1, ω2) = +∞ Xn1=0 with fn1,n2(R) =Z d2R′ 2πl2 B = (2πl2 B)−2 n1!n2! ∂n1+n2 ∂sn1 1 ∂sn2 2 (24) +∞ Xn2=0Z d2R fn1,n2(R + r) R 2πl2 B ×Θn1,n2 (R, ω1, ω2) 2 (cid:19) Fn2(cid:18)R′ + 2 (cid:19) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)s1=s2=0 BBs1,s2(cid:3)(cid:1) Fn1(cid:18)R′ − exp(cid:0)−R2/(cid:2)2l2 1 − s1s2 R ,(25) where we have used expression (5) for the kernel Fn(R). Then, introducing the polar coordinates for the position R and performing the angular integral in Eq. (24), we ar- rive at the final expression of the disorder averaged LDoS- LDoS correlations (which now obviously is only function of the intertip distance r = r): χ(r, ω1, ω2) = +∞ +∞ Xn1=0 Xn2=0 where l−2 RdR hn1,n2(R, r) B Z +∞ ×Θn1,n2 (R, ω1, ω2) , 0 (26) hn1,n2 (R, r) =Z 2π 0 dθ 2π B)−2 n1!n2! (2πl2 = fn1,n2(cid:16)pR2 + r2 + 2rR cos θ(cid:17) ∂n1+n2 ∂sn1 1 ∂sn2 l2 2 "I0(cid:18) rR BBs1,s2(cid:3)(cid:1) BBs1,s2(cid:19) #s1=s2=0 × exp(cid:0)−(cid:2)R2 + r2(cid:3) /(cid:2)2l2 1 − s1s2 6 Eq. (23). Therefore, the main contributions to the cor- relations arise when Sn1,n2(R) ∼ Γn1Γn2 . In contrast, the functions hn1,n2 (R, r) are indepen- dent of the characteristic features of the disorder and have a pure geometric origin (as already discussed, they contain information on the spatial overlap of the quan- tum cyclotron rings). For r = 0, we have the relation hn1,n2(R, 0) = fn1,n2(R) from which a simple physical interpretation can be easily drawn. Using the semiclas- sical physical picture of the kernel Fn(R) put forward previously, we understand that nonzero contributions for fn1,n2(R) in Eq. (25) are picked up only from the (quan- tum broadened on scale lB) cyclotron orbits with radii RL n2 that live around the points ±R/2 sepa- rated by the distance R. This statement can be proved on more mathematical grounds starting from expression (28) (28). Taking the limits n1 and n2 → ∞ in Eq. as done in Ref. 15, we obtain an approximate formula, which reads n1 and RL (27) hn1,n2(R, r) ≃ l2 B (2πl2 B)2 Z +∞ 0 dq qJ0(RL n1 q) J0(RL n2 q) with I0(x) the modified Bessel function of the first kind. An alternative writing of Eq. (27) obtained by going to the Fourier space, which turns out to be more suitable for the consideration of large n1 and n2, is as follows: hn1,n2 (R, r) = dq q Fn1 (q) Fn2 (q)J0(Rq) l2 B (2πl2 B)2 Z +∞ 0 ×J0(rq). (28) The above expressions (26) through (28) are exact for arbitrary 2D smooth disorder (i.e., ξ ≫ lB). V. DISCUSSION OF THE LDOS-LDOS CORRELATIONS A. Spatial dependence of the correlations Let us now analyze the LDoS-LDoS correlations on the basis of Eq. (26). We see that the correlations result from the combination of two functions hn1,n2(R, r) and Θn1,n2 (R, ω1, ω2), which are quite different in nature. Obviously, only the functions Θn1,n2 (R, ω1, ω2) defined in Eq. (23) contain the information about the energy de- pendence. For sufficiently widely separated Landau lev- els, the overlap between two states with two given ener- gies ω1 and ω2 vanishes for most of the Landau level pairs due to the sharp energy Gaussians cutoff in Eq. (23), so that essentially only the one specific pair of Landau lev- els (n1, n2) that is associated to the cyclotron energies (En1 , En2 ) closest to the tip voltages (ω1, ω2) yields a nonzero contribution in the sum over Landau levels in- dices in Eq. (26). Furthermore, the LDoS-LDoS corre- lator χ(r, ω1, ω2) clearly vanishes whenever the disorder correlator Sn1,n2(R) is small compared to Γn1 Γn2 due to the exact cancellation by the square averaged term in n1 + RL n1 + RL n2(cid:12)(cid:12). n1 − RL precisely as anticipated in the discussion of Fig. 1. n2(cid:12)(cid:12) < r < RL n2 or R < (cid:12)(cid:12)RL n1 − RL ×J0(Rq) J0(rq). (29) integral (29) strictly vanishes when R > For r = 0, RL In the more real- istic case of finite n1 and n2, this semiclassical limit shows that the resulting overlap of two quantum orbital motions turns out to be significant under the inequal- n2 . The functions hn1,n2(R, r) being symmetrical in R and r, we understand that for R = 0 the contributions to the LDoS-LDoS cor- relations arise when the distance r between the two tip n1 + RL n2 , n2(cid:12)(cid:12) < R < RL ities (cid:12)(cid:12)RL positions is such that (cid:12)(cid:12)RL n1 − RL The spatial dependence of the two-point correlator χ(r, ω1, ω2) resulting from the numerical computation of the integral over the distance R in Eq. (26) is shown 4 in the regime of widely separated Landau in Fig. peaks (we have taken in what follows ωc = 5 v). In these two figures, we have chosen the situation where the LDoS-LDoS correlations are maximal, that is, we have considered that the energies ωi (here i = 1, 2) corre- spond exactly to Landau level energies Eni . The case of equal energies ω1 = ω2 = ω is first investigated in Fig. 4. According to the previous discussion, the dominant contributions among the different possible pairs of Lan- dau level indices are the diagonal ones corresponding to n1 = n2 = n. For the lowest Landau level energy (n = 0) shown with the solid line in Fig. 4, the correlations de- crease in a monotonic way as a function of the distance r between the two tip positions with a characteristic decay length of the order of the disorder correlation length ξ (here all the distances are expressed in units of the mag- netic length lB and we have taken ξ = 5 lB). When the energy of the first Landau level is probed (dotted line of Fig. 4 corresponding to ω = E1), the spatial dependence of the correlations is still decreasing with the same decay n = 2√2nlB ≈ 2.8lB for n = 1, associated to length ξ, but it now exhibits a mild peak for the position r close to 2RL the overlap of the quantum cyclotron rings for the n = 1 Landau states. In the second Landau level (at ω = E2), besides the peak close to 2RL n = 4lB for n = 2, an ad- ditional peak is clearly seen in the r dependence of the LDoS-LDoS correlations, see dashed line in Fig. 4. This is because the wave functions for n > 1 have n zeros, hence additional rings of high probability density. ω = E0 ω = E1 ω = E2 ) ω , ω , r ( χ 0.6 0.4 0.2 0 0 1 2 3 r/lB 4 5 6 FIG. 4: (color online) LDoS-LDoS correlations at equal en- ergies (ω1 = ω2 = ω) as a function of the tips distance r (in units of lB) for different energies ω. Here the correla- tions are expressed in units of (2πl2 Bv)−2 with the parameters ωc/v = 5 and lB/ξ = 0.2. The spatial dependence of the correlations presents peaks whose number corresponds to the Landau level index of the probed energy (for instance, for ω = E2 two peaks are seen). 7 This nonmonotonic spatial dependence observed for equal energies was anticipated in the Introduction of the paper, and can be understood more quantitatively as fol- lows. When the disorder potential is smooth on the scale of lB (i.e., ξ ≫ lB) the two functions involved in the integrand of Eq. (26) are, in fact, characterized by two very different characteristic length scales. Indeed, the function hn1,n2(R, r) decays with R on the typical length n = √2nlB which, for indices n not too big, turns scale RL out to be much smaller than the characteristic length n )2 + ξ2) for the spatial vari- ations of the functions Θn,n (R, ω, ω). Therefore, within the first few Landau levels a good approximation to the integral (26) is provided by the Laplace's method owing to the inequality ξ ≫ RL scale (of the order of p(RL n . Using the formula dx I0(ax)e−bx2 Z +∞ 0 =r π 4b e a2 8b I0(cid:18) a2 8b(cid:19) , (30) we obtain the approximate analytical expression for the LDoS-LDoS correlations at identical tip energies (ω1 = ω2 = ω) for ω close to the energy En χ(r, ω, ω) ≃ (2πl2 B)−2 2πv2 ( √πξ 2√2lB e−(ω−En)2/(2v2) 1 (n!)2 ∂2n 1 ∂sn 2"pBs1,s2 1 − s1s2 ∂sn I0(cid:18) r2 BBs1,s2(cid:19) e 4l2 − 4l2 B r2 Bs1 ,s2#s1=s2=0 − e−(ω−En)2/v2) . Consequently the nonmonotonous behavior of the cor- relations seen in Fig. 4 is connected with the oscilla- tions of the function hn,n(0, r) within Eq. (26), and is fully described analytically by formula (31) in the regime ξ ≫ lB. As seen in Fig. 5 for two large values of ξ/lB and for energies taken at the second Landau level, the agree- ment between the numerical evaluation of the integral in Eq. (26) and the analytical approximation (31) is quite excellent. We note also that the correlations disappear (for generic values of the frequency) in the clean limit v → 0 despite the v−2 prefactor due to the exponential terms in Eq. (31), while they become singular precisely at the Landau level frequency. We would like to comment here on the semiclassical n with n → ∞ (and vanishing lB) for the In that case, we replace the n ) in Eq. (22), and obtain limit of fixed RL LDoS-LDoS correlations. kernel Fn(R) by 1 2πRL n δ(R−RL (31) numerically Fig. 6. Not surprisingly, the quantum oscil- lations at r < 2RL n disappear, yet a kink subsists for the exactly tangent cyclotron trajectories at r = 2RL n , while a logarithmically diverging correlation occurs at r = 0. These are classical remnants of the quantum effects dis- cussed so far, and this robustness can again be expected according to the geometrical argument of Fig. 1 (now in the classical limit). These characteristic features can be understood analytically. Indeed, inserting Eq. (29) within formula (26), and performing Laplace's method on the R-dependent integral (this is valid in the case ξ ≫ RL), we find (2πl2 B)−2 2πv2 ξ 2Z +∞ 0 dqJ0(rq)[J0(RLq)]2−1!. χ(r, En, En) ≃ (32) Expression (32) is clearly logarithmically divergent at 1.5 1 0.5 ) 2 E , 2 E , r ( χ 0 0 ξ/lB = 14, numerics ξ/lB = 14, analytics ξ/lB = 5, numerics ξ/lB = 5, analytics 1 2 r/lB 3 4 FIG. 5: (color online) LDoS-LDoS correlations at equal en- ergies taken at the second Landau level (ω1 = ω2 = E2) as a function of the tips distance r (in units of lB) for two dif- ferent values of the disorder correlation length ξ/lB = 14 and ξ/lB = 5, with a comparison of the numerical evaluation of expression (26) and the analytical formula (31). 1 ) n E , n E , r ( χ 0.75 0.5 0.25 0 0 semiclassics 1 2 r/RL n 3 FIG. 6: (color online) LDoS-LDoS correlations at equal ener- gies taken at the Landau level (ω1 = ω2 = En) as a function of the tips distance r (in units of RL n ) in the semiclassical approximation (lB → 0) to integral (22). r = 0, but continuous at r = 2RL (the situation of tangent cyclotron orbits). The spatial derivative of the correlator can be analyzed in the limit r → 2RL, and gives a finite derivative for r < 2RL and a diverging one for r > 2RL, explaining the kink feature seen in Fig. 6. The onset of the logarithmic divergence at r = 0 can be understood starting from the quantum expression (31), which alternatively reads for r = 0 and ω = En: χ(0, En, En) = (2πl2 B)−2 2πv2 " ξ √2lBZ +∞ du(cid:2)Ln(cid:0)u2(cid:1)(cid:3)2 0 e−u2 − 1#. (33) The above equation is again obtained from Laplace's method in the case ξ ≫ RL n to analytically perform the integral over R in Eq. (22), and used the explicit expres- sion (6) for the kernel Fn, instead of the derivative trick performed in Eq. (31). Using the asymptotic formula for the Laguerre polynomial in the large n limit as in Ref. 8 n with RL n = √2nlB. 15, we get χ(0, En, En) ∝ log(n)/RL We recover the logarithmic divergence for n → ∞ and n , while this expression vanishes as log(n)/√n fixed RL for n → ∞ and fixed lB, in agreement with Fig. 4, where a slight decrease with increasing n is seen at r = 0. We thus stress that quantum mechanics (finite lB) always regularizes the singular classical behavior, and that max- imal correlations are, in fact, obtained (at a given mag- netic field, hence fixed lB) for the lowest Landau levels. Let us come back to the quantum case of finite lB and investigate the effect of energy detuning in the spatial de- pendence of the LDoS correlator. For energies ω1 6= ω2 a different spatial structure from the situation of equal energies can be seen in Fig. 7, which corresponds to the cases (ω1, ω2) = (E0, E1) and (ω1, ω2) = (E1, E2), the solid and dotted lines, respectively. Unlike the equal energy cases of Fig. 4, the correlations are no more max- imally obtained for the tips distance r = 0 but for an intermediate tip distance of the order RL 2 , as can be guessed again from the geometrical interpretation of Fig. 1, in the case RL 2 . We note that the spatial dependence of the correlations between the first and sec- ond Landau levels is also characterized by an extra mild peak, a reminiscent feature of the equal energy case. 1 6= RL 1 − RL ω = E0 ω = E1 0.4 ) c ω + ω , ω , r ( χ 0.3 0.2 0.1 0 0 1 2 3 r/lB 4 5 6 FIG. 7: (color online) LDoS-LDoS correlations at unequal energies (ω1 = ω and ω2 = ω + ωc) as a function of the tips distance r (in units of lB) for different energies ω. We used the same parameters as in Fig. 4. B. Energy dependence of the correlations We now study the energy dependence of the LDoS- LDoS correlations. We have represented in Fig. 8 the correlator χ(r, ω1, ω2) as a function of the energy ω2 = ω for identical tip positions (i.e., for r = 0) when the first tip energy is pinned to the first Landau level (ω1 = E1). We note the presence of different peaks in the correlations corresponding to the sequence ω = En of the Landau lev- els energy peaks in the local density of states. However, it is worth noting that the LDoS-LDoS correlation peaks are much sharper than the LDoS peaks because of the sensitive matching of the spatial overlaps between the ) ω , 1 E , 0 = r ( χ 0.6 0.4 0.2 0 0 1 2 ω/ωc 3 4 FIG. 8: (color online) LDoS-LDoS correlations for the tip dis- tance r = 0 and for the energy ω1 pinned to the first Landau level (ω1 = E1) as a function of the energy ω2 = ω. The other parameters are the same as in Fig. 4. The highest peak is obtained for ω2 = ω1 and is accompanied by satellite peaks corresponding to the neighbouring energy levels of the first Landau level. quantum rings associated to the states in Landau level n1 = 1 and n2 = n. Besides the strongest peak ob- tained for ω = E1, a succession of lateral peaks occurs for all other cyclotron energies En, showing that nondiag- onal elements of hn1,n2 (R, r) are also strongly correlated. More interestingly, the regions surrounding each of these sharp peaks with positive correlations are characterized by negative values for the correlator χ(0, E1, ω2), thus corresponding to anticorrelations. This crossover from positive to negative correlations as a function of the en- ergy can be easily understood on the basis of Eq. (23) at ξ ≫ RL, which is then well approximated by a small-R expansion Θn1,n2(R, ω1, ω2) ≃ e− ξ2 4R2 (ω1 −ω2+En2 −En1 )2 2v2 2R 1 2πv2 ξ − e− (ω1−En1 )(ω2 −En2 ) 2v2 ×e− (ω1−En1 )2 +(ω2 −En2 )2 2v2 !. (34) 9 To obtain this expression, we have developed the dis- order correlator Sn1,n2(R) [Eq. (19)] and the linewidth Γn in the large ξ limit. As can be seen in Eq. (34), a competition between two exponential terms occurs de- pending on the relative values of ω1 and ω2. For δω ≡ ω1 − ω2 + En2 − En1 = 0, the first term in the right-hand side (r.h.s) of Eq. (34), which is large and positive (of the order of ξ/R), dominates over the second exponential contribution, and one gets positive LDoS-LDoS correla- tions, as already noted. For δω slightly different from zero, the first term decreases exponentially fast (note the large ξ2/R2 prefactor within the exponential), resulting in the sharp peak observed in Fig. 8 near coinciding en- ergies. At increasing δω, the second term in the r.h.s of Eq. (34), which has a small and negative amplitude, is characterized by a slower exponential decrease, and then gives the main contribution to the correlations. Coming back to the initial Eq. (1) for the correlator χ(r, ω1, ω2), we conclude that the square average density dominates quickly at increasing energy detuning. Analytical insight can be obtained by further assum- ing the classical regime of RL ≫ lB (hence high Landau levels). Using then the approximation (29) within for- mula (26) together with Eq. (34) (which is valid in the limit ξ ≫ RL) we get the semiclassical approximation for the LDoS correlations at coinciding tip position r = 0 χ(0, ω1, ω2) = dq qJ0(RL n1 q)J0(RL n2 q)Z +∞ 0 dR J0(Rq)e− (δω)2 2v2 ξ 2πv2(2πl2 B)2 Xn1,n2Z +∞ −(cid:10)ρ(0, ω1)(cid:11)(cid:10)ρ(0, ω2)(cid:11). 0 ξ2 4R2 e− (ω1 −En1 )(ω2 −En2 ) 2v2 (35) Here the double sum over (n1, n2) is again constrained by the external frequencies (ω1, ω2), and reduces to a single term in case of sharply defined LDoS-LDoS corre- lation peaks, giving rise to a single peak at ω1 − ω2 = En1 − En2 . Let us focus first on the case where n1 = n2, n2 = RL. We see that the above in- so that RL tegrand in Eq. (35) then behaves as 1/q for vanishing δω, because J0(RLq)2 looses its oscillatory character n1 = RL at large momentum q. One obtains thus a logarithmi- cally diverging peak χ(0, ω1, ω2) ∼ log v/δω, related to the log lB/r spatial divergence found previously in the semiclassical limit at small intertip distance. How- ever, for nonzero n ≡ n1 − n2, the LDoS-LDoS correla- tion peak at ω1 − ω2 = En is no more logarithmically diverging when δω → 0. This is because the product Bnq/RL)/q for large J0(RL n2 q) behaves as cos(l2 n1 q)J0(RL B n n1,2 = RL± l2 momentum q, as can be seen using the expression for the cyclotron radii RL 2RL in the limit n ≪ n1, n2, where RL ≡ √n1 + n2lB. In order to capture the rele- vant energy scale at small δω, we can make the change in variables k = (ξδω/v)q and z = (v/ξΩn)R with B/(RLξ) into Eq. (35), and use the asymptotic Ωn = nvl2 form of the Bessel functions for δω ≪ ξv/RL. After in- tegration over k, this provides the deviation of the LDoS correlations from the nth peak value (at δω = 0) δχ = (2πl2 B)−2 2π2v2 ξ RL Z +∞ 1 dz e− δω 8z2 − 1 Ωn2 1 √z2 − 1 . (36) The new energy scale Ωn therefore sets the width of the nth correlation peak. Since Ωn ≪ v in the regime ξ ≫ RL ≫ lB, we recover the fact that the LDoS-LDoS correlations are more sharply defined than the average LDoS peaks. The linear increase of Ωn with n also ex- plains the progressive smearing of the correlations at in- creasing energy detuning of the tips (see Fig. 8). We now discuss the situation of strongly overlapping Landau level, ωc ≪ v =phV 2i, so that the average den- sity of states becomes structureless. In that case, the sharper peaks in the LDoS-LDoS correlations can sur- vive for smooth disorder (ξ ≫ lB), under the condition vlB/ξ ≪ ωc for the lowest Landau levels. A similar re- sult was already found by Rudin et al.17, who studied the energy dependence of the local LDoS-LDoS correlations in a weak magnetic field and long classical orbits RL ≫ ξ in the diffusive regime. The disorder dependence of cor- relator (31) in Ref. 17 was also in 1/v2 as in the present paper. It is worth stressing that the condition to obtain sharp peaks in the LDoS-LDoS correlations corresponds pre- cisely to the absence of local Landau level mixing. In- deed, transitions between adjacent Landau levels pro- vide18 a typical energy scale l2 B∇V 2/ωc, so that our cal- culation is controlled when the parameter (l2 B/ξ2)(v2/ω2 c ) is small. Clearly, large overlap in the average DoS can be compatible with no mixing, because this quantity re- lates to global properties (long-wavelength fluctuations) of the smooth disorder. In contrast, only the correlations of the LDoS can feel the local interplay of disorder and Landau quantization, and reveal whether the Landau in- dex n stays a good quantum number or not. Interestingly, a logarithmic singularity for the n = 0 peak and an energy width Ωn proportional to n2 for the nth peak were obtained in the semiclassical diffusive regime17. This is only slightly different from our results in the semiclassical limit, showing the continuity of the present physics from low to high magnetic fields. We emphasize again that quantum effects at finite magnetic length regularize the spurious divergences generically re- lated to the semiclassical approximation. Finally, we note that some of the energy-related fea- tures on the LDoS correlations discussed above have al- ready been reported experimentally8,10 in heavily doped three-dimensional (3D) GaAs semiconductors. For in- 10 stance, conductance anticorrelations were observed10 in the presence of Landau levels, while the narrowing of the LDoS fluctuations has been found8 at weak mag- netic field. However, these experimental studies were performed using resonant tunneling impurity, hence at a fixed position, asking for a different analysis from what was performed here (besides the 3D character of the stud- ied samples). Indeed, averaging was performed either over the applied voltage8 or magnetic field10, and in the later case, the Landau levels are simply lost. Spatial av- eraging in a STM configuration, as proposed in our work, should allow a greater control of the LDoS correlations (in terms of the applied magnetic field and tip voltages). Moreover, this offers a way to investigate spatial cor- relations of the LDoS, that could not be assessed with previous experimental techniques. VI. CONCLUSION We have studied theoretically the two-point correla- tions of the local density of states in a disordered two- dimensional electron gas under a large magnetic field. A rich spatial dependence of the correlations was found, which can be qualitatively explained by geometrical over- laps of the two quantum cyclotron rings that roughly de- scribe circular wave functions. The energy behavior of the correlations was shown to provide sharp peaks when the frequency detuning matches integer multiple of the cyclotron frequency, similar to the low magnetic field re- sults of Rudin et al.17. These sharp peaks in the LDoS correlations reveal that Landau levels correspond to well- defined quantum numbers, information that cannot be easily gathered from the average density of states only, where large overlaps are usually reported experimentally1 even at large magnetic fields. We have also emphasized here that LDoS correlations can be either positive or neg- ative, depending on the degree of frequency mismatch. We would also like to mention that at energies strictly co- inciding with the centers of Landau levels (quantum Hall critical point), the LDoS correlations exhibit a power-law behavior at long distance, reflecting the multifractality of critical wave functions19,20. This more complex behav- ior is beyond the scope of the present paper, which did not consider quantum tunneling between closed drift or- bits. Such effects are, however, only relevant at very low temperatures for a smooth potential. We end up by noting that recent experimental pro- gresses for electron gases confined at the surface of InSb semiconductor1 and also in graphene2,3 allow high spatial and energy resolution measurements of the local density of states. The disorder averaged LDoS and its corre- lations can, in principle, be straightforwardly obtained from the experimental data by averaging over large scale spatial maps. Because both the width of the disorder dis- tribution and the correlation length of the random land- scape can be extracted from the knowledge of the average LDoS, our predictions could be tested even quantitatively without extra fitting parameter. In the case of graphene, extension of the present work can be straightforwardly done following the results of Ref. 5. The spinorial form of the wavefunction is expected to lead to more complex spatial structures, yet with similar general behavior to that discussed here. Acknowledgments We are thankful for the hospitality of APCTP at POSTECH, where this work was initiated. M. E. R. 1 K. Hashimoto, C. Sohrmann, J. Wiebe, T. Inaoka, F. Meier, Y. Hirayama, R. A. Romer, R. Wiesendanger, and M. Morgenstern, Phys. Rev. Lett. 101, 256802 (2008). 2 D. L. Miller, K. D. Kubista, G. M. Rutter, M. Ruan, W. A. de Heer, P. N. First, and J. A. Stroscio, Science 324, 924 (2009). 3 D. L. Miller, K. D. Kubista, G. M. Rutter, M. Ruan, W. A. de Heer, M. Kindermann, P. N. First, and J. A. Stroscio, Nature Phys. 6, 811 (2010). 4 F. Wegner, Z. Phys. B 51, 279 (1983). 5 T. Champel and S. Florens, Phys. Rev. B 82, 045421 (2010). 6 L. Saminadayar, D. C. Glattli, Y. Jin, and B. Etienne, Phys. Rev. Lett. 79, 2526 (1997). 7 T. Schmidt, R. J. Haug, V. I. Fal'ko, K. v. Klitzing, A. Forster, and H. Luth, Phys. Rev. Lett. 78, 1540 (1997). 8 J. P. Holder, A. K. Savchenko, V. I. Fal'ko, B. Jouault, G. Faini, F. Laruelle, and E. Bedel, Phys. Rev. Lett. 84, 1563 (2000). 9 B. Jouault, M. Gryglas, G. Faini, U. Gennser, A. Cavanna, M. Baj, and D. K. Maude, Phys. Rev. B 73, 155415 (2006). 10 J. Konemann, P. Konig, T. Schmidt, E. McCann, V. I. Fal'ko, and R. J. Haug, Phys. Rev. 64, 155314 (2001). 11 T. Champel and S. Florens, Phys. Rev. B 80, 125322 (2009). 12 M. E. Raikh and T. V. Shahbazyan, Phys. Rev. B 51, 9682 (1995). 13 The equivalence between Eqs. (5) and (6) can be estab- lished straightforwardly using the definition of the La- guerre polynomials in terms of the generating function g(x, z) = exp [−xz/(1 − z)] /(1 − z). 14 A. A. Koulakov, M. M. Fogler, and B. I. Shklovskii, Phys. Rev. Lett. 76, 499 (1996); A. H. MacDonald, H. C. A. Oji, and K. L. Liu, Phys. Rev. B 34, 2681 (1986). acknowledges the support of DOE Grant No. DE-FG02- 06ER46313. 11 15 The formula Fn(R) ≃ 1 2πRL n be derived from Eq. (4) by going first to Fourier space δ(cid:0)R − RL n(cid:1) when n ≫ 1 can Vn(R) = Z d2q (2π)2 V (q) Fn(q) e−iq·R, (37) where V (q) and Fn(q) are the Fourier transforms of the potential V (R) and of the kernel Fn(R), respectively. The latter is written explicitly in Eq. (11). For large Landau- level indices, one then uses Eq. (6) and the asymptotic formula (38) n = √2nlB (which re- Introducing the new length scale RL places lB → 0), one finds that Eq. (37) reads for large n n(cid:17)i = J0 (cid:0)2√z(cid:1) . lim n→∞hLn (cid:16) z Vn(R) ≃ Z = Z d2q (2π)2 d2 η 2πRL n V (q)J0(RL n q) e−iq·R δ(cid:16)η − RL n(cid:17) V (R + η), (39) where we have dropped out the Gaussian factor e−l2 which vanishes for lB → 0. (1993). 16 M. E. Raikh and T. V. Shahbazyan, Phys. Rev. B 47, 1522 B q2/4 17 A. M. Rudin, I. L. Aleiner, and L. I. Glazman, Phys. Rev. B 58, 15698 (1998). 18 T. Champel, S. Florens, and L. Canet, Phys. Rev. B 78, 125302 (2008). 19 J. T. Chalker, Physica A 167, 253 (1990). 20 F. Evers and A. D. Mirlin, Rev. Mod. Phys. 80, 1355 (2008).
1503.05375	1	1503	2015-03-18T12:52:39	Strain-tunable topological quantum phase transition in buckled honeycomb lattices	[ "cond-mat.mes-hall" ]	Low-buckled silicene is a prototypical quantum spin Hall insulator with the topological quantum phase transition controlled by an out-of-plane electric field. We show that this field-induced electronic transition can be further tuned by an in-plane hydrostatic biaxial strain $\varepsilon$, owing to the curvature-dependent spin-orbit coupling (SOC): There is a $Z_2$ = 1 topological insulator phase for biaxial strain $\|\varepsilon\|$ smaller than 0.07, and the band gap can be tuned from 0.7 meV for $\varepsilon = +0.07$ up to a fourfold 3.0 meV for $\varepsilon = -0.07$. First-principles calculations also show that the critical field strength $E_c$ can be tuned by more than 113\%, with the absolute values nearly 10 times stronger than the theoretical predictions based on a tight-binding model. The buckling structure of the honeycomb lattice thus enhances the tunability of both the quantum phase transition and the SOC-induced band gap, which are crucial for the design of topological field-effect transistors based on two-dimensional materials.	cond-mat.mes-hall	cond-mat	a Strain-tunable topological quantum phase transition in buckled honeycomb lattices Jia-An Yan1,∗ Mack A. Dela Cruz1, Salvador Barraza-Lopez2, and Li Yang3 1. Department of Physics, Astronomy, and Geosciences, Towson University, 8000 York Road, Towson, MD 21252, USA 2. Department of Physics, University of Arkansas, Fayetteville, Arkansas 72701, USA 3. Department of Physics, Washington University, St Louis, MO 63005, USA (Dated: June 26, 2018) Low-buckled silicene is a prototypical quantum spin Hall insulator with the topological quan- tum phase transition controlled by an out-of-plane electric field. We show that this field-induced electronic transition can be further tuned by an in-plane hydrostatic biaxial strain ε, owing to the curvature-dependent spin-orbit coupling (SOC): There is a Z2 = 1 topological insulator phase for biaxial strain ε smaller than 0.07, and the band gap can be tuned from 0.7 meV for ε = +0.07 up to a fourfold 3.0 meV for ε = −0.07. First-principles calculations also show that the critical field strength Ec can be tuned by more than 113%, with the absolute values nearly 10 times stronger than the theoretical predictions based on a tight-binding model. The buckling structure of the honeycomb lattice thus enhances the tunability of both the quantum phase transition and the SOC- induced band gap, which are crucial for the design of topological field-effect transistors based on two-dimensional materials. Two-dimensional (2D) quantum spin Hall (QSH) in- sulator [1 -- 7] is characterized by an insulating bulk and gapless edge states at its boundaries [1, 2]. These edge states are topologically protected from backscattering of non-magnetic defects or impurities due to time-reversal symmetry, thus providing enticing concepts for novel quantum electronic devices with low energy dissipation [6, 7]. Quantized conductance through QSH edge states were originally reported on HgTe/CdTe [4, 5], and on InAs/GaSb [8, 9] quantum wells too. There is an intense drive to realize QSH insulators with controllable quantum phase transitions and tunable elec- tronic and spin properties [10, 11]. The intrinsic spin- orbit coupling (SOC) in graphene is weak [2, 12 -- 14], but other 2D materials that may realize this phenomena include honeycomb lattices of bismuth atoms on a sili- con surface [15], an electric-field-induced QSH phase on few-layer black phosphorus [16], a new structural phase of a transition-metal dichalcogenide [17, 18], and sil- icene/germanene [10, 11, 19 -- 24], among others [25 -- 28]. Mechanical strain and curvature (i.e., shape) [13, 29] modify the electronic and spin properties of buckled hon- eycomb lattices, and a deep understanding of their effects is emerging [30 -- 33]. In a classic work, the dependence of the SOC strength λSOC on curvature is established to arise from (1) on-site spin-flips among σ− and π− bands due to the intrinsic coupling among spin and s− and p−electronic orbitals, and (2) a subsequent hybridiza- tion of the hopping π−electron with electrons from bands with σ−symmetry in the presence of curvature [13]. This important phenomenon lies beyond those descriptions of the electronic structure of silicene and other group-IV 2D materials that are based on π−electrons only [10, 11]. Curvature is of central importance for the discussion of 2D-based topological field-effect transistors because it raises the intrinsic gap, and hence the temperature at which these devices could operate. On the other hand, a controllable and tunable topological quantum phase tran- sition will greatly facilitate the device fabrication and op- eration [18]. Working on silicene, we generalize Ezawa's result ∆(Ez) = 2lEz − λSOC for the dependence of the energy band gap ∆(Ez) on the electric field (E−field) Ez, in which the SOC strength λSOC is assumed to be a constant, into: ∆(Ez, ε) = 2l(ε)Ez − λSOC (ε), (1) where λSOC evolves with strain due to a curvature- induced hybridization among s and p electrons. Pristine silicene is a QSH insulator with a band gap ∆(Ez = 0, ε = 0) of about 1.5 meV (18 K) [20, 21]. This band gap is tunable by an E−field Ez perpendicu- lar to the buckled atomic layer [10, 22, 23] as an electro- static potential difference is established between the two Si atoms in the unit cell due to their height difference 2l. Silicene behaves as a Z2 = 1 non-trivial QSH insulator below a critical Ez strength Ec, and becomes a triv- ial band insulator for values of Ez > Ec [10, 11]. In addition, silicene is likely to be fabricated on substrates [35 -- 42] and may be subject of in-plane strain ε already. A fundamental understanding of effects of the strain on the topological quantum phase transition is important for the design of quantum electronic devices based on silicene and other 2D materials [15, 16, 18, 24]. Calculations were carried out using density-functional theory [43, 44] as implemented in the Quantum ESPRESSO code [45] with the GGA-PBEsol exchange- correlation functional [46]. The SOC was included in the fully relativistic Rappe-Rabe-Kaxiras-Joannopoulos (RRKJ) ultrasoft pseudopotential scheme [47] with a nonlinear core correction. The cutoff energy in the plane wave expansion is 50 Ry. A Monkhorst-Pack uniform k- grid of 36×36×1 is employed. A vacuum region of 20 A is introduced along the out-of-plane (z) direction to elim- inate spurious interactions among periodic images. The (a) y x (c) ) V e ( y g r e n E 3 2 1 0 -1 -2 -3 ) V e ( y g r e n E 0.006 0.003 0.000 -0.003 -0.006 e (b) ) (d) 3 2 1 0 -1 -2 -3 V e ( y g r e n E 0.006 0.003 0.000 -0.003 -0.006 E K M G K E = 0 V/nm E = 1 V/nm z (a) ) V e ( y g r e n E ) V e ( y g r e n E (b) ) V e m ( ) ,e 0 = E ( z K M G K e = 0.0 e = 0.05 e =-0.05 FIG. 1. (Color online) The electronic properties of silicene are independently tuned by (a) an in-plane biaxial strain ε, and (b) an out-of-plane E−field Ez. Subplots (c) and (d) are band dispersions under typical values of ε or Ez, respectively. Insets are zoom-ins of the band dispersion near the K−point (the Fermi level is set to zero; note the overlapping bands for ε = 0.00 and 0.05 on (c)). Here, we will explore the combined effects of ε and Ez on its electronic structures. E−field Ez is induced by a saw-tooth potential along the z−direction (c.f., Fig. 1(b)). Biaxial strain is applied to the silicene lattice, and ε is defined as ε = (a−a0)/a0 (see Fig. 1(a)). Here a and a0 = 3.85 A [24, 27, 48] are the strained and unstrained lattice constants, respectively. There is a 1.5 meV gap opening at the K−point that is induced by the SOC for ε = 0 and Ez = 0. When in- dependently applied (c.f., Figs. 1(a) and 1(b)) strain and E−field lead to distinct changes on the electronic band dispersions, as seen in Figs. 1(c) and 1(d), respectively. In-plane strain preserves inversion symmetry, leading to a gap opening that is more apparent for compressive strain (see the inset of Fig. 1(c), ε = −0.05), thus suggesting a relation among λSOC and ε that -- as a matter of fact -- has been discussed in the context of graphene some time ago [13]. On the other hand, Ez lifts inversion symmetry thus removing the band degeneracy, as seen on the inset of Fig. 1(d) for Ez = 1 V/nm (dashed lines). The effects of E−fields on the electronic properties of silicene have been thoroughly discussed in the past, so we continue exploring the effect of strain on Fig. 2. There are three effects of strain on the electronic struc- ture that are conveyed by Fig. 2(a): (i) A renormaliza- e = -0.07 K G K M G e = 0.02 K 1.5 1.0 0.5 0.0 -0.5 -1.0 -1.5 1.5 1.0 0.5 0.0 -0.5 -1.0 -1.5 e = -0.04 K G K M G e = 0.04 K 1.5 1.0 0.5 0.0 -0.5 -1.0 -1.5 1.5 1.0 0.5 0.0 -0.5 -1.0 -1.5 1.5 1.0 0.5 0.0 -0.5 -1.0 -1.5 1.5 1.0 0.5 0.0 -0.5 -1.0 -1.5 2 e = -0.02 K G K M G e = 0.07 K G K M G G K M G G K M G 3.5 3.0 2.5 2.0 1.5 1.0 0.5 D (Ez=0,e ) = -0.82 + 5.25l(e ) -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 e (%) 0.8 0.7 0.6 0.5 0.4 0.3 0.2 ) Å ( ) (e l 2 FIG. 2. (a) Typical band dispersions for various magnitudes of ε: A semicondoctor-to-metal transition occurs when ε > 0.07, as highlighted by the tilted (red) arrows. Insets help highlight the band dispersions near the Fermi level around the K−point. (b) Band gap ∆ (left axis) and the height difference 2l due to buckling (right axis) as a function of biaxial strain ε, for an Ez = 0 display a linear relationship in silicene, that has been explicitly indicated. tion of the Fermi velocity near the K−point [34]. (ii) An upward shift of the valence-band maxima (VBM) under compressive strain, and a downward shift of the conduction-band minima (CBM) under tensile strain at the Γ−point. (iii) A gap opening at the K−point that is especially evident under compressive strain. Effects (ii) and (iii) are responsible for the semiconductor-to-metal transition, and are the focus of the ensuing discussion. For sufficiently large compressive strain, the VBM crosses the Fermi level at the Γ−point (c.f., Fig. 2(a) for ε = −0.07; this crossing has been highlighted by a tilted red arrow). Similarly, the CBM at the Γ−point nearly touches the Fermi level for a tensile strain of ε = +0.07: Silicene undergoes a transition to a metallic phase for strain beyond ε ∼ 0.07. Although arising under dif- ferent mechanisms, these effects due to strain are remi- niscent of the topological-metal phase transition that is induced by an electric field in few-layer black phosphorus [16]. We next discuss the band dispersion seen around the G G D M e t a l ( V B M ) QSH 3 ) V e m ( (cid:39) 0 (cid:16)10 i n B a s n d u l a t o r (cid:72) ( 0 % ) in s B a n d ulator M (cid:16) 0 . 5 10 e t al ( C E 0 . 5 M ) m ) B 0 . 0 ( V /n z FIG. 3. (Color online) Topological quantum phase diagram of silicene with respect to in-plane biaxial strain ε and out- of-plane E−field Ez. The vertical axis is the band gap ∆. The critical E−fields Ec, at which there is a phase transition from a topological insulator into a band insulator, have been indicated by vertical white arrows. The metallic state has a zero value of ∆ that is shown in red, and the area marked by QSH represents the topological spin Hall state phase. K−point in the insets of Fig. 2(a). To do so, we first report the SOC-induced band gap ∆ and the buckling height 2l as a function of ε in Fig. 2(b): ∆ increases monotonically as ε decreases from ε = +0.07 down to ε = −0.07, meaning that compressive strain enhances the band gap ∆ while tensile strain decreases the gap. Since the energy band gap never closes in going from pristine silicene -- a Z2 = 1 topological insulator -- to strained silicene with ε < 0.07, so these systems share the same topological classification according to the adia- batic continuity argument for transformations of the elec- tronic bands. The above phenomenon can be understood from the change of the buckling height 2l among two silicon atoms due to an in-plane strain (l = l(ε)) shown in Fig. 2(b). Buckling (and hence l) increases with compressive strain, thus enhancing the overlap among σ and π orbitals, re- sulting in a strain-dependent SOC [13, 20]. The band gap ∆ increase is almost linear on l: ∆ ∼ 5.25l. A positive ε reduces l, making the structure more planar (graphene- like), so the overlap among σ and π orbitals decreases, bringing λSOC down into its "intrinsic value" (i.e., its value under a zero curvature) [13, 20]. However, when ε = +0.07, the CBM at the Γ−point nearly touches the Fermi level (tilted arrow on subplot in Fig. 2(a)), so that a further increase of ε induces a transition into a metallic phase. The strain-dependent SOC may be more qualitatively 3 explained with the following matrix form at atom A [21]: HA,SOC = λSOC 2  0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 -i 0 0 0 0 -i 0 +i 0 0 0 1 -i 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 -i 0 0 0 0 +i 0 +i 0 0 0 1 +i 0 0 0 0 0    , (2) on the basis set {s ↑i, px ↑i, py ↑i, pz ↑i, s ↓i, px ↓i, py ↓i, pz ↓i}. This SOC matrix produces spin flips among pzi (π) and pxi, pyi orbitals (that belong on σ− bands prior to the orbital hybridization) and λSOC could be obtained as a fitting parameter. The SOC matrix at atom B HB,SOC has an identical form. This intrinsic SOC is further enhanced by hopping [13]. After indicating the dependence among ∆ and ε seen on the first-principles data (Figs. 1 and 2), we study the combined effects of strain and E−field on the topolog- ical phase transition, and display ∆ as a function of the Ez and ε. ∆(Ez, ε) has a characteristic W −shape [10, 18, 20, 21, 24] as seen in Fig. 3. For a given value of ε, ∆ decreases to zero as Ez increases. For a given mag- nitude of ε smaller than 7%, the value of Ez for which ∆ = 0 is known as the critical field Ec(ε): As the field increases further beyond Ec(ε), the energy band gap ∆ reopens and increases again, but the electronic state is a trivial insulator because, according to the bulk-boundary correspondence principle, the topological phase transi- tion occurs at the critical field strength Ec(ε) when the band gap closes. Fig. 3 clearly shows that the critical field strength Ec(ε) increases with compressive strain: Strain tunes the quantum phase transition due to a buckling- dependent SOC. Fig. 3 can also be seen as a phase diagram that indi- cates the different quantum phases that can be reached by jointly tuning Ez and ε: Between −0.07 < ε <+0.07, silicene is a QSH insulator for any applied E−field in the range of E < Ec(ε). On the other hand, silicene changes to an ordinary band insulator with E > Ec(ε). This phase diagram is clearly visible in Fig. 3. c Finally, we discuss the critical field strength Ec. In Fig. 4, the first-principles data E DFT are shown as a func- tion of the theoretical predictions from Eq. (1): E th c = λSOC /l. The DFT results are nearly 10 times larger than the theoretical values, indicating a strong screening in sil- icene. In Ref. [23], the field-induced band gap is found to be suppressed by a factor of about eight due to the high polarizability of silicene. Here, we show that the screening also significantly enhances the critical E−field strength required to induce a quantum phase transition. c ∼ 9.6E th More precisely, a linear fit in Fig. 4 yields E DFT c . In summary, using first-principles methods, we show that the biaxial strain ε can be utilized to tune the spin- ) m n / V ( T F D E c 0.36 0.32 0.28 0.24 0.20 0.16 0.12 0.025 Ec DFT ~ 9.6 Ec th 0.035 0.040 0.045 0.030 Ec th (V/nm) (Color online) The critical E−field strength E DFT FIG. 4. calculated from DFT is shown as a function of the theoretical value E th c based on Eq. (1). The red line indicates a linear fit of the data. c orbit coupling in silicene and hence its topological quan- tum phase transition. At ε∼0.07, silicene undergoes a transition from topological insulator into a metallic phase. Within the range of −0.07 <ε< +0.07, pristine silicene remains a QSH insulator with a strain-dependent SOC that increases under a compressive strain. The crit- ical electric field strength is significantly enhanced by the screening, nearly 10 times larger than the theoretical pre- dictions from a tight-binding model. These phenomena highlight the interplay between the mechanical strain and E−field on the electronic properties of low-buckled hon- eycomb lattices. J.A.Y. acknowledges the Faculty Development and Re- search Committee grant (OSPR No. 140269) and the FCSM Fisher General Endowment at the Towson Uni- versity. M.A.D.C. is partially supported by FCSM Fisher General Endowment at the Towson University. S.B.L. Acknowledges funding from the Arkansas Biosciences In- stitute and NSF-XSEDE (TG-PHY090002). ∗ [email protected] [1] C. L. Kane and E. J. Mele, Phys. Rev. Lett. 95, 226801 (2005). 4 [7] X.-L. Qi, S.-C. Zhang, Rev. Mod. Phys. 83, 1057 (2011). [8] C. Liu, T. L. Hughes, X. L. Qi, K. Wang, S. C. Zhang, Phys. Rev. Lett. 100, 236601 (2008). [9] L. Du, I. Knez, G. Sullivan, R.-R. Du, http://arxiv.org/abs/1306.1925. [10] M. Ezawa, New J. Phys. 14, 033003 (2012). [11] M. Ezawa, Phys. Rev. Lett. 109, 055502 (2012). [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] D. Huertas-Hernando, F. Guinea, and A. Brataas, Phys. Rev. B 74, 155426 (2006). [14] Y. Yao, F. Ye, X.-L. Qi, S.-C. Zhang, and Z. Fang, Phys. Rev. B 75, 041401(R) (2007). [15] M. Zhou, W. Ming, Z. Liu, Z. Wang, P. Li, and F. Liu, Proc. Natl. Acad. Sci. (USA) 111, 14378 (2014). [16] Q. Liu, X. Zhang, L. B. Abdalla, A. Fazzio, and A. Zunger, http://arxiv.org/abs/1411.3932. [17] M. A. Cazalilla, H. Ochoa, and F. Guinea. Phys. Rev. Lett. 113, 077201 (2014). [18] X. Qian, J. Liu, L. Fu, and J. Li, Science 346, 1344 (2014). [19] S. Cahangirov, M. Topsakal, E. Akturk, H. Sahin, and S. Ciraci, Phys. Rev. Lett. 102, 236804 (2009). [20] C. C. Liu, W. X. Feng, and Y. G. Yao, Phys. Rev. Lett. 107, 076802 (2011). [21] C. C. Liu, H. Jiang, and Y. G. Yao, Phys. Rev. B 84, 195430 (2011). [22] Z. Ni, Q. Liu, K. Tang, J. Zheng, J. Zhou, R. Qin, Z. Gao, D. Yu, and J. Lu, Nano Lett. 12, 113-118 (2012). [23] N. D. Drummond, V. Z´olyomi, and V. I. Fal'ko, Phys. Rev. B 85, 075423 (2012). [24] J. E. Padilha, L. Seixas, R. B. Pontes, A. J. R. da Silva, and A. Fazzio, Phys. Rev. B 88, 201106(R) (2013). [25] Y. Ma, Y. Dai, M. Guo, C. Niu, and B. Huang, J. Phys. Chem. C 116, 12977 (2012). [26] Y. Xu, B. Yan, H.-J. Zhang, J. Wang, G. Xu, P. Tang, W. Duan, and S.-C. Zhang, Phys. Rev. Lett. 111, 136804 (2013). [27] P. Rivero, J.-A. Yan, V. M. Garcia-Suarez, J. Ferrer, and S. Barraza-Lopez. Phys. Rev. B 90, 241408(R) (2014). [28] P. Tang, P. Chen, W. Cao, H. Huang, S. Cahangirov, L. Xian, Y. Xu, S.-C. Zhang, W. Duan, and A. Rubio. Phys. Rev. B 90, 121408(R) (2014). [29] M. A. H. Vozmediano, M. I. Katsnelson, and F. Guinea. Phys. Rep. 496, 109 (2010). [30] A. A. Pacheco Sanjuan, Z. Wang, H. Pour Imani, M. Vanevic, and S. Barraza-Lopez. Phys. Rev. B 89, 121403(R) (2014). [31] A. A. Pacheco Sanjuan, M. Mehboudi, E. O. Harriss, H. Terrones, and S. Barraza-Lopez. ACS Nano 8, 1136 (2014). [32] J. Guan, Z. Jin, Z. Zhu, C. Chuang, B.-Y. Jin, and D. [2] C. L. Kane, E. J. Mele, Phys. Rev. Lett. 95, 146802 Tom´anek. Phys. Rev. B 90, 245403 (2014). (2005). [3] B. A. Bernevig, S.-C. Zhang, Phys. Rev. Lett. 96, 106802 (2006). [4] B. A. Bernevig, T. L. Hughes, S.-C. Zhang, Science 314,1757 (2006). [5] 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). [6] M. Z. Hasan, C. L. Kane, Rev. Mod. Phys. 82, 3045 (2010). [33] T. Castle, Y. Cho, X. Gong, E. Jung, D. M. Sussman, S. Yang, and R. D. Kamien. Phys. Rev. Lett. 113, 245502 (2014). [34] F. de Juan, M. Sturla, and M. A. H. Vozmediano. Phys. Rev. Lett. 108, 227205 (2012). [35] B. Aufray, A. Kara, S. Vizzini, H. Oughaddou, C. Le- andri, B. Ealet, and G. Le Lay, Appl. Phys. Lett. 96, 183102 (2010). [36] B. Lalmi, H. Oughaddou, H. Enriquez, A. Kara, S. Vizzini, B. Ealet, and B. Aufray, Appl. Phys. Lett. 97, 5 223109 (2010). [37] P. Vogt, P. DePadova, C. Quaresima, J. Avila, E. Frantzeskakis, M. C. Asensio, A. Resta, B. Ealet, and G. LeLay, Phys. Rev. Lett. 108, 155501 (2012). [38] A. Fleurence, R. Friedlein, T. Ozaki, H. Kawai, Y. Wang, and Y. Yamada-Takamura, Phys. Rev. Lett. 108, 245501 (2012). [39] L. Chen, C.-C. Liu, B. Feng, X. He, P. Cheng, Z. Ding, S. Meng, Y. Yao, K. Wu, Phys. Rev. Lett. 109, 056804 (2012). [40] B. Feng, Z. Ding, S. Meng, Y. Yao, X. He, P. Cheng, L. Chen, and K. Wu, Nano Lett., 12, 3507-3511 (2012). [41] H. Jamgotchian, Y. Colignon, N. Hamzaoui, B. Ealet, J. Y. Hoarau, B. Aufray, and J. P. Bib´erian, J. Phys. Condens. Matter 24, 172001 (2012). [42] C.-L. Lin, R. Arafune, K. Kawahara, N. Tsukahara, E. Minamitani, Y. Kim, N. Takagi, and M. Kawai, App. Phys. Exp. 5, 045802 (2012). [44] W. Kohn and L. J. Sham, Phys. Rev. 140, A1133 (1965). [45] P. Giannozzi, S. Baroni, N. Bonini, M. Calandra, R. Car, C. Cavazzoni, D. Ceresoli, G. L. Chiarotti, M. Cococ- cioni, I. Dabo, A. Dal Corso, S. Fabris, G. Fratesi, S. de Gironcoli, R. Gebauer, U. Gerstmann, C. Gougous- sis, A. Kokalj, M. Lazzeri, L. Martin-Samos, N. Marzari, F. Mauri, R. Mazzarello, S. Paolini, A. Pasquarello, L. Paulatto, C. Sbraccia, S. Scandolo, G. Sclauzero, A. P. Seitsonen, A. Smogunov, P. Umari, R. M. Wentzcovitch, J. Phys. Condens. Matter 21, 395502 (2009). [46] J. P. Perdew, A. Ruzsinszky, G. I. Csonka, O. A. Vydrov, G. E. Scuseria, L. A. Constantin, X. Zhou, and K. Burke, Phys. Rev. Lett. 100, 136406 (2008); Erratum Phys. Rev. Lett. 102, 039902(E) (2009). [47] A. M. Rappe, K. M. Rabe, E. Kaxiras, and J. D. Joannopoulos, Phys. Rev. B 41, 1227(R) (1990); Erra- tum Phys. Rev. B 44, 13175 (1991). [48] F. Bechstedt, L. Matthes, P. Gori, and O. Pulci, App. [43] P. Hohenberg and W. Kohn, Phys. Rev. 136, B864 Phys. Lett. 100, 261906-261908 (2012). (1964).
1608.06311	1	1608	2016-08-22T20:49:48	Disorder Induced Phase Transitions of Type-II Weyl Semimetal	[ "cond-mat.mes-hall", "cond-mat.dis-nn" ]	Weyl semimetals are a newly discovered class of materials that host relativistic massless Weyl fermions as their low-energy bulk excitations. Among this new class of materials, there exist two general types of semimetals that are of particular interest: type-I Weyl semimetals, that have broken inversion or time-reversal symmetry symmetry, and type-II Weyl semimetals, that additionally breaks Lorentz invariance. In this work, we use Born approximation to analytically demonstrate that the type-I Weyl semimetals may undergo a quantum phase transition to type-II Weyl semimetals in the presence of the finite charge and magnetic disorder when non-zero tilt exist. The phase transition occurs when the disorder renormalizes the topological mass, thereby reducing the Fermi velocity near the Weyl cone below the tilt of the cone. We also confirm the presence of the disorder induced phase transition in Weyl semimetals using exact diagonalization of a three-dimensional tight-binding model to calculate the resultant phase diagram of the type-I Weyl semimetal.	cond-mat.mes-hall	cond-mat	a Disorder Induced Phase Transitions of Type-II Weyl Semimetal Moon Jip Park1,2, Bora Basa3, and Matthew J. Gilbert2,3 1 Department of Physics, University of Illinois, Urbana, IL 61801 2 Micro and Nanotechnology Laboratory, University of Illinois, Urbana, IL 61801 and 3 Department of Electrical and Computer Engineering, University of Illinois, Urbana, IL 61801 (Dated: August 24, 2016) Weyl semimetals are a newly discovered class of materials that host relativistic massless Weyl fermions as their low-energy bulk excitations. Among this new class of materials, there exist two general types of semimetals that are of particular interest: type-I Weyl semimetals, that have broken inversion or time-reversal symmetry symmetry, and type-II Weyl semimetals, that additionally breaks Lorentz invariance. In this work, we use Born approximation to analytically demonstrate that the type-I Weyl semimetals may undergo a quantum phase transition to type-II Weyl semimetals in the presence of the finite charge and magnetic disorder when non-zero tilt exist. The phase transition occurs when the disorder renormalizes the topological mass, thereby reducing the Fermi velocity near the Weyl cone below the tilt of the cone. We also confirm the presence of the disorder induced phase transition in Weyl semimetals using exact diagonalization of a three-dimensional tight-binding model to calculate the resultant phase diagram of the type-I Weyl semimetal. Introduction- Weyl semimetals (WSM), which are characterized by the gapless bulk states whose Fermi sur- faces are either nodal points or lines, have been an intense area of research1–8. The bulk nodal points of a WSM pos- sess non-degenerate band crossings that are robust under a small perturbations to the Hamiltonian. The low en- ergy excitations of these materials are Weyl fermions that are described by two component spinors. The topological nature of the WSM are revealed through examination of the monopoles of Berry curvature present in the Brillouin zone(BZ) referred to as Weyl nodes. In the WSM, the net monopole charge integrated over the entire BZ is zero as the WSM possesses an equal number of the nodes with the positive and the negative charges9;10. In the non- interacting theory, these nodal points can only be elimi- nated by coming into contact with an oppositely charged Weyl node in momentum space, thereby annihilating one another. WSM have been predicted in a number of differ- ent materials and structures each of which is character- ized by the symmetries broken. The most studied WSM are type-I WSM (WSM1), characterized by the presence of broken inversion or time-reversal symmetry, and type- II WSM (WSM2), which possess broken Lorentz invari- Inversion broken WSM1, such as T aAs4;11, are ance. characterized by the presence of disconnected Fermi arcs at the surface1 that give rise to unconventional trans- port signatures such as quantum anomalous Hall (QAH) effect12 and the chiral anomaly13. More recently, WSM2, such as M oT e2 and W T e2, have been discovered14–17. WSM2 are characterized by an exotic hyperboloid Fermi surface, where the nodes are tilted in the BZ. Due to the tilted nodes, WSM2 exhibit transport properties that are distinct from the WSM1 including the absence of the chi- ral anomaly at certain magnetic field angles18, magnetic break down resulting in a collapse of Landau levels19, and anisotropy of the dynamical conductivity20. While the gapless states are protected by the un- derlying topology of the phases, the topology of the band structure may be altered when the finite disor- der is present21–26. This is due to the renomalization of the topological mass that determines the band topol- ogy. The effect of disorder within topological materials has been studied in a number of different contexts includ- ing time-reversal topological insulators21–26, where it is shown that the symmetry preserving disorder may in- duce a transition between topological insulator and triv- ial insulator. Furthermore, symmetry preserving disorder may also induce a transition between the weak topolog- ical insulator to the strong topological insulator phases. Similarly, the effects of disorder have been examined in WSM127–30 in which WSM1 is shown to undergo a tran- sition between normal insulator(NI) and 3D quantum anomalous hall insulators(QAHI) in the presence of both charge and magnetic disorder. While the effect of disorder in WSM1 is well- established, the stability of WSM2 in the presence of dis- order has not been considered. In this work, we show that WSM1 with small but non-zero tilt undergoes a quantum phase transition to WSM2 when charge and magnetic dis- orders are present. We illustrate this phase transition by calculating the topological mass renormalization that oc- curs within the first order Born approximation. We find that the topological mass is renormalized while the tilt of the Weyl cones in remains invariant. As a result, the Fermi velocity near the Weyl cone is also renormalized and leads to the possible phase transition between WSM1 and WSM2. Additionally, we confirm our results using numerical exact diagonalization of a three-dimensional tight binding model. To analyze the effect of the disor- der self-energy contribution to the numerically obtained Green's function, we utilize the spectral function, which enables to understand the change of the band structure even in the presence of the disorder31. Furthermore, we find that the reverse transition from WSM2 to WSM1 is also possible depending on the value and sign of the topological mass. Our work reveals the rich phase dia- gram of WSM in the presence of disorder and will aid in the experimental characterization of WSM materials. Model -We begin by writing the Hamiltonian a Weyl fermion using the lowest order expansion of the momen- tum near the Weyl cone as, H = vi,jkiσj + γtilt,ikiI2 (1) (cid:88) i,j where vi,j is a matrix which specifies the Fermi veloc- ity, and σi(I2) is the i-th pauli matrix and I2 is the 2 × 2 identity matrix. The second term, γtilt,i, tilts the cone in i-direction and whose presence the breaks Lorentz invariance5. The dispersion of the Hamiltonian in Eq. (1) is given as, E = γtilt,iki +(cid:112)kivijvjkkk (2) In this work, v is a diagonal matrix corresponding to the velocity in each direction. It should be noted that the choice of the velocity does not effect our conclusions. To consider the disorder effect, we rewrite the Weyl Hamil- tonian in the real space in lattice regularized form as, HW = H0 + Htilt (3) where H0 is the minimal two band model of WSM1, which breaks the time-reversal symmetry3;12, and may be written as H0 = txsin(kx)σx + tysin(ky)σy +[m0(2 − cos(kx) − cos(ky)) + (mz − tzcos(kz))]σz. (4) The second term in Eq. (3) is the tilt term, Htilt, that breaks the Lorentz invariance can be generally written up to quadratic order in the lattice regularized form as, Htilt = at,isin(ki)I2 + bt,icos(ki)I2 (5) Using Eq. (3), WSM2 is characterized by the choice of parameters in which the tilt within any direction domi- nates the Fermi velocity of H0 near the cone, γtilt,i > vi, where the individual tilts at the cones in the lattice Hamiltonian are given as γtilt = (atcos(Q) ± btsin(Q)). In this model, the Weyl cones are located at Q = (0, 0,±acos(mz/tz)) in the BZ and the Fermi velocity k=Q = tx,y in each of the cones is given as vx,y = ∂E ∂kx,y k=Q = tzsin(Q). m0 gaps out the spec- and vz = tz trum at X and Y points. Here, we choose the direction of the tilt to be in the z-direction as the purpose of this work is to observe the phase transition resulting from the renormalized topological mass, mz, in the z direc- tion. As the goal of this work is to examine the effects of disorder on the resultant phases in WSM, a momentum space representation of the Hamiltonian is not useful as the inclusion of disorder forbids the use of a momentum ∂E ∂kz 2 (cid:88) H = δ=x,y,z space representation. Therefore, it is necesary to utilize the real space representation of the full Hamiltonian is given as, † c (i,j,k)+δhδci,j,k + c † i,j,k(2m0I2 + mzσz)ci,j,k +h.c. (6) where i, j, k are the coordinates corresponding to the x, y, z directions respectively, and h(x,y) = itx,y 2 σx,y, hz = tz In the remainder of this work, we place a tilde on the top of each parameter so as to indicate it has been normalized respect to tz. 2 σz + iat+bt I2. 2 Phase in the Clean Limit-In Fig. 1, we plot the phase diagram for the WSM as a function of the value of the topological mass and the tilt parameter. The phase dia- gram of the WSM in the clean limit is characterized by the locations of the Weyl nodes. When a non-zero mz is present, the Weyl cones are separated in the BZ by the momentum vector (0, 0, 2Q). As long as the nodes are well-separated in momentum space, the WSM phases are stable in the single particle picture. However, when the nodes with the opposite monopole charge meet one another at the same point in the BZ, they annihilate and produce a gap in the spectrum32 resulting in an insulat- ing phase. This transition is mathematically defined as the point in the BZ where Q is ill-defined, namely mz > 1 and mz < −1 as is shown in the boundary with the solid lines in Fig. 1. There are two distinct insulating phases: the NI phase ( mz > 1) and QAHI phase ( mz < −1). The QAHI phase can be thought as stacks of Chern in- sulators with a non-zero Hall conductance3. These two distinct phases of insulator can be heuristically under- stood by considering a slice of the Hamiltonian at each momentum in kz. In this case, each slice looks like a gapped Dirac fermion except at the gapless nodes at the (0, 0,±Q). At these points, there exists a non-zero Hall conductance in the interval of kz ∈ (−Q, Q), where the outside of the interval the WSM has the zero Hall con- ductance. As we decrease mz, the Weyl cones shift in momentum space towards the (0, 0,±π) and the region in momentum space with a non-zero Hall conductance extends. Finally, when the Weyl nodes annihilate in the momentum space at (0, 0,±π) which occurs at mz = −1, the gapped bulk is fully specified by the non-zero Hall conductance. In the opposite limit, where the nodes an- nihilate at (0, 0, 0) for mz = 1, the kz region with zero Hall conductance extends. In this situation, when the nodes meet in the zero momentum space, the Hamilto- nian is characterized by zero Hall conductance resulting in the NI phases. In addition to the metal-insulator transitions, there ex- ist WSM1-WSM2 transitions as we increase the mz in the presence of non-zero γtilt. This phase transition oc- curs when the increased mz lowers the Fermi velocity 3 FIG. 2: The phase diagram of the Weyl semimetal with the effect of disorder included as calculated from the first Born approximation with the parameters, tx = ty = at = 1. In this figure, we represent the different phases of WSM1, WSM2 and the insulating phase. We observe that, depending on the initial negative values of mz, as we increase the disorder there exists a phase transition from WSM1 to WSM2 and finally to an insulating phase. However, we also note that for a range of values of positive mz, we find that insulating metals may be driven through both the WSM2 and WSM1 phases with increasing disorder. tion, ¯G, that is given as, 1 1 ¯G = = (cid:104) E − HW − Σdis + iη E − (HW + Hdis) + iη (cid:105), (8) where <> indicates the average expectation value over the random disorder configurations and η is the infinites- imal broadening term. To calculate the self energy, Σdis, we use the Dyson equation to dress the single particle Green's function given as, ¯G = G0 + G0Σ ¯G (9) 1 where G0 = E−HW +iη is the Green's function of the bare Hamiltonian without disorder. Then, applying the Born approximation, the correction of the self-energy term is equivalent to33, [(cid:82) W/2 Σ = [I2][< riG0ri >] (10) −W/2 2d] (cid:90) W = W 2 12 d3k 1 E − H + iη . When we decompose the self-energy into its various directional contributions, these correspond to terms in which the chemical potential and the topological mass near E = 0 are renormalized as21;30;33, µre = µ + limk→0Re(Σ0), mi,re = mi + limk→0Re(Σi). (11) (12) Where Σi is the self-energy correction is decomposed into each of the i-th directional pauli matrices. We observe FIG. 1: Plot of the phase diagram of the WSM in the clean limit. There are the two distinct insulating phase regions cor- responding to different values of the topological mass: QAHI when mz < −1 and trivial insulator for mz > 1. In this phase diagram, we have ignored bt, as the WSM2 transition occurs generally when bt > 1. below the finite value of the tilt term, γtilt. This condi- tion can be analytically written for our specific model as vf = tzsin(Q) < γtilt. In Fig. 1, we plot the full phase diagram of the WSM in the clean limit as a function of the value of γtilt and mz showing the transitions between WSM1, WSM2, and insulator phases as calculated using the analysis presented. The Effect of Disorder -After establishing the phases of the WSM in the clean limit, we now consider the inclusion of the disorder, the form of which is given as, Hdis = r,sc† r,scr,s (7) r,s where r and s are the coordinate in the lattice and spinor index respectively. r,s is a uniformly distributed random number in the range of [−W/2, W/2] utilized to mimic the random on-site disorder potential to be added into Eq. (3). Note that this disorder configuration does not preserve time-reversal symmetry, therefore we are con- sidering both charge and magnetic disorder. Within a given distribution of the disorder, we begin by specify- ing the correlation function of the disorder, whose av- erage value zero, in the following form (cid:104)i,s(cid:105) = 0. The two point correlation of the disorder energy is given as, (cid:104)i1,s1i2,s2(cid:105) = 12 δi1,i2δs1,s2. To more clearly see the effect of the disorder, we calculate the disorder averaged self-energy term in Green's func- δi1,i2δs1,s2 = W 2 [(cid:82) W/2 −W/2 2d] W (cid:88) where the integral expression in the above equation, α, is derived from the evaluation of the Eq. (10) as, 4 that Re(Σx,y) vanishes since the numerator of the Eq. (10) is an odd function of the momentum. Therefore, within the Born approximation, the value of the topolog- ical mass terms, mx and my, are invariant even with the disorder included in our analysis. Using Eq. (10), Σz is derived as, Σz ≈ +W 2α. (13) (cid:90) d3k (2π)3 α = − 1 12 m0(2 − cos(kx) − cos(ky)) + (mz − tzcos(kz))2 + txsin(kx)2 + tysin(ky)2 − atsin(kz) + btcos(kz)2 m0(2 − cos(kx) − cos(ky)) + (mz − tzcos(kz)) ≈ −1 48m0π log( (mz +(cid:112)m2 t + (bt − tz)2))(mz +(cid:112)m2 0π4 m2 z − (a2 z − (a2 t + (bt + tz)2)) ). (14) (15) To obtain Eq. (15), we have kept terms only up to quadratic order of kx,y contributions of the integral. In similar manner as outlined in Eq. (11), a renormaliza- tion of the chemical potential also occurs when btilt,i (cid:54)= 0 since the btilt term is even function in the momentum space. However, the change of the chemical potential does not alter the phase of the WSM. We only focus on the renormalization of the topological mass. In Eq. (15), we notice that the sign of α is always negative (of m0), therefore, the disorder renormalizes the value of mz to decreasing values as the magnitude of the disorder is increased as, mz → mz + αW 2 (16) Furthermore, the Fermi velocity in z direction is also renormalized as vf = tzsin(acos(mz/tz)) → tzsin(acos((mz + αW 2)/tz)) (17) which results in the renormalization of the effective Fermi velocity in z direction. Therefore, while the tilt is invari- ant in the presence of the disorder, there is a decrease (increase) of the Fermi velocity when mz > 0( mz < 0). We find that when the effective Fermi velocity becomes smaller than the tilt at mz < 0, WSM1 undergoes the dis- order induced quantum phase transition to WSM2 and then to an insulating phase with increasing disorder, as shown in Fig. 2 indicated by the arrow. Additionally, the reverse transition from an insulating phase to WSM2 and finally to WSM1 with increasing disorder is possible if we start from mz ≈ 1. Numerical Calculation-While the first-order Born ap- proximation offers an analytical insight of the disorder induced phase transition, to confirm the result from (3). the Born approximation, we numerically investigate the disorder effect via exact diagonalization of the three- dimensional tight-binding Hamiltonian of Eq. In the presence of the disorder, we calculate the momentum space spectral function to observe the phase transitions with increasing disorder. The spectral function calcula- tion gives an estimate of the tilt and the topological mass with a finite broadening of the states due to the presence of the disorder31. As the value of the disorder increases, we use the largest value of the spectral function at a given kz to identify the resultant change of the dispersion. The spectral function in the momentum space is defined as34, A(ω, k) = φn,i(k)2 ω − En + iη (18) 2(cid:88) i=1 Where η is the infinitesimal imaginary number and i is the band index of the eigenstate. En and φn,i(k) are the energy eigenvalue and the eigenstate of the Hamil- tonian with disorder respectively. The spectral function is calculated using a system size of 8 × 8 × 60 sites in the real space lattice. We numerically distinguish the WSM1, WSM2, and insulating phases by identifying the resultant changes of the dispersions and the tilt at the cones, while the change of the tilt is calculated from the slopes of the dispersions at the cone. We identify the insulating phases when the spectral function contains a gapped dispersion, and, similarly, we identify the WSM2 phase when the resultant dispersion shows the charac- teristic tilted cones. Fig. 3 shows the disorder induced transition from the WSM1 to the WSM2 and, finally, to the insulating phase. To understand these different disorder induced transitions, we begin with Fig. 3 (a), which uses ( mz, W ) = (−0.27, 0), where we present the spectral function which contains degenerate crossings of two bands at k ≈ (0, 0,±2) where the Weyl nodes are 5 and shows the tilted Weyl cones. Again, the increase of W reduces the renormalized mz until when the location of the nodes, Q = ±(acos( mz), are ill-defined to gap out the dispersion. Fig. 3 (d) shows this transition, as the disorder, W , reaches the value of 4. Fig. 3 (d) shows the spectral function of the insulating phase which are fully gapped out. In result, the behavior of the transitions shown in the numerical calculation can be understood from the decrease of mz and the resulting change of the Fermi velocity due to the disorder induced renormaliza- tion, in which it eventually shifts the full phase diagram to the positive mz direction. In Fig. 3 (e), we show the complete disorder induced phase diagram of the WSM as a function of W and mz in which we derive by repeating the calculations with var- ious values of mz. The numerical phase diagram agrees in its behavior with the phase transitions predicted us- ing the Born approximation in Fig. 2. The general in which mz reduces as the disorder increases, trend, is similar to the previous study30. However, in con- trast to the previous study, we find that there always exists a finite region of WSM2 phase before the WSM1 phase undergoes the transition to an insulating phase, when atilt is non-zero. This intermediate region of the WSM2 phase is guaranteed to exist because the Fermi velocity eventually vanishes and the tilt dominates be- fore the insulating transition. Due to the WSM1-WSM2 transition, WSM1 phase near mz = −cos(atan(at)) is unstable to the arbitrary weak disorder to the transi- tion to the WSM2 phase. On the opposite side of the phase diagram where mz is positive, WSM2 phase near mz = cos(atan(at)) is unstable to the WSM1 phase tran- sition as the renormalized mz decreases the Fermi veloc- ity. On both sides of the phase boundaries, the WSM2 phase is unstable to the finite disorder, resulting the transitions to WSM1( mz = cos(atan(at))) and QAHI ( mz = −1), as the renormalized Fermi velocity and mz is significantly modified. As the disorder increases large enough, Anderson localization can occur where the per- turbation theory fails and the bulk gap collapses30. Conclusion- In conclusion, we have studied the effect of disorder on the resultant phase diagram of WSM with a non-zero tilt to elucidate the boundaries of the different physical regimes as a function of disorder and topologi- cal mass. We have illustrated these various phase transi- tions both analytically, using the first Born approxima- tion, and numerically, via exact diagonalization calcula- tions. We find that the renormalization of the topological mass changes the effective Fermi velocity of the nodes re- ducing both the mass and resultant Fermi velocity with increasing values of disorder. The resulting change of the Fermi velocity leads to phase transitions between WSM1, WSM2, and insulator. Moreover, our results show that the disorder induced WSM2 phase always occurs before the metal-insulator transition of WSM1. Therefore, we assert that the WSM2 phase naturally occurs before the Spectral function and disorder induced phase di- FIG. 3: agram derived from exact diagonalization. In each of the plots at the top (a),(b),(c) and (d), we show the numeri- cally calculated spectral function as a function of kz. The plots (a),(b),(c), and (d) use the following parameters for ( mz, W ) : (−0.27, 0), (−0.27, 3), (−0.78, 0), (−0.78, 4) respec- tively. We find that the system undergoes a phase transitions to WSM1-WSM2-QAHI, as predicted from the Born approxi- mation. (e) The full phase diagram calculated from the exact diagonalization. This figure is generated with the parameters, tx = ty = at = 1. The parameters used in (a)-(d) are marked with different shapes on (e) respectively. located in momentum space along the z-direction. The spectral function has no broadening as the correspond- ing phase and dispersion are calculated in the clean limit. We find that the two Weyl cones have no tilt, which in- dicates WSM1 phase. Gradual increase of W reduces the renormalized mz according to Eq. (16), until γtilt dominates the Fermi velocity and resulting in WSM2 phase. Eventually, when W reaches up to 3 in Fig. 3 (b), the calculated dispersion shows the tilted Weyl cones along z direction, therefore it signals the disorder induced phase transition from WSM1 to WSM2. In ad- dition to the WSM1-WSM2 phase transition, Fig. 3 (c) and (d) show a transition from WSM2 phase to insulat- ing phase. Fig. 3 (c), with the choice of the parameters ( mz, W ) = (−0.78, 0) shows WSM2 phase in the clean limit in which the spectral function has no broadening disorder induced transition between WSM1 and the in- sulating phases in the known WSM materials. 6 [28] S. Bera, J. D. Sau, and B. Roy, Phys. Rev. B 93, 201302 (2016). [29] B. Roy and S. D. Sarma (2016), arxiv:1511.0636. [30] C.-Z. Chen, J. Song, H. Jiang, Q.-f. Sun, Z. Wang, and X. C. Xie, Phys. Rev. Lett. 115, 246603 (2015). ACKNOWLEDGEMENT [31] R. Queiroz and A. P. Schnyder, Phys. Rev. B 91, 014202 (2015). [32] C. Fang, M. J. Gilbert, X. Dai, and B. A. Bernevig, Phys. Rev. Lett. 108, 266802 (2012). [33] A. Altland and B. D. Simons, Condensed Matter Field theory (Cambridge, 2010). [34] S. Datta, Quantum Transport: Atom to Transistor (Cambridge University Press, 2013). This work is supported by NSF CAREER ECCS- 1351871. [1] X. Wan, A. M. Turner, A. Vishwanath, and S. Y. Savrasov, Phys. Rev. B 83, 205101 (2011). [2] Y. Kim, M. J. Park, and M. J. Gilbert, Phys. Rev. B 93, 214511 (2016). [3] A. A. Burkov and L. Balents, Phys. Rev. Lett. 107, 127205 (2011). [4] B. Q. Lv, H. M. Weng, B. B. Fu, X. P. Wang, H. Miao, J. Ma, P. Richard, X. C. Huang, L. X. Zhao, G. F. Chen, et al., Phys. Rev. X 5, 031013 (2015). [5] A. A. Soluyanov, D. Gresch, Z. Wang, Q. Wu, M. Troyer, X. Dai, and B. A. Bernevig, Nature 527, 495 (2015). [6] G. Y. Cho, J. H. Bardarson, Y.-M. Lu, and J. E. Moore, Phys. Rev. B 86, 214514 (2012). [7] S. Murakami, New Journal of Physics 9, 356 (2007). [8] P. Delplace, J. Li, and D. Carpentier, EPL (Europhysics Letters) 97, 67004 (2012). [9] H. Nielsen and M. Ninomiya, Phys. Lett. B 105, 219 (1981). [10] H. Nielsen and M. Ninomiya, Phys. Lett. B 130, 389 (1983). [11] S.-Y. Xu, I. Belopolski, N. Alidoust, M. Neupane, G. Bian, C. Zhang, R. Sankar, G. Chang, Z. Yuan, C.-C. Lee, et al., Science 349, 613 (2015), ISSN 0036-8075. [12] K.-Y. Yang, Y.-M. Lu, and Y. Ran, Phys. Rev. B 84, 075129 (2011). [13] S. A. Parameswaran, T. Grover, D. A. Abanin, D. A. Pesin, and A. Vishwanath, Phys. Rev. X 4, 031035 (2014). [14] N. Xu and et al. (2016), arXiv:1604.02116. [15] A. Liang and et al. (2016), arXiv:1604.01706. [16] C. Wang and et al. (2016), arXiv:1604.04218. [17] F. Y. Bruno and et al. (2016), arXiv:1604.02411. [18] M. Udagawa and E. Bergholtz (2016), arXiv:1604.08457. [19] T. E. O'Brien, M. Diez, and C. W. J. Beenakker, Phys. Rev. Lett. 116, 236401 (2016). [20] M. Koshino (2016), arXiv:1605.08282. [21] C. W. Groth, M. Wimmer, A. R. Akhmerov, J. Tworzyd(cid:32)lo, and C. W. J. Beenakker, Phys. Rev. Lett. 103, 196805 (2009). [22] K. Kobayashi, T. Ohtsuki, and K.-I. Imura, Phys. Rev. Lett. 110, 236803 (2013). [23] P. Titum, N. H. Lindner, M. C. Rechtsman, and G. Re- fael, Phys. Rev. Lett. 114, 056801 (2015). [24] S. Ryu and K. Nomura, Phys. Rev. B 85, 155138 (2012). [25] J. Li, R.-L. Chu, J. K. Jain, and S.-Q. Shen, Phys. Rev. Lett. 102, 136806 (2009). [26] H.-M. Guo, G. Rosenberg, G. Refael, and M. Franz, Phys. Rev. Lett. 105, 216601 (2010). [27] H. Shapourian and T. L. Hughes, Phys. Rev. B 93, 075108 (2016).
1201.2982	1	1201	2012-01-14T03:26:50	Non-local spin valve in Van der Pauw cross geometry with four ferromagnetic electrodes	[ "cond-mat.mes-hall" ]	We consider a non-local spin valve in a Van der Pauw cross geometry with four ferromagnetic electrodes. Two antiparallel ferromagnets are used as (charge) source and drain while the detector circuit involves measuring the voltage between two collinear ferromagnets with parallel or antiparallel magnetizations. We find a potentially large increase of the non-local spin voltage. The setup displays several additional interesting properties: (i) infinite GMR for the non-local resistance (if a symmetry requirement for the device is met); (ii) ON-OFF switch effect, when the injector electrodes are parallel instead of antiparallel; (iii) insensitivity to offset voltages. The device can additionally be used as a Direct Spin Hall Effect probe and as a reprogrammable magneto-logic gate implementing basic operations (NOR, NAND, inverter, AND, OR,...).	cond-mat.mes-hall	cond-mat	Non-local spin valve in Van der Pauw cross geometry with four ferromagnetic electrodes. K.-V. Pham∗ Laboratoire de Physique des Solides, Univ. Paris-Sud, CNRS, UMR 8502, F-91405 Orsay Cedex, France Abstract We consider a non-local spin valve in a Van der Pauw cross geometry with four ferromagnetic electrodes. Two antiparallel ferromagnets are used as (charge) source and drain while the de- tector circuit involves measuring the voltage between two collinear ferromagnets with parallel or antiparallel magnetizations. We find a potentially large increase of the non-local spin voltage. The setup displays several additional interesting properties: (i) infinite GMR for the non-local resistance (if a symmetry requirement for the device is met); (ii) ON-OFF switch effect, when the injector electrodes are parallel instead of antiparallel; (iii) insensitivity to offset voltages. The device can additionally be used as a Direct Spin Hall Effect probe and as a reprogrammable magneto-logic gate implementing basic operations (NOR, NAND, inverter, AND, OR, etc). 2 1 0 2 n a J 4 1 ] l l a h - s e m . t a m - d n o c [ 1 v 2 8 9 2 . 1 0 2 1 : v i X r a 1 I. INTRODUCTION. Pure spin manipulation is an important topic in spintronics due to possible applications for programmable logic and memory. Non-local spin valves1 -- 6 provide an example of pure spin current generation. In the latter a pure spin current generated at a ferromagnetic - paramagnetic interface reaches a ferromagnetic probe in the absence of charge current; the spin accumulation created in the probe vicinity can then be detected as a charge voltage by virtue of Johnson-Silsbee charge-spin coupling1,7,8. Such pure spin currents have already proved useful to switch magnetizations9,10. This is opening promises for further applications in logic, sensing and memory devices but to that end it is desirable to increase the signals: the spin voltages are typically in the µV range while mV would be more suitable to ensure sufficient SNR (signal to noise ratio). While spin valves are in everyday use in hard-drives, in order to reach or even go beyond the 1 T bit/inch2 density in hard-drives11 -- 13, novel spin valves are required with a low RA resistance times area product (typically RA ≤ 0.1 Ωµm2)12 -- 14 while sustaining a mA current and mV voltage; this is beyond TMR (tunneling magnetoresistance) sensors capability since they are too resistive. Metallic spin valves are therefore more appropriate. However in order for them to have a suitable SNR it is also necessary that the read-heads function with a large enough contrast12: ∆R A ≥ 5 mΩµm2. Metallic spin valves with larger GMR ratio ∆R/R are therefore required. At first sight non-local metallic spin valves are not obvious candidates for larger MR (magnetoresistance) ratios: they indeed underperform when compared to their local coun- terparts; using the same materials and dimensions a local CP P spin valve is expected to have a larger ∆R A since spin confinement is better15 -- 17. The goal of this paper is to discuss a non-local spintronics device with potentially: • enhanced spin voltage in the mV range for currents ∼ mA (so that the non-local resistance variation ∆Rnl is in the Ohm range) with realistic density currents j < 108 A/cm2 addressing the needs of industry. • enhanced non-local GMR ratio: ∆Rnl/Rnl (up to 100% for the pessimistic ratio; or up to infinity for the optimistic ratio), helping quite generally for better SNR and perhaps making them suitable candidates as sensors or read-heads for hard-drive areal 2 densities larger than 1 T bit/inch2. Regarding the enhancement of the non-local signal, spin valves with tunnel junctions have been reported with non-local resistance in the Ω range but due to a polarization decreasing rapidly when the current is larger than ∼ µA , the spin voltage remains small in the usual µV range18,19. However much progress has been reported recently in pure metallic lateral spin valves (∼ 10 µV for nanopillars10) or lateral valves with very low resistance tunnel junctions (using a thin nm MgO layer)20 -- 22, reaching in the latter case the 100 µV range with RA ∼ 0.2 Ωµm2 so that already ∆R A is of the order of a few 1 mΩµm2. We propose to go even further in the improvement by relying on two ideas: (i) use two injectors instead of a single one, which should at face value double the signal; (ii) enhance the spin confinement by making good use of tunnel barriers, thin enough to stay close to the metallic regime but resistive enough to hinder spin leakage. The idea of minimizing the spin relaxation volume has been expressed in particular in17,23 and explains the large signals seen in spin valves using carbon nanotubes24. Our basic setup applies these ideas by using four collinear ferromagnetic terminals. In a standard lateral spin valve2,3,25 the charge current flows from a ferromagnetic electrode to a paramagnetic drain; the injector electrode is connected by a lateral wire to another ferromagnetic electrode used as a detector. In our setup we propose to replace the param- agnetic drain by a ferromagnet antiparallel to the terminal acting as current source; the two antiparallel electrodes are connected by a paramagnetic metal with thin tunnel barriers in order to better confine spin. The two antiparallel ferromagnets act as spin sources although in terms of charge one is a source and the other a drain: this effectively doubles the spin accumulation in the lateral wire while the tunnel junctions make sure spin is confined. We further change the standard detection setup by using a ferromagnetic counter- electrode instead of a paramagnetic one. The advantage of using two ferromagnets is evidenced when the two detector electrodes are placed symmetrically with respect to the injectors (source and drain): provided they are otherwise identical terminals this implies that when their magnetizations are parallel, their voltage difference should be identically zero by symmetry. This is how we reach an infinite non-local GMR ratio. We will also address the issue of voltage offsets plaguing non-local setups1,2,4,5,15,16: while voltages generated by spin accumulation are clearly observed, some additional voltages of various origins are also usually seen. These offset (or baseline) voltages have been credited to 3 charge current inhomogeneities26 -- 30 (which impact the calculations done for non-local setups since they usually assume one dimensional drift-diffusion equations7,31,32), or to heating (notably Joule and Peltier heating33 -- 35). They may or may not be a nuisance but at any rate they prevent observation of pure non-local voltages. The device we discuss in this paper can be made insensitive to these offset voltages when the two detector electrodes are identical and symmetric since the offsets will cancel out when the voltage difference is measured. This is an additional advantage of our device. The geometry of our device is that of a Van der Pauw cross as in the Jedema and coll. seminal experiments4. A close device within a pure lateral geometry will be discussed elsewhere36. In Section II we introduce the Van der Pauw geometry with four ferromagnetic terminals and give general expressions for the non-local voltage. The basic functionalities of the device are discussed, and notably we will show that the device can perform logic operations (notably as a NOR or NAND gate), be reprogrammed to perform other functions (AND, XOR and inverter gates), displays a potentially interesting ON-OFF switch effect, and when used as a standard 1-bit read-head shows an infinite GMR for the non-local resistance. Use as a Direct Spin Hall Effect probe will also be discussed. The next section III studies in detail the impact of the transparency of interfaces and of the number of ferromagnetic terminals (two or three out of four) on spin confinement, re- sulting in small or large non-local signals. The signals expected are systematically compared to those in the standard lateral geometry. The last section IV discusses the main setup with four ferromagnetic symmetric terminals since it displays the previously mentioned properties of (i) immunity to offset voltages and (ii) infinite GMR ratio for the non-local resistance. Issues pertaining to the use as a sensor are briefly touched upon. The bulk of calculations are relegated to the Appendices. Appendix B revisits the bipolar spin switch calculations by including spin leakage in the measuring electrodes. 4 II. VAN DER PAUW SETUP. A. Geometry and notations. 1. Geometry. We consider in this section a four-terminal device in a Van der Pauw geometry (see Fig. 1). The terminals are ferromagnets F 1 − F 4 positioned as in the Figure 1; we allow for arms of unequal lengths. In sections III-III B we will allow some of these terminals to be paramagnetic through a suitable choice of parameters. One-dimensional assumption. We will assume that width and thickness of all arms are much smaller than their length. Experimentally, current inhomogeneities due to depar- tures from strict one-dimensional flow can arise; however the basic functionalities of our device are for the most part independent of that assumption although quantitative predic- tions may accordingly lose accuracy. Injector electrodes. F 1 injects a charge current which is collected in terminal F 2. We will designate them collectively as injector electrodes; when the need to differentiate them shows up, we will say that F 1 is the source or injector electrode while F 2 is the drain or collector electrode. Detector electrodes. The detection sub-setup consists in terminals F 3 and F 4 hooked to a voltmeter (or a potentiometer or an ammeter). The latter will measure the non-local voltage as a function of the magnetization orientations of each terminal. Orientation. We define points O(x = 0; z = 0) the origin and center of the Van der Pauw cross, A(x = 0; z = L1), B(x = 0; z′ = L2) , C(x = L3; z = 0) and D(x = 0; z′ = L4) where each arm has been for later convenience oriented away from O (axis Ox, Ox′, Oz and Oz′ ) following Jedema and coll.4. The four paramagnetic arms are: I − IV (resp. OA, OB, OC, OD). The charge current flowing through F 1 − I − II − F 2 is Ic and flows from top to bottom (is therefore negative relative to arm I, but positive, relative to arm II). The spin accumulation is defined as: ∆µ = (µ↑ − µ↓) 2e 5 (1) Figure 1. Van der Pauw cross with four ferromagnetic terminals. The arm lengths are respectively L1, L2, L3 and L4. The arrows represent the magnetization direction. When the device is used as a basic spin-valve, only one terminal can switch its magnetization (here F 3). The arms are oriented away from origin O. (where for later convenience we have divided by the electron charge e). The spin currents are oriented away from origin O (therefore are counted positive on a given arm when flowing away from O). 2. Spin parameters. Arms parameters. The central cross is a normal metal. Its parameters are its spin resistance RN = ρ∗ N lN /AN where lN is the spin diffusion length, AN is the cross-section and ρ∗ N is the resistivity. We define the lengths of each arm relative to the spin diffusion length as (for i = 1 − 4): li = . (2) Li lN Ferromagnet parameters. For each ferromagnetic terminal F 1 − F 4 (i = 1 − 4), one defines the conductivity polarization PF,i (= βi in Valet-Fert notation31), spin resistance 6 RF,i = ρ∗ F,ilF,i/AF,i where lF,i is the spin diffusion length, AF,i is the cross-section and ρ∗ F,i = (ρi↑+ρi↓) /4 . (NB: note on terminology; we will call throughout the paper 'spin resistance' the char- acteristic resistance found as the product of the resistivity times the spin diffusion length divided by the cross-section ). Interface parameters. At the interface between the ferromagnets and the paramagnetic arms we assume there is a spin dependent interface resistance so that one can define for each interface Fi − N (i = 1 − 4) a conductance polarization Pci (= γi in Valet-Fert notation), a spin resistance Rci = (Ri↑+Ri↓) /4. For simplicity we will neglect all spin flips at interfaces so that spin relaxation occurs solely in the bulk of the device. 3. Spin resistance mismatch. We define spin resistance mismatch parameters at F - N interfaces as: X = RF + Rc RN . (3) Three limits can be singled out: X < 1: this corresponds to the limit of a transparent junction, which as we will see later in detail (Appendix A 2 e) is very leaky in terms of spin: this favors large spin currents at the cost of reduced spin accumulations in the central paramagnet. X > 1: spin confining or tunneling regime, for which spin accumulation increases but spin current decreases (in magnitude) (see Appendix A 2 e). For X ∼ 10, for which the contact resistance is moderate (about 10 Ω) we will say that we are in the weak tunneling limit. This is the most interesting limit in terms of applications to all-metallic read-heads or sensors. For Rc = 102 − 104 Ω which are usual values in tunnel junctions, X ∼ 102 − 104 which we will qualify as strong tunneling limit. Although the strong tunneling regime can be described by our equations, we will focus primarily in the discussions on the transparent and weak tunneling regime where resistances are in the metallic range which interests us for sensor applications. X = 1: spin impedance matching. The naming for this border situation will be justified below in the discussion on effective spin resistance (section II A 5). 7 4. Effective polarizations. The following definition will also prove useful. We define for each terminal (i = 1− 4) an effective polarization as: where: and: Pef f,i = gP Ri δ+ i gP Ri = (PF iRF i + PciRci) /RN exp li ± The effective polarization can be rewritten as: 2 δ± i = (Xi + 1) (Xi − 1) 2 exp−li. Pef f,i = PF iRF i + PciRci RN sinh li + (RF i + Rci) cosh li ; (4) (5) (6) (7) clearly, Pef f,i ≤ 1. Upon magnetization reversal of the electrode, the effective polarization is an odd function: Pef f,i −→ −Pef f,i. In the limit of short arm length li ≪ 1: Pef f,i −→ PF iRF i + PciRci RF i + Rci (8) which is a weighted average of the electrode bulk and interface polarizations. When li −→ ∞ the effective polarization vanishes exponentially which translates the complete spin relaxation in the arm: Pef f,i = 2 PF iRF i + PciRci RN + (RF i + Rci) exp−li. (9) The effective polarization therefore varies between 0 and its maximum value PF iRF i+PciRci RF i+Rci which is bounded from above by sup (PF i; Pci). 5. Effective spin resistances. We also define an effective spin resistance for each arm of length li (i = 1 − 4) as: 8 Ref f,i(li, Xi) = RN = RN δ+ i δ− i Xi cosh li + sinh li Xi sinh li + cosh li ; (10) (11) Ref f,i(Xi) is an increasing function of spin resistance mismatch Xi. It is shown in Appendix A 2 b that the spin current Is,i(O) at the cross center O on arm i and the spin accumulation there are related by ∆µ(O) = −Ref f,i Is,i(O) (for arms III − IV ; for arms I − II a more general relation taking into account the spin injection at F 1 and F 2 holds). This is an analog of Ohm's law for spin which explains our identification of Ref f,i as a spin resistance. (Note that the analogy is not complete: the relation for spin is a local one (expressed here at point O), while Ohm's law holds for a voltage difference and is therefore non-local. This results of course from the non-conservation of spin current.) We also define a total effective spin resistance for the device: Ref f = . 1 Ref f,i (12) 1 Pi=1−4 The previous expression admits obvious generalization to an arbitrary number n ≥ 4 of arms. As can be seen from its definition, the total effective resistance Ref f is related to the arms spin resistances Ref f,i by the analog of a parallel resistance addition law. We will show later (section II B 2) that the spin voltage is proportional to the total effective spin resistance Ref f so that large effective resistances are desirable; this will also effectively demonstrate for our geometry the parallel addition law for spin resistances. (For a discussion of spin resistance addition law at nodes we refer the reader to37.) Since the total effective spin resistance is the sum of four resistances in parallel, whenever one is much smaller than the others, it will short the other arms: spin leakage will be stronger so that spin accumulation will be reduced. The length dependence of the effective resistance for various values of the spin resistance mismatch X is shown in Fig. 2-3 . At large distance the effective resistance converges exponentially fast to Rn the spin resistance of the paramagnetic arm: Ref f,i −→ RN 9 (13) which reflects the fact that the spin relaxation is dominated by the paramagnet bulk. In the limit li −→ 0 Ref f,i −→ RF i + Rci (14) (which is sensible since the paramagnet is then too short for spin relaxation to occur). When li ≪ 1, the effective resistance remains close to RF i + Rci if the distance l obeys: l (cid:12)(cid:12)X 2 − 1(cid:12)(cid:12) < X. When X ≫ 1, the condition becomes l ≪ 1/X which reflects a steeper exponential decrease. The effective spin resistance for a given arm is therefore comprised between RN and RF,i + Rc,i: RN ≤ Ref f,i ≤ RF i + Rci (15) (or the reverse inequality if RN ≥ RF i + Rci). As a rule the effective spin resistance will be larger for large interface or ferromagnet spin resistance (Rc and RF ); since the ferromagnet spin resistance RF is in general much smaller than the paramagnet spin resistance RN due to short spin diffusion lengths, large interface resistances Rc are required to achieve large effective resistances Ref f,i. For X = 1, one observes that the effective spin resistance does not depend any more on the arm length li and is equal to RN . One then has (on arms III or IV ) ∆µ(O) = −RN Is,i(O) which is the relation one would get from an infinite arm. Everything happens as if the interface had been washed away: this is the reason why we qualified the case X = 1 as corresponding to spin impedance matching in II A 3. In the transparent regime (X < 1), the effective spin resistance is larger at large distance, which is an interesting feature for the design of large non-local circuits. This advantage is circumvented by the exponential decrease of the effective polarization (the non-local resis- tance will be shown to be proportional to both in II B 2) so that in terms of large signals the transparent regime is not interesting, neither in the short-distance nor the large-distance limit. 10 Figure 2. Effective resistance for one terminal (normalized to RN ) as a function of arms length l (relative to paramagnet spin relaxation length) for various impedance mismatches X = 1 − 10 in the (weak) tunneling regime. B. General expression of the spin voltage in Van der Pauw geometry. 1. Spin accumulation at cross center. Solving the one-dimensional drift-diffusion equations (see Appendix A for details of the calculations) leads to the following results. The spin accumulation in the center of the cross is found as: ∆µ(O) = (Pef f,1 − Pef f,2) Ref f Ic We can define a total effective polarization at injector electrodes F 1 and F 2: Pinjector = Pef f,1 − Pef f,2. (16) (17) The minus sign in front of Pef f,2 comes from the fact that F 2 is a charge drain: in terms of spin accumulation it therefore acts contrariwise to electrode F 1. This also shows that to ensure maximum signal it is better to have opposite orientations for F 1 and F 2 (antiparallel 11 Figure 3. Effective resistance for one terminal (normalized to RN ) as a function of arms length l (relative to paramagnet spin relaxation length) for various impedance mismatches X = 0.1 − 1 in the transparent regime (X < 1). injector electrodes). This is easy to understand: when injector electrodes are parallel, F 2 acts as a spin sink for the spins injected by F 1; therefore the spin accumulation should decrease. But when F 2 is antiparallel to F 1, spin leakage is frustrated; although F 2 is a charge drain, it acts as an additional spin source. It is noteworthy that the spin accumulation does not depend on the magnetization orien- tations of detector electrodes F 3 and F 4 (although it does depend on its parameters through Ref f ). This is sensible: spin injection is ensured by F 1 and F 2 not F 3 and F 4. The general structure of the spin accumulation in terms of F 1 and F 2 relative orientation can be understood simply. Suppose we flip all spins of the setup. Then: ∆µ −→ −∆µ since we have exchanged spin up and spin down electrons. This implies that the spin accumulation must be an odd function of electrode polarizations. This is easily checked on Eq. (16): when both F 1 and F 2 are flipped, their effective polarizations get reversed Pef f −→ −Pef f and therefore the spin accumulation is reversed. Another way to reach the same conclusion is to notice that the physics of the setup should be invariant when both 12 current and magnetizations are reversed. Flipping all spins or reversing the charge current should lead to the same spin accumulation, namely a reversed one. An important consequence of the structure of Eq. (16) is that the spin accumulation at the cross center vanishes when F 1 and F 2 are identical and parallel so that Pef f,1 = Pef f,2. This follows clearly from symmetry: when F 1 and F 2 are symmetric with respect to the cross center and have identical parameters, the spin accumulation on the line F 1−F 2 should be antisymmetric and the spin current should be symmetric when F 1 and F 2 are parallel (the opposite when they are antiparallel). This can be understood by reversing the current: if one reverses the current, on the one hand, the spin accumulation at the cross center should not change since F 1 and F 2 are identical and parallel (by watching Fig. 1 after a π rotation); on the other hand, reversing the current must reverse the spin accumulation since reversing the current is equivalent to reversing all spins. The only way out is for the spin accumulation at the cross center to vanish. This result does not depend on the assumption of one-dimensional flow and follows directly from symmetry. We will see later that this enables an ON-OFF switch function onto the device. 2. Spin voltage and non-local resistance. Spin voltage. The spin voltage (or non-local voltage) is the voltage drop between terminals F 3 and F 4: Vnl = − [µF 3(+∞) − µF 4(+∞)] /e. Straightforward calculations (see Appendix A 2 d) lead to: Vnl = −∆µ(O) (Pef f,3 − Pef f,4) . (18) (19) The behaviour under magnetization reversal is easy to understand: when the magnetiza- tions of both detector electrodes are switched, their coupling to the spin accumulation gets reversed so that the non-local spin voltage should change sign. The spin voltage is therefore odd under magnetization switching of both detector electrodes. Inserting the expression of the spin accumulation in Eq. (16): Vnl = (Pef f,1 − Pef f,2) (Pef f,3 − Pef f,4) Ref f Ic (20) This expression factors out neatly in three contributions: 13 (i) the geometry dependent effective spin resistance Ref f =(cid:0)Pi=1−4 R−1 (ii) a total effective polarization for injector and collector electrodes (F 1 and F 2) ef f,i(cid:1)−1; Pinjector = Pef f,1 − Pef f,2; (iii) and a total effective polarization for the two detector electrodes (F 3 and F 4): Pdetector = Pef f,3 − Pef f,4. (21) (22) Non-local resistance. The non-local resistance (sometimes called a transresistance) is defined as the ratio of the non-local voltage to the charge current flowing through the injector electrode F 1 to the collector electrode F 2 : Rnl = Vnl Ic = Ref f Pef f,injector Pef f,detector. (23) To achieve a large signal it is therefore necessary to have a large total effective spin resistance for the cross Ref f and to have on the one hand antiparallel source and drain terminals, on the other hand antiparallel detector electrodes. Ref f is largest when all arms effective spin resistances are also large, which is the case if spin resistance mismatches are in the tunneling regime according to the discussion in II A 5 and if the arms length is short enough. Two situations may arise: i) one or several spin resistance mismatches are in the transparent regime (X ≤ 1). Then Ref f ∼ RN or smaller. No enhancement of the spin voltage is to be expected when compared with the usual lateral setup. We will say that we have an open geometry23 which leaks spin (larger spin currents but smaller spin accumulations). ii) all spin resistance mismatches are in the spin confining regime (X > 1). Then Ref f ∼ Rc in the limit of short length for the arms (and assuming for simplicity mismatches roughly equal Xi ∼ X ∼ Rc/RN . In such a geometry which will be qualified as closed23 the signal is therefore potentially much larger than in an open geometry. Let us compare the non-local resistance to local resistances in the same device. i) Firstly a local resistance can be measured between source and drain (F 1 and F 2); as shown in Appendix A 3 a: Rlocal,12 = Ref f [Pef f,1 − Pef f,2]2 + R0 {P1−2} (24) 14 where R0 {P1−2} = Xi=1−2 ρ∗ F i (cid:0)1 − P 2 F i(cid:1) zi + ρ∗ N li +Rci+ F iRciRF i − P 2 [P 2 ciR2 RF i + Rci ci − 2PF iPciRF iRci] . (25) (zi are the locations of probes in the electrodes, see Appendix A 3 a). In general Rlocal,12 ≈ R0 since the GMR effect is a few percents; the non-local resistance which is commensurate with Rlocal,12 − R0 is therefore much smaller than Rlocal,12. It is more meaningful to compare the variations upon magnetization switching ∆R = RAP − RP ; for the non-local signal, we have set in the following F 1 and F 2 antiparallel while switching F 4 magnetization: ∆Rnl ∆Rlocal,12 = (Pef f,1 + Pef f,2) Pef f,4 2Pef f,1 Pef f,2 (26) (where for Rlocal, either terminal F 1 or F 2 have been switched). In the case of identical source and drain electrodes (F 1 and F 2) this reduces to: ∆Rnl ∆Rlocal,12 = Pef f,4 Pef f,1 ; (27) if the ratio is larger than unity (Pef f,4 > Pef f,1) there is an amplification of non-local MR variation versus local MR (or the converse if Pef f,4 > Pef f,1). ii) If we then compare to the local resistance found when current flows from F 3 to F 4, one gets instead: ∆Rnl ∆Rlocal,34 = (Pef f,1 + Pef f,2) 2Pef f,3 ; One can again get an MR amplification (or reduction if Pef f,3 is sufficiently large). But if all electrodes are identical, all these MR ratios are then equal to unity ∆Rnl ∆Rlocal = 1 (28) (29) which means the non-local measurement performs equally well as local measurements in terms of raw resistance variation, with one proviso: non-local resistances have smaller base- lines. Indeed for Rnl, the baseline or smallest signal is: Rnl,min = Ref f (Pef f,1 − Pef f,2) (Pef f,3 − Pef f,4) while for Rlocal,12: Rlocal,min = R0 {P1−2} + Ref f (Pef f,1 − Pef f,2)2 15 which is larger on account of the R0 term. iii) It is worthwhile to compare to CPP GMR for a spin valve with similar dimensions: ∆RCP P = 4RN (PcRc+PF RF )2 2RN (RF +Rc) cosh l+[(RF +Rc)2+R2 N] sinh l for identical and infinite ferromagnetic layers separated by the same distance L = l lN . In terms of X = (RF + Rc) /RN this can be recast as: ∆RCP P = 8R−1 N (PcRc + PF RF )2 (X + 1)2 exp l − (X − 1)2 exp −l so that for a non-local device with identical terminals: ∆Rnl ∆RCP P = 1 2 . (30) The non-local signal is smaller by a factor 2 which is easy to understand: in the cross geometry the spin relaxation volume is doubled when compared with a spin valve with a paramagnetic layer of identical length because of the side arms. A systematic comparison of the non-local resistance in the cross geometry with the stan- dard lateral geometry is left to sections III A-IV. 3. Non-local charge current. The voltage between the detector electrodes F 3 and F 4 actually acts as an electromotive force (emf) of magnetic origin; when F 3 and F 4 are shorted, a charge current therefore appears. This non-local charge current is induced by the spin accumulation generated by the remote source and drain F 1 and F 2. Note that since the spin voltage is an electromotive force, it can be measured either through a (nano-)voltmeter or with a potentiometer: in the latter case, the advantage is that there is no current at all during the measurement. If however it proves advantageous that the signal be a current (for chaining the non-local device to a bipolar transistor for instance rather than a MOSFET), the non-local current is found (see Appendix A 3 b) as: where Rlocal,34 is the local resistance measured when one drives a current between F 3 and Inl = − Rnl Rlocal,34 Ic (31) F 4. This can be rewritten as: Inl = − Ref f (Pef f,1 − Pef f,2) (Pef f,3 − Pef f,4) Ref f [Pef f,3 − Pef f,4]2 + R0 {P3−4} Ic (32) 16 where: R0 {P3−4} = Xi=3−4 ρ∗ F i (cid:0)1 − P 2 F i(cid:1) zi + ρ∗ N li + Rci [P 2 F iRciRF i − P 2 + ciR2 RF i + Rci ci − 2PF iPciRF iRci] . (33) (it can be checked that R0 ≥ 0); zi (i = 3 − 4) are the locations on F 3 and F 4 of the wires which short them together (in the following we have chosen to focus on Rnl rather than Inl). C. Main properties of the device. The device functionalities depend on the terminals for which the magnetization has been fixed (prevented from switching); additional symmetry requirements can also add properties. We first discuss functionalities pertaining to sensing or data storage. As explained in the introduction the non-local resistance and spin voltage are contami- nated by offsets, the origin of which is still under debate (current inhomogeneities; thermal origin). This may adversely affect the measured signals and in the following we will take care to indicate the potential impact of these offsets. 1-bit reading. This is the basic functionality of the spin valve as a sensor. Suppose the orientations of all terminals but one (say F 3) are pinned (for definiteness we assume: F 1 :↑, F 2 :↓ and F 4 :↓ ) so that F 3 acts as a sensing electrode. We make no special assump- tions on the terminals (later on some conditions will be imposed for further functionalities). One recovers a spin-valve behaviour (see Fig. 4) with two distinct values of the non-local resistance which uniquely determine the orientation of terminal F 3: RP = Ref f (Pef f,1 + Pef f,2) (Pef f,3 − Pef f,4) RAP = Ref f (Pef f,1 + Pef f,2) (Pef f,3 + Pef f,4) (the P/AP index refer to F 3 magnetization orientation relative to F 4). The spin valve can therefore be used to read a 1-bit information. In the standard non-local spin valve and if we neglect voltage offsets, the spin voltage changes sign when one terminal is flipped so that for antiparallel and parallel alignment RAP = −RP (this is recovered here in the limit Pef f,4 = 0, when F 4 is a paramagnet); this is not the case here although depending on the relative values of Pef f,3 and Pef f,4 there can still be a change of sign (Fig. 4-b, when Pef f,3 > Pef f,4). 17 Figure 4. Non-local resistance variation when the orientation of a single terminal (here F 3) is switched by an external magnetic field (here H3). All other terminals are pinned. Injector F 1 and collector F 2 are antiparallel with orientation as shown on the graph. (a) case Pef f,3 < Pef f,4: standard spin valve effect (1-bit reading); (b) case Pef f,3 > Pef f,4: spin valve effect with spin voltage change of sign.). For the Van der Pauw cross under the most general conditions the difference with the standard non-local setup is therefore minor; it remains to see if larger signals can be achieved. This will be the topic of sections III-IV where we will show that tunnel contacts at the four terminals can greatly enhance the non-local resistance. Furthermore this functionality is clearly affected by offset voltages which will shift the signals and change the GMR ratios (adversely if the offset is positive). We now discuss a simple way to circumvent these offsets. Offset free 1-bit reading with infinite non-local GMR. The offset issue is easily fixed if we assume F 3 and F 4 are identical electrodes placed symmetrically with respect to the rest of the setup. Due to symmetry offset voltages are neutralized (assuming offset voltages are spin independent) since they will shift both voltages V3 and V4 in the same manner so that the spin voltage Vnl = V3 − V4 is free from offsets. Additionally: RP = Ref f (Pef f,1 + Pef f,2) (Pef f,3 − Pef f,4) = 0 in parallel alignment 18 since Pef f,3 = Pef f,4 while Rnl 6= 0 in antiparallel alignment (we have assumed that F 1 and F 2 are antiparallel and pinned). One has achieved an infinite GMR for the non-local resistance since the ratio GMR = (RAP − RP ) /RP −→ ∞ (using the optimistic ratio; the pessimistic ratio would be 100 %). There is an intrinsic contrast which is protected by symmetry from the voltage offsets. Such a maximized MR ratio is clearly helpful for SNR in terms of Johnson noise (or shot noise in the strong tunneling limit) since the latter scales as ∆R/√R; indeed if we compare with a CPP spin valve with a ∆R/R = 1 − 10 % GMR ratio, this would imply at identical ∆R (∆R = ∆Rlocal = ∆Rnl) an increase of SNR by 10− 20 dB. (Indeed: SNR for non-local device would be ∝ 1/√∆R so that SNRnon−local/SNRCP P ∝pRlocal/∆R). The difference is quite significant given that under operation one expects in general at least 30 dB SNR. In terms of geometry requirements, note that the symmetry between detector electrodes is required only on the scale of a few spin relaxation lengths lF (on the ferromagnet side) since the spin accumulation is washed at larger distance. The property is also clearly independent of the precise geometric arrangement, does not depend on the assumption of one-dimensional flow and will be valid for other geometries than the cross studied in this paper, provided the two detectors are arranged symmetrically with respect to the injectors. Experimentally this is very useful since this gives a lot of flexibility in terms of design. 3-bit reading or storage. For that function one terminal is pinned, while the other three are free and play the role of input signals (the non-local resistance measured between F 3 and F 4 is as previously the output signal). Eq. (23) shows the non-local resistance can assume 8 different values when the electrodes orientation are changed and the maximum value for Rnl is reached when on the one hand F 1 and F 2 are antiparallel, and on the other hand F 3 and F 4 are also antiparallel. The orientations of the three non-pinned terminals are uniquely determined by the 8 distinct values of the spin voltage. This implies that the device encodes 3 bits in principle (3 bit spin valve). Fig. 5 shows the eight outputs signals as a function of the input variables (for illustrative purposes F 3 is varied by an external field H3 on the graphs, showing the output variation with the change of one input). The property survives offset voltages which come as an additive contribution to the voltage. However resolution may be adversely affected by offsets. Note that although there 19 Figure 5. Non-local resistance variation as a function of terminals magnetization orientation showing 3-bit sensor or storage. For illustrative purposes F 3 magnetization is varied in each figure (a-d) (through an applied field H 3) to show the output signal variation with the change of one input. (a) Injector F 1 and collector F 2 are antiparallel. (b) They are switched to parallel: the signal collapses (to some extent). (c) Signal is reversed when injector and collector are both switched from antiparallel alignment in (a). (d) Injector and Collector are parallel (but with directions opposite to (b) ). are four ferromagnets and a priori 4 bits could be stored, the spin voltage can assume only 8 values, not 24 = 16. This is because when all spins are reversed, the spin voltage is unchanged in Eq. (23). Offset-free 2-bit reading. By the same token applied previously to detectors for the 1-bit reading it is possible to convert the 3-bit reading function into a protected 2-bit reading function: suppose that detector terminals F 3 and F 4 are symmetric (namely, have identical parameters and are placed symmetrically with respect to the device) but that F 1 and F 2 differ (Pef f,1 6= Pef f,2). Let us pin terminals 3 and 4 in antiparallel orientation (F 3 :↑ and F 4 :↓), while only F 1 and F 2 are allowed to switch magnetizations. The non-local spin voltage then takes 4 distinct values depending on the orientations of terminals 1 and 2: 20 Rnl = Ref f (Pef f,3 + Pef f,4) (±Pef f,1 ± Pef f,2). These values uniquely determine the 2-bit state of F 1 and F 2. The main advantage when compared with the 3-bit function discussed previously is that since terminals F 3 and F 4 are symmetric, there can be no offset voltages since they automatically cancel out when measuring the voltage drop between F 3 and F 4 (if offset voltages are spin-independent; if it is not the case, they will add up to the spin voltage). Note that if F 1 and F 2 are pinned while F 3 and F 4 are not pinned but identical (to avoid offsets), the spin voltage only assumes three values since RP = 0; to store 2 bits without offsets it is therefore necessary to pin F 3 and F 4, not the other way around. ON-OFF switch for the 1-bit read-out or storage function. When source F 1 and drain F 2 are identical ferromagnets (same distance from origin, same conductivities, polarizations, interface resistances, spin diffusion length), Pef f,1 = Pef f,2. This then implies that when F 1 and F 2 are parallel, the spin accumulation at the cross center vanishes ∆µ(O) = 0 (34) so that whatever the orientation of F 3 and F 4, RP = 0 = RAP . The spin voltage has been killed and we have disabled the read-head or 1-bit storage. The property is clearly interesting in terms of logic if the device is chained to another device for instance a MOSFET whose gate is controlled by the spin voltage (after suitable amplification). That property should survive offset voltages since symmetry protects it. But one might wonder if the spin voltage measured at F 3 and F 4 will still vanish if we take into account departures from strict one-dimensional charge and spin flow. It is clear indeed that even if we take into account the 3-dimensional nature of the device but remain in a quasi-one-dimensional approximation, the spin accumulation on the side arms (zones III − IV ) will remain small (it may be non zero at edges) and will never diffuse as far as F 3 and F 4 provided the width of each arm is much smaller than its length (w ≪ l): indeed when F 1 and F 2 are parallel, the spin accumulation is an odd function of z in the direction F 1− F 2; therefore, on the side arms (III − IV ), there will be some spilling of spin accumulation with opposite signs on opposite edges, close to the origin O. But if w ≪ l, spin diffusion will mix these opposite spin accumulations which will cancel out when one reaches the detector terminals (F 3 and F 4). The spin voltage measured at F 3 and F 4 should therefore still vanish. Direct Spin Hall Effect probe. The geometry lends itself easily to probing the Spin 21 Hall Effect6,38,39. Imagine F 1 and F 2 are normal electrodes and that the magnetizations of F 3 and F 4 are perpendicular to the plane of the cross. On the arm AB (F 1−I−O−II−F 2 or zOz′) a spin accumulation may appear due to Hall coupling αH = σxx/σxy on the width w (for −w/2 ≤ x ≤ w/2): αH ∆V L1 + L2 ∆µ(x, z) = x. (35) When we reach the cross center, there can be a spilling of charge current lines to the side arms, but we will neglect this effect by assuming that the arms width are much smaller than their lengths (w ≪ Li for i = 1 − 4). When we move on the side arms (III − IV ), the spin accumulation decreases in magnitude. This decrease can be estimated using drift-diffusion equations as: ∆µIII(C) = X3 δ+ 3 ∆µ(x = w 2 , z = 0) [see Appendix A 2 d, Eq. (A48)] with a similar expression for arm IV : ∆µIV (D) = X4 δ+ 4 ∆µ(x = − w 2 , z = 0) ∆µIII(C) can be re-expressed as: and: ∆µIII(C) = X3 δ+ 3 αH ∆V L1 + L2 w 2 ∆µIV (D) = − X4 δ+ 4 αH ∆V L1 + L2 w 2 . The spin voltage is therefore (see Appendix A 2 d): Vnl = − (Pef f,3 + Pef f,4) ∆µ(x = αH ∆V L1 + L2 = − (Pef f,3 + Pef f,4) w 2 w 2 ) . (36) (37) (38) (39) (40) (41) Note that for identical and antiparallel electrodes F 3 and F 4, the signal therefore vanishes since Pef f,3 = −Pef f,4 and that for maximal non-local signal, parallel magnetizations are required. Let us go back to our initial setup with four ferromagnetic electrodes. If magnetizations are perpendicular to the plane of the Van der Pauw cross, the spin Hall effect will add up to the non-local spin voltage arising from spin injection. But if we choose a symmetric setup with identical parallel electrodes F 3 and F 4, offsets cancel out in Vnl = V3 − V4 and VP = 0 22 as explained previously. Therefore the only remaining signal is that of the Direct Spin Hall Effect. This provides an interesting all-electrical alternative to the observation of Direct Spin Hall Effect, which initially was observed optically in semiconductors40,41 although later on some electrical detection schemes have been used in metallic systems to investigate both Direct and Inverse Spin Hall Effect42 -- 44. D. Implementation of magneto-logic gates. Many proposals exist in the literature for the use of magnetoelectronics circuits as logic gates45; this has prompted a lot of activity in the field of semiconductor spintronics since integration to existing processes would be optimal46. Our device is metallic and therefore not the best candidate as a spin transistor since there is no amplification: this renders the chaining of gates more delicate for instance (unless one uses hybrid designs combining pure magnetoelectronics devices with conventional transistors). However, the device still possesses obvious capability as a programmable magneto-logic gate as we now demonstrate. Let us associate bit 0 with down ↓ magnetization and bit 1 with up ↑ magnetiza- tion. F 1 and F 2 are assumed to be identical; F 3 and F 4 are also identical (same arm length, same polarizations, spin diffusion length, etc); only the magnetization orientations are allowed to differ. The non-local resistance can then assume only three values: 0 and ±R0 = ±4Ref f Pef f,1 Pef f,3. For logic operations we will consider two conventions: (i) associate bit 0 to zero resistance, and bit 1 to Rnl = +R0. (ii) or (opposite convention) associate bit 1 to zero resistance, and bit 0 to Rnl = +R0. Note that in what follows we have discarded all configurations for which Rnl = −R0 to avoid ambiguities in the bit association and keep only those for which there are only two possible outputs: 0 and R0. For the first convention assigning the bit content of Rnl, the expression for Rnl [see Eq. (23)] can therefore be rewritten in terms of bits as the Boolean equation: (F 1 − F 2) (F 3 − F 4) = Rnl where to simplify notation we have conflated Rnl and its bit content. With the opposite convention, (F 1 − F 2) (F 3 − F 4) = Rnl 23 (42) (43) Note that in order to have well defined HIGH and LOW states the associated voltages (or non-local resistances) must be sufficiently different which is not ensured in the presence of offset voltages. That's why we choose symmetric terminals to get rid of these offsets. We first try to reproduce basic binary Boolean functions. We need therefore to pin two terminals; the other two terminals will represent the variables treated by the device in the following manner (for instance): imagine each terminal is screened to prevent any magnetic field applied to a given electrode to influence any other electrode; we then apply external magnetic fields to either of the two non-pinned terminals but not to the other twos, whose magnetizations are therefore fixed during the whole operation. NOR gate. We choose convention of Eq. (42) for bit coding. We pin F 2 and F 4 (F 2 :↑ and F 4 :↑) so that F 2 = 1 and F 4 = 1 in bit terms. To have a non-zero resistance among the four possible bit configurations of F 1 and F 3, only F 1 = 0 and F 3 = 0 are allowed. The non-local resistance is then Rnl = +R0 which we associate with HIGH state or bit 1: this means that F 1 F 3 = Rnl; this can also be recovered through algebra by using Eq. (42) which in our case is: so that: (F 1 − 1) (F 3 − 1) = Rnl F 1 F 3 = Rnl. (44) (45) This is precisely a NOR gate since F 1 F 3 = F 1 + F 3 by De Morgan theorem. This is a very important property since NOR has functional completeness: any Boolean function can be implemented by using a combination of NOR gates (only NAND possesses the same property). OR gate. In the same configuration as for the NOR gate, if LOW state (bit 0) is now associated with Rnl = +R0 and HIGH state (bit 1) to Rnl = 0 [convention of Eq. (43)], one obviously gets an OR gate. Of course since the association of HIGH and LOW has been reversed with the previous case, the two settings are incompatible since they correspond to opposite bit assignment. AND gate. We still pin F 2 and F 4 (F 2 :↓ and F 4 :↓) so that F 2 = 0 and F 4 = 0 in bit terms. Using Eq. (42) this implies F 1 F 3 = Rnl which is an AND operation (using the first convention for Rnl bit content). This gate is compatible with the NOR gate but not the OR gate which is produced with a different convention for bit coding. 24 NAND gate. By changing the bit coding of Rnl, the AND gate turns into a NAND gate with the same configuration for F 2 and F 4. As mentioned previously, the achievement of a NAND is quit noteworthy since it has functional completeness in Boolean algebra. AB or A : B (A not implied by B) gate. We now pin F 2 and F 3 and choose F 2 = 1 and F 3 = 0 in bit terms. Then Eq. (42) (first convention) turns into F 1 F 4 = Rnl. A + B or A ⇐ B gate. Using the same configuration for F 2 and F 3 as previous gate but exchanging the conventions for the bit content for Rnl leads to an inverted gate: indeed F 1 F 4 = Rnl implies F 1 + F 4 = Rnl by De Morgan theorem. AB or A ; B gate. We pin F 2 and F 3 and choose F 2 = 0 and F 3 = 1 in bit terms. Then Eq. (42) (first convention) turns into F 1 F 4 = Rnl. A + B or A ⇒ B gate. Choosing the second convention of Eq. (43) with F 2 = 0 and F 3 = 1 leads to F 1 F 4 = Rnl which is equivalent to F 1 + F 4 = Rnl by De Morgan theorem. FALSE gate. We pin F 1 and F 2 in parallel configuration. Then Rnl = 0 (in terms of resistance not bit) whatever the configuration of F 3 and F 4. If we adopt the first convention, the zero resistance translates into bit 0. This is therefore a FALSE gate. TRUE gate. In the same terminals configuration as previous gates, if we adopt the opposite convention for the bit content of Rnl, one then gets a TRUE function. The six other possible binary operations (out of sixteen) can not be built as easily using a single cross device; this leaves out the NOR and XNOR gates from the list of the basic gates in use in electronic logic (AND, OR, NOR, NAND, XOR, XNOR, buffer, inverter). A way to achieve them would be to chain gates since all Boolean functions can be recovered using only a NOR (or NAND) gate. We now turn to the two other basic gates, buffer and inverter. Buffer. For that function only one terminal is not pinned. We choose to pin F 1 = 1, F 2 = 0 and F 4 = 0. Using Eq. (42) this implies F 3 = Rnl. This can also be realized by pinning F 1 = 0, F 2 = 1 and F 4 = 1. Using Eq. (43) this implies F 3 = Rnl. Inverter. When pinning F 1 = 1, F 2 = 0 and F 4 = 0, using Eq. (43) implies F 3 = Rnl. This can also be realized with Eq. (42) by pinning F 1 = 0, F 2 = 1 and F 4 = 1 which leads to F 3 = Rnl. The NOR, AND, A : B, A ; B , FALSE, buffer and inverter gates use the same bit convention for Rnl (HIGH state for Rnl = +R0, LOW state for Rnl = 0) and can be implemented together in the same circuit; since they differ only by the assignment of pinned 25 terminals, this means that one can easily turn a gate into another one (reprogrammable logic) through local application of an external magnetic field or through spin transfer torque. In summary, the non-local cross with four ferromagnetic terminals is a versatile spintronics device which can be used either as a standard spin valve: when some symmetry requirements are met, it displays the important property of an infinite GMR for the non-local resistance; it can be immunized against offset voltages observed in many non-local setups. Finally, it can be used as a magneto-logic gate as well as probe the Direct Spin Hall Effect. III. VAN DER PAUW CROSS WITH TWO OR THREE FERROMAGNETS. Which principles should guide us in order to achieve large signals in non-local setups? As stressed time and again in the literature15,17,23, they are quite simple: larger spin accu- mulations in the paramagnet can be generated if (i) the paramagnet volume in which spins can relax is small (in comparison with l3 N ) and if (ii) spin back-flow to the ferromagnets is hindered by large enough interface resistances. One can classify geometries as open or closed according to the (non-)fulfillment of these prescriptions23: it has been argued in the latter reference that at small enough volume and for large enough tunnel barriers the spin accumulation can be greatly enhanced because it then scales with the (large) interface resistances Rc. But whenever the geometry is open, the spin accumulation scales with the much smaller spin resistance of the paramagnetic channel RN which in practice is often in the Ohm range. Although prescriptions (i) and (ii) have been stressed repeatedly, spin leakage in the current or voltage probes is often overlooked although they may alter significantly the signal (an example is provided in Appendix B). The goal of this section is to examine the transition from open to closed geometries as a strategy to enhance the non-local signal used for instance to read or sense 1 bit, and to study its interaction with the doubling of spin injector electrodes. The signal generated in our cross geometry will be systematically compared with the spin voltage in the standard lateral device. The following general features will be a guide: a large signal requires a large total effective spin resistance for the device Ref f . Since the latter can be interpreted as resulting from the addition of four spin resistances Ref f,i (i = 1 − 4) in parallel corresponding to the four arms of the cross, one will expect a small signal whenever one or several of these spin resistances 26 are significantly smaller than the others since they will short the other arms. If all effective spin resistances Ref f,i are commensurate, in order to have a large Ref f,i, it is then necessary that Rc,i ≫ RN simultaneously for all four terminals as explained in section II B 2. (As an aside remark we wish to bring to the reader's attention that the expressions derived in the literature47,48 for the bipolar spin switch transistor of Johnson remarkably show the enhancements characteristic of closed geometries over open geometries with a typical scaling with interface resistance Rc at small enough distance between the ferromagnetic terminals. As discussed in Appendix B this is due to neglecting spin leakage to the current drain; when this leakage is taken into account, the signal is actually found to scale with RN as in open geometries.) A. Van der Pauw cross with two ferromagnets. 1. Open geometry with two ferromagnets. For the sake of comparison we first consider the case when the detector electrodes are paramagnets as in the cross arms I − IV . When F 1 and F 3 are identical ferromagnets at the same distance of origin (l1 = l3) while F 2 and F 4 are identical paramagnets with spin resistance RN , the spin resistance mismatches at each terminal are X1 = X3 = X; X2 = X4 = 1 while: gP R1 = gP R3 =gP R gP R2 = gP R4 = 0 and defining δ± F for arms I and III: δ± F = δ± 1/3 = (X + 1) 2 exp l1 ± (X − 1) 2 exp−l1. δ± 2/4 = exp l2/4 27 (46) (47) (48) (49) (50) so that: Rnl = = F δ+ σ1σ3 RN (cid:16)gP R(cid:17)2 F(cid:1)2i 2hδ− F +(cid:0)δ+ σ1σ3 RN (cid:16)gP R(cid:17)2 (cid:2)(1 + X)2 exp l + (X 2 − 1)(cid:3) where l = 2L1/lN the total length separating the ferromagnetic terminals (in units of the spin diffusion length in the paramagnet) and where we have defined σ1 = ±1, σ3 = ±1 to index the majority spin directions of electrodes F 1 and F 3 (relative to an absolute axis). In the standard geometry, since one has a spin valve it is customary to quote the resistance variation when one ferromagnet magnetization is switched; this is twice the maximum value RP : (51) (52) δRnl,0 = RP − RAP = 2 RN (cid:16)gP R(cid:17)2 (cid:2)(1 + X)2 exp l + (X 2 − 1)(cid:3). This generalizes the expression found in the literature for the non-local resistance variation: more precisely the result quoted by Jedema and coll.4 corresponds to the case of vanishing interface resistance, so that X = RF /RN (= 1/M using Jedema and coll. notations4). Although the geometry is open (in the sense that spin current can leak easily since terminals F 2/F 4 do not hinder its flow (X2 = X4 = 1) ), it is interesting to observe that the larger the resistance mismatch X at the ferromagnets, the larger the signal (for instance, if we set P = 1 in Eq. (52) one gets that δRnl,0 ∝ X 2/(cid:2)(1 + X)2 exp l + (X 2 − 1)(cid:3) which is an increasing function of X). This means that large resistance mismatches are already beneficial and increase the spin accumulation although there is some spin leakage. The largest value is at short distance when the denominator of Eq. (52) is 2X (X + 1). In the tunneling regime X ≫ 1, δRnl,0 ∝ RN . In the opposite limit X ≪ 1, the signal will be even smaller since it scales as Rc + RF ≪ RN : and δRnl,0(l −→ 0) ≥ inf (Pc, PF )2 (Rc + RF ) δRnl,0(l −→ 0) ≤ sup (Pc, PF )2 (Rc + RF ) . Such a scaling which is at most ∼ RN or even below (≪ RN ) is characteristic of open geometries. 28 Let us compare to the non-local resistance variation for the lateral geometry with iden- tical parameters (same distance l between injector and detector, same spin resistance mis- match X at ferromagnets, same cross-section for the paramagnet channel connecting the ferromagnets)6: We plot on Fig. 6 the ratio of the cross signal versus the one in the lateral geometry: δRnl,lateral = 2 4RngP R (cid:2)(2X + 1)2 exp l − exp −l(cid:3) . = (cid:2)(2X + 1)2 exp l − exp −l(cid:3) 2(cid:2)(1 + X)2 exp l + (X 2 − 1)(cid:3) . (53) (54) m(X, l) = δRnl,0 δRnl,lateral In the short distance limit: m(X, l −→ 0) −→ 1 which is expected since the cross and the lateral geometries are then identical. At large distance, the signal is larger in the cross geometry whenever X > 1/√2 (which includes the tunneling regime at the ferromagnets and also part of the transparent regime). The impact of a large value of X is moderate (about 10 % between X = 10 − 100) in stark contrast to what we will observe in a closed geometry. The general behaviour is easy to understand: in the tunneling regime spin current leakage is less pronounced in the cross geometry since the paramagnetic drain is further away (while the detector ferromagnet and paramagnetic counter-electrodes are at the same distance). In the transparent regime, by a similar reasoning one gets the opposite. 2. Spin confining geometry with two ferromagnets. It has been argued by Jaffres and coll.23 that spin confinement tends to increase non-local signals (resistance or voltage) since spin accumulation is stronger whenever spin leaking is hindered by large tunnel barriers. We can study this by considering that electrodes F 2 and F 4 are normal paramagnets but that there is a resistance mismatch Y = RN,0 + Rc,0 RN 29 (55) Figure 6. Non-local resistance in cross setup compared with lateral setup. l is the distance in units of spin diffusion length between injector and detector in either setup. In tunneling regime (X > 1) and in part of the transparent regime, the signal is larger in the cross geometry. where RN,0 and Rc,0 are the spin resistance of F 2 and F 4 and the interface resistance between them and the central cross. One can therefore interpolate between an open geometry (the standard geometry, i.e. Y = 1) and a closed geometry (Y ≫ 1 and X ≫ 1). We will now study the transition from one regime to the other. Let us assume that terminals F 1 and F 3 are identical ferromagnets at identical distance l1 = l3 = l0 of the cross center O, that F 2 and F 4 are identical paramagnets also at the same distance l2 = l4 = l0 from O (ABCD is then a square); the total distance from the injector F 1 to the detector F 3 is l = 2l0. We also set: and: X1 = X3 = X; X2 = X4 = Y gP R1 = gP R3 =gP R, gP R2 = gP R4 = 0. 30 (56) (57) (58) There are several obvious ways to create this spin resistance mismatch Y : one is to deposit a tunnel barrier between the terminals F 2 and F 4 and the central cross (Rc,0 6= 0); another is to use the same paramagnet for both F 2, F 4 and the central cross but have a different cross-section (Rc,0 = 0 but RN,0 = ρ∗ N lN /AN,0 6= RN = ρ∗ N lN /AN ). One ends up with: Rnl = where σ1σ3 RN (cid:16)gP R(cid:17)2 F(cid:1)2i 2hδ− N(cid:0)δ+ F + δ− N δ+ F δ+ δ± F = δ± 1/3, δ± N = δ± 2/4. [The definitions for δ± i are given in Eq. (6).] Let us define ∆2F (X, Y, l) = (59) (60) (61) (62) and so that: 2 [(Y + 1) exp l0 + (Y − 1) exp −l0]−1 {(Y + 1) exp l0 (cid:2)(X + 1)2 exp 2l0 + (X 2 − 1)(cid:3) − (Y − 1) exp −l0(cid:2)(X − 1)2 exp−2l0 + (X 2 − 1)(cid:3)(cid:9) ∆0(X, l) =(cid:2)(2X + 1)2 exp 2l0 − exp(−l)(cid:3) δRnl = 4 RN (cid:16)gP R(cid:17)2 ∆2F The signal will be largest at short distance for which: and: inf (Pc, PF )2 RN XY X + Y ≤ δRnl(l −→ 0) δRnl(l −→ 0) ≤ sup (Pc, PF )2 RN XY X + Y so that δRnl(l −→ 0) scales as RN XY X+Y . Therefore in the strong tunneling regime (X ≫ 1 and Y ≫ 1), δRnl scales as inf(X, Y ) RN ≫ RN yielding much larger signals than in the lateral geometry. But whenever either of X or Y is in the transparent regime, δRnl will scale as the smaller of the two and so will be at most at 31 the scale of RN as predicted for open geometries (see II B 2). In order to check the impact of increasing the spin resistance mismatches at the various terminals to the signal in the lateral geometry we consider a non-local resistance ratio m(X, Y, l) per: m(X, Y, l) = δRnl(X, Y, l) δRnl,lateral(X, l) = ∆0(X, l) ∆2F (X, Y, l) . The relative increase of the signal is then m(X, Y, l) − 1. At l = 0: m(X, Y, l = 0) −→ Y (X + 1) X + Y . At large distance: m(X, Y, l) −→ (2X + 1)2 2 (X + 1)2 (63a) (63b) (64) (65) which is larger than 1 whenever x ≥ 1/√2 and more generally belongs to the interval [0.5; 2]. The large distance behaviour is of course less interesting since the non-local signal is weaker, so we will discuss in detail only the short distance behaviour. At short distances, large values are achieved whenever both parameters X and Y are large (spin confining regime, closed geometry). For X ≫ Y ≫ 1, m(X, Y, l = 0) ∼ Y ; while for Y ≫ X ≫ 1, m(X, Y, l = 0) ∼ X+1; finally for X ∼ Y ≫ 1, m(X, Y, l = 0) ∼ (X + 1) /2. This third situation is probably the easiest to achieve: indeed if we want to gain at least an order of magnitude the other cases would imply that one of the mismatches is at least two order of magnitudes larger while when X ∼ Y ≫ 1 they both have the same magnitude. For a realistic value X ∼ 10 in the weak tunneling regime (where the contact resistance is usually metallic) this yields an enhancement by a factor up to 5 (see Section IV for a discussion of realistic parameters X ∼ 10). We plot in the next figures (Fig. 7-9) the length dependence for various values of X and Y in the weak tunneling regime (keeping moderate values (X, Y ) ≤ 10) covering the three situations described in the previous paragraph (larger X , larger Y or equal X = Y ). 32 Figure 7. Ratio of non-local resistance variation in cross versus lateral setup as a function of distance l between injector F 1 and detector F 3 (in units of spin diffusion length lN ) for spin resistance mismatches Y = 10 (at terminals F 2/F 4) and X = 1 − 10 (at terminals F 1/F 3). F 1 and F 3 are identical ferromagnetic metals while F 2 and F 4 are paramagnets in this section with a spin resistance mismatch with the central paramagnet. Still in the weak tunneling regime we also plot on Fig. 10 the non-local resistance ratio for fixed distance l = 0.1 − 0.3 as a function of spin resistance mismatch X (assuming Y = X). Values in the range l = 0.1 − 0.3 are quite reasonable experimentally (for instance for Cu at ambient temperature lN = 300 nm while a distance L = 100 nm is within reach of lithography). At such distances the enhancement can still be several hundred of percents as can be seen in Fig. 10 unless X < 1 (transparent regime). The transparent regime is less interesting in terms of an increase of the signal since the geometry is now open. Yet whenever Y is larger than 1, even if X is in the transparent regime, one still gets an increase of the signal at short distance (Fig. 11). But if both parameters are smaller than unity, the signal gets reduced by several tens of percent when 33 Figure 8. Ratio of non-local resistance variation in cross versus lateral setup as a function of distance l between injector F 1 and detector F 3 (in units of spin diffusion length lN ) for spin resistance mismatches Y = 1 − 10 (at terminals F 2/F 4) and X = 10 (at terminals F 1/F 3). F 1 and F 3 are identical ferromagnetic metals while F 2 and F 4 are paramagnets in this section with a spin resistance mismatch with the central paramagnet. compared against the lateral setup as can be seen on Fig. 12-13. To summarize this section devoted to the cross geometry with two ferromagnets (one as charge source and spin injector, the other as spin accumulation detector), we have confirmed the impact on spin confinement of tunnel barriers even of moderate strength (X = 10 might correspond to Rc ∼ 10 Ω since RN is usually in the Ohm range). In the transparent limit, the device under-performs when compared to the standard lateral spin valve and should be avoided. 34 Figure 9. Ratio of non-local resistance variation in cross versus lateral setup as a function of distance l between injector F 1 and detector F 3 (in units of spin diffusion length lN ) for spin resistance mismatches Y = 1 − 10 (at terminals F 2/F 4) and X = Y (at terminals F 1/F 3). F 1 and F 3 are identical ferromagnetic metals while F 2 and F 4 are paramagnets in this section.. B. Van der Pauw cross with three ferromagnets. We now use two ferromagnets as spin injectors. When the (charge) source and drain are in antiparallel orientation the signal is enhanced because both ferromagnets acts as spin sources. But when they are parallel, one is a spin source and the other a spin sink so that spin accumulation is reduced. If the electrodes are identical ferromagnets, then the signal will be doubled when compared with the case where there is only one spin injector electrode. This is exactly what we found since Rnl ∝ Pef f,1 − Pef f,2 = 2Pef f,1 if electrodes are identical and antiparallel. We assume that F 1− F 2− F 3 are identical electrodes (F 1 and F 2 antiparallel while F 3 can switch from one orientation to the other) and F 4 is a paramagnet. The spin resistance mismatches are set as: X1 = X2 = X3 = X; X4 = Y. (66) 35 Figure 10. Non-local resistance ratio in cross versus lateral setup at fixed distances l = 0.1 − 0.3 between injector F 1 and detector F 3 (in units of spin diffusion length lN ) for spin resistance mismatches X = Y . We also assume l1 = l2 = l3 = l4 and use the same definitions as in previous section: (67) (68) (69) Then: gP R1 = gP R2 = gP R3 =gP R gP R4 = 0 2σ1σ3 RN (cid:16)gP R(cid:17)2 F(cid:1)2i h3δ− N(cid:0)δ+ F + δ− N δ+ Rnl = F δ+ where σ1 = ±1 and σ3 = ± refer to the majority spin direction of ferromagnets F 1 and F 3 relative to an absolute axis. Let us define ∆3F (X, Y, l) = [2 (Y + 1) exp l0 + 2 (Y − 1) exp −l0]−1 × {(Y + 1) exp l0 (cid:2)2 (X + 1)2 exp 2l0 − (X − 1)2 exp−2l0 +(cid:0)X 2 − 1(cid:1)(cid:3) − (Y − 1) exp −l0(cid:2)2 (X − 1)2 exp−2l0 − (X + 1)2 exp 2l0 +(cid:0)X 2 − 1(cid:1)(cid:3)(cid:9)(70) 36 Figure 11. Ratio of non-local resistance variation in cross versus lateral setup as a function of distance l between injector F 1 and detector F 3 for spin resistance mismatches Y = 10 (at terminals F 2/F 4) in the weak tunneling regime and X = 0.1 − 1 (at terminals F 1/F 3) in the transparent regime. Then: δRnl = RP − RAP = 4 RN (cid:16)gP R(cid:17)2 ∆3F where P/AP refer to the direction of F 3 relative to F 1. Therefore the non-local resistance (71) (72) as in section III A roughly scales as: δRnl ∼ C RN X 2 ∆3F (X, , Y, l) (where C is a constant; we have used upper and lower bounds on(cid:16)gP R(cid:17)2 above). The largest signal is found at small distance when: δRnl −→ RN XY 3Y + X . We observe a characteristic quadratic dependence of the numerator against a linear one in the denominator as a function of spin resistance mismatch: this is what ensures that in the 37 Figure 12. Ratio of non-local resistance variation in cross versus lateral setup as a function of distance l between injector F 1 and detector F 3 for spin resistance mismatches Y = 0.5 (at terminals F 2/F 4) and X = 0.1 − 1 (at terminals F 1/F 3). All electrodes are in the transparent regime. tunneling limit, δRnl can scale as Rc (here as the smaller of RN X ≫ RN or RN Y ≫ RN . But as soon the geometry is open due to one or both spin resistance mismatches in the transparent limit, δRnl scales as Rn or below. The ratio of the non-local resistance with regards to the standard lateral setup is then: m(X, Y, l) = ∆0 ∆3F (73) By varying the spin resistance mismatch Y at terminal F 4 one can interpolate between an open geometry (Y ≤ 1) and a closed one (Y ≫ 1 while X ≫ 1). Let us make some general comments comparing the three ferromagnets cross setup with the two ferromagnets cross discussed in the previous section III A 2. When X = Y , one gets ∆2 = 2∆3 so that m3F (X, Y = X, l) = 2m2F (X, Y = X, l). The non-local signal is exactly twice that found for the setup with two ferromagnets. Indeed when all terminals have the same spin resistance mismatch, the effective resistances Ref f are 38 Figure 13. Ratio of non-local resistance variation in cross versus lateral setup as a function of distance l between injector F 1 and detector F 3 for spin resistance mismatches Y = 0.1 − 1 (at terminals F 2/F 4) and X = Y (at terminals F 1/F 3). All electrodes are in the transparent regime. identical in both setups since the parameters are identical; but the total effective polarization at injector Pef f,1 − Pef f,2 = 2Pef f,1 is doubled in the 3 ferromagnet setup (Pef f,2 = 0 in the two ferromagnet setup). For X > Y the effective resistance (see II B 1) for the 3 ferromagnet setup becomes larger than that of the 2 ferromagnet setup (see III A 2; this is quite clear since Ref f,i(Xi) is an increasing function of Xi so that Ref f,3F =(cid:20) Ref f,2F =(cid:20) 3 Ref f,1(X) 2 Ref f,1(X) + + 1 Ref f,1(Y )(cid:21)−1 Ref f,1(Y )(cid:21) .−1 > 2 This means that the non-local signal will be more than doubled with respect to the 2 ferromagnet setup studied in III A 2. This effect does not require a totally closed geometry to occur. For X < Y , the 3 ferromagnet geometry is by the same arguments less spin confining since we have three terminals with a smaller spin resistance against only two in the 2 ferro- 39 m(X, Y = 1, l) = 2 (cid:2)(2X + 1)2 exp l − exp−l(cid:3) ×(cid:2)2 (X + 1)2 exp l − (X − 1)2 exp−l +(cid:0)X 2 − 1(cid:1)(cid:3)−1 X + 1(cid:19)2 m(X, Y = 1, l) −→(cid:18)2X + 1 . (74) (75) (76) magnet setup. However the doubling of effective polarizations remain which may mitigate the decrease of effective spin resistance. But as a rule in order to achieve stronger signals one should seek the condition X ≥ Y . 1. Open geometry (Y = 1). We first consider the case when the paramagnetic terminal has no tunnel barrier but is perfectly matched in terms of spin resistance to the central cross. This is in the terminology of Ref.23 an open geometry. Nevertheless as in the 2 ferromagnet geometry studied in III A, we will find that spin accumulation can still be enhanced even though not all the terminals are tunnel barriers. The spin resistance mismatch Y is set to unity: X4 = Y = 1. The non-local resistance ratio to the two ferromagnet cross (with Y = 1) is then: At large distance its limit is: which is always larger than 1 whether X is in the spin confining (X > 1) or the transparent regime (X < 1). At small distance the ratio tends to: m(X, Y = 1, l) −→ 4 X + 1 X + 3 (77) which is larger than 2 and can be as large as 4 in the tunneling regime (provided X > 1) and is comprised in the interval [4/3; 2] for X ≤ 1 (transparent regime). So although the geometry is open, the use of tunneling junctions does have some spin confining effect, which here is reinforced by the use of two injector electrodes resulting in an enhancement of the non-local resistance which can be up to 300% increase in the tunneling regime. For a distance l = 0.2 and X = 2 − 10, m = 2.45 − 3.30 (145 − 230 % increase). As can be seen in the next figure (Fig. 14) the enhancement can be seen at all distances so that this geometry is already better than the lateral one in terms of spin confinement. 40 Figure 14. Ratio of non-local resistance in cross geometry (with three ferromagnets) versus lateral setup as a function of distance l between injector and detector. Spin resistance mismatch is X = 1 − 10 at terminals F 1 − F 2 − F 3 (weak tunneling regime). In the transparent regime (X < 1), the doubling effect coming from the two spin injectors compensate partly the spin leakage due to transparent junctions so that there is always at least a 33 % increase at short distance. The enhancement is still present at large distance and can then reach up to 125 % as can be seen in Fig.15. 2. General case. We now conjugate the effects of spin confinement by tunnel barriers and the doubling of injector terminals. At small distance one gets: m(X, Y = 1, l) −→ 4Y X + 1 X + 3Y 41 (78) Figure 15. Ratio of non-local resistance in cross geometry (with three ferromagnets) versus lateral setup as a function of distance l between injector and detector. Spin resistance mismatch is X = 0.1− 1 at terminals F 1− F 2− F 3 (transparent regime). There is still an enhancement of the signal. while at large distances: X + 1(cid:19)2 m(X, Y = 1, l −→ ∞) −→(cid:18)2X + 1 < 4. (79) Large values are achieved whenever both parameters X and Y are large (both in the tunneling regime). Let us focus first on that tunneling regime. There are then three interesting limits: (i) X ≫ Y ≫ 1, which implies m(X, Y, l = 0) ∼ 4Y ; (note also that the spin confinement is more pronounced in that limit than in the 2 ferromagnet setup due to a larger effective spin resistance); (ii) Y ≫ X ≫ 1 implying m(X, Y, l = 0) ∼ 4 (iii) finally for X ∼ Y ≫ 1, m(X, Y, l = 0) ∼ X + 1 (see Fig. 16-18). For a realistic value X = Y ∼ 10 (see section IV C) this yields an enhancement by a factor 3 (X + 1); 42 Figure 16. Ratio of non-local resistance in cross geometry (with three ferromagnets) versus lateral setup as a function of distance l between injector and detector. Spin resistance mismatch is X = 10 at terminals F 1 − F 2 − F 3 and Y = 1 − 10 at paramagnet F 4. All electrodes are in the tunneling regime. up to 11 reaching therefore an order of magnitude. At a finite distance the enhancement remains considerable ranging from m = 3.3− 6.6 at l = 0.2 for X = 10 and Y = 1− 10 (Fig. 16). If one exchanges X and Y (with large Y = 10 and X = 1 − 10), one gets weaker signals. This is normal, since when Y is larger than X, spin confinement is disfavored since there are then three terminals with spin resistance mismatch smaller than the fourth (Fig. 17). When X and Y are about equal and large, X ∼ Y ≫ 1 the relative increase (∼ X + 1) for a given mismatch X is smaller than when Y ≫ X ≫ 1 (m ∼ 4 3 (X + 1)); however the latter condition might be more inconvenient to realize for a modest gain (for instance, if we 43 Figure 17. Ratio of non-local resistance in cross geometry (with three ferromagnets) versus lateral setup as a function of distance l between injector and detector. Spin resistance mismatch is varied (X = 1 − 10) at terminals F 1 − F 2 − F 3 and set at Y = 10 at paramagnet F 4. All electrodes are in the tunneling regime.. aim at an order of magnitude increase, X ∼ 10 this would imply Y ∼ 100 (in the strong tunneling limit) yielding m ∼ 14 at l = 0; while already for X = Y ∼ 10, m ∼ 11 at l = 0 (see Fig. 18). In practice the distance between injector and collector can not be reduced arbitrarily. Nevertheless the ratio m (X, Y, l) may remain large as can be seen in the next figure (Fig. 19) which shows the relative increase of signal at distances l = 0.1 − 0.3. For X ∼ Y = 10, m ∼ 5.7 − 8.0 for l = 0.1 − 0.3 (which are quite reachable at low temperature for l = 0.1 or even room temperature for l = 0.3). This is a doubling of the signal when compared to the values we found for X ∼ Y = 10 in the two ferromagnet closed geometry (of section III A 2). (As previously mentioned, this stems from the fact that the total effective polarization is doubled due to the use of two spin injectors while the effective resistances are identical.) 44 Figure 18. Ratio of non-local resistance in cross geometry (with three ferromagnets) versus lateral setup as a function of distance l between injector and detector. Spin resistance mismatch is varied X = 1 − 10 at terminals F 1 − F 2 − F 3 but Y = X at paramagnet F 4. All electrodes are in the tunneling regime.. Turning now to the transparent regime, we still observe an enhancement of the signal (see Fig.20): as before this is entirely due to the doubling of spin injector terminals since there is spin leakage at all terminals. For X = Y ≤ 1, the ratio m varies between X + 1 and 2 for an increase up to 125 % at large distance. To summarize this section, we have seen the impact of doubling the number of spin injectors conjugated to spin confinement resulting from large spin resistance mismatches. Although this geometry with three ferromagnets is quite interesting in terms of the large signals which can be achieved, in practice it is better to consider a four ferromagnet setup which will have all the qualities of the 3 ferromagnet Van der Pauw cross with some additional 45 Figure 19. Ratio of non-local resistance in cross geometry (with three ferromagnets) versus lateral setup at distances l = 0.1 − 0.3 between injector and detector. Spin resistance mismatches are set to identical values at all terminals X = Y properties: protection from voltage offsets (which may be detrimental to SNR) and infinite GMR ratio for the non-local resistance. IV. VAN DER PAUW CROSS WITH FOUR FERROMAGNETS. We now consider now the main device of this paper, a four ferromagnet Van der Pauw setup with the following symmetries: (i) identical injector and collector electrodes (F 1 and F 2), (ii) identical detector electrodes (F 3 and F 4). Besides an enhancement of the spin voltage such a device is protected against voltage offsets and offers an infinite non-local GMR ratio. Other general properties are described in section II C: notably such a setup will also display an ON-OFF switch effect (for the basic 1-bit basic sensing function). 46 Figure 20. Ratio of non-local resistance in cross geometry (with three ferromagnets) versus lateral setup as a function of distance l between injector and detector. Spin resistance mismatches are set to identical values at all terminals X = Y in the transparent regime (X = 0.1 − 1). A. Non-local resistance and infinite GMR. We consider a rhombus geometry for the cross: OA = OB = l1 while OC = OD = l3. The resistance mismatch parameters for the device are set as: X1 = X2 = X; X3 = X4 = Y while: The non-local resistance [Eq. (23)] becomes: . gP R1 = gP R2 =gP R, gP R3 = gP R4 =gP R ′ Rnl = Ref f (σ1 − σ2) (σ3 − σ4) Pef f,1 Pef f,3 (80) (81) (82) (83) where σi = ±1 refer to the majority spin direction of terminal i relative to an absolute axis. 47 Eventually: Rnl = where Rn (σ1 − σ2) (σ3 − σ4) gP RgP R [(X + 1) (Y + 1) exp l − (X − 1) (Y − 1) exp −l] ′ l = l1 + l3. (84) (85) Upon magnetization switching of one detector electrode the variation in non-local resistance is therefore: δRnl = [(X + 1) (Y + 1) exp l − (X − 1) (Y − 1) exp −l] ′ 4RngP RgP R B. Amplitude of non-local resistance. The signal is largest at small distance where the non-local resistance obeys: inf (Pc, PF ) inf (P ′ c, P ′ F ) 4Rn XY (X + Y ) ≤ δRnl(l −→ 0) (86) with a similar upper bound. Again one recovers the characteristic quadratic dependence on spin resistance mismatches at the numerator (linear at denominator), which imply that if both X and Y are large, δRnl will scale with them, much beyond RN . If one of the spin mismatches (say X) is in the transparent regime, it will set the scale, below or at most at RN . Let us compare in details with the standard lateral setup; we will set gP R = gP R and X = Y for simplicity. (When X 6= Y the same trends against the case X = Y are seen as in the three ferromagnet geometry.) ′ Then the non-local resistance ratio becomes: m(X, l) = δRnl δRnl,lateral = (cid:2)(2X + 1)2 exp l − exp −l(cid:3) (cid:2)(X + 1)2 exp l − (X − 1)2 exp−l(cid:3) . At small distance, one gets the enhancement: m(X, l) −→ X + 1. (87) This is the same enhancement as in the 3 ferromagnet setup when X = Y (see III B 2). We first consider the tunneling regime (X > 1). We plot in the following figures (Fig. (21-22) ) the length and resistance mismatch dependence of the ratio m(X, l). For X = 1, 48 Figure 21. Ratio of non-local resistance in cross setup (with four ferromagnets) versus standard lateral setup, as a function of distance l between injector and detector. Spin resistance mismatch X is in the spin confining regime (X = 1 − 10). we observe that m ≈ 2 at all distances: in the case of open cross geometry with two ferromagnets (see section III A 1) already one had m ∼ 1; the doubling here stems therefore from the two spin injectors. For X = 10, the enhancement remains strong at increasing distance: for instance for l = 0.1 − 0.3, m ∼ 500 − 770 %. For completeness we now turn to the transparent regime (X < 1) although it is less interesting than the tunneling regime if one is to achieve large signals. We observe the same trend in Fig. 23 as before with an increase in several tens of % and up to 100 % at large distance, which stems from the double spin injectors. At shorter distance, the spin current leakage is more pronounced. 49 Figure 22. Ratio of non-local resistance in cross setup (with four ferromagnets) versus standard lateral setup, at fixed distance l = 0.1− 0.3 between injector and detector. Spin resistance mismatch X ≥ 1 is in the spin confining regime. C. Discussion and comparison to experiments. Can we expect signals in the mV range for non-local voltages? Already the state-of-art has reached ∼ 100 − 200 µV in lateral spin valves21; the peculiarity of these lateral spin valves is that they use thin tunnel MgO barriers (1 − 6 nm) between the ferromagnets (Permalloy NiF e) and the paramagnetic channel which is silver Ag. Due to the thinness of MgO the interface resistances are much smaller than in usual tunnel junctions, with an interface resistance times cross section product in the range RI A ∼ 0.2 Ωµm2. While for as deposited samples the signals are already higher than usual (∼ 10 µV ), after annealing the signals can jump by an order of magnitude; this has been accounted on oxygen vacancies which increase after annealing. Let us see how much we can get in similar conditions but with the cross geometry with four ferromagnetic electrodes: we will borrow the parameters found of Fukuma and coll.21 for lateral spin valves based on P y − MgO − Ag junctions; the size of the junctions is 50 Figure 23. Ratio of non-local resistance in cross setup (with four ferromagnets) versus standard lateral setup, as a function of distance l between injector and detector. Spin resistance mismatch X is in the transparent regime (X = 0.1 − 1). All terminals are identical in this plot. A = 0.022 µm2; the separation between detector and injector is L = 300 nm . At 10 K, for a 1 mA current, the largest spin voltage is measured at 112 µV with the following parameters: Pc = 0.44 for the interface conductance polarization (noted PI by Fukuma and coll.21 ), PF = 0.35 for P y electrodes, lF = 5 nm, lN = 1100 nm for the spin relaxation lengths of P y and silver Ag, RN = ρ∗ N lN /A = 0.89 Ω for Ag spin resistance, F lF /A = 0.09 Ω (RN iF e = 0.08 Ω in Fukuma and coll.21 is related to our spin F ) ), Rc = 12.13 Ω (our interface spin resistance Rc I ) with RIA = 0.2152 Ωµm2 and RF = ρ∗ resistance RF by RN iF e = RF (1 − P 2 is related to RI of Fukuma and coll. by Rc = RI/(1 − P 2 A = 0.022 µm2). Then the spin impedance mismatch is as large as: while: X = RF + Rc RN = 13.73 l = L/lN = 0.27. 51 For these values and using the measured ∆Vnl = 110 µV , ∆Rnl = 110 mΩ, I = 1 mA this implies: which would imply a spin voltage variation as large as: m(X, l) = 660 % or: ∆V ∼ 725 µV ∆Rnl ∼ 725 mΩ (since I = 1 mA); if we use the symmetric four-terminal cross device, the baseline signal vanishes (for parallel orientation of detector electrodes), so that R A = ∆R A = 16 mΩµm2 which compared with CPP GMR spin valves is large in terms of ∆R A product (a few mΩµm2) but small in terms of customary R A values (10 − 100 mΩµm2)12. Still at 10 K, the highest voltage reported is 200 µV for a larger current I = 3.5 mA so that the non-local resistance is smaller ∆Rnl = 63 mΩ; using our setup we predict one would have achieved signals in the range: ∆Vnl ∼ 1.320 mV ∆Rnl ∼ 415 mΩ ∆R A ∼ 9.1 mΩµm2 j ∼ 1.6 107A/cm2. The current density is therefore reasonable although the mV mark for the spin voltage is reached! The non-local resistance variation is however smaller (Fukuma and coll.21 attributes the decrease to an enhanced spin relaxation in Ag due to phonons). If we put into perspective these predictions, we may remember that the first measurements of non-local resistance in bulk metal wires1 by Johnson and Silsbee yielded ∼ nΩ, were thereafter improved in thin films2 to about 10 µΩ; Fukuma and coll.21 have pushed to ∼ 100 mΩ. The largest non-local resistance variation observed is however already in the Ohm range in lateral tunnel junctions18,19, with one major catch: currents must remain small (∼ µA) due to spin depolarization at larger currents, which results in a spin voltage 52 still in the customary µV range. In contrast, in metallic lateral spin valves larger currents can be achieved21,22,49 -- 54. Turning now to room temperature, there is few change in spin impedance mismatch X but due to a much smaller lN ∼ 300 nm One still finds a sizable enhancement: X = 13.44 l = 1. m(X, l) = 415 % which would lead to the following numbers (using original experimental values of Fukuma and coll.21 ∆Vnl = 51 µV , ∆Rnl = 51 mΩ, I = 1 mA): ∆Vnl ∼ 210 µV ∆Rnl ∼ 210 mΩ ∆R A ∼ 4.5 mΩµm2 j ∼ 5 106A/cm2. There is still room for improvement at room temperature before reaching the mV mark. One obvious way would be to decrease the distance between injector and detector; for instance suppose L = 100 nm which is quite reasonable in terms of lithography, then l = 0.33; taking the previous room-temperature value for X (X = 13.44) would yield: so that m(X, l) = 605 % ∆Vnl ∼ 310 µV ∆Rnl ∼ 310 mΩ ∆R A ∼ 6.8 mΩµm2 j ∼ 5 106A/cm2. If we lower to a still achievable 50 nm for which l = 0.17 then: m(X, l) = 795 % 53 and: ∆Vnl ∼ 405 µV ∆Rnl ∼ 405 mΩ ∆R A ∼ 8.9 mΩµm2 j ∼ 5 106A/cm2. (at such length L = 50 nm m = 1150% at 10K so that ∆Vnl ∼ 1.250 mV again in the mV mark at I = 1 mA). Are there any other ways to further improve the signal at room-temperature? We will show elsewhere36 that with an all-lateral structure even stronger enhancements are found due to better spin confinement. In the latter example one would get as much as m ∼ 1070 % at room temperature (or m ∼ 1320 % at 10 K). Application to sensor and read-heads technology? Reaching the 1 T bit/inch2 mark up to 10 T bit/inch2 areal density requires new recording methods to push beyond the super-paramagnetic limit; bit patterning and thermally assisted magnetic recording are currently investigated13. But such large areal densities are also very challenging for current sensor technologies. The transition from AMR, CIP GMR, TMR and now CPP GMR has accompanied dramatic and steady increase of the areal density of hard-drives up to about 0.5 − 0.7 T bit/inch2 (as of late 2011). Further increase to reach the 1 T bit/inch2 has motivated the latest transition from TMR read-heads to CPP metallic spin valves because TMR is limited to R A ∼ 1 Ωµm2 which will not be small enough to accommodate the high data rates required for the desired areal densities12. In contrast CPP GMR read-heads can have R A < 0.1 Ωµm2 which is the minimum required for an areal density 1 T bit/inch2. An additional requirement for CPP spin valves is to have larger ∆R A > 5 mΩµm2 instead of the usual< 1 mΩµm2 which are too small to ensure good SNR (signal to noise ratio). The source of noises are usually Johnson thermal noise in metallic spin valves (both CIP and CPP) or shot noise in TMR valves. But at the shrinking sizes relevant for larger areal densities additional relevant sources of noise appear to be STT (spin transfer torque) noise, mag-noise (noise due to thermal fluctuations in the free layer) and amplifier noise12,14,55. Non-local devices have been barred for practical use due to small signals and have mostly been interesting for fundamental and pedagogical use. As we have shown this may prove less 54 of a concern by using geometries with better spin confinement and with multiple injectors. Such non-local devices used as sensors would hold several advantages over their standard counterparts while retaining the small RA of CPP metallic spin valves: - smaller thickness of the sensing detector (two or three layers since the reference layer is located elsewhere and can have a different coercive field by playing on its geometry); - diminished sensitivity to STT noise since the largest momentum transfer is between source and drain electrodes; - no self-field since no current flows through the detectors (Oersted fields are at injectors). - In the case of a symmetric cross geometry, the possibility to achieve 100 % or maximum MR ratio (using the pessimistic ratio) so that R = ∆R is also advantageous regarding Johnson and shot noise providing a 10− 20 dB boost in SNR (as explained above in section II C). In addition the maximum MR ratio property allows to satisfy both requirements sought in CPP spin valves: small enough R A < 0.1 Ωµm2 for high data rates, but large enough ∆R A > 5 mΩµm2 for good SNR. (The latter figure would need to be adjusted to take into account the peculiarities of the non-local device but we use it as a reasonable rule-of-thumb). The main constraints for a large SNR will remain mag-noise, amplifier noise as well as the ability to scale down the non-local setup to par with the 20 − 30 nm track width relevant11,14 for areal densities above 1 T bit/inch2. Although we expect STT noise to be less of a nuisance, angular momentum transfer still occurs from injectors to detectors and a quantification of critical currents adapted to the geometry would be interesting. The angular response is thus an important aspect to explore in view of a potential use as a sensor and is discussed elsewhere36. Although the geometry studied in this paper might seem quite peculiar, there is actually a lot of flexibility in terms of design, provided the following recommendations are enforced: (i) confine spin with large interface resistances (when compared with the paramagnet spin resistance) wherever spin current can leak; (ii) in the smallest volume for the paramagnet connecting the ferromagnetic terminals (when compared to l3 N ); (iii) to ensure infinite MR (optimistic) ratio, use two identical ferromagnetic electrodes as detectors within a symmetric arrangement with respect to injectors (the geometry of the ferromagnetic electrodes need not be completely identical: they just need to be identical to 55 within a few spin diffusion lengths lF of the contact to the paramagnet; this is an important practical remark since this allows to have different coercive fields for the various electrodes while keeping the properties stemming from symmetry); (iv) for further gain in the signal, use two antiparallel ferromagnetic electrodes. With these prescriptions non-local devices will perhaps be able to reach the realm of applications. V. CONCLUSION AND PROSPECTS. We have presented a variant of non-local spin valves based on a Van der Pauw cross setup: its basic characteristic is its reliance on four collinear ferromagnetic terminals. By playing on the numerous magnetization configurations of the four ferromagnets and on their symmetries several functionalities have been described: (i) an improved non-local spin valve (when used to read 1 bit) with an infinite GMR for the non-local resistance; (ii) ON-OFF switch effect where magnetization at the injectors can control the read-off at detectors; (iii) 3-bit storage or sensing and offset-free 2-bit reading; (iv) direct Spin Hall measurement; (v) use as programmable magneto-logic gates. In addition to these functionalities the amplitude of the non-local resistance has the potential to be much increased in such a setup due to the conjunction of (i) spin confinement by tunnel barriers; (ii) the use of two electrodes as spin injectors instead of only one. We have also studied the separate or combined impact of both features by considering setups with two or three ferromagnets: the use of two injectors basically doubles the signal while the use of tunnel barriers hinder spin leaking from the Van der Pauw cross, thus increasing spin accumulation within it. The latter effect is already observed even when not all terminals are connected by tunnel barriers to the Van der Pauw cross. An important parameter is the spin resistance mismatch at each ferromagnetic-paramagnetic interface (at the four terminals), which is the ratio between the spin resistances of the ferromagnet and the paramagnet. The larger this parameter, the better the spin confinement in the Van der Pauw cross. Additional functionalities will be discussed elsewhere: non-collinear properties which are essential for sensor applications; as well as caloritronic ones. A related non-local device in the lateral geometry for which signals are much stronger will be also investigated36. 56 Appendix A: Computing the Spin Voltage. We derive in this Appendix the general expression of the spin voltage by relying on the one-dimensional drift-diffusion equations7,31,32. 1. Basic equations. a. Spin accumulation vectors. We orient spin currents away from the origin O of the cross. In each electrode (F 1 − F 4) the spin accumulation and spin currents are then: F 1: ∆µF 1(z) = ∆µF 1(A) exp− z − L1 lF 1 ∆µF 1(A) RF 1 exp− z − L1 lF 1 Is,F 1(z) = −PF 1Ic − so that: Is,F 1(A) = −PF 1Ic − ∆µF 1(A) RF 1 . (A1) (A2) (A3) F 2: without loss of generality we assume F 2 antiparallel to F 1 which allows to have symmetric equations (for the parallel case one should take opposite polarizations) Is,F 2(B) = −PF 2Ic − ∆µF 2(B) RF 2 where Is,F 2(B) is counted positive when flowing in the direction z′ = −z; F 3: Is,F 3(C) = − ∆µF 3(C) RF 3 where Is,F 3(B) is oriented in the direction x. F 4: (A4) (A5) ∆µF 4(D) Is,F 4(D) = − RD4 where Is,N 4(D) is oriented in the direction x′ = −x. In the arms (I − IV ) we define spin accumulation vectors which are related to the spin (A6) accumulation and spin currents by: I: 57 ∆µI(z) = KI . (cid:18)exp . (cid:18)exp Is,I(z) = KI RN z ; exp − lN z lN ; − exp − z lN(cid:19) , lN(cid:19) z where KI is a constant two-component vector. II: following our convention of orienting away from the origin III: IV: ; exp − ; − exp − KII RN Is,II(z′) = ∆µII(z′) = KII . (cid:18)exp . (cid:18)exp ∆µIII(x) = KIII . (cid:18)exp . (cid:18)exp ∆µIV (x′) = KIV . (cid:18)exp . (cid:18)exp Is,IV (x′) = Is,III(x) = KIII RN KIV RN z′ lN z′ lN x lN x lN x′ lN x′ lN x z′ z′ lN(cid:19) , lN(cid:19) lN(cid:19) , lN(cid:19) lN(cid:19) , lN(cid:19) x′ x′ x ; exp− ; − exp− ; exp− ; − exp − (A7) (A8) (A9) (A10) (A11) (A12) (A13) (A14) b. Conditions at ferromagnetic - paramagnetic interfaces and at cross center. Interfaces. At interfaces we will neglect all spin flip but assume interface resistance is spin dependent. This implies that spin current at points A − D is continuous: Is,F 1(A) = Is,I(A) (A15) and so forth at points B − D. At points A − D the interface resistance induces a discontinuity in spin accumulations: ∆µF 1(A) − ∆µI(A) = Rc1 Is(A) + Rc1 Pc1 Ic ∆µF 2(B) − ∆µII(B) = Rc2 Is(B) + Rc2 Pc2 Ic 58 (A16) (A17) ∆µF 3(C) − ∆µIII(C) = Rc3 Is(C) ∆µF 4(D) − ∆µIV (D) = Rc4 Is(D) (A18) (A19) Cross center. At O, the spin accumulations are continuous: ∆µI(O) = ∆µII(B) = ∆µIII(C) = ∆µIV (O) (A20) and spin current continuity implies: 0 = Is,I(O) + Is,II(O) + Is,III(O) + Is,IV (O) (A21) so that for α = I − IV : Kα . (1; 1) ≡ ∆µ(O); 0. Kα . (1; −1) = Xα (A22) (A23) 2. Solution of the equations. a. Spin accumulation vectors Kα We first solve for all four spin accumulation vectors in the arms (I − IV ). From the equations (A3, A15, A16) at interface F 1 - I one gets: ∆µI(A) + (Rc1 + RF 1) Is(A) = − (PF 1RF 1 + Pc1Rc1) Ic (A24) After substitution in terms of spin accumulation vector KI: KI . ((1 + X1) exp l1; (1 − X1) exp −l1) = − (PF 1RF 1 + Pc1Rc1) Ic where: and X1 = RF 1 + Rc1 RN L1 lN . l1 = 59 (A25) (A26) (A27) KII obeys a similar equation by changing the index 1 with index 2 since we have assumed that the magnetization of F 2 is antiparallel to that of F 1. KII . ((1 + X2) exp l2; (1 − X2) exp −l2) = − (PF 2RF 2 + Pc2Rc2) Ic while the equations for arms III − IV obtain by cancelling the charge current: KIII . ((1 + X3) exp l3; (1 − X3) exp −l3) = 0, KIV . ((1 + X4) exp l4; (1 − X4) exp −l4) = 0. (A28) (A29) (A30) with X2/3/4 and l2/3/4 defined as X1 and l1. Using the continuity of spin accumulation and spin current at O yields a second equation for each spin accumulation vector: Kα . (1; 1) = ∆µ(O), (α = I, .., IV) (A31) Solving for the Kα vectors in terms of ∆µ(O) leads to the following expressions: ∆µ(O) 2δ+ 1  KI = + (X1 − 1) exp −l1 (X1 + 1) exp l1 (PF 1RF 1 + Pc1Rc1) 2δ+ 1  Ic  −1 1   , (X3 − 1) exp−l3 (X3 + 1) exp l3 KII = 1 ←→ 2 KIII = ∆µ(O) 2δ+ 3  KIV = 3 ←→ 4, while: where: δ± i = (Xi + 1) 2 exp li ± (Xi − 1) 2 exp−li. 60 (A32) (A33) (A34) (A35) (A36) b. Spin currents at cross center. The relation between spin currents at O on each arm and the spin accumulation ∆µ(O) is derived using: Is,I(O) = 1 RN KI . (1; −1) so that: Is,I(O) = − with a similar relation for arm II where: ∆µ(O) Ref f,1 − Pef f,1 Ic Ref f,i = RN δ+ i δ− i Pef f,i = gP Ri δ+ i and: For arm III, one has simply: gP Ri = [PF iRF i + PciRci] /RN Is,III(O) = − ∆µ(O) Ref f,3 and a similar relation for arm IV . c. Spin accumulation at cross center. (A37) (A38) (A39) (A40) (A41) (A42) Plugging the expressions of the spin currents into the spin current continuity equation at O [Eq. (A21)] allows determination of ∆µ(O): ∆µ(O) = − (Pef f,1 + Pef f,2) Pi=1−4 R−1 ef f,i 1 Ic. (A43) Observe that the expression of ∆µ(O) does not depend on the polarizations of the fer- romagnetic electrodes used as detectors (F3 and F4): it is fully determined by the injectors 61 only (F 1 and F 2). The previous expression has been derived by assuming F 1 and F 2 an- tiparallel, which reflects into the polarizations signs. In the more general case of arbitrary collinear orientation, one has: ∆µ(O) = − (Pef f,1 − Pef f,2) Ref f Ic (A44) where positive polarizations are defined relative to an absolute axis (negative polarizations in the opposite direction). The largest spin accumulation in magnitude is achieved for antiparallel injector and collector electrodes. This can be plugged in the expressions for the spin current: with a similar expression for arm II while for arm III (or IV with appropriate substitutions): Ref f,1 Is,I(O) =(cid:20)(Pef f,1 − Pef f,2) Ref f Is,III(O) =(cid:20) (Pef f,1 − Pef f,2) Ref f − Pef f,1(cid:21) Ic (cid:21) Ic. Ref f,3 (A45) (A46) d. Spin voltage. The voltage is measured as: Vnl = − [µF 3(+∞) − µF 4(+∞)] /e. (A47) This voltage depends on the orientations of the four ferromagnetic terminals F 1 − F 4 so that in the most general case it can assume sixteen distinct values (24); but since the setup is invariant under a joint flipping of all the magnetizations this number is reduced to eight. We will need the chemical potential variation at interfaces: µF 3(C) − µIII(C) = −Rc3 Pc3 Is(C); µIV (D) − µF 4(D) = Rc4 Pc4 Is(D). Leaving out the factor e: −Vnl = [µF 3(+∞) − µF 3(C)] + [µF 3(C) − µIII(C)] + [µIII(C) − µIII(O)] + [µIV (O) − µIV (D)] + [µIV (D) − µF 4(D)] + [µF 4(D) − µF 4(+∞)] = PF 3 ∆µF 3(C) − Rc3 Pc3 Is(C) − PF 4 ∆µF 4(D) + Rc4 Pc4 Is(D) 62 so that: RF 3 −Vnl =(cid:20)PF 3RF 3 + Pc3Rc3 +(cid:20)PF 4RF 4 + Pc4Rc4 =(cid:20)PF 3RF 3 + Pc3Rc3 −(cid:20)PF 4RF 4 + Pc4Rc4 RF 4 + Rc4 RF 3 + Rc3 RF 4 ∆µF 3(C)(cid:21) ∆µF 4(D)(cid:21) ∆µIII(C)(cid:21) ∆µIV (D)(cid:21) The spin accumulations ∆µIII(C) and ∆µIV (D) are derived easily from the spin accu- mulation vectors KIII and KIV [Eq. (A34-A35)]: ∆µIII(C) = KIII . (exp l3; exp −l3) ∆µ(O) = X3 (A48) δ+ 3 and: One thus gets a voltage drop: ∆µIV (D) = 3 ←→ 4 Vnl = −∆µ(O) gP R3 4 ! 3 − gP R4 = −∆µ(O) (Pef f,3 − Pef f,4) δ+ δ+ e. Spin currents and spin accumulations at detector terminals. The spin current at point C is given by: so that: Is(C) = 1 RN KIII . (exp l3; − exp −l3) Is(C) = − ∆µ(O) RN δ+ 3 = (Pef f,1 − Pef f,2) Ref f Ic RN δ+ 3 . Let us study Is(C)as a function of X3 while keeping all other parameters fixed. Is(C) ∝ 1 aδ+ 3 + δ− 3 ∝ 1 2 {X3 [a cosh l3 + sinh l3] + [a sinh l3 + cosh l3]} 63 (A49) (A50) (A51) (A52) where: a =Xi6=3 R−1 ef f,i > 0. The spin current is therefore (in magnitude) a decreasing function of X3: in order to have large spin currents it is therefore better to have transparent junctions rather than large tunnel barriers. The spin accumulation at C behaves in an opposite manner: ∆µIII(C) = ∆µ(O) X3 δ+ 3 X3 3 + δ− 3 aδ+ ∝ ∝ X3 [a cosh l3 + sinh l3] + [a sinh l3 + cosh l3] X3 which is an increasing function of X3. A large accumulation at the interface with detector F 3 is favored by large spin impedance mismatch X3. The spin currents and accumulations at source and drain can be studied in the same manner with identical conclusions: large X leads to larger spin accumulations but smaller spin currents; with an opposite result for small X. 3. Local resistance and non-local charge current. The resistance can also be measured locally along the charge current path from source to drain; if additionally we close the circuit between the two detector electrodes a charge current is generated. a. Local resistance at injectors. The voltage probes at source and drain are positioned at z1 and z2 which we will assume to be very large (≫ lF i the spin relaxation lengths in the ferromagnets). Vlocal = − [µF 1(z1) − µF 2(z2)] . (A53) (we have left out the factor e). We will need the dependence of the chemical potentials as well as the interface disconti- nuities: 64 µF 1(z) = −ρ∗ µF 2(z′) = ρ∗ F 1(cid:0)1 − P 2 F 2(cid:0)1 − P 2 F 1(cid:1) Ic z + C1 − PF 1∆µF 1(z) F 2(cid:1) Ic z′ + C2 − PF 2∆µF 2(z′) where C1−2 are constants and: µF 1(A) − µI(A) = −Rc1 (Ic + Pc1Is(A)) µF 2(B) − µII(B) = Rc2 (Ic − Pc2Is(B)) − Vlocal = [µF 1(z1) − µF 1(A)] + [µF 1(A) − µI(A)] + [µI(A) − µII(B)] + [µII(B) − µF 2(B)] + [µF 2(B) − µF 2(z2)] Since: eventually: − Vlocal =(cid:20) PF 1RF 1 + Pc1Rc1 −(cid:20) PF 2RF 2 + Pc2Rc2 RF 2 + Rc2 RF 1 + Rc1 − R0 Ic ∆µI(A)(cid:21) ∆µII(B)(cid:21) where R0 ({P1−2}) is an even function of polarizations {P1−2} = PF i, Pci, i = 1 − 2: R0 {P1−2} = Xi=1−2 ρ∗ F i (cid:0)1 − P 2 F i(cid:1) zi + ρ∗ N li +Rci+ [P 2 F iRciRF i − P 2 ciR2 RF i + Rci ci − 2PF iPciRF iRci] . (It can be checked that R0 ≥ 0 as it should be.) Since: ∆µI(A) = KI . (exp l1; exp −l1) = ∆µ(O) δ+ 1 X1 (with a similar expression for ∆µII(B)) − Vlocal = ∆µ(O) [Pef f,1 − Pef f,2] − R0Ic = −Ref f [Pef f,1 − Pef f,2]2 Ic − R0Ic 65 (A54) (A55) (A56) (A57) (A58) (A59) (A60) (A61) The resistance variation when one flips terminal F 2 is therefore: ∆Rlocal =RAP − RP = VAP − VP Ic = 4 Pef f,1 Pef f,2 Ref f (A62) b. Local resistance at detectors and non-local charge current. The previous expressions will help us to derive the expression of the non-local charge current flowing through the detectors F 3 and F 4 when they are in closed circuit instead of being in open circuit. Suppose one drives a current I ′ c through terminals F 3 and F 4 but not through F 1 and F 2 exchanging the roles of injectors and detectors (Ic = 0); V34 = Ref f [Pef f,3 − Pef f,4]2 I ′ c + R0 {P3−4} I ′ c We can define the local conductance at detectors: I ′ c = g34V34 (A63) (A64) where g34 =(cid:2)Ref f [Pef f,3 − Pef f,4]2 + R0 {P3−4}(cid:3)−1 . (A65) We now switch on the charge current Ic at source and drain F 1 and F 2; when I ′ c = 0, V34 = Vnl = Rnl Ic = Ref f (Pef f,1 − Pef f,2) (Pef f,3 − Pef f,4) Ic. (A66) When both Ic and I ′ c are switched on, by superposition one gets the general expression of the current flowing through detector electrodes: If we short F 3 and F 4, V34 = 0 so that the non-local charge current is: I ′ c = g34 (V34 − Rnl Ic) . or: Inl = −g34 Rnl Ic Inl = − Ref f (Pef f,1 − Pef f,2) (Pef f,3 − Pef f,4) Ref f [Pef f,3 − Pef f,4]2 + R0 {P3−4} Ic 66 (A67) (A68) (A69) Figure 24. Left: geometry used by Hershfield-Zhao and Fert-Lee for the bipolar spin switch calcu- lations. Right: geometry in our simplified one-dimensional treatment which includes spin leaking to the bottom. Appendix B: Revisiting the bipolar spin switch. The original theory by Johnson of the bipolar spin switch was refined by Hershfield and Zhao48 who included spin relaxation in the ferromagnets, and also by Fert and Lee47 who studied the (adverse) impact of interface spin flips. If we discard the effect of spin flips, the non-local resistance variation was found by Hershfield and Zhao to be: ∆Rnl = [(X + 1) (Y + 1) exp l − (X − 1) (Y − 1) exp −l] ′ 4RngP RgP R (B1) where we have renamed parameters according to the definitions used in this paper: X and gP R = (PcRc + PF RF ) /RN are defined at injector while Y and gP R detector. Fert and Lee have the same expression in the casegP R =gP R ferromagnets at injector and detector). ′ ′ pertain to the and X = Y (identical What is noteworthy is that this expression is actually identical to that we found for the four ferromagnetic terminal cross geometry. In the limit of large X and Y , ∆Rnl might therefore be quite large as has been discussed repeatedly in this paper due to spin confinement. But a look at the geometry may cast doubt on this result (see Fig. 24 ): there should be some spin leakage in the setup even if both injector and detector are in tunnel contacts with the central paramagnet, because spin is leaving along with charge on the bottom 67 paramagnetic arm used as (charge) current drain. Hershfield and Zhao argued that spin leakage could be neglected in the bottom arm since in Johnson's original experiment the voltage pads are much larger than the spin relaxation length (the pads area was 0.01 mm2 ≫ l2 N ) but it is difficult to see how this can be relevant for the bottom arm which is the current drain. The argument however applies to the spin leaking through the voltage probes. An expression which takes into account spin flow in the bottom arm is actually easy to derive in line with the calculations done with the cross geometry. Referring to Fig. 24 we consider a three arm geometry (arms I − III) with two ferro- magnetic terminals F 1 and F 2 with current flowing from F 1 to III through I. We allow a finite size for arms I and II (L1 = l1 lN and L2 = l2 lN ) for a more general expression. We find: where: ∆Rnl = 2Ref f Pef f,1 Pef f,2 R−1 ef f =(cid:20) 1 R′ N + 1 Ref f,1 + 1 Ref f,2(cid:21) (B2) (B3) The effective polarizations Pef f,i for terminals F 1/F 2 and spin resistances Ref f,i for arms I and II are defined as in the bulk of the paper (with X1 = X and X2 = Y ): R−1 ef f = R−1 N (cid:20)RN R′ N + δ− 1 δ+ 1 + δ− 2 δ+ 2 (cid:21) (B4) We have allowed for different cross sections AN and A′ N along x′Ox and Oz′ so that the spin resistance along x′Ox is RN = ρ∗ N lN /AN but along Oz′ is R′ N = ρ∗ N lN /A′ N . The total spin resistance is easy to interpret: it corresponds to parallel addition of the spin resistance of the three arms with spin resistance R′ N for arm III and spin resistances Ref f,1 and Ref f,2 for arms I and II respectively. Suppose we neglect arm III spin resistance so that: R−1 ef f = 1 Ref f,1 + 1 Ref f,2 . (B5) One then recovers Fert-Lee and Hershfield-Zhao expression [Eq. (B1)] when l1 = l2 = l/2, which therefore does imply a neglect of spin leakage to the bottom terminal. 68 When is it allowed to do so? The condition is: The effective resistance varies between RN and RF + Rc; the condition then reduces to: R′ N ≫ (Ref f,1, Ref f,2) (B6) R′ N ≫ (RF 1 + Rc1, RF 2 + Rc2, RN ) . (B7) which is favored by small cross-section A′ N ≪ AN . Note that the condition RN ≪ R′ N is however not enough to recover Eq. (B1). So more generally when the condition in Eq. (B7) is not met, one gets (when l1 = l2 = l/2): ′ ∆Rnl = 8RngP RgP R × {2 (X + 1) (Y + 1) exp l − 2 (X − 1) (Y − 1) exp−l + RN R′ N [(X + 1) (Y + 1) exp l + (X − 1) (Y − 1) exp−l + 2 (XY − 1)]}−1 (B8) which is significantly different from Eq. (B1) . Let's define the spin mismatch Z = R′ N /RN . At small distance where the signal will be largest: ∆Rnl ∼ XY X −1 + Y −1 + Z −1 . The scale will be set by the smaller of the spin impedance mismatches. If anyone of them is in the transparent regime, one will recover the scaling characteristic of open systems: ∆Rnl ≤ RN as it should be. We have neglected in the above spin leakage in the voltage probes but it is easy to generalize the present considerations to include it, should it become experimentally relevant. ∗ [email protected] 1 M. Johnson and R. H. Silsbee, Physical Review Letters 55, 1790 (1985). 2 M. Johnson, Physical Review Letters 70, 2142 (1993). 3 M. Johnson, Science 260, 320 (1993). 4 F. J. Jedema, A. T. Filip, and B. J. van Wees, Nature 410, 345 (2001). 5 F. J. Jedema, H. B. Heersche, A. T. Filip, J. J. A. Baselmans, and B. J. van Wees, Nature 416, 713 (2002). 69 6 S. Takahashi, S. Maekawa, and H. Imamura, in Concepts in Spin Electronics, Series on Semicon- ductor Science and Technology No. 13 (Sadamichi Maekawa, New-York, 2006) pp. 343 -- 371(29). 7 M. Johnson and R. H. Silsbee, Physical Review B 35, 4959 (1987). 8 M. Johnson and R. H. Silsbee, Physical Review B 37, 5312 (1988). 9 T. Kimura, Y. Otani, and J. Hamrle, Physical Review Letters 96, 037201 (2006). 10 T. Yang, T. Kimura, and Y. Otani, Nat Phys 4, 851 (2008). 11 M. Takagishi, K. Yamada, H. Iwasaki, H. N. Fuke, and S. Hashimoto, IEEE Transactions on Magnetics 46, 2086 (2010). 12 K. Nagasaka, Journal of Magnetism and Magnetic Materials 321, 508 (2009). 13 R. Wood, Journal of Magnetism and Magnetic Materials 321, 555 (2009), <ce:title>Current Perspectives: Perpendicular Recording</ce:title>. 14 J. Katine and E. E. Fullerton, Journal of Magnetism and Magnetic Materials 320, 1217 (2008). 15 A. Fert, J. George, H. Jaffres, and G. Faini, Journal of Physics D: Applied Physics 35, 2443 (2002). 16 J.-M. George, A. Fert, and G. Faini, Physical Review B 67, 012410 (2003). 17 A. Fert, J. George, H. Jaffres, and R. Mattana, IEEE Transactions on Electron Devices 54, 921 (2007). 18 S. O. Valenzuela, D. J. Monsma, C. M. Marcus, V. Narayanamurti, and M. Tinkham, Physical Review Letters 94, 196601 (2005). 19 S. O. Valenzuela and M. Tinkham, Applied Physics Letters 85, 5914 (2004). 20 T. Wakamura, K. Ohnishi, Y. Niimi, and Y. Otani, Applied Physics Express 4, 063002 (2011). 21 Y. Fukuma, L. Wang, H. Idzuchi, S. Takahashi, S. Maekawa, and Y. Otani, Nat Mater 10, 527 (2011). 22 Y. Fukuma, L. Wang, H. Idzuchi, and Y. Otani, Applied Physics Letters 97, 012507 (2010). 23 H. Jaffres, J.-M. George, and A. Fert, Physical Review B 82, 140408 (2010). 24 L. E. Hueso, J. M. Pruneda, V. Ferrari, G. Burnell, J. P. Valdes-Herrera, B. D. Simons, P. B. Littlewood, E. Artacho, A. Fert, and N. D. Mathur, Nature 445, 410 (2007). 25 F. J. Jedema, M. S. Nijboer, A. T. Filip, and B. J. van Wees, Physical Review B 67, 085319 (2003). 26 M. Ichimura, Journal of Applied Physics 95, 7255 (2004). 27 J. Hamrle, T. Kimura, Y. Otani, K. Tsukagoshi, and Y. Aoyagi, Physical Review B 71, 094402 (2005). 28 J. Hamrle, T. Kimura, T. Yang, and Y. Otani, Journal of Applied Physics 98, 064301 (2005). 70 29 M. Johnson and R. H. Silsbee, Physical Review B 76, 153107 (2007). 30 J. Bass and W. P. Pratt, Journal of Physics: Condensed Matter 19, 183201 (2007). 31 T. Valet and A. Fert, Physical Review B 48, 7099 (1993). 32 E. Rashba, The European Physical Journal B - Condensed Matter 29, 513 (2002). 33 S. Garzon, I. Zutic, and R. A. Webb, Physical Review Letters 94, 176601 (2005). 34 F. Casanova, A. Sharoni, M. Erekhinsky, and I. K. Schuller, Physical Review B 79, 184415 (2009). 35 F. L. Bakker, A. Slachter, J.-P. Adam, and B. J. van Wees, Physical Review Letters 105, 136601 (2010). 36 K.-V. Pham, unpublished (2012). 37 T. Kimura, J. Hamrle, and Y. Otani, Physical Review B 72, 014461 (2005). 38 M. I. D'Yakonov and V. I. Perel', Soviet Journal of Experimental and Theoretical Physics Letters 13, 467 (1971). 39 J. E. Hirsch, Physical Review Letters 83, 1834 (1999). 40 Y. K. Kato, R. C. Myers, A. C. Gossard, and D. D. Awschalom, Science 306, 1910 (2004). 41 J. Wunderlich, B. Kaestner, J. Sinova, and T. Jungwirth, Physical Review Letters 94, 047204 (2005). 42 S. O. Valenzuela and M. Tinkham, Nature 442, 176 (2006). 43 T. Kimura, Y. Otani, T. Sato, S. Takahashi, and S. Maekawa, Phys. Rev. Lett. 98, 156601 (2007). 44 L. Vila, T. Kimura, and Y. C. Otani, Phys. Rev. Lett. 99, 226604 (2007). 45 I. Zutic and M. Fuhrer, Nat Phys 1, 85 (2005). 46 J. Fabian, A. Matos-Abiaguea, C. Ertlera, P. Stanoa, and I. Zutic, Acta Physica Slovaca 57, 565 (2007). 47 A. Fert and S.-F. Lee, Physical Review B 53, 6554 (1996). 48 S. Hershfield and H. L. Zhao, Physical Review B 56, 3296 (1997). 49 Y. Ji, A. Hoffmann, J. E. Pearson, and S. D. Bader, Applied Physics Letters 88, 052509 (2006). 50 Y. Ji, A. Hoffmann, J. S. Jiang, J. E. Pearson, and S. D. Bader, Journal of Physics D: Applied Physics 40, 1280 (2007). 51 Y. Ji, A. Hoffmann, J. S. Jiang, and S. D. Bader, Applied Physics Letters 85, 6218 (2004). 52 T. Kimura, Y. Otani, and J. Hamrle, Physical Review B 73, 132405 (2006). 71 53 T. Kimura and Y. Otani, Physical Review Letters 99, 196604 (2007). 54 A. van Staa, J. Wulfhorst, A. Vogel, U. Merkt, and G. Meier, Physical Review B 77, 214416 (2008). 55 S. Takahashi and S. Maekawa, Physical Review B 67, 052409 (2003). 72
1307.3828	3	1307	2013-08-28T09:19:32	Thermal spin injection and accumulation in CoFe/MgO tunnel contacts to n-type Si through Seebeck spin tunneling	[ "cond-mat.mes-hall" ]	We report the thermal spin injection and accumulation in crystalline CoFe/MgO tunnel contacts to n-type Si through Seebeck spin tunneling (SST). With the Joule heating (laser heating) of Si (CoFe), the thermally induced spin accumulation is detected by means of the Hanle effect for both polarities of the temperature gradient across the tunnel contact. The magnitude of the thermal spin signal scales linearly with the heating power and its sign is reversed as we invert the temperature gradient, demonstrating the major features of SST and thermal spin accumulation. We observe that, for the Si (CoFe) heating, the thermal spin signal induced by SST corresponds to the majority (minority) spin accumulation in the Si. Based on a quantitative comparison of thermal and electrical spin signals, it is noted that the thermal spin injection through SST can be a viable approach for the efficient injection of spin accumulation	cond-mat.mes-hall	cond-mat	Thermal spin injection and accumulation in CoFe/MgO tunnel contacts to n-type Si through Seebeck spin tunneling Kun-Rok Jeon1,2, Byoung-Chul Min3, Seung-Young Park4, Kyeong-Dong Lee1,5, Hyon-Seok Song1, Youn-Ho Park3, and Sung-Chul Shin1,6,* 1Department of Physics and Center for Nanospinics of Spintronic Materials, Korea Advanced Institute of Science and Technology (KAIST), Daejeon 305-701, Korea 2National Institute of Advanced Industrial Science and Technology (AIST), Spintronics Research Center, Tsukuba, Ibaraki 305-8568, Japan 3Center for Spintronics Research, Korea Institute of Science and Technology (KIST), Seoul 136-791, Korea 4Division of Materials Science, Korea Basic Science Institute (KBSI), Daejeon 305-764, Korea 5Department of Material Sciences and Engineering, Korea Advanced Institute of Science and Technology (KAIST), Daejeon 305-701, Korea 6Department of Emerging Materials Science, Daegu Gyeongbuk Institute of Science and Technology (DGIST), Daegu 711-873, Korea *e-mail: [email protected]; [email protected] We report the thermal spin injection and accumulation in crystalline CoFe/MgO tunnel contacts to n-type Si through Seebeck spin tunneling (SST). With the Joule heating (laser heating) of Si (CoFe), the thermally induced spin accumulation is detected by means of the Hanle effect for both polarities of the 1 temperature gradient across the tunnel contact. The magnitude of the thermal spin signal scales linearly with the heating power and its sign is reversed as we invert the temperature gradient, demonstrating the major features of SST and thermal spin accumulation. We observe that, for the Si (CoFe) heating, the thermal spin signal induced by SST corresponds to the majority (minority) spin accumulation in the Si. Based on a quantitative comparison of thermal and electrical spin signals, it is noted that the thermal spin injection through SST can be a viable approach for the efficient injection of spin accumulation. Understanding the interplay between heat and spin transport is a fundamental and intriguing subject that also offers unique possibilities for emerging electronics based on the combination of thermoelectrics and spintronics1-4. Especially in semiconductor (SC) spintronics5-8, where the injection, control, and detection of non- equilibrium spin populations (i.e. spin accumulation) in non-magnetic SCs are the main building blocks5-9, the functional use of heat provides a new route to inject and control the spin accumulation in SCs without a charge current4,7,8. Recently, Le Breton et al.10 introduced a conceptually new mechanism for the injection of spin accumulation (Δμ), in which a temperature difference (or heat flow) across a ferromagnet (FM)/oxide/SC tunnel contact can induce Δμ into SC through Seebeck spin tunneling (SST). It was found that the SST effect, involving the thermal transfer of the spin angular momentum from FM to SC without a tunneling charge current, is a purely interface-related phenomenon of the ferromagnetic tunnel contact and the resultant Δμ is governed by the energy derivative of its tunnel spin polarization (TSP)10. As shown in a recent report11, the SST and thermal spin accumulation (Δμth) 2 stem from the spin-dependent Seebeck coefficient of the ferromagnetic tunnel contact. It was also shown that the spin-dependent Seebeck coefficient gives rise to the tunnel magnetothermopower (or tunnel magneto Seebeck effect), i.e., the dependence of the thermopower of a magnetic tunnel junction (MTJ) on the relative magnetic configuration of the two FM electrodes, as observed in MgO-based12,13 and Al2O3- based14 MTJs. Therefore, the SST effect offers a notable thermal approach for the generation of Δμ in SCs as well as for the functional use of heat in spintronic devices. Whereas the SST and Δμth phenomena in p-type Si have been intensively studied using a Ni80Fe20/Al2O3 tunnel contact10, the demonstration of these phenomena in n- type Si with different tunnel contact materials has not been realized thus far. Furthermore, considering the fact that the sign and magnitude of the induced Δμth are crucially dependent on the energy derivative of TSP of the ferromagnetic tunnel interface10,11, the investigation of those in the different material system is highly desirable for a complete understanding of the SST mechanism and the successful implementation of its functionality in SC spintronics. Here, we report the achievement of thermal spin injection and accumulation in crystalline CoFe/MgO contacts to n-type Si through SST. Using the Joule heating (laser heating) of Si (CoFe) and Hanle measurements, we explicitly demonstrate the major features of SST and thermal spin accumulation that the magnitude of thermally injected spin signal scales linearly with the heating power and its sign is reversed when the temperature gradient across the tunnel contact is reversed. It is also observed that, for the Si (CoFe) heating, the thermal spin signal induced by the SST corresponds to the majority (minority) spin accumulation in the Si. Based on a quantitative comparison of the thermal and electrical spin signals, the thermal spin injection through SST is 3 suggested as an effective route to inject the spin accumulation. Figure 1 schematically illustrates the device geometry and measurement scheme used in the present study. We fabricated a device consisting of highly (001) textured Co70Fe30(5 nm)/MgO(2 nm) tunnel contacts to n-Si(001) in which the n-Si channel is heavily As-doped (nd ~ 2.5×1019 cm-3 at 300 K)15. A Si-active channel with a lateral dimension of 300×500 μm2 and three identical tunnel contacts (a-c, 100×300 μm2) were defined by standard photolithographic and Ar-ion beam etching techniques. These contacts are separated by about 100 μm from each other, which is much longer than the spin-diffusion length. The magnetic easy axis of the Co70Fe30 (hereafter abbreviated as CoFe) contacts are along the [110] direction of Si parallel to the long axes of the contacts. For electrical isolation (at the sides of the tunnel contacts) and for the contact pads, approximately 120-nm-thick Ta2O5(115 nm)/Al2O3(2 nm) layers and 110-nm-thick Au(100 nm)/Ti(10 nm) layers, respectively, were grown by a sputtering technique. Details of the sample preparation as well as the structural and electrical characterization are available in the literature15. To detect thermally-induced spin accumulation through the SST process, we generated the temperature difference (ΔT≡TSi–TCoFe) across the tunnel contact a employing two different heating methods (Joule heating and laser heating), which are the identical to those used in a previous study16. For the SC heating (Fig. 1(a)), we applied a heating current (Iheating) through the Si channel using two contacts b and c, which causes Joule heating and raises TSi with respect to TCoFe (ΔT>0). For the FM heating (Fig. 1(b)), the Au bond pad was heated using a laser beam with a wavelength of 532 nm and a maximum power of 200 mW. The laser beam spot (with a diameter of 5-10 µm and a skin depth of ~3 nm) on the 100-nm-thick Au pad is located approximately 300 µm from 4 one edge of the tunnel contact (note that the size of the Au pad is large enough and the position of the laser beam spot is far enough away to prevent the direct illumination of the Si layer). Hence, a part of the heat generated from the laser beam passes through the contact, resulting in TCoFe>TSi (ΔT<0). In an open-circuit geometry, where the tunneling charge current (Itunnel) is zero, the 10,11. The first term measured voltage between the contact a and d is given by V=Vth+ΔVTH Vth is the thermovoltage maintaining the zero net charge current (Itunnel=0) and the second term ΔVTH is the SST voltage due to the induced Δμth in the SC. The ΔVTH can be detected by means of the Hanle effect17,18. When we apply a magnetic field (Bz) transverse to the spins in the SC, the induced Δμth is suppressed via spin precession and dephasing. This results in a voltage change (ΔVTH), directly proportional to Δμth, with a Lorentzian line shape as a function of Bz. As noted in recent reports19,20, the induced Δμth can also be detected with an in-plane magnetic field (Bx) parallel to the film plane, giving rise to the inverted Hanle effect. Both the normal and inverted Hanle effects arising respectively from the suppression and recovery of the spin accumulation are indicative of the presence of spin accumulation. The full spin accumulation is given by the sum of the normal and inverted Hanle signals. Therefore, the two measurement schemes (Fig. 1) using the normal and inverted Hanle effects17-20 provide a concrete means of demonstrating the SST and resultant Δμth in the SC. Figure 2 shows the measured voltage changes (ΔVTH ∝Δμth) under perpendicular (Bz) and in-plane (Bx) magnetic fields corresponding to the normal and inverted Hanle effects, respectively, while heating the Si side (ΔT>0) as described above. As shown in the ΔVTH–Bz plots of Figs. 2(a) and 2(b), large normal Hanle signals (ΔVTH, normal) with a Lorentzian line shape were observed at 300 K (base T). The inverted Hanle signals 5 (ΔVTH, inverted)19,20 in Bx (Figs. 2(c) and 2(d)), roughly 1.3-1.4 times larger than the magnitude of ΔVTH, normal, were also clearly measured. It is worth noting that the line shapes of Hanle curves (ΔVTH) and the ratio of ΔVTH, inverted to ΔVTH, normal are consistent with those of electrical Hanle signals (ΔVEH) obtained by electrical spin injection/extraction using the same contact a21. The observed thermal Hanle signals (ΔVTH, normal, ΔVTH, inverted) show identical curve irrespective of the polarity in Iheating, indicating that the sign and magnitude of the induced Δμth in the Si are identical for both heating currents. Figures 2(e)-2(h) summarize the magnitude of the thermal Hanle signals (ΔVTH, normal, ΔVTH, inverted) as a function of Iheating (up to ±10 mA). These figures clearly show that ΔVTH, normal and ΔVTH, inverted scale quadratically (linearly) with Iheating (Iheating 2). These results strongly support that the observed Hanle signals mainly come from the thermally driven spin accumulation in Si, which scales linearly with the ΔT10,11. We performed a decisive test10 using the Hanle measurements in larger perpendicular magnetic fields (Bz) (see Fig. 1) to rule out any possibility that the observed spin accumulation is induced by the origin of the spin caloric effect (e.g., the spin Nernst effect)4 in Si. It is possible that the spin Nernst effect4 produces a spin current (along the z-axis) transverse to a thermal gradient (in the x- or y-axis direction) via the spin orbit interaction of Si (see Fig. 1). This would cause spin accumulation at the Si interface with the spins pointing parallel to the film plane (note that the direction of the accumulated spins is transverse to the thermal gradient and to the spin current). Figure 3(a) illustrates the expected characteristic behavior10 of the Δμth signal as a function of Bz for two different origins ((i) and (ii)) of Δμth. (i) If the spins were accumulated by origin of the spin Nernst effect in Si without thermal spin injection from 6 FM to Si, they would be unrelated to the magnetization (M) of FM10. This gives rise to a simple reduction of Δμth signal in small values of Bz (the blue line in Fig. 3(a)) because the direction of the induced spins is always orthogonal to Bz irrespective of the M of FM. (ii) In contrast, if the spins induced in Si were thermally injected from FM, their direction would coincide with that of the M of FM10,19,20 even with larger values of Bz. This results in distinctly different behavior of the Δμth signal with larger Bz values (the red line in Fig. 3(a)) as compared to the case (i). As Bz is increased, the Δμth value is sharply reduced due to the Hanle effect; thereafter, it gradually increases when the M of FM rotates out of the film plane. When the M and the induced spins in the Si are fully aligned with Bz higher than μ0Ms of FM (where μ0 is the permeability of the free space and Ms is the saturation magnetization), the Δμth signal eventually becomes saturated. Figure 3(b) shows the obtained thermal Hanle signals (ΔVTH, normal, ΔVTH, inverted; normalized) under applied magnetic fields (Bz, Bx) up to 50 kOe with an Iheating value of ±10 mA (note that each Hanle curve is the average of two curves measured with positive and negative Iheating values). The ΔVTH, normal signal (closed symbol) clearly reveals the distinctive behavior with Bz; the Hanle effect in a small Bz (<2 kOe), the increase of ΔVTH, normal due to the rotation of M in an intermediate Bz (2-22 kOe), and the saturation of ΔVTH, normal in a large Bz (>22 kOe). This result excludes any possible origin for thermal spin accumulation in our system that does not involve spin injection from CoFe to Si. It was noted earlier that the difference of the ΔVTH, normal value in the saturation region from the peak value at zero Bz is associated with the initial spin precession and dephasing effect caused by the local magnetostatic fields due to the finite roughness of the CoFe/MgO interface, as proven by the inverted Hanle signal (ΔVTH, inverted, open symbol) under an in-plane magnetic field (Bx)19,20. 7 The sign of Δμth can be determined by a direct comparison10,19 with that of electrical Hanle signals (ΔVEH, normal, ΔVEH, inverted) obtained from three-terminal Hanle (TTH) measurements22-24 for the same contact a21.The sign of ΔVEH, normal in Bz is negative for a negative electrical voltage (Vel) (VCoFe–VSi<0, electron spin injection) and positive for a positive Vel (VCoFe–VSi>0, electron spin extraction). As shown in Figs. 2(e) and 2(g), the sign of ΔVTH, normal (in Bz) is identical to that of the former ΔVEH, normal (Vel<0), indicating that Δμth produced by the thermal spin injection with Si heating (TSi>TCoFe) has the same sign as Δμel induced by electrical spin injection. Given the positive TSP of bcc (Co)Fe/MgO(001) tunnel interfaces25,26, the induced Δμth by SST when TSi>TCoFe corresponds to the majority spin accumulation (Δμ>0) in the Si, where a larger number of electrons have spin angular momentum parallel to the direction of the M of FM. Based on SST theory10,11, the positive sign of Δμth (or majority spin accumulation) for the SC heating indicates that the sign of the energy-derivative of the TSP of the CoFe/MgO tunnel interface at the Fermi-level (EF) is negative. This corresponds to the case in which the TSP value of the CoFe/MgO interface varies weakly for energies below EF but decays significantly for energies above EF. It should also be noted that the same sign of the experimental results (Δμ>0 for SC heating) was observed in n-type Ge with Co70Fe30/MgO19 and Co60Fe20B20/MgO27 contacts and in p- type Si with a Ni80Fe20/Al2O3 contact10. We estimated that the associated Δμth in the CoFe/MgO tunnel contact to n-type Si is (+)0.37 meV with the maximum Iheating (±10 mA) through the typical conversion formula of Δμth=(–2e)ΔVTH/γ10,11,22-24 based on the existing theory9,28. In this calculation, we used the total measured ΔVTH of (–)0.13 mV, given by the sum of ΔVTH, normal and ΔVTH, inverted, and the assumed TSP (γ) of (+)0.7 for a crystalline CoFe/MgO tunnel 8 interface24,25. Another major feature of the SST and Δμth is that, for a given energy-dependence of TSP of the ferromagnetic tunnel contact, the sign of the thermal spin signal is reversed when ΔT is reversed10,11. In order to demonstrate this, we employed a laser beam to heat the FM instead of Joule heating16. This approach provides a simple way to heat the FM (TCoFe>TSi, ΔT<0) effectively and minimize the contribution of spurious effects such as current-in-plane (CIP) tunneling and anisotropic magnetoresistance (AMR) on the Hanle signal. Figures 4(a) and 4(b) show the measured voltage signals (V) as a function of perpendicular (Bz, Fig. 4(a)) and in-plane (Bx, Fig. 4(b)) magnetic fields while varying the incident laser power (Pinc. laser) from 0 to 100 mW at 300 K (base T). At zero Pinc. laser (black symbols), no magnetic field dependence of the signal is detected, as expected. With an increase of Pinc. laser from 20 to 100 mW (purple to red symbols), clear normal and inverted Hanle signals are observed and the amplitudes of both Hanle signals gradually increase. As depicted in the ΔVTH–Pinc. laser plots (Figs. 4(c) and 4(d)), the obtained Hanle signals (ΔVTH, normal, ΔVTH, inverted) scale almost linearly with Pinc. laser, indicating that the obtained signals stem from the thermal spin injection and accumulation in Si. More importantly, the observed sign of ΔVTH is obviously reversed with respect to the Si heating case. The ΔVTH, normal (ΔVTH, inverted) value is positive (negative) for TCoFe>TSi (see Figs. 4(c) and 4(d)) whereas it is negative (positive) for TCoFe<TSi (see Figs. 2(e) and 2(f)). This result clearly demonstrates another key feature of SST and Δμth that the thermal spin signal is reversed when ΔT is reversed10,11. In addition, an analysis based on the sign of Hanle signals allows us to exclude 9 another possible origin of the spin signal, such as the spin-polarized hot-electron injection via conventional Seebeck effect, in the laser-heating experiment. Thermal excitation by laser heating can produce an electron flow in the Au pad and the CoFe layer. Nevertheless, this hot-electron flow cannot be an origin of the obtained Hanle signals, since the observed sign of ΔVTH, normal (a positive sign, minority spin accumulation; see Fig. 4(c)) with the laser heating of CoFe is opposite to the expected sign of ΔVEH, normal (a negative sign, majority spin accumulation) due to the injection of the spin-polarized hot- electron into Si, whose transmission in FM is known to be larger for majority spins than for minority spins29. This strongly suggests that the obtained Hanle signals with laser heating do not arise from the hot-electron flow in the ferromagnetic tunnel contact. For a quantitative analysis and for a comparison study, we plotted the magnitude of the Hanle signals (ΔVTH, ΔVEH) obtained in the same contact a (Fig. 1) caused by the thermal and electrical spin accumulation (see Figs. 4(a) and S(c)) as a function of the driving term. The thermal driving term is S0ΔT (=Vth)11, which should be compared to the electrical driving term of RcontactItunnel (=Vel)11. Here, S0 is the charge thermopower (or charge Seebeck coefficient), Rcontact is the contact resistance, and Itunnel is the tunneling current across the contact. Figures 5(a) and 5(b), respectively, show the ΔVTH, normal–Vth, eff and ΔVEH, normal–Vel plots. It is noted that, because the reference contact d in our measurement scheme is located far away from the ferromagnetic tunnel contact a (see Fig. 1), the offset voltage (or VB=0–ΔVTH) in Fig. 4(a), which mainly arises from the Seebeck effect across the MgO tunnel barrier and partly from the Seebeck effects inside the Au and n-Si electrodes, should be considered as an effective value for the Vth (or Vth, 10 eff). A noteworthy aspect of these plots is that the thermal spin accumulation (red symbol in Fig. 5(a)) requires about 100 times smaller value of Vth, eff to obtain a similar magnitude of ΔVHanle, normal compared to the electrical spin accumulation case (blue symbol in Fig. 5(b)). This means that the proportionality factor between the spin accumulation and the driving term for the thermal spin injection is much larger than that for the electrical spin injection. According to the model11, the proportionality factor of the electrical spin injection is limited by the absolute value of TSP, which cannot be larger than one (or 100 %) by definition. However, such a restriction does not exist for the proportionality factor of the thermal spin injection, which is governed by the energy derivative of TSP, since there is no limit for the energy derivative of TSP (note that, in principle, the proportionality factor for thermal spin accumulation can be arbitrarily large for suitably engineered ferromagnetic tunnel contacts)11. Therefore, the experimental result supports the theoretical proposition11 that, for a ferromagnetic tunnel contact with large energy-derivative of TSP and proper thermal interface resistance, the thermal spin injection through SST can be an effective way to inject spin accumulation into SCs. In conclusion, we have achieved the thermal spin injection and accumulation in crystalline CoFe/MgO contacts to n-type Si through SST. With the Joule heating (laser heating) of Si (CoFe) and subsequent Hanle measurements, we have demonstrated the major features of SST and thermal spin accumulation that the magnitude of a thermally injected spin signal scales linearly with the heating power and its sign is reversed as the temperature gradient across the tunnel contact is inverted. A quantitative analysis and a comparison study of the thermal and electrical spin signals suggest that the thermal spin injection through SST is a viable approach for the efficient injection of the spin 11 accumulation. This work was supported by the DGIST R&D Program of the Ministry of Education, Science and Technology of Korea (11-IT-01); by the KIST institutional program (2E22732 and 2V02720) and by the Pioneer Research Center Program (2011- 0027905); and by the KBSI grant No.T33517 for S-YP. The authors would like to thank H. Saito and R. Jansen at AIST for the high-field measurements by PPMS and for the helpful discussions. One of the authors (K-RJ) acknowledges a JSPS Postdoctoral Fellowship for Foreign Researchers. References 1M. Johnson and R. H. Silsbee, Phys. Rev. B 35, 4959 (1987). 2M. Hatami, G. E. W. Bauer, Q. Zhang, and P. J. Kelly, Phys. Rev. Lett. 99, 066603 (2007). 3G. E. W. Bauer, A. H. MacDonald, and S. Maekawa, Solid State Commun. 150, 459 (2010). 4G. E. W. Bauer, E. Saitoh, and B. J. van Wees, Nature Mater. 11, 391 (2012). 5I. Žutić, J. Fabian, and S. Das Sarma, Rev. Mod. Phys. 76, 323 (2004). 6A. Fert, Rev. Mod. Phys. 80, 1517 (2008). 7R. Jansen, Nature Mater. 11, 400 (2012). 8R. Jansen, S. P. Dash, S. Sharma, and B. C. Min, Semicond. Sci. Technol. 27, 083001 (2012). 9A. Fert and H. Jaffrès, Phys. Rev. B 64, 184420 (2001). 10J. C. Le Breton, S. Sharma, H. Saito, S. Yuasa, and R. Jansen, Nature 475, 82 (2011). 12 11R. Jansen, A. M. Deac, H. Saito, and S. Yuasa, Phys. Rev. B 85, 094401 (2012). 12M. Walter, J. Walowski, V. Zbarsky, M. Munzenberg, M. Schafers, D. Ebke, G. Reiss, A. Thomas, P. Peretzki, M. Seibt, J. S. Moodera, M. Czerner, M. Bachmann, and C. Heiliger, Nat. Mater. 10, 742 (2011). 13N. Liebing, S. Serrano-Guisan, K. Rott, G. Reiss, J. Langer, B. Ocker, and H. W. Schumacher, Phys. Rev. Lett. 107, 177201 (2011). 14W. Lin, M. Hehn, L. Chaput, B. Negulescu, S. Andrieu, F. Montaigne, and S. Mangin, Nat. Commun. 3, 744 (2012). 15K. R. Jeon, B. C. Min, I. J. Shin, C. Y. Park, H. S. Lee, Y. H. Jo, and S. C. Shin, Appl. Phys. Lett. 98, 262102 (2011); K. R. Jeon, C. Y. Park, and S. C. Shin, Cryst. Growth Des. 10, 1346 (2010). 16K. R. Jeon, B. C. Min, S. Y. Park, K. D. Lee, H. S. Song, Y. H. Park Y. H. Jo, and S. C. Shin, Sci. Rep. 2, 962 (2012). 17M. Johnson and R. H. Silsbee, Phys. Rev. Lett. 55, 1790 (1985). 18M. Johnson and R. H. Silsbee, Phys. Rev. B 37, 5326 (1988). 19S. P. Dash, S. Sharma, J. C. Le Breton, J. Peiro, H. Jaffr`es, J.-M. George, A. Lemaıtre, and R. Jansen, Phys. Rev. B 84, 054410 (2011). 20K. R. Jeon, B. C. Min, Y. H. Park, Y. H. Jo, S. Y. Park, C. Y. Park, and S. C. Shin, Appl. Phys. Lett. 101, 022401 (2012); K. R. Jeon, B. C. Min, Y. H. Park, S. Y. Park, and S. C. Shin, Phys. Rev. B 87, 195311 (2013). 21See supplementary material at [URL will be inserted by AIP] for the analysis on the electrical Hanle signals obtained by electrical spin injection/extraction using the same contact. 22X. Lou, C. Adelmann, M. Furis, S. A. Crooker, C. J. Palmstrøm, and P. A. Crowell, 13 Phys. Rev. Lett. 96, 176603 (2006). 23S. P. Dash, S. Sharma, R. S. Patel, M. P. de Jong, and R. Jansen, Nature 462, 491 (2009). 24C. H. Li, O. M. J. van ‘t Erve and B. T. Jonker, Nat. Commun. 2, 245 (2011). 25S. S. P. Parkin, C. Kaiser, A. Panchula, P. M. Rice, B. Hughes, M. Samant and S. H. Yang, Nature Mater. 3, 862 (2004). 26S. Yuasa, T. Nagahama, A. Fukushima, Y. Suzuki, and K. Ando, Nature Mater. 3, 868 (2004). 27A. Jain, C. Vergnaud, J. Peiro, J. C. Le Breton, E. Prestat, L. Louahadj, C. Portemont, C. Ducruet, V. Baltz, A. Marty, A. Barski, P. Bayle-Guillemaud, L. Vila, J.-P. Attané, E. Augendre, H. Jaffrès, J.-M. George, and M. Jamet, Appl. Phys. Lett. 101, 022402 (2012). 28M. Tran, H. Jaffr`es, C. Deranlot, J. -M. George, A. Fert, A. Miard, and A. Lemaıtre, Phys. Rev. Lett. 102, 036601 (2009). 29R. Jansen, J. Phys. D: Appl. Phys. 36, R289 (2003). Figure captions FIG. 1. (Color online) Schematic illustration of the device geometry and measurement scheme for Seebeck spin tunneling and thermal spin accumulation with (a) the Joule heating of Si and (b) the laser heating of CoFe. FIG. 2. (Color online) (a)-(d) Obtained thermal Hanle signals (ΔVTH, normal, ΔVTH, inverted) under applied magnetic fields (Bz, closed circles; Bx, open circles) for heating currents (Iheating) of ±5 and ±10 mA at 300 K (base T), in the case of Si heating (ΔT>0). (a)/(b) 14 and (c)/(d) represent ΔVTH, normal in Bz and ΔVTH, inverted in Bx for an Iheating value of ±5/±10 mA, respectively. (e)-(h) Summary of the magnitudes of ΔVTH, normal and ΔVTH, inverted as a function of Iheating (Iheating 2). (e) ΔVTH, normal and (f) ΔVTH, inverted with Iheating, together with a quadratic fit. (g) ΔVTH, normal and (h) ΔVTH, inverted with Iheating 2, together with a linear fit. FIG. 3. (Color online) (a) Expected characteristic behavior of the thermal spin accumulation (Δμth) signal as a function of the perpendicular magnetic field (Bz) for two different origins ((i) and (ii)) of Δμth: (i) Δμth by thermal spin injection from CoFe to Si (red line) and (ii) Δμth without thermal spin injection from FM to Si (blue line). (b) Obtained thermal Hanle signals (ΔVTH, normal, ΔVTH, inverted; normalized) under applied magnetic fields (Bz, Bx) up to 50 kOe with the heating current (Iheating) of ±10 mA. Note that each Hanle curve is the average of two curves measured with positive and negative Iheating values. FIG. 4. (Color online) (a) Obtained thermal spin signals (ΔVTH, normal, ΔVTH, inverted) under (a) perpendicular (Bz) and (b) in-plane (Bx) magnetic fields as a function of the incident laser power (Pinc. laser) at 300 K (base T) in the case of CoFe heating (ΔT<0). (c) ΔVTH, normal and (d) ΔVTH, inverted as a function of Pinc. laser, together with a linear fit. FIG. 5. (Color online) (a) Hanle signal (ΔVTH) obtained by thermal spin accumulation with the laser heating of CoFe (see Fig. 4(a)) as a function of the effective thermovoltage (Vth, eff), defined as VB=0–ΔVTH. (b) Hanle signal (ΔVTH) obtained by electrical spin accumulation using the same contact a (see Fig. S(c)) as a function of the 15 electrical voltage (Vel), defined as VB=0–ΔVEH. 16 (a) Ta2O5/Al2O3 2O5/A 5/A Al OOAl222O222O33 Iheating + + V - d dddddddddddd b bbbb aaa MM M M Au/Ti AuAuAuAu AAAA c ccccc CoFe MgO n-Si 5 nm 5 2 nm 2 ~100 nm 00~1 z ~500 µm ~5 B y y y x x x (b) + + ++ V - - ~300 µm ~3000~3300000 µmmµmµmmmmmmm 3000 µmm Ta2O5/Al2O3 2O5 5/AAl222O222O33 d dddddddddddddddddd SC heating FM heating Laser heating La Laser beam spot Au/Ti AuAuAuAu a aa M M MM 100 nm Fig. 1 ) V m ( f f o V - V , H T V ' 0.00 -0.02 -0.04 -0.06 -0.08 -0.10 -0.12 0.00 -0.02 -0.04 -0.06 -0.08 -0.10 -0.12 T =300 K, base V (+)= B=0 +1.6 mV, V (-)= B=0 -0.6 mV I heating I heating =+5 mA =-5 mA (a) ΔVTH, normal T =300 K, base V (+)= B=0 +3.3 mV, (-)= V B=0 -5.1 mV I heating I heating =+10 mA =-10 mA -4 -2 0 B (kOe) z 2 (b) 4 ΔVTH, normal 0.15 0.12 0.09 (c) I heating I heating =+5 mA =-5 mA ) V m ( t a s V - V , H T V ' 0.06 0.03 0.00 0.15 0.12 0.09 ΔVTH, inverted I heating I heating =+10 mA =-10 mA (d) 0.06 0.03 0.00 ΔVTH, inverted 0.00 ) V m ( H T V ' -0.02 -0.04 -0.06 (e) ) V m ( H T V ' 0.10 0.08 0.06 0.04 0.02 0.00 'V TH, normal Fit 'V aI2 heating -12 -8 -4 I heating 0 4 (mA) 8 12 -12 -8 0.00 ) V m ( H T V ' -0.02 -0.04 -0.06 (g) ) V m ( H T V ' 0.10 0.08 0.06 0.04 0.02 0.00 'V TH, normal Fit 'V aP heating 'V TH, inverted Fit 'V aI2 heating (f) 8 12 -4 I heating 0 4 (mA) 'V TH, inverted Fit 'V aP heating -4 -2 0 B (kOe) x 2 4 0 30 I heating 60 2 (mA2) 90 0 30 I heating 60 2 (mA2) (h) 90 Fig. 2 1.2 1.0 0.8 0.6 ) s t i n u . b r a ( h t μ Δ 0.4 0.2 0.0 Δμth by spin injection from FM Δμth w/o spin injection from FM Increase of Δμth due to rotation of M No Increase of Δμth Inverted Hanle effect (Bx//M) Normal Hanle effect (a) ) d e z i l a m r o n ( H T V ' 1.2 1.0 0.8 0.6 0.4 0.2 0.0 T =300 K, I base =+/-10 mA heating Rotation of M Inverted Hanle effect (Bx//M) Normal Hanle effect (b) -2 -1 0 1 Bz/μ0Ms (arb. units) 2 -40 -20 20 40 0 B (kOe) z Fig. 3 Normal Hanle effect (a) P = inc. laser 100 mW 90 80 70 60 50 40 30 20 0 T base =300 K 225 200 175 150 125 100 75 50 25 0 ) V P ( V 225 200 175 150 ) V P ( V 125 100 75 50 25 0 -6 -4 -2 0 2 4 6 B (kOe) z -6 -4 -2 0 2 4 6 B (kOe) x Inverted Hanle effect (b) 100 12 'V TH, normal Linear fit 90 80 70 60 50 40 30 20 0 T base =300 K ) V P ( H T V ' ) V P ( H T V ' 8 4 0 0 -4 -8 (c) 20 40 60 80 100 P (mW ) inc. laser 0 (d) -12 'V TH, inverted Linear fit 0 20 40 60 80 100 P (mW ) inc. laser Fig. 4 0.04 0.02 0.00 ) V m ( H T V ' -0.02 -0.04 Majority spin accumulation Minority spin accumulation Majority spin accumulation Minority spin accumulation 'V EH, normal 'V TH, normal Linear fit Laser heating 0.04 0.02 0.00 ) V m ( H E V ' -0.02 (a) -1.0 T =300 K base -0.04 (b) T=300 K -0.5 V th, eff 0.0 (mV) 0.5 1.0 -140 -70 0 V (mV) el 70 140 Fig. 5
1511.08989	1	1511	2015-11-29T09:59:30	A Variational Approach to Extracting the Phonon Mean Free Path Distribution from the Spectral Boltzmann Transport Equation	[ "cond-mat.mes-hall" ]	The phonon Boltzmann transport equation (BTE) is a powerful tool for studying non-diffusive thermal transport. Here, we develop a new universal variational approach to solving the BTE that enables extraction of phonon mean free path (MFP) distributions from experiments exploring non-diffusive transport. By utilizing the known Fourier solution as a trial function, we present a direct approach to calculating the effective thermal conductivity from the BTE. We demonstrate this technique on the transient thermal grating (TTG) experiment, which is a useful tool for studying non-diffusive thermal transport and probing the mean free path (MFP) distribution of materials. We obtain a closed form expression for a suppression function that is materials dependent, successfully addressing the non-universality of the suppression function used in the past, while providing a general approach to studying thermal properties in the non-diffusive regime.	cond-mat.mes-hall	cond-mat	A Variational Approach to Extracting the Phonon Mean Free Path Distribution from the Spectral Boltzmann Transport Equation Vazrik Chiloyana, Lingping Zenga, Samuel Hubermana, Alexei A. Maznevb, Keith A. Nelsonb, Gang Chena1 aDepartment of Mechanical Engineering, Massachusetts Institute of Technology, Cambridge, Massachusetts 02139, USA bDepartment of Chemistry, Massachusetts Institute of Technology, Cambridge, Massachusetts 02139, USA Abstract The phonon Boltzmann transport equation (BTE) is a powerful tool for studying non-diffusive thermal transport. Here, we develop a new universal variational approach to solving the BTE that enables extraction of phonon mean free path (MFP) distributions from experiments exploring non-diffusive transport. By utilizing the known Fourier solution as a trial function, we present a direct approach to calculating the effective thermal conductivity from the BTE. We demonstrate this technique on the transient thermal grating (TTG) experiment, which is a useful tool for studying non-diffusive thermal transport and probing the mean free path (MFP) distribution of materials. We obtain a closed form expression for a suppression function that is materials dependent, successfully addressing the non-universality of the suppression function used in the past, while providing a general approach to studying thermal properties in the non-diffusive regime. Corresponding author: [email protected] 1 The Boltzmann transport equation (BTE) is widely used in analyzing heat transfer at length scales and time scales for which Fourier’s law breaks down. In particular, there has been a growing interest recently in developing numerical and analytical solutions to the BTE to model thermal transport in phonon spectroscopy experiments [1–11] to extract phonon mean free path (MFP) distribution. The thermal conductivity accumulation function has been utilized as an elegant metric for understanding which MFP phonons contribute predominantly to thermal transport in a material [12–14]. Various experimental tools such as time-domain thermoreflectance [4,6,8,15,16] (TDTR), frequency-domain thermoreflectance [17,18] (FDTR), and transient thermal grating [3,5,7,19] (TTG) techniques have been utilized extensively recently in order to probe and observe non-diffusive transport by using ultrafast time scales or ultrashort length scales and gain key insight into the material’s MFP spectrum. When the length scales in a system become comparable to the MFPs in a material, the effective thermal conductivity is reduced compared to its bulk, diffusive limit value [20,21]. A suppression function S! is used to quantify this reduction or suppression of thermal conductivity, defined as keff = 1 3 !m" 0 C!v!!!S!d! . The suppression function provides the ability to extend the notion of thermal conductivity beyond the diffusive regime in which it is defined from Fourier’s law [20,22]. By utilizing the suppression function for a given experimental geometry, one can obtain the material’s phonon MFP distribution from the experimentally measured thermal conductivity [5,6,22]. To obtain the effective thermal conductivity, the thermal signal from the experiment is fitted to the results of the Fourier law. The suppression function is calculated through modeling of the given experimental geometry with the BTE. However, one key assumption in this method is the universality of the suppression function, i.e. the ability to 2 express the suppression function as S! = S !! / L ( ) so that it depends only on the ratio of MFP to a characteristic length for a given experimental configuration, but not otherwise on the material properties. This assumption allows one to obtain effective thermal conductivities by solving for the suppression function from the gray BTE, i.e. the BTE equation with a single MFP [23]. This assumption has been shown to be not strictly valid in the past [23,24] and will be further shown in this work, with an approach that addresses this shortcoming. The BTE is notoriously difficult to solve, especially for complex geometries, which presents difficulty in calculating the effective thermal conductivity of materials in a given experimental geometry. So far almost exclusively, numerical solutions are implemented that directly attempt to solve the BTE, and are then fitted to the Fourier solution to extract the effective thermal conductivity and corresponding suppression function for the experimental geometry. Experimental methods have also been utilized that rely on first-principles material property data to obtain a calibrated suppression function [6]. The key insight in our work here is to utilize the temperature distribution obtained from the Fourier diffusion equation directly in the BTE for the given experimental geometry to facilitate, hence significantly simplify, its solution. Furthermore, we develop a variational approach to yield solutions to the spectral BTE. By utilizing the temperature field derived from the Fourier solution and the variational method, we obtain solutions that are both simple yet can reproduce the exact numerical results from the BTE in terms of obtaining the effective thermal conductivity. Our approach provides a more direct, universal methodology for extracting the effective thermal conductivity and corresponding suppression function to enable the extraction of intrinsic material properties such as the phonon MFP distribution from non-diffusive experiments. 3 The variational approach utilized here for the BTE is analogous to the variational method in quantum mechanics, used for improving one’s trial solution for the ground state energy of a given system [25,26]. The variational principle has been applied to the BTE previously in calculating the cross plane heat flux in a thin film [27]. Allen [27] utilized a specific error metric, one that tries to best enforce uniform heat flux through the slab to ensure energy conservation, to approximately calculate the thermal flux between a hot wall and cold wall. Furthermore, variational techniques have been applied to solving the BTE at the initial stage of the partial differential equation itself [28,29]. In this approach, first described by Ziman [30], the partial differential equation for the phonon distribution function is solved utilizing the variational principle, and the variational parameter is calculated by optimizing the entropy. In this work, we develop the application of the variational principle upon the integral equation for the temperature profile, derived from the BTE, and solve for the variational parameter by minimizing the residual error in the equation. Although we anticipate that this approach can be applied also directly to the BTE, the technique developed here is applied at the stage of the temperature equation for two reasons. One being that analytically solving the BTE up to the temperature equation decreases inaccuracies that can build up from utilizing approximations earlier on in the solution. Second, the temperature equation allows for the direct utilization of the Fourier solution as the trial function, with effective thermal conductivity (or other properties such as interfacial resistance or boundary temperature slip) as the parameters used to minimize the error of the variational trial function. 4 In this work, we apply the variational technique for the one-dimensional TTG experimental geometry as an example. In the TTG experiment, two laser beams are crossed in order to generate a sinusoidal heating profile on a sample, with a spatial periodicity of length !. Once heated, the sample is allowed to relax and the thermal decay profile is measured to yield information about the transport within the material. At grating periods on the order of micrometers, non-diffusive transport has been observed [3,7,19]. Given the success of this experiment in probing non-diffusive transport and the opportunity to yield MFP data using reconstruction techniques that have been developed [31], the ability to model this experiment is critical. Furthermore, the relative simplicity of the geometry makes it more accessible for theoretical modeling. The TTG in the one-dimensional case has been studied in a two-fluid framework, and with simplifying assumptions about the scattering of high and low frequency phonons, an analytical suppression function has been calculated [19] and utilized in MFP reconstruction [31], but there is a concern that this model is only valid at the onset of non-diffusive transport. Collins et al. solved the problem with a numerical approach to obtain the exact solution both in the gray case as well as the full spectral case for the BTE for Si and PbSe [23]. Deviation of the two-fluid model from the exact numerical solution was shown for PbSe [23]. Hua and Minnich obtained the Fourier transform of the thermal decay analytically, and were able to recover the two-fluid model suppression function in the weakly non-diffusive limit [24]. However, there is no closed form expression for the thermal decay rate ! and the suppression function S that matches numerical results. 5 Utilizing the notation by Collins et al., we begin with the spectral BTE in one dimension under the relaxation time approximation [23] : !g! !t +µv! !g! !x = g0 " g! "! (1) where g! is the phonon energy density per unit frequency interval per unit solid angle above the reference background energy, related to the distribution function and density of states as g! = !!D !( ) 4! f! ! f0 T0( ( ) . µ is the direction cosine, v! is the group velocity, !" is the ) relaxation time, and g0 is the equilibrium energy density, given by g0 ! 1 4! C! T "T0 ( ) in the linear response regime. In the TTG experiment, the temperature initially has a sinusoidal profile ) = T0 +h t( )Tmeiqx in complex form where Tm is the initial amplitude and in general obeys T x,t of the spatial variation, q = 2!/" is the grating wavevector, and h t( ) is the non-dimensional ( temperature that describes the decay of the initial temperature profile. Solving Eq. (1) and utilizing the equilibrium condition in the spectral case [32] to close the problem yields the integral equation for the non-dimensional temperature obtained previously [23,24]: h t( ) !m! 0 C! "! d! = !m! 0 C! "! b! t( )d! + t! 0 h "t( ) !m! 0 C! 2 b! t # "t "! ( )d! d "t (2) where we have defined for simplicity b! t( ) ! e" !"sinc qv!t ( t ) . This integral equation is easily solved with a Laplace transform, and the temperature profile can be solved for with an inverse transform as obtained by Hua and Minnich [24]. Other methods to solving the BTE is to either obtain a numerical solution by solving the integral equation by finite differences [8,23], or by utilizing Monte Carlo techniques [31,33,34]. We depart from these established approaches by 6 treating the unknown temperature distribution as a variational function and rewrite Eq. (2) by shifting all terms to one side of the equation to define: H t( ) = !m! 0 C! "! b! t( )d! "h t( ) !m! 0 C! "! d! + t! 0 h #t( ) !m! 0 C! 2 b! t " #t "! ( )d! d #t (3) If the function we guess for the temperature profile is the exact temperature profile that solves the BTE, then this function will be identically zero everywhere. The function H has a physical interpretation as it has been defined from the integral equation which comes from the equilibrium condition of the BTE; it represents the error in energy conservation, and can be thought of as an artificial heat source/sink (up to constant factors such as 4! and Tm ). In the exact case it should be zero everywhere, but since our trial function will not be the exact solution, we would like to optimize the function that makes H t( ) as close to zero as possible to minimize the error in our trial solution. The optimization procedure can be done in several ways. One method is to mathematically define an error metric and minimize the error in order to calculate the variational parameter. Some common examples of error metrics are least squares Er !( ) = !" 0 H 2 t( )dt and least absolute error Er !( ) = !" 0 H t( ) dt . Another approach is to require certain physical conditions to be met, and we can impose one physical condition for every variational parameter available in the trial solution. In this work, we use the following simple physical constraint to solve the variational parameter: !" 0 H t( )dt = 0 (4) 7 Since H represents an artificial heat generation rate, the integral in Eq. (4) represents the total net heat generated over all time and solving it to be zero imposes the requirement that even if the energy conservation does not exactly hold at every moment in time as it does for the exact solution, it is satisfied over the entire decay time. The key point is to optimize the trial function, either by imposing physical conditions to be met or mathematical error functions to be minimized and there are various ways to do so. Typically in the variational approach, one uses a trial function that is known from intuition about the system, and the trial function is optimized to minimize the chosen error function. In solving the BTE with this variational approach, the Fourier solution provides this trial function, especially since our goal is to extract the effective thermal conductivity (or in other cases properties such as interfacial thermal resistance, diffusivity, etc.) of the system. For the one-dimensional TTG, the exact temperature solution of the diffusion equation is T x,t ) = T0 +Tmeiqxe!!q2t where ! is the thermal diffusivity. Therefore, we take for the trial ( function h t( ) = e!!t with !eff ="/ q2 . The elegance of this approach is that it immediately utilizes the Fourier temperature field appropriately modified as an input, and optimizes to find the modified properties such as effective thermal conductivity that solves the BTE with minimized error, converting the difficult task of solving an integral equation for the temperature distribution into a simple task of performing integration. We note that in other heat transfer configurations, there can be multiple parameters in the trial solution, such as temperature slip that can occur due to the boundary resistance as well as effective thermal conductivity in the 8 non-diffusive regime. Here, we demonstrate the technique on this simple case where only one variational parameter will be needed for simplicity. Using the trial function, we can solve the condition of Eq. (4) to obtain for the thermal decay rate: != d" !m( 0 1 C" " 1! arctan $"( $ $" #" # 1 arctan "!( d!C! "! ) !m( 0 % ) ' & (5) where we have defined the non-dimensional Knudsen number !" = q!" = 2#!" /$ where ! is the grating period. As described, while different mathematical error metrics can be chosen or different physical conditions, we will show that the simple physical constraint of Eq. (4) does an excellent job in recovering the exact numerical results. We utilize the normalized effective thermal conductivity for simplicity, defined as keff kbulk = C! q2kbulk . The bulk thermal conductivity is given by kbulk = 1 3 !m" 0 C!v!!!d! . In Fig. 1, we compare results obtained for Si and PbSe, for which previous approaches for obtaining the effective thermal conductivity include assuming a constant MFP distribution [23], an exact numerical solution [23,24], and the two-fluid model [19]. The spectral numerical results are obtained by fitting the exact solution of Eq. (2), obtained by finite differences, to the Fourier exponential profile [23]. The spectral variational results are plotted from Eq. (5). The gray variational results are obtained by taking the gray limit of Eq. (5), extracting a gray suppression function, and 9 inputting into the effective thermal conductivity, i.e. keff , gray = 1 3 !m" 0 C!v!!!Sgray "!( )d! . Note that this gray approximation is identical to the ‘frequency integrated gray medium’ approach performed numerically by Collins et al. [23]. Silicon is known to have a wide range of MFPs, and shows that the gray solution derived suppression function does a poor job in reproducing the exact numerical results. We also look at PbSe, which has a narrower range of MFPs. Note that the optimized solution agrees excellently with the exact numerical solution, which demonstrates the predictive power of the variational approach. Here the two-fluid model deviates from the exact solution at smaller grating periods. The gray suppression function performs better for this material due to its narrower range of MFP’s as compared to silicon [23]. FIG. 1 Effective thermal conductivity of silicon (a) and PbSe (b). Here the effective thermal conductivity is plotted to compare the variational technique with the exact numerical technique and various approximations. The variational technique for the full spectral BTE demonstrates excellent agreement with the exact numerical solution. From the definition of the effective thermal conductivity and the thermal decay rate of Eq. (5), we extract the suppression function: 10 S! = arctan "!( 3 " 2 1! $ "! # C#C d# #m( 0 % ) ' !" & arctan !#( !" (6) ) where C is the heat capacity obtained by integrating the spectral heat capacity C = !m! 0 C!d! . We note that although the numerator is dependent only on the ratio of MFP to the grating spacing and hence universal, the denominator depends in general on the material properties, This result is significant not only because it shows the suppression function is not universally dependent on a ratio of MFP to a length in the system, but also because we now have a way to properly address this problem analytically and more generally, numerically. The numerator is equal to the suppression function previously derived by Maznev et al. [19] and has been called the weakly quasiballistic suppression function [24]. Hua & Minnich have shown that there is in fact a correction to the suppression function in the full form, but their expression depends on the thermal decay time which intrinsically depends on the temperature solution to the BTE [24]. Our optimized solution provides the suppression function and illuminates its dependence on the grating period as well as its material property dependence. Furthermore, we can determine the validity domain of the two-fluid model by comparing the denominator of the optimized expression to unity. Thus, the following quantitative metric is obtained for the validity of the two-fluid model, d! "m" 0 C!C arctan "!( ) !" !1 <<1. One could Taylor the suppression function of Eq. (6) expand for large values of the grating period relative to MFP to get an expression that is a first order correction to the two-fluid approximation for the thermal decay rate. In Fig. 2, we show the denominator of Eq. (6) for both Si and PbSe. We find that the two-fluid model can predict the effective thermal conductivity of Si with less than 5% error for grating spacings of 1 11 micron or higher. For PbSe, the two-fluid model has less than 5% error for grating spacings of 0.1 micron or higher. The cutoff grating spacing is larger for Si than for PbSe because Si has a MFP distribution that has a larger maximum MFP value than for PbSe, as shown in Fig. 1, hence demonstrates earlier deviation from the exact result. FIG. 2: Denominator of optimized spectral suppression function for Si (red) and PbSe (blue), yielding a metric for the validity of the two-fluid model. The variational method can, of course, be applied to the gray case, for which we extract a suppression function that only depends on the Knudsen number. We take Eq. (6) and assume a constant MFP distribution to obtain: Sgray !"( ) = 3 " 2 1! $ !" # arctan !"( ) !" % ' & ( * ) +* !" arctan !"( , * - .* ) (7) The term in the curly brackets is the additional factor we have obtained compared to the two- fluid model. The gray suppression function demonstrates a weaker suppression (higher effective thermal conductivity) than the two-fluid model due to this additional factor. The gray suppression function of Eq. (7) excellently reproduces the results of the normalized gray medium 12 effective diffusivity obtained numerically previously by Collins et al. [23]. However, we have shown that indeed this approach of inputting the gray suppression function into the effective thermal conductivity expression is not universal, and performs rather poorly, especially for silicon as shown by the gray variational results from Fig.1. In summary, we have developed a variational approach that yields a new way of extracting the effective thermal conductivity of the system by exploiting knowledge of the Fourier solution. In general, this approach to solving the temperature equation for the spectral BTE can directly yield the effective thermal conductivity from quasi-ballistic phonon transport experiments without brute force numerical solution of the BTE. We demonstrate the power of this approach by calculating the thermal decay rate as well as an analytical suppression function for one- dimensional transient grating experiments. Our spectral suppression function yields the exact suppression of thermal conductivity. We have shown that the suppression function is not universal, and utilizing the gray solution to the BTE does not perform well in reproducing the exact spectral data. Moreover, the variational approach developed here can be used as a universal technique for solving the BTE and obtaining both experimental observables, such as measured heat flux or thermal decay rate, as well as the effective thermal conductivity. This technique can be extended beyond the TTG problem, and can be used to calculate the effective thermal conductivities and suppression functions in different experimental geometries. This technique and our results here will allow for a better understanding of transport beyond the diffusive limit. 13 Acknowledgment: The authors would like to thank Dr. Kimberlee C. Collins for providing the material property data for Si and PbSe. This material is based upon work supported as part of the “Solid State Solar-Thermal Energy Conversion Center (S3TEC)”, an Energy Frontier Research Center funded by the U.S. Department of Energy, Office of Science, Office of Basic Energy Sciences under Award Number: DE-SC0001299/DE-FG02-09ER46577. 14 References [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] M. E. Siemens, Q. Li, R. Yang, K. A. Nelson, E. H. Anderson, M. M. Murnane, and H. C. Kapteyn, Nat. Mater. 9, 26 (2010). A. J. Minnich, J. A. Johnson, A. J. Schmidt, K. Esfarjani, M. S. Dresselhaus, K. A. Nelson, and G. Chen, Phys. Rev. Lett. 107, 095901 (2011). J. A. Johnson, A. A. Maznev, J. Cuffe, J. K. Eliason, A. J. Minnich, T. Kehoe, C. M. S. Torres, G. Chen, and K. A. Nelson, Phys. Rev. Lett. 110, 025901 (2013). K. M. Hoogeboom-Pot, J. N. Hernandez-Charpak, X. Gu, T. D. Frazer, E. H. Anderson, W. Chao, R. W. Falcone, R. Yang, M. M. Murnane, H. C. Kapteyn, and D. Nardi, Proc. Natl. Acad. Sci. 112, 201503449 (2015). J. Cuffe, J. K. Eliason, A. A. Maznev, K. C. Collins, J. A. Johnson, A. Shchepetov, M. Prunnila, J. Ahopelto, C. M. Sotomayor Torres, G. Chen, and K. A. Nelson, Phys. Rev. B 91, 245423 (2015). Y. Hu, L. Zeng, A. J. Minnich, M. S. Dresselhaus, and G. Chen, Nat. Nanotechnol. (2015). J. A. Johnson, J. K. Eliason, A. A. Maznev, T. Luo, and K. A. Nelson, J. Appl. Phys. 118, 155104 (2015). L. Zeng and G. Chen, J. Appl. Phys. 116, 064307 (2014). C. Hua and A. J. Minnich, J. Appl. Phys. 117, 175306 (2015). C. Hua and A. J. Minnich, Phys. Rev. B 90, 214306 (2014). A. J. Minnich, Phys. Rev. B 92, 085203 (2015). C. Dames and G. Chen, Thermal Conductivity of Nanostructured Thermoelectric Materials (CRC Press, 2005). 15 [13] [14] [15] [16] [17] [18] [19] [20] [21] [22] [23] [24] [25] [26] [27] [28] G. C. Asegun S. Henry, J. Comput. Theor. Nanosci. 5, 141 (2008). F. Yang and C. Dames, Phys. Rev. B 87, 035437 (2013). K. C. Collins, A. A. Maznev, J. Cuffe, K. A. Nelson, and G. Chen, Rev. Sci. Instrum. 85, 124903 (2014). D. G. Cahill, Rev. Sci. Instrum. 75, 5119 (2004). K. T. Regner, S. Majumdar, and J. A. Malen, Rev. Sci. Instrum. 84, 064901 (2013). K. T. Regner, D. P. Sellan, Z. Su, C. H. Amon, A. J. H. McGaughey, and J. A. Malen, Nat. Commun. 4, 1640 (2013). A. A. Maznev, J. A. Johnson, and K. A. Nelson, Phys. Rev. B 84, 195206 (2011). G. Chen, Nanoscale Energy Transport and Conversion: A Parallel Treatment of Electrons, Molecules, Phonons, and Photons: A Parallel Treatment of Electrons, Molecules, Phonons, and Photons (Oxford University Press, 2005). G. Chen, J. Heat Transfer 118, 539 (1996). A. J. Minnich, Phys. Rev. Lett. 109, 205901 (2012). K. C. Collins, A. A. Maznev, Z. Tian, K. Esfarjani, K. A. Nelson, and G. Chen, J. Appl. Phys. 114, 104302 (2013). C. Hua and A. J. Minnich, Phys. Rev. B 89, 094302 (2014). J. Sakurai, Modern Quantum Mechanics (Addison Wesley, 1993). D. J. Griffiths, Introduction to Quantum Mechanics (Pearson Education, 2005). P. B. Allen, arXiv:1308.2890 (2013). L. E. G. Ah-Sam, H. H. Jensen, and H. Smith, J. Stat. Phys. 3, 17 (1971). 16 [29] [30] [31] [32] [33] [34] N. K. Hindley and F. Garcia-Moliner, Phys. Rev. 140, A343 (1965). J. Ziman, Electrons and Phonons: The Theory of Transport Phenomena in Solids (Oxford Classic Texts in the Physical Sciences) (Oxford University Press, London, 1960). A. J. Minnich, Phys. Rev. Lett. 109, 205901 (2012). A. Majumdar, J. Heat Transfer 115, 7 (1993). J.-P. M. Péraud and N. G. Hadjiconstantinou, Phys. Rev. B 84, 205331 (2011). J.-P. M. Péraud and N. G. Hadjiconstantinou, Appl. Phys. Lett. 101, 153114 (2012). 17
1211.7273	1	1211	2012-11-30T14:47:32	How reliable are Hanle measurements in metals in a three-terminal geometry?	[ "cond-mat.mes-hall" ]	We test the validity of Hanle measurements in three-terminal devices by using aluminum (Al) and gold (Au). The obtained Hanle and inverted Hanle-like curves show an anomalous behavior. First, we measure Hanle signals 8 orders of magnitude larger than those predicted by standard theory. Second, the temperature and voltage dependences of the signal do not match with the tunneling spin polarization of the ferromagnetic contact. Finally, the spin relaxation times obtained with this method are independent of the choice of the metallic channel. These results are not compatible with spin accumulation in the metal. Furthermore, a scaling of the Hanle signal with the interface resistance of the devices suggests that the measured signal is originated in the tunnel junction.	cond-mat.mes-hall	cond-mat	How reliable are Hanle measurements in metals in a three-terminal geometry? Oihana Txoperena1, Marco Gobbi1, Amilcar Bedoya-Pinto1, Federico Golmar,1 Xiangnan Sun1, Luis E. Hueso1,2 and Fèlix Casanova1,2 1CIC nanoGUNE, 20018 Donostia-San Sebastian, Basque Country (Spain) 2IKERBASQUE, Basque Foundation for Science, 48011 Bilbao, Basque Country (Spain) We test the validity of Hanle measurements in three-terminal devices by using aluminum (Al) and gold (Au). The obtained Hanle and inverted Hanle-like curves show an anomalous behavior. First, we measure Hanle signals 8 orders of magnitude larger than those predicted by standard theory. Second, the temperature and voltage dependences of the signal do not match with the tunneling spin polarization of the ferromagnetic contact. Finally, the spin relaxation times obtained with this method are independent of the choice of the metallic channel. These results are not compatible with spin accumulation in the metal. Furthermore, a scaling of the Hanle signal with the interface resistance of the devices suggests that the measured signal is originated in the tunnel junction. Spintronics is a rapidly growing research area that aims to use and manipulate not only the charge, but also the spin of the electron.1 A spintronic device capable of creating, transporting, manipulating and detecting spins is a long-sought goal. There is special interest in purely electrical spin injection and detection devices with long, active semiconductor channels, to integrate the spin functionality into conventional electronics. A simple device to study spin injection and transport in semiconductors uses a three-terminal (3T) geometry, in which spin accumulation is induced and probed by a single magnetic tunnel contact.2,3 The detection is possible because a transverse magnetic field reduces the spin accumulation due to dephasing during the precession, the so-called Hanle effect3,4. Since the 3T geometry does not require submicron-sized fabrication processes, this type of devices have become very popular.2,3,5-16 However, many of the results associated to this method are still controversial. For example, large disagreements between the theoretically predicted and experimentally measured spin accumulation have been found.5 Experimental values well above the theoretical ones were first reported in Fe/Al2O3/GaAs structures.6 Since then, similar results have been observed in various semiconductors.3,5,7-11 Moreover, some other reported features such as an anomalous bias dependence of the Hanle signal7,8,12,13 or the unclear origin of the inverted Hanle effect11,14, put the 3T-measurements strongly into question. In order to gain insight about the reliability of this method, we have applied it to metals with well-known spin transport properties such as aluminum (Al)4,17-21 and gold (Au).21- 24 In this letter, we report that measurements done with a 3T geometry lead to Hanle- and inverted Hanle-like features which are not compatible with spin accumulation in the studied metals. In addition, the measured signals scale with the interface resistance of the tunnel barrier of the devices, suggesting that the origin of these anomalous signals might arise from the tunnel barrier itself. We produced devices with different non-magnetic metals: samples 1-3 with the structure Al/AlOx/Py, where the Al (Py) thickness is 15 nm (10 nm), and samples 4-6 with the ferromagnetic metal at the bottom Py/AlOx/Au, with 10 nm of Py and 12 nm of Au. The devices were fabricated in a UHV e-beam evaporation chamber (base pressure <1·10-9 mbar), using an integrated shadow masking system. The device geometry is sketched in Fig. 1(a). We used two different strategies for the tunnel barrier fabrication: a plasma exposure to oxidize the Al in a single run (Sample 1); and a three-step deposition of a 6 Å Al layer with subsequent oxidation in an O2 atmosphere without plasma (Samples 2-6). The contact area ranged between 250 x 250 µm2 and 250 x 500 µm2. Electrical measurements were performed by a dc method using a dc current source and a nanovoltmeter,17 while the temperature and field control is done in a commercial Quantum Design PPMS cryostat. I V ) 40 ) 40 2 6 2 6 m m µµµµ 20 µµµµ 20 · · ΩΩΩΩ ΩΩΩΩ ) A ) A 0 9 9 0 µµµµ 0 4 0 µµµµ 4 ( 1 1 ( I I · -20 · -20 ( ( A A · · -40 -40 t t n n configuration and the magnetic field direction for both Hanle (𝐵! ) and inverted Hanle (𝐵!! ) effect are -0.5 0.0 0.5 0 100 200 300 i 0 100 200 300 i -0.5 0.0 0.5 R R V (V) V (V) T (K) T (K) FIG. 1. (a) Scheme and dimensions of the device used for the Hanle measurements. The electrical A = w!w! , as a function of temperature for Sample 1 (open circles) and Sample 2 (solid triangles). shown. The magnetization vector of Py strip (M), parallel to its easy axis, is also shown. (b) I-V data for Sample 1 (open circles) and Sample 2 (solid triangles), measured at 10 K. Red solid lines are best fits to the standard Simmons model. (c) Zero bias resistance of the tunnel junction, multiplied by its area First, we characterize the tunnel junctions obtained by different fabrication processes. In both the height of the tunnel barriers and their thickness: 𝜙 = 2.8 V and 𝑑 = 1.5 nm for Fig. 1(b), we show the current-voltage data of the tunnel junctions of Al samples Sample 1, and 𝜙 = 0.8 V and 𝑑 = 2.7 nm in the case of Sample 2. The difference in the (Samples 1 and 2) measured at 10 K. The measured data is fitted following the standard Simmons model for tunneling.25 From the fittings (red curves in Fig. 1(b)), we obtain Fig. 1(c) shows the resistance of the tunnel junction (𝑅!"# ) measured at zero bias and multiplied by its total area (𝐴) as a function of temperature for Samples 1 and 2. In both barrier parameters is due to the different fabrication strategy of the junctions, as expected.26,27 Samples 3-6 show barrier parameters similar to Sample 2 (not shown). cases, we observe a weak decrease with increasing temperature, as expected from a tunnel junction with no pinholes.27 The values of resistivity of the metal stripes at 10 K, measured with four probes, are 11.3 µΩ·cm (Sample 1, Al), 20.2 µΩ·cm (Sample 2, Al) effect when a transverse magnetic field (𝐵! ) is applied. The spins, accumulated at the and 3.5 µΩ·cm (Sample 4, Au). With the measurement configuration shown in Fig. 1(a), we should observe the Hanle interface due to electrical injection from the Py contact into the metal, start to precess around the transverse magnetic field. Due to the dephasing during this precession, the spin accumulation is gradually reduced with an approximately Lorentzian shape.3,4 The ∆𝑉 𝐵! = ∆!!"#$% extra voltage at the interface associated to the spin accumulation (spin voltage) can be written as: !!(!!!!" )! . (1) ∆𝑉!"#$% is the value of spin voltage in the absence of 𝐵! , 𝜏!" the spin relaxation time, and 𝜔! = 𝑔𝜇! 𝐵!/ℏ is the Larmor frequency, where g is the Landé g-factor, 𝜇! is the Bohr magneton and ℏ is the reduced Planck’s constant. Commonly, the spin voltage is normalized to the injected current and, thus, ∆𝑅!"#$% = ∆𝑉!"#$% /𝐼 . In Fig. 2 (solid symbols) we show two examples of a Hanle-like curve measured at 10 K, obtained after apply an in-plane magnetic field (𝐵!! ).11 It has been reported that, due to the roughness subtracting the quadratic background from the data.7 Fig. 2(a) corresponds to Sample 1 (measurements done at I=-100 µA), whereas Fig. 2(b) shows the same experiment on injected spins precess even in the absence of 𝐵! . Therefore, the precession is suppressed Sample 2 (I=-5 µA). We should also observe a change in the measured voltage if we by applying 𝐵!! and the spin accumulation increases until it saturates for large enough fields.11 This is the so-called inverted Hanle effect, with associated signal ∆𝑅!"# . The of the interface, local magnetic fields appear on the non-magnetic material and make the data measured at 10 K on Sample 1 (performed at I=-100 µA) and Sample 2 (I=-5 µA) effect curve. We call this difference the total Hanle signal, ∆𝑅!"! = ∆𝑅!"#$% + ∆𝑅!"# . are represented by the open symbols in Fig. 2(a) and (b), respectively. According to At 10 K and previously described biases, ∆𝑅!"#$% values are 26% and 21% of ∆𝑅!"! Ref. 11, the total spin accumulation is proportional to the difference between the saturation value in the inverted Hanle effect curve and the minimum value of the Hanle for Sample 1 and 2, respectively. These percentages are in agreement with those reported in Ref. 11. Fig. 2 shows that very similar results are obtained for both samples, suggesting that properties of the tunnel oxide barrier are not relevant. by Δ𝑅!"! = 𝛾 ! 𝑅! , in the limit 𝑅!"# ≫ 𝑅! ≫ 𝑅! , which is our case. 𝛾 is the tunneling /𝑉!(! )! spin polarization and 𝑅!(! ) = 𝜌!(! ) 𝜆!(! ) The standard theory of spin injection and accumulation in the diffusive regime28,29 states ! magnetic (ferromagnetic) side of the interface, where 𝜌!(! ) are the resistivities, 𝜆!(! ) that the total spin signal associated to spin accumulation at the interface is given the spin diffusion lengths and 𝑉!(! )! is the spin resistance of the non- that, for the device sketched in Fig. 1(a), 𝑤! , 𝑤! ≫ 𝜆 ≫ 𝑑 is satisfied, then 𝑉!! =𝑤! · 𝑤! · 𝑑 . Therefore, the total spin signal can be expressed as: the effective volumes of spin relaxation.30 Assuming that the spin injection occurs homogeneously along all the contact area and Δ𝑅!"! = !! !! !!! !!!!! . (2) Taking into account the 𝜆! values obtained from the fitting to Eq. 1 (see below) and that of the order of 10!! Ω, whereas the measured values are at least 8 orders of magnitude typical γ values for Py/Al2O3 interfaces are 0.02-0.25,18,19 the total spin signal should be higher (see Fig. 2). (cid:10)(cid:5)(cid:11)(cid:1) 8 40 (cid:10)(cid:6)(cid:11)(cid:1) 4 20 (cid:12) (cid:9) (cid:10) (cid:8) (cid:10) (cid:1) ) (cid:1)(cid:1)(cid:1)(cid:1) ( R (cid:2)(cid:2)(cid:2)(cid:2) I(cid:1) ) (cid:1)(cid:1)(cid:1)(cid:1) ( R (cid:2)(cid:2)(cid:2)(cid:2) (cid:4)(cid:9)(cid:1) (cid:2)(cid:7)(cid:3)(cid:8)(cid:1) (cid:2)(cid:7)(cid:1) (cid:12) (cid:9) (cid:5) (cid:7) (cid:11) (cid:1) (cid:12) (cid:9) (cid:4) 0 (cid:2) (cid:7) 0 (cid:6) (cid:3) -1.0 -0.5 0.0 0.5 1.0 (cid:1) 1.0 -1.0 -0.5 0.0 0.5 B (kOe) B (kOe) measured at 10 K and -100 µA. Red solid line is the Lorentzian fit of the data to Eq. 1. Hanle (Δ𝑅!"#$% ), inverted Hanle (Δ𝑅!"# ) and total Hanle (Δ𝑅!"! ) signals are defined. (b) Same as (a) for Sample 2, FIG. 2. (a) Hanle and inverted Hanle-like curves (solid and open circles, respectively) for Sample 1, measured at 10 K and -5 µA. The device scheme is shown as an inset. spin signal could be caused by the decrease of 𝜏!" due to the broadening of the Hanle Discrepancies between theoretical and experimental values of spin signal have previously been observed and discussed for semiconductors.3,5-11 The small theoretical curve,3,11 although, in our case, this effect is by far not enough to explain the discrepancy. The role of localized states near the tunnel junction/semiconductor an strong temperature dependence of 𝑅!"# · 𝐴, or a barrier height 𝜙 larger than expected interface has also been deeply debated lately.3,5-11 In our samples, since Al is a conductor, such states could only be created within the oxide tunnel junction due to fabrication conditions. However, evidences of the existence of localized states, such as are not found,31 neither in Sample 1, with the tunnel barrier grown by plasma, nor in sample 2, grown by natural oxidation. Finally, Dash et al.3 analyze the possibility of having lateral inhomogeneities at the tunnel junction. In this case, electrons mostly tunnel through the thinnest regions of the junction, so-called hot spots. This scenario is the theoretical spin signal. We can recalculate 𝑉!! by assuming the existence of N hot probable in our tunnel barriers due to the inherent roughness of the AlOx surface spots on the tunnel barrier. If the size of these spots is smaller than 𝜆! , and the distance (r.m.s.=0.7 nm for Sample 1 and 0.4 nm for Sample 2). In the presence of hot spots, the between them is longer than 2𝜆! , then Δ𝑅!"! = 𝛾 ! 𝜌! 𝑁𝜋𝑑 . In the limiting unrealistic effective volume of spin accumulation would be reduced, leading to an enhancement of case with N=1, which gives the smallest effective volume, we find that Δ𝑅!"! ≈ 10!! Ω, and the theoretical Hanle signal is still two orders of magnitude lower than the experimental one. Therefore, the enormous difference between standard theory and cannot be explained on the basis of Eq. 2. Fig. 3(a) shows ∆𝑅!"! as a function of experiments cannot be explained by the existence of hot spots. Next, we show that the temperature and bias dependencies of the total Hanle signal proportional to the resistivity,21 then Δ𝑅!"! ∝ 𝛾 !𝜌!!! , where 𝛾 can be expressed as32 𝛾 = 𝛾! (1 − 𝑇 𝑇! ! ! ), being 𝑇! the Curie temperature of the ferromagnet. Using this expression for 𝛾 and the experimental values for 𝜌!!! , we fitted Δ𝑅!"! (T) to extract 𝑇! temperature both for Sample 1 and 2. Since the spin diffusion length is inversely for Sample 1 and 2 (Fig. 3(a)). For Sample 1, we obtain 𝑇! = 516 ± 13 𝐾 , which 𝑇! = 159 ± 7 𝐾 , in disagreement with a tunneling spin injection from Py. Fig. 3(b) shows the voltage-dependent measurements of Δ𝑅!"! at 10 K. For both Sample 1 and 2, we observe that the signal becomes undetectable at low bias voltage ( 𝑉 ≼ 0.5 𝑉 in could be in agreement with literature values for Py.33 However, for Sample 2, we obtain Sample 1 and 𝑉 ≼ 0.025 𝑉 in Sample 2). This gap in Δ𝑅!"! (V) at low bias cannot be polarization 𝛾 is the only bias-dependent parameter in Eq. 2, and it is largest at low explained by the standard theory of spin injection. Indeed, the tunneling spin bias.20,34 (cid:4)(cid:2)(cid:5)(cid:1) 10 ) (cid:1)(cid:1)(cid:1)(cid:1) 8 ( T O T 6 R (cid:2)(cid:2)(cid:2)(cid:2) 4 300 40 20 0 -1.0 -0.5 (cid:4)(cid:3)(cid:5)(cid:1) 0.5 1.0 0.0 V (V) 40 30 20 ) (cid:1)(cid:1)(cid:1)(cid:1) ( T O T R (cid:2)(cid:2)(cid:2)(cid:2) 0 100 200 T (K) FIG. 3. (a) Total Hanle signal as a function of temperature, measured at -100 µA for Sample 1 (open circles) and at -5 µA for Sample 2 (solid triangles). Red solid lines are fits of the data to Eq. 2. (b) Total Hanle signal as a function of the applied bias at 10 K, for Sample 1 (open circles) and Sample 2 (solid triangles). Eq. 1 to extract the value for Al: in the case of Sample 1, we obtain 𝜏!" = (80 ± 2) ps at 10 K and (73 ± 9) ps at 300 K. For Sample 2, the Lorentzian curve vanishes above 100 K, obtaining 𝜏!" = (82 ± 3) ps at 10 K and (43 ± 13) ps at 100 K. These Concerning the spin relaxation time, the obtained Lorentzian curves have been fitted to From 𝜏!" , spin diffusion length values 𝜆!" can be calculated as 𝜆!" = 𝐷𝜏!" , where D is the diffusion coefficient. The obtained values in Sample 1 are 𝜆!" = (439 ± 5) nm at 10 K and (370 ± 20) nm at 300 K, whereas in Sample 2 we obtain 𝜆!" = (325 ± values are the same for all biases (including injection into Al and extraction from Al). 6) nm at 10 K and 232 ± 35 nm at 100 K. Therefore, while there is no agreement neither with the amplitude, nor with temperature dependence and bias dependence of the accumulation, the spin diffusion length values obtained in the two samples seem to be in agreement with literature.4,17-21 In order to clarify this controversy, similar measurements are performed in Au, another the Lorentzian curve (red solid line in Fig. 4) is (144±5) ps, which is two orders of metal whose spin transport properties are also well-known21-24 but are very different to the ones in Al. Fig. 4 shows the Hanle and inverted Hanle-like curves measured at 10 K for I=-50 µA in Sample 4. The spin relaxation time value extracted from the fitting of magnitude higher than the expected values for Au (𝜏!"~ 1 ps).21-24 Moreover, total Hanle amplitudes above the theoretically expected values (see Fig. 4) and anomalous voltage-dependence of signal amplitude (not shown) have also been measured in this sample. Although the spin relaxation times obtained for Al (Samples 1 and 2) are reasonable,4,17-21 similarities between the results reported in Al and Au 3T devices (comparable spin relaxation time, amplitude of signals above the expected values, and anomalous bias dependence of the total Hanle signal) evidence that the measured Hanle and inverted Hanle-like curves are not originated by spin accumulation in the metal. 2 1 ) (cid:1)(cid:1)(cid:1)(cid:1) ( R (cid:2)(cid:2)(cid:2)(cid:2) I(cid:1) (cid:2)(cid:6)(cid:1) (cid:2)(cid:5)(cid:3)(cid:7)(cid:1) (cid:4)(cid:8)(cid:1) 0 1.0 0.5 0.0 -1.0 -0.5 B (kOe) FIG. 4. Hanle and inverted Hanle-like curves (solid and open circles, respectively) measured in Sample 4 Fig. 5 shows the spin RA product (∆𝑅!"! · 𝐴) as a function of the RA product of the at 10 K and -50 µA. Red solid line is the Lorentzian fit of the data to Eq. 1. The device scheme is shown as an inset. tunnel junctions (𝑅!"# · 𝐴) of all 3T devices used in this work, which employ different scaling between ∆𝑅!"! · 𝐴 and 𝑅!"# · 𝐴, with a power law exponent of 1.31, is observed metals and tunnel barriers obtained through different fabrication processes. A clear 𝑅!"# ≫ 𝑅! ≫ 𝑅! .28,29 The scaling suggests again that the Hanle and inverted Hanle-like in our Py/AlOx/metal devices with different RA products. This result, which spans over 2 orders of magnitude, is not predicted by the standard theory in the condition curves do not arise from the modulation of the spin accumulation in the metal by an external magnetic field, but from the tunnel junction itself. Recent reports on 3T measurements in Si- and Ge-based devices have shown similar disagreements. Aoki et al.14 report a Lorentzian curve in both Hanle and inverted Hanle configuration, with a corresponding spin relaxation time of ~50 ps, which cannot be associated to spin accumulation in Si when compared to non-local four-terminal measurements. Uemura et al.15 and Sharma et al.35 report a tunnel barrier thickness dependence of Hanle signals, observing a scaling of the spin RA product with the RA product of the junction as well. It is worth noting that the spin relaxation time obtained in Ref. 15 for n-Si is 150 ps, very close to our values both for Al and Au. 106 105 Sample 1 Sample 2 Sample 3 Sample 4 Sample 5 Sample 6 1010 ) 2 m µµµµ · (cid:1)(cid:1)(cid:1)(cid:1) ( A · T O T R (cid:2)(cid:2)(cid:2)(cid:2) 109 Rint·A ((cid:1)(cid:1)(cid:1)(cid:1)· µµµµm2) respectively. Red line is the logarithmic fit to the data. FIG. 5. Spin RA product as a function of the RA product of the tunnel junction, for different samples measured at 10 K and their optimum bias conditions. Solid and open symbols are for Al and Au samples, In conclusion, we have tested the reliability of three-terminal Hanle measurements in metals such as Al and Au, with well-known and different spin transport properties. Our results indicate that the obtained Hanle and inverted Hanle-like curves are not related to spin accumulation in the metal channel. First, and most important, the spin relaxation times obtained for both metals are comparable, even if much shorter values are expected for Au. Furthermore, these values are also comparable to those reported in other systems such as semiconductors using the same configuration.14,15 Second, the total Hanle signal is several orders of magnitude higher than the value predicted from the standard theory. Such a difference cannot be explained by any of the sources previously discussed in literature for semiconductors.3,5-11 Third, the temperature and voltage dependences of the signal are not in agreement with the tunneling spin polarization of the used ferromagnetic contact. Last, we show that the total Hanle signal is proportional to the interface resistance of the tunnel barrier, a dependence which cannot be explained by models proposed so far and which suggests that the observed effect originates at the tunnel barrier. Finally, we would like to emphasize that great care should be taken (e.g., use of control samples) when measuring the Hanle effect using three-terminal geometries with highly resistive tunnel barriers to extract spin injection and transport properties from any type of materials, including metals and semiconductors. This work is supported by the European Union 7th Framework Programme (NMP3-SL- 2011-263104-HINTS, PIRG06-GA-2009-256470 and the European Research Council Grant 257654-SPINTROS), by the Spanish Ministry of Science and Education under Project No. MAT2009-08494 and by the Basque Government under Project No. PI2011-1. We would also like to acknowledge Prof. Casey W. Miller for fruitful discussions. REFERENCES 1 I. Žutić, J. Fabian, and S. Das Sarma, Rev. Mod. Phys. 76, 323-386 (2004). 2 X. Lou, C. Adelmann, M. Furis, S. A. Crooker, C. J. Palmstrøm, and P. A. Crowell, Phys. Rev. Lett. 96, 176603 (2006). Nature (London) 416, 713 (2002). 3 S. P. Dash, S. Sharma, R. S. Patel, M. P. de Jong, and R. Jansen, Nature (London) 462, 491 (2009). 4 F. J. Jedema, H. B. Heersche, A. T. Filip, J. J. A. Baselmans, and B. J. Van Wees, 5 R. Jansen, S. P. Dash, S. Sharma, and B. C. Min, Semic. Sci. Technol. 27, 083001 (2012). 6 M. Tran, H. Jaffrès, C. Deranlot, J.-M. George, A. Fert, A. Miard, and A. Lemaître, Phys. Rev. Lett. 102, 036601 (2009). 7 C. H. Li, O. M. J. van ’t Erve, and B. T. Jonker, Nature Commun. 2, 245 (2011). 8 O. M. J. van ‘t Erve, A. L. Friedman, E. Cobas, C. H. Li, J. T. Robinson, and B. T. Jonker, Nature Nanotechnol. 7, 737 (2012). 9 R. Jansen, A. M. Deac, H. Saito, and S. Yuasa, Phys. Rev. B 85, 134420 (2012). 10 A. Jain, J.-C. Rojas-Sanchez, M. Cubukcu, J. Peiro, J. C. Le Breton, E. Prestat, C. Vergnaud, L. Louahadj, C. Portemont, C. Ducruet, V. Baltz, A. Barski, P. Bayle- Guillemaud, L. Vila, J.-P. Attané, E. Augendre, G. Desfonds, S. Gambarelli, H. Jaffrès, J.-M. George, and M. Jamet, Phys. Rev. Lett. 109, 106603 (2012). 11 S. P. Dash, S. Sharma, J. C. Le Breton, J. Peiro, H. Jaffrès, J.-M. George, A. Lemaître, and R. Jansen, Phys. Rev. B 84, 054410 (2011). 12 Y. Ando, K. Kasahara, S. Yamada, Y. Maeda, K. Masaki, Y. Hoshi, K. Sawano, M. Miyao, and K. Hamaya, Phys. Rev. B 85, 035320 (2012). 13 Y. Ando, K. Kasahara, K. Yamane, Y. Baba, Y. Maeda, Y. Hoshi, K. Sawano, M. Miyao, and K. Hamaya, Appl. Phys. Lett. 99, 012113 (2011). 14 Y. Aoki, M. Kameno, Y. Ando, E. Shikoh, Y. Suzuki, T. Shinjo, and M. Shiraishi, Phys. Rev. B 86, 081201(R) (2012). 15 T. Uemura, K. Kondo, J. Fujisawa, K.-I. Matsuda, and M. Yamamoto, Appl. Phys. Lett. 101, 132411 (2012). 16 K.-R. Jeon, B.-C. Min, I.-J. Shin, C.-Y. Park, H.-S. Lee, Y.-H. Jo, and S.-C. Shin, Appl. Phys. Lett. 98, 262102 (2011). 17 F. Casanova, A. Sharoni, M. Erekhinsky, and I. K. Schuller, Phys. Rev. B 79, 184415 (2009). 18 S. O. Valenzuela and M. Tinkham, Appl. Phys. Lett. 85, 5914 (2004). 19 A. van Staa, J. Wulfhorst, A. Vogel, U. Merkt, and G. Meier, Phys. Rev. B 77, 214416 (2008). 20 S. O. Valenzuela, D. J. Monsma, C. M. Marcus, V. Narayanamurti, and M. Tinkham, Phys. Rev. Lett. 94, 196601 (2005). 21 J. Bass and W. P. Pratt Jr., J. Phys.: Condens. Matt. 19, 183201 (2007). 22 J. Ku, J. Chang, S. Han, J. Ha, and J. Eom, J. Appl. Phys. 99, 08H705 (2006). 23 J.-H. Ku, J. Chang, H. Kim, and J. Eom, Appl. Phys. Lett. 88, 172510 (2006). 24 T. Kimura, J. Hamrle, and Y. Otani, Phys. Rev. B 72, 014461 (2005). 27 J. J. Åkerman, R. Escudero, C. Leighton, S. Kim, D. A. Rabson, R. W. Dave, J. M. 25 J. G. Simmons, J. Appl. Phys. 34, 1793 (1963). 26 J. S. Moodera, E. F. Gallagher, K. Robinson, and J. Nowak, Appl. Phys. Lett. 70, 3050 (1997). Slaughter, and I. K. Schuller, J. Magn. Magn. Mater. 240, 86-91 (2002). 29 A. Fert and H. Jaffrès, Phys. Rev. B 64, 184420 (2001). 28 A. Fert, J.-M. George, H. Jaffrès, and R. Mattana, IEEE Trans. Electron. Dev. 54, 921-932 (2007). 30 H. Jaffrès, J.-M. George, and A. Fert, Phys. Rev. B 82, 140408(R) (2010). 31 Y. Xu, D. Ephron, and M. R. Beasley, Phys. Rev. B 52, 2843 (1995). 32 C. H. Shang, J. Nowak, R. Jansen, and J. S. Moodera, Phys. Rev. B 58, R2917-R2920 (1998). 33 P. Yu, X. F. Jin, J. Kudrnovský, D. S. Wang, and Bruno, Phys. Rev. B 77, 054431 (2008). 34 B. G. Park, T. Banerjee, J. C. Lodder, and R. Jansen, Phys. Rev. Lett. 99, 217206 (2007). 35 S. Sharma, A. Spiesser, S. P. Dash, S. Iba, S. Watanabe, B. J. van Wees, H. Saito, S. Yuasa, and R. Jansen, arXiv:1211.4460 (2012).
1602.01367	1	1602	2016-02-03T17:03:05	Superconducting quantum metamaterials as active lasing medium: Effects of disorder	[ "cond-mat.mes-hall", "quant-ph" ]	A metamaterial formed by superconducting circuits or quantum dots can serve as active lasing medium when coupled to a microwave resonator. For these artificial atoms, in contrast to real atoms, variations in their parameters cannot be avoided. In this paper, we examine the influence of disorder on such a multi-atom lasing setup. We find that the lasing process evolves into a self-organized stationary state that is quite robust against disorder. The reason is that photons created by those atoms which are in or close to resonance with the resonator stimulate the emission also of more detuned atoms. Not only the number of photons grows with the number of atoms, but also the width of the resonance as function of the detuning. Similar properties are found for other types of disorder such as variations in the individual coupling. We present relations how the allowed disorder scales with the number of atoms and confirm it by a numerical analysis. We also provide estimates for the sample-to-sample variations to be expected for setups with moderate numbers of atoms.	cond-mat.mes-hall	cond-mat	Superconducting quantum metamaterials as active lasing medium: Effects of disorder Martin Koppenhofer,1 Michael Marthaler,1 and Gerd Schon1, 2 1Institut fur Theoretische Festkorperphysik, Karlsruhe Institute of Technology, D-76131 Karlsruhe, Germany 2Institute of Nanotechnology, Karlsruhe Institute of Technology, D-76344 Eggenstein-Leopoldshafen, Germany (Dated: October 23, 2018) A metamaterial formed by superconducting circuits or quantum dots can serve as active lasing medium when coupled to a microwave resonator. For these artificial atoms, in contrast to real atoms, variations in their parameters cannot be avoided. In this paper, we examine the influence of disorder on such a multi-atom lasing setup. We find that the lasing process evolves into a self- organized stationary state that is quite robust against disorder. The reason is that photons created by those atoms which are in or close to resonance with the resonator stimulate the emission also of more detuned atoms. Not only the number of photons grows with the number of atoms, but also the width of the resonance as function of the detuning. Similar properties are found for other types of disorder such as variations in the individual coupling. We present relations how the allowed disorder scales with the number of atoms and confirm it by a numerical analysis. We also provide estimates for the sample-to-sample variations to be expected for setups with moderate numbers of atoms. PACS numbers: 42.70.Hj,78.45.+h,78.67.Pt,85.25.-j I. INTRODUCTION Lasers are the standard sources of coherent light with a wide range of applications [1]. Their basic components are a resonator that stores photons and selects particular modes, an optically active medium that emits photons co- herently into the resonator by stimulated emission, and a pumping process that establishes a population inver- sion in the medium [2]. A large variety of systems can serve as optically active medium. This includes natural atoms or semiconductor devices [3], but further systems have been proposed and studied experimentally. These include strongly coupled single or few Josephson qubits [4 -- 7] and semiconductor quantum dot systems [8 -- 10]. In these setups, the low number of atoms is compensated by strong coupling. Their frequencies are in the GHz regime, accordingly they are sometimes called "maser" instead of laser. These systems may find useful applications, e.g., as miniaturized on-chip sources of coherent microwaves in low temperature experiments. There are other on-chip microwave sources which have been studied, including voltage-biased Josephson junctions [11 -- 13] or nonlinear resonators close to the quantum regime [14 -- 16]. How- ever these devices emit incoherent radiation unless driven by a coherent microwave source. Lasing devices based on qubits also show unconventional properties, such as dressed-state lasing [17, 18] or photon-number squeezed light [19, 20]. So far, experimental realizations of lasers based on su- perconducting or quantum dot qubits have only used sin- gle or few artificial atoms. A way to reach higher out- put power is to increase the number M of atoms, e.g. by using superconducting quantum metamaterials. Such materials with 10 − 100 qubits have already been pro- duced and studied [21]. A drawback of using solid-state circuits is the fact that they invariably suffer from disor- der, either due to the variations in the fabrication process or in the environment. As a result, the level-splitting ǫj (j = 1, . . . M ) and hence the detuning from the resonator frequency ∆j = ǫj/−ω, as well as other parameters such as the coupling strength to the resonator or the local driving, vary for different artificial atoms. We therefore examine the influence of this (quasi-static) disorder on the multi-atom lasing. Our analysis reveals that the lasing process is rather robust against disorder over a wide range of disorder vari- ances. The origin of this effect is the following: Those atoms which are above the lasing threshold, e.g., close enough to resonance, start emitting photons into the res- onator. These photons enhance the process of stimulated emission, which is proportional to their number hni, also for atoms which are still below the threshold. Hence these atoms start participating in the lasing process, and hni grows further. In parallel to this growth, also the range of parameters such as detuning or coupling strength which are sufficient for lasing increases. In this paper, we will study the effects of disorder in the detuning, the coupling strength, and the pumping strength of the individual atoms on the photon number and the lasing thresholds. After presenting the model and the basic relations, we first compare the single- and the multi-atom setups of an ordered system. At this stage we observe already the increase of the allowed range of detuning and reduced requirement on the coupling strength. In the next sections quantitative results are presented for a Gaussian and a box distribution of the disorder in the various parameters. Since probably most experiments in the near future will be carried out with not too large numbers of atoms (M . 100) one should expect significant sample-specific deviations from the av- erage behavior. We therefore also study these statistical properties. In Sect. V we reformulate the problem, which provides further insight into the mechanism responsible for the enhanced stability against disorder. II. THE MODEL In the stationary state, this set of equations can be cast into a fixed point equation for the quantum statistical 2 We start from the Tavis-Cummings Hamiltonian for M two-level systems (atoms), labeled by j = 1, . . . M , with level-splitting energy ǫj and coupled with strength gj to a common resonator with frequency ω, H = ωa†a + 1 2 M Xj=1 ǫjτ j z + M Xj=1 gj(cid:16)τ j +a + τ j −a†(cid:17) . (1) Here a and a† are the photon operators of the radiation field and τ j ν are Pauli matrices acting on the two states of atom j. The influence of a dissipative environment, leading to relaxation and decoherence processes is conveniently ac- counted for by a quantum master equation for the density matrix ρ with the appropriate Lindblad terms d dt ρ = − κ LRρ = LQ,jρ = + + i [H, ρ] + LRρ + LQ,jρ , M Xj=1 Γ↑,j Γ↓,j +τ j −ρτ j 2 (cid:0)2aρa† − a†aρ − ρa†a(cid:1) , 2 (cid:16)2τ j − − τ j 2 (cid:16)2τ j + − τ j 2 (cid:0)τ j + − ρτ j − − ρτ j z − ρ(cid:1) . +ρτ j z ρτ j −τ j ϕ,j Γ∗ +τ j −τ j −ρ(cid:17) +ρ(cid:17) (2) (3) (4) Here Γ↑,j and Γ↓,j are the effective excitation and de- cay rates of atom j, Γ∗ ϕ,j its pure dephasing rate, and κ the decay rate of the resonator. Their combination defines the relaxation rate Γ1,j = Γ↑,j + Γ↓,j and the total dephasing rate Γϕ,j = Γ1,j/2 + Γ∗ ϕ,j. Below, we use the abbreviation Γκ,j = Γϕ,j + κ/2. By adjusting the rates Γ↑,j > Γ↓,j we can also account for the pump- ing of the lasing medium. Its strength is characterized by the value of the atomic polarization in equilibrium, D0 = (Γ↑,j − Γ↓,j)/Γ1,j. From the master equation we obtain the generalized Maxwell-Bloch equations by using what is known as semi- quantum approximation [22]. It amounts to neglecting M d d +τ j −E ≈ 0 for i 6= j, and direct interatomic couplings , Dτ i za†a(cid:11) ≈(cid:10)τ i factorizing (cid:10)τ i igj(cid:16)Dτ j Xj=1 dt(cid:10)a†a(cid:11) = z(cid:11) = 2igj(cid:16)Dτ j dt(cid:10)τ j − Γ1,j(cid:0)(cid:10)τ j +aE = −(cid:16)Γϕ,j + dtDτ j − i z(cid:11)(cid:10)a†a(cid:11). This yields −a†E(cid:17) − κ(cid:10)a†a(cid:11) , (5) +aE −Dτ j +aE(cid:17) −a†E −Dτ j z(cid:11) − D0,j(cid:1) , +aE 2 − i∆j(cid:17)Dτ j 2 (cid:0)1 + (2 hni + 1)(cid:10)τ j z(cid:11)(cid:1) . (7) (6) gj κ d average photon number hni =(cid:10)a†a(cid:11), D0,j(cid:0)hni + 1 κ,j + ∆2 Γ2 hni = βj M Xj=1 Here we introduced the parameters 2 2(cid:1) + 1 j + αj(cid:0)hni + 1 2(cid:1) . (8) αj = 4g2 j Γκ,j Γ1,j , and βj = 2g2 j Γκ,j κ . The term hni on the right-hand side of Eq. (8) accounts for the stimulated emission. III. PHOTON NUMBER AND ALLOWED DETUNING OF THE ORDERED SYSTEM It is instructive to first consider the case without disor- der. In this case, the fixed point equation has the solution M = X +rX 2 + hni0 1 X = − 4 Γ1 4g2 κ + ∆2 Γ2 Γκ M Γ1 4κ + n0(∆) = M Γ1 4κ (D0 + 1) , (9) D0 − n0(∆) 2 , , (10) where the superscript 0 refers to the absence of disor- der. The quantity n0(∆) is the photon saturation num- ber known already from the semi-classical theory of lasing [6, 23]. Within this theory the threshold for lasing is n0(∆) < M Γ1 2κ D0 . (11) In Fig. 1 we display how the photon number hni0 M (g, ∆) depends on the coupling strength g and de- tuning ∆ for a many-atom setup with M = 100 and the single-atom laser with M = 1. Fig. 1(a) illustrates the dependence on the coupling strength g. While the many- atom system shows a sharp, kink-like transition to the lasing state above gmin, the transition is washed out for the single- and few-atom setup. However, even for M = 1 it remains remarkably sharp. The crossover occurs at gmin(M ) =s 1 M κ Γ2 ϕ + ∆2 2D0 Γϕ 1 √M . ∝ (12) As a function of the detuning, illustrated in Fig. 1(b), we note a gradual decrease of hni0 M (g, ∆) with increasing ∆, followed by a a sharp transition to a low, close to thermal population. The crossover occurs at ∆max(M ) = Γϕs2g2M D0 κΓϕ − 1 ∝ √M . (13) 200 M / 0 M i n h 100 0 0 200 gmin(1) 0.0005 g/ω (a) M / 0 M i n h 100 ∆max(1) 0 0 0.25 ∆/ω (b) 0.5 FIG. 1: (Color online) Properties of the stationary quantum statistical average photon number hni0 M (g, ∆)/M per atom in an ordered lasing setup with M atoms. (a): Plots as function of the coupling strength g for atoms on resonance, (b): Plots as function of the atomic detuning ∆ ∆ = 0. at g = 0.002. Solid red curves represent hni0 1 (g, ∆), dashed blue curves hni0 , ∆)/M and dash-dotted green curves hni0 M (g, ∆)/M . Parameters are Γ↑ = 0.006, Γ↓ = 0.002, Γ∗ϕ = 0.001, κ = 0.00001 and M = 100. All rates and cou- plings are measured in units of ω. M ( g √M The analytic results (12) and (13) follow from Eq. (11), which is, like Eq. (10), obtained within the semi-classical approximation. Strictly, these approximations are valid for very large numbers of M , only. Remarkably they provide good estimates for the thresholds also for small M [5]. The results shown in Fig. 1 also illustrate a remarkable scaling relation [5], namely 1 M hni0 M(cid:18) g √M , ∆(cid:19) ≈ hni0 1 (g, ∆) . (14) The scaling relation, combined with Eqs. (12) and (13), displays the following properties: (i) in the lasing state the number of photons hni0 M grows linearly with the num- ber of atoms M , (ii) for different M we obtain the same qualitative dependence on the coupling strength provided that it is rescaled as g/√M , (iii) we furthermore note that the width of the resonance as function of detuning increases with the number of atoms proportional to √M . 3 In other words, for a fixed coupling strength, a system with many atoms can tolerate a stronger detuning and still show a transition to the lasing state than a single- atom laser. As we will show in the following section, this is the reason for the robustness against disorder which we observe. In the rest of the paper we focus on disordered systems. Their stationary quantum average photon number will be denoted by hniM . 0.001 IV. DISORDERED LASING MEDIUM We will now study the lasing transition in a system with many atoms M ≫ 1 with disorder in either the coupling strength, the detuning or the pumping. Accord- ingly, we average Eq. (8) over the appropriate normalized probability distribution, e.g., M Xi=1 ··· = MZZZ d∆ dg dD0 p(∆, g, D0) . . . . Having carried out the integration, we are left with a fixed point equation for hniM , depending on M and the distribution p. For clarity we concentrate in the following on disorder in only one lasing parameter at a time. In general, the problem needs to be solved numerically. However, we can proceed analytically by using a Gaus- sian distribution pG or a box distribution pB, pG(x) = pB(x) = 1 exp(cid:18)− √2πσx b (cid:20)Θ(cid:18)x − µ + 1 σ2 1 2 b x (cid:19) , (x − µ)2 2(cid:19) − Θ(cid:18)x − µ − b 2(cid:19)(cid:21) , b = √12 σx , where x is the variable to be averaged over, µ is its mean value and σx its standard deviation. Below, we will present results for the disorder aver- ages of the quantum statistical expectation values hniM . On the other hand, an experimental realization of quan- tum metamaterial-based lasing most likely will not have a very large number M of artificial atoms, and the prob- ability distribution p might not be sampled sufficiently to be well described by the integrals. Instead, for a given realization there will be deviations from the mean value hniM . We will analyze these fluctuations numerically by randomly generating setups of M atoms with lasing pa- rameters distributed according to a given probability dis- tribution p, and solving Eq. (8) numerically for each of these setups. The solution for such a random system of M atoms will be denoted by hniM . They show variations from sample to sample. A. Disorder in the detuning In this subsection we examine disorder in the atomic detuning ∆ = ǫ/ − ω. Its average is chosen to be zero. 30000 20000 0 0 1 i n h 10000 0 0 200 100 1 i n h 0 0 0.1 ∆/ω 0.2 0.2 σ∆/ω 0.4 FIG. 2: (Color online) Influence of disorder in the detuning ∆ on the quantum statistical average photon number hni100 in the resonator. Solid lines are calculated for a Gaussian distri- bution with mean ∆ = 0 and standard deviation σ∆, dotted lines represent a box distribution with the same parameters. The average photon number decreases remarkably slowly with increasing disorder. This cannot be explained by averaging the single-atom resonance curves hni1 (∆), examples of which are shown in the inset, over the same distribution of detun- ing. Averaging of the single-atom results would lead to the curves plotted with thin lines. Plot parameters are M = 100, Γ↑ = 0.006, Γ↓ = 0.002, Γϕ = 0.001 and κ = 0.00001; blue curves g = 0.004, red curves g = 0.002 and green curves g = 0.001. All rates and couplings are measured in units of ω. Since ǫ ≥ 0 we have ∆ ≥ −ω. This constraints the width of the box distribution to σ∆ ≤ ω/√3. Also in the case of a Gaussian distribution we choose sufficiently narrow dis- tributions to minimize the effect of unphysical values of the detuning. With these restrictions, we average Eq. (8) analytically and obtain hni = M β(cid:20)D0(cid:18)hni + ζhni =sΓ2 κ + α(cid:18)hni + 1 2(cid:19) + 2(cid:19) . 1 1 2(cid:21) I(cid:0)ζhni(cid:1) , (15) The integrals I(ζ) for the two types of distributions are given by 1 2 IG(ζ) =r π √3ζσ∆ IB(ζ) = 1 ζσ∆ 2σ2 exp(cid:18) ζ 2 ∆(cid:19) erfc(cid:18) ζ arctan √3σ∆ ζ ! . √2σ∆(cid:19) , Here, erfc is the complementary error function. These stationary fixpoint equations can be solved nu- merically for hniM (σ∆). Results for M = 100 atoms are shown by the thick upper curves in Fig. 2 for three val- ues of g and the two types of distribution pG/B(∆). We note that disorder in the detuning decreases hni100(σ∆) only weakly over a broad range of the variance σ∆. This behavior cannot be explained as an average of 4 t n u o c t n u o c 25 0 50 25 0 0 -0.2 0.2 (hni50 − hni50)/hni50 0 -0.004 0.004 (hni800 − hni800)/hni800 t n u o c t n u o c 25 0 50 25 0 0 -0.2 0.2 (hni100 − hni100)/hni100 0 -0.004 0.004 (hni1600 − hni1600)/hni1600 FIG. 3: (Color online) Fluctuations hniM around hniM for disorder in the detuning due to non-perfect sampling of a Gaussian disorder distribution with σ∆ = 0.2. Histograms are created for 10000 systems, randomly chosen with the Gaussian distribution, for M = 50, 100, 800, 1600 atoms, respectively. Results are hni50 = 7479, hni100 = 16976, hni800 = 156184, hni1600 = 316100 with standard deviations 386, 352, 182 and 136 photons, respectively. Plot parameters are ∆ = 0, g = 0.002, Γ↑ = 0.006, Γ↓ = 0.002, Γ∗ϕ = 0.001, κ = 0.00001. All rates and couplings are measured in units of ω. the single-atom laser resonances hni1 (∆), which are dis- played in the inset. Averaging them over the distribu- tions pG/B(∆) leads to the thin lower curves in Fig. 2, which obviously are much narrower in the disorder vari- ance than the collective behavior of the M atoms. We conclude that for disorder in the detuning, the setup with M atoms shows lasing in a much broader range of detun- ings than what we obtain from the single-atom results hni1 averaged over the same probability distribution. We will further illustrate this behavior in Sec. V. In addition we observe that for weak coupling strength g = 0.001 ω, the average contribution of each atom in the M -atom setup in the limit σ∆ → 0 is hni100/100 ≈ 200, whereas hni1 (∆ = 0) ≈ 195. That implies that the lasing activity per individual atom is enhanced in the M -atom setup as compared to the single-atom case, although for the parameters considered this is a rather weak effect. We also note that the results shown for the Gaussian and the box distribution nearly coincide. Both distribu- tions were chosen to have the same average and second- order moment. In addition hniM (σ∆) depends only on even moments of pG/B(∆). As a result, for sufficiently narrow distributions, σ∆ ≪ ω, both distributions yield similar results. We conclude this subsection with an analysis of the sample-to-sample variations of setups with a finite num- ber M of atoms. For this purpose, we consider ensembles with random parameters chosen according to the Gaus- sian distribution pG(∆) and solve Eq. (8) numerically. Results of hniM , varying around its mean value hniM , are shown in Fig. 3 for the variance σ∆ = 0.2 ω and M = 50, 100, 800 and 1600 atoms. On long enough time scales, when the quasi-static parameters vary in an experiment, we expect that the lasing intensity will vary accordingly. B. Disorder in the coupling strength Similar to the previous case, we examine disorder in the coupling strength g between the atoms and the resonator. The condition g ≥ 0 imposes rigorous constraints on the box distribution, σg ≤ g/√3, and similar approximate conditions for the Gaussian distribution. After averaging, the fixed point equation (8) becomes M Γ1 2κ (cid:18)D0 + hni = chni =s Γ1(Γ2 κ + ∆2) 4(hni + 1/2)Γκ , 1 2 hni + 1(cid:19) I(cid:0)chni(cid:1) , (16) with the integrals IG(c) = 1 − πcV (g, σg, c) , IB(c) = 1 − c b ×(cid:20)arctan(cid:18) µ + b/2 (cid:19)(cid:21) . Again, we have b = √12 σg, and V is the Voigt function (cid:19) − arctan(cid:18) µ − b/2 c c V (g, σg, x) 1 2 2πσg √2σg ! + c.c. .  erfc x − ig √2σg !2 exp  x − ig =r π Fig. 4 shows numerical solutions hniM (σg) of these fix- point equations for M = 100 atoms at resonance, ∆ = 0, and for two non-zero values of atomic detuning (thick curves). The mean coupling strength g = 0.001 ω is cho- sen to be close to the lasing threshold of a single atom in resonance. The position of g is indicated at the single- atom resonance curve in the inset by a black dot. For disorder in the coupling strength, similar to what we found above, the properties of hniM cannot be ex- plained by averaging the single-atom resonance curves hni1 (g) over the distribution of pG/B(g) (which would result in the thin lines). The averaging yields a smaller average photon number because a part of the atoms are coupled with g < g and therefore do no contribute (sig- nificantly) to the lasing process. On the other hand, hni100/100 ≈ 200, hence all atoms are actually partic- ipating in the lasing process at their maximum contri- bution irrespective of their actual individual coupling strength g. This effect is even more pronounced for ∆ = 0.1 or ∆ = 0.15 when the single-atom resonance 20000 15000 0 0 1 i n h 10000 5000 0 0 200 100 1 i n h 0 0 0.0125 g/ω 0.025 0.0002 σg/ω 5 0.0004 FIG. 4: (Color online) Influence of disorder in the cou- pling strength g on the stationary quantum statistical average photon number hni100 in the resonator (thick lines). Solid lines are calculated for a Gaussian distribution with mean g = 0.001 ω and standard deviation σg, dotted lines represent a box distribution with the same parameters. The results cannot be explained by averaging the single-atom resonance curves hni1 over the same distribution of coupling strength. This would yield the results represented by thin lines. The inset shows the corresponding single-atom lasing resonance curves hni1. The average g is indicated by a black circle. Plot parameters are M = 100, Γ↑ = 0.006, Γ↓ = 0.002, Γϕ = 0.001 and κ = 0.00001. Red curves ∆ = 0, blue curves ∆ = 0.1 and green curves ∆ = 0.15. The thin blue and green curves coincide. All rates and couplings are measured in units of ω. curves and their naıve average predict no lasing activity at all. However, the multi-atom setup is still operating at approximately 80 % or 50 % of its maximum photon number, respectively. In Sec. V we will provide further explanations of these properties. In a typical lasing experiment, the pumping rates are chosen such that the laser operates far above the lasing threshold of a single resonant atom. Then, the decrease in hniM for increasing ∆ is even less pronounced than shown in Fig. 4. In this figure we also observe again that the results obtained for a Gaussian and a box distribution coincide for a wide range of variance σg. We conclude with an analysis of the fluctuations in hniM around the mean value hniM due to disorder in g for a finite system size M , by the same procedure as described above for the variations in the detuning. The histograms of hniM in Fig. 5 show a main peak around hniM and a tail representing a few ensembles with much lower hniM . This tail arises because we choose g close to the lasing transition: Some atoms have such weak cou- pling strengths that they cannot participate in the lasing process. The corresponding systems have effectively a reduced M . As these systems occur rarely, an ensemble of 10000 systems is not sufficient to produce a smooth distribution. The standard deviation of the main peak is 0.24, 0.17, 0.07 and 0.05 photons, respectively. 10000 t n u o c 100 1 10000 t n u o c 100 1 -0.001 0 (hni50 − hni50)/hni50 10000 t n u o c 100 1 10000 t n u o c 100 1 -0.001 (hni100 − hni100)/hni100 0 -0.00003 0.00000 (hni800 − hni800)/hni800 -0.0006 (hni1600 − hni1600)/hni1600 0 FIG. 5: (Color online) Fluctuations hniM around hniM for disorder in the coupling strength due to a non-perfect sam- pling of a Gaussian distribution with σg = 0.0004 ω. His- tograms are created for 10000 systems, randomly chosen with the Gaussian distribution, with M = 50, 100, 800, 1600, re- spectively. hni50 = 9998, hni100 = 19998, hni800 = 159998, hni1600 = 319998. Plot parameters are g = 0.002, Γ↑ = 0.006, Γ↓ = 0.002, Γ∗ϕ = 0.001, κ = 0.00001. All rates and couplings are measured in units of ω. C. Disorder in the pumping The stationary value of the atomic polarization D0 = (Γ↑ − Γ↓)/(Γ↑ + Γ↓) is a function of the pumping and relaxation rates. These rates appear in Eq. (8) via the expressions D0 but also Γ1 = Γ↑ + Γ↓. In this subsection, we concentrate on the effect of disorder in D0 while as- suming that Γ1 is fixed. A motivation is provided by the system analyzed in Ref. 4, where Γ↑ − Γ↓ ∝ cos(θ) , Γ1 = Γ↑ + Γ↓ ∝ 1 2(cid:0)1 + cos2(θ)(cid:1) , and θ is the mixing angle of Coulomb and Josephson en- ergy. Typically, θ . π/2, so that D0 fluctuates propor- tional to θ − π/2 whereas Γ1 is approximately constant. Averaging Eq. (8) over the disorder in D0, we arrive at the fixpoint equation hni = 2(cid:1) + 1 M β(cid:2)D0(cid:0)hni + 1 2(cid:3) κ + ∆2 + α(hni + 1 Γ2 2 ) . (17) Its solution is of the same form as Eq. (9), except that D0 is replaced by D0. For typical lasing parameters above the threshold, Eq. (11) and the relation Γ1D0 ≫ κ hold. Then, Eq. (9) reduces to a linear dependence on D0, hniM = M Γ1 2κ D0 − n0(∆) . 6 This means that, in contrast to the previously discussed examples, disorder in D0 is properly accounted for by averaging over the single-atom results hni1. V. DISCUSSION The physical origin for the robustness of the system against disorder is an increased stimulated emission of each individual atom if there are additional photons hnaddi in the cavity originating from the lasing activity of other atoms. For ordered systems, the enhanced lasing activity, i.e., the growth of the average photon number, hniM ∝ M , and the increased range of allowed detuning is explicitly derived from Eq. (8). For disordered sys- tems with disorder in the detuning or coupling strength, we found similar behavior of the average quantity hniM . To gain further insight how hnaddi additional photons in the resonator broaden and enhance the lasing activity of each individual atom hnii, we split hni = M Xj=1 hnji = hnii +(cid:10)ni add(cid:11) with (cid:10)ni add(cid:11) = M Xj=1 j6=i hnji . (18) (19) Accordingly, we split Eq. (5) into M equations for hnii, what can be interpreted as the contributions of atom, i, which is solved by , (20) hnii = βi D0,i(hnii +(cid:10)ni κ,i + ∆2 Γ2 add(cid:11) + 1/2) + 1/2 add(cid:11) + 1/2) i + αi(hnii +(cid:10)ni If we sum this relation over all i = 1, . . . M we recover Eq. (8), but it is also valid for the single-atom case, M = 1. Since on the right hand side (cid:10)ni add(cid:11) appears always together with hnii, which is the term describing for a single atom the stimulated emission, the relation displays the property that each individual atom acquires an enhanced lasing activity by the presence of additional photons in the resonator. Eq. (20) is solved by hnii = X +s X 2 + 4 − (cid:10)ni add(cid:11)2 X = − 1 Γ1,i 2κ (cid:20)D0,i(cid:18)(cid:10)ni add(cid:11) + 1 2(cid:19) + 1 2(cid:21) , + Γ1,i 4κ D0,i − n0 2 . (21) This set of M equations for hnii has to be solved self- consistently via the relation Eq. (18). This is what we had done (effectively) in the previous parts of the paper both for an ordered as well as for disordered systems. To illustrate the enhancement effect we can also sim- ply assume that there exist additional photons, wher- ever they come from. Fig. 6 compares plots of hnii for 200 i i n h 100 0 0 200 i i n h 100 0 0 0.25 ∆/ω (a) 0.02 g/ω (b) 0.5 0.04 FIG. 6: (Color online) (a): hnii as a function of detuning for fixed g = 0.001. (b): hnii as a function of coupling strength for fixed ∆ = 0.2. Solid red curves represent the case without additional photons in the resonator, dashed blue curves the same in the presence of (cid:10)ni add(cid:11) = 4000 additional photons in the resonator cavity. Their presence increases the lasing activity of each individual atom as well as the range of allowed detuning, and decreases the threshold in coupling strength. Plot parameters are Γ↑ = 0.006, Γ↓ = 0.002, Γ∗ϕ = 0.001, κ = 0.00001. (cid:10)ni add(cid:11) = 0 and 4000. Fig. 6(a) demonstrates the broad- ening of the lasing resonance curve as a function of the detuning due to the presence of additional photons, as well as a slight enhancement at ∆ = 0, consistent with the observations made in Fig. 2. Fig. 6(b) demonstrates the lowering of the lasing threshold coupling strength, consistent with the observations made in Fig. 4. For disordered setups with a finite-width disorder distribution, we observed that atoms above the lasing threshold, e.g., close enough to resonance, "drag" oth- ers, which appeared to be below, also into a lasing state, and finally a self-organized stationary state is established. But we found an enhancement of the lasing window also in the case of ordered systems. Here we like to point 7 out that the reformulation presented in this section can reproduce also this property. To understand it we con- sider M identical atoms which would all be off-resonant in the single-atom setup, ∆ > ∆max(1). It is important to note that the semi-quantum model does not exhibit a sharp transition to the lasing state. Therefore, each of the atoms produces a small but non-vanishing contribu- tion hnii to the total photon number hni0 M . This possibly very small contribution is then enhanced by the presence of all other ones, which may be sufficient to drive the system into the self-consistent broadened state. VI. CONCLUSION In this paper, we showed that a multi-atom lasing setup with M ≫ 1 atoms is rather robust against dis- order in the individual atomic parameters, e.g. detuning, coupling strength to the resonator, or pumping strength. If an atom is coupled to a cavity that contains additional photons not originating from the atom itself, its lasing activity is nevertheless enhanced due to stimulated emis- sion. This leads to a growth of the number of photons scaling with M but also to a broadening of the reso- nance conditions, with the maximum allowed detuning scaling proportional to √M . Therefore, multiple atoms connected to a common resonator can effectively drag each other into resonance and generate a self-consistent stationary state that is robust against disorder. The average total photon number hniM of the setup can be calculated by performing the averages implied by the fixed point equation (8). We have performed these averages for two types of distributions, box and Gaus- sian, with similar results. Since currently systems with relatively low numbers of artificial atoms (M ≤ 100) are of interest for lasing experiments we also examined the fluctuations around hniM due to the imperfect sampling of the parameter distribution. This provides estimates for sample-to-sample fluctuations in such lasing setup, as well. The conclusion from our analysis is that imperfections in the control of material parameters do not prohibit the construction of multi-atom lasing setups. This will help the construction of miniaturized on-chip radiation sources for low-temperature microwave experiments. VII. ACKNOWLEDGEMENTS acknowledge fruitful We J. Braumuller. This work was supported by the DFG Research Grants SCHO 287/7-1 and MA 6334/3-1. discussions with [1] W. E. Lamb, W. P. Schleich, M. O. Scully, and C. H. [2] M. O. Scully, M. S. Zubairy, Quantum Optics, Cambridge Townes, Rev. Mod. Phys. 71, S263 (1999). University Press, 1997. 8 [3] G. H. B. Thompson, Physics of semiconductor laser de- vices, Wiley, 1980. M. Marthaler, and G. Johansson, Phys. Rev. Lett. 115, 027004 (2015). [4] O. Astafiev, K. Inomata, A. O. Niskanen, T. Yamamoto, Yu. A. Pashkin, Y. Nakamura, and J. S. Tsai, Nature 449, 588 (2007). [14] F. Mallet, F. R. Ong, A. Palacios-Laloy, F. Nguyen, P. Bertet, D. Vion, and D. Esteve, Nat. Phys. 5, 791 (2009). [15] A. Kamal, J. Clarke, and M. H. Devoret, Nat. Phys. 7, [5] S. Andr´e, V. Brosco, A. Shnirman, and G. Schon, Phys. 311 (2011). Rev. A 79, 053848 (2009). [16] S. Andre, L. Guo, V. Peano, M. Marthaler, and G. Schon, [6] S. Andr´e, P.-Q. Jin, V. Brosco, J. H. Cole, A. Romito, A. Shnirman, and G. Schon, Phys. Rev. A 82, 053802 (2010). [7] D. A. Rodrigues, J. Imbers, and A. D. Armour, Phys. Rev. Lett. 98, 067204 (2007). [8] L. Childress, A. S. Sørensen, and M. D. Lukin, Phys. Rev. A 69, 042302 (2004). [9] P. Q. Jin, M. Marthaler, J. H. Cole, A. Shnirman, and G. Schon, Phys. Rev. B 84, 035322 (2011); J. Jin, M. Marthaler, P. Q. Jin, D. Golubev, and G. Schon, New. J. Phys. 15, 025044 (2013). [10] Y.-Y. Liu, J. Stehlik, C. Eichler, M. J. Gullans, J. M. Taylor, and J. R. Petta, Science 347, 285 (2015). [11] M. Hofheinz, F. Portier, Q. Baudouin, P. Joyez, D. Vion, P. Bertet, P. Roche, and D. Esteve, Phys. Rev. Lett. 106, 217005 (2011). [12] V. Gramich, B. Kubala, S. Rohrer, and J. Ankerhold, Phys. Rev. Lett. 111, 247002 (2013). [13] J. Leppakangas, M. Fogelstrom, A. Grimm, M. Hofheinz, Phys. Rev. A 85, 053825 (2012). [17] J. Hauss, A. Fedorov, C. Hutter, A. Shnirman, and G. Schon, Phys. Rev. Lett. 100, 037003 (2008). [18] G. Oelsner, P. Macha, O. V. Astafiev, E. Il'ichev, M. Grajcar, U. Hubner, B. I. Ivanov, P. Neilinger, and H.- G. Meyer, Phys. Rev. Lett. 110, 053602 (2013). [19] M. Marthaler, J. Leppakangas, and J. H. Cole, Phys. Rev. B 83, 180505 (2011). [20] M. Koppenhofer and M. Marthaler, arXiv:1510.04168 (2015). [21] P. Macha, G. Oelsner, J.-M. Reiner, M. Marthaler, S. Andr´e, G. Schon, U. Hubner, H.-G. Meyer, E. Il'ichev, and A. V. Ustinov, Nat. Comm. 5, 5146 (2014). [22] P. Mandel, Phys. Rev. A 21, 2020 (1980). [23] The semi-classical approximation is valid in the limit of very large photon numbers, but the parameter n0(∆) ap- pears also in the present semi-quantum approximation used to study systems with finite M .
1011.3787	1	1011	2010-11-16T19:01:12	Barrier-Free Tunneling in a Carbon Heterojunction Transistor	[ "cond-mat.mes-hall" ]	Recently it has been experimentally shown that a graphene nanoribbon (GNR) can be obtained by unzipping a carbon nanotube (CNT). This makes it possible to fabricate all-carbon heterostructures that have a unique interface between a CNT and a GNR. Here we demonstrate that such a heterojunction may be utilized to obtain a unique transistor operation. By performing a self-consistent non-equilibrium Green's function (NEGF) based calculation on an atomistically defined structure, we show that such a transistor may reduce energy dissipation below the classical limit while not compromising speed - thus providing an alternate route towards ultra low-power, high-performance carbon-heterostructure electronics.	cond-mat.mes-hall	cond-mat	Barrier-Free Tunneling in a Carbon Heterojunction Transistor Youngki Yoona) and Sayeef Salahuddinb) Department of Electrical Engineering and Computer Sciences, University of California, Berkeley, CA 94720 ABSTRACT Recently it has been experimentally shown that a graphene nanoribbon (GNR) can be obtained by unzipping a carbon nanotube (CNT). This makes it possible to fabricate all-carbon heterostructures that have a unique interface between a CNT and a GNR. Here we demonstrate that such a heterojunction may be utilized to obtain a unique transistor operation. By performing a self-consistent non-equilibrium Green’s function (NEGF) based calculation on an atomistically defined structure, we show that such a transistor may reduce energy dissipation below the classical limit while not compromising speed – thus providing an alternate route towards ultra low-power, high-performance carbon-heterostructure electronics. E-mail: a)[email protected], b)[email protected] 1 Conventional transistors are thermally activated. A barrier is created whose height is modulated to control current flow. This modulation of the barrier changes the number of electrons exponentially following the Boltzmann factor exp(qV/kBT). This in turn means that to change the current by one order of magnitude (or one decade) a voltage of at least 2.3kBT/q (60 mV at room temperature) is necessary. This leads to the fact that the rate of change in voltage with respect to current on a log scale, often termed as subthreshold swing, has a classical limit of 60 mV/decade. Supply voltage requirement in a conventional transistor is constrained by this limit, currently leading to a situation that the future of electronics will be severely limited by the large voltage requirement and ensuing energy dissipation.1 It has been proposed that the classical constraint on supply voltage can be avoided by resorting to inter-band quantum tunneling so that the current flow is modulated by a tunneling barrier instead of a thermal barrier.2,3 However, conventional tunneling field-effect transistors (TFETs) suffer seriously from the very presence of a tunneling barrier as it significantly limits the current flow and thereby the speed at which the transistor can be operated. In this paper, we show that by utilizing a heterojunction created at the interface of a graphene nanoribbon (GNR)4,5 and a metallic carbon nanotube (CNT)6, a unique transistor structure may be possible, where current is not dictated by the Boltzmann factor, yet the tunneling barrier is greatly diminished. The basic idea is illustrated in Fig. 1(a). In a conventional TFET, the current must tunnel through the barrier (shown by the dashed line), the amplitude of the current being severely limited by the significant resistance posed by the barrier. One could, on the other hand, imagine that if the tunneling barrier is replaced by a material with zero bandgap (shown by the 2 shaded region), a large current should flow unhampered due to the absence of any tunneling barrier. However, to obtain this situation in practice, a very unique combination of materials is needed: (i) a material with zero bandgap whose band alignment could be modulated by a gate (which means ordinary metals would not work), and (ii) good wavefunction overlap of the zero bandgap material with left p and right i (intrinsic) regions to ensure a large current. Both these conditions can be satisfied with a heterostructure of CNT and GNR obtained by partially unzipping a metallic CNT. Notably, both the requirements, namely (i) the control of bands in a metallic CNT by doping or gate voltage2,7 and (ii) unzipping a CNT to produce a CNT-GNR interface4,5, have already been experimentally demonstrated. Here we consider a single-walled (6,0) CNT that has been predicted to be metallic by both a simple pz tight-binding model8 and ab-initio studies9. It is energetically stable and CNTs of even smaller diameters have already been experimentally deomonstrated10. The two ends of the CNT are connected to semiconducting n = 12 armchair-edge GNR (a-GNR) unzipped from the (6,0) CNT, assuming no loss of carbon atoms [Fig. 1(b)]. The schematic of a transistor made out of this interface is shown in Fig. 1(c). We shall call this a carbon-based heterojunction FET (HFET), which eventually combines the concepts of tunneling FET and conventional FET in the same device structure. Note that any GNR partially unzipped from a reasonably small-sized metallic CNT is expected to be semiconducting in nature (See supplementary material). The n = 12 a- GNR is assumed to be hydrogen-terminated, having a bandgap of Eg = 0.6 eV. Once the atomistic structure is constructed, device characteristics are calculated from a self- consistent solution of 3-D Poisson and non-equilibrium Green’s function (NEGF) based 3 transport with atomistic resolution. A nearest-neighbor tight-binding approximation was used with a pz orbital basis set (t0 = –2.7 eV) in real space approach (See supplementary material). An infinite-potential-well boundary condition was used along the width of the semiconducting n = 12 a-GNR, whereas a periodic boundary condition was applied along the circumferential direction of the metallic (6,0) CNT (See supplementary material for details of the Hamiltonian matrix and the equations solved). A modified tight-binding parameter (t = 1.12 t0) was used for the edges of GNR to treat edge-bond relaxation.11 The effect of doping was treated by the number of dopants per each carbon atom (NS/D, NCNT). In practice, GNR doping can be achieved by chemically functionalizing the edges.12 Figure 2 summarizes the simulated results. Transfer and output characteristics are shown in Figs. 2(a) and 2(b), respectively. Notably, near the threshold voltage Vt (≈ 0.15 V), which is defined where a significant current starts to show up in the linear scale, the current changes by more than 4 orders of magnitude with a change in voltage of only 100 mV, thus providing a subthreshold swing that is much below the classical limit of 60 mV/decade [Fig. 2(a)]. To compare the ON current, we also simulated a TFET based on all-semiconducting GNR with other device parameters such as source/drain doping density, gate geometry, gate oxide, source/drain extension, and power supply voltage unchanged (See supplementary material). An HFET can deliver up to ~70 times larger current than a TFET at the ON state (VG = Vt + 0.15). Since no bandgap exists inside the metallic CNT region of the HFET [Fig. 2(c)], the carrier transport takes place through the gapless energy states [Fig. 2(d)] of the p-n junction created inside the metallic CNT [left panel in Fig. 2(g)], where the transmission probability can be as large as unity due to the absence of back scattering stemming from the linear energy dispersion relation near the 4 Fermi level.6 The maximum value of the simulated transmission probability of the HFET is 0.9978 [Fig. 2(h)], which is very close to the theoretical maximum value of 1. By contrast, the maximum transmission probability is 0.015 for a GNR TFET, where a current flow is fundamentally limited by the tunneling barrier posed by the bandgap of GNR (Fig. S1). Therefore, the simulated characteristics show that the HFET can (i) provide a subthreshold swing less than the classical limit and (ii) at the same time efficiently eliminate the tunneling barrier so that the drive current can be significantly large. Looking at Fig. 2(a), one would see an asymmetry in the current of the HFET with respect to the gate voltage. To understand this, we plotted the energy-resolved current spectrum in Figs. 2(d), 2(e) and 2(f) at positive, zero and negative gate voltages, respectively. At VG ≥ 0, electrons are injected near the source Fermi level [Figs. 2(d) and 2(e)]. For VG < 0, however, the spectrum of current exists around the drain Fermi level, which indicates hole flow [Fig. 2(f)]. It is clear from these observations that the mechanism of current flow is fundamentally different at negative gate voltages from that at positive gate voltages. At a large positive gate voltage, current flows through the ‘barrier-free’ channel [Fig. 2(d)], whereas for negative voltages, the current must tunnel through a barrier by inter-band tunneling [Fig. 2(f)]. In addition, the use of a partial gate further adds to the suppression of current at negative voltages. As a result, at VG = –0.2 V, the current is still 3 orders of magnitude smaller than the current at a similar positive VG [Fig. 2(a)]. Ordinarily, a band bending will exist at the CNT-GNR interface near the source (x = 15 nm) due to the misalignment of Fermi levels of the two regions. For the simulated device structure, we have used a large doping (NCNT = 10 NS/D) partially inside the metallic CNT to show how this 5 equilibrium band bending can be minimized (Fig. 3). In practice, electrostatic doping by a separate gate is widely used to emulate the effect of local doping.2 Note that if significant band bending remains, the carriers must tunnel through a barrier at the source/channel interface and hence the current density will be limited [Fig. 3(b)]. If an ordinary metal were used, such reduction of band bending would not be possible, since the bands inside the metal cannot be effectively controlled by electrostatic means. As for the junction at the other side (x = 35 nm), there is no equilibrium band bending since it is a junction of intrinsic semi-metal and intrinsic semiconductor. The small tip existing at this interface is caused by the gate bias as observed at regular Schottky contacts. This has minimal influence on the current as the size of the barrier is very small. By artificially changing the curvature of the carbon heterostructure, we have verified that such variation has a minimal effect on electrostatics (See supplementary material for details: Fig. S2). However, edge roughness can result in bandgap variations in the GNR region and thus degrade performance. It is expected, though, that in the unzipped GNR, the edge roughness would be significantly reduced.4,5 Furthermore, various scattering phenomena and surface morphology may also adversely affect the transistor performance. To summarize, we have shown that it may be possible to use the CNT-GNR heterostructure to obtain a unique transistor action such that (i) at low voltages the device acts like a tunneling transistor, reducing the voltage requirement below the classical limit and (ii) at high voltages the tunnel barrier is effectively diminished, thus allowing a large ON current. Therefore, such a heterostructure may make it possible to combine the best practices of both the tunneling FET and 6 the conventional FET in the same device structure and thus guide the way in designing ultra low- power carbon-heterostructure electronics. The authors thank Cheol-Hwan Park for helpful discussions. This work was supported in part by FCRP center on Functional Engineered and Nano Architectonics (FENA). 7 REFERENCES L. B. Kish, Physics Letters A 305 (3-4), 144-149 (2002). J. Appenzeller, Y. M. Lin, J. Knoch, and P. Avouris, Physical Review Letters 93 (19), 196805 (2004). W. Y. Choi, B. G. Park, J. D. Lee, and T.-J. K. Liu, IEEE Electron Device Letters 28 (8), 743-745 (2007). D. V. Kosynkin, A. L. Higginbotham, A. Sinitskii, J. R. Lomeda, A. Dimiev, B. K. Price, and J. M. Tour, Nature 458 (7240), 872-876 (2009). L. Jiao, L. Zhang, X. Wang, G. Diankov, and H. Dai, Nature 458 (7240), 877-880 (2009). T. Ando, T. Nakanishi, and R. Saito, Journal of the Physical Society of Japan 67 (8), 2857-2862 (1998). C. Zhou, J. Kong, E. Yenilmez, and H. Dai, Science 290 (5496), 1552-1555 (2000). R. Saito, G. Dresselhaus, and M. S. Dresselhaus, Physical properties of carbon nanotubes. (Imperial College Press, London, 1998). I. Cabria, J. W. Mintmire, and C. T. White, Physical Review B 67 (12), 121406(R) (2003). N. Wang, Z. K. Tang, G. D. Li, and J. S. Chen, Nature 408 (6808), 50-51 (2000). Y. W. Son, M. L. Cohen, and S. G. Louie, Physical Review Letters 97 (21), 216803 (2006). X. Wang, X. Li, L. Zhang, Y. Yoon, P. K. Weber, H. Wang, J. Guo, and H. Dai, Science 324 (5928), 768-771 (2009). 1 2 3 4 5 6 7 8 9 10 11 12 8 FIGURE CAPTIONS Figure 1. Carbon-based heterojunction field-effect transistor (HFET). (a) Schematic band structure at ON (red solid line) and OFF states (blue solid line). Proposed device has a semi-metallic zero-bandgap region (shaded). EFS and EFD are Fermi levels at the source and the drain, respectively. The dashed lines show the bandgap of normal p-i-n structure in the absence of semi-metal at ON state. (b) Atomistic configuration of a carbon nanotube (CNT) and graphene nanoribbon (GNR) heterostructure. (c) Schematic device structure. The channel is a carbon-based heterostructure that consists of a metallic (6,0) CNT and semiconducting n = 12 armchair-edge GNRs (a-GNRs). Double-gate geometry with 1.5 nm thick HfO2 gate oxide (κ = 16). Lengths of source/drain extensions, metallic CNT, gate-controlled channel, and unbiased channel are LS/D = 15, LCNT = 20, LG = 20, LU = 15 nm, respectively. Source (drain) GNR is p-doped (n-doped) with doping density, NS/D = 1.810-3 /atom, which is equivalent to 0.1 /nm. Effective doping density at the left half of the metallic CNT is NCNT = 10 NS/D. Power supply voltage is VDD = 0.4 V. Figure 2. (a) ID – VG characteristics on a log scale (blue curve with left axis) and on a linear scale (green curve with right axis). Red dots indicate several important points such as threshold voltage (c), ON state (d), and OFF states (e, f). Subthreshold swing, S  dV d G  log 10 I D  are shown for various gate bias ranges. It shows an ambipolar conduction by both electrons and holes. Current saturates beyond VG = 0.3 V as a potential hump appears [see the arrow-indicated region in Fig. 2(d)] due to the partial gate structure. (See supplementary material for details: Fig. S2). The 9 minimum leakage current is shifted to VG = 0 by using appropriate metal work function engineering ( ms = 0.15 eV). (b) ID – VD plot. The output characteristic is analogous to that of an ordinary transistor except in the low voltage where a tunneling behavior can be clearly observed. (c) Local density-of-states (LDOS), shown on a log scale, at threshold voltage, Vt = 0.15 V. Metallic CNT does not have bandgap and states exist at the entire energy levels. The solid lines show the bandgap of the semiconducting GNR. The dashed line shows the self-consistent electrostatic solution of where the conduction band and the valence band touch each other within the metallic region. (d) Current spectrum (red strip) at ON state (VG = 0.3 V). (e-f) Current spectrums (red strips) at OFF states. The different energy levels of the current spectrums indicate a transition from the electron conduction (e, VG = 0 V) to the hole conduction (f, VG = –0.05 V). (g) Zoom-in plot of the spatial distribution of current as a function of energy for the region shown inside the box of Fig. 2d (left). The right panel shows energy-resolved current density. The sharp feature appearing at E ≈ -0.11 eV is due to the resonant tunneling states originating from the barrier profile at this specific voltage. (h) Transmission probability, T as a function of energy, E and gate voltage, VG. The largest value of T is 0.9978. Figure 3. Effect of doping in the metallic CNT. (a) Band profiles for CNT doping density, NCNT = NS/D (insufficient doping, red dashed lines) and NCNT = 10 NS/D (used in our simulation, blue solid lines). The left half of the metallic CNT is 10 doped to reduce the inherent band bending at the heterojunction. Red strip shows current spectrum for NCNT = NS/D at VG = 0.3 V. (b) Energy-resolved current density at VG = 0.3 V. 11 12 13 14
1812.09815	1	1812	2018-12-24T03:09:56	Observation of Moir\'e Excitons in WSe2/WS2 Heterostructure Superlattices	[ "cond-mat.mes-hall" ]	Moir\'e superlattices provide a powerful tool to engineer novel quantum phenomena in two-dimensional (2D) heterostructures, where the interactions between the atomically thin layers qualitatively change the electronic band structure of the superlattice. For example, mini-Dirac points, tunable Mott insulator states, and the Hofstadter butterfly can emerge in different types of graphene/boron nitride moir\'e superlattices, while correlated insulating states and superconductivity have been reported in twisted bilayer graphene moir\'e superlattices. In addition to their dramatic effects on the single particle states, moir\'e superlattices were recently predicted to host novel excited states, such as moir\'e exciton bands. Here we report the first observation of moir\'e superlattice exciton states in nearly aligned WSe2/WS2 heterostructures. These moir\'e exciton states manifest as multiple emergent peaks around the original WSe2 A exciton resonance in the absorption spectra, and they exhibit gate dependences that are distinctly different from that of the A exciton in WSe2 monolayers and in large-twist-angle WSe2/WS2 heterostructures. The observed phenomena can be described by a theoretical model where the periodic moir\'e potential is much stronger than the exciton kinetic energy and creates multiple flat exciton minibands. The moir\'e exciton bands provide an attractive platform to explore and control novel excited state of matter, such as topological excitons and a correlated exciton Hubbard model, in transition metal dichalcogenides.	cond-mat.mes-hall	cond-mat	Observation of Moiré Excitons in WSe2/WS2 Heterostructure Superlattices Chenhao Jin1†, Emma C. Regan1,2†, Aiming Yan1,3, M. Iqbal Bakti Utama1,4, Danqing Wang1,2, Ying Qin5, Sijie Yang5, Zhiren Zheng1, Kenji Watanabe6, Takashi Taniguchi6, Sefaattin Tongay5, Alex Zettl1,3,7, Feng Wang1,3,7* 1 Department of Physics, University of California at Berkeley, Berkeley, California 94720, United States. 2 Graduate Group in Applied Science and Technology, University of California at Berkeley, Berkeley, California 94720, United States 3 Material Science Division, Lawrence Berkeley National Laboratory, Berkeley, California 94720, United States. 4 Department of Materials Science and Engineering, University of California at Berkeley, Berkeley, California 94720, United States. 5 School for Engineering of Matter, Transport and Energy, Arizona State University, Tempe, Arizona 85287, United States 6 National Institute for Materials Science, 1-1 Namiki, Tsukuba, 305-0044, Japan. 7 Kavli Energy NanoSciences Institute at University of California Berkeley and Lawrence Berkeley National Laboratory, Berkeley, California 94720, United States. † These authors contributed equally to this work * Correspondence to: [email protected] Abstract: Moiré superlattices provide a powerful tool to engineer novel quantum phenomena in two- dimensional (2D) heterostructures, where the interactions between the atomically thin layers qualitatively change the electronic band structure of the superlattice. For example, mini-Dirac points, tunable Mott insulator states, and the Hofstadter butterfly can emerge in different types of graphene/boron nitride moiré superlattices, while correlated insulating states and superconductivity have been reported in twisted bilayer graphene moiré superlattices1-12. In addition to their dramatic effects on the single particle states, moiré superlattices were recently predicted to host novel excited states, such as moiré exciton bands13-15. Here we report the first observation of moiré superlattice exciton states in nearly aligned WSe2/WS2 heterostructures. These moiré exciton states manifest as multiple emergent peaks around the original WSe2 A exciton resonance in the absorption spectra, and they exhibit gate dependences that are distinctly different from that of the A exciton in WSe2 monolayers and in large-twist-angle WSe2/WS2 heterostructures. The observed phenomena can be described by a theoretical model where the periodic moiré potential is much stronger than the exciton kinetic energy and creates multiple flat exciton minibands. The moiré exciton bands provide an attractive platform to explore and control novel excited state of matter, such as topological excitons and a correlated exciton Hubbard model, in transition metal dichalcogenides. A moiré superlattice can form between two atomically thin materials with similar lattices, and its period varies continuously with the twist angle between the constituent layers. The periodic moiré pattern introduces a new length and energy scale, providing a powerful new way to control quantum phenomena in 2D heterostructures1-12. The most striking moiré superlattice phenomena emerge in the "strong-coupling" regime, where the periodic moiré potential dominates over the relevant kinetic energy and qualitatively changes the electronic band structure and the electron wavefunction in the heterostructure. Recently, it was reported that "strong-coupling" moiré superlattices can generate flat electronic bands, leading to exotic phases such as correlated insulating states and superconductivity in magic-twist-angle bilayer graphene and tunable Mott insulator states in trilayer graphene/boron nitride heterostructures1-6. Moiré superlattices also offer exciting opportunities to engineer the band structure of collective excitations, such as excitons in 2D semiconducting heterostructures. Monolayer transition metal dichalcogenides (TMDCs) are direct bandgap semiconductors that feature strong light-exciton interactions and dramatically enhanced electron-electron interactions. Exciton binding energies in monolayer TMDCs can be hundreds of meV -- orders of magnitude larger than what is seen in typical semiconductors like silicon or GaAs16,17 -- which leads to well defined dispersive exciton bands in the Brillouin zone. Recently it was predicted that moiré superlattices in the "strong coupling" regime could lead to moiré exciton minibands in TMDC heterostructures13-15, which are distinctly different from the separate electron and hole minibands due to the strong electron- hole correlation. Here we report the first experimental observation of moiré excitons in nearly aligned WSe2/WS2 heterostructures. The moiré superlattice splits the WSe2 A exciton resonance into multiple peaks that all exhibit comparable oscillator strengths in the absorption spectrum. Furthermore, the emergent exciton peaks show distinct doping dependences that are different from that of the A exciton in WSe2 monolayers and in large-twisting-angle WSe2/WS2 heterostructures. This unusual behavior can be understood using an empirical model for moiré excitons with a peak-to- peak exciton moiré potential of 250 meV. The periodic potential energy is much larger than the exciton kinetic energy of 8 meV within the first mini-Brillouin zone, and it completely changes the exciton dispersion in the moiré superlattice, leading to flat low-energy exciton bands with highly localized exciton density of states. The near-aligned WSe2/WS2 moiré superlattice can therefore potentially host a variety of novel excitonic states, such as topological exciton bands and a strongly-correlated exciton Hubbard model13-15,18,19. Figure 1a and b show an optical microscopy image and a side-view schematic of a representative WSe2/WS2 heterostructure device (D1). The results measured from device D1 are reproducible in all near-aligned heterostructures that we fabricated (see supplementary information). The WSe2/WS2 heterostructure is encapsulated in thin hexagonal boron nitride (hBN) layers. Few layer graphite (FLG) flakes are used for both the bottom gate and the electrical contacts to the heterostructure. The carrier concentration in the heterostructure can be tuned continuously with the back gate voltage Vg. All of the two-dimensional materials were first mechanically exfoliated from bulk crystals and then stacked together by a dry transfer method using a polyethylene terephthalate (PET) stamp (see Methods). The whole stack was then transferred onto a 90 nm SiO2/Si substrate. The relative twist angle between the WSe2 and WS2 layers was determined optically using polarization-dependent second harmonic generation (SHG) measurements. Characteristic six-fold SHG patterns are clearly observed for the WSe2 layer (green color) and the WS2 layer (yellow color) in Fig. 1c, from which we can determine the relative twist angle between the two layers to be <0.5 degree (see supplementary information). For a near-zero twist angle heterostructure, the lattice mismatch between the two layers is dominated by the intrinsic lattice constant difference of about 4% (Ref. 20) , which leads to a moiré periodicity (cid:1838)(cid:3014)(cid:3406) 8 nm (Fig. 1d). Similar moiré superlattices have been observed experimentally using scanning tunneling microscope (STM) in aligned WSe2/MoS2 heterostructures21,22, which have lattice constants almost identical to the WSe2/WS2 heterostructure. To confirm the formation of the moiré superlattice in our WSe2/WS2 heterostructure, we prepared another device (D2) on a transmission electron microscopy (TEM) grid and collected high resolution TEM images of the device (see methods). Rich sets of diffraction patterns are observed in the Fourier transform of the TEM image in Fig. 1e, which shows the main diffraction peaks from the WSe2, WS2, and hBN layers as well as side peaks from the local reconstruction of the atomic structures due to layer-layer interactions. A zoomed- in image (Fig. 1f) shows a well-defined hexagonal lattice in the center region that corresponds to a real space periodicity of ~ 8 nm. This indicates that a periodic lattice distortion with ~ 8 nm periodicity exists in the heterostructure in real space, which is consistent with strong layer-layer interaction and significant lattice reconstruction within the moiré superlattice observed in previous STM studies21,22. We probe the moiré excitons in WSe2 with optical spectroscopy at a temperature of 10 Kelvin. Figure 2a shows the photoluminescence (PL) spectrum of device D1 (blue curve) and a reference monolayer WSe2 sample (green curve) in both linear (main panel) and logarithmic (inset) scale. The heterostructure PL features a single peak at 1.409eV, corresponding to the emission from the interlayer exciton, and does not show any emission from WSe2 A exciton. This indicates an efficient interlayer charge transfer across the whole measured region that leads to strong quenching of the WSe2 PL23,24. The exciton absorption in the same heterostructure region is directly probed through reflection contrast measurements (top panel in Fig. 2b), where a slowly varying background has been subtracted to better resolve the resonances (see supplementary information). The absorption spectrum from D1, a nearly aligned heterostructure, is strikingly different from that of a large-twist-angle WSe2/WS2 heterostructure measured at the same condition (lower panel in Fig. 2b). We focus on the spectral range between 1.6 to 1.8 eV as it is well-separated from all WS2 resonances. While the large-twist-angle heterostructure shows only a WSe2 A exciton peak at 1.715 eV, three prominent peaks emerge in device D1 at 1.683, 1.739, and 1.776 eV, labeled as resonance I, II, and III, respectively. All three resonances show strong absorption, with peak II and peak III having oscillator strengths at 20% and 50% of the peak I value. To better understand these new exciton peaks, we measured the photoluminescence excitation (PLE) spectrum of the device D1 (black dots in Fig. 2c) by monitoring the interlayer exciton emission intensity as the excitation photon energy was swept from 1.6 to 2.1 eV. The excitation spectrum shows perfect correspondence to the results from reflection spectroscopy (blue line in Fig. 2c). In particular, the new exciton peaks between 1.6 to 1.8 eV all give rise to strong enhancement of the interlayer exciton emission at 1.409 eV, indicating that they arise from the strongly coupled WSe2/WS2 heterostructure. To further investigate the nature of the emergent exciton resonances, we measure their doping dependence (Fig. 3a). The horizontal and vertical axes represent the photon energy and gate voltage Vg, respectively, and the color corresponds to reflection contrast. The charge neutral point is approximately at Vg=0, and positive and negative Vg values correspond to electron- and hole-doping, respectively. The three main peaks in the WSe2 A exciton range show rich behavior, with dramatic spectral changes upon both electron and hole doping. The strong gate-dependence upon electron doping is particularly remarkable: Due to the type-II band alignment in WSe2/WS2 heterostructures, doped electrons reside mostly in the WS2 layers and tend to have relatively weak effects on the intra-layer A exciton resonance in WSe2 (Ref. 25,26). Indeed, previous studies of large-twist-angle WSe2/WS2 heterostructures shows that the WSe2 A exciton resonance only experiences a slight redshift upon electron doping of the heterostructure26. In contrast, the exciton peaks in D1, a nearly aligned heterostructure with a large moiré superlattice, show unusual dependences on electron doping that varies for different peaks (Fig. 3b). Both peak I and peak III are strongly modified at increasing electron concentration: Peak I shows a strong blueshift and transfers its oscillator strength to another emergent peak at lower energy (I'), and peak III also shows a strong blueshift with diminished oscillator strength. On the other hand, peak II remains largely unchanged except for a small energy shift. The strong effect of electrons in WS2 on certain exciton transitions in WSe2 indicates dramatically enhanced interlayer electron-exciton interactions through the moiré superlattice. In addition, the strikingly different gating behavior of the exciton peaks cannot be explained by any established electron-exciton interactions in monolayers, such as dielectric screening effects or trion formation, which affect all exciton peaks in a similar fashion27-29. Instead, it indicates that the exciton peaks I, II, and III correspond to very different exciton states within the moiré superlattice. Both the emergence of multiple exciton peaks around the WSe2 A exciton resonance and their peculiar electron doping dependence can be understood within an empirical theory in which a periodic moiré exciton potential in the "strong coupling" regime is introduced. We follow the theoretical model in Ref. 13 and describe the center-of-mass motion of WSe2 A excitons using the Hamiltonian (cid:1834)(cid:3404)(cid:1834)(cid:2868)(cid:3397)(cid:3533)(cid:1848)(cid:3037)exp (cid:4666)(cid:1861)(cid:2184)(cid:2192)∙(cid:2200)(cid:4667) , (cid:4666)1(cid:4667) (cid:2874) (cid:3037)(cid:2880)(cid:2869) The exciton dispersion in the mini-Brillouine Zone (mBZ) can be directly calculated from this photons have negligible momentum, only the lowest energy exciton can interact with light, information). Owing to the three-fold rotational symmetry and Hermitian requirement, only one where (cid:1834)(cid:2868) is the low energy effective Hamiltonian for the A exciton 1s state in monolayer WSe2. (cid:1848)(cid:3037) describes the effective potential on the exciton created by the moiré pattern; its momentum is given by the reciprocal lattice vectors of the moiré superlattice, (cid:2184)(cid:3037) (see supplementary component in (cid:1848)(cid:3037) is independent and can be defined as (cid:1848)(cid:2869)(cid:3404)(cid:1848)exp(cid:4666)(cid:1861)(cid:2032)(cid:4667). low energy (Fig. 4a). These two bands are degenerate at (cid:2011) point, and have parabolic and linear giving a single strong peak at (cid:1831)(cid:3404)(cid:1831)(cid:2868) in the absorption spectrum (Fig. 4d). The moiré potential can mix exciton states with momenta that differ by (cid:2184)(cid:3037), leading to additional absorption peaks from the (cid:2011) point states of higher-energy minibands. When the moiré potential is weak ((cid:1848)(cid:3404)5 meV, Fig. 4b), the exciton dispersion remains largely dispersion, respectively, as a consequence of the intervalley exchange interaction13,30. Because model (see Fig. 4, a-c). Without the moiré potential, the exciton shows two continuous bands at unchanged. Therefore, the emergent side peak in absorption always appears at ~30 meV above the main peak, regardless of the exact form of the moiré potential (Fig. 4e). Furthermore, the amplitude of the side peak is orders of magnitude smaller than the main peak due to the weak mixing between states. These features pose sharp contrast to the experimental absorption spectrum and cannot explain our observations. On the other hand, a larger moiré potential that corresponds to the "strong coupling" regime dramatically modifies the exciton dispersion, (see Fig. 4c). As a result, the energy of the moiré exciton states in different minibands (labeled I to III in Fig. 4c), as well as of the corresponding absorption peaks (peak I to III in Fig. 4f), become sensitively dependent on the moiré potential. In addition, the strong mixing between different exciton states make their oscillator strengths comparable to each other. By taking (cid:1848)(cid:3404)25 meV and (cid:2032)(cid:3404)15°, the simulated absorption spectrum can reproduce our experimental observation (Fig. 4f, see also supplementary information). This moiré potential has a peak-to-peak amplitude of ~ 250 meV, which is much larger than the exciton kinetic energy of ~ 8 meV within the first mini-Brillouin zone (see supplementary information). The dramatic change in the exciton dispersion in momentum space implies that the exciton center-of-mass wavefunction is also strongly modified in real space. Figures 4g-i show the distribution of the exciton probability density for states I to III in the moiré superlattice. The originally homogeneous wavefunction distribution is dramatically changed by the moiré potential. For example, the lowest energy state (state I) is trapped around the moiré potential minimum (labeled as point ) in a length scale much smaller than the moiré superlattice (Fig. 4g). As a result, excitons in different moiré periods are well separated, forming an effective exciton lattice with significantly reduced hopping between neighboring lattice sites, which is consistent with the significantly reduced bandwidth in their momentum dispersion. The peculiar wavefunction distribution of moiré excitons in the "strong coupling" regime introduces a new degree of freedom that is determined by the exciton location in the moiré superlattice. Interestingly, both peak I and III are centering around the same point , while peak II has its largest amplitude at a different point Fig. 4, h-i). The difference in real space position between moiré exciton states can account for their distinctive doping-dependence: The doped electrons will also have localized density of states in real space21,22. If the gate-induced electrons in WS2 are also localized at point  in the moiré superlattice, they will predominantly change the exciton peaks I and III and leave exciton peak II little affected, as observed in the experiment. We note that a complete description of the moiré exciton optical spectra will require a much more sophisticated model that fully accounts for the lattice relaxation and corrugation, as well as the interlayer electronic states hybridization in the heterostructure moiré superlattice, which is beyond the scope of this study. Nevertheless, our simple moiré exciton model captures most of the salient features observed experimentally, and it shows that WS2/WSe2 heterostructures exhibit sufficiently strong interlayer interaction to enter the "strong coupling" regime for excitons, where the moiré excitons become spatially concentrated at well-separated points and form a quantum array in an extended moiré superlattice13-15. The significantly reduced exciton bandwidth also makes this artificial exciton lattice a promising platform for realizing exotic phases such as a topological exciton insulator and a strongly-correlated exciton Hubbard model. Methods: Heterostructure preparation for optical measurements: WSe2/WS2 heterostructures were fabricated using a dry transfer method with a polyethylene terephthalate (PET) stamp31. Monolayer WSe2, monolayer WS2, few-layer graphene, and thin hBN flakes were exfoliated onto silicon substrates with a 90 nm SiO2 layer. Polarization-dependent SHG was used to determine the relative angle between the WS2 and WSe2 flakes (see text and supplementary information for details). A PET stamp was used to pick up the top hBN flake, the WS2 monolayer, the WSe2 monolayer, several few-layer graphene flakes for electrodes, the bottom hBN flake, and the few-layer graphene back gate in sequence. The angle of the PET stamp was adjusted between picking up the WS2 and the WSe2 to assure a near-zero twist angle between the flakes. The PET stamp with the above heterostructure was then stamped onto a clean Si substrate with 90 nm SiO2, and the PET was dissolved in dichloromethane for 12 hours at room temperature. The PET and samples were heated to 60 °C during the pick-up steps and to 130 °C for the final stamp process. Contacts (~75 nm gold with ~5 nm chromium adhesion layer) to the few-layer graphene flakes were made using electron-beam lithography and electron-beam evaporation. Heterostructure preparation for TEM: WS2/WSe2 heterostructures were prepared for TEM characterization using a modified dry transfer technique with a PET stamp. WS2 monolayers, WSe2 monolayers, and thin hBN flakes were exfoliated and SHG measurements were used to determine flake orientation, as described above. A PET stamp was used to pick up the top hBN flake, the WS2 monolayer, the WSe2 monolayer, and the bottom hBN flake in sequence. A Ted Pella Quantifoil TEM grid with 2 µm holes (657-200-AU) was placed on a silicon chip that was attached to the transfer stage. The PET stamp was lowered until it was in contact with the TEM grid, and then the temperature was raised to 80 °C until the stamp and the grid were well contacted, as seen through an optical microscope. The PET stamp and TEM grid were then placed in dichloromethane for 12 hours at room temperature to dissolve the PET. High-resolution TEM imaging and FFT analysis: High-resolution TEM images of the hBN- encapsulated WS2/WSe2 heterostructure were taken under 200 keV accelerating voltage for the electron beam. A fast Fourier transform with high-pass filter was performed on each high- resolution TEM image to show the superlattice periodicity in the WS2/WSe2 heterostructure in reciprocal space. Data availability. The data that support the findings of this study are available from the corresponding author upon reasonable request. Acknowledgements: This work was supported primarily by the Director, Office of Science, Office of Basic Energy Sciences, Materials Sciences and Engineering Division of the U.S. Department of Energy under contract no. DE-AC02-05-CH11231 (van der Waals heterostructures program, KCWF16). The device fabrication was supported by the NSF EFRI program (EFMA-1542741). PLE spectroscopy of the heterostructure is supported by the US Army Research Office under MURI award W911NF-17-1-0312. Growth of hexagonal boron nitride crystals was supported by the Elemental Strategy Initiative conducted by the MEXT, Japan and JSPS KAKENHI Grant Numbers JP15K21722. S.T. acknowledges the support from NSF DMR 1552220 NSF CAREER award for the growth of WS2 and WSe2 crystals. Author contributions: C.J. and E.C.R. contributed equally to this work. F.W. and C.J. conceived the research. C.J. and E.C.R. carried out optical measurements. A.Y. and A.Z. performed TEM measurements. C.J., F.W., E.C.R. and D.W. performed theoretical analysis. E.C.R., M.I.B.U., D.W. and Z.Z. fabricated van der Waals heterostructures. Y.Q., S.Y. and S.T. grew WSe2 and WS2 crystals. K.W. and T.T. grew hBN crystals. All authors discussed the results and wrote the manuscript. Hunt, B. et al. Massive Dirac Fermions and Hofstadter Butterfly in a van der Waals Heterostructure. Science 340, 1427-1430, (2013). Ponomarenko, L. A. et al. Cloning of Dirac fermions in graphene superlattices. Nature 497, 594-597, (2013). Dean, C. R. et al. Hofstadter's butterfly and the fractal quantum Hall effect in moire superlattices. Nature 497, 598-602, (2013). Cao, Y. et al. Unconventional superconductivity in magic-angle graphene superlattices. Nature 556, 43-+, (2018). Cao, Y. et al. Correlated insulator behaviour at half-filling in magic-angle graphene superlattices. Nature 556, 80-+, (2018). Chen, G. et al. Gate-Tunable Mott Insulator in Trilayer Graphene-Boron Nitride Moiré Superlattice. Arxiv, 1803.01985, (2018). Shi, Z. W. et al. Gate-dependent pseudospin mixing in graphene/boron nitride moire superlattices. Nat Phys 10, 743-747, (2014). Spanton, E. M. et al. Observation of fractional Chern insulators in a van der Waals heterostructure. Science 360, 62-66, (2018). Wallbank, J. R., Patel, A. A., Mucha-Kruczynski, M., Geim, A. K. & Falko, V. I. Generic miniband structure of graphene on a hexagonal substrate. Phys Rev B 87, (2013). Song, J. C. W., Samutpraphoot, P. & Levitov, L. S. Topological Bloch bands in graphene superlattices. P Natl Acad Sci USA 112, 10879-10883, (2015). Gorbachev, R. V. et al. Detecting topological currents in graphene superlattices. Science 346, 448-451, (2014). Lee, M. et al. Ballistic miniband conduction in a graphene superlattice. Science 353, 1526-1529, (2016). References 1 2 3 4 5 6 7 8 9 10 11 12 14 16 17 18 19 20 21 13 Wu, F. C., Lovorn, T. & MacDonald, A. H. Topological Exciton Bands in Moire Heterojunctions. Phys Rev Lett 118, (2017). Yu, H. Y., Liu, G. B., Tang, J. J., Xu, X. D. & Yao, W. Moire excitons: From programmable quantum emitter arrays to spin-orbit-coupled artificial lattices. Sci Adv 3, (2017). 15 Wu, F. C., Lovorn, T. & MacDonald, A. H. Theory of optical absorption by interlayer excitons in transition metal dichalcogenide heterobilayers. Phys Rev B 97, (2018). Chernikov, A. et al. Exciton Binding Energy and Nonhydrogenic Rydberg Series in Monolayer WS2. Phys Rev Lett 113, (2014). Ye, Z. L. et al. Probing excitonic dark states in single-layer tungsten disulphide. Nature 513, 214-218, (2014). Greiner, M., Mandel, O., Esslinger, T., Hansch, T. W. & Bloch, I. Quantum phase transition from a superfluid to a Mott insulator in a gas of ultracold atoms. Nature 415, 39-44, (2002). Fisher, M. P. A., Weichman, P. B., Grinstein, G. & Fisher, D. S. Boson Localization and the Superfluid-Insulator Transition. Phys Rev B 40, 546-570, (1989). Schutte, W. J., Deboer, J. L. & Jellinek, F. Crystal-Structures of Tungsten Disulfide and Diselenide. J Solid State Chem 70, 207-209, (1987). Zhang, C. D. et al. Interlayer couplings, Moire patterns, and 2D electronic superlattices in MoS2/WSe2 hetero-bilayers. Sci Adv 3, (2017). Pan, Y. et al. Quantum-Confined Electronic States Arising from the Moire Pattern of MoS2-WSe2 Heterobilayers. Nano Lett 18, 1849-1855, (2018). Hong, X. P. et al. Ultrafast charge transfer in atomically thin MoS2/WS2 heterostructures. Nat Nanotechnol 9, 682-686, (2014). 24 Wang, K. et al. Interlayer Coupling in Twisted WSe2/WS2 Bilayer Heterostructures Revealed by Optical Spectroscopy. Acs Nano 10, 6612-6622, (2016). Kang, J., Tongay, S., Zhou, J., Li, J. B. & Wu, J. Q. Band offsets and heterostructures of two-dimensional semiconductors. Appl Phys Lett 102, (2013). Jin, C. H. et al. Imaging of pure spin-valley diffusion current in WS2-WSe2 heterostructures. Science 360, 893-896, (2018). 27 Wang, F. et al. Interactions between individual carbon nanotubes studied by Rayleigh scattering spectroscopy. Phys Rev Lett 96, (2006). Raja, A. et al. Coulomb engineering of the bandgap and excitons in two-dimensional materials. Nat Commun 8, (2017). 29 Mak, K. F. et al. Tightly bound trions in monolayer MoS2. Nat Mater 12, 207-211, (2013). Yu, H. Y., Liu, G. B., Gong, P., Xu, X. D. & Yao, W. Dirac cones and Dirac saddle points of bright excitons in monolayer transition metal dichalcogenides. Nat Commun 5, (2014). 31 Wang, L. et al. One-Dimensional Electrical Contact to a Two-Dimensional Material. 22 23 25 26 28 30 Science 342, 614-617, (2013). Figures: Figure 1│Moiré superlattice in near-zero twist angle WSe2/WS2 heterostructure. a, b, Optical microscope image (a) and side-view illustration (b) of a representative near-zero twist angle heterostructure (device D1). c, The polarization-dependent SHG signal measured on the monolayer WSe2 (green circles) and WS2 (yellow circles) regions in device D1 and the corresponding fittings (green and yellow curves). The SHG results confirm that the WSe2 and WS2 twist angle is smaller than our experimental uncertainty of 0.5 degree. d, Illustration of the moiré superlattice in real space. The superlattice vectors, a1 and a2, have a length of ~ 8 nm. e, f, Fourier transform of the TEM image of another near-zero twist angle WSe2/WS2 heterostructure (device D2) (e) and the zoom-in plot at the center region (f). Representative first order diffraction points are labelled by circles in e for top and bottom hBN (light blue) and the WSe2/WS2 heterostructure (red), respectively. Two well-defined hexagonal lattices are observed in the center region, and the inner one (arrow in f) corresponds to a periodic lattice distortion with ~ 8 nm periodicity, consistent with the formation of a moiré superlattice. Scale bar: 2 nm-1. Figure 2│Moiré exciton states in WSe2/WS2 moiré superlattice. a, PL spectrum of device D1 (blue) and a reference monolayer WSe2 sample (green curve) in both linear (main panel) and logarithmic (inset) scale. The complete disappearance of monolayer PL in the heterostructure indicates efficient interlayer coupling across the whole measured region. b, Reflection contrast spectrum of device D1 (blue color, upper panel) compared to a large-twist-angle WSe2/WS2 heterostructure device (navy color, lower panel). The latter only shows a single resonance in the energy range between 1.6 to 1.8 eV from the WSe2 A exciton state. In contrast, the moiré superlattice formed in device D1 gives rise to three prominent peaks with comparable oscillator strength in this range (labelled as I to III), corresponding to different moiré exciton states. c, Comparison between the interlayer exciton photoluminescence excitation spectrum (black dots) and the reflection spectrum (blue curve). Strong enhancement of interlayer exciton photoluminescence is observed when excited at all moiré exciton states, indicating that all states are from the strongly-coupled WSe2/WS2 heterostructure. Figure 3│Doping dependence of the moiré exciton resonances. a, Gate-dependent reflection contrast spectrum of device D1 with both electron- (positive Vg) and hole- (negative Vg) doping. White dashed box encloses the photon energy range near the WSe2 A exciton, where the three prominent moiré exciton states (labeled as I, II, and III) appear. b, Detailed reflection contrast spectra in the WSe2 A exciton range on the electron-doping side, which reveal unusual gate- dependence of the moiré exciton states: Peak I shows a strong blueshift and transfers its oscillator strength to another emerging peak at lower energy (I'), and peak III shows a strong blueshift with diminished oscillator strength. On the other hand, peak II remains largely unchanged except for a small energy shift. These observations cannot be explained by any established electron-exciton interactions in TMD monolayers. Figure 4│ Moiré excitons in the "strong-coupling" regime. a-c, WSe2 A exciton dispersion in the mini-Brillouin zone with a moiré potential parameter V=0 meV (zero coupling, a), 5meV (weak couping, b), and 25 meV (strong coupling, c), and the corresponding absorption spectrum (d-f). A broadening of 2 meV is used in calculating the absorption spectrum. Inset in (a) illustrates the mini-Brillouin zone in momentum space and the high-symmetry points. The absorption spectrum features a single resonance at energy E0 at zero moiré potential (d), and shows a small side peak fixed at ~ 30 meV under a weak moiré potential (e). These features cannot explain our experimental observation. On the other hand, the exciton dispersion is strongly modified in the "strong-coupling" regime due to the strong mixing between different exciton states (c), which gives rise to multiple moiré exciton peaks with comparable oscillator strength in the absorption spectrum (peak I to III in f) from different moiré mini-bands (state I to III in c). The experimentally observed reflection contrast can be reproduced by taking (cid:1848)(cid:3404)25 meV and (cid:2032)(cid:3404)15°. g-i, Real space distribution of exciton center-of-mass wavefunction in the "strong-coupling" regime. The strong moiré potential traps the lowest-energy exciton state I around its minimum point  (g). Interestingly, state III is also centered at point  but state II is centered at a different point (h, i), which can account for the remarkably different gate dependence between the moiré exciton states. Supplementary Materials for Observation of Moiré Excitons in WSe2/WS2 Heterostructure Superlattices Chenhao Jin†, Emma C. Regan†, Aiming Yan, M. Iqbal Bakti Utama, Danqing Wang, Ying Qin, Sijie Yang, Zhiren Zheng, Kenji Watanabe, Takashi Taniguchi, Sefaattin Tongay, Alex Zettl, Feng Wang* † These authors contributed equally to this work * Correspondence to: [email protected] 1. Results from additional near-aligned heterostructures 2. Determination of the relative twist angle between WSe2 and WS2 layers 3. Subtraction of background in reflection contrast spectra 4. Dependence of moiré exciton absorption spectra on the moiré potential 5. Spatially localized exciton center-of-mass wavefunctions 1. Results from additional near-aligned heterostructures The moiré excitons observed in device D1 and described in the text are reproducible in all near- aligned heterostructures that we fabricated, with crystals from several commercial and academic sources. For example, Fig. S1 shows the reflection contrast and PLE measurement results from another near-aligned heterostructure device D3 when it is slightly n-doped. Four resonances (I', I, II, III) are clearly observed between 1.6 and 1.9 eV in both reflection contrast and PLE spectra, whose energy and amplitude match well with the corresponding resonances observed for device D1. I' I II III 1.2 0.8 0.4 ) . . u a ( y t i s n e t n i L P / R R d -0.4 0.0 0.4 0.8 0.0 1.6 1.7 1.8 1.9 Energy (eV) 2.0 2.1 Fig. S1. Reflection contrast (blue) and PLE (black) spectra of another near- aligned heterostructure D3 at slight n-doping. 2. Determination of the relative twist angle between WSe2 and WS2 layers The crystal orientation of WSe2 and WS2 flakes can be obtained from the second harmonic generation (SHG) polarization dependence. However, since the SHG patterns of both materials have six-fold rotational symmetry, the case of AA stacking (~ 0 twist angle) and AB stacking (~ 60 twist angle) cannot be differentiated by separately measuring the SHG of each material. On the other hand, the two cases can be distinguished by directly measuring the SHG of the heterostructure: in AA (AB) stacking case, the second harmonic field of the two layers will constructively (destructively) interfere, giving SHG signal stronger (weaker) than monolayers. Figure S2 shows the SHG intensity from WSe2 alone, WS2 alone, and heterostructure regions in device D1 measured with a 900 nm incident beam and the same experimental configuration. The WSe2 and WS2 regions show similar SHG intensity, while the heterostructure region show SHG intensity approximately four times larger than the monolayer. This result indicates that the twist angle between WSe2 and WS2 layers is near zero, i.e. AA-stacking. 1250 1000 750 500 250 s t n u o c G H S 0 420 HT WSe2 WS2 440 460 Wavelength (nm) 480 Fig. S2. SHG signal of device D1 measured on the WSe2 alone (red), WS2 alone (blue), and heterostructure (black) regions with the same experimental configuration. 3. Subtraction of background in reflection contrast spectra original spectrum background / R R d -0.4 0.0 0.4 0.8 1.6 1.8 2.0 2.2 2.4 Energy (eV) Fig. S3. Original reflection contrast (black) and background (red) obtained from polynomial fitting. the slow-varying The WSe2/WS2 heterostructure is encapsulated in two pieces of hBN and placed on 90 nm SiO2/Si substrate. This multi-film structure introduces interference between reflections at different interfaces, leading to a background signal from the real part of the dielectric function of the heterostructure32, see Fig. S3. Because the phase difference between the multi-reflections in the interference depend on the light wavelength, this background signal will vary with photon energy. To better resolve the exciton absorption resonances (i.e. the imaginary part of dielectric function), we obtain a slowly-varying background signal through polynomial fitting of the spectra using regions away from resonances (red curve in Fig. S3). The same background is used universally to obtain background-subtracted spectra at charge neutral (Fig. 2 in text) and at different doping levels (Fig. 3 in text). The validity of the background subtraction is also confirmed by the comparison to PLE spectrum (Fig. 2c in text) because the latter only depends on the imaginary part of dielectric function and is background free. 4. Dependence of moiré exciton absorption spectra on the moiré potential The low-energy effective Hamiltonian of the A-exciton in monolayer WSe2 is described by (cid:1834)(cid:2868)(cid:3404)(cid:4678)(cid:1831)(cid:2868)(cid:3397)(cid:1328)(cid:2870)(cid:2173)(cid:2870)2(cid:1839)(cid:4679)(cid:2028)(cid:2868)(cid:3397)(cid:1836)(cid:2173)(cid:2028)(cid:2868)(cid:3397)(cid:1836)(cid:2173)(cid:3427)cos(cid:3435)2(cid:2038)(cid:2173)(cid:3439)(cid:2028)(cid:3051)(cid:3397)sin(cid:3435)2(cid:2038)(cid:2173)(cid:3439)(cid:2028)(cid:3052)(cid:3431), where (cid:2173) is exciton total momentum, (cid:2038)(cid:2173) is the polar angle of (cid:2173) in the momentum space, (cid:1839)(cid:3406)(cid:1865)(cid:2868) is the total mass of the electron and the hole, (cid:1836)(cid:3404)0.04eV∙nm describes the intra- and inter-valley exchange interaction, and (cid:2028)(cid:3037) (cid:4666)(cid:1862)(cid:3404)0,(cid:1876),(cid:1877),(cid:1878)(cid:4667) is the Pauli matrices for valley pseudospin13. Without the moiré potential, only the (cid:2173)(cid:3404)0 exciton is bright due to the negligible momentum of photons, which has energy (cid:1831)(cid:3404)(cid:1831)(cid:2868). With the moiré potential, the total Hamiltonian of the exciton becomes: (cid:1834)(cid:3404)(cid:1834)(cid:2868)(cid:3397)(cid:1848)(cid:3040)(cid:3404)(cid:1834)(cid:2868)(cid:3397)(cid:3533)(cid:1848)(cid:3037)exp (cid:4666)(cid:1861)(cid:2184)(cid:3037)∙(cid:2200)(cid:4667) . intuitively understood from perturbation theory. The first order correction to the wavefunction (cid:2874) (cid:3037)(cid:2880)(cid:2869) Φ(cid:3435)(cid:2184)(cid:3037)(cid:3439)(cid:2869)(cid:3398)Φ(cid:3435)(cid:2184)(cid:3037)(cid:3439)(cid:2868)(cid:3404)(cid:3452)Φ(cid:3435)(cid:2184)(cid:3037)(cid:3439)(cid:2868)(cid:4698)(cid:1848)(cid:3040)(cid:4698)Φ(cid:4666)0(cid:4667)(cid:2868)(cid:3456) (cid:1831)(cid:3435)(cid:2184)(cid:3037)(cid:3439)(cid:3398)(cid:1831)(cid:2868) When (cid:1848) is small (i.e. the "weak coupling" regime), the effect of the moiré potential can be dictates that states at (cid:2173)(cid:3404)(cid:2184)(cid:3037) will be mixed with the (cid:2173)(cid:3404)0 state in the following way: Φ(cid:4666)0(cid:4667)(cid:2868)~ (cid:1848)(cid:3037)4(cid:1831)(cid:3040) Φ(cid:4666)0(cid:4667)(cid:2868), where Φ(cid:4666)(cid:2173)(cid:4667) is the exciton center-of-mass wavefunction at momentum (cid:2173) and (cid:1831)(cid:3040)(cid:3404)(cid:1328)(cid:2870)(cid:1854)(cid:3037)(cid:2870)/(cid:4666)8(cid:1839)(cid:4667) is the exciton kinetic energy within the first mini-Brillouin zone. (cid:1831)(cid:3040) ~ 8 meV in a WSe2/WS2 moiré superlattice with periodicity of ~ 8 nm. Because the wavefunction of excitons at (cid:2173)(cid:3404)(cid:2184)(cid:3037) longer completely forbidden and has oscillator strength ~(cid:3627)(cid:1848)(cid:3037)/(cid:4666)4(cid:1831)(cid:3040)(cid:4667)(cid:3627)(cid:2870)(cid:3404)(cid:4670)(cid:1848)/(cid:4666)4(cid:1831)(cid:3040)(cid:4667)(cid:4671)(cid:2870), giving an additional absorption peak at 4(cid:1831)(cid:3040)(cid:3404)30 meV above the main peak. The magnitude of the moiré Figure S4a shows the simulated exciton absorption spectra with phase (cid:2032)(cid:3404)15° and different larger (cid:1848). The quantitative dependence deviates from the square scaling law at large (cid:1848), which is be (cid:1848)~25 meV. expected since the perturbation treatment fails in the "strong-coupling" regime. By comparing the experimental results to the simulation, we can extract the magnitude of the moiré potential to potential can therefore be obtained by examining the amplitude of this side peak in the absorption spectrum. now contains part of the bright exciton wavefunction, their transition from the ground state is no magnitudes of the moiré potential. The side peak amplitude shows a monotonic increase with a 0.05 0.00 -0.05 ) . u . a ( e c n a b r o s b A V = 1 meV V = 5meV V = 10meV V = 15meV V = 25meV V = 40meV b ) . . u a ( e c n a b r o s b A 0.05 0.00 -0.05 -0.10  30 75 0 45 90 15 60 -0.10 -100 -50 0 50 100 -100 -50 0 50 100 meV meV are vertically shifted for visual clarity. Fig. S4. a, Simulated exciton absorption spectra with different moiré absorption spectra in the "weak coupling" regime but will affect the spectra in the "strong coupling" regime, where higher order mixing between exciton states become important and different mixing paths start to interfere, as shown in Fig. S4b. The complicated dependence of potential magnitude (cid:1848). (cid:2032)(cid:3404)15°. b, Simulated exciton absorption spectra with different moiré potential phase (cid:2032) and (cid:1848)(cid:3404)25 meV. Different curves The phase (cid:2032) of the moiré potential plays a more subtle role: It has negligible effect on the the spectra on (cid:2032) makes it difficult to have an accurate determination through comparison to the experiment. However, different (cid:2032) will not qualitative change the properties of the system, e.g. Figure S5 shows the real-space distribution of the moiré potential using the parameters (cid:1848)(cid:3404) 25 meV and (cid:2032)(cid:3404)15°. After summing up the six components, the peak-to-peak amplitude of the moiré potential reaches ~ 250 meV, which is much larger than (cid:1831)(cid:3040). As a result, the lowest between "weak coupling" regime and "strong coupling" regime. 5. Spatially localized exciton center-of-mass wavefunction energy excitons are trapped around the potential minimum point (labelled as ), as discussed in the text. We note that the exciton wavefunction discussed here refers to the center-of-mass envelop function for 1s exciton state, which is spatially homogeneous without the moiré superlattice. This should not be confused with the relative motion between the electron and the hole within an exciton that defines atom-like levels, e.g. 1s and 2p states. Fig. S5. Real space distribution of the moiré potential with (cid:1848)(cid:3404)25 meV and (cid:2032)(cid:3404)15°. The potential minimum is labelled as . References: 32 Wang, F. et al. Gate-variable optical transitions in graphene. Science 320, 206-209, (2008).
1302.5379	1	1302	2013-02-21T19:23:18	Electronic transport on carbon nanotube networks: a multiscale computational approach	[ "cond-mat.mes-hall" ]	Carbon nanotube networks are one of the candidate materials to function as malleable, transparent, conducting films, with the technologically promising application of being used as flexible electronic displays. Nanotubes disorderly distributed in a film offers many possible paths for charge carriers to travel across the entire system, but the theoretical description of how this charge transport occurs is rather challenging for involving a combination of intrinsic nanotube properties with network morphology aspects. Here we attempt to describe the transport properties of such films in two different length scales. Firstly, from a purely macroscopic point of view we carry out a geometrical analysis that shows how the network connectivity depends on the nanotube concentration and on their respective aspect ratio. Once this is done, we are able to calculate the resistivity of a heavily disordered networked film. Comparison with experiment offers us a way to infer about the junction resistance between neighbouring nanotubes. Furthermore, in order to guide the frantic search for high-conductivity films of nanotube networks, we turn to the microscopic scale where we have developed a computationally efficient way for calculating the ballistic transport across these networks. While the ballistic transport is probably not capable of describing the observed transport properties of these films, it is undoubtedly useful in establishing an upper value for their conductivity. This can serve as a guideline in how much room there is for improving the conductivity of such networks.	cond-mat.mes-hall	cond-mat	Electronic transport on carbon nanotube networks: a multiscale computational approach Luiz F. C. Pereira 1, M. S. Ferreira 2,∗ School of Physics, Trinity College Dublin, Dublin 2, Ireland. 3 1 0 2 b e F 1 2 ] l l a h - s e m . t a m - d n o c [ 1 v 9 7 3 5 . 2 0 3 1 : v i X r a Abstract Carbon nanotube networks are one of the candidate materials to function as malleable, transparent, conducting films, with the technologically promising application of being used as flexible electronic displays. Nanotubes disorderly dis- tributed in a film offers many possible paths for charge carriers to travel across the entire system, but the theoretical description of how this charge transport occurs is rather challenging for involving a combination of intrinsic nanotube properties with network morphology aspects. Here we attempt to describe the transport properties of such films in two different length scales. Firstly, from a purely macroscopic point of view we carry out a geometrical analysis that shows how the network connectivity depends on the nanotube concentration and on their respective aspect ratio. Once this is done, we are able to calculate the resistivity of a heavily disordered networked film. Comparison with experiment offers us a way to infer about the junction resistance between neighbouring nanotubes. Furthermore, in order to guide the frantic search for high-conductivity films of nanotube networks, we turn to the microscopic scale where we have developed a computationally efficient way for calculating the ballistic transport across these networks. While the bal- listic transport is probably not capable of describing the observed transport properties of these films, it is undoubtedly useful in establishing an upper value for their conductivity. This can serve as a guideline in how much room there is for improving the conductivity of such networks. Keywords: carbon nanotube networks, Green functions, electronic transport 1. Introduction Carbon nanotubes present remarkable physical proper- ties and since their discovery, have shown much promise in terms of applications [1]. Nanotube network films are be- ing heavily investigated as the basic building material for a new generation of flexible electronic devices. In particular, thin metallic films are currently the most promising mate- rial to enable the construction of large scale flexible elec- tronic displays [2 -- 4]. Carbon nanotube (CNT) films are produced by deposition of solutions containing dispersed CNT bundles. Figure 1 shows a typical nanotube network film produced by this technique. In general, it is possi- ble to control the density of the films by controlling the amount of solution deposited. The fabrication of novel electronic displays requires flexible, transparent, highly conductive films and nanotube networks seem to possess all the right ingredients. Flexi- bility is a basic property of CNTs. Transparency can be achieved by controlling the density of nanotubes in the net- work. Finally, while individual CNTs are known to have remarkable transport properties, the electronic transport ∗Corresponding author [email protected] [email protected] in network films takes place through individual nanotubes interconnected by junctions. In this case, electrons experi- ence low resistance when travelling along a nanotube, but are faced with significant resistance at the junctions. Fur- thermore, transparency and conductivity are intimately connected in thin metallic films. A very dense film might show high electrical conductivity but low transparency. On the other hand, a sparse network may show high trans- parency but poor conductivity. It is essential to search for optimum ways of increasing the conductivity but main- taining a good level of optical transparency. To achieve that, it is necessary to obtain a clear understanding of how the electronic transport occurs in such networked films, a task in which computer models are extremely useful. Modelling complex structures such as carbon nanotube network films is a huge challenge. A successful theoretical model needs to account for the topology and the disordered nature of the arrays formed by individual nanotubes (and bundles). It is also necessary to account for the electronic coupling between individual components throughout the films, since electronic transport takes place through a tun- nelling network spanning the whole film. Finally, it is nec- essary to account for the electronic structure of individual carbon nanotubes. The problem presents two clear length scales. One which accounts for the geometrical aspects of nanotube Preprint submitted to Nano Communication Networks June 23, 2018 percolation theory predictions. Moving on to the micro- scopic description of the problem, it will be shown how Green function methods can be used to calculate electronic transport on disordered nanotube networks. Finally, by combining both length scales in a single formalism, we show how the resulting multiscale model can be applied to calculate the maximum conductance of network films. 2. Random Networks of Finite Rods The complex structure presented by carbon nanotube network films can be modelled by standard geometric per- colation models. Single nanotubes and small diameter bundles are modelled by rigid rods of specific length and diameter. The structure of the film itself can be repre- sented by a 3D box which encloses the disordered array of interconnected rods. It is important to define a box with appropriate dimensions, such that most rods are located inside the box. It is also important to reproduce other aspects present on the real experiments, such as the place- ment of metallic electrodes on the sides of the films for electrical resistance measurements. Once the aspect ratio of the rods, defined as length- to-diameter ratio, and the dimensions of the enclosing box are defined, it is also necessary to specify the total number of rods inside the box, i.e. the number of nanotubes in the film. The total number of rods considered along with the dimensions of the box determine the density of the film. The density of a specific CNT film can also be expressed in terms of its volumetric fraction occupied by nanotubes. The volume fraction of a typical laboratory produced nan- otube film is related to the volume of CNT solution used for deposition [5, 6], and can be directly measured. In a computer simulation, the volume fraction depends on the total number of rods distributed inside the box, and is calculated directly for each configuration generated. Each rod is randomly placed and randomly oriented inside the box. The arbitrary orientation direction is cho- sen in a 3D space with no constraints, and individual rods are allowed to overlap with other rods. An arbitrary rod might also have part of its volume located outside the box, and this portion is not included when calculating the vol- ume fraction of the film. Each ensemble of randomly dis- tributed/oriented rigid rods gives rise to a complex random network of interconnected 1-dimensional elements, capable of capturing the most important morphological features found on a nanotube network. Figure 2 shows a typical configuration generated with N = 300 identical rigid rods of length-to-diameter ratio equal to 20. It is possible to identify interconnections between rods throughout the network by considering the centre-to-centre distance between each pair of rods. If the smallest centre- to-centre distance between a given pair is smaller than the sum of their radii, then this pair is said to be connected. By calculating the distance between each possible pair of rods in the network, all the connections are mapped and a list of neighbours for each rod is constructed. In general, Figure 1: Scanning electron micrograph of a typical carbon nanotube film produced by deposition of a solution containing dispersed nan- otube bundles. Image is courtesy of J. N. Coleman's research group, Trinity College Dublin. network films, as well as the complex topology of these structures. We shall refer to this as the macroscopic scale. The other length scale considers that carbon nanotubes are made of individual atoms arranged in the usual hexag- onal structure. This shall be referred to as the micro- scopic scale. Once the relevant length scales have been clearly identified, it is possible to develop a multiscale model which accounts for both with the appropriate ac- curacy. The geometrical aspects of individual carbon nanotubes can be properly captured by a simple model where these are represented by rods of specified length and diameter. Choosing a random position and orientation for each rod, within a containing volume, a network of interconnected el- ements is generated, which mimics the topology of the net- works in CNT films. Naturally, percolation theory meth- ods can be used to investigate the transport character- istics of these networks. On the microscopic scale, the electronic structure of a carbon nanotube can be satisfac- torily described by simple model Hamiltonians. The tight binding approximation, considering only one orbital per carbon atom, is enough to provide a description of the relevant electronic states. Furthermore, standard Green function methods can be used to describe the electronic structure and electronic transport characteristics across the nanotube networks. From a theoretical viewpoint, the description of the transport properties of such networks requires a multi- disciplinary effort involving classical electromagnetism, quan- In this work we attempt to tum physics and statistics. use tools of all these multidisciplinary fields of study to present a multiscale approach that can describe the elec- tronic transport on carbon nanotube films. It begins with a description of the model utilised to account for the com- plex topological network formed by CNTs. It will be shown how the connectivity of the network depends on the den- sity of the network as well as the dimensions of individual nanotubes. Considering a limit where the network resis- tance is exclusively due to nanotube junctions, it will be shown how the proposed model compares with standard 2 considered when calculating the volume fraction. Rods are allowed to overlap with other rods, which might lead to an overestimation of the volume fraction, especially in the case of very dense networks. However, this possible overestimation of the volume fraction does not affect the morphology of the resulting networks. Electrodes are located on the 2ℓ × 2ℓ faces of the box, and rods crossing one of those faces are connected to that electrode. The box is defined with a separation large enough to avoid an individual rod being connected to both electrodes, in order to avoid short circuits. For a fixed number of rods, and a box of fixed dimensions, a very large number of configurations can be generated by ran- domly distributing the rods. Therefore, a large number of independent configurations must be considered, and a careful statistical analysis of the results must be carried out. All the results reported in the next section are ob- tained by averaging a large number of independent reali- sations, without discarding any configuration. In order to determine which pairs of rods are intercon- nected it is necessary to calculate the minimum distance between each rod and all the rods around it. Identifying which rods are in contact with the electrodes also requires testing the criteria for all rods close to the respective faces. Both procedures can be very time consuming from the computational point of view, and the computation time grows very rapidly with the total number of rods consid- ered. In all simulations we have considered networks with a maximum of 1800 individual rods. 2.2. Results Let us begin by analysing the relation between the to- tal number of rods and the volume fraction of the films simulated. As previously mentioned, experimental studies have found a direct relationship between the volume of so- lution deposited and the volume fraction of the resulting network [5]. Figure 3 shows the relationship between the volume fraction and the total number of rods randomly distributed in the film. The data is shown for different aspect ratios, which is controlled by varying the diame- ter of the rods. Each data point is an average over sev- eral independent configurations and error bars are smaller than symbol sizes. Solid lines shown in the graph are lin- ear least-square fits to each data set, and confirm a direct proportionality between the volume fraction and the total number of rods in the network. The existence of such a direct relation between the number of rods and the volume fraction provides a direct way of comparing and combin- ing simulation results with experimental measurements. On a real CNT film there are several thousand individual nanotubes, whereas computational resources limit simula- tions to only a few thousand rods. However, morphologi- cal properties such as connectivity and density of junctions are directly related to the volume fraction, which connects experiment and simulation at least on a qualitative level. Now we shall proceed to analyse how the connectivity of the network depends on the concentration of rods. For Figure 2: Illustrative representation of rigid rods randomly placed inside a rectangular box. There is a total of N = 300 identical rods inside the box, each with aspect ratio ℓ/D = 20. each rod has an arbitrary number of neighbours, but the number of connections (neighbours) per rod follows some probability distribution with a well defined mean. The average number of connections per node is defined as the connectivity of the network. Networks with a low den- sity of rods, i.e. a small volume fraction, are expected to have low connectivities, while networks with a large vol- ume fraction will have high connectivities. In other words, the density of junctions in a network film depends directly on its volume fraction. In order to investigate electronic transport across ran- dom nanotube networks, it is necessary to include the placement of metallic electrodes in model films. Contacts to electrodes are crucial to electronic transport studies, and must be addressed with caution. In experimental re- alisations, metallic electrodes are placed at opposing ends of a film and the resistance is measured between these electrodes in a usual two probe method. In this work elec- trodes are modelled by two opposing faces of the 3D box. A rod crossing one of the specified electrode faces is con- sidered to be in contact with that particular electrode. In this manner, a map of all the rods connected to the elec- trodes is constructed. 2.1. Simulation Details Individual carbon nanotubes are represented by rigid rods of length ℓ and diameter D, with a well defined aspect ratio given by ℓ/D. The film is mimicked by a rectangu- lar cuboid, whose dimensions are defined in terms of the rod length ℓ. In all the results reported here the box has dimensions given by 2ℓ × 2ℓ × 4ℓ. For each rod, a random position for one of its ends is generated inside the box, and the position of the other end is determined by the orien- tation of the rod. The total number of rods considered in each film is N , and along with ℓ and D these are the parameters of the simulations. The orientation of the rods is chosen at random from all the possible directions in a 3D space. A particular rod can have one of its endings located outside the box, and the portion of the rod located outside of the film is not 3 0.4 0.3 0.2 0.1 n o i t c a r F e m u l o V 0 0 Aspect Ratio 10 12.5 20 25 40 600 Number of Rods 1200 1800 Figure 3: Dependence of the volume fraction on the total number of rods. The lines shown in the graph are linear least-square fits to each data set, and error bars are smaller than symbol sizes. 8 6 Aspect Ratio 10 12.5 20 25 40 4 > α < 2 0 0 > α < 10 5 0 0.1 0.2 Volume Fraction 0.3 0.4 Figure 4: Average network connectivity as a function of volume frac- tion. Inset: Universal scaling of the connectivity with Vf × ℓ/D. Error bars are smaller than symbol sizes. a film of fixed dimensions and containing rods of identical aspect ratio, it is natural to expect that the average con- nectivity will increase with the volume fraction. If there are more rods inside a fixed finite volume, more rods will be in contact. On the main panel of figure 4 it is shown how the average connectivity per rod hαi, increases with the volume fraction Vf , for each aspect ratio considered. From the data it is also clear that for a fixed volume frac- tion, networks composed of high aspect ratio rods have a higher connectivity when compared to lower aspect ratio rods. This is an interesting feature, which is in accordance with experimental observations that films made of high aspect ratio bundles present a lower overall resistivity [6]. Based on a rod percolation model similar to the one considered here, a purely geometric argument has been used to propose that the mean number of junctions per 0 15 2 Vf x AR 4 6 nanotube (or bundle) is expected to behave as [7] 0.15 0.1 0.05 0 0.15 0.1 0.05 0 0.15 0.1 0.05 ) α ( P ) α ( P ) α ( P Vf = 17.9% Vf = 26.8% Vf = 35.9% 0 0 4 8 12 α 16 20 Figure 5: Connectivity probability distribution for random networks of finite rods with aspect ratio ℓ/D = 10, and indicated volume frac- tion. Vertical solid lines indicate the calculated average connectivity. Each data point is an average over 100 independent configurations. hαi = β × Vf hℓi hDi , (1) where β is a constant independent of the nanotube (or bun- dle) diameter and length. Also, Vf is the volume fraction, hℓi is the average bundle length, and hDi is the average bundle diameter. In order to verify if this hypothesis is satisfied by the present model, the average connectivity is plotted as a function of the scaled variable Vf × ℓ/D, in the inset of figure 4. Remarkably, the results are in good agreement with the predicted relation, for all the aspect ratios considered. Besides providing a direct verification of equation (1), from the linear least-square fit on the in- set of figure 4 the value of the constant β = 2.14 ± 0.03 is calculated, which is independent of the other factors in the equation. It is also instructive to analyse the probability distri- bution of connections per rod in the network. Figure 5 presents the distribution for the probability that an arbi- trary rod will have α connections. The data is obtained by considering random configurations with rods of aspect ratio ℓ/D = 10. The vertical solid lines indicate the cal- culated average connectivity hαi for each volume fraction specified. From the data it can be clearly seen that the average number of connections per rod increases as the concentration of the network is increased, while the prob- ability distribution becomes broader. The broadening of the distribution indicates that as the density of the net- works increases, they actually become less homogeneous. Another important factor extracted from the computer model considered is the number of rods connected to the electrodes. In figure 6 (main panel) it is shown that the 4 > E α < 80 60 40 20 0 0 0.1 0.2 Vf x (AR)2 40 60 20 > E α < 90 60 30 10 12.5 20 25 0.3 0.4 Volume Fraction 0.5 0.6 Figure 6: Number of rods connected to electrodes as a function of the volume fraction. Inset: Universal scaling of the electrode con- nectivity with Vf × (ℓ/D)2. number of rods crossing the electrode faces increases with the volume fraction of the film. Similar to the average connectivity of the network, the actual increase of hαEi with Vf depends on the aspect ratio of the rods, albeit in this case the dependence is more pronounced. In order to be useful when compared to experimental data, it is desirable to find a universal scaling dependence for hαEi in terms of the parameters Vf , ℓ and D, similar to the one in equation (1). Following an empirical approach, it has been found that the relation D(cid:19)2 hαEi = βE × Vf(cid:18) ℓ , (2) provides an excellent collapse of the data on a single straight line, as shown on the inset of figure 6. Moreover, the pro- portionality constant is found to be βE = 1.35 ± 0.07. In practice, equations (1) and (2) provide numerical values for the connectivity and for the number of connec- tions to the electrodes in a model network film, for any set of parameters chosen. Furthermore, the expressions ob- tained for hαi and hαEi can be used in combination with experimental data in order to estimate physical proper- ties of real carbon nanotube films [8]. Finally, equations (1) and (2) can also be coupled with electronic transport models to provide calculations for the electrical conductiv- ity of network films. In the next section we present a methodology to cal- culate the resistance of a network composed of Ohmic re- sistors. This methodology is later coupled with the re- sults presented above, to provide calculations for the re- sistance of nanotube films from a purely classical point of view. Later on, the results presented above will be cou- pled with a quantum transport model, providing a more realistic microscopic picture of electronic transport across carbon nanotube films. 5 3. Resistive Networks In carbon nanotube films, the resistance of the nan- otubes itself is negligible when compared to the junction resistance. In this case, a nanotube network film resembles a networks of perfect conductors connected by Ohmic re- sistors. Calculating the resistance between arbitrary nodes on a lattice of resistors is a problem that has been inves- tigated before. Consider for example a resistive lattice where each vertex contains a resistance R0, i.e. between every two adjacent nodes there is an Ohmic resistor R0. Different methodologies have been developed in order to tackle this problem. Venezian [9] introduced a method to calculate the re- sistance between two adjacent nodes on an infinite square grid of equal resistors based on a superposition of po- tentials and currents of two one-terminal configurations. Atkinson and van Steenwijk [10] generalised the original method to 3D and higher-dimensions. Cserti and collabo- rators developed a methodology based on the application of lattice Green functions [11, 12], which provides values for the equivalent resistance in infinite networks. Build- ing on the method developed by Cserti, it is possible to propose yet another approach specifically aimed at deal- ing with finite networks of resistors. The methodology consists on the direct application of Ohm's and Kirchoff's circuit laws to finite resistive networks. When a current is injected in an arbitrary node and extracted from another arbitrary site, the expression for the equivalent resistance is given in terms of the potential difference between these sites. Consider a resistive network where between every two adjacent sites there is a resistor R0. The current that can be extracted from a site denoted by A is given by the net current from its nearest neighbours, which can be written as IA = 1 R0 [(VA − VA+1) + (VA − VA−1) + · · ·] , (3) where VA is the electrical potential on site A, and A ± 1 represent two of the nearest neighbours of site A. It is possible to rewrite equation (3) as IA = 1 R0 "αAVA − VA+δ# , αA Xδ (4) where δ represents the nearest neighbours of site A, and αA its total number of neighbours. In matrix notation, it can be written in a compact form as I = 1 R0 MV. (5) Only two elements of I are non-zero, the ones correspond- ing to the injection and extraction of current, being one positive and one negative. The matrix denoted by M has the general form M =  0 0 · · · α1 −1 0 −1 α2 −1 · · · 0 −1 α3 −1 · · · ... . . . ... ... ... , (6)   where αA is again the number of nearest neighbours of site A, and MA,B = −1 if A and B are neighbours or zero otherwise. Notice that for a true random lattice, each site has an arbitrary number of neighbours. At this point, the matrix equation (5) is re-arranged to read V = R0M−1I. (7) By performing the calculation above, we obtain the value of the electrostatic potential in each and every node of the network. Moreover, from V one calculates the poten- tial difference across every resistor in the network, which provides a direct way of calculating the equivalent resis- tance between any two arbitrary points of the network. Furthermore, by constructing the appropriate matrix M, it is possible to calculate the resistance between arbitrary nodes of any resistive structure, all that is required is to describe the matrix M according to the specific network connectivity. The resistance between two arbitrary sites A and B is given by RA,B = VA − VB I . (8) Which can finally be expressed in terms of specific ele- ments of the inverse of M as RA,B = R0( [M −1]A,A + [M −1]B,B − [M −1]A,B − [M −1]B,A ). (9) Therefore, the resistance between two arbitrary points in any type of lattice can be calculated as long as we know the number of connections, and which sites are interconnected. It is not hard to generalise this result for the case where the resistance between nearest neighbours is not identical, or even for a distribution of different resistances. In order to do so, the 1 term in Eq. (3) would have to be replaced by R0 the specific resistance connecting the corresponding sites, and could be absorbed into the definition of M. 3.1. Application The method described above can be directly applied to model carbon nanotube films where the resistance be- tween the two electrodes is calculated through equation (9). Generating a random configuration of rods and cal- culating the connectivities of the arrangement, the matrix M is written and the resistance between the electrodes is calculated. It is important to recall this approach is valid when the resistance along the nanotubes is very small, and the resistance of the film is dominated by the junctions. Figure 7: Schematic representation of the model used to calculate the resistance of carbon nanotube networks. Individual CNTs and bundles are represented by rods of finite length. Ohmic resistors R0 are place at each junction. Total resistance across the film is calculated as the equivalent resistance between electrodes (shaded lateral rectangles). A random array of nanotubes (represented by finite- length rigid rods) is generated inside a containing box of specified dimensions, and the volume fraction is calculated. The junctions between rods in close proximity are mapped and a list of neighbours of each rod is constructed. Sim- ilarly, rods that cross the electrode faces are identified, and that specific electrode is added to their list of neigh- bours. The inter-tube junction resistances experienced by conduction electrons, are represented by Ohmic resistors (R0), placed between every two rods in contact. Finally, the total resistance between the electrodes is calculated, and for the purpose of analysis, it is converted into con- ductance by taking its inverse. Figure 7 illustrates in a schematic way the set up used in our calculations. As previously mentioned, nanotube films are commonly produced by deposition of solutions containing dispersed CNTs. The disordered network has a density (and volume fraction) which is directly proportional to the volume of solution used. A small volume of deposited solution will generate networks with a low density of connections and few percolating paths for the charge to move from one elec- trode to the other. In this case, films with high resistance will be produced. On the other hand, if the volume of solution used is large, it is expected that the films pro- duced will have a low overall resistance, since there will be several percolating paths for the conduction electrons to move across the network. It is more convenient to express the charge transport results in terms of the inverse resistance, i.e. conductance, defined simply as Γ = 1/R. Percolation theory predicts that for small values of Vf , the conductance behaves ap- proximately as Γ ∼ (Vf − Vc)t, (10) where Vc is the critical value of the volume fraction. The critical exponent t, depends only on the dimensionality of the embedding space. In our model the rods are randomly oriented in a 3D space, and for this case, the theory pre- dicts t = 1.94 [13]. Beyond the percolation threshold, i.e. 6 8 6 ) 1 - 0 R ( Γ 4 2 0 0 Γ ~ (Vf - Vc)1.94 AR=10, Vc=4.66% AR=20, Vc=2.64% AR=40, Vc=1.41% Aspect Ratio 10 20 40 10 ) 1 - 1 0 R ( Γ 0.1 0.1 0.2 Vf 0.3 0.4 0.01 1000 ) 1 - 0 R ( Γ 20 15 10 5 0 0 10000 < α > N 20000 < α > N 10000 Figure 8: Network conductance as a function of the volume fraction near the percolation region. Solid lines are a fit of equation (10) with the appropriate 3D exponent t = 1.94. Figure 9: Film conductance increases linearly with density of con- nections for large enough densities. Log-log plot shows convergence to linear regime. for networks with a volume fraction considerably larger than the critical value Vc, the conductance is found to in- crease linearly with Vf [14, 15]. The behaviour of Γ in the percolation region can be analysed in detail in order to verify the agreement with the standard theory. In figure 8, we fit equation (10) with the expected scaling exponent t = 1.94, to the data near the percolation region. The agreement with theory is re- markable, and from the fit we also extract the value of the critical volume fraction for each aspect ratio considered. The calculated values of Vc are inversely proportional to ℓ/D, which is an intuitive result and is in agreement with previous theoretical predictions [14, 16, 17]. In section 2 it has been shown how the connectivity of a random network of rods scales linearly with the product of the volume fraction with the aspect ratio of individual rods. It is natural to expect the conductivity of the net- work to increase with its connectivity, which is indirectly demonstrated by the dependence of the conductance with the volume fraction (see figure 8). Moreover, it is also pos- sible to analyse the dependence of the film conductance with the density of junctions, which is defined as the total number of junctions per unit volume, and can be written in terms of the network connectivity as NJ = hαiN 2 1 V , (11) where N is the total number of rods inside the box of volume V . Figure 9 shows a plot of the network conductance as a function of hαiN = 2NJV on a log-log scale, and on a linear scale on the inset, for three different aspect ratios. Once again our calculations are limited by the computa- tional resources required for simulations with the rods of large aspect ratio. However, it is possible to infer from the log-log plot that, at least for high enough concentrations of junctions, the conductance seem to converge to the lin- ear dependence represented by the dashed straight line. Furthermore, in the inset we show the excellent collapse of all the data to the same straight line in a linear scale Γ/R−1 0 = m × 2NJ V . In combination with experimental measurements, the model described above has been used to estimate the inter-tube junction resistance [8]. The result obtained is in broad agreement with independent exper- imental results [18, 19]. Nonetheless, it is important to point out that in real CNT films the junction resistance can vary widely, depending on the chirality of the nan- otubes involved as well as the diameter of the bundles, and an exact determination of these contact resistances is still an open problem. 4. Microscopic Modelling The approach presented in the previous sections anal- yses the system in a macroscopic scale. Individual CNTs were modelled as perfectly conducting rigid rods of finite aspect ratio, and tunnel barriers at nanotube junctions were mimicked by Ohmic resistors. The film resistance was calculated as the equivalent resistance between op- posing electrodes separated by a disordered network of identical resistors. As previously mentioned, the method provides results in reasonably good agreement with exper- iments and previous percolation model studies. It is also possible to approach electronic transport prob- lems from an atomistic viewpoint, most notably by means of ab-initio density functional theory calculations (DFT) coupled with non-equilibrium Green function methods [20, 21]. However, the computational cost of these calculations can be considerably high. Moreover, all current DFT im- plementations require at least some degree of periodicity in order to handle large systems (≈ 103 atoms), which unfortunately is clearly not present in disordered CNT networks. Nonetheless, ab-initio methods have been used to investigate electrical resistance at single wall nanotube (SWNT) junctions [22], and more recent studies have ex- tended these results considering the presence of O2 and N2 7 molecules [23] near the junction. The use of first principles methods for obtaining the quantum conductance between individual nanotubes is able to shed some light on the quality of the junction, but it is unable to reproduce the general observed features in network films because it is not capable of including disorder effects to the scale required. The alternative of carrying out a fully atomistic transport calculation within a heavily disordered environment is very desirable but it is currently too computationally demand- ing. It is therefore necessary to compromise if one wishes to combine the two features. One possible approach consists of using computation- ally inexpensive semi-empirical methods to describe the electronic structure of carbon nanotubes. Tight binding model Hamiltonians are a very efficient and convenient way to describe electronic structure of materials in gen- eral, and are well suited to deal with carbon nanotubes. By describing the electronic structure of individual nan- otubes with a simplistic model Hamiltonian, we can afford to account for disorder effects by considering hundreds, if not thousands, of CNTs, and several possible configura- tions for the structure of the random tunnelling networks formed in the films. Furthermore, this approach is par- ticularly well suited to treat electronic transport across networks on a purely ballistic regime. By neglecting all sources of decoherence-inducing scattering, it is possible to calculate the conductance of ideal network films. The conductance calculated in such idealised scenario repre- sents an upper bound for the transport performance of real films, and can be used as an indication of how much room there is for improving the conductivity of these materials. 4.1. Tight Binding Hamiltonian and Green Functions A general semi-empirical tight binding Hamiltonian can be written as Hν =Xi ν, iiǫ0hν, i +Xhi,ji ν, iiγ0hν, j, (12) where ν, ii represents an electron on atom i of a wire la- belled by ν, ǫ0 is the on-site energy, γ0 is the hopping inte- gral parameter, and the sum hi, ji is over nearest-neighbouring atoms. This general form can describe several types of nanostructures, depending on the choice of atomic struc- ture and on the orbital degrees of freedom represented by the states ii. Associated with the Hamiltonian Hν , there are the retarded(+) and advanced(−) single-particle Green functions, which are defined as G± = [E ± ıη − Hν ]−1, (13) where η is a small positive imaginary part added to the electronic energy E in order to avoid singularities at eigen- values of the Hamiltonian. Electronic transport calculations typically require the presence of particle reservoirs, which represent source and drain for the conduction electrons. In the case of transport 8 simulations of carbon nanotube networks, semi-infinite nan- otubes are employed to mimic the presence of particle reservoirs. These semi-infinite CNTs are solely described by the GF elements relative to the atoms on its final unit cell. In order to calculate surface GF elements of semi- infinite nanotubes, we apply efficient recursion methods, which are numerically stable and computationally inex- pensive. 4.1.1. Network Green Functions Following the general method applied on electronic trans- port calculations, the system is separated into three parts: two leads and a central region [24, 25]. The major differ- ence between our particular approach and the usual one is that each one of the three parts of the system are them- selves composed of several wires. The leads are represented by a number of semi-infinite nanotubes, which act as parti- cle reservoirs. Meanwhile, the central region of the system consists of the disordered network itself, represented by an ensemble of finite-sized CNTs. We begin by constructing the Green function matrices of a completely disconnected system. The inter-tube junctions are introduced in the form of electronic hopping terms between CNTs in the network, and between the network and the electrodes. Fi- nally, the GF of a fully connected network of nanotubes is calculated with the Dyson equation. Beginning from a disconnected tripartite system, one can write the complete Green function as a super matrix of the form g± =  g± L 0 0 0 g± C 0 0 0 g± R   , (14) where g± L,R are the GF matrix of the left and right elec- trodes, and g± C is the GF of the central scattering re- gion. At this stage, the network is an ensemble of non- interacting nanotubes. In this initial description of the disconnected system, each of the matrices in g± are in block diagonal form, with one block for each individual CNT. The electrodes act as charge reservoirs which are represented by semi-infinite nanotubes. In other words, whenever a CNT is contacted to either electrode we will represent it by a semi-infinite object. Therefore, there are hαEi diagonal blocks given by surface GF of semi-infinite carbon nanotubes. In general, the electrode GFs can be written as g± L,R(E) =  S±(E) 0 0 0 ... S±(E) 0 ... 0 0 S±(E) ... · · · · · · · · · . . . , (15)   where S±(E) are advanced (or retarded) Green functions of the surface elements of semi-infinite wires, which mimic the particle reservoirs. Similarly, the GF of the central scattering region con- sisting of a disconnected network with N wnanotubes is represented by a block diagonal matrix with N blocks, and it is written as 3. Make hν, jVCC ν′, j′i = γ′ ; 4. Repeat until total number of connections equals hαiN/2. g± C (E) =  G±(E) 0 0 ... 0 G±(E) 0 0 0 ... G±(E) ... · · · · · · · · · . . . , (16)   where we recall that G±(E) is the Green function of a finite-sized pristine monatomic linear chain or carbon nan- otube. In general, each diagonal block G±(E) is a matrix whose dimension equals the total number of atoms on the individual wires. Individual nanotubes are connected by introducing a perturbing potential bridging any two CNTs that should be in contact. This inter-tube coupling appears in the form of an electronic hopping term γ′ in the Hamiltonian of the complete system, and is given by Vν,ν ′ = ν, jiγ′hν′, j′, (17) where ν and ν′ label the connecting wires, and j and j′ identify the atoms in each one. Notice that ν and ν′ must be different while j and j′ are allowed to have the same value. These individual hopping terms can be grouped to- gether in the form of a general perturbation potential writ- ten as V =  0 VLC 0 V † LC VCC VCR 0 V † CR 0   , where once again each term represents a matrix itself. On the construction of the perturbing potential we as- sume that semi-infinite CNTs in the same electrode do not interact with each other. Furthermore, we assume that the central scattering region is large enough, such that there can be no direct hopping terms between semi-infinite wires located on different electrodes. The first assumption can be easily relaxed, and it should not have any significant effects on the results. The second assumption, however, should not be relaxed. The physical meaning of the sec- ond constraint, i.e., [VLR] = [VRL] = [0], is to avoid short circuits on the system, which could manifest as spuriously high values for the conductance, and significantly affect the results. The term VCC is by far the most significant one, and also the one that relies at the heart of our methodology. In our model, [VCC ] is a matrix constructed from a random network, generated from an ensemble of nodes with a given average number of connections per node. Equivalently, a random graph can also be defined by its total number of sites N , and the total number of connections on the graph hαiN/2. The process to construct VCC is as follows: 1. Select two unconnected wires at random: ν, ν′ ; 2. Select one random atom in each wire to carry on the interaction: j, j′ ; 9 (18) 4.2. Kubo Formula Applied to Networks A similar procedure is also applied to construct [VLC ] and [VCR], where the interaction of the electrodes with the finite-sized nanotubes is chosen by a randomised process. The inter-tube hopping term, represented as γ′ is defined in analogy with the usual tight binding hopping integral parameter γ0, and roughly represents the coupling between individual atoms in interconnected CNTs. Remember that Equations (15) and (16) describe the GF of the completely disconnected network. The Green function of the interconnected disordered systems can eas- ily be calculated with Dyson's equation, and can be written in matrix form as g±, G± = g± + g±V (cid:0)1 − g±V(cid:1)−1  G± =  .  G± L G± CL G± G± G± 0 0 C G± RC G± CR LC R (19) (20) Equation 20, is the Green function of the whole intercon- nected network, including left- and right-hand side elec- trodes, central scattering region, and propagators between these three regions. Therefore, G± provides all necessary information with respect to the electronic structure of the disordered network, and can be used to calculate the elec- tronic conductance of the system with the Kubo formula. The zero-bias conductance of the system is calculated with the Kubo formula, which provides a simple expression for the conductance by calculating the net electronic cur- rent across a reference plane in the system, often referred to as the cleavage plane, which is located between any two parts of the system. The current is expressed in terms of a few Green function matrix elements of the system, and the calculation becomes simple enough. In analogy to the standard formalism first introduced by Caroli et al. [24], the system is divided in three parts. The location of the cleavage plane is arbitrary and can be chosen in the most convenient location for the case in question. Here it is placed between the central region and the right lead, as illustrated in Fig. 10. Translating Fig. 10 into words, regions L and R contain hαEi semi-infinite nanotubes, and region C contains a disordered array of N finite CNTs, forming a random network with an average number of connections per tube given by hαi. Having placed the cleavage plane between the regions C and R, the Kubo formula for the zero-bias conductance is written as [26, 27] Γ(EF ) = (cid:18) 4e2 h (cid:19) ℜe{Tr[ GC VCR GRV † CR − VCR GRC VCR GRC ]}, (21) Figure 10: Schematic representation of the system, divided in left and right leads (L,R), and a central scattering region (C). The cleavage plane is placed between C and R. where all the GFs are calculated at the Fermi energy EF , and G(E) are defined as (22) G(E) = 1 2ı(cid:2)G−(E) − G+(E)(cid:3) . 4.2.1. Reducing Green Function Matrices Even though it is possible to write an explicit expres- sion for the Green function of the interconnected network as in Equation (20), from which the matrix elements re- quired by the Kubo formula in Equation (21) can be ob- tained, it can still be an extremely challenging task. The difficulty arrises from the size of the matrices involved. Consider for example, a network with N = 100 wires, In this case, g± where each wire has L = 1000 atoms. C is a matrix of (N × L)2 = 1010 elements, and [VCC ] is a matrix of same dimension. Storing matrices of this size on the memory of computer clusters becomes a demanding computational task in itself, and inverting matrices of this dimension takes a considerable time, even if one applies advanced linear algebra routines. The solution to this limitation is facilitated by two gen- eral properties of the system. The matrix g± is block di- agonal and V is a sparse matrix. Therefore, the prod- uct g±V on Equation (19) is also sparse, or at least has many vanishing terms that will not affect G±, i.e., several of the matrix elements in G± do not depend on the per- turbation, whenever the corresponding element vanishes. In practical terms, the corresponding row/column in the matrix G± C will not involve any V -dependent terms and is simply given by G±. A connection between atom j of tube ν and atom j′ of tube ν′ will only involve the corre- sponding matrix elements of the g-matrix. Therefore, if a given CNT is connected to three neighbours at intra-tube sites i, j and k, the matrix elements G± i,j,, G± i,k, G± j,k (and the respective adjoint elements) are the only matrix elements appearing in the expression for the net- work GF of Eq.(19). This can be an enormous simplifica- tion when compared with the full diagonalisation method. Rather than inverting enormous matrices of dimension of N × L, this technique allows us to express the network GF in terms of inverse operations of much smaller matri- ces, and whose sizes are defined by the total number of connections in the system, i.e. matrices of order N × hαi. k,k, G± i,i, G± j,j, G± Therefore, because of the particular structure in Dyson's equation, it is possible to reduce the Green function ma- trices, such that only the elements affected by the pertur- bation V are stored. Consider, for example, a CNT with several thousand atoms. If only 20 of these atoms interact with neighbouring nanotubes, then only the elements cor- responding to these atoms and the propagators between them need to be stored, since only these will be affected by the perturbation. In this case we can perform calcu- lations up to a thousand wires, each with a few thousand atoms. 5. Multiscale Model of CNT Networks The results from the geometrical modelling presented in section 2 can now be combined with our quantum trans- port model, providing a multiscale approach to the study of electronic transport in carbon nanotube network films. In particular, by considering an idealised ballistic trans- port regime it is possible to evaluate the upper limit for the conductance of CNT network films. It has been ar- gued that the construction of transparent, flexible dis- plays requires thin films with an electrical conductivity σdc ≥ 6 × 105 S/m [28]. However, the production of CNT films with such high conductivities has been very difficult to achieve, and it is possible that nanotube networks are not the most suitable material to be used in the next gen- eration of flexible displays [29, 30]. In section 2 results for the connectivity of disordered networks of rods were presented. Universal relations for the dependence of the network connectivity with experi- mentally accessible quantities such as the volume fraction of the films and the aspect ratio of individual wires, were obtained. In particular, it was found that for a random network of rods with individual length ℓ and diameter D, inside a rectangular cuboid film of dimensions given by 2ℓ × 2ℓ × 4ℓ, the average connectivity per rod is given by hαi = (2.14 ± 0.03) × Vf ℓ D . (23) Furthermore, by considering that metallic electrodes are located on the opposing square faces of the film, and that any rod crossing these surfaces is taken to be in contact to that specific electrode, we found that the number of rods crossing each electrode can be written as D(cid:19)2 hαEi = (1.35 ± 0.07) × Vf(cid:18) ℓ . (24) Therefore, once the aspect ratio of the wires and the vol- ume fraction of the film are chosen, hαi and hαEi are auto- matically defined. These parameters are then fed into the general quantum transport model presented in the pre- vious section, which yields the conductance for a network with the specified geometrical properties. Finally, since we have considered disordered networks to be defined inside a film of specified geometry, it is then possible to convert 10 the network conductance into film conductivity. A result which can be directly compared with experimental mea- surements. Let us begin by investigating the conductance of net- works made of (5, 5) carbon nanotubes, which have a metal- lic band structure. The diameter of a (5, 5) CNT is ≈ 0.68 nm, while the length depends on the number of unit cells considered. In a (5, 5) SWNT, each unit cell has 4×5 = 20 atoms. We consider nanotubes of aspect ratio equal to 11, 16.5, 22, 27.4, which correspond to lengths of 7.5 nm, 11.2 nm, 14.9 nm, 18.6 nm, respectively. For each of these aspect ratios, it is possible to vary the total number of carbon nanotubes in the films, such that the network vol- ume fraction ranges from 0.065 up to 0.325. Moreover, through equations (23) and (24), the combinations of ℓ/D with Vf provide networks with hαi in the range 3.6 − 7.5, and hαEi = 27 − 77. In figure 11 it is shown how the network conductance depends on the volume fraction for the respective aspect ratios considered. Each data point is an average over at least 50 independent configurations, and the error bars represent the statistical error in the ensemble. Similarly to the resistive network results presented in section 3.1, there is a clear improvement in the conductance with a higher concentration of nanotubes. Moreover, it is also clear that higher network conductance values are achieved when the film is composed of nanotubes with larger aspect ratios. The dependence of the film conductance on the aspect ratio of individual nanotubes is in agreement with previous experimental studies [31, 32]. One possible way to account for this trend is to consider that, on average, a longer nanotube will have a larger distance between its junctions with other tubes. Therefore, on nanotubes with higher ℓ/D ratio the electrons will, on average, travel a longer distance along the tubes before experiencing heavy scattering caused by the inter-tube junctions. It is also possible to analyse how the conductance de- pends on the total number of junctions in a network film of volume V , which is defined as NJ V = hαiN/2. Figure 12 shows the averaged network conductance as a function of the total number of junctions in simulated films. The dashed lines are least-square fits to each data set, and show a linear dependence of the conductance with the number of junctions. We again see an increase in the overall film conductance as the aspect ratio of individual nanotubes is increased. Moreover, the model reproduces the approxi- mate linear trend observed in experimental reports [8]. It is possible to convert the calculated network conduc- tance into film conductivity by considering the geometry of the films. Employing the usual relation between con- ductance and conductivity, it is possible to write σ = Γ L A = Γ 4ℓ 4ℓ2 = Γ ℓ . (25) At this point, it is important to remember that Γ rep- resents the conductance of an idealised network of finite- sized carbon nanotubes, in a fully ballistic electronic trans- 11 ) h / 2 e 2 ( Γ 16 14 12 10 8 0 0.05 0.1 0.15 0.2 Volume Fraction Aspect Ratio 11 16.5 22 27.4 0.25 0.3 Figure 11: Conductance versus volume fraction for networks of (5, 5) armchair SWNTs. The legends indicate the aspect ratio of each data set. Higher aspect ratio CNTs provide better conductance, in accordance with experiments [31]. ) h / 2 e 2 ( Γ 16 14 12 10 8 0 (5,5) SWNTs Aspect Ratio 11 16.5 22 27.4 1000 2000 3000 4000 < α >N = 2NJV 5000 6000 Figure 12: Averaged film conductance as a function of total number of nanotube junctions. Dashed straight lines are least-square linear fits to simulation data, in accordance with experimental results of ref. [8]. port regime. In the calculation of the conductance, all decoherence-inducing factors (certainly present in experi- mental realisations) were neglected. The calculated con- ductance is an upper limit for the conductance of a real film made of similar CNTs. The conductivity calculated σ, also represents an upper bound for the maximum electrical conductivity of real nanotube network films. Based on an empirical analysis of the conductivity data, it is found that it scales with a combined variable of the parameters Vf , ℓ and D. This dependence allows for a direct comparison of experimental data against this supe- rior limit estimate. A comparison between experimental and simulation results, provides a direct indication of how much it is possible to further improve the transport perfor- mance of the laboratory produced films. Defining a scaled variable x = Vf × ℓ/D2, figure 13 shows the dependence of the film conductivity (in units of siemens per meter) with x. The data points represented by circles correspond to nanotubes of varying aspect ratio, and includes all the data in figure 12. The data corresponding to networks of finite-sized (5, 5) SWNTs shows considerable dispersion, even after averaging over several configurations. Nonethe- less, it is possible to identify a general trend on the data which stresses the linear dependence of the conductivity with x, represented by the least-squares linear fit indicated by the dashed (black) line. The dispersion on the data for CNTs is caused by the finite-length of the nanotubes considered. Since the Green function of independent nan- otubes is calculated numerically, it quickly becomes too computationally demanding to increase the length of the nanotubes much further. In order to overcome this computational limitation, one can investigate the conductance of networks made of ide- alised nanowires, represented by linear monatomic chains. In order to compare the results with carbon nanotubes, an orbital degeneracy is introduced in the chain Hamiltonian such that there are two independent conducting channels. These monatomic chains present a maximum conductance of 4e2/h at the Fermi energy, just like carbon nanotubes. It is possible to write an analytic expression for the Green function of linear chains [33], and very long chains can be simulated with little computational effort. In order to min- imise the finite-size effects on the local density of states, chains with L = 16000 atoms where considered. Also shown in figure 13 is data obtained for the conduc- tivity of networks composed of ideal nanowires (squares), with varying volume fraction and aspect ratio. In this case we can see a much smaller dispersion on the calcu- lated conductivity. Furthermore, by fitting a straight line to the nanowire conductivity we find σu = M × 4.25 × 10−5 × Vf ℓ D2 (in S/m), (26) where the factor M accounts for the number of conduction channels in the wires, for the case considered here M = 2. Because of all the idealisations applied to our calculations of the conductivity, the expression above is deemed to be 12 ) / m S ( σ 3.5×105 3.0×105 2.5×105 2.0×105 1.5×105 1.0×105 0.0 Upper Bound Ideal NWs (5,5) SWNTs 2.0×109 x (m-1) 4.0×109 Figure 13: Maximum conductivity of carbon nanotube films as a function of scaled variable x = Vf × ℓ/D2. the upper limit for the conductivity of random network films. It is instructive to test the expression in equation (26) for typical SWNT values, namely ℓ = 1µm and d = 1.2 nm. For Vf = 30%, the predicted upper bound would be σu = 1.8 × 107S/m if CNTs of these dimensions could be fully dispersed to form the network. However, carbon nan- otubes are known to bundle together, which means that in reality wire diameters are considerably larger. On the other hand, larger-diameter bundles have more nanotubes on the surface leading to more current-carrying channels per wire. Taking all this into consideration, we can com- pare our expression with the highest-conductivity case re- ported so far (Vf = 30%, ℓ = 5 µm, d = 20 nm) [3, 28, 34]. Our prediction of σu = 9 × 106 S/m is only one order of magnitude superior to the measured value of σ = 6 × 105 S/m. Bearing in mind that the upper bound here ob- tained assumes a number of ideal conditions that are ex- perimentally unavoidable, this might be a clear indication that we are approaching a saturation point in the conduc- tivity of nanotube network films. Finally, although our focus has been on disordered net- works comprised of carbon nanotubes, we can extend our results to deal with other wires. This could represent the case of networks made of other conducting materials, such as noble-metal nanowires, for instance. In this case the number of conducting channels M depends linearly on the wire diameter and the overall conductivity of the network is likely to scale inversely with D, with a proportionality constant that depends on the specifics of the material in question. If the conductivity of CNT network films is ap- proaching its saturation point, it is likely that wires other than nanotubes may occupy the post of ideal components for disordered network films. Indeed, it has been shown recently [19] that silver nanowire network thin films might be a strong contender in the race for the next generation of transparent electrodes. References 6. Concluding Remarks In this work we have shown how a relatively simple methodology can be applied to model the complex features present in the morphological structure of carbon nanotube network films. A multiscale approach has been applied to deal with the despairingly different length scales present. From a macroscopic point of view, individual nanotubes are modelled as rigid rods of finite length and diameter. A network of interconnected rods represents the film and the volume fraction of these networks allows for a connection between theory and experiment. On the other hand, from a microscopic approach carbon nanotubes are described by a semi-empirical Hamiltonian, and Green function meth- ods can be applied to calculate the electronic conductance. Merging both scales in a unique multiscale approach is pos- sible when the connectivities derived from the macroscopic model are used as input into the microscopic conductance calculation. The volumetric fraction of the film occupied by rods was found to be linearly dependent on the number of rigid rods in the network. The average number of connections per rod in a random network was found to scale universally with a combined variable given by the product of the vol- ume fraction with the aspect ratio of the individual rods considered. Similarly, the number of rods connected to electrodes placed in opposing faces of the film was found to scale with the volume fraction times the square of the aspect ratio. We have presented a method, based on previ- ously developed methodologies, to calculate the equivalent resistance between arbitrary points on a network of resis- tors. The rigid rods percolation model was combined with resistance calculations in order to obtain the conductance of model carbon nanotube network films. We have shown how our results are in perfect agreement with standard percolation theory, and how the conductance of nanotube films is affected by the morphology of the networks. Direct comparison of simulation results with experimental mea- surements are found to be in agreement, and provide es- timates for the junction resistance in real world nanotube films. These estimates are compatible with independent measurements reported in the literature. We have shown that an electronic structure descrip- tion based on semi-empirical tight binding Hamiltonians is capable of accounting for the heavily disordered envi- ronments found on nanotube networks. In conjunction with results from the geometrical percolation model, we have presented simulation results for the conductance of single-walled carbon nanotube networks. Finally, by con- sidering a fully ballistic transport regime on films of ideal nanowires, and comparing the ensuing results with typi- cal experimental evaluations we have shown that carbon nanotube network films might be approaching their perfor- mance limit, and that perhaps metallic nanowires repre- sent a more promising alternative to future developments. 13 [1] R.H. Baughman, A.A. Zakhidov, and W.A. de Heer. Car- Science, bon nanotubes-the route toward applications. 297(5582):787 -- 92, August 2002. [2] E.S. Snow, J.P. Novak, P.M. Campbell, and D. Park. Random networks of carbon nanotubes as an electronic material. Applied Physics Letters, 82(13):2145, 2003. [3] Z. Wu, Z. Chen, X. Du, J.M. Logan, J. Sippel, M. Nikolou, K. Kamaras, J.R. Reynolds, D.B. Tanner, A.F. Hebard, and A.G. Rinzler. Transparent, conductive carbon nanotube films. Science, 305(5688):1273, August 2004. [4] Q. Cao and J.A. Rogers. Ultrathin Films of Single-Walled Car- bon Nanotubes for Electronics and Sensors: A Review of Fun- damental and Applied Aspects. Advanced Materials, 21:29 -- 53, January 2009. [5] L. Hu, D. S. Hecht, and G. Gruner. Percolation in Transparent and Conducting Carbon Nanotube Networks. Nano Letters, 4(12):2513 -- 2517, December 2004. [6] L. Hu, D. S. Hecht, and G. Gruner. Carbon Nanotube Thin Films: Fabrication, Properties, and Applications. Chemical Re- views, 110(10):5790 -- 5844, July 2010. [7] F. Blighe, P.E. Lyons, S. De, W.J. Blau, and J.N. Coleman. On the factors controlling the mechanical properties of nanotube films. Carbon, 46:41 -- 47, January 2008. [8] P.E. Lyons, S. De, F. Blighe, V. Nicolosi, L.F.C. Pereira, M.S. Ferreira, and J.N. Coleman. The relationship between network morphology and conductivity in nanotube films. Journal of Applied Physics, 104(4):044302, 2008. [9] G. Venezian. On the resistance between two points on a grid. American Journal of Physics, 62(11):1000, 1994. [10] D. Atkinson and F.J. Van Steenwijk. Innite resistive lattices. American Journal of Physics, 67(6):486 -- 492, 1999. [11] J. Cserti. Application of the lattice Greens function for calculat- ing the resistance of an infinite network of resistors. American Journal of Physics, 68(10):896, 2000. [12] J. Cserti, G. David, and A. Piroth. Perturbation of infinite networks of resistors. American Journal of Physics, 70(2):153, July 2002. [13] D. Stauffer and A. Aharony. Introduction to percolation theory. Taylor & Francis, 1992. [14] G. Pike and C. Seager. Percolation and conductivity: A com- puter study. I. Physical Review B, 10(4):1421 -- 1434, August 1974. [15] C. Seager and G. Pike. Percolation and conductivity: A com- puter study. II. Physical Review B, 10(4):1435 -- 1446, August 1974. [16] S. Munson-McGee. Estimation of the critical concentration in an anisotropic percolation network. Physical Review B, 43(4):3331 -- 3336, February 1991. [17] A. Celzard, E. McRae, C. Deleuze, M. Dufort, G. Furdin, and J.F. Marech´e. Critical concentration in percolating sys- tems containing a high-aspect-ratio filler. Physical Review B, 53(10):6209 -- 6214, March 1996. [18] M.S. Fuhrer, J. Nygard, L. Shih, M. Forero, Y.G. Yoon, M.S.C. Mazzoni, H.J. Choi, J. Ihm, S.G. Louie, A. Zettl, and P.L. McEuen. Crossed nanotube junctions. Science, 288:494 -- 497, April 2000. [19] P.N. Nirmalraj, P.E. Lyons, S. De, J.N. Coleman, and J.J. Boland. Electrical connectivity in single-walled carbon nan- otube networks. Nano letters, 9(11):3890, November 2009. [20] M. Brandbyge, J.-L. Mozos, P. Ordej´on, J. Taylor, and K. Stok- bro. Density-functional method for nonequilibrium electron transport. Physical Review B, 65(16):165401, March 2002. [21] A.R. Rocha, V.M. Garc´ıa-Su´arez, S.W. Bailey, C.J. Lambert, J. Ferrer, and S. Sanvito. Towards molecular spintronics. Nature materials, 4(4):335, April 2005. [22] Y.G. Yoon, M. Mazzoni, H. Choi, J. Ihm, and S.G. Louie. Struc- tural Deformation and Intertube Conductance of Crossed Car- bon Nanotube Junctions. Physical Review Letters, 86(4):688 -- 691, January 2001. [23] D. Mowbray, C. Morgan, and K. Thygesen. Influence of O2 and N2 on the conductivity of carbon nanotube networks. Physical Review B, 79(19):195431, May 2009. [24] C. Caroli, R. Combescot, P. Nozieres, and D. Saint-James. Di- rect calculation of the tunnelling current. Journal of Physics C: Solid State Physics, 4:916, 1971. [25] S. Datta. Electronic transport in mesoscopic systems. Cam- bridge University Press, Cambridge, UK, 1997. [26] J. Mathon, A. Umerski, and M. Villeret. Oscillations with Co and Cu thickness of the current-perpendicular-to-plane giant magnetoresistance of a Co/Cu/Co (001) trilayer. Physical Re- view B, 55(21):14378, 1997. [27] J. Mathon. Tight-binding theory of tunneling giant magnetore- sistance. Physical Review B, 56(18):11810, 1997. [28] E.M. Doherty, S. De, P.E. Lyons, A. Shmeliov, P.N. Nirmal- raj, V. Scardaci, J. Joimel, W.J. Blau, J.J. Boland, and J.N. Coleman. The spatial uniformity and electromechanical stabil- ity of transparent, conductive films of single walled nanotubes. Carbon, 47(10):2466 -- 2473, August 2009. [29] L.F.C. Pereira, C.G. Rocha, A. Latge, J.N. Coleman, and M.S. Ferreira. Upper bound for the conductivity of nanotube net- works. Applied Physics Letters, 95(12):123106, 2009. [30] Michael Segal. Nanotubes: Effective networking. Nature Nan- otechnology, (doi:10.1038/nnano.2009.316), October 2009. [31] D. S. Hecht, L. Hu, and G. Gruner. Conductivity scaling with bundle length and diameter in single walled carbon nanotube networks. Applied Physics Letters, 89(13):133112, 2006. [32] D.H. Shin, H.C. Shim, J.-W. Song, S. Kim, and C.-S. Han. Conductivity of films made from single-walled carbon nanotubes in terms of bundle diameter. Scripta Materialia, 60(8):607 -- 610, April 2009. [33] E.N. Economou. Green's functions in quantum physics. Springer, Berlin, third edition, 2006. [34] H.-Z. Geng, K.K. Kim, K.P. So, Y.S. Lee, Y. Chang, and Y.H. Lee. Effect of acid treatment on carbon nanotube-based flexible transparent conducting films. Journal of the American Chem- ical Society, 129(25):7758, June 2007. 14
1605.00994	2	1605	2016-08-26T16:26:33	Magnetic-field-induced relativistic properties in type-I and type-II Weyl semimetals	[ "cond-mat.mes-hall", "cond-mat.mtrl-sci" ]	We investigate Weyl semimetals with tilted conical bands in a magnetic field. Even when the cones are overtilted (type-II Weyl semimetal), Landau-level quantization can be possible as long as the magnetic field is oriented close to the tilt direction. Most saliently, the tilt can be described within the relativistic framework of Lorentz transformations that give rise to a rich spectrum, displaying new transitions beyond the usual dipolar ones in the optical conductivity. We identify particular features in the latter that allow one to distinguish between semimetals of different types.	cond-mat.mes-hall	cond-mat	a Magnetic-field-induced relativistic properties in type-I and type-II Weyl semimetals Serguei Tchoumakov, Marcello Civelli and Mark O. Goerbig Laboratoire de Physique des Solides, CNRS UMR 8502, Univ. Paris-Sud, Universit´e Paris-Saclay F-91405 Orsay Cedex, France (Dated: August 29, 2016) We investigate Weyl semimetals with tilted conical bands in a magnetic field. Even when the cones are overtilted (type-II Weyl semimetal), Landau-level quantization can be possible as long as the magnetic field is oriented close to the tilt direction. Most saliently, the tilt can be described within the relativistic framework of Lorentz transformations that give rise to a rich spectrum, displaying new transitions beyond the usual dipolar ones in the optical conductivity. We identify particular features in the latter that allow one to distinguish between semimetals of different types. The discovery of massless ultrarelativistic electrons in graphene [1, 2] has triggered a tremendous inter- est in novel types of semimetallic phases in condensed- matter systems [3–5]. A particularly intriguing class is that of electronic systems with tilted Dirac cones that were first investigated in the framework of the quasi- twodimensional (2D) organic crystal α-(BEDT-TTF)2I3 [6]. In contrast to graphene where the Dirac cones are sit- uated at high-symmetry points [7], tilted Dirac cones in α-(BEDT-TTF)2I3 arise and migrate in wavevector space as a function of pressure applied to the system. One of the most remarkable consequences of the tilt is unveiled in the presence of a magnetic field B. Indeed, the tilt renormalizes the cyclotron frequency by the same factor as an inplane electric field E that was investigated in a co- variant description by Lukose et al. [8]. This hints at an intimate relation between the tilt of the Dirac cones and relativistic transformations that was later investigated in the framework of magnetotransport [9] and magnetoop- tics [10]. Indeed, one may naturally ask what happens when the Dirac cones are overtilted such that the original recip- rocal isoenergy trajectories are no longer closed ellipses but open hyperbolas [7]. In the presence of a magnetic field, the tilted Dirac cones may be characterized with the help of Lorentz transformations. As we discuss in more detail below, this can be achieved if one identifies the tilt with an effective electric field E. Below the critical tilt, we are confronted with a "weak" electric field that allows for a Lorentz boost to a frame of reference where E effectively vanishes. In this so-called magnetic regime [11], the electronic orbits are then still closed (cyclotron orbits) and their energy is quantized into Landau level (LLs) [12]. In contrast, such a boost is not possible for overtilted cones that corresponds to a "strong" electric field. Via a Lorentz transformation, one can now sim- ply find a frame of reference where the B-field vanishes. This electric regime is characterized by open orbits that prevent LL quantization. Whereas overtilted cones have not been found in 2D materials, following the proposal by Soluyanov et al. [13], several 3D systems have been identified during the last months as possible candidates for representing such phase [14–17]. Furthermore, these phases, coined type-II Weyl semimetals (WSM), have been classified from a topological point of view [18, 19]. In the present paper, we show that the above classifi- cation needs to be revisited in 3D systems that present a remarkably rich behavior. Indeed, we find that a type- II WSM can undergo a transition from the magnetic to the electric regime as a function of the angle between the magnetic field and the tilt direction. Whereas the elec- tric regime is at first sight the natural regime of a type-II WSM, the latter can nevertheless show LL quantization– the tilt is then most prominent in the direction of the B-field, and the original overtilt is well represented by a onedimensional band in the latter direction. This situa- tion needs to be contrasted with that of at type-I WSM that is always in the magnetic regime regardless of the B-field direction. However, one can, at least in prin- ciple, induce transitions between WSM of the two dif- ferent types by the application of an additional electric field with a nonzero component perpendicular to the B- field. Finally, we investigate in detail the magneto-optical signatures in the magnetic regime, which allow for a dis- tinction between type-I and type-II WSM. Most saliently, we find, similarly to tilted Dirac cones in 2D materials [10], that a magnetic field applied in a direction different from the tilt yields magnetooptical transitions beyond the usual dipolar ones that couple LLs with adjacent in- dices, n → (n ± 1). Tilted Dirac cones in 3D systems can be modeled con- veniently by the Hamiltonian H = ω0 · k1 + 3(cid:88) µ=1 vµkµ σµ, (1) in terms of the three Pauli matrices σµ (µ = 1, 2, 3) and the anisotropic Fermi velocities vµ. Here and in the fol- lowing, we use a system of units where = 1. Further- more, we combine the tilt velocities ω0 = (ω0x, ω0y, ω0z) into the tilt parameter vector t = (2) which allows us to distinguish a type-I WSM (t < 1) from a type-II WSM (t > 1), in analogy with the 2D , ω0yvy , ω0zvz (cid:18) ω0xvx , (cid:19) case [7]. In order to account for a magnetic field in the z- direction, we use the Peierls substitution k → (qx, qy, kz), with qx/y = kx/y + eAx/y(r) and B = ∇ × A(r). As a consequence of the r-dependence of the vector po- tential, the x- and y components of the new momenta become operators that no longer commute, [qx, qy] = −isign(vxvyB)/l2 B, in terms of the magnetic length lB = 1/(cid:112)eB. In the following, we consider the tilt to have no component in the y-direction–this can always be achieved via a rotation and a possible rescaling of the momenta– and the Hamiltonian thus reads (see Supplementary Ma- terial) HB = (ω⊥qx + ω0zkz)1 + v⊥(qx σx + qy σy) + vzkz σz, (3) where v⊥ = (cid:112)vxvy is the average Fermi velocity in the xy-plane, and ω⊥ = v⊥(cid:112)(ω0x/vx)2 + (ω0y/vy)2 is a rescaled tilt velocity. While the Hamiltonian (3) can in principle, within a lengthy calculation, be solved by the introduction of the usual ladder operators a± = lB(qx ± iqy)/ 2, similarly to the 2D case [20, 21], a more elegant method consists of using a hyperbolic transformation to change the eigen- value equation (HB − E1)Ψ(cid:105) = 0 into √ (cid:16) e θ 2 σx HBe θ 2 σx − Eeθσx (cid:17) Ψ(cid:105) = 0, (cid:104) where Ψ(cid:105) = N exp(−θσx/2)Ψ(cid:105), and N is a normaliza- tion constant required since the hyperbolic transforma- tion does not preserve the norm of the wave functions. This transformation is nothing other than a Lorentz boost in the x-direction in terms of the relativistic pa- rameter tanh θ = β [8], which reads β = −ω⊥/v⊥ in the magnetic regime (β < 1) and β = −v⊥/ω⊥ in the electric regime (β > 1) [22]. As mentioned in the intro- duction, only the magnetic regime allows for quantized LLs, and we concentrate henceforth on this regime. In the transformed frame of reference, the eigenvalue equa- tion (4) reads γ(ω0z kz − E)1 + v⊥(qx σx + qy σy) + vz kz σz (5) where γ = (1 − β2)−1/2 is the dilatation factor. The tilde indicates transformed wave vectors, qx = qx/γ + γβ(ω0zkz − E)/v⊥, qy = qy, and kz = kz, in agreement with a Lorentz boost in the x-direction [11]. One notices that the eigenvalue equation (5) now only contains the noncommutative wavevector compo- nents qx and qy in the offdiagonal matrix elements. It can therefore easily be solved by the introduction of the standard ladder operators, but with a renormal- ized B-field because the commutation relations between the transformed wavevector components read [qx, qy] = −isign(vxvyB)/γl2 B and [qx, kz] = [qy, kz] = 0. One thus (cid:105) Ψ(cid:105) = 0, 2 FIG. 1. (Color online) Graphical representation of the tilt- parameter vector t. The unit sphere t = 1 represents the boundary between type-I (blue arrow) and type-II (red ar- rows) WSM. In the presence of a magnetic field along the z−axis, the projection of a tilt-parameter vector on the xy plane identifies the regime, the magnetic regime (i.e. inside the green disc) or in the electric regime (outside the disc). The vectors tm1, tm2, and te2 indicate a type-I WSM, a type- II WSM in the magnetic regime, and a type-II WSM in the electric regime, respectively. (4) finds the LLs (cid:115) 1 γ 2eBv2⊥ n for n > 0, En,λ(kz) = ω0zkz + λ γ E0(kz) = [ω0z − sign(vxvyB)vz/γ] kz z + v2 z k2 for n = 0, (6) where λ = ±1. Let us first discuss the different regimes. As mentioned above, LL quantization is only possible in the magnetic regime. Similarly to the tilt parameter (2), this can conveniently be described in terms of the inplane tilt vector, (cid:19) (cid:18) ω0xvx , ω0yvy t⊥ = t × B B = . (7) Its norm is precisely β and can be related to the tilt parameter (2) and the angle α between the magnetic field and the tilt direction, via β = t⊥ = t sin α. These geometric relations are shown in Fig. 1, where the sphere t = 1 indicates the border between a type-I (inside) and a type-II WSM (outside). Similarly, the magnetic regime is represented by the inside (green) of the circle x + t2 t2 y = 1 while the electric regime is situated outside. Whereas a type-I WSM, represented by the blue vector tm1 inside the sphere, has always a projection to the xy-plane inside the sphere, i.e. in the magnetic regime, one needs to distinguish two situations for type-II WSM. Below a critical angle αc, given by sin αc < 1/t, (8) txtzWSM-IIWSM-I111magneticelectrictm2αtytm1te2Bthe projection of the tilt vector (tm2 in Fig. 1) is within the sphere and thus in the magnetic regime. However, if the condition (8) is not satisfied, the projection (of te2) is outside the unit circle, and the system is thus in the electric regime. β = 0.75, in units of v⊥/lB (cid:39) 26 meV × v⊥[106m/s]/(cid:112)B[T]. FIG. 2. (Color online) LL spectrum of a tilted Weyl cone for The green lines are n > 0 LLs and the blue (red) line repre- sents the n = 0 state for sign(vxvyvzB) = 1 (sign(vxvyvzB) = −1). The left figure corresponds to the case of a type-I Weyl semimetal where the n = 0 LL group velocity changes sign with chirality. On the contrary, the right figure is for a type-II Weyl semimetal, where the n = 0 slope representing the LL group velocity direction is independent of chirality. √ The LL spectrum (6) is represented in Fig. 2 for a type-I WSM [panel (a)] and a type-II WSM [panel (b)] in the magnetic regime. One retrieves the known 2D results for kz = 0 [7, 12, 20], i.e. a graphene-like LL spectrum with En,± ∝ ±(1 − β2)3/4 Bn, where the spacing is reduced by the relativistic factor (1 − β2)3/4. In the 3D case, these LLs evolve into 1D bands with a dispersion in kz. It is precisely this dispersion that bares information about the underlying WSM type since their bands are also tilted and can be seen as gapless (n = 0) or gapped (n (cid:54)= 0) Dirac cones with a tilt below (type-I) or above (type-II) the critical value. Their tilt is characterized by the 1D tilt parameter tz = ω0z/v(cid:48) z, in terms of the velocity v(cid:48) z =(cid:112)1 − β2vz, t cos(α) (cid:112)1 − t2 sin(α)2 tz = . (9) One thus finds that for any angle α one has tz < 1 for a type-I and tz > 1 for a type-II WSM, i.e. the type of the 1D bands is the same as the original 3D WSM in the absence of a magnetic field. 3 A naturally arising question is whether one can also in- duce a transition from the magnetic to the electric regime in a type-I WSM. From a theoretical point of view, this can in principle be achieved by an electric field with a nonzero component perpendicular to B. Let us consider for simplicity an electric field E in the y-direction that yields a supplementary term V = eEy1 to Hamiltonian In the Landau gauge A = −Byex, the modified (1). B − (E/B)kx1, where H(cid:48) Hamiltonian reads HE,B = H(cid:48) B has the same structure as Eq. (3) if we replace the tilt velocity [9] ω0 → ω = ω0−E × B/B2 or equivalently the tilt parameter t → tE , with (cid:18) (E × B)x vx , tE = t − 1 B2 (E × B)y vy (E × B)z vz , . (10) (cid:19) (cid:88) A type-I WSM can therefore undergo a transition from the magnetic to the electric regime for a judicious choice of the electric field if the criterion (8) is satisfied in terms of the tilt parameter tE , inducing a breakdown of the LL spectrum [8, 9]. From an experimental point of view, however, the situation is more involved because of the screening of the electric field due to free carriers both in the bulk and at the surfaces. Further studies are thus required to establish the feasibility of such experiments. We show however that distinctive signatures of type-I and type-II WSM are clearly accessible in magnetoopti- cal (reflectivity) experiments [23] in the magnetic regime, with no electric field involved. Within linear-response theory, the reflection of light at a polarization l is pro- portional to the real part of the optical conductivity Re σll(ω) = l · Vj,j(cid:48)2 σ0 2πl2 Bω × [f (Ej) − f (Ej(cid:48))] δ(ω − ωjj(cid:48)) j,j(cid:48) (11) where the subscripts j, j(cid:48) denote the quantum numbers (λ, n, kz), σ0 = e2/2π is the quantum of conductance, and ωjj(cid:48) = Ej − Ej(cid:48) is the energy difference between the final and initial states. Furthermore, f (Ej) is the Fermi-Dirac distribution for a Fermi level that we choose at zero energy (at the position of the Weyl point) and the matrix element is given by Vj,j(cid:48) = (cid:104)Ψj∇kHBΨj(cid:48)(cid:105). This formula is integrated analytically for two elliptic polarizations (see Supplementary Material). One of the most salient features of the matrix elements in Eq. (11) is related to the nonorthogonality of the spinor com- ponents as a consequence of the relativistic boost [10], i.e. when β (cid:54)= 1. Indeed, the harmonic-oscillator wave functions acquire an energy-dependent shift in their posi- tion due to the Lorentz transformation, which thus yields energy-dependent overlap functions (see Supplementary Material). As a consequence, the usual dipolar selection n → (n ± 1) are violated, and as a function of β addi- tional interband peaks arise in the optical conductivity at energies 2eBv⊥(cid:0)√ √ n(cid:1) . √ ωmn = (1 − β2)3/4 m + (12) -2-112-1.5-1.0-0.50.51.01.5E/(v/lB)vzqzlBv(a)-2-112-1.5-1.0-0.50.51.01.5E/(v/lB)vzqzlBv(b)4 √ This is shown in Fig. 3 for a type-I [panel (a)] and a type-II WSM [panel (b)], where we depict the optical conductivity as a function of frequency for several val- ues of β. For β = 0, one retrieves the typical spectrum of a WSM discussed in the literature [24, 25], where the ω − ωmn divergence due to the peaks have the usual 1/ 1D character of the LL bands as a function of kz. These are also visible at β (cid:54)= 0 and are the main difference with respect to the 2D case discussed in Ref. [10]. Fur- thermore, the optical conductivity of an undoped type-II WSM shows additional peaks at low energies that are due to allowed transitions between LLs of the same original band. Indeed, LLs corresponding to the family λ = +, which are always at positive energies in a type-I WSM, have parts of bands situated at negative energies, whereas those of with λ = − become positive for certain values of kz. Therefore, transitions between LLs of the same family [red arrows in the inset of panel (a)] are no longer blocked by the Pauli principle and contribute to the op- tical conductivity up to the energy z + 1 . (13) ωl = (1 − β2)3/4 √ 2eBv⊥ √ 2tz −(cid:112)t2 (cid:112)t2 z − 1 This yields the unusually large optical conductivity at low frequency and reflects the large density of states of a type-II WSM at zero B-field, in comparison with a type-I WSM. The effect is different from that previously described in Ref. [24] for a type-I WSM where low- frequency peaks in the optical conductivity are obtained upon doping, whereas here the system is undoped. In conclusion, we have shown that 3D WSM show a rich variety of phases in the presence of a magnetic field that unveil the underlying relativistic symmetry. In contrast to 2D systems, the magnetic and the electric regimes, with and without LL quantization, respectively, do no longer coincide with the type of the WSM in 3D. Indeed, a type-II WSM can show LL quantization as long as the angle between the tilt direction and the magnetic field satisfies the condition (8)–in this case, the type-II character is visible in the 1D Landau bands as a function of the wave vector kz in the direction of the B-field. Most saliently, we have shown that the relativistic parameter β has a clear fingerprint in the optical conductivity, which can be probed in magnetooptical measurements, discern- ing different types of WSM. We would finally emphasize that our results are readily generalized to the case of a pair (or more) of Weyl points in a more realistic band structure. If the two cones are related by a discrete sym- metry (e.g. time reversal), the tilts of the two cones are in opposite directions. In this case, one obtains two copies of the same LL spectrum, since the parameter β depends only on the modulus of the tilt vector t⊥. For more pairs of Weyl points, the tilt vectors are no longer necessarily of the same modulus, and the condition (8) for the transition between magnetic and electric regime is FIG. 3. (Color online) Optical conductivity for the e+-elliptic polarization of undoped type-I [panel (a)] and type-II WSM [panel (b)], for different values of β (= 0, 0.1, 0.2, 0.4 and 0.6). The curves have been displaced up for increasing value of β. In the case β = 0 one observes the usual Landau level spectrum with n → (n± 1) transitions while for higher values of the in-plane tilt, more transitions are allowed. Insets are a reminder of the LL structure and the arrows indicate inter-LL transitions visible in the optical conductivity. then not unique for all points. During the writing of the present paper, we became aware of two other preprints on magnetic-field properties of type-II WSM. The findings of Ref. [26] are similar to ours but the authors consider only the optical conductivity for β = 0, i.e. when the tilt is in the same direction as the B-field, in which case one has no violation of the dipolar selection rule n → (n± 1) and no renormalization of the LL spacing. The authors of Ref. [27] study the chiral anomaly furthermore in a lattice tight-binding model and extract numerically the optical conductivity of a type-II WSM in the magnetic regime. We acknowledge fruitful discussions with Milan Orlita and Marek Potemski. β00.10.20.40.602468102468101214Re[σ+(ω)]ℏωlB/vF-3-2-1123qz-4-224E(a)β00.10.20.40.602468102468101214Re[σ+(ω)]-3-2-1123qz-4-224E(b)ℏωlB/vF5 245131 (2013). [25] C.J. Tabert, J.P. Carbotte, E.J. Nicol, Phys. Rev. B 93, 085426 (2016) . [26] Z. Yu, Y. Yao, and S.A. Yang, arXiv:1604.04030. [27] M. Udagawa and E.J. Bergholtz, Phys. Rev. Lett. 117, 086401 (2016). [1] 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). [2] Y. Zhang, Y.-W. Tan, H. L. Stormer, and P. Kim, Nature 438, 201 (2005). [3] X. Wan, A.M. Turner, A. Vishwanath and S.Y.Savrasov, Phys. Rev. B 83, 205101 (2011). [4] A. A. Burkov and L. Balents, Phys. Rev. Lett. 107, 127205 (2011). [5] S.-Y. Xu, I.Belopolski, N. Alidoust, M. Neupane, G. Bian, C.Zhang, R.Sankar, G.Chang, Z.Yuan, C.-C. Lee, S.-M.Huang, H. Zheng, J.Ma, D.S. Sanchez, B.Wang, A.Bansil, F.Chou, P.P. Shibayev, H.Lin, S.Jia, M. Z. Hasan, Science 349, 6248 (2015). [6] S. Katayama, A. Kobayashi, and Y. Suzumura, J. Phys. Soc. Jpn. 75, 054705 (2006); A. Kobayashi, S. Katayama, Y. Suzumura, and H. Fukuyama, ibid. 76, 034711 (2007). J.-N. Fuchs, G. Montambaux and [7] M.O. Goerbig, F.Pi´echon, Phys. Rev. B 78, 4 0454145 (2008). [8] V. Lukose, R. Shankar and G. Baskaran, Phys. Rev. Lett. 98, 11 116802 (2007). [9] M.O. Goerbig, J.-N. Fuchs, G. Montambaux and F.Pi´echon, EPL 85, 5 57005 (2009). [10] J. S´ari, M.O. Goerbig and C.Toke, Phys. Rev. B 92, 3 035306 (2015). [11] J. D. Jackson, Classical Electrondynamics (Wiley, New York), 3rd ed. (1999). [12] For a review, see M. O. Goerbig, Rev. Mod. Phys. 83, 1193 (2011). [13] A.A. Soluyanov, D. Gresch, Z. Wang, Q. Wu, M. Troyer, X. Dai, and B.A. Bernevig, Nature 527, 495-498 (2015). [14] F.Y. Bruno, A. Tamai, Q.S. Wu, I. Cucchi, C. Barreteau, A. de la Torre, S. McKeown Walker, S. Ricco, Z. Wang, T.K. Kim, M. Hoesch, M. Shi, N.C. Plumb, E. Giannini, A. A. Soluyanov and F. Baumberger, arXiv:1604.02411. [15] Y. Wu, N.H. Jo, D. Mou, L. Huang, S. L. Bud'ko, P. C. Canfield and A. Kaminski, arXiv: 1604.05176 [16] L. Huang, T. M. McCormick, M. Ochi, Z. Zhao, M. Suzuki, R. Arita, Y. Wu, D. Mou, H. Cao, J. Yan, N. Trivedi and A. Kaminski, arXiv: 1603.06482 [17] A. Tamai, Q. S. Wu, I. Cucchi, F. Y. Bruno, S. Ricco, T.K. Kim, M. Hoesch, C. Barreteau, E. Giannini, C. Bernard and A. A. Soluyanov, F. Baumberger, arXiv: 1604.08228. [18] L. Muechler, A. Alexandradinata, T. Neupert, and R. Car, arXiv:1604.01398. [19] T. M. McCormick, arxiv:1604.03096. I. Kimchi, and N. Trivedi, [20] T. Morinari, T. Himura, and T. Tohyama, J. Phys. Soc. Jpn. 78, 023704 (2009). [21] N. M. R. Peres and E. V. Castro, J. Phys.: Condens. Matter 19, 406231 (2007). [22] M. Le Bellac and J.M. L´evy-Leblond, II Nuovo Cimento B 14, 2 217-238 (2008). [23] A. Akrap, M. Hakl, S. Tchoumakov, I. Crassee, J. Kuba, M.O. Goerbig, C.C. Homes, O. Caha, J. Novak, F. Teppe, S. Koohpayeh, L. Wu, N.P. Armitage, A. Nateprov, E. Arushanov, Q.D. Gibson, R.J. Cava, D. van der Marel, C. Faugeras, G. Martinez, M. Potemski and M. Orlita, arxiv:1604.00038. [24] P. E. C. Ashby and J. P. Carbotte, Phys. Rev. B 87, SUPPLEMENTARY MATERIAL MINIMAL MODEL FOR TILTED WEYL SEMIMETAL We consider the model of a tilted Weyl Hamiltonian with an anisotropic Fermi velocity 3(cid:88) H = ω0 · k1 + vµkµ σµ, 6 (S1) µ=1 where σµ (µ = 1, 2, 3) are the three Pauli matrices, the vµ are the anisotropic Fermi velocities and ω0 is the tilt velocity. Upon rescaling of the momenta qµ = vµ v⊥ kµ for µ ∈ {1, 2}, (S2) the Hamiltonian reads (S3) where the anisotropy has been absorbed in the tilt parameters, ω1,µ = v⊥ω0,µ/vµ for µ ∈ {1, 2} and ω1,3 = ω0,z. The Hamiltonian can still be simplified by the following rotation H = ω1 · q1 + v⊥(qx σx + qy σy) + vzkz σz,  2 σz 2 σz He−i φ 2 σzΨφ(cid:105) HB = ei φ Ψ(cid:105) = ei φ q(cid:48) x = cos(φ)qx + sin(φ)qy y = cos(φ)qy − sin(φ)qx q(cid:48) q(cid:48) z = kz , (S4) with tan(φ) = ω0,y/ω0,x, one finds the minimal form of the Hamiltonian HB, Eq. (3) of the main text. We emphasize that this rotation, which does not depend on the wave vectors and keeps kz invariant, leaves the commutation rules between the x- and y-components of the wave vector unchanged, even in the presence of a magnetic field. We consider this form of the Hamiltonian in the main text. EIGENSTATES IN THE MAGNETIC REGIME Magnetic regime In the magnetic regime introduced in the main text (5), the momenta fulfil the following commutation rules [we consider sign(vxvyB) = 1] which indicate that the magnetic field is lowered by B → Beff. =(cid:112)1 − β2B. We use the following ladder operators [qy, kz] = 0 , [qx, kz] = 0, γl2 B , (S5) [qx, qy] = − i and solve the Landau levels equation (5) for n > 0,  a(E, kz) =(cid:112) γ a†(E, kz) =(cid:112) γ   (cid:16) ±(cid:16) Ψn,±(cid:105) = 1√ 2 En,±(kz) = ω0,zkz ± 1 γ (cid:113) 2 lB [qx(E, kz) − iqy] , 2 lB [qx(E, kz) + iqy] , (cid:17) 1  , z + 2eBv2⊥ v2 z k2 γ n 2 n − 1(cid:105) 2 n(cid:105) (cid:17) 1 1 ± vzkz ∆En 1 ∓ vzkz ∆En (S6) (S7) (cid:113) where ∆En = v2 z k2 z + 2eBv2⊥ γ n, and for n = 0 E0(kz) = (cid:16) (cid:20) 00(cid:105) Ψ0(cid:105) = (cid:21) . ω0,z − vz γ (cid:17) kz 7 (S8) The states in original basis are obtained by Ψ(cid:105) = (1/N )e θ 2 σx Ψ(cid:105), i.e. the Lorentz boost mixes the components. Number states In Eqs. a†(E, kz)a(E, kz), that depends explicitly on energy E and momentum kz. the number states with their E- and kz-dependence. (S7) and (S8), the n(cid:105) states correspond to the eigenstates of the number operator, n(E, kz) = In the following we explicitly write Two number states n1, E1, kz,1(cid:105) and n2, E2, kz,2(cid:105) are in general not orthogonal. Indeed, the difference between two ladder operators a(E1, kz,1) and a(E2, kz,2) defined in (S6) of different energies, E1 and E2, and momenta, kz,1 and kz,2, is a scalar where a(E1, kz,1) − a(E2, kz,2) = α1,2, α1,2 = α(E1 − E2, kz,1 − kz,2), α(E, kz) = β (1 − β2)3/4 ω0,zkz − E √ 2eBvF (S9) (S10) (S11) . The scalar shift α1,2 signifies that the cyclotron orbits at different energies and momenta are displaced from each other [1]. For this reason the 0, E, kz(cid:105)-states (i.e. such that a(E, kz)0, E, kz(cid:105) = 0) depend on energy and momentum. If one applies (S9) to 0, E2, kz,2(cid:105) one finds the following coherent-state equation a(E1, kz,1)0, E2, kz,2(cid:105) = α1,20, E2, kz,2(cid:105), (S12) because a(E2, kz,2)0, E2, kz,2(cid:105) = 0 by construction. One then deduces [1] the following relation between two number basis of different (E, kz) 0, E2, kz,2(cid:105) = e− α1,22 2 αp 1,2√ p! p, E1, kz,1(cid:105). (S13) ∞(cid:88) p=0 This coherent-state property has consequences on the scalar product of n, E, kz(cid:105)-state which is by construction (cid:104)n2, E2, kz,2n1, E1, kz,1(cid:105) (S14) We replace a(E2, kz,2) by its expression in Eq. (S9) and use the binomial theorem (a + b)n2 = (cid:80)n2 Also we replace (cid:104)0, E2, kz,2 by its expression as a coherent state of a(E1, kz,1) as in (S12). The terms that appear in the sum of the binomial theorem are of the form (cid:1)akbn2−k. (cid:0)n2 k=0 = k (cid:104)0, E2, kz,2a(E2, kz,2)n2n1, E1, kz,1(cid:105). 1√ n2! (cid:104)α1,2a(E1, kz,1)n2−kn1, E1, kz,1(cid:105) (cid:115) (cid:104)α12k + n1 − n2, E1, kz,1(cid:105) n1! = = (k + n1 − n2)! √ ∗k+n1−n2 n1!α 12 (k + n1 − n2)! e− α122 2 . (S15) If one introduces this term back in the sum, one recognizes the generalized Laguerre polynomial L(α) n (x) = (cid:80)n k=0 (cid:0)n+α n−k (cid:1) (−x)k k! and one finds (cid:104)n2, E2, kz,2n1, E1, kz,1(cid:105) (cid:114) n2! n1! = ∗n1−n2 1,2 α L(n1−n2) n2 (α1,22)e− α1,22 2 . The nonorthogonality (S16) of the spinor components has drastic consequences on the Landau levels spectroscopy of tilted Dirac cones, as we expose in detail here. MAGNETO-OPTICS The electron-light coupling is of the form Light-matter interaction (S17) where A is the vector potential corresponding to the oscillating electric field E(ω). For convenience we study this coupling term in the orientation defined by Eq. (S4) and in the basis defined by the Lorentz boost in Eq. (5) of the main text Hint = eA · ∇k H, (cid:16)∇k HB (cid:17) (cid:16) θ 2 σx HB,int = A · e = A · ∇k = (A · e+) ve σ+ + (A · e−) ve σ− + (A · ez) vz σz, 2 σx HBe θ 2 σx θ 2 σx e θ e (cid:17) where we use the fact that the boost β is independent of the momenta. We introduced two convenient – but not orthogonal, i.e. e+ · e− (cid:54)= 0 – elliptical polarization modes e± defined as 8 (S16) (S18) (S19) (S20) (S21) (S22)  ,  vx/γ∓ivy  , 0 e± = ez = 1 ve  0 0 1 (cid:113) where ve = v2 x/γ2 + v2 y and the Cartesian axis are in the directions defined in Eq. (S4). We also defined σ± as σ± = 1 2 (σx ± iσy) . Optical conductivity (S23) We consider optical measurements in the Faraday geometry where the light polarisation l (i.e. A ∼ l) satisfies l · ez = 0. The light reflection at polarization l is proportional to the real part of the optical conductivity σll(ω), defined as j·l = σll(ω)E·l. We use the expression (11) introduced in the main text obtained from linear response theory and consider the chemical potential µ = 0, zero temperature, and ω > 0. In the case of a type-I Weyl semimetal, the band occupation term is non zero for λ = − → + transitions only. Once integrated, one finds the following expression for the optical conductivity Re(cid:2)σI ll(ω)(cid:3) = v2 e σ0nB 2πωvz ω − (1 − β2)3/4v⊥ Θ (cid:104) (cid:104) n∈N (cid:32) l · e+2(cid:88) (cid:88) (cid:114)(cid:104) (cid:20) nm Re(cid:2)(l · e+) (γω)2 − 2eBv2⊥ − 4eBv2⊥ n,m∈N∗ (γω)2 − 2eBv2⊥ (cid:26) √ Θ + γ 2 ∗ (n + m) γ 0 √ 2eBn Rn−1 √ (cid:105) (cid:105)2 (cid:16) 2eBv2⊥ (cid:21)(cid:2)l · e+2Rn−1 (l · e−)(cid:3) Rn (n + m) ω − (1 − β2)3/4v⊥ 2eB ( γ + (cid:27)(cid:19) m (ω)2 + l · e−2Rn m−1(ω)Rn−1 m (ω) , (ω)2 √ (cid:17)2 n) m + [(n − m)2 − (n + m)2] √ (cid:105) m−1(ω)2(cid:3) γ 9 (S24) where Θ(x) is the Heaviside function, nB = 1/2πl2 energy ω, B, and Rs p(ω) is the overlap function introduced in Eq. (S16) at p(ω) = (cid:104)s, En,+, kzp, Em,−, kz(cid:105) Rs (cid:115) = αp−s ω L(p−s) s (αω2)e− αω2 2 , s! p! αω = − β (1 − β2)3/4 ω√ 2eBv⊥ . (S25) (S26) (S27) The expression in the case of a type-II Weyl semimetal also includes λ = ± → ± transitions the contributions of which are similar to the previous − → + transitions. We do not show the corresponding expression here but one has to introduce the band edge energy scales ω(edge) nm,λλ(cid:48) = (1 − β2)3/4 √ 2eBv⊥ (cid:34) λtz√ n − λ(cid:48)(cid:112)n + [t2 (cid:112)t2 z − 1 z − 1] m (cid:35) (S28) . This expression of the band edge is the reason for the low-frequency intraband peak width (13) introduced in the main text. One other property of the band edge is that it is given by ω(edge) m0,λs where s = sign(vxvyvzω0zB) for the transitions involving the n = 0 state. This is not discussed in the main text but it can shift the edges depending on the direction of the magnetic field. [1] M. O. Scully and Z.M. Suhail, Quantum Optics. Cambridge University Press p.50-51 (1997)
1703.03538	1	1703	2017-03-10T04:03:22	Spin-orbit dynamics of single acceptor atoms in silicon	[ "cond-mat.mes-hall" ]	Two-level quantum systems with strong spin-orbit coupling allow for all-electrical qubit control and long-distance qubit coupling via microwave and phonon cavities, making them of particular interest for scalable quantum information technologies. In silicon, a strong spin-orbit coupling exists within the spin-3/2 system of acceptor atoms and their energy levels and properties are expected to be highly tunable. Here we show the influence of local symmetry tuning on the acceptor spin-dynamics, measured in the single-atom regime. Spin-selective tunneling between two coupled boron atoms in a commercial CMOS transistor is utilised for spin-readout, which allows for the probing of the two-hole spin relaxation mechanisms. A relaxation-hotspot is measured and explained by the mixing of acceptor heavy and light hole states. Furthermore, excited state spectroscopy indicates a magnetic field controlled rotation of the quantization axes of the atoms. These observations demonstrate the tunability of the spin-orbit states and dynamics of this spin-3/2 system.	cond-mat.mes-hall	cond-mat	a Spin-orbit dynamics of single acceptor atoms in silicon J. van der Heijden,1, ∗ T. Kobayashi,1 M.G. House,1 J. Salfi,1 S. Barraud,2 R. Lavieville,2 M.Y. Simmons,1 and S. Rogge1, ∗ 1School of Physics and Australian Centre of Excellence for Quantum Computation and Communication Technology, UNSW, Sydney, Australia 2University of Grenoble-Alpes and CEA, LETI, MINATEC, 38000 Grenoble, France Two-level quantum systems with strong spin-orbit coupling allow for all-electrical qubit control and long-distance qubit coupling via microwave and phonon cavities, making them of particular interest for scalable quantum information technologies. In silicon, a strong spin-orbit coupling exists within the spin-3/2 system of acceptor atoms and their energy levels and properties are expected to be highly tunable. Here we show the influence of local symmetry tuning on the acceptor spin- dynamics, measured in the single-atom regime. Spin-selective tunneling between two coupled boron atoms in a commercial CMOS transistor is utilised for spin-readout, which allows for the probing of the two-hole spin relaxation mechanisms. A relaxation-hotspot is measured and explained by the mixing of acceptor heavy and light hole states. Furthermore, excited state spectroscopy indicates a magnetic field controlled rotation of the quantization axes of the atoms. These observations demonstrate the tunability of the spin-orbit states and dynamics of this spin-3/2 system. Single impurities in semiconductors offer an extensive variety of spin systems [1, 2], desirable for the emerging fields of quantum technologies, such as quantum com- munication [3], quantum sensing [4, 5], and quantum information [6, 7]. Impurities with spin-1/2, such as phosphorus in silicon, are promising two-level spin sys- tems for quantum technologies [8]. Alternatively, im- purities with spin-3/2 offer the possibility to isolate a two-level system whose quantum properties strongly de- pend on the specific splitting and mixing of the spin-3/2 states, thereby providing a highly tunable quantum sys- tem. Several spin-3/2 impurities have been investigated in recent studies, such as single Co atoms [9, 10] and Si vacancies in silicon carbide [11, 12], as well as spin-3/2 hole quantum dots [13–15]. However, the adjustability of impurity spin-3/2 systems has not yet been demon- strated. Here this tunability is shown using the until now little explored spin-3/2 system of a single acceptor in sil- icon, which has a level configuration and environmental couplings predicted to yield highly tunable qubits [16– 19]. Of particular interest is the possibility to tune the acceptor system to a regime with a large dipole coupling whilst being robust against charge noise [18]. By using a silicon CMOS transistor to access the single acceptor regime [20], which allows excellent control over the en- vironment of the atom, this work demonstrates control over the level configuration of a single acceptor in silicon and probes the resulting influence on the spin-dynamics of the acceptor states. A predicted strong spin-orbit coupling for the heavy hole (mj = ±3/2) and light hole (mj = ±1/2) states makes acceptors in silicon good candidates for spin-orbit qubits [16–18]. This type of qubit is advantageous for the all-electrical qubit control by local gates [15, 21–23] and the long-range qubit coupling via the interaction with ∗ [email protected]; [email protected] microwave resonators [24–26]. The level configuration of the acceptor atom is predicted to strongly depend on the local strain, electric field, magnetic field, and the prox- imity of the atom to an interface [19, 27, 28]. As a conse- quence the essential properties of possible two-level quan- tum systems created from these levels, such as the electric dipole coupling, relaxation time and coherence time, can be drastically changed [16–18]. So far spin-dynamic ex- periments on acceptor ensembles in bulk silicon have been hindered by a large inhomogeneous broadening, originat- ing from the random heavy-light hole splitting induced by each atom's local environment [29]. By addressing indi- vidual acceptor atoms, which is essential to operate them as qubits, this effect of inhomogeneous broadening is cir- cumvented. Single acceptor measurements have demon- strated the effects of electric field [30], local strain [31], a nearby interface [32], magnetic field [20], and Hubbard- like interactions between two acceptors [33]. Here we demonstrate the ability to tune the level con- figuration and, as a consequence, the hole spin dynamics of an acceptor atom, thereby showing the fundamental versatility of this spin-3/2 system. For this study a two- hole system confined to two boron atoms in a state-of- the-art silicon CMOS transistor (see Supplementary in- formation A for fabrication details) is investigated using a combination of single hole transport [20, 34, 35] and dispersive radio-frequency (rf) gate reflectometry tech- niques [36–40] (see Fig. 1a). The crossover point between one hole on each acceptor, the (1,1) state, and both holes on one acceptor, the (2,0) state, is used to probe the spin state of the double hole system, making use of spin selective tunneling [41]. Magnetic field dependent mea- surements of the relaxation rate and the excited state spectrum of the two-hole system show that the quantiza- tion axis of the angular momentum can be rotated with the applied field, which strongly tunes the mixing and splitting between the heavy and light hole states. The amount of control over the level configuration of a sin- gle acceptor atom demonstrated here is an essential step 2 (grey arrows in Fig. 1b) imply the charging of several charge states of the same localized state, here likely to be an unintentional quantum dot at the Si/SiO2 inter- face [42]. In contrast, a boron atom can bind no more than two holes [43]. A single charging line, unaccompa- nied with any counterparts of equal slope (black arrow in Fig. 1b), indicates the charging of the first hole onto a single boron atom. The location of this atom is esti- mated by comparing the capacitive couplings to the four transistor electrodes (see Supplementary information B), which reveal a position not directly beneath the top gate, but instead close to the drain lead. This verifies that this charging line does not come from the charging of a gate- defined quantum dot. A sudden electrostatic shift in the charging signal of this acceptor, shown in Fig. 2a, is the signature of the charging of another localized site nearby. Further mea- surements of this state, presented later in this paper, show that this site is a second boron atom. The neg- ative ∆φrefl found within the break of the line indicates the tunneling of holes between the two atoms. A de- tailed study of the reflected signal in the frequency do- main shows that the acceptor-lead tunnel rate is compa- rable to the resonant frequency (583 MHz) of the LC- circuit (Supplementary information C). In contrast, the inter-acceptor tunnel rate is found to be much faster than this resonant frequency. The tunneling of holes from ei- ther lead to the second acceptor atom is not observed, likely due to tunnel rates much slower than the resonant frequency. Two-hole system The magnetic field dependence of the inter-acceptor tunneling, shown in Fig. 2b, reveals the number of holes bound to the acceptor atoms. We observe the signa- ture of Pauli spin blockade in the reflected signal [40, 44] when a magnetic field is applied along the axis of the nanowire that forms the top gate of the transistor (y- direction in Fig. 1a). Approximately at 0.25 T the re- flectometry signal starts to weaken due to spin-selective tunneling associated with the increasing thermal popula- tion of a triplet state. This observation of spin-selective tunneling indicates the presence of an even number of holes bound to the two acceptor atoms, from which it is concluded that the inter-acceptor tunneling corresponds to the (1,1)↔(2,0) transition. To describe the two-hole system we use a theoretical framework to construct the (1,1) and (2,0) states from single hole states (Supplementary information D). The four lowest energy states of a single hole bound to an ac- ceptor in silicon can be described by angular momenta of ±3/2 (heavy holes) and ±1/2 (light holes). A distortion of the symmetry of the wave function can lift the degen- eracy of heavy and light holes [30–32]. Our experiments are consistent with heavy hole ground states for both ac- ceptors, with energy splitting ∆i LH between the heavy FIG. 1. Detection of a single acceptor atom. a, Scanning electron microscope image of the used tri-gate transistor in a schematic diagram of the experimental setup for transport and rf gate reflectometry measurements. b, Change in re- flected phase from the LC-circuit (red) compared to the mea- sured source-drain current (blue), both at 1 mV VD. The signals of a boron atom (black arrow) and an unintentional quantum dot (grey arrows) are identified. towards an acceptor qubit. Identification of individual acceptors RF gate reflectometry [36–40] is used to detect in- dividual acceptors. The charging of a boron atom in the nano-transistor is observed as a shift of the phase, ∆φrefl, or the amplitude, ∆Arefl, of the reflection from a LC-circuit connected to the top-gate, when driven near the resonance frequency with VRF. This shift is caused by a change in admittance of the LC-circuit induced by tunneling of holes when a localized site in the transistor changes its charge occupation [38]. In a map of topgate voltage (VTG) and backgate voltage (VBG) in the pres- ence of a small drain voltage (VD = 1 mV), shown in Fig. 1b, some charging lines appear far below the onset of hole transport current ISD. These lines are identified as the charging of localized sites which are only strongly tunnel coupled to one of the electrodes. The slopes of these lines correspond to the capacitive coupling ratio to the top and back gate. Multiple lines with equal slopes 145 nm BOXsubstratebias-tee40 mK4 KVRFI/Q VTG VBGVDL = 270 nH Cp~ 0.3 pFISDbias-teexyz-7600-5050I [pA]-4-2024VTG [mV]-800-820-780VBG [V]-840-860Δφrefl [°]25VTG boronquantum dotba3 FIG. 2. Two-hole system. a, Reflected amplitude and phase response showing a break in the acceptor-lead charging line (red), caused by the charging of a nearby acceptor atom (dashed lines). b, c, Measured and calculated magnetic field dependence of the inter-acceptor tunnelling phase response. The magnetic field yielding the singlet-triplet crossing BS→T and singlet- quadruplet crossing BS→Q are plotted in c as a function of detuning by dashed curves. d, Schematic overview of the relevant two-hole states. The (1,1) and (2,0) SHH states [blue], (1,1) THH states [green], and (2,0) QLH states [red] are shown. While this subset of two-hole states is sufficient to describe our experiments, a full description of the spin-orbit states can be found in Supplemetary information D. Insets show a depiction of the potential landscape and possible spins involved in these states. HH, T0 HH, and T+ and light hole states of acceptor i. The total energy of the two-hole system at zero magnetic field is shown as a function of the energy level detuning between the atoms in Fig. 2d. To interpret our measurements it is sufficient to only consider the lowest spin manifold for the (1,1)-configuration (right side of Fig. 2d), where one heavy hole resides on each acceptor, forming a singlet state, SHH, and three triplet states, T− HH. For the (2,0)-configuration (left side of Fig. 2d), which is the doubly occupied acceptor A+ state, the exchange en- ergy between holes with the same angular momentum (JHH, JLL) or different angular momenta (JLH=JHL) are treated in the limit where JHH, JLL > ∆LH > JLH, which is the only regime consistent with our data. In analogy to the two-hole states of neutral group II acceptors [45] and of two closely spaced boron atoms [33], the lowest energy manifold in the (2,0)-configuration consists of six states, split by ∆LH and JLH. The ground state is the heavy hole singlet, SHH. Four states consisting of one heavy and one light hole, Qi LH, with i denoting the angular momentum, are ∆LH higher in energy and split by JLH into singlet and triplet states, as shown in Fig. 2d. The sixth state is the light hole singlet state, SLL, another ∆LH higher in energy. At zero magnetic field the tunnel coupling t between the (1,1) and (2,0) SHH states gives rise to the inter- acceptor reflectometry signal, where t is measured by a microwave excitation experiment to be 4.3±0.3 GHz (Supplementary information C) [46]. At a finite magnetic field, which we label BS→T, the (1,1) T− HH state becomes the ground state at the zero detuning point. Since the Pauli principle forbids tunneling between the atoms in the T− HH state, the reflectometry becomes suppressed in this regime. Using the tunnel coupling t and the ther- mal occupation of the two-hole states (see Supplementary information E), this magnetic field dependence of the re- flectometry signal is simulated [40], as shown in Fig. 2c. BS→T is found to be 0.25±0.05 T, which provides an effective Land´e g-factor (g∗=gmj) of the T− HH state of 1.2±0.3, using the Zeeman energy of g∗µBB, where µB is the Bohr magneton. Noteworthy is that in bulk sili- HH state is expected to have mj=3 and g ∼ 1, con the T− which indicates that in our experiment the g∗ is strongly suppressed in this low magnetic field regime. Such a sit- uation occurs when the quantization axis of the angular momentum is not aligned with the applied magnetic field, dacbVBG [V]VTG [mV]Δφrefl [°]0.0-2.0(2,0)(2,1)(1,1)(1,0)ΔLHJLH(2,0) QLH (1,1) THH (2,0) SHH (1,1) SHH εEnergy2t= heavy hole= light hole(1,1)(2,0)ΔArefl [%]0.0-0.5-1.5ε1.92.0-779-778-777ε [μeV]Δφrefl [°]VTG [mV]-778.8-778.4-778.00.01.00.0-1.2-0.4-0.8Magnetic field [T]0.20.40.60.80.01.0Magnetic field [T]0.20.40.60.8-1.0-1.0040BS TBS Q-40experimenttheory04 FIG. 3. Spin relaxation mechanisms of the two-hole system. a, Time averaged rf reflectometry amplitude (red) and phase (blue) measurement with continuous pulses applied to the drain, at a magnetic field of 1.5 T. The inter-acceptor tunneling is − HH ground state. b, Relaxation displaced from the break of the acceptor-lead tunneling signal by the Zeeman-energy of the T time extracted from the time dependent phase signal for a pulse between the points 0 and 1. c, Magnetic field dependence −1 of the relaxation rate T 1 . d, Schematic view of the relevant relaxation rates in this system at the zero detuning point (see inset). The data in c is fit to a combination of the ΓS/Q→T rate [green line] and ΓS→T rate [blue line] resulting in the black line. This is compared to a fit using the ΓS→Q rate [red line] instead of the ΓS→T rate. indicating that this quantization does not come from the magnetic field but instead from either local strain, elec- tric field, or a nearby interface. state. This ability to sense the two-hole spin state could be utilized in a readout scheme for a single hole spin state, in analogy to the scheme first demonstrated by Koppens et al. [47]. Relaxation mechanisms A pulsed measurement technique was used to measure the spin dynamics of the two-hole system. The reflec- tometry signal originating from the inter-acceptor tun- neling in the SHH state can be recovered at magnetic fields above BS→T by pulsing the system into the (2,0) region. Such a recovered signal is indicated by a white circle in Fig. 3a, where the time-averaged amplitude and phase response of the reflected signal are displayed for an applied voltage pulse of 0.8 mV to the drain for 40 out of every 200 ns and at a magnetic field of 1.5 T. This indicates a finite population of the SHH excited state at the zero detuning point, which is utilised to explore the spin-dynamics of the two-hole system using time resolved spectroscopy. Fig. 3b shows the time dependence of the phase response with a voltage pulse applied between the 0 and 1 points, as marked in Fig. 3a, and averaged over ∼ 106 pulse sequences. The observed exponentially de- caying signal corresponds to the relaxation of the SHH Fig. 3c shows the relaxation rate T−1 1 of the SHH state as a function of magnetic field. In the high magnetic field regime (>1.5 T), T−1 increases monotonically with mag- netic field. This T−1 1 1 behavior is well approximated by a power-law dependence on magnetic field, which is usually expected for spin-lattice relaxation processes [16, 28]. On the other hand, in the low magnetic field regime (<1.5 reaches a maximum at ∼0.8 T. This implies a T), T−1 different spin-dynamic mechanism, likely arising from the specific interactions within this spin-3/2 acceptor system. 1 From the magnetic field dependence of T−1 1 we de- termine what relaxation mechanisms are relevant in the probed double-acceptor system. Fig. 3d illustrates the magnetic field dependence of the two-hole spin eigenen- ergies at zero detuning. While the (1,1) T− HH level is the ground state at magnetic fields above BS→T, the (2,0) Q2− LH level also falls below the SHH state above another critical magnetic field BS→Q. The relaxation rates from the SHH state to the T− LH states, ΓS→T and ΓS→Q, are expressed by a power law dependence on mag- HH and Q2− bacdε0ε1B = 1.5 T T1 = 390 ± 40 nsVBG [V]1.92.02.1VTG [mV]-777-778-779-780Time [μs]1.00.01.50.5(2,0) QLH(1,1) THH -SHHMagnetic FieldEnergyMagnetic Field [T]1231.52.51012025T1-1 [μs-1]0.0Δ phase [º]1.0ε0ε1ΓS/Q TΓS QΓS TΓS Q (ħω)3ΓS/Q T (ħω)3ΓS/Q T + ΓS TBS TBS Q2 tSQεEB = 1.5 T Δφrefl [°]0-0.8ΔArefl [%]-0.4-0.8-0.40Relaxation mechanismsΓS T (ħω)5BS Q 2 -netic field: (cid:16)(cid:0)B − BS→T[Q] (cid:1)2 + (2tST[Q])2(cid:17)γST[Q]/2 1 HH [(2,0) Q2− ΓS→T[Q] = AST[Q] (1) where AST[Q] is the relaxation amplitude for the S → T[Q] process and tST[Q] is the tunnel coupling between the SHH and the (1,1) T− LH] state. While tST is expected to be negligible, the spin-orbit interaction for light and heavy holes is expected to give a sizable tSQ. Theoretical reports predict different magnetic field power laws for these relaxation process with γST=5 (Kramers degenerate) and γSQ=3 (non-Kramers degenerate) [16, 18]. Both these relaxation rates monotonically increase above ∼1.5 with magnetic field like the observed T−1 T. However, this dataset is insufficient to conclude which relaxation process is dominant in this high magnetic field range and the best fit with both the ΓS→T and ΓS→Q are shown in Fig. 3c (Supplementary information F). The non-monotonic magnetic field dependence of T −1 in the low magnetic field regime indicates the existence of another relaxation mechanism. The heavy-light hole mixing plays an important role in the spin relaxation around the resonance between the SHH and (2,0) Q2− LH state. In the vicinity of BS→Q, the finite tSQ hybridises the SHH and (2,0) Q2− LH states, resulting in new spin eigen- states, both containing a finite light hole component. This light hole character opens a light-to-heavy hole re- laxation path to the (1,1) T− HH state, which is expected to be much faster than the heavy-to-heavy hole relax- ation process between SHH and T− HH [16, 18]. Since this relaxation mechanism strongly depends on the light-hole component in the hybridised states (see Supplementary information G), the relevant transition rate ΓS/Q→T can be written as: 1 (cid:16) SQ (B − BS→T)3 2t2 (B − BS→Q)2 + (2tSQ)2(cid:17) , (2) ΓS/Q→T = AQT where AQT is the relaxation amplitude between the Q2− LH and THH states, which, as mentioned, is expected to be much larger than AST. The non-monotonic magnetic field dependence of the Lorentzian peak and the large re- laxation amplitude AQT give a satisfactory explanation for the observed T−1 1 hotspot around 0.8 T. A combination of the monotonic rate ΓS→T and non- monotonic rate ΓS/Q→T is used to fit the full magnetic field dependence of T−1 1 , as shown by a solid black line in Fig. 3c. In this fit BS→Q is found at 0.70±0.02 T, tSQ is found in magnetic field units to be 0.04±0.015 T and we extracted a ratio between the relaxation amplitudes AQT/AST ∼ 1.2 × 103. The here observed spin-relaxation hotspot can be used as a rapid initialization of the spin-state of the accep- tor atom, where the measured ratio AQT/AST of more than three orders of magnitude demonstrates the excel- lent tunability of the acceptor system. Similar hotspots, caused by the mixing of spin and orbital degrees of free- dom, have been predicted and measured for systems with 5 a strong spin-orbit coupling in III-V semiconductors [48– 50]. Furthermore, tSQ is expected to mainly originate from the spin-orbit coupling of the heavy and light hole states. Therefore, the strong mixing between the SHH and Q2− LH states, as observed in our experiment, points to the presence of a strong spin-orbit coupling for the acceptor states and thereby shows the potential to use them as electrically drivable spin-orbit qubits. Two-hole excited state spectrum To understand the environment of the acceptor atoms we investigate the two-hole excited state spectrum. Due to their fast relaxation rate, these excited states are not accessible by the reflectometry technique and instead we use single hole transport spectroscopy [20, 34, 35] to probe the two-hole energy spectrum. In Fig. 4a the transconductance signal as a function of VTG and VBG at zero magnetic field shows a bias triangle (contour marked by a solid line), demonstrating sequential hole tunneling via both acceptors in series. Lines parallel to the trian- gle baseline (arrows in Fig. 4a) are attributed to transi- tions from the (1,1) ground state to (2,0) excited states (Supplementary information D). The observed (2,0) spec- trum matches the expected energy spectrum of a doubly charged acceptor state, shown in Fig. 4b. This supports our identification of the second confinement site as an acceptor atom, which is further validated since our mea- surement is not compatible with the regular spaced ex- cited state spectrum expected for quantum dots. The (2,0) ∆LH and JLH are extracted as 110±10 µeV and 36±5 µeV respectively. The measured ∆LH reflects the effect of local fields on the symmetry of the acceptor states. However, the direction of the resulting quantization axis for the angu- lar momenta is unknown, which makes it impossible to predict the exact Zeeman interaction for low magnetic fields [27, 28]. The extracted JLH reflects the mixing between light and heavy holes, associated with the spin- orbit coupling. We can compare this to tSQ, taking into account that this tunnel coupling carries an additional component, as the SHH state at zero detuning is in a superposition of (1,1) and (2,0) states. Converting tSQ to an energy scale using 1.5 (cid:46) g∗ (cid:46) 3, gives tSQ in the range of 3 to 7 µeV. Although somewhat lower than JLH, likely caused by the rotation of the quantization axis at this magnetic field as discussed later, both these param- eters show a strong spin-orbit coupling, which is crucial to envision electrical manipulations of the acceptor spin states. Next we discuss the magnetic field dependence of the transition energy spectrum, as shown in Fig. 4c, which is measured on a cut through the bias triangle at VBG=1.4 V (dotted line in Fig. 4a). The baseline of the trian- gle, indicated with a solid line in Fig. 4c, changes its slope twice. The first change around 0.25 T indicates the singlet-triplet crossing at BS→T, in agreement with 6 FIG. 4. Excited states of the two-hole system measured by transport spectroscopy. a, Transconductance dI/dVT G as a function of the top and back gate voltages at a 10 mV VD. The bias triangle, marked by solid black lines, shows three excited states in addition to the ground state as indicated by arrows. b, The level configuration of the A+ state, corresponding to the found excited states. c, Magnetic field dependence of the transconductance at a cut through the bias traingle at 1.4 V VBG, as indicated by the dotted line in a. At the bottom arrows indicate the excited state spectrum as found in a. Two changes in the slope of the ground state at 0.25 and 0.5 T are the result of the BS→T and BS→Q crossings. d, Model of the transition energies in the high magnetic field limit. Top panel: Overview of the the (1,1)→(2,0) transitions at a magnetic field of 1.3 T. Bottom panel: Simulated magnetic field behaviour of the transition energy spectrum, using the g-factors and ∆LH extracted from the transition energies in c and using the same line broadening as found in this experiment. the observation of spin blockade in Fig. 2b. The sec- ond change around 0.50 T indicates the change of the (2,0) ground state from SHH to Q2− LH, corresponding to the same anti-crossing that generates the relaxation hotspot in Fig. 3b, but here occurring in the negative detuning region and therefore at a lower magnetic field. In the high magnetic field regime (>1 T) we observe well established resonance lines. The number of resolv- able features is larger than at zero magnetic field, in- dicating spin split excited states. In this regime the Zeeman energy dominates over ∆LH and therefore the quantization axis of the angular momenta of the two-hole states align with the external magnetic field. Ignoring any mixing terms other than the SHH mixing, the mag- netic field dependence of the transition energies can be easily described by using mj = ±3/2 for heavy holes and mj = ±1/2 for light holes to calculate the Zeeman shifts, as shown in Fig. 4d. Using this model, the g-factors and ∆LH can be extracted by fitting the slopes and position of the transition lines in Fig. 4c respectively. For the doubly occupied acceptor we find g3/2 of 0.85 and g1/2 of 1.07 and ∆LH of 30 µeV. For the single occupied acceptor only the heavy hole states play a role in the transitions for which we extract g3/2 of 1.05. Our observations allow for the characterization of the acceptor system in both the low and high magnetic field regimes, which shows the extent of tuning of the acceptor states. Discussion The detected level configurations indicate a symmetry transition of the acceptor spin-3/2 system as a function of magnetic field. The g-factors found in the high mag- netic field regime are close to those found for bulk boron acceptors [51, 52], which is in stark contrast to the ef- fective g-factor found at low magnetic fields in the spin blockade experiment of Fig. 2b. Furthermore the ∆LH of the doubly occupied acceptor is about four times smaller in the high magnetic field regime than at zero magnetic field. These changes demonstrate that the quantization- axis in the high magnetic field regime differs from the quantization axis at zero magnetic field, thereby strongly influencing the g-factor and heavy-light hole splitting. A bacd-0.010.010VBG [V]1.01.21.41.61.82.0VD = 10 mV-774-768-772-770VTG [mV]dI/dVTG [μS]Energy(2,0) A+ statesSLLSHHQLHJLHΔLHΔLHVTG [mV]Magnetic Field [T]0.03.01.02.00.010dI/dVTG [μS]VD = 10 mV-771-769BS T-0.01Magnetic Field [T]0.03.01.02.0detuning [meV]Energy [meV]B = 1.3 T0.0-0.1-0.2-0.3BS Q1.30-0.10-0.050.05-0.15-0.20-0.40.25 T0.50 TSHHQLHQLHQLHQLHSLL-2 -+2 +SHHTHHTHH-0further indication of the rotating quantization axis is the decaying amplitude of the (1,1) T− HH to (2,0) SHH tran- sition, which completely disappears at a magnetic field around 2 T. The observation of the reduced mixing be- tween these states is well explained by the enhancement of the spin-blockade effect when the angular momentum becomes quantized along the magnetic field. The here demonstrated possibility to change the quantization axis and strongly influence the g-factor, heavy-light hole split- ting and relaxation rate of the acceptor states by control- ling the atoms environment with an external field, is a fundamental requirement for reaching the optimal condi- tions for acceptor-based qubits [17, 18]. In conclusion, the interactions between heavy and light hole states of individual acceptor atoms in silicon have been analysed, establishing the evaluation of the effect of local fields on the wave function symmetry and thereby on the level configurations and spin-dynamics of accep- tor atoms. Mixing between light and heavy holes is found to strongly influence the spin-dynamics of the two-hole system, leading to a hotspot in the spin-relaxation rate. Furthermore it is shown that the quantization axis of the holes can be rotated by applying a magnetic field, thereby significantly tuning the g-factor and heavy-light hole splitting. We have shown access to and control over the heavy and light hole states of individual acceptors and found evidence of a strong spin-orbit coupling in this system, thereby demonstrating the fundamental princi- ples to implement acceptor atom spin-orbit qubits. SUPPLEMENTARY INFORMATION A. Experimental setup The silicon tri-gate transistor used in this work is fab- ricated using standard CMOS technology, similar to pre- viously reported devices [35, 39]. The channel, defined by the top gate, is 11 nm high, 42 nm wide and 54 nm long. On both sides of this gate, 25 nm long Si3N4 spacers form barriers to the highly doped source and drain re- gions. With a background boron doping of 5 × 1017/cm3 in the silicon channel, about 10 boron atoms are expected under the gate and about 10 under the spacers. In a dilution refrigerator at a base temperature of 40 mK, hole transport and rf gate reflectometry measure- ments give complementary insights in the static and dy- namic behaviour of the two-hole system of interest. The hole temperature was measured from quantum-dot trans- port at ∼ 300 mK. The noise floor of the transport mea- surement (∼100 fA) allows us to detect currents with a total tunnel rate down to 500 kHz. A surface mount 270 nH inductor, attached to the top gate, forms a tank cir- cuit together with a parasitic capacitance of ∼0.3 pF. Tunnel events in the transistor with a tunnel rate higher than or equal to the resonant frequency of 583 MHz re- sult in a change of the reflected amplitude and/or phase of this resonator. Furthermore, low intensity light was 7 directed to the silicon chip through an optical fiber to enable the use of the lower doped substrate as a back gate. B. Location of the acceptors In order to estimate the location of the two atoms, we have extracted the relative capacitive coupling to the different electrodes. The hole temperature of ∼300 mK gives kbT (cid:29) Γ for the acceptor-lead tunnel rate, thereby justifying a thermally broadened fit [40]. The reflectom- etry line width shows a relative coupling to the top gate (αTG) and drain (αD) of 0.50±0.01 eV/V and 0.41±0.02 eV/V respectively, thereby placing this atom roughly halfway between these two electrodes, under the Si3N4 spacer. This relatively weak coupling to the top gate supports our identification of the acceptor atom, as gate- induced quantum dots are known to have gate couplings above ∼0.7 eV/V [39, 42]. The second boron atom has to be located closer to the centre of the channel, as the tunnelling to both electrodes is too slow to be detected. C. Estimation of tunnel rates In the rf gate reflectometry experiments shown in Figs. 1, 2, and 3, a driving frequency of 583 MHz is used, which is close to the resonant frequency and ideal to distinguish the acceptor-lead and inter-acceptor tran- sitions. The full frequency response of the reflected signal from the LC-circuit reveals more information about the tunnel rates involved in the experiment [53]. Frequency domain measurements of the amplitude and phase re- sponse around the resonance of the LC circuit are shown in Fig. 5. These measurements show that the acceptor- lead signal mainly changes the amplitude of the reflected signal, while the inter-acceptor signal mainly reduces the resonant frequency. This difference in response of the LC-circuit can be ex- plained by a difference in tunnel rate between these two processes [40]. When the tunnel rate is much faster than the resonant frequency (Γ (cid:29) ω), the tunnel process will predominantly add capacitance to the circuit (dispersive response), which in turn shifts the resonance to a lower frequency. This is observed for the inter-acceptor tunnel- ing process, thereby revealing a tunnel rate much faster than the resonant frequency. When the tunnel rate is of the order of the resonant frequency (Γ ∼ ω), the tunnel process will predominantly add resistance to the circuit (dissipative response). This added resistance can either bring the circuit closer or further away from its ideal matching condition, depending on the initial impedance of the circuit, thereby either reducing or increasing the re- flection amplitude. A reduction in reflection amplitude is observed for the acceptor-lead tunneling process, thereby revealing a tunnel rate of the order of the resonant fre- quency. 8 FIG. 6. Inter-accpetor continuous-wave microwave excitation experiment. The change in phase shift of the inter-acceptor tunnel signal as a function of microwave frequency, using a constant microwave power applied to the drain electrode of the transistor. The resonant dip reveals the tunnel coupling between the atoms measured as t = 4.3±0.3 GHz. The in- set shows a schematic representation depicting the excitation from the bonding to the anti-bonding singlet state with mi- crowaves. opposite response of the resonator [46], the observed ∆φrefl is expected to decrease when the occupation in the anti-bonding state is increased and completely disappear when the occupations in the system become unpolarised. In order to keep the input power of the applied mi- crowaves constant during the experiment, the splitting of the acceptor-drain signal is also probed during the ex- periment. The splitting of this signal has no dependence on the microwave frequency and is solely dependent on the microwave power applied at the drain. Using this measurement we can calibrate for any frequency depen- dent losses in the system and use the same input power for every frequency. The response of the inter-acceptor tunnel signal is shown in Fig. 6 as ∆φrefl/∆φ0, with ∆φ0 being the signal without microwaves. The resonance dips at a frequency of the 8.6±0.6 GHz, which demonstrates a tunnel coupling between the atoms of 4.3±0.3 GHz. The width of this resonance indicates a T ∗ 2 for these charge states of ∼ 200 ps, similar to what was found for a two- electron system on a hybrid donor-quantum dot [46]. D. Two-hole states In bulk silicon the single hole ground state of an ac- ceptor atom is fourfold degenerate, consisting of heavy hole and light hole states. The effect of strain, electric field and magnetic field on these states is well described by Bir et al. [27, 28]. Strain and electric field split the FIG. 5. Frequency domain analysis of the reflectometry sig- nal. a, (b,) Amplitude (phase) response of the tank circuit as a function of frequency. c, (d,) Amplitude (phase) shift relative to the off-resonant signal as a function of frequency on a line cut through the (1,1)-(2,0) transition, with VBG = 1.93 V and VT G = -777 mV. e, (f ) Amplitude (phase) shift for the linecuts on the red (inter-acceptor singal) and blue (acceptor-lead singal) dotted lines in c (d). To find the tunnel coupling between the two accep- tor atoms, we probe the resonance between the bonding and anti-bonding singlet SHH states, as shown in the in- set of Fig. 6, in accordance with a measurement on two electrons on a hybrid donor-quantum dot system [46]. Here we make use of the bias-tee coupled to the drain lead, as shown in Fig. 1a. The inter-acceptor signal is probed with rf gate reflectometry, while continuous wave microwaves are applied to the drain electrode. When the transition between the bonding and anti-bonding SHH states is excited by the applied microwaves, the occu- pation of the bonding state is lowered, while the occu- pation of the anti-bonding state is increased. As the bonding and anti-bonding singlet states give rise to an 01-1ΔA [a.u.]-2ΔΦ [º]02-2-4ecbadfAmplitude [a.u]Phase [º]080-80Frequency [MHz]583585584582Frequency [MHz]575590Voltage drain [mV]Voltage drain [mV]Frequency [MHz]5755900-0.8-0.40-0.8-0.4Δ Phase [º]Δ Amplitude [a.u.]Frequency [MHz]58058559057520-2-40Frequency [MHz]583585584582Frequency [MHz]580585590575off-resonanceinter-acceptoracceptor-leadoff-resonanceinter-acceptoracceptor-leadinter-acceptorinter-acceptoracceptor-leadacceptor-leadFrequency [GHz]1002030 Δφrefl / Δφ00.20.40.60.81.00.0 Energyε2tμ-wavest = 4.3 ± 0.3 GHzheavy and light hole manifolds, leading to a splitting of ∆LH at zero magnetic field. A magnetic field splits all four spin-states, where this Zeeman effect strongly de- pends on the local environment of the atom. The two atoms probed in the experiments are found to have heavy hole ground states. The ∆LH is assumed to be different for the two acceptors, as they experience a different lo- cal environment. As an example, the Zeeman effect is calculated for two different local environments where the heavy holes form the ground state at zero magnetic field, shown in Fig. 7. For the first example we consider the sit- uation where a compressive strain of 1.2×10−4 is present parallel to the magnetic field and an electric field of 0.6 MV/m is present perpendicular to the magnetic field, see FIG. 7. Examples of the Zeeman effect on A0-states. a, Mag- netic field dependence of the energy levels of a hole bound to a single boron acceptor in silicon with a compressive strain of 1.2×10−4 parallel with the magnetic field and an electric field of 0.6 MV/m perpendicular to the magnetic field. b, Mag- netic field dependence of the energy levels of a hole bound to a single boron acceptor in silicon with a compressive strain of 0.6×10−4 parallel with the magnetic field and an electric field of 1.2 MV/m perpendicular to the magnetic field. Dashed lines show the linear projection of the high magnetic field Zeeman effect, showing a change in ∆LH in b. 9 Fig. 7a. In Fig. 7b the situation with half of the strain and double the electric field is considered. In this second example, we note that if a linear approximation is taken for the hole-spin states in the high magnetic field region and traced back to zero magnetic field (dashed lines), a different heavy-light hole splitting is found than at zero magnetic field. This effect, caused by a rotation of the quantization axis of the angular momentum of the hole, has been observed in our experiment presented in Fig. 4. For the doubly occupied and positively charged accep- tor state (A+) the exchange energy between two holes from the same manifold far exceeds the heavy-light hole splitting (JHH, JLL (cid:29) ∆LH). Due to this exchange in- teraction, only 6 out of the 16 possible two-hole states can be bound to a boron atom in silicon [54]. A simi- lar sixfold manifold is found for two-holes bound to two closely placed boron atoms [33] and the neutral states of two holes bound to group II acceptors [45, 55]. As these six states are build from combinations of different spin states, one heavy hole singlet state, one light hole singlet state and four light/heavy-quadruplet states are found, which we will write as: √ SHH(cid:105) = 1/ 2 (⇑⇓(cid:105)− ⇓⇑(cid:105)) , √ 2 (↑↓(cid:105)− ↓↑(cid:105)) , SLL(cid:105) = 1/ √ LH(cid:105) = 1/ 2 (⇓↓(cid:105)− ↓⇓(cid:105)) , Q2− √ LH(cid:105) = 1/ Q− 2 (⇓↑(cid:105)− ↑⇓(cid:105)) , √ LH(cid:105) = 1/ 2 (⇑↓(cid:105)− ↓⇑(cid:105)) , Q+ √ LH(cid:105) = 1/ Q2+ 2 (⇑↑(cid:105)− ↑⇑(cid:105)) , (3) (4) (5) (6) (7) (8) where ⇑,⇓ represent the heavy hole states and ↑,↓ rep- resent the light hole states. Furthermore, when an ex- change interaction between the heavy and light holes is introduced, JLH, the QLH state splits into a singlet and triplet state [18]. In Fig. 8a an example of the mag- netic field dependence of the A+-states of an acceptor is shown, using the A0 states from Fig. 7b and neglecting any exchange between the light and heavy holes (JLH = 0). The colors in the figure show the heavy-hole (red) and light-hole (blue) character of these states, where the states built from one heavy and one light hole turn pur- ple. The splitting between the heavy hole singlet and the light hole singlet (ESLL - ESHH = 2∆LH) is found to be different in the high magnetic field limit compared to the zero magnetic field splitting, in line with was found in Fig. 7b. In the situation where each hole is bound to a different acceptor, the two hole state can be described as a product state of the single particle states, leading to an energy spectrum of 16 states divided into 4 fourfold degenerate manifolds at zero magnetic field. Each fourfold manifold splits into a singlet and a triplet state when an exchange interaction is introduced. In our experiments the heavy- hole manifold has the lowest energy at zero magnetic field ba Energy [meV]Bz [T]0-0.50.5 Energy [meV]0-0.50.50123Bz [T]0123εxx = 1.2 x 10-4 Ez = 0.6 MV/mεxx = 0.6 x 10-4 Ez = 1.2 MV/m10 the ground state manifold, the ground state in the (1,1) region remains the T− HH state for any positive magnetic field. In the transport measurements shown in Fig.4 the (1,1)→(2,0) transition energies at zero magnetic field and as a function of magnetic fields up to 3 T are probed. In these experiments, a voltage of +10 mV is applied to the drain, while the current is measured at the source. This bias direction gives a transport direction for the holes from the drain to the source. From the rf gate reflectom- etry experiments, as shown in Fig. 2, it is deduced that the acceptor with a charge transition from 0 to 1 hole has a stronger tunnel coupling to the drain (detectable with rf reflectometry) than the acceptor with a charge transition from 1 to 2 holes (not detectable with rf re- flectometry). It follows that the probed sequential tun- nelling involves the (1,1)→(2,0) transition, where the two observed bias triangles are formed by the two following hole charge transfer processes: (1, 0) → (1, 1) → (2, 0) → (1, 0), (2, 1) → (1, 1) → (2, 0) → (2, 1). (13) (14) The current measured within the bias triangle region is around 5 pA at the baseline at zero magnetic field and de- clining to the measurement noise floor of ∼ 100 fA at the other side of the triangle, indicating a total tunnel rate of the sequential tunnel process around 30 MHz and lower. We recall that the resonant tunnel-rate between the (1,1) and (2,0) SHH states was measured at 4.3 GHz and the tunnel-rate from the drain to the acceptor was found to be around the resonant frequency of 583 MHz, indicat- ing that the tunnel-rate between the second acceptor and the source is the limiting tunnel process. However, we note that due to off-resonant tunneling or tunneling be- tween two states consisting of different spins, the inter- acceptor tunnel rate can be strongly suppressed, thereby possibly becoming the limiting tunnel process. Further- more, the ratio between this inter-acceptor tunnel rate and the relaxation rate of each excited state in the (1,1)- configuration determines if this excited state will have any effect on the measured current. The transport mea- surement presented in Fig. 4 can be explained by only taking the heavy-hole manifold of the (1,1)-configuration into account. This indicates that any state which in- cludes any light-hole has a fast relaxation to the heavy hole ground state, which matches with the fast light-to- heavy hole relaxation process observed as a relaxation hotspot in Fig. 3. E. Pauli spin blockade in rf gate reflectometry The shift in phase response of the reflected signal as shown in Fig. 2b is ascribed to the addition of a quantum capacitance (CQ) to the LC-circuit [36–40] when holes can tunnel between the two atoms, given by: TGq24t2(cid:0)2 + 2t2(cid:1)−3/2 CQ = 1 2 ∆α2 , (15) FIG. 8. Examples of the Zeeman effect on two-hole states. The light or heavy hole character of each state is represented by the line color, where red represents heavy hole character and blue represents light hole character. a, A+ energies for the same local environment as in Fig. 7b and JLH = 0. A clear difference in splitting between the SHH and SLL states in the low and high magnetic field limit is visible. b, The (1,1)-configuration product states of the two single hole states described in Fig.7a and 7b. and can be described as: √ 2 (⇑,⇓(cid:105)− ⇓,⇑(cid:105)) , SHH(cid:105) = 1/ HH(cid:105) =⇓,⇓(cid:105), T− √ HH(cid:105) = 1/ T0 HH(cid:105) =⇑,⇑(cid:105). T+ 2 (⇑,⇓(cid:105)+ ⇓,⇑(cid:105)) , (9) (10) (11) (12) The energy splitting to the next two manifolds is given by the ∆LH of each acceptor, where these two-hole states consist of one heavy and one light hole state. Finally, the light-hole manifold is split from the heavy-hole man- ifold by the sum of both ∆LH's. In Fig. 8b an example of such a (1,1) level configuration is shown for a com- bination of the A0-states presented in Fig. 7a and 7b. Although many level crossings are observed, we highlight the fact that in this situation where the heavy holes form ba Energy [meV]Bz [T]0-0.80.8 Energy [meV]0-0.80.80123Bz [T]0123εxx = 0.6 x 10-4 Ez = 1.2 MV/m-0.40.4-0.40.4(2,0) A+ states(1,1) statesQLH2-QLH-QLH+QLHSHHSLL2+11 where ∆αTG is the difference in capacitive coupling from the top gate to each acceptor, found to be 0.14 eV/V in the transport measurements, q is the electron charge, is the energy detuning between the two acceptors, and t is the tunnel coupling between the (1,1) and (2,0) SHH states. All other tunnel couplings between (1,1) and (2,0) states are considered to be much smaller and therefore not contributing to a change in the detected reflected signal. The total phase shift (∆φ) is determined by the difference in occupation, given by the Boltzmann distri- bution, of the lower and upper branch of the heavy hole singlet state: e(−ES− /kT) − e(−ES+ /kT) (cid:80) e(−Ei/kT) ∆φ = CQ , (16) where ES− and ES+ are the energies of the bonding and anti-bonding singlet states respectively and Ei describes the energies of all the involved states. Using Eq. 15 and 16, the magnetic field dependence of the phase shift caused by the inter-acceptor tunneling is calculated as shown in Fig. 2c. We note that although the crossing of the QLH state and the SHH state has some influence on the Boltzmann distribution of the SHH states, this influ- ence is not distinguishable in the measurement. However, for completeness, we have displayed the BS→Q crossing point in Fig. 2c, as found in the relaxation-hotspot mea- surement together with an estimated g-factor of the QLH state. F. Relaxation mechanisms At magnetic fields much higher than the BS→Q mix- ing point the dominant relaxation mechanism comes from either the relaxation to the (1,1) T− HH state or the relax- ation to the (2,0) Q2− LH state, given by Eq. 1. In Fig. 9a the fits of the relaxation rate data with a combination of Eqs. 1 and 2 is given for the relaxation process to the T− HH state (blue) and the Q2− LH state (red) are shown. Our data is insufficient to determine if a (ω)3 or (ω)5 re- laxation mechanism is dominant at high magnetic fields. The relaxation amplitudes extracted from these two fits, which are interpreted as the relaxation time at a field of 1 T relative to the corresponding level-crossing, is found to be 0.5 µs for ASQ and 5 µs for AST. Both of these values are in good agreement with the theoretical predictions for light-heavy hole relaxation (T1 = 0.14 µs at 1T [16]) and heavy-heavy hole relaxation (T1=40 µs at 0.5T [17]). Fig. 9b shows the extracted relaxation rates on a larger scale of 1/T1, demonstrating the more than 3 orders of magnitude increase in relaxation rate at the hotspot. Furthermore, it reveals that if this hotspot behavior can be moved to a higher magnetic field, by increasing the heavy-light hole splitting or decreasing the g-factor, re- laxation times longer than 1 ms could be achieved at magnetic fields around 0.5T. FIG. 9. Relaxation rate models. a, Comparison between the fit of the relaxation rate data with a model combining the − HH re- hotspot-relaxation mechanism with either the SHH to T laxtion (blue) or with the SHH to Q2− LH relaxation (red). Both fits have a comparable value for R2. b, Comparison of the modeled relaxation mechanisms on a larger scale of magnetic field and T1. baMagnetic Field [T]1231.52.510120251/T1 [μs-1]BS QMagnetic Field [T]1/T1 [μs-1]120.51010010.10.010.001R20.840.83ΓS QΓS/Q T + ΓS TRelaxation mechanismsΓS Q (ħω)3ΓS/Q T (ħω)3Relaxation mechanismsΓS T (ħω)5ΓS/Q T +ΓS QΓS/Q T + ΓS TΓS/Q T +G. Relaxation hotspot hole character at the zero energy point as: The observed relaxation hotspot is explained by the mixing of the bonding SHH state and the (2,0) Q2− LH state, which allows a light-to-heavy hole relaxation to the (1,1) T− HH state. This relaxation process is expected to have a (ω)3 dependence and furthermore depends linearly on the light hole character of the mixed state. Here we show the calculation of the light hole character of the probed state. Around the anti-crossing we can write the energies of the lower (-) and upper (+) states as: (cid:113) E± = ∆E 2 ± 1 2 (∆E)2 + (2tSQ)2, (17) where ∆E is the energy splitting between the states in the absence of mixing (EQ2− - ESHH), ESHH is taken as the reference energy (E = 0). We define the amount of light hole character of each state as the chance to measure it as the QLH state in the basis of QLH and SHH, written as ξ± and given by: LH ξ± = ∓ 1 2 (cid:113) 2 ∆E (∆E)2 + (2tSQ)2 . (18) (cid:17) E± The relaxation measurement is performed at zero detun- ing, which is the original position of ESHH in the ab- sence of mixing. The state probed here can be writ- ten as a combination of the + and - states, depending on their relative energy difference with the measurement . We note that E−ξ− = E+ξ+ = point E−+E+ (∆E)2 + (2tSQ)2 and also that (E− + E+) is (∆E)2 + (2tSQ)2, which gives the total light (cid:16) (cid:113) (cid:113) equal to t2 SQ/ 12 (19) ξE=0 = E−ξ− + E+ξ+ E− + E+ = as used in Eq. 2. 2t2 SQ (∆E)2 + (2tSQ)2 , H. Acceptor-based spin-orbit qubits The key properties of acceptor-based spin-orbit qubits, such as the dipole coupling and decoherence time between the chosen qubit states, are predicted to strongly depend on changes to the symmetry of the silicon crystal and the confinement of the hole [16–19, 27, 28]. Unstrained silicon ensemble measurements have established an elec- tric dipole coupling of 0.26 Debye [51] and relaxation and coherence times in the micro-second regime [29]. These acceptor systems are promising for fast electrically-driven spin-orbit qubits with a strong coupling to cavities and phonons [16], although a longer coherence time would be desirable. An optimal regime, where the coherence and relaxation times are prolonged while the strong electric dipole coupling is maintained, has been predicted in the small strain regime, giving a small splitting and strong mixing between the heavy and light hole states [18]. The possibility to lengthen the acceptor relaxation time has been demonstrated in experiments on intentionally heav- ily strained bulk samples, where T1 times up to 100 ms are measured [56, 57]. The access to and control over sin- gle atoms, as has been demonstrated here, is necessary to probe the physics in an environment with a small local strain, which is the crucial regime for the realisation of acceptor qubits. ACKNOWLEDGMENTS The device has been designed and fabricated by the TOLOP Project partners, see http://www.tolop.eu. This work was supported by the ARC Centre of Ex- cellence for Quantum Computation and Communica- tion Technology (CE110001027) and the ARC Discovery Project (DP120101825). [1] Koenraad, P. M. & Flatte, M. E. Single dopants in semi- conductors. Nat. Mater. 10, 91–100 (2011). [2] Zwanenburg, F. A. et al. Silicon quantum electronics. Rev. Mod. Phys. 85, 961–1019 (2013). [3] Kurtsiefer, C., Mayer, S., Zarda, P. & Weinfurter, H. Stable solid-state source of single photons. Phys. Rev. Lett. 85, 290–293 (2000). [4] Tetienne, J.-P. et al. Nanoscale imaging and control of domain-wall hopping with a nitrogen-vacancy center mi- croscope. Science 344, 1366–1369 (2014). [5] Bonato, C. et al. Optimized quantum sensing with a sin- gle electron spin using real-time adaptive measurements. Nat. Nanotechnol. 11, 247–252 (2016). [6] Muhonen, J. T. et al. Storing quantum information for 30 seconds in a nanoelectronic device. Nat. Nanotechnol. 9, 986–991 (2014). [7] Smolenski, T. et al. Magnetic ground state of an individ- ual Fe2+ ion in strained semiconductor nanostructure. Nat. Commun. 7, 10484 (2016). [8] Kane, B. E. A silicon-based nuclear spin quantum com- puter. Nature 393, 133–137 (1998). [9] Otte, A. F. et al. The role of magnetic anisotropy in the Kondo effect. Nat. Phys. 4, 847–850 (2008). [10] Toskovic, R. et al. Atomic spin-chain realization of a model for quantum criticality. Nat. Phys. 12, 656–660 (2016). [11] Kraus, H. et al. Room-temperature quantum microwave emitters based on spin defects in silicon carbide. Nat. Phys. 10, 157–162 (2014). [12] Widmann, M. et al. Coherent control of single spins in silicon carbide at room temperature. Nat. Mater. 14, 164–168 (2015). [13] Hu, Y., Kuemmeth, F., Lieber, C. M. & Marcus, C. M. Hole spin relaxation in Ge-Si core-shell nanowire qubits. Nat. Nanotechnol. 7, 47–50 (2012). [14] Prechtel, J. H. et al. Decoupling a hole spin qubit from the nuclear spins. Nat. Mater. 15, 981–986 (2016). [15] Maurand, R. et al. A CMOS silicon spin qubit. Nat. Commun. 7, 13575 (2016). [16] Ruskov, R. & Tahan, C. On-chip cavity quantum phon- odynamics with an acceptor qubit in silicon. Phys. Rev. B 88, 064308 (2013). [17] Salfi, J., Tong, M., Rogge, S. & Culcer, D. Quantum computing with acceptor spins in silicon. Nanotechnology 27, 244001 (2016). [18] Salfi, J., Mol, J. A., Culcer, D. & Rogge, S. Charge- insensitive single-atom spin-orbit qubit in silicon. Phys. Rev. Lett. 116, 246801 (2016). [19] Abadillo-Uriel, J. C. & Calder´on, M. J. Interface effects on acceptor qubits in silicon and germanium. Nanotech- nology 27, 024003 (2016). [20] van der Heijden, J. et al. Probing the spin states of a single acceptor atom. Nano Lett. 14, 1492–1496 (2014). [21] Nowack, K. C., Koppens, F. H. L., Nazarov, Y. V. & Vandersypen, L. M. K. Coherent control of a single elec- tron spin with electric fields. Science 318, 1430–1433 (2007). [22] 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). [23] Kawakami, E. et al. Electrical control of a long-lived spin qubit in a Si/SiGe quantum dot. Nat. Nanotechnol. 9, 666–670 (2014). [24] Petersson, K. D. et al. Circuit quantum electrodynamics with a spin qubit. Nature 490, 380–383 (2012). [25] Gustafsson, M. V. et al. Propagating phonons coupled to an artificial atom. Science 346, 207–211 (2014). [26] Viennot, J. J., Dartiailh, M. C., Cottet, A. & Kontos, T. Coherent coupling of a single spin to microwave cavity photons. Science 349, 408 (2015). [27] Bir, G., Butikov, E. & Pikus, G. Spin and combined res- onance on acceptor centres in Ge and Si type crystals-I: Paramagnetic resonance in strained and unstrained crys- tals. J. Phys. Chem. Solids 24, 1467–1474 (1963). [28] Bir, G., Butikov, E. & Pikus, G. Spin and combined reso- nance on acceptor centres in Ge and Si type crystals-II: The effect of the electrical field and relaxation time. J. Phys. Chem. Solids 24, 1475–1486 (1963). [29] Song, Y. P. & Golding, B. Manipulation and decoherence of acceptor states in silicon. Europhys. Lett. 95, 47004 (2011). [30] Calvet, L. E., Wheeler, R. G. & Reed, M. A. Observation of the linear Stark effect in a single acceptor in Si. Phys. Rev. Lett. 98, 096805 (2007). 13 [31] Calvet, L. E., Wheeler, R. G. & Reed, M. A. Effect of local strain on single acceptors in Si. Phys. Rev. B 76, 035319 (2007). [32] Mol, J. A. et al. Interface-induced heavy-hole/light-hole splitting of acceptors in silicon. Appl. Phys. Lett. 106, 203110 (2015). [33] Salfi, J. et al. Quantum simulation of the Hubbard model with dopant atoms in silicon. Nat. Commun. 7, 11342 (2016). [34] Li, R., Hudson, F. E., Dzurak, A. S. & Hamilton, A. R. Pauli spin blockade of heavy holes in a silicon double quantum dot. Nano Lett. 15, 7314–7318 (2015). [35] Voisin, B. et al. Electrical control of g-factor in a few- hole silicon nanowire MOSFET. Nano Lett. 16, 88–92 (2016). [36] Petersson, K. D. et al. Charge and spin state readout of a double quantum dot coupled to a resonator. Nano Lett. 10, 2789–2793 (2010). [37] Colless, J. I. et al. Dispersive readout of a few-electron double quantum dot with fast rf gate sensors. Phys. Rev. Lett. 110, 046805 (2013). [38] Verduijn, J., Vinet, M. & Rogge, S. Radio-frequency dispersive detection of donor atoms in a field-effect tran- sistor. Appl. Phys. Lett. 104, 102107 (2014). [39] Gonzalez-Zalba, M. F., Barraud, S., Ferguson, A. J. & Betz, A. C. Probing the limits of gate-based charge sens- ing. Nat. Commun. 6, 6084 (2015). [40] House, M. G. et al. Radio frequency measurements of tunnel couplings and singlet-triplet spin states in Si:P quantum dots. Nat. Commun. 6, 8848 (2015). [41] Johnson, A. C. et al. Triplet-singlet spin relaxation via nuclei in a double quantum dot. Nature 435, 925–928 (2005). [42] Voisin, B. et al. Few-electron edge-state quantum dots in a silicon nanowire field-effect transistor. Nano Lett. 14, 2094–2098 (2014). [43] Aleksandrov, V. N., Gershenzon, E. M., Mel'nikov, A. P., Rabinovich, R. I. & Serebryakova, N. A. Interaction of A+(D−) centers in semiconductors with charged and neu- tral impurities. JETP Lett. 22, 282–283 (1975). [44] Betz, A. C. et al. Dispersively detected Pauli spin- blockade in a silicon nanowire field-effect transistor. Nano Lett. 15, 4622–4627 (2015). [45] Kartheuser, E. & Rodriguez, S. Group-theoretical study of double acceptors in semiconductors under uniaxial stress. Phys. Rev. B 8, 1556–1570 (1973). [46] Urdampilleta, M. et al. Charge dynamics and spin block- ade in a hybrid double quantum dot in silicon. Phys. Rev. X 5, 031024 (2015). [47] Koppens, F. H. L. et al. Driven coherent oscillations of a single electron spin in a quantum dot. Nature 442, 766–771 (2006). [48] Bulaev, D. V. & Loss, D. Spin relaxation and decoherence of holes in quantum dots. Phys. Rev. Lett. 95, 076805 (2005). [49] Stano, P. & Fabian, J. Orbital and spin relaxation in sin- gle and coupled quantum dots. Phys. Rev. B 74, 045320 (2006). [50] Srinivasa, V., Nowack, K. C., Shafiei, M., Vandersypen, L. M. K. & Taylor, J. M. Simultaneous spin-charge re- laxation in double quantum dots. Phys. Rev. Lett. 110, 196803 (2013). [51] Kopf, A. & Lassmann, K. Linear Stark and nonlinear Zeeman coupling to the ground state of effective mass ac- ceptors in silicon. Phys. Rev. Lett. 69, 1580–1583 (1992). [52] Stegner, A. R. et al. Isotope effect on electron paramag- netic resonance of boron acceptors in silicon. Phys. Rev. B 82, 115213 (2010). [53] Hile, S. J. et al. Radio frequency reflectometry and charge sensing of a precision placed donor in silicon. Appl. Phys. Lett. 107, 093504 (2015). [54] Smit, G. D. J., Rogge, S., Caro, J. & Klapwijk, T. M. Group-theoretical analysis of double acceptors in a mag- netic field: Identification of the Si:B+ ground state. Phys. Rev. B 69, 085211 (2004). [55] Bhattacharjee, A. K. & Rodriguez, S. Group-theoretical study of the Zeeman effect of acceptors in silicon and germanium. Phys. Rev. B 6, 3836–3856 (1972). [56] Ludwig, G. & Woodbury, H. Solid-State Physics, vol. 13 (Academic Press, Inc. New York, 1962). [57] Dirksen, P., Henstra, A. & Wenckebach, W. T. An ESR hole burning study of dynamic nuclear polarisation of 29 Si in Si:B. J. Phys. Chem. Solids 1, 8535 (1989). 14
1512.07973	1	1512	2015-12-25T07:35:34	Study of morphology effects on magnetic interactions and band gap variations for 3$d$ late transition metal bi-doped ZnO nano structures by hybrid DFT calculations	[ "cond-mat.mes-hall" ]	Using density functional theory (DFT) based electronic structure calculations, the effects of morphology of semiconducting nano structures on the magnetic interaction between two magnetic dopant atoms as well as a possibility of tuning band gaps have been studied in case of the bi-doped (ZnO)$_{24}$ nano structures with the impurity dopant atoms of the 3$d$ late transition metals (TM) - Mn, Fe, Co, Ni and Cu. To explore the morphology effect, three different structures of the host (ZnO)$_{24}$ nano-system having different degrees of spatial confinement, have been considered : a two dimensional (2D) nanosheet, an one dimensional (1D) nanotube and a finite cage-shaped nanocluster. The present study employs hybrid density functional theory to accurately describe the electronic structure of all the systems. It is shown here that the magnetic coupling between the two dopant atoms, remains mostly anti-ferromagnetic in course of changing the morphology from the sheet geometry to the cage-shaped geometry of the host systems, except for the case of energetically most stable bi-Mn doping, which shows a transition from ferromagnetic to anti-ferromagnetic coupling with decreasing aspect ratio of the host system. The effect of the shape change, however, has a significant effect on the overall band gap variations of both the pristine as well as all the bi-doped systems, irrespective of the nature of the dopant atoms and provides a means for easy tunability of their optoelectronic properties.	cond-mat.mes-hall	cond-mat	Study of morphology effects on magnetic interactions and band gap variations for 3d late transition metal bi-doped ZnO nano structures by hybrid DFT calculations Soumendu Datta,1,∗ Gopi Chandra Kaphle,2 Sayan Baral1 and Abhijit Mookerjee1,†,‡ 1Department of Condensed Matter Physics and Material Sciences, S.N. Bose National Centre for Basic Sciences, JD Block, Sector-III, Salt Lake City, Kolkata 700 098, India 2 Central Department of Physics, Tribhuvan University, Kathmandu, Nepal (Dated: July 2, 2021) Using density functional theory (DFT) based electronic structure calculations, the effects of mor- phology of semiconducting nano structures on the magnetic interaction between two magnetic dopant atoms as well as a possibility of tuning band gaps have been studied in case of the bi-doped (ZnO)24 nano structures with the impurity dopant atoms of the 3d late transition metals (TM) - Mn, Fe, Co, Ni and Cu. To explore the morphology effect, three different structures of the host (ZnO)24 nano-system having different degrees of spatial confinement, have been considered : a two dimen- sional (2D) nanosheet, an one dimensional (1D) nanotube and a finite cage-shaped nanocluster. The present study employs hybrid density functional theory to accurately describe the electronic structure of all the systems. It is shown here that the magnetic coupling between the two dopant atoms, remains mostly anti-ferromagnetic in course of changing the morphology from the sheet ge- ometry to the cage-shaped geometry of the host systems, except for the case of energetically most stable bi-Mn doping, which shows a transition from ferromagnetic to anti-ferromagnetic coupling with decreasing aspect ratio of the host system. The effect of the shape change, however, has a significant effect on the overall band gap variations of both the pristine as well as all the bi-doped systems, irrespective of the nature of the dopant atoms and provides a means for easy tunability of their optoelectronic properties. PACS numbers: 36.40.Cg, 73.22.-f,71.15.Mb,75.30.Hx,75.50.Pp I. INTRODUCTION The II-VI compound semiconductor ZnO with a direct wide band gap of 3.37 eV and a large excitonic binding energy of 60 meV for the bulk hexagonal wurtzite crys- tal structure at the ambient temperature and pressure condition,[1] has continued to attract great attention till today, both experimentally and theoretically. The reason for this huge interest in ZnO materials is mainly due to its versatile properties in piezoelectric, optical, magneto- electronics, highly efficient blue light emitting diodes (LEDs) and microwave devices.[2] TM-atom doped ZnO was also predicted as one of the most promising materi- als for room temperature (RT) dilute magnetic semicon- ductor (DMS)[3 -- 5] to provide more new functionalities such as spin-based information storage, data processing, spin-polarized laser and so on. Therefore, the TM-atoms doped ZnO as well as other II-VI and III-V semicon- ductors have also been intensively explored for the spin- tronics applications.[6, 7] Another interesting fact is that ZnO is transparent to visible light and can be made elec- trically highly conductive too by the introduction of de- fects. Moreover, ZnO is bio-safe as well as biocompatible ∗Electronic mail: [email protected] †Visiting Faculty, Department of Physics, Lady Brabourne College, Kolkata ‡Visiting Distinguished Professor, Department of Physics, Presi- dency University, Kolkata and therefore, can be used for biomedical applications without coating.[8] On the other hand, one of the main limitations of ZnO for its practical applications, is that its band gap is too large to effectively use visible light. Therefore, it is important to have band gap reduction in ZnO. Being a wurtzite crystal structure, one striking feature of the bulk ZnO is the appearance of polar sur- faces during its growth along the c-axis. The hexagonal wurtzite crystal structure consists of a number of alter- nating planes composed of tetrahedrally coordinated O2− and Zn2+ ions, stacked along the c-axis. The oppositely charged ions produce positively charged (0001)-Zn and negatively charged (000¯1)-O polar surfaces, resulting in a normal dipole moment and spontaneous polarization along the growth direction. Nano structured ZnO has received particular attention in current research due to intriguing nanosized effects on their properties. The interesting fact about nano structured ZnO systems, is that they have a diverse group of growth morphologies, such as nanocombs,[9] nanohelixes/nanosprings,[11] nanotubes/nanorods,[10] nanobelts/nanoribbons/nanorings[12, and nanocages.[14] The morphology of the ZnO nanos- tructures is largely directed by the polarity and satu- rated vapor pressure of the solvents. Recently, doping concentration driven morphological evolution has also been reported for the Fe doped ZnO nanostructure.[15] Therefore, the morphology of the ZnO nanostructures can be varied from one shape to another by controlling the growth kinetics through the adjustment of the preparation method and preparation conditions.[16] 13] These various morphologies and matured growth meth- ods in addition with the intrinsic surface and quantum confinement effects of the nano structures, lead to easy preparation of various nano-ZnO based high-tech functional devices. For examples, ZnO nanostructures have been used widely in field-effect transistors,[17] light emitters,[18] lasers,[19] dye sensitized solar cells[20] and sensing.[21] Nanostructured DMSs have also been of current interest for the future spintronics applica- tions. The possibilities of RT ferromagnetism in ZnO nanostructures, have been realized by the dopings with atoms of TM elements, sp/d0 elements[22] or capping the surface of the host system with N or S containing ligands.[23] There have been many reports which indi- cate that surface effects and defects, however, play an important role in stabilizing ferromagnetic couplings in these systems.[24] In fact, the magnetism in the doped ZnO nanosystems is rather subtle and depending on the material preparation conditions, there have been considerable controversy in the reported experimental results of the RT ferromagnetism and its real origin.[25] It, therefore, clearly indicates the fact that the research field involving the ZnO nanostructures, in its infancy but nanostructured materials built up from such nanoparticles are expected to have exceptional proper- ties which are promising in microelectronic devices,[26] energy conversion and storage,[27] catalysis[28] and drag delivery[29]. For this reason, it is also essential to identify the exact role of the shape of the ZnO nano particles and to gain a precise understanding of the nature of bonding between the host systems and the added foreign impurities. is still Morphology control of semiconducting nano systems are very demanding for enhancing the light absorp- tion and shortening transfer distance of photo-generated carriers.[30, 31] Recently, photo-catalytic properties of various ZnO nano structures and the effects of aspect ratio due to the morphological changes, have also been studied.[32] In several recent experimental works, the morphology-controlled synthesis of ZnO nanostructures has been demonstrated in terms of the interplay of several associated effects like polar charges, surface area, elastic deformation and so on[1, 33]The variations of the dimen- sionality for the ZnO nanostructures, on top of their finite size effects, incur several unique properties. Therefore, comparison of their properties with respect to that of the well-studied bulk, surface and thin-film systems related to ZnO, would be useful as well as interesting. Theo- retical study on the magnetism and band gap variation of the ZnO nanosystems with shape changes is, however, very limited. One possible reason may be the absence of any well defined recipe to construct nano crystals with a defined aspect ratio, apart from a few exceptions.[34, 35] In the present work, we report our results on first prin- ciples electronic structure calculations to investigate the morphology effects of host system on the possibilities of engineering the magnetic couplings and band gap nar- rowing of the ZnO nano structures upon diluted substi- 2 tutional doping of the two 3d late TM atoms of same type at the two metal sites of the host systems. To explore the morphology effects of a (ZnO)24 nano system, we have considered its three structural forms of different shapes - a 2D mono-layer-type nanosheet, an 1D nanotube and one finite quasi-spherical cage-shaped nanocluster, which have the same number of atoms with stoichiometric com- position. The three structures of the host system are, rather, distinguished by different degrees of spatial con- finement. We study the substitutional bi-doping with each of the 3d late transition metal atoms Mn, Fe, Co, Ni and Cu at the metal sites of the (ZnO)24 nano system for all the three structures of the host system. We have employed sophisticated hybrid functional as prescribed by Heyd, Scuseria and Ernzerhof (HSE)[36] for accurate description of the electronic structure. Our study re- veals that the effect of shape change is dramatic in case of the band gap variation of both the pristine as well as the doped systems in going from one morphology to the others. The effect of this morphology change on the magnetic properties of the doped systems, is, however, comparatively minor as the magnetic coupling remains mostly anti-ferromagnetic irrespective of the morphology changes and the notable difference arises mainly in the nature of the spatial separation of the two dopant impu- rity atoms in cases of the most favorable bi-doped sys- tems. We believe, our results here will not only be useful to provide a basis to rationalize the diverse experimen- tal reports in the past on the TM-doped semiconductor nano particles, but may also guide the future experiments to opt a specific synthesis method for the desired opto- electronic properties. II. COMPUTATIONAL DETAILS The calculations reported in this study, were per- formed using DFT within the framework of pseudo po- tential plane wave method, as implemented in the Vienna abinitio Simulation Package (VASP).[37] 3p, 3d as well as 4s electrons for the transition metal atoms and 2s as well as 2p electrons for the oxygen atoms, were treated as valance electrons. The interaction between the valance electron and ion core has been described by the projected augmented wave (PAW) pseudo potential [38, 39] and exchange correlation energy density functional was con- sidered under the framework of generalized gradient ap- proximation (GGA) as formulated by Perdew, Burke and Ernzerhof (PBE).[40] Energy cut-off of plane wave expan- sion was set to 450 eV. The convergence of the energies with respect to the cut-off value were checked. Symme- try unrestricted geometry optimizations were performed using the conjugate gradient and the quasi-Newtonian methods until all the force components were less than a threshold value of 0.001 eV/A. The convergence of self consistency was achieved with a tolerance in total energy of 10−4 eV. It is well known that calculations with the GGA-PBE level of theory generally underestimates the 3 FIG. 1: (Color online) Three different morphologies of (ZnO)24 nano structures used in this work : (a) 2D extended sheet, (b)1D elongated tube and (c) cage-like ZnO cluster of finite dimension. Green colored larger balls correspond to Zn atoms and violet colored smaller balls correspond to O atoms. band gap and formation energy of transition metal ox- ide semiconductors.[41] To overcome these deficiencies, all electronic property calculations have been carried out using HSE hybrid density functional,[36] which mixes 25 % of the exact nonlocal exchange of Hartree-Fock the- ory with the semilocal PBE functional. The GGA-PBE relaxed structures were used as input structures for the HSE calculations. Each of the three structures, has been modeled with a periodic super-cell. A rectangular paral- lelepiped super-cell for the nanosheet geometry, a square parallelepiped super-cell for the nanotube geometry and a simple cubic super-cell for the nanocage geometry were used with periodic boundary condition. The vacuum size was set at 12 A between the system and its neighboring images, which essentially makes the interaction between them negligible. The vacuum separation was considered along the vertical direction for the nanosheet structures, along the cross-sectional direction for the nanotube struc- tures and along the all directions around each nanocage structure. The Brillouin zone has been sampled with a sufficient number of Γ-centered k-point sets generated based on the Monkhorst and Pack scheme.[42] Full re- laxation of the geometries have been done with k-point mesh of 4×4×1 for the nanosheet geometry, 4×1×1 for the nanotube geometry. For the nanocage geometry, the reciprocal space integrations were carried out at the Γ point. Spin-polarized calculations were performed for all the transition metal doped systems. III. RESULTS AND DISCUSSIONS A. Pure nano-structures Fig. 1 shows the optimized minimum energy structure (MES) for each of the three morphologies of the pristine (ZnO)24 system. The optimized structure of the (ZnO)24 nano-sheet having a length of 26 A and a width of 9.8 A consists of hexagonal (ZnO)3 building blocks with the in- finite periodicity all along the plane. The parallelogram shape for the planar sheet has been considered to have the provision of including both the near and far spatial separations between the two dopant atoms in the case of bi-dopings in this structure. Our calculated average Zn-O bond length for the optimal nanosheet is 1.88 A in accordance with the earlier result,[43] while its value for the bulk wurtzite ZnO is 1.98 A corresponding to its two lattice constants of a = 3.249 A and c = 5.207 A.[44] In the optimized nanosheet structure, the average values of the Zn-O-Zn and O-Zn-O angles remain at 120◦. The (ZnO)24 nanotube has a regular zigzag (6,0) form and again consists of (ZnO)3 hexagons. The optimized struc- ture of the nanotube has a length of around 11 A and diameter of 7 A. The average Zn-O bond-length of the optimized nanotube structure, is 1.91 A. The Zn-O-Zn bond angles along the axial direction are around 109.3◦- 109.5◦ and along the cross sectional direction are around 119.2◦-119.8◦. The corresponding values for the O-Zn- O bond angles are around 112.5◦-112.7◦ along the axial direction and around 133.2◦-133.6◦ along the cross sec- tional direction. The optimal structure of the fullerene- like (ZnO)24 nanocage, on the other hand, consists of sev- eral sections of (ZnO)4 octagons, (ZnO)3 hexagons and (ZnO)2 squares as reported in earlier works.[45, 46] The appearance of the octagon, hexagon and square shaped building blocks together in case of the optimized (ZnO)24 nanocage structures, arises in order that Zn-Zn or O-O direct bonds are not formed. The diameter of the optimal pristine cage-shaped structure is ≈ 11 A and the average value of Zn-O bond-lengths is 1.94 A, which is slightly larger than that of the pristine nanosheet as well as the nanotube structures. It is to be noted that recently, the planar structure of a flat honeycomb nanosheet[47] and formation of the smooth single-walled ZnO nanotube[48] as well as the cage-shaped structure for larger (ZnO)n clusters with n >8,[46] have also been predicted to be energetically more favorable than their reconstructed ge- ometries. The different geometrical structures of the three sys- tems, are primarily characterized by different aspect ra- tio (i.e length/width ratio) which is decreasing along nanosheet→nanotube→nanocage. It would be interest- ing to distinguish them in terms of their microscopic elec- tronic properties. Fig. 2 shows the s, p and d orbitals projected density of states (PDOS) summed over all the constituent atoms for the optimal structures of the three morphologies in case of the pristine (ZnO)24 nano sys- tem. It is seen that the top of the valence band (TVB) is mostly contributed by the Zn−d and O−p orbitals. On the other hand, the bottom of the conduction band (BCB) is mainly contributed by the O−p orbitals and also partially by the O−s and Zn−s orbitals. We note that the change in morphology affects the types of orbital hybridization for the constituent atoms of the (ZnO)24 system. We find out that the optimal structure for each of the three morphologies of the pristine (ZnO)24 nano structure, is characterized by the dominance of the differ- ent kinds of orbital hybridization. In order to understand the trend in the hybridization among the various orbitals of the constituents in the optimal structures, we have an- alyzed a hybridization index parameter as shown in the right panels of the Fig. 2. We define the hybridization w(I) i,k w(I) where k and l are the orbital indices - s, p and d, wI i,k (wI i,l) is the projection of i-th Kohn-Sham orbital onto the k (l) spherical harmonic centered at atom I, inte- grated over a sphere of specified radius. First of all, it is seen from the Fig. 2 that the p-d hybridization index is the most dominating for each of the three pristine struc- tures, with the maximum value for the optimal (ZnO)24 nanosheet structure. In addition, the optimal (ZnO)24 nanosheet is associated with the maximum value of the s-d hybridization index too, which favors a planar geom- etry. The (ZnO)24 nano-cage, on the other hand, is asso- ciated with the enhanced s-p hybridization index which shows a decreasing trend along cage→tube→sheet struc- tures. The optimized nanotube structure, however, al- ways adopts the intermediate value for each of the three hybridization index parameters. index parameter[49, 50] hkl as, hkl = occ(cid:80) 40(cid:80) I=1 i=1 ; i,l The optimized geometries of the three pristine (ZnO)24 nano structures are found to be nonmagnetic. Our calcu- lated band gaps [i.e. HOMO-LUMO gap] for the pristine systems, show a nice trend with the change of aspect ratio. The calculated band gaps using the HSE func- tional, are 3.39 eV, 3.0 eV and 2.22 eV for the opti- mal nanosheet, nanotube and the cage-shaped (ZnO)24 nano structures respectively. While the band gaps of the tube and cage structures are dependent on their respec- tive diameters,[51] our calculated band gap for the ZnO monolayer is in close agreement with the earlier result of 3.57 eV by GW calculations.[52] It is also important to note that our calculated band gaps are of direct type for all the three pristine structures, in accordance with the direct wide-band gap of 3.3 eV for the bulk wurtzite ZnO.[53] Interestingly, this fact of possessing same kinds of band gaps for the bulk as well as monolayer phases of the ZnO system, is in contrast with the cases of transi- 4 FIG. 2: (Color online)Plot of HSE method calculated s, p and d orbital projected density of states (for the nanosheet, nanotube and cage shaped structures in the left panels) and s- d, p-d as well as s-p hybridization indexes (from top to bottom in the right panels) corresponding to the MESs of the three pure (ZnO)24 nano structures. A smearing width of 0.1 eV is used. The energy along x-axis of the left panels, is shifted with respective to the Fermi energy of the respective system. tion metal dichalcogenides, namely MoX2, WX2 (X = S, Se, Te) which are indirect band gap semiconductors in bulk phase, whereas their monolayers have direct band gaps.[54, 55] The trend of the decreasing band gaps with the decreasing aspect ratios, as mentioned above for the (ZnO)24 nano systems, is, however, in accordance with the trend in band gap variation of another II-VI semi- conductor, namely CdS nanocrystals with the change of aspect ratio.[56] We note that the decrease in band gap of the pristine (ZnO)24 nano structures in going from the nanosheet to the cage-shaped geometry, arises mainly from the downward shifting of the BCB with the decreas- ing aspect ratio. B. Bi-doped nano-structures The bi-dopings with the TM atoms tunes the band structure of the host systems such that it promotes band gap engineering, in addition to inducing the spin polar- ization in the doped systems. In this section, we have an- alyzed the magnetic properties as well as variation of the band gaps of the doped systems and ultimately figured out the effects of the morphological changes of the host system with decreasing aspect ratio from the nanosheet to the nanocage geometry. 1. Understanding the magnetic interactions The stability of TM bi-doped ZnO systems, is found to be sensitive to the magnetic coupling or chemical bond- ing between the dopant atoms,[57]which in turn is ex- pected to be influenced by the change in morphology of the host systems because of the different degrees of spa- tial confinement of the various morphologies. In order to explore the effects of the change in morphology of the host system on the magnetic interactions between the two magnetic dopant atoms, we have first examined the pre- ferred magnetic orderings of a pair of each magnetic atom M = Mn, Fe, Co, Ni and Cu in the three morphologies of (ZnO)24 nano system and thereafter, have tried to under- stand its origin. In a recent work on Mn-doped ZnO nano clusters,[58] it was shown that substitutional doping at the Zn site is energetically more favorable compared to endohedral or exohedral doping. Therefore, we have con- sidered in our study here only the substitutional doping with the TM atoms. The substitution at two Zn sites of the host system by the two dopant M atoms, thereby corresponds to a doping concentration of 8.4 atom %. To determine the energetically most favorable magnetic coupling, we have performed calculations for both the ferromagnetic as well as anti-ferromagnetic couplings of the two dopants. Moreover, for each of the ferromagnetic and anti-ferromagnetic orderings, both the near as well as far spatial separations of the two dopant atoms have been explored. In case of the near spatial separation, the two TM dopant atoms are separated by one oxygen atom, while for the case of far separation, we consider as much separation as possible in the chosen geometries. It is important to note here that the solubility of the dopant atoms in the (ZnO)24 host systems at the given growth conditions, can be determined by calculating formation energy for the bi-doped systems. The formation energy is the energy needed to insert the two dopant atoms (taken from a reservoir) into the (ZnO)24 systems after remov- ing two Zn atoms from the host system (to a reservoir). We, therefore, define the formation energy for the two M atoms substitutional doping at two Zn sites, as (cid:52)Ef = E(Zn24O24)−E(Zn22O24M2)−2E(Zn)+2E(M ), where E(Zn22O24M2) is the total energy of the optimal bi-doped system and E(M) [E(Zn)] is the total energy of an iso- lated M atom [ Zn atom ] which resembles the chemical potential of the respective elements in case of bulk-like system. According to our definition, formation energy should be positive for a stable structure. Generally, high formation energy leads to a low solubility. Note that the values of the single atom energies in the expression of the formation energy, depends very much on the growth con- ditions under which the bi-doped systems are prepared. In our calculations of formation energies, we have taken the isolated atoms in gas phase, as the reservoir of the respective element. The left panels of the Fig. 3 show the plot of our calculated formation energies for the optimal ferromagnetic as well as optimal anti-ferromagnetic con- figurations. The preferred spatial separations between 5 the two impurity dopant atoms in case of the optimal FM as well as AFM orderings, are also conveniently men- tioned in the Fig. 3 by a notation like '(1st, 2nd)'. The notation 1st denotes the type of favorable spatial sepa- ration for the optimal FM ordering out of far (denoted by F) as well as near (denoted by N) spatial separations and the notation 2nd denotes the type of favorable spa- tial separation for the optimal AFM ordering. Therefore, each of the 1st and 2nd notations, can adopt either 'F' or 'N' symbol. It is seen that the bi-Cu doping has the least formation energy and therefore, it will be the most favorable case than the other bi-dopings for each of the three host systems. We have also performed the forma- tion energy calculations with respect to the crystalline bulk phases of the metal atoms i.e α-Mn, body centered cubic Fe, hexagonal closed pack Co, face-centered cubic (FCC) Ni, FCC Cu and hexagonal closed pack Zn struc- tures as the reservoir of the respective element. We find that the overall trend in the variation of the formation energies remains the same in the both cases i.e decreasing formation energy for the dopants with increasing atomic numbers. For the Cu bi-doping in each of the three mor- phologies in particular, the formation energy, however, becomes negative with respect to the reservoir of Cu in equilibrium with the bulk FCC Cu. Note that the nega- tive formation energy (according to our definition) of the Cu bi-doped (ZnO)24 system, implies relative difficulty for fabrication of this system in experiment as additional energy is needed. Further, we use the energy difference (cid:52)E between the optimal AFM and FM orderings (i .e (cid:52)E = EAF M -EF M ) as an indicator of the magnetic stability. A positive (neg- ative) (cid:52)E indicates that the ground state is FM (AFM). Our estimated values of (cid:52)E for each doped system, are also shown in the right panels of the Fig. 3. It is, there- fore, seen that overall the AFM coupling between the two dopant atoms is energetically more favorable, except the case of bi-Mn doping in case of the 2D nanosheet geom- etry of the (ZnO)24 nano structure. It is also interesting to note that the optimal AFM couplings in the sheet ge- ometry, always favor a near spatial separation of around 3.14 - 3.26 A between the two dopant atoms for all the bi-doping cases. In case of the most stable structure for the bi-Mn doping in the (ZnO)24 nano sheet structure, the two Mn atoms favor to couple ferromagnetically with a far spatial separation of 6.52 A. This is also indicat- ing a fact that the a homogeneous distribution of Mn atoms in the ZnO nanosheet structure, would favor fer- romagnetism. Note that our observation for the mag- netic couplings of the bi-doping cases in the nanosheet structure, is in accordance with the previous results on dopings in a ZnO thin film, which indicated FM coupling for Mn-doped ZnO thin film and the AFM coupling for each of the Fe, Co, Ni doped ZnO thin films.[59, 60]In the case of the Mn-doped ZnO thin film, the magnetic cou- pling has, however, been reported to depend very much on the Mn doping concentrations. For the cases of our tube-shaped (ZnO)24 nano structure, the energy differ- 6 FIG. 3: (Color online) Plot of formation energies as calculated by HSE method for the optimal ferromagnetic (FM) and anti- ferromagnetic (AFM) coupling configurations of the dopants elements (left panels) and energy difference, (cid:52)E between the optimal AFM and FM structures for a given dopant (right panels) in case of bi-doped (ZnO)24 nano structures. The symbols within parenthesis, denote the types of spatial sep- aration between the two dopants for the optimal FM as well as AFM couplings. ence, (cid:52)E is found to be very small of the order of the RT energy (i.e. 25 meV), indicating a borderline case which means both the FM and AFM couplings are almost de- generate. Above all, the properties of a nanotube struc- ture, are very much dependent on the size and aspect ratio of the tube.[61, 62] However, it is not our purpose to explore it here as we are focusing on the overall vari- ation in the properties with the change of aspect ratio. Finally, the trend for the TM bi-doping in case of the cage-shaped (ZnO)24 nano structure is, however, solely of AFM type including the case for the bi-Mn doping. It is also interesting to note that the most stable AFM coupling for each bi-doping case in the cage structure, always adopts a far spatial separation between the two dopants. Note that our findings of favorable magnetic interactions of the bi-dopings in the cage structure, are also in accordance with the available earlier reports from the literature. Liu et al.[63] showed that doping of Mn into the Zn12O12 cluster, stabilizes AFM ground state for small Mn-Mn separation, while FM and AFM states are degenerate for large Mn-Mn distance. Another work on Fe doped ZnO nano clusters, indicates a clean dom- inance of AFM coupling in a neutral defect-free cluster, whereas defects under suitable conditions, can induce FM interaction between the dopant atoms.[64] To point out precisely the overall effects of the change of morphology on the magnetic interactions, it is seen that the mag- netic coupling in the most stable structure of the Mn FIG. 4: (Color online) Plot of TDOS (blue line) and PDOS (shaded region) of the most stable structure for the TM bi- doped (ZnO)24 nanostructures in each of the three morpholo- gies calculated within HSE method. The green colored shaded region shows the TM-d PDOS. For clarity, the TM-d PDOS is scaled by a factor of two for each system. TDOS of the pristine systems are also shown at the top panels. A smear- ing width of 0.1 eV is used. The energy along x-axis is shifted with respective to the Fermi energy of the respective system, as indicated by a vertical dashed line passing through zero energy. The positive and negative values of DOS are for the majority and minority spin channels, respectively. bi-doped systems, changes from far FM coupling in case of the sheet geometry to far AFM coupling in case of the cage shaped geometry. For the other bi-doping cases, the ground state structures favor AFM coupling for both the sheet as well as cage shaped geometries of the host system. The only difference in case of the magnetic cou- plings for the MESs of the bi-dopings with the Fe, Co, Ni and Cu ions between the sheet and cage structures of the host system, arises in the spatial separation of the two dopant atoms. This is in the sense that the AFM coupling of the two dopants at a near spatial separation is transformed to an AFM coupling at far spatial sepa- rations while transforming the shape of the host system from the sheet to the cage geometry. It is also interest- ing to note that the value of energy difference, (cid:52)E is exceptionally larger in case of the bi-Cu doping in the cage-shaped structure compared to its value in case of the sheet morphology. While studying the local structure around the TM dopants in the MESs of the doped systems, we note that the (cid:104)TM-O(cid:105) bond length with the neighboring oxygen atoms, is slightly larger than the (cid:104)Zn-O(cid:105) bond-length of the corresponding pure host system in case of the Mn bi- doping, while it is slightly smaller in case of the bidopings of Fe, Co, Ni and Cu atoms. Overall, the (cid:104)TM-O(cid:105) bond 7 ried out to calculate the atom-centered magnetic mo- ments of the constituent atoms in the MES of all the doped systems for the three morphologies. Fig. 5 shows the spin density surface plot for the MESs of three rep- resentative bi-doped systems, namely Mn, Co and Cu bi-doped systems in the sheet and cage morphologies. In the most optimized structures for the bidopings in the sheet geometry, our calculations show that the TM-TM distances vary as 6.52 A, 3.18 A, 3.24 A, 2.89 A and 3.12 A for TM = Mn, Fe, Co, Ni and Cu respectively. We note that the two dopant atoms carry an average local magnetic moment of magnitude 4.45 µB, 3.61 µB, 2.65 µB, 1.66 µB and 0.74 µB per dopant site for Mn, Fe, Co, Ni and Cu respectively. The moments of the neighboring O atoms are, however, very small within an amount of 0.04-0.07 µB per oxygen atom. In the spe- cific case of the FM ground state for the bi-Mn doping in the (ZnO)24 nano sheet structure, we note that the nearest neighbor oxygen atoms around the TM-dopant atoms, carry an average moment of 0.04 µB per O-site and they are also ferromagnetically coupled with the two dopant atoms. Therefore, two Mn atoms are interacting via ferromagnetically coupling to the oxygen atoms. The ferromagnetic coupling for Mn-doped DMS systems, has been understood previously in terms of hole mediated RKKY interaction or strong p-d indirect interaction for both the bulk as well as nano systems.[66, 67] On the other hand, in case of the AFM ground states for the bidopings with Fe, Co, Ni and Cu in the sheet structure with near spatial separation, we note that the two dopant atoms are connected to each other through one common oxygen atom which has zero magnetic moment i.e this oxygen atom is behaving as a nonmagnetic atom. There- fore, the two dopant atoms and the moments associated with the other neighboring oxygen atoms, interact with each other through a superexchange interaction involving the intermediate nonmagnetic oxygen atom, which re- sults in AFM coupling within each other. In case of the bi-dopings in the nanotube structure of (ZnO)24 systems, the values of the local magnetic moments for the dopant atoms remain almost the same as that of the respective cases of the bi-doped nanosheet structure. Turning our attention to the MESs of the bi-doped cage structures, we note that the TM-TM separations are 8.9 A, 8.4 A, 5.9 A, 6.1 A and 5.7 A for TM = Mn, Fe, Co, Ni and Cu dopings, while the average local magnetic moments at the dopant site retain values of 4.42 µB, 3.60 µB, 2.65 µB, 1.65 µB and 0.68 µB respectively. The favorable mag- netic coupling between the two dopant atoms, is still of AFM type as it was for the MESs in case of the most bi- doped nanosheet structures. The striking point to note is that contrary to the bi-doped nanosheet structures, the two dopant atoms in the MESs of the bi-doped nanocage structures, like to stay away from each other, which sig- nifies a long range nature of the magnetic interaction and it may be regarded as a possible consequence of the finite size effect of the cage morphology. FIG. 5: (Color online) Spin density surface plot of the most stable structures for the Mn, Co and Cu bi-doped (ZnO)24 nanostructures in the nanosheet (left) and nanocage (right) morphologies calculated within HSE method. The isosurface value of spin charge is 0.01 e−/A3. The red (blue) colored surface indicates the positive (negative) spin. lengths are larger in the doped nanocage structure com- pared to that in the doped nanosheet structure for each TM bi-doping. The significant overlaps of the valence d-orbital of the dopant atoms with the p-orbital of its neighboring oxygen atoms in the doped (ZnO)24 system, results into a change in the electronic configurations of the interacting atoms and thereby induces a spin polar- ization to them. In order to understand the magnetic properties of the bi-doped systems in more details, we have plotted in Fig. 4 the spin-polarized total density of states (TDOS) and the dopants TM-d PDOS for the most stable structure of each doped system. It is seen that the hybridization between each dopant atom and its neighboring oxygen atoms, results in the splitting of the energy levels near the Fermi energy. In case of the MES of bi-Mn doping in the sheet structure, the majority spin channel is completely filled and the minority spin chan- nel is completely empty. For the MESs of all the other bi-doping cases in the sheet as well cage structures, the TDOS is symmetric around the Fermi energy, which is a typical signature of the AFM coupling. The top of the valence band of the doped systems, is contributed mainly by the dopant states and therefore, the bidopings also play role in band gap engineering as discussed in the following Section III B 2. Mulliken population analysis[65] of spin has been car- 8 from the dopant's electronic configuration. In case of bi-Mn doping, each dopant atom is mainly character- ized by half-filled d-orbital. While considering the bi- doping cases along Mn→Fe→Co→Ni→Cu, the electronic states of each dopant atom is gradually filled by one ex- tra electron. This increases the splitting of the d-orbital of the dopant atoms and shifts the TVB towards lower energy. Consequently, the TVB gradually moves down- ward along bi-Mn doped system towards bi-Cu doped system as shown by the dashed lines in the Fig. 6 for the three structures. It is, therefore, obviously seen that the calculated band gaps of the doped systems, show a wide variation with the change of morphology of the host systems, as both the effects of changing morphology as well as the electronic structure of the dopant atoms come into play. IV. SUMMERY AND CONCLUSIONS In summary, first principles electronic structure calcu- lations have been performed to investigate the effects of morphological changes of the (ZnO)24 host system, on the magnetic properties and band gap variations in case of the pristine system as well as bi-dopings in it with the 3d late transition metal dopant atoms. The pris- tine systems are nonmagnetic and orbital hybridizations of the constituent atoms play significant role in stabi- lizing the respective morphologies. The doped systems are spin polarized. The magnetic couplings in the most stable structures are mostly anti-ferromagnetic in case of dopings in the sheet as well as cage morphologies which can be attributed to the superexchange interaction, while the optimal ferromagnetic and anti-ferromagnetic cou- plings are almost degenerate for the bi-dopings in the tube morphology. Separation between the two dopant atoms for the most preferred substitution, is found to be sensitive to the morphology of the host systems - the anti-ferromagnetic couplings between the two dopants in the sheet structure, favor short-ranged interaction, while the anti-ferromagnetic couplings are long ranged in case of the bi-dopings in the cage-shaped structure. Band gaps show wide variation for both the pristine as well as the doped systems and it shows overall decreasing trend while moving along sheet→tube→cage structures. The reduction in band gap variations with the decreasing as- pect ratio, results mainly from the downward shifting of the conduction band minimum. Our results demonstrate that morphology variation of the semiconducting host system, will be very promising for band gap engineering in future optoelectronic applications. Acknowledgments We are grateful to Prof. T. Saha-Dasgupta for pro- viding the computational facilities as well as for many simulating discussions. S. D. thanks Department of Sci- FIG. 6: (Color online) Plot of calculated band gaps (upper panels) and band-edge energies (lower panels) by HSE meth- ods for the MESs of the pure as well as TM bi-doped (ZnO)24 nanostructures in the three morphologies. The green (orange) colored shaded regions in the bottom panels represent the va- lence (conduction) energy region. The dashed lines in the lower panels show the overall trend in the band gap variation of the bi-doped systems. 2. Understanding the trend in band gap variations - Our calculated band gap [the highest occupied molec- lowest unoccupied molecular orbital ular orbital (HOMO-LUMO) gap] using HSE functional for the MESs of the pristine as well as the doped systems of all the three structures, are shown in Fig. 6. The positions of the TVB and BCB are also shown in the Fig. 6. We find that the band gaps in general reduce with the decreasing aspect ratio of the host systems i.e chang- ing the morphology from the nanosheet to the nanocage structures for both the pristine as well as the doped sys- tems. This means that a bi-doped system, in general, possesses the maximum band gap in the sheet geometry and the least for the cage-shaped geometry. This reduc- tion in band gaps of the doped systems arises mainly from the downward shifting of the BCB of the system while moving along the nanosheet→nanotube→nanocage structure, as clearly seen from the band edge positions of each doped system in the lower panels of the Fig. 6. Also note that the band gaps for the doped sys- tems are, in general, reduced compared to that of the pristine system within a given morphology having fixed degree of spatial confinement. It can be readily under- stood from the PDOS plot in the Fig. 4 which shows the appearance of the dopant states in the mid gap re- gion of the pristine (ZnO)24 system and thereby result- ing into the lowering of effective band gaps of the doped systems. Interestingly, the band gaps of the doped sys- tems within a given morphology of the host system, show an overall increasing trend from the bi-Mn doped sys- tem to the bi-Cu doped system. It can be understood ence and Technology, India for support through INSPIRE Faculty Fellowship, Grant No. IFA12-PH-27. 9 [1] Z. R. Tian, J. A. Voigt, J. Liu, B. Mckenzie, M. J. Mc- dermott, M. A. Rodriguez, H. Konishi and H. Xu, Nat. Mater. 2, 821 (2003). [2] D. C. Look, Mater. Sci. Eng. 80, 383 (2001); U . Ozgur, Ya. I. Alivov, C. Liu, A. Teke, M. A. Reshchikov, S. Dogan, V. Avrutin, S.-J. Cho and H. Morkoc, J. Appl. Phys. 98, 041301 (2005). [3] S. W. Jung, S.-J. An, Gyu-Chul Yi, C. U. Jung, Sung-Ik Lee and S. Cho, Appl. Phys. Lett. 80, 4561 (2002); J. R. Neal, A. J. Behan, R. M. Ibrahim, H. J. Blythe, M. Ziese, A. M. Fox and G. A. Gehring, Phys. Rev. Lett. 96, 197208 (2006). [4] M. Venkatesan, C. B. Fitzgerald, J. G. Lunney and J. M. D. Coey, Phys. Rev. Lett. 93, 177206 (2004) ; Nick S. Norberg, K. R. Kittilstved, J. E. Amonette, R. K. Kukkadapu, D. A. Schwartz and D. R. Gamelin, J. Am. Chem. Soc. 126, 9387 (2004). [5] P. V. Radovanovic, N. S. Norberg, K. E. McNally, and D. R. Gamelin, J. Am. Chem. Soc. 124, 15192 (2002). [6] S. J. Pearton, D. P. Norton, M. P. Ivill, A. F. Hebard, J. M. Zavada, W. M. Chen and I. A. Buyanova, J. Elec- tronic Mater. 36, 462 (2007); P. Sharma, A. Gupta, F. J. Owens, A. Inoue, K. V. Rao, J. Mag. Mag. Mater. 282, 115 (2004). [7] W. Prellier, A. Fouchet and B. Mercey, J. Phys. : Con- dens. Matter 15, 1583 (2003); K. Sato and H. Katayama- Yoshida, Semicond. Sci. Technol. 17, 367 (2002). [8] J. Wang, X. Sun, A. Wei, Y. Lei, X. Cai, C. Li, Z. Dong, Appl. Phys. Lett. 88, 233106 (2006). [9] K. Yu, Q. Zhang, J. Wu, L. Li, Y. Xu, S. Huang and Z. Zhu, Nano Res. 1, 221 (2008); C. S. Lao, P. X. Gao, R. S. Yang, Y. Zhang, Y. Dai, Z. L. Wang, Chem. Phys. Lett. 417, 359 (2005). [10] Y. J. Xing, Z. H. Xi, Z. Q. Xue, X. D. Zhang, J. H. Song, R. M. Wang, J. Xu, Y. Song, S. L. Zhang and D. P. Yu, Appl. Phys. Lett. 83, 1689 (2003); Y. Sun, G. M. Fuge, N. A. Fox, D. J. Riley and M. N. R. Ashfold, Advanced Materials 17, 2477 (2005). [11] X. Y. Kong and Z. L. Wang, Nano Lett. 3, 1625 (2003); P. X. Gao and Z. L. Wang, Small 1, 945 (2005). [12] C. S. Lao, J. Liu, P. Gao, L. Zhang, D. Davidovic, R. Tummala and Z. L. Wang, Nano Lett. 6, 263 (2006). [13] X. Y. Kong, Y. Ding, R. Yang and Z. L. Wang, Science 303, 1348 (2004); W. L. Hughes and Z. L. Wang, Appl. Phys. Lett. 86, 043106 (2005). [14] A. Dmytruk, I. Dmitruk, I. Blonskyy, R. Belosludov and Y. Kawazoe, Microelectron. J. 40, 218 (2009); X. Wang, B. Wang, L. Tang, L. Sai and J. Zhao, Phys. Lett. A 374, 850 (2010). [15] A. Sahai, Y. Kumar, V. Agarwal, S. F. Olive-M´endez and N. Goswami, J. Appl. Phys. 116, 164315 (2014). [16] J. Zhang, L. D. Sun, J. L. Yin, H. L. Su, C. S. Liao, C. H. Yan, Chem. Mater. 14, 4172 (2002). [17] M. S. Arnold, P. Avouris, Z. W. Pan, Z. L. Wang, J. Phys. Chem. B 107, 659 (2003). [18] W. I. Park, G. C. Yi, Adv. Mater. 16, 87 (2004); M. T. Chen, M. P. Lu, Y. J. Wu, J. Song, C. Y. Lee, M. Y. Lu, Y. C. Chang, L. J. Chou, Z. L. Wang, L. J. Chen, Nano Lett. 10, 4387 (2010). [19] S. Chu, G. Wang, W. Zhou, Y. Lin, L. Chernyak, J. Zhao, J. Kong, L. Li, J. Ren, J. Liu, Nat. Nanotechnol. 6, 506 (2011). [20] M. Law, L. E. Greene, J. C. Johnson, R. Saykally, P. D. Yang, Nat. Mater. 4, 455 (2005). [21] A. Wei, L. Pan and W. Huang, Materials Science and Engineering B 176, 1409 (2011); X. D. Bai, P. X. Gao, Z. L. Wang and E. G. Wang, Appl. Phys. Lett. 82, 4806 (2003). [22] H. Pan, J. B. Yi, L. Shen, R. Q. Wu, J. H. Yang, J. Y. Lin, Y. P. Feng, J. Ding, L. H. Van, J. H. Yin, Phys. Rev. Lett. 99, 127201 (2007); T. S. Herng, S. P. Lau, L. Wang, B. C. Zhao, S. F. Yu, M. Tanemura, A. Akaike, K. S. Teng, Appl. Phys. Lett. 95, 012505 (2009). [23] Q. Wang, Q. Sun and P. Jena, J. Chem. Phys. 129, 164714 (2008). [24] R. Podila, W. Queen, A. Nath, J. T. Arantes, A. L. Schoenhalz, A. Fazzio, G. M. Dalpian, J. He, S. J. Hwu, M. J. Skove and A. M. Rao, Nano Lett. 10, 1383 (2010). [25] Q. Wang, Q. Sun, P. Jena and Y. Kawazoe, Phys. Rev. B 79, 115407 (2009). [26] O. Lupan, G. Chai, L. Chow, Microelectro. Eng. 85, 2220 (2008). [27] B. Q. Zhang, C. S. Dandeneau, X. Zhou and G. Cao, Adv. Mater. 21, 4087 (2009); X. Zhang, X. Huang, C. Li and H. Jiang, Adv. Mater. 25, 4093 (2013). [28] R. Y. Hong, J. H. Li, L. L. Chen, D. Q. Liu, H. Z. Li, Y. Zheng, J. Ding, Powder Technology 189, 426 (2009). [29] H. M. Xiong, Adv. Mater. 25, 5329 (2013); J. W. Ras- mussen, E. Martinez, P. Louka and D. G. Wingett, 7, 1063 (2010). [30] T. M. Gur, S. F. Bent and F. B. Prinz, J. Phys. Chem. C 118, 21301 (2014). [31] Z. C. Tu, X. Hu, Phys. Rev. B 74, 035434 (2006); D. Look, D. Reynolds, J. Sizelove, R. Jones, C. Litton, G. Cantwell, W. Harsch, Solid State Commun. 105, 399 (1998); Z. Fu-Chun, Z. Zhi-Yong, Z. Wei-Hu, Y. Jun- Feng, Y. Jiang-Ni, Chin. Phys. Lett. 26, 016105 (2009). [32] F. Ahmed, N. Arshi, M. S. Anwar, R. Danish and B. H. Koo, RSC Adv. 4, 29249 (2014). [33] L. Yin and Y. Bando, Nat. Mater. 4, 883 (2005); C. Ye, X. Fang, Y. Hao, X. Teng and L. Zhang, J. Phys. Chem. B 109, 19758 (2005). [34] T. Akiyama, K. Nakamura and T. Ito, Phys. Rev. B 73, 235308 (2006). [35] S. Datta, M. Kabir and T. Saha-Dasgupta, Phys. Rev. B 86, 115307 (2012); S. Datta, M. Kabir, T. Saha-Dasgupta and D. D. Sarma, J. Phys. Chem. C 112, 8206 (2008). [36] J. Heyd, G. E. Scuseria and M. Ernzerhof, J. Chem. Phys. 124, 219906 (2006). [37] Vienna ab initio simulation package (VASP), Technische Universitat Wien, 1999; G. Kresse and J. Hafner, Phys. Rev. B 47, 558 (1993); G. Kresse and J. Furthmuller, Phys. Rev. B 54, 11169 (1996). [38] P. E. Blochl, Phys. Rev. B 50, 17953 (1994). [39] G. Kresse and D. Joubert, Phys. Rev. B 59, 1758 (1999). [40] J. P. Perdew, K. Burke and M. Ernzerhof, Phys. Rev. 10 Lett. 77, 3865 (1996). [41] A. Janotti, J. B. Varley, P. Rinke, N. Umezawa, G. Kresse, and C. G. Van de Walle, Phys. Rev. B 81, 085212 (2010); E. R. Batista, J. Heyd, R. G. Hennig, B. P. Uberuaga, R. L. Martin, G. E. Scuseria, C. J. Umrigar and J. W. Wilkins, Phys. Rev. B 74, 121102 (2006). [42] H. J. Monkhorst and J. D. Pack, Phys. Rev. B 13, 5188 (1976); J. D. Pack and H. J. Monkhorst, Phys. Rev. B 16, 1748 (1977). [43] H. Guo, Y. Zhao, N. Lu, E. Kan, X. C. Zeng, X. Wu and J. Yang, J. Phys. Chem. C 116, 11336 (2012). [44] Y. N. Xu and W. Y. Ching, Phys. Rev. B 48, 4335 (1993). [45] M. K. Yadav, M. Ghosh, R. Biswas, A. K. Raychaudhuri, A. Mookerjee and S. Datta, Phys. Rev. B 76, 195450 (2007). Group III (Springer-Verlag, Heidelberg, 2002). [54] K. Mak, C. Lee, J. Hone, J. Shan and T. F. Heinz, Phys. Rev. Lett. 105, 136805 (2010). [55] W. S. Yun, S. W. Han, S. C. Hong, I. G. Kim and J. D. Lee, Phys. Rev. B 85, 033305 (2012); Y. Ding, Y. Wang, J. Ni, L. Shi, S. Shi and W. Tang, Physica B 406, 2254 (2011). [56] B. A. Simmons, S. Li, V. T. John, G. L. McPherson, A. Bose, W. Zhou and J. He, Nano Lett. 2, 263 (2002). [57] P. Gopal and N. A. Spaldin, Phys. Rev. B 74, 094418 (2006); N. Ganguli, I. Dasgupta and B. Sanyal, Appl. Phys. Lett. 94, 192503 (2009). [58] H. T. Liu, S. Y. Wang, G. Zhou, J. Wu and W. H. Duan, J. Chem. Phys. 124, 174705 (2006). [59] Q. Wang, Q. Sun, P. Jena and Y. Kawazoe, Phys. Rev. [46] B. Wang, X. Wang, G. Chen, S. Nagase and J. Zhao, J. B 79, 115407 (2009). Chem. Phys. 128, 144710 (2008). [47] C. Tusche, H. L. Meyerheim and J. Kirschner, Phys. Rev. Lett. 99, 026102 (2007); M. Topsakal,S. Cahangirov, E. Bekaroglu and S. Ciraci, 80, 235119 (2009). [48] B. Wang, S. Nagase, J. Zhao and G. Wang, Nanotech- nology 18, 345706 (2007); Y. Li, Z. Zhou, Y. Chen and Z. Chen, J. Chem. Phys. 130, 204706 (2009). [49] H. Hakkinen, M. Moseler and U. Landman, Phys. Rev. Lett. 89, 033401 (2002); S. Datta, M. Kabir and T. Saha- Dasgupta, Phys. Rev. B 84, 075429 (2011). [50] S. Datta and T. Saha-Dasgupta, J. Phys. : Condens. Matter 25, 225302 (2013); S. Datta and T. Saha- Dasgupta, J. Phys. : Condens. Matter 26, 185004 (2014). [51] G. Stan, C. V. Ciobanu, P. M. Parthangal, and R. F. Cook, Nano. Lett. 7, 3691 (2007); Q. Zhao, T. Xie, L. Peng, Y. Lin, P. Wang, L. Peng and D. Wang, J. Phys. Chem. C 111, 17136 (2007). [52] M. Topsakal, S. Cahangirov, E. Bekaroglu and S. Ciraci, Phys. Rev. B 80, 235119 (2009); Z. C. Tu,J. Comput. Theor. Nanosci. 7, 1182 (2010). [53] Condensed Matter, Landolt-Bornstein, New Series, [60] H. Y. Xu, Y. C. Liu, C. S. Xu, Y. X. Liu, C. L. Shao and R. Mu, J. Chem. Phys. 124, 074707 (2006). [61] G. Stan, C. V. Ciobanu, P. M. Parthangal and R. F. Cook, Nano Lett. 7, 3691 (2007). [62] Q. Han, B. Cao, L. Zhou, G. Zhang and Z. Liu, J. Phys. Chem. C 115, 3447 (2011); M. Lucas, W. Mai, R. Yang, Z. L. Wang and E. Riedo, Nano Lett. 7, 1314 (2007). [63] H. Liu, S. Wang, G. Zhou, J. Wu and W. Duan, J. Chem. Phys. 124, 174705 (2006). [64] N. Ganguli, I. Dasgupta and B. Sanyal, Appl. Phys. Lett. 94, 192503 (2009). [65] R. S. Mulliken, J. Chem. Phys. 23, 1833 (1955). [66] U. Ilyas, R. S. Rawat, T. L. Tan, P. Lee, R. Chen, H. D. Sun, L. Fengji, and S. Zhang, J. App. Phys. 111, 033503 (2012); T. Dietl, H. Ohno, and F. Matsukura, Phys. Rev. B 63, 195205 (2001). [67] B. K. Rao and P. Jena, Phys. Rev. Lett. 89, 185504 (2002); C.-W. Zhang, C. Han, S.-S. Yan and F.-B. Zheng, Euo. Phys. Lett. 95, 47011 (2011).
1206.2163	1	1206	2012-06-11T11:20:16	Resonant plasmonic effects in periodic graphene antidot arrays	[ "cond-mat.mes-hall", "cond-mat.other" ]	We show that a graphene sheet perforated with micro- or nano-size antidots have prominent absorption resonances in the microwave and terahertz regions. These resonances correspond to surface plasmons of a continuous sheet "perturbed" by a lattice. They are excited in different diffraction orders, in contrast to cavity surface plasmon modes existing in disconnected graphene structures. The resonant absorption by the antidot array can essentially exceed the absorption by a continuous graphene sheet, even for high antidot diameter-to-period aspect ratios. Surface plasmon-enhanced absorption and suppressed transmission is more efficient for higher relaxation times of the charge carriers.	cond-mat.mes-hall	cond-mat	Resonant plasmonic effects in periodic graphene antidot arrays 1 Instituto de Ciencia de Materiales de Arag´on and Departamento de F´ısica de la Materia Condensada, A. Yu. Nikitin1,∗ F. Guinea2, and L. Martin-Moreno1† 2 Instituto de Ciencia de Materiales de Madrid, CSIC, Cantoblanco, E-28049 Madrid, Spain CSIC-Universidad de Zaragoza, E-50009, Zaragoza, Spain We show that a graphene sheet perforated with micro- or nano-size antidots have prominent absorption resonances in the microwave and terahertz regions. These resonances correspond to surface plasmons of a continuous sheet "perturbed" by a lattice. They are excited in different diffraction orders, in contrast to cavity surface plasmon modes existing in disconnected graphene structures. The resonant absorption by the antidot array can essentially exceed the absorption by a continuous graphene sheet, even for high antidot diameter-to-period aspect ratios. Surface plasmon- enhanced absorption and suppressed transmission is more efficient for higher relaxation times of the charge carriers. PACS numbers: 42.25.Bs, 41.20.Jb, 42.79.Ag, 78.66.Bz Graphene now is widely known not only for its amazing transport properties,1 but also for a signif- icant potential in photonics.2 Most applications re- quire that the electromagnetic (EM) field is first ab- sorbed by graphene. In a large part of the EM spec- trum the two-dimensional conductance of one mono- layer σ is small, so that α = 2πσ/c (cid:28) 1. Then the absorption coefficient A for a free-standing graphene can be approximately written at normal incidence as A (cid:39) 2Re(α). In the optical regime graphene behaves as an absorbing dielectric (its effective refractive index is a purely imaginary number): αopt = πα0/2, with α0 = 1/137 being the fine-structure constant. This leads to the classical result Aopt = πα0 (cid:39) 2.3%. In both terahertz (THz) and microwave frequency ranges the situation is rather different. The conductivity is contributed predominantly by the intra-band elec- tronic transitions, so that graphene behaves like a two-dimensional metal. For low temperatures its con- ductance can be approximated by a Drude-Lorentz model3 -- 5 αTHz = 2iα0µ/(ω + iτ−1) depending upon the chemical potential µ, frequency ω and relaxation time τ . In this case for ωτ (cid:28) 1 the absorption coeffi- cient reads ATHz (cid:39) 4α0µ/(τ ω2) and can become of order of tenths of a percent. Obviously, it is challeng- ing to use graphene as an efficient absorber. Fortunately, one of the promising points of graphene is that it supports surface waves [graphene surface plasmons (GSPs)].6 -- 18 GSPs allow for a strong sub- wavelength confinement of the EM fields and thus for a much higher absorption. Previously it has been demonstrated that periodic arrays of graphene ribbons,12,14 discs15 and rings16 can resonantly absorb THz radiation due to excitation of GSP cavity modes. However, this kind of structures is not "electrically- continuous" preventing therefore a direct current pass- ing through the samples. An alternative solution can be a graphene periodic antidot array (GPAA),19 -- 24 which presents both continuity (to provide the electric transport) and periodicity (to transfer the momentum to GSPs from the incident wave). The antidots in a graphene sheet can be considered as regions of zero conductivity due to either (i) actual holes in graphene, (ii) the influence of an external voltage or (iii) appro- priate chemical doping. GPAAs could open interesting possibilities for elec- tromagnetic control via GSP-assisted effects. This can be an important point for designing photo-electric de- vices, e.g. tunable absorbers that can act at the same time as conducting electrodes. FIG. 1: (Color online) The schematic of the studied sys- tem: a plane electromagnetic wave impinging onto a peri- odic antidot array in a graphene monolayer placed onto a dielectric substrate. In this Letter we report on GSP-enhanced ab- sorption and suppressed transmission in continuous GPAAs with circular antidots. We characterize reso- nances, presenting various dependencies of absorption A, transmission T , and reflection R coefficients upon several parameters: period L, antidot diameter d and relaxation time τ . We consider a square GPAA placed on the substrate with dielectric permittivity ε. The system is illuminated by a normal-incident monochro- matic plane wave with wavevector k from the vacuum half-space. The electric field of the incident wave is parallel to one of the Bragg's vectors of the lattice. The scheme of the structure is shown in Fig. 1. Graphene is modeled by a conductivity σ, computed within the random-phase approximation,3 -- 5 which is valid pro- vided that the mean free path is shorter than the su- perlattice scale. In this work we assume room temper- ature, T = 300 K. FIG. 2: (Color online) The spectra of absorption, reflection and transmission coefficients for square GPAAs. (a) A, R, T for the suspended GPAA and GPAA on the substrate with ε = 3 and τ = 40 ps. RF and TF corresponding to the continuous graphene sheet (Fresnel coefficients) are shown by the discontinuous curves. (b) Absorption spectra for GPAAs at different τ for ε = 1. The spectra for continuous graphene monolayers are shown by the dashed lines. The insets to (b) show the colorplots for the spatial distribution of the electric field: modulus (below) and the real part of the projection on the direction of the incident wave electric field shown by an arrow (above). In all panels µ = 0.2eV , L = 2d = 5µm. An example of the wavelength spectra for A, T and R is shown in Fig. 2 (a) corresponding to L = 2d = 5µm. The simulations have been conducted by using the finite elements method (FEM).26 We choose this value of L for demonstration purposes, in order to have prominent resonances in the window 20 − 100µm; the dependence with L will be addressed later on. Both 2 √ (cid:113) the case of free-standing GPAA and GPAA on the sub- strate are shown. While absorption and reflection co- efficients have a set of resonance maxima, the trans- mission spectra possess pronounced minima. The po- sitions of the resonances correspond approximately to √ the condition of GSP excitation in different diffraction n2 + m2 = kp, where G = 2π/L is the orders, i.e. G shortest Bragg's vector, and kp is the GSP wavevec- tor. For convenience, these resonances will be further labeled as [n, m]. In suspended GPAAs with suffi- ciently small antidots a good approximation for kp can be kp (cid:39) k L )2] being the average value of the normalized graphene conduc- tance in PGAA. The characterization of the resonance positions for large antidots will be presented below. For graphene in a nonsymmetric dielectric environ- ment the dispersion relation reads ε1 + 2α = 0, qz1 1 − ¯α−2, with ¯α = α · [1 − π + ε2 qz2 4 ( d ε1,2 − k2 p/k2 and ε1,2 are the dielec- where qz1,2 = tric constants of the surrounding media.9 This equa- tion cannot be resolved in algebraic functions and therefore is treated numerically. The positions of the three highest resonance wavelengths estimated from the above "undressed" dispersion law are shown by the discontinuous vertical lines in Fig. 2 (a). The min- imum/maximum corresponding to the longest wave- length resonance [1, 0] is redshifted with respect to the unperturbed resonance wavelength, while the next two (lower-wavelength) ones, i.e. [1, 1] and [2, 0] are blue- shifted. In a symmetric configuration the maximal lim- iting value for A is 1/2, but it can be increased, for instance, placing a metal layer below the GPAA.14,15 In order to better visualize the resonance picture, the inset to Fig. 2 (b) shows the spatial field distri- bution for both electric field modulus and the real part of the electric field component along the polar- ization of the incident wave. The colorplots are taken at the wavelengths corresponding to [1, 0] and [1, 1] res- onances at λ = 53.26 µm and λ = 42.06 µm, respec- tively. Let us point out that the plasmonic wavelengths λp corresponding to a continuous monolayer for these values of λ are λp = 6.61 µm and λp = 4.12 µm. Notice that GSP resonances in PGAA are reminis- cent to short-range surface plasmon resonances studied in hole arrays in ultra-thin metallic films.25 However, in case of metallic structures, absorption and reflection peaks (transmission dips) appear in the optical region. Owing to much higher values of τ ,27,28 suspended samples can present a special interest. Nowadays, plac- ing graphene on a substrate severely reduces the mo- bility. For this reason, in this Letter we mainly con- centrate on the free-standing GPAA. However, for aca- demic purposes, in order to see other effects of the sub- strate not related to the change of mobility, in Fig. 2 (a) we compare the spectra for symmetric and nonsym- metric configurations at the same value of τ . From this comparison it follows that the spectra are similar, and 203040506070809010011000.20.40.60.81 λ, μmAbsorptionAbsorption, Reflection, Transmission[1,0][1,0][1,1][1,1][2,0][1,0][1,1][2,0]E(a)(b)203040506070 τ, ps40510.25[1,1][1,0]0.10.20.30.4 1ε=3ε=FTFRTRAthe only significant difference between them consists in a displacement of the resonance for GPAA to longer wavelength compared to the free-standing case. This is related to the higher values of the GSP wavevector for a nonsymmetric dielectric environment. Let us turn to the behavior of the absorption coefficient with respect to the change of τ . is strongly dependent upon the period due to Bragg's origin of the GSP resonance. 3 FIG. 3: (Color online) (c) Absorption spectra for GPAAs at different antidot sizes d. The main figure is for L = 5µm, while the inset is for L = 1µm. The curve corresponding to d/L = 0 renders A for the continuous monolayer. ε = 3, τ = 40 ps, µ = 0.2eV . In Fig. 2 (b) the absorption spectra for different re- laxation rates τ are compared. It is seen that higher τ provide higher visibility of the resonances, especially for lower wavelengths. An important message here is that the GSP-induced absorption in GPAAs can not only be higher than the absorption corresponding to lower τ , but can also largely exceed the absorption in the continuous graphene monolayer. Since resonances for different τ show the same behavior in function of geometric parameters of GPAAs, we will further con- centrate on the case of the maximal predicted value for τ in continuous films.27 The resonances are very sensitive to the geometrical aspect ratio d/L. In Fig. 3 the evolution of the longest- wavelength resonance [1, 0] with respect to the antidot diameter change is illustrated for L = 5 µm. When the antidot size increases, the resonance suffers a red- shift and the intensity of the absorption peak increases. However, in the interval of d/L = (0.3, 0.8) the reso- nance position is quite stable. Further increasing d/L up to the values d/L ∼ 1 leads to the fast displacement of the peak. A point to note is that even for large d/L, for which the graphene area is smaller than the one of the antidot one, the maximal value of the resonance peak remains hight. The same behavior takes place for other values of the period. As an example, the case of L = 1 µm is shown in the inset to Fig. 3. Notice that in contrast to the arrays of graphene ribbons or discs, the resonance position for GPAA at a fixed value of d FIG. 4: (Color online) Characterization of the resonances in GPAAs with respect to the aspect ratio d/L. (a) Maxi- mal value of absorption Amax as a function of d/L at a fixed L for the resonances [1, 0] and [1, 1]. (b) The positions of the maxima for [1, 0] and [1, 1] resonances at different d/L for a fixed L. (c) The positions of the maxima for [1, 0] and [1, 1] resonances at different d/L for a fixed d. In (b), (c) continuous lines represent the estimated position of the resonance, while the symbols correspond to the full calcu- lations. In all panels ε = 1, µ = 0.2 eV and τ = 40 ps. In order to gain a further insight into the proper- ties of GSP resonances in GPAAs, let us analyze in more detail the dependencies of the absorption max- ima Amax and resonance wavelength λmax upon d/L. We consider GPAAs with the L values down to 1µm and values of d down to 100nm. In Fig. 4 (a),(b) Amax and λmax both for [1, 0] and [1, 1] resonances are shown as a function of d, at several fixed periods. For the [1, 0] resonance, Amax quickly reaches high values, af- ter which it saturates, remaining practically constant for all periods. In contrast, Amax for the [1, 1] reso- nance has an optimum close to d/L = 0.7, with this value being independent upon L. 455055606570758010−310−210−1λ, μmλ, μmAbsorption dL10−310−210−120242832 00.10.20.30.40.5 0.10.20.30.40.50.60.70.8204060801051100120maxAdL10305070 max,mλμmax,mλμ,mLμ[1,0][1,1][1,0][1,1]1mdμ=(b)(c)(a)4 The dependency of λmax upon d and L changes in different regions of d/L. When d/L (cid:46) 0.1, the dispersion relation of the GP and therefore λmax, is governed by the average value of α, see the coinci- dence of the points and the curves in the region of small d/L in Fig. 4 (b,c). However, for larger d/L there is a modification of the GSP dispersion branches. The degenerated resonance diffraction orders become strongly coupled (directly and also due to multiple scattering processes through both evanescent and ra- diative diffraction orders), leading to an increase of the bandgap width and a shift of the bandgap center. In other words, the dependence of λmax deviates from the estimation based on the average conductance, see Fig. 4 (b,c). To summarize, in this Letter we have shown that graphene periodic antidot arrays (GPAAs) can pro- vide a strong electromagnetic response in both mi- crowave and THz regions, where GSPs are excited. The absorption peaks and transmission dips in the an- tidot structure are more pronounced for less absorbing graphene samples, characterized by higher relaxation rates of charge carriers. The system discussed here can be realized in CVD graphene, where large sam- ples can be routinely fabricated.29 The resonances in antidot arrays are due to excitation of dressed GSPs under the Bragg's condition. This makes the difference with electrically-isolated graphene periodic structures (stripes, discs, etc.), where the resonances are due to cavity GSP-modes. The continuous graphene struc- tures possess thus a supplementary tunability parame- ter L. Additionally, GPAAs can have applications for devices, in which the same graphene layer can be both an electric conductor and radiation absorber. The authors are grateful to M. L. Nesterov for help- ful discussions regarding numeric simulations, and to Institute for Biocomputation and Physics of Com- plex Systems (BIFI) of Zaragoza for computational reasons. We acknowledge support from the Span- ish MECD under Contract No. MAT2009-06609-C02, FIS2008-00124, Consolider CSD2007-00010, and Con- solider Project "Nanolight.es". A.Y.N. acknowledges the Juan de la Cierva Grant No. JCI-2008-3123. ∗ Electronic address: [email protected] † Electronic address: [email protected] 1 A. H. Castro Neto, F. Guinea, N. M. R. Peres, K. S. Novoselov, and A. K. Geim, Rev. Mod. Phys. 81, 109 (2009). 2 F. Bonaccorso, Z. Sun, T. Hasan, and A. C. Ferrari, Nature Photon. 4, 611 (2010). 3 B. Wunsch, T. Stauber, F. Sols, and F. Guinea, New J. of Phys. 8, 318 (2006). 4 E. H. Hwang and S. Das Sarma, Phys. Rev. B 75, 205418 (2007). 5 L. A. Falkovsky, Phys.-Uspekhi 51, 887 (2008). 6 Kenneth W. -K. Shung, Phys. Rev. B 34, 979 (1986). 7 G. Campagnoli and E. Tosatti, in Progress on Elec- tron Properties of metals, edited by R. Girlanda et al., Kluwer, Dordrecht, p. 337 (1989). 8 O. Vafek, Phys. Rev. Lett. 97, 266406 (2006). 9 G. W. Hanson, J. Appl. Phys. 103, 064302 (2008). 10 M. Jablan, H. Buljan, and M. Soljaci´c, Phys. Rev. B 80, 245435 (2009). 11 A. Vakil and N. Engheta, Science 332, 1291 (2011). 12 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, Nature Nanotech. 6, 630 (2011). 13 G. G´omez-Santos and T. Stauber, Phys. Rev. B 84, 165438 (2011). 14 A. Yu. Nikitin, F. Guinea, F. J. Garcia-Vidal, and L. Martin-Moreno, Phys. Rev. B 85, 081405 (2012). 15 S. Thongrattanasiri, F. H. L. Koppens, and F. J. Garc´ıa de Abajo, Phys. Rev. Lett. 108, 047401 (2012). 16 P. Liu, W. Cai, L. Wang, X. Zhang, and J. Xu, Appl. Rev. Lett. 100, 153111 (2012). 17 J. Chen, M. Badioli, P. Alonso-Gonz´alez, S. Thongrat- tanasiri, F. Huth, J. Osmond, M. Spasenovic, A. Cen- teno, A. Pesquera, P. Godignon, A. Zurutuza, N. Ca- mara, J. Garcia de Abajo, R. Hillenbrand, F. Koppens, e-print arXiv:1202.4996v2. 18 Z. Fei, A. S. Rodin, G. O. Andreev, W. Bao, A. S. McLeod, M. Wagner, L. M. Zhang, Z. Zhao, G. Dominguez, M. Thiemens, M. M. Fogler, A. H. Castro- Neto, C. N. Lau, F. Keilmann, D. N. Basov, e-print arXiv:1202.4993v1. 19 T. G. Pedersen, C. Flindt, J. Pedersen, N. A. Mortensen, A. P. Jauho, and K. Pedersen, Phys. Rev. Lett. 100, 136804 (2008) 20 M. Kim, N. S. Safron, E. Han, M. S. Arnold, and P. Gopalan, Nano Lett. 10, 1125 (2010). 21 J. Bai, X. Zhong, S. Jiang, Yu Huang, and X. Duan, Nature Nanotech. 5, 190 (2010). 22 O. Akhavan, ACS Nano 4, 4174 (2010). 23 B. Krauss, P. Nemes-Incze, V. Skakalova, L. P. Biro, K. von Klitzing, and J. H. Smet, Nano Lett. 10, 4544 (2010). 24 M. Begliarbekov, O. Sul, J. Santanello, N. Ai, X. Zhang, E. H. Yang, and S. Strauf, Nano Lett. 11, 1254 (2011). 25 S. G. Rodrigo, L. Martin-Moreno, A. Yu. Nikitin, A. V. Kats, I. S. Spevak, and F. J. Garcia-Vidal, Opt. Lett. 34, 4 (2009). 26 A commercial software comsol has been used. 27 E. H. Hwang, S. Adam, and S. D. Sarma, Phys. Rev. Lett. 98, 186806 (2007). 28 K. I. Bolotin, K. J. Sikesb, Z. Jianga, M. Klimac, G. Fudenberga, J. Honec, P. Kima, and H. L. Stormer, Sol. State Commun. 146, 351 (2008). 29 S. Bae, H. Kim, Y. Lee, X. Xu, J.-S. Park, Yi Zheng, J. Balakrishnan, T. Lei, H. R. Kim, Y. Il Song, Y.-J. Kim, K. S. Kim, B. Ozyilmaz, J.-H. Ahn, B. H. Hong, and S. Iijima, Nature Nanotechnology 5, 574 (2010)
1504.02449	2	1504	2015-08-20T09:05:30	Tuning edge state localization in graphene nanoribbons by in-plane bending	[ "cond-mat.mes-hall" ]	The electronic properties of graphene are influenced by both geometric confinement and strain. We study the electronic structure of in-plane bent graphene nanoribbons, systems where confinement and strain are combined. To understand its electronic properties, we develop a tight-binding model that has a small computational cost and is based on exponentially decaying hopping and overlap parameters. Using this model, we show that the edge states in zigzag graphene nanoribbons are sensitive to bending and develop an effective dispersion that can be described by a one-dimensional atomic chain model. Because the velocity of the electrons at the edge is proportional to the slope of the dispersion, the edge states become gradually delocalized upon increasing the strength of bending.	cond-mat.mes-hall	cond-mat	a Tuning edge state localization in graphene nanoribbons by in-plane bending S.G. Stuij1, P. H. Jacobse2, V. Jurici´c1, and C. Morais Smith1 1Institute for Theoretical Physics, Center for Extreme Matter and Emergent Phenomena, Utrecht University, Leuvenlaan 4, 3584 CE Utrecht, the Netherlands 2Debye Institute for Nanomaterials Science, Condensed Matter and Interfaces, Utrecht University, P.O. Box 80.000, 3508 TA Utrecht, the Netherlands (Dated: October 1, 2018) The electronic properties of graphene are influenced by both geometric confinement and strain. We study the electronic structure of in-plane bent graphene nanoribbons, systems where confinement and strain are combined. To understand its electronic properties, we develop a tight-binding model that has a small computational cost and is based on exponentially decaying hopping and overlap parameters. Using this model, we show that the edge states in zigzag graphene nanoribbons are sensitive to bending and develop an effective dispersion that can be described by a one-dimensional atomic chain model. Because the velocity of the electrons at the edge is proportional to the slope of the dispersion, the edge states become gradually delocalized upon increasing the strength of bending. I. INTRODUCTION ples of graphene spirals.18 -- 20 Many of graphene's remarkable features stem from two facts. The first is that its low energy quasiparticles are linearly dispersive and can be effectively described as Dirac fermions;1 the second is that graphene is a two- dimensional ultrathin membrane that holds promises to revolutionize the current nanotechnology.2 In addition, this 2D membrane can be cut into 1D structures, so- called graphene nanoribbons (GNRs), which exhibit dif- ferent transport properties, depending on their termina- tion. Armchair terminated GNRs are usually gapped and therefore insulating. By virtue of their band gap, they can be used to create field-effect transistors.3,4 On the other hand, zigzag GNRs (ZGNRs) show local- ized edge states that may be spin-polarized.5 Although armchair-type GNRs have been successfully synthesised using bottom up approaches,6 -- 9 ZGNRs still remain elu- sive. Recently, patterned graphene with zigzag edges10 and GNRs with mixed armchair and zigzag terminations extending through a few lattice constants11 have been ex- perimentally realized. Despite the fact that the edges are not completely of zigzag type, they turned out to be of sufficient quality to confirm the prediction that the edge states become magnetized.12,13 Besides the geometrical confinement, another research area that has attracted much attention recently is the study of elastic deformations in graphene. Interest in this topic originated mainly from the theoretical predic- tion that strain may couple to the Dirac fermions as a pseudo-magnetic field (a magnetic field that preserves time-reversal symmetry). The subject was initially stud- ied in the light of deformations in carbon nanotubes.14 After the rise of graphene, this research direction grew in prominence by the vision of using strain as a way to tune graphene's properties and use it in develop- ing an all-graphene electronics.15 The pursuit of strain engineering16 was pioneered by the experimental obser- vation of "pseudo"-Landau levels in strained graphene,17 and has been recently corroborated by fascinating exam- In this paper, we study graphene systems that are both geometrically confined and strained, thus combin- ing the two research areas through a specific example: in-plane bent GNRs. These systems have been theo- retically investigated using a model based on density- functional theory.21 In addition, they have been proposed as a graphene geometry where strain couples as a uni- form pseudo-magnetic field.22 Recently, these systems have been experimentally realized by pushing a GNR with the tip of a scanning tunneling microscope.23 Al- though the experimentally synthesized bent samples are armchair terminated, here we concentrate on in-plane bent ZGNRs and study the dependence of the electronic behavior on the bending angle. We furthermore inves- tigate the dependence of the electronic structure on the type of bending. Our studies complement recent investi- gations of the mechanical properties of these systems.24 As a main result, we find that the bending leads to an increased dispersion in the otherwise almost flat edge states. The bending breaks the symmetry between the inner and the outer edges, causing an effective compres- sion of the inside and elongation of the outside edge. To distinguish the contribution of each edge state to the dis- persion, we compare our findings to straight ZGNRs un- der compressive or tensile strain. Our results show that by tuning the bending angle, the edge states become dis- persive and hence delocalized. From a more abstract perspective, we can view the de- formations in graphene in terms of Japanese paper-art. Within this analogy, straight GNRs emerge as the art of paper cutting graphene. On the other hand, origami, the traditional Japanese art of paper folding, is connected to the study of strain in graphene. These two come together in graphene kirigami,24 in which cutting and folding are combined. Here, the cutting refers to the specific ter- mination of the GNR, as well as to the fact that the hexagonal unit cells are empty (cut), and can thus be deformed in a variety of ways. A bent GNR is a very spe- cific and not so complicated type of graphene kirigami, but precisely due to its relative simplicity, it is possible to study its electronic properties in depth. For this reason, the system is a good probe to understand how the elec- tronic behavior arising from confinement and termination is affected by strain. Therefore, we may generalize the notion of kirigami to more complicated graphene nanos- tructures, and apply a similar approach to understand their electronic properties. Knowing what is to be ex- pected in this simple case, may help us understand more complicated situations. This paper is organised as follows. In section II we introduce a tight-binding model with exponentially de- caying hopping and overlap parameters, that we argue is suitable to study confined strained graphene systems. To the best of our knowledge, this particular tight-binding model has not been used previously to study GNRs, but turns out to capture all the relevant features of the band structure. We then introduce two types of bend- ing, which allow us to optimize the computational cost. In section III, we apply our model to study the effects of bending on the edge state and find that their localiza- tion can be tuned by bending. Conclusions are provided in section IV. II. A MINIMAL TIGHT-BINDING MODEL FOR BENT GNRS A. Three-parameter tight-binding model for strained confined graphene systems The electronic structure of graphene is usually derived using a tight-binding model with one pz-orbital per site. If we assume a graphene system with n sites positioned at ri, the single-electron wavefunction is given by ψ(cid:105) = ci φi(cid:105) . (1) i=1 φi(cid:105) are in general site-dependent basis states, Here, which are assumed to be normalized. The vector c = (c1, . . . , cn)T thus completely specifies the electron state. The Schrodinger equation can then be reduced to the n × n matrix equation (SE + T )c = ESc, (2) where E is the energy associated with the state speci- fied by c. Here, we have split the Hamiltonian matrix H, the elements of which are given by Hij = (cid:104)φi Hφj(cid:105), into the so-called hopping matrix (T ), the diagonal on- site energy matrix (E), and the overlap matrix (S), such that H = SE + T . The elements of the overlap ma- trix are given by Sij = (cid:104)φiφj(cid:105). The matrix E is diago- nal, with the elements corresponding to on-site energies, E φj(cid:105) = j φj(cid:105), as also defined in Ref. 25. Note that a more standard convention defines the on-site energy as the expectation value of the energy in a certain state and the hopping matrix as non-diagonal elements of the n(cid:88) 2 Hamiltonian matrix. However, the convention used here allows us to treat the on-site energy as a simple shift in E if the on-site energy is the same for each state. In general, we now have n(n + 1) parameters, the ele- ments of the matrices. In tight-binding, these parameters may be found by fitting to a reference calculation, rather than calculating them explicitly as integrals over basis functions. However, a model with n(n + 1) parameters is impossible to fit when n is not very small. Therefore, additional assumptions are made in order to reduce the parameter space. In graphene, translational symmetry allows one to use periodic boundary conditions. Since there is no longer a difference between individual sites, the on-site energy, hopping, and overlap parameters be- come site independent. A common procedure is to consider a two-parameter model that only takes nearest-neighbor (NN) and next- nearest-neighbor (NNN) hopping into account, and as- sumes orthogonal basis states. In this case, the site- independent on-site energy 0 is left unspecified, as it leaves the eigenvectors invariant and produces only an absolute shift in the spectrum.1 However, we are inter- ested in a model for the graphene system that can de- scribe a bent GNR. For such a model, we have to specify the dependence of the hopping and overlap parameters on the distance, and, at the same time, the parameters of the model should not change when the system is geometri- cally confined, e.g. when graphene is confined to a GNR. This last condition would allow us to fit the parameters to a graphene reference calculation and not to a reference calculation for the specific GNR we study. We find that instead of the usual convention, a non-orthogonal model better satisfies these two conditions. First, we introduce the model and later argue why it compares positively to an orthogonal model. The tight-binding model we use is based on non- orthogonal site-independent basis states, which in real space are given by (cid:104)rφi(cid:105) = φ(r − ri). Next to that, we assume that the hopping and overlap parameters be- tween these states are such that tij = t(rj − ri) and Sij = s(rj − ri) where s, t are exponentially decaying functions, given by (cid:40) (cid:40) t(r) = s(r) = t0eκ(1−r/a) 0 s0eκ(1−r/a) 1 r > 0 r = 0 r > 0 r = 0 , . (3) Here, a is the NN distance of graphene and t0 and s0 are the values of the NN-hopping and overlap parameter, respectively. Note that the on-site hopping parameter is zero and that the overlap of an orbital with itself is one. The dimensionless constant κ determines the fall-off rate of the hopping. Although this procedure introduces a discontinuity in the overlap that cannot be physically realistic, we will assume that the strain sizes are small enough, such that this effect can be neglected. We fur- ther assume that the hopping and overlap parameters are proportional to each other, which implies that the parameter κ is the same for both. This model satisfies the first condition we mentioned, a dependence of the hopping and overlap parameters on the distance, better than an orthogonal model. This can be seen by noting that in studies of strained graphene, exponentially decaying functions have been used for pa- rameters corresponding to orthogonal basis states.26,27 However, efforts to reproduce the asymmetric band struc- ture of graphene using up to 20 fitted hoppings have re- sulted in subsequent parameters sometimes having op- posite signs and clearly not following a trend that can be described with an exponential decay.28 On the other hand, if we relax the orthogonality condition, hopping and overlap are approximately exponentially decaying.29 When overlap is ignored in our parametrization (s0 = 0), the model would be reduced to the one used in Ref. 26. Such a model does not reproduce the correct particle-hole asymmetry. Nevertheless, for low energies the overlap becomes less important and it would yield a good esti- mate of the spectrum. Orthogonal models which involve a non-exponential dependence on distance have also been used. Ref. 21, for instance, introduces a separate linear dependence for both the NN and NNN hopping. One reason why this model is disadvantageous is that it has four fitting parameters instead of three, as in our case. An even more important reason for adopting the non- orthogonal approach is that these parameters are less dependent on the specific confinement than orthogonal parameters, thus better satisfying the second condition. To understand this, we note that in a quantum-confined graphene system we cannot expect all the hopping pa- rameters to have the same value as the bulk parameters, since now the edge needs to be taken into account. For or- thogonal states this is due, in part, to the fact that these states are a linear combination of pz-orbitals obtained us- ing an orthogonalization scheme, like the Lowdin one.30 These states are not the same on the edge and in the bulk, which also results in a difference of on-site energy and hopping between bulk and edge. Therefore, it is more realistic to assume non-orthogonal basis states for the tight-binding model. This allows us to get the param- eters from fitting to a graphene reference calculation and then apply it to the specific confined structure in which we are interested. A model based on nonorthogonal-basis states would be more universal than an orthogonal one for that reason. In Ref. 21, an orthogonal tight-binding model is used and indeed we see that different hopping values are assumed for different GNRs: NNN hopping is zero for AGNRs and non-zero for ZGNRs. A more pre- cise way to treat the edge effect requires the introduction of a different hopping at the edge.31,32 However, for the sake of simplicity, we neglect this effect here. We have argued that the parameters of the model can be obtained by fitting to a reference graphene spectrum. In the periodic graphene case, Bloch's theorem is used to reduce Eq. (2) to a 2×2 matrix equation, with wave func- tions labeled by the wavevector k in the Brillouin zone 3 FIG. 1. (Color online.) Plots of the dispersion relation for graphene along the line connecting the Γ− M − K − Γ points of the Brillouin zone. The green-dashed curve corresponds to the two-parameter orthogonal dispersion of Ref. 1, where the NN and NNN hopping parameters are t = −3.00236 eV and t(cid:48) = 0.20509 eV, and 0 has been chosen such that the K points are at zero energy. The blue-dashed curve depicts the orbital dispersion with exponentially decaying hopping and overlap parameter, described by Eq. (3), with values t0 = −2.8 eV, s0 = 0.2 , κ = 2.6, and 0 = −1.28 eV, chosen such that the zero energy is at the K points. The red-dashed curve corresponds to an orthogonal-basis dispersion taking into account the first 15 hoppings,28 and the black-solid line corresponds to DFT calculations made using the Quantum- Wise software and a Huckel type basis set.33 of graphene. In that case, the solution of this equation is equivalent to the one described in Ref. 29. By fit- ting to a reference first-principle spectrum, we find that t0 = −2.8 eV, s0 = 0.2, and κ = 2.6 gives a reasonable match, which is also not very far off from the parame- ters used in Ref. 29. Although a more elaborated fitting method would allow us to find parameters that reproduce the reference spectrum more closely, we settle with these because we are mostly interested in global features and not in extremely precise quantitative results. The dispersion of graphene along a line connecting high-symmetry points of the Brillouin zone is shown in Fig. 1. In this figure, different graphene dispersions ob- tained from different models are compared. One can ob- serve the results obtained from our three-parameter non- orthogonal model (blue-dashed line), the two-parameter orthogonal model of Ref. 1 (green-dashed line), and an orthogonal model where the first 15 hopping parameters of Ref. 28 are used (red-dashed line). The figure also de- picts the energy dispersion from a first-principle calcula- tion of graphene that was made using the QuantumWise software (black-solid line).33 From the figure, we can ob- serve that the 15 parameter orthogonal basis model re- produces very well the dispersion relation obtained by first-principle calculations. The two-parameter orthogo- nal and three-parameter non-orthogonal models capture the essential features, but differ markedly at the M point for the chosen parameters. This is not surprising, as it has been shown that the behaviour around the M point is strongly influenced by higher-order hoppings.34 4 (cid:88) This symmetry can be seen as a type of modified pe- riodic boundary condition.21 Because the Hamiltonian commutes with the rotation operator by an angle θ1d, we write a 1D Bloch-type wavefunction for a bent GNR in terms of a continuous quantum number k. In real space, Eq. (1) then assumes the form ψΘ,k(r) = ei(cid:96)kcΘ,k m φ(r − R−(cid:96)θ1d r(cid:48)1d m )). (5) (cid:96),m m = FΘ(r1d Here, k ∈ [0, 2π], m runs over the atoms in the bent unit cell r(cid:48)1d m ), and (cid:96) runs over the number of unit cells in the ribbon. The vector cΘ,k = (cΘ,k 2N )T therefore completely determines the electron state for a certain wavevector k and bending parameter Θ. Namely, the components cj of Eq. (1) are given by cj = ei(cid:96)kcΘ,k m , with j related to (cid:96), m such that rj = r1d m + (cid:96)a. From the time-independent Schrodinger equation (2), we can derive a matrix equation for the vector of orbital compo- nents cΘ,k, , . . . , cΘ,k 1 (SΘ,k)−1T Θ,kcΘ,k = (EΘ (6) Here, Tk and Sk are 2N × 2N matrices with components k − 0)cΘ,k. (cid:88) (cid:88) (cid:96) eik(cid:96)s(cid:0)R−(cid:96)θ1d r(cid:48)1d eik(cid:96)t(cid:0)R−(cid:96)θ1d r(cid:48)1d n m − r(cid:48)1d m − r(cid:48)1d n (cid:1), (cid:1), SΘ,k mn = T Θ,k mn = (7) (cid:96) where t and s are defined as in Eq. (3) and EΘ k is the spectrum of the eigenstates. In our calculations, we use the values for s0 and κ derived from graphene. The on- site energy is set to zero, giving a Fermi level close to, but not exactly at zero. After the calculation, the spectra are shifted by an amount 0 to place the Fermi level at zero. As can be seen from Eq. (6), the dispersion scales linearly with t0 when the scale is normalized around the Fermi level, and we can thus calculate the dispersion in terms of t0 without having to explicitly specify its value. The tight-binding model using non-orthogonal basis and exponentially decaying hopping and overlap in combina- tion with lattice-preserving bending may be used as a minimal model to study bent GNRs because it only re- quires three parameters and equations with matrices of size 2N × 2N . C. Two types of bending A realistic geometry for a bent GNR may be extracted from a molecular dynamics simulation, where bending affects both bond lengths and bond angles. The exact type of bending then depends on the ratio of the spring constants of the respective deformations. From previous work, it is known that the bond length in the graphene lattice is much stiffer than the bond angle.35,36 This observation prompts us to explore a limiting scenario, FIG. 2. (Color online.) Parameters characterizing the straight and the strained GNR. The parameters L and W characterize the straight ribbon, while the inner and outer radius Rin and Rout specify perfectly circular bending. Here, we also define the bending radius R along the center of the rib- bon, the width of the ribbon after bending W (cid:48) = Rout − Rin, the length L(cid:48) along the center of the bent ribbon, the length of a 1D unit vector a(cid:48) along the center of the bent unit cell, the curvature θ1d of a unit cell, the total curvature θfull of the ribbon and the bending parameter Θ = W/2R. The region enclosed by the dotted line is the 1D unit cell of the bent ribbon. B. Lattice-preserving bending To find a minimal model that can describe the ge- ometry of bent GNRs, we first introduce the concept of lattice-preserving bending. This type of deformation can be described by the parameters defined in Fig. 2. We quantify the degree of bending using the dimensionless parameter Θ, defined as Θ = W/2R, where W is the width of the undistorted ribbon and R is the radius of the circular deformation. For W (cid:48) ≈ W , this is approximately equal to the parameter used in Ref. 21. For straight rib- bons, we can define a 1D unit cell with sites labelled by m}, and a 1D lattice vector a. All sites m, given by {r1d can then be decomposed in ri = r1d m + (cid:96)a for some num- ber (cid:96). This allows us to reduce the size of the matrices in Eq. (2) using the 1D Bloch's theorem to 2N × 2N , with N the number of dimer lines of the ribbon (num- ber of sublattice pairs in the unit cell, which is always even for ZGNRs), see for instance Ref. 13. However, the 1D translational symmetry that allows this procedure is broken after bending. A lattice-preserving bending is a type of bending deformation that still allows us to reduce the matrices in Eq. (2) to size 2N × 2N . This is possi- ble because a lattice-preserving bending FΘ satisfies the discrete rotational symmetry FΘ(ri + a) = R−θ1d FΘ(ri), (4) where R−θ1d is the matrix that represents a clockwise rotation by angle θ1d, and ri and a are the lattice sites and 1D lattice vector of the straight ribbon, respectively. θfullRinRRoutW'L'a'L'unit cellθ1dLWΘ=W/2R=0.165 It is not straightforward to give a closed formula for the bond length-preserving bending. However, we can construct the profile of the deformation by applying FN N (Ri, Θ) on specific ribbon sites Ri recursively, see Appendix. The bondlength-preserving bending is sim- ilar to the width-preserving one, but has a non-linear strain profile from the bottom to the top of the ribbon. At the inner edge, the ribbon experiences not only lon- gitudinal compressive strain, but also transverse tensile strain. At the outer edge, on the other hand, a com- pressive transverse strain is present. It is also important to note that the total width becomes reduced, as can be seen in Fig. 3. This reduction of width needs to be taken into account when comparing effects of the bond length-preserving with the width-preserving bending. As a consequence of the reduction of width, the longitudinal strains at the inner (in) and outer edge (out) are not identical for the two types of bending. III. RESULTS: TUNABLE EDGE STATE DISPERSION We have calculated the dispersion relation for bent ZGNRs by solving Eq. (6) numerically both for width- preserving and for bond length-preserving bending. In Fig. 4, the dispersion relation for different values of the bending parameter is depicted. Since we argued that bending introduces a profile of elastic deformation with effective compressive strain on the inside and ten- sile strain on the outside, it is useful to compare it to the effects of a uniform longitudinal strain , defined as = ∆L/L, where ∆L is the length deformation intro- duced by the strain, and L is the length of the unde- formed nanoribbon. Fig. 5 depicts the effect of positive (tensile) and negative (compressive) longitudinal strain on a N = 4 ZGNR subjected to a width-preserving uni- form strain deformation. We can see that the energy of the edge states increases (decreases) for negative (pos- itive) strain. When we compare the two cases with a ribbon bent using width-preserving bending, we observe that the energy increase in the edge state that experi- ences compression is roughly equal to the energy increase in both edge states of a longitudinally compressed rib- bon. Similarly, we find a good agreement for the outer edge state with both edge states of a ribbon experiencing tensile strain. These observations indicate that the dis- persion of ribbons bent by Θ is quantitatively related to the dispersion of a uniformly strained ribbon with strain = ±Θ, a result consistent with Ref. 21. Plotting the wavefunctions of the edge states confirms that the low-energy state resides on the outside, as shown in Fig. 6. Here, the orbital components of the eigenstates of the edge states, cj, as defined in Eq. (5), are plotted for increasing Θ. The width-preserving bending scheme was used in generating the plots. First, we note that the edge states are localized on one sublattice at both edges, forming a symmetric and antisymmetric combination of FIG. 3. (Color online.) A bent ZGNR with N = 5 and Θ = 0.15 for bond length-preserving (red dots) and width preserving (blue dots). The width of the ribbon after bond length-preserving bending is W (cid:48) B and after width-preserving bending is W (cid:48) W . where bending is completely absorbed in bond-angle dis- tortions, and which we call bondlength-preserving bend- ing. In addition, we consider a distortion which we call width-preserving bending, where the atomic positions are rotated around a concentric point. The width-preserving bending is the same deformation as has been used in Ref. 22. Notice that the bond length-preserving bend- ing obeys the rules of graphene kirigami, since the paper can be folded (bond-angle deformations), but it cannot be strained (bond-length deformations). The fixing of the bond lengths in the bond length-preserving bending leaves the NN hopping unchanged, so that any perturba- tion in the electronic structure can mainly be ascribed to modifications of the NNN hopping. In contrast, bond lengths are allowed to change in the width-preserving bending scheme, so it may be expected that the changes in the dispersion are mainly due to changes in NN hop- pings. Comparing the effects of these two types of bend- ing on the spectrum, therefore, allows us to decouple the effects of NN and NNN distortions. Both bending deformations are depicted in Fig. 3. We can explicitly describe the width-preserving bending by the deformation function (cid:18) sin(rx/R) cos(rx/R) (cid:19) Fw(r, Θ) = (ry + R) , (8) with r = (rx, ry). This deformation assumes that the ribbon is positioned such that the middle of the GNR is on the x-axis and the ribbon lies in the xy-plane. Hence, the y coordinate of the undeformed site is in the inter- val [−W/2, W/2]. One can easily verify that this bend- ing satisfies the definition of a lattice-preserving bending Fw(r + a1d, Θ) = R−θ1d Fw(r, Θ). This deformation is a width-preserving bending in the sense that the distances between sites in the direction along the width of the GNR remain unchanged. Another feature of this bending is that the strain in the direction along the ribbon width increases linearly from the inner to the outer edge. This, in conjunction with the fact that the bending considered here equally compresses on the inside as it stretches out- side, yields a line of zero stress exactly in the middle of the ribbon. W'BW'wstates localized on either edge. The states are almost degenerate, which would allow us to form orthonormal combinations that are still eigenstates of the Hamilto- nian with the same eigenenergy. In fact, since interac- tion effects arising from the Coulomb repulsion are not accounted for, we may expect these effects to favor a dif- ferent combination in the two-dimensional Hilbert space of eigenstates. Intuitively, the effect of electron-electron repulsion should be to split the symmetric and antisym- metric states into two states that are localized on their respective edges, both singly occupied. With increasing bending, we observe that the nearly degenerate states that initially reside on both edges in our model transform into a high-energy state localized on the inner edge and a low-energy state localized on the outer edge. It is interesting to note that this already occurs for the very small bending parameter of Θ = 10−4, indi- cating that for this strength of bending, the symmetric FIG. 4. (Color online.) Dispersion relation and DOS for a N = 4 ZGNR as a function of the bending parameter Θ for width-preserving bending (blue) and bond length-preserving bending (red). The thin-black line corresponds to a straight ribbon. On the left side, we show the spectrum over the complete BZ, for Θ varying from 0 at the top to 0.15 at the bottom panel, with steps of 0.05. On the right side, we zoom in on the edge state with k ranging from 2π/3 to 4π/3 and depict the lattice of the GNR. All plots have the same scale as shown in the bottom. Calculations were made using non-orthogonal parameters with exponential decay, given by Eq. (3), with s0 = 0.2 , κ = 2.6, and 0 such that the Fermi energy (dotted line) of the straight ribbon lies at zero. The DOS is calculated using a Lorentzian broadening with a width of 0.03 eV (DOS in arbitrary units). 6 and antisymmetric states mix in order to form the states localized on a single edge, energetically more favorable. We also find a significant dependence of the localization length of both edge states on the momentum k. When we plot, for example, the edge states for a wave vector of k = 7π/8, the wave function appears to spread more into the bulk of the ribbon than for the value k = π, as shown in Fig. 6. Although not shown here, the results for bond length-preserving bending show that the edge state for k = 7π/8 is also less localized than for k = π. However, for the same degree of bending, the effect is much less pronounced than for width-preserving bending. Another striking observation is that the two edge states do not only split but also develop opposite curvature, as shown in Fig. 4. The top band is curved upward, but at its center a small downward curvature develops, such that there is a local maximum at k = π, whereas the opposite occurs for the lower band. This is in contrast with what we observe for positive or negative uniform strain in Fig. 5. In that case, the edge states are only shifted, but retain the same dispersion as in the strain- free ribbon. A minimal model that captures this behaviour, and in particular fits the dispersion of the edge states around the point k = π, is a tight-binding model of a 1D chain of sites with a single NN hopping th/l and an on-site energy h/l. Here, the superscripts refer to the higher- energy band and lower-energy band, which are localized on the inner and outer edge, respectively. The effective dispersion obtained from the 1D NN tight-binding model reads Eh/l k (Θ) = h/l 0 (Θ) + 2th/l(Θ) cos(k). (9) Inspection of the zoomed in panels of Fig. 4 suggest that this effective model can describe the shape of the bands in the region around k = π reasonably well. A positive or negative th/l relates to the dispersion that exhibits, respectively, an upwards or downwards curvature around momentum k = π. Before we compare this effective model quantitatively with the tight-binding results, we need to mention the effect of the width of the ribbon on the edge states. As a ribbon becomes narrower, the edge state localized on one side with k closer to k = π starts to hybridize with the edge state localized on the other edge. On the other hand, when one starts bending a ribbon the edge states start moving closer in energy to the bulk states. This can be seen in Fig. 4. After a certain bending, the va- lence band maximum hybridizes with the lower-energy edge states, as well as the conduction band minimum hy- bridizes with the higher-energy edge states. Since wider ribbons have a smaller bulk band gap, these effects are more pronounced. These effects are shown in Fig. 7, where we plot the lower-energy edge state for three dif- ferent widths of the ribbon, N = 4, 14, and 30, using the same bending parameter, Θ = 0.1, and k−value k = 6π/8. We observe that the two edge states of the N = 4 ribbon hybridize with each other, and are there- E(t0)k02π/34π/32πk2π/34π/3Θ=0.05Θ=0Θ=0.1Θ=0.15Θ=0Θ=0.05Θ=0.1Θ=0.150-2240-0.20.2DOS7 FIG. 5. (Color online.) Dispersion relation for a N = 4 ZGNR for uniform strain = −0.1 (left panel, red line) and = 0.1 (middle panel, blue line). The thinner-black line in the left and middle picture corresponds to the straight ribbon. The right panel shows the dispersions for uniform strain = −0.1 (red), = 0.1 (blue), and width-preserving bending with Θ = 0.1 (black). Calculations were performed using non-orthogonal parameters with exponential decay, given by Eq. (3) with s0 = 0.2 , κ = 2.6, and 0 such that the Fermi energy (dotted line) of the straight ribbon lies at zero. fore not localized anymore. The edge states of the ribbon with N = 30 also hybridize, but instead with bulk states, and are not localized anymore either. The ribbon with N = 14, however, still shows localized edge states for the same regime of parameters. These two opposite effects make the comparison between different ribbon sizes very intricate. We have chosen to analyze the N = 14 ribbon in more detail because this one has the optimal width to avoid spurious hybridization effects of the first or sec- ond kind. Our observations are expected to hold also for ribbons of different width, if care is taken to account for these hybridization effects. We fit the parameters of the effective 1D dispersion of Eq. 9 to the tight-binding calculations for a ribbon of width N = 14. In Fig. 8, we plot the fitted parameters for different values of the bending Θ. We observe that both the lower- and the higher-energy edge states start out with the same positive hopping parameter. Interest- ingly, in both bending schemes, th crosses zero, implying that for a certain bending parameter the band becomes dispersionless. This is an important observation because many-body effects can be expected to become even more relevant for that bending parameter. By comparing how the parameters change with respect to the type of bending used we can identify whether the NN or the NNN hopping is more important. The effective parameters for the state on the inner edge decrease for both types of bending. However, for the outer edge the effective parameters increase for bondlength-preserving bending, but decrease for width-preserving bending. The main difference between the two bending methods is that in the width-preserving bending also the NN distance is modified. Therefore, we can conclude that for the outer- edge state NN effects are more important than for the inner-edge. General behavior of the inner-edge state, however, can be captured by only considering the effect of the NNN hopping. If we compare the effective param- eters for the inner-edge state between the two bending methods in more detail, we observe that the effective pa- rameters for bond length-preserving bending show a lin- ear dependence on Θ, while this dependence for width- preserving bending is more complicated. One reason for this behavior could be the fact that the width- and bond length-preserving bending produce a small difference in strain on the edges (in, out). To check whether this can account for the difference, we also plot the effective parameters as a function of the strain (smaller plots in Fig. 8). We can clearly see that the general behavior does not change. Therefore, the difference should be sought in effects of the NN hopping. Changes in the NN distance influences the hybridization between the opposite edges and the hybridization of the edge state with bulk states. These effects might explain why the effective parameters of width-preserving bending exhibit a nonlinear depen- dence on the bending. Furthermore, the effect of the per- turbation of the NN distance also depends on the width of the ribbon, which additionally complicates the problem. Because of all this, in the following we focus only on the effective parameters of bond-length-preserving bending. For bond length-preserving bending (see plots in red in Fig. 8), the effective hopping at the inner edge (higher-E) linearly decreases and changes sign, whereas the hopping at the outer edge (lower-E) linearly increases. We could try to understand this behaviour by assuming a perfectly localized edge state. The inner edge experiences a neg- ative strain, so the hopping becomes more negative and the 1D dispersion would curve downwards. This indeed corresponds to what we observe in Fig. 8. On the same token, the hopping at the outer edge should decrease, be- cause the distances between the lattice sites increase, and therefore a flat band should develop. However, the oppo- site behaviour is visible in Fig. 8. This can be understood by noting that the changes due to bending at the outer edge are determined by the weight of the wavefunction on sites closer to the bulk. This is because these sites are closer to each other and therefore contribute more to the energy. This together with the fact that sites close to the bulk have a sizeable weight implies that our assumption k02π/34π/32πk02π/34π/32πk02π/34π/32πε=0.1ε=-0.1, ε=0.1, Θ=0.1 E(t0)ε=-0.102π/34π/32πk02π/34π/32πk02π/34π/32πε=0.1ε=-0.1,ε=0.1,Θ=0.1E(t0)ε=-0.10-224of the localized states does not apply. The fact that the edge state becomes less localized as the momentum moves further away from k = π is crucial here. This enhances the effect that can already be seen for straight ribbons, where the edge states are dispersive at the momenta away from k = π, and causes an increasing positive effective hopping. In conclusion, we can understand the behaviour as a competition between two effects due to NNN hopping and strain: For the edge state localized on the outer edge, the first effect is always dominant and becomes even more rele- vant after bending. For the inner edge, the second effect overcomes the first after a certain bending parameter. This is the reason why the dispersion of the inner edge has to go through a point at which it is dispersionless. This also clarifies our earlier observation that the outer edge state is more sensitive to changes in the NN hop- ping. The outer edges are more delocalized, and therefore more sensitive to the effects of the NN hoppings. 8 1. An effective positive hopping for increasing neg- ative strain because of the increasing delocalized nature of the edge state as the momentum moves further away from k = π. 2. An effective negative hopping for increasing nega- tive strain because the edge state is localized. FIG. 6. (Color online.) Edge states in the real space. Compo- nents c(cid:96),m of edge states for a section of the ribbon at k = π and k = 7π/8 are mapped to the corresponding ri points of the bent GNR for different values of bending-parameter Θ. The width-preserving bending scheme was used here in com- bination with our standard orbital hopping parameters. c(cid:96),m are related to the eigenvector through the definition Eq. (5) and satisfy Eq. (6). The coefficients c(cid:96),m are complex num- bers that are depicted in two ways. The first is by dots of which the diameter is proportional to the absolute value of c(cid:96),m and the color corresponds to the phase, as indicated by the color code. Additionally the coefficients c(cid:96),m are repre- sented by a vector in the complex plane (see small black lines at the center of the dots). The phase is chosen such that the lattice site at the left bottom of the picture has phase zero. The top (bottom) rows of the k = π and k = 7π/8 panels correspond to the high-energy (low-energy) edge states. The ribbon has width N = 6. FIG. 7. (Color online.) Real space depiction of lower energy edge states for ribbons of with N = 4, N = 14, N = 30. Construction is the same as explained in Fig. 6. For all three ribbons the bending parameter is Θ = 0.1 and k = 6π/8. The N = 4 ribbon shows no clear edge states because the states are hybridized across the entire ribbon, while the N = 30 ribbon shows localized edge states, which are, however, hybridized with ones in the bulk . k=πΘ=0.0001Θ=0Θ=0.2k=7π/8Low Eedge stateHigh Eedge stateLow E edge stateHigh E edge stateColor code for complex phase φφ=0φ=π/2φ=πφ=3π/2N=14N=4N=309 FIG. 8. (Color online.) Best fit values of the effective 1D chain parameters th, h 0 (lower rectangular panel) versus the bending parameter Θ. These parameters are defined in Eq. (9). The fitting has been performed for width-preserving bending (blue dots) and bond length-preserving bending (red dots). The right picture in each panel shows these parameters with respect to the strain on the inner edge in for the higher-energy edge state and with respect to the strain on the outer edge out for the lower-energy edge states. The relation between in, out and Θ is explained in the text. The fitting was done for a ZGNR of width N = 14, and is based on data points chosen in the region around k = π given by k ∈ [2.41, 3.86]. Error bars are obtained from the standard deviation between the fitted spectra and the numerics on the lattice. All calculations are performed using our standard set of exponentially decaying orbital hopping parameters. 0 (upper rectangular panel) and tl, l IV. CONCLUSIONS We show here that a tight-binding model with ex- ponentially decaying hopping and overlap can be used as a minimal model with three parameters to study a graphene-based system that is both geometrically con- fined and strained. To obtain geometries of bent nanorib- bons that serve as the input of the tight-binding model, we develop two types of bending, bond length-preserving and width-preserving. We would like to point out that bond-length preserving bending geometry, generated us- ing a recursive algorithm, shows a particularly strong analogy with the Japanese art of kirigami. Both types of bending are lattice-preserving, causing the resulting geometry to exhibit rotational symmetry (the unit cell is rotated by θ1d to generate the entire bent GNR), and therefore allowing us to reduce the tight-binding model to the numerically inexpensive problem of solving a matrix equation with 2N × 2N matrices, with 2N the number of sites in the unit cell of the GNR. The different types of bending allow us to decouple the effects of perturba- tions of the NN and NNN parameters of the tight-binding model. We have investigated the qualitative features of the dispersion relation upon bending. Our calculations show that bending leads to nontrivial effects on the edge states of ZGNRs, resulting from the broken symmetry between the top and bottom edges. We observe that both width- preserving and bond length-preserving bending predict a splitting of the two edge states (without considering in- teractions). A lower-energy edge state localizes on the outer edge and a higher-energy edge state on the inner edge. In fact, there is an emergent band structure around the point k = π of the edge states that can be fitted to the tight-binding dispersion of a 1D chain with an effec- tive hopping and on-site energy parameter. The higher- energy edge state has an effective hopping parameter that changes sign as the bending is cranked up from Θ = 0.11 to Θ = 0.17, with the exact value where the effective hop- ping vanishes depending on the type of bending. Hence, there is a critical degree of bending at which the band is effectively flat and interaction effects are expected to become prominent. Since the charge carrier velocity is proportional to the slope of the dispersion, the degree of localization of the edge states can be tuned with bend- ing. By comparing the two bending methods, we can conclude that effects on the dispersion of the inner-edge state are dominated by changes in NNN hopping. For the outer edge state, changes in NN hopping also become important. The effects due to NN hopping changes, how- ever, are less universal and depend on width and bending method. The effects of the NNN hopping on the emergent band structure at the edges can be explained by a com- petition between the decreasing localization of the elec- tronic states with the momenta away from k = π and the εh0(t0)εoutεoutΘhigher E edge state hopping (N=14)εl0(t0)th(t0)εin0-0.05-0.1-0.15-0.2higher E edge state on-site energy (N=14)lower E edge state hopping (N=14)εh0(t0)tl(t0)ΘεoutεoutΘΘhigher E edge state hopping (N=14)εl0(t0)th(t0)εin0-0.05-0.10-0.15-0.2εin0-0.05-0.10-0.15-0.2higher E edge state on-site energy (N=14)lower E edge state on-site energy (N=14)lower E edge state hopping (N=14)00.050.100.150.200.20.4-0.2-0.400.20.4-0.2-0.4Θ00.050.100.150.200.20.4-0.2-0.4εin0-0.05-0.1-0.15-0.200.20.4-0.2-0.40-0.05-0.1-0.15-0.20-0.05-0.1-0.15-0.20.40.60.80.20lower E edge state on-site energy (N=14)Θtl(t0)00.050.100.150.20.40.60.80.20Θ00.050.100.150.20.40.60.80.200.40.60.80.20localized character of the edge state. A next step would be to include interaction effects, as these are important for edge states, especially when the bending gives rise to the flat bands. Furthermore, motivated by our work, it would be important to understand how bending would affect the magnetic polarization of the edge states de- tected recently.12 We hope that our results will stimulate further research in these directions. V. ACKNOWLEDGMENTS. We are grateful to Ingmar Swart for fruitful discus- sions on electronic structure theory, graphene nanorib- bons, and their interplay with scanning probe microscopy experiments. APPENDIX Recursion formula for bond length-preserving bending of ZGNR We construct the bond length-preserving bending, FN N , for a ZGNR. First, we construct the bent 1D unit cell. The orientation is chosen such that the first site in the 1 = (0, R− W (cid:48)/2). Note bent unit cell is positioned at r(cid:48)1d that we do not know W (cid:48) and R yet, but they will be obtained using a recursive procedure outlined below. We can now recursively generate the next atoms in the de- formed 1D unit cell using the following rule: (cid:40) r(cid:48)1d m = m−1 if i is even if i is odd, f (r(cid:48)1d (r(cid:48)1d m−1 m−1, θ1d)R−θ1d/2(cid:98)r(cid:48)1d m−1 + a)(cid:98)r(cid:48)1d (cid:113) m−1, θ1d) = r(cid:48)1d (a)2 − r(cid:48)1d f (r(cid:48)1d + m−1 cos(θ1d/2) m−12 sin2(θ1d/2). 10 Here, (cid:98)r(cid:48)1d 2N − r(cid:48)1d m−1 is the unit vector in the direction of r(cid:48)1d m−1. We still assume that the distance along the middle of the GNR remains unchanged, and therefore θ1d = Θa(cid:48)/(W/2). If we follow this recursion until r(cid:48)1d 2N , where N is the number of A sites in the 1D unit cell, we have generated the deformed 1D unit cell r(cid:48)1d m . However, we started with r(cid:48)1d 1 defined in terms of the bent GNR width W (cid:48), which was unknown. We can now use the identity W (cid:48) = r(cid:48)1d 1 , which is an equation with W (cid:48) on both sides, to write out the recursion explicitly. How- ever, this is a rather involved equation. We can, on the other hand, easily find a good approximation iteratively for W (cid:48). We start with the assumption that W (cid:48) ≈ W . Then, after running the recursion, we calculate the W (cid:48) of that ribbon. If it differs by more than a set test value from the previous recursion, we use that value of W (cid:48) to generate a new unit cell. This iterative procedure runs until the test condition, that gives the minimal differ- ence between a new and old width, is satisfied. Note also that this deformation does not work for every Θ, as for large enough bending the square root in the definition will become complex. This is understandable, as there should be a maximum bending at which the lattice sites on the outer edge of the ribbon are all separated by a. Once the bent unit cell is generated, the complete bent GNR is obtained by copying the unit cell through multi- ples of rotations by θ1d. Thus, we can describe the bond length-preserving bending as FN N (r1d m + (cid:96)a, Θ) = R−(cid:96)θ1d r(cid:48)1d m . Here, (cid:96) runs over the number of unit cells in the ribbon. We explicitly use that the lattice sites of a straight GNR can be described by a site in the 1D unit cell plus a multiple of a, the lattice vector of the straight ribbon. One can show, using simple trigonometry, that each site now has 3 neighbors that are at a distance equal to a, as shown in Fig. 3. Due to the construction, it is obvious that the rotational symmetry is satisfied and thus this is a lattice-preserving bending. 1 A. H. Castro Neto, F. Guinea, N. M. R. Peres, K. S. Novoselov, and A. K. Geim, Rev. Mod. Phys. 81, 109 (2009). 2 A. K. Geim and K. S. Novoselov, Nat. Mater. 6, 183 (2007). 3 X. Wang, Y. Ouyang, X. Li, H. Wang, J. Guo, and H. Dai, Phys. Rev. Lett. 100, 206803 (2008). 4 P. B. Bennett, Z. Pedramrazi, A. Madani, Y.-C. Chen, D. G. de Oteyza, C. Chen, F. R. Fischer, M. F. Crommie, and J. Bokor, Appl. Phys. Lett. 103, 253114 (2013). 5 M. Fujita, K. Wakabayashi, K. Nakada, and K. Kusakabe, J. Phys. Soc. Jpn. 65, 1920 (1996). 6 J. Cai, P. Ruffieux, R. Jaafar, M. Bieri, T. Braun, S. Blankenburg, M. Muoth, A. P. Seitsonen, M. Saleh, X. Feng, K. Mullen, and R. Fasel, Nature (Londen) 466, 470 (2010). 7 Y.-C. Chen, D. G. de Oteyza, Z. Pedramrazi, C. Chen, F. R. Fischer, and M. F. Crommie, ACS Nano 7, 6123 (2013). 8 A. Narita, X. Feng, Y. Hernandez, S. A. Jensen, M. Bonn, H. Yang, I. A. Verzhbitskiy, C. Casiraghi, M. R. Hansen, A. H. R. Koch, G. Fytas, O. Ivasenko, B. Li, K. S. Mali, T. Balandina, S. Mahesh, S. De Feyter, and K. Mullen, Nat. Chem. 6, 126 (2014). 9 A. Basagni, F. Sedona, C. A. Pignedoli, M. Cattelan, and M. Sambi, J. Am. Chem. L. Nicolas, M. Casarin, Soc. 137, 1802 (2015). 10 Z. Shi, R. Yang, L. Zhang, Y. Wang, D. Liu, D. Shi, E. Wang, and G. Zhang, Advanced Materials 23, 3061 (2011). 11 P. Han, K. Akagi, F. Federici Canova, H. Mutoh, S. Shi- raki, K. Iwaya, P. S. Weiss, N. Asao, and T. Hitosugi, 11 ACS Nano 8, 9181 (2014). 12 G. Magda, X. Jin, I. Hagym´asi, P. Vancs´o, Z. Osv´ath, P. Nemes-Incze, C. Hwang, L. P. Bir´o, and L. Tapaszt´o, Nature (Londen) 514, 608 (2014). 13 Y. Son, M. Cohen, and S. Louie, Phys. Rev. Lett. 97, 216803 (2006). 23 J. van der Lit, P. H. Jacobse, D. A. M. Vanmaekelbergh, and I. Swart, New J. Phys. 17, 053013 (2015). 24 Z. Qi, D. K. Campbell, and H. S. Park, Phys. Rev. B 90, 245437 (2014). 25 M. O. Goerbig, Rev. Mod. Phys. 83, 1193 (2011). 26 V. M. Pereira, A. H. Castro Neto, and N. M. R. Peres, 14 C. L. Kane and E. J. Mele, Phys. Rev. Lett. 78, 1932 Phys. Rev. B 80, 045401 (2009). (1997). 15 B. Amorim, A. Cortijo, F. de Juan, A. G. Grushin, F. Guinea, A. Gutirrez-Rubio, H. Ochoa, V. Parente, R. Roldn, P. San-Jos, J. Schiefele, M. Sturla, and M. A. H. Vozmediano, arXiv:1503.00747. 16 V. M. Pereira and A. H. Castro Neto, Phys. Rev. Lett. 103, 046801 (2009). 17 N. Levy, S. A. Burke, K. L. Meaker, M. Panlasigui, A. Zettl, F. Guinea, A. H. C. Neto, and M. F. Crommie, Science 329, 544 (2010). 18 S. M. Avdoshenko, P. Koskinen, H. Sevin¸cli, A. A. Popov, and C. G. Rocha, Sci. Rep. 3 (2013). 19 T. Korhonen and P. Koskinen, AIP Adv. 4, 127125 (2014). 20 X. Zhang and M. Zhao, Sci. Rep. 4 (2014). 21 P. Koskinen, Phys. Rev. B 85, 205429 (2012). 22 F. Guinea, A. K. Geim, M. I. Katsnelson, Novoselov, Phys. Rev. B 81, 035408 (2010). and K. S. 27 R. M. Ribeiro, V. M. Pereira, N. M. R. Peres, P. R. Brid- don, and A. H. Castro Neto, New J. Phys. 11, 115002 (2009). 28 J. Jung and A. H. MacDonald, Phys. Rev. B 87, 195450 (2013). 29 S. Reich, J. Maultzsch, C. Thomsen, and P. Ordej´on, Phys. Rev. B 66, 035412 (2002). 30 J. G. Aiken, J. A. Erdos, and J. A. Goldstein, Int. J. Quant. Chem. 18, 1101 (1980). 31 K. Sasaki, Y. Shimomura, Y. Takane, and K. Wak- abayashi, Phys. Rev. Lett. 102, 146806 (2009). 32 Y. Hancock, A. Uppstu, K. Saloriutta, A. Harju, and M. J. Puska, Phys. Rev. B 81, 245402 (2010). 33 M. Brandbyge, J. Mozos, P. Ordej´on, J. Taylor, and K. Stokbro, Phys. Rev. B 65, 165401 (2002). 34 C. Bena and L. Simon, Phys. Rev. B 83, 115404 (2011). 35 C. Li and T.-W. Chou, Int. J. Solids Struct. 40, 2487 (2003). 36 P. Zhao and G. Shi, Tech Science SL. 5, 49 (2011).
1606.00997	1	1606	2016-06-03T07:57:18	The granularity effect in amorphous InGaZnO$_4$ films prepared by rf sputtering method	[ "cond-mat.mes-hall" ]	We systematically investigated the temperature behaviors of the electrical conductivity and Hall coefficient of two series of amorphous indium gallium zinc oxides (a-IGZO) films prepared by rf sputtering method. The two series of films are $\sim$700\,nm and $\sim$25\,nm thick, respectively. For each film, the conductivity increases with decreasing temperature from 300\,K to $T_{\rm max}$, where $T_{\rm max}$ is the temperature at which the conductivity reaches its maximum. Below $T_{\rm max}$, the conductivity decreases with decreasing temperature. Both the conductivity and Hall coefficient vary linearly with $\ln T$ at low temperature regime. The $\ln T$ behaviors of conductivity and Hall coefficient cannot be explained by the traditional electron-electron interaction theory, but can be quantitatively described by the current electron-electron theory due to the presence of granularity. Combining with the scanning electron microscopy images of the films, we propose that the boundaries between the neighboring a-IGZO particles could make the film inhomogeneous and play an important role in the electron transport processes.	cond-mat.mes-hall	cond-mat	The granularity effect in amorphous InGaZnO4 films prepared by rf sputtering method Hui Zhang,1 Xin-Jian Xie,2 Xing-Hua Zhang,2 Xin-Dian Liu,1 and Zhi-Qing Li1, ∗ 1Tianjin Key Laboratory of Low Dimensional Materials Physics and Preparing Technology, 2School of Material Science and Engineering, Hebei University of Technology, Tianjin 300130, China Department of Physics, Tianjin University, Tianjin 300072, China (Dated: September 26, 2018) We systematically investigated the temperature behaviors of the electrical conductivity and Hall coefficient of two series of amorphous indium gallium zinc oxides (a-IGZO) films prepared by rf sputtering method. The two series of films are ∼700 nm and ∼25 nm thick, respectively. For each film, the conductivity increases with decreasing temperature from 300 K to Tmax, where Tmax is the temperature at which the conductivity reaches its maximum. Below Tmax, the conductivity decreases with decreasing temperature. Both the conductivity and Hall coefficient vary linearly with ln T at low temperature regime. The ln T behaviors of conductivity and Hall coefficient cannot be explained by the traditional electron-electron interaction theory, but can be quantitatively described by the current electron-electron theory due to the presence of granularity. Combining with the scanning electron microscopy images of the films, we propose that the boundaries between the neighboring a-IGZO particles could make the film inhomogeneous and play an important role in the electron transport processes. Amorphous indium gallium zinc oxide (a-IGZO), which simultaneously possesses high optical transparency in vis- ible range, high carrier mobility and low temperature fab- rication process, is being considered as the good candi- date of the active channel material in thin-film transistors (TFTs).1 -- 4 Compared with polycrystalline metal oxides, the amorphous oxides do not have grain boundary. The grain boundary could seriously affect the carrier mobility and uniformity of oxide materials, which in turn would influence the performance of TFTs.2,4 -- 6 Hence the amor- phous oxide material is more suitable for TFTs than the polycrystalline one. However, even for the amorphous film, it often experiences the discontinuous to continuous process in the first stage of film growth. Thus the bound- aries between amorphous particles are also inevitable in the amorphous films. In this Letter, we measured the variations in resistivity and Hall coefficient with temper- ature for the thick (∼700 nm) and thin (∼25 nm) a-IGZO films from 300 down to 2 K. We found the electron- electron interaction (EEI) due to the presence of gran- ularity governs the temperature behaviors of longitudi- nal and Hall transport of the a-IGZO films. Combining the scanning electron microscopy (SEM) images of the films, we conclude that the subtle influence of bound- aries between the the neighboring amorphous particles in a-IGZO films on the electronic transport processes can not be neglected. The samples were deposited on glass substrates by rf sputtering method. A commercial InGaZnO4 ceramic target with purity of 99.99% was used as the sputtering source. The atomic ratio of In, Ga, and Zn is 1 : 1 : 1. The base pressure of the sputtering chamber was below ∼1 × 10−5 Pa. The sputtering was carried out in an ar- gon (99.999% in purity) atmosphere with the pressure of 0.6 Pa, and the sputtering power was maintained at 100 W during the deposition process. To obtain films with high conductivity, the substrate temperature was set as 688 and 788 K, respectively. We deposited IGZO films with thickness ∼700 and ∼25 nm at each deposi- tion temperature. The film notations are listed in Ta- ble I. Hall-bar-shaped samples were deposited for the measurements of Hall coefficient RH and conductivity σ, by using mechanical masks. The thicknesses of the films were measured with a surface profiler (Dektak, 6 M) for those films with thickness of ∼700 nm, and with the low- angle x-ray diffraction for those films with thickness of ∼25 nm, respectively. The crystal structures of the films were measured by using a x-ray diffractometer (XRD, D/MAX-2500v/pc, Rigaku) with Cu Kα radiation. The results indicate that all films are amorphous. The sur- face morphologies of the films were characterized by the SEM (S-4800, Hitachi). The electrical conductivity and Hall effect were measured using a physical property mea- surement system (PPMS-6000, Quantum Design), by em- ploying the standard four-probe method. The aluminium electrodes with thickness ∼280 nm were deposited before the electrical transport property measurements. Figure 1 shows the SEM images of the films. Although these IGZO films are amorphous, they are composed of amorphous IGZO particles. For those ∼700 nm thick films, the mean particle size of the film deposited at 788 K is significantly larger than that of the film deposited at 688 K. In addition, there are distinct boundaries between the neighboring particles. These boundaries could affect the uniformity as well as the subtle electron transport processes of the amorphous films. Figure 2 shows the conductivity varies as a function of temperature from 300 down to 2 K for the films. The room temperature conductivities of the films lie between 4.05 × 102 and 7.20 × 102 Ω−1 cm−1. Hence all the films have relative high conductivity. Upon decreasing tem- perature from 300 K, the conductivity of each film in- creases with decreasing temperature, reaches its max- imum at Tmax (Tmax varies from 92 to 115 K for our films), and then decreases with further decreasing tem- perature. The enhancement of conductivity with de- 2 TABLE I. Relevant parameters for the a-IGZO films. TS is the substrate temperature during deposition, t is the mean-film thickness, T1 is the maximum temperature for the ∆σ ∝ ln T law hold. σ0 and gT are the adjustable parameters in Eq. (2), σ0 represents the conductivity without the EEI effec. n∗ is the measured effective carrier concentration at T2, and T2 is the maximum temperature below which RH varies linearly with ln T . cd and E0 are the adjustable parameters in Eq. (3). Film No. 1 2 3 4 TS (K) 688 788 688 788 t (nm) 653.5 683.7 24.5 25.6 T1 (K) 25 30 15 15 σ0 (10−4 S) 5.05 4.90 7.00 4.42 gT 17.3 15.5 11.6 7.2 T2 (K) 45 45 40 35 n∗ (1026 m−3) 1.04 1.00 1.73 1.10 cd 1.38 1.24 1.51 1.29 E0 (10−22 J) 7.07 7.07 5.07 4.36 creasing temperature above Tmax indicates that these a- IGZO films possess metal (more strictly, degenerate semi- conductor) characteristics in transport properties.7,8 The reasons why the a-IGZO films have the metal-like trans- port properties deserve further investigation. In homo- geneous disordered conductors, the decrease of conduc- tivity with decreasing temperature at liquid helium tem- peratures generally originates from the weak-localization (WL) and EEI effects.9 -- 14 In three dimensional disor- dered conductors, the corrections to the conductivity due to WL and EEI effect are given as12 -- 14 ∆σ(T ) = e2 2π2 α τ 1/2 ϕ + e2 4π2 1.2 √2 (cid:18) 4 3 − F(cid:19) √kBT D 3 2 , (1) where α is a constant, e is the electronic charge, is the Planck constant divided by 2π, kB is the Boltzmann constant, D is the diffusion constant, τϕ is electron de- phasing time, F is the the screening factor averaged over the Fermi surface. The first term on the right hand side of Eq. (1) arises from the WL effect, while the second term originates from the interaction effect. To determine F independently, we measured the conductivity varies as a function of temperature under a magnetic field of 7 T, which is plotted in the inset of Fig. 2. Since the WL ef- fect would be suppressed by the high field,12,14 one would expect that the conductivities vary linearly with √T at √T law is low temperature regime. However, the ∆σ ∝ not observed in the inset of Fig. 2, which means other mechanisms should govern the temperature behavior of conductivity at liquid helium temperatures. As mentioned above, there are distinct boundaries be- tween the neighboring a-IGZO particles. The conduc- tivities of the boundaries would be much less than that inside of the particles. Thus the low temperature con- duction processes of the a-IGZO films would be simi- lar to that of granular metals, where a granular metal means a metal-dielectric composite lying above the per- colation threshold.15 -- 17 According to the recent theories, the EEI effect in granular metals is distinct from that of homogeneous disordered conductors. Specifically, in granular metals and the strong intergrain coupling limit [gT ≫1, where gT = GT /(2e2/) is the dimensionless in- tergranular tunneling conductance, GT is the intergrain FIG. 1. SEM micrographs of a-IGZO films (a) No.1, (b) No.2, (c) No.3, and (d) No.4. ) 1 - m c 1 - 2 0 1 ( 5.0 4.8 4.6 4.4 4.2 4.0 No.1 No.2 No.2 No.1 T0.5 (K0.5) 0 2 4 6 8 10 12 (a) 1.01 1.00 0.99 0.98 ) K 0 0 1 ( / (b) No.3 No.4 8.0 7.6 7.2 4.0 ) 1 - m c 1 - 2 0 1 ( 0 100 200 300 T (K) 0 100 200 300 T (K) FIG. 2. (Color online) The conductivity as a function of tem- perature for a-IGZO films (a) Nos.1 and 2, and (b) Nos.3 and 4. The inset of (a): normalized conductivity σ /σ(100 K) versus ln T under a field of 7 T for films No.1 and No.2. tunneling conductance], the conductivity17 -- 20 and Hall coefficient21,22 can be written as and σ = σ0(cid:20)1 − 1 2πgT d ln(cid:18) gT Ec kBT (cid:19)(cid:21) RH = 1 n∗e (cid:20)1 + cd 4πgT ln(cid:18) E0 kBT (cid:19)(cid:21) , (2) (3) sample No.1 and No.2 are listed in Table I. From Table I, one can see that the gT values are far greater than 1. Hence the necessary condition (gT ≫ 1) for the validity of Eqs. (2) and (3) is satisfied.17,23 In addition, the ex- tracted cd values are close to 1 (more precisely, slightly greater than 1), which is consistent with the theoretical prediction.23,26 The obtained values of σ0 are also rea- sonable. Hence Eqs. (2) and (3) are safely applicable for films No.1 and No.2. 3 where n∗ is the effective carrier concentration, cd is a numerical lattice factor, δ is the mean level spac- ing in the metallic particle, Ec is the charging energy, E0=min(gT Ec, ETh), and ETh is the Thouless energy, σ0 is the conductivity without the Coulomb interaction, and d is the dimensionality of the granular array. Eq. (2) is valid in the temperature range gT δ/kB < T ≪ Ec/kB, while Eq. (3) is valid in gT /kBδ . T . E0/kB. Re- cently, the validity of Eq. (2)23 -- 31 and Eq. (3)28 -- 31 has been experimentally tested. ) 1 - m c 1 - 2 0 1 ( 7.8 7.7 7.6 4.3 4.2 4.1 (a) (b) No.3 No.4 No.3 No.4 1 10 T (K) 100 1 100 10 T (K) ) C / 3 m 8 - 0 1 ( H R -3.6 -3.7 -3.8 -5.6 -5.8 -6.0 5.00 (a) ) 1 - m c 1 - 2 0 1 ( 4.95 4.85 4.80 4.75 1 (b) No.1 No.2 -5.9 -6.0 -6.1 -6.2 -6.3 -6.4 ) C / 3 m 8 - 0 1 ( H R No.1 No.2 10 T (K) 100 1 100 10 T (K) FIG. 3. The conductivity and Hall coefficient versus loga- rithm of temperature for films No.1 and No.2. The solid straight lines in (a) and (b) are least-squares fits to Eq. (2) and Eq. (3), respectively Figure 3 (a) and 3 (b) show the conductivities and Hall coefficients of the ∼ 700 nm thick films variation with temperature at low temperature regime. Clearly, the conductivity (Hall coefficient) varies linearly with log- arithmic temperature from 2 (2) to ∼25 K (∼45 K) for each film. The experimental σ (RH ) versus T (T ) data are least-squares fitted to Eq. (2) [Eq. (3)], and the re- sults are plotted as solid curves in Fig. 3 (a) [Fig. 3 (b)]. When comparing the σ-T data with Eq. (2), σ0 and gT are set as the adjustable parameters and the charging en- ergy Ec is taken to be ≈10kBT1, where T1 ≈ 25 K is the maximum temperature below which the ∆σ ∝ ln T law holds.23,28 For the RH -T fitting processes, cd and E0 are the adjustable parameters, gT has been independently determined above, and n∗ is taken the value of carrier concentration at T2 (T2 is the maximum temperature be- low which RH varies linearly with ln T ). The obtained fitting parameters, together with the basic parameters of FIG. 4. The conductivity and Hall coefficient versus loga- rithm of temperature for films No.3 and No.4. The solid straight lines in (a) and (b) are least-squares fits to Eq. (2) and Eq. (3) Figure 4 (a) and 4 (b) show variations of conductiv- ity and Hall coefficient with temperature for the two ∼25 nm, thin films, respectively. Clearly, both the con- ductivity and Hall coefficient vary linearly with ln T at low temperature regime. Inspection of Eqs. (2) and (3) indicates that the ln T behaviors of σ and RH are inde- pendent of the granular array dimensionality. From the SEM images, one can see that the a-IGZO particle size lies between 20 and 30 nm, thus it is reasonable to assume that the ∼25 nm films are covered by only one layer of a-IGZO particles. Here we treat the ∼25 nm films as two-dimensional (2D) granular arrays [d = 2 in Eq.(2)]. The experimental σ(T ) and RH (T ) data are least-squares fitted to Eq.(2) and Eq.(3), respectively, and the results are shown as solid curves in Fig.4. The detailed fitting processes are identical to that used in ∼700 nm films. The obtained values of parameters σ0, gT , cd and E0 are listed in Table I. The values of the parameters are also reasonable. In 2D disordered homogeneous conductors, the EEI effect can also cause a ln T correction to the conductivity. Specifically, the change of the sheet conductivity can be written as14,32 ∆σ(cid:3)(T ) = e2 2π2 (cid:18)1 − 3 4 F ′(cid:19) ln(cid:18) T T0(cid:19) , (4) where T0 is an arbitrary reference temperature, F ′ is the electron screening factor. The σ(T ) data in Fig. 4 (a) are also compared with Eq. (4). Certainly, the experimental 4 σ(T ) data follow the predication of Eq. (4). However, the least-squares fits would lead to positive F ′ values for both films ( F ′ ≃ −0.17 and −0.13 for films No.3 and No.4, respectively). For the 2D disordered homoge- neous conductor, the electron screening factor is required 0 ≤ F ′ ≤ 1.9 Thus the measured ∆σ ∝ ln T cannot be ascribed to the conventional EEI effect being applicable to the disordered homogeneous conductor. This result, on the other side, indicates that the boundaries between the neighboring a-IGZO particles play an important role in the electron transport processes. In summary, we deposited ∼700 nm and ∼25 nm a- IGZO films by rf-sputtering methods. The SEM images indicated the films are composed of a-IGZO particles and there are distinct boundaries between neighboring parti- cles. All films reveal metallic characteristics in electron transport properties. At low temperature regime, both the conductivity and Hall coefficient vary linearly with ln T , which cannot be explained by the traditional EEI effect. We found that these ln T behaviors of σ and RH result from the EEI effect due to the presence of granu- larity. The subtle inhomogeneity of the a-IGZO is caused by the boundaries between neighboring amorphous par- ticles. This work was supported by The National Ba- sic Research Program of China through Grant No. 2014CB931703, the NSF of China through Grant No. 11174216, and Research Fund for the Doctoral Program of Higher Education through Grant No. 20120032110065. ∗ Author to whom correspondence should be addressed. (2002). Electronic address: [email protected] 18 K. B. Efetov and A. Tschersich, Phys. Rev. B 67, 174205 1 K. Nomura, H. Ohta, A. Takagi, T. Kamiya, M. Hirano, (2003). and H. Hosono, Nature 432, 488 (2004). 19 I. S. Beloborodov, K. B. Efetov, A. V. Lopatin, and V. M. 2 T. Kamiya, K. Nomura, and H. Hosono, Sci. Technol. Adv. Vinokur, Phys. Rev. Lett. 91, 246801 (2003). Mater. 11, 044305 (2010). 3 H. Yabuta, M. Sano, K. Abe, T. Aiba, T. Den, H. Kumomi, and H. Hosono, Appl. Phys. Lett. 89, 112123 (2006). 4 T. Kamiya, K. Nomura, and H. Hosono, J. Disp. Technol. 5, 273 (2009). 20 I. S. Beloborodov, A. V. Lopatin, V. M. Vinokur, and K. B. Efetov, Rev. Mod. Phys. 79, 469 (2007). 21 M. Y. Kharitonov and K. B. Efetov, Phys. Rev. Lett. 99, 056803 (2007). 22 M. Yu. Kharitonov and K. B. Efetov, Phys. Rev. B 77, 5 C. Kim, S. Kim and C. Lee, Jpn. J. Appl. Phys. 44, 8501 045116 (2008). (2005). 23 L. Rotkina, S. Oh, J. N. Eckstein, and S. V. Rotkin, Phys. 6 K.Nomura,H. Ohta, K. Ueda, T. Kamiya, M. Hirano, and Rev. B 72, 233407 (2005). H. Hosono, Science 300, 1269 (2003). 7 J. M. Ziman, Electrons and Phonons (Clarendon Press, Oxford, 1960), p.364. 8 C. Kittel, Introduction to Solid State Physics, 7th ed, (Wi- ley, New York, 1996). 9 B. L. Altshuler and A. G. Aronov, in Electron-Electron Interactions in Disordered Systems, edited by A. L. Efros and M. Pollak (Elsevier, Amsterdam, 1985). 10 G.Bergmann, Int. J. Mod. Phys. B 24, 2015 (2010). 11 G.Bergmann, Phys. Rep. 107 (1984) 1. 12 B. L. Altshuler, D. Khmelnitzkii, A. I. Larkin, and P. A. 24 P. Achatz, W. Gajewski, E. Bustarret, C. Marcenat, R. Piquerel, C. Chapelier, T. Dubouchet, O. A. Williams, K. Haenen, J. A. Garrido, and M. Stutzmann, Phys. Rev. B 79, 201203 (2009). 25 Y. C. Sun, S. S. Yeh, and J. J. Lin, Phys. Rev. B 82, 054203 (2010). 26 R. Sachser, F. Porrati, C. H. Schwalb, and M. Huth, Phys. Rev. Lett. 107, 206803 (2011). 27 F. Porrati, R. Sachser, C. H. Schwalb, A. S. Frangakis, and M. Huth, J. Appl. Phys. 109, 063715 (2011). 28 Y. J. Zhang, Z.Q. Li and J. J. Lin, Phys. Rev. B 84, 052202 Lee, Phys. Rev. B 22, 5142 (1980). (2011). 13 S. P. Chiu, J. G. Lu, and J. J. Lin, Nanotechnology 24, 29 Y. Yang, Y. J. Zhang, X. D. Liu, and Z. Q. Li, Appl. Phys. 245203 (2013). Lett. 100, 262101 (2012). 14 P. A. Lee and T. V. Ramakrishnan, Rev. Mod. Phys. 57, 30 Y. J. Zhang, K. H. Gao, and Z. Q. Li, Appl. Phys. Lett. 287 (1985). 106, 101602 (2015). 15 P. Sheng, Philos. Mag. B 65, 357 (1992). 16 B. Abeles et al., Adv. Phys. 24, 407 (1975). 17 K. B. Efetov and A. Tschersich, Europhys. Lett. 59, 114 31 Y. N. Wu, Y. F. Wei, Z. Q. Li, and J. J. Lin, Phys. Rev. B 91, 104201 (2015). 32 B. L. Altshuler, A. G. Aronov, and P. A. Lee, Phys. Rev. Lett. 44, 1288 (1980).
0904.3193	3	0904	2010-06-02T14:42:46	Cryogenic instrumentation for fast current measurement in a silicon single electron transistor	[ "cond-mat.mes-hall", "cond-mat.mtrl-sci" ]	We present a realisation of high bandwidth instrumentation at cryogenic temperatures and for dilution refrigerator operation that possesses advantages over methods using radio-frequency single electron transistor or transimpedance amplifiers. The ability for the low temperature electronics to carry out faster measurements than with room temperature electronics is investigated by the use of a phosphorous-doped single-electron transistor. A single-shot technique is successfully implemented and used to observe the real time decay of a quantum state. A discussion on various measurement strategies is presented and the consequences on electron heating and noise are analysed.	cond-mat.mes-hall	cond-mat	Cryogenic instrumentation for fast current measurement in a silicon single electron transistor Journal of Applied Physics T. Ferrus,1 D. G. Hasko,2 Q. R. Morrissey,3 S. R. Burge,3 E. J. Freeman,3 M. J. French,3 A. Lam,2 L. Creswell,2 R. J. Collier,2 D. A. Williams,1 and G. A. D. Briggs4 1Hitachi Cambridge Laboratory, J. J. Thomson Avenue, CB3 0HE, Cambridge, United Kingdom∗ 2Cavendish Laboratory, University of Cambridge, J. J. Thomson Avenue, CB3 0HE, Cambridge, United Kingdom 3Rutherford Appleton Laboratory, Chilton, Didcot, OX11 0QX, Oxon, United Kingdom 4Department of Materials, University of Oxford, Parks Road, Oxford OX1 3PH, United Kingdom (Dated: November 3, 2018) We present a realisation of high bandwidth instrumentation at cryogenic temperatures and for dilution refrigerator operation that possesses advantages over methods using radio-frequency single electron transistor or transimpedance amplifiers. The ability for the low temperature electronics to carry out faster measurements than with room temperature electronics is investigated by the use of a phosphorous-doped single-electron transistor. A single-shot technique is successfully implemented and used to observe the real time decay of a quantum state. A discussion on various measurement strategies is presented and the consequences on electron heating and noise are analysed. PACS numbers: 71.30.+h, 71.55.Gs, 72.10.Fk, 72.15.Rn, 72.20.Ee, 73.20.Fz, 73.23.Hk I. INTRODUCTION. The ability to carry out high speed current measure- ments on nanoscale solid state devices is essential for quantum information, in implementations where rapid decoherence limits the timescale for readout.[1] Conven- tionally, such a current measurement would make use of an ideal room temperature operating transimpedance amplifier and the measurement speed would be limited by a low pass filter effect due to the output impedance of the nanodevice and the capacitance of the connect- ing cable. In measurements at low temperature, this cable capacitance is significant due to the large length between the room temperature transimpedance ampli- fier and the nanodevice. This combined with the high output impedance of semiconductor single electron tran- sistors [2] results in a practical measurement bandwidth of a few 100 Hz. Alternatively, a high speed conductance measurement may be carried out using radiofrequency techniques.[3] This offers a much higher measurement bandwidth, but requires the nanodevice to be connected to the room temperature measurement system using high bandwidth cables. In previous work,[4] it has been shown that a low pass filter is required on all connections to the nanodevice in order to prevent electron heating. The electron heating is manifest in degradation of the Coulomb oscillations. Another measurement technique is to integrate the current onto a capacitor for a fixed pe- riod of time and then to measure the increase in voltage across the capacitor. This approach has much greater complexity compared to the transimpedance amplifier, since the capacitor must be periodically discharged. But, ∗Electronic address: [email protected] such a function may be implemented without the use of high value resistors and so can operate under cryo- genic conditions. By operating the measurement circuit at low temperature, the cable to the nanodevice can be considerably reduced in length, so that the low pass fil- ter effect is diminished and a much higher measurement bandwidth is possible. In this paper, we describe the use of a custom complementary metal-oxide-semiconductor integrated circuit that operates at 4.2 K (LTCMOS) and is compatible with dilution refrigerator operation. The capabilities of this measurement circuit are investigated through the characterisation of a phosphorous-doped sili- con single electron transistor (SET) [5, 6] which is known to be sensitive to microwave irradiation.[7] A significant improvement in measurement speed was found compared to the conventional approach using room temperature transimpedance amplifiers (∼ 104 times faster [8]). Fi- nally we show that such an approach enables the real time evolution of system states to be monitored by a sin- gle shot technique [9] due to the reduction in microwave heating compared to the conventional approach. II. LOW TEMPERATURE INSTRUMENTATION AND TEST DEVICE A. Fast measurement techniques A transimpedance amplifier outputs a voltage propor- tional to its input current and can easily be built from an operational amplifier and a single resistor. However, in a non-ideal transimpedance amplifier, the measurement bandwidth is limited by the time constant resulting from the feedback resistor of the amplifier and the stray ca- pacitance. The value of this resistor increases with the sensitivity so that the amplifier generally imposes a mea- surement speed limit for the lower current ranges. Op- erating the transimpedance amplifier at low temperature reduces the capacitance but causes a considerable varia- tion in the characteristics of the high value resistors used in the feedback circuit and a significant heat loading es- pecially when in close proximity to the device.[10] Alternatively, a radiofrequency (RF)-based SET gives a measurement bandwidth up to few tens of MHz. The SET device is embedded into a resonant circuit and the RF reflectivity from the source contact is measured. The variation of charge in the device then modifies the con- ductance of the SET so modulating the amplitude of the reflected wave. Because of the high frequency operation, the 1/f noise, due to the background charge, is reduced. However, these methods cannot be used with devices such as phosphorous-doped silicon SETs that are sensitive to microwaves.[4, 7] Indeed the use of RF induces a signif- icant increase in the electron temperature of the device but filtering is not applicable in this situation. Also RF may induce irreversible charge movements or charge tun- nelling in the SET island that affect capacitances. Small currents may also be measured using a charge integration circuit. By avoiding the need for a high value resistance, the charge integrator is made more compati- ble with cryogenic operation. The operation of a charge integrator is significantly more complicated than that of a transimpedance amplifier. Switching circuits are needed to periodically discharge the integration capacitor and these additional circuits are conveniently accommodated within a purpose made application specific device, des- ignated for cryogenic operation. By appropriate choice of the integration capacitance and switching times, the charge integration circuit may be operated at signifi- cantly greater speed than an equivalent transimpedance amplifier. B. Measurement setup 1. LTCMOS ASIC The measurement unit consists of a custom LTCMOS Application Specific Integrated Circuit (ASIC) and a room temperature Data AcQuisition (DAQ) and commu- nication box. The LTCMOS ASIC supplies the gate (Vg) and source-drain voltages (VSD) and measures the SET drain current (ISD). A high speed and sensitive current measurement is achieved by using the SET current to charge a very small value capacitor for a predetermined time. The voltage level on the capacitor is then sam- pled by the DAQ and communication box, and then dis- charged periodically. 100 fF, 1 pF, and 10 pF capacitors and sampling periods as low as 25 µs may be selected un- der computer control to adjust the speed and sensitivity of the measurement. The maximum sampling time and input current are limited by the saturation of the capac- itor voltage measurement circuit. The measurement res- olution and noise are a function of signal level, sampling time, and measuring circuit capacitance. Parasitic input 2 capacitances (such as that due to the cable between the LTCMOS ASIC and the device being tested) also degrade the noise performance. For improved measurement qual- ity it is therefore desirable to minimise the cable length to the SET and use a cryogenic measurement system. The DAQ and communications box at room temperature acts as a controller for the LTCMOS ASIC plus a data digi- tiser and recorder for the integrated signal data that is sent back from the LTCMOS ASIC. This data is then transmitted via gigabit optical Ethernet to a host com- puter for storage and analysis (Fig. 1). The measurement units are integrated into a custom designed probe that is constructed from thin walled stainless steel tubing to minimise heat transfer from the room temperature parts into the cryostat. The probe is intended to be inserted into a dilution unit and is divided into a number of levels corresponding to the structures of an Oxford Instruments Superconductivity KelvinOxTM 400 dilution refrigerator. Copper plates and clamps at the mixing chamber, cold plate, still, 1 K pot and 4 K plate levels allow thermal cou- pling of signal cables to avoid thermal loading to lower levels. The heat load from the LTCMOS ASIC is also transferred to the helium bath via the 4.2 K plate by this method. A custom designed device carrier is mounted at the bottom of the probe to hold the device package and microwave feed and provide thermal coupling to the mixing chamber. 2. Mounting for microwave experiments Devices were mounted onto 20 pin leadless header packages that are suitable for signal frequencies up to ∼1 GHz, according to the manufacturer specifications. However the inductance due to the bond wires (30 µm di- ameter and 1 mm length) and the resistance of the doped silicon leads (sheet resistance ∼1 kΩ.(cid:3)−1), in combina- tion with the stray capacitance to the substrate, causes a significant attenuation at high frequencies. While this bandwidth is more than sufficient to carry the gate, source and drain voltages as well as current measurement signals, it is not fast enough to carry a microwave manip- ulation pulse, which may include signal frequencies be- tween 1 to 20 GHz so that the microwave coupling to the device needs to be implemented differently. In previous work, this difficulty has been overcome by embedding the device in a microwave printed circuit board and by using coplanar waveguides for the high frequency signals.[11] However, it is difficult to match the impedance to the one of the device in this approach, due to the problem of providing a 50 Ω terminating resistance close to the nanostructure. As a result, standing wave effects make the microwave power coupled to the device depend on frequency. In addition, it is known that the signal leads need to be low-pass filtered (typically preventing frequen- cies above ∼1 MHz from propagating to the device) to avoid significant degradation of the Coulomb characteris- tics through electron heating (see Sec. II. C. 2.). The ap- proach used here employs an open ended semi-rigid cylin- drical waveguide (RG402) to weakly couple microwaves in the frequency range from ∼1-20 GHz to the device in a global manner. By controlling the distance between the device and the end of the waveguide standing wave effects may be avoided. The base of the header package is a thick gold layer (much thicker than the skin depth in the relevant frequency range) and so provides an effi- cient reflector of microwaves. Part of the incident wave is reflected back from the base of the header package into the waveguide with a phase change of π radians from the incoming signal. The remainder of the microwave signal is able to pass between the end of the waveguide and the header package base. This part of the microwave signal is also subject to a reflection due to the propagation from an environment with 50 Ω waveguide impedance to an en- vironment with a characteristic impedance of 377 Ω.(cid:3)−1. This second reflection is in phase with the incoming signal in contrast to the reflection from the base of the header package. By adjusting the gap between the end of the waveguide and the base of the header package (∼1 mm), these two reflections can be matched in amplitude so that the cavity behaves as a 50 Ω termination. As this effect is determined by geometry rather than electrical proper- ties, the impedance matching is independent of frequency and temperature over a wide range. 3. Microwave Experimental setup Measurements were controlled by a LabView program running on a computer that integrated the LTCMOS in- terface, with an Agilent E8257D-520 PSG Analog Signal Generator and an Agilent B1130A pulse pattern genera- tor linked via a GPIB bus (Fig. 1). The microwave source provides continuous wave (cw) signals to 1 Hz precision at power levels of up to +15 dBm; the frequency may be changed on a time-scale of a few milliseconds. The cw microwave signal is passed through a Herotek single pole single throw S1D0518A4 PIN switch, which has a switch- ing time of about 30 ns. The PIN switch is controlled by a B1130A Pulse Pattern Generator (PPG), allowing arbitrary on-off switching patterns to be implemented within the limitations imposed by the switching time of the PIN switch and the memory capacity of the PPG. A trigger signal generated by the DAQ and communication box is used to synchronize the start of a sequence of cur- rent measurements with the initiation of a pulse sequence from the PPG. Most of the signal path makes use of flexi- ble SMA terminated microwave cable so that some signal loss occurs. The microwave power available at the de- vice from this setup was sufficient for cw measurements, but for pulsed measurements an additional amplification stage, based on a Mini-Circuits ZX60-5916M-S preampli- fier and an output stage using a BFM21 FET, was used to increase the microwave power to about +25 dBm (in the range up to a few GHz). Most of the measurements were performed at 4.2 K by immersing the probe in a liquid Trigger E8257D microwave source PIN Switch 3 Room temperature control electronics B1130A pulse pattern generator Microwave power amplifier LTCMOS TTL control Cryostat Filters GPIB bus Computer pattern generator SET Fibre optic link FIG. 1: Schematic of the measurement setup. Arrows indi- cates directional communication between apparatus. helium dewar for preliminary assessment and to provide an efficient thermal coupling to the 4.2 K bath. Com- plementary experiments were carried in the internal vac- uum chamber of the dilution refrigerator but placed addi- tional constraints on power dissipation especially during microwave experiments. C. Device fabrication and DC characteristics 1. SET as a test device A highly doped silicon SET was chosen to test the LTCMOS since the characteristics of these devices have been widely explored previously using a variety of room temperature measurement systems. The SETs used in this work are fabricated from a standard silicon-on- insulator (SOI) substrate having a 30 nm-thick highly- doped silicon layer. Phosphorous atoms were ion im- planted, resulting in a post-annealing concentration of ∼ 1020 cm−3. High resolution electron beam lithography and reactive ion etching were used to pattern a single is- land of diameter ∼60 nm. The 30-nm width constrictions that separate the island to the connecting source and drain leads are depleted of electrons and act as tunnel barriers. The formation of such barriers has been widely investigated and will not be discussed here.[12] Thermal oxidation is then used to reduce the island size and tunnel barrier widths as well as for surface passivation, resulting in a significant reduction in random telegraph switching. The devices are controlled by an in-plane gate that is formed using the same SOI layer. Several devices were processed identically and behaved similarly. The follow- ing results are presented for a specific device that was chosen for its high reproducibility in time as well as for its high signal to noise ratio. 2. DC characteristics and electrical noise heating All measurements were performed using the 1 pF cur- rent measurement capacitor in the LTCMOS ASIC which gave the best signal-to-noise ratio (SNR) of the three possible values. The visibility of Coulomb blockade os- cillations at 4.2 K is poor with a SNR of about 18:1 for VSD=2 mV and a significant background conductivity (Fig. 2) when compared with conventional room tem- perature equipment. From the dependencies in gate and source drain bias, it is possible to extract estimates for the gate and drain level arms, e.g. αg and αD: ) A n ( D S I 0.3 0.2 0.1 0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 -0.7 4 0.6 0.3 0.0 -0.3 -0.6 -0.9 -1.2 -1.5 -1.8 -2.1 -2.4 -0.4 -0.2 0.0 0.2 0.4 -0.4 -0.2 0.0 0.2 0.4 Vg (V) Vg (V) αg = Cg/CΣ and αD = CD/CΣ (1) where Cg, CD and CΣ are the gate, drain and total capacitance extracted from the Coulomb diamond slopes. We find αD ∼ 0.53 and αg ∼ 8.7 × 10−3 and that both values are approximately independent of gate volt- age. The SET diameter determined from the scanning electron microscope (SEM) is about 60 nm, so that the first excited state in the dot is expected to be at 1.1 meV above the ground state, a value somewhat larger than the thermal broadening at helium temperature (∼ 0.4 meV). Under this condition, the electron temperature Te can be estimated from : αg ∼ 3.5kBTe/ (eW ) (2) where W is the width of the Coulomb peak. This approximation is valid as long as the intrinsic linewidth of the Coulomb peak is negligible compared with the thermal broadening, e.g. T > 4.2 K in the present case. For Vg=-0.039 V and VSD=2 mV, we find Te ∼ 9.0 K implying significant electron heating. In the absence of deliberate microwave irradiation, high frequency electromagnetic noise was found to be the major source of electron heating in the present ex- perimental setup. Apart from electrical noise coming from the LTCMOS itself, the room temperature con- trol electronics that contains digital circuitry operating at clock speeds of 25 MHz and 125 MHz (plus a 1 GHz internal clock to the optical transceiver) may contribute to the noise that propagates down to the device. For equivalent measurements, made using conventional room temperature electronics, such as Keithley 236 source- measurement units for example, this problem may be cir- cumvent by using 2 MHz cut-off frequency low pass filters (MiniCircuits BLP1.9) on each lead connecting the de- vice. Unfortunately, the high capacitance of the BLP1.9 circuit prevents its use with the charge integrating cir- cuit of the LTCMOS circuit as the noise gain is greatly increased. In this case, we used a single stage resistor- capacitor low-pass filter made of a ceramic capacitor and a metal film resistor because of the materials being tem- perature independent. To extend the high frequency cut- off, ferrite beads were attached to the filter inputs and FIG. 2: Coulomb blockade oscillations when lines are not filtered (left) and after adding the low temperature RLC filter to the circuit (right). The measurement was taken at 4.2 K with source-drain biases from -2 mV (bottom) to 2 mV (top). outputs. The ferrite material was chosen so as to sup- press noise over a wide range of frequencies as well as to enhance its efficiency at low temperature. Filters were placed near the device to remove electromagnetic noise as well as to thermally anchor the leads. Unlike a normal inductor, ferrite beads absorb the energy and dissipate it as heat. However the amount of heat is negligible and does not contribute to electron heating effects. The cut- off frequency at 4.2 K was 2 MHz so that single shot mea- surements were not affected by the filters. The use of this cryogenic filter improved the SNR of the Coulomb block- ade oscillations to 64:1 and resulted in the appearance of blockade regions with zero current. For Vg=-0.083 V and VSD=2 mV, we find αD ∼ 0.44, αg ∼ 3.9 × 10−2 Te ∼ 4.6 K and the charging en- ergy was about 4.3 meV. We also find Cg ∼ 2.0 aF and CΣ ∼ 37.0 aF and a corresponding dot diameter of about 57 nm, assuming a spherical dot. This value is in good agreement with the dot size determined from SEM imag- ing. The electron temperature was thus efficiently re- duced to a value close to the lattice temperature during liquid helium immersion measurement (Fig. 2). For measurements in the dilution refrigerator, the sam- ple is located in a internal vacuum chamber so that the lower operating temperature reduces the thermal conduc- tivity and specific heat capacity in the device. These ef- fects make the problem of electron heating in the dilution refrigerator much more severe than for liquid helium im- mersion measurements. A base temperature, measured by the mixing chamber thermometer, of 27 mK was ob- tained for dc measurements and 60 mK for microwave experiments due the additional heat loading from the waveguide. The electron temperature was estimated to be Te ∼ 300 mK for VSD=-2.25 mV, 150 mK for -1.75 mV and about 80 mK for 1.25 mV after correction for the source drain offset of 0.75 mV for dc measurements. 3. Continuous microwave spectroscopy and microwave heating In previous work, the CW microwave excitation has been found to consist of a large number of resonances that are detectable with the LTCMOS when the mi- crowave power P at the source is greater than about -20 dBm (see Fig. 3a and b).[4] These resonances ex- hibit a wide range of amplitudes and quality factors with the largest amplitudes corresponding to the lowest Q val- ues (Fig. 3c). Far from being a unique case, the pres- ence of resonances with high Q value have also been ob- served in TiO2 sol-gel based high-k gate dielectric silicon MOSFETs where they result from resonant excitation between trapped states.[13] Previously resonant features have also been observed in a SET. Manscher noticed an increase of source-drain current in an Al/Al2O3/Al SET as a result of enhancement of the cotunneling effect due to microwave induced heating.[14] However, in the res- onances observed here, we noticed the presence of both a relative increase and a relative decrease in the SET current, which is inconsistent with an explanation based on heating. Also, in our setup, a variation in the mi- crowave power does not induce a relative change in the current through a modification of the microwave coupling constant because of the method used to couple the mi- crowave to the device. Finally, figure 3c shows that a same mechanism is responsible for both high and low Q resonances.In the case of a highly doped silicon SET, the resonances are likely to result from a tunnelling process that is well isolated from the environment (Fig. 3b). From the Lorentzian shape of the resonances and their width that is five or six orders of magnitude lower than kBT , it has been shown that the most likely electronic process that is taking place is phononless transport like in frequency driven two-level tunneling systems [15] or trap assisted tunnelling via localised states in or around the SET island. These localised states are favoured by the presence of phosphorous donor in the device despite their metallic- like density. Because of surface segregation of dopants from the silicon that was consumed by thermal oxida- tion, a significant number of donors are embedded into a silicon oxide matrix leading to numerous electrons be- ing localised in states on the sidewalls of the SET.[16] This effect is enhanced by the large surface state den- sity found on the non-(100) surfaces forming the side- walls. The presence of these trapped electrons induces sidewall depletion that may result in the creation of tun- nel barriers in small width silicon region. The electron displacement, due to a tunnelling process occurring be- tween localized states, modifies the electrostatic environ- ment in the SET island so that the magnitude of the measured current is changed. Localised states in the di- rect transport pathway are expected to lead to low Q resonances with large amplitude due to the short resi- dence time of the electron. Whereas more remote states are expected to lead to the higher Q, but lower ampli- 5 tude, resonances. Electron displacement can lead to an increase or decrease (or combination of these) in the SET current depending on the proximity and orientation of the localized states with the respective tunnel barriers to the island. Coupling to the phonon bath is weak and electronic processes happen over a timescale shorter than the thermalization time, so that the resonance widths ∆ν0 could be related to the energy loss mechanism in the system using Q = πν0/∆ν0 ∼ πT0ν0, where T0 is the energy relaxation time of the system [17] and ν0 the resonance frequency. In the Coulomb blockade regime, the conductivity os- cillations are known to be sensitive to electron temper- ature through a change in the Coulomb peak lineshape and an increase in the background conductivity. As mi- crowave irradiation inevitably leads to electron heating, the visibility of the oscillations can be used to deter- mine the maximum usable microwave power before no- ticeable heating occurs (Fig. 4a). As demonstrated by Manscher,[14] in a Al/A2O3/Al SET, the co-tunneling effect is modified in the presence of an ac signal and the conductivity is linear (respectively quadratic) with mi- crowave power in the conducting (respectively blockade) regime to the second order perturbation (Fig. 4b and c). Away from the resonance conditions as described previ- ously, most of the microwave absorption occurs in the substrate by coupling to the low electron density in the lightly doped silicon. Because of the efficient coupling between the silicon substrate and the phonon bath, the temperature of the silicon will then rise until the thermal losses match the incoming microwave power. We experimentally found that a 10 % increase in the temperature of the silicon substrate would lead to a no- ticeable change in the Coulomb blockade oscillations. In the case of a direct immersion into liquid helium at 4.2 K, the thermal impedance is dominated by the Kapitza resistance between the semiconductor and the liquid cryogen.[18] From the value of the Kapitza coeffi- cient between silicon and 4He [19] and for a device dimen- sion of 2.5 mm × 2.5 mm × 0.25 mm, we estimate that power needed to cause such an increase of temperature is about 4 mW for that geometry. As the cooling power of helium gas is much greater than for the liquid, this may underestimate the heat load required since even a mod- est rise in temperature would result in boiling off liquid helium. The situation in the dilution refrigerator is very different. As the sample is in vacuum, the heat load is taken mainly by the sample wires (200 µm diameter and 10 cm long), which are thermally anchored to the mixing chamber. At the lowest operating temperature (about 60 mK) and by using the value of thermal conductivity for pure copper, the minimum power is reduced down to 0.1 µW. The lower thermal conductivity to the thermal anchoring and the reduced tolerance to temperature rise at the lower temperatures in the dilution then greatly reduce the maximum power that can be coupled to the device. = 5.8 GHz VSD = 4 mV (a) -0.4 -0.3 (b) -0.2 Vg (V) (c) -0.1 0.0 ) A p ( D S I 0.2 0.1 0.0 6 15 12 9 6 3 0 -3 0.8 ~ 92 s 0.9 1.0 1.1 1.2 1.3 t (ms) ) A p ( ) f f o ( D S I - ) 0 ( D S I 2.716 GHz 2.718 GHz 125 100 75 50 25 0 0.0 0.5 1.0 1.5 2.0 2.5 t (ms) FIG. 4: Time evolution of the relative SET current off reso- nance (νoff = 2.716 GHz) and on resonance (ν0 = 2.718 GHz). The inset shows the real-time decay of the current. ical average over 1000 identical and independent single- shot measurements. The delay in the trigger signal causes the microwave pulse to arrive at t ∼ 0.8 ms (Fig. 5). The signal measured in single shot mode also contains some very large damped oscillations, which do not depend on the nature of the microwave pulse. They are even present in the absence of the microwave pulse and are probably due to broadband radiofrequency excitation due to the digital signals in the control circuits. By tuning the mi- crowave frequency to on or off resonance, the effects of these background oscillations may be removed. The dif- ference between these is an exponential decay with a time constant τ ∼ 90 µs. This value is more than one order of magnitude longer than that found from the cw linewidth measurement. As the time-averaged power in the pulsed measurement is a factor of 1000 less than that used in the cw measurement, this difference in relaxation time may result from the much greater microwave-induced electron heating in the cw measurement. The lifetime τ corre- sponds to the time for the microwave-excited electron to relax to its original state or to escape. Long lifetimes are consistent with glassy behaviour,[20] a characteris- tic manifestation of Coulomb interaction in systems with localized states. The disorder in the highly doped sil- icon system is probably due to the random placement of phosphorous donors. The dynamics of transport and relaxation in this system results from the competition between disorder and Coulomb interactions. B. Measurement methods comparison 1. Single shot spectroscopy versus cw and pulse microwaves A comparison of the lifetime T0 deduced from the res- onance linewidth in a cw measurement and the lifetime τ obtained from a single shot measurement, indicates that ) A n ( D S I ) A n ( D S I 0.3 0.2 0.1 0.0 -0.1 -0.2 -0.3 0.6 0.4 0.2 0.0 T = 60 mK 0.0 0.5 P (mW) 1.0 0.0 0.5 P2 (mW2) 1.0 FIG. 3: a) Gate oscillations at 4.2 K under microwave illumi- nation with P = 0 dBm (top) down to -15 dBm (bottom). b) Variation of the current with microwave power in the conduct- ing region( Coulomb blockade maxima as designed by symbols in a)). c) Variation in the blockade regime where cotunneling is expected to be the dominant transport mechanism. Sym- bols refer to maxima and minima in the Coulomb blockade oscillations as designed in a). III. AN EXAMPLE OF FAST MEASUREMENT : THE SINGLE-SHOT TECHNIQUE. A. Single shot measurement The high Q value and resonant frequency suggests that the lifetime for the associated excitation can be as long as 100 µs for some resonances. With the larger measure- ment bandwidth afforded by the use of the LTCMOS, the real-time evolution of the source-drain current after the application of a short microwave pulse can be measured. This technique is also known as a single shot measure- ment. This procedure consists of sending a single mi- crowave excitation pulse of duration tp whilst carrying out current measurements at intervals tm with the con- dition tp ≤ tm ≪ τ ≪ tr where tr the time between repeated single shot operations and τ the quantum state lifetime. To this purpose, a well-isolated resonance on a level background at a frequency ν0 ∼ 2.718 GHz was selected from a cw measurement at a power of +9 dBm, with a charge integration time of 200 µs and averaging the results over 200 measurements (Fig. 5). This results in a signal-to-noise ratio of about 100:1. The width of the chosen resonance was about 207 kHz, for which the corresponding relaxation time is T0 ∼ 5 µs. A single shot measurement was then performed by reduc- ing the charge integration time to 20 µs and using a mi- crowave pulse of duration of 2 µs. The microwave power was increased to about +25 dBm using a power amplifier to compensate for the decrease in the signal amplitude. These conditions also greatly reduced the SNR compared to the cw measurement, so we proceeded to a to a numer- 7 the time dependence is extracted from current spec- troscopy by the use of a Fourier transform. This method allows ultra-fast measurement with a bandwidth of the order of 100 MHz, but cannot be used in devices that are sensitive to radio-frequency signals as is the case for doped or undoped silicon SETs.[21] Also, high frequen- cies are incompatible with the signal filtering that is es- sential in reducing the electron temperature. A time average measurement has also been used when the evolution of a quantum state could not be monitored in real-time due to measurement bandwidth limitations. In quantum computing architectures such as Gorman's, [8] this method requires the measured current to be aver- aged over a large number of identical gate voltage pulses, with the condition that the next pulse is applied be- fore the excitation, due to the previous pulse, can fully decay.[22] This limits the time between successive pulses and assumes long term stability of the system under in- vestigation. By contrast, the single shot measurement technique allows the real-time evolution of the system to be recorded after excitation by a single pulse, starting and finishing with the system in a fully relaxed state. IV. CONCLUSIONS. We have realised a high bandwidth low temperature CMOS amplifier circuit based on charge integration for the use in dilution refrigerators. This allowed to imple- ment the single shot measurement technique and observe slow-decaying electronic states in a phosphorous-doped silicon single electron transistor. In order to address the issue of electron heating usually occurring during contin- uous microwave spectroscopy, we have developed a single shot spectroscopy technique that permits the observation of high quality factor resonances up to 400,000 at 60 mK that were not observable by other methods. V. ACKNOWLEDGEMENTS. The authors thank EPSRC (UK) for funding under grant number GR/S24275/01 and for a Professorial Re- search Fellowship for G. A. D. B. GR/S15808/01. They are also grateful to V. Mikheev and P. Noonan from Ox- ford Instrument for their involvement in the design and the realisation of the dilution refrigerator. This work was partly supported by Special Coordination Funds for Pro- moting Science and Technology in Japan. Two of the authors (T. F. and D. G. H.) contributed equally to this work. -0.55 -0.60 -0.65 -0.70 -0.75 -0.80 -0.85 ) A n ( D S I -0.65 Q ~ 370,000 -0.70 -0.75 2.7654 T0 ~ 43 s 2.7656 -0.90 2.7650 2.7655 2.7660 2.7665 2.7670 2.7675 (GHz) FIG. 5: Single-shot mode spectroscopy showing a very high Q resonance at 60 mK. Measurements were taken with P =7 dBm, a gate voltage of 34 mV, a source voltage of 2 mV and a pulse duration of 2 µs the Q value may be considerably underestimated by cw spectroscopy. Indeed, the continuous application of mi- crowaves, at power levels needed for signal detection, sig- nificantly raises the electron temperature and so reduces the visibility of the resonances due to the enhancement of thermally driven relaxation and consequently the Q value. This problem is more acute in a dilution refrig- erator compared with the liquid helium immersion ex- periment because of the difference in specific heat capac- ity and thermal conductivity of the system as discussed in section II. C. 3. The strategy to maintain a reason- able SNR during single shot measurements is quite com- plex and generally requires the time delay to be adjusted between to successive measurements. This reduces the overall measurement speed but allow the device to ex- change heat with the mixing chamber through the cop- per wires. In single shot measurements, the time av- eraged microwave power is greatly decreased by using pulses, even though the instantaneous power is signifi- cantly higher than in the cw case. This problem is cir- cumvented by using the single-shot approach for spec- troscopy. The signal amplitude at each frequency, is taken from the first current measurement point after the arrival of the microwave pulse, in single shot mode, i.e. at t = 825 µs. This method resulted in the observation of resonances with Q factors as high as 400,000 measured in the dilution refrigerator and at a temperature of 60 mK (Fig.6) with negligible microwave induced heating. 2. Single shot technique versus standard time dependence measurement Other methods are proposed in order to implement time dependence measurements and may have even higher bandwidth. This is the case of RFSETs [3] where 8 [1] D. P. DiVincenzo, Fortschr. Phys. 48, 771 (2000) [2] D. V. Averin and K. K. Likharev, J. Low Temp. Phys. 62, 345 (1986); T. A. Fulton and G. J. Dolan, Phys. Rev. Lett. 59, 109 (1987) [3] W. Lu, Z. Ji, L. Pfeiffer, K. W. West and A. J. Rim- berg, Nature 423, 422-425 (2003);R. J. Schoelkopf, P. Wahlgren, A. A. Kozhevnikov, P. Delsing and D. E. Prober, Science 280, 1238 (1998) [4] D. G. Hasko, T. Ferrus, Q. R. Morrissey, S. R. Burge, E. J. Freeman, M. J. French, A. Lam, L. Cresswell, R. J. Collier, D. A. Williams and G. A. D. Briggs, Appl. Phys. Lett. 93, 19, 192116 (2008) [5] T. A. Fulton and G. J. Dolan, Phys. Rev. Lett. 59, 109 (1987); P. Lafarge, H. Pothier, E. R. Williams, D. Esteve, C. Urbina and M. H. Devoret, Z. Phys. B 85, 327 (1991) [11] T. Obata, M. Pioro-Ladrire, T. Kubo, K. Yoshida, Y. Tokura and S. Tarucha, Rev. Sci. Instr. 78, 104704 (2007) [12] S. C. Vitkavage and E. A. Irene, J. Appl. Phys. 68, 5262 (1990) [13] M. Ziaur Rahman Khan, D. G. Hasko, M. S. M. Saifullah and M. E. Welland, Appl. Phys. Lett. 93, 193501 (2008) [14] M. H. Manscher, M. T. Savolainen, and J. Mygind, IEEE Trans. Appl. Supercond. 13, 2 (2003) [15] B. I. Shklovskii and A. L. Efros, Sov. Phys. JETP, 54, 218 (1981) [16] J. Eldridge and D. Kerr, J. Electrochem. Soc. 118 986 (1971); P. Balk and J. Eldridge, Proc. IEEE 57, 1558 (1969) [17] C. N. Banwell and E. M. McCash, Fundamentals of molecular spectroscopy, 4th ed., McGraw-Hill (1994) [6] C. W. J. Beenakker, Phys. Rev. B 44, 1646 (1991) [7] L. A. Creswell, D. G. Hasko and D. A. Williams, Micro- [18] G. L. Pollack, Rev. Mod. Phys. 41, 48 (1969) [19] J. R. Olson and R. O. Pohl, J. Low Temp. Phys., 94, 539 electronic Engineering 84, 1614 (2007) (1994) [8] J. Gorman, D. G. Hasko and D. A. Williams, Phys. Rev. [20] B. I. Shklovskii and A. L. Efros, Electronic Properties of Lett. 95, 090502 (2005) [9] M. A. Nielsen and I. L. Chuang, textitQuantum Com- putation and Quantum Information (Cambridge U. P., Cambridge, 2000); T. Meunier, I. T. Vink, L. H. Willems van Beveren, F. H. L. Koppens, H. P. Tranitz, W. Wegscheider, L. P. Kouwenhoven, and L. M. K. Vander- sypen, Phys. Rev. B 74, 195303 (2006) [10] J. Pettersson, P. Wahlgren, P. Delsing, D. B. Haviland, T. Claeson, N. Rorsman and H. Zirath, Phys. Rev. B 53, R13272 (1996); E. H. Visscher, J. Lindeman, S. M. Verbrugh, P. Hadley, J. E. Mooij and W. van der Vleuten, Appl. Phys. Lett. 68, 2014 (1996) Doped Semiconductors (Springer-Verlag Berlin 1984) [21] E. Prati, M. Fanciulli, A. Calderoni, G. Ferrari and M. Sampietro, J. Appl. Phys. 103, 104502 (2008); M. Ziaur Rahman Khan, D. G. Hasko, M. S. M. Saifullah, and M. E. Welland, J. Vac. Sci. Technol. B 26, 1887 (2008) [22] Y. A. Pashkin, T. Yamamoto, O. Astafiev, Y. Nakamura, D. V. Averin and J. S. Tsai, Nature 421, 823 (2003); L. C. L. Hollenberg A. S. Dzurak, C. Wellard, A. R. Hamilton, D. J. Reilly, G. J. Milburn and R. G. Clark, Phys. Rev. B 69, 113301 (2004) ) A p ( D S I ) A p ( D S I 225 150 75 0 -75 -150 2.0 40 30 20 10 0 -3 (a) 2.2 2.4 2.6 2.8 3.0 (GHz) 3.2 3.4 3.6 3.8 4.0 (b) -2 0 -1 - 0(MHz) 1 2 (c) 105 104 103 102 10-2 10-1 100 101 102 103101 Amplitude (pA) Q
1506.06908	2	1506	2015-11-09T08:41:04	Resonators coupled to voltage-biased Josephson junctions: From linear response to strongly driven nonlinear oscillations	[ "cond-mat.mes-hall", "cond-mat.supr-con" ]	Motivated by recent experiments, where a voltage biased Josephson junction is placed in series with a resonator, the classical dynamics of the circuit is studied in various domains of parameter space. This problem can be mapped onto the dissipative motion of a single degree of freedom in a nonlinear time-dependent potential, where in contrast to conventional settings the nonlinearity appears in the driving while the static potential is purely harmonic. For long times the system approaches steady states which are analyzed in the underdamped regime over the full range of driving parameters including the fundamental resonance as well as higher and sub-harmonics. Observables such as the dc-Josephson current and the radiated microwave power give direct information about the underlying dynamics covering phenomena as bifurcations, irregular motion, up- and down conversion. Due to their tunability, present and future set-ups provide versatile platforms to explore the changeover from linear response to strongly nonlinear behavior in driven dissipative systems under well defined conditions.	cond-mat.mes-hall	cond-mat	Resonators coupled to voltage-biased Josephson junctions: From linear response to strongly driven nonlinear oscillations S. Meister, M. Mecklenburg, V. Gramich,∗ J. T. Stockburger, J. Ankerhold, and B. Kubala Institut fur komplexe Quantensysteme and IQST, Universitat Ulm, Albert-Einstein-Allee 11, 89069 Ulm, Germany (Dated: August 20, 2018) Motivated by recent experiments, where a voltage biased Josephson junction is placed in series with a resonator, the classical dynamics of the circuit is studied in various domains of parameter space. This problem can be mapped onto the dissipative motion of a single degree of freedom in a nonlinear time-dependent potential, where in contrast to conventional settings the nonlinearity appears in the driving while the static potential is purely harmonic. For long times the system ap- proaches steady states which are analyzed in the underdamped regime over the full range of driving parameters including the fundamental resonance as well as higher and sub-harmonics. Observables such as the dc-Josephson current and the radiated microwave power give direct information about the underlying dynamics covering phenomena as bifurcations, irregular motion, up- and down con- version. Due to their tunability, present and future set-ups provide versatile platforms to explore the changeover from linear response to strongly nonlinear behavior in driven dissipative systems under well defined conditions. PACS numbers: 74.50.+r, 05.45.-a, 74.40.De I. INTRODUCTION The nonlinear properties of Josephson junctions (JJs) have made such devices a key circuit element for clas- sical and quantum electronics. Accordingly, there has been a long tradition of studying non-linear phenomena in driven superconducting circuits, starting as early as the 1960s with the discovery of Shapiro steps1. While Shapiro-steps have remained a tool in exploring new directions in Josephson physics, for instance in atomic point contacts2 -- 6, other nonlinear phenomena, like syn- chronization, have been investigated in arrays of JJs: as a test-bed for generic theory models to capture syn- chronization phenomena7,8, but as importantly with the prospect of applications as sources of more intense coher- ent radiation9, cf. also new developments using intrinsic arrays10 -- 12. More recently, the nonlinearity of the JJ was ex- ploited as the crucial factor in enabling the high sensi- tivity of Josephson bifurcation amplifiers, achieving sub- stantial improvements towards reaching quantum-limited measurement processes13 -- 16. Most of the features of Josephson bifurcation amplifiers, in fact, only rely on (and can consequently be described by) any type of nonlinearity17 -- 22, e.g., Duffing-type models, so that only recently the full nonlinear potential of the JJ has become of interest in this field23. A recent addition to the field of driven nonlinear JJs24 -- 26 are experiments on a dc-biased JJ connected to a resonator27 -- 29. In this sort of setup, charge trans- fer through the JJ leads to excitations in the resonator, and therefore allows to convert energy carried by charge quanta into quantum microwave photons. In these de- vices measurement of both the Josephson current and the emitted microwave radiation is possible, a distinct advan- tage in comparison to other recently proposed transport setups30,31, which show similar nonlinear features like bifurcations, period multiplication and up- and down- conversion32 -- 34. Such nonlinear effects will, in fact, dom- inate the system's dynamics and therewith the charac- teristics of the Josephson current and the emitted mi- crowave radiation for driving beyond a linear regime of weak Josephson coupling. While quantum properties of the system, in particu- lar of the emitted microwave radiation have been inves- tigated widely, both theoretically35 -- 45 and within new experiments46, due to the complexity and richness of non- linear effects a deeper understanding of the purely clas- sical dynamics of the system is instructive, but also of relevance for current experimental activities29, as we will discuss below. This is especially the case, as the nonlin- earity enters the system in a peculiar way. It does not stem from a nonlinear (static) potential, but rather from the manner of coupling the drive (JJ) to the resonator. The Josephson phase in this setup is thus not fixed by the external voltage, but appears as a dynamical degree of freedom manifest in a time-dependent effective poten- tial determining the phase dynamics. We note in passing that related nonlinear phenomena have recently been ex- plored for Josephson phase slip devices47. In this paper we present analytical and numerical in- vestigations in the regime, where the system's dynamics is described by classical Josephson equations. While the features found are to an extent common to a wide class of nonlinear classical systems, which specific effects in what distinct manner are realized and how they are observed in this new type of nonlinear system is an intriguing open question. To tackle it, we will first present the system under study in more detail and introduce the analytical methods used in Sec. II. The following sections cover the fundamental resonance (Sec. III) and higher order reso- nances (Sec. IV). The influence of a thermal environment at finite temperatures is investigated in Sec. V, before we conclude in Sec. VI. II. CIRCUIT DYNAMICS We consider a circuit (see Fig. 1), where a Josephson √ junction (JJ) is placed in series with a resonator with only LC being relevant. a single mode of frequency ω0 = 1/ The total impedance Zt(ω) seen by the tunneling Cooper pairs consists of the combination of the capacitance CJ of the JJ and the parallel LC resonance with finite Q factor. Since experimentally CJ (cid:28) C, one has Zt(ω) ≈ Z(ω) with Z(ω) = 1 C ω i(ω2 − ω2 0) + ωω0/Q . (1) Based on Kirchhoff rules, equations of motion for the circuit, biased by a dc-voltage V , are found, ϕ Q ϕ+ω2 0ϕ+ω0 +(EJ /φ2 0C) sin(φ) = 0 , φ = ϕ+ωJ t , (2) expressed in terms of the resonator's phase variable ϕ = −(2e/¯h)(cid:82) dtVLC(t), and the Josephson phase φ = (2e/¯h)(cid:82) dtVJ (t) with V = VLC + VJ .1,48 Here, ωJ = 2eV /¯h denotes the driving frequency, φ0 = ¯h/2e the re- duced flux quantum, and the time derivative · = d/dt. This set of equations can be cast into an equation of motion for a fictitious particle with an effective mass, m = φ2 0C, moving in a harmonic potential and coupled to an external time-periodic, position dependent force, i.e., m ϕ + mω2 0ϕ + m ω0 Q ϕ + EJ sin(ϕ + ωJ t) = ¯ξ(t) . (3) To incorporate finite temperature effects, as discussed below, thermal noise ¯ξ at temperature T is added.49 It is related to the resonator damping via the fluctuation- dissipation theorem as (cid:104) ¯ξ(t) ¯ξ(t(cid:48))(cid:105)β = 2m(ω0/Q)kBT δ(t− t(cid:48)) and (cid:104) ¯ξ(cid:105)β = 0. To explore the dynamics of the above Langevin equa- tion, it is convenient to work with dimensionless units, where times are scaled with ω0 and energies with mω2 0. This then leads to ϕ + γ ϕ + ϕ + λ sin(Ωt + ϕ) = ξ(t) , (4) where the dimensionless friction coefficient γ is related to the Q-factor of the resonator via γ = 1/Q and we further introduced the dimensionless driving ampli- tude λ = EJ /mω2 0 and driving frequency Ω = ωJ /ω0. Throughout the first part of the paper we concentrate on the limit T = 0 (ξ ≡ ¯ξ/mω2 0 = 0) and discuss the impact of thermal noise later. This form of the equation of motion, (4), is the starting point for studying the dynamics of the voltage-biased cir- cuit (cf. Fig. 1) in the rest of this paper. In particular, we 2 Figure 1. Circuit diagram of a voltage-biased Josephson junc- tion in series with a resonator with relevant mode frequency ω0 described by an effective impedance Z(ω) as specified in (1). Using a SQUID-geometry1 for the JJ allows for tuning the effective Josephson energy EJ (Φx) by an external mag- netic flux Φx. (cid:18) ϕ2 2 d dt (cid:19) + ϕ2 2 are interested in the long-time limit, where the balance between the dissipative and the driving part of Eq.(4) has pushed the system into time-periodic steady-state orbits. Considering the energy balance of the resonator obtained from Eq.(4), = −γ ϕ2 − λ ϕ sin(ϕ + Ωt) (5) = Pdiss + PJJ→HO , we easily identify the power dissipated from the res- onator, Pdiss, and the power injected into the resonator via the driving, PJJ→HO ≡ IJ VLC, which will both be considered in detail below. The dissipated power to- gether with the dimensionless Josephson current IJ = λ sin(ϕ + Ωt) constitute the main observables, which can be accessed experimentally, either averaged over many oscillation periods, or time- or frequency resolved. Note that the physical current results from the dimensionless current by multiplying φ0/L. Due to the nonlinearity present in (4) the structure of steady-state orbits will depend sensitively on the damp- ing and amplitude and frequency of the driving, giving rise to the full wealth of nonlinear phenomena such as bifurcations, up- and down-conversion etc. What dis- tinguishes the situation under consideration here from most other driven nonlinear systems is that the nonlin- earity appears not in form of a static potential energy but rather as part of the driving force. In fact, it turns out that the effective time-dependent potential giving rise to (4), namely, Veff (ϕ) = 1 2 ϕ2 − λ cos(ϕ + Ωt) , (6) can sometimes be illuminating to achieve a better under- standing of the fictitious particle dynamics. Of course, for nonlinear, time-dependent problems such as (4) explicit solutions can, in general, only be ob- tained numerically. However, analytical progress, at least 3 Figure 2. (color online) Tuning the dimensionless Josephson coupling λ the steady-state dynamics at the fundamental resonance, Ω = ωJ /ω0 = 1, accesses different ranges discussed in Secs. III A, III B, III C, which are characterized by typical phase space portraits. Within the domain III B, the dotted vertical line separates the range of regular dynamics from the range λ2 < λ, where the transition towards irregular dynamics occurs. The solid line illustrates the corresponding behavior of the dimensionless dc-current through the JJ with the shadow indicating its standard deviation when starting the steady-state dynamics with thermal initial conditions. The specific data shown for γ = 0.01 exemplify the generic behavior in the underdamped regime. In this regime, the parameters λ1/γ, λ∗ 1 and λ2 are not very sensitive to varying the friction strength. (cid:48)(cid:88) k∈MΩ for the steady state, can be made in limiting domains of parameter space. Physically, one expects a steady-state solution oscillating with the frequency of the drive, as well as higher and possibly subharmonics. Putting such an ansatz into (4), the nonlinear sin-term will again pro- duce higher harmonics with the coefficients of the various frequency components appearing in Bessel-functions. Formally, one can write a generic ansatz, ϕ(t) = ϕ0 + 1 2 ϕk eikΩt+iθk , (7) where ϕk = ϕ−k and θk = −θ−k so that solutions are real-valued. The sum runs over a suitable set of rational numbers MΩ with the prime indicating that k = 0 is excluded. For example, below we will study the fundamental resonance at Ω = 1 implying MΩ = Z. Other situations are driving at Ω = n (n integer) with MΩ = {ν/nν ∈ Z} and at Ω = 1/n (n integer) with MΩ = Z. Of course, in a perturbative treatment only a finite number of coefficients is taken into account. Now, inserting this ansatz into (4) yields a nonlinear equation for the steady-state Fourier coefficients, i.e., (cid:2)ϕkeiθk (−Λk + iγΩk)(cid:3) eikΩt (cid:48)(cid:88) (cid:2)eiΩteiϕ0F+(t) − e−iΩte−iϕ0 F−(t)(cid:3) = 0, 1 2 k (8) ϕ0 + + λ 2i with Λk = k2Ω2 − 1 and F±(t) = (cid:81) F ± k (ϕk, t). These k>0 latter functions F ± k (ϕk, t) = ∞(cid:88) l=−∞ il(±1)lJl(ϕk)eil(kΩt+θk) (9) contain Bessel functions Jl(·) of the first kind and of in- teger order. Eq. (8) serves as a starting point for per- turbative treatments in various ranges of the dynamics in order to gain a deeper understanding of the numerical findings based on (4). III. FUNDAMENTAL RESONANCE We will start in this section with the dynamics near and at the fundamental resonance where the driving fre- quency matches the Josephson frequency so that Ω = 1. According to (4), in absence of noise only two dimen- sionless parameters are left which determine the nature of steady-state orbits, namely, the friction γ and the driv- ing amplitude λ. Current experimental realizations are operated in the underdamped regime with a fixed γ (cid:28) 1 (Q-factors vary from 10 to about 1000) and varying driv- ing strengths. In the classical domain that we consider here, this leads from simple linear to strongly nonlinear phase space patterns, the structure of which is reflected in specific observables such as charge current and photon flux. Before we study the details, let us give a brief qualita- tive account of the different dynamical ranges when λ is varied (cf. Fig. 2): (i) In the regime of weak driving (Sec. III A), the dynam- ics changes from linear to nonlinear towards a threshold λ1, where a first bifurcation occurs. Perturbative treat- ments capture this transition fairly accurate. While the oscillation amplitude already reflects nonlinearities of the system, phase-space orbits in this regime are basically ellipses with only small deviations from the harmonic limit. Physically, the initial quadratic rise of the dc- current through the JJ flattens with increasing λ until it saturates at λ = λ1. (ii) For an intermediate range λ1 < λ < λ2 (Sec. III B) the dynamics is dominated by the full nonlinearity of the driving force but remains still regular. Within this do- main further bifurcations occur, however, without affect- ing the dc-current through the JJ which stays basically at its value somewhat above λ1. Phase-space orbits in- creasingly loose their harmonic-like ellipsoidal structure and become potato-shaped with deformations in form of 'dips', thus reflecting the fact that the effective time- dependent potential Veff (ϕ) (6) turns from mono-stable to multi-stable. 4 (iii) With even further increasing λ > λ2 (also Sec. III B), the system displays strong sensitivity to initial conditions (multiple steady-state orbits) and eventually irregular phase space patterns. When this domain is approached, the dc-current through the JJ grows substantially in par- allel with larger current variations depending on the driv- ing amplitude. (iv) For extremely strong driving λ > λ3 (cid:29) 1 and finite friction (Sec. III C), the dynamics turns into a regular motion again with a large time scale separation between the motion happening within one of the local wells in Veff (t, ϕ) and global dynamics exploring Veff over a wide range of ϕ. A. Weak driving regime As argued above, phase-space orbits in the weak driv- ing regime remain basically ellipses, while the amplitude becomes nonlinear until a bifurcation occurs. Thus, the steady-state orbit (7) is dominated by Fourier coefficients ϕ0, ϕ±1. The corresponding equations are easily obtained by projecting Eq. (8) on the respective Fourier modes as and with Λ1 = Ω2 − 1 ϕ0 + λJ1(ϕ1) cos(ϕ0 − θ1) = 0 ϕ1[cos(θ1)Λ1 + γΩ sin(θ1)] − λ[J0(ϕ1) sin(ϕ0) − J2(ϕ1) sin(ϕ0 − 2θ1)] = 0 ϕ1[− sin(θ1)Λ1 + γΩ cos(θ1)] − λ[J0(ϕ1) cos(ϕ0) + J2(ϕ1) cos(ϕ0 − 2θ1)] = 0 . (10a) (10b) (10c) This set of equations for {ϕ0, ϕ1, θ1} can be solved ap- proximately for small ϕ0, ϕ1 by exploiting that for the Bessel functions one has Jk(x) ∼ O(xk) for x (cid:28) 1. One first gains from (10a) in leading order ϕ0 ≈ −λJ1(ϕ1) cos(θ1) ≈ −λϕ1 cos(θ1) so that driving depen- dent terms in (10b) are of higher order in the small pa- rameter λ. This yields the phase of oscillations, tan(θ1) = − Λ1 Ωγ (11) with θ1 = 0 at resonance. Inserting this result into (10c) leads close to resonance and in leading order in ϕ0 to λ(cid:112)Λ2 1 + Ω2γ2 ϕ1 = [J0(ϕ1) + J2(ϕ1)] . (12) The known result for a driven harmonic oscillator is re- gained for ϕ1 (cid:28) 1 while for somewhat larger driving nonlinearities in the Bessel functions tend to play a role. This type of orbit, with amplitude named ϕI 1 henceforth, which continually evolves from the harmonic-oscillator type of solution, is the only stable orbit until at a crit- ical driving strength λ1 a second solution appears with amplitude ϕII 1 . In contrast to the harmonic-oscillator type orbit ϕI 1, the new orbit ϕII 1 exists even in absence of dissipation, and it is in this limit that it can easily be found analyt- ically: putting γ = 0 in (10b), (10c) taken at resonance Λ1 = 0, we assume ϕ0 = 0 to find the phase θII 1 = π/2 from (10a). Its amplitude then follows via (10c) from J0(ϕ1) − J2(ϕ1) = 2 d dϕ1 (13) as ϕ1,II (cid:39) 1.841 independent of the driving amplitude. For finite friction and away from resonance orbits are obtained numerically from (4). J1(ϕ1) = 0 It turns out that in a range beyond the threshold λ1 the type-II orbit is the only stable one, while the harmonic-oscillator type-I orbits become unstable at this bifurcation point. The threshold λ1 is determined by the condition that the amplitudes of both solutions match, ϕI 1 , i.e., according to (12) λ1/γ = ϕII 1 /[2J0(ϕII In phase space both types of solutions display ellipsoidal orbits. The transition from type-I to type-II solutions at the bifurcation point λ1 has also been found in the classical limit of a quantum de- scription within rotating-wave approximation in36,37. 1(λ1) = ϕII 1 )] ≈ 2.912. 5 Figure 3. (color online) Left: Mean dc-current (cid:104)IJ(cid:105) through the JJ in steady state and on resonance Ω = 1 as a function of the scaled driving amplitude λ/γ for (from bottom to top) γ=0.005 (blue), 0.01 (black, cf. Fig. 2), 0.02 (red). Right: Power transfer PJJ→HO from the JJ to the resonator during one period of the oscillations in the steady state at t∞ = n·2π,n (cid:29) 1, n ∈ N. The power transfer is shown for γ = 0.01 and at four different driving strengths (also marked in the left panel): λ/γ=0.1 (A), 2 (B), 3 (C) and 3.5 (D). While the harmonic-oscillator type-I solutions gain energy from the drive nearly during the whole oscillation cycle (A and B), for type-II oscillations the energy gained during part of the cycle is partly flowing back to the drive in other parts (C and D). In consequence, increasing the driving strength beyond λ1/γ ≈ 2.9 does not further increase oscillation amplitude and current. An intuitive understanding of the saturation of the oscillation amplitude, when the driving strength is in- creased beyond λ1 and, indeed, of the nature of the type- II orbit is offered by the numerical results in Fig. 3. How the resonator acts back onto the charge transfer in the JJ below/above the threshold is monitored by the en- ergy transferred from the JJ to the resonator [cf. (5)], i.e., the power PJJ→HO(t) = −λ ϕ(t) sin[ϕ(t) + Ωt] in the right panel of Fig. 3. Sufficiently below the threshold λ1/γ ≈ 2.9, energy is nearly unidirectionally injected into the resonator, i.e., the drive pushes energy into the oscil- lator at each time of the oscillation cycle. The phase shift of the type-II oscillations, however, results in an oscilla- tion which extracts energy from the drive during one part of the oscillation cycle, and pushes energy back during another part. Increasing the driving strength beyond λ1 will thus increase energy in- and back-flow, but will not further increase the net gain over a full cycle; hence, the Figure 4. (color online) Top: Dissipated power Pdiss from the driven JJ+resonator system for γ = 0.01, Ω = 1 and various driving strengths λ1 < λ < λ2 around the second bifurcation λ∗ 1. Bottom: Snap-shots of the effective time- dependent potential Veff (ϕ) (6) for various driving strengths (see top for the color code) beyond the first threshold λ > λ1 and at times Ωt = (2n + 1)π, n ∈ N. saturation of the oscillation amplitude at ϕ1,II (cid:39) 1.841. Experimentally, the transition from type-I to type-II or- bits is seen as a saturation in the dc-current through the JJ, i.e., (cid:104)IJ(cid:105) = λ(cid:104)sin[ϕ(t)+Ωt](cid:105)Ω , where the time average (cid:104)·(cid:105)Ω is taken in steady state and over several oscillation periods, see left panel of Fig. 3. The dc-current result- ing from this averaging is (approximately) proportional 1(cid:105) [analytically found from Eq. (12) and (13)]. Physi- (cid:104)ϕ2 cally, this proportionality origins from balancing dissipa- tion (proportional to the stored energy) and power input. Accordingly, the quadratic dependence (cid:104)IJ(cid:105) ∝ λ2 ∝ E2 J in the regime of very weak driving passes over to a lin- ear dependence for somewhat stronger driving but still before the threshold. This changeover is reminiscent of the changeover from the perturbative domain of sequen- tial (Coulomb blockade) to the regime of coherent charge transfer (phase coherent regime). Note that the various numerically found current curves in Fig. 3 approximately just scale with γ, but for stronger damping the bifur- cation threshold is shifted below the result λ1/γ ≈ 2.9 calculated for the γ → 0 limit above. B. Beyond the first threshold Beyond the first threshold λ > λ1, type-II orbits dic- tate the dynamics until a second bifurcation occurs at BBAACCDD−0.050Pdiss−2020121.40.90.56 Figure 5. (color online) Various classes of steady-state orbits (outer panels) depending on the initial conditions (inner panels) for γ = 0.01 and strong λ = 1 (left) and very strong driving λ = 2.455 (right). The color code for the initial conditions in phase space corresponds to a particular type of steady-state orbit in phase space. Red dots result from Poincar´e plots and indicate points of return for steady-state orbits after one period 2π/Ω. 1 ≈ 0.8 with only a very weak dependence on the fric- λ∗ tion strength in this underdamped regime. The emer- gence of this class of orbits is characterized by a substan- tial deformation of the ellipsoidal phase-space structure (see Fig. 2), which is related to a changeover of the ef- fective potential Veff (ϕ) (6) from being essentially mono- stable to dominantly multi-stable (see Fig. 4). A particle moving (rather weakly damped) in the time- dependent effective potential will encounter a deep well during its passage through ϕ = 0 in one direction, while on the way back in the second part of its oscillation cycle (see lower panel of Fig. 4), the well is less pronounced (corresponding to the situation below λ∗ 1) or even turns into a barrier around ϕ = 0 (situation well above λ∗ 1). In- deed, in the power dissipated from the driven system into the reservoir Pdiss(t) = −γ ϕ2(t) (upper panel of Fig. 4), the maximal amplitude at Ωt = 2nπ (n integer) shows a local maximum right around the bifurcation value λ∗ 1, whereas the speed of the fictitious particle (and concur- rently Pdiss) at Ωt = (2n + 1)π is more and more reduced and even develops a local minimum when passing through ϕ = 0. While this second bifurcation has almost no effect on the dc-current (cf. Fig. 2), its appearance is detectable in the discussed features of the dissipated power. With further increasing the driving λ > λ∗ 1 and initial conditions close to the phase space origin, further bifur- cations occur that we do not need to discuss in detail here. It is important to note though that each bifurca- tion is associated with a change in stability meaning that only the newly emerging orbits determine the dynamics beyond each bifurcation threshold. Independent of the appearance of new orbits, the dc-current IJ,dc stays ba- sically constant over a wide range of driving amplitudes λ1 < λ < λ2 ≈ 1.6 (see Fig. 2). Similar to λ∗ 1 the numer- ical value of λ2 depends only very weakly on the friction strength in the underdamped regime. For the driving strengths considered so far, it has been sufficient to consider initial conditions {ϕ(0), ϕ(0)} close to the phase space origin only. Now, that steady-state orbits tend to explore larger domains in phase space, one may wonder about the impact of initial conditions lo- cated in these regions. This is illustrated in Fig. 5 (left), where classes of steady-state orbits are studied depending on their initial conditions for moderately strong driving λ = 1, i.e., λ1 < λ < λ2. conditions {ϕ(0), ϕ(0)}, one now asymptotically finds three different classes of steady-state orbits in phase space. This dynam- ical multi-stability reflects the multi-stability of the effec- tive potential Veff (ϕ) due to a strong nonlinearity. How- ever, in this regime of driving-strengths, λ1 < λ < λ2, the existence of multiple types of steady-state orbits will usually not be of experimental relevance. This is due to the fact, that multi-stability only occurs if the fictitious particle leaves local wells in Veff which requires at least an energy of 2λ . This could, in principle, be done by prepar- ing the system initially (a difficult task though) such that EHO( ϕ(0), ϕ(0)) > 2λ: See the dark blue area in the left panel of Fig. 5, indicating initial conditions around the phase space center which lead to one and the same for different Indeed, sets of initial steady-state orbit. Starting from an equilibrated circuit, however, domains of initial conditions leading to addi- tional steady-state orbits, do play a role only at highly elevated temperatures, kBT /mω2 0 > 2λ. For the current experimental situation, this regime is not relevant. For λ > λ2 ≈ 1.6, one enters again a qualitatively new regime. It is characterized by a sharp rise of the dc- current through the JJ by almost an order of magnitude (see Fig. 2). Now, even for initial conditions close to the phase space origin, multiple types of stable steady-state orbits exist, both exploring large domains in phase space, see Fig. 5 (right). Some of these orbits are covered only after multiples of the fundamental period 2π/Ω, in con- trast to the regime λ < λ2 [cf. single (multiple or single) Poincar´e-plot points in the orbits of the left (right) panel of Fig. 5]. Hence, the sensitivity with respect to initial conditions grows substantially, thus marking the onset of irregular, chaotic-like behavior for sufficiently large driv- ing amplitudes λ > λ2. A detailed analysis of properties and characteristics of possible chaotic dynamics in this domain is beyond the scope of this paper and will be presented elsewhere. C. From multi-well to elevator dynamics Interestingly, in the regime of extremely strong driv- ing, λ > λ3 (cid:29) 1, (and for finite friction) regular dynam- 7 ics dominates again (see Fig. 6). At this driving strength the periodic part of Veff is pronounced enough to create a multitude of local minima in the superimposed quadratic potential, see Fig. 6(b). We mention in passing that this multi-well pattern of Veff has some analogy to the po- tential profile of superconducting quantum interference devices (SQUIDs). In contrast to SQUIDs, however, the potential Veff (ϕ), cf. Eq. (6), combines a static quadratic with a time-dependent sinusoidal potential. For somewhat stronger friction [upper The dynamics is then easily understood; most simply in the completely underdamped and the strongly over- damped cases. In the completely underdamped case [up- per left of Fig. 6(a)] the particle essentially undergoes a simple oscillation in the quadratic potential over a wide ϕ region, running with high energy over the potential wells, which then causes slight wiggles in the phase-space orbit. right of Fig. 6(a)], a fictitious particle runs periodically through a cycle of localized and de-localized motion: Starting some- where in the low energy sector, it gets trapped quickly in one of the local minima of the potential close to the global minimum of Veff (ϕ) (6). It is then transferred up in energy by the driving term ∝ cos(ϕ + Ωt) while being trapped in this local well. During this process the poten- tial barrier of the respective local well shrinks until the particle can escape to run towards the global minimum while loosing substantial energy so that it gets trapped close to the global energy minimum again. This type of "elevator" dynamics leads to increasingly simpler phase space patterns towards the overdamped regime. Phys- ically, during the trapping period ϕ grows almost lin- early with ϕ ∼ 1/Ω, whereas the motion towards the global minimum is associated with an almost instanta- neous drop in amplitude accompanied by a large increase in momentum. As a consequence, based on the second Josephson relation V (t) ∝ ϕ(t), one expects to observe strong voltages pulses with frequencies much lower than the driving frequency Ω. Figure 6. (color online) (a) Phase space portraits of steady- state orbits with Poincar´e points (red) in the regime of ex- tremely strong driving with λ = 30 >∼ λ3 and γ = 0.01 (top left), 0.025 (top right), 2.5 (bottom right), 5 (bottom left). Sketched in (b) is the particle dynamics in the effective poten- tial Veff (ϕ), see Eq. (6), corresponding to phase space orbits in the right panels of (a). For moderate damping the particle is trapped in one of the local wells, 'elevated' upwards in Veff (ϕ) [the motion towards large negative ϕ-values with small ϕ in (a)] until its escape and consequent retrapping. IV. HIGHER ORDER RESONANCES So far we have considered driving the system at (or close to) the eigenfrequency, ω0 ≡ 1, of the resonator. At sufficiently strong driving, the system's response then contains higher harmonics. It is, thus, also interesting to drive at frequencies Ω (cid:54)≈ 1, where higher or subhar- monics of the drive can become resonant with the eigen- frequency. Physically, the drive is detuned, of course, simply by changing the applied dc-voltage bias. Figure 7 shows the steady-state power spectrum, ϕ(ω)2, i.e., the response of the system at frequency ω if driven at Ω. Such spectra have been recently in- vestigated experimentally in Ref. [29]. Resonances [i.e., stable steady-state solutions of (4) or (8) with large am- plitudes] are found for driving frequencies Ωn = n for in- teger n (cid:54)= 0, where the system responds at ω = 1, 2, 3, . . ., and for driving frequencies Ω 1 = 1/n with response fre- n 8 Figure 8. (color online) DC-current through the JJ vs. driv- ing frequency Ω and driving amplitude λ for γ = 0.1. For weak driving the system is nearly linear and dominated by the fundamental resonance, Ω = ω0 = 1, while for stronger driving the subharmonic resonances at Ω = Ω 1 = ω0/n and n the multi-photon resonances at Ω = Ωn = nω0 become ap- parent. Note, the sharp onset of two-photon processes at λc ≈ γΩ = 0.2 described in Subsec. IV A, while subharmonic resonances increase smoothly and shift with increasing λ as discussed in Subsec. IV B (cf. also Fig. 9). photon quanta ¯hΩac, i.e., 2evdc/¯h = nΩac, n ≥ 1 being integer. For the present situation, however, one could argue in two ways (we temporarily return to physical di- mensions): (i) The dc-voltage which determines the driv- ing frequency Ω = 2eV /¯h must be multiples of the res- onator frequency ω0 implying Ω = nω0, n ≥ 1 integer, or (ii) the excitation of the resonator by one energy quantum ¯hω0 requires multiples of ac-photon quanta n¯hΩ, n ≥ 1 integer, i.e., Ω = ω0/n. Along these lines, one could inter- prete either the subharmonic Ω 1 or the higher harmonic n Ωn resonances as Shapiro steps. The fundamental differ- ence to the conventional set-up to observe Shapiro steps is that there the dynamics of the Josephson phase φ is fixed by the external voltage according to 2eV (t) = φ/¯h, while here the Josephson phase appears as a dynamical degree of freedom fixed by the dynamics of the resonator phase ϕ [see Eq. (2)]. A. Integer multi-photon processes Now, we start with multi-photon resonances Ωn = n, where the system asymptotically responds mostly with frequency ω = 1 (down-conversion). In particular, we focus on the generic case Ω = 2 (see Fig. 7) which can be interpreted as a parametric resonance. The general argument for the appearance of this type of resonances can be directly read off from the equation for the Fourier mode amplitudes (8): For Ω = n we seek orbits with a time dependence dominated by cos(t) im- plying that Fourier coefficients ϕ 1 with kn = 1 dominate the expansion (7). Accordingly, in (8) F± ≈ F ± ) and (ϕ 1 n 1 n n Figure 7. (color online) a) Power spectrum, ln( ϕ(ω)2), of steady-state orbits ϕ(t) for a range of driving frequencies Ω in the underdamped regime γ = 0.1 and at moderate driving amplitude λ = 0.2. For the sake of presentation the spectrum is numerically broadened by taking the Fourier transform of a finite time signal. b) Dominant components ϕk2 of ϕ(t) [see (7)], corresponding to lines ω = kΩ in a). Panel c) shows a cut of the power spectrum at the (shifted) subharmonic resonance Ω = 0.365 ≈ Ω 1 at a larger driving amplitude λ = 0.7, where the system dominantly responds with ω = 3Ω ≈ 1. 3 quencies ω = 1/n, 2/n, 3/n, . . .. The situation, where the system is driven with Ω and responds with ω < Ω is called down-conversion, the situation, where it responds with ω > Ω up-conversion. Apparently, for a driving fre- quency Ω2 the system dominantly responds with ω = 1 while contributions of higher harmonics are weak. In con- trast, driving with Ω 1 one observes a response in which 2 dominantly frequencies with ω = 1/2 but also the first few higher harmonics ω = 1, 3/2, . . . are present. While the power spectrum, ϕ(ω)2, shown here gives a direct intuitive link to the system dynamics, comparable in- formation can also be extracted from the directly mea- surable spectrum of light emission. Before we turn to details of its resonant features below, let us address the consequences of varying the dc-bias (and thus the driv- ing frequency) for the dc-current through the JJ. As de- picted in Fig. 8, current peaks appear indeed at driving frequencies Ωn and Ω ≈ Ω 1 with n ≥ 1 integer with a characteristic shift towards Ω > Ω 1 in the subharmonic n domain. These features may remind of Shapiro steps1 known from JJs subject to dc- and ac-voltages of the form V (t) = vdc + vac cos(Ωact), an interpretation which for the present situation is somewhat confusing though: The conventional argument for the existence of Shapiro steps is that resonances appear whenever the energy gap induced by the dc-voltage 2evdc matches multiples of ac- n a)b)c)k1/31/2n n n upon projecting onto orbits with e±it, one arrives at time- which in- independent equations for the coefficients ϕ 1 clude Bessel functions Jn−1(ϕ 1 ) and Jn+1(ϕ 1 ) [see (9)]. In the particular case of Ω = 2, we may write for small amplitudes λ sin[ϕ(t)+2t] ≈ λ[sin(2t)+ϕ(t) cos(2t)], thus giving in the equation of motion (4) effectively rise to a parametrically driven harmonic oscillator scenario. For this system steady-state solutions only exist below a crit- ical, damping dependent value λc = 2γ, while above that threshold oscillation amplitudes will grow infinitely in time. Of course, here the nonlinearity of the full problem prevents this divergence to occur. Then, the simplest ansatz, ϕ(t) ≈ ϕ 1 ), taking only oscillations at the oscillator's resonance frequency into account, yields for the amplitudes cos(t + θ 1 2 2 [J1(ϕ 1 2 ) + J3(ϕ 1 2 )]/ϕ 1 2 = γ/λ . (14) this equation only exist above Solutions of the parametric-resonance threshold, λ > λc = 2γ, with a ≈ −π/4 (cf. also [36]). The full solution with phase θ 1 an additional contribution ϕ1 (cid:28) ϕ 1 and θ1 ≈ π/2 2 shows similar threshold behavior with some quantitative corrections (cf. the threshold in Fig. 8 and Fig. 9). 2 B. Subharmonic resonances In the driving-frequency range Ω < 1 below the fun- damental resonance, the system will dominantly respond with a few higher harmonics ω = Ω, 2Ω, . . . , see Fig. 7. Resonances appear approximately at driving frequencies Ω 1 , where one of these harmonics matches the eigenfre- n quency ω0 = 1. This resonant response results, in par- ticular, in a strong contribution to the dc-current close to Ω 1 , where a shift of the resonance with increasing n driving strength can be observed, see Fig. 8. As a specific example, we consider the shift of the resonance close to Ω = 1/2 for γ → 0. Taking then an ansatz for the steady-state orbits (7) of the form ϕ(t) ∼ ϕ1 cos(Ωt+θ1)+ϕ2 cos(2Ωt+θ2) with Ω ≈ Ω 1 and considering (8) we find, that the Fourier coefficients for this type of steady-state orbits have to be gained from nonlinear equations including products of Bessel func- tions J0(ϕ2)J1(ϕ1), J2(ϕ2)J1(ϕ1). Assuming for suffi- ciently strong driving a dominant response with 2Ω ≈ 1 and thus ϕ1 (cid:28) 1, ϕ2, we take into account only the low- est order of the Bessel functions for ϕ1 and find the res- onance condition Λ2 = −λ cos θ1ϕ1J1(ϕ2)/ϕ2 > 0 from the 2Ω-projection of Eq. (8). That this shift is indeed positive, as seen in Fig. 8, can be analytically confirmed for this limit by analyzing the Ω-projection. Increasing the driving strength, the shift that depends on the mixing between the driving frequency and the first higher har- monics grows further. Remarkably, at least for moderate driving similar shifts do not occur for the Ωn-resonances (including the fundamental one), as apparent from the 2 9 Figure 9. (color online) Time-averaged steady-state energy in the resonator for γ = 0.1 and at different driving strengths: λ =0.1 (black), 0.15 (blue), 0.2 (green), 0.5 (red). For the two- photon resonance, Ω = 2, there is a rather sharp threshold at λ >∼ λc = 2γ, while subharmonic resonances, Ω ≈ 1/n gradually increase and shift with increased driving strength. numerical data in Fig. 8 and analytically following Eq. (10) and Sec. IV A. Note, that the multi-photon processes and subhar- monic resonances discussed here and shown in Figs. 7, 8 all occur for comparatively weak driving; for strong driving similarly rich behavior for driving at sub- and higher-harmonic resonance frequencies may be expected, as found in Sec. III B, III C for the fundamental reso- nance. C. Energy transfer As already discussed for the fundamental resonance in Sec. III A, the energy transfer from the driving source to the resonator is another tool to reveal details of the nonlinear dynamics. Experimentally, it is accessible as (mean) photon emission from the resonator. Here, we depict resonance curves of the time averaged resonator energy EHO = (cid:104) ϕ2/2 + ϕ2/2(cid:105)Ω, see Fig. 9. As expected, the resonances discussed in the previous sections appear . in form of pronounced peaks at frequencies Ωn and Ω 1 n In the subharmonic regime the shift in the resonances towards Ω > Ω 1 for larger λ is seen as well. We note also n the threshold λ > λc for the occurrence of the parametric resonance at Ω2. V. THERMAL NOISE In actual experimental realizations, noise stemming from various sources is always present and may sensi- tively influence the dynamics of the JJ+resonator device. In the Langevin equation (4) we restricted ourselves to +012EHO10-2100514131210 Figure 11. (color online) Mean dissipated power at temper- ature kBT = 0 (top two panels) and kBT = 0.1 (bottom) with γ = 0.1 and Ω = 2 for various driving strengths λ=0.15 (black), 0.2 (blue), 0.5 (green). The threshold for the para- metric amplification in absence of noise is λc ≈ 2γ = 0.2. denotes the average over initial conditions according to a thermal distribution. In fact, the size of current fluctua- tions directly indicates ranges in parameter space, where the underlying asymptotic dynamics displays either bi- furcations or irregular behavior, cf. Fig. (2). Thermal noise according to the full Langevin dynam- ics has a similar impact. Here, we focus on two do- mains, namely, the domain around the first bifurcation λ ≈ λ1 for Ω = 1 (see Sec. III A) and the domain around the parametric resonance λ ≈ λc ≈ 2γ for Ω = 2 (see Sec. IV A). For the first case, in Fig. 10 the mean steady-state am- plitude max [(cid:104)ϕ(t → ∞)(cid:105)ξ] is shown. Even in the weak driving regime the linear response of the system gets in- fluenced by temperature as thermal fluctuations become larger and increasingly explore the nonlinearity of the ef- fective potential. More substantial deviations occur for λ → λ1, where the bifurcation is increasingly smeared out and the overall oscillation amplitude decreases at el- evated temperatures. In the second case, the parametric-resonance thresh- old, the situation is a bit more intricate. Here, thermal noise may mix higher harmonics in the Fourier expansion (7) with the consequence that down-conversion already occurs prior to the threshold λc, see Fig. 11. Accordingly, already for driving strengths below the zero-temperature threshold, thermal noise excites oscillations with the res- onant frequency ω = Ω2/2 = 1 with a drastically in- creased amplitude compared to the response at the driv- ing frequency ω = Ω2 in the absence of noise (e.g., an increase by about two orders of magnitude for the black Figure 10. (color online) Mean steady-state oscillation ampli- tude, ϕmax = max [(cid:104)ϕ(t → ∞)(cid:105)ξ], vs. driving strength aver- aged over 10000 realizations of thermal noise at temperatures kBT =0 (black), 0.01 (green-dashed), 0.1 (blue), 0.25 (red) for friction parameter γ = 0.01. thermal noise related to the finite photon lifetime in the resonator via the fluctuation-dissipation theorem. An- other major source of noise may be local voltage fluc- tuations at the JJ that may induce charge localization (Coulomb blockade). However, for the present situation the impact of these fluctuations is of minor relevance. Al- though assuming a purely classical regime implies, that temperatures are high as compared to quantum fluctua- tions, the circuit is nonetheless operated at temperatures sufficiently low compared to other energy scales of the system, i.e., in dimensional units kBT < mω2 0, EJ . The simplest way to include finite temperature effects is to perform a thermal averaging over initial conditions, i.e., we start initially with a thermal distribution of phase space variables but then follow a deterministic time evo- lution according to (4) for ξ = 0. While somewhat in- consistent, this scenario accounts for the fact that precise initial conditions are experimentally not feasible. From a purely theoretical perspective, it allows to analyze the sensitivity of steady-state orbits onto initial conditions. The full description of thermal noise works with the Langevin equation (4) and seeks for phase space orbits averaged over many noise realizations (cid:104)ϕ(t)(cid:105)ξ,(cid:104) ϕ(t)(cid:105)ξ. In this latter situation, asymptotically thermal fluctuations may induce transitions between various steady-state or- bits even when their respective initial conditions are well separated in phase space. The main effect of an initial thermal distribution is apparent in the regime λ > λ2 (see Fig. 2), where bare steady states tend to depend sensitively on the initial conditions. Accordingly, when averaged over an initial thermal distribution, phase space structures in steady state are washed out. This in turn gives rise to relatively large current fluctuations (cid:104)(Idc −(cid:104)Idc(cid:105)0)2(cid:105)0, where (cid:104)···(cid:105)0 600.010.020.03'max00.511.52lines in Fig. 11). Above-threshold oscillations are some- what reduced by temperature. Nonetheless, a transition remains clearly visible, for example, in the mean dissi- pated power (cid:104)Pdiss(t)(cid:105)ξ. Furthermore, an offset emerges such that (cid:104)Pdiss(t)(cid:105)ξ > 0 for all times, which can be re- lated to the thermal energy continuously injected into the system. VI. CONCLUSION AND DISCUSSION In this paper we analyzed the classical dynamics of a circuit, where a single relevant resonator mode interacts with a dc-voltage biased JJ. This problem can be mapped onto the dissipative dynamics of a fictitious particle mov- ing in a nonlinear, time dependent potential, where in contrast to conventional settings the nonlinearity appears as part of the driving, while the static part of the po- tential is purely harmonic. In the regime of moderate to large Q-factors (underdamped regime) and weak thermal noise, steady-state orbits and corresponding observables are determined by basically only two parameters, namely, the dimensionless driving strength (Josephson coupling) λ = EJ /mω2 0 and the dimensionless driving frequency Ω (in units of ω0). At the fundamental resonance Ω = ωJ /ω0 = 1 this sys- tem displays a changeover from a linear response regime for weak driving towards a strongly nonlinear behav- ior for strong driving. The various dynamical domains leave their signatures in the dc-current flowing through the JJ and in the microwave power emitted from the resonator and are thus directly accessible experimen- tally. Resonances are also found when driving with either higher harmonics (Ω = n, n integer) or sub-harmonics (Ω ≈ 1/n, n integer), while the system responds with the fundamental frequency thus corresponding to pro- cesses of down- and up conversion. These features can also be detected by either monitoring the dc-Josephson current or the radiated microwaves. Due to its high de- gree of tunability the resonator+JJ circuit thus allows to study the full wealth of classical nonlinear dynamics in one-dimensional driven, dissipative systems. The im- pact of weak thermal noise is most prominent close to bifurcations of steady-state orbits. At this point let us discuss a typical set of parame- ters for circuit designs that allows to access the physics discussed above: The classical regime with weak ther- 11 0 (cid:29) kBT (cid:29) ¯hω0. For a res- mal noise requires that mω2 onance frequency of ω0 ∼ 5 GHz, this can be realized with an LC-circuit with C ∼ 5 pF operated at temper- atures T ∼ 150 mK. One then has mω2 0 ∼ 0.08 meV, kBT ∼ 0.015 meV, and ¯hω0 ∼ 0.003 meV. In the phase regime of the JJ, one further needs EJ (cid:29) EC which applies for a typical EJ ∼ 1 meV. An external mag- netic flux allows to tune this maximal Josephson cou- pling down to a factor of about 100, i.e., within the range EJ ∼ 0.01 . . . 1 meV. Present resonator designs have typical photon lifetimes over a wide range of Q- factors Q ≈ 10 . . . 104 which coincides with the (strongly) underdamped regime. Theoretically, at the fundamental resonance new dy- namical domains are associated with driving parame- ters λ1, λ∗ 1, λ2, and λ3 as depicted in Fig. 2 which, given the above parameters, translates into the following cou- pling energies: EJ,1 ≈ 0.02 meV, E∗ J,1 ≈ 0.05 meV, EJ,2 ≈ 0.1 meV, EJ,3(λ ≈ 20) ≈ 1.3 meV. Apparently, the first three coupling energies are easily accessible ex- perimentally with only EJ,3 lying at the edge of the range. The challenge here is that to cover the full range of driv- ing amplitudes within one set-up from the linear response regime EJ (cid:28) EJ,1 to EJ,3, requires to vary EJ by a factor of somewhat more than 100. Note that the superconducting gap 2∆ defines an up- per limit on the applied voltage 2eV < 2∆ which for Al junctions corresponds to ωJ (cid:28) 500 GHz, thus allowing also for driving at higher harmonics Ωn. Following this discussion, we are confident that the classical, nonlinear dynamics analyzed in this work is indeed accessible in re- alistic circuits similar to those in reference [29] and opens the door to study the interplay of driving, dissipation and resonances under well controlled conditions. ACKNOWLEDGMENTS This work benefitted from discussions with M. Blencowe, F. Portier, and A. Rimberg. JA and BK thank the Department of Physics and Astronomy, Darmouth College, Hanover, USA, for the kind hospitality. Finan- cial support from the Carl-Zeiss-Foundation (SM), the Harris-Foundation (JA), and the Deutsche Forschungs- gemeinschaft through AN336/6-1 is gratefully acknowl- edged. ∗ Present address: Fraunhofer-Institut fur Angewandte Fest- korperphysik, Tullastrasse 72, 79108 Freiburg, Germany 1 A. Barone and G. Patern`o, Physics and applications of the Josephson effect (Wiley, New York [u.a.], 1982). 2 M. F. Goffman, R. Cron, A. Levy Yeyati, P. Joyez, M. H. Devoret, D. Esteve, and C. Urbina, Phys. Rev. Lett. 85, 170 (2000). 3 J. C. Cuevas, J. Heurich, A. Mart´ın-Rodero, A. Levy Yey- ati, and G. Schon, Phys. Rev. Lett. 88, 157001 (2002). 4 M. Chauvin, P. vom Stein, H. Pothier, P. Joyez, M. E. Huber, D. Esteve, and C. Urbina, Phys. Rev. Lett. 97, 067006 (2006). 5 F. S. Bergeret, P. Virtanen, A. Ozaeta, T. T. Heikkila, and J. C. Cuevas, Phys. Rev. B 84, 054504 (2011). 6 L. Bretheau, C¸ . Girit, L. Tosi, M. Goffman, P. Joyez, H. Pothier, D. Esteve, and C. Urbina, Comptes Rendus 12 Physique 13, 89 (2012). 7 K. Wiesenfeld, P. Colet, and S. H. Strogatz, Phys. Rev. Lett. 76, 404 (1996). 8 K. Wiesenfeld, P. Colet, and S. H. Strogatz, Phys. Rev. E 57, 1563 (1998). 9 P. Barbara, A. B. Cawthorne, S. V. Shitov, and C. J. Lobb, Phys. Rev. Lett. 82, 1963 (1999). 10 L. Ozyuzer, A. E. Koshelev, C. Kurter, N. Gopalsami, Q. Li, M. Tachiki, K. Kadowaki, T. Yamamoto, H. Mi- nami, H. Yamaguchi, T. Tachiki, K. E. Gray, W.-K. Kwok, and U. Welp, Science 318, 1291 (2007). 11 H. B. Wang, S. Gu´enon, J. Yuan, A. Iishi, S. Arisawa, T. Hatano, T. Yamashita, D. Koelle, and R. Kleiner, Phys. Rev. Lett. 102, 017006 (2009). 12 U. Welp, K. Kadowaki, and R. Kleiner, Nature Photonics 7, 702 (2013). 13 I. Siddiqi, R. Vijay, F. Pierre, C. M. Wilson, M. Metcalfe, C. Rigetti, L. Frunzio, and M. H. Devoret, Phys. Rev. Lett. 93, 207002 (2004). 14 I. Siddiqi, R. Vijay, F. Pierre, C. M. Wilson, L. Frunzio, M. Metcalfe, C. Rigetti, R. J. Schoelkopf, M. H. Devoret, D. Vion, and D. Esteve, Phys. Rev. Lett. 94, 027005 (2005). 15 R. Vijay, M. H. Devoret, and I. Siddiqi, Review of Scien- tific Instruments 80, 111101 (2009). 16 N. Boulant, G. Ithier, P. Meeson, F. Nguyen, D. Vion, D. Esteve, I. Siddiqi, R. Vijay, C. Rigetti, F. Pierre, and M. Devoret, Phys. Rev. B 76, 014525 (2007). 17 A. Verso and J. Ankerhold, Phys. Rev. E 82, 051116 (2010). 217005 (2011). 28 M. Blencowe, A. D. Armour, and A. J. Rimberg, in Fluc- tuating Nonlinear Oscillators: From Nanomechanics to Quantum Superconducting circuits, edited by M. Dykman (Oxford University Press, Oxford, 2012). 29 F. Chen, J. Li, A. D. Armour, E. Brahimi, J. Stettenheim, A. J. Sirois, R. W. Simmonds, M. P. Blencowe, and A. J. Rimberg, Phys. Rev. B 90, 020506 (2014). 30 C. L´opez-Mon´ıs, C. Emary, G. Kiesslich, G. Platero, and T. Brandes, Phys. Rev. B 85, 045301 (2012). 31 R. Hussein, A. Metelmann, P. Zedler, and T. Brandes, Phys. Rev. B 82, 165406 (2010). 32 G. L. Baker and J. P. Gollub, Chaotic dynamics: an intro- duction, 2nd ed. (Cambridge Univ. Pr., Cambridge, 1996). 33 J. Guckenheimer and P. Holmes, Nonlinear oscillations, dynamical systems, and bifurcations of vector fields, 4th ed. (Springer, New York [u.a.], 1993). 34 G. Iooss and D. D. Joseph, Elementary stability and bi- furcation theory, 2nd ed. (Springer, New York; Heidelberg [u.a.], 1997). 35 C. Padurariu, F. Hassler, and Y. V. Nazarov, Phys. Rev. B 86, 054514 (2012). 36 A. D. Armour, M. P. Blencowe, E. Brahimi, and A. J. Rimberg, Phys. Rev. Lett. 111, 247001 (2013). 37 V. Gramich, B. Kubala, S. Rohrer, Phys. Rev. Lett. 111, 247002 (2013). and J. Ankerhold, 38 J. Leppakangas, G. Johansson, M. Marthaler, and M. Fo- gelstrom, Phys. Rev. Lett. 110, 267004 (2013). 39 J. Leppakangas, G. Johansson, M. Marthaler, and M. Fo- gelstrom, New Journal of Physics 16, 015015 (2014). 18 V. Peano and M. Thorwart, Phys. Rev. B 70, 235401 40 A. D. Armour, B. Kubala, and J. Ankerhold, Phys. Rev. (2004). B 91, 184508 (2015). 19 I. Serban and F. K. Wilhelm, Phys. Rev. Lett. 99, 137001 41 B. Kubala, V. Gramich, and J. Ankerhold, Physica Scripta (2007). 2015, 014029 (2015). 20 M. Marthaler and M. I. Dykman, Phys. Rev. A 73, 042108 42 S. Dambach, B. Kubala, V. Gramich, and J. Ankerhold, (2006). 21 S. Kohler, T. Dittrich, and P. Hanggi, Phys. Rev. E 55, 300 (1997). 22 M. Dykman, ed., Fluctuating Nonlinear Oscillators: From Nanomechanics to Quantum Superconducting circuits (Ox- ford University Press, 2012). 23 P. Jung, S. Butz, M. Marthaler, M. Fistul, J. Leppakangas, V. Koshelets, and A. Ustinov, Nature Communications 5, 3730 (2014). 24 N. R. Werthamer and S. Shapiro, Phys. Rev. 164, 523 (1967). 25 O. Astafiev, K. Inomata, A. Niskanen, T. Yamamoto, Y. Pashkin, Y. Nakamura, and J. Tsai, Nature 449, 588 (2007). 26 F. Chen, A. J. Sirois, R. W. Simmonds, and A. J. Rimberg, Applied Physics Letters 98, 132509 (2011). 27 M. Hofheinz, F. Portier, Q. Baudouin, P. Joyez, D. Vion, P. Bertet, P. Roche, and D. Esteve, Phys. Rev. Lett. 106, Phys. Rev. B 92, 054508 (2015). 43 M. Trif and P. Simon, Phys. Rev. B 92, 014503 (2015). 44 L. Guo and M. Marthaler, (2014), arXiv:1410.3795. 45 J. Leppakangas, M. Fogelstrom, A. Grimm, M. Hofheinz, M. Marthaler, and G. Johansson, Phys. Rev. Lett. 115, 027004 (2015). 46 O. Parlavecchio, M. Westing, M. Trif, P. Simon, J. Anker- hold, B. Kubala, J. Leppakangas, G. Johansson, J. P. Al- timiras, C., D. Vion, P. Roche, D. Esteve, M. Hofheinz, and F. Portier, Emission of non classical radiation by in- elastic Cooper pair tunneling, in preparation. 47 A. M. Hriscu and Y. V. Nazarov, Phys. Rev. B 83, 174511 (2011). 48 G. Wendin and V. Shumeiko, in Handbook of Theoretical and Computational Nanotechnology, edited by M. Rieth and W. Schommers (American Scientific Publishers, 2006). 49 H. Risken, The Fokker-Planck equation: methods of solu- tion and applications, 2nd ed. (Springer, Berlin; Heidelberg [u.a.], 1996).
1508.05259	2	1508	2016-05-22T09:06:22	Qubit quantum-dot sensors: noise cancellation by coherent backaction, initial slips, and elliptical precession	[ "cond-mat.mes-hall", "quant-ph" ]	We theoretically investigate the backaction of a sensor quantum dot with strong local Coulomb repulsion on the transient dynamics of a qubit that is probed capacitively. We show that the measurement backaction induced by the noise of electron cotunneling through the sensor is surprisingly mitigated by the recently identified coherent backaction [PRB 89, 195405] arising from quantum fluctuations. This renormalization effect is missing in semiclassical stochastic fluctuator models and typically also in Born-Markov approaches, which try to avoid the calculation of the nonstationary, nonequilibrium state of the qubit plus sensor. Technically, we integrate out the current-carrying electrodes to obtain kinetic equations for the joint, nonequilibrium detector-qubit dynamics. We show that the sensor-current response, level renormalization, cotunneling, and leading non-Markovian corrections always appear together and cannot be turned off individually in an experiment or ignored theoretically. We analyze the backaction on the reduced qubit state - capturing the full non-Markovian effects imposed by the sensor quantum dot on the qubit - by applying a Liouville-space decomposition into quasistationary and rapidly decaying modes. Importantly, the sensor cannot be eliminated completely even in the simplest high-temperature, weak-measurement limit: The qubit state experiences an initial slip that persists over many qubit cycles and depends on the initial preparation of qubit plus sensor quantum dot. A quantum-dot sensor can thus not be modeled as a 'black box' without accounting for its dynamical variables. We furthermore find that the Bloch vector relaxes (T1) along an axis that is not orthogonal to the plane in which the Bloch vector dephases (T2), blurring the notions of T1 and T2 times. Finally, the precessional motion of the Bloch vector is distorted into an ellipse in the tilted dephasing plane.	cond-mat.mes-hall	cond-mat	a Qubit quantum-dot sensors: Noise cancellation by coherent backaction, initial slips, and elliptical precession M. Hell(1,2), M. R. Wegewijs(1,2,3), and D. P. DiVincenzo(1,2,4) (1) Peter Grunberg Institut, Forschungszentrum Julich, 52425 Julich, Germany (2) JARA- Fundamentals of Future Information Technology (3) Institute for Theory of Statistical Physics, RWTH Aachen, 52056 Aachen, Germany (4) Institute for Quantum Information, RWTH Aachen, 52056 Aachen, Germany (Dated: February 14, 2018) We theoretically investigate the backaction of a sensor quantum dot with strong local Coulomb repulsion on the transient dynamics of a qubit that is probed capacitively. We show that the mea- surement backaction induced by the noise of electron cotunneling through the sensor is surprisingly mitigated by the recently identified coherent backaction [M. Hell, M. R. Wegewijs, and D. P. Di- Vincenzo, Phys. Rev. B 89, 195405 (2014)] arising from quantum fluctuations. This indicates that a sensor with quantized states may be switched off better than naively expected. This renormaliza- tion effect is missing in semiclassical stochastic fluctuator models and typically also in Born-Markov approaches, which try to avoid the calculation of the nonstationary, nonequilibrium state of the qubit plus sensor. Technically, we integrate out the current-carrying electrodes to obtain kinetic equations for the joint, nonequilibrium detector-qubit dynamics. We show that the sensor current response, level renormalization, cotunneling broadening, and leading non-Markovian corrections al- ways appear together and cannot be turned off individually in an experiment or ignored theoretically. We analyze the backaction on the reduced qubit state – capturing the full non-Markovian effects imposed by the sensor quantum dot on the qubit – by applying a Liouville-space decomposition into quasistationary and rapidly decaying modes. Importantly, the sensor cannot be eliminated completely even in the simplest high-temperature, weak-measurement limit since the qubit state experiences an initial slip depending on the initial preparation of qubit plus sensor quantum dot. The slip persists over many qubit cycles, i.e., also on the time scale of the qubit decoherence in- duced by the backaction. A quantum-dot sensor can thus not be modeled as usual as a "black box" without accounting for its dynamical variables; it is part of the quantum circuit. We furthermore find that the Bloch vector relaxes (rate 1/T1) along an axis that is not orthogonal to the plane in which the Bloch vector dephases (rate 1/T2), blurring the notions of relaxation and dephasing times. Moreover, the precessional motion of the Bloch vector is distorted into an ellipse in the tilted dephasing plane. PACS numbers: 73.63.Kv, 73.63.-b, 03.65.Yz I. INTRODUCTION The ongoing effort to mitigate the qubit decoherence due to environmental noise detrimental to quantum com- puting has made substantial progress by identifying well- isolated two-level systems [1–4] and developing efficient decoupling techniques [5–7]. However, active readout el- ements must be integrated into any quantum computer and "noise" from such sensors may soon become a rel- evant source of errors. Therefore, the unavoidable dis- turbance of the qubit evolution during a readout process gains importance, both for single-shot qubit readout [8, 9] as well as continuous qubit monitoring [10]. Another question of central importance for qubit manipulations is which parameters should be varied to switch off a sen- sor most effectively when no measurement is intended to be made. In a single-shot measurement, the goal is to achieve a strong, projective measurement that dephases the qubit state as quickly as possible. However, any measurement still takes a finite time and relaxation processes [9], exci- tation processes [11], and incoherent detector dynamics limit the detector efficiency [12–15]. For continuously monitoring the qubit evolution, by contrast, one has to realize a weak measurement. The aim is here to disturb the qubit evolution as weakly as possible to retain the (partial) purity of the quantum state [16], while avoid- ing quantum jumps and the quantum Zeno effect [17]. Understanding the backaction of a weak measurement is therefore of great interest. In this paper, we focus on the noninvasive, weak- measurement backaction exerted on a qubit by a capaci- tively coupled sensor quantum dot (SQD) [9, 18]. SQDs are attractive qubit detectors due to their strong tun- ability and higher signal-to-noise ratio as compared to quantum point contacts [19, 20] and dispersive readout schemes [21]. This derives from the fact that an SQD is an interacting quantum system. For smaller quantum dots (QDs) than typically used for readout, the elec- trons may even occupy discrete energy levels. The re- quired readout current easily leads to a strong nonequi- librium and nonstationary sensor state. This altogether makes the description of the backaction arising from an SQD on a qubit challenging. Thus, typical weak-coupling approaches to decoherence assuming an environment in equilibrium with a continuous spectrum [22–28] cannot be applied here and naive extensions are prone to er- rors as we will illustrate. Understanding the intertwined evolution of SQD and qubit is of key importance to un- derstand the measurement backaction [13, 29–32]. Out of these challenges arises the question which phys- ical effects have to be included for a consistent descrip- tion of the measurement backaction of a sensor QD on a qubit. For the qubit, we consider the simplest model that involves capacitive readout, a charge qubit, realized as a double quantum dot. An important aspect lies in the type of setup considered, namely that of indirect de- tection: one measures the conductance of a sensor QD in the attached electrodes, while only the SQD capacitively interacts with the qubit, see Fig. 1. To maximize the sensitivity, the SQD is operated at the threshold to the Coulomb blockade regime. Here, the conductance of the SQD shows the strongest response to small qubit-induced level shifts. Previous studies addressing the backaction of an SQD on a qubit in this regime focused mainly on the lowest-order approximation in the tunnel coupling Γ of the SQD to the attached electrodes [13, 16, 29–31, 33], strictly valid when operated in the single-electron tunnel- ing (SET) regime. There are two main reasons for going beyond this approximation. The first reason relates to the backaction on the qubit and its dependence on the level position, experimentally controlled by gate voltages. This is important since the level position is one of the key experimental control pa- rameters by which one can try to switch off the sen- sor backaction. In this approximation, the leading-order rates (SET ∝ Γ) become exponentially small when the level position (ε) of the SQD is tuned away from the electrochemical potentials of the electrodes (µr). Thus, when one is interested in the backaction at the onset of Coulomb blockade, next-to-leading order ∝ Γ2 cotunnel- ing processes should also be accounted for because they are only algebraically suppressed, scaling as 1/(ε − µr). One would thus naively expect that the backaction is suppressed only inversely proportional to the detuning from resonance. Yet, level-renormalization effects should also be considered; they lead to level shifts that depend logarithmically on the level position ε and therefore the response of the level renormalization to the measurement perturbation also scales as 1/(ε − µr). A central finding of our study is that level-renormalization effects in fact mitigate the naively expected cotunneling decoherence. A second reason for going beyond the lowest-order ap- proximation comes in view when one takes into account the experimentally used sensor signal: if one accounts for the first nonvanishing contributions ∝ λΓ/T that pro- duce a nonzero sensor signal, one has to include also renormalization effects since they appear in the same or- der. Basically, if one has time to measure, one has time to fluctuate as well. As already emphasized in Ref. [32], in- corporating terms ∝ λΓ/T is another reason that forces us to keep also cotunneling processes ∝ Γ2/T since in the weak-measurement limit λ ≪ Γ the latter are larger. Only when one is not interested in the sensor current, 2 one can consistently neglect cotunneling and renormal- ization effects by taking the high-temperature limit: as we show, they must either be kept or neglected together. The above-mentioned processes combine in a nontriv- ial way to give three types of backaction on the qubit [32]. First, both SET and cotunneling processes con- tribute to a stochastic switching of the SQD charge state. This switching generates a randomly fluctuating effective magnetic field acting on the qubit Bloch or isospin vector [26, 27]. This "noise" – called here the stochastic back- action – leads to a shrinking of the Bloch vector, i.e., to decoherence. In addition, there is a dissipative back- action, which is the flip-side of the measurement action: it arises whenever one accounts for a nonzero response of the sensor SET tunnel rates to the qubit state and therefore a nonzero sensor signal. Finally, there is a co- herent backaction, the most striking finding of Ref. [32]. It arises from the above-mentioned level-renormalization response and translates into torque terms involving the qubit Bloch vector. These torques and related precession effects are similar to those emerging in various other QD transport setups: It is well-known that tunneling pro- cesses can produce exchange fields leading to an (iso)spin precession in the context of spintronics [34, 35], double dots [36], molecular quantum dots [37, 38], and super- conducting devices [39]. All these level-renormalization effects arise from quantum fluctuations of electrons by tunneling into the attached electrodes. In this respect, a qubit coupled to a sensor QD is not different. An interesting question is how these different types of backaction relate to the information gained during the measurement process. Clearly, the quasistationary time-dependent current through the sensor contains in- formation about the qubit state and at the same time causes decoherence of the qubit. However, besides this fundamentally unavoidable backaction, the decoherence induced by the detector can be stronger. This can be for- mulated in terms of general inequalities relating the noise of the measured operator (here the position of the qubit electron) to the noise of the measurement signal (here the current) and additional noise cross terms [77]. The situation considered in this paper is far away from the quantum limit (meaning the above-mentioned inequality is far from being satisfied with equality). A part of our work actually focuses on a simple limit when only the stochastic backaction is accounted for but dissipative and coherent backaction are neglected. The SQD then acts rather as an "ordinary" environment and no information is obtained during the operation – a situation relevant when the detector is supposed to be switched off. Yet, even in this simple limit, there are ef- fects beyond the scope of the picture developed in Ref. 77: the fast relaxation due to switching on the sensor may affect the qubit state also in a way that depends on the initial dynamical variables of the sensor. This is not cap- tured at all by the noise inequalities mentioned above. The explicit time evolution of the sensor, which is not considered in Ref. 77, is thus also crucial to understand the backaction on the qubit. An interesting question aris- ing from this insight is how this backaction effect has to be assessed in view of the information gain. Our work could thus spur new activity on the topic of information gain during the measurement. The impact of the three backaction effects on the full transient dynamics of the qubit so far remained an out- standing question that we address in this article (the analysis in Ref. [32] was restricted to the stationary state). Our analysis is divided into two parts. Part 1. The main point of this paper is to eliminate the electrodes' degrees of freedom [33] and to analyze the transient dynamics within the resulting physical pic- ture of the coupled SQD-qubit dynamics. In this way, we can deal with the nontrivial interplay of the SQD- qubit coherence, strong local Coulomb interaction in the SQD, nonequilibrium conditions imposed by the attached electrodes, as well as both leading (SET) and next-to- leading order effects in the tunneling (cotunneling). The necessary inclusion of the latter furthermore forces us to extend Ref. [32] by including also the leading mem- ory effects on the sensor-qubit system due to the tunnel- ing to the electrodes, which is necessary for the study of the transient qubit dynamics. (For the stationary state, which we studied in Ref. [32], they can be be ignored without making additional approximations.) The impor- tance of memory effects for the dynamics when going beyond weak coupling is known especially since Ref. 78, see also Refs. [40, 41, 79–82] and progress for strong backaction [83, 84]. Non-Markovian corrections have also been studied in related contexts, such as the backaction of a quantum point contact on a double dot [81, 83] or quantum-feedback control based on quantum measure- ments [85, 86]. The various effects of non-Markovian processes remain, however, an uncharted territory [87]. The central result in our case are the kinetic equations (26) for the system of qubit plus quantum-dot sensor. The equations reveal the above three-fold nature of the backaction of the sensor QD on the qubit; importantly, the relevant energy scale for this backaction is not sim- ply the internal capacitive interaction λ (SET-induced stochastic backaction) but additionally involves the en- ergy scale Γλ/T (dissipative and coherent backaction in- volving transport processes). Our kinetic equation furthermore allows us to iden- tify slowly evolving quasistationary modes – containing the qubit evolution – and faster evolving decay modes re- flecting the dissipative SQD dynamics due to its coupling to the electrodes. The coupling between these modes generates the total backaction on the qubit and is medi- ated by all three types of backaction. To account for all these backaction effects, it is indispensable to keep the capacitive interaction (λ 6= 0) when integrating out the electrodes. In this aspect, our work differs from the oth- erwise closely related approach of Ref. [88], which starts out from the assumption that the electrodes affect exclu- sively the SQD. There, all backaction effects derive only from the internal interaction, i.e., the stochastic backac- 3 tion. A surprising finding of our analysis is that the total backaction exhibits a strong reduction when tuning the SQD towards the Coulomb blockade regime: we find that the coherent backaction actually cancels the cotunneling ("broadening") corrections in the coupling of the quasis- tationary to the decay modes. This eliminates the naively expected leading power-law dependence ∝ 1/(ε − µr) of the backaction, affecting also the decoherence time scales. This indicates that a sensor with quantized orbital states can be switched off more efficiently by controlling its gate voltage than naively expected, the first important experi- mental implication of this article. This requires, however, to prepare the sensor state in a controlled way to avoid a slip of the qubit state (see part 2 below). Instead, It is important to emphasize already here that the coherent backaction, which is responsible for this mit- igation, is not an independent mechanism that can be "added" to counteract cotunneling noise. it arises together with cotunneling as an integral part of quantum fluctuation effects of the qubit-sensor system when consistently describing all types of backaction. No- tably, we show that this mitigation is not captured by widely-used classical stochastic fluctuator models and can also be easily overlooked in Born-Markov approaches that integrate out the entire environment of the qubit (i.e., including the SQD). The effect of the exchange of electrons between the SQD and the electrodes can thus not be fully captured by classical switching of the SQD charge state. We also review and compare in detail our results with earlier works and pinpoint a number of lim- itations of standard approaches. The prominent role of renormalization effects under- lying the coherent backaction distinguishes a QD sensor with few, discrete energy levels from what can be ex- pected for a sensor with a continuous energy spectrum. The recent study [89] showed that similar torque terms appearing in a spintronic context are much suppressed in single-electron transistors (continuous spectrum) as op- posed to QDs (discrete spectrum). In the former case, renormalization effects tend to nullify when averaging over their continuous energy spectrum. This motivates the extensive analysis of the detection of a qubit state by a sensor QD undertaken in this paper. Our work raises the interesting question to which extent backaction ef- fects due to renormalization effects are suppressed in a single-electron transistor. Part 2. One might think that following the above de- scription one can in a second step eliminate the sensor QD from the description to obtain an effective theory for the qubit only. However, already on general grounds, this is questionable: specific to our indirect detection problem is that the environment of the qubit is not sta- tionary. Moreover, initial correlations between SQD and qubit – both microscopic systems – might exist. When integrating out the environment, the factorizability and the stationarity of the environment are, however, often invoked to eliminate the so-called slip of the initial condi- tion for the subsystem [90] coming from the initial state of the environment and its short-time transient evolution. To illustrate this point, we analyze in more detail the simpler high-temperature limit where the complications due to cotunneling and coherent backaction can be con- sistently ignored. Even in this high-temperature, weak- measurement limit the qubit develops a slip of order λ/Γ on a time scale t < 1/Γ, which is beyond the control over the qubit system alone. The slip depends explicitly on the initial qubit-sensor state. The second experimen- tal implication of our work is that the dynamical state of the sensor and its correlations with the qubit cannot be ignored and must be brought under experimental con- trol. This slip effect is cumulative, e.g., it results in phase shifts that still affect the qubit on much longer time scales t ∼ 1/λ ≫ 1/Γ relevant to the readout. By contrast, the relevant time scales T1 (relaxation) and T2 (dephasing) of the transient qubit dynamics for times t ≫ 1/Γ do not depend on the initial sensor QD state. The precession axis of the qubit Bloch vector also turns out to be independent of the initial state. In the simple high-temperature limit, we furthermore identify an addi- tional effect of the (purely stochastic) backaction, which is to induce a tilt of the Bloch vector precession axis. We find that the circular isospin precession becomes slightly elliptical in the presence of the detector, adding as a fingerprint oscillations to the exponential decay of the qubit-state purity. This mixes the notions of relaxation and dephasing as we will see. It is an interesting question how these effects behave at low temperature and strong qubit-sensor coupling where they have received little at- tention so far. Outline. After this topical outline, we now present the organization of the sections of the paper and the key equations. In Sec. II, we briefly review the generic indi- rect readout model of Ref. [32] and discuss the dynami- cal variables that are needed to describe the mixed quan- tum state of the joint qubit-sensor system. This requires two isospins, which capture both the reduced qubit state as well as the correlations with the sensor QD. After this, we outline the key technical challenges of our approach in Sec. III, deferring details to the App. A, and we present the time-local kinetic equation (26) for the coupled sensor QD plus qubit system. In App. A, we fur- ther discuss the importance of including non-Markovian corrections to retain the positivity of the reduced den- sity operator. Without further approximations, we iden- tify the relevant unperturbed modes (λ = 0) with the electrodes integrated out. From the representation of the kinetic equation (56) in these modes we prove the exact cancellation between the coherent backaction and the cotunneling ("broadening") noise [Eq. (62)]. We fur- thermore study the implications for the dependence of the total backaction on various experimentally relevant parameters (tunneling-rate asymmetries, bias, and gate voltages). From the formal solution of the effective qua- sistationary mode evolution [Eqs. (71) and (72)], with details given in Appendix B, we infer that the qubit evo- 4 lution is non-Markovian and exhibits a slip of the ini- tial condition that we characterize in Appendix C. Ini- tial slips generally go hand in hand with non-Markovian dynamics as Refs. [79, 90–94] and the references therein point out. Initial entanglement between a qubit and its environment, one cause of initial slips, can drastically af- fect the qubit coherence [94]. In Sec. IV, we attempt to integrate out the sensor QD to derive an effective Liouvillian [Eq. (80)] that effectively incorporates its fast switching dynamics. We then focus on the analytically tractable case of high temperature. This suffices to illustrate the general importance of initial slips in the context of detector backaction [Eq. (81)], the breakdown of orthogonality of relaxation and dephasing qubit modes, and the exponentially damped oscillatory but elliptical precession of the qubit Bloch vector. In this high-temperature limit, we obtain tangible expressions for the qubit relaxation and dephasing rates, expanded to leading order in λ/Γ. In Sec. V, we compare our results with semiclassical stochastic fluctuator models as well as Born-Markov and exact quantum approaches to provide further insight into the origin of the coherent backaction. In the accompanying Appendix D we show how the co- herent backaction affects the qubit phase evolution in a way that is not accounted for by semiclassical stochas- tic fluctuator approaches. We summarize our findings in Sec. VI. FIG. 1: Sensor quantum dot (SQD) tunnel-coupled to source and drain electrodes and capacitively coupled to a qubit, whose different logical states involve two possible positions, left and right, in a double quantum dot. If the qubit electron is left (a) or right (b), the Coulomb repulsion to the SQD electron is larger or smaller, respectively, compared to the full delocalization of the qubit electron. II. INDIRECT DETECTION A. Model We analyze the indirect detection setup sketched in Fig. 1: the readout (HI ) of a double-quantum dot charge qubit (HQ) by a proximal sensor quantum dot (HS), which in turn is read out (HT ) by the conductance of the transport current in one of the attached electrode reservoirs (HR). The model we use, discussed in detail in Ref. [32], thus consists of three "layers" with their respective interactions: H = HQ + (HI + HS) + (HT + HR). (1) This models the essential physics found in many exper- iments on QD qubits and can be extended to supercon- ducting qubits [45] as well as to spin qubits if measured by spin-to-charge conversion [9, 19]. Qubit. The qubit is realized as a charge qubit, a single electron occupying a double quantum dot. This electron can reside either on the left dot, denoted by the state Li, or on the right dot, denoted by the state Ri. The qubit state is represented by the ensemble average τ = hτi of an operator τ (corresponding to an isospin τ /2) with components τi = Xl,l′=L,R (σi)ll′lihl′. (2) Here, σi denotes the Pauli matrix for i = x, y, z. The av- erage z component τz quantifies the imbalance between the probabilities for finding the qubit electron in the left orbital rather than in the right orbital, while τx and τy quantify coherences between the left and right occupa- tion. The general form of the Hamiltonian of the isolated qubit is HQ = Ω · τ /2, (3) in which the qubit field Ω is used in applications to con- trol the qubit evolution: it induces coherent tunneling of the qubit electron between the two dots (Ωx, Ωy) com- bined with a detuning (Ωz). In our analysis, Ω is con- stant in time and later on, when we discuss tangible re- sults, we will chose Ω = Ωex. However, unless stated otherwise, we first keep Ω general. Sensor quantum dot. The sensor quantum dot (SQD or sensor QD) is modeled as a single, interacting, spin- degenerate orbital level with Hamiltonian HS = εn + U n↑n↓. Here, nσ = d†σdσ is the number operator for electrons with spin σ =↑,↓, where dσ denotes the cor- responding field operator, and n = n↑ + n↓ is the total electron number operator. We take the Coulomb repul- sion energy U here to be the largest energy scale (except for the bandwidth 2W of the electrodes), in accordance with typical experimental situations. We therefore ex- clude the double occupation of the sensor QD orbital in the following. This allows us to reduce the model to HS = ε P 1, P 1 = Pσ σihσ, (4) 5 if we accordingly adjust the high-energy cutoffs in the electrodes [discussed below Eq. (31)]. In the considered subspace, we can replace n = P 1. For the readout one tunes the level position ε = −Vg by a gate voltage Vg close to the electrochemical potentials of one of the elec- trodes. While the spin of the qubit electron is irrelevant (it was not written above since the readout couples to the charge, see below), it is important to include the spin de- gree of freedom of the SQD because the spin degeneracy enters into the tunneling rates. Electrodes. The final stage of the readout involves the electrodes, treated as noninteracting reservoirs of elec- trons with spin: (5) HR = Pr,k,σ ωrkσc†rkσcrkσ, with the field operators crkσ (c†rkσ) acting on the elec- trons in orbital k with spin σ in the source (r = s) and the drain (r = d), respectively. These are each held in equilibrium with a common temperature T , but at differ- ent electrochemical potentials µs = Vb/2 and µd = −Vb/2 by a applying a bias voltage Vb. Readout. The indirect readout of the qubit state using the sensor QD involves two couplings: the first one is the capacitive interaction of the SQD electron charge n with the charge polarization τz of the qubit: HI = nλ · τ /2. (6) The measurement vector λ specifies both the basis in which one measures and the measurement strength λ: λ = λez. (7) Thus, depending on the qubit state, the sensor QD level experiences an energy/gate-voltage offset of at most ±λ/2 [see Figs. 1 (a) and 1(b)]. This in turn affects the conductance measured in one of the electrodes due to tunneling to and from the SQD: (8) HT = Pr,k,σ trd†σcrkσ + H.c. The strength of this second coupling involved in the read- out is quantified by the tunnel rates Γr = 2πtr2νr, where we take both the tunneling amplitude tr and the density of states νr to be spin (σ)- and energy (k)- independent within the electrode bandwidth 2W . The threefold layered structure of this indirect detec- tion (negligible direct coupling of the electrodes to the qubit) is reflected in our theoretical analysis of the mea- surement backaction. In Sec. III C, we first eliminate the "outer" detection layer – the electrodes – in favor of effec- tive equations describing the joint SQD-qubit dynamics [99]. In Sec. IV, we then attempt to integrate out the "inner" detection layer – the SQD – to find the effective qubit evolution. B. Weak-measurement and weak-tunneling limit We consider a sensor with a fast response, i.e., the in- ternal dynamics of qubit plus sensor QD is slow as com- pared to the electron tunneling dynamics induced by the attached electrodes: ∆ ∼ λ, Ω ≪ Γ. (9) This condition means physically that many electrons pass through the SQD during its interaction time with the qubit (λ ≪ Γ) - a weak measurement is performed. Moreover, if Ω lies in the x-y plane, the internal qubit evolution describes a coherent tunneling of the qubit elec- tron with dwell times of electrons in the SQD that are much smaller than the period of a qubit cycle (Ω ≪ Γ). Each electron sees a "snapshot" of the SQD-plus-qubit state. If we assumed Γ . Ω instead (but still weak mea- surement λ ≪ Γ), the readout would be too slow to re- solve any qubit evolution. As we see below in Sec. III, the leading-order response of the tunnel rates to a measurement-induced gate volt- age offset ∼ λ is given by Γλ/T . This, in return, induces a dissipative backaction affecting the polarization of the qubit that also scales as Γλ/T . Moreover, when condi- tion (9) holds, the tunneling also affects the isospin co- herences of the qubit-SQD state. This is well-known from the analysis of two-level systems coupled to a reservoir: Here, the density-operator coherences in the energy basis matter when the levels (here split by ∆) are degenerate on the scale of the coupling (here Γ) to the environment. In our case, the two levels correspond to the sensor QD and the qubit, each being a two-state system. One has to carefully identify which coherences are relevant, which is done below in Sec. II C. These coherences are affected by tunneling processes: The simple physical intuition be- hind this is that if an electron on the sensor has time to interact with the qubit and change the current, it cer- tainly has time to fluctuate into the electrodes. This leads to a response of the level renormalization and re- sults in a coherent backaction which scales as Γλ/T , i.e., in the same way as the response of the sensor tunneling rates resulting in the dissipative backaction. This is a central result of Ref. [32] and here we explore its effect on transient dynamics. It should thus be noted that the energy scale for back- action on the qubit is not simply λ (from the internal interaction HI ) but also λΓ/T , the scale of effective dis- sipative and coherent coupling between sensor QD and qubit, which are induced by tunneling processes. In a way, these couplings account for an indirect interaction of the qubit with the electrodes extending the approach of Ref. [88]. This is thus another relevant perturbative scale for a weak-measurement expansion, besides the scale λ itself, see Ref. [32] for a detailed exposition. Another crucial point for this work is that if terms of order Γλ/T are taken into account and λ ≪ Γ, then at least cotunneling terms scaling as Γ2/T must also be accounted for (if not even higher-order tunneling terms). We can neglect higher-order tunneling processes beyond cotunneling if we restrict the temperatures such that Γ2/T ≪ λ. (10) 6 This condition means that the cotunneling-induced noise imposes only a weak perturbation of the qubit. Taken together, we employ here a weak-coupling limit in two ways, namely that of weak measurement and weak tun- neling: Γ/T ≪ ∆/Γ ≪ 1 . (11) Note that by Eq. (10) this imposes a stronger condi- tion than the usual weak-tunneling assumption Γ/T ≪ 1 alone. We next discuss the dynamical variables needed to describe the measurement backaction. C. Charge-specific isospins and qubit decoherence To describe both the backaction of the sensor on the qubit as well as to compute the signal current through the sensor QD, one needs at least the reduced density opera- tor ρ(t) of the combined qubit plus SQD system obtained by tracing over the electrodes. Even though we do not analyze the sensor signal here, the signal is of course of high interest to gain insight into, e.g., the efficiency of the measurement [12–15]. Studying the backaction in a situation where the sensor signal current is not negligi- ble, is an experimentally highly relevant situation, which we pursue in this paper. The relevant part of the SQD-qubit density operator ρ(t) can be expanded as follows [32]: ρ(t) = 1 2Xn P n ⊗"pn(t)1 +Xi τ n i (t)τi# . (12) Here, P n denotes the projector onto the charge states n = 0, 1 of the sensor QD. The numbers pn(t) = tr( P nρ(t)) give the probability for the sensor QD to be in the respec- tive charge state n = 0, 1, which for any time t sum up to one due to the probability conservation: p0 + p1 = 1. The only irrelevant coherences (off-diagonal matrix ele- ments in the energy basis) of ρ are those involving differ- ent charge states on the sensor. These can be shown to decouple from the relevant part due to the charge conservation by the tunneling. However, all remaining qubit-SQD density matrix elements including their co- herences must be kept in (12). These are the six num- i (t) = tr( P n τiρ(t)), which are the averages of the bers τ n isospin components i = x, y, and z for the two sensor QD charge states n = 0 or 1, respectively. To describe the correlated SQD-qubit system, one thus needs two charge- specific isospins τ 0 and τ 1. Based on Eq. (12), it is con- venient to introduce the following column representation of the density operator: ρ =   p0 p1 τ 0 τ 1 .   (13) In this form, one distinguishes the charge and isospin part, respectively, but the isospin part is still kept ba- In other words, we may represent τ 0 sis independent. and τ 1 in a different orbital basis than the one used in definition (2) if we transform the directions of λ and Ω accordingly. We now further explore the physical meaning and im- portance of the two charge-specific isospins. By construc- tion, they sum up to the total isospin, τ = τ 0 + τ 1, (14) which is often of main interest. This is the usual Bloch vector that describes the state of the qubit, i.e., its re- duced density operator. One can easily show that a single Bloch vector can describe the joint SQD-qubit state only if it is factorizable. In other words, if ρ = ρS ⊗ ρQ = Xn pn P n! ⊗ 1 2 1 +Xi τi(t)τi! , (15) then the charge-specific isospins are given by τ 0 = p0τ and τ 1 = p1τ , respectively, by comparing Eq. (15) with Eq. (12). The other combination of the two isospins, δ := p0τ 1 − p1τ 0, (16) quantifies the nonfactorizability of the qubit-sensor den- sity operator: ρ = ρS ⊗ ρQ ⇔ δ = 0. (17) We emphasize that the nonfactorizability is crucial to describe the readout and its backaction on the qubit state τ . We will see in Sec. III C 1 below Eq. (54) that the deviation δ has no impact on the qubit evolution only if the qubit and sensor are strictly decoupled (λ = 0). For the coupled case (λ 6= 0), which is of interest, we have to keep the individual dynamics of τ 0 and τ 1. The relevance of the two isospins for the decoherence can be seen explicitly from the equation of motion for τ (t)2, which characterizes the purity of the isospin state: d dt [τ2] = 2 τ · τ = −2λ · (τ 0 × τ 1), (18) where we inserted after the first equality τ = Ω × τ + λ × τ 1. (19) Equation (18) is an exact result which can be obtained from the Heisenberg equation of motion for τ with re- spect to the full Hamiltonian (1). The only essential as- sumption on which Eq. (18) relies is the indirect readout structure of our setup, i.e., [HT , HQ] = 0. It thus holds generally for any Ω and any tunneling Γ. Preservation of this exact equation imposes an exact isospin sum rule [discussed below Eq. (26)], which any kinetic equation for ρ should satisfy: This was only realized recently, see 7 Refs. [46] and [32], in particular, see Appendices D and E. Equation (18) shows that the reduction of the purity of the qubit state appears only due to noncollinearities of τ 0 and τ 1. These noncollinearities develop because of the readout: the isospins τ 0 and τ 1 are subject to different effective "magnetic" fields depending on the charge state n of the sensor QD. To see this, we rewrite HQ + HI = Beff(n) · τ /2 with an effective field acting on the isospin τ , Beff = Ω + λδn, (20) where δn = n − hni and hni = tr( P 1ρ) = p1. Here, the first part is the mean field , Ω = Ω + hniλ, (21) which the isospin experiences due to the internal isospin field Ω and the average field caused by the mean charge hni = p1 on the sensor QD with respect to the exact total density operator ρ. The mean-field contribution Ω to Beff is the same for both charge-specific isospins, τ 0 and τ 1, and therefore not responsible for the qubit- state decay. The mean SQD occupation hni = p1 merely tilts the qubit precession axis and changes its frequency contributing to the detuning of the qubit. Note that the average is here an ensemble average but not a time average since p1 = p1(t) can change in time with the state ρ(t). The qubit decay is induced by the second, fluctuating contribution λδn to Eq. (20), where δn = n − hni is the charge-state dependent deviation from the mean field. This generates a noncollinearity of τ 0 and τ 1, which reduces the purity of the qubit state by Eq. (18). Even though our approach does not make use of the decomposition into a mean-field and a fluctuating part, we can identify both effects in our results in Sec. III E. We will first identify p1 = p1 st with the state of the SQD in the stationary limit, in which the ensemble and the time average are equal. We further connect more precisely the decoherence rates to the components of the fluctuating part λδn along and perpendicular to the mean field Ω in Sec. IV A (in accordance with the literature [26, 27]). This accounts for what we call stochastic backaction on the qubit by the sensor QD. This effect is also present for single-electron transistor sensors with a continuum of electronic levels, but (classically) quantized charge states. However, there are also a dissipative and a coherent backaction effect [32] (see Sec. II B). As we discuss below, they modify the relative orientations of τ 0 and τ 1 and therefore affect the qubit decay as well. This mechanism - first noted in Ref. [32] - derives from a renormaliza- tion effect induced by the interplay of the readout inter- actions (λ) and the tunneling on and off the sensor QD (Γ) as discussed in Sec. III. It results in isospin torques similar to those encountered in spintronic QD setups. As mentioned in the Introduction (Sec. I), the prominent role of renormalization effects distinguishes a QD sen- sor with few, discrete energy levels from sensors with a continuous energy spectrum. Finally, we note from Eq. (18) that the Bloch vector may not just shrink exponentially. We will find that τ 0 and τ 1 perform different precessional motions (due to both Beff and the coherent backaction being dependent on the charge state of the sensor), which implies that the component of τ 0 × τ 1 along the measurement vec- tor λ also oscillates in time. Thus, the rate of decay of the purity is not purely exponential but additionally oscillates in time as explicitly confirmed by our analy- sis in Sec. IV D. This illustrates that the motion of the charge-specific isospins is closely related to the qubit de- cay. Accounting for the interplay of their dynamics turns out to be the key to set up a correct description of the transient qubit dynamics that includes all the different types of backaction. III. QUBIT-SENSOR QUANTUM DOT DYNAMICS A. Outline The indirect measurement setup introduced in the pre- vious section poses several challenges for the theoreti- cal treatment of the measurement backaction. A central complication is that the environment of the qubit (the SQD plus the electrodes) is not in a simple equilibrium state since the detection is done by nonequilibrium trans- port. But even when specializing to near-equilibrium conditions, one has to treat the SQD as a strongly in- teracting quantum system with spin degeneracy. Both the nonequilibrium conditions and the interactions in the SQD prevent a simple direct approach where one aver- ages over the environmental degrees of freedom, leaving only the qubit degrees of freedom. Moreover, to obtain the sensor current, we need to retain the sensor degrees of freedom as well. As we discuss in Sec. V B, specifying the environmental state is the main difficulty when try- ing to directly calculate the evolution of the qubit density operator for an indirect detection setup. Therefore, we integrate out only the electrodes to ob- tain the density operator ρ for the joint qubit-SQD sys- tem. The resulting equation, which is of the form d dt ρ(t) = −iLρ(t), (22) is given below [Eq. (26)] and is the first main equation of this work. Our main conclusion is that this provides a systematic description of the measurement backaction: in the weak tunneling, weak measurement limit Γ/T ≪ λ/Γ ≪ 1, it does not get any simpler without making drastic concessions. Yet, a mere reformulation of Eq. (26) [see Eq. (56)] already provides important insights into the measurement backaction. Still, we will also describe an attempt to eliminate the SQD degrees of freedom in the high-temperature limit where corrections Γ/T can be dropped. This results in an effective Liouvillian Leff that reproduces the outcome 8 of Eq. (22) for the evolution of total average isospin op- erator, τ (t) = hτi = τ 0(t) + τ 1(t), in the long-time limit, d dt τ (t) = −iLeffτ (t) (t − t0 ≫ 1/Γ), (23) with initial time t0. The preparations for this step pro- vide interesting insights into Eq. (22). However, Eq. (23) turns out to be invalid for small times t − t0 . 1/Γ. The error made when still using Eq. (23) to compute τ (t) starting from t = t0 for t− t0 ≫ 1/Γ can be compensated by a correction to the initial condition τ (t0), a so-called initial slip. This correction depends on the initial qubit- sensor state in an essential way, preventing the sensor from being integrated out completely. Before we discuss the details, let us first outline the further challenges posed in deriving the above two equations. The derivation of Eq. (22) has to include various ef- fects: First, since we incorporate the measurement back- action terms ∼ Γλ/T , we must also include next-to- leading order tunnel processes ∼ Γ2/T into the kinetic equations for the SQD-qubit evolution (see Sec. II B). We have given such a consistently expanded kinetic equation in Ref. [32]. However, there we employed an additional Markov approximation with respect to the electrodes, which is valid to obtain the stationary long-time limit studied in Ref. [32]. Here, by contrast, we are interested in the transient dynamics, where non-Markovian effects induced by the electrodes must be accounted for, as we explain in Sec. III B 1. We include the required leading non-Markovian correction perturbatively in the tunnel coupling Γ along the lines of Refs. [40, 41, 47, 78–80]. We present and explain the resulting time-local kinetic equations in Secs. III B 2–III B 4. We next analyze in Sec. III C how the qubit is affected by the measurement within the resulting description. For this purpose, we first solve in Sec. III C 1 the kinetic equa- tions for zero capacitive interaction λ = 0. An important step is to identify a set of quasistationary modes that contain the degrees of freedom of the qubit only, i.e., τ = τ 0 + τ 1. This identification remains valid also for nonzero capacitive interaction λ 6= 0. The time scale ∼ 1/Ω for the evolution of these modes – connected with the slow qubit dynamics – is well separated from that for the evolution of the residual decaying modes. Those are strongly damped on a short time ∼ 1/Γ due to the fast tunneling dynamics of the SQD. We then introduce new dynamical variables to analyze the coupling between the quasistationary and decay modes in Sec. III C 2 for nonzero capacitive interaction λ. This will reveal the mitigation of the measurement backaction by the coher- ent backaction, the first key result of the paper. Finally, we derive the evolution of the quasistationary modes by effectively incorporating the impact of the decay modes (see Sec. III E). Importantly, the resulting equations are not independent of the detector evolution and even in the long-time limit an explicit dependence on the initial overlap with the decay modes remains as we will see in Sec. IV B. Kinetic equation 1. Integrating out the electrodes Whenever the time evolution of an open system is con- sidered, non-Markovian features arise from the memory of the environment, that is, its correlation functions de- cay within a nonzero correlation time τC [48, 49]. When integrating out the environment (here the electrodes, see Fig. 1), the time evolution of the reduced density opera- tor ρ(t) of the open system (here the SQD plus qubit) is governed by a time-nonlocal kinetic equation: ρ(t) = −iLQSρ(t) +Z t t0 dt′W (t − t′)ρ(t′). (24) Here, LQS• = [HQ + HS + HI ,•] is the internal Liouvil- lian of the reduced system with "•" denoting the operator the Liouvillian acts on. Moreover, all effects of the envi- ronment are contained in a kernel W that we compute by a real-time diagrammatic approach [50, 51]. If the initial value ρ(t0) is specified, Eq. (24) can be used to compute ρ(t) without explicitly keeping track of the state of the electrodes. A key assumption enabling such a closed de- scription of ρ(t) is that the reservoir is stationary, i.e., [HR, ρR] = 0, which is satisfied here because we assume the electrodes to be a thermal equilibrium state ρR (see, e.g., Ref. [50]). When calculated to leading order in Γ, the kernel roughly decays as W (t − t′) ∼ Γe−(t−t′)/τC with cor- relation time τC ∼ 1/T [49]. Within the Markov ap- proximation with respect to the electrodes, one replaces W (t − t′) = W (i0)δ(t − t′), where on the right-hand side W (z) = R ∞ 0 eiztW (t) denotes the Laplace-transformed kernel. This yields a time-local kinetic equation dρ/dt ≈ In (−iLQS + W (i0))ρ(t) when inserted into Eq. (24). general, (−iLQS + W (i0))ρst = 0 determines the ex- act stationary state, a fact which is often overlooked but easily shown [50]. Physically, this makes sense since 9 a nearly constant state cannot "remember" much. As ρ(t) approaches the constant stationary state ρst, non- Markovian corrections in Eq. (25) become weaker [for fixed t′ the memory kernel W (t−t′) decays as t increases]. To go beyond this Markovian approximation to obtain the transient dynamics, we include the non-Markovian corrections induced by the electrodes perturbatively in Γ/T . To do so, we insert the Taylor expansion for the reduced density operator, ρ(t′) = ρ(t) + ρ(t)(t′ − t) + . . . , (25) recursively into Eq. (24), as explained in Refs. [40, 41]. As we argue in Appendix A 1, the derivatives dnρ/dtn are C ∼ 1/T n on the order of Γn, and we estimate (t′−t)n . τ n within the correlation time of the kernel. Thus, higher- order terms in the expansion (25) correspond to higher orders in the tunneling expansion in Γ/T . Truncating the expansion after the leading-order memory correction (n = 1), one can derive a time-local kinetic equation for ρ(t) as we show in Appendix A 1. The above treatment is closely related to the tech- niques developed for full counting statistics [78–80] and to the recent study in Ref. [47]. There is also a concep- tual connection to time-convolutionless master equations [52–54]: In the latter approach, the full density opera- tor evaluated at time t′ is obtained by evolving the full density operator at time t backwards in time before inte- grating out the electrodes, resulting also in an effectively time-local kinetic equation. 2. Kinetic equation Including all terms of order ∆, Γ, as well as Γ2/T and ∆Γ/T , where ∆ ∼ Ω, λ, as well as the leading memory corrections we obtain the kinetic equation expressed here in the representation (13) of ρ (no time arguments writ- ten): p0 p1 τ 0 τ 1 d dt    =   −2γ0 +γ1 +2γ0 −γ1 +2cλ +cλ −2γ0 + (Ω − κλ)× −2cλ −cλ +2γ0 + κλ× +2cλ· −2cλ· +cλ· −cλ· +γ1 + κλ/2× −γ1 + (Ω + λ − κλ/2)× p0 p1 τ 0 τ 1       (26) When computing the matrix product with the col- umn vector in the above equation, the dot "·" (cross "×") in the entries of the matrix indicates that a three- dimensional scalar (vector) product is to be formed with the corresponding entries of ρ. The above equation is valid under the weak-coupling assumption Γ/T ≪ λ/Γ ≪ 1 introduced in Sec. II B such that corrections of order Γ3/T 2, Γ2∆/T 2, and Γ∆2/T 2 can be neglected. The above kinetic equation is the first central equation of this paper. It goes beyond a simple master equation by including all relevant coherences (see Sec. II C) and extends the kinetic equation of Ref. [32], which is Marko- vian with respect to the electrodes, to access the transient dynamics by including the kernel frequency dependence. The kinetic equation (26) respects the probability conser- vation, p0 + p1 = 0, and also the recently found [32] exact 10 must be set to W ∼ U , where U is the large but finite lo- cal Coulomb interaction energy of the SQD (we excluded the doubly occupied state from the SQD Hilbert space). The term in the second line of Eq. (27) relates to the cotunneling processes through the SQD, which incorpo- rates the derivative of the renormalization function, φ′r = ∂φ(x) ∂x (cid:12)(cid:12)(cid:12)(cid:12)x=(ε−µr)/T , (32) which is also plotted in Fig. 2. The contribution from each electrode r changes its sign close to the resonance ε = µr and takes its extremal values of φ′r ≈ ∓0.143 at ε − µr ≈ ±1.911T . While the terms in the first line of Eq. (27) depend exponentially on the distance to the res- onance ε−µr, the cotunneling term is only algebraically suppressed, since [100] φ′r ≈ − q +f−q = f + (27) 1 π T ε − µr for ε − µr ≫ T . (33) and 2f + q +1 > 1 in Eq. (27). When these terms are added together, they result (for each electrode r) in a temperature-broadened step function, which approaches its asymptotes algebraically. Therefore, this must be ac- counted for when studying the qubit-sensor dynamics at the onset of Coulomb blockade where typically the read- out is performed. τ 0 + τ 1 = Ω × τ + λ × τ 1. The isospin sum rule (19), latter derives from the conservation of the total isospin, τ = τ 0 + τ 1, when electrons tunnel from the electrodes into the SQD and vice versa, a generic feature [46] of in- direct measurement models of type (1). We next discuss the expressions and physical significance of the four new coefficients γ0, γ1, c, and κ occurring in Eq. (26); for the definition of λ and Ω see Eqs. (7) and (3), respectively. 3. Stochastic, dissipative, and coherent backaction First, Eq. (26) incorporates the SQD switching rates r with contributions from each junction γ0,1 = Pr γ0,1 r = s, d reading ΓrΓq 2T (f±r )′φq  ηΓrf±r + Xq=s,d ΓrΓq 2T φ′r(2f + γ0,1 r = Xr=s,d ∓ Xq=s,d q + f−q )  , Let us first focus on where 0, 1 corresponds to ±. the meaning of the three different physical terms in Eq. (27). The first term in the first line of Eq. (27) is the sequential tunneling contribution, whose depen- dence on the voltages is governed by the Fermi func- tions f±r = f±((ε − µr)/T ) for electrode r = s, d with f +(x) = 1/(ex − 1) and f−(x) = 1 − f +(x). We com- ment on the non-Markovian correction factor η [Eq. (36)] in Sec. III B 4. The second term is a correction to the se- quential tunneling rate accounting for a renormalization of the level position ε, incorporating the derivative of the Fermi function, (f±r )′ = ∂f±(x) ∂x (cid:12)(cid:12)(cid:12)(cid:12)x=(ε−µr)/T and the renormalization function, φr = φ((ε − µr)/T ), with , (28) (29) (30) φ(x) = PZ +Λ −Λ dy π f +(y) x − y π (cid:20)− Re ψ(cid:18) 1 2 1 x = + i 2π(cid:19) + ln(cid:18) Λ 2π(cid:19)(cid:21) . (31) Indeed, combining this second term with the first, f±r (ε) + (f±r )′(ε)(δ/T ) ≈ f±r (ε + δ), one identifies the shift δ = Pq Γqφq/2. The function φ(x) is plotted in Fig. 2 and shows a maximum at x = 0 with logarithmic tails. In Eq. (30), P denotes the principal value of the integral with a cutoff Λ = W /T , yielding the real part of the digamma function ψ with a logarithmic correction. The latter depends on the electrode bandwidth W , which FIG. 2: Renormalization function φ(x), Eq. (30), and its derivative φ′(x) = dφ(x)/dx. For illustration purposes we chose Λ = 15 to be rather small, noting that Λ only shifts the φ vertically and drops out in φ′ [see Eq. (31)]. All the above-mentioned tunneling processes contribute to a stochastic switching of the SQD occupation n, which results via the capacitive interaction HI = nλ · τ /2 in a fluctuation of the effective field Bn eff = Ω + nλ acting on the qubit as explained in Sec. II C [see Eq. (20)]. The importance of the capacitive interaction to produce this stochastic contribution to the total measurement backaction becomes apparent when rewriting the kinetic equation (26) in terms of quasistationary and decaying dynamical variables (see Sec. III C 2). It causes the decoherence of the qubit already in lowest order as we show in Sec. V. Before we enter the detailed analysis of the next-to- leading order corrections, let us right away indicate their importance for the stochastic backaction on a more qualitative level. In Fig. 3, we compare the evolution of the x-component τx(t) of the isospin, obtained by solving the kinetic equations (26), when higher-order terms are included (red) or neglected by hand (green). The figure illustrates that noise from O(Γ2) terms indeed contributes to the qubit decoherence as naively expected. On a quantitative level, however, one would expect an algebraic suppression of the measurement backaction with ε based on Eq. (27) when entering the Coulomb blockade regime. It will turn out in Sec. III C 3 that this expectation is incorrect, i.e., the backaction is weaker than expected. We further see from Fig. 3 that the oscillation period of the qubit is notably changed due to next-to-leading order corrections. This is due to the mean field, Ω = Ω +hniλ, acting on the qubit in the presence of the sensor QD. The average occupation hni = p1 st on the sensor is significantly modified by higher-order tunneling terms (see Sec. IV B 1). In addition to the stochastic backaction, there is also a dissipative backaction of the SQD on the qubit: These terms are related to the isospin-charge conversion rates ∼ cλ with coefficient Γr 2T (−f + r )′. c = Xr (34) This coupling appears in two ways. The isospins influ- ence the SQD dynamics (allowing for the readout) and vice versa the SQD occupation probabilities directly in- fluence the isospins (backaction). This dissipative back- action drives the sensor QD and qubit into a correlated state in the stationary limit, in contrast to the stochastic backaction, which only changes the occupation probabil- ities of the qubit state. For the parameters chosen for Fig. 3, the dissipative backaction has a negligible impact on the qubit evolution and therefore we show no compar- ison. The reason for this suppression is that the SQD is already mildly Coulomb blockaded for these parameters and the dissipative backaction is exponentially peaked around the resonance as Eq. (34) shows. The dissipative backaction therefore only becomes relevant close to reso- nance. Finally, there is a third type of backaction: the tunneling gives rise to isospin torque terms ∼ κλ, where Γr T φ′r, κ = Xr (35) 11 x (t) + τ 1 FIG. 3: Modification of the sensor backaction due to the next-to-leading order Γ2 stochastic backaction and the co- herent backaction. Shown is the x-component of the isospin, τx(t) = τ 0 x (t) as a function of time. The isospin is obtained by solving the kinetic equations (26) when all terms are taken into account (blue), when higher-order Γ2 correc- tions are neglected (green), or when the coherent backaction ∝ κ is neglected (red). One can see that the isospin precession period is larger if next-to-leading-order contributions are ne- glected since the isospin experiences a different mean field as a consequence of the differing average occupation hni. The time is given for all three curves in the same unit, the inverse of the natural frequency for the full result, Ω = pΩ2 + (hniλ)2, which is the precession frequency including next-to-leading st = 2γ0/γ from the sta- corrections. We insert here hni = p1 tionary solution of the full kinetic equation (26) at λ = 0 [see Eq. (49)]. The parameters are Γs = Γd = 0.2T , λ = 0.02T , Ω = 0.001T , Vb = 0, Vg = −3T , W = 1000T . Initially, the sensor is empty, p0(0) = 1 − p1(0) = 1, and conditional upon this, the qubit isospin vector is prepared perpendicular to the measurement vector, τ 0(0) = ex and τ 1(0) = 0. incorporates the derivative of the renormalization func- tion shown in Fig. 2. This signals the coherent nature of this contribution to the backaction: it characterizes the response of the sensor QD level renormalization to a change in the state of the qubit. Similar to the cotunnel- ing corrections, the coherent backaction gains importance with the onset of Coulomb blockade [32]. The importance of coherent backaction for the qubit evo- lution stands out in Fig. 3. We can see that neglecting the coherent backaction (red) noticably (but artificially) enhances the qubit decoherence as compared to the full solution (blue). This points to a cancellation effect be- tween coherent backaction and cotunneling noise that we discuss in detail in Sec. III C 3. We further note that the coherent backaction has a negligible effect on the qubit oscillation period [the period is the same for both the curves including O(Γ2) corrections]. The coherent back- action can thus not be interpreted as a simple correction to the qubit mean field (21); it is the joint system of qubit and sensor QD that is renormalized and not just the qubit system. We finally emphasize that even though Fig. 3 shows theoretical results when different contribu- tions of the kinetic equation are neglected, they cannot be switched off individually in a real experiment - there they always appear together and have to be taken into account altogether. 4. Impact of non-Markovian corrections It remains to discuss the three ways in which non- Markovian corrections induced by the electrodes are con- tained in Eq. (26) and how the latter differs from the [32]). Markovian kinetic equations (Eq. First, the non-Markovian corrections modify the leading- order SQD tunneling rates [see Eq. (27)] just by introduc- ing the prefactor (18) of Ref. η = 1 +Xr Γr T φ′r. (36) Since the correction η − 1, the cotunneling broadening term [see Eq. (27)], and the coherent backaction coef- ficient κ [see Eq. (35)], all depend on the same factor Pr Γr/T φ′r with algebraic tails, the non-Markovian ef- fects should clearly be accounted for [101]. The correc- tion factor η − 1 is an appreciable quantitative correc- tion that yields a contribution of O(Γ2/T ) to the switch- ing rates [40]. However, in contrast to the cotunneling and coherent backaction, the correction η − 1 is mul- tiplied with the exponentially scaling SET contribution [see Eq. (27)] and therefore it has no qualitative impact here. By contrast, the second type of non-Markovian cor- rection affects the coherent backaction terms by qualita- tively changing them in general relative to the Markov approximation. The direction of the tunneling-induced isospin torque terms changes: In Ref. [32], we also found a contribution to the coherent backaction ∝ κΩ. These terms are canceled out here up to O(ΓΩ/T ) = O(Γλ/T ). This is expected on physical grounds as the backaction is mediated by the capacitive interaction λ and there- fore we expect these terms to vanish when setting λ = 0. We emphasize that the isospin torque terms ∝ κΩ × τ n do not affect the stationary state, which we studied in Ref. [32], and therefore all the conclusions drawn in Ref. [32] remain valid. The third effect of the non-Markovian corrections is a sign change of the isospin torque terms ∼ κλ in the last column of the matrix (26) as compared to Ref. [32]. Both the above modifications of the coherent backac- tion have important physical consequences illustrated in Appendix A 2: If one naively computes the transient dy- namics of the SQD-qubit system using the equations of Ref. [32], which neglect non-Markovian terms induced by the electrodes, one obtains exponentially increasing tran- sient modes leading to a violation of the positivity of the density operator. Moreover, within the Markovian ap- proximation the coherent backaction strongly enhances the measurement backaction in the Coulomb-blockade 12 regime for a large parameter regime, while the coher- ent backaction suppresses the measurement backaction for nearly all parameter values when non-Markovian cor- rections are correctly accounted for (see Appendix A 2). This clearly illustrates that non-Markovian corrections go hand in hand with renormalization effects, which in an indirect measurement set up go hand in hand with the cotunneling effects of the sensor rates. All of these are of vital importance for describing the indirect measure- ment. C. Coupling of modes With the kinetic equations (26) now in hand we can proceed to analyze the measurement backaction, but still without integrating out the sensor. To achieve this goal, we make use of the separation of different time scales in the coupled evolution of SQD and qubit in the weak- coupling, weak-measurement limit Γ/T ≪ ∆/Γ ≪ 1. To identify these time scales, we first solve in Sec. III C 1 the unperturbed problem of the decoupled SQD-qubit system (λ = 0) as described by Eq. (26). This produces eigen- modes which are well-separated in energy by Γ ≫ Ω and correspond to the wide-band limit for the sensor quan- tum dot. It turns out that one needs to compute the evolution of only a part of the modes – referred to as the quasistationary modes in the following – to construct the evolution of the total isospin τ . In Sec. III C 2, we restore the coupling λ and by sim- ply writing the kinetic equation (26) in the basis of these eigenmodes, we can immediately extract λ2/Γ as the rele- vant time scale for the qubit decoherence time. However, this is not the full story of the backaction: there is a prefactor which strongly affects this time scale. We ana- lytically identify a nontrivial competition of the coherent backaction and the cotunneling-induced stochastic back- action determining this prefactor. Finally, introducing an exact [relative to Eq. (26)] projection of the dynam- ics onto the quasistationary modes in Sec. III E we gain further insight, still without integrating out the sensor. This projection incorporates the effect of the coupling between the modes and provides the starting point for deriving an effective equation for the isospin evolution in Sec. IV. 1. Quasistationary and decay modes for λ = 0 We first solve the kinetic equation (26) for λ = 0 in which case the dynamics of the occupation probabilities p0 and p1 decouples from the dynamics of the isospins τ 0 and τ 1 as shown by the "unperturbed" time-evolution 13 we find − iL0 = Xp hiΩpVq p Vq†p + (iΩp − γ)Vd p Vd†p i , (41) with p = c, 0,±, and eigenfrequencies Ωc = Ω0 = 0, Ω± = ± Ω, (42) and Vk†p and Vk p are the left and right eigenvectors, re- spectively, using dyadic notation. The indices p = c, 0,± and k = q, d label in total eight different modes. Be- fore we discuss the physical meaning of these modes, we first note the following general property: since L0 is di- agonalizable (although L†0 6= L0), all the left and right eigenvectors are mutually biorthonormal, Vk†p · Vk′ p′ = δkk′ δpp′ , and they satisfy the completeness relation Xk=q,d Xp=c,0,± Vk p Vk†p · = 1. (43) (44) This also follows explicitly by using Eq. (40) and the expressions below. This can be exploited to expand the state vector ρ(t), defined by Eq. (13), as follows: (p0 + p1)Vq +(p1 stp0 − p0 stp1)Vd ρ = = (45) α α(cid:1) Vd (46) stτ 0 α + τ 1 α α(cid:1) Vq α − p0 α(0)Vq α stτ 1 eiΩαtvq c +Xα (cid:0)τ 0 c +Xα (cid:0)p1 c +Xα c +Xα Vq + e−γtvd c (0)Vd e(iΩα−γ)tvd α(0)Vd α, α(0) = Vk†α · ρ(0), vk (47) taking the initial time t0 = 0 here. Importantly, equality (45) is generally valid for any state ρ(t), while the second equality (46) holds only if ρ(t) = e−iL0tρ(0). We now discuss the explicit form of the modes. The most fundamental one is the stationary charge mode with the conjugated left eigenvector and the right eigen- vector Vq c =   1 1 0 0   c =   p0 st p1 st 0 0   , Vq , (48) expressed in the occupation probabilities of the SQD in the stationary limit and for zero coupling λ = 0, p0 st = γ1 γ , p1 st = 2γ0 γ , (49) stτ 0 − p0 FIG. 4: Sketch of the complex eigenvalues and the associated dynamics of the joint qubit plus sensor QD system without readout (λ = 0). The four upper sketches depict the evolution of the total isospin τ = τ 0 +τ 1 described by the coefficients of the four quasistationary modes in the expansion of ρ [see Eq. (45)]. Moreover, the lower four sketches depict the evolution of the weighted difference δ = p1 stτ 1 of the charge- specific isospins associated with the coefficients of the decay- ing isospin modes [see Eq. (45)]. The indices α = +1, 0, −1 label the three different polarizations for the precessional mo- tion. The unit vector e0 = Ω is the "bare" qubit field and the vectors e1 and e2 can be chosen arbitrarily in the trans- verse plane. The two sketches on the right show the different dynamics of the occupation probabilities p0 and p1 for the stationary and the decaying charge mode, respectively. Note that p0 and p1 are not the coefficients in Eq. (45) but only their combinations p0 + p1 = 1 and δst = p1 If the readout is included (λ 6= 0), the modes become coupled. Since L 6= L†, the transitions between the modes are not al- ways possible in both directions as indicated by the arrows. stp0 − p0 stp1. generator L0 := Lλ=0: − iL0 =   −2γ0 +γ1 +2γ0 −γ1 0 −2γ0 + Ω× 0 0 0 +2γ0 0 0 0 0 +γ1 −γ1 + Ω×   (37) This can be brought into diagonal form easily by noting that the cross product operation is diagonal in the basis of the complex unit vectors [Eq. (B.30)] e0 = Ω = Ω/Ω, e± = (e1 ∓ ie2) /√2. (38) (39) These are constructed from a right-handed orthonormal system (e0, e1, e2) with e0 = Ω and an arbitrary choice of unit vectors e1 and e2 in the plane perpendicular to e0. The complex unit vectors satisfy the orthonormality and completeness relations (α = 0,±): e†α · eα′ = δαα′, Pα eαe†α· = 1, (40) where here the dot · denotes the scalar product taken with the object to its right. Writing L0 in diagonal form, . where τ n coefficients are given by α := e†α · τ [see Eqs. (38) and (39)] and the introducing the often recurring rate combination γ := 2γ0 + γ1. (50) as seen by inserting Eq. (27) into Eq. (50) and noting Γ/T ≪ 1. The conjugated left and right eigenvector of the charge decay mode read, respectively: 14 2 st The right zero eigenvector corresponds to a physical state, a valid density operator, which is factorizable, 1. c = ρS,st ⊗ 1 ρq In this state, the SQD is stationary, P n, and the qubit is in the completely ρS,st = Pn pn mixed state with zero Bloch vector (τ 0 = τ 1 = 0, c.f. Fig. 4, upper left). Any valid solution of the λ = 0 kinetic equation ρ(t) = −iL0ρ(t) always involves this stationary charge mode superposed with other modes. These addi- tional modes contain the isospin precession as we explain in the next paragraph. As one can see from expansion (45), the coefficient vq c (t) = 1 for all t and irrespective of the initial condition because the corresponding left zero eigenvector is just the trace operation, guaranteeing that ρ(t) has unit trace for all times t: tr(ρ(t)) = Xn pn = Vq†c · ρ(t) = 1. (51) The remaining seven "modes" have zero trace (see Eq. (43) with k = q, p = c) and therefore cannot repre- sent proper density operators on their own. These modes cannot be excited alone: They always appear in combi- nation with the stationary charge mode. In this respect these modes differ from modes encountered in, e.g., pure- state unitary evolution problems. There are three more quasistationary modes (k = q), for which the SQD remains in the stationary state ρS,st but the qubit state is not completely mixed, i.e., the isospin τ is polarized (see upper right of Fig. 4). The re- lated conjugated left and right eigenvectors, respectively, read: Vq α =   0 0 eα eα   , Vq α =   0 0 p0 eα st p1 eα st   . (52) The expansion (45) shows that the coefficients of these quasistationary isospin modes are connected with the to- tal isospin τ = τ 0 + τ 1. If the mode α = 0 is excited, the total isospin τ points along the qubit axis e0 and does not precess. If the other two modes α = +(−) are ex- cited, the total isospin τ precesses (counter)clockwise in the plane perpendicular to e0 with frequency Ω. Thus, if the isospin was nonzero initially, expansion (45) involves at least one of these three modes in addition to the qua- sistationary charge mode. In the case λ = 0, the former are not damped and the magnitude of the total isospin remains unchanged, reproducing exactly the free unitary evolution of the qubit. In addition to these four (quasi)stationary modes, there are four more decay modes (k = d) that are expo- nentially damped in time. As Fig. 4 illustrates, the eigen- values of these modes are well-separated from the quasis- tationary modes in the complex plane since Ω ≪ γ ∼ Γ Vd c =   +p1 st −p0 0 0 st   c =   +1 −1 0 0   , Vd . (53) If only this mode is excited in addition to the funda- mental stationary charge mode, the SQD state devi- ates from the stationary state, i.e., the coefficient δst := p1 stp1 6= 0 in Eq. (45), while the qubit remains in stp0 − p0 the completely mixed state (τ 0 = τ 1 = 0). Clearly, such deviations from the stationary SQD state decay on the short time scale 1/γ set by the SQD tunneling dynam- ics (see Fig. 4). Finally, there are three decaying modes with conjugated left and right eigenvectors, respectively, (α = 0,±) in which the isospins are polarized: 0 0 , Vd . (54) Vd α =   0 0 +p1 st −p0 st eα eα   α =   +eα −eα   stτ 0 − p0 The coefficients of these modes characterize the decay of stτ 1 of the charge- the weighted difference δst = p1 specific isospins [see Eq. (45)]. The weighted difference δst and the sum τ = τ 0 +τ 1 are linearly independent and together uniquely determine τ 0 and τ 1. Note that δst in- volves the stationary occupation probabilities in contrast to δ defined by Eq. (17). Similar to the total isospin for the quasistationary modes, the difference δst can point along e0 = Ω without any precessional motion for α = 0 or it can precess (counter)clockwise in the plane perpen- dicular to e0 for α = +(−) (c.f. lower right of Fig. 4). We see now that for λ = 0, the total isospin τ decouples from the motion of all other degrees of freedom and par- ticularly also from δst. In contrast to τ , the difference δst is strongly susceptible to the tunneling dynamics of the SQD, i.e., the switching between charge states n = 0 and n = 1. This generates the noncollinearities of τ 0 and τ 1, which are responsible for the qubit decoherence for nonzero capacitive coupling [see discussion in Sec. II C, Eq. (18) ff.]. What is important for the following is that that the expansion (45) carries over to the case of nonzero coupling λ 6= 0; even in this case, it suffices to compute the evolution of the quasistationary modes to obtain the evolution of the total isospin [102]. This ob- servation provides the starting point of the subsequent discussion where we first investigate how the quasista- tionary modes – "containing" the qubit dynamics – are coupled to the decay modes and then even explicitly elim- inate the decay modes (except for their initial state). 2. Coupling of quasistationary and decay modes We now turn the capacitive interaction back on, λ 6= 0, and investigate what we can say about the evolution us- 15 modes into the decay modes and back. These virtual processes are responsible for the qubit relaxation and de- coherence; the multiplet of quasistationary eigenvalues in Eq. (41) acquires an imaginary part Ωα → Ωα + iγα for α = 0,±, which induces a shrinking of the isospin Bloch vector. From Eq. (56) and Fig. 4, we expect that 1/T1, 1/T2 ∼ ΛqdΓ−1Λdq ∝ λ2/Γ, i.e., if the couplings Λqd and/or Λdq are small, then so will the backaction be. Before we make this more precise [see Eq. (69)], we investigate the detailed form of these couplings, which contains the first main result of the paper. 3. Mitigation of cotunneling noise by coherent backaction The transition matrix from the quasistationary modes into the decay modes reads [see Appendix B 3] Λdq = (cid:18) 0 (1 + p0 st)cλ (1 + p0 −rλ× (cid:19) , st)cλ· (61) (62) ing the above discussion of the eigenmodes of the Liou- villian L0. We can identify new dynamical variables that characterize the evolution in the quasistationary and de- cay subspace, respectively. Based on Eq. (45), we intro- duce Xq = (cid:18) p0 + p1 τ 0 + τ 1 (cid:19) , Xd = (cid:18) p1 stp0 − p0 stτ 0 − p0 p1 stp1 stτ 1 (cid:19) , (55) and rewrite the kinetic equation as d dt(cid:18) Xq Xd (cid:19) = −i(cid:18) Lqq 0 + Λqq Λdq Λqd 0 + Λdd. (cid:19)(cid:18) Xq Xd (cid:19) . Ldd (56) This is the second main equation of our study. The Λkk′ blocks are given and discussed below except for Λdd, which is not needed here [it is given by Eq. (B.34) in Appendix B]. First, the action of the unperturbed Liou- villian on these variables reads trivially 0 − iLqq −iLdd 0 = (cid:18) 0 0 = (cid:18) −γ 0 Ω× (cid:19) , 0 −γ 1 + Ω×(cid:19) , 0 (57) with the transition factor (58) r = p0 stp1 st − κ(cid:18) 1 2 st(cid:19) , st − p0 p1 0 = Ldq and Lqd 0 = 0 for the λ = 0 solution. The perturba- tion Λ has two effects. It first introduces a direct action on the quasistationary variables: − iΛqq = (cid:18) 0 0 p1 0 stλ×(cid:19) . (59) This produces the mean-field backaction, which amounts to a tilting of the internal qubit field Ω as anticipated in Sec. IV B: adding Λqq to Lqq 0 gives the effective qubit field Ω = Ω + p1 stλ. (60) The term Λqq, i.e., the term linear in λ thus does not lead to dissipative dynamics of the isospin contrary to what one might naively expect. The isospin decoherence is at least quadratic in λ (see below). We also note that the effective qubit field Ω is modified by O(Γ2/T ) pro- cesses (cotunneling broadening, level shift), which affect the stationary occupation probability p1 st [see Eq. (49)]. Moreover, if one is close to resonance, the probability p1 st may be a sizable fraction of 1, which it approaches in the Coulomb blockade regime for ε ≪ µs, µd. Since we allow for λ ∼ Ω, this implies that the correction p1 stλ leads to a large change in the qubit frequency and the direction of the qubit axis for a large range of parameters. The second effect of the perturbation is due to the coupling of the quasistationary variables to the decay- ing variables due to the off-diagonal blocks Λqd and Λdq. As a consequence, the decaying variables cannot be just ignored after a time ∼ 1/γ since they are permanently excited by virtual transitions from the quasistationary while the transition matrix back into quasistationary modes is given by Λqd = (cid:18) 0 0 −λ×(cid:19) . 0 (63) We first note that Λ, like L0, is non-Hermitian, Λ† 6= Λ, since the qubit-sensor evolution is nonunitary due the tunneling. As a consequence, the two transition matrices are markedly different: while transitions from the decay modes to quasistationary modes (Λqd) do not depend on the parameters of the SQD (level position, bias voltage, and tunneling rates Γ), transitions from the quasistation- ary modes (Λdq) exhibit a strong dependence on the sen- sor QD parameters that we discuss below. That Λqd is entirely induced by the readout interaction HI can be shown to be a consequence of the probability conserva- tion together with the conservation of the isospin during tunneling processes. The latter is specific to the indirect measurement setup. st/2 − p0 With Eq. (61) in hand, we can now precisely pin- point what we mean by stochastic backaction: the di- agonal component of Λdq can be split into a first term, p0 stp1 stλ, associated with the stochastic backaction, and a second term, −κ(p1 st)λ, associated with the co- herent backaction as signaled by the factor κ. The com- bination p0 st appears as a simple consequence of the charge fluctuations of the SQD, which are characterized by hn2i − hni2 = (1 − hni)hni = p0 st for a two-level stp1 system (see Sec. V A). The rates γ0,1 determining p0 st incorporate both the effect of the SET tunneling as well as that of next-to-leading order corrections. The stochas- tic term is multiplied with λ because to act back on the stp1 stp1 qubit, the "tunneling noise" has to act together with the internal interaction HI = nλ · τ /2 to evoke a λ-induced transition mediated by Λdq. The most striking finding is that Λdq is strongly sup- pressed when tuning the SQD towards Coulomb block- ade. To see this, we first note that Eq. (61) in- corporates the isospin-to-charge conversion rates (34), r )′, which depend on the derivative of the Fermi function. These rates are thus exponentially suppressed in the Coulomb blockade regime. We next inspect the diagonal element of Λdq. It is useful to first consider the expansion of the transition factor r to zeroth order in Γ/T : c ∼ Pr 2T (−f + Γr r = 2Γ+Γ− T(cid:19) , (2Γ+ + Γ−)2 + O(cid:18) Γ (64) st or p1 with Γ± :=Pr Γrf±r and 2Γ+ + Γ− > Γ :=Pr Γr. The terms in Eq. (64) derive only from the stochastic part p0 stp1 st in Eq. (62). Thus, in the single-electron tunneling approximation, the factor r is exponentially suppressed with gate voltage since either p0 st becomes exponen- tially small when going off-resonance. One would expect that this exponential dependence is removed by includ- ing cotunneling corrections and the coherent backaction, which scale algebraically with ε − µr and start to domi- nate over the single-electron tunneling rates as one moves into the Coulomb blockade regime. In our calculation, we include these terms as well, but still obtain an ex- ponential suppression of the transition factor. Indeed, expanding Eq. (62) to the next order in Γ/T , we find r = 1 T (2Γ+ + Γ−)2 (cid:20)2Γ+Γ−(cid:18)1 + Pr Γrφ′r (cid:19) (Γ+)′Pr Γrφr 2Γ+ + Γ− (cid:21) + (Γ− − 2Γ+)(Γ− + Γ+) +O(cid:18) Γ2 T 2(cid:19) , (65) with Γ′ = dΓ+(ε)/dε =Pr Γr(f + r )′/T . Clearly, Γ+, Γ−, and Γ′ are determined by the Fermi functions and their derivatives. Thus, transitions from the quasistationary modes into the decay modes become exponentially small when tuning the SQD into the Coulomb blockade regime. What this implies is that any deviation from the expo- nential suppression must be due to even higher order tunneling contributions (i.e., beyond cotunneling) and thus must be a higher power law ∼ 1/(ε − µr)n with n ≥ 2. The experimentally important conclusion that we can draw from this is that the sensor can be switched off better with the gate voltage than naively expected [see Eq. (27)]. What has happened in Eq. (65) is that the coherent backaction κ ∼Pr Γrφ′r/T ∝ 1/(ε − µr), which depends algebraically on ε, has completely canceled out the alge- braically scaling cotunneling corrections to the stochas- tic backaction in the first term of Eq. (62). Hence, in Eq. (61), the coherent backaction term (Γλ/T ) counter- acts the change in the stochastic backaction term due 16 to a change in the sensor QD rates by the cotunneling (λ/Γ × Γ2/T ∼ Γλ/T ). This can be seen by explic- itly comparing Eq. (64) to Eq. (65) and is clearly visible in Fig. 3. We emphasize that for this cancellation also non-Markovian effects induced by the electrodes are im- portant, which modify the coherent backaction (see Sec. III B 4). Without these, the transition factor r exhibits a different dependence on the level position ε that can lead to a violation of positivity of the qubit-SQD density oper- ator as we discuss further in Appendix A. Moreover, the cancellation does not imply that the backaction is pre- cisely the same as when only accounting for the lowest- order Γ approximation (see Fig. 3); the renormalization of the level position and non-Markovian contributions, both scaling exponentially, still modify the backaction. By formulating the problem in Eq. (56) in terms of the λ = 0 eigenmodes one most clearly sees how the cotunneling and coherent backaction, formally terms of different order, conspire to effectively cancel out. Note also that the dissipative backaction (through c) appears on its own. This highlights the importance of keeping track of all three types of backaction that are revealed only after integrating out the electrodes coupled to the sensor QD to obtain our central Eq. (26). It should be noted that the dissipative backaction couples the qua- sistationary modes to the decay modes and therefore is not relevant for the leading-order λ2/Γ dephasing times as Appendix B 3 b shows; see also the discussions after Eq. (34) and after Eq. (111) below. Rather, the dissi- pative backaction must be kept to be able to calculate the response of the dissipative sensor current of which it represents the flip side, as explained in the introduction. As emphasized in Sec. III B 3, we were careful through- out our analysis to include all terms which depend on the function φ′r with algebraic tails that could possibly can- cel out. In Appendix B 4, we further discuss the cancel- lation in view of our weak-measurement, weak-coupling assumption (see also Sec. II B). An important conclusion, which we draw in Sec. V, is that this cancellation of cotunneling noise and coherent backaction cannot be understood within simple classi- cal fluctuator model. Although this approach could, in principle, be extended to account for the cotunneling- induced noise by modifying the switching rates, it seems not possible to include the coherent backaction. More- over, other approaches that aim at directly calculating the qubit Bloch vector τ must make an assumption about the qubit environment, in particular the sensor QD. Here one is liable to miss the above cancellation as we also dis- cuss in Sec. V B 2. It is furthermore interesting to observe that this cancel- lation appears even though the coherent-backaction in- duced torque terms in the kinetic equations (26) scale with λ, while the cotunneling corrections do not. How- ever, to affect the qubit, the "cotunneling noise" has to act together with the internal interaction HI ∼ nλ · τ to evoke a λ-induced transition mediated by Λdq. This is why they both affect the measurement backaction to first order in λ. Note also that the transition factor r is not only independent of the SQD-qubit interaction λ (it ap- pears as a factor in Λdq) but also of the internal qubit field Ω. This means that the relative importance of coherent backaction over the stochastic backaction cannot be al- tered by measuring weaker or stronger (i.e., by changing λ); in the weak measurement limit, these effects phys- ically come together and should be calculated together [103]. Other experimental parameters alter this compe- tition and their effect is studied in the next section. In short, the delicate interplay of the qubit plus sensor renormalization ∝ Γ and sensor cotunneling rates ∝ Γ2 in indirect capacitive detection may be rationalized as follows. By keeping track of the sensor-quit coherences (since both are quantum systems), we find that coherent effects counteract decoherence, which is not really that unexpected. This may in fact present a key difference of a sensor with quantized levels from a single-electron transistor with a continuous spectrum. A comparison of both types of detectors regarding the importance of renormalization effects is therefore an interesting future task. Finally, we emphasize that for the above cancella- tion the modeling of the qubit as a charge qubit is not relevant as long as the isospin couples to the charge of the SQD. D. Experimental control over backaction strength Since Eq. (65) is a key result of this paper, character- izing – together with c – the strength of the backaction [beyond the mean-field effect in Eq. (60)], we now in- vestigate its dependence on experimental parameters in some detail. 1. Gate-voltage dependence of transition factor r. In Fig. 5(a), we plot the dependence of the transition factor r on gate voltage. The figure shows two curves, one including the coherent backaction (blue, κ 6= 0) and the other excluding the coherent backaction (green, κ = 0). Clearly, the factor r is the largest in the sequential tun- neling regime of the SQD (Vg . Vb/2). Here, transitions from the quasistationary modes into the decay modes are induced by the fast succession of tunneling electrons, which impose a strong noise on the qubit. The coherent backaction and cotunneling are negligible in this regime. This drastically changes when tuning the SQD into the Coulomb blockade regime (Vg & Vb/2). The full tran- sition factor r is actually exponentially suppressed with gate voltage [linear on the scale of Fig. 5(a)]. By con- trast, the gate-voltage dependence is markedly nonexpo- nential when neglecting the coherent backaction, charac- teristic of cotunneling noise, see also the discussion of Eq. (65). Experimentally, we expect the cancellation to be reflected in the voltage dependencies of the qubit relaxation and 17 dephasing rates provided the sensor can be made the dominant environment (which should be the case for a good qubit, for which noise from manipulation "chan- nels" can be switched off). Measuring the qubit deco- herence rates could clearly distinguish between an al- gebraic and exponential dependence in an experiment. Here, we expect the measurement-induced decay rates to scale exponentially into the Coulomb blockade regime until higher-order tunneling processes at least of order Γ3/T 2 become relevant. They can lead to a crossover to an algebraic scaling ∝ 1/ε − µrn with n > 1 deep in the Coulomb blockade regime, see Appendix B 4. In any case, our numerical examples illustrate that a QD detector can be switched off more efficiently with a gate voltage than naively expected. 2. Bias-voltage dependence of r We continue with the discussion of the bias dependence of the transition factor r, which we show in Fig. 5(b). For small bias voltages, the SQD is Coulomb-blockaded and the coherent backaction strongly suppresses r. When the bias is increased, sequential tunneling sets in and when the level position ε ≈ Vb/2 is resonant with the elec- trochemical potential of the drain, the transition factor saturates. Here, the correction due to the coherent back- action actually becomes positive, as shown in the inset in Fig. 5(b). Yet, one should note that in the sequen- tial tunneling regime the coherent backaction has only a small impact. For even larger bias voltages, the correc- tion from the coherent backaction drops to zero. This is in accordance with the general finding that renormaliza- tion effects can be neglected in the large-bias limit [31, 32] because κ =Pr Γrφ′r/T ∼Pr Γr/(ε−µr) ∝ 1/Vb is sup- pressed [see Eq. (33)]. 3. Tunnel coupling dependence of transition factor transition factor r We finally discuss the impact of the coherent back- action when changing the tunnel couplings and their asymmetries. Since the coherent backaction is linear in κ ∼ Γ/T [see Eq. (35)], increasing the average tunnel cou- pling ¯Γ = (Γs +Γd)/2 and lowering the temperature both increase renormalization effects in a trivial way (within the limit Γ/T ≪ ∆/Γ ≪ 1). By contrast, the asym- metry of the tunneling rates in the generic experimental situation, quantified by g := Γs − Γd Γs + Γd , (66) may have a nontrivial effect. Controlling the asymmetry has also been suggested [13] as an experimental strategy for optimizing sensor efficiency in the limit Γd/Γs > 1. 18 ponential suppression with gate voltage effected by the coherent backaction, found in Fig. 5(a), is simply rigidly shifted horizontally without changing its shape consid- erably. This can be understood from the fact that the maximum of r as a function of Vg is shifted by an asym- metry g due to a basic effect of Coulomb interaction [104]. In the vicinity of this maximum, leading-order processes dominate and cotunneling and the coherent backaction effects can be ignored. One then finds the rule that for large asymmetries (g ≫ 1), the SQD level position ε = −Vg effectively lies closer to resonance with the elec- trochemical potential of the more strongly coupled elec- trode. Now, for positive bias Vb, as in Fig. 6(a), the electrochemical potential µs = +Vb/2 (µd = −Vb/2) of the source (drain) is resonant with the QD level for neg- ative (positive) values of Vg. Thus, when the coupling to the source is larger (smaller) than that to the drain, the maximal transition factor r is achieved for negative (positive) Vg as Fig. 6(a) demonstrates. FIG. 5: Voltage dependence of the transition factor r, Eq. (62), which determines the measurement-induced backac- tion of the sensor QD on the qubit (e.g., decoherence rates). We include the coherent backaction in the blue curves (κ 6= 0), while we exclude it by hand for the green curves (κ = 0). We show the dependence of r in (a) on gate voltage Vg for bias voltage Vb = 2.5T and in (b) on bias voltage Vb for gate volt- age Vg = 5T . In all plots we use Γs = Γd = ¯Γ = 0.02T and bandwidth W = 1000T . The inset in (b) shows the difference of the two curves for κ = 0 and κ 6= 0 and illustrates that the coherent backaction can also enhance the backaction for larger bias voltages. Note that r is independent of λ and Ω [choosing, e.g., λ ∼ Ω ∼ 0.1¯Γ would be a parameter combina- tion consistent with the conditions for the kinetic equations (26) to be valid: ¯Γ/T ≪ λ/Γ ≪ 1]. In the stationary limit, we previously found [32] that for Γd/Γs < 1 the impact of the coherent backaction is strongly enhanced. The effect of asymmetries on the transition factor r [Eq. (62)] strongly depends on the chosen bias and gate voltages. To illustrate this point, we plot in Fig. 6(a) the relative change in r, K = rκ=0 − rκ6=0 rκ6=0 , (67) due to the coherent backaction as a function of gate volt- age Vg for the different values of g as indicated. When introducing a nonzero junction asymmetry g 6= 0, the ex- FIG. 6: Relative change K in the transition factor r, Eq. (62), by the coherent backaction as a function of (a) the gate voltage Vg for Vb = 2.5T and (b) of the bias voltage Vb for Vg = 1.5T . The different curves correspond to the indicated values of the tunneling asymmetry g = (Γs−Γd)/(Γs+Γd) for fixed average tunneling rate ¯Γ = (Γs + Γd)/2 = 0.02T and bandwidth W = 1000T . By contrast, Fig. 6(b) shows that the impact of the asym- metry g on K as function of the bias voltage is more complicated close to resonance. In general, K decreases with Vb, as Fig. 5(b) already showed for g = 0. In the limit Vb → ∞, one expects κ ∼ 1/Vb → 0 and there- fore K → 0. Moreover, we find K > 0 for small bias [generally valid] and for large bias [specific to the param- eters chosen in Fig. 6(b)]. Thus, both for large and small bias the coherent backaction suppresses the measurement backaction. However, for intermediate bias voltages, K shows strong drops as Fig. 6(b) illustrates. This sup- pression appears since κ and (p1 st) change their sign in the vicinity to resonance at ε = µr and there- fore become both very small. This suppresses K and for a small intermediate bias regime K can become negative as pointed out already in Sec. III D 2. Figure 6(b) reveals that the position and even the existence of these drops and sign changes depends crucially on the asymmetry g and the gate and bias voltage polarity. In particular, for Vg > 0, as assumed in Fig. 6(b), the source cannot become resonant with the level for positive bias. Accord- ingly, the drop is absent for g = 0.5 in Fig. 6(b), i.e., when the source is coupled more strongly to the SQD than the drain. st/2 − p0 In summary, the coherent backaction starts to can- cel the cotunneling-induced stochastic backaction on the flanks of the SQD resonances where one enters Coulomb blockade – the optimal spot for sensing – with correc- tions depending on the junction asymmetry, see Fig. 6. Junction asymmetries commonly found in experiments strongly affect the relative importance of the coherent backaction terms and thus also the qubit decay rates. E. Effective evolution of quasistationary modes Since the qubit isospin is contained in the quasista- tionary degrees of freedom [see Eq. (55)], it is of interest to find a description only for their evolution. Impor- tantly, we do this without making further approxima- tions, i.e., we reproduce exactly the result for Xq(t) as one does when solving Eq. (26) for ρ. For all practical purposes, one should solve the latter equation [or equiva- lently Eq. (56)]: Eqs. (71) and (72) below merely serve to highlight some properties of this solution, in particular, its non-Markovian nature and the impracticability for re- ally getting rid of the sensor dynamics. It also serves as a basis for Sec. IV. Equation (56) is a coupled set of differential equa- tions for Xq(t) and Xd(t). The equations can be solved most conveniently by Laplace transformation, defined 0 dteiztf (t) for any (well-behaved) time- dependent function f (t). This yields by f (z) = R ∞ Xq(z) = i z − Lqq eff(z) Xq eff (z), (68) with a Laplace-frequency dependent effective Liouvillian 19 0 + Λqq + Λqd eff(z) = Lqq Lqq 1 z − Ldd 0 − Λdd and matrices given by Eqs. (57) – (59), (61), (63), and (B.34). The frequency-dependent initial condition is given by Λdq, (69) Xq eff (z, 0) = Xq(0) + Λqd 1 z − Ldd 0 − Λdd Xd(0), (70) where we take the initial time to be t = 0. The inter- mediate steps of the derivation of Eqs. (68) – (70) are given in Appendix B. There we show how to reproduce the above result in the general framework of a projec- tion approach following that of Nakajima and Zwanzig [48, 55–57]. The projection approach simply separates the dynamics in the complementary subspaces spanned by the quasistationary and decaying modes. Importantly, this projection technique treats all the different types of backaction we discussed so far on the same footing, which allows us to go beyond the approaches of Refs. [88] and [33]. Transforming the solution (68) back to time space and exploiting the convolution theorem yields [see Appendix B] dt′Πqq Xq(t) = Z t eff (t − t′)Xq Xq eff (t′) = Xq(0)δ(t′ − 0) + Πqd 0 eff(t′), eff (t′)Xd(0), (71) (72) where δ(t′ − 0) indicates a δ-function with infinitesimal shift and Πqq Πqd eff (t) = Z ∞ eff (t) = Z ∞ −∞ −∞ dz 2π dz 2π e−izt e−iztΛqd , i z − Lqq eff(z) 1 z − Ldd 0 − Λdd (73) . (74) Equations (71) and (72) are the third set of main equa- tions. We stress that they are generally valid in the sense that they do not involve any approximations be- yond those needed for the validity of the kinetic equa- tions (26): Eq. (71) exactly reproduces Xq(t) as obtained from the solution of Eq. (56). The expression for Xq(t) simplifies drastically for times t ≫ 1/γ if one accounts only for the leading-order contributions in ∆/γ. Such an expansion is valid only in the weak-measurement limit ∆/γ and complies with our kinetic equations in in the high-temperature limit Γ/T ≪ ∆/Γ as we discuss fur- ther below in Sec. IV. Before doing so, let us first note a couple of general properties of the above solution. Non-Markovian dynamics and time scales. The effective Liouvillian (69) confirms the discussion in Sec. III C 2: the unperturbed evolution in the quasista- tionary subspace is perturbed directly by Λqq, which can be absorbed into a redefinition ¯Lqq 0 + Λqq that just leads to the mean-field tilting of the qubit axis. 0 = Lqq Moreover, the third term in Eq. (69) gives an explicit expression for the indirect and in general non-Markovian perturbation of the quasistationary evolution by virtual transitions via the decay modes. This interpretation fol- lows most clearly from an alternative derivation in time space given in Appendix B 2 and illustrates that the time delay between these transitions - giving rise to non- Markovian effects - converts into the frequency depen- dence of this expression. This term entails the effect of the stochastically fluctuating deviations from the mean field (see Sec. II C) and separates it from the mean-field effect contained in Λqq. Since the denominator in this ex- pression does not grow exponentially (it rather tends to be constant because Lqq 0 ∼ γ), the voltage dependence of this term is largely determined by that of Λdq, which we discussed above. The third term in Eq. (69) has, in general, several ef- fects: the eigenvalues of ¯Lqq 0 in general acquire (i) a real part leading to an additional shift of the qubit frequency, and (ii) an imaginary part, which corresponds to the re- laxation and dephasing rates of the qubit. To extract both effects, one has to inverse the Laplace-transformed function (68), which leads to an integral that should be computed by the applying the residue theorem. The residues are determined by the zeros of the denominator satisfying zp = Lqq eff (zp) and determine the time scales of the qubit evolution. (iii) Moreover, the third term may not commute with ¯Lqq 0 . This induces transitions between the unperturbed eigenstates and leads to a rotation of the qubit eigenbasis. (iv) Finally, since the third term is not Hermitian due to the decoherence that it induces (in contrast to ¯Lqq 0 ), the qubit eigenaxes may not be mu- tually orthogonal any more. This renders the circular precession induced by ¯Lqq slightly elliptical as we illus- 0 trate below in Sec. IV. The effective Liouvillian (69) also allows for a compar- ison with earlier results. We note that our approach is conceptually quite similar to that discussed in Ref. [88]. However, Ref. [88] employs the clearly stated additional assumption that the effect of the electrodes is only to modify the SQD dynamics (assumed to obey a Lindblad equation) without affecting the (effective) coupling to the qubit. In our formulation, this would mean that the cou- pling to between the quasistationary and decay modes is mediated only by the interaction HI = nλ · τ /2, i.e., only the stochastic backaction. This means that in the approach of Ref. [88] both the dissipative and coher- ent backaction are neglected. Within this approximation one can thus not calculate the experimentally measur- able signal current; it must be consistently set to zero. (The signal current was not calculated nor of interest in Ref. [88]). Moreover, Ref. [88] focuses only on the time scales and does not consider initial-slip effects that we next turn to. Initial slip. Equations (70) and (72) show that the decay modes affect the quasistationary modes not only through the effective Liouvillian but also through the effective initial state [79, 90–93]: the latter is not just 20 given by the initial quasistationary variables contained in Xq(0). In addition, there is a term in (72) that ac- counts for a initial contribution from the decaying sub- space, Xd(0), followed by a time integral over transitions into the quasistationary subspace. This leads to an ini- tial slip that affects the quasistationary modes. Like the time evolution (71), the slip (72) of the initial state has a a time-nonlocal expression in terms of the initial state of the decay mode Xd(0) [in the Laplace transform (70) the corresponding second slip term has frequency depen- dence]. An experimentally relevant question is how to elimi- nate or minimize the initial slip since it can induce errors even for a perfectly prepared initial qubit state. In Ap- pendix C, we show that initial qubit-SQD states ρ(0) that exhibit no initial slip form a subset of measure zero in the set of all valid initial density operators ρ(0). In general, a sufficient condition for zero slip is that Xd(0) = 0, i.e., using Eq. (55): δst(0) = p1 δst(0) = p1 stp0(0) − p0 stτ 0(0) − p0 stp1(0) = 0, stτ 1(0) = 0. (75) (76) To find all initial states with zero slip, one has to compute 0 − Λdd)−1. Assuming the the kernel of matrix Λqd(z − Ldd frequency z does not hit a pole of the denominator (for example when considering the Markov approximation for 0 − Λdd)−1 exists and has full z = 0), the inverse (z − Ldd rank. Thus, to determine the dimension of the kernel, it suffices to determine the rank of Λqd, which is 2 since 0 0 −λ×(cid:19)(cid:18) x 0 (cid:19)(77) Λqd(cid:18) x y λ (cid:19) = (cid:18) 0 for any x, y ∈ R and using Eq. (63). Thus, the matrix 0 − Λdd)−1 has a kernel of dimension 2, which Λqd(z − Ldd means that the set of initial states has zero measure since Xd can be taken from a four-dimensional set. y λ(cid:19) = (cid:18) 0 One way to eliminate the initial slip is to switch off the capacitive interaction before t = 0, i.e., λ(t) = 0 for −1/γ ≪ t < 0. Then the detector can establish a sta- tionary state and the initial SQD-qubit state factorizes: Xd(0) = 0 [see the λ = 0 solution (46] and see also Ap- pendix C). By contrast, if one switches off the current through the sensor, Γ → 0, before t = 0, then one starts with a sensor in a definite charge state p1(0) = 0 or 1, which is highly nonstationary for typical operation pa- rameters of the SQD (tuned close to resonance for high sensitivity, one finds usually p0 st). The backaction- induced initial slip thus leads to an essential difference between two ways of switching off a sensor, which should be taken into account in designing detection protocols. st ∼ p1 The magnitude of the slip in general depends on fre- quencies only for z & Ldd 0 + Λdd ∼ γ. This means that the initial value Xd(0) influences Xq(t) for times t ∼ 1/γ [through the integral (72)]. In Sec. IV C, we investigate the slip magnitude in more detail in a high-temperature limit to leading order in λ/γ; importantly, we find that even in this simple case there is a slip effect of order λ/γ which affects the overall qubit dynamics by, e.g., phase- shifting the solution in a way depending on the sensor initial state. From the above we can generally conclude within the regime of validity of the kinetic equation (26) that due to the backaction-induced initial slip additional errors are generated; since the zero-slip states are sets of measure zero, it is clear that preparation errors of the qubit-sensor state will invariably lead to an initial slip. Models of such possible errors are actually relevant for a different branch of quantum information. In quantum-error correction, one deals with decoherence from the environment in a phenomenological way by introducing additional bit-flip errors [95]. This requires assumptions to be made about the type and statistics of the different possible errors. Our work thus provides in this context a possible scenario how such errors may arise and how they can be modeled, after, e.g., a measurement has been performed. This will become more concrete in the next section, where we dis- cuss simplified equations, which have, however, only a limited range of applicability. IV. HIGH-TEMPERATURE QUBIT DYNAMICS The kinetic equation, either in representation (26) or (56), form central results of this paper. They fully suf- fice to compute the transient dynamics of the charge- specific isospins τ 0(t) and τ 1(t). From this result, the total isospin τ (t) = τ 0(t) + τ 1(t), i.e., the reduced qubit state can be constructed. However, there are several rea- sons to attempt to obtain a closed description in terms of τ (t) only. First, from Eq. (26), it is not directly clear on which time scales τ (t) evolves or decays. Second, an effective qubit description plays an important role, for example, in quantum error correction. It is an interesting question how far Eq. (26) can actually be reduced to a closed equation for the reduced density operator for the qubit alone. One indication that this requires additional as- sumptions is that (26) and (56) can be solved only if the full initial state vector [either in form Eq. (13) or Eq. (55)] is specified and not just the sum τ 0(0) + τ 1(0). We have already seen that the initial values of the other degrees of freedom produce an initial slip [see Eq. (70)] and will see below that even in lowest nonvanishing order this slip cannot be avoided. The third consideration is related to this and concerns the minimization of backac- tion in quantum-information processing: One would like to know the effective qubit eigenmodes, e.g., to construct initial states that are least sensitive to backaction by set- ting experimental parameters. To investigate all this further systematically, an effec- tive theory for the qubit evolution in a simple limit is useful. In this section, we consider the regime where the coherent backaction and O(Γ/T ) corrections to the stochastic backaction can both be neglected. We stress that this is for the purpose of illustration mostly since 21 the latter effects are relevant under typical experimen- tal conditions. This simplification allows us to perform a expansion of the effective Liouvillian (69) and the initial slip (70) to leading order in λ/γ and we investigate the resulting transient qubit evolution here in some detail. In Sec. III D 1, we give tangible analytical expressions for the relaxation and dephasing rates as well as for the qubit precession frequency. In Sec. IV B, we assess the accu- racy of the approximate theory by comparing with the numerical solution of the full kinetic equation (26). We further discuss the slip of the initial condition of the qubit isospin in Sec. IV C which relates to a "kick" that the qubit experiences during the relaxation time of the sensor QD. Finally, we show in Sec. IV D that the measurement backaction forces the isospin to precess about a tilted axis in an elliptical way. The eccentricity is connected to oscil- lations in the decay of the purity of the qubit state. This illustrates concretely that our density-operator approach goes beyond standard master-equation approaches as we discuss in Sec. V. A. Effective Liouvillian, initial slip, and mode vectors The kinetic equation [Eq. (26) or Eq. (56)] was de- rived using the weak-coupling, weak-measurement limit, Γ/T ≪ ∆/Γ ≪ 1 where ∆ ∼ λ, Ω. This prevents one from just expanding in λ/Γ since that would imply tak- ing λ/Γ ≪ Γ/T . However, if we consistently neglect the corrections Γ/T (the cotunneling corrections to the stochastic backaction as well as dissipative and coherent backaction), then we can take λ/Γ → 0 and expand in this parameter. We refer to this as the high-temperature limit (since it is the large temperature that allows one to take the infinitely weak-measurement limit). It should be noted that in this approximation the current through the sensor QD is zero, i.e., at this level of the theory one is not accounting for the actual backaction effects due to the current measurement (rates ∼ Γλ/T ), but only for the leading effect of the tunnel coupling. Below, the station- ary occupations p0,1 = γ1,0/γ are given by their leading [see order expressions (SET rates) γ0,1 = Pr=s,d Γrf±r Eq. (27)] and γ = 2γ0 + γ1. We thus simplify the isospin evolution obtained from the effective Liouvillian (69) and for concreteness assume from hereon Ω = Ωex, (78) perpendicular to the capacitive interaction vector λ = λez. This means that if we ignored the mean-field tilting Ω → Ω (which we do not), the qubit would oscillate in the measurement basis. As we explain in detail in Appendix B, the isospin evo- lution contained in Eqs. (69) and (70) can be simplified by performing a Markov approximation, i.e., by replac- ing z = 0. This Markov approximation with respect to memory induced by the sensor QD (after integrating out the electrodes) is valid in the weak-measurement limit λ/Γ ≪ 1 (see Appendix B 3 c). We note that in the high-temperature limit also non-Markovian corrections due to tunneling processes (Γ) are consistently neglected as next-to leading order Γ2/T corrections are not ac- counted for. With z = 0, the Laplace-transform inverse of Eq. (68) can be easily performed. Expanding the de- nominator in powers of ∆/Γ and extracting the isospin from Xq, we find τ (t) = e−iLeff tτ eff (0) + O(∆2/γ2). (79) In this approximation the stationary state τ (∞) = 0. The effective Liouvillian in Eq. (79) reads in diagonal form: − iLeff = Xα=0,± (i Ωα − γα)eeff,αe†eff,α, (80) where the eigenvalues, the left and right eigenvectors, and the effective initial state τeff (0) are specified below (81)–(93). Equation (79) is valid for times by Eqs. 1/γ ≪ t ≪ γ2/∆3 as Appendix B shows and also our numerical checks below confirm. The lower limit indicates that we consider the wide-band limit [105] with respect to the sensor QD band-width γ (by setting γ ≫ z → i0 above), whereas for times t ≫ γ/∆2, corrections of O(∆3/γ2) to the effective Liouvillian (80) accumulate and the error made for τ (t) may become sizable. 1. Initial slip The effective initial condition appearing in Eq. (79) reads (see Appendix B): τ eff(0) = τ (0) − 1 γ λ × (p1 stτ 0(0) − p0 stτ 1(0)). (81) This shows that even in this simple limit the qubit description is still not closed. Although the entire sensor variables (electrodes plus sensor QD) have been eliminated, the initial condition does not depend only on τ (0) (see Appendix C). Instead, it additionally requires the specification of the component of the initial qubit-sensor state ρ(0) in the decaying subspace. both initial charge-specific isospins τ n(0) are Thus, needed to compute τ (t). In contrast to the general case discussed in Sec. III E, Eq. (81) only relies on stτ 1(0) and does not involve δst(0) = p1 δst(0) = p1 stp1(0). The reason is that the denominator in Eq. (74) is approximated here by −γ and Λqd does not act on the charge sector [see Eq. (63)]. As mentioned in the general discussion of the initial slip, Eq. (72), initial qubit-sensor states ρ(0) with zero initial slip form a zero-measure subset of all possible initial states. As a result, the initial slip adds to preparation errors, an error that depends on the sensor dynamical In the present simple limit, the initial slip is state. a time-local expression, Eq. (81), in contrast to the stτ 0(0) − p0 stp0(0) − p0 22 general case, Eq. (72). If the measurement is not weak any more, i.e., if λ ∼ γ, one can expect the dynamics of the charge-specific isospins to become important on the entire time scale of the qubit decay and not just through an effective slip of the initial condition. Still, even in the weak-measurement limit studied here, Eq. (81) shows explicitly that the slip is of non-negligible order λ/γ. 2. Qubit time scales The simple formulas (79)-(81) for the relevant preces- sion, relaxation, and dephasing time scales form the third central set of equations of the paper. We now discuss their contents. The eigenvalues in Eq. (80) contain the effective qubit frequencies Ωα = α Ω, α = 0,±, which up to O(∆3/γ2) read Ω = (cid:12)(cid:12)(cid:12) Ω(cid:12)(cid:12)(cid:12) = qΩ2 + (p1 stλ)2. Equation (83) is precisely the length of the mean isospin field announced earlier in Eq. (21), but now with hni = p1 st, Ω = Ω + p1 stλ = Ωek, (84) whose unit direction vector is relevant for the following: (82) (83) (85) (86) Moreover, we will need the perpendicular unit vector, ek = (cid:16)Ω Ω + p1 stλλ(cid:17) / Ω. e⊥ = (cid:16)Ωλ − p1 stλ Ω(cid:17) / Ω, lying in the plane spanned by Ω and λ. Our calculation thus confirms the intuitive picture explained in Sec. II C: The mean-field effect of the average SQD charge hni = p1 st is just to tilt the qubit axis (see also below in Sec. IV B 1) and does not rely on the tunneling-induced fluctuations ∼ δn, see Eq. (20). Tunneling influences the mean sen- sor charge only indirectly as noted in the discussion of Eq. (21). Compared to the first term of Eq. (41), the eigenvalues of the quasistationary modes have acquired small dissi- pative parts [see Eq. (80)]: γ0 = 1 T1 , γ± = 1 T2 . (87) Here, the relaxation rate is given up to O(∆4/γ3) by 1 T1 = r (λ · e⊥)2 γ = r(cid:18) Ω Ω(cid:19)2 λ2 γ , (88) which is quadratic in the component of the measurement vector λ perpendicular to the average isospin field Ω with the transition factor r given up to zeroth order in Γ/T as given by Eq. (64). The dephasing rate 1/T2 is expressed compactly in terms of the pure dephasing rate 1/Tφ = 1/T2 − 1/(2T1) [27] up to order O(∆4/γ3) as Ω (cid:19)2 λ2 = r(cid:0)λ · ek(cid:1)2 = r(cid:18) λp1 1 Tφ (89) γ γ st , which is quadratic in the projection of λ on the unit vector ek = Ω/ Ω along the average isospin field. In the following, we refer to both relaxation and dephasing as decoherence because both drive the qubit into a mixed state. Note that in both decay rates, the transition factor r appears, which links to the discussion of the previous sections. If higher-order ∆/γ terms are included into the relaxation and dephasing rate, additional terms ap- pear that depend on the dissipative backaction terms ∼ c stemming from the off-diagonal elements in Eq. (63). It is easy to see that the relaxation and pure dephasing time T1 and Tφ are positive [106] , since the transition factor is given to lowest order by r ≈ 2Γ+Γ−/(2Γ+ + Γ−)2 > 0 [see Eq. (64)]. Since Tφ > 0, the ratio T2/2T1 further satisfies the relation [27]: T2 2T1 = 1 stλ/Ω)2 < 1. 1 + 2(p1 (90) Equations (88) and (89) again confirm our intuitive ex- pectation from Sec. II C that only the fluctuating part of the SQD charge ∼ λδn (involving here virtual transi- tions into the decay modes) is responsible for the qubit decoherence. The energy scale λ2/γ for the decoherence rates exhibits the expected quadratic scaling with the weak coupling λ and inverse scaling with the large detec- tor band width γ, as discussed further below in Sec. V A. The decay time is thus slow compared to the time scale of the relaxation ∼ 1/γ of the sensor QD and that of the intrinsic evolution of the SQD-qubit system ∼ 1/∆. We emphasize that this picture and our approach hold only in the limit ∆ ≪ Γ: it breaks down if the tunneling be- comes to strong relative to either the measurement λ or the qubit internal field Ω. [107] One should note that the corrections to the decoher- ence rates γα [see Eqs. (88) and (89)] are of O(∆4/γ3), while the corrections to the qubit frequency (82) already appear in lower O(∆3/γ2). The reason is that these quantities behave differently under a simultaneous re- versal of the orientation of λ and Ω as one can see from a simple physical argument: mapping λ → −λ and Ω → −Ω corresponds to spatially mirroring the de- tection setup about the vertical axis in Fig. 1. This clearly inverts the sense of the precessional motion, i.e., one has Ωα (−λ,−Ω) = −Ωα (λ, Ω) but the qubit de- cay cannot depend on mirroring the setup, i.e., we have γα (−λ,−Ω) = γα (λ, Ω). This implies that corrections to γα must be of even order in ∆ and therefore at least of fourth order in ∆, in agreement with our calculation. 3. Mode vectors To complete the specification of the effective Liouvil- lian (80), we give the explicit formulas for the unit vectors eeff,α. Expressed in ek and e⊥ given by Eq. (85) and (86), respectively, they read: 23 λ2 γ Ω λ2 γ Ω Ωλp1 st ek × e⊥, Ω2 Ω2 − Ω2/2 Ω Ω + Ω Ω2 ek × e⊥, (91) (92) (93) eeff,0 ∝ ek + r eeff,1 ∝ e⊥ + r eeff,2 ∝ ek × e⊥ Ω st +r ek + Ω + Ω λ2 γ Ω Ω2 λp1 Ω2 − Ω2 + Ω Ω/2 e⊥! , where ∝ indicates that we suppress the normalization constants. As before [see Eqs. (38) and (39)], we define eeff,± = (eeff,1 ∓ ieeff,2) /√2 and we note that ek × e⊥ = Ω × λ [see Eqs. (85) and (86)]. In stark contrast to the unperturbed case, the real unit vectors eeff,0, eeff,1, and eeff,2 form a real nonorthogonal basis. This implies that to decompose a vector in the basis {eeff,α}, one needs to take the scalar product with the dual basis denoted {eeff,α}, see Fig. 7. The dual basis vectors eeff,α are non-unit vectors orthogonal to the plane spanned by eeff,β and eeff,γ, eeff,α = eeff,β × eeff,γ eeff,α · (eeff,β × eeff,γ) , (94) where (α, β, γ) is a cyclic permutation of (0, 1, 2). We refer to this nonorthogonality of the eigenvectors in the following simply as a distortion of the isospin modes. FIG. 7: Distortion of isospin mode vectors: (a) Three- dimensional sketch of the effective qubit axes eeff,α [α = 0, ±, Eqs. (91) – (93)] and its dual basis eeff,α. Note that eeff,0 has a nonzero projection on the plane spanned by eeff,1 and eeff,2, while eeff,0 is orthogonal to this plane. Likewise, a vector in the eeff,1-eeff,2 plane has a nonzero component along eeff,0. The distortion of the mode vectors scales with the small ratio of the magnitude of the decoherence rates λ2/γ rel- ative to the effective qubit frequency Ω ∼ ∆ and is fur- thermore suppressed by the transition factor r when go- ing off-resonance [see Eq. (62)]. Since we allow for λ ∼ Ω, the remaining factors in Eqs. (91)–(93) can be of order 1. This distortion also has tangible physical consequences: Eq. (91) shows that the precessional motion of the qubit isospin τ becomes tilted. It is also not circular any more, but becomes slightly elliptical instead as we investigate in more detail in Sec. IV D. detector backaction. It is valid in zeroth order in λ and Ω relative to Γ, where we neglect all relaxation and deco- herence rates λ2/γ ≪ λ. Expanding the expressions (91) – (93) for the mode vectors to the corresponding zeroth order yields an orthonormal basis, 24 4. Other approaches At this point, it is instructive to compare with some other approaches. At first sight, the mode distortion may appear peculiar and one may wonder why it does not show up in other approaches. In fact, the mode distortion disappears in the limit λ/Ω → 0, i.e., when the capacitive coupling becomes smaller than all other en- ergy scales. In this limit, the modes become orthogonal with e0 = Ω. This diminishes also the mean-field effect. This limit is equivalent to the frequently made secular approximation that Davies has shown to be exact in this strict weak-coupling limit [58, 59]. Here, however, we consider the more general situation that the capacitive coupling λ can be of the same magnitude as the internal qubit energy scale Ω. The secular approximation is not applicable in this case as we showed previously [32] in accordance with other works [46]. This is furthermore signalled by the observation that the secular approx- imation conflicts with the isospin conservation when electrons tunnel between the electrodes and the sensor QD. The strength of the mode distortion thus reflects the importance of nonsecular corrections (coherences). We next compare to Ref. [33], whose approach is similar to ours [108]. This study applies a two-step procedure to derive a closed, effective description of the qubit dynamics, starting from a generalized master equation for qubit plus SQD, as we do. To treat the limit Ω ≫ λ2/γ, it is additionally assumed that λ · e⊥ ≪ (cid:12)(cid:12)λ · ek(cid:12)(cid:12), i.e., perpendicular fluctuations of Beff along e⊥ are small as compared to longitudinal fluctuations along the mean field Ω. Perpendicular fluctuations are treated there perturba- tively. In this way coherences between isospin states quantized along ek are only effectively included. By contrast, we allow for comparable fluctuations in the direction of both ek and e⊥, which is a more general case. therefore B. Accuracy of effective isospin dynamics Before we illustrate the effects of the initial slip and the mode-vector distortion, we discuss the accuracy of our Liouville-space perturbation theory as compared to the full solution of the kinetic equations (26) in the high- temperature limit and up to times times t ∼ γ/λ2 ≫ 1/γ. 1. Undistorted, mean-field qubit modes As a starting point for this discussion, we first discuss our result in view of the rough mean-field picture of the eeff,0 ≈ ek, eeff,1 ≈ e⊥, and eeff,2 ≈ ek × e⊥, (95) with ek and e⊥ given by (85) and (86), respectively. Thus, in this approximation the dual basis {eeff,α} coin- cides with {eeff,α}. In analogy to the unperturbed case discussed in Sec. III C 1, the isospin just precesses cir- cularly about ek; however, the precession frequency – Ω instead of Ω – and the precession axis – along ek = Ω/ Ω instead of Ω – are different. This rough mean-field pic- ture of the measurement is therefore to tilt the "bare" isospin field to the mean isospin field (84) by an angle [see Eq. (21)]: tan θ = p1 stλ Ω . (96) In our concrete charge-qubit model, this means that the capacitive readout simply detunes the charge qubit due the gating effect of the sensor QD with mean charge p1 st. Here the mean charge is identified with the ensem- ble averaged charge, see the related discussion in Sec. II C. Since we only require ∆ ≪ γ but impose no con- straint on the ratio λ/Ω, this angle can be large. In Fig. 8, we illustrate this effect by showing the evolution of the three isospin components in the basis (ex, ey, ez) = ( Ω, λ × Ω, λ) on a long time scale t ≫ 1/γ when the isospin initially points into the direction of Ω, i.e., per- pendicular to λ. If the coupling λ was switched off, we would expect the isospin not to precess at all and to remain stable along Ω. By contrast, the oscillations of all components in Fig. 8 clearly demonstrates that the isospin revolves about a very different axis, roughly pointing in the direction of Ω + λ in line with Eq. (96) for the parameters employed here. 2. Accuracy of weak-measurement expansion (79) based on our perturbation theory, The green curves in Fig. 8 depict the difference be- tween the isospin evolution obtained by solving the ki- netic equations (26) and the evolution computed from Eq. indicat- ing that both agree well (plotted is the error multiplied by 100). The remaining deviation is mostly due to a small phase shift between the full and perturbative so- lution that accumulates in time. The origin lies in ig- nored corrections of order of ∼ ∆3/γ2 to the effective qubit frequency (82). In Fig. 8, this accumulates after the shown time ∆t ∼ 25 · (2π/ Ω) to a phase difference ∆ϕ . 25 · (λ/4¯Γ)2 ∼ 0.01, which is just visible. How- ever, we emphasize that the accuracy of the decay rates 25 y The approximate analytical solution (79) is valid over the entire time scale shown in Fig. 9(a) except for In Fig. 9(b), we illustrate very small times t < 1/γ. how the isospin computed from the full kinetic equations (26) (red) approaches the approximate analytical solu- tion (79) (blue): all curves for the full solution (red) start from the same value for τy(0) = 0 but immedi- ately develop differently depending on the initial SQD charge p1(0). On a time scale ∼ 1/γ, they approach the approximate analytical solutions τ ana (0) (blue), which are offset by the initial slip (81). Figure 9(b) confirms that precisely due to this slip the analytic solution ac- curately approximates the full numerical one for times t ≫ 1/γ. The latter time scale is expected since the approximate curve relies on the SQD wide-band limit. (The analytic solution may even be unphysical for times t . 1/γ: In certain cases, including Fig. 9, one may find τ eff (0) > 1.) If one, however, neglects the initial slip (81) one ob- tains a curve similar to the blue one in Fig. 9 but with zero vertical offset for time t = 0. This clearly leads to a nonnegligible deviation from the full solution for ini- tial conditions with nonstationary sensor. After a time t ∼ 1/γ, the qubit phase is advanced by Ω/γ which can be of the same order as the phase angle ∼ λ/γ of the initial slip (81) (depending on the relation λ, Ω, within the restriction λ, Ω ≪ Γ). We stress that this slip leads to a cumulative effect: Even at long times t ≫ 1/γ the approximate solution without slip remains offset relative to the full solution. Altogether, this shows clearly that one cannot get rid of the detector completely – even though we describe the qubit state using only the Bloch vector τ (t). It is difficult to eliminate the slip by a choice of the initial state of the coupled qubit-sensor system as mentioned (see Sec. III E) and further discussed in Appendix C. Im- portantly, one should note that it may not be possible to remove the initial-slip backaction by any qubit prepara- tion (i.e., just its reduced density operator τ ). For quan- tum error correction, it is thus important to model the failure of the preparation not only of the qubit dynamical state, but also the dynamical state of the sensor QD and their mutual correlations. D. Distortion of isospin mode vectors The rough mean-field picture also breaks down when accounting for the backaction effect on the isospin modes vectors: the eigenvectors eeff,α are modified from Eq. (95) to Eqs. (91)–(93) when taking into account the finite decoherence rate λ2/γ. This leads to both a tilting of the qubit axis and elliptical isospin precession. FIG. 8: Time evolution of the isospin vector components (a) τx, (b) τy, and (c) τz for times t ≫ 1/γ. The blue curves show the analytic solution τ ana given by Eq. (79), whereas the green curves show 100 times the error with the τ obtained from the solution of the full kinetic equations (26). The coordinate system is chosen as (ex, ey, ez) = ( Ω, λ × Ω, λ). The initial state of sensor plus qubit is p0(0) = 0, τ 0(0) = 0, p1(0) = 1, and τ 1(0) = Ω. The remaining parameters are Γs = Γd = ¯Γ = 10−3T , λ = Ω = 0.1¯Γ = 10−4T , Vb = Vg = 2.5T , and W = 1000T . For these parameters, we find γ ≈ 3.75 · 10−3T and Ω ≈ 1.37 · 10−4T and therefore 1/γ ≈ (2π/ Ω)/200 ≪ t ≪ γ2/∆3 ≈ 300(2π/ Ω) is well fulfilled for the times shown above. is higher as discussed below Eq. (62) and therefore the exponentially decaying envelope of the isospin evolution agrees with much larger accuracy up to longer times. The rough mean-field picture introduced above com- plies with the physical picture developed in prior works [26–28, 33]. It also forms the basis of a simple classi- cal understanding of the qubit decoherence in fluctuator models, to which we compare our results in Sec. V A. However, there are important corrections to this picture even in the high-temperature limit, which we next dis- cuss. C. Effect of initial slip A first illustration of the corrections to the mean- field picture is the effect of the slippage of the initial condition, Eq. (81). To illustrate this effect, we com- pare in Fig. 9 the solution for the isospin for two dif- ferent initial states. We start from a factorizable ini- tial state ρQS = ρQ ⊗ ρS with a fixed total isospin τ (0) = Ω along the "bare" internal field [determining ρQ = (cid:0)1Q + τ (0) · τ(cid:1) /2] while changing the initial con- dition for the SQD through the sensor charge equal to p1(0) [determining ρS = (1 − p1(0)) P 0 + p1(0) P 1]. We show the outcome for τ ana for the two cases of an ini- tially empty SQD (p1(0) = 0) and a SQD hosting an electron (p1(0) = 1). Figure 9(a) exhibits a phase shift between the two isospin evolutions that persists over an entire qubit cycle (and in fact for all future times, which are not shown here). y tionally tilts the relaxation mode vector from ek → e0,eff through the virtual-transition terms ∝ λ2/γ. 26 FIG. 10: Distortion of isospin modes: (a) Comparison of the components of the high-temperature approximation (79) for the isospin τ (t) along the mean-field qubit axis, ek = Ω/ Ω, (black), along the tilted relaxation mode vector eeff,0 (red), and along its dual eeff,0, (blue) as a function of time. Plotted are ln(cid:0)ek · τ(cid:1) + t/T1 (black), ln (eeff,0 · τ ) + t/T1 (red), and ln ((eeff,0/ eeff,0) · τ ) + t/T1 (blue), respectively. The initial state is given by p1(0) = 1 − p0(0) = 0, τ (0) = τ 0(0) = Ω and all other parameters as in Fig. 9. Note that the decay of the precessional component is hardly visible here because T2 ≈ T1 for the parameters chosen here and because of the short time window shown in this figure (see Fig. 8) (b) Sketch of the orientation of ek, eeff,1, and eeff,1 relative to the eeff,1- eeff,2 plane. The thick green line indicates the precession of the isospin in a plane parallel eeff,1-eeff,2 plane and shifted along eeff,0. The precessing part of the total isospin leads to an oscillation along the components along eeff,0 (as indicated) and also along e. Only the component along eeff,0 does not oscillate. However, the component of τ (t) along the tilted relax- ation mode vector eeff,0 shows also an oscillation (super- posed on an exponentially decaying contribution) as the red curve in Fig. 11(a) illustrates. This is an effect of the mode distortion: the plane spanned by the unit vectors eeff,1 and eeff,2 is not anymore orthogonal to eeff,0. Since the precessing part of the isospin lies in this plane, the projection of the isospin τ (t) on eeff,0 becomes oscillatory as sketched in Fig. 10(b). These oscillations are there- fore damped with the dephasing rate 1/T2, which differs from the relaxation rate 1/T1 that sets the time scale for the exponential decay of the nonoscillatory contribution. Thus, the nonorthogonality of the isospin mode vectors mixes relaxation and dephasing in a nontrivial way. The only component of τ (t) that does not oscillate for an arbitrary initial state is the one along eeff,0 as the blue curve in Fig. 10(a) shows. The reason is that the dual vector eeff,0 ∝ eeff,1 × eeff,2 is normal to the eeff,1 - eeff,2 plane [see definition (94)]. This normal component is also independent of the dephasing time: it simply decays exponentially with the relaxation rate 1/T1. However, the dual vector eeff,0 should not be con- If we initially have fused with the zero mode eeff,0: y y y y FIG. 9: Effect of initial slip: time evolution of the isospin com- ponent τ ana computed from the analytical expression (79). (a) Component τ ana for a full qubit cycle. We assume a factorizable initial state of SQD plus qubit factorizes by taking τ n(0) = pn(0) Ω for n = 0, 1. Both solutions show a phase shift with respect to each other that does not die out, i.e., it persists even over many qubit cycles. In inset (b), we compare τ ana (blue) with the component τy com- puted from the full kinetic equations (26) (red). The high- temperature approximation τ ana approaches the full solution τy on the time scale ∼ 1/γ during which the SQD approaches stationarity. The inset (c) shows τ ana after one qubit cy- cle for the two initial conditions and illustrates that the off- set of both curves has not changed appreciably as compared to the evolution close to t = 0. The coordinate system is chosen as (ex, ey, ez) = ( Ω, λ × Ω, λ). The parameters are Γs = Γd = ¯Γ = 10−3T , λ = Ω = 0.1¯Γ = 10−4T , Vb = 3T , Vg = T , W = 1000T , resulting in γ = 2γ0 + γ1 ≈ 3.30 · 10−3T [see Eq. (27)] and Ω ≈ 1.27 · 10−4T [see Eq. (83)]. y 1. Tilting of precession axis As a first consequence, the effective precession axis eeff,0 acquires an additional tilting beyond the mean-field effect. This manifests as a nonzero component of eeff,0 along ek × e⊥ = Ω × λ, which is perpendicular to both the intrinsic qubit precession axis Ω and the measure- ment vector λ and therefore perpendicular to Ω. By virtue of Eq. (91) this rotates the qubit axis relative to ek by an angle χ ≈ r λ2 γ Ω Ωλp1 st Ω2 (97) plus higher-order corrections. This tilt becomes notice- able close to resonance, where detection is performed, as we illustrate in Fig. 10(a), where we plot the projection of τ (t) onto the mean-field axis ek = Ω/ Ω. In addition to an exponential decay with the relaxation rate 1/T1, this component acquires an additional oscillatory component for a general initial state. This simply indicates that we are looking at the component along a vector that is not the zero eigenmode of the qubit: the backaction addi- τ eff (0) ∝ eeff,0, then the isospin still has a precessional component since eeff,0 is not the relaxation mode vector. If one aims to prepare the qubit in a state whose Bloch vector direction is stable under the time evolution that includes the measurement backaction, one should take τ eff (0) = F e0,eff, (98) where F is suitable real constant. [109] For the initial state (98), we indeed find pure exponential decay to the stationary state τ (∞) = 0: τ (t) = F e−t/T1 eeff,0. (99) The above illustrates that the notion of "exciting a qubit mode" has to be treated with care due to both backaction-induced initial slip and due to mode vector distortion. Due to the initial slip discussed above, one has to prepare the qubit-sensor state [see Eq. (72), Eq. (81), and Appendix C] very carefully in order to achieve the initial condition (98). 2. Elliptical precession The second qualitative consequence of the distortion of the qubit modes due to the finite decoherence rate con- cerns the precessional motion in the eeff,1 – eeff,2 plane with normal eeff,0. The trajectory of the isospin vector τ (t) in this plane is changed from a circle to an ellipse. We illustrate this in Fig. 11(a) for an effective initial isospin τ eff(0) = F eeff,1 lying in this plane (again for a suitable real constant F ). Applying Eq. (79), the evolu- tion of the isospin can be expressed as: τ (t) = e−t/T2 F hcos( Ωt)eeff,1 + sin( Ωt)eeff,2i (100) 4(cid:17) v1 4(cid:17) v2i .(101) = e−t/T2 F hp1 − ǫ/2 cos(cid:16) Ωt + +p1 + ǫ/2 sin(cid:16) Ωt + In the rewritten form, the part in the bracket describes an elliptical motion with linear eccentricity π π ǫ = 2eeff,1 · eeff,2 = 1 ΩT1 = r λ2 γ Ω Ω2 Ω2 , (102) which is maximal near the resonance of the sensor QD [due to r, see Eq. (62)] and proportional to the scale λ2/γ for the decoherence rate relative to the qubit frequency Ω. The precession plane is spanned by two orthonormal vectors, v1,2 = eeff,1 ± eeff,2 √2 ± ǫ , (103) which are at the same time unit vectors along the prin- cipal axes of the precession ellipse as sketched in Fig. 11(b). 27 (a) Magnitude of the high- FIG. 11: Elliptical precession. temperature approximation of the isospin τ (t) (green). We take the initial condition p1(0) = 1−p0(0) = 0, τeff (0) = eeff,1 here such that τ (t) evolves in the eeff,1 - eeff,2 plane. From Eqs. (92) and (93), it is easy to see that eeff,1 has a much larger component along λ than eeff,2, leading to a stronger decay initially (see explanation in the text). The oscillating deviations of the isospin magnitude from the exponential de- phasing, τ eff (0)e−t/T2 (blue), reveal the elliptical precession. All other parameters are as in Fig. 9. Although the effects are weak in our controlled perturbative calculations, they in- dicate qualitatively new features that can be expected to grow for stronger readout couplings. (b) Two-dimensional sketch of the ellipse with principal axis v1 and v2 [see Eq. (103)] de- scribed by the tip of the isospin for initial condition as in (a). The exponential shrinking with rate 1/T2 is not indicated for simplicity. Equation (101) shows that the magnitudes of both prin- cipal axes shrink exponentially with rate 1/T2. The mag- nitude of the isospin [green curve in Fig. 11(a)] thus os- cillates around a pure exponential decay with that rate [blue curve in Fig. 11(a)]. The oscillations are a signature of the elliptical distortion of the precessional motion. If there were no mode distortion, the above picture would be true also for slipped initial states out of the e1,eff – eeff,2 plane. However, due to the mode distor- tion arbitrary initial states must again considered with care because the projection of τ (t) onto the eeff,1 - eeff,2 precession plane has an additional contribution arising from the component of τ along eeff,0 [see Fig. 10(b)]: this causes the center of the ellipse to be shifted away from origin. This shift of the center decays exponentially towards the origin with the relaxation rate 1/T1, which is again different from the decay rate of the precession, which is damped with the dephasing rate 1/T2. Relation to state purity. The discussed elliptical isospin motion reflects the exponentially damped but os- cillatory decrease of the qubit-state purity due to the readout by the sensor QD. This can be roughly under- stood as follows. The dephasing is the strongest when the isospin and the measurement vector λ are perpendic- ular to each other because this corresponds to the charge qubit electron being delocalized between the two sites. Here, small fluctuations in the detuning, induced by the stochastic switching of the SQD, then introduce a strong dephasing. By contrast, no dephasing appears when the isospin and the measurement vector λ are collinear, i.e., when the charge qubit electron is localized in one of the QDs. This interpretation is consonant with the situation in Fig. 11(a), which shows the magnitude of the qubit Bloch vector as a function of time. The qubit Bloch vec- tor has initially a large overlap with λ, leading to a sup- pressed decay in the first quarter of a precession period 2π/ Ω. The oscillatory reduction of the qubit purity was anticipated also in our discussion of the exact relation (18), d dt [τ (t)2] = −2λ · [τ 0(t) × τ 1(t)]. (104) If τ (t) precesses, its charge-specific components τ n(t) also precess and their components along λ change in time, resulting in a nonexponential purity decay [110]. Finally, we note that these oscillations of the purity decay are not an effect of non-Markovian corrections [see Eqs. (71) and (72)] induced by the sensor QD onto the qubit by capacitive interaction λ: they may even be re- produced by modeling the SQD charge as a classical fluc- tuator (see, e.g., Refs. [60, 61]). In fact, to obtain Eqs. (79) and (81), we employ a Markovian approximation with respect to the sensor on the qubit (by setting z = 0, see discussion in Appendix B 3 c). The oscillations of the qubit decay should also not be mistaken for coher- ence revivals, i.e., an increase in the purity of the qubit state. Figure 11(a) clearly shows that the magnitude of the qubit Bloch vector decreases for all times, i.e., the information is permanently transferred to the envi- ronment during the measurement process. However, the rate of information loss is nonmonotonic, which simpler approaches might not predict (see, e.g., Ref. 77). This shows most clearly that our perturbation theory goes be- yond the simple "mean-field" detector picture discussed in Sec. IV B. V. COMPARISON TO OTHER APPROACHES In this final Section we compare our results for the measurement backaction with the results of prior works. One of the central results that we obtained – the strong suppression of the measurement backaction when tun- ing the sensor into the Coulomb blockade regime – is surprising: One expects to underestimate the measure- ment backaction by a too simplified treatment which ig- nores cotunneling noise. We accounted for this cotun- neling noise contribution ("broadening" contribution to the rates) but our result (65) revealed that it is canceled out by the coherent backaction, a renormalization effect. Such a cancellation should of course be viewed very crit- ically and we have carefully traced its origins. It also raises the question, which physical assumptions and ap- proximations may have caused it not to be noticed before. This is discussed here. We first show in Sec. V A that semiclassical approaches can only reproduce the contributions to the decay times 28 induced by the stochastic backaction, but they fail to account for corrections from the coherent backaction. This also quite easily happens within a standard weak- coupling Bloch-Redfield approach aimed at finding a qubit-only description as we discuss in Sec. V B. Both these approaches are correct only under the assumption of the weak-coupling, high-temperature limit, which may experimentally be violated and does not allow the sen- sor QD current to be calculated. Finally, we show that our findings are in accordance with exact quantum treat- ments of models of decoherence due to noninteracting two-level fluctuators in equilibrium, insofar a compari- son is possible. A. Semiclassical stochastic approaches A popular way to study the decoherence a qubit suffers from the coupling to its environment is a semiclassical qubit-fluctuator model [24, 26–28, 62]. In this approach, the qubit is considered subjected to noise generated by a randomly switching two-level system. The basic idea is to replace the occupation number n of the SQD in the interaction Hamiltonian, HI = nλ · τ /2, by a classical random process n → ξ(t)/2 [111]. Applied to our case, the SQD introduces a fluctuating effective magnetic field acting on the qubit [see Eq. (21)]: 1 H eff Q = 2(cid:0)Ω + 1 2 ξ(t)λ(cid:1) · τ . (105) Since our primary interest lies in the interplay of the coherent backaction with the cotunneling-induced back- action, we can take Ω = 0 and drop the spin degree of freedom of the SQD electrons and therefore the effect of the Coulomb interaction in the following. Neither of these assumptions are critical for reproducing these ef- fects (see Appendix D). The measurement then induces a dephasing of superpositions of the qubit states Li and Ri, physically related to the double-QD electron residing either in the left or right QD. This can be characterized by the decay of the off-diagonal elements of the qubit density matrix, the visibility [23, 24, 60]: hτ+(t)i = DhLe−i R t × hRe+i R t 0 dt1H eff 0 dt1H eff Q (t1)Li Q (t1)RiE . (106) Here, the bracket h. . .i denotes the average over many re- alizations of the random process ξ(t) with hξ(t)i = 2p1 st. Splitting the qubit Hamiltonian HQ = hHQi + δHQ according to Eq. (21) into a mean-field part, hHQi = 1 2 p1 4 δξ(t)λ · τ , with δξ(t) = ξ(t) − hξ(t)i and hδξ(t)i = 0, Eq. (106) can be recast as stλ · τ , and a fluctuating part, δHQ = 1 hτ+(t)i = De− iλ 2 R t 0 dt1δξ(t1)E . (107) Hence, the decay of the coherences depends only on the "amplitude" of the fluctuations, while the average of the fluctuations, hξ(t)i, is irrelevant. This mean-field part just induces a constant shift of the qubit energy level, which has been absorbed into the average hHQi, see our discussion of the mean-field picture in Sec. IV B 1. For simplicity, let us first take δξ(t) to be a Gaussian random process. In this case, Eq. (107) can be rewritten as [24] −iλ 2 R t 0 dt1δξ(t1)E = e− λ De resulting in an exponential decay of hτ+(t)i = e−t/T2. The dephasing time, 0 dt2hδξ(t1)δξ(t2)i,(108) 2 8 R t 0 dt1 R t 1 T2 = 1 Tφ = (cid:18) λ 2(cid:19)2 Sδξ(0) 2 , (109) can be related to the noise power spectrum Sδξ(ω) of the Gaussian random process δξ(t) [24], Sδξ(ω) := Z +∞ −∞ dτ e−iωτhδξ(0)δξ(τ )i, (110) where we used that the correlator hδξ(t1)δξ(t2)i only depends on the time difference τ = t2 − t1 because a Gaussian random process is time-translational invariant. We emphasize that the relation (108) holds exactly for a Gaussian process, whose entire statistics is fixed by two- point correlation functions. For a general random process, however, higher-order time correlators may contribute to Eq. (108) [24, 28, 63] and therefore the spectral function is not sufficient to characterize a random process completely. A prominent example is a Poisson process inducing random telegraph noise. Such a process would actually be a better model for a capacitively coupled QD stochastically switching its occupation due to tunneling processes. Random tele- graph noise has also been extensively studied to explain the origin of flicker (1/f ) noise in superconducting qubits [24–26, 28, 62] and Gaussian noise in semiconductor QDs [64, 65]. This noise results from an ensemble of fluctuat- ing background charges, each of which may be compared with our sensor QD, except that they do not carry a cur- rent on average. The non-Gaussian behavior of the Pois- son process comes to light only if a few fluctuators dom- inate the decoherence of the qubit state, which entails a nonexponential decay of the coherences [22, 24, 60, 61] and even coherence revivals have been predicted [60, 61]. The impact of charge fluctuations of an individual QD have also been modeled by classical random telegraph process [24, 66]. In general, δξ switches between two val- ues δξu (upper), and δξl (lower) with different switching rates 1/τl for δξu → δξl, and 1/τu for δξl → δξu. This gives rise to different probabilities pu to find the value δξu, and pl to find the value δξl, respectively. In the weak-coupling limit, λ ≪ γ = 1/τu + 1/τl, it turns out that the dephasing is still exponential and the dephasing rate reads as [112]: 1 T2 ≈ [1 − (pu − pl)2] (λ/2)2 γ . (111) 29 st and pl → p0 Identifying pu → p1 st with the stationary oc- cupation probabilities of the SQD for our case and not- ing (λ · e)2 = 1 for Ω = 0, Eq. (111) reproduces the the dephasing rate (89), which we obtained in the high- temperature limit. st and p1 Yet, in the classical fluctuator model, we have not ex- plicitly assumed high temperatures. Thus, if one naively extends the above result to include cotunneling correc- tions, one gets a faulty result. The cotunneling changes the occupation probabilities p0 st and the switch- ing rate γ, but there is no way in which additional co- herent backaction terms could appear [see the transition factor r, Eq. (62)]. As a consequence, one might overes- timate the qubit dephasing rate at the onset of Coulomb blockade. In addition, the classical approach does also not account for the dissipative backaction, which is less important for the dephasing times as compared to the coherent backaction [113]. The reason why the above classical approach is not able to reproduce the coherent backaction is that the electron number n is a classical variable with a definite value at each instant of time, meaning the SQD is fluctuating be- tween the states 0ih0 and 1ih1, where ni denotes the SQD state with n electrons on it. In this way, it can only produce the stochastic backaction. Quantum coher- ences 0ih1 and 1ih0 involved in virtual processes of the SQD (quantum fluctuations), which play a role dur- ing the tunneling, are disregarded here. However, as we illustrate in Appendix D, these coherences – included in our calculation – are crucial for obtaining the coherent backaction. (Note that despite this, such coherences can not appear in the real quantum state, the relevant den- sity operator ρ [see Eq. (12)] due to charge conservation.) It turns out that the related quantum fluctuations induce an additional "phase kick" while a charge transition in the SQD takes place. This phase shift partially com- pensates the phase shift from the stochastic backaction while the SQD is in a specific charge state. This makes plausible why quite generally one may expect a coherent backaction that mitigates measurement backaction when employing a quantum sensor for the indirect detection of a qubit. Thus, a purely classical understanding of the indirect measurement backaction due to a sensor QD is incomplete. B. Bloch-Redfield approach Another frequently employed method to study deco- herence is the Bloch-Redfield approach [48, 67], also in the context of qubits [26, 27]. In the original presenta- tion [67], this approach is first developed for a semiclassi- cal [114] and after this also for a quantum perturbation. In both cases, one considers the limit of weak coupling between a quantum system and its environment and an additional Born-Markov approximation is made. Impor- tantly, the approach generally predicts an exponential decay into a stationary state irrespective of the statis- tics of the environmental fluctuations. The correspond- ing relaxation matrix of the reduced density matrix is furthermore related to the noise power spectrum of the perturbation. 1. Stochastic backaction in spectral function For our case, the Bloch-Redfield approach confirms the dephasing rate (109) also within a quantum treatment: one simply has to replace in Eq. (110) the classical aver- age hδξ(t)δξ(t + τ )i by the quantum-ensemble average: dτ e−iωτ 4 hδn(t)δn(t + τ )i . (112) Sδ n(ω) := Z +∞ −∞ For an indirect readout model of interest here, one treats the SQD as a noninteracting two-level quantum system in contact with a thermal reservoir. Applying the results of Ref. [25] for a single fluctuator, the spectral function reads [115] Sδ n(ω) = [1 − (p1 st − p0 st)2] 2γ γ2 + ω2 (113) leading to Eq. (111) when inserting Sδ n(ω = 0) into Eq. (110). Hence, also the Bloch-Redfield approach – even accounting for a quantum environment – does not reproduce the coherent backaction and may therefore be applied only in the high-temperature limit to derive de- phasing and relaxation times. 2. Difficulty of capturing dissipative and coherent backaction The reason why the Bloch-Redfield approach fails to account for the dissipative and coherent backaction is not specific to this approach. It is rather a problem that ap- pears in principle for any procedure that tries to treat the sensor in an indirect measurement setup as a given environment without a nonequilibrium dynamics of its own. Such an approach always needs to make some as- sumptions about the sensor density operator, which, as we have seen, does not seem to have a simple structure allowing for an educated guess based on general physical principles. This means that one should simply calculate the joint quantum state of sensor QD and qubit, which is what we have done. It is, however, instructive to further understand the problems encountered when one tries to avoid this by making (too) simple approximations of this joint state. (i) The Bloch-Redfield approach (as well as many other density-operator approaches) applied in the weak- coupling limit involves a factorization assumption [68] for the state of the qubit and its environment (here the sen- sor QD plus the reservoir): ρtot(t) ≈ ρQ(t)⊗ ρSR(t). The factorization assumption ignores the dissipative backac- tion valid only in the high-temperature limit. Otherwise, 30 the dissipative backaction leads to a nonzero stationary qubit Bloch vector that reflects the nonfactorisability of the qubit-environment state even in the stationary limit. By contrast, to find the decoherence rates in the long- time limit t − t0 > 1/Γ, the factorization assumption may still work. (ii) The next critical point is then to find a proper de- scription for the unknown evolution ρSR(t) of the envi- ronment. An assumption frequently made is that the qubit environment is in a stationary state, ρSR(t) = ρSR,st, for example, an equilibrium state. However, sta- tionarity of the entire qubit environment (SQD plus elec- trodes) is actually never reached whenever a measure- ment is performed, for which the electrodes must be held at finite bias to produce a nonzero measurement current. Here one should be careful to note that while the reduced sensor QD density operator may become stationary after some time, this is not true for the joint SQD-electrode state. In our approach, the reduced sensor-qubit system can become stationary after eliminating the electrodes, which are stationary. (iii) Even if we further simplify the problem and as- sume zero bias voltage (e.g., to compute the measure- ment backaction in linear response) and ρSR,eq is station- ary, then it is still difficult to compute the equilibrium state ρSR,eq since the sensor QD is a strongly interact- ing system. Naive assumptions made about ρSR,eq are prone to errors. Consider for instance the approximation ρSR,eq = ρS,st ⊗ ρR,0, where ρS,st denotes the stationary SQD state and ρR,0 is the grand-canonical equilibrium state of the reservoirs. (One may be inclined to make this approximation since weak coupling Γ often implies such a factorization.) If we use this state and average the two- point charge correlator in Eq. (112), this involves only charge-diagonal SQD states 0ih0 and 1ih1 – similar to the semiclassical approach of Sec. V A. What goes wrong here is that ρS,st ⊗ ρR,0 is not the correct equilibrium state if we go beyond the lowest-order Γ approximation in the tunneling. For larger Γ, the hybridization between both systems cannot be neglected any more [68]. During tunneling processes the total system can be in virtual in- termediate states involving sensor QD coherences 0ih1 and 1ih0 and corresponding charge coherences in the electrode. These intermediate virtual states explicitly appear in the calculation of the coherent backaction (see Appendix D). Thus, even the quantum-ensemble averag- ing procedure that brings us from Eq. (112) to Eq. (113) misses the coherent backaction since it relies on a weak- coupling expansion between SQD and electrodes. This explains why for an indirect weak measurement setup, the procedure employed in this paper and also in Refs. [29, 30] seems unavoidable. By integrating out the electrodes first one incorporates their effect on the joint sensor-qubit system. As we have seen, this reveals that the qubit experiences a stochastic, dissipative, as well as a coherent backaction effect. This problem of describing the nonstationary environment is specific to indirect measurement setups and not encountered when the qubit is directly coupled to, e.g., a stationary envi- ronment. This is the case, for example, for a bath of har- monic oscillators as in the spin-boson model [27, 69, 70], where the environment may indeed be described by a sta- tionary equilibrium state. Yet, previous studies for other types of environments also show that the approximations made to integrate out the entire qubit envronment can be too crude, including both a spin-boson model [97] and a driven two-level fluctuator [96]. In particular, Ref. [96] explicitly compares results for different levels of approx- imations showing the breakdown of Markov and secular approximation beyond the weak qubit-environment cou- pling. C. Nonperturbative quantum solutions Our study indicates that renormalization effects, based on quantum coherences between the qubit and its detec- tor, are vital for the description of the measurement back- action even in the weak-coupling limit. We next compare our results to prior studies that treat the qubit decoher- ence arising from single fluctuators coupled to fermionic reservoirs fully quantum-mechanically. Such studies employ various approaches, such as an exact numerical evaluation of the visibility (106) using electrodes of finite size [60], a Heisenberg equation-of- motion technique [22], or a Keldysh path-integral formal- ism [22, 23]. All these approaches are nonperturbative both in the measurement interaction λ and the sensor tunneling Γ. However, in contrast to our model these studies are limited to noninteracting fluctuators (here: the SQD) in equilibrium with a single reservoir. Thus, they cannot access the situation of a (nonequilibrium) signal current through the fluctuator, an essential aspect of the indirect detection that we do consider. Moreover, they only consider qubit energy splittings along the mea- surement vector, in our notation HQ = Ωτz/2, that is, they only study pure dephasing in which the qubit Bloch vector has no precessional motion. This leads to drastic simplifications employed in the derivation of these ap- proaches which limit their applicability. References [60, 61] highlight coherence revivals for the short-time transients in the strong measurement regime λ ≫ Γ. This nonexponential decay reflects the non- Gaussian statistics of the quantum telegraph process. Moreover, these studies also find oscillatory corrections to the dephasing even in the weak-measurement regime λ < Γ in agreement with our work (see Fig. 11). The dependence of the dephasing rate on the level position is not reported and thus cannot be compared. In Ref. [22], a path integral method is used to study the qubit decoherence due to two-level fluctuators. Their dephasing rate also includes terms containing the real part of the digamma function ψ, which determines the renormalization function φ [Eq. (30)]. It appears both in the cotunneling rates [see in Eq. (27)] as well as in the coherent backaction [Eq. (35)]. In contrast to Ref. 31 [22], our expressions depend on the derivative φ′, which is arises from our expansions in λ and Γ. However, a direct comparison is not possible because tangible results of Ref. [22] are actually given only for the sequential tunneling regime where renormalization effects are neglected [116]. Reference [23] also uses a path-integral approach and considers also the case when the QD level ε is tuned strongly away from resonance ε− µ/T ≫ 1. Expanding their result (Eq. (15) of Ref. [23]) for the long-time limit of the dephasing rate in the weak-measurement limit, we obtain [117]: 1 T2 = T 4π λ2Γ2 (ε − µ)4 + O(cid:0)(λ/Γ)4(cid:1) . (114) Thus, the dephasing rate drops algebraically in the fourth power with the level position. This is consistent with our results in the sense that also here the decoherence does not scale as expected from the cotunneling noise. In our case, we were only able to show that the algebraic scaling in 1/(ε−µ) is suppressed at least to the second power (see Sec. III D 1). However, our analysis revealed the physical origin of this behavior by identifying what cancels the expected cotunneling noise contribution, namely the co- herent backaction. Note that the result (114) does not rely on the high-temperature assumption Γ/T ≪ λ/Γ as in our case, which is why their result can be expanded in orders λ/Γ while accounting also for higher-order tun- neling corrections [e.g., O(Γ3/T 2) and higher, see Ap- pendix B 4]. This implies that to compare concretely with their result (114), we would have to include higher- order tunneling terms, which, however, we do not expect to restore power laws with lower exponents. This is chal- lenging for our model since we account for Coulomb in- teractions and nonequilibrium conditions. This is beyond the scope of the present paper. In summary, various aspects of our assumptions and findings seem to be in accordance with previous approx- imate as well as exact quantum-mechanical treatments and shed new light on them. Our study extends these ap- proaches by simultaneously dealing with a nonstationary, nonequilibrium, Coulomb-blockaded sensor QD (fluctua- tor), which is fully quantum-correlated with the qubit (non-factorizing density operator ρ, virtual off-diagonal charge coherences during tunneling). The latter leads to the coherent backaction as an integral part of the total backaction together with stochastic and dissipative back- action, leading to the cancellation of cotunneling noise. VI. SUMMARY AND OUTLOOK We studied an indirect detection setup, in which a charge qubit is capacitively probed (λ) by a sensor quan- tum dot (SQD). The SQD is in turn tunnel coupled (Γ) to electrodes in which the time-dependent conductance is measured. Electrons in the sensor QD occupy a single quantized orbital in which they strongly interact. Kinetic equations. We considered the weak-tunneling, in which weak-measurement limit Γ/T ≪ λ/Γ ≪ 1, quantum fluctuation effects are important on the time scale of the qubit-sensor interaction. We derived a ki- netic equation [Eq. (26)] by integrating out the current- carrying electrodes to obtain an effective theory for the composite qubit-SQD system. This revealed three types of backaction on the qubit: (i) a stochastic backaction due to random fluctuations of the qubit detuning, (ii) a dissipative backaction (coefficient c), the flip side of the modulation of the sensor tunnel current by the qubit, and (iii) a coherent backaction due to the level renormal- ization of the composite qubit-SQD system (coefficient κ). We showed the importance of the effects of single- electron tunneling (SET), as well as its cotunneling broadening and level-shift corrections. We also included the leading non-Markovian correction from the electrodes induced by the tunnel coupling (Γ) to the SQD (linear kernel frequency dependence). Moreover, our approach captures all non-Markovian effects introduced by the sen- sor QD on the qubit subsystem, which are mediated by the capacitive interaction. Suppression of cotunneling-induced backaction. By rewriting the kinetic equation in the basis of quasista- tionary and decay modes (defined by the λ = 0 limit), we found that the interplay of these types of backaction leads to a nontrivial cancellation; whereas the dissipative backaction (c) independently couples these modes, the stochastic backaction and coherent backaction (κ) par- tially cancel. In particular, the change in the stochastic backaction due to cotunneling broadening is canceled by the coherent backaction [Eq. (62)]. The expected alge- braic decay ∝ 1/(ε − µr) of the backaction (determining the decoherence rates) is thus suppressed, implying that the actual power law must at least have a higher expo- nent. Experimentally, this is important since it indicates that a SQD can be switched off by applying a gate volt- age better than expected. By identifying the underlying physics, we suspect this to be a crucial difference with the backaction of sensors with dense level spectra, such as single-electron transistors. Thus, the less-than-expected backaction due to the coherent backaction is beneficial for switching the sensor on / off, provided one takes care to prepare sensor state to avoid initial slip errors (see next paragraph). Initial slip. We derived effective equations for the re- duced qubit density operator [Eq. (71)], which are exact relative to the kinetic equations (26). In particular, we keep all non-Markovian effects induced by the sensor and account for a slip of the initial condition [Eq. (72)]. This slip depends on both the initial sensor QD state and the initial quantum correlations with the sensor QD. It is important for the long-time qubit evolution and thus the sensor QD needs also to be considered as part of the dynamical quantum circuit. The dynamical state of the sensor QD is relevant for qubit error propagation, e.g., the initial slip may introduce errors even for perfectly 32 prepared qubit initial states. This provides a concrete ex- ample for errors usually phenomenologically introduced in quantum-error correction. Such a sensor QD cannot be considered (without further evidence) as a "black box" in a quantum circuit which is merely characterized by static parameters (e.g., relaxation and dephasing times). This is different from the treatment of the macroscopic electrodes coupled to the sensor QD. The electrodes, in which the current measurement is performed, can instead be assumed to be stationary, which eliminates initial-slip effects on qubit-SQD system (provided initial quantum correlations are neglected). Specializing to the high- High-temperature limit. temperature limit Γ/T → 0 and to times larger than the SQD relaxation time 1/Γ, we obtained the qubit evo- lution [Eq. (80)], which neglects non-Markovian effects induced by the sensor QD. We connected dephasing and relaxation times ∝ λ2/Γ with the component of the mea- surement vector λ along the mean-field qubit axis Ω and perpendicular to it, respectively. We demonstrated the importance of the initial slip ∝ λ/Γ [Eq. (81)] even in this simplest limit: the set of initial states of the qubit- SQD system without slip is only a subset of zero mea- sure. Generally, the magnitude of the slip increases the "less factorizable" the qubit-SQD state is and the "more nonstationary" the sensor QD is before the detection is started. Due to the latter, the initial slip should be reck- oned with in particular when the sensor QD is initial- ized in a fixed charge state. This happens, for example, when switching the sensor off by tuning the gate volt- age far away from resonance or by reducing the tunnel coupling Γ to the electrodes. By contrast, switching off the capacitive interaction λ leads to an initially station- ary and factorizable state, which is favorable for avoiding an initial slip. This difference in the backaction between various parameters for switching off a sensor QD is an important experimental implication of our study. The need to control not only the qubit but also the readout device carefully may not only be a nuisance for engineer- ing quantum circuits but could also provide additional means of controlling qubits. For example, one could con- sider to switch the SQD during the readout to another readout point to compensate for manipulation errors de- tected in a weak measurement process. Mode distortion. The analysis of the isospin dynamics showed that additionally the qubit eigenmode vectors are distorted due to the small but finite value of the decoher- ence rates ∼ λ2/Γ. This corrects the simple mean-field picture, in which the qubit axis Ω = Ω + p1 stλ is only influenced by the average occupation of the SQD. The distortion reflects the breakdown of the secular approxi- mation – often made in derivations – because the capac- itive coupling λ can be of the same order as the internal qubit splitting Ω. Importantly, the distortion must be included to satisfy an isospin sum rule that follows from the conservation of the isospin by the tunneling of sen- sor electrons [32, 46]. The distortion is thus enforced by a general principle. The experimentally relevant conse- quence of this distortion for the qubit evolution is two- fold: first, the usual circular Bloch-vector precession be- comes slightly elliptical , with a shape that is not altered in time as the size shrinks with dephasing rate 1/T2. Sec- ond, this "precession plane" is not orthogonal any more to the "relaxation axis," along which the Bloch vector decays with the relaxation rate 1/T1. The relaxation axis is moreover slightly tilted relative to the mean-field axis Ω. Finally, the projections of the isospin on the mean-field axis, the relaxation axis, and the precession plane all show a superposition of relaxation and preces- sional dephasing motion. The measurement backaction thus mixes the effects of relaxation and dephasing, even in this simple Markovian limit. This effect may generally appear in indirect-coupling setups, which are typical for detection setups. The tilting of the qubit modes is related in a broader context to the concept of "gauge qubits" [98]: two phys- ically distinct qubits can be each subject to strong de- coherence but the joint Hilbert space formed by both qubits may contain a two-level subsystem with low deco- herence. Locating such subsystems is interesting for de- veloping strategies for quantum-error correction. In our case, the qubit mode tilting reflects an "admixing" of the sensor QD degrees of freedom to the low-decoherence subspace, which mostly overlap with the qubit degrees of freedom. This mixing can be strongly enhanced when the coupling of the quasistationary and decay modes be- comes stronger, which effectively happens, e.g., due to enhanced quantum-fluctuation effects at lower tempera- tures. Investigating this mixing further would therefore be interesting also in the context of quantum-error cor- rection strategies. Comparison with other approaches. We compared the above results at various points with existing approaches and explained why potential differences are expected within their validity: (a) Semiclassical stochastic approaches cannot capture the coherent backaction because the starting assumption of classical charge fluctuations on the sensor QD already excludes relevant quantum coherences of the qubit-sensor density operator. Including these into the description leads to additional "phase kicks" that counteract the stochastic "phase kicks." Both together, when averaged, lead to a mitigated decoherence. It thus seems that one should quite generally reckon with coherent backaction, which can mitigate the measurement backaction when employing a quantum-dot sensor for the indirect detec- tion of a qubit. (b) Density operator approaches that try to integrate out the sensor together with the attached electrodes run into problems as well because one needs to "guess" the time-dependent, current-carrying sensor state as well as the quantum correlations with the qubit. These can lead to non-Markovian behavior and affect the initial slip of the qubit. All of this is systematically calculated in our approach. The more advanced approach of Ref. [88], which first calculates the nonequilibrium SQD state in 33 the absence of the qubit (an approximation carefully pointed out in Ref. [88]), misses both the dissipative and coherent backaction. Extending such an approach, e.g., to include cotunneling broadening would thus lead to inconsistencies since only one part of two canceling ef- fects is taken into account. (c) Nonperturbative quantum solutions of related mod- els agree with the cancellation in the backaction we found here as does a separate calculation for a noninteracting limit of our model (U = 0). (d) It is also interesting to compare with prior studies not aiming at sensor backaction. In particular, Ref. 78 highlighted the importance of a competition of next- to-leading-order effects and non-Markovian corrections, which are closely related to the non-Markovian correc- tions of the first type affecting the dissipative switching rates (36) discussed in Sec. III B 4. However, what is imporant here are non-Markovian corrections of the sec- ond type affecting the coherent backaction. We stress that although these non-Markovian corrections give an important part of the coherent backaction, they are not identical to it: the coherent backaction already arises in the stationary limit that we studied in Ref. 32 where non- Markovian effects can be neglected. In this way, we see that the general line of thought emphasized in Ref. 78 ex- tends to the coherent backaction, which is also a second- order effect, namely first order both in the tunneling (Γ) and the measurement interaction (λ). Outlook. All this shows that the coherent backaction and other quantum fluctuation effects (cotunneling, level renormalization) are intrinsic effects of a quantum sen- sor: they can neither be "added" or "controlled" indepen- dently in an experiment, nor should they be neglected in a calculation. In view of the above, further studies that can address the experimentally relevant lower temper- ature dynamics of weak measurements using quantum- dot sensors are necessary. We expect the qualitative ef- fects that we identified to be present and quantitatively stronger under experimental conditions where, e.g., the capacitive interaction and tunnel coupling may not be weak anymore (e.g., because of the trade-off between a significant current signal and low backaction). Besides the commonly discussed relaxation and dephasing back- action, also the qubit initial slip, nonorthogonal mode distortion, and sensor-induced memory effects will be en- hanced, neither of which has received much attention so far. Acknowledgments We acknowledge stimulating discussions with H. Bluhm, L. Schreiber, J. Konig, V. Meden, C. Klockner, D. Kennes, and P. Samuelsson. We thank A. Braggio for helpful feedback on the preprint version. We are grateful for support from the Alexander von Humboldt founda- tion. Appendix A: Importance of non-Markovian corrections induced by the electrodes In this Appendix, we explain how to account for non- Markovian corrections arising from the memory induced by the electrodes on the qubit-sensor subsystem and il- lustrate their importance for a correct description of the detector backaction. One should clearly distinguish from this further non-Markovian behavior induced by the SQD – with the effect of the electrodes incorporated – on the qubit subsystem. This is discussed separately in Ap- pendix B 3 c. In Appendix A 1, we show how the non- Markovian corrections due to the electrodes can be in- corporated within the real-time diagrammatic formalism based on a perturbative weak-tunneling (Γ) expansion. Within the leading non-Markovian correction included in this paper, the kinetic equation remains a time-local and first-order differential equation for the density operator. Based on this, we then perform the weak-measurement In Appendix A 2, we show that neglecting expansion. the non-Markovian correction leads to a violation of the positivity of the SQD-qubit density operator and to an overestimation of the measurement backaction. These unphysical features are removed when the leading non- Markovian correction is included. 1. Incorporating non-Markovian effects in real-time diagrammatics In Ref. [32], we started from a kinetic equation for the reduced density operator ρ(t) of the joint system of SQD plus qubit, obtained in the standard way by integrating out the electrodes' degrees of freedom: 0 ρ(t) = −iLQSρ(t) +Z t dt′W (t − t′)ρ(t′). (A.1) Here, LQS• = [HQ+HS+HI ,•] is the internal Liouvillian of the reduced system with "•" denoting the operator the Liouvillian acts on. The effects of the electrodes are incorporated in the kernel W , which we evaluated using the real-time diagrammatic approach [50, 51]. In Ref. [32], we were only interested in the stationary solution of Eq. (A.1), ρ(t) = ρst, which obeys ρst = 0. This solution 0 dt′W (t − 0 dτ W (τ ). In the long-time limit t → ∞, when the stationary state is approached, the time-integrated kernel is given by the zero-frequency limit of the Laplace transform depends only on the time-integrated kernel R t t′) = R t W (z) = Z ∞ 0 dτ eizτ W (τ ). (A.2) Hence, the exact ρst is the same as the stationary so- lution of the approximate, time-local Markovian kinetic equation, dρ dt (t) = (−iLQS + W )ρ(t), (A.3) 34 with W = W (z = i0). Weak-tunneling expansion. Yet, in the present paper, we study the nonstationary time evolution of the qubit- sensor density operator ρ(t) and we show now that this implies that memory effects induced by the electrodes through the kernel W have to be included. We will refer to all the effects of the frequency dependence of W (z), Eq. (A.2), as non-Markovian corrections induced by the electrodes [even when effectively a time-local equation re- sults, see Eq. (A.7)]. An approach to include such cor- rections within the real-time diagrammatic formalism has been given in Ref. [41], and applied to study decay rates in Refs. [71, 72]. The basic idea is to perform a Tay- lor expansion of ρ(t′) in the integral in Eq. (A.1) around time t, ρ(t′) = 1 n! dnρ(t) dtn (t′ − t)n, ∞ Xn=0 (A.4) and to subsequently perform the integration over t′. This results in a well-defined expression if the kernel decays faster than any polynomial in t − t′, which is usually fulfilled because the kernel decays exponentially W (t − t′) ∼ e−(t−t′)/τC on the time scale of the inverse temperature τC ∼ 1/T [118]. If we consider only times t ≫ τc, one may compute 0 dt′W (t − t′)(t′ − t)n = R t R t 0 dτ W (τ )(−τ )n by replacing t → ∞ on the right-hand side since all contributions from τ > t are negligibly small. Taking advantage of the Laplace transform (A.2), we obtain from Eq. (A.1) dρ(t) dt = −iLQSρ(t) +Xn 1 n! ∂nW dnρ(t) dtn , (A.5) with the nth derivative of the Laplace-transformed kernel (A.2) at zero frequency with respect to −iz: ∂nW = . (A.6) ∂nW (z) ∂(−iz)n(cid:12)(cid:12)(cid:12)(cid:12)z=i0 To include the leading-order non-Markovian corrections, we truncate the sum on the right-hand side of Eq. (A.5) and keep only the terms with n 6 1. This yields the following time-local differential equation of first order in time: dρ(t) dt = 1 1 − ∂W (−iLQS + W )ρ(t) = −iLρ(t),(A.7) whose solution can be written as ρ(t) = e−iLtρ(0) (A.8) for initial state ρ(0). When dropping the non-Markovian correction 1/(1 − ∂W ) in Eq. (A.7), one recovers the Markovian generator of Eq. (A.3). We next show that this solution with L defined through Eq. (A.7) accounts for all non-Markovian effects up to O(Γ2/T ) provided one is seeking for exponential solu- tions of Eq. (A.5). We do not discuss algebraic or loga- rithmic time dependencies here, which may also appear [73]. Thus, let us substitute the exponential ansatz ρ(t) = eAtρ(0), (A.9) into Eq. (A.5). We obtain the following equation for A: A = −iLQS + W + (∂1W )A + . . . . (A.10) When all derivatives ∂nW are dropped from the above equation, we recover the Markovian generator A = −iLQS + W . Non-Markovian corrections therefore en- ter through terms ∼ ∂nW that capture the frequency dependence of the kernel [78–80]. Equation (A.9) shows that the derivatives of the den- sity operator scale as dnρ/dtn ∼ An ∼ Γn since LQS ∼ ∆ ≪ Γ and W ∼ Γ here. As a consequence, the expansion (A.4) of the density operator in corrections from higher-order time derivatives is not independent of the perturbative expansion of the kernel W in pow- ers of Γ. When expanding A in powers of Γ, we find that the nth order derivative of the kernel scales as ∂nW An ∼ (Γ/T n)·Γn plus higher-order corrections [119]. Thus, when expanding the kernel W = W Γ + . . . only up to first order in Γ, Markovian corrections must be i.e., AΓ = −iL + W Γ. Yet, when expand- ignored, ing up to order Γ2/T as considered here, one is al- lowed to omit terms (∂nW )An ∼ (Γ/T n) · Γn only for n > 2 from Eq. (A.10), while one has to keep the first derivative ∂1W . Solving for A, we obtain A = −iL = (1 − ∂1W )−1(−iL + W ) + O(Γ3/T 2) in agreement with Eq. (A.7). Further expanding the inverse and the kernel, we find for the effective generator: − iL = (1 + ∂1W Γ)(−iLQS + W Γ) + W Γ2/T +(Γ3/T 2). (A.11) Weak-measurement expansion. The next step is the ex- pansion of −iL in ∆/T , where ∆ ∼ λ, Ω denotes the small energy scale of the detection and intrinsic qubit frequency. While we have to keep the first-order terms in ∆/T for W Γ = W Γ,0 + W Γ,∆ + O(Γ∆2/T 2), (A.12) we can neglect the ∆ dependence of ∂1W Γ = ∂W Γ,0 + O(Γ∆/T ), W Γ2/T = W Γ2/T,0 + O(Γ2∆/T 2), (A.13) (A.14) and obtain for the effective, non-Markovian generator − iL = (1 + ∂1W Γ,0)(−iLQS + W Γ,0) +W Γ,∆ + W Γ2/T,0 +O(Γ3/T 2, Γ2∆/T 2, Γ∆2/T 2). (A.15) We now note that in Ref. [74], we already computed the kernels W Γ,0, W Γ,∆, and W Γ2,0 and therefore we have 35 to merely evaluate ∂W Γ,0 to obtain non-Markovian cor- rections up to the order considered here. The frequency derivative of ∂W Γ,0 [see Eq. (A.6)] can be easily con- verted into energy derivatives (∂/∂ǫ) [120], which results in the kinetic equation (26) in the main text. It should be noted that ∂W Γ,0 is not simply the derivative of the SET contribution: the imaginary factor in Eq. (A.6) changes the role of imaginary and real parts, which are related to δ functions and principal-value parts in the frequency integrals, respectively. While in the SET contributions only the δ functions remain and the principal-value parts cancel out, this is opposite for the non-Markovian cor- rection ∂W Γ,0. These principal-value parts evaluate to the renormalization function φ(x) [see Eq. (30)], which is of central importance in our work and explains how the non-Markovian corrections can affect the coherent back- action as noted in Sec. III B 4. To evaluate ∂W Γ,0, one thus must first compute W Γ,0, then apply ∂ and only af- ter that take the relevant matrix elements (restricted by charge conservation). This completes the derivation of our non-Markovian Liouvillian, accounting consistently for all terms up to O(∆, Γ, Γ2/T, Γ∆/T ). To find the solution (A.9) of the kinetic equation ρ(t) = Aρ(t), we directly solve for ρ(t) without ex- panding ρ(t) in Γ/T . Not solving the kinetic equation order-by-order [121] avoids well-known problems with ill- defined coherences [75] and nonequilibrium occupations close to the Coulomb blockade regime [51]. As a con- sequence, our solution can comprise of terms of higher order in ∆/T and Γ/T . In particular, the stationary solution obtained from Aρst = 0 differs formally from that obtained by solving the corresponding Markovian equation A∂W =0ρst = 0. However, the deviations are of O(Γ3/T 2, Γ2∆/T 2, Γ∆2/T 2) and therefore consistently negligible in the perturbative limit considered here. 2. Retaining positivity and effect on coupling of quasistationary and decay modes In this appendix, we elucidate the importance of non- Markovian corrections induced by the electrodes to de- scribe the indirect detection setup. We explicitly illus- trate that without these corrections the kinetic equations (26) for the SQD-qubit system possess exponentially in- creasing solutions in time and the measurement backac- tion is qualitatively different. In Ref. [32], we reported that the Markovian kinetic equation when used to calculate a time-dependent solu- tion for the density operator ρ(t) for the joint SQD-qubit system violate the positivity condition, even though the stationary state showed no such problems. (It should be noted that for the stationary state to be positive, in addition to the dissipative and coherent backaction also O(Γ2/T ) effects were required, already hinting at the can- cellation effect that we discuss in the present paper.) The 36 full equation (A.15) instead, all eigenvalues have a non- positive real part for all gate voltages as shown in Fig. 12(b). Thus, no positivity violation can occur here. We investigated this thoroughly by numerically exploring a large parameter regime. We note, as already pointed out in Ref. [32], that the inclusion of next-to-leading order Γ2/T corrections in the tunneling is also crucial to avoid such positivity problems. To assess the importance of non-Markovian corrections for the qubit backaction, we next consider the effective Liouvillian (69) when starting from the Markovian ap- proximation of the generator (A.16). As we discussed in Sec. III C 3, the importance of the coherent backac- tion can be assessed from the transition matrix from the quasistationary into the decay modes. In both cases, the transition matrix takes the form Λdq = (cid:18) (1 + p0 0 (1 + p0 st)e†α · (cλ) −e†α · (r × eα′ ) (cid:19) st)cλ · eα′ (A.17) where in the Markovian approximation the vector r reads rM = κ(1 + p0 stp1 st + κ 1 2 (1 + p0 st)Ω +(cid:2)p0 and when including the leading non-Markovian correc- tion [see Eq. (61)], the vector r reads st)(cid:3) λ (A.18) FIG. 12: (a) and (b): Two largest real parts b0, b1 of the eigenvalues of the SQD-qubit generator −iL, as a function of gate voltage Vg. We use the Markovian approximation (A.16) of the generator −iL∂W =0 in (a) and the full non- Markovian generator (A.15) in (b). (c) and (d): Component of r [Eq. (A.17)] along λ. We show the projection rM = rM · λ = p0 st) [see Eq. (A.18)] within the Markovian approximation in (c) and rNM = rNM · λ = p0 st/2 − p0 st) [see Eq. (A.19)] including non-Markovian corrections in (d). In (c) and (d), the blue curves include the coherent backaction (κ 6= 0), while the green curves exclude them by hand (κ = 0). The parameters in all plots are Vb = 0, Γs = Γd = ¯Γ = 10−2T , λ = Ω = 0.1¯Γ = 10−3T , and W = 1000T . 2 (1 + p0 st − κ(p1 st + κ 1 stp1 stp1 positivity problems for time-dependent solutions arising in the Markovian approximation with respect to the elec- trodes can be readily inferred by diagonalizing the gen- erator (A.15) in that approximation: − iL∂W =0• = Xi (bi + iai)Ai tr S,Q [ Ai•]. (A.16) Positivity violation occurs when at least one eigenvalue exists with a positive real part, i.e., bi > 0. This leads to exponentially growing contributions ρ(t) ∼ ρst + ebitρi [32]. Figure 12(a) shows the two largest real parts of the eigenvalues of the Markovian generator (A.16) for typical parameters considered in this paper. It illustrates that a positivity violation may appear when the SQD is tuned deep into the Coulomb blockade regime for Vg/T ≫ 1 [see Fig. if we in- clude the leading non-Markovian correction and use our [32]]. By contrast, 7(b) in Ref. (A.19) stp1 rNM = (cid:2)p0 2 p1 st − p0 st − κ(cid:0) 1 st(cid:1)(cid:3) λ, which is the central result (62) discussed in detail in the main text. The non-Markovian corrections lead to two important differences between Eqs. (A.18) and (A.19). First, the coherent backaction leads to a different gate- voltage dependence of the transition matrix when non- Markovian corrections are neglected. To illustrate this, we plot in Figs. 12(c) and 12(d) the component of rM and rNM along the measurement vector λ. Due to the cancellation effect explained in the main text, the coherent backaction suppresses the transition factor in the non-Markovian case, leading to an exponential gate- voltage dependence. This is different for the Markovian case; here, the term arising from the coherent backac- tion, κ(1 + p0 st), changes sign at resonance and enhances the transition rate into the decay modes for positive gate voltages [see Fig. 12(d)]. Moreover, the transition factor scales there algebraically with Vg in stark contrast to the exponential dependence in the non-Markovian case. The second difference between Eqs. (A.18) and (A.19) is that the vector rM in the Markovian case also has a component along Ω. This component emerges because in the Markovian approximation the coherent backaction appears as torque terms in the kinetic equations that are proportional to κ (Ω + λ/2) [see Eqs. (25) and (26) in Ref. [32]]. By contrast, in the non-Markovian case the torque terms involve only the vector λ [see Eq. (26)]. Both these differences clearly show that non- Markovian corrections induced by the electrodes have crucial consequences for the total backaction due to the capacitive interaction λ and must be accounted for con- sistently. Appendix B: Liouvillian perturbation theory a. Liouville-space projection technique 37 In this appendix, we derive the effective Liouvillian (69) that we investigate in the main text to study the measurement backaction on the qubit evolution. We first keep our formulation as general as possible to bring out the generic features and to indicate that this procedure can be applied also to more complicated indirect detec- tion models. An important prerequisite is that one can identify a subspace of interest in Liouville space whose dynamics takes place on a time scale that is well sepa- rated from the dynamics in the complementary subspace. In our situation, this is related to the slow dynamics ∼ 1/∆ of the subspace of the quasistationary modes as compared to the fast dynamics ∼ 1/γ in the subspace of the decay modes. Using the well-established projection technique [56, 57] along the lines of Chapter 17 in Ref. [55], we outline how to obtain an effective Liouvillian that mediates the dynamics in the subspace of interest only. In contrast to most cases found in the literature, the unperturbed problem onto which we project already exhibits non-Hamiltonian dynamics, i.e., L• 6= [H,•], see discussion after Eq. (B.37). Therefore, we review these steps here to highlight that the initial slip is a general phenomenon that appears when projecting out a comple- mentary subspace. This in fact prevents a full elimina- tion of these degrees of freedom unless special conditions apply. In the high-temperature limit, the projection can be analytically performed and we can obtain the effective Liouvillian and initial slip perturbatively in the coupling strength between the relevant subspace and its comple- ment. We give here two complementary approaches. The first one is given in frequency space in Appendix B 1, which allows for a very compact procedure. However, to un- derstand the physical meaning of the involved approx- imations, we also show how to obtain our results in a time-space formulation in Appendix B 2. In Appendix B 3, we then apply our general Liouville-space projection technique to the detection setup considered in the main text. We finally derive the effective transient evolution of the qubit, Eqs. (79)–(81), by removing the SQD degrees of freedom as far as possible. 1. Derivation in frequency space Consider a system with density operator ρ whose time evolution is generated by a Liouvillian L, i.e., ρ(t) = −iLρ(t), (B.1) with initial condition ρ(0) given at time t = 0. We refer to L as the Liouvillian although it might not have a sim- ple commutator structure and can have dissipative, non- Hermitian parts. To solve the above equation, one can transform it to Laplace space. The Laplace transform of 0 dteiztf (t). This yields using Eq. (B.1): a function f (t) is defined as f (z) = R ∞ − izρ(z) − ρ(0) = −iLρ(z). (B.2) Next, we are only interested in the evolution of ρ in a subspace a of the entire Liouville space, defined by a pro- jection superoperator P a, which satisfies (P a)2 = P a. In contrast to Sec. III C, our formulation here is completely basis independent to emphasize the generality. Projecting the kinetic equation (B.2) onto the subspace a and its complement b with projector P b = 1 − P a, we find − izρa(z) − ρa(0) = −iLaaρa(z) − iLabρb(z),(B.3) −izρb(z) − ρb(0) = −iLbaρa(z) − iLbbρb(z).(B.4) Solving the second equation for ρb(z), ρb(z) = i z − Lbb (ρb(0) − iLbaρa(z)), (B.5) and inserting this into the first equation, we can formally write the exact solution as ρa(z) = i z − Laa eff(z) ρa eff (z). (B.6) This incorporates a frequency-dependent effective Liou- villian, Laa eff(z) = Laa + Lab 1 z − Lbb Lba, and the frequency-dependent initial condition eff(z) = ρa(0) + Lab ρa 1 z − Lbb ρb(0). (B.7) (B.8) To transform Eq. (B.6) back to time space we apply the inverse Laplace transform: ρa(t) = Z +∞ −∞ dz 2π e−iztρa(z). (B.9) We start with the derivation in frequency space by ap- plying a general projection technique. This results in the most general expression for the time evolution, for which we give a perturbative expansion subsequently. Identifying the subspace a (b) with the subspace of the quasistationary (decay) modes labelled by q (d), Eqs. (B.6) – (B.8) yield Eqs. (68) – (70) of the main part. To compute the solution in time space, the integral (B.9) has to be solved by applying the residue theorem. Since Laa eff(z) can be represented by a finite matrix in our case, we only have isolated poles satisfying zp = Laa eff(zp). If the coupling between the subspaces a and b is absent, these poles are given by the eigenvalues of Laa. Importantly, Eq. (B.8) shows that the component of the initial state in the complementary b-subspace has been transformed into a correction to the initial state in the targeted a subspace. This is referred to as a slippage of the initial condition [90]. In general, if the coupling of the b-subspace to the a-subspace is nonzero (Lab 6= 0), this initial slip is present unless the initial state has no projection on the complementary space, ρb(0) = P bρ(0) = 0. In our problem, P b projects onto density operators ρ, which are either not factorizable or for which the reduced state of the SQD is not station- ary (see Appendix C). This has to be compared with the generalized master equation (24) for the joint qubit-SQD state: As noted there, when deriving the general kinetic equation one in fact projects onto a stationary (equilib- rium) state of the reservoirs. One then often assumes (as we do) an initially stationary reservoir state that factor- izes with the SQD-qubit state. This eliminates the initial slip. b. Perturbative solution To next find a perturbative solution – required to ob- tain the high-temperature equations (80) and (81) of the main text – we assume that the eigenvalues of Laa are well separated from those of Lbb in the complex plane as compared to the coupling mediated by Lab and Lba. Thus, if l := max(Lab,Lba) ≪ g := Lbb − Laa, (B.10) we can neglect the coupling-induced shift of the poles zp in the denominator of Eq. (B.7) to lowest order in l/g. This is the basis of our perturbative expansion of Laa eff in orders of l/g. For the measurement setup studied in this paper, the situation is even a bit simpler because the eigenvalues in the quasistationary subspace satisfy ω := Lqq ≪ g, (B.11) which allows us to insert z = 0 into Eqs. (B.7) and (B.8). This means that we carry out a Markovian approxima- tion (i) for the effective Liouvillian Laa eff = Laa eff(z = 0) = Laa − Lab 1 Lbb Lba (B.12) and (ii) for the effective initial condition: ρa eff (0) = ρa(0) − Lab 1 Lbb ρb(0). (B.13) One can next readily transform back to time space, which yields the effective time-evolution equation, ρa(t) = −iLaa effρa(t), (B.14) 38 with the "slipped initial condition" (B.13). Equations (B.12) – (B.14) reproduce Eqs. (79) – (81) of the main text. We emphasize that this Markov approximation is pro- foundly different from the Markov approximation dis- cussed in Appendix A. The latter accounts for memory effects of the electrodes (kernel frequency dependence), which are already contained in L and therefore included into the projection approach from the start. In the present case we neglected the frequency dependence in Eqs. (B.8) and (B.7) that is generated by considering a subsystem of the system described by L. We address this point further in Appendix B 3 where we discuss this for our concrete detection problem. eff tρa The Markovian approximation (B.14) neglects correc- tions to the effective Liouvillian of O([max(l, ω)]3/γ2) (B.14), ρa(t) = and therefore the solution of Eq. e−iLaa eff (0), is limited to times t ≪ γ2/[max(l, ω)]3. Moreover, there is a restriction for Eq. (B.14) to be valid for small times: one has to require t > 1/γ basically to ensure that the projection of the density operator on the decaying subspace, ρb(t), becomes negligibly small and effectively contributes only through the slipped ini- tial condition Eq. (B.13). This becomes clearer from our complementary time-space derivation below. 2. Derivation in time space To gain further physical insight into the approxima- tions made in the above derivation, we re-derive the above results now in time space. We start here from decomposing Eq. (B.1) into its components in the two complementary subspaces a and b: ρa(t) = −iLaaρa(t) − iLabρb(t), ρb(t) = −iLbaρa(t) − iLbbρb(t). The second equation is formally solved by (B.15) (B.16) ρb(t) = e−iLbbtρb(0) −iZ t 0 dt′e−iLbb(t−t′)Lbaρa(t′), (B.17) which inserted into the first equation yields the integro- differential equation ρa(t) = −iLaaρa(t) −LabZ t −iLabe−iLbbtρb(0). 0 dt′e−iLbb(t−t′)Lbaρa(t′) (B.18) This equation incorporates three terms: The first line represents the internal evolution in subspace a, which re- produces the full time evolution of ρa if subspaces a and b are decoupled (Lab = Lba = 0). However, if the coupling is nonzero, the second line accounts for virtual transitions from the subspace a into subspace b at time t′ < t, fol- lowed by a period of internal evolution in the subspace b (mediated by Liouvillian Lbb) and a final transition back into subspace a again at time t. The third line is related to the initial slip: it accounts for the contribution to ρa(t) that stems from the initial projection of the density op- erator ρb(0) on subspace b, combined with a transition into subspace a at time t. To next derive the effective Liouvillian (B.12) and ini- tial condition (B.13) from Eq. (B.18) for the measure- ment setup under study, we make use of the separation of time scales governing the dynamics: This allows us to ne- glect the time evolution in the quasistationary subspace in Eq. (B.18), i.e., we perform a Markov approximation with respect to the memory induced by the decaying sub- space (identifying now a = q and b = d), ρq(t′) ≈ ρq(t). (B.19) Inserting Eq. (B.19) into Eq. (B.18) yields ρq(t) = −i"Lqq eff + Lqd e−iLddt Ldd Ldq# ρq(t) −iLqde−iLddtρd(0), (B.20) with the effective Liouvillian eff = Lqq − Lqd 1 Lqq Ldd Lqd + O(cid:16) ωl2 g2 , l3 g2(cid:17) . (B.21) The higher-order terms in Eq. (B.21) are due to non- Markovian corrections ∆ρq(t′) = ρq(t′) − ρq(t) to Eq. (B.19). They can be estimated as follows: Since the exponential in the integral Eq. (B.18) decays on a time scale 1/g, we only need to account for corrections ∆ρq(t′) for times t′ satisfying t−t′ . 1/g. Integrating Eq. (B.18), this yields corrections ∆ρq ∼ ω/g, l/g. Multiplied with the order l2/g of the corrections from the decaying sub- space, we obtain the estimate of the higher-order terms in Eq. (B.21) in accordance with the result obtained in frequency space. The time-local, Eq. (B.20) is not yet fully Markovian in the sense that it still contains explicitly time-dependent terms in the time-evolution generator, which therefore becomes frequency dependent in Laplace space [122]. In the long-time limit t ≫ 1/g this time dependence drops out: As we now argue, one can omit the second term in the bracket in Eq. (B.20) while the second line of Eq. (B.20) must not be neglected. One may drop the first term since it gives a correction to the derivative ∼ l2/g on a time scale 1/g, i.e., they result in an accumulated correction ∆ρq(t) on the order of ∼ l2/g2, which can be neglected. This is usually achieved in standard deriva- tions of master equations by setting t → ∞ in the inte- gral in Eq. (B.18) [48]. By contrast, the corrections from the second line of Eq. (B.20) are of lower order l/g as in- tegrating Eq. (B.20) shows. In many cases, these terms do not appear as one often assumes ρd(0) = 0 from the start. Here these terms must be kept and the solution of Eq. (B.20) can therefore be approximated as 39 ρq(t) = e−iLqq eff th ρq(0) dt′e+iLqq eff t′ 0 −iZ t +O(cid:16) l2 g2(cid:17) . Lqde−iLddt′ ρd(0)(cid:21) (B.22) Again, one can exploit that the exponentials in the sec- ond line of Eq. (B.22) vary on a different time scale: While the factor e−iLddt′ is nonzero only on a short time scale ∼ 1/g, the factor e+iLqq changes on a much longer time scale max(ω, l2/g), and we may therefore expand e+iLqq ≈ 1 + O(ω/g, l2/g2) in Eq. (B.22). In the long- time limit t → ∞, we can then approximate the integral well by setting its upper bound t → ∞, resulting in eff t′ eff t′ ρq(t) = e−iLqq eff tρq eff (0) + O(cid:16) ωl g2(cid:17) , g2 , l2 (B.23) with the slipped initial state eff(0) = ρq(0) − Lqd 1 ρq Ldd ρd(0). (B.24) We have thus arrived again at Eq. (B.12) and (B.13), re- spectively, particularly emphasizing that Eq. (B.23) and (B.24) describe the time evolution correctly as long as 1/g ≪ t ≪ g2/[max(l, ω)]3; otherwise, the correction terms to the effective Liouvillian (B.21) can accumulate to a large error in Eq. (B.23). This is also borne out by numerical checks that we performed. 3. Effective Liouvillian for indirect detection We next apply the above Liouville-space projection technique to the indirect detection setup of the SQD- qubit system studied in the Sec. III and IV of the main text. In Appendix B 3 a we first make the connection between the projections just discussed and the dynami- cal variables considered in the main text. After this, we provide in Appendix B 3 b some important intermediate steps in the derivation of the effective Liouvillian (80). We comment in Appendix B 3 c on the Markov approxi- mation with respect to the memory induced by the sensor QD on the qubit. Finally, we discuss the validity of our effective Liouvillian in view of the perturbative expansion of the kinetic equations in Appendix B 4 and comment on the U = 0 limit of our problem. a. Definition of projections and representing matrices of the Liouvillian To make a connection between the projection ρq = P qρ of the SQD-qubit density operator on the quasistation- ary subspace and the dynamical variables introduced in Sec. III C 2, we exploit the eigenbasis of the Liouvillian L0 [Eq. (37)] with left and right eigenvectors V k p and V k p , respectively. In the following, " =" denotes that we rep- resent a basis-independent object on the left hand side by its components on the right hand side with respect to the eigenbasis of L0. Representing, for example, the density operator ρ in this basis, we obtain ρ = (cid:18) Xq Xd (cid:19) =   1 τ 0 α + τ 1 α stp0 − p0 p1 stp1 stτ 1 p1 stτ 0 α − p0 α   , (B.25) with Xq and Xd given by Eq. (55) in the main text. In contrast to the representation (56) in Sec. III C 2, work- ing in the eigenbasis of L0 also fixes a particular ba- sis for the isospin part, namely the polarization basis which we order as e− = (e1 + ie2) /√2, e0 = Ω/Ω, and e+ = (e1 − ie2) /√2. The latter vectors are constructed from the right-handed orthonormal basis e0, e1 = λ/λ, and e2 = e0× e1 = Ω×λ/Ωλ. This yields the three com- α = e†α · τ n (α = −, 0, +) in Eq. (B.25) that ponents τ n make up the isospins τ n = Pα τ n eα. The projection of ρ on the quasistationary and the decaying subspace are next represented as α 0 (cid:19) , Xd (cid:19) . ρq = P qρ = (cid:18) Xq ρd = P dρ = (cid:18) 0 )pp′ = Vk†p · L0 · Vk′ (B.26) (B.27) 0 The matrix (Lkk′ p′ of the unperturbed Liouvillian [see Eq. (41)] expressed in terms of its left and right eigenvectors V k p , respectively, is specified completely by the two diagonal blocks p and V k 0 0 iαΩδαα′ (cid:19) , − iLqq −iLdd 0 = (cid:18) 0 0 = (cid:18) −γ 0 0 (−γ + iαΩ)δαα′ (cid:19) , (B.28) (B.29) whence the off-diagonal blocks vanish: Lqd 0 = 0. Here and below we use the shorthand notation Mαα′ for a 3 × 3 matrix M in the above-mentioned polarization basis. In this polarization basis the cross product Ω× is diagonal: 0 = Ldq 40 Next expressing the four blocks of the perturbation Λ = L − L0 in this basis, one finds stsαα′ (cid:19) , 0 −λsαα′ (cid:19) , 0 λp1 (B.32) (B.31) 0 0 0 (1 + p0 st)λcα (1 + p0 st)λc∗α′ −rλsαα′ (cid:19) , λcα (p0 λc∗α′ st − 3κ/2)λsαα′ (cid:19) . (B.33) (B.34) Λqq = (cid:18) 0 Λqd = (cid:18) 0 Λdq = (cid:18) Λdd = (cid:18) 0 The components of the matrix sαα′ are proportional to the (signed) volume of the parallel epiped spanned by the vectors eα, eα′, and λ = λ/λ, sαα′ = e†α ·(cid:16)λ × eα′(cid:17) = iλ √2  0 +1 0 +1 0 −1 0 −1 0   , (B.35) (B.36) (B.37) and the remaining factor cα is given by the projected isospin-to-charge conversion vector cα = c(cid:16)e†α · λ(cid:17) = c√2  1 0 1   , with c given by Eq. (34). Finally, Eq. (B.33) incorporates the transition factor r = p1 stp0 st − κ(cid:0) 1 2 p1 st − p0 st(cid:1) , whose dependence on the SQD parameters we thoroughly discussed in Sec. III C 3 in the main text. Equation (B.31) together with Eq. (B.35) reveals that the "direct" perturbation of the quasistationary modes for λ 6= 0 is not diagonal in the unperturbed eigenbasis. This expresses the fact that the mean-field Ω = Ω + p1 stλ is tilted with respect to Ω. Moreover, since the Liou- villian is not Hermitian, L 6= L†, the transition matrices are also not simply related by Hermitian conjugation, i.e., Λdq 6= (Λqd)†. This means that transitions between qua- sistationary and decay modes are not always possible in both directions (see Fig. 4), in contrast to Hamiltonian dynamics. In particular, all transitions into the station- ary charge modes (q, c) are forbidden (the first rows of Λqq and Λqd are zero). This is a consequence of the sum rules guaranteeing probability and isospin conservation by tunneling (see Ref. [32]). Physically, the eigenvalue of the stationary charge mode must stay pinned to zero. We further find from the above equations [see Eqs. (B.10)-(B.11)]: Ω× = Pαα′(cid:2)e†α · (Ω × eα′ )(cid:3) eαe†α′ =   −iΩ 0 0 0 0 0 0 0 +iΩ  , (B.30) g = Lbb − Laa ∼ γ, l = max(Lab,Lba) ∼ λ, ω = Lqq (B.38) (B.39) ∼ ∆ ∼ λ, Ω. (B.40) Thus, the time scales on the quasistationary subspace and the decaying subspace separate as required by Eq. (B.11) and satisfied by our assumption ∆ ≪ γ. Moreover, the weak-coupling assumption (B.10) is ful- filled in the weak-measurement limit λ ≪ γ. For further discussion of the consistency see also Appendix B 4. b. Computation of effective Liouvillian Exploiting Eqs. (B.26) and (B.14), the effective evolu- tion of the quasistationary modes can be expressed as Xq(t) = (cid:18) 0 τα(t)(cid:19) = −iLqq eff Xq(t) = Lqq eff(cid:18) 1 τα(t) (cid:19) . (B.41) Consistently accounting for terms up to second order in ∆ in the high-temperature limit, the effective Liouvillian reads Lqq eff = Lqq 0 + Λqq − iΛqd 1 γ Λqd + O(cid:16) ∆3 γ 2(cid:17) . (B.42) Inserting Eqs. (B.28) – (B.34), we obtain − iLqq eff = (cid:18) 0 Iα −i(Leff)αα′ (cid:19) , 0 with (B.43) (B.44) λ2r γ (s · s)αα′ (B.45) Iα = O(cid:16) ∆3 T(cid:17) γ 2 × Γ (Leff )αα′ = iαΩδαα′ + λp1 stsαα′ − +O(cid:16) ∆3 γ 2(cid:17) . The injection term Iα of Eq. (B.44) is thus negligible in the high-temperature limit considered here [123]. In Eq. (B.45) the first two terms are responsible for the qubit precession with frequency ∼ ∆, while the third term induces the isospin decay with the a rate ∼ λ2/γ, which coincides with Eq. (62) in the main text. Inserting Eq. (B.43) back into Eq. (B.41), we obtain for the total isospin evolution: τ (t) = Xα ταeα = Xα = −iLeffτ (t) eα(−iLeff)αα′ τα′ (t) (B.46) with Leff = (Leff)αα′ eαe†α′ = Xα=0,± (i Ωα − γα)eeff,αe†eff,α, (B.47) with Ωα, γα, and eeff,α, thus establishing Eqs. (89) in the main part of the paper. (80) – Finally, the slipped initial state following from Eq. (B.24) reads 41 (B.48) (B.49) 1 τeff,α(0)(cid:19) Xq eff (0) = (cid:18) = Xq(0) + Λqd 1 γ = (cid:18) Xd eff(0) + O(cid:16) ∆2 γ 2(cid:17) stτ 1 − p1 γ sαα′ (p0 1 stτ 0)α(0)(cid:19) (τ 0 + τ 1)α(0) − λ γ 2(cid:17) + O(cid:16) ∆2 α = e†α · τ n. Computing τeff(0) = Pα τeff,α(0)eα with τ n from Eq. (B.50), we arrive at Eq. (81) from the main text. The set of states with zero slip is discussed in de- tail in Appendix C and it is shown that the "size" of this set grows with the "distance" of the sensor from the stationary state. (B.50) This completes the derivation of the effective isospin evolution discussed in Sec. IV A. c. Markov approximation relative to the sensor QD Regarding the discussion of non-Markovian effects, we have to distinguish between two different memory ef- fects, namely those arising from the electrodes (imposed on the qubit-SQD system) and those arising from the sensor quantum dot (imposed on the qubit only). As explained in Appendix A 2, we account for the leading non-Markovian effect on the composite qubit-sensor QD system from the electrodes, which is induced by the tun- nel coupling; these effects are contained in the effective Liouvillian L in the kinetic equations (26), which forms the starting point of the above analysis. Their effect is to modify the coupling of the quasistationary and decay modes through the coherent backaction terms in the ki- netic equation (26) as explored in Appendix A 2 and Sec. III B 4. By contrast, the frequency dependence of the effective Liouvillian Leff for the reduced qubit system, Eq. (B.7), incorporates non-Markovian effects on the qubit due to memory of the sensor quantum dot after the electrodes were integrated out (thus effectively of SQD plus elec- trodes). These are accounted for by our general Laplace- space approach given in Appendix B 1 in an exact way relative to Eq. (26). In principle, these non-Markovian effects can be already studied based on the solutions of our full kinetic equations (26). However, the expressions (71) and (72) provide a more convenient starting point to gain further insight on an analytical level. In the main text, we focused for further illustration on the high-temperature limit that allows us to make the Markov approximation (B.19) when expanding the effec- tive Liouvillian (80) to lowest order in λ/γ. In this limit, SQD-induced memory effects on the qubit are thus ne- glected. Moreover, we neglect O(Γ2/T ) corrections to the SQD tunneling rates and, thus, non-Markovian correc- tions from the electrodes are also consistently neglected. This is reasonable because if memory effects of the SQD are not accounted for, then memory effects from the elec- trodes should have no effect a fortiori. 4. Validity of perturbation expansion In this part of the Appendix, we collect various re- marks on the validity and the limitations of our pertur- bation theory. a. Effective Liouvillian (69) As stated in Sec. II B, our kinetic equations are appli- cable as long as Γ/T ≪ ∆/Γ ≪ 1. One may now wonder whether it is permissible to expand the denominator in Eq. (69) in orders of ∆/Γ and to truncate this expansion. In general, the answer is no, for the following reason: To lowest order in ∆/Γ, the effective Liouvillian scales as λ2/Γ. The coherent backaction and and the cotunneling terms then yield corrections of order λ2/Γ × Γ/T ; how- ever, higher-order corrections in ∆/Γ are at least of order λ2/Γ × ∆/Γ and are therefore not negligible once we in- clude terms of order λ2/Γ × Γ/T [124]. Therefore, even though we start from the weak-measurement limit, one must not expand the denominator in Eq. (69) from the start. However, if we first neglect all Γ/T corrections (dis- sipative and coherent backaction and Γ2/T corrections to the SQD rates), then one can consistently expand the effective Liouvillian in ∆/T . The lowest-order approxi- mation to this is investigated in Sec. IV and requires high temperatures so that the Γ/T corrections are sufficiently small. b. Kinetic equation (26) The reader may also wonder whether one should not include terms of order ∆2 into the kinetic equation (26) since the lowest order to the effective Liouvillian scales at least quadratically in ∆. However, such terms appear only in combination with additional tunneling processes as the internal interaction L ∼ ∆ is treated without ap- proximation in the kinetic equations. Thus, terms in- cluding ∆2 must be at least of order Γ∆2/T 2 and are therefore strongly suppressed in the high-temperature limit that we consider in this paper. For example, if these terms appear in Λqq, they lead to corrections of or- der Γ∆2/T 2 ≪ ∆2/Γ × Γ/T since Γ/T ≪ 1. They are even less important if they appear in Λqd, Λdq, Λqq, where they would lead to terms of order ∆2/Γ(∆Γ/T 2) when inserted into Eq. (69). Moreover, one may also wonder whether terms of or- der Γ3/T 2 should be included into the kinetic equation 42 (26) since they can also modify the stochastic backac- tion. However, in the weak-coupling, weak-measurement limit considered in this paper, those result in corrections of even higher-order ∆2/Γ(Γ2/T 2) ≪ ∆2/T . Yet, deep in the Coulomb blockade regime (which is beyond the present scope), these terms should play an important role since the transition factor r [Eq. (62)] - which controls the strength of the total backaction together with c - is exponentially suppressed by virtue of the effective cancel- lation of Γ2/T and Γλ/T terms explained in Sec. III C 3. We expect higher-order corrections to cause deviations from this exponential suppression of the transition fac- tor r, resulting in an algebraic scaling ∼ 1/(ε − µr)n but with an exponent of higher than that of cotunneling broadening, n > 1, in agreement with other theoretical work (see Sec. V C). We emphasize that the main point of our work is just to explain physically the cancellation of the expected leading power law n = 1. A much more elaborate analysis is needed to find the actual power law (including the complications of a nonstationary detector with strong local Coulomb interaction, etc.). c. Exact result for U = 0 Finally, to further support the cancellation in Eq. (62) that we found here for large Coulomb interaction U and perturbatively in Γ, we computed in a separate calcula- tion the effective Liouvillian for a noninteracting sensor QD (U = 0) nonperturbatively in Γ but only to lead- ing order in λ. This releases the condition Γ/T ≪ λ/Γ and can be used to investigate also the opposite regime λ/Γ ≪ Γ/T (≪ 1). These results actually confirm the mitigation of the cotunneling-induced stochastic backac- tion by the coherent backaction also in this case. Yet, it remains an interesting future question to understand the role of the coherent backaction in the low-temperature, strong qubit-sensor coupling regime when also the strong interaction U is accounted for. Appendix C: Initial states without slip In this Appendix, we characterize the set of initial qubit-sensor QD density operators with zero initial slip [see Eqs. (74) and (81)] in the reduced dynamics of the qubit in the high-temperature limit. We show that this imposes a strong condition: Initial states with zero slip form a set of measure zero, thus making the initial slip a relevant source of errors in indirect detection unless the qubit-SQD quantum state (not just the qubit state) is under accurate control. In addition, increasing the non- stationarity of the sensor reduces these sets of zero slip. We also formulate the constraint in terms of relations be- tween the reduced qubit state and the composite qubit- SQD state and discuss how it relates to the factorizability of the qubit-SQD state. 43 st = 1. stp1(0), respectively. The radii are equal if and only if the reduced sensor state ρS is sta- tionary: the condition p0 stp0(0) is equivalent to pn(0) = pn st (n = 0, 1) due to the normalization condi- stp1(0) = p1 r1(cid:12)(cid:12) 6 R1 := p0 and (cid:12)(cid:12) tion Pn pn(0) =Pn pn From this set of valid initial states let us now construct those which have zero slip. According to Eq. (C.5), this requires the sum of the vectors r0 and r1 to lie on the line defined by the measurement vector: r0 + r1 = xλ with any x ∈ R. For the construction, sketched in Fig. 13, first draw a sphere with radius R0 (blue) and draw the line xλ through its origin (black). For each vector r0 in this sphere draw a second sphere of radius R1 centered at its tip (red). The set of vectors r1 that give zero- slip state are just given by the intersection of this second sphere with the line xλ. This is a set of measure zero. Moreover, from the figure it is clear that the construction is possible only if r0 is inside a cylinder of radius R1 with the line xλ as its axis. Whenever r0 lies outside this cylinder, it is not possible at all to construct a zero-slip initial state. (i) Stationary state pn(0) = pn The radii Rn are controlled by the initial reduced sen- sor state through pn(0) and there are two extreme limits: In this case for every r0 one can find an r1 giving a zero-slip state. Still, the subset has measure zero in the total set of possible states. st: R0 = R1. (ii) Integral charge state p0,1(0) = 0, p1,0(0) = 1: R0,1 = 0, R1,0 = p1,0 st . In these two cases, the charge- specific isospin τ 0,1(0) = τ (0) coincides with the total isospin, while τ 1,0(0) = 0. The slip is τ eff (0) − τ (0) = ∓λ × p1,0 st τ (0)/γ and can be avoided only if τ 0,1(0) is collinear to λ so that the tip of r0,1 lies on the line xλ to allow for r1,0 = 0 as sketched in Fig. 13(a). As the sensor deviates from the stationary state, the radii R1 and R0 differ and the above possibilities for con- structing initial zero-slip states are reduced. Since in any case the zero-slip states are sets of measure zero, it is clear that most preparation errors of the sensor QD will lead to an initial slip on the qubit, i.e., the backaction-induced initial slip generates additional errors beyond the control over the qubit. Restriction on the qubit isospin. It is instructive to further clarify the restrictions that the above imposes on the reduced qubit state, a density operator completely characterized by τ (0), relative to the composite qubit- sensor state ρ(0). To this end we change the variables to τ (0) = τ 0(0) + τ 1(0) stp0(0) − p0 δst(0) = p1 δ(0) = p1(0)τ 0(0) − p0(0)τ 1(0) stp1(0) (C.6) (C.7) (C.8) The conditions imposed by the positivity ρ(0) on τ (0) and δ(0) are not easy to formulate and will be ignored in the following (note that τ (0) 6 1 is only necessary, not sufficient). Rewriting the zero-slip condition (81) by FIG. 13: Geometric restrictions on the qubit-sensor density operator imposed by requiring zero initial slip. We show a 2D cross section of the 3D construction described in the text, i.e., the circles represent spheres and the dotted lines indicate the boundaries of a cylinder around the line xλ. Kinematic restrictions. It is most convenient for the following considerations to work with the representation of the initial qubit-SQD state, ρ(0) = Xn P n ⊗ 1 2(cid:0)pn(0)1Q + τ n(0) · τ(cid:1) , (C.1) in terms of the occupation probabilities pn(0) and the charge-specific isospins τ n(0). We recall that for this state to be a valid quantum state, pn(0) and τ n(0) have be real (for ρ† = ρ), Pn pn(0) = 1 (for tr ρ = 1), and the magnitudes τ n(0) 6 pn(0) (C.2) have to be restricted (for positivity, ρ > 0). We note that the reduced state of the SQD and the qubit, (C.3) (C.4) ρS := tr Q ρQ := tr S pn(0) P n, ρ(0) = Xn ρ(0) = 1 2(cid:0)1Q + τ (0) · τ(cid:1) , are completely specified by the occupation probabilities pn(0) (in fact by only one) and the Bloch vector τ (0) = τ 0(0) + τ 1(0), respectively. When pn(0) = pn st, the SQD is stationary, but this does not imply that the qubit-SQD state state ρ(0) is factorisable (see below). Charge-specific isospins. We first investigate the slip based on the charge-specific isospins τ n(0) since these have simple kinematic constraints, allowing for an easy, complete characterization. The zero-slip condition ob- tained from Eq. (81), 0 = λ ×(cid:0)r0 + r1(cid:1) , (C.5) stτ 0(0) is expressed using rescaled isospin vectors r0 = p1 stτ 1(0). For an arbitrary initial state, the and r1 = −p0 two three-dimensional vectors r0 and r1 are taken from a six-dimensional set that is constrained only by the pos- itivity of ρ(0): the vectors r0 and r1 have to lie within stp0(0) spheres of different radii given by (cid:12)(cid:12) r0(cid:12)(cid:12) 6 R0 := p1 inserting τ 0,1(0) = p0,1(0)τ (0) ± δ(0) yields 0 = λ × [δ(0) + δst(0)τ (0)] (C.9) Noting that δ(0) = 0 corresponds to a factorizable state [see Eq. (17)] and δst(0) = 0 to a stationary reduced sensor state (see above), we have four cases in which there is zero initial slip: (i) nonfactorizable, nonstationary initial state δ(0) 6= 0, δst(0) 6= 0: The qubit state and the qubit-SQD correla- tions must be fine tuned such that δ(0)+δst(0)τ (0) = xλ or some x ∈ R. (ii) nonfactorizable, stationary initial state δ(0) 6= 0, δst(0) = 0: The qubit state can be arbitrary, but the qubit-SQD state must have have very special correlations such that δ(0) ∝ λ. (iii) factorizable, nonstationary initial state δ(0) = 0, δst(0) 6= 0: The qubit state must be prepared in a measurement-basis state τ (0) ∝ λ. (iv) factorizable, stationary initial state δ(0) = 0, δst(0) = 0: no conditions, there is no slip. Equation (C.9) emphasizes that the set of states with- out slip does not simply coincide with factorizable initial states ("uncorrelated states"). It further shows that the more the sensor deviates from the stationary state, now quantified by δst(0), the larger the initial slip becomes. This concludes our discussion of the states with zero slip and, as the previous analysis showed, the zero-slip states are subsets of measure zero of the set of possible initial states. The experimental possibilities to avoid the slip [case (iv)] are discussed in Sec. III E and Sec. VI. Appendix D: Renormalization-induced qubit phase kicks In this final Appendix, we explain how the coherent backaction appears in the visibility (106) as an addi- tional contribution that cannot be understood in a semi- classical stochastic picture discussed in Sec. V A. We show that, loosely speaking, the coherent backaction in- duces additional "phase kicks" on the qubit that par- tially "undo" the phase kicks induced by the stochastic backaction that result in decoherence. As a result, the decoherence is mitigated. We emphasize from the start that these phase kicks are completely unrelated to the initial slip: They do not lead to a phase shift of coherent isospin precession but instead only affect the net qubit decoherence. Our objective in the following is to merely further il- lustrate the physical origin of the coherent backaction by calculations that are certainly not rigorous. The proper treatment is achieved by our kinetic equation (26) which is based on the systematic real-time diagrammatic ap- proach. Still, we believe the following may be instructive. We start from the expression (106) for the visibility, D(t) = tr [(+ih− ⊗ 1SR) ×e−iHt(+ih− ⊗ ρSR)eiHt(cid:3) , (D.1) 44 assuming the initial qubit-environment state factorizes and H is given by Eq. (1). We furthermore neglect the spin degree of freedom for the SQD and thereby also the Coulomb interaction effect, i.e., the SQD takes two charge states 0i and 1i. As noted in Sec. V A and Appendix B 4, neither is essential for the coherent back- action. We next split up the Hamiltonian H = H0 + HT and expand Eq. (D.1) in the tunneling HT of electrons between the SQD and the reservoir and vice versa. In O(Γ) ∼ O(H 2 T ), the following term contributes (there are more terms; we just consider a relevant one): D(t) ∼ Z dt1dt2 [h+e−iH0(t−t1)HT e−iH0t1+i tr S,R 06t1,t26t ⊗ ρSRh−eiH0(t−t2)HT eiH0t2−i] (D.2) Importantly, the tunneling process for the ket-evolution h+ ← h+ may happen at a time t1 different from the time t2 for the tunneling process of the bra evolution h− → −i. The coherent evolution in between is re- sponsible for an additional phase shift. To see this, we assume the SQD to be initially unoccupied, i.e., ρSR = 0ih0 ⊗ ρR,0, and to end up in a singly occu- pied state 1ih1. We assume again for simplicity Ω = 0. Evaluating the trace over the electrodes, we get D(t) ∼ Z dt1dt2 06t26t16tXr Z dωf + ×he−i(ε+ λ 2 +ω)(t1−t2)i , 2 −ω)(t1−t2) + e−i(−ε+ λ r (ω)Γr(ω)e−i λ 2 (t−t1) (D.3) where the two terms in the bracket relate to the two cases, t1 > t2, and t1 6 t2 in Eq. (D.2), respectively. It is easy to show that the terms in the second line arise from SQD-qubit coherences of the type 1, +ih0,− and 0, +ih1,−, respectively. Finally, the SQD reaches the singly occupied real state 1ih1, giving rise to the phase shift e−iλ(t−t1). The latter is related to the stochastic backaction (since the time t1 is random), while the for- mer phase shift contains the coherent backaction as we explain next. For this purpose, we redefine τ := t1 − t2, assume large t, and introduce a small imaginary part into the exponentials to ensure convergence. Since τ can take nearly all positive values for large t, the coherent phase shift becomes approximately 2 −ω−i0)τ + e−i(−ε+ λ 0 ∼ Z ∞ = (−i)(cid:20) ≈ −2 Im 1 dτhe−i(ε+ λ ε + λ/2 − ω − i0 − ε − ω + i0 − iλ ∂ ∂ε 1 where the last step holds in first order in λ. We ignore the first part, which is independent of λ, and insert the second part into Eq. (D.3), which yields a term propor- tional to the renormalization function. When taking the 1 2 +ω−i0)τi ε − λ/2 − ω + i0(cid:21) ε − ω + i0 1 , Re (D.4) time derivative of Eq. (D.3), it reproduces the coherent backaction: Γrλ T φ′r(cid:18) ε − µr D ∼ (−i)Xr T (cid:19) = − iκ (D.5) which coincides with the effect of the term τ 1 = κλ× τ 0 in the kinetic equation (26). What our heuristic argument clarifies is why a semiclas- sical approach as discussed in Sec. V A is not capable of reproducing the coherent backaction: this approach cru- cially relies on the assumption that the charge state of the SQD is a i.e. n(t) is just a fixed (but random) function of t. Thus, the transi- classical variable, 45 tion 0 → 1 happens on both the bra- and the ket-part of the evolution at the same time t1 = t2 (while the time t1 is random). In this case, the coherent phase shift just vanishes. This amounts formally to replacing 1/x → −iπδ(x) and "by hand" dropping the principal value integral term P (1/x) that contains renormaliza- tion effects. In the single-step Born-Markov approach discussed in Sec. V B, the problem is to find a proper description of the environmental state that contains the intermediate coherences 1ih0 and 0ih1. Our above non- rigorous derivation of the coherent backaction terms thus illustrates that virtual fluctuations in nanoscale sensors can have a real impact on the measurement backaction. [1] J. I. Cirac and P. Zoller, Phys. Rev. Lett. 74, 4091 235307 (2001). (1995). [2] A. M. Tyryshkin, S. A. Lyon, A. V. Astashkin, and A. M. Raitsimring, Phys. Rev. B 68, 193207 (2003). [3] 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). [4] J. Koch, T. M. Yu, J. Gambetta, A. A. Houck, D. I. Schuster, J. Majer, A. Blais, M. H. Devoret, S. M. Girvin, and R. J. Schoelkopf, Phys. Rev. A 76, 042319 (2007). [5] H. Bluhm, S. Foletti, I. Neder, M. Rudner, D. Ma- halu, V. Umansky, and A. Yacoby, Nature Phys. 7, 109 (2010). [6] J. Bylander, S. Gustavsson, F. Yan, F. Yoshihara, K. Harrabi, G. Fitch, D. G. Cory, Y. Nakamura, J.-S. Tsai, and W. D. Oliver, Nature Physics 7, 565 (2011). [7] X. Rong, Y. Wang, J. Yang, J. Zhu, W. Xu, P. Feng, X. Wen, J. Su, and J. Du, Chinese Science Bulletin 56, 591 (2011). [8] R. Hanson, L. H. W. van Beveren, I. T. Vink, J. M. Elzerman, W. J. M. Naber, F. H. L. Koppens, L. P. Kouwenhoven, and L. M. K. Vandersypen, Phys. Rev. Lett. 94, 196802 (2005). [9] C. Barthel, M. Kjaergaard, J. Medford, M. Stopa, C. M. Marcus, M. P. Hanson, and A. C. Gossard, Phys. Rev. B 81, 161308 (2010). [10] E. Il'ichev, N. Oukhanski, A. Izmalkov, T. Wagner, M. Grajcar, H.-G. Meyer, A. Y. Smirnov, A. Maassen van den Brink, M. H. S. Amin, and A. M. Zagoskin, Phys. Rev. Lett. 91, 097906 (2003). [11] E. Onac, F. Balestro, L. H. W. van Beveren, U. Hart- mann, Y. V. Nazarov, and L. P. Kouwenhoven, Phys. Rev. Lett. 96, 176601 (2006). [12] A. Korotkov and D. Averin, Physical Review B 64 (2001). [18] R. J. Schoelkopf, P. Wahlgren, A. A. Kozhevnikov, P. Delsing, and D. E. Prober, Science 280, 1238 (1998). [19] J. M. Elzerman, R. Hanson, L. H. W. van Beveren, B. Witkamp, L. M. K. Vandersypen, and L. P. Kouwen- hoven, Nature 430, 431 (2004). [20] D. J. Reilly, C. M. Marcus, M. P. Hanson, and A. C. Gossard, Appl. Phys. Lett. 91, 162101 (2007). [21] J. I. Colless, A. C. Mahoney, J. M. Hornibrook, A. C. Doherty, H. Lu, A. C. Gossard, and D. J. Reilly, Phys. Rev. Lett. 110, 046805 (2013). [22] E. Paladino, L. Faoro, G. Falci, and R. Fazio, Phys. Rev. Lett. 88 (2002). [23] A. Grishin, I. Yurkevich, and I. Lerner, Physical Review B 72 (2005). [24] J. Bergli, Y. M. Galperin, and B. L. Altshuler, New Journal of Physics 11, 025002 (2009). [25] A. Shnirman, G. Schon, I. Martin, and Y. Makhlin, Physical review letters 94, 127002 (2005). [26] G. Ithier, E. Collin, P. Joyez, P. Meeson, D. Vion, D. Es- teve, F. Chiarello, A. Shnirman, Y. Makhlin, J. Schriefl, et al., Physical Review B 72 (2005). [27] L. Chirollli and G. Burkard, Adv. in Phys. 57, 225 (2008). [28] Y. M. Galperin, B. L. Altshuler, and D. V. Shant- sev, in Fundamental Problems of Mesoscopic Physics (Springer, 2004), p. 141165. [29] A. Shnirman and G. Schon, Phys. Rev. B 57, 15400 (1998). [30] Y. Makhlin, G. Schon, and A. Shnirman, Rev. of Mod. Phys. 73, 357 (2001). [31] S. A. Gurvitz and D. Mozyrsky, Phys. Rev. B 77, 075325 (2008). [32] M. Hell, M. R. Wegewijs, and D. P. DiVincenzo, Phys. Rev. B 89, 195405 (2014). [33] Y. Makhlin, G. Schon, and A. Shnirman, Physical Re- [13] S. A. Gurvitz and G. P. Berman, Phys. Rev. B 72, view Letters 85, 4578 (2000). 073303 (2005). [34] J. Konig and J. Martinek, Phys. Rev. Lett. 90, 166602 [14] A. A. Clerk and A. D. Stone, Phys. Rev. B 69, 245303 (2003). (2004). [35] M. Braun, J. Konig, and J. Martinek, Phys. Rev. B 70, [15] A. Korotkov, Physical Review B 78 (2008). [16] N. Oxtoby, H. Wiseman, and H.-B. Sun, Physical Re- view B 74, 045328 (2006). 195345 (2004). [36] B. Wunsch, M. Braun, J. Konig, and D. Pfannkuche, Phys. Rev. B 72, 205319 (2005). [17] H.-S. Goan and G. Milburn, Physical Review B 64, [37] A. Donarini, M. Grifoni, and K. Richter, Phys. Rev. 46 Lett. 97, 166801 (2006). [70] A. Shnirman, Y. Makhlin, and G. Schn, Physica Scripta [38] M. G. Schultz, Phys. Rev. B 82, 155408 (2010). [39] M. Governale, M. G. Pala, and J. Konig, Phys. Rev. B 77, 134513 (2008). [40] J. Splettstoesser, M. Governale, J. Konig, and R. Fazio, Phys. Rev. B 74, 085305 (2006). [41] J. Splettstoesser, M. Governale, J. Konig, and M. Buttiker, Phys. Rev. B 81, 165318 (2010). [88] C. Emary, Physical Review A 78, 032105 (2008). [89] S. Lindebaum and J. Knig, Physical Review B 84, 2002, 147 (2002). [71] L. D. Contreras-Pulido, J. Splettstoesser, M. Governale, J. Konig, and M. Buttiker, Phys. Rev. B 85, 075301 (2012). [72] J. Schulenborg, J. Splettstoesser, M. Governale, and L. D. Contreras-Pulido, Phys. Rev. B 89, 195305 (2014). [73] M. Pletyukhov, D. Schuricht, and H. Schoeller, Phys. Rev. Lett. 104, 106801 (2010). [74] M. Hell, S. Das, and M. R. Wegewijs, Phys. Rev. B 88, 235409 (2011). 115435 (2013). [90] P. Gaspard and M. Nagaoka, The Journal of chemical [75] I. Weymann, J. Konig, J. Martinek, J. Barna´s, and physics 111, 5668 (1999). G. Schon, Phys. Rev. B 72, 115334 (2005). [45] M. H. Devoret and J. M. Martinis, Experimental As- [76] E. Bonet, M. M. Deshmukh, and D. C. Ralph, Phys. pects of Quantum Computing p. 163203 (2005). Rev. B 65, 045317 (2002). [46] J. Salmilehto, P. Solinas, and M. Mottonen, Physical Review A 85, 032110 (2012). [47] C. Karlewski and M. Marthaler, arXiv preprint arXiv:1402.6205 (2014). [48] H.-P. Breuer and F. Petruccione, The Theory of Open Quantum Systems (Oxford University Press, Oxford, 2002). [77] A. A. Clerk, M. H. Devoret, S. M. Girvin, F. Marquardt, and R. J. Schoelkopf, Reviews of Modern Physics 82, 1155 (2010). [78] A. Braggio, J. Konig, and R. Fazio, Phys. Rev. Lett. 96, 026805 (2006). [79] C. Flindt, T. Novotn´y, A. Braggio, M. Sassetti, and A.-P. Jauho, Phys. Rev. Lett. 100, 150601 (2008). [49] R. B. Saptsov and M. R. Wegewijs, Phys. Rev. B 90, [80] C. Flindt, T. Novotn´y, A. Braggio, and A.-P. Jauho, 045407 (2014). [50] H. Schoeller, Eur. Phys. J. Special Topics 168, 179 (2009). [51] M. Leijnse and M. R. Wegewijs, Phys. Rev. B 78, 235424 (2008). [52] C. W. Bates Jr, Physical Review 188, 529 (1969). [53] M. Tokuyama and H. Mori, Progress of Theoretical Phys. Rev. B 82, 155407 (2010). [81] A. Braggio, C. Flindt, and T. Novotn´y, Journal of Statistical Mechanics: Theory and Experiment 2009, P01048 (2009). [82] M. C. Goorden, M. Thorwart, and M. Grifoni, Phys. Rev. Lett. 93, 267005 (2004). [83] U. Hartmann and F. K. Wilhelm, Phys. Rev. B 75, Physics 55, 411429 (1976). 165308 (2007). [54] C. Timm, Phys. Rev. B 83, 115416 (2011). [55] E. Fick and G. Sauermann, The quantum statistics of dynamic processes, Springer series in solid-state sciences (Springer-Verlag, 1990). [56] S. Nakajima, Progress of Theoretical Physics 20, 948959 (1958). [84] D. M. Kennes, O. Kashuba, M. Pletyukhov, H. Schoeller, and V. Meden, Phys. Rev. Lett. 110, 100405 (2013). [85] P. Rebentrost, I. Serban, T. Schulte-Herbruggen, and F. K. Wilhelm, Phys. Rev. Lett. 102, 090401 (2009). [86] F. F. Floether, P. de Fouquieres, and S. G. Schirmer, [57] R. Zwanzig, The Journal of Chemical Physics 33, 1338 New J. Phys. 14, 073023 (2012). (1960). [58] E. B. Davies, Communications in Mathematical Physics 39, 91 (1974). [59] E. B. Davies, Annales de l'institut Henri Poincar´e (B) 11, 265 (1975). [60] B. Abel and F. Marquardt, Physical Review B 78 [87] S. J. Glaser, U. Boscain, T. Calarco, C. P. Koch, W. Kockenberger, R. Kosloff, I. Kuprov, B. Luy, S. Schirmer, T. Schulte-Herbruggen, et al., arXiv preprint arXiv:1508.00442 (2015). [88] C. Emary, Phys. Rev. A 78, 032105 (2008). [89] S. Lindebaum and J. Konig, Phys. Rev. B 84, 235409 (2008). (2011). [61] C. Neuenhahn, B. Kubala, B. Abel, and F. Marquardt, [90] P. Gaspard and M. Nagaoka, J. Chem. Phys. 111, 5668 physica status solidi (b) 246, 1018 (2009). (1999). [62] J. Schriefl, Y. Makhlin, A. Shnirman, and G. Schn, New [91] F. Haake and M. Lewenstein, Phys. Rev. A 28, 3606 Journal of Physics 8, 1 (2006). (1983). [63] Y. Makhlin and A. Shnirman, Physical Review Letters 92 (2004). [64] A. Berthelot, C. Voisin, C. Delalande, P. Roussignol, R. Ferreira, and G. Cassabois, Advances in Mathemat- ical Physics 2010, 1 (2010). [65] A. V. Kuhlmann, J. Houel, A. Ludwig, L. Greuter, D. Reuter, A. D. Wieck, M. Poggio, and R. J. War- burton, Nature Phys. 9, 570 (2013). [66] T. Itakura and Y. Tokura, Physical Review B 67 (2003). [67] A. G. Redfield, IBM Journal of Research and Develop- ment 1, 19 (1957). [68] S. Koller, M. Grifoni, M. Leijnse, and M. R. Wegewijs, Phys. Rev. B 82, 235307 (2010). [69] E. Paladino, M. Sassetti, G. Falci, and U. Weiss, Phys- ical Review B 77 (2008). [92] U. Geigenmuller, U. Titulaer, and B. Felderhof, Physica A: Statistical Mechanics and its Applications 119, 41 (1983). [93] F. Haake and R. Reibold, Phys. Rev. A 32, 2462 (1985). [94] F. Wilhelm, New J. Phys. 10, 115011 (2008). [95] M. A. Nielsen and I. L. Chuang, Quantum Computa- tion and Quantum Information (Cambridge University Press, Cambridge, United Kingdom, 2010). [96] H. Gassmann, F. Marquardt, and C. Bruder, Phys. Rev. E 66, 041111 (2002). [97] H. Gutmann, F. K. Wilhelm, W. M. Kaminsky, and S. Lloyd, in Experimental Aspects of Quantum Comput- ing (Springer, 2005), pp. 247–272. [98] D. Kribs and D. Poulin, Operator Error Correction (Cambridge University Press, Cambridge, United King- dom, 2013). [99] The first step of our approach is in principle generally possible at least for weak tunnel coupling Γ, i.e., not limited to the weak-measurement situation we study in this paper. [100] Here we use Re ψ(1/2 + ix/2π) ≈ ln x for x ≫ 1. [101] Note that we investigate the impact of the coherent backaction by setting here and below "by hand" κ = 0 in our results. Although one can express the non- Markovian correction factor as η = 1 + κ, we do not set η = 1 since this would affect the stochastic backac- tion and would not lead to the comparison we intend to make. [102] Note that the dynamics of the quasistationary modes does not contain information about the response of the sensor QD to the qubit since the coefficient of the sta- tionary charge mode Vq c trivially equals 1. To describe the sensor response, one must compute the dynamics of δst = p1 stp1, see Eq. (55) (while the isospin degrees of freedom may be projected out for that pur- pose). stp0 − p0 [103] The coherent backaction does not only affect the transi- tions into the decaying subspace but also affects the evo- lution in the decay space by modifying Λdd in Eq. (56). However, these terms yields only a small correction as compared to the large free evolution term Lqq 0 ∼ γ. [104] It is known that at finite temperature the condition for a resonance is not ε = µr, but there is an offset linear in T with a coefficient that grows with junction asym- metry. This is related to Coulomb interaction effects on the QD and appears already in O(Γ) [76]. This causes the maximum of the decoherence rates to lie at nonzero Vg (see Fig. 5) and also causes the shift with changing junction asymmetry g in Fig. 6. [105] The kinetic equations for sensor QD plus qubit are con- sidered here in the wide-band limit with respect to the bandwidth W of the electrodes to which the sensor is coupled. [106] The transition factor satisfies r > 0 even when cotun- neling and coherent backaction terms are included, see Eq. (65). [107] Note that the decay rates would diverge in the limit λ/Γ → ∞. [108] We stress that Ref. [33] is based on a kinetic equation to leading order in Γ, i.e., cotunneling and renormalization effects are excluded from the start. Thus, the authors do not address the same questions as we do here. [109] Note that τ eff (0) differs from the initial isospin τ (0) due to the initial slip (81) and therefore F is not necessarily limited to the range [−1, +1]. [110] We note that applying Eq. (104) requires a calculation of the charge-specific isospins from the full kinetic equa- tions (26). Using only the projections quasistationary subspace, Xq(t) = (1 τ (t))†, see Eq. (B.26), would just yield zero when inserted into Eq. (104). [111] We introduce a factor 1/2 for convenience to adjust the amplitude of the fluctuations to ∆ξ = 2, as usually used for a random telegraph process. 47 [112] We adjusted the notation in Ref. [66] to our notation by replacing 1/τ0 = 1/τu + 1/τl = γ and JT → λ/2. [113] The reason is that the dissipative backaction does not affect the coupling between the quasistationary and de- cay modes [see Eqs. (61) – (63)]. [114] In Redfield's notation in Ref. [67], the random pertur- bation is for our model given by G(t) = 1 2 ξ(t)λ · τ . [115] To compare with Eq. (7) of Ref. [25], we note that for our model there is no direct hybridization ∆j = 0 be- tween the orbital levels of the SQD, i.e., cos(θj) = 1, and sin(θj) = 0. Furthermore, the relaxation rate Γ1,j to the stationary state is identified with γ here. [116] This is a consequence of the approximation K2,j(λ) ≈ K2,j (0) employed below Eq. (3) in Ref. [22]. [117] We express the parameters g and ε in Ref. [23] in terms of our notation as g = λ/Γ and ε = 2ε/Γ. [118] The correlation time is set by the time dependence of the contraction functions in the diagrammatic expansion (reservoir correlation functions), which drop on the time scale of the inverse temperature 1/T , see Eqs. (93) and (100) in Ref. [49]. Note that the contribution from the stationary state ρst just vanishes, i.e., W (t − t′)ρst = 0 for any time t′. [119] This can be understood as follows: ∂ ∂ ∂z i.e., 1 z/T +Xi−(LQS −µi)/T such z/T +Xi−(LQS −µi)/T , derivative the finite fre- is given by Eq. (A6) in Ref. [32] i0 → z/T . Each frequency deriva- it involves ∼ quency kernel by replacing tive in ∂nW acts on a propagator, expressions 1 which can T dimensionless as be for each z- and a energy T ∂nW ∼ 1 derivative. This yields 1 T n (cid:0)Qi is a dimensionless function that contains the frequency integrals. Thus, ∂nW ∼ Γ/T n since W ∼ Γ (plus higher-order corrections). ∂[(LQS −µ)/T ]n I (cid:16) LQS −µ ∂(LQS −µi)/T rewritten ratios factor schematically (cid:17), where I of 1/T as 1 a Γi T (cid:1) ∂n T kernel [120] The first-order schematically i.e., ∂1W Γ(z) ∼ reads W Γ(z) ∼ R dxf ±(x)/[z + x − L], ∂/∂LR dxf ±(x)/[z + x − L]. [121] To obtain a solution ρ(t) that is also consistently ex- panded to all orders of ∆ and Γ, one should insert the expanded density operator ρ(t) = ρ0,0(t) + ρΓ,0(t) + ρ0,∆ + . . . together with the expanded kernels into the kinetic equation ρ(t) = −iLρ(t), and compare left- and right0hand side in all orders. [122] See also the discussion of microscopic derivations of master equations in Ref. [48]. [123] We checked that when accounting for the leading-order expression in Eq. (B.44) for lower temperatures, one reproduces the stationary state of Ref. [32] with nonzero stationary isospin τst. [124] We note that we still consider the weak-measurement regime (λ/Γ ≪ 1). The validity of our kinetic equations is just limited to high temperatures implying Γ/T ≪ λ/Γ.
1910.01295	1	1910	2019-10-03T03:47:33	Nonlinear transport of ballistic Dirac electrons tunneling through a tunable potential barrier in graphene	[ "cond-mat.mes-hall", "cond-mat.mtrl-sci" ]	Dirac-electronic tunneling and nonlinear transport properties with both finite and zero energy bandgap are investigated for graphene with a tilted potential barrier under a bias. For validation, results from a finite-difference based numerical approach, which is developed for calculating transmission and reflection coefficients with a dynamically-tunable (time-dependent bias field) barrier-potential profile, are compared with those of both an analytical model for a static square-potential barrier and a perturbation theory using Wentzel-Kramers-Brillouin (WKB) approximation. For a biased barrier, both transmission coefficient and tunneling resistance are computed and analyzed, indicating a full control of the peak in tunneling resistance by bias field for a tilted barrier, gate voltage for barrier height, and energy for incoming electrons. Moreover, a finite energy gap in graphene is found to suppress head-on transmission as well as skew transmission with a large transverse momentum. For a gapless graphene, on the other hand, filtering of Dirac electrons outside of normal incidence is found and can be used for designing electronic lenses. All these predicted attractive transport properties are expected extremely useful for the development of novel electronic and optical graphene-based devices.	cond-mat.mes-hall	cond-mat	Nonlinear transport of ballistic Dirac electrons tunneling through a tunable potential barrier in graphene Farhana Anwar1,2∗, Andrii Iurov3,1†, Danhong Huang4, Godfrey Gumbs5,6 and Ashwani Sharma1,2,4 1Center for High Technology Materials, University of New Mexico, 1313 Goddard SE, Albuquerque, NM 87106, USA 2Department of Electrical and Computer Engineering, University of New Mexico, Albuquerque, NM 87106, USA 3Department of Physics and Computer Science, Medgar Evers College of City University of New York, Brooklyn, NY 11225, USA 4Air Force Research Laboratory, Space Vehicles Directorate, Kirtland Air Force Base, NM 87117, USA 5Department of Physics and Astronomy, Hunter College of the City University of New York, 695 Park Avenue, New York, NY 10065, USA 6Donostia International Physics Center (DIPC), P de Manuel Lardizabal, 4, 20018 San Sebastian, Basque Country, Spain (Dated: October 4, 2019) Dirac-electronic tunneling and nonlinear transport properties with both finite and zero energy bandgap are investigated for graphene with a tilted potential barrier under a bias. For validation, results from a finite-difference based numerical approach, which is developed for calculating trans- mission and reflection coefficients with a dynamically-tunable (time-dependent bias field) barrier- potential profile, are compared with those of both an analytical model for a static square-potential barrier and a perturbation theory using Wentzel-Kramers-Brillouin (WKB) approximation. For a biased barrier, both transmission coefficient and tunneling resistance are computed and analyzed, indicating a full control of the peak in tunneling resistance by bias field for a tilted barrier, gate voltage for barrier height, and energy for incoming electrons. Moreover, a finite energy gap in graphene is found to suppress head-on transmission as well as skew transmission with a large trans- verse momentum. For a gapless graphene, on the other hand, filtering of Dirac electrons outside of normal incidence is found and can be used for designing electronic lenses. All these predicted attractive transport properties are expected extremely useful for the development of novel electronic and optical graphene-based devices. I. INTRODUCTION Graphene, a one atom-thick allotrope of carbon, being a conductor with exceptionally large mobility at a large range of ambient temperatures, makes a strong case for a number of ballistic transport nanodevices. It has unique electronic properties due to its linear energy dispersion with zero bandgap, as well as a spinor two-component wave function. These unique characteristics give rise to some highly unusual electronic and transport properties. 1 -- 4 These peculiar properties result in a fact that a potential barrier becomes transparent to electrons arriving at normal incidence regardless of its height or width. This effect, known as Klein tunneling, 1 restricts the switching-off capability (complete pinch-off of electric current) for logical applications, and makes graphene difficult to achieve logical functionalities without use of chemical modification or patterning. 5 -- 8 On the other hand, such a situation also offers a unique possibility to fabricate various ballistic devices and circuits in which electrons experience focusing by one or several potential barriers. Practically, a zero-bandgap two-dimensional material acquires an important advantage over metals in its capability to tune the conductivity by means of either chemical doping or a gate voltage with a desired geometrical pattern. 9 -- 11 Interestingly, the induced planar barrier structure within a grraphene sheet can be realized by applying a gate voltage, either a static or a transient one. This is quite different from the design of a high-electron-mobility transistor. For example, by using different inhomogeneous profiles of static bias voltage, various structures , such as bipolar (p−n, n− p, p− n− p, n− p− n, etc.) as well as unipolar junctions (n− n(cid:48), p− p(cid:48)) can be facilitated 12 -- 18 to achieve desired voltage dependence of electrical conductance. In spite of the considered junction being abrupt or graded, its angular selectivity for carrier transport makes it a unique one in comparison with conventional semiconductor junctions, e.g., metaloxidesilicon field-effect transistors. 9 1 0 2 t c O 3 ] l l a h - s e m . t a m - d n o c [ 1 v 5 9 2 1 0 . 0 1 9 1 : v i X r a ∗ [email protected] † [email protected], [email protected] 2 Other significant roles played by our proposed tunable junctions include Veselago lens, 19,20 Fabry-Perot interfer- omete,r 13 subthermal switches, 5 Andreev reflections, 21 by exploiting optics-like behavior of ballistic Dirac electrons. Therefore, in order to design future-generation graphene-based electronics, it is crucial to gain a full understanding of mechanism for ballistic transports across various types of potential barriers in graphene. Negative refractive index with a single ballistic graphene junction, which is associated with electron-hole switching, has already been observed experimentally, 5,15 and it strongly affects the operation of an electric switch. 5,22,23 Various theoretical methods have been adopted aiming to obtain electron transport in graphene, including trans- fer matrix, 24 -- 26 non-equilibrium Green's function, 6,27 tight-binding model, 28,29 as well as semi-classical Wentzel- Kramers-Brillouin (WKB) approximation. 14,17,28 However, there are still few of studies on electronic transport prop- erties using finite-difference method 30 for numerical calculation with an arbitrary potential profile. A crucial advan- tage of this numerical method is a possibility to take into account of random local disorder potentials within barrier materials of roughness at two barrier edges. A number of fabricated optical devices face such a situation, which detri- ments the device performance 8,23 while trying to accomplish ballistic p − n junction characteristics experimentally. Alternatively, some smooth p − n and n − p − n junctions in graphene were realized and analyzed theoretically in Refs. [7,8,12,13,15,31,32]. The remaining part of this paper is organized as follows. In Sec. II, we introduce finite-difference method for calculating transmission coefficient of Dirac electrons in graphene in the presence of a biased potential barrier, along with numerical results of transmission coefficient as functions of incident angles and electron energy, as well as tunneling resistance as a function incident electron energy, with various values of bias field. We present in Sec. III analytical results within the WKB approximation for both large and small bias-field limits, accompanied by numerical results for transmission coefficient as functions of both incident angles and electron energy. Finally, our concluding remarks are presented in Sec. IV. II. FINITE-DIFFERENCE METHOD FOR TUNNELING OF DIRAC ELECTRONS In this Section, we lay out the formalism, and present and discuss our numerical results based on a finite-difference method. The main advantage of this method is its capability to obtain exact electron wave functions for arbitrary potential profiles. 30 We will consider both cases with a zero or finite energy gap for graphene. Technically, an energy gap (∼ 200 meV) could be introduced by placing a graphene sheet on top of either insulating silicon-based 33 or hexagonal boron- nitride substrate. 34 It could also be realized by patterned hydrogen adsorption 35 or imposing a circularly-polarized off-resonance laser field.36,37 This gap opening leads to substantial modifications of electronic, transport and collective properties of graphene, e.g., plasmon dispersions. 38 -- 40 A. Electronic States of Gapped Graphene with energy dispersion εγ(k) = γ(cid:112)(vF k)2 + ∆2 For gapped graphene, there exists a finite energy bandgap EG = 2∆G between the valence and conduction bands G, where γ = ±1 correspond to electron and hole state, respec- tively. The Hamiltonian matrix associated with this dispersion possesses an additional Σz term on top of the Dirac Hamiltonian for gapless graphene, 31,40 yielding (1) where r = (x, y), Σx,y,z are two-dimensional Pauli matrices, Σ0 is a (2 × 2) unit matrix, and VB(x) is a spatially- nonuniform barrier potential. Hg(r) = −ivF Σx,y · ∇r + VB(x) Σ0 + ∆G Σz , In general, the scattering-state solution for the Hamiltonian in Eq. (1) has a two-component (spinor) type of wave B(x)]T , where γ = sign[ε0(k) − VB(x)] = ±1 represents the function Ψγ(r) = exp(ikyy) Φγ(x) = exp(ikyy) [φγ electron-hole index and ε0(k) is the given energy of an incident electron. A(x), φγ For the case with a constant barrier potential V0, however, the Hamiltonian in Eq. (1) can be greatly simplified as (2)  V0 + ∆G vF k− vF k+ V0 − ∆G  , H(0) g (k θk) = where k = (k(0) in Eq. (2) gains the explicit form 31,41 x , ky) and k± = k(0) 3 x ± iky. In this case, the scattering-state wave function related to the Hamiltonian Ψ(0) γ (r) = 1(cid:112)2γ δε0(k)  (cid:112)δε0(k) + ∆G γ(cid:112)δε0(k) − ∆G eiθk  exp(ik(0) x x + ikyy) , (3) x ), δε0(k) ≡ ε0(k) − V0 ≥ ∆G for γ = +1, while δε0(k) ≤ −∆G for γ = −1. Here, two where θk = tan−1(ky/k(0) components of the wave function in Eq. (3) are not the same but they are still interchangeable for electrons and holes with γ = ±1. B. Finite-Difference Method for Tunneling Dirac Electrons For the Hamiltonian in Eq. (1), a pair of scattering-state equations within the barrier region are obtained as dφB(x) dx dφA(x) + ky φB(x) = − ky φA(x) = i vF i vF [ε0(k) − VB(x) + VD δ(x − xD) − ∆G] φA(x) , [ε0(k) − VB(x) + VD δ(x − xD) + ∆G] φB(x) . dx (4) Here, we consider a titled potential barrier under an applied electric field Edc, which gives rise to VB(x) = V0 − eEdc x in the barrier region, where V0 and Edc can be either positive or negative. Additionally, ky of electrons remains conserved during a tunneling process along the x direction. Moreover, a single disorder at 0 < x = xD < WB is assumed within the barrier region with a constant trap-potential amplitude −VD. Mathematically, we can divide the electron wave function corresponding to three separated regions. To the left of the potential barrier x < 0, we acquire the wave function  1 eiθk  exp(ik(0)  1 −eiθk  exp(−ik(0) x x) , (5) Φ<(x) = s(ε0) x x) + r(ε0) where s(ε0) and r(ε0) represent incoming and reflected wave-function amplitudes. To the right of the potential barrier x > WB, on the other hand, the wave function is found to be  1 eiθk(cid:48)  exp(ik(cid:48) xx) , Φ>(x) = t(ε(cid:48) 0) (6) where t(ε(cid:48) 0) is the transmitted wave-function amplitude. Results in Eqs. (5) and (6) can be applied to construct boundary conditions on both sides of a potential barrier. For the wave function within the barrier region, a finite-difference method can be employed to seek for a numerical solution of Eq. (4). Following the procedure adopted in Ref. [30] for a two-dimensional electron gas, we discrete the whole barrier region 0 ≤ x ≤ WB into NB (odd integer) equally spaced slabs, and each slab has the same width ∆0 = WB/NB. Therefore, two coupled differential equations in Eq. (4) can be solved simultaneously through a backward-iteration procedure in combination with two continuity boundary conditions at x = WB and x = 0. Especially, for ∆G = 0 we find the following backward iterative relation for 1 ≤ j ≤ NB + 1 and xj = (j − 1)∆0  φA(xj−1) φB(xj−1)  =  φA(xj) φB(xj) − ky∆0  φA(xj) −φB(xj)  + i∆0 vF (cid:2)ε0(k)− V0 + eEdcxj + VDδ(xj − xD)(cid:3) φB(xj) φA(xj)  . (7) By using Eq. (6), the first continuity boundary condition at xNB +1 = WB = NB∆0 leads to  φA(xNB +1) φB(xNB +1)  = t(ε(cid:48) 0(k(cid:48)))  1 eiθk(cid:48)  exp(ik(cid:48) xNB∆0) exp NB +1(cid:88) j=2  (cid:113)κ(xj) Θ(−κ(xj))/ xj(cid:54)=xD 4 , (8) ky∆0 (cid:113) where κ(xj) = [(1/vF ) (ε0(k) − V0 + eEdcxj)]2 − k2 y = ε0(k) + eEdcWB, and Θ(x) is a step function, i.e., Θ(ξ) = 1 for ξ > 0 while zero for others. Physically, the last exponential term in Eq. (8) does not affect the transmission coefficient if κ(xj) > 0, corresponding to a semi-classical regime. However, this term can significantly reduce the transmission coefficient, but not the reflection coefficient, if κ(xj) < 0, connecting to a quantum-tunneling regime. The backward iteration in Eq. (7) can be performed all the way down to x1 = 0. 0(k(cid:48)) = vF (cid:48)2 x + k2 y, ε(cid:48) k In a similar way, using Eq. (5) and another continuity boundary condition at x1 = 0, we find 1 4 r(ε0)2  s(ε0)2  ,  a2 + b2 + 2Re(ab∗eiθk)  =  1  + r(ε0)  1  = s(ε0)  φA(x1)  ≡ a2 + b2 − 2Re(ab∗eiθk) φB(x1) eiθk −eiθk  a b  . (9) (10) where we have defined the notations The transmission coefficient T (k, θk Edc), which is defined as the ratio of the transmitted to the incident probability current densities, 1,2 is given by T (k, θk Edc) = 0)2 t(ε(cid:48) s(ε0)2 , (11) since electrons on both sides of the potential barrier have the same group velocity vF . Numerically, it is easy to set 0) ≡ 1, then to find s(ε0) through Eq. (10) after having performed all the backward iterations, and finally obtain t(ε(cid:48) the ratio in Eq. (11). Using the transmission coefficient in Eq. (11), we are able to compute the tunneling electric current J0 per length, yielding (cid:88) k J0 = 4e A T (k, θk Edc) vF cos θk [f0(ε0(k)) − f0(ε0(k) + eEdcWB)] , (12) where A is the graphene sheet area, f0(x) = {1 + exp [(x − u0)/kBT ]}−1 is the Fermi function for thermal-equilibrium electrons at temperature T , and u0(T ) is the chemical potential of electrons. For a weak electric field, we have eEdcWB (cid:28) ε0(k), which leads to J0 ≈ 4e2vF U0A T (k, θk Edc) cos θk − ∂f0(ε0(k)) ∂ε0 , (13) were U0 = EdcWB represents the voltage drop across the potential barrier. If T is low, i.e., kBT (cid:28) EF with EF as the zero-temperature u0 or Fermi energy, we find (cid:88) k (cid:21) J0 ≈ 4e2vF U0A T (k, θk Edc) cos θk δ(vF k − EF ) = = Edc kF WB dθk T (kF , θk Edc) cos θk (cid:88) (cid:18) 2e2 k h (cid:19) 1 π π/2(cid:90) −π/2 π/2(cid:90) dk k δ(k − kF ) 0 −π/2 dθk T (k, θk Edc) cos θk (14) (cid:20) (cid:19) ∞(cid:90) U0 π h (cid:18) 2e2  , √ where kF = two-terminal sheet tunneling conductivity σ(kF ,Edc) (in units of 2e2/h), given by 42 πn0 is the Fermi wave vector and n0 is the areal electron density. Finally, we obtain the nonlinear σ(kF ,Edc) = J0Edc = kF WB π dθk T (kF , θk Edc) cos θk . (15) Specifically, for normal incidence of electrons with θk ≡ 0, we simply get σ0(kF ,Edc) = (kF WB) T (kF Edc). To simulate disorder effects on the tunneling of Dirac electrons, we introduce a normal distribution function and re- place the transmission coefficient T (kF , θk Edc) ≡ T (kF , θk Edc, xD) in Eqs. (11) and (15) by its average T (k, θk Edc), yielding T (k, θk Edc) = 1 ND dxD T (kF , θk Edc, xD) ρ(xD σ0) ≈ ∆0 ND T (kF , θk Edc, x∗ s) ρ(x∗ s σ0) , WB(cid:90) 0 NB(cid:88) s=2 5 (16) (17) (18) (19) π/2(cid:90) −π/2 where x∗ s = (s − 1)∆0, the introduced distribution function is assumed to be ρ(x∗ s σ0) = 1(cid:112)2πσ2 0 exp (cid:20) − (x∗ s − WB/2)2 2σ2 0 (cid:21) with the standard deviation σ0 = ∆0 and WB/2 = [(NB + 1)/2]∆0. In addition, the normalization factor in Eq. (16) is given by WB(cid:90) NB(cid:88) ND = dxD ρ(xD σ0) ≈ ∆0 ρ(x∗ s σ0) . For convenience, in numerical calculations we further approximate the delta-function in Eq. (7) by 0 s=2 with a broadening parameter Γ = ∆0. δ(xj − xD) ≡ δ(xj − x∗ s) ≈ Γ/π (xj − x∗ s)2 + Γ2 C. Results for Dirac-Electron Tunneling in Graphene In the previous subsections II A and II B, we have established a finite-difference numerical scheme and applied it to study tunneling transport of carriers through a biased potential barrier in graphene. As a validation, we first compare our finite-difference results with those from an analytical solution 2 for a square potential barrier VB(x) = V0. Figure 1 displays a comparison for calculated transmission coefficients T (ε, φk Edc = 0) as a function of incident angle φk using either an analytical solution 2 (black solid curves) or our finite-difference method presented in subsection II B (red dashed curves). The results in this figure clearly indicate that our finite-difference method in subsection II B is valid and can be applied to arbitrary potential profiles VB(x) including a biased potential barrier. The numerical results of T (ε, φk Edc) based on finite-difference method are presented in Fig. 2 as a function of incident angle φk for various values of bias field Edc. Our results indicate that the Klein paradox, i.e., T (ε, φk Edc) = 1 at φk = 0, persists for all considered bias field values, either positive or negative. When Edc is very small, large-angle resonant tunneling only occurs for Edc > 0 or a reduced potential barrier but not for Edc < 0 or an enhanced potential barrier. As Edc becomes large, however, resonant tunneling are squeezed into a narrow angle region around φk = 0 (see Fig. 1 for a comparison). Such variations observed in T (ε, φk Edc) can be attributed to the modification of a barrier potential profile VB(x) by a bias field compared with a square potential barrier VB(x) = V0. Figure 3 displays density plots of T (ε, φk Edc) as functions of both incident energy ε and incident angle φk with six different values for bias field Edc. We take the case with Edc = 0 as a starting point, where the Klein paradox and 6 FIG. 1: (Color online) Comparison of calculated transmission coefficients T (ε, φk Edc = 0) as a function of incident angle φk based on either an analytical solution (black-solid curves) or finite-difference approach (red-dashed curves) for four different barrier thickness WB = 1 nm (upper-left), 50 nm (upper-right), 110 nm (lower-left) and 5 nm (lower-right), where V0 = 285 meV and ε/V0 = 1.25 are chosen for calculations. FIG. 2: (Color online) Polar plots for transmission coefficient T (ε, φk Edc) as a function of incident angle φk for different bias values Edc, where both results for an enhanced potential barrier Edc < 0 (left) and a reduced potential barrier Edc > 0 (right) are shown in this figure for a full comparison. Here, ε = 400 meV (left) and 80 meV (right) are chosen. The other parameters are the same as those in Fig. 1. 7 FIG. 3: (color online) Density plots of T (ε, φk Edc) as functions of both ε and φk, WB = 110 nm, V0 = 285 meV and different values of Edc are assumed. FIG. 4: (Color online) Ratio of tunneling resistance R/R0 (inverse conductance) for ballistic electrons in graphene, calculated from Eq. (15), through a biased potential barrier with different values of bias fields. Here, R0 is the resistance for normal incidence with φk = 0 and results for both positive (left) and negative (right) biases are presented for comparisons. collimation effect exist with many sharp resonances (branching and needling features) observed. As Edc increases from zero to 250 V/cm, these resonant branching and needling features are greatly obscured although the Klein paradox persists. On the other hand, as negative Edc increases from zero to −350 V/cm, both branching and needling regions expand significantly to higher incident energy range of electrons. The calculated tunneling coefficient T (ε, φk Edc) can be put into Eq. (15) to find tunneling conductivity or resistivity (its inverse) of ballistic electrons through a biased potential barrier in graphene. Here, the conductivity strongly depends on the bias field Edc due to nonlinear nature of tunneling transport. For ballistic Dirac electrons in the absence of a potential barrier, their conductivity should be integer multiple of 2e2/h, as indicated by Eq. (15). In the presence of scattering by impurities or phonons, the occurring resistive force can give rise to a bias-dependent 8 FIG. 5: (Color online) Transmission coefficient T (ε, φk V0, α) at Edc = 50 V/cm as a function of incident angle φk for various barrier widths WB. conductivity which is accompanied by a joule heating of electrons. Here, however, a bias-dependent conductivity is induced by a tunneling barrier which elastically and coherently reflects incoming electrons, leading to a destructive interference. Such a behavior can be attributed to a strong bias modulation (α dependence) of tunneling coefficient T (ε, φk Edc) in Eq. (15). In Fig. 4, we present the calculated resistance ratios as functions of incident electron energy for different positive and negative biased potential barriers. We find from this figure that the resistance peak height decreases with increasing positive Edc and the peak position shifts down to lower incident energy ε at the same time. For increasing negative Edc, on the other hand, the peak position shifts upward with ε but the peak height remains nearly unchanged. Furthermore, the resistance peak is broadened with increasing Edc, and the broadening effect becomes stronger for negative Edc values. As shown in Fig. 5, the transmission coefficient T (ε, φk Edc) at Edc = 50 V/cm is suppressed only for large incident angles φk with increasing barrier width WB due to enlarged switching from a semi-classical regime to a quantum- tunneling regime inside barrier region, as well as due to interference effect in reflections from both barrier edges. Meanwhile, T (ε, φk Edc) for φk around zero remains unchanged, leading to enhanced collimation of Dirac-electron tunneling. III. WKB APPROXIMATION FOR WAVE FUNCTION AND TRANSMISSION In Section II, we have demonstrated a finite-difference approach for calculating Dirac-electron tunneling through an arbitrary potential barrier. In order to gain physics behind nonlinear transport of ballistic Dirac electrons tunneling, we introduce WKB approximation so as to analyze the dynamics of tunneling Dirac electrons in an explicit form beyond the numerical solution. For this purpose, let us consider a tilted potential barrier, as shown in Fig. 7, with the potential VB(x) = V0 + αx, while VB(x) = 0 outside of the barrier region. For this case, the effective x−dependent wave vector k(x) can be written as k(x) = ε − VB(x) vF ε − V0 vF = − ax , (20) where ε is the energy of an incoming particle, which is conserved for an elastic scattering with the barrier region, and a = α/(vF ). For a square potential barrier, as considered in Ref. [2], we simply set a = 0 and will use it to build up our perturbation theory below. We first introduce a unitary transformation for a gapless Dirac Hamiltonian, i.e., a π/2-rotation around the x−axis, as employed in Ref. [14], for simplification. This leads to the mixed eigen-function Φ(x ky) = [φ+, φ−]T , where √ φ± = (φB ± φA)/ 2. If VB(x) ≡ V0 for a constant potential barrier, we find the eigen-function  eikyy =  φ+ φ−  eiθk + 1 eiθk − 1 1 2  eik(0) Ψ0(x, y ky) = x x+ikyy , (cid:113) 9 (21) where θk = tan−1(ky/k(0) wave-function amplitude is independent of x and y. x ) is the in-plane angle in the momentum space, k(0) x = [(1/vF )(ε − V0)]2 − k2 y , and the In the most general case, the wave-function amplitudes ψ±(x, y) satisfy the following equations 14 ∓ i ∂ψ± ∂x ∓ ∂ψ∓ ∂y = k(x) ψ± . (22) Throughout our derivation, the translational symmetry in the y direction is always kept since our potential VB(x) varies only along the x−axis. Therefore, we can simply write down ψ±(x, y) = exp(ikyy) φ±(x). This simplify Eq. (22) into (23) where ∂xφ(x) ≡ dφ(x)/dx. As a special case, one can easily verify that the solution Ψ0(x, y ky) in Eq. (21) satisfies the above equation as VB(x) = V0 or a = 0 in Eq. (20) is taken for k(x). ∓ i ∂xφ±(x) ∓ iky φ∓(x) = k(x) φ±(x) , A. WKB Semi-Classical Approach The general form of semi-classical WKB expansion for a tunneling-electron wave function Ψ(x, y ky) can be ex- pressed as 43 (cid:90) where S∆(x) = x Ψ(x, y ky) = e(i/) S∆(x) ∞(cid:88) (−i)s Ψs(x, y ky) , s=0 (24) dξ kx(ξ) represents an action. Here, we will only consider the leading s = 0 term in Eq. (24) and obtain (cid:21) where kx(ξ) = (1/vF )(cid:112)[ε − VB(ξ)]2 − (vF k∆)2 and k∆ = (1/vF )(cid:112)(vF ky)2 + ∆2 (cid:27) more, we have also introduced the following two dimensionless quantities in Eq. (25) Ψ±(x, y ky) = eiΘ∆(x ky) ± 1 C∆(x ky) S∆(x) (cid:20) i (cid:16) (cid:17) exp 1 2 1 C∆(x ky) = kx(x) Θ∆(x ky) = tan−1 ∆G[ε − ∆G − VB(x)] (vF )2 ky k−(x) + i vF k+(x) ε − ∆G − VB(x) (cid:21) , (cid:26) (cid:20) eikyy , (25) G is independent of ξ. Further- , (26) where k±(x) = kx(x) ± iky. It is straightforward to verify that the above solution becomes equivalent to that of gapless graphene as ∆G = 0, given by 14 Ψ±(x, y ky) = k+(x) ± k(x) 2(cid:112)k(x) kx(x) exp x(cid:90) i  eikyy . dξ kx(ξ) 10 (27) where k(x) has already been given by Eq. (20). As an electron moves uphill with increasing potential, the sum of its potential and kinetic energies remains as a constant. Therefore, the kinetic energy of the electron decreases on its way. For this case, we need define a turning point of a semi-classical trajectory, at which kx(x) = 0 but the total kinetic energy is still positive due to ky (cid:54)= 0. We first find the turning point x0 = (ε − V0)/α from k(x) = 0 in Eq. (20), where a Dirac electron turns into a Dirac hole. Moreover, the range corresponding to x − x0 < ξc becomes a classically forbidden region in which kx(x) become imaginary, where ξc = vF ky/α for ∆G = 0. If this forbidden region lies entirely within the biased potential barrier region, the transmission coefficient T (α ky) is found to be 43 −2 x0+ξc(cid:90) x0−ξc (cid:113) dξ k2 y  = exp (cid:18) (cid:19) − 4 a k2 y T (α ky) (cid:118) exp , (28) which is a clear manifestation of the conservation of the Klein paradox for a biased potential barrier layer. For the case with ∆G > 0, the result in Eq. (28) could be generalized to −2 x0+ξc(cid:90) (cid:113) dξ k2 ∆(ξ) x0−ξc  = exp (cid:26) − 4 a y + (∆G/vF )2(cid:3)(cid:21) (cid:2)k2 , (29) T∆(α ky) (cid:118) exp (cid:113) ξc = 1 α (vF ky)2 + ∆2 G . Therefore, Klein paradox will not exist for any α and ky values. In addition, an exact solution for the wave function in this case could also be obtained by using the results in Ref. [14], as demonstrated in Appendix A. B. Perturbative Solution at Low Bias We would like to emphasize that all the results obtained in previous subsection suffer a limitation, i.e., they are valid only if the electron-to-hole switching occurs inside the barrier region. However, this becomes invalid if either the slope α of a potential profile or the barrier width WB becomes very small. To seek for a perturbative solution within a barrier layer, we first assume a very small slope α to ensure a = α/vF (cid:28) k2(x). We further assume ε > VB(x) so that particle-hole switching will not occur. As a result, the wave function takes the form ψA,B(x, y) = φA,B(x) exp(ikyy) and Eq. (23) can be applied to find solution for φA,B(x). In this case, however, a π/2-rotation for φ±(x) is not needed. For VB(x) = V0 + αx, the electron momentum is k(x) = k0 − ax, where k0 = (ε − V0)/vF and a = α/vF . From this, we find ∂xk(x) = −a, which becomes a small parameter in expansion. Based on these assumptions, we acquire a second-order differential equation with respect to the first wave-function component φA(x), yielding ∂2 xφA(x) + a k0 − ax ∂xφA(x) + (k0 − ax)2 − aky k0 − ax − k2 y φA(x) = 0 . (30) Considering the fact that a (cid:28) 1, we approximate the above equation as (cid:18) a k0 (cid:19) + a2x k2 0 (cid:18) ky k0 (cid:19) (cid:18) + a2 x2 − xky k2 0 (cid:19)(cid:21) φA(x) = 0 . (31) ∂2 xφA(x) + ∂xφA(x) + 0 − k2 k2 y − a + 2k0x (cid:21) (cid:20) (cid:20) 11 A (x) + ··· , and include only the terms up to the first non-vanishing linear correction to φA(x). Therefore, we get the 0th and 1st order equations, respectively, Now, we look for a perturbative solution of Eq. (31) in the form of φA(x) = φ(0) A (x) + a2 φ(2) A (x) + a φ(1) A (x) +(cid:0)k2 A (x) +(cid:0)k2 xφ(0) ∂2 xφ(1) ∂2 y 0 − k2 0 − k2 y (cid:1) φ(0) (cid:1) φ(1) a0 : a1 : A (x) = 0 , 1 k0 A (x) + (cid:18) ky k0 (cid:19) + 2k0x ∂xφ(0) A (x) − φ(0) A (x) = 0 . Moreover, making use of the relation in Eq. (23) for two components of the wave function, i.e., φB(x) = ∂xφA(x) − ky φA(x) i(k0 − ax) ≡ φ(0) B (x) + a φ(1) B (x) , we find φ(0) B (x) = φ(1) B (x) = −i k0 −i k2 0 (cid:104) (cid:110) (cid:105) A (x) A (x) − ky φ(1) (cid:104) A (x) − ky φ(0) ∂xφ(1) ∂xφ(0) k0 , (cid:105) (cid:104) A (x) + x ∂xφ(0) A (x) − kyφ(0) A (x) . (cid:105)(cid:111) For the 0th order solution, we are dealing with the bias-free case having a = 0 or a square potential barrier VB(x) = V0. From Eq. (32) we easily find its solution φ(0) A (x) = c(0) 1 eikxx + c(0) 2 e −ikxx with kx = (cid:113) which is a superposition of the forward and backward plane waves. 2 In this case, from Eq. (35) the corresponding solution for the second component of the wave function is given by (cid:18) kx + iky (cid:19) γ k0 (cid:16) φ(0) B (x) = c(0) 1 ≡ γ 1 eiθk eikxx − c(0) c(0) 2 e eikxx + c(0) 2 −iθk e 0 − k2 k2 y , (cid:19) −ikxx e (cid:18)−kx + iky −ikxx(cid:17) γ k0 , (32) (33) (34) (35) (36) (37) (38) (39) where γ = sign (ε − V0) = ±1 is the electron-hole index within the barrier region and θk = tan−1 (ky/kx) for Dirac electrons inside the barrier region. Assuming ε > V0, we always have γ > 0 and no electron-hole switching will occur. Two constants c(0) 1 in Eq. (38) can be determined by boundary conditions at both sides of a barrier layer. and c(0) 2 The incoming wave function can be written as 2,31,32  eik(0)  eiφk/2 y and φk = tan−1(cid:16) e−iφk/2 1√ 2 x x , (cid:17) Φi(x) = (cid:113) where VB(x < 0) = 0, k(0) Here, the transversal electron wave vector ky remains to be a constant during the whole tunneling process. is the incident angle of Dirac electrons. ky/k(0) x x = (ε/vF )2 − k2 We first notice that c(0) 1 and c(0) 2 boundary condition at x = 0, giving rise to in Eqs. (37) and (38) are not normalized, and they can be determined by the first c(0) 1 = (cid:2)γ + ei(φk+θk)(cid:3)(cid:0)1 + e2iφk(cid:1) ei(2kxWB +θk) (cid:0)1 + e2iφk(cid:1)(cid:0)γ eiθk − eiφk(cid:1) 2 + γ ei(φk+θk)(cid:105) D(kx, φk θk) D(kx, φk θk) , , c(0) 2 = D(kx, φk θk) = γ + ei(φk+θk)(cid:104) + 2 ei(2kxWB +φk+θk) [γ cos (θk − φk) − 1] . (40) 1 and c(0) Here, c(0) in Eq. (37) play the role of transmission and reflection amplitudes within the barrier region. Using the result in Eq. (40) and the second boundary condition at x = WB as well, we can further calculate the transmission coefficient t(0) as 2 t(0) = γ cos(kxWB) cos θk cos φk + i sin(kxWB) (γ sin θk sin φk − 1) γ e−ikxWB cos θk cos φk , (41) which is identical to the corresponding results for a square potential barrier in Ref. [2], as expected. In a similar way, we can find the 1st order solution from Eq. (33) for φ(1) A (x), yielding 12 φ(1) A (x) = c(1) F(x kx, θk) = (cid:16) − kxe −ikxx + F(x kx, θk) 1 eikxx + c(1) 2 e e−ikxx 4k3 x + i c(0) (cid:110) 1 e2ikxx(cid:16) − ikxe+θk (2kxx + i) + k0 [2kxx (kxx + i) − 1] −θk (2kxx − i) + k0 [2kxx (1 + ikxx) − i] c(0) 2 with (cid:17) (cid:17)(cid:111) . (42) Here, the two new undetermined constants c(1) 1 and c(0) 2 barrier region. By using these calculated first wave-function components φ(0) is straightforward to find the second wave-function component φ(1) becomes a bit tedious to write out. are completely different from the zero-order constants c(0) 1 in Eq. (40), and they represent the first-order corrections to transmission and reflection amplitudes inside the A (x) in Eqs. (37) and (42), it B (x) from Eq. (36) although its explicit expression A (x) and φ(1) and c(1) 2 Now, we are able to determine the coefficients c(1) 1 in Eq. (42) and the correction to the transmission coefficient t(0) in Eq. (41). For this, we would like to write the transmission and reflection coefficients as t = t(0) +a t(1) and r = r(0) + a r(1), corresponding to the wave functions in Eqs. (36) and (42). Using the two boundary conditions at x = 0 and x = WB, we arrive at two equations for r(1) and t(1), given by and c(1) 2  1 −e−iφk  eik(0) r(1)  1 eiφk  =  φ(1)  φ(1) A (x = 0 kx, θk) B (x = 0 kx, θk) φ(1) A (x = WB kx, θk) B (x = WB kx, θk) φ(1)  ,  eikxWB . and t(1) x WB = (43) Finally, the transmission amplitude T (ε, φk V0, α) can be simply found from T (ε, φk V0, α) = t(0) + at(1)2, where t(0) and t(1) are given by Eqs. (41) and (43), respectively. The numerical results from Eqs. (28) and (29) for a large electric bias are presented in panels (a) and (b) of Fig. 6. Here, a large graphene gap ∆G significantly suppresses T (ε, φk V0, α) for all values of a, as shown in Fig. 6(a), while the increase of electric bias a enhances T (ε, φk V0, α) for all values of ∆G, as seen in Fig. 6(b). From Figs. 6(c) and 6(e), we find that the full transmission for a head-on collision remains unchanged even under an electric bias a (cid:54)= 0. For small a values, the electron-hole transition does not take place within the barrier region. Instead, a finite a only slightly modifies the resonances of oblique tunneling but not the Klein paradox for the head-on collision. IV. SUMMARY AND REMARKS In summary, we have developed a numerical approach for accurately calculating nonlinear tunneling transport of ballistic Dirac electrons through an arbitrary barrier potential in graphene. Here, our barrier-potential profile mimics a conventional MOSFET configuration where the square-barrier potential comes from the gate potential, and the barrier tilting connects to a source to drain applied bias. In addition, the barrier-potential profile can be tuned by applying a bias, and meanwhile the barrier height can be independently controlled by a gate voltage. Our research results can be applied to both sharp and smooth in-plane p-n junctions of either bipolar or unipolar devices. In the 13 FIG. 6: (Color online) Transmission amplitude T (ε, φk) in graphene for fixed V0/EF = 2 and various potential biases specified by different a values. Panels (a) and (b) display T (ε, φk) from Eq. (29) for gapped graphene as functions of ∆G and ak2 F , respectively, with fixed ky = 0.5 kF . Panels (c)-(e) present density plots for T (ε, φk) from Eqs. (41) and (43) for gapless graphene ∆G = 0 as functions of ε/EF and φk in (c), (e) and functions of WBkF and φk in (d), (f ) with ε/EF = 1 for a = 0 −2 F (bottom row). Here, EF = vF kF = 6.28 meV is taken for the energy unit and kF is the unit for (middle row) and a = 0.1 k wave vector. ballistic limit, we propose a mechanism by using barrier profile modulation for Dirac-electron tunneling, allowing an n-p-n junction to be smoothly converted into a p-n junction with a proper choose of an applied bias. In order to gain insight about the tunneling mechanism of Dirac electrons, we have introduced a perturbation theory for electron transmission through a slanted potential barrier with a small tilting compared to the inverse barrier width and characteristic electron momenta. In addition, we have derived a set of equations, corresponding to different orders of expansion parameter, and obtained analytical solutions of these equations. Furthermore, we have demonstrated how the tunneling resonances of a square potential barrier are affected under a finite bias voltage. Physically, we have extended a previously developed WKB theory for electron transmission in the opposite limit of a large bias, in which electron-to-hole switching occurs within the barrier region. Finally, a finite energy gap in graphene is included and we have shown that both head-on and skew transmissions will be suppressed exponentially due to existence of an energy gap and a large transverse momentum. Another interesting implication is tunable filtering of Dirac electrons for nearly normal incidence, which might be utilized to design electronic lenses. The uniqueness of our mechanism is that we can specify a range for incident electron energies for focusing. Moreover, the electron resistance could be reduced and conductance minima can be shifted in energy just by controlling bias polarity, barrier height and the strength of bias field. Therefore, our model ()e()f()c0.0()a()b0.80.20.40.40.30.20.10.02.01.01.251.50()d0.04.01.02.00.04.01.02.02.05.08.014.02.05.08.014.00.60.50.40.30.214 FIG. 7: (Color online) WKB schematics for the biased potential VB(x) = V0 + αx in the region of 0 ≤ x ≤ WB and two classical turning points at x = x0 ± ξc. system can be employed to tune the refractive index of such a potential barrier in ballistic-electron optics. All these revealed properties are expected extremely valuable for the development of novel electronic and optical graphene-based devices. Appendix A: Exact Wave Function In this Appendix, we seek for an exact solution for the electron/hole wave function in the finite-slope region of a barrier, as seen in Fig. 7, with potential VB(x) = V0 + αx. If two boundaries of the barrier region stay far away from the electron-to-hole crossing point, i.e., k(x0) ≈ 0, the wave function could be written as 14 (cid:26) (cid:21) (cid:20) F(η, ζ) G(η, ζ) (cid:21)(cid:27) (cid:20) F (cid:63)(η, ζ) G(cid:63)(η, ζ) + c2 Ψ(B)(x ky) = c1 where η(x) = (x − x0) a, the symbol (cid:63) means taking complex conjugation, and the two arbitrary constants c1 and c2 will be fixed by the boundary conditions on each side of the barrier region. Moreover, two functions F(η, ζ) and G(η, ζ) in Eq. (A1) can be expressed by a Kummer confluent hypergeometric function M(a, b z) as 14 √ √ a, ζ(ky) = ky/ eikyy , (A1) (cid:18) (cid:19) (cid:18) (cid:19) F(η, ζ) = exp − i 2 G(η, ζ) = −ζη exp (cid:18) η2 M − i 2 η2 (cid:18) − i 4 M ζ 2, 1 2 1 − i 4 (cid:19) , (cid:12)(cid:12)(cid:12) iη2 . (A2) The wave functions outside of the barrier region are easily obtained for the incoming and reflected waves, yielding (cid:19) (cid:12)(cid:12)(cid:12) iη2 ζ 2, 3 2 (cid:20) −e−iφk (cid:21) ±1 (cid:21) (cid:20) eiφk ±1 Ψ(L)(x k) = 1 2 eik(0) x x eikyy + rk 2 where φk = tan−1(ky/k(0) x ). Similarly, for the transmitted wave we have (cid:21) (cid:20) eiθk(cid:48) ±1 Ψ(R)(x k (cid:48) ) = tk(cid:48) 2 eikxx eikyy , −ik(0) e x x eikyy , (A3) (A4) where θk(cid:48) = tan−1(ky/kx). The transmission coefficient tk(cid:48) and the reflection coefficient rk can be obtained by matching the wave functions at two boundaries at x = 0 and x = WB, i.e., Ψ(L)(x = 0 k) = Ψ(B)(x = 0 ky) and Ψ(B)(x = WB ky) = Ψ(R)(x = WB k ). Therefore, we acquire four equations for these two-components wave functions, which can be used to determine four unknowns rk, c1, c2, and tk(cid:48), and the calculated tk(cid:48) will be further applied for evaluating the transmission Tk(cid:48) = tk(cid:48)2. (cid:48) Here, we would like to emphasize that although the obtained solution in Eq. (A1) is exact, it holds true only for a very thick potential barrier with 0 (cid:28) x0 (cid:28) WB. Additionally, using this approach we can not address the limiting case with a small slope a → 0. For the boundaries of a very thick potential barrier with a substantial slope α, we find vary large absolute value of η(x) = (x − x0) √ a, and the wave function is calculated as 15 (cid:21) (cid:20) 0 η→∞ Ψ(B)(x, ky) = η→−∞ Ψ(B)(x, ky) = eikyy lim lim 1 (cid:20) exp (cid:105)(cid:20) 0 (cid:21) y/a(cid:1). This result is the same as that obtained from the a which gives rise to the transmission Tk(cid:48) = exp(cid:0)−πk2 (cid:104)− π (cid:20) 0 (cid:21)(cid:27) (cid:20) i − i 2 − i 2 η2(x) eikyy , exp (cid:26) η2(x) + const k2 y 2a (A5) exp η2(x) , exp 1 1 2 (cid:20) (cid:21) (cid:21) (cid:21) semi-classical theory. 1 M. Katsnelson and K. Novoselov, Solid State Communications 143, 3 (2007). 2 A. C. Neto, F. Guinea, N. M. Peres, K. S. Novoselov, and A. K. Geim, Reviews of modern physics 81, 109 (2009). 3 S. D. Sarma, S. Adam, E. Hwang, and E. Rossi, Reviews of modern physics 83, 407 (2011). 4 K. S. Novoselov, A. K. Geim, S. Morozov, D. Jiang, M. I. Katsnelson, I. Grigorieva, S. Dubonos, Firsov, and AA, nature 438, 197 (2005). 5 K. Wang, M. M. Elahi, L. Wang, K. M. Habib, T. Taniguchi, K. Watanabe, J. Hone, A. W. Ghosh, G.-H. Lee, and P. Kim, Proceedings of the National Academy of Sciences 116, 6575 (2019). 6 T. Low and J. Appenzeller, Physical Review B 80, 155406 (2009). 7 M. S. Jang, H. Kim, Y.-W. Son, H. A. Atwater, and W. A. Goddard, Proceedings of the National Academy of Sciences 110, 8786 (2013). 8 Q. Wilmart, S. Berrada, D. Torrin, V. H. Nguyen, G. F`eve, J.-M. Berroir, P. Dollfus, and B. Pla¸cais, 2D Materials 1, 011006 (2014). 9 K. S. Novoselov, E. McCann, S. Morozov, V. I. Falko, M. Katsnelson, U. Zeitler, D. Jiang, F. Schedin, and A. Geim, Nature physics 2, 177 (2006). 10 Y. Zhang, Y.-W. Tan, H. L. Stormer, and P. Kim, nature 438, 201 (2005). 11 T. Ohta, A. Bostwick, T. Seyller, K. Horn, and E. Rotenberg, Science 313, 951 (2006). 12 V. V. Cheianov and V. I. Falko, Physical review b 74, 041403 (2006). 13 A. V. Shytov, M. S. Rudner, and L. S. Levitov, Physical review letters 101, 156804 (2008). 14 E. Sonin, Physical Review B 79, 195438 (2009). 15 S. Chen, Z. Han, M. M. Elahi, K. M. Habib, L. Wang, B. Wen, Y. Gao, T. Taniguchi, K. Watanabe, J. Hone, et al., Science 353, 1522 (2016). 16 J. Cayssol, B. Huard, and D. Goldhaber-Gordon, Physical Review B 79, 075428 (2009). 17 P. E. Allain and J.-N. Fuchs, The European Physical Journal B 83, 301 (2011). 18 V. T. Phong and J. F. Kong, arXiv preprint arXiv:1610.00201 (2016). 19 V. V. Cheianov, V. Fal'ko, and B. Altshuler, Science 315, 1252 (2007). 20 D. Dahal and G. Gumbs, Journal of Physics and Chemistry of Solids 100, 83 (2017). 21 C. Beenakker, Physical review letters 97, 067007 (2006). 22 R. N. Sajjad and A. W. Ghosh, Applied Physics Letters 99, 123101 (2011). 23 M. M. Elahi, K. Masum Habib, K. Wang, G.-H. Lee, P. Kim, and A. W. Ghosh, Applied Physics Letters 114, 013507 (2019). 24 J. Tworzyd(cid:32)lo, C. Groth, and C. Beenakker, Physical Review B 78, 235438 (2008). 25 A. R. Hern´andez and C. H. Lewenkopf, Physical Review B 86, 155439 (2012). 26 G. Xu, X. Xu, B. Wu, J. Cao, and C. Zhang, Journal of Applied Physics 107, 123718 (2010). 27 T. Low, S. Hong, J. Appenzeller, S. Datta, and M. S. Lundstrom, IEEE Transactions on Electron Devices 56, 1292 (2009). 28 R. Logemann, K. Reijnders, T. Tudorovskiy, M. Katsnelson, and S. Yuan, Physical Review B 91, 045420 (2015). 29 N. Ghobadi and Y. Abdi, Current Applied Physics 13, 1082 (2013). 30 D. Huang, A. Singh, and D. Cardimona, Physics Letters A 259, 488 (1999). 31 A. Iurov, G. Gumbs, O. Roslyak, and D. Huang, Journal of Physics: Condensed Matter 24, 015303 (2011). 32 A. Iurov, G. Gumbs, O. Roslyak, and D. Huang, Journal of Physics: Condensed Matter 25, 135502 (2013). 33 S. Y. Zhou, G.-H. Gweon, A. Fedorov, d. First, PN, W. De Heer, D.-H. Lee, F. Guinea, A. C. Neto, and A. Lanzara, Nature materials 6, 770 (2007). 34 G. Giovannetti, P. A. Khomyakov, G. Brocks, P. J. Kelly, and J. Van Den Brink, Physical Review B 76, 073103 (2007). 35 R. Balog, B. Jørgensen, L. Nilsson, M. Andersen, E. Rienks, M. Bianchi, M. Fanetti, E. Laegsgaard, A. Baraldi, S. Lizzit, et al., Nature materials 9, 315 (2010). 36 O. Kibis, Physical Review B 81, 165433 (2010). 37 K. Kristinsson, O. V. Kibis, S. Morina, and I. A. Shelykh, Scientific reports 6, 20082 (2016). 38 N. Horing, A. Iurov, G. Gumbs, A. Politano, and G. Chiarello, Low-dimensional and nanostructured materials and devices (2016). 39 A. Iurov, G. Gumbs, D. Huang, and V. Silkin, Physical Review B 93, 035404 (2016). 40 P. Pyatkovskiy, Journal of Physics: Condensed Matter 21, 025506 (2008). 41 A. Iurov, G. Gumbs, D. Huang, and L. Zhemchuzhna, Journal of Applied Physics 121, 084306 (2017). 42 V. Vargiamidis and P. Vasilopoulos, Applied Physics Letters 105, 223105 (2014). 43 V. Zalipaev, C. Linton, M. Croitoru, and A. Vagov, Physical Review B 91, 085405 (2015). 16
1103.4899	2	1103	2011-04-26T08:49:27	Dirac Electrons in Molecular Solids	[ "cond-mat.mes-hall" ]	Electrons in solids are characterized by the energy bands, which indicate that electrons are considered to be "elementary particles" with specific effective masses and g-factors reflecting features of each solid. There are cases where these particles obey dispersion relationship similar to those of Dirac electrons. Examples include graphite and bismuth both of which are known for many years, together with graphene, a single layer of graphite, recently addressed intensively after its realization. Another recent example is a molecular solid, alpha-ET2I3, which is described by an equation similar to Weyl equation with massless Dirac cones but the coordinate axis is tilted because of the location of cones at off-symmetry points. Orbital susceptibility of such Dirac electrons in graphite and bismuth has been known to have striking features not present in ordinary band electrons but resulting from the inter-band matrix elements of magnetic field. Results of theoretical studies on not only orbital susceptibility but also Hall effect of such Dirac electrons in molecular solids with tilting are introduced in this paper.	cond-mat.mes-hall	cond-mat	Dirac Electrons in Molecular Solids Hidetoshi Fukuyama Faculty of Science and Research Institute for Science and Technology Tokyo University of Science, Tokyo 162-8601 Akito Kobayashi Institute for Advanced Research, Nagoya University, Nagoya 464-8602 Yoshikazu Suzumura Department of Physics, Nagoya University, Nagoya 464-8602 [email protected] Abstract. Electrons in solids are characterized by the energy bands, which indicate that electrons are considered to be “elementary particles” with specific effective masses and g-factors reflecting features of each solid. There are cases where these particles obey dispersion relationship similar to those of Dirac electrons. Examples include graphite and bismuth both of which are known for many years, together with graphene, a single layer of graphite, recently addressed intensively after its realization. Another recent example is a molecular solid, -ET2I3, which is described by an equation similar to Weyl equation with massless Dirac cones but the coordinate axis is tilted because of the location of cone s at off-symmetry points. Orbital susceptibility of such Dirac electrons in graphite and bismuth has been known to have striking features not present in ordinary band electrons but resulting from the inter-band matrix elements of magnetic field. Results of theoretical studies on not only orbital susceptibility but also Hall effect of such Dirac electrons in molecular solids with tilting are introduced in this paper. 1. Introduction Energy bands of electrons in solids have diversity reflecting particular features of solids and disclose specific energy-momentum relationship, which can be viewed as particular “elementary particles” at least locally in momentum space with effective masses and g-factors generally with spatial anisotropy. There are cases where these elementary particles obey equations similar to Dirac electrons. In this context graphite[1] and bismuth (Bi)[2, 3] have been known for many years; 2x2 Weyl equation for the former and 4x4 Dirac equation for the latter. Main physical interests in these systems had been the orbital susceptibility in weak magnetic field especially in bismuth (Bi) [4]. The difficulty with Landau-Peierls formula based on the pioneering work by Peierls [5], which takes only intra-band contribution of the effects of magnetic field on Bloch electrons, had long been noticed in view of experimental findings and roles of inter-band effects have been recognized [6]. Accordingly since the first derivation of orbital susceptibility by Wilson [7] there have been many efforts to derive exact formula of orbital susceptibility of Bloch electrons [8]. Theoretical understanding of experimentally observed anomalous orbital susceptibility has been achieved for graphite[9] and Bi[10] by taking these inter-band contributions of vector potential based on k-p Hamiltonian (Luttinger–Kohn representation) [11] which is suited to explore such intricate effects in a precise and transparent way. A simple and exact formula [12] for the orbital susceptibility for Bloch electrons has been derived based on this k-p Hamiltonian. Besides orbital susceptibility, which is thermodynamic, there is general interest in such inter-band contribution on transport properties, typically weak field Hall conductivity and then Hall effect [13, 14]. These have been explicitly studied for the recent example of Dirac electrons in graphene [15, 16], a single layer of graphite. Based on the proper scheme [17] in extracting correctly (in a gauge-invariant way) the contributions in the linear order of vector potential in the Kubo formula Hall conductivity has been evaluated [18] together with conductivity and orbital susceptibility at absolute zero as a function of chemical potential under the assumption of energy independent electron damping (lifetime). The results indicate that Hall conductivity vanishes and change signs at the crossing of Dirac cone. This feature at the crossing energy is very natural in view of the symmetry between “electrons” and “holes”. This fact is, however, completely opposite to the conventional understanding based on the band electrons, since the effective carrier density would be vanishing if the Fermi energy is located at the crossing. 2. Experimental puzzles in molecular solids, -ET2I3 There had been various puzzles in the experimental results in -ET2I3 [19], which is one of polytypes of charge–transfer molecular solids consisting of ET molecules, ET2X, with X being anions. At ambient pressure the resistivity of-ET2I3 shows a sharp increase at around 150K as the temperature is lowered [19]. This metal-insulator transition is suppressed by external pressure and vanishes completely at around 20 GPa leading to almost temperature independent resistivity below room temperature. The measurement of Hall effect in this pressure region of temperature-independent resistivity discloses that the effective carrier number decreases very strongly (several orders of magnitude between room temperature and few degree K) as temperature is lowered , whereas the mobility increases accordingly. Even more, there are cases where Hall constant changes sign at low temperature very sharply [20], which is sample-dependent. 3. Electronic states of molecular solids, -ET2I3: tilted Weyl particles It is now established that electronic states of molecular solids are, in general, well described by the tight-binding approximation based on molecular orbitals [21]. Molecular solids of charge transfer type with particular ratio of 2:1 between cations and anions, which lead to the quarter-filling of holes, have a great diversity in electronic properties. Microscopic and systematic theoretical understanding of the diversity is now possible by the extended Hubbard model which is based on the tight -binding approximation in terms of molecular orbitals for the band structure together with Coulomb interactions U on the same molecules and V between them [22,23]. The characteristic electronic states of quarter-filled band are due to coordinated effects of dimerization and frustration (from square to triangular) of lattice structures resulting in two typical insulating states, dimer Mott and charge ordering [23]. Typical example of the former is the type which shows competition between antiferromagnetic Mott insulator and superconductivity [24] and the realization of spin liquid [25] for the first time, while -ET2I3 is the example of the latter[26]. For -ET2I3, using the transfer energy experimentally obtained under uniaxial strain [27], the gapless behaviour in the density of states has been disclosed [28]. Further, detailed analysis of such band structure within the tight-binding approximation [29] and later by first-principles calculations [30, 31 ] have led to the remarkable finding of the possible accidental degeneracy of bands (Dirac cones). This finding has led to the breakthrough of microscopic understanding of mysteries of this system. Since Dirac cones in this case are located at two points (with inversion symmetry) in the Brillouin zone away from symmetry points in contrast to graphenes, the band energy is described by the following equation similar to Weyl equation but with diagonal component linear in wave vector , k, measured from the crossing point leading to “tilting “ of cones[32]. (1) Here ℏ=h/2 taken as unit with h being the Plank constant and are 2x2 matrices with 0 being unit matrix and the Pauli matrix for =x, y, z, respectively. It turns out that velocity v are given by v0= (v0,0), vx=(vx,0), vy=0, vz=(0,vz) and that vx is close to vz and then vx=vz is assumed in the explicit calculations. The existence of finite v0 represents tilting, which is characterized by the dimensionless parameter,=v0/vc. Here vc is defined by vc= vx=vz [32]. It turns out that ~0.8 in the case of  -ET2I3. 4. Effects of tilting on orbital susceptibility and Hall conductivity The effects of tilting on conductivity and Hall conductivity have been studied [33] in terms of proper formulas applied to the Hamiltonian, eq. (1). The results are shown in Fig.1 and 2 at absolute zero as a function of chemical potential under the assumption of energy independent electron damping (lifetime). Here X=, with  and being the chemical potential and damping, respectively. In a +xx b, where both the case of conductivity, there are two different types of contributions, i.e. xx=xx a and xx b have contributions from only at the Fermi energy (on-shell contributions) but have xx 0 (the solid Fig. 1. The conductivityxx /xx 0 (the dashed line) and xx b /xx a/xx 0 line), xx 0=e2/22. The (the dotted line) for =0, where xx inset A shows -dependences of enhancement a() /xx a() (the dashed line) and factors, xx b() (the dotted line). The inset B b() /xx xx shows X-dependences of total xx for several degrees of tilting, the solid line(the dashed line) and 0.9(the dot-dashed line). [33] 0 Fig. 2. The Hall conductivity xy /xy a /xy 0 (the dashed line) (the solid line), xy 0 (the dotted line) in the absence b /xy and xy 0=e3vc 2H/4 2c0 2 . The of tilting, where xy inset A shows dependences of a() /xy a() (the enhancement factors, xy b() /xy b() (the dashed line) and xy dotted line). The inset B shows X-dependences of total xy for several degrees of tilting. [33] zyxH,,,0vk different dependences on . (There is a finite anisotropy in the conductivity but it is small as is expected from the weak  dependences of xx , especially near the crossing, X=0, as seen in the inset a +xy b. It is to be noted, of Fig.1.) Similarly the Hall conductivity has two contributions, xy=xy a contains all contributions below however, that there is a qualitative difference from conductivity; xy b has only on-shell contributions. In Fig.3 is shown the result for orbital the Fermi energy whereasxy susceptibility, in which case tilting affects only the over-all prefactor. In all these singular behaviour toward  =1 is due to the fact that the velocity in one direction is vanishing in this limit. It is seen that finite tilting affects both orbital susceptibility and Hall conductivity quantitatively but qualitative features are unchanged except close to =1. It is to be noted that the present assumption of energy-independent damping of electrons leads to increase of conductivity as a function of energy. This result will lead to the increase of conductivity as a function of temperature in the case where the Fermi energy is located at the crossing energy at the absolute zero. This is not consistent with the experimental results. This will imply that the damping is increasing as a function of temperature (energy) almost linearly. If this is the case, the Hall coefficient will be strongly suppressed as the temperature is increased as seen in the experiment, although it is a future problem to explore these dependences on energy at absolute zero and on temperature at fixed carrier density. Regarding the sharp change of sign in Hall coefficient at low temperatures, it has been indicated [33] that the possible small amount of -) of the order of only deficiency of anions (I3 ppm together with asymmetry of energy dependences of density of states relative to the crossing energy can be the cause. Fig. 3. The orbital susceptibility  as the function of X in the absence of tilting, where  e2vc 2/3 2c20. The inset shows dependence of enhancement factor  () / (). [33] 5. A possible Berezinski-Kosterlitz-Thouless phase transition under strong magnetic field So far the orbital susceptibility and Hall conductivity have been studied in weak magnetic field. Under strong magnetic field of the order of 10 Tesla applied perpendicularly, resistivity measurement indicates interesting temperature dependences[19], which shows the existence of two characteristic temperatures, T0 and T1, for its sharp change. It has been proposed [34] that these characteristic temperatures will correspond to the crossover temperature of the formation of valley -order similar to the XY ferromagnetic order and the Berezinski-Kosterlitz-Thouless (BKT) transition at lower temperature as will be briefly explained below. Magnetic field applied perpendicularly to the plane leads to the quantized Landau levels. Note that there exists a particular Landau level at the crossing energy (N=0) independent of the strength of magnetic field. The finite tilting does not affect this qualitatively [35] and the energy level separation between N=0 and N=1 is of the order of 50K at H=10T. The spin Zeeman splitting is much smaller than the Landau level splitting since the g-factor is g~2. Because there are two Dirac cones in the Brillouin zone, these states (valleys) are doubly degenerate (characterized by the pseudo-spin) if two valleys are independent. Once mutual Coulomb interactions are taken into account, electrons in different valleys interact and get scattered among them. In order to treat such many-body effects in the presence of Landau quantization, ordinary Landau wave functions, which are plane wave and then infinitely spread in one direction, is not convenient and then orthonormal basis sets, which are localized in space similar to “Wannier functions”, introduced some time ago are employed. It turns out that finite tilting plays crucial roles by introducing processes of mixing different valleys. If the chemical potential is fixed to the crossing energy and this mixing is strong enough compared with the spin Zeeman splitting, some kind of valley order state described as ferromagnetism of pseudo-spins is possible at T=T0 in the mean-filed approximation. The order parameters in this case are complex and their phases represent degrees of freedom leading to vortices which can have BKT phase transition at lower temperature T=T1. Theoretical analysis concludes T1/T0 ~1/4. If T0 and T1 are associated with the two step changes of resistivity in the experiment [19] this ratio between T0 and T1 is close to experimental results. A phase transition similar to BKT had been indicated in resistivity measurement of graphene [36] at much lower temperatures. Since tilting is absent in graphene, some mechanism other than the present tilting-driven inter-valley mixing by Coulomb interactions, e. g. an electron-lattice interaction [37], will be operating in graphene. Acknowledgment One of authors (HF) dedicates hearty gratitude to late Professor Ryogo Kubo for his guidance in very early stage based on the deep insight into inter-band effects of magnetic field on Bloch electrons. Authors are grateful to Mark Goerbig for collaborations for the subject in Sec.5. References [1] P. R. Wallace, Phys. Rev. 71, 622 (1947) [2] M. H. Cohen and E. I. Blount, Phil. Mag. 5, 115 (1960) [3] P. A. Wolff, J. Phys. Chem. Solids 25, 1057 (1964). [4] D. Schoenberg and M. Z. Uddin, Proc. Roy. Soc. A156, 687,701 (1936). [5] R. Peierls, Z. Phys. 80, 763 (1933). [6] E. N. Adams, Phys. Rev. 89, 633 (1953). [7] A. H. Wilson, Proc. Camb. Phil. Soc. 49, 292 (1953). [8] For example, G. H. Wannier and U. N. Upadyaya, Phys. Rev. 136, A803 (1964); R. Kubo. Proc. Intern. Conf. Magnetism, Notthingam 1964 (The Institute of Physics and Physical Society) p.35, and references therein. [9] J. W. McClure, Phys. Rev. 104, 666 (1956). [10] H. Fukuyama and R. Kubo, J. Phys. Soc. Jpn. 28, 570 (1970). [11] J. M. Luttinger and W. Kohn, Phys. Rev. 97, 869 (1955). [12] H. Fukuyama, Prog. Theor. Phys. 45, 704 (1971). [13] R. Kubo and H. Fukuyama, Proc. of the “10th International Conference on the Physics of Semiconductors” (Ed. E. P. Keller and J. C. Hensel and F. Stern, Published by the United State Atomic Energy Commission, 1970). [14] H. Fukuyama, Ann. Phys. (Leipzig) 15, 520 (2006). [15] 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). [16] T. Ando: J. Phys. Soc. Jpn. 74, 777 (2005). [17] H. Fukuyama, H. Ebisawa and Y. Wada, Prog. Theor. Phys. 42, 494 (1969). [18] H. Fukuyama, J. Phys. Soc. Jpn. 76, 043711 (2007). [19] N. Tajima, S. Sugawara, M. Tamura, Y. Nishio and K. Kajita, J.Phys. Soc. Jpn. 75, 0510101 (2005). [20] N. Tajima, S. Sugawara, M. Tamura, R. Kato, Y. Nishio and K. Kajita, private communications. [21] T. Mori, A. Kobayashi, T. Sasaki, H. Kobayashi, G. Saito and H. Inokuchi, Chem. Lett. 13, 957 (1984) and references therein. [22] H. Kino and H. Fukuyama, J.Phys. Soc. Jpn. 65, 2157 (1996). [23] H. Seo, C. Hotta and H. Fukuyama, Chem. Rev. 104, 5005 (2004). [24] K. Kanoda, J. Phys. Soc. Jpn. 75, 051007 (2005). [25] K. Kanoda and R. Kato, Annu. Rev. of Condens. Matter Phys. 2011. 2:167-88 [26] H. Seo, J. Phys. Soc. Jpn. 69, 805 (2000). [27] R. Kondo, S. Kagoshima, and J. Harada, Rev. Sci. Instrum. 76, 090302 (2005). [28] A. Kobayashi, S. Katayama, K. Noguchi, and Y. Suzumura, J. Phys. Soc. Jpn 73, 3135 (2004). [29] S. Katayama, A. Kobayashi, and Y. Suzumura, J. Phys. Soc. Jpn 75, 054705 (2006). [30] H. Kino and T. Miyazaki, J. Phys. Soc. Jpn 75, 034704 (2006). [31] S. Ishibashi, T. Tamura, M. Kohyama, and K. Terasaki, J. Phys. Soc. Jpn 75, 015005 (2006). [32] A. Kobayashi, S. Katayama, Y. Suzumura, and H. Fukuyama,J. Phys. Soc. Jpn 76, 034711 (2007). [33] A. Kobayashi, Y. Suzumura, and H. Fukuyama, J. Phys. Soc. Jpn 77, 064718 (2008). [34] A. Kobayashi, Y. Suzumura, H. Fukuyama, and M. O. Goerbig, J. Phys. Soc. Jpn 78, 11471, (2009). [35] T. Morinari, T. Himura, and T. Tohyama, J. Phys. Soc. Jpn 78, 023704 (2009). [36] J. G. Checkelsky, L. Li, and N. P. Ong, Phys. Rev. B 79, 115434 (2009). [37] K. Nomura, S. Ryu, and D-H Lee : Phys. Rev. Lett. 103 (2009) 216801.
1208.1893	1	1208	2012-08-09T12:56:38	Temperature dependent reversal of voltage modulated light emission and negative capacitance in AlGaInP based multi quantum well light emitting devices	[ "cond-mat.mes-hall" ]	We report a reversal in negative capacitance and voltage modulated light emission from AlGaInP based multi-quantum well electroluminescent diodes under temperature variation. Unlike monotonically increasing CW light emission with decreasing temperature, modulated electroluminescence and negative capacitance first increase to a maximum and then decrease while cooling down from room temperature. Interdependence of such electronic and optical properties is understood as a competition between defect participation in radiative recombination and field assisted carrier escape from the quantum well region during temperature variation. The temperature of maximum light emission must coincide with the operating temperature of a device for better efficiency.	cond-mat.mes-hall	cond-mat	Temperature dependent reversal of voltage modulated light emission and negative capacitance in AlGaInP based multi quantum well light emitting devices Kanika Bansal and Shouvik Datta Division of Physics, Indian Institute of Science Education and Research, Pune - 411008, Maharashtra, India Abstract We report a reversal in negative capacitance and voltage modulated light emission from AlGaInP based multi-quantum well electroluminescent diodes under temperature variation. Unlike monotonically increasing CW light emission with decreasing temperature, modulated electroluminescence and negative capacitance first increase to a maximum and then decrease while cooling down from room temperature. Interdependence of such electronic and optical properties is understood as a competition between defect participation in radiative recombination and field assisted carrier escape from the quantum well region during temperature variation. The temperature of maximum light emission must coincide with the operating temperature of a device for better efficiency. 1 Improvement in the performance efficiency of electroluminescent diodes (ELDs) is the primary goal of physicists as well as engineers working in the field of optoelectronics. Continuous research is going on to achieve better efficiencies for a vast variety of technological applications1. For efficiency improvement as well as from basic understanding point of view, it is necessary to study the interdependence of electronic and optical properties of ELDs under their operating condition of light emission. With this motivation we studied steady state behavior of commercially available red-ELDs (AlGaInP based multi-quantum well laser diodes) during light emission below the lasing threshold. Conventional characterization techniques based on depletion approximation and electrostatic description of the junction may not work within their usual interpretation in this high forward bias regime beyond2 the range of diffusion capacitance. We probed the steady state dielectric response of the ELDs by studying both voltage modulated electroluminescence2 (VMEL) and junction impedance. In our earlier work2 with these diodes (for frequencies ≤ 100 kHz) at room temperature we showed that under sufficiently high forward bias and low frequency, reactive component of the impedance acquires ‘inductive like’ behavior commonly referred as negative capacitance (NC), which is exciting, technologically important3,4 and yet less understood phenomenon. Interestingly our VMEL results showed2 that the modulated light emission is closely related to NC and both depend systematically on applied modulation frequencies. Frequency dependent NC and VMEL were understood as the outcome of dynamic competition between radiative recombination process in the quantum well (QW) and the steady state response from shallow defects. This produces a transient change in the charge carrier reservoir available for recombination and triggers a compensatory inductive like response2. In the work we assumed that at constant room temperature steady state population of charge carriers inside the QW is unaffected by any escape 2 of carriers over the QW barrier. Results presented in this letter also support this assumption. The observed low frequency response can be a cause of efficiency reduction during high frequency direct modulation2 (~GHz) used in applications like optical communication. This is because defect mediated steady state contribution of carriers undergoing radiative recombination will be missing in action at such high frequency. Hence we had emphasized the need to carry out such low frequency steady state characterizations to optimize the device design for better light emission efficiency. Under low frequency modulation, NC is observed by other groups in ELDs5-7 and also in different semiconductor devices8. Time domain study of frequency dependent NC has also been done9-11. However for the first time, to our knowledge, we are reporting the temperature variation of both NC and VMEL and their interdependence. Unlike usual12 continuous wave (CW) light emission which expectedly decreases continuously with increasing temperature, both NC and VMEL go through a maximum during temperature scan. Position of this maximum systematically shifts towards higher temperature with increasing forward bias. This occurrence of peak or maxima in modulated light emission immediately indicates that the device can have maximum efficiency at different temperatures for different biases. Thus, we suggest that for efficient working of ELDs under direct modulation, the temperature of maximum light emission must be tailored to their operational temperature. This can be achieved either by careful choice of materials and/or by clever device engineering. Fig. 1a shows the variation of VMEL or modulated light emission for commercially available laser diode (Sanyo, DL3148-025) with temperature at modulation frequency of 1 kHz under different current levels below lasing threshold. Experiments were done at constant forward bias current (Idc), which allows us to ignore any effect of change in the total charge carrier 3 density on the mechanisms we are probing at various temperatures (accuracy of 0.5K). Details of the experimental setup and devices can be found in ref. 2. For a given Idc, when the sample is cooled down from room temperature, VMEL signal increases and interestingly, within a certain range of temperatures (~140K – 210K), reverses the nature and starts to decrease giving a maximum in light emission. This temperature of maximum emission (TMax), shifts systematically to higher temperature side as the bias increases. However CW light emission (fig. 1b) expectedly increases with cooling. In general, efficiency of radiative recombination improves with decreasing temperature in QW because of the decreased radiative recombination lifetime13. Absence of any maximum in the case of CW light emission indicates that the observed features in VMEL are only due to steady state response of charge carriers to applied voltage modulation. This is evident in fig. 1c which shows the frequency variation of the temperature dependent VMEL measured at Idc = 250 µA. For any given temperature, lower frequency produces higher VMEL signal which is in agreement with our earlier reported results2. Noticeably, TMax is relatively independent of the frequency (for frequencies ≤ 100 kHz). Figures 2a and 2b show temperature variation of capacitance and conductance respectively for different injection levels at a modulation frequency of 1 kHz. These measurements are done under the circuit model containing capacitance and conductance in parallel as described in ref. 2. In capacitance also we observed similar temperature dependent reversal giving a temperature (TMax) where NC is maximum. Here also TMax shifts to higher temperature with increasing forward bias. Similar changes are seen in conductance too. Fig. 2c shows that like VMEL, TMax is independent of modulation frequency in case of NC too. NC is more prominent for lower frequencies for all the temperatures. Figures 1 and 2 certainly indicate the correlation between measured optical (VMEL) and electronic (NC) properties. We will explain this correlated behavior in terms of 4 active participation of defects in radiative recombination dynamics and continuous escape of charge carriers from the quantum well region. Defect states can only contribute to the VMEL and NC signal if their thermal rate of carrier exchange with the band edges (1/ ) is comparable to applied frequency (f). Here (1/ ) is related to the thermal activation energy (ETh) of the defect level by where kB is the Boltzmann constant, is the thermal prefactor, hence we can write: (1) Steady state charge carrier (e.g. electrons) contribution from defects is given as , g(E) is the sub-bandgap defect density. Limits of integration represent2 the total defect states which can respond to the applied signal as given by equation 1 and also by the position of quasi Fermi level (EFn) and band edges (conduction band EC in case of electrons). Interestingly, at a given bias and frequency, defect contribution to charge carrier density available for radiative recombination increases with the increasing temperature. This is why we see counterintuitive increase in the magnitude of VMEL and NC signal as shown by figures 1a and 2a, when temperature is increased from a quite low value ( 60 K). However this increase is rather slow for relatively low temperatures (≤ 100 K) due to the inability of charge carriers to respond to the applied voltage modulation. In such case one would expect that for even lower temperatures both VMEL and NC will saturate. However, with further heating VMEL and NC reverse the behavior and start to decrease above certain temperature (> TMax). Likely cause is the increasing loss of charge carriers available for radiative recombination due to their field assisted thermal escape from the active QW region. This is when the QW escape process dominates over shallow defect mediated contribution to steady state radiative 5 1exp()ThBEkTln(/).ThBEkTf()()FnCThEDefectsEEnEgEdErecombination. In addition to this escape, typical activation of deep defects with very large response time may also cause the net loss of charge carriers in the QW within the measurement time window. Schematic of these processes is shown in fig. 3a. It is interesting to note that thermally activated QW escape rate and consequently the TMax seems to be independent of modulation frequency within our range of measurements. In fact, the independence of TMax within the range of applied modulation frequency (fig. 1c and 2c) favors the presence of very fast QW escape processes around room temperature to explain the reversal observed in VMEL and NC. To probe the escape process we measured the variation of CW light emission with temperature for different Idc values (fig. 3b) and also the variation in Idc with temperature while maintaining certain light emission intensity (fig 3c). Continuous charge carrier escape from QW is a complimentary process to light emission in a sense that it depletes the number of charge carriers available for radiative recombination. In our case, a major contribution to this Idc comes from charge injection to active region, QW escape, radiative recombination and trapping by defects etc. Current density of escape JE for electrons in a QW is14 (2) where L is the width of the QW, nQW is the carrier density inside the well, e is the electronic charge and 1/ E is the rate of thermal escape. As the temperature increases, 1/ E for charge carriers in the QW increases, and escape process starts to dominate. This is also evident in fig. 3b where falling rate of radiative recombination is the outcome of increasing temperature at a fixed Idc due to higher consumption rate of charge carriers by the thermally activated QW escape process. At a particular temperature more light emission is observed for higher Idc as expected. 6 QWEEneLJSimilarly in fig. 3c, as the temperature increases, higher rate of charge injection in terms of Idc is required to maintain constant light emission intensity. This is again due to the increase in thermally activated QW escape process with increasing temperature. At a particular temperature, required Idc to maintain light emission increases with increasing light emission intensity as is also discussed in the context of fig. 3b. Straight line behavior in Arrhenius like plots of figures 3b and 3c show thermal activation rate of light emission and injected charge respectively. Calculated values of respective activation energies EI and EII are shown in the insets. The rate for any thermally activated QW escape process can be given as13 (3) where Q is the factor related to L and effective mass of the charge carrier inside the well and EB is the effective thermal barrier height. In the present case of active light emitting device under voltage modulation, thermal rate of one process is intricately connected with the other competing processes like charge injection, recombination, escape and defect trapping etc. Many of these processes are also intrinsically thermally activated with different barrier heights. Hence it is difficult to directly associate the calculated activation energies solely to any particular barrier height of field assisted thermal escape from QW. However we understand EI as the activation energy of light emission which goes down with increasing bias ( Idc) suggesting that the QW escape process which is complimentary to light emission decreases with increasing Idc. This happens due to the field assisted behavior of QW escape process which is also evident in figures 1a and 2a. In absence of any forward bias, built-in field of the p-n junction efficiently separates thermally escaped charge carriers15, however with increasing forward bias (Idc) this net field contribution to field assisted thermally activated escape of carriers from QW decreases. As a 7 1exp2BBEBkTEQkTresult TMax shifts towards higher temperature for higher Idc as can be seen in figures 1a and 2a. EII is understood as the activation energy of charge injection which remains relatively unchanged with light emission intensity within our range of measurements and mainly depends on p-n junction parameters of ELD. To sum up, as the temperature increases, field assisted thermal escape process gets strengthened and dominates over defect mediated radiative recombination process, consequently we see a decrease in the magnitude of VMEL and NC signals (figures 1a and 2a) above a certain temperature TMax. In conclusion, we report counter intuitive decrease in the magnitude of both VMEL and NC from red electroluminescent diodes below a temperature TMax which deviates from the usual monotonically increasing nature of light emission with decreasing temperature. Here TMax is the temperature where modulated light emission is maximum. This is caused by the interplay of active participation of shallow defect levels in radiative recombination and the continuous charge carrier escape processes from the quantum well active region. This TMax systematically shifts to higher temperature side with increasing forward bias and is relatively independent of modulation frequency within the range of our measurements. Similar qualitative features are also observed in negative capacitance. This correlated behavior of optical and electronic processes is important to understand the device behavior during modulated light emission. From application point of view, the temperature of maximum light emission must be brought to operating temperature of the light emitting device for its efficient performance under direct modulation. Although here we presented results on AlGaInP based multi quantum well laser diodes below lasing threshold, we expect these results and analyses to be valid for a wider range of light emitting devices. To further understand the correlation between observed results, a rigorous theoretical frame work beyond depletion approximation and electrostatic description of the junction is required along 8 with the experiments for varying quantum well parameter, defect dynamics and modulation parameters. Authors wish to thank IISER-Pune for startup funding of the laboratory infrastructure and Dept. of Science and Technology, India for DST Nano Unit grant SR/NM/NS-42/2009. KB is thankful to CSIR, India for research fellowship. 9 Figure 1: Temperature scan of light emission from laser diode at different Idc. (a) Voltage modulated light emission at 1 kHz showing maxima at certain temperatures (TMax). Increasing bias shifts TMax to higher temperature. For cost effective performance, this maximum should be around operational temperature of the ELD. (b) Usual CW light emission showing a continuous increase with decreasing sample temperature. (c) Frequency dependence of temperature scan of modulated light emission at 250 µA. Note that TMax is relatively independent of applied modulation frequencies. 10 60901201501802102402703003300.000.020.040.060.080.100.120.14(c) 1 kHz 10 kHz 30 kHZ Modulated Light Emission (a.u.)Temperature (K)60901201501802102402703003300.00.10.20.30.4 50 300 450 600 800 1 m Modulated Light Emission (a.u.) Temperature (K)(a)60901201501802102402703003300.00.10.20.30.40.5(b) 50 300 450 600 800 1 m CW Light Emission (a.u.)Temperature (K) Figure 2: Temperature scan of (a) negative capacitance and (b) conductance for laser diode at 1kHz for different Idc. A maximum is observed at certain temperature (TMax) which shifts to higher side for higher bias. (c) Frequency dependence of the temperature scan of negative capacitance at 250 µA. Again TMax is relatively independent of applied modulation frequency. 11 609012015018021024027030033002468101214161820 50 300 450 600 800 1 mConductance (mS)Temperature (K)(b)6090120150180210240270300330-14-12-10-8-6-4-20 50 300 450 600 800 1 mCapacitance (nF)Temperature (K)(a)6090120150180210240270300330-1.5-1.0-0.50.0 1 kHz 10 kHz 30 kHz Capacitance (nF)Temperature (K)x1/2(c) Figure 3: (a) Schematic band diagram of the electronic processes occurring in QW. Electrons (holes) are injected from the n (p) neighborhood of the junction. These electrons (holes) can undergo radiative recombination or defect mediated recombination through some defect with activation energy ETh. With increasing temperature, field assisted thermal escape of charge carriers starts to dominate, thereby reducing the number of charge carriers inside the well (nQW). EFn shows the quasi Fermi level for electron within QW for small forward bias. Quasi Fermi level for holes is not shown here for simplicity. (b) Showing increase in the rate of light emission with decreasing temperature and increasing Idc. Absolute value of the activation energy for light emission EI as calculated from the straight line portion of the Arrhenius plot is given in the inset. (c) Variation in the rate of charge injection in the form of Idc with temperature to maintain certain CW light emission intensity. Increasing light emission requires higher Idc. Calculated activation energy for charge injection EII is shown in the inset. 12 345678910111213141510-510-410-310-20.20.40.60.81.01.21.430405060708090 EI(meV)Idc (mA)(b) CW Light Emission (a.u.)1000/T (1/K)Increasing Idc345678910111213141510-410-30.040.080.120.160.2030405060708090 EII(meV)Light Intensity (a. u.)(c) Idc (A)1000/T (1/K)Increasing Light EmissionReferences: 1 E. F. Schubert, Light Emitting Diodes (Cambridge University Press, 2006). 2 K. Bansal and Shouvik Datta, J. Appl. Phys. 110, 114509 (2011). 3 S. Salahuddin and S. Datta, Nano Lett. 8, 405 (2008). 4 J. Shulman, Y. Y. Xue, S. Tsui, F. Chen and C. W. Chu, Phys. Rev. B 80, 134202 (2009). 5 J. Bisquert, G. Garcia-Belmonte, A´. Pitarch, H. J. Bolink, Chem. Phys. Lett. 422, 184 (2006). 6 Y. Li, C. D. Wang, L. F. Feng, C. T. Zhu, H. X. Cong, D. Li, G. Y. Zhang, J. Appl. Phys. 109, 124506 (2011). 7 L. F. Feng, D. Li, C. Y. Zhu, C. D. Wang, H. X. Cong, X. S. Xie and C. Z. Lu, J. Appl. Phys. 102, 063102 (2007). 8 See references 2 to 7 and work sited therein. 9 F. Lemmi and N. M. Johnson, Appl. Phys. Lett. 74, 251 (1999). 10 M. Ershov, H. C. Liu, L. Li, M. Buchanan, Z. R. Wasilewski, and A. K. Jonscher, IEEE Trans. Electron. Devices, 45, 2196 (1998). 11 R. Tadros-Morgane, G. Vizdrik, B. Martin and H. Kliem, J. Appl. Phys. 109, 014501 (2011). 12 In the context of photoluminescence efficiency, which is also CW emission within our frame work, some reports show non-monotonic features with temperature variation. For example: R. H. Bube, Phys. Rev. 81, 633 (1951). 13 A. M. Fox, D. A. B. Miller, J. E. Cunningham and W. Y. Jan, IEEE J. Quantum Electron. 27, 2281 (1991). 14 J. Nelson, M. Paxman, K. W. J. Barnham, J. S. Roberts and C. Button, IEEE J. Quantum Electron. 29, 1460 (1993). 15 Shouvik Datta, B. M. Arora and S. Kumar, Phys. Rev. B 62, 13604 (2000). 13
1509.07949	1	1509	2015-09-26T07:37:36	Spin Dynamics and Relaxation in Graphene Dictated by Electron-hole Puddles	[ "cond-mat.mes-hall" ]	The understanding of spin dynamics and relaxation mechanisms in clean graphene and the upper time and length scales on which spin devices can operate are prerequisites to realizing graphene spintronic technologies. Here we theoretically reveal the nature of fundamental spin relaxation mechanisms in clean graphene on different substrates with spin-orbit Rashba fields as low as a few tens of micron eV. Spin lifetimes ranging from 50 picoseconds up to several nanoseconds are found to be dictated by substrate-induced electron-hole characteristics. A crossover in the spin relaxation mechanism from a Dyakonov-Perel type for SiO2 substrates to a broadening-induced dephasing for hBN substrates is described. The energy dependence of spin lifetimes, their ratio for spins pointing out-of-plane and in-plane, and the scaling with disorder provide a global picture about spin dynamics and relaxation in ultraclean graphene in presence of electron-hole puddles.	cond-mat.mes-hall	cond-mat	Spin dynamics and relaxation in graphene dictated by electron-hole puddles Dinh Van Tuan1, Frank Ortmann,2, Aron W. Cummings1, David Soriano1 and Stephan Roche,1,3 1ICN2 - Institut Catala de Nanociencia i Nanotecnologia, Campus UAB, 08193 Bellaterra (Barcelona), Spain 2 Institute for Materials Science, Dresden Center for Computational Materials Science, Technische Universitat Dresden, 01062 Dresden, Germany 3ICREA, Instituci´o Catalana de Recerca i Estudis Avanc¸ats, 08070 Barcelona, Spain (Dated: October 13, 2018) 5 1 0 2 p e S 6 2 ] l l a h - s e m . t a m - d n o c [ 1 v 9 4 9 7 0 . 9 0 5 1 : v i X r a The understanding of spin dynamics and relaxation mechanisms in clean graphene and the upper time and length scales on which spin devices can operate are prerequisites to realizing graphene spintronic technolo- gies. Here we theoretically reveal the nature of fundamental spin relaxation mechanisms in clean graphene on different substrates with spin-orbit Rashba fields as low as a few tens of µeV. Spin lifetimes ranging from 50 picoseconds up to several nanoseconds are found to be dictated by substrate-induced electron-hole charac- teristics. A crossover in the spin relaxation mechanism from a Dyakonov-Perel type for SiO2 substrates to a broadening-induced dephasing for hBN substrates is described. The energy dependence of spin lifetimes, their ratio for spins pointing out-of-plane and in-plane, and the scaling with disorder provide a global picture about spin dynamics and relaxation in ultraclean graphene in presence of electron-hole puddles. PACS numbers: 72.80.Vp, 73.63.-b, 73.22.Pr, 72.15.Lh, 61.48.Gh The tantalizing prospect of graphene spintronics was initi- ated by Tombros and coworkers [1], who first reported long spin diffusion length in large area graphene. The small spin- orbit coupling (SOC) in carbon, plus the absence of a hyper- fine interaction, suggested unprecedented spin lifetimes (τs) at room temperature (from µs to ms) [2, 3, 4, 5, 6, 7]. However, despite significant progress in improving graphene quality, resolving contact issues, and reducing sub- strate effects [1, 8, 9, 10, 11, 12, 13, 14, 15], the measured τs are orders of magnitude shorter, even for high-mobility samples. Extrinsic sources of SOC, including adatoms [16, 17, 18, 19] or lattice deformations [20, 21], have been proposed to explain this discrepancy. Moreover, the na- ture of the dominant spin relaxation mechanism in graphene is elusive and debated. The conventional Dyakonov-Perel (DP) [22] and Elliot-Yafet (EY) [23] mechanisms, usually describing semiconductors and disordered metals, remain inconclusive in graphene because neither effect can con- vincingly reproduce the observed scaling between τs and the momentum relaxation time τp [8, 11]. Although gen- eralizations of both mechanisms have been proposed, they do not allow an unambiguous interpretation of experiments [6, 20, 21, 24, 25]. It should be noted that the achieved room-temperature spin lifetime in graphene is already long enough for the ex- ploration of spin-dependent phenomena such as the spin Hall effect [26, 27], or to harness proximity effects as induced for instance by magnetic oxides [28] or semiconducting tung- sten disulphide [29]. However, a comprehensive picture of spin dynamics of massless Dirac fermions in presence of weak spin-orbit coupling fields is of paramount importance for further exploitation and manipulation of spin, pseudospin and valley degrees of freedom [7, 30, 31, 32]. In this study, we show numerically that a weak uniform Rashba SOC (tens of µeV), induced by an electric field or s /τ the substrate, yields spin lifetimes from 50 ps up to sev- eral nanoseconds. The dominant spin relaxation mechanism is shown to be dictated by long range potential fluctuations (electron-hole puddles) [33]. For graphene on a SiO2 sub- strate, such disorder is strong enough to interrupt the spin precession driven by the uniform Rashba field, resulting in motional narrowing and the DP mechanism. We also find the (cid:107) s (cid:39) 1/2, demonstrating the anisotropy of the in- ratio τ⊥ plane Rashba SOC field. For the case of a hexagonal boron nitride (hBN) substrate, the role of electron-hole puddles is reduced to an effective energy broadening and the spin life- time is limited by pure dephasing [34, 35]. These situations, however, share a common fingerprint -- a M-shape energy dependence of τs that is minimal at the Dirac point. Taken together, our results provide deeper insight into the funda- mentals of spin lifetimes in graphene dominated by electron- hole puddles. Results Disorder and Spin dynamics. Electron-hole puddles are real-space fluctuations of the chemical potential, induced by the underlying substrate, which locally shift the Dirac point [33, 36, 37]. Since measured transport properties usually result from an aver- age around the charge neutrality point, it is generally dif- ficult to access the physics at the Dirac point. As shown by Adam and coworkers [33], electron-hole puddles can be modeled as a random distribution of long range scatterers, j=1 j exp[−(r − Rj)2/(2ξ2)], where ξ = 10 and 30 nm denote the effective puddle ranges for SiO2 and hBN substrates, respectively [36, 38], and j is randomly chosen within [−∆, ∆]. Based on experimental data, typical V (r) = (cid:80)N impurity densities are ni = 1012 cm−2 (0.04%) for SiO2 and ni = 1011 cm−2 (0.004%) for hBN substrates [36, 39]. From such information, we can tune ∆ to obtain suitable disorder profiles for the onsite energy of the π-orbital. Fig. 1 (main frame) shows the onsite energy distribution corre- sponding to hBN and SiO2 substrates, where the Gaussian profiles give standard deviations of σ = 5.5 and 56 meV, re- spectively. This allows us to extract ∆ = 50 meV for SiO2 and ∆ = 5 meV for hBN. The inset of Fig. 1 shows the en- ergy landscape for a sample with 0.04% Gaussian impurities (SiO2 case). FIG. 1: Onsite energy distribution of the carbon atoms in the graphene sample, which mimics the chemical potential induced by hBN (green) and SiO2 (black) substrates together with their Gaus- sian fitting lines. Inset: Real space vizualization of the energy landscape for a graphene sample with 0.04% Gaussian impurities (SiO2 case). To fully characterize the role of electron-hole puddles, we evaluate the transport time τp using a real-space order-N ap- proach, which computes the diffusion coefficient D(E, t). We extract τp from the saturation of D(E, t) since τp = F (E) [40]. For numerical convenience, the Dmax(E)/2v2 calculations are made using ∆ = 0.1γ0 (with γ0 the nearest neighbor hopping parameter), thus in absence of intervalley scattering [41], while the final values for hBN and SiO2 sub- strates are extrapolated from the numerical results using the scaling law √ τp ∼ πn∗ξ K0 eπn∗ξ2 I(πn∗ξ2) √ where I1(x) is the modified Bessel function of the first kind, K0 = 40.5ni(∆/t)2(ξ/ 3a)4 is a dimensionless pa- rameter dictating the strength of the Gaussian potential, and the carrier density n∗ is modified from the pristine graphene (1) , 2 FIG. 2: (a) Transport times for graphene on SiO2 and hBN sub- strates (solid black and red curves, respectively). The dashed line shows the spin precession time. (b) Time-dependent spin polariza- tion for out-of-plane (solid red line) and in-plane (solid black line) spin injection for the SiO2 substrate, plus the fits to the exponential damping (dashed lines). The blue curves show the same informa- tion for the hBN substrate with out-of-plane injection. density n by n∗ = n + K0 2π2ξ2 [33, 42, 43]. The computed τp are shown in Fig. 2(a) for both substrates. For SiO2, τp is on the order of a few ps, while for hBN τp is more than two orders of magnitude larger. The spin precession time, TΩ = π¯h/λR, is shown for comparison. Spin dynamics and lifetimes in the presence of electron-hole puddles. We now analyze the spin dynamics for puddles corre- sponding to the SiO2 and hBN substrates. The blue curve in Fig.2(b) shows the time-dependent spin polarization for the hBN substrate (ni = 0.004%) at the Dirac point for an initial out-of-plane polarization, P hBN⊥ (t) (see Methods). The po- larization exhibits oscillations with period TΩ = π¯h/λR (cid:39) 55 ps, corresponding to the spin precession induced by the Rashba field. Simultaneously, the polarization decays in time, and by fitting P hBN⊥ (t) = cos (2πt/TΩ) e−t/τs, both TΩ and the spin relaxation time τs can be evaluated. (t) for the SiO2 substrate (ni = 0.04%) with initial spin polarization in-plane (α =(cid:107)) and out-of-plane (α =⊥). In contrast to the hBN case, for which P hBN⊥ (t) exhibits significant precession, the disorder strength of electron-hole puddles for SiO2 is sufficient to in- terrupt spin precession. As a result, the polarization for SiO2 (cid:107)/⊥ (t) = e−t/τs. The absence of preces- is better fit with P SiO2 sion for P SiO2⊥ (t) compared to P hBN⊥ (t) is consistent with the ratio between transport time and precession frequency, since τ SiO2 /TΩ (cid:28) 1 whereas τ hBN Fig.2(b) also shows P SiO2 /TΩ > 1. To scrutinize the origin of the dominant relaxation mecha- nism, we first examine the spin lifetimes τs for the SiO2 case α p p 050100150200t (ps)-1-0.8-0.6-0.4-0.200.20.40.60.81Pα (t)-400-2000200400E (meV)0.1110100100010000τp (ps)α=, SiO2α=⊥,α=⊥, hBNSiO2hBNTΩ=π h_ /λRa)b)SiO23 Crossover in spin relaxation behavior for hBN and SiO2 substrates. Fig.4 provides a global view of our results, where we plot τs vs. 1/τp for the SiO2 and hBN substrates (black and red symbols respectively) at the Dirac point and at E = −200 meV (closed and open symbols respectively). For low defect densities (hBN substrate), τs decreases strongly with decreasing τp. However, with increasing defect den- sity (SiO2 substrate) this trend reverses and τs scales al- most linearly with 1/τp, according to the DP relationship τs = ν · (TΩ/2π)2/τp. At E = −200 meV, ν = 1, fitting the usual DP theory. At the Dirac point, the scaling is some- what weaker, with ν = 1/4. These results are reminiscent of those summarized in Fig. 5(a) of Drogeler et al. [13], where spin lifetimes of graphene devices on SiO2 scaled inversely with the mobility, while devices on hBN appear to show the opposite trend. While the SiO2 results of Fig.4 show DP behavior, the nature of the spin relaxation for weak electron-hole pud- dles is less clear. The fact that τs and τp decrease together suggests the EY mechanism, but we find τs ≤ τp near the Dirac point and τs (cid:28) τp at higher energies. This contrasts with the usual picture of EY relaxation, where charge car- riers flip their spin when scattering off impurities, giving τs = τp/α, where α (cid:28) 1 is the spin flip probability [6]. Instead, this situation matches that described in Ref. [44]; when τp > TΩ, the spin precesses freely until phase infor- mation is lost during a collision, in analogy to the collisional broadening of optical spectroscopy. More collisions result in a greater loss of phase, reducing τs with decreasing τp. We verify this by removing the real-space disorder (setting ∆ = 0) and modeling the electron-hole puddles with an ef- fective Lorentzian energy broadening η∗. The results are shown in Fig.4 (main frame, blue dashed line), where we plot τs vs. η∗ at E = −200 meV (top axis). For small η∗, the scaling matches well with the real-space simulations of hBN, indicating that the puddles can be represented as a uniform energy broadening (See supplementary material). Larger values of η∗ lead to stronger mixing of different spin dynamics and τs saturates at very large η∗. There, the scal- ing of τs vs. η∗ clearly fails to replicate the DP behavior seen in the real-space simulations, since the effective broadening model does not induce the momentum scattering necessary for motional narrowing [44]. We finally explain the downturn of τs at the high energy wings of the M-shaped τs behavior in the hBN case. We compare the spin dynamics in the TB model (Eq. (2) in Methods) and the low-energy model in the absence of pud- dles (∆ = 0). In this regime τp (cid:29) TΩ, and spin dephas- ing and relaxation are driven by a combination of energy broadening and a nonuniform spin precession frequency. For the TB model, spin dynamics are calculated with the real- space approach and with a standard k-space approach and FIG. 3: Spin lifetimes for out-of-plane (a) and in-plane (b) spin injection for SiO2 substrate at impurity densities of 0.04% (black solid curves), 0.08% (red dashed curves), and 0.16% (blue dotted curves ). (c) Spin lifetime with out-of-plane spin injection for the hBN substrate at impurity densities of 0.004% (black curve) and 0.016% (red curve). when rotating spin polarization (out-of-plane vs. in-plane), and varying the density of impurities (0.04%, 0.08%, and 0.16%). Fig.3 shows the extracted τs in the out-of-plane (a) and in-plane (b) cases. The energy dependence of τs ex- hibits an M-shape increasing from a minimum at the Dirac point, with a saturation and downturn of τs for E ≥ 200 meV. The values of τs range from 50 to 400 ps depending on the initial polarization and impurity density. We observe an increase of τs with ni, which shows that a larger scattering strength reduces spin precession and dephasing, resulting in a longer spin lifetime, as described by the so-called motional (cid:107) narrowing effect [44]. Additionally, the ratio τ⊥ s (not shown) changes from 0.3 to 0.45 when ni is varied from 0.04% to 0.16%. Such behavior is expected when enhanced scattering drives more randomization of the direction of the (cid:107) Rashba SOC field, which ultimately yields τ⊥ s = 0.5 in the strong disorder limit [2, 3]. These results are fully con- sistent with the DP spin relaxation mechanism [20, 21, 44]. s /τ s /τ Fig.3(c) shows τ⊥ s for the hBN substrate (ni = 0.004% and 0.016%) where a similar M-shape is observed. While τ⊥ s (BN ) is similar to τ⊥ s (SiO2) near the Dirac point, it is much larger at higher energies, reaching nearly 1 ns (for λR = 37.4 µeV). A striking difference is that the scaling of τs with ni is opposite to that of the SiO2 case, with an in- crease in puddle density resulting in a decrease in τs, which indicates a different physical origin. For hBN, this behav- ior is reminiscent of the EY mechanism, but we will argue below that its origin is a different one. 50100150200τs⊥ (ps)-400-2000200400E (meV)0200400600800τs⊥ (ps)0.004%0.016%-400-2000200400E (meV)100150200250300350400τs (ps)0.04%0.08%0.16%c)a)b)SiO2hBN⊥⊥4 the Dirac point), see Supplementary material. This suggests that dephasing is the limiting factor of spin lifetimes in the ultraclean case. We finally note that by taking λR = 5 µeV (electric field of 1 V/nm [4]), a spin lifetime of τs (cid:39) 1.4 ns is deduced at the Dirac point, whereas at higher energies τs could reach about 7 ns. Discussion Our results show a clear transition between two different regimes of spin relaxation, mediated solely by the scattering strength of the electron-hole puddles. For hBN substrates, spin relaxation is dominated by dephasing arising from an effective energy broadening induced by the puddles, and τs scales with τp. In contrast, for SiO2 substrates dephasing is limited by motional narrowing, leading to a DP regime with τs ∝ 1/τp. Remarkably, both regimes exhibit similar values of τs at the Dirac point and a similar M-shape energy dependence (Fig.3), making it a signature of spin relaxation in graphene for all puddle strengths. The crossover between both mechanisms occurs when τp (cid:39) TΩ, which might have been realized in some experiments. This could explain some conflicting interpretations of experimental data in terms of either Elliot-Yafet or Dyakonov-perel mechanisms [11]. Our findings suggest alternative options for determining the spin relaxation mechanism in graphene from experimen- tal measurements. Indeed, the typical approach, to examine how τp and τs scale with electron density and to assign either the EY or DP mechanism accordingly, is not always appro- priate. For example, the EY mechanism in graphene is given F · τp, such that τs and τp would scale oppositely by τs ∝ E2 with respect to electron density if τp ∝ 1/EF [6]. Similarly, for our results the scaling of τp and τs with energy suggest an EY mechanism near the Dirac point and a DP mechanism at higher energies, but Figs. 3 and 4 indicate a richer behav- ior. Therefore, to determine the spin relaxation mechanism it would be more appropriate to study how τs and τp scale with defect density or mobility at each value of the electron density. Finally it should be noted that our simulations are per- formed using a constant Rashba spin-orbit coupling interac- tion, λR, which is different from the experimental situation where λR will be increased at higher charge density owing to larger applied external electric field. This might explain why, especially for hBN substrate, the simulations show a larger variation of τs in energy than the gate voltage depen- dent spin lifetimes reported in the experiments [13, 14]. FIG. 4: Low-energy spin lifetimes versus 1/τp (for initial out-of- plane spin polarization). Squares (circles) are for graphene on hBN (SiO2) substrate. Closed (open) symbols are for spin relaxation at the Dirac point (at E = −200 meV). The blue dashed line shows the spin lifetime assuming only energy broadening (top axis). Inset: spin lifetime in absence of puddles computed using the TB model in real space (red circles) or k-space (blue solid line), and the low- energy model in k-space (green dashed line), with η = 13.5 meV. give identical τs (inset of Fig.4, red circles and blue solid line), indicating the equivalence of the real- and k-space approaches in the clean limit when accounting for the full TB Hamiltonian. We observe that while for all models, the spin lifetime shows a minimum at the Dirac point (in agree- ment with experimental data, and explained by a strong spin- pseudospin coupling [34, 35]), spin transport simulations with the widely used low-energy Hamiltonian H(0) (see Methods for H(0) and green dashed line in Fig. 4 inset for re- sults) clearly cannot capture the saturation and downturn of τs(E), i.e. its full M-shape. To qualitatively reproduce the M-shape of τs(E), the first-order term of the Rashba Hamil- 2 [kx(σxsy + σysx) + ky(σxsx − σysy)], needs to tonian, λRa be included in H(0). This term introduces stronger dephas- ing at higher energy, driven by the anisotropy of the Rashba spin-orbit interaction [35]. In addition to their different energy dependence, the TB and low-energy models also yield very different spin life- times. A value of τs = 10 ns is obtained at the Dirac point for the low-energy model, which is two orders of magnitude larger than τs from the TB Hamiltonian, indicating a strong spin dephasing induced by the high-order k-terms. Inter- estingly, by studying the changes of τs(E) with respect to the Rashba SOC strength, we observe the scaling behavior τs(E) ≈ β(E)TΩ ≈ β(E) π¯h , meaning the spin relaxes af- ter a finite number of precession periods β (β (cid:39) 4.5 close to λR METHODS Model of homogeneous SOC and electron-hole puddles. [2] D. Huertas-Hernando, F. Guinea, and A. Brataas, Phys. Rev. [3] C. Ertler, S. Konschuh, M. Gmitra, and J. Fabian, Phys. Rev. B 74, 155426 (2006). B 80, 041405 (2009). 5 The tight-binding (TB) Hamiltonian for describing spin dynamics in graphene is given by H = −γ0 + iVR † i cj + i c (cid:88) † i (cid:126)z · ((cid:126)s × (cid:126)dij)cj, 2√ 3 VI (cid:104)(cid:104)ij(cid:105)(cid:105) c (cid:88) (cid:88) (cid:104)ij(cid:105) (cid:104)ij(cid:105) † i(cid:126)s · ((cid:126)dkj × (cid:126)dik)cj c (2) where γ0 is the nearest-neighbor π-orbital hopping, VI is the intrinsic SOC, and VR is the Rashba SOC. In the low-energy limit, this Hamiltonian is often approximated by a contin- uum model describing massless Dirac fermions, H(0) = ¯hvF (cid:126)σ · (cid:126)k + λI σzsz + λR ((cid:126)σ × (cid:126)s)z, where vF is the Fermi velocity, ¯h(cid:126)k is the momentum, (cid:126)s((cid:126)σ) are the spin (pseu- dospin) Pauli matrices, λR = 3 3VI. The value λI = 12 µeV is commonly used for the intrin- sic SOC of graphene [4] while the Rashba SOC is electric field-dependent. Here, we let λR = 37.4 µeV, taken from an extended sp-band TB model for graphene under an electric field of a few V/nm [4, 5]. Higher-order SOC terms in the continuum model beyond H(0) allow an extension to higher energy [45]. 2 VR, and λI = 3 √ Spin dynamics methodology. The time-dependent spin polarization of propagating wavepackets is computed through [35] (cid:126)P (E, t) = (cid:104)Ψ(t)(cid:126)sδ(E − H) + δ(E − H)(cid:126)s Ψ(t)(cid:105) 2(cid:104)Ψ(t)δ(E − H)Ψ(t)(cid:105) , (3) where (cid:126)s are the Pauli spin matrices and δ(E − H) is the spectral measure operator. The wavepacket dynamics are obtained by solving the time-dependent Schrodinger equa- tion [40], starting from a state Ψ(t = 0)(cid:105) which may have either out-of-plane (z-direction) or in-plane spin polariza- tion. An energy broadening η is introduced for expand- ing δ(E − H) through a continued fraction expansion of the Green's function [40], and mimics an effective disorder. This method has been used to investigate spin relaxation in gold-decorated graphene [35]. Here, we focus on the expec- tation value of the spin z-component Pz(E, t) = P⊥(E, t) and the spin x-component Px(E, t) = P(cid:107)(E, t). [1] N. Tombros, C. Jozsa, M. Popinciuc, H. Jonkman, and B. Van Wees, Nature 448, 571 (2007). [4] M. Gmitra, S. Konschuh, C. Ertler, C. Ambrosch-Draxl, and J. Fabian, Phys. Rev. B 80, 235431 (2009). [5] C. R. Ast and I. Gierz, Phys. Rev. B 86, 085105 (2012). [6] H. Ochoa, A. H. Castro Neto, and F. Guinea, Phys. Rev. Lett. 108, 206808 (2012). [7] W. Han, R. K. Kawakami, M. Gmitra, and J. Fabian, Nature Nanotechnology 9, 794807 (2014). [8] K. Pi, W. Han, K. M. McCreary, A. G. Swartz, Y. Li, and R. K. Kawakami, Phys. Rev. Lett. 104, 187201 (2010). [9] T.-Y. Yang, J. Balakrishnan, F. Volmer, A. Avsar, M. Jaiswal, J. Samm, S. R. Ali, A. Pachoud, M. Zeng, M. Popinciuc, et al., Phys. Rev. Lett. 107, 047206 (2011). [10] A. Avsar, T.-Y. Yang, S.Bae, J. Balakrishnan, F. Volmer, M. Jaiswal, Z. Yi, S. R. Ali, G. Gntherodt, B. H. Hong, et al., Nano Letters 11, 2363 (2011). [11] P. J. Zomer, M. H. D. Guimaraes, N. Tombros, and B. J. van Wees, Phys. Rev. B 86, 161416 (2012). [12] B. Dlubak, M.-B. Martin, D. C., S. B., X. S., R. Mattana, M. Sprinkle, C. Berger, W. de Heer, F. Petroff, et al., Nature Physics 8, 557 (2012). [13] M. Drogeler, F. Volmer, M. Wolter, B. Terrs, K. Watanabe, T. Taniguchi, G. Gntherodt, C. Stampfer, and B. Beschoten, Nano Letters 14, 6050 (2014). [14] M. H. D. Guimaraes, P. J. Zomer, J. Ingla-Ayn´es, J. C. Brant, N. Tombros, and B. J. van Wees, Phys. Rev. Lett. 113, 086602 (2014). [15] M. Venkata Kamalakar, C. Groenveld, A. Dankert, and S. P. Dash, Nature Communinations 6, 6766 (2015). [16] A. H. Castro Neto and F. Guinea, Phys. Rev. Lett. 103, 026804 (2009). [17] D. V. Fedorov, M. Gradhand, S. Ostanin, I. V. Maznichenko, A. Ernst, J. Fabian, and I. Mertig, Phys. Rev. Lett. 110, 156602 (2013). [18] M. Wojtaszek, I. J. Vera-Marun, T. Maassen, and B. J. van Wees, Phys. Rev. B 87, 081402 (2013). [19] D. Kochan, M. Gmitra, and J. Fabian, Phys. Rev. Lett. 112, [20] D. Huertas-Hernando, F. Guinea, and A. Brataas, Phys. Rev. 116602 (2014). Lett. 103, 146801 (2009). [21] P. Zhang and M. W. Wu, New Journal of Physics 14, 033015 [22] M. I. Dyakonov and V. I. Perel, Soviet Physics Solid State 13, [23] Y. Yafet, Solid State Physics (F. Seitz and D. Turnbull, 1963). [24] B. Dora, F. Muranyi, and F. Simon, Europhysics Letters 92, (2012). 3023 (1972). 17002 (2010). [25] S. Roche and S. O. Valenzuela, Journal of Physics D: Applied Physics 47, 094011 (2014). [26] J. Balakrishnan, G. K. W. Koon, M. Jaiswal, A. H. Cas- tro Neto, and B. Ozyilmaz, Nature Physics 9, 284287 (2013). [27] J. Balakrishnan, G. K. W. Koon, A. Avsar, Y. Ho, J. H. Lee, M. Jaiswal, S.-J. Baeck, J.-H. Ahn, A. Ferreira, M. A. Cazalilla, et al., Nature Communications 5, 4748 (2014). [28] Z. Wang, C. Tang, R. Sachs, Y. Barlas, and J. Shi, Phys. Rev. Lett. 114, 016603 (2015). [29] A. Avsar, J. Y. Tan, J. Balakrishnan, G. K. W. Koon, J. Lahiri, A. Carvalho, A. S. Rodin, T. Taychatanapat, E. C. T. O'Farrell, G. Eda, et al., Nature Communications 5, 4875 [30] S. Roche, J. kerman, B. Beschoten, J.-C. Charlier, M. Chshiev, S. P. Dash, B. Dlubak, J. Fabian, A. Fert, M. Guimares, et al., 2D Materials 2, 030202 (2015). [31] Y.-W. Son, M. L. Cohen, and S. G. Louie, Nature 444, 347 (2010). (2004). [32] D. Pesin and A. H. MacDonald, Nature Materials 11, 409416 113405 (2010). (2014). (2006). (2012). 6 physics Letters) 79, 57003 (2007). [43] J. W. Kłos and I. V. Zozoulenko, Phys. Rev. B 82, 081414 [44] I. Zuti´c, J. Fabian, and S. Das Sarma, Rev. Mod. Phys. 76, 323 [45] P. Rakyta, A. Korm´anyos, and J. Cserti, Phys. Rev. B 82, [33] S. Adam, S. Jung, N. N. Klimov, N. B. Zhitenev, J. A. Stroscio, and M. D. Stiles, Phys. Rev. B 84, 235421 (2011). [34] E. I. Rashba, Phys. Rev. B 79, 161409 (2009). [35] D. Van Tuan, F. Ortmann, D. Soriano, S. Valenzuela, and S. Roche, Nature Physics 10, 857 (2014). [36] J. Martin, N. Akerman, G. Ulbricht, T. Lohmann, J. H. Smet, K. von Klitzing, and Y. A., Nature Physics 4, 144 (2008). [37] A. Deshpande, W. Bao, F. Miao, C. N. Lau, and B. J. LeRoy, Phys. Rev. B 79, 205411 (2009). [38] Y. Zhang, V. Brar, C. Girit, V. Zettl, and M. Crommie, Nature Physics 5, 722 (2009). [39] J. Xue, J. Sanchez-Yamagishi, D. Bulmash, P. Jacquod, A. Deshpande, K. A. Watanabe, T. Taniguchi, P. Jarillo- Herrero, and B. J. LeRoy, Nature Materials 10, 282 (2011). [40] S. Roche, Phys. Rev. B 59, 2284 (1999). [41] F. Ortmann, A. Cresti, G. Montambaux, and S. Roche, EPL (Europhysics Letters) 94, 47006 (2011). [42] A. Rycerz, J. Tworzydo, and C. W. J. Beenakker, EPL (Euro- ADDENDUM This work has received funding from the European Union Seventh Framework Programme under grant agreement 604391 Graphene Flagship. S.R. acknowledges the Span- ish Ministry of Economy and Competitiveness for funding (MAT2012-33911), the Secretaria de Universidades e In- vestigacion del Departamento de Economia y Conocimiento de la Generalidad de Cataluna and the Severo Ochoa Pro- gram (MINECO SEV-2013-0295). F.O. would like to ac- knowledge the Deutsche Forschungsgemeinschaft (grant OR 349/1-1). Inspiring discussions with Sergio O. Valenzuela, Shaffique Adam, and Jaroslav Fabian are deeply acknowl- edged.
1807.04382	3	1807	2019-01-15T06:20:20	Twisted Bilayer Graphene: A Phonon Driven Superconductor	[ "cond-mat.mes-hall", "cond-mat.supr-con" ]	We study the electron-phonon coupling in twisted bilayer graphene (TBG), which was recently experimentally observed to exhibit superconductivity around the magic twist angle $\theta\approx 1.05^\circ$. We show that phonon-mediated electron electron attraction at the magic angle is strong enough to induce a conventional intervalley pairing between graphene valleys $K$ and $K'$ with a superconducting critical temperature $T_c\sim1K$, in agreement with the experiment. We predict that superconductivity can also be observed in TBG at many other angles $\theta$ and higher electron densities in higher Moir\'e bands, which may also explain the possible granular superconductivity of highly oriented pyrolytic graphite. We support our conclusions by \emph{ab initio} calculations.	cond-mat.mes-hall	cond-mat	Twisted Bilayer Graphene: A Phonon Driven Superconductor Biao Lian,1 Zhijun Wang,2 and B. Andrei Bernevig2, 3, 4 1Princeton Center for Theoretical Science, Princeton University, Princeton, New Jersey 08544, USA 2Department of Physics, Princeton University, Princeton, New Jersey 08544, USA 3Dahlem Center for Complex Quantum Systems and Fachbereich Physik, Freie Universitat Berlin, Arnimallee 14, 14195 Berlin, Germany 4Max Planck Institute of Microstructure Physics, 06120 Halle, Germany (Dated: January 16, 2019) We study the electron-phonon coupling in twisted bilayer graphene (TBG), which was recently experimentally observed to exhibit superconductivity around the magic twist angle θ ≈ 1.05◦. We show that phonon-mediated electron electron attraction at the magic angle is strong enough to induce a conventional intervalley pairing between graphene valleys K and K(cid:48) with a superconducting critical temperature Tc ∼ 1K, in agreement with the experiment. We predict that superconductivity can also be observed in TBG at many other angles θ and higher electron densities in higher Moir´e bands, which may also explain the possible granular superconductivity of highly oriented pyrolytic graphite. We support our conclusions by ab initio calculations. Twisted bilayer graphene (TBG) is a highly tunable condensed matter system that exhibits a rich physics. The TBG system is engineered by stacking one graphene layer on top of another at a relative twist angle θ, a procedure which produces a Moir´e pattern superlattice potential. In recent experiments [1, 2], it was observed that TBG at low Moir´e unit cell filling exhibits uncon- ventional insulator and superconductor phases near the magic angle θ = 1.05◦. At this angle, the low energy electron bands of the superlattice are predicted to be ex- tremely flat [3, 4]. The Fermi energy of the system is below 10meV, while by comparison the superconductor critical temperature Tc ∼ 1K is relatively high. Since then, some theoretical studies have been devoted to un- derstanding the insulator and superconductor phases of the TBG [5 -- 33]. A closely related system, the highly oriented pyrolytic graphite (HOPG) which contains nu- merous twisted interfaces, was also earlier reported show- ing evidences of granular superconductivity [34 -- 36], and is suggested to share a similar superconductivity mecha- nism as that in TBG [9, 37]. Here we study electron-phonon mediated superconduc- tivity of TBG. We show that the TBG Moir´e pattern ex- hibits a strong electron-phonon coupling, which can lead to a conventional superconductivity with high Tc at cer- tain twist angles and electron densities. In particular, our calculation estimates a Tc of order 1K at the magic angle θ = 1.05◦ around a filling of 2 electrons per super- lattice unit cell, in agreement with the TBG experiment [1]. Most importantly, we make the falsifiable prediction that Tc can be higher at larger electron densities and certain ranges of the twist angle θ, for instance in the second Moir´e band near θ = 0.6◦, and in the second or higher Moir´e bands for θ (cid:38) 1◦. This may explain the possible superconductivity of HOPG where the interface twist angles are mostly not at the magic angle. At last, we conjecture the insulating phase observed at 2 electrons per unit cell is a Bose Mott insulator [38]. The Moir´e superlattice of TBG is as shown in Fig. 1a, which has two lattice vectors D1 and D2 of length Dj = FIG. 1. a. Moir´e pattern of the TBG, where the AA stack- ing centers form a triangular superlattice, while AB and BA stacking centers form a dual hexagon superlattice (A and B denote the honeycomb sublattices). b. When graphene layer 1 (blue) undergoes a uniform displacement u relative to graphene layer 2 (red), the superlattice sites displaces a dis- tance (cid:101)u = γu perpendicular to u, with γ = 1/2 tan(θ/2). a0/[2 sin(θ/2)], where a0 = 0.246 nm is the graphene lattice constant. The superlattice potential significantly modifies the electron band structure of graphene [4]. The large electron-phonon coupling of TBG can be in- tuitively understood from Fig. 1. We denote the phonon field in TBG layer j (j = 1, 2) as u(j)(r), namely, the atomic displacement at coordinate r in layer j. We then define the relative displacement u = u(1) − u(2), and the center of mass displacement uc = (u(1)+u(2))/2. The key observation is that for small θ, a small in-plane relative displacement u can significantly affect the superlattice. For example, when layer 1 of the TBG undergoes a uni- form translation u relative to layer 2 as shown in Fig. 1a, the AA stacking positions will move by (cid:101)u = γu perpendicular to u, where γ = 1/[2 tan(θ/2)] (Fig. 1b). If the in-plane relative displacement u is nonuniform, it will induce a large superlattice deformation due to the amplification factor γ ≈ θ−1 (cid:29) 1 ([39, 40] Sec. I). Ac- cordingly, the low energy electrons will experience a large variation of superlattice potential, yielding a strong cou- 9 1 0 2 n a J 5 1 ] l l a h - s e m . t a m - d n o c [ 3 v 2 8 3 4 0 . 7 0 8 1 : v i X r a uuu~θD1D2AAu~u~u~AAAAABBAab2 be further folded into a 2× 2 effective Dirac Hamiltonian [4] (cid:101)H K(k) = (cid:18) 1 − 3α2 (cid:19) 1 + 6α2 vσ∗ · k , (2) where α = w/vkθ. In total, there are 4 Dirac fermions at K(cid:48) M and 4 Dirac fermions at KM near zero energy, due to the valley K, K(cid:48) and spin ↑,↓ 4-fold degeneracy (further momentum hoppings are needed in Eq. (1) to obtain the Dirac fermions at KM ). The Dirac fermions at valley K(cid:48) have an opposite helicity, described by Eq. (2) with σ∗ · k → σ · k. The magic angle θ ≈ 1.05◦ is given by α2 = 1/3, where the Fermi velocity of the Dirac band becomes zero. Numerical calculations at the magic angle show the entire band width of the lowest two bands can be as low as 1 meV [4]. The coupling between electrons and long wavelength phonons can be obtained by examining the change of elec- tron band energies under uniform lattice deformations. Under the superlattice deformation induced by a relative displacement u, one can show that the momentum vec- tors qj (Fig. 2b) are deformed by δq1 = γkθ(∂xux, ∂yux), and δq2,3 = γkθ(±√ 3 2 ∂yuy) ([40] Sec. I). This induces a change of kj = k − qj in the Hamiltonian (1), and thus perturbs the electron band energies. The variations of v and w are subleading com- pared to δqj, and will be ignored here. The variation in the folded 2 × 2 Hamiltonian (2), namely, the electron- phonon coupling Hep(k) = δ(cid:101)H(k), can be derived to be 2 ∂yux ± √ 2 ∂xuy − 1 2 ∂xux,− 1 3 ([40] Sec. I) H η,ζ,s ep H η,ζ,s C3 (k) = H η,ζ,s (k) = g1αηγvψ C3 (k) + H η,ζ,s SO(2)(k) , † k(cid:48)[kx(∂yux + ∂xuy) + ky(∂xux − ∂yuy)]ψk , SO(2)(k) = γvψ H η,ζ,s † k(cid:48)[g2α(ησxkx − σyky)(∂yux − ∂xuy) + g3α(ησxky + σykx)(∂xux + ∂xuy)]ψk , 1+6α2 . ψk and ψ (3) where index η = ±1 is for graphene valley K, K(cid:48), ζ = ±1 M , s = ±1 is for spin ↑,↓, is for Moir´e BZ valley KM , K(cid:48) and we have defined g1α = 9α2(1+3α2) (1+6α2)2 , g2α = (1+6α2)2 † and g3α = 3α2 k are the Dirac electron an- nihilation and creation operators, and k = (k + k(cid:48))/2 is the average momentum of the initial and final elec- tron state before and after phonon emission (absorption). Note that Hep is independent of ζ and s, and contains two parts HC3 and HSO(2), which are C3z and SO(2) rota- tionally invariant about z axis, respectively. Besides, Hep respects the TBG 2-fold rotation symmetry C2x about x axis, which transforms (ux, uy) to (−ux, uy). 9α2 We also need to know the phonon spectrum of the TBG. Previous studies show that the coupling between in-plane phonons of the two layers of TBG is extremely small [41], so we can approximate the interlayer coupling Illustration of the graphene BZs of two layers, FIG. 2. a. and their relation to the Moir´e BZ. b. Under phonon induced superlattice deformations, qj are deformed, which leads to the change in electron band energies. pling with the in-plane relative displacement phonon field u. In contrast, the center of mass displacement uc has no amplified effect on the superlattice, and has much weaker couplings to electrons ([40] Sec. I). Therefore, we shall only focus on the electron-phonon coupling of the relative displacement field u. The low energy band structure of the TBG can be cal- culated using the continuum model in momentum space constructed in Ref. [4]. Fig. 2a shows the hexagonal graphene Brillouin zones (BZs) of the two layers, which are relatively twisted by θ. When the two layers are de- coupled, the low energy electrons of each layer are Dirac fermions at K and K(cid:48) points, which are described by Hamiltonian hK(k) = v(σxkx − σyky) = vσ∗ · k and (k) = −vσ · k, respectively. Here σx,y,z are the hK(cid:48) Pauli matrices for sublattice indices, v ≈ 610 meV·nm is the graphene Fermi velocity, and momentum k is mea- sured from the Dirac point. In addition, each Dirac band has a 2-fold spin degeneracy, and we assume zero spin or- bit coupling. The K (K(cid:48)) points of the two layers differ by momentum vectors qj (−qj) as shown in Fig. 2a (j = 1, 2, 3), which constitute the edges of the hexag- onal superlattice Moir´e BZ. Their lengths are given by qj = kθ = 8π sin(θ/2)/3a0. When the interlayer hopping is introduced, a state of momentum k in layer 1 can hop with a state of momen- tum p(cid:48) in layer 2 if k − p(cid:48) = qj or higher superlattice reciprocal vectors [4]. If we only keep the nearest hop- pings, to the lowest order, the Hamiltonian at valley K and near Moir´e BZ K(cid:48) M point (Fig. 2a) has the truncated form [4]  hK(k) wT1 † wT 1 † wT 2 † wT 3 0 0 hK(k1) wT2 0 hK(k2) wT3 0 0 0 hK(k3)  , (1) H K(k) = M point, kj = k − qj (j = where k is measured from K(cid:48) 1, 2, 3), the matrices Tj are given by T1 = 1 + σx, T2 = √ 1 − 1 2 σy, and w ≈ 110 2 σy, T3 = 1 − 1 meV is the nearest momentum hopping amplitude. In the vicinity of k = 0 and zero energy, the Hamiltonian (1) can 2 σx − √ 2 σx + 3 3 Γq1q2q3KK'θq1q2q3q + δq33q + δq22q + δq11k space deformationMoire BZGraphene BZab(1)K'(2)(1)K(2)KMK'Mkkyx3 1+6α2 v(k−kF ) being the band energy at k, while ck,I − 1−3α2 † and c k,I are electron annihilation and creation operators in the Dirac hole band I. To simplify the result, we take the approximation cL = cT (both around 104m/s). For θ near the magic angle (α2 ≈ 1/3), we find the interaction in the lowest two Moir´e bands is ω2 fηη(cid:48)(ϕk, ϕk(cid:48)), (7) where M is the Carbon atomic mass, p = k − k(cid:48), ϕk = arg(kx + iky) is the polar angle of k, and ηη(cid:48) kk(cid:48) = φη† k(cid:48) φη 2 being the Dirac hole band wave function at valley η. The function fηη(cid:48) is given by √ k = (1,−ηe−iηϕk)T / −k with φη −k(cid:48)φη(cid:48) V II(cid:48) kk(cid:48) (ω) Ωs ≈ 2v2k2 ω2 − ω2 F ηη(cid:48) kk(cid:48) kφη(cid:48)† 9M c2 T  −1 − 2 cos(ϕk − ϕk(cid:48)), (η = η(cid:48)) (cid:12)(cid:12)(cid:12)1 − η e2iϕk−e2iϕk(cid:48) .(η = −η(cid:48)) e−iϕk−e (cid:12)(cid:12)(cid:12)2 −iϕk(cid:48) fηη(cid:48)(ϕk, ϕk(cid:48)) = p,T p,T C3 and H η,ζ,s (8) The interaction V II(cid:48) kk(cid:48) (ω) is independent of spin s and Moir´e valley ζ. At low energies ω < ωp,T , fη,−η > 0 indicates the in- tervalley interaction between K and K(cid:48) (η = −η(cid:48)) is at- tractive. In contrast, fη,η is on-average negative, and one can prove that the intravalley interaction (η = η(cid:48)) is re- pulsive in all pairing channels ([40] Sec. IIC). This is due to the fact that the hole (or electron) band projections of H η,ζ,s SO(2) in Eq. (3) are odd and even under k, k(cid:48) → −k,−k(cid:48), or under η → −η, respectively ([40] Sec. IIC). Assume an electron state (wave packet) kK,ζ,s(cid:105) around momentum k at valley K experiences a phonon- induced lattice potential (cid:104)H K,ζ,s SO(2)(k)(cid:105) = UC3 + USO(2). By symmetry, the state − kK,ζ(cid:48),s(cid:48)(cid:105) at the same valley K will feel a potential −UC3 + USO(2), while the state − kK(cid:48),ζ(cid:48),s(cid:48)(cid:105) in the opposite valley K(cid:48) will feel a potential UC3 + USO(2). Therefore, two electrons kK,ζ(cid:48),s(cid:48)(cid:105) and − kK(cid:48),ζ(cid:48),s(cid:48)(cid:105) in opposite valleys feel the same phonon-induced lattice potential, which induces an effective attraction between them. In contrast, two elec- trons kK,ζ(cid:48),s(cid:48)(cid:105) and − kK,ζ(cid:48),s(cid:48)(cid:105) in the same valley K feel different potentials, so the effective attraction between them is weaker or even absent. Therefore, the intervalley Cooper pairing is preferred. (k)(cid:105) + (cid:104)H K,ζ,s C3 This does not yet uniquely determine the form of pair- ing. Since the Dirac bands are degenerate with respect to indices ζ and s, the intervalley pairing could be either spin-singlet Moir´e valley-triplet, or spin-triplet Moir´e valley-singlet. Here we shall simply assume the pairing is time reversal invariant, which is generically more robust under non-magnetic disorders [42]. This forces a pairing between opposite Moir´e valleys and opposite spins, and yields an intervalley pairing amplitude ([40] Sec. III) = sδs,−s(cid:48)δζ,−ζ(cid:48)δη,−η(cid:48)(cid:101)∆(ηk) , ∆ηη(cid:48),ζζ(cid:48),ss(cid:48) where (cid:101)∆(k) is a real function of ϕk = arg(kx + iky). The Numerically, (cid:101)∆(k) can be solved and has the shape (9) k FIG. 3. a. The process two electrons of momentum k and −k exchanges a phonon of momentum p = k(cid:48)−k, which mediates the electron-electron interaction in Eq. (7). b. Illustration of phonon-induced potentials for electrons at valleys K and K(cid:48) (we have plotted them at Moir´e valleys KM and K(cid:48) M , re- spectively, but the potentials do not depend on Moir´e valley), where US and ±UC3 are contributed by H η,ζ,s SO(2) and H η,ζ,s , respectively. The dashed circles represent the Fermi surfaces. c. Intervalley pairing (cid:101)∆(k) solved numerically as a function C3 of ϕ(k), which is s-wave ([40] Sec. III). as zero. In this approximation, the TBG in-plane phonon spectrum is simply that of two isolated graphene mono- layers folded into the Moir´e BZ. The lowest bands of phonon field u is described by Hamiltonian Hph = † † p,Lap,L + ωp,T a p,T ap,T , (4) (cid:88) (cid:16)ωp,La p (cid:17) † † p,L and ap,T , a where ap,L, a p,T are the annihilation and creation operators of longitudinal and transverse polar- ized phonons, respectively. The frequencies ωp,L = cLp and ωp,T = cT p are acoustic, with p = p. cL, cT are the longitudinal and transverse sound speeds of mono- layer graphene. The phonon field u at long wavelengths is u(r) = (ipup,L + iz × pup,T ) , (5) (cid:88) eip·r√ (cid:113) Ω p NsΩs 2M ωp,χ † −p,χ) for χ = L, T polar- where up,χ = (ap,χ + a izations, Ω and Ωs are the unit cell areas of the graphene lattice and Moir´e superlattice, respectively, and Ns is the number of supercells. There are also many optical phonon bands in the Moir´e BZ corresponding to short wavelength components of u, but here we will only focus on the lowest acoustic phonon bands in Eq. (4), since Hep in Eq. (3) is derived for long wavelength deformations. We now assume that the Fermi surfaces are given by k = kF in the Dirac hole (or electron) bands, and cal- culate the phonon mediated electron-electron interaction near the Fermi surfaces. The Bardeen-Cooper-Schrieffer (BCS) channel of the interaction takes the generic form: H (ph) int = V II(cid:48) kk(cid:48) (ω) NsΩs † † −k(cid:48),I(cid:48)c−k,I(cid:48)ck,I , k(cid:48),I c c (6) (cid:88) k,k(cid:48) where I = (η, ζ, s) denotes indices for the eight Dirac cones, the frequency ω = (ξk(cid:48) − ξk)/ with ξk = K, KMK', K'Mq1q2q3USO(2)+UC3USO(2) - UC3k-kk-kUSO(2) -UC3USO(2)+UC3bakk'-k'-kpc∆(k)~ϕ(k)kykx(cid:112)(4πne)1/2e2F /I ∼ 10ωD [46, 47]. This agrees well the plasma frequency, which is 4 roughly ωpe ∼ with the experimentally observed Tc. We do emphasize that our Tc estimation is very rough, with inaccuracies from both λ, µ∗ c and the McMillan formula itself for large λ. The above electron-phonon coupling calculation can be easily generalized to other twist angles and electron den- sities. We still keep only the nearest momentum hoppings in the continuum model of TBG, but truncate the Hamil- tonian at sufficiently high momentum to obtain more ac- curate band structures. We then numerically calculate the energy change in each electron band under small de- formations of qj, and verify it is comparable to our ab initio results ([40] Sec. V). Subtracting the contribution from Moir´e BZ deformations ([40] Sec. IV), we can es- timate the electron-phonon coupling of each band and the BCS coupling strength λ ≈ NDVkk. Fig. 4a and 4b show the DOS ND and BCS coupling strength λ with respect to the number of electrons filling per superlattice unit cell n = neΩs at θ = 1.05◦. The DOS is predom- inantly high for the first conduction and valence bands (n < 4). However, the BCS coupling λ for n < 4 and for n > 4 are of the same order, despite the fact that the DOS is much lower (∼ 0.05 meV−1 · Ω−1 s ) at n > 4. Numerically, this is because the energy suscep- tibility to deformations of a Moir´e band becomes large when the band energy is large ([40] Sec. IV). This im- plies possible BCS superconductivity at higher n. Fig. 4c shows Tc from the McMillan formula with respect to angle θ and number of electrons per unit cell n. There is only a narrow superconducting region at n < 4 near the magic angle, which correspond to the superconductivity observed in TBG. In contrast, the superconductivity oc- curs in a wide range of θ (cid:38) 1◦ for n near 8 and higher. There are also parameter spaces for θ < 1◦ where Tc is of order of 1K, e.g., 4 (cid:46) n (cid:46) 8 near θ = 0.8◦, 2 (cid:46) n (cid:46) 12 near θ = 0.6◦, and 8 (cid:46) n (cid:46) 12 around θ = 0.3◦. The vast superconductivity region indicates the Moir´e pattern generically enhances the electron-phonon coupling of all bands. This may explain the possible superconductiv- ity of HOPG, which contains numerous Moir´e interfaces with different electron densities n and twist angles θ. Lastly, we comment that the strong phonon-mediated attraction may favor a Bose Mott insulator [38] for the TBG insulating phase observed at n = 2 [1, 2]. This is because the attraction may pair the electrons into charge 2e bosons (Cooper pairs), and for n = 2, the system has one boson per unit cell, thus may form a Bose Mott insulator [38], with a possible charge density wave order. Such a phase will have a resistivity around h/(2e)2 ≈ 6kΩ at the superconductor-insulator transition, due to charge 2e carriers [48 -- 50]. However, the experimentally observed quantum oscillations [1] indicates these Cooper pairs have to break down for magnetic fields above 1T, if this explanation is correct. In summary, we have shown the electron-phonon cou- pling is strong in TBG, and can lead to BCS supercon- FIG. 4. a. The density of states ND as a function of number of electrons per superlattice unit cell n, showing the lowest conduction and valence bands are pretty flat. b. The esti- mated BCS coupling strength λ ≈ NDVkk(cid:48) (0) as a function of n. c. The superconducting Tc with respect to n and θ estimated from the McMillan formula. shown in Fig. (3)c, which is nodeless and dominated by s- wave. We note that an earlier phonon study [13] obtained both s-wave and d-wave, while a recent atomistic study supports s-wave [43]. Substituting the realistic parameters into Eq. (7) and taking kF ∼ kθ, we find that the phonon mediated inter- valley attraction is of order of magnitude −V II(cid:48) kk(cid:48) (0)/Ωs ∼ 1 meV around the magic angle, which is comparable to the Fermi energy F and the Debye frequency of acous- tic Moir´e phonon bands ωD ∼ cT kθ ≈ 2 meV. When the optical phonon contributions are included, the at- traction could be further enhanced. Since the density of states (DOS) is as large as ND (cid:38) 1 meV−1 · Ω−1 at the magic angle, the BCS coupling strength λ ≈ NDV II(cid:48) kk(cid:48) (cid:38) 1 is strong. The screened Coulomb inter- action takes the form Ve(q) = 2πe2/q(q), where (q) is the screened dielectric function at momentum q. Here we simply adopt the (two-dimensional) Thomas-Fermi III) (q) ≈ I (1 + qT F /q), approximation ([40] Sec. where I ≈ 2 ∼ 10 is the dielectric constant of undoped graphene, and qT F ≈ 2πe2(∂ne/∂µ)/I = 2πe2ND/I is the Thomas-Fermi momentum (ne and µ are the electron density and chemical potential, respectively). With ND (cid:38) 1 meV−1 · Ω−1 around the magic angle, qT F (cid:38) 50kθ (cid:29) q, so the screened Coulomb potential Ve(q) ≈ 2πe2/I qT F ∼ N−1 D , yielding a Coulomb cou- pling strength µc ≈ NDVe(q) ∼ 1. If we adopt the McMillan formula for superconductor Tc [44, 45], taking λ = 1.5 and µc = 1, we obtain s s (cid:21) (cid:20) ωD 1.45kB − 1.04(1 + λ) ≈ 0.9K (10) exp Tc = λ − µ∗ at the magic angle, where µ∗ is the reduced Coulomb coupling strength, ωpe c = µc/[1 + µc ln(ωpe/ωD)] is c (1 + 0.62λ) (c)-30-20-10102030n024ND (s-1meV-1)0-30-20-10102030n0240θ=1.05θ=1.05(a)(b)n08-816-1624-2410.80.60.40.221.81.61.41.2θ00.50.20.11Tc (K)ductivity with a Tc in agreement with the experiment [1]. We find the intervalley pairing between valleys K and K(cid:48) is favored in the flat bands near the magic angle, which is topologically trivial. Besides, we predict that supercon- ductivity can be achieved at many other angles and at higher electron fillings (Fig. 4c), and we expect our pre- diction to be verified by higher doping TBG experiments in the future [51, 52]. [1] Yuan Cao, Valla Fatemi, Shiang Fang, Kenji Watanabe, Takashi Taniguchi, Efthimios Kaxiras, and Pablo Jarillo- Herrero, "Unconventional superconductivity in magic- angle graphene superlattices," Nature 556, 43 (2018). Jason Y. Luo, [2] Yuan Cao, Valla Fatemi, Ahmet Demir, Shiang Fang, Spencer L. Tomarken, Javier D. Sanchez-Yamagishi, Kenji Watanabe, Takashi Taniguchi, Efthimios Kaxiras, Ray C. Ashoori, and Pablo Jarillo- Herrero, "Correlated insulator behaviour at half-filling in magic-angle graphene superlattices," Nature 556, 80 (2018). [3] E. Su´arez 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, 121407 (2010). [4] Rafi Bistritzer and Allan H. MacDonald, "Moir´e bands in twisted double-layer graphene," Proceedings of the National Academy of Sciences 108, 12233 -- 12237 (2011), http://www.pnas.org/content/108/30/12233.full.pdf. [5] Noah F. Q. Yuan and Liang Fu, "Model for the metal- insulator transition in graphene superlattices and be- yond," Phys. Rev. B 98, 045103 (2018). [6] Hoi Chun Po, Liujun Zou, Ashvin Vishwanath, and T. Senthil, "Origin of mott insulating behavior and su- perconductivity in twisted bilayer graphene," Phys. Rev. X 8, 031089 (2018). [7] Cenke Xu and Leon Balents, "Topological superconduc- tivity in twisted multilayer graphene," Phys. Rev. Lett. 121, 087001 (2018). [8] B. Roy and V. Juricic, "Unconventional superconduc- tivity in nearly flat bands in twisted bilayer graphene," ArXiv e-prints (2018), arXiv:1803.11190 [cond-mat.mes- hall]. [9] G. E. Volovik, "Graphite, graphene and the flat (2018), JETP Letters band superconductivity," 10.1134/S0021364018080052. [10] B. Padhi, C. Setty, and P. W. Phillips, "Wigner Crystal- lization in lieu of Mottness in Twisted Bilayer Graphene," ArXiv e-prints (2018), arXiv:1804.01101 [cond-mat.str- el]. [11] J. F. Dodaro, S. A. Kivelson, Y. Schattner, X. Q. Sun, and C. Wang, "Phases of a phenomenological model of twisted bilayer graphene," Phys. Rev. B 98, 075154 (2018). [12] G. Baskaran, "Theory of Emergent Josephson Lattice in Neutral Twisted Bilayer Graphene (Moi\'re is Differ- (2018), arXiv:1804.00627 [cond- ent)," ArXiv e-prints 5 ACKNOWLEDGMENTS Acknowledgments. BL is supported by Princeton Cen- ter for Theoretical Science at Princeton University. ZW and BB are supported by the Department of Energy Grant No. de-sc0016239, the National Science Founda- tion EAGER Grant No. noaawd1004957, Simons Inves- tigator Grants No. ONRN00014-14-1-0330, No. ARO MURI W911NF-12-1-0461, and No. NSF-MRSEC DMR- 1420541, the Packard Foundation, the Schmidt Fund for Innovative Research. mat.supr-con]. [13] Fengcheng Wu, A. H. MacDonald, and Ivar Mar- tin, "Theory of phonon-mediated superconductivity in twisted bilayer graphene," Phys. Rev. Lett. 121, 257001 (2018). [14] Hiroki Isobe, Noah F. Q. Yuan, and Liang Fu, "Uncon- ventional superconductivity and density waves in twisted bilayer graphene," Phys. Rev. X 8, 041041 (2018). [15] T. Huang, L. Zhang, and T. Ma, "Antiferromagneti- cally ordered Mott insulator and d + id superconductiv- ity in twisted bilayer graphene: A quantum Monte carlo study," ArXiv e-prints (2018), arXiv:1804.06096 [cond- mat.supr-con]. [16] Y.-Z. You and A. Vishwanath, "Superconductivity from Valley Fluctuations and Approximate SO(4) Symme- try in a Weak Coupling Theory of Twisted Bilayer Graphene," ArXiv e-prints (2018), arXiv:1805.06867 [cond-mat.str-el]. [17] X.-C. Wu, K. A. Pawlak, C.-M. Jian, and C. Xu, "Emergent Superconductivity in the weak Mott insulator phase of bilayer Graphene Moir\'e Superlattice," ArXiv e-prints (2018), arXiv:1805.06906 [cond-mat.str-el]. [18] Y.-H. Zhang, D. Mao, Y. Cao, P. Jarillo-Herrero, and T. Senthil, "Moir\'e Superlattice with Nearly Flat Chern Bands: Platform for (Fractional) Quantum Anoma- lous Hall Effects and Unconventional Superconductivity," ArXiv e-prints (2018), arXiv:1805.08232 [cond-mat.str- el]. [19] Jian Kang and Oskar Vafek, "Symmetry, maximally localized wannier states, and a low-energy model for twisted bilayer graphene narrow bands," Phys. Rev. X 8, 031088 (2018). [20] Mikito Koshino, Noah F. Q. Yuan, Takashi Koretsune, Masayuki Ochi, Kazuhiko Kuroki, and Liang Fu, "Max- imally localized wannier orbitals and the extended hub- bard model for twisted bilayer graphene," Phys. Rev. X 8, 031087 (2018). [21] Dante M. Kennes, Johannes Lischner, and Christoph Karrasch, "Strong correlations and d+id superconductiv- ity in twisted bilayer graphene," Phys. Rev. B 98, 241407 (2018). [22] L. Zhang, "Low-energy Moir\'e Band Formed by Dirac Zero Modes in Twisted Bilayer Graphene," ArXiv e- prints (2018), arXiv:1804.09047 [cond-mat.mes-hall]. [23] J. M. Pizarro, M. J. Calder´on, and E. Bascones, "The nature of correlations in the insulating states of twisted bilayer graphene," ArXiv e-prints (2018), arXiv:1805.07303 [cond-mat.str-el]. [24] Francisco Guinea and Niels R. Walet, "Electrostatic effects, band distortions, and superconductivity in twisted graphene bilayers," Proceedings of the Na- tional Academy of Sciences 115, 13174 -- 13179 (2018), https://www.pnas.org/content/115/52/13174.full.pdf. [25] Alex Thomson, Shubhayu Chatterjee, Subir Sachdev, and Mathias S. Scheurer, "Triangular antiferromag- netism on the honeycomb lattice of twisted bilayer graphene," Phys. Rev. B 98, 075109 (2018). [26] Masayuki Ochi, Mikito Koshino, and Kazuhiko Kuroki, "Possible correlated insulating states in magic-angle twisted bilayer graphene under strongly competing in- teractions," Phys. Rev. B 98, 081102 (2018). [27] Xiao Yan Xu, K. T. Law, and Patrick A. Lee, "Kekul´e valence bond order in an extended hubbard model on the honeycomb lattice with possible applications to twisted bilayer graphene," Phys. Rev. B 98, 121406 (2018). [28] Teemu J. Peltonen, Risto Ojajarvi, and Tero T. Heikkila, "Mean-field theory for superconductivity in twisted bi- layer graphene," Phys. Rev. B 98, 220504 (2018). [29] M. Fidrysiak, M. Zegrodnik, and J. Spa(cid:32)lek, "Unconven- tional topological superconductivity and phase diagram for an effective two-orbital model as applied to twisted bilayer graphene," Phys. Rev. B 98, 085436 (2018). [30] Liujun Zou, Hoi Chun Po, Ashvin Vishwanath, and T. Senthil, "Band structure of twisted bilayer graphene: Emergent symmetries, commensurate approximants, and wannier obstructions," Phys. Rev. B 98, 085435 (2018). [31] J. Gonz´alez and T. Stauber, "Kohn-Luttinger supercon- ductivity in twisted bilayer graphene," ArXiv e-prints (2018), arXiv:1807.01275 [cond-mat.mes-hall]. [32] Ying Su and Shi-Zeng Lin, "Pairing symmetry and spontaneous vortex-antivortex lattice in superconduct- ing twisted-bilayer graphene: Bogoliubov-de gennes ap- proach," Phys. Rev. B 98, 195101 (2018). [33] Huaiming Guo, Xingchuan Zhu, Shiping Feng, and Richard T. Scalettar, "Pairing symmetry of interacting fermions on a twisted bilayer graphene superlattice," Phys. Rev. B 97, 235453 (2018). [34] Pablo Esquinazi, "Invited review: Graphite and its hid- den superconductivity," Papers in Physics 5 (2013). [35] A. Ballestar, J. Barzola-Quiquia, T. Scheike, and P. Esquinazi, "Josephson-coupled superconducting re- gions embedded at the interfaces of highly oriented py- rolytic graphite," New Journal of Physics 15, 023024 (2013). [36] Ana Ballestar, Tero T Heikkila, and Pablo Esquinazi, "Size dependence of the josephson critical behavior in py- rolytic graphite tem lamellae," Superconductor Science and Technology 27, 115014 (2014). [37] P. Esquinazi, T. T. Heikkila, Y. V. Lysogorskiy, D. A. Tayurskii, and G. E. Volovik, "On the superconductiv- ity of graphite interfaces," JETP Letters 100, 336 -- 339 (2014). [38] Matthew P. A. Fisher, Peter B. Weichman, G. Grin- stein, and Daniel S. Fisher, "Boson localization and the superfluid-insulator transition," Phys. Rev. B 40, 546 -- 570 (1989). [39] See the Supplemental Video. [40] See Supplemental Material for details. [41] Alexandr I. Cocemasov, Denis L. Nika, and Alexander A. Balandin, "Phonons in twisted bilayer graphene," Phys. Rev. B 88, 035428 (2013). 6 [42] P.W. Anderson, "Theory of dirty superconductors," Journal of Physics and Chemistry of Solids 11, 26 -- 30 (1959). [43] Young Woo Choi and Hyoung Joon Choi, "Strong electron-phonon coupling, electron-hole asymmetry, and nonadiabaticity in magic-angle twisted bilayer graphene," Phys. Rev. B 98, 241412 (2018). [44] W. L. McMillan, "Transition temperature of strong- coupled superconductors," Phys. Rev. 167, 331 -- 344 (1968). [45] P. B. Allen and R. C. Dynes, "Transition temperature of strong-coupled superconductors reanalyzed," Phys. Rev. B 12, 905 -- 922 (1975). [46] E. H. Hwang and S. Das Sarma, "Dielectric function, screening, and plasmons in two-dimensional graphene," Phys. Rev. B 75, 205418 (2007). [47] S. Das Sarma and E. H. Hwang, "Collective modes of the massless dirac plasma," Phys. Rev. Lett. 102, 206412 (2009). [48] Matthew P. A. Fisher, G. Grinstein, and S. M. Girvin, "Presence of quantum diffusion in two dimensions: Uni- versal resistance at the superconductor-insulator transi- tion," Phys. Rev. Lett. 64, 587 -- 590 (1990). [49] Ali Yazdani and Aharon Kapitulnik, "Superconducting- insulating transition in two-dimensional a-moge thin films," Phys. Rev. Lett. 74, 3037 -- 3040 (1995). [50] Myles A. Steiner, Nicholas P. Breznay, and Aharon Kapitulnik, "Approach to a superconductor-to-bose- insulator transition in disordered films," Phys. Rev. B 77, 212501 (2008). [51] Kyounghwan Kim, Ashley DaSilva, Shengqiang Huang, Babak Fallahazad, Stefano Larentis, Takashi Taniguchi, Kenji Watanabe, Brian J. LeRoy, Allan H. MacDonald, and Emanuel Tutuc, "Tunable moir´e bands and strong correlations in small-twist-angle bilayer graphene," Pro- ceedings of the National Academy of Sciences 114, 3364 (2017). [52] Dillon Wong, Yang Wang, Jeil Jung, Sergio Pezzini, Ash- ley M. DaSilva, Hsin-Zon Tsai, Han Sae Jung, Ramin Khajeh, Youngkyou Kim, Juwon Lee, Salman Kahn, Sajjad Tollabimazraehno, Haider Rasool, Kenji Watan- abe, Takashi Taniguchi, Alex Zettl, Shaffique Adam, Allan H. MacDonald, and Michael F. Crommie, "Lo- cal spectroscopy of moir´e-induced electronic structure in gate-tunable twisted bilayer graphene," Phys. Rev. B 92, 155409 (2015). [53] L. D. Landau, L. P. Pitaevskii, A. M. Kosevich, and E. M. Lifshitz, "Theory of elasticity," (2012). [54] Song, Wang, Fang and Bernevig, in preparation. [55] P. E. Blochl, "Projector augmented-wave method," Phys. Rev. B 50, 17953 -- 17979 (1994). [56] G. Kresse and D. Joubert, "From ultrasoft pseudopoten- tials to the projector augmented-wave method," Phys. Rev. B 59, 1758 -- 1775 (1999). [57] G. Kresse and J. Hafner, "Ab initio molecular dynamics for liquid metals," Phys. Rev. B 47, 558 -- 561 (1993). [58] G. Kresse and J. Furthmller, "Efficiency of ab-initio total energy calculations for metals and semiconductors using a plane-wave basis set," Computational Materials Science 6, 15 -- 50 (1996). [59] P. Hohenberg and W. Kohn, "Inhomogeneous electron gas," Phys. Rev. 136, B864 -- B871 (1964). [60] 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). 7 SUPPLEMENTARY MATERIAL FOR "TWISTED BILAYER GRAPHENE: A PHONON DRIVEN SUPERCONDUCTOR" I. DERIVATION OF ELECTRON-PHONON COUPLING AT MAGIC ANGLE In this section we first briefly recall the continuum model in momentum space formulated in Ref. [4], and then use it to derive the coupling between electrons and interlayer phonons is derived. a · a(j) Fig. 5 illustrates the real space configuration of two graphene layers which have a relative twist angle θ. We denote the lattice vectors of layer j (j = 1, 2) as a(j) 1 and G(j) 2 , which satisfy G(j) are equal to the lattice constant a0 = 0.246nm. Each layer j consists of two sublattice positions A and B, which are located at τ (j) B in the unit cell of layer j. Without loss of generality, we can choose sublattice A in layer j as the origin of the unit cell of layer j, so that τ (j) A = 0, and τ (j) i are rotated by θ from one another, and so do τ (1) and τ (2). More explicitly, the above vectors in components are given by 2 , and the reciprocal lattice vectors of layer j as G(j) b = 2πδab. The norms of the lattice vectors a(j) B = τ (j) as shown in Fig. 5. The vectors a1 A and τ (j) 1 and a(j) i and a2 i a(j) 1 = a0R(j) θ/2 (cid:33)T √ 3 2 (cid:32) − 1 2 , (cid:32) G(j) 1 = 4π√ 3a0 R(j) θ/2 − 1 2 , for layer j = 1, 2, where R(1) θ/2 = , θ/2 √ 3 2 a(j) 2 = a0R(j) (cid:33)T (cid:18) cos(θ/2) − sin(θ/2) 2 = G(j) , sin(θ/2) cos(θ/2) (cid:19) (cid:32) √ ,− − 1 2 3 2 (cid:33)T (cid:32) , 4π√ 3a0 R(j) θ/2 − 1 2 , , R(2) θ/2 = sin(θ/2) − sin(θ/2) cos(θ/2) (cid:19) , τ (j) B = a0√ 3 R(j) θ/2 (0, 1)T , (11) (12) 3 2 τ (j) A = (0, 0)T , √ (cid:33)T (cid:18) cos(θ/2) , (cid:17) which are the rotation matrices of angle ±θ/2, respectively. The interlayer hopping t(r) between two atoms in different layers is generically a function of their in-plane distance r. When transformed into momentum space, the electron hopping from momentum p(cid:48) and sublattice β in layer 2 to momentum k and sublattice α in layer 1 takes the form (cid:88) (cid:16) R(1),R(2) tk+G(1) n1,m1 Ω T αβ kp(cid:48) = (cid:88) = 1 N (cid:88) n1,m1 n2,m2 R(1) + τ (1) α − R(2) − τ (2) β t eip(cid:48)·(R(2)+τ (2) β )−ik·(R(1)+τ (1) α ) δk+G(1) n1,m1 ,p(cid:48)+G(2) n2 ,m2 eiG(2) n2,m2 ·τ (2) β −iG(1) n1,m1 ·τ (1) α , (13) √ where α, β = A, B are sublattice indices, k and p(cid:48) are measured from Γ point of the graphene Brillouin zone (BZ), runs Ω = over all n, m ∈ Z reciprocal lattices in layer j. 0/2 is the area of graphene unit cell, tk is the Fourier transform of t(r), and G(j) 1 + mG(j) n,m = nG(j) 3a2 2 We first consider the low energy physics near the K points of the graphene BZs of the two layers (see main text Fig. 2), namely, k and p(cid:48) near the Dirac point momenta K(1) 2 )/3, respectively. It is shown that the hopping tk decays exponentially with respect to k [4], so a good approximation D , and is to keep only the 3 leading nearest hopping terms tk+G(1) approximate them to tK(1) n,m around the magnitude K(1) with k + G(1) D = −(G(1) D = −(G(2) 2 )/3 and K(2) 1 + G(2) 1 + G(1) n,m . We then define the three vectors D q1 = K(2) D − K(1) D , q2 = C3zq1 = K(2) D + G(2) , q3 = C 2 3zq1 = K(2) D + G(2) 2 − K(1) D − G(1) 2 , where C3z is the 3-fold rotation about z axis. Explicitly, qj in components are 1 D − G(1) 1 − K(1) (cid:32)√ (cid:33)T 3 2 , 1 2 (cid:32) √ 3 2 , 1 2 − (cid:33)T , q1 = kθ(0,−1)T , q2 = kθ , q3 = kθ 8 FIG. 5. a. Illustration of the real space TBG lattice, where the twist angle is θ = 10◦. b. The graphene lattice vectors and reciprocal vectors in each layer. c. The graphene BZs of the two layers. where kθ = qj = (8π/3a0) sin(θ/2). Under the above nearest hopping approximation, an electron state with momentum p(cid:48) in layer 2 can hop to an electron state with momentum k in layer 1 if k − p(cid:48) = qj. Since we are interested in the low energy band structure near the graphene BZ K (or K(cid:48)) point, hereafter we set the origin of k and p(cid:48) at K (or K(cid:48)) point of each layer. The effective Hamiltonian of each layer is given by (k) = −vσ· k at K(cid:48) point of each layer (related hK(k) = v(kxσx − kyσy) = vσ∗ · k at K point of each layer, and hK(cid:48) to hK(k) via the time reversal transformation T ), where v is the fermi velocity, and σx,y,z are the Pauli matrices for sublattice indices. Note that the Dirac fermions at graphene valleys K and K(cid:48) have opposite helicities. Therefore, to the lowest order, the TBG Hamiltonian at K point in the vicinity of layer 1 momentum k = 0 (which is the K(cid:48) point of the Moir´e BZ) truncated at the nearest hoppings is [4] M H K,K(cid:48) M (k) = hK θ/2(k) † wT 1 † wT 2 † wT 3 wT1 hK−θ/2(k − q1) 0 0 wT2 wT3 0 hK−θ/2(k − q2) 0 0 0 hK−θ/2(k − q3) (14)   , √ 3 2 σy .  = 0,k, ψT 1,k, ψT 2,k, ψT where the basis is (ψT 3,k)T , and ψ0,k and ψj,k (j = 1, 2, 3) are the 2-component column spinors in the AB sublattice index basis at momentum k in layer 1 and momentum k − qj in layer 2, respectively. hK±θ/2(k) is hK(k) rotated by a ±θ/2 angle, and w = tK(1) /Ω is the interlayer hopping which can be chosen as real. The hopping ·τ (2) β −iG(1) matrix Tj are defined by (Tj)αβ = eiG(2) where (n1, m1) = (0, 0), (n2, m2) = (1, 0) and (n3, m3) = (0, 1), and G(i) satisfy K(i) Explicitly, Tj are given by (which is nothing but the phase factor part of Eq. (13)), n,m in layer i is defined below Eq. (13). They D = 4π/3a0, thus correspond to the nearest interlayer hoppings in the momentum space. nj ,mj = K(i) D + G(i) ·τ (1) nj ,mj nj ,mj D α T1 = 1 + σx , T2 = 1 − 1 2 σx − 3 2 σy , T3 = 1 − 1 2 σx + √ Since the twist angle θ is small, to the lowest order, we shall ignore the ±θ/2 rotation of hK±θ/2(k) in Eq. (14). Under this approximation, the system has a particle-hole symmetry, and the low energy eigenstates are given by † j ψ0,k (j = 1, 2, 3), where hj is short hand for hK(−qj). The low energy Hamiltonian is a 2 × 2 ψj,k ≈ −wh−1 j T Hamiltonian in the ψ0 space [4]: (cid:101)H K,K(cid:48) M (k) = (cid:104)ΨH K,K(cid:48) M (k)Ψ(cid:105) (cid:104)ΨΨ(cid:105) v 1 + 6α2 = σ∗ · k + w2 3(cid:88) j=1 Tjh−1 † j σ∗ · kh−1 j T j 1 − 3α2 1 + 6α2 vσ∗ · k , (15) effective Hamiltonian vanishes. Note that (cid:101)H K,K(cid:48) in hj), so we shall also denote it as (cid:101)H K,K(cid:48) We note that the Dirac Hamiltonian (cid:101)H K,K(cid:48) where α = w/vkθ. The first magic angle θ = 1.05◦ is given by α2 = 1/3, where the Dirac velocity of the above M (k) depends on the momenta q1, q2 and q3 (which implicitly appear M point of the Moir´e BZ (MBZ) (see Fig. 2 of the main text), and comes from the K point of both layer 1 and layer 2 of TBG. Similarly, there is also a Dirac Hamiltonian around layer 2 momentum p(cid:48) = 0 coming from the K point of M (k) in Eq. (15) around layer 1 momentum k = 0 is at the K(cid:48) M (k, q1, q2, q3) later to emphasize its dependence on qj. τ(1)τ(2)ABABlayer 1layer 2a1(1)a2(1)G2(1)G1(1)a1(2)a2(2)G2(2)G1(2)KD(2)KD(1)abcgraphene BZxyboth layer 1 and layer 2 of TBG, which is at the KM point of the MBZ. In total, there are eight Dirac fermions labeled by graphene BZ valley K, K(cid:48) (which is the same for both layers, no coupling between K and K(cid:48) exists in the M and spin ↑,↓ indices. The band structure obtained in the above Hamiltonian of Eq. (15)), Moir´e valley KM , K(cid:48) model is thus 4-fold degenerate everywhere in the MBZ with respect to graphene valley K, K(cid:48) and spin ↑,↓, and has 4 Dirac fermions at Moir´e valley KM and another 4 Dirac fermions at Moir´e valley K(cid:48) M . In particular, the helicity of the Dirac fermions only depends on whether they come from graphene valley K or K(cid:48). 9 When small distortions are added to the graphene lattices, the above effective Hamiltonian will change. Since distortions can be expressed using phonon fields, the change in the effective Hamiltonian simply gives the low energy electron-phonon coupling term. The leading contribution to the change of Hamiltonian is due to the distortion of momentum vectors qj. In the main text we have argued that interlayer phonon waves change the Moir´e pattern dramatically. To prove this from microscopics requires some more calculations. This can be shown explicitly as follows. Denote the in-plane displacement of atoms of layer j at r as u(j)(r) = . The displacement u(j) is nothing but the in-plane phonon field in layer j. In the continuum limit, the variation of the lattice vectors a(j) a(j) 2 under the displacement field u(j) are simply given by δa(j) Explicitly, one has · ∇)u(j)(r) and δa(j) 1 and · ∇)u(j)(r). 2 (r) = (a(j) 1 (r) = (a(j) u(j) x (r), u(j) (cid:17)T y (r) (cid:16) 2 1 (cid:16) · ∇(cid:17) (cid:20) δa(j) b = a(j) b u(j) = T · a(j) b (cid:19)(cid:21)T(cid:32) (cid:18) ∂x ∂y (cid:33) (cid:32) u(j) x u(j) y = a0 ∂xu(j) x ∂xu(j) y ∂yu(j) x ∂yu(j) y R(j) θ/2 (cid:19) (cid:18) − 1 2±√ 3 2 , (16) where the ± signs are for b = 1 (lattice vector a(j) order the distortion of reciprocal vectors of layer j satisfies δG(j) a solution 1 ) and b = 2 (lattice vector a(j) b + G(j) · a(j) 2 ), respectively. Accordingly, to linear b = 0 (a, b = 1, 2), which has a a · δa(j) (cid:33) (cid:33) a = −∇(cid:16) δG(j) (cid:17) (cid:19)(cid:16) (cid:18) ∂x ∂y u(j) · G(j) a = − (cid:17) (cid:32) u(j) · G(j) a = − 4π√ 3a0 ∂xu(j) x ∂yu(j) x ∂xu(j) y ∂yu(j) y R(j) θ/2 (cid:19) (cid:18) −√ 3 2± 1 2 , where the ± signs are for index a = 1 (reciprocal vector G(j) a ·τ (j) +G(j) particular, since τ (j) = (a(j) invariant, so the interlayer hopping matrix Tj = eiG(2) Only qj in the continuum model Hamiltonian change. 2 )/3, one has δG(j) 1 −a(j) nj ,mj 1 ) and a = 2 (reciprocal vector G(j) a ·δτ (j) = 0, which implies G(j) 2 ), respectively. In 1,2·τ (j) = ±2π/3 remains remains unchanged under the deformation. ·τ (1) nj ,mj α ·τ (2) β −iG(1) With the expressions for δG(j) a , it is straightforward to derive the variation of qj from their definitions: (cid:34)(cid:32) (cid:33)(cid:18) cos θ 2 sin θ 2 (cid:32) (cid:19) − ∂xu(2) x ∂yu(2) x ∂xu(2) y ∂yu(2) y (cid:19)(cid:35) (cid:33)(cid:18) cos θ 2− sin θ 2 ∂xu(1) x ∂yu(1) x ∂xu(1) y ∂yu(1) y 2 − δG(2) 3 δq1 = = kθ δq2 = δG(1) 1 + δG(1) (cid:0)γ∂xux + ∂xuc (cid:34)√ −2δG(1) 4π 3a0 2 = 1 − δG(2) (cid:1)T , 1 − δG(2) 2 (cid:0)γ∂xux + ∂xuc y y , γ∂yux + ∂yuc y 1 + δG(1) 2 + 2δG(2) = kθ δq3 = 3 2 3 (γ∂xuy − ∂xuc x) − 1 2 2 − δG(2) 1 − 2δG(1) 1 + 2δG(2) δG(1) √ 3 2 (cid:34) = kθ − 3 2 (γ∂xuy − ∂xuc x) − 1 2 (cid:1) , − 1 2 (cid:0)γ∂yux + ∂yuc y (cid:1) + √ 3 2 (cid:35)T (γ∂yuy − ∂yuc x) , (17) (cid:0)γ∂xux + ∂xuc y (cid:1) , − 1 2 (cid:0)γ∂yux + ∂yuc y (cid:1) − √ 3 2 (cid:35)T (γ∂yuy − ∂yuc x) , where we have defined γ = [2 tan(θ/2)]−1, the relative displacement u = u(1) − u(2), and the center of mass displace- ment uc = (u(1) + u(2))/2. For small angles θ, we have γ ≈ 1/θ (cid:29) 1, so δqj is dominated by the relative displacement phonon field u. We therefore will ignore the contribution of center of mass displacement uc hereafter. Fig. 6 shows the graphene lattices and Moir´e patterns for θ = 5◦ before and after a relative shear deformation l ∂lul)δab/2 (a, b, l = 1, 2) is the relative shear tensor. One Σxx = −Σyy = 0.02, where Σab = (∂aub + ∂bua)/2 − ((cid:80) can see that a small relative deformation greatly affects the Moir´e pattern. 10 FIG. 6. a. Graphene lattice plaquettes of two layers of TBG (at AA stacking center) before deformation (plotted for θ = 5◦). b. Graphene lattice plaquettes of two layers of TBG (at AA stacking center) with a relative shear deformation Σxx = −Σyy = ∂xux − ∂yuy = 0.02, while expansion and rotation deformations are zero. c. The Moir´e pattern superlattice of TBG before deformation. d. The Moir´e pattern superlattice of TBG after the shear deformation Σxx = −Σyy = 0.02. One can see the superlattice is greatly affected, although the graphene lattices only have a 2% shear deformation. With the deformed vectors qj, the electron-phonon coupling Hamiltonian is simply given by the variation of the effective 2 × 2 Hamiltonian (cid:101)H K,KM in Eq. (15), namely, H K,K(cid:48) ep (k) = (cid:101)H K,K(cid:48) M (k, q1 + δq1, q2 + δq2, q3 + δq3) − (cid:101)H K,K(cid:48) j σ∗ · kδh−1 j T (cid:17) − 1 − 3α2 † j + h.c. M (1 + 6α2)2 (k) = (cid:101)H K,K(cid:48) (cid:16) 3(cid:88) H K,K(cid:48) ep M vw2 1 + 6α2 = Tjh−1 j=1 M (k, q1 + δq1, ) M (k, q1, q2, q3) 3(cid:88) j=1 vw2σ∗ · k † 0Tjδ(h−2 j )T † j ψ0 , ψ (18) (cid:35) (cid:35) (cid:41) (cid:41) (cid:40)(cid:34)√ (cid:40)(cid:34)√ 3 4 3 4 where we have defined hj = hK(−qj), and δhj = hK(−qj − δqj) − hj. The first term and the second term in the MΨ(cid:105) and δ(cid:104)ΨΨ(cid:105) of Eq. (15), respectively. Explicitly, one can show above result come from the variations δ(cid:104)ΨH K,K(cid:48) that δh−1 δh−1 1 = γ(vkθ)−1(∂xuxσx + ∂yuxσy) , 2 = γ(vkθ)−1 (∂yux − ∂xuy) + (∂xux + ∂yuy) ∂xuy − ∂yux + σx + ∂yuy √ σy , 1 4 ∂xux − 1 4 (19) (cid:35) (cid:35) (cid:34) (cid:34) 1 4 δh−1 3 = γ(vkθ)−1 (∂xuy − ∂yux) + If we denote the above expression as δh−1 −2α2Tj(kxAj,y + kyAj,x). Besides, we note that h−2 so one has δ(h−2 (cid:34) j ) = (v)−2δqj−2I2, and one can show that δq1−2 = 2γk−2 θ ∂yux, δq2−2 = γk−2 θ 3 4 1 4 3 4 ∂yuy σx + ∂yux + ∂xuy + (∂xux + ∂yuy) ∂xux − 1 4 1 4 j σ∗ · kδh−1 j = (vkθ)−1(σxAj,x + σyAj,y), we find w2Tjh−1 j T (cid:35) † j = j = (v)−2qj−2I2 (where I2 is the 2 × 2 identity matrix), √ σy . 1 2 ∂yux − 3 2 (∂xux − ∂yuy) , 3 2 3 4 3 4 √ (cid:34) δq3−2 = γk−2 θ 1 2 ∂yux − 3 2 ∂xuy − (∂xux − ∂yuy) √ 3 2 (cid:35) ∂xuy + . D2D1D2+δD2D1+δD1undistorted graphene latticeundistorted TBG superlatticegraphene lattice with relative shearΣxx=-Σyy=0.02 (2%)TBG superlattice with relative shearΣxx=-Σyy=0.02 (2%)abcd11 Further, we set ψ0 in Eq. (18) to be an eigenstate of σ∗ · k (with eigenvalue either +k or −k). With these results, one then finds the electron-phonon coupling in Eq. (18) to be (cid:110) 9α2(1 + 3α2) H K,K(cid:48) ep M + (k) = −γv (1 + 6α2)2 σ∗ · k(∂yux − ∂xuy) + (1 + 6α2)2 9α2 [kx(∂yux + ∂xuy) + ky(∂xux − ∂yuy)] (cid:111) 1 + 6α2 z · (σ∗ × k)(∂xux + ∂yuy) 3α2 (20) , where the momentum k in the above electron-phonon coupling should be understood as the average momentum of the electron before and after phonon emission (absorption), and explicitly one has z · (σ∗ × k) = σxky + σykx. Note that the three terms kx(∂yux + ∂xuy) + ky(∂xux − ∂yuy) , σ∗ · k(∂yux − ∂xuy) , z · (σ∗ × k)(∂xux + ∂yuy) are contributed to by the relative shear tensor Σab, relative rotation R = ∂xuy − ∂yux, and relative expansion Θ = ∂xux + ∂yuy, respectively. Since the shear tensor is a rank 2 tensor in the 2D space, the first term contributed by shear (k times the shear tensor) is a rank 3 tensor. In particular, the first term can be rewritten as kx(∂yux + ∂xuy) + ky(∂xux − ∂yuy) = Im[(kx + iky)(∂x + i∂y)(ux + iuy)] . Under a rotation of angle φ about z axis, each of the three terms kx + iky, ∂x + i∂y and ux + iuy gains a phase factor eiφ. Therefore, this term is clearly only C3z rotationally invariant about z axis (i.e., φ = 2π/3). In contrast, both σ∗ · k and z · (σ∗ × k) are 2D scalars, and both the relative rotation R and the relative expansion Θ are also 2D scalars, so the second and the third terms contributed by relative rotation and expansion are SO(2) rotationally symmetric about z axis. One may ask why we only have term Im[(kx +iky)(∂x +i∂y)(ux +iuy)] but not Re[(kx +iky)(∂x +i∂y)(ux +iuy)], both of which respect C3z symmetry. This is because the TBG also has a 2-fold rotation symmetry C2x about the x axis. Under C2x, we have (∂x, ∂y) → (∂x,−∂y) and (kx, ky) → (kx,−ky), while the relative displacement field transforms differently as (ux, uy) → (−ux, uy) due to the exchange of two layers. Since Re[(kx + iky)(∂x + i∂y)(ux + iuy)] flips sign under C2x, it is forbidden in the TBG electron phonon coupling. Similarly, one can check that terms like σ∗ · k(∂xux + ∂yuy) and z · (σ∗ × k)(∂yux − ∂xuy) are also forbidden by C2x. When the TBG is at the magic angle, namely α2 = 1/3, the second term in Eq. (18) vanishes, and Eq. (20) becomes H K,K(cid:48) ep M (k) = − γv 3 [2kx(∂yux + ∂xuy) + 2ky(∂xux − ∂yuy) + σ∗ · k(∂yux − ∂xuy) + (σxky + σykx)(∂xux + ∂yuy)] . (21) In the same way one could obtain the electron-phonon couplings at other valley/Moir´e valley/spin indices, which we shall not repeat here. Instead, we give a symmetry analysis derivation of electron-phonon coupling at other valley/Moir´e valley/spin indices in the next section above Eq. (31). The results are as shown in Eq. (31) as well as in the main text Eq. (3). II. THE PHONON HAMILTONIAN AND PHONON MEDIATED ELECTRON-ELECTRON INTERACTION In this section, we first describe the Hamiltonian of the relative in-plane displacement phonon mode, and then consider the phonon mediated electron-electron interaction. A. Phonon Hamiltonian In principle, the relative in-plane displacement phonons are optical phonons. However, if the two graphene lattices twisted by angle θ are not commensurate, a uniform relative in-plane displacement u does not cost energy, so these phonons are still acoustic. In fact, since the interlayer Van der Waals interaction between atoms in two graphene layers is much weaker than the intralayer atomic interaction, the in-plane polarized phonons of the two layers are nearly decoupled, and thus the relative in-plane displacement phonons are almost acoustic [41]. As a good approximation, we shall ignore the coupling between phonons in different layers. From elastic dynamics [53], we know the in-plane of the expansion scalar Θ(j) =(cid:80) l l ∂lu(j) and the traceless shear tensor Σ(j) deformation (i.e., in-plane phonon) energy of the TBG in the continuum limit can be expressed as a quadratic function )δab/2 of layer j = 1, 2, where a, b, l = x, y. In particular, Θ(j) and Σ(j) ab occupy the spin 0 and spin 2 representations of the SO(2) rotation group about z axis, respectively (one can easily verify Σxx ± iΣxy have spin ±2 under SO(2), respectively, where Σxx and Σxy are the two independent components of Σab). Due to the C6z rotation symmetry of monolayer graphene, the symmetry allowed terms (scalars) in the deformation energy of layer j must occupy the spin 0 (mod 6) representations of SO(2). Therefore, the only allowed quadratic terms (scalars) are Θ(j)2 ab Σ(j) ab which have spin 0 (there are no interlayer terms since we have approximated the two layers as decoupled). Therefore, the Hamiltonian (to quadratic order) for in-plane displacement u(j) (with two layers decoupled) can be written as ab = (∂au(j) b + ∂bu(j) and Σ(j) l ∂lu(j) l a )/2 − ((cid:80) 12 Hph = (∂tu(j))2 + KelΘ(j)2 1 2 + µelΣ(j) ab Σ(j) ab , (22) (cid:21) (cid:20) M Ω (cid:90) 2(cid:88) j=1 d2r √ 3a2 where M is the Carbon atom mass, Ω = 0/2 is the graphene unit cell area, while Kel and µel are the bulk modulus and shear modulus of monolayer graphene, respectively. The first term is the kinetic energy of the Carbon atoms (note that there are 2 atoms in each graphene unit cell), while the second and the third terms are the elastic potential energy. To separate the relative deformation u = u(1) − u(2) and the center-of-mass deformation uc = (u(1) + u(2))/2, l ∂lul)δab/2, b + ∂buc l )δab/2, respectively. ph, where the relative phonon wave part is we define the relative expansion scalar and shear tensor as Θ =(cid:80) and the center-of-mass expansion and shear Θc =(cid:80) (cid:21) l ∂lul and Σab = (∂aub + ∂bua)/2 − ((cid:80) The phonon Hamiltonian can then be written as two parts Hph = H r a)/2 − ((cid:80) ab = (∂auc ph + H c l and Σc l ∂luc l ∂luc = d2r (∂tu)2 + [(∂xux)2 + (∂yuy)2] + [(∂xuy)2 + (∂yux)2] + and the center-of-mass phonon wave part is (∂tu)2 + 1 4 Kel + µel KelΘ2 + ΣabΣab µel 2 µel 4 4 (cid:90) (cid:20) 2M Ω H c ph = d2r (∂tuc)2 + Kel(Θc)2 + 2µelΣc abΣc ab (cid:27) ∂xux∂yuy , Kel 2 (23) (24) (cid:21) . (cid:20) M 2Ω (cid:90) (cid:26) M 2Ω H r ph = d2r (cid:90) (cid:19)(cid:18) ux (cid:19) (cid:18) ux (cid:19) (cid:18) Kel∂2 1 2 Since the relative phonon wave dominates the electron-phonon coupling, we shall consider only the relative phonon part H r ph in Eq. (23). Explicitly, the equation of motion of H r ph is M Ω ∂2 t ∇(∇ · u) + Kel 2 ∇2u . µel 2 x + µel(∂2 Kel∂x∂y x + ∂2 y) Kel∂x∂y y + µel(∂2 = = uy uy Kel∂2 x + ∂2 y) real space, the canonical momentum π(r) = M The Hamiltonian of relative phonon waves can then be quantized following the standard canonical method. (25) Given the momentum −i∇ = p = (px, py), there are two eigenvectors: up,L ∝ p and up,T ∝ z × p, which satisfies ∇(∇ · up,L) = ∇2up,L and ∇(∇ · up,T ) = 0, respectively. Accordingly, their eigenfrequencies are ωp,L = cLp and and transverse sound speed of monolayer graphene, and p = p. ωp,T = cT p, respectively, where cL =(cid:112)(Kel + µel)Ω/2M and cT =(cid:112)µelΩ/2M being the longitudinal sound speed When transformed into the momentum space, the canonical momentum πp = (cid:82) d2r√ up =(cid:82) d2r√ (cid:88) Ω ∂tu(r) satisfies [ua(r), πb(r(cid:48))] = iδabδ(r − r(cid:48)) (where a, b = x, y). e−ip·rπ(r) and displacement e−ip·ru(r) satisfies [up,a, πp(cid:48),b] = iδabδp,−p(cid:48), where As is the total area of the sample. We can decompose up = up,L + up,T into a longitudinal part up,L and a transverse part up,T (which are perpendicular to each other), and similarly we can also do so for the canonical momentum πp = πp,L + πp,T (which are also perpendicular to each other, up,T · up,L = 0). One can then show the phonon Hamiltonian in the momentum space becomes † † p,Lap,L + ωp,T a p,T ap,T (cid:16)ωp,La (p · up)(p · u−p) + πp · π−p + p2up · u−p (cid:20) Ω (cid:88) , (26) ph = (cid:17) (cid:21) H r In = As As Kel 2 p p,χ are the phonon annihilation and creation operators with momentum p and polarization χ satisfying p,χ] = 1. The relative displacement phonon field u in the Shrodinger picture is given by u(r) = eip·r√ NsΩs (ipup,L+iz× pup,T ) , up,L = † (ap,L+a −p,L) , up,T = † (ap,T +a −p,T ) , (27) (cid:115) Ω 2M ωp,T 2M p where ap,χ, a† [ap,χ, a† (cid:88) p µel 2 (cid:115) Ω 2M ωp,L 13 where p = p/p is the unit vector along momentum p, while z is the unit vector along the out of plane direction (z× p is the transverse direction), and for later convenience we have rewritten the total area of the sample As as NsΩs, with Ns being the number of superlattice unit cells, and Ωs = Ω/[4 sin2(θ/2)] being the superlattice unit cell area. In the Heisenberg picture, the phonon field u is time t dependent, namely, u(r, t) = eiH r pht, with u(r) defined in Eq. (27) above. Accordingly, one can obtain the canonical momentum as π(r, t) = M phtu(r)e−iH r Ω ∂tu(r, t). Since the interlayer phonon coupling is ignored, the phonon bands in the MBZ is simply obtained by folding p into the superlattice MBZ. Among these phonon bands, the lowest two bands (one longitudinal and one transverse) are acoustic (under the approximation that the two layers are decoupled) and cause the long wavelength deformation of the Moir´e pattern superlattice, while the higher bands are optical and mainly lead to short wavelength deformations within each unit cell of the superlattice. Since our electron phonon coupling is derived in the long wavelength deformation limit, for now we shall restrict ourselves to the lowest acoustic phonon band in the MBZ. We will briefly discuss the contribution of optical phonon bands at the end of this supplementary section. B. Electron-Phonon Coupling for other valley/Moir´e valley/spin For convenience, hereafter we define the graphene BZ valley K, K(cid:48) as index η = ±1, the Moir´e valley KM , K(cid:48) M as index ζ = ±1, and the spin up and down as index s = ±1. Since there is no spin-orbit coupling, the electron-phonon coupling does not flip spin and is independent of spin index s. In last section we derived the electron-phonon coupling for (η, ζ) = (+1,−1). Now we use symmetry arguments to obtain the expression of electron-phonon coupling at other indices (η, ζ). Identical results can be obtained by performing brute force calculations. First, TBG has the time-reversal symmetry. A time-reversal transformation T changes (η, ζ, s) → (−η,−ζ,−s) and k → −k. Since for spinless (no spin-orbit coupling) fermions the time reversal T is simply complex conjugation, one finds the electron-phonon coupling at (−η,−ζ) is given by (k) =(cid:2)H η,ζ ep (−k)(cid:3)∗ H−η,−ζ ep , (28) where we have found H K,K(cid:48) ep M (k) in Eq. (21), and z∗ stands for the complex conjugate of z. Secondly, TBG also has a C2x symmetry, which is the 2-fold rotation about x axis (see Fig. 5a for definitions of x and y axes). From the main text Fig. 2a and supplementary Fig. 5, one can see the C2x transformation interchanges M , changes momentum (kx, ky) (measured from the KM point) to (kx,−ky) (measured from Moir´e valley KM and K(cid:48) the K(cid:48) M point) in momentum space, interchanges layer 1 with layer 2, and interchanges sublattice indices A and B. Meanwhile, the valley K or K(cid:48) remains invariant. Since the Pauli matrices in Eq. (14) (and afterwards) are in the sublattice basis, the interchange of sublattice indices A and B leads to a transformation of Pauli matrices x , namely, (σx, σy) → (σx,−σy). Therefore, the band Hamiltonian of the continuum model at Moir´e σ → σxσσ−1 valley KM and valley K is given by (cid:104) (cid:105) H K,KM (k) = U H K,K(cid:48) M (kx,−ky) U−1 =  hK−θ/2(k) † wT 1 † wT 3 † wT 2 wT1 hK θ/2(k + q1) wT3 0 0 0 hK θ/2(k + q3) 0 hK θ/2(k + q2) (29)  , wT2 0 0 M (k) is given by Eq. (14), and U = σx ⊗ I4 is the transformation matrix for the interchange of sublattices where H K,K(cid:48) A and B under C2x (i.e., ψj,k → σxψj,k for each ψj,k (1 ≤ j ≤ 4) in the basis of Hamiltonian (ψT 3,k)T , recall that ψj,k are defined in the 2D Hilbert space of sublattice A and B). If one approximate θ/2 = 0 in hK±θ/2 (as we have assumed in the first supplementary section), and interchanges electron basis ψ2,k ↔ ψ3,k, one finds H K,KM (k) and M (k) only differ by a sign flip qj → −qj (j = 1, 2, 3). From Eq. (18) one sees the expression of electron-phonon H K,K(cid:48) coupling H K,K(cid:48) (k) ep (where k of H K,K(cid:48) M and KM , respectively), and H η,ζ ep (k) is invariant under ζ → −ζ. Alternatively, one can achieve this conclusion by directly applying C2x on H K,K(cid:48) (k) we derived in Eq. ep x ,−u(j) (20). In particular, C2x changes the displacement field in layer j as (u(j) y ), and interchanges layer 1 with layer 2, so the relative displacement field u = u(1) − u(2) changes as (ux, uy) → (−ux, uy). Besides, it changes (∂x, ∂y) → (∂x,−∂y). Therefore, one finds (k) is quadratic in qj (thus invariant under qj → −qj), so we conclude H K,K(cid:48) (k) are measured from K(cid:48) y ) → (u(j) (k) and H K,KM (k) = H K,KM x , u(j) 0,k, ψT 1,k, ψT 2,k, ψT ep ep ep ep M M M M H K,KM ep (k) = σx H K,K(cid:48) ep M (kx,−ky) x = H K,K(cid:48) σ−1 ep M (k) , (30) (cid:20) (cid:12)(cid:12)(cid:12)∂y→−∂y,ux→−ux (cid:21) 14 M ep where H K,K(cid:48) (k) is given in Eq. (20) (obtained under the approximation θ/2 = 0 in hK±θ/2 in the first supplementary section). This indicates the electron-phonon coupling H η,ζ ep (k) is also independent of the Moir´e valley ζ. We can then obtain the electron-phonon coupling Hamiltonian for all indices, which in the second quantized language takes the form SO(2) , H η,ζ,s H η,ζ,s C3 = g1α ep = H η,ζ,s C3 + H η,ζ,s γηv√ NsΩs SO(2) = − γv√ (cid:88) (cid:88) k,k(cid:48) NsΩs k,k(cid:48) H η,ζ,s p ψη,ζ,s† k p ψη,ζ,s† k (cid:8)(cid:2)2¯kx px py + ¯ky(p2 (cid:2)g2α (cid:0)ησx ¯kx − σy x − p2 y)(cid:3) up,L +(cid:2)2¯ky px py − ¯kx(p2 (cid:1) up,T + g3α (cid:1) up,L (cid:0)σy ¯kx + ησx ¯ky ¯ky x − p2 y)(cid:3) up,T (cid:3) ψη,ζ,s , k(cid:48) (cid:9) ψη,ζ,s k(cid:48) , (31) where the numerical factors are from Eq. (20): g1α = 9α2(1 + 3α2) (1 + 6α2)2 , g2α = 9α2 (1 + 6α2)2 , g3α = 3α2 1 + 6α2 , (32) k(cid:48) is the electron annihilation operator at momentum k with η, ζ, s indices, k − k(cid:48) = p (the momentum while ψη,ζ,s of the ∂aub terms), and we have defined ¯k = (k + k(cid:48))/2 is the average electron momentum before and after phonon emission (absorption), which comes from Eq. (20). Physically, this comes from the Fourier transformation of the real space hopping amplitude t(r, r(cid:48))c†(r)c(r(cid:48)), where t(r, r(cid:48)) is induced by the displacement field u(r). In particular, for the magic angle α2 = 1/3, we have g1α = 2/3 and g2α = g3α = 1/3. Besides, as we have shown below Eq. (20), is only C3z rotationally invariant, while the term H η,ζ,s the term H η,ζ,s SO(2) is SO(2) rotationally invariant. Finally, we is odd in K, K(cid:48) valley index η, since the electron band Hamiltonian and thus the electron-phonon note that H η,ζ,s coupling undergoes a sign flip and complex conjugate under η → −η. C3 C3 The electron-phonon coupling H η,ζ,s ep we obtained is independent of the Moir´e valley ζ and the spin s. However, we note that the independence of Moir´e valley holds only under the approximation θ/2 = 0 in hK±θ/2(k). If we do ep will be weakly ζ dependent, where the ζ dependent part will be of a factor O(θ) not make this approximation, H η,ζ,s smaller than the ζ independent part given in Eq. (20). Since we are interested in small angles θ ∼ 1◦ ∼ 0.02 rad, we shall ignore the weak ζ dependence of H η,ζ,s . Besides, since the continuum model of TBG has no coupling between valleys K and K(cid:48), the electron-phonon coupling we obtained has a definite K, K(cid:48) valley index η (i.e., initial and final states of the electron are in the same valley η). ep C. Phonon-Mediated Electron-Electron Interaction eigenstate of (ησxkx − σyky) (which is proportional to the Dirac Hamiltonian (cid:101)H η,ζ(k) = 1−3α2 We now calculate the phonon mediated electron-electron interaction near the Fermi surface. To be concrete, we shall assume the Fermi surface is at k = kF in the hole Dirac bands (where superconductivity is observed), and project the electron operators to the vicinity of the Fermi surface. Such a Fermi surface might not be very accurate since the band is quite flat, but we shall simply model the Fermi surface by a circle with a large kF of a low velocity k of the Dirac hole band state kη,ζ,s(cid:105) is assumed to be the negative eigenvalue 2D Dirac fermion. The wave function φη 1+6α2 v(ησxkx − σyky)), namely, φη 2 at valley η, where ϕk = arg(kx + iky) is the polar angle of momentum k (we note that for α2 < 1/3 this is the valence band of the Dirac Hamiltonian, while for α2 > 1/3 this in fact becomes the conduction band of the Dirac Hamiltonian, and for α2 = 1/3 the conduction or valence band becomes ill-defined unless higher order terms in k are included. Here we shall ignore these complications and take the wave function φη k). One can then rewrite the Dirac annihilation and creation operators ψη,ζ,s kck,η,ζ,s † and ψη,ζ,s† k,η,ζ,s are the electron annihilation and creation operators in the hole Dirac band with indices η, ζ, s. The Dirac hole band Hamiltonian can then be written as † = φη† k c k,η,ζ,s, where ck,η,ζ,s and c √ k = (1,−ηe−iηϕk)T / in Eq. (31) as ψη,ζ,s , ψη,ζ,s† = φη k k k k where ξk = − 1−3α2 We then make the approximation k ≈ k(cid:48) ≈ kF , based on which we find the two terms in H η,ζ,s 1+6α2 v(k − kF ) is the the band energy relative to the Fermi level. (cid:18) (cid:19) (cid:34) (cid:35) (cid:1) (ησxkx − σyky) φη† k ¯kx − σy ¯ky k(cid:48) = φη† k x − σyk(cid:48) 2 y k(cid:48) = φη† φη k −k 2 − k(cid:48) 2 (cid:0)ησx (cid:1) φη SO(2) are approximately k(cid:48) ≈ −kF φη† φη k φη k(cid:48) , † k,η,ζ,sck,η,ζ,s , (cid:101)H η,ζ,s(k) = ξkc (cid:0)ησxk(cid:48) + 2 (cid:0)ησx φη† k ¯ky + σy ¯kx (cid:1) φη k(cid:48) = φη† k (cid:34) (ησxky + σykx) 2 + (cid:0)ησxk(cid:48) y + σyk(cid:48) 2 x (cid:35) (cid:1) k(cid:48) = i(k(cid:48) − k)φη† φη k σzφη k(cid:48) ≈ 0 , 15 k = −kφη k, and (ησxky + σykx) φη k = −iσz (ησxkx − σyky) φη k = where we have used the definition (ησxkx − σyky) φη ikσzφη −(cid:2)2¯ky px py − ¯kx(p2 (cid:18) keiϕk − k(cid:48)eiϕk(cid:48) k. Besides, we note that the two terms in H η,ζ,s (cid:19) x − p2 y)(cid:3) + i(cid:2)2¯kx px py + ¯ky(p2 (cid:19) (cid:18)keiϕk + k(cid:48)eiϕk(cid:48) x − p2 C3 = ke−iϕk − k(cid:48)e−iϕk(cid:48) × 2 of Eq. (31) satisfy y)(cid:3) = (px + ipy)2(¯kx + i¯ky) = ≈ kF 2 e2iϕk − e2iϕk(cid:48) e−iϕk − e−iϕk(cid:48) , (cid:19) (cid:18) px + ipy px − ipy (¯kx + i¯ky) (33) where we have used p = k− k(cid:48) and ¯k = (k + k(cid:48))/2. Therefore, they can be written as the real part and imaginary part e2iϕk−e2iϕk(cid:48) of the complex quantity kF −iϕk(cid:48) , respectively. This enables us to approximate the electron-phonon interaction e−iϕk−e 2 projected in the Dirac hole band as (cid:18) e2iϕk − e2iϕk(cid:48) e−iϕk − e−iϕk(cid:48) (cid:19)(cid:21) up,T + η g1α 2 Im (cid:18) e2iϕk − e2iϕk(cid:48) e−iϕk − e−iϕk(cid:48) (cid:19) (cid:27) up,L (34) (cid:26)(cid:20) (cid:88) p H η,ζ,s ep ≈ γvkF√ NsΩs × φη† k φη g2α − η g1α 2 Re k,k(cid:48) † k,η,ζ,sck(cid:48),η,ζ,s . k(cid:48)c k C3 C3 e−iϕk−e e−iϕk−iπ−e and H η,ζ,s (proportional to g1α) and by H η,ζ,s −iϕk(cid:48) −iπ = − e2iϕk−e2iϕk(cid:48) −iϕk(cid:48) → e2iϕk+2iπ−e2iϕk(cid:48) +2iπ interaction. We treat the electron energy (cid:101)H η,ζ,s(k) and the phonon energy H r In particular, under momentum reversal k, k(cid:48) → −k,−k(cid:48), the polar angles ϕk → ϕk +π and ϕk(cid:48) → ϕk(cid:48) +π), so we have e2iϕk−e2iϕk(cid:48) −iϕk(cid:48) . Therefore, the terms in the projected electron-phonon coupling e−iϕk−e contributed by H η,ζ,s SO(2) (proportional to g2α) are odd and even under momentum reversal k, k(cid:48) → −k,−k(cid:48), respectively, which will be useful in later calculations. We note that this odd/evenness results from the projection of electron-phonon coupling onto a single band. In the original expression of Eq. (31) before projection, one may thought both H η,ζ,s SO(2) are odd in k since they are both linear in k. However, this naive expectation ignores the effect of the Dirac wave functions ψη,ζ,s in the expression of Eq. (31), thus is incorrect. One can then use the standard second order perturbation theory to calculate the phonon-mediated electron electron ph as the unperturbed Hamiltonian, and the electron-phonon interaction H η,ζ,s as the perturbation. The electron electron interaction is then induced by the emission and absorption of a phonon between two electrons. Consider the initial state Ψ0(cid:105) = k1,I , k2,I(cid:48)(cid:105) = † † k2,I(cid:48)0(cid:105) of two electrons which have momenta k1 and k2, respectively, where I = (η, ζ, s) and I(cid:48) = (η(cid:48), ζ(cid:48), s(cid:48)) are c k1,I c short hand for their Dirac cone indices, and 0(cid:105) is the particle vacuum. The initial state energy is E0 = ξk1 + ξk2. Assume the final state is Ψf(cid:105) = k3,I , k4,I(cid:48)(cid:105), where the total momentum k3 + k4 = k1 + k2 is conserved, and the final state energy Ef = ξk3 + ξk4 = E0 remains unchanged. There are two intermediate states with an emitted phonon † † p,χ0(cid:105) where a phonon with momentum p and with polarization χ: one is Ψ1,χ(cid:105) = k3,I , k2,I(cid:48), p,χ(cid:105) = c k2,I(cid:48)a† k3,I c polarization χ is emitted from the first electron, while the other is Ψ2,χ(cid:105) = k1,I , k4,I(cid:48),−p,χ(cid:105) where a phonon with momentum −p and polarization χ is emitted from the second electron, with the momentum p = k1 − k3 = k4 − k2. The energy of the two intermediate states are E1,χ = ξk3 + ξk2 + ωp,χ and E2,χ = ξk1 + ξk4 + ω−p,χ, respectively, where ωp,χ is the phonon energy. The phonon of the intermediate state is then absorbed by the other electron, resulting in the final state Ψf(cid:105). According to the second order perturbation theory, the interaction between the two electrons is given by V II(cid:48) (cid:33) epΨ1,χ(cid:105)(cid:104)Ψ1,χH I epΨ0(cid:105) ep (35) k3k4,k1k2 NsΩs = (cid:32)(cid:104)ΨfH I(cid:48) (cid:32)(cid:104)ΨfH I(cid:48) (cid:88) (cid:88) χ χ = E0 − E1,χ epΨ1(cid:105)(cid:104)Ψ1H I ξk3 − ξk1 − ωp,χ + (cid:104)ΨfH I epΨ2,χ(cid:105)(cid:104)Ψ2,χH I(cid:48) epΨ0(cid:105) (cid:33) E0 − E2,χ epΨ2(cid:105)(cid:104)Ψ2H I(cid:48) ξk4 − ξk2 − ω−p,χ epΨ0(cid:105) , epΨ0(cid:105) (cid:104)ΨfH I + where I = (η, ζ, s) and I(cid:48) = (η(cid:48), ζ(cid:48), s(cid:48)) are notations for the Dirac cone indices of the two electrons. Since the electron ep is independent of spin s, the above electron-electron interaction is independent of s and s(cid:48). phonon interaction H I Besides, in the continuum model H I (31)). Since our goal is to study the BCS superconductivity, we shall focus on the Cooper channel of the interaction, namely, we shall set the momenta of the two electrons as opposite to each other, k1 = −k2 = k and k3 = −k4 = k(cid:48). The Cooper channel ep is also independent of Moir´e valley ζ (see Eq. electron-electron interaction is then (cid:104)(cid:104)k(cid:48) (cid:88) η,ζ,s,−k(cid:48) η(cid:48),ζ(cid:48),s(cid:48)H η,ζ,s η,ζ,s,−k(cid:48) ep χ V ηη(cid:48) kk(cid:48) (ω) NsΩs = (cid:104)k(cid:48) + η(cid:48),ζ(cid:48),s(cid:48)H η(cid:48),ζ(cid:48),s(cid:48) ep kη,ζ,s,−k(cid:48) k(cid:48) η,ζ,s,−kη(cid:48),ζ(cid:48),s(cid:48), p,χ(cid:105)(cid:104)k(cid:48) ω − ωp,χ η(cid:48),ζ(cid:48),s(cid:48),−p,χ(cid:105)(cid:104)kη,ζ,s,−k(cid:48) −ω − ωp,χ η,ζ,s,−kη(cid:48),ζ(cid:48),s(cid:48), p,χH η,ζ,s ep kη,ζ,s,−kη(cid:48),ζ(cid:48),s(cid:48)(cid:105) η(cid:48),ζ(cid:48),s(cid:48),−p,χH η(cid:48),ζ(cid:48),s(cid:48) ep kη,ζ,s,−kη(cid:48),ζ(cid:48),s(cid:48)(cid:105) where ω = (ξk(cid:48) − ξk)/, and we have replaced ω−p,χ by ωp,χ since they equal to each other. Accordingly, in the second quantized language, the phonon-mediated electron-electron interaction in the Cooper channel is 16 (cid:105) (36) , V ηη(cid:48) kk(cid:48) (ω)c † k(cid:48),η,ζ,sc † −k(cid:48),η(cid:48),ζ(cid:48),s(cid:48)c−k,η(cid:48),ζ(cid:48),s(cid:48)ck,η,ζ,s . (37) (cid:88) (cid:88) (cid:88) (cid:88) k,k(cid:48) η,η(cid:48) ζ,ζ(cid:48) s,s(cid:48) H (ph) int = 1 NsΩs We now calculate V ηη(cid:48) kk(cid:48) using the projected electron-phonon Hamiltonian H η,ζ,s Eqs. (27) and (34), we find the matrix elements for intermediate states (defined as kI , k(cid:48) with phonon polarizations χ = L and χ = T are: ep which we obtained in Eq. (34). By p,χ0(cid:105)) † † I(cid:48), p,χ(cid:105) = c k(cid:48),I(cid:48)a† k,I c (cid:114) pΩ (cid:114) pΩ (cid:20) η,ζ,s,−kη(cid:48),ζ(cid:48),s(cid:48), p,TH η,ζ,s (cid:104)k(cid:48) γvkF√ NsΩs φη† k(cid:48) φη 2M cT ep k kη,ζ,s,−kη(cid:48),ζ(cid:48),s(cid:48)(cid:105) = (cid:104)k(cid:48) (cid:18) e2iϕk − e2iϕk(cid:48) (cid:19)(cid:21) η,ζ,s,−k(cid:48) η(cid:48),ζ(cid:48),s(cid:48)H η,ζ,s ep g2α − η g1α 2 Re e−iϕk − e−iϕk(cid:48) , η,ζ,s,−kη(cid:48),ζ(cid:48),s(cid:48), p,LH η,ζ,s (cid:104)k(cid:48) γvkF√ k × η NsΩs φη† k(cid:48) φη 2M cL ep (cid:18) e2iϕk − e2iϕk(cid:48) kη,ζ,s,−kη(cid:48),ζ(cid:48),s(cid:48)(cid:105) = (cid:104)k(cid:48) g1α 2 e−iϕk − e−iϕk(cid:48) Im (cid:19) η,ζ,s,−k(cid:48) , η(cid:48),ζ(cid:48),s(cid:48)H η,ζ,s ep = = kη,ζ,s,−k(cid:48) η(cid:48),ζ(cid:48),s(cid:48),−p,T(cid:105) kη,ζ,s,−k(cid:48) η(cid:48),ζ(cid:48),s(cid:48),−p,L(cid:105) (cid:104)kη,ζ,s,−k(cid:48) γvkF√ NsΩs = (cid:20) η(cid:48),ζ(cid:48),s(cid:48),−p,TH η(cid:48),ζ(cid:48),s(cid:48) (cid:114) pΩ ep φη(cid:48)† −k(cid:48)φη(cid:48) −k 2M cT kη,ζ,s,−kη(cid:48),ζ(cid:48),s(cid:48)(cid:105) = (cid:104)k(cid:48) g2α + η(cid:48) g1α 2 Re (cid:18) e2iϕk − e2iϕk(cid:48) (cid:19)(cid:21) η,ζ,s,−k(cid:48) η(cid:48),ζ(cid:48),s(cid:48)H η(cid:48),ζ(cid:48),s(cid:48) ep e−iϕk − e−iϕk(cid:48) , (cid:104)kη,ζ,s,−k(cid:48) = − γvkF√ (cid:114) pΩ NsΩs 2M cL η(cid:48),ζ(cid:48),s(cid:48),−p,LH η(cid:48),ζ(cid:48),s(cid:48) ep kη,ζ,s,−kη(cid:48),ζ(cid:48),s(cid:48)(cid:105) = (cid:104)k(cid:48) (cid:18) e2iϕk − e2iϕk(cid:48) (cid:19) η,ζ,s,−k(cid:48) e−iϕk − e−iϕk(cid:48) , φη(cid:48)† −k(cid:48)φη(cid:48) −k × η(cid:48) g1α 2 Im η(cid:48),ζ(cid:48),s(cid:48)H η(cid:48),ζ(cid:48),s(cid:48) ep k(cid:48) η,ζ,s,−kη(cid:48),ζ(cid:48),s(cid:48), p,T(cid:105) (38) k(cid:48) η,ζ,s,−kη(cid:48),ζ(cid:48),s(cid:48), p,L(cid:105) where we have used the fact that ϕ−k = ϕk + π, and recall that Ω and Ωs are the graphene unit cell area and † superlattice unit cell area, respectively. Since H η,ζ,s −p,χ) (see Eq. (27)), the amplitude of creating a phonon state p,χ(cid:105) is always the same as that of annihilating a phonon state − p,χ(cid:105) (when the other quantum numbers are the same), so we have the above equal relations between every two matrix amplitudes. To further simplify the result, we approximate cL ≈ cT , both of which are of the order of magnitude 104m/s, so that ωp,L ≈ ωp,T = cT p. Under this approximation, the prefactors of all the matrix elements in Eq. (38) become identical, i.e., is proportional to up,χ ∝ (ap,χ + a (cid:113) pΩ (cid:113) pΩ . By defining = ep 2M cT 2M cL (cid:20) 1 + cos(ϕk − ϕk(cid:48)) (cid:21) 2 , (39) one can then show the interaction in Eq. (36) is kφη(cid:48)† −k(cid:48)φη(cid:48) k(cid:48) φη F pηη(cid:48) kk(cid:48) kk(cid:48) = φη† ηη(cid:48) (cid:18) (cid:18) e2iϕk − e2iϕk(cid:48) (cid:40) e−iϕk − e−iϕk(cid:48) 2M cT Ωs γ22v2Ωk2 = (cid:40)(cid:20) V ηη(cid:48) kk(cid:48) (ω) Ωs g2α − η g1α Re 2 F ηη(cid:48) γ22v2Ωk2 kk(cid:48) 2M c2 T Ωs = ω2 p,T ω2 − ω2 p,T g2 2α + g1αg2α 2 + ω − ωp,T 1 (cid:19)(cid:21)(cid:20) g2α + η(cid:48) g1α 2 2 1 )(ϕk−ϕk(cid:48) ) −k = ei( η+η(cid:48) (cid:19) (cid:19)(cid:21) (cid:18) e2iϕk − e2iϕk(cid:48) (cid:19) (cid:18) e2iϕk − e2iϕk(cid:48) e−iϕk − e−iϕk(cid:48) −ω − ωp,T × Re (η(cid:48) − η)Re e−iϕk − e−iϕk(cid:48) (cid:20) Im (cid:18) e2iϕk − e2iϕk(cid:48) (cid:12)(cid:12)(cid:12)(cid:12)2(cid:41) (cid:12)(cid:12)(cid:12)(cid:12) e2iϕk − e2iϕk(cid:48) e−iϕk − e−iϕk(cid:48) e−iϕk − e−iϕk(cid:48) (cid:19)(cid:21)2(cid:41) . (40) − ηη(cid:48) g2 1α 4 − ηη(cid:48) g2 1α 4 Therefore, the intra valley interaction with η = η(cid:48) is V ηη kk(cid:48) Ωs = γ22v2Ωk2 T Ωs M c2 F ηη kk(cid:48) ω2 p,T ω2 − ω2 p,T (cid:26) = − N0(ω)eiη(ϕk−ϕk(cid:48) ) [1 + cos(ϕk − ϕk(cid:48))] [1 + cos(ϕk − ϕk(cid:48))] [1 + cos(ϕk − ϕk(cid:48))] 2α − g2 (cid:26) 1α g2 2 2α − g2 1α g2 2 (cid:27) 17 (41) , (cid:27) where we have used the fact that N0(ω) = γ22v2Ωk2 with ω < ωp,T , we have the coefficient N0(ω) > 0. η = −η(cid:48) = 1 is p,T −ω2 , and used Eq. (39) for the expression of ηη(cid:48) = 2−2 cos 2(ϕk−ϕk(cid:48) ) 2−2 cos(ϕk−ϕk(cid:48) ) = 2[1 + cos(ϕk − ϕk(cid:48))], and we have defined kk(cid:48). We note that for low frequency processes In a similar way, we find the inter valley interaction with −iϕk(cid:48) T Ωs M c2 ω2 ω2 p,T F (cid:12)(cid:12)(cid:12) e2iϕk−e2iϕk(cid:48) e−iϕk−e (cid:12)(cid:12)(cid:12)2 V KK(cid:48) kk(cid:48) Ωs = F KK(cid:48) γ22v2Ωk2 kk(cid:48) M c2 T Ωs = − N0(ω) [1 + cos(ϕk − ϕk(cid:48))] (cid:12)(cid:12)(cid:12)(cid:12)g2α − g1α 2 ω2 p,T p,T ω2 − ω2 (cid:12)(cid:12)(cid:12)(cid:12)g2α − g1α 2 e2iϕk − e2iϕk(cid:48) e−iϕk − e−iϕk(cid:48) (cid:12)(cid:12)(cid:12)(cid:12)2 . e2iϕk − e2iϕk(cid:48) e−iϕk − e−iϕk(cid:48) (cid:12)(cid:12)(cid:12)(cid:12)2 (42) For θ near magic angle α2 ≈ 1/3, one has g1α ≈ 2/3 and g2α ≈ 1/3, and the interaction takes the form shown in main text Eq. (8). In particular, for low frequencies ω < ωp,T , one finds that the intervalley interaction is attractive, while the intravalley interaction is repulsive. To see this, consider two electrons (near the Fermi surface) at valley η and η(cid:48), whose 2-body wave function (with total momentum zero assumed) can generically be written as Ψηη(cid:48)(cid:105) = † † −k,η(cid:48),ζ(cid:48),s(cid:48)0(cid:105) , βleilϕkc k,η,ζ,sc (cid:88) (cid:88) l∈Z k=kF (cid:90) 2π (cid:88) (cid:88) (cid:90) 2π (cid:90) 2π 0 where the physical meaning of l ∈ Z is the relative angular momentum between the two electrons. For two electrons in the same valley η = η(cid:48) = K (with wave function ΨKK(cid:105)), the intravalley interaction energy is given by β∗ l βleilϕk−il(cid:48)ϕk(cid:48) EKK = (cid:104)ΨKKH (ph) int = −N0(ω)ei(ϕk−ϕk(cid:48) ) 0 ΨKK(cid:105) = dϕk 2π (cid:90) 2π (cid:19) (cid:0)g2 (cid:0)β−22 + β02(cid:1) + dϕk 2π β−12 + 1 2 2α l∈Z l(cid:48)∈Z dϕk(cid:48) 2π V KK kk(cid:48) Ωs [1 + cos(ϕk − ϕk(cid:48))] (cid:26) (cid:1)(cid:0)β−22 + β02(cid:1) + (cid:0)β−32 + β12(cid:1)(cid:21) 2α dϕk(cid:48) 2π 1α − g2 0 0 , 2α − g2 1α g2 2 g2 1α 4 1 9 (cid:20)(cid:18) 3 (cid:20) 2 9 1α − g2 g2 4 β−12 + 1 6 = N0(ω) ≈ N0(ω) (cid:27) [1 + cos(ϕk − ϕk(cid:48))] (cid:0)β−32 + β12(cid:1)(cid:21) β∗ l βleilϕk−il(cid:48)ϕk(cid:48) (43) where in the last line we have assumed θ is near the first magic angle so that g1α ≈ 2/3 and g2α ≈ 1/3. For ω < ωp,T , one has N0(ω) > 0, and one finds the intravalley interaction energy is in fact always positive, i.e., EKK > 0. Therefore, two electrons in the same valley always repulse each other (similarly one can show this for η = η(cid:48) = K(cid:48)). in Eq. (42) is always real and negative for ω < ωp,T , i.e., N0(ω) > 0. Therefore, if one considers two electrons in valley η = K and η(cid:48) = K(cid:48), respectively, one has the intervalley interaction energy In contrast, one notes that the intervalley interaction V KK kk(cid:48) which is attractive (for ω < ωp,T ). Therefore, we conclude the intervalley Cooper pairing will be favored. EKK(cid:48) = (cid:104)ΨKK(cid:48)H (ph) int ΨKK(cid:48)(cid:105) < 0 , (44) Physically, the fact that the intervalley interaction is more attractive than intravalley interaction can be under- (8)). Recall that the electron-phonon interaction stood as follows (see also the paragraph below main text Eq. SO(2)(k), where H η,ζ,s projected to the vicinity of the Fermi surface contains two parts, H η,ζ,s and H η,ζ,s SO(2) are odd and even with respect to the electron momentum k (average momentum before and after phonon emission/absorption), respectively. Now we consider an electron wave packet state with average momentum k as (k) = H η,ζ,s (k) + H η,ζ,s C3 C3 ep kη,ζ,s(cid:105) = (cid:80) 18 † k β(k − k)c k,η,ζ,s0(cid:105), where β(k − k) is a wave packet function peaked at k. Under a certain lattice deformation (i.e., given a nonzero configuration of the displacement field u(r)), the electron-phonon interaction will generate a background lattice potential, which we assume is Uep = UC3 + USO(2) for an electron kK,ζ,s(cid:105) at valley K, where UC3 = (cid:104)kK,ζ,sH K,ζ,s SO(2) are odd and even with respect to k, an electron state − kK,ζ,s(cid:105) in the same valley K with opposite momentum would feel a background ep = −UC3 + USO(2). On the other hand, an electron state − kK(cid:48),ζ,s(cid:105) in the other valley K(cid:48) would lattice potential U(cid:48) feel the same potential Uep = UC3 + USO(2) as that of the electron kK,ζ,s(cid:105). kK,ζ,s(cid:105), and USO(2) = (cid:104)kK,ζ,sH K,ζ,s SO(2)kK,ζ,s(cid:105). Since H η,ζ,s and H η,ζ,s C3 C3 In general, if two electrons feel the same background potential due to deformed lattice, they will tend to get closer in the space, thus leading to a phonon-mediated electron-electron attraction. In the above, we have shown that two electrons kK,ζ,s(cid:105) and − kK(cid:48),ζ,s(cid:105) in opposite valleys K and K(cid:48) feel the same background potential UC3 + USO(2), thus an effective attraction will be produced between them. Instead, two electrons kK,ζ,s(cid:105) and − kK,ζ,s(cid:105) in the same valley feel different background potentials UC3 + USO(2) and −UC3 + USO(2), thus the effective attraction between them will be weaker or even absent. Therefore, the intervalley interaction is more attractive than the intravalley interaction. By substituting the realistic system parameters v ≈ 106 m/s, cT ≈ 104 m/s, w = 110 meV, a0 = 0.246 nm, and θ ≈ 1.05◦ into the phonon-induced inter valley interaction (Eq. (42)), and take kF ≈ kθ as an order of magnitude estimation, we find the interaction contributed by the acoustic MBZ phonon bands near magic angle is V KK(cid:48) kk(cid:48) Ωs (0) ∼ − 4γ22v2Ωk2 T Ωs 9M c2 θ ≈ − 42v2k2 θ 9M c2 T ∼ −1meV , (45) which is comparable to both the electron band width around the first magic angle (of order 1 ∼ 10meV) and the band width of the lowest acoustic MBZ phonon bands ωD ∼ cT kθ ≈ 2meV. Therefore, the electron phonon coupling is relatively strong. Furthermore, when the optical phonon bands are taken into account, the total phonon induced attractive interaction should be further enhanced. Since our electron phonon coupling Hep is derived in the long wavelength phonon limit, it does not apply for optical phonon bands. Nevertheless, here we give a very rough discussion of the optical phonon band contributions to the electron-electron interaction. A very crude approximation is to assume the electron phonon coupling matrix elements Mep,χ of all optical phonon bands χ are roughly the same as that of the lowest acoustic bands. As we have discussed, the optical phonon bands are obtained by folding the phonon bands in the graphene BZ of the two layers into the MBZ. Equivalently, before folding, we can say that the lowest acoustic phonon bands are in the first MBZ, while the post-folding optical phonon bands are in the second, third and higher MBZ. Since the n-th MBZ is roughly nkθ away from Γ point of the first MBZ, the energy of the optical phonon band in the n-th MBZ is roughly ncT kθ. Besides, the number of n-th MBZs is roughly 2πn (roughly speaking, all the n-th MBZs together form a circle at radius nkθ), so there are roughly 2πn transverse and longitudinal polarized optical phonon bands with energy ncT kθ. The upper limit of n is around 1/θ, where nkθ reaches the boundary of the original graphene BZ. The modified total electron-electron interaction will then be enhanced to Mep,T2 + Mep,L2 1/θ(cid:88) 2πn −cT kθ ≈ 2π θ V ac kk(cid:48) , kk(cid:48) ∼(cid:88) V tot χ Mep,χ2 −ωp,χ ∼ n n=1 where χ here denotes all the MBZ phonon bands (acoustic and optical), ωp,χ is the eigenfrequency of phonon band χ, and V ac kk(cid:48) is the interaction contributed solely by the acoustic phonon bands we derived earlier in this section. Namely, the optical phonon bands contribute an additional factor 2π/θ to the electron-electron interaction if all the optical phonon bands contribute, which is a large number for small angles θ. This is clearly an overestimation, since high energy optical phonon bands are expected not to participate in the low energy superlattice physics. In practice, we expect this factor of interaction enhancement by optical phonons to be much smaller than 2π/θ (but obviously greater than 1). III. SCREENED COULOMB POTENTIAL AND BCS SUPERCONDUCTIVITY AT MAGIC ANGLE The phonon-mediated electron-electron interaction can induce conventional BCS superconductivity. The density of for a 1 meV band width, which gives a BCS states of the lowest bands at magic angle is around ND (cid:38) 1meV−1 · Ω−1 coupling strength λ ≈ NDVkk(cid:48)(0) (cid:38) 1, which is relatively strong. s To determine the superconductor critical temperature, we also need to estimate the screened Coulomb potential between electrons. Here we shall simply adopt the Thomas-Fermi approximation in two dimensions (2D) for an order of magnitude estimation, and do not discuss the accuracy of the approximation. The Coulomb potential without screening is given by Ve(r) = e2/I r, whose Fourier transform is Ve(q) = 2πe2/I q, where q = q is the Fourier wave vector, and I ≈ 2 ∼ 10 is the dielectric constant of a charge neutral graphene. The 2D Poisson's equation for the bare Coulomb potential can then be written as qIVe(q) = 2πρ(q), (46) where ρ(q) is the bare charge density. In the presence of free electrons, a Coulomb potential Ve will induce a local charge density ρf (r) ≈ −e2(∂ne/∂µ)Ve(r), or in Fourier space ρf (q) ≈ −e2(∂ne/∂µ)Ve(q), where ne is the electron number density, µ is the chemical potential, and thus ∂ne/∂µ is the density of states ND. One can then write the charge density as ρ = ρs + ρf , where ρs is the screened charge. From the Poisson's equation we have 19 2πρs(q) = qIVe(q) − 2πρf (q) = [qI + 2πe2(∂ne/∂µ)]Ve(q) = q(q)Ve(q) , so we find the screened dielectric function of the form (q) ≈ I (cid:18) 1 + qTF q (cid:19) , independent. This is because the 2D plasma frequency ωpe =(cid:112)(4πne)1/2e2F /I (cid:38) 20meV at the magic angle is where qTF = 2πe2(∂ne/∂µ)/I = 2πe2ND/I is the Thomas-Fermi screening momentum. Near the magic angle, ND (cid:38) 1 meV−1 · Ω−1 s , which yields a qTF (cid:38) 50kθ (cid:29) q, so the screened Coulomb potential is approximately Ve(q) ≈ 2πe2/I qT F ∼ N−1 D , and the Coulomb coupling strength µc ≈ NDVe(q) ∼ 1. We note that due to large density of states ND, the screened Coulomb potential is much smaller than the bare Coulomb potential. In particular, the screened Coulomb potential is comparable to Vkk(cid:48). However, the phonon induced attraction Vkk(cid:48)(ω) is frequency ω dependent and large as ω approaches ωT,p, while the screened Coulomb potential can be approximated as frequency much larger than the Debye frequency ωD ≈ cT kθ ≈ 2meV, where F is the Fermi energy (of order 1 ∼ 10meV around magic angle) [47]. We then adopt the McMillan formula [44, 45] to give a proper estimation of the BCS superconductivity critical temperature taking into account both the phonon induced attraction and the Coulomb repulsion: (49) c = µc/[1 + µc ln(ωpe/ωD)] is the reduced Coulomb coupling strength, and ωpe/ωD (cid:38) 10 around the magic where µ∗ angle. If we take the BCS coupling strength λ ≈ NDVkk(cid:48) ≈ 1.5, and µc ≈ 1, we find the superconductivity critical temperature from the McMillan formula is Tc ≈ 0.9K, close to the experimentally measured value. c (1 + 0.62λ) kBTc = λ − µ∗ exp , Next, we discuss the pairing amplitude of the inter valley pairing. The pairing amplitude is defined as ωD 1.45 − 1.04(1 + λ) (cid:20) (cid:21) where V ηη(cid:48) kk(cid:48) is as given in Eq. (40). At zero temperature, the BCS self-consistency gap equation is given by (47) (48) (50) (51) ∆ηη(cid:48),ζζ(cid:48),ss(cid:48) k = (cid:88) k(cid:48) ∆ηη(cid:48),ζζ(cid:48),ss(cid:48) k = − 1 2 V ηη(cid:48) kk(cid:48) (cid:104)c−k(cid:48),η(cid:48),ζ(cid:48),s(cid:48)ck(cid:48),η,ζ,s(cid:105) , (cid:88) V ηη(cid:48) kk(cid:48) ∆ηη(cid:48),ζζ(cid:48),ss(cid:48) , k(cid:48) k(cid:48) EBdG ξk(cid:48)<ωD k(cid:48) is the Bogoliubov-de Gennes (BdG) band energy, while ξk = vF (k − kF ) is the band energy with the where EBdG Fermi velocity vF = − 1−3α2 1+6α2 v. Since we have shown the intervalley pairing is favored, we shall ignore the intravalley interaction (which is in fact repulsive as we have shown in Eq. (43)), and only keep the intervalley interaction V KK(cid:48) . kk(cid:48) Furthermore, for simplicity we shall adopt the BCS approximation which assumes the electron-electron interaction is a constant function of ω for frequency ω < ωD and zero otherwise (ωD is the Debye frequency), namely, we approximate in Eq. (42) to −1 for ω < ωD, and 0 for ω ≥ ωD. Qualitatively, this does the frequency dependence factor not affect the shape of the pairing function. We shall also assume θ is near the magic angle, so that α2 ≈ 1/3, and accordingly g1α ≈ 2/3 and g2α ≈ 1/3. The intervalley interaction in Eq. (42) is then approximated as ω2−ω2 ω2 p,T p,T V KK(cid:48),ζζ(cid:48),ss(cid:48) kk(cid:48) ≈ −V0KK(cid:48) kk(cid:48) (cid:20) 1 + cos(ϕk − ϕk(cid:48)) = − V0 2 (cid:12)(cid:12)(cid:12)(cid:12)1 − e2iϕk − e2iϕk(cid:48) (cid:21)(cid:12)(cid:12)(cid:12)(cid:12)1 − e2iϕk − e2iϕk(cid:48) e−iϕk − e−iϕk(cid:48) e−iϕk − e−iϕk(cid:48) (cid:12)(cid:12)(cid:12)(cid:12)2 (cid:12)(cid:12)(cid:12)(cid:12)2 , (52) where V0 ≈ γ22v2Ωk2 9M c2 F T Ωs , and we have used Eq. (39). 20 With the 4-fold degeneracy from Moir´e valley ζ and spin s, the pairing can be either Moir´e valley triplet spin singlet, or Moir´e valley singlet spin triplet. Here we shall simply assume the pairing is time-reversal invariant (which is more robust than time-reversal violating pairings in the presence of disorder [42]). Since the time-reversal symmetry T brings valley K → K(cid:48), Moir´e valley KM → K(cid:48) M , and spin s → −s (see Eq. (28)), a time reversal invariant pairing will be between opposite spins and opposite Moir´e valleys. Such a time reversal invariant intervalley pairing then takes the form where (cid:101)∆(k) satisfies ∆ηη(cid:48),ζζ(cid:48),ss(cid:48) k (cid:101)∆(k) = NDV0 (cid:90) 2π =NDV0 dϕk(cid:48) 2π 0 (cid:90) 2π (cid:90) ωD (cid:20) 1 + cos(ϕk − ϕk(cid:48)) dϕk(cid:48) 2π −ωD dξ 0 2 ∼ sδs,−s(cid:48)δζ,−ζ(cid:48)δη,−η(cid:48)(cid:101)∆(ηk) , (cid:20) 1 + cos(ϕk − ϕk(cid:48)) (cid:21)(cid:12)(cid:12)(cid:12)(cid:12)1 − e2iϕk − e2iϕk(cid:48) (cid:113)(cid:101)∆(k(cid:48))2 + ξ2 (cid:21)(cid:12)(cid:12)(cid:12)(cid:12)1 − e2iϕk − e2iϕk(cid:48) (cid:12)(cid:12)(cid:12)(cid:12)2(cid:101)∆(k(cid:48)) sinh e−iϕk − e−iϕk(cid:48) e−iϕk − e−iϕk(cid:48) −1 2 (cid:101)∆(k(cid:48)) (cid:113)(cid:101)∆(k(cid:48))2 + ξ2 (cid:12)(cid:12)(cid:12)(cid:12)2 (cid:101)∆(k(cid:48)) , 2 ωD (53) (54) where in this case the BdG band energy is EBdG (cid:101)∆(k) is real and has the shape as shown in the main text Fig. [3c]. For generic values of NDV0 > 0, we find the pairing Eq. (54) can then be numerically solved by iteration. As an example, for NDV0 = 0.5, we find the pairing amplitude k(cid:48) = k(cid:48). amplitude is always real and nodeless, thus is dominated by s-wave pairing and is topologically trivial. Finally, we note that in the absence of disorders, the pairing is not necessarily time-reversal invariant, and the pairing amplitude could be either Moir´e valley singlet spin triplet or Moir´e valley singlet spin triplet, which are degenerate. IV. NUMERICAL CALCULATION FOR OTHER ANGLES AND ELECTRON DENSITIES: PREDICTION OF OTHER SUPERCONDUCTING ANGLES The above calculations of electron-phonon coupling can be generically applied to any twist angle θ and electron density. We still adopt the continuum model with nearest hoppings [4], namely, an electron with momentum k in layer 1 can hop with an electron with momentum p(cid:48) in layer 2 iff k− p(cid:48) = qj (j = 1, 2, 3). However, instead of truncating at the smallest four momenta (k and k − qj) in Eq. (14), we truncate the momentum of the Hamiltonian to sufficiently high momenta (momentum shells, i.e., second, third and higher MBZs), so that the band structure (which can be solved numerically) is more accurate. Besides, in the calculation we do not approximate θ/2 in hK±θ/2(k) to zero, and this leads to a small particle-hole asymmetry to the band structure [54]. Fig. 8a and 8b show two examples of Moir´e BZ band structure calculated from the continuum model for θ = 1.05◦ and θ = 1.20◦, respectively. We can then calculate the density of states ND from the band structure. In the main text Fig. [4a] we have shown ND at θ = 1.05◦ as a function of number of electrons per superlattice unit cell n = neΩs. Here in the upper panel of Fig. 8c, we show ND at θ = 1.20◦ as another example. Fig. 8d shows the Log10(ND·meV·Ωs) in a wide range of θ and n. We then add small deformations to qj induced by ∂aub (a, b = x, y) (according to supplemental Eq. (17)), and numerically calculate the band structure under different deformations. To first order, the change of band energy of a band Em generically takes the form (cid:18) ∂xux − ∂yuy (cid:19) (cid:18) ∂xuy − ∂yux (cid:19) (cid:18) ∂xux + ∂yuy (cid:19) (cid:18) ∂xuy + ∂yux (cid:19) δEm(k) = b1(k) 2 + b2(k) 2 + b3(k) 2 + b4(k) 2 , (55) which is contributed by the band energy response to relative shear (first two terms), rotation and expansion between the two layers, respectively. This is nothing but the electron-phonon coupling of band Em in the long wavelength limit. The four coefficients bj (1 ≤ j ≤ 4) are real and can be numerically extracted out from band structure variation under small deformations. One effect we need to exclude in calculating δEm(k) is the following: when one makes a change qj → qj + δqj, the Moir´e BZ is deformed, so the folding of the original graphene BZ momentum into the Moir´e BZ is also changed (Fig. 7). Therefore, for higher Moir´e bands (which originates from folding of the graphene BZ), the definition of momentum k in the Moir´e BZ is changed under deformations. This additional change of electron momentum is simply a redefinition and is unphysical, thus should be eliminated. To be precise, consider the electron state in the m-th 21 FIG. 7. A higher Moir´e band state with Moir´e momentum k in the 1st Moir´e BZ (MBZ, black solid hexagon) originates from the folding of the state with graphene momentum pG in the graphene BZ. Under the deformation qj → qj + δqj, the MBZ is deformed (into the red hexagon); accordingly, the graphene momentum pG will be folded to a different Moir´e momentum k + δk. This implies a redefinition of k in higher Moir´e bands. Moir´e band (at graphene valley K) which takes a generic form k, m(cid:105)M = k,l1,l2,jk + l1g1 + l2g2 − (j − 1)q1(cid:105)j , W m (56) (cid:88) (cid:88) l1,l2∈Z j=1,2 where k is the momentum in the Moir´e BZ, j is the layer index, g1 = q2−q3 and g2 = q3−q1 are the reciprocal vectors of the Moir´e superlattice, and p(cid:105)j is the basis of Dirac electron in layer j with an original graphene BZ momentum k,l1,l2,j denotes the coefficient of basis k + l1g1 + l2g2 − (j − 1)q1(cid:105)j. p measured from graphene valley K. Besides, W m (cid:88) (cid:88) Therefore, the state k, m(cid:105)M carries an average graphene momentum (cid:17) k + l1g1 + l2g2 − (j − 1)q1 . (cid:104)pG(cid:105) = (cid:104)k, mpGk, m(cid:105)M = k,l1,l2,j2(cid:16) W m (57) l1,l2∈Z j=1,2 Upon the uniform deformation qj → qj + δqj, one should keep the electron's graphene momentum (cid:104)pG(cid:105) invariant and then examine the energy variation, because the absorption or emission of a long wave length phonon should not change the electron momentum. Namely, one need to shift the Moir´e BZ momentum k to k + δk so that 0 = δ(cid:104)pG(cid:105) ≈ δk + l1δg1 + l2δg2 − (j − 1)δq1 , (58) (cid:17) (cid:88) (cid:88) k,l1,l2,j2(cid:16) W m l1,l2∈Z j=1,2 where δg1 = δq2 − δq3 and δg2 = δq3 − δq1 (see Fig. 7). Since the wave function W m k,l1,l2,j can be calculated numerically, we are able to compute δk. Accordingly, one should evaluate the energy variation in Eq. (55) as (Fig. 7) δEm(k) = E(cid:48) m(k + δk) − Em(k) , (59) where Em(k) and E(cid:48) m(k) are the energy dispersion of the m-th band before and after deformation, respectively. We note that in our analytical calculations of electron-phonon coupling near the Dirac points of the lowest two Moir´e flat bands (Eq. (31)), one has δk ≈ − 2α2 1+6α2 (δq1 + δq2 + δq3) = 0, so we need not consider this momentum shift δk problem there. For numerical calculations of higher bands, however, δk is in general nonzero. From Eq. (27) we have the displacement field after quantization u(r) =(cid:80) , where p is the phonon eip·rχup,χ √ momentum, up,χ = † −p,χ) for polarization χ, and χ is the polarization 2M cχp (ap,χ + a vector. The electron-phonon coupling in band Em is then simply the expression of δEm(k) in Eq. (55) with the displacement field u(r) expressed in terms of phonon operators (as above), which in the momentum space takes the form † −p,χ) = (ap,χ + a 2M ωp,χ (cid:113) Ω (cid:113) Ω p,χ NsΩs Hep(k) ≈ 4(cid:88) (cid:88) j=1 χ=L,T (cid:115) Ωp 2NsΩsM cχ Nj,χ(p)bj(k) † −p,χ) , (ap,χ + a (60) where k should be understood as the average of initial momentum k and final momentum k(cid:48) of the electron, Ω and Ωs are the graphene unit cell area and the superlattice unit cell area, and Nj,χ(p) are dimensionless coefficients depending on the unit vector p = p/p (direction of phonon momentum) and are of order 1. These coefficients Nj,χ(p) can be obtained by substituting the expression of u(r) in Eq. (27) into Eq. (55), and then rewrite the results into the form of Eq. (60). 1st MBZ2nd MBZ3rd MBZpGkk+δk22 FIG. 8. a. TBG band structure at θ = 1.05◦ from the continuum model. b. TBG band structure at θ = 1.20◦ from the continuum model. c. TBG band structure at θ = 1.80◦ from the continuum model. d. Density of states ND and BCS coupling strength λ with respect to n calculated for θ = 1.20◦. In particular, the peaks in ND around n = 8 indicates a Van Hove singularity in the second or third conduction (valence) bands (the second band and third band have a significant overlap in energy, as shown in panels a and b here, so n = 8 is not an in-gap filling factor). e. Density of states ND and BCS coupling strength λ with respect to n calculated for θ = 1.80◦. The shape of the ND function agrees well with the STM measurement of large angle TBG [52]. f. Log10(ND·meV·Ωs) plotted as a function of θ and n, where one can clearly see ND is large at the magic angle, and is generically larger for smaller angles. The electron-electron interaction Vkk(cid:48)(ω) can then be estimated from Eq. (35). For zero frequency ω = 0, the interaction Vkk(cid:48) can be crudely estimated as (cid:88) χ=L,T Vkk(cid:48) Ωs = 2Ns−ωp,χ (cid:88) Ωp 2NsΩsM cχ j,j(cid:48) Nj,χ(p)Nj(cid:48),χ(p)bj(k)bj(cid:48)(k) ≈ − Ω M c2 T Ωs bj(¯k)2 (61)  4(cid:88) j=1  , where for order of magnitude estimation we have simply approximated Nj(cid:48),χ ∼ 1 and ignored the cross terms between bj(k) and bj(cid:48)(k). We also approximated cL = cT . Given a Fermi energy F , the BCS coupling strength λ ≈ NDVkk(cid:48) can be numerically estimated as (cid:90) λ ∼ 1 2ωD ξk−F ≤ωD d2k (2π)2 [−V (k)] , (62) where ωD is the Debye frequency, and we have written Vkk(cid:48) = V (k) for short. Note that λ is well-defined in the limit ωD → 0. The main text Fig. 4b shows the BCS coupling strength λ we estimated for θ = 1.05◦, while the lower panels of Fig. 8d and Fig. 8e here shows λ for θ = 1.20◦ and θ = 1.80◦, respectively. We find the BCS coupling strength λ in higher bands n > 4 can also be generically as large as order 1, although the density of states therein is much lower (∼ 0.05meV−1 · Ω−1 s ). Here we give a heuristic understanding and estimation. First, for angles θ near 1◦, numerical calculation shows the second and higher MBZ bands have band widths ∼ 100 meV, and this is determined by the two characteristic TBG energy scales which are of the same order: the interlayer hopping w = 110 meV and the graphene kinetic energy vkθ ∼ 200 meV. Roughly speaking, the Moir´e bands are obtained by folding the original graphene band structure into the MBZ, while the interlayer hopping w couples different Moir´e bands. If we treat w as a perturbation, the energy of the m-th band Em(k) in the 2nd perturbation theory is roughly Em(k) ∼ vk(m) G + z w2 EW , (63) MMK-100-50050100energy (meV)ΓΓθ=1.05MMMMMMMK-100-50050100energy (meV)ΓΓθ=1.20MMMMMadfebND (meV )-1s-100.20.4θ=1.200123-20-30-100102030nλθ=1.20n08-816-1624-2410.80.60.421.81.61.41.2θ0.2210-1-2Log (N )10DMKMMMMMM-200-1000100200Energy (meV)Γθ=1.8000.511.5-20-30-100102030nλθ=1.80ND (meV )-1s-100.020.04θ=1.80c23 G is the original graphene momentum, z is a numerical factor, and EW ∼ 100 meV is the energy scale of where k(m) the band width as well as the energy separation between neighbouring bands. Since each Moir´e band is coupled to three other bands via qj (j = 1, 2, 3), we can estimate the numerical factor z as z ∼ 3. Under a lattice deformation, the three vectors qj changes to qj +δqj (j = 1, 2, 3), where δqj ∼ γkθ∂aub (a, b = x, y) as given in Eq. (17). This changes the energy separation between neighbouring bands EW by an amount ∼ vδqj. Keeping the graphene momentum k(m) G invariant in Eq. (63) (which is the requirement of Eq. (58)), we have the energy variation of band Em(k) to be roughly δEm(k) ∼ z w2 E2 W vδqj ∼ z w2 E2 W vγkθ∂aub . (64) This gives an estimation of the coefficients in Eq. (55) as bj(k) ∼ z w2 mediated interaction is of order E2 W vγkθ. By Eq. (61), we then find the phonon Vkk(cid:48) Ωs ∼ − Ωz2γ2(vkθ)2 M c2 T Ωs w4 E4 W ∼ − z2 M c2 T w4 E2 W , (65) where we have used vkθ ∼ EW , and γ2 ≈ Ωs/Ω (recall that γ ∼ 1/θ, while Ω and Ωs are the graphene unit cell area and the Moir´e supercell area, respectively). On the other hand, the density of states ND also has a generical grow trend as the Fermi energy increases (for electron densities n > 4), as can be seen in the upper panels of Fig. 8d and 8e. This is because as the energy increases, there are more and more Moir´e bands falling into the same energy interval. This can also be understood in the w → 0 limit, in which case the TBG becomes two decoupled monolayer graphene, and the density of state ND will be that of the graphene Dirac electrons, which grows as the energy increases. From s ∼ 0.05 meV−1 · Ω−1 the numerical results of the upper panel of Fig. 8d, we simply estimate ND as ND ∼ 5E−1 for second or higher bands near the magic angle. Therefore, if we take z ≈ 3 (for the three nearest neighbour couplings of momenta qj), we find the BCS coupling strength in higher bands to be around W · Ω−1 s λ ∼ −NDVkk(cid:48) ∼ 5z2 M c2 T w4 E3 W ∼ 0.5 . (66) The density of states ND could be higher near von Hove singularities, where the BCS coupling strength could be higher. This estimation indicates that the electron-phonon coupling is generically enhanced by the Moir´e pattern (by the amplification factor γ), and does not necessarily require the presence of flat bands. Based on λ estimated in the above, we use the McMillan formula to estimate Tc in a wide range of θ and n, as is shown in the main text Fig. 4c. Despite the fact that Tc is estimated in a very rough way, the parameter space where we predict superconductivity may occur at large n is stable. V. AB INITIO CALCULATIONS To verify the continuum model gives the correct order of magnitude of electron-phonon coupling, we also run ab initio calculations of the variation of band structure of TBG under deformations where the lattices of the two layers remain commensurate (which we shall explain below), and compare it with the spectrum calculated from the continuum model. We performed ab initio calculations based on the density functional theory with the projector augmented wave (PAW) method [55, 56] as implemented in VASP package [57, 58]. The local density approximation (LDA) was adopted for the exchange-correlation functional [59]. The kinetic energy cutoff of the plane wave basis was set to 300 eV. Only the convergence of band energy at Γ point in the MBZ is used as the criteria for the convergence of self-consistent calculations. Fig. 9 illustrates a graphene lattice, where each site here denotes the center of a hexagon plaquette of graphene. The primitive lattice vectors are denoted as a and b, and the lattice constant is set to be a0 = 2.456 A. We define the lattice vector Rm,n = ma + nb. Now assume this graphene lattice is the lattice of layer 1 of TBG. By stacking the graphene lattice of layer 2 on top of it (which is not shown in Fig. 9), we arrive at a TBG. To construct a commensurate TBG (undistorted), we first assume the lattices of layer 1 and layer 2 differ by a rotation of angle θ about point o in Fig. 9 (which we define as the origin), namely, the hexagon plaquette centers of layer 1 and layer 2 coincide at point o. Now consider two lattice vectors Ai = R2i+1,i+1 and Bi = R2i+1,i away from the origin o with i ≥ 0 being an integer, which correspond to points 1 and 2 as illustrated in Fig. 9. The two vectors have equal lengths Ai = Bi = LA i = (cid:112)3(2i + 1)2 + 1 2 a0 . 24 FIG. 9. Illustration of the commensurate Moir´e pattern constructed from supercells, and the construction of slightly deformed Moir´e pattern. The numbers 1, 2, 3, 4 and 1(cid:48), 2(cid:48), 3(cid:48), 4(cid:48) label several particular sites (red solid points) useful in defining the commensurate configuration. i = R−i−1,i and B(cid:48) Similarly, the vectors A(cid:48) i = R−i,i+1 from the origin o correspond to points 1(cid:48) and 2(cid:48) in Fig. 9, which also have equal lengths and are related to Ai and Bi by a 2π/3 rotation. We then assume the layer 2 lattice is given by rotating the layer 1 lattice so that Bi (B(cid:48) i). Such a TBG then forms a commensurate Moir´e pattern superlattice which is periodic in real space. As shown in Ref. [60], the twist angle θ = θi of such a commensurate configuration satisfies cos θi = 3i2+3i+0.5 i) is rotated to Ai (A(cid:48) 3i2+3i+1 , or equivalently, θi = 2 arctan 1√ 3(2i + 1) . The spacial periods, namely, the superlattice vectors, are then given by Ai and A(cid:48) i, and the superlattice unit cell i. We call such a Moir´e pattern the (11(cid:48)22(cid:48)) configuration, of which the is the parallelogram with edges Ai and A(cid:48) definition involves the four points 1, 1(cid:48) and 2, 2(cid:48). As an example, we consider the commensurate undistorted TBG configuration (11(cid:48)22(cid:48)) with i = 10, and calculate the band structure with ab initio. The distance between the two layers is d0 (∼ 3.35 A). The electronic band structure from ab initio for i = 10 is shown as red solid lines in both Fig. 10c and 10d (identical between these two figures), where the twist angle θ = θ10 = 3.15◦, and the high symmetry points in the figure should be understood as those of the MBZ. To generate an example of deformed Moir´e pattern, we consider two different lattice vectors Ci = R2i+1,i−2 and Di = R2i+1,i−3 away from the origin o, which correspond to points 3 and 4 in Fig. 9, respectively (The reason we choose points 3 and 4 will be explained later). We also define another two vectors C(cid:48) i = R−i+3,i+4, which are Ci and Di rotated by 2π/3, and correspond to points 3(cid:48) and 4(cid:48), respectively. We shall still assume the rotated and deformed lattice plotted in Fig. 9 is the lattice of layer 1. Next, we assume the lattice of layer 2 is relative to the lattice of layer 1, so that vector Di and D(cid:48) i of layer 2 coincide with Ci and C(cid:48) i of layer 1, respectively. This is again a commensurate configuration, and the superlattice unit cell are the parallelogram with edges Ci and C(cid:48) i in layer 1. (which coincide with the parallelogram with edges Di and D(cid:48) i in layer 2, after layer 2 is rotated and deformed). We call this configuration (33(cid:48)44(cid:48)). In particular, the lengths of vectors are i = R−i+2,i+3 and D(cid:48) Ci = C(cid:48) i = LC i = a0 , Di = D(cid:48) i = LD i = (cid:112)3(2i + 1)2 + 52 2 (cid:112)3(2i + 1)2 + 72 2 a0 , which are not equal, and the twist angle is now the angle between Ci and Di, which is 5√ 3(2i + 1) Therefore, this configuration (33(cid:48)44(cid:48)) involves both a relative rotation δθ = θ(cid:48) 2(LD i − θi and a relative expansion Θ ≈ i compared to the undistorted configuration (11(cid:48)22(cid:48)). The relative shear Σab is, however, zero, since 7√ 3(2i + 1) θ(cid:48) i = arctan − arctan . i − LC i )/LC 𝑨𝑖 𝑩𝑖 𝑪𝑖 𝑫𝑖 𝑨′𝑖 𝑩′𝑖 𝑪′𝑖 𝑫′𝑖 25 FIG. 10. Variation of band energies under two deformation configurations "3344" and "3142" at θ = 3.15◦ from the continuum model (a and b), and from commensurate TBG ab initio calculations (c and d). The energy variations are comparable in order of magnitude in the continuum model and in ab initio (slightly larger in ab initio). the deformation in this case is isotropic. We can then write down the relative deformation field u for (33(cid:48)44(cid:48)) (compared to the undistorted TBG configuration) as(cid:18) ∂xux ∂xuy (cid:19) ∂yux ∂yuy  LD i i −LC LC i i − θi θ(cid:48) ≈  . θi − θ(cid:48) i i −LC LD LC i i (67) (68) We can also construct a configuration with nonzero relative shear. To do this, we can assume the lattice of layer 2 is rotated and deformed relative to the lattice of layer 1 so that vector Di and B(cid:48) i of layer 2 coincide with Ci and A(cid:48) of layer 1, respectively, which we shall name as configuration (31(cid:48)42(cid:48)). The superlattice unit cell is then given by the parallelogram with edges Ci and A(cid:48) i in layer 2 after relative deformation). This configuration then has relative rotation, shear and expansion, which can be seen in the following calculation of relative displacement field u. The relative displacement field u of (31(cid:48)42(cid:48)) relative to the undistorted TBG (11(cid:48)22(cid:48)) can be solved as follows: Define U(r) = u(r) + θir × z, which the relative displacement compared to the untwisted (AA stacking) bilayer graphene (The undistorted TBG (11(cid:48)22(cid:48)) has displacement θir × z relative to the untwisted bilayer graphene). The deformation field U then satisfies (Ci · ∇)U ≈ Di − Ci, and (Ai · ∇)U ≈ Bi − Ai. We then find the relative displacement field compared to undistorted TBG (11(cid:48)22(cid:48)) for large i is i in layer 1 (which coincide with the parallelogram with edges Di and B(cid:48) i (cid:18) ∂xux ∂xuy ∂yux ∂yuy (cid:19) ≈  LD i i −LC LC i −LC i LD √ i 3LC i  . θi − θ(cid:48) θi−θ(cid:48) i√ 3 i We now explain why we choose lattice points 3 and 4 (3(cid:48) and 4(cid:48)) to define the deformed TBG. In our ab initio M and KM points calculations, this is to ensure the K point of graphene BZ of layers 1 and 2 coincide with the K(cid:48) MMK-0.4-0.200.20.4Energy (eV)ΓΓMMK-0.4-0.200.20.4Energy (eV)ΓΓabMKM-0.4-0.200.20.4Energy (eV)ΓMKM-0.4-0.200.20.4Energy (eV)Γcd26 of MBZ when folded into the deformed MBZ, respectively. This allows the band structure to be compared with that of the continuum model (in which K point of two layers always coincide with K(cid:48) M and KM ). To see this, consider a configuration that has superlattice vectors (which spans the superlattice unit cell parallelogram) R1 = m1a + n1b and R2 = m2a + n2b, where a and b are the lattice vectors of the lattice of layer 1. The reciprocal vectors of the superlattice gi (i = 1, 2) satisfies gi · Rj = 2πδij. On the other hand, the momentum of the K point of layer 1 is KD as we defined in the first section, which satisfies KD · a = KD · b = 2π/3. Therefore, we find KD · R1 = 2π(m1 + n1)/3 , KD · R2 = 2π(m2 + n2)/3 . M point of the MBZ is located at momentum K(cid:48) This shows KD = [(m1 + n1)g1 + (m2 + n2)g2]/3. Note that K(cid:48) M = −(g1 +g2)/3. Therefore, in order for KD to coincide with K(cid:48) M point, one has to have m1 +n1 ≡ m2 +n2 ≡ −1 (mod 3) (so that KD and K(cid:48) M differ by integer multiples of superlattice reciprocal vectors). One can easily see this is satisfied for configuration (11(cid:48)22(cid:48)) where R1 = Ai and R2 = A(cid:48) In order to find another configuration satisfying this i. condition, one has to change m1 + n1 and m2 + n2 by multiples of 3. The configurations (33(cid:48)44(cid:48)) and (31(cid:48)42(cid:48)) are two such configurations which have small deformations. We then use ab initio to calculate the band structure for configurations (33(cid:48)44(cid:48)) and (31(cid:48)42(cid:48)) with i = 10 (see the black dashed lines in Fig. 10c and 10d), respectively, and compare with that of the undistorted configuration (11(cid:48)22(cid:48)) (the red solid lines in Fig. 8c and 8d). Accordingly, we calculate the deformation of TBG band structure from the continuum model (at graphene valley K) for configurations (33'44') and (31'42') (using deformations in Eqs. (67) and (68)) at θ = θ10 = 3.15◦, respectively, and the results are shown in Fig. 10a and 10b, where the red solid lines are the original band structure, and the black dashed lines are the deformed band structure. In the figure, the high symmetry points should be understood as those of the MBZ. First, we note the band structures before deformations match well between the continuum model and the ab initio. The continuum model bands appear to be less in number than that of ab initio, which is because we have only plotted the bands of continuum model at valley K. Besides, for both configurations, one can see that the deformed band energies in the continuum model also have the same order of magnitude as that in ab initio. Therefore, our estimation of electron-phonon coupling strength from the continuum model in the previous sections has the correct order of magnitude.
1302.4768	3	1302	2013-10-01T00:33:08	The single-electron transport in a three-ion magnetic molecule modulated by a transverse field	[ "cond-mat.mes-hall" ]	We study single-electron transport in a three-ion molecule with strong uniaxial anisotropy and in the presence of a transverse magnetic field. Two magnetic ions are connected to each other through a third, nonmagnetic ion. The magnetic ions are coupled to ideal metallic leads and a back gate voltage is applied to the molecule, forming a field-effect transistor. The microscopic Hamiltonian describing this system includes inter-ion hopping, on-site repulsions, and magnetic anisotropies. For a range of values of the parameters of the Hamiltonian, we obtain an energy spectrum similar to that of single-molecule magnets in the giant-spin approximation where the two states with maximum spin projection along the uniaxial anisotropy axis are well separated from other states. In addition, upon applying an external in-plane magnetic field, the energy gap between the ground and first excited states of the molecule oscillates, going to zero at certain special values of the field, in analogy to the diabolical points resulting from Berry phase interference in the giant spin model. Thus, our microscopic model provides the same phenomenological behavior expected from the giant spin model of a single-molecule magnet but with direct access to the internal structure of the molecule, thus making it more appropriate for realistic electronic transport studies. To illustrate this point, the nonlinear electronic transport in the sequential tunneling regime is evaluated for values of the field near these degeneracy points. We show that the existence of these points has a clear signature in the I-V characteristics of the molecule, most notably the modulation of excitation lines in the differential conductance.	cond-mat.mes-hall	cond-mat	The single-electron transport in a three-ion magnetic molecule modulated by a transverse field Javier I. Romero∗ and Eduardo R. Mucciolo† Department of Physics, University of Central Florida, Orlando, Florida 32816, USA (Dated: March 26, 2021) We study single-electron transport in a three-ion molecule with strong uniaxial anisotropy and in the presence of a transverse magnetic field. Two magnetic ions are connected to each other through a third, nonmagnetic ion. The magnetic ions are coupled to ideal metallic leads and a back gate voltage is applied to the molecule, forming a field-effect transistor. The microscopic Hamiltonian describing this system includes inter-ion hopping, on-site repulsions, and magnetic anisotropies. For a range of values of the parameters of the Hamiltonian, we obtain an energy spectrum similar to that of single-molecule magnets in the giant-spin approximation where the two states with maximum spin projection along the uniaxial anisotropy axis are well separated from other states. In addition, upon applying an external in-plane magnetic field, the energy gap between the ground and first excited states of the molecule oscillates, going to zero at certain special values of the field, in analogy to the diabolical points resulting from Berry phase interference in the giant spin model. Thus, our microscopic model provides the same phenomenological behavior expected from the giant spin model of a single-molecule magnet but with direct access to the internal structure of the molecule, thus making it more appropriate for realistic electronic transport studies. To illustrate this point, the nonlinear electronic transport in the sequential tunneling regime is evaluated for values of the field near these degeneracy points. We show that the existence of these points has a clear signature in the I − V characteristics of the molecule, most notably the modulation of excitation lines in the differential conductance. I. INTRODUCTION From the various types of magnetic states in matter, ferromagnetism and its microscopic causes is one of the most intriguing topics. A very singular class of mag- netic systems are single-molecule magnets (SMMs).1 -- 3 Molecules in this class typically have a large net spin ground state and exhibit unusual attributes such as quan- tum tunneling of the magnetization (QTM),4,5 a rela- tively large decoherence time,6 and Berry-phase interfer- ence effects in the presence of magnetic fields.7,8 Com- monly, SMMs are composed of transition metal ions (open 3d or 4f shells) bridged by ligand atoms and molecules. The phenomenological Hamiltonian that de- scribes the magnetic properties of SMMs is the so-called giant spin approximation (GSA) model. For instance, in its simplest form, the GSA Hamiltonian can be written as z + E (cid:0)S2 x − S2 y (cid:1) , HGSA = −D S2 (1) The net magnetization has a preferential direction (an easy axis z in this example) as a consequence of the dominant uniaxial anisotropy (D (cid:28) E). The in-plane transverse anisotropy allows the total spin to transit be- tween different values of Sz. In this particular example, the molecule has a predominant second-order anisotropy signaling a rhombic symmetry. A more microscopic description of SMMs comes from considering interactions at the ion level. In this case, one uses instead a multi-spin Hamiltonian of the type Hms = iz + ei ix − S2 iy Jij (cid:126)Si· (cid:126)Sj. (2) (cid:88) (cid:2)−diS2 i (cid:0)S2 (cid:1)(cid:3)−(cid:88) i(cid:54)=j Here each magnetic site of the molecule has local uniax- ial and transverse anisotropies di and ei, respectively. In addition, there is an effective isotropic ferromagnetic in- teraction between pairs of sites parametrized by Jij > 0, which contributes to the energy splitting of states with different total spin. (Some molecules are better de- scribed by an anisotropic Jij.) We note that the on-site anisotropy terms (proportional to di and ei) in Eq. (2) are meaningful only when the total spin of each site is Si ≥ 1. The accuracy of the GSA and the multi-spin models was investigated in Ref. 9 by studying the connection between molecular and single-ion uniaxial anisotropies. Anisotropies in SMMs are very sensitive to the orienta- tion of the ligands with respect to each magnetic ion, thus the energy spectrum of SMMs varies greatly with magnetic site symmetries. For instance, in the case of zero transverse anisotropies (ei = 0 and, consequently, E = 0), the Hamiltonians in Eqs. (1) and (2) yield a two-fold degenerate ground state involving spin states parallel (Sz = S) and anti-parallel (Sz = −S) to the uniaxial anisotropy axis. If the transverse anisotropy terms (ei and E) are nonzero, then the rotational Sz symmetry is broken and the two-fold degeneracy is lifted, with the ground and first excited states now being anti- symmetric and symmetric combinations of the Sz = ±S states. In general, the parameters used in the multi-spin approach depend on the intra-site or inter-site interac- tions between orbitals of the ligands and transition met- als, but due to the large Hilbert space involved and the strong interplay between the many microscopic parame- ters, it is very challenging to take into account explicitly the overlap between orbitals and their symmetries for large magnetic molecules. One often employs the em- 3 1 0 2 t c O 1 ] l l a h - s e m . t a m - d n o c [ 3 v 8 6 7 4 . 2 0 3 1 : v i X r a pirical Goodenough-Kanamori rules10,11 that dictate the nature of interaction between magnetic ions. In addi- tion, one also includes an effective interaction intermedi- ated by diamagnetic atoms (the Anderson superexchange interaction).12 In recent years, interest emerged in exploring how the magnetic properties of SMMs may affect the molecule's electronic transport properties. Effects due to the QTM,13 Berry-phase interference,14,15 and their inter- play with the Kondo resonance16,17 have been proposed. Several authors have also used first-principles calcula- tions to investigate the electronic configuration, the mag- netic properties, and coherent transport in SMMs.18 -- 21 In addition, many important contributions have been made in studying the sequential transport for different molecu- lar setups,22,23 as well as sequential transport dependence on spin-orbit coupling, the Jahn-Teller effect, and ligand charge variations.24 -- 26 In this paper, we investigate the dependence of sequen- tial tunneling transport on an transverse magnetic field applied to a three-ion model that contains the essential microscopic details necessary for reproducing a SMM be- havior. The microscopic model comprises two magnetic ions bridged by a third diamagnetic ion and takes into account the valence, ligand fields, and orbital energies of the ions, as well as direct and exchange Coulomb inter- actions present in the molecule. Our goal is to develop a simple phenomenological microscopic model of a SMM that goes beyond the giant spin Hamiltonian model and is amenable to realistic electronic transport studies. We explore the energy spectrum of this system for sev- eral points in the parameter space of the Hamiltonian. We find that for a certain parameter range a magnetic bistability, similar to that observed for SMMs, develops. This bistability is characterized by a ground state degen- eracy point and is driven by a magnetic field perpendic- ular to the molecule's main anisotropy axis. The modu- lation of the transition between magnetic states through a transverse magnetic field is one of the most unusual features of SMMs and is fully reproduced by our model. In the context of the giant spin model of an SMM, this modulation is understood as the result of Berry-phase interference of multiple spin tunneling paths, which in turn lead to the appearance of diabolical points in the molecule's spectrum.27 In contrast, Our model contains many degrees of freedom and there are no easily identi- fied topological phases. Instead, the modulation in our model arises from changes in electronic correlations when a finite transverse field is present. We evaluate the in- coherent electronic transport through the molecule near this bistability. At the degeneracy points, states with op- posite spins are decoupled, resulting in a clear qualitative change in the differential conductance of the molecule. The paper is organized as follows: In Sec. II we de- scribe the model Hamiltonian for the three-ion magnetic molecule. In Sec. III we present our choice of model pa- rameters and the energy spectrum and symmetry of the states of the molecule. In Sec. IV we evaluate the sequen- 2 tial current and differential conductance of the molecule for in-plane magnetic fields ranging between zero and the nearest degeneracy point. In Sec. V we discuss the ef- fects of spin tunneling modulation by a magnetic field on transport. Finally, in Sec. VI, we analyze our results and draw some conclusions. Figure 1: (Color online) Scheme of a simple three-ion SMM. For simplicity, we do not depict the rhombic ligand environ- ment surrounding ions a and c. The magnetic ions labeled by a and c interact through the diamagnetic ligand ion b with a bond angle of θ = π/2, so that the dx2−y2 orbitals (green/solid lobes) overlap with the px and py orbitals (pur- ple/dashed lobes) of the ligand separately. II. THE THREE-ION MODEL OF A SMM A minimal realistic molecular core capable of repro- ducing the main features of a SMM such as large to- tal spin (S > 1/2), uniaxial and in-plane anisotropies, and transverse field modulation consists of two transition metal ions bridged by a diamagnetic ligand ion. Con- sider the system shown schematically in Fig. 1. The two transition metal ions a and c have a 3d8 electronic configuration and a total spin S = 1 each. They in- teract magnetically through a superexchange interaction intermediated by a diamagnetic O2− ion (electronic con- figuration 2p6), represented by b in Fig. 1. The five-fold degeneracy of the 3d orbitals in the magnetic ions is bro- ken due to the bonding to ligands. To simulate such an effect, we assume that a weak orthorhombic ligand field acts on a and c, inducing local uniaxial and trans- verse anisotropies on each site. Thus, for an orthorhom- bic symmetry (D2h point group), the ground states of ions a and c have two occupied unpaired single-particle orbitals (a1,2 and c1,2, respectively) which are symmetric and antisymmetric combinations of dx2−y2 and d3z2−r2 d states. We also consider a 90o angle between their bonds. As a result, the dx2−y2 components of the single-particle orbitals in both ions a and c overlap separately with the px,y orbitals b1,2 of ion b. Under this assumption, we use the appropriate Slater-Koster two-center integrals, which then yields Epx,dx2−y2 = Epy,dx2−y2 = (pdσ) 3/2. This configuration favors a ferromagnetic superexchange in- teraction. In our case the Epz,d3z2−r2 and Epx,y,dx2−y2 two-center integrals are zero. √ Figure 2: Scheme of the electron hopping (double-arrow lines) between magnetic ions and the diamagnetic ion. For a 3d8 ion in a weak orthorhombic ligand field (i.e., with a small distor- tion to rhombic symmetry), the unpaired single-particle or- bitals (say for ion a) are ψa2 = κ1 dx2−y2 + κ2 d3z2−r2 and ψa1 = κ1 d3z2−r2 − κ2 dx2−y2 , where κ2 2 = 1. Since the rhombic contribution to the ligand field is small, we regard the mixing amplitude κ2 between the initially degenerate un- paired orbitals as a small parameter: κ2 (cid:28) κ1. 1 + κ2 The scheme of these selective orbital overlaps is shown in Fig. 2. The features of our model are captured by the effective Hamiltonian Hmol = Ha + Hb + Hc + Hab + Hbc, (3) where Hα, with α = a,c denotes the Hamiltonian for a magnetic ion, Hα = εM,i nαi,σ + UM nαi,↑ nαi,↓ (cid:88) (cid:88) (cid:1) , (cid:88) i=1,2 σ=↑,↓ nα1,σ nα2,σ(cid:48) − JM (cid:126)Sα1 · (cid:126)Sα2 x,α − S2 (cid:88) y,α nbi,↑ nbi,↓ i=1,2 σ=↑,↓ (cid:88) i=1,2 + U(cid:48) M − d S2 σ,σ(cid:48)=↑,↓ (cid:88) z,α + e(cid:0)S2 (cid:88) i=1,2 + U(cid:48) O (cid:88) σ,σ(cid:48)=↑,↓ nb1,σ nb2,σ(cid:48) − JO (cid:126)Sb1 · (cid:126)Sb2. (5) (4) and Hb describes the diamagnetic ion, Hb = εO nbi,σ + UO For simplicity we consider that the a and c ions have identical rhombic symmetries and therefore they have the same interaction strengths, orbital energies, and anisotropy parameters. We allow for a crystal field split- ting of the magnetic ion orbitals, but assume that the orbitals in the diamagnetic ion are degenerate. In Eq. 3 (3), Hab and Hbc describe electron hopping between the magnetic ions and the diamagnetic ion, Hab = t and Hbc = t (cid:88) (cid:88) σ=↑,↓ a2,σ cb1,σ + H.c.(cid:1) (cid:0)c† (cid:16) (cid:17) † c b2,σ cc2,σ + H.c. σ=↑,↓ (6) , (7) with α = a, c. In Eqs. (4), (6), and (7), nαi,σ = c† αi,σcαiσ, where c† αi,σ (cαiσ) creates (annihilates) an electron with z spin projection σ in the orbital i of ion α = a, b, c and satisfy the standard fermionic anticommutation re- lations. The total spin operator associated to an orbital i in ion α is (cid:126)Sαi while (cid:126)Sα = (cid:126)Sα1 + (cid:126)Sα2 . The first term on the r.h.s of Eqs. (4) and (4) ac- counts for the single-particle orbital energies, while the second and third terms represent the on-site and intra- site Coulomb repulsion within ion α. The fourth term enforces Hund's first rule, maximizing the total spin on each ion (Jα > 0). The last two terms on the r.h.s. of Eq. (4) are the uniaxial and transverse on-site anisotropies produced by the rhombic ligand field environment and by the spin-orbit coupling within each magnetic ion. III. ADJUSTING THE MODEL PARAMETERS The model Hamiltonian in Eq. (3) has a large number of parameters that have to be properly adjusted in or- der to produce the phenomenology expected of a SMM. Below, we discuss our choices of parameter values. We first consider the Hamiltonian in the absence of trans- verse anisotropies (eα = 0). Then we include non-zero eα terms, and an in-plane external magnetic field in or- der to produce degeneracy points in the spectrum of the molecule. A. Total Sz spin in the absence of transverse anisotropy The lowest energy states of a SMM typically have a large total spin S, with the maximum spin projection Sz = S happening at the ground state, followed by excited states corresponding to decreasing spin projec- tions. A convenient way to produce such a spectrum in our model is to start with Sz as a good quantum number by setting eα = 0. Then, for different choices of the pa- rameters εM,i, εO, UM, U(cid:48) O, JM, JO and d, we calculate the total Sz spin of the molecule as a function of t. For the states with maximum possible spin, S = 2 for our three-ion model, we compare their composition to that expected for a state obtained by adding two S = 1 spin states. Even though U(cid:48) O are nonzero in a real molecule, we have found that they have little qual- itative importance in our results. Thus, for the sake of M, UO, U(cid:48) M and U(cid:48) simplicity and in order to decrease our parameter space, we have set them equal to zero. Realistic values for some parameters can be obtained from the review by Imada, Fujimori, and Tokura28 and the recent work by Kim and Min29 using NiO systems as a reference. Thus, from the data in Table III of Ref. 28 we set the on-site Coulomb repulsion between d electrons to UM = 7 eV. Typical values for the on-site Coulomb repulsion of p-electrons are in the range from 4 − 7 eV (see Refs. 30 and 31). Here, we assume UO = 4 eV. From table III of Ref. 28, the Hund's rule parameter for the magnetic ion is set to JM = 0.95 eV. For the diamagnetic ion, we choose a slightly larger value, JO = 1.5 eV in order yield a spectrum similar to that of a ferromagnetic SMM. The orbital energies in the magnetic 1 −εO = 0.66 ions are obtained from Fig. 59 of Ref. 28: εM 2 − εO = 0.72 eV. For convenience, we shift the eV and εM total energy such that εO = 0. The energy splitting between the two orbitals of the magnetic ions, εM 1 = ∆εcrystal is due to the crystal field splitting produced by the surrounding ligands. In Ref. 32, rhombic crystal field parameters are found to be of the order of meV or smaller; we choose the splitting to be about 10 meV. 2 − εO Based on the EPR measurements in Ref. 9, the uni- axial anisotropy parameter is set to d = 0.6 meV (equiv- alent to 7 K). Figure 3 shows Sz for the ground state (two-fold degenerate), first excited state (two-fold degen- erate), and second excited state (nondegenerate) versus the hopping parameter t. It is clear that for the param- eter we chose, the ground state has the highest possible total spin, S = 2. For t > 0.75 eV, states with total spin projection Sz = 1 become the first excited states, while the state with total spin projection zero moves up to the second excited state position. This is the typical case for a SMM within the description provided by the GSA Hamiltonian of Eq. (1). B. Anisotropy and degeneracy points We now consider the inclusion of local transverse anisotropies that make the eigenstates of the Hamiltonian to be symmetric and antisymmetric combinations of Sz(cid:105) states. In addition, we consider the effect of an applied transverse (in-plane) magnetic field on the molecule's en- ergy spectrum. The transverse field Hamiltonian is given by (cid:33) (cid:32)(cid:88) Hfield = b gx,αSx,α cos φ + gy,α Sy,α sin φ , (8) α=a,c where b = µB (cid:126)B and (cid:126)B = (B cos φ, B sin φ, 0) is the external magnetic field. Following Refs. 33 and 34, a reasonable estimate of the anisotropic g-factors would be gx,α = 2.270 and gy,α = 2.269. We set the hopping amplitude to t = 0.8 eV, while the transverse anisotropy parameter is set to e = 5 µeV, which is less than 1% of the uniaxial anisotropy parameter d. We do not consider 4 Figure 3: (Color online) Total spin projection Sz of the molecule. For 0 < t < 1.2 eV, the lowest energy level is a ± 2(cid:105) state, the first excited state has Sz = 0, while the next (degenerate) excited states have Sz = ±1. We find the typ- ical SMM behavior in the 0.7 eV < t < 1 eV range, where the first excited states have Sz = ±1 and the highest energy level has Sz = 0. The crossing in the graph denotes the point where total Sz spin changes for the first and second excited eigenstates. For t = 0.77 eV,the energy splitting between the two lowest eigenstates (Sz = ±2 and Sz = ±1) is of the order of the uniaxial anisotropy parameters dα, with the value of ∆ = 0.07eV, while the splitting between Sz = ±1 and Sz = 0 is ∆ = 0.5 meV. coupling of the magnetic field to the middle ion since it would change the energy levels of the high energy sector by a very small amount, given that b (cid:28) U . The goal is to find φ and b such that the in-plane anisotropy brought by the external magnetic field compensates the intrinsic in-plane anisotropy of the magnetic ions, removing the splitting between states with maximum projection Sz, thus restoring degeneracy to the ground state. Figure 4: (Color online) Splitting of the two lowest energy levels versus the transverse field angle φ. The angle is mea- sured with respect to the positive x axis. We see that for various values of the magnetic field, the lowest splitting oc- curs at φ = π. In Fig. 4, we show the resulting splitting of the two lowest energy levels versus the angle φ. The angle φ = π creates symmetric spin paths for the tunneling of the molecule's magnetization, leading to maximum destruc- tive interference. In Fig. 5 the splitting is shown as a (cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)gsΕ1Ε20.20.40.60.81.0teV0.51.01.52.0Sz(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:230)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)0.51.01.52.0Φ(cid:61)nΠ0.050.100.150.200.25(cid:68)gs,Ε1ΜeVb=12b=14b=17ΜeVΜeVΜeVfunction of the applied field magnitude b at this partic- ular field direction (φ = π). We observe that the second degeneracy point occurs approximately at twice the value of the first one. We note that the periodic modulation of the splitting with b is as characteristic feature of SMMs (see Refs. 4,7,27). We now consider how the symmetry of the eigenstates is affected by changes in the magnitude of the magnetic field. For this purpose, we define the symmetry coeffi- cient CΣ = C1 + C2, 1 2 (9) where C1,2 are the amplitudes of the two degenerate states along the Sz = ±2 basis states. If CΣ = 0 the eigenstate is antisymmetric, whereas if CΣ = C1,2 the eigenstate is symmetric. The results are presented in Fig. 6. We observe that the ground state changes from symmetric to antisymmetric at the first degeneracy point (left dashed line), while the first excited state does the reverse. The energy splitting between the linear combina- tion of Sz states (symmetric and antisymmetric) vanishes at the degeneracy points. Figure 5: (Color online) Splitting of the two lowest energy levels versus the magnitude of the magnetic field b at an an- gle of φ = π. We observe two degeneracy points where the splitting between the two lowest energy levels goes to zero. The first point occurs at b ≈ 18 µeV (B ≈ 0.3 tesla), while the second one occurs at b ≈ 34 µeV (B ≈ 0.6 tesla). IV. SEQUENTIAL TRANSPORT THROUGH THE MOLECULE While a fully-coherent transport approach can be used, it is not essential in describing theoretically the effect of the transverse magnetic field modulation on the SET I- V characteristics, which can be fully captured by a rate equation approach in the dc limit. We study electronic transport in our model system by connecting the molecule to two reservoirs of noninter- acting electrons. The current through the molecule can be controlled by applying a voltage difference between the reservoirs, as well as by changing the total charge 5 Figure 6: (Color online) Symmetry of the ground state (filled squares) and first excited state (empty squares) versus the magnetic field magnitude b. At b = 0 the ground and first excited states are gs(cid:105) = −0.68(+2(cid:105)+−2(cid:105))+0.27×(contam.) and 1(cid:105) = 0.69( + 2(cid:105)−− 2(cid:105)) + 0.22× (contam.), respectively, where (contam.) represents a contribution from spin states other than ± 2(cid:105). As the transverse magnetic field grows, the amplitude CΣ tends to decrease, allowing for an increasing admixture with Sz = 0 states. At the second degeneracy point (right dashed line), the symmetry of the states changes again. of the molecule through an applied back-gate voltage. We define a general charge-spin state by (n, S) where n and S denote the excess charge and the total spin of the molecule, respectively. For simplicity, we only consider transitions between two charge states (n = 0, 1). Ini- tially, the molecule is in the state (0, 2) where it has a 3d8 − 2p6 − 3d8 electronic configuration. We then allow one electron to hop from the reservoirs into the molecule and restrict it to be localized either on the a1 or c1 or- bitals, bringing the molecule to the (1, 3 2 ) state, which comprises both 3d9 − 2p6 − 3d8 and 3d8 − 2p6 − 3d9 elec- tronic configurations. We assume that the coupling to the reservoirs is such that there is equal probability for an electron to land in either one of the two magnetic ions. Since one of the ions changes its oxidation state when an electron is added, its anisotropy terms and the g-factors change as well (see Refs. 13, 16, 35, 36, and 37). Concerning model parameters, we assume a reduction in the local uniaxial anisotropy of the ion upon changing its charge state and set d(cid:48) α = 0.3 meV. It is worth noting that in the context of the GSA, the anisotropy parame- ters change their values upon varying the charge state of the molecule (see Ref. 40). In our transport analysis, d(cid:48) α does not play an essential role since we will focus on the contributions coming from the two lowest spin states of the (1,3/2) charge-spin sector which can only have pure (or combinations of) Sz = 3 2 z-spin components. The modified transverse anisotropy e(cid:48) α does not take part in the Hamiltonian for the (1, 3 2 ) configuration since the to- tal spin of the ion receiving an electron becomes Sα = 1 2 , turning the local ground state of the ion a Kramers dou- blet. Finally, for the sake of simplicity, we assume that b1b'1010203040bΜeV0.020.040.060.080.100.120.14(cid:68)gs,Ε1ΜeV(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:224)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)(cid:225)gsΕ1gssΕ1asΕ1sgsas10203040bΜeV0.20.40.60.81.0C(cid:83)the same g factors used for the (0, 2) configuration. In order to evaluate single-electron transport, we start by diagonalizing the Hamiltonian for different charge- spin configurations. We then solve a set of coupled dif- ferential equations for the time evolution of the quan- tum state probabilities of the molecule. This approach is suitable for describing incoherent, sequential transport across the molecule. In the context of our model, its use is justifiable because the phase of the itinerant electron does not play an essential role in the existence of degen- eracy points in the molecule's spectrum. In other words, the situation is similar to that of a quantum dot where external parameters such as plunger gate voltages can tune the states available to an itinerant electrons tunnel- ing in and out of the system regardless of the electron's phase coherence. The only necessary assumption for the rate equation approach to be valid is that the molecule should be weakly coupled to the leads and the temper- ature sufficiently low. These conditions can be cast as γ, kBT (cid:28) ∆ (see notation below). The rate equation describing the time evolution of the probability of the molecule to be in an arbitrary state m is given by38 (cid:88) m(cid:48) dpm dt = −pm (cid:88) m(cid:48) Γm→m(cid:48) + pm(cid:48) Γm(cid:48)→m, (10) where Γm→m(cid:48) is the transition rate between different eigenstates m and m(cid:48). The first term on the r.h.s of Eq.(10) comprises outgoing terms describing the transi- tion from an initial state m to a final state m(cid:48) by either taking away or adding an extra electron. The second term represents processes where a final state m(cid:48) transi- tions back to an initial state m (again, by means of either taking away or adding an extra electron). Equation (10) can be written in matrix form as = Γ · P, (11) where P is a vector containing the probabilities {pm} and Γ is a matrix with entries {Γmm(cid:48)} which are related to the the transition rates in the following way: m (cid:54)= m(cid:48) Γmm(cid:48) = Γm(cid:48)→m dP dt if Γmm = − (cid:88) 6 the Fermi distribution of the left (L) or right (R) reser- voirs. The electrochemical potential of the molecule, i.e., the energy required to go from a state m to state a m(cid:48), is defined as µmm(cid:48) = m − m(cid:48) − eηVg, (14) where m and m(cid:48) are the energy eigenvalues correspond- ing to eigenstates m and m(cid:48), respectively, and Vg is the backgate voltage (we set the lever arm coefficient η = 1). The coefficients γτ β,β(cid:48) are the tunneling rates between two eigenstates β and β(cid:48) of the molecule and are given in the Golden rule approximation by γτ β,β(cid:48) = 2πρ T σ,τ ββ(cid:48)2 (15) (cid:88) σ where ρ is the density of states of both reservoirs (here considered constant) and T σ,τ ββ(cid:48) = † tj,τ(cid:104)βc jσβ(cid:48)(cid:105) (16) (cid:88) j denote the tunneling matrix elements. The operator c† αj σ creates an electron with spin σ on the single-particle or- bital j of ion α. To simplify the calculations we assume that tj,τ is independent of the orbital, thus we drop the j index, also we assume that tL = tR. Since different charge configurations of the molecule involve states with different spin quantum numbers, selection rules for spin transitions become important in the transport calcula- tions. The tunneling matrix elements satisfy the selec- tion rules Sm − Sm(cid:48) = 1/2, Szm − Szm(cid:48) = 1/2 and Nm − Nm(cid:48) = 1, for any two eigenstates of the Hamilto- nian belonging to different charge-spin sectors. The net steady-state current through the left lead (we omit the L subscript) is given by ΓL m→m(cid:48)pm, (17) I = e(cid:88) mm(cid:48) where m, m(cid:48) are indexes for different charge states. In (17), multiply the r.h.s it by −1 if an electron is Eq. going out of the molecule and into the left reservoir. Γm→m(cid:48). (12) V. EFFECT OF SPIN TUNNELING MODULATION ON TRANSPORT m(cid:54)=m(cid:48) In the stationary regime, we look for the steady-state probabilities pm such that dpm/dt = 0. In matrix form, this is achieved by finding the eigenvector P of the ma- trix Γ with a zero eigenvalue. The transition rates from a state m to a state m(cid:48) when adding (or subtracting) electrons (e−) to (from) the molecule are given by e− in, [1 − fτ (µmm(cid:48) − eVτ )] , e− out m(cid:48),m fτ (µm(cid:48)m − eVτ ), γτ m,m(cid:48) (cid:26) γτ m→m(cid:48) = Γτ (13) where the index τ = L, R refers to the left and right reservoirs, respectively. Here, fτ (x) = 1/(1 + ex/kB T ) is Transport is evaluated for different values of the field between in the range [0 : b2], where bS denotes the mag- nitude of the field for which a degeneracy point occurs for the charge-spin sector with total spin S = 2. In order to see signatures of magnetization tunneling interference on transport, we only consider transitions between the two lowest energy levels of each charge-spin sector, as shown in Fig. 7. The ground and first excited eigen- states of charge sector n are labeled as gs(n)(cid:105) and (n) 1 (cid:105), respectively. In Fig. 7, we see that for b = 0 the eigen- states of the (0, 2) configuration are linear combinations of Sz = ±2(cid:105) spin states, while the ground state of the 2 ) sector is two-fold degenerate with the Sz = ± 3 2(cid:105) (1, 3 spin states having the same energy. For b = b2 the tunnel splitting of the (0, 2) configuration goes to zero and the ground state becomes the two-fold degenerate Sz = ±2(cid:105) spin states. As for the (1, 3 2 ) sector, an energy splitting (∆b2 ) is induced by the magnetic field, with the eigen- 2(cid:105) spin states being now a linear combination of Sz = ± 3 states. In both cases, the corresponding selection rules allow transitions between all eigenstates. 7 Figure 8: Splittings between the two lowest energy levels for the S = 3/2 (solid line) and S = 2 (dashed line) total spin sectors. The first degeneracy point for S = 2 is at b2 ≈ 18 µeV (equivalent to 0.3 tesla). cally engineering the SMM ligands, arriving at such low temperatures in single-electron transistor setups is quite challenging. Figure 7: (Color online) Allowed transitions (dashed arrows) between the two lowest energy levels for the (0, 2) and (1, 3/2) charge-spin sectors. In (a), ∆0 is the splitting between eigen- states for zero field (b = 0) while in (b) ∆b2 is the splitting when the field is tuned to the degeneracy point (b = b2). Interesting physics can also be found if all combina- tions of Sz states corresponding to a particular total spin S of the molecule are considered. This is because anisotropy contributions tend to contaminate the lowest energy levels with a small admixture of Sz = 0 states, opening transitions in the transport that would other- wise have been forbidden by selection rules. A study of these contributions to transport within the GSA can be found in Ref. 16. Since we are primarily interested in how the modulation of the transverse anisotropy split- ting affects transport, we only consider the two lowest energy Sz states. For other states, the energy cost to ac- cess different transitions would be too large compared to the energy gap generated by the transverse anisotropy. We tune the gate voltage Vg so that the ground states energies for the (0, 2) and (1, 3 2 ) charge-spin sectors are aligned and vary Vg slightly around this point. Since the tunnel splitting due to transverse anisotropies is very small, of the order of the µeV (see Fig. 8), a very small temperature as well as a very low coupling γτ β,β(cid:48) are re- quired to resolve features in the electronic transport that can be associated to the tunneling of the magnetization. Thus, we set the temperature in the reservoirs to T = 0.1 mK and choose the product ρT τ ββ(cid:48)2 so that both kBT and γτ ββ(cid:48) are much smaller than the energy level separa- tion within the molecule. We note that while, in prac- tice, a small value for γτ β,β(cid:48) can be achieved by chemi- The left-lead current (I) to/from the molecule is shown in Fig. 9. For b = 0, transitions between excited and ground states are seen in the current steps at zero bias voltage and at Vb/kBT ≈ ±13. When b = b2, these steps can be seen instead at Vb/kBT ≈ ±61. Figure 10 shows a plot of the differential conductance (dI/dVb) as a function of the bias voltage and the magnetic field. Resonance peaks seen at b(cid:48) = 0 and b(cid:48) = 1 correspond to the current steps of Fig. 9. As one approaches the degeneracy point (b(cid:48) = 1), the conductance peak cor- responding to the gs(1)(cid:105) ↔ (0) 1 (cid:105) transition is shifted towards the larger conductance peak (at V (cid:48) b = 0). At the degeneracy point, this peak is absorbed by the zero- bias conductance peak, increasing the current flow be- tween the ground states of the molecule. In addition, we notice that the conductance decreases when the field is close to zero. We also observe that new resonances appear as a consequence of the field-induced energy gap of the (1, 3 2 ) charge-spin sector. These can be seen in the conductance peaks coming out of the zero-bias and 1 (cid:105) transition resonance peaks in the b(cid:48) = 0 gs(1)(cid:105) ↔ (0) plane. For some values of the field, conductance peaks arise as resonances merge with each other. These peaks appear when the electrochemical potentials in the dot are = µ(0) the same i.e., µgs(0),(1) , 1 ,(1) and µgs(1),(0) . Figures 11 and 12 show contour 1 ,(1) plots of the differential conductance versus bias and gate voltage. The positive slope line seen for b = 0 in Fig. 11 1 (cid:105) transition and is elim- corresponds to the gs(1)(cid:105) ↔ (0) inated upon tuning the field to the degeneracy point. At the same time, the applied magnetic field creates an en- ergy gap allowing the gs(0)(cid:105) ↔ (1) 1 (cid:105) transition, which corresponds to the negative slope line in Fig. 12. , µgs(0),gs(1) = µ(0) = µ(0) 1 ,(1) 1 1 1 1 1 In Figs. 11 and 12, in order to show in detail the fine features that appear upon crossing a degeneracy point, (a)(b)(cid:68)0(cid:68)b2S=3/2S=2010203040bΜeV0.10.20.30.40.50.60.7(cid:68)gs,Ε1ΜeV8 Figure 11: Contour plot of the differential conductance dI/dVb as a function of bias and gate voltage at b = 0. Figure 9: Current through left lead versus bias voltage at b = 0 (solid line) and b = b2 (dashed line) for T = 1 mK (current is shown in arbitrary units). The bias voltage is plotted in units of kBT /e. The gate voltage for each curve is such that the ground states for the two charge sectors involved are aligned. the gate voltage Vg was shifted and rescaled: V (cid:48) Vs) × 103, where eVs = 8.169 eV. g = (Vg − VI. CONCLUSION We have studied the incoherent electronic transport through an anisotropic magnetic molecule using a micro- scopic model that provides spectral properties similar to Figure 12: Contour plot of the differential conductance, dI/dV in arbitrary units, as a function of bias and gate volt- age at b = b2. those of a multi-ion single-molecule magnet. By describ- ing the molecule's core as a set of a few multi-orbital quantum dots, we open the door to a better understand- ing of the interplay between internal degrees of freedom of the molecule and its transport properties. This can- not be done within the giant spin approximation which is usually the starting point for characterizing the behavior of SMMs. Another advantage of our model is its simplicity: by reducing the degrees of freedom to a manageable number, the model makes it possible to study in more detail the in- fluence of the molecule's geometry and ion arrangement, as well as of the electron path across the molecule, on transport. In addition, by keeping the number of degrees of freedom small, it might be possible to go beyond the incoherent regime and use this model to study strongly Figure 10: Plot of the differential conductance dI/dVb as a function of the dimensionless bias voltage V (cid:48) b = Vb/Vmax (here Vmax = 75kBT /e) and the dimensionless magnetic field b(cid:48) = b/b2. The gate voltage is varied so that the ground states of the two charge sectors, (0, 2) and (1, 3 2 ), are kept aligned. The differential conductance is given in arbitrary units. b(cid:61)0b(cid:61)b2(cid:45)60(cid:45)40(cid:45)20204060Vb(cid:45)0.6(cid:45)0.4(cid:45)0.20.20.40.6Ia.ucorrelated phenomena such as the Kondo effect. These are currently out of the reach of fully atomistic calcula- tions, such as those of Refs. 20 and 21. Our model captures the essential phenomenology of a SMM, including the quantum tunneling of the net mag- netization. We have illustrated this point by showing a modulation of the non-linear I-V characteristics upon the application of an transverse magnetic field. The ap- pearance and disappearance of resonance lines is a clear indication of the existence of degeneracy points in the molecule's spectrum at certain values and directions of the transverse field. This behavior is similar to what is expected for a SMM in the giant spin approximation, where the destructive interference between tunneling tra- jectories of the giant spin create a periodic dependence on the transverse field. However, the lack of a well-defined topological phase in our model prevents us from making a direct connection between the modulation we observe and Berry phase interference. Our model does have some limitations. For instance, we restrict the orbitals that participate in the hopping terms of the molecule's Hamiltonian. In addition, not all configurations are included and the interactions with lig- ands is only taken into account in at a phenomenological level. The energy spectrum in our model is very sensitive to the choice of parameters and there is a complex interplay between the different interactions present in the model. In general, on-site anisotropies, Coulomb and overlap in- teractions intra and inter transition metals are not trivial to estimate. We have tried to use realistic or reasonable values whenever possible. These parameters depend on the electronic structure of the magnetic ions as well as on the geometrical configuration of the ligands surrounding the magnetic core. The bond angle between two magnetic ions is also of key importance in our calculations. For the particular three-ion model we studied, we chose a right bond angle to be 90 degrees and magnetic ions with par- allel anisotropies. This allowed us to neglect the sigma overlap between the dx2−y2 orbitals of the magnetic ions and the pz orbital of the bridging diamagnetic ion, as well as to prevent any contamination by d3z2−r2 orbitals. The result was an effective superexchange ferromagnetic interaction that competed with the Sz = 0 ground state favored by the local uniaxial anisotropies. Even with all 9 these constraints, we found that the model Hamiltonian yielded a high-spin, tunneling-split ground state, as it is typical for SMMs. A crucial parameter for our study is the on-site in-plane anisotropy e, which controls the splitting and competes with the in-plane magnetic field. Our calculations show that at very low temperatures certain transitions are suppressed when a transverse mag- netic field is tuned to a special direction and value that make the ground state twofold degenerate. This has a di- rect effect on non-equilibrium electronic transport across the molecule. At small enough bias voltages, the effect of the external field when tuned in-between the degen- eracy points is to modulate the access to excited states. This in turn shifts the peaks in the dI/dV response in an amount proportionally to the tunnel splitting of spin states. The fact that we have to rely on very low tempera- ture (in the mili-Kelvin range) to visualize these features indicates that they will be challenging to observe experi- mentally. Molecules with a large tunnel splitting are bet- ter suited for exploring the interplay between transport and quantum tunneling of the magnetization. This will require a relatively strong in-plane anisotropy as com- pared to the uniaxial anisotropy, which goes somewhat against the usual synthesis effort, which aims at increas- ing the uniaxial anisotropy. It is worth noting that recent advances in experimental setups indicate that small split- tings may be observable.39,40 We plan to further explore our microscopic formula- tion to consider more complex molecules with multi-ion transition-metal cores. We are currently extending our approach to investigate field modulation effects in the electronic transport of mononuclear SMMs, which consist of one rare-earth ion within a ligand cage. These systems have been synthesized with success in recent years (see for instance Refs. 41 -- 44). Acknowledgments This work was supported by NSF ECCS under Grant No. 1001755. The authors would like to thank Enrique del Barco, George Martins, and Edson Vernek for valu- able comments. ∗ Electronic address: [email protected] † Electronic address: [email protected] 1 G. Christou, D. Gatteschi, D. N. Hendrickson, and R. Ses- soli, MRS Bulletin 25, 66 (2000). 2 D. Gatteschi, R. Sessoli, and J. Villain, Molecular Nano- magnets (Oxford University Press, New York, 2007). 3 J. Friedman and M. P. Sarachik, Annu. Rev. Condens. Matter. Phys. 1, 109 (2010). 4 J. R. Friedman, M. P. Sarachik, J. Tejada, and R. Ziolo, Phys. Rev. Lett. 76, 3830 (1996). 5 L. Thomas, F. Lionti, R. Ballou, D. Gatteschi, R. Sessoli, and B. Barbara, Nature (London) 383, 145 (1996). 6 A. Ardavan et al., Phys. Rev. Lett 98, 057201 (2007). 7 W. Wernsdorfer and R. Sessoli, Science 284, 133 (1999). 8 H. M. Quddusi et al., Phys. Rev. Lett. 106, 227201 (2011). 9 S. Hill et al., Dalton Trans. 39, 4693 (2010). 10 J. B. Goodenough, Phys. Rev. 100, 564 (1955). 11 J. Kanamori, J. Phys. Chem. Solids 10, 87 (1959). 12 P. W. Anderson, Phys. Rev. 79, 350 (1950). 13 C. Romeike, M. R. Wegewijs, W. Hofstetter, and H. 10 Schoeller Phys. Rev. Lett. 96, 196601 (2006). 14 G. Gonzalez and M. N. Leuenberger, Phys. Rev. Lett. 98, 256804 (2007). 27 J. von Delft and C. L. Henley, Phys. Rev. Lett. 69, 3226 (1992); A. Garg, Europhys. Lett. 22, 205 (1993). 28 M. Imada, A. Fujimori, and Y. Tokura, Rev. Mod. Phys. 15 M. N. Leuenberger and E. R. Mucciolo, Phys. Rev. Lett. 70, 1039 (1998). 97, 126601 (2006). 16 C. Romeike, M. R. Wegewijs, and H. Schoeller, Phys. Rev. 29 B. H. Kim and B. I. Min, New J. Phys. 13, 073034 (2011). 30 V. Yu. Yushankhai and R. Hayn, Europhys. Lett. 47, 116 Lett. 96, 196805 (2006). (1999) 17 G. Gonzalez, M. N. Leuenberger, and E. R. Mucciolo, 31 I. A. Nekrasov, M. A. Korotin, and V. I. Anisimov, Phys. Rev. B 78, 054445 (2008). 18 K. Park, E.-C. Yang, and D. N. Hendrickson, J. Appl. Phys. 97, 10M522 (2005). 19 S. Barraza-Lopez, K. Park, V. Garcia-Suarez, and J. Fer- rer, Phys. Rev. Lett. 102, 246801 (2009). 20 K. Park, S. Barraza-Lopez, V. M. Garcia-Suarez, and J. Ferrer, Phys. Rev. B 81, 125447 (2010). 21 J. F. Nossa, M. F. Islam, C. M. Canali, and M. R. Peder- son, Phys. Rev. B 85, 085427 (2012). 22 F. Elste and C. Timm, Phys. Rev. B 71, 155403 (2005). 23 J. Lehmann and D. Loss, Phys. Rev. Lett. 98, 117203 (2007). 24 S. Herzog and M. R. Wegewijs, Nanotechnology 21, 274010 (2010). 25 F. Reckermann, M. Leijnse, and M. R. Wegewijs, Phys. Rev. B 79, 075313 (2009). preprint, arXiv:cond-mat/0009107. 32 A. Bose and R. Chatterjee, Proc. Phys. Soc. 82, 23 (1963). 33 Z.-H. Zhang et al., Mod. Phys. Lett. B 23, 1415 (2009). 34 F. Wang et al., J. Magn. Magn. Mat. 329, 88 (2013). 35 R. Basler et al., Inorg. Chem. 44, 649 (2005). 36 W. L. Feng et al., J. Magn. Magn. Mater. 323, 528 (2011) 37 A. S. Zyazin et al., Nanolett. 10, 3307 (2010). 38 C. W. J. Beenakker, Phys. Rev. B 44, 1646 (1991). 39 M. Urdampilleta, S. Klyatskaya, J.-P. Cleuziou, M. Ruben, and W. Wernsdorfer, Nature Mat. 10, 502 (2011). 40 E. Burzuri, A. S. Zyazin, A. Cornia, and H. S. J. van der Zant, Phys. Rev. Lett. 109, 147203 (2012). 41 M. A. Aldamen et al., J. Am. Chem. Soc. 130, 8874 (2008). 42 M. A. Aldamen et al., Inorg. Chem. 48, 3467(2009). 43 J. Titel et al., J. Am. Chem. Soc. 133, 15814 (2011). 44 M. Martinez-Perez et al., Phys. Rev. Lett. 108, 247213 26 C. Romeike, M. R. Wegewijs, M. Ruben, W. Wenzel, and (2012). H. Schoeller, Phys. Rev. B. 75, 064404 (2007).
1905.12193	2	1905	2019-05-30T05:25:29	Intrinsic resistance peaks in AB-stacked multilayer graphene with odd number of layers	[ "cond-mat.mes-hall", "cond-mat.mtrl-sci", "cond-mat.soft" ]	The intrinsic resistance peak (ridge) structures were recently found to appear in the carrier density dependence plot of the resistance of the AB-stacked multilayer graphene with even numbers of layers.The ridges are due to topological changes in the Fermi surface. Here, these structures were studied in AB-stacked multilayer graphene with odd numbers of layers (5 and 7 layers) by performing experiments using encapsulated high-quality graphene samples equipped with top and bottom gate electrodes.The intrinsic resistance peaks that appeared on maps plotted with respect to the carrier density and perpendicular electric field showed particular patterns depending on graphene's crystallographic structure, and were qualitatively different from those of graphene with even numbers of layers. Numerical calculations of the dispersion relation and semi-classical resistivity using information based on the Landau level structure determined by the magnetoresistance oscillations, revealed that the difference stemmed from the even-odd layer number effect on the electronic band structure.	cond-mat.mes-hall	cond-mat	Intrinsic resistance peaks in AB-stacked multilayer graphene with odd number of layers Tomoaki Nakasuga1, Taiki Hirahara1, Kota Horii1, Ryoya Ebisuoka1, Shingo Tajima1, Kenji Watanabe2, Takashi Taniguchi2 and Ryuta Yagi1 1Graduate School of Advanced Sciences of Matter (AdSM), Hiroshima University, Higashi-Hiroshima 739-8530, Japan, 2National Institute for Materials Sciences (NIMS), Tsukuba 305-0044, Japan. ABSTRACT The intrinsic resistance peak (ridge) structures were recently found to appear in the carrier density dependence plot of the resistance of the AB-stacked multilayer graphene with even numbers of layers. The ridges are due to topological changes in the Fermi surface. Here, these structures were studied in AB-stacked multilayer graphene with odd numbers of layers (5 and 7 layers) by performing experiments using encapsulated high-quality graphene samples equipped with top and bottom gate electrodes. The intrinsic resistance peaks that appeared on maps plotted with respect to the carrier density and perpendicular electric field showed particular patterns depending on graphene's crystallographic structure, and were qualitatively different from those of graphene with even numbers of layers. Numerical calculations of the dispersion relation and semi-classical resistivity using information based on the Landau level structure determined by the magnetoresistance oscillations, revealed that the difference stemmed from the even-odd layer-number effect on the electronic band structure. (Corresponding Author [email protected]) 1 Introduction Since the discovery of massless Dirac fermions in graphene, a number of scientific investigations [1-3] have tried to elucidate its physical property and potential applicability. The electronic properties of graphene depend strongly on the crystallographic structure. While monolayer graphene has a single band with a massless dispersion relation [4-6], bilayer graphene has with a massive dispersion relation [4,7- 10]. As the number of layers increases, many different stacking structures become possible, each of which is expected to have a particular band structure. In particular, AB- stacked graphene shows regularity in the evolution of its band structure. 2N (N: integer) layer graphene has N bilayer-like band(s), and 2N+1 layer graphene has N bilayer-like band(s) and a monolayer-like band, as shown in Fig. 1a [8,11-17]. Detailed band calculations have suggested that the band structure of multilayer graphene is much more complicated than those shown in Fig. 1 a. The band structure of graphene in the high-energy regime has been studied in optical spectroscopic experiments [18-24]. Moreover, while the low-energy band structure, which affects transport phenomena, has not been fully revealed by optical measurements, the advent of techniques for making high-quality graphene samples has made it possible to probe the low-energy band structure by using Shubnikov- de Haas oscillations [4,25-35]. Recently, high-quality multilayer graphene was found to show intrinsic resistance peaks in its carrier density dependence [31,33,36]. Detailed measurements using graphene samples with top and bottom gate electrodes have uncovered intrinsic resistance ridges structure specific to the band structure of AB-stacked 4-layer [31,33] and 6-layer graphene [36]. These intrinsic resistance ridges are considered to be a promising means of probing the band structure of two-dimensional materials. The ridges (or peaks) are related to the topological changes in the Fermi surface [31]; the complicated dispersion relations of multilayer graphene are tunable with a perpendicular electric field, and the resultant shape of the Fermi surface (energy contour of the dispersion relations) varies depending on the chemical potential. The ridges in AB-stacked graphene with even number of layers graphene are principally due to two reasons [33,36]. One is the nearly flat band structure created at the bottoms of the bilayer-like band by the perpendicular electric field; the field opens a band gap at the bottoms of these bands, and make them approximately flat. This results in a the heavy band mass [7,10,37-40]. Conspicuous split ridges appear on the map of resistivity as a function of carrier density and perpendicular electric flux density. The other reason is formation of mini-Dirac cones [41], which are created by the perpendicular electric field [31,33,36]. Trigonal warping locally closes the energy gap created by the perpendicular electric field, and thereby, mini-Dirac cones are created. Mini-Dirac points appear as sharp resistance ridges. In AB-stacked 4-layer graphene, they appear at the charge neutrality point [31,33], while and in AB-stacked 6- layer graphene, they appear not only at the charge neutrality point but also at non-zero carrier densities [36]. In this paper we will examine the intrinsic resistance peak structure of AB-stacked multilayer graphene with odd numbers of layers. We found that the resistance ridges are qualitatively different from graphene with even number of layers. We will show that this difference originates from the dispersion of the bilayer- like band which is hybridized with the monolayer-like band. The bottoms of the bilayer- like bands form particular structure that is qualitatively different from that of the even numbers of layers. 2 Experimental Our samples consisted of high-quality graphene flakes encapsulated with thin flakes of h-BN, and equipped with top and bottom gate electrodes, as shown schematically in Fig. 1 a and b. The graphene flakes were prepared by mechanical exfoliation of high-quality Kish graphite crystals. Thin h-BN flakes were also prepared using a similar method. A stack consisting of graphene layers and h-BN layers was formed by using the transfer technique described in Ref. [42] and Ref [25]. The graphene sample was formed on a Si substrate (covered with SiO2) which was heavily doped and remained conducting at low temperature. The substrate served as the bottom gate electrode. The top gate was formed by transferring graphene with a few layers onto the top of the encapsulated graphene. The resulting stack, consisting of the graphene and h-BN flakes, was patterned into a Hall bar by using reactive ion etching with a low pressure mixture of CF4 and O2 gas. Therefore, the top gate electrode and the effective sample area had an exact geometry (Fig. 1(b) and 1(c)). Electrical contact with graphene was attained by using the edge contact technique reported in Ref. [42]. Electrical contact with top gate electrode was carefully made at a point where the graphene sample to be measured was not underneath the h-BN, so as not to make a direct connection with the graphene to be measured. The number of layers and the stacking of the graphene were verified by various methods. We identified their effect on the characteristic Landau level structures [17], which are specific to a particular number of layers and stackings (see the Appendix). 3. Results and Discussion (1) AB-stacked 5-layer graphene AB-stacked 5-layer graphene is a typical example of odd layer multilayer graphene with the multiple bilayer like bands. It is expected to have two bilayer-like bands and a monolayer band [11-13,16-17,43-45]. Figure 1(d) shows the back gate voltage ( 𝑉𝑏 ) dependence of the resistivity for different top gate voltages (𝑉𝑡), which was measured at 𝑇 = 4.2 K. The mobility (𝜇) was calculated with the simple formula 𝜇 = 1/𝑛𝑡𝑜𝑡𝑒𝜌 using data for the 𝑉𝑏 dependence of resistivity (𝜌) with 𝑉𝑡 = 0. It was about 1.9 × 105 cm2/Vs in the electron regime, and 1.1 × 105 cm2/Vs in the hole regime at large carrier densities. Here, 𝑛𝑡𝑜𝑡 is the total carrier density carrier density. It is clear that varying the top gate changed the overall shape of the resistance traces with respect to 𝑉𝑏. The data with 𝑉𝑡 = 0 shows conspicuous double-peak structures whose peak resistivities are approximately the same. These structures would originate from the bottoms of the bilayer-like bands [31,33]. With increasing 𝑉𝑡𝑔, the shapes of the traces change into ones with a main peak and a small side peak. The resistivity of the main peak increases with 𝑉𝑡𝑔, until it saturates and then slightly decreases for large 𝑉𝑡𝑔. This behavior is reminiscent that of AB-stacked 4-layer [31,33] and 6-layer graphene [36] and is strikingly different from the behavior of bilayer [46-49] or AB-stacked trilayer graphene[49-53]. Bilayer graphene shows insulating behavior as the top gate voltage increases, while trilayer graphene shows the opposite behavior; the resistivity of the peaks appearing near the charge neutrality point decreases with increasing top gate voltage. To investigate the above mentioned properties further, we measured the top and bottom gate voltage dependence of the resistivity in detail. The results are summarized in the map of resistivity with respect to 𝑉𝑡 and 𝑉𝑏, as shown in Fig 2(a). Resistivity peaks appear as ridges. Salient resistance ridges on a linear line from the upper left to lower right satisfy the condition of charge neutrality, which corresponds to the large peaks in Fig 2(d) for 𝑉𝑡 > 0. In addition, side peaks which are parabolic in shape are discernible in the figure. Because the peak structure would result from a variation in the dispersion relation arising from the perpendicular electric field as in AB-stacked 4- and 6-layer graphene [31,33,36], we replotted the map as a function of total carrier density (𝑛𝑡𝑜𝑡) and electric flux density (𝐷⊥) perpendicular to the graphene. The total carrier density can be calculated by summing the carrier densities induced by the top and bottom gate voltages as 𝑛𝑡𝑜𝑡 = (𝐶𝑡(𝑉𝑡 − 𝑉𝑡0) + 𝐶𝑏(𝑉𝑏 − 𝑉𝑏0))/(𝑒). (1) Here, 𝐶𝑡 and 𝐶𝑏 are the specific capacitances of the top and bottom gate electrodes, respectively. 𝑉𝑡0 and 𝑉𝑏0 represent the shift in gate voltage due to carrier doping associated with the top and bottom gate electrodes. The effect of the perpendicular electric field can be estimated using electric flux density induced by the top and bottom gate voltages, which is given by 𝐷⊥ = (𝐶𝑡(𝑉𝑡 − 𝑉𝑡0) -- 𝐶𝑏(𝑉𝑏 − 𝑉𝑏0)) / 2. (2) From the charge neutrality condition in Fig. 2(a), one can estimate the ratio of the capacitances (= 𝑉𝑡/𝑉𝑏) to be about 3.8. The specific capacitances were calculated from the Landau level structure measured under the condition, 𝐷⊥=0, to be 𝐶𝑡 = 395 𝑎F/μm2 and 𝐶𝑏 = 104 𝑎F/μm2. Figures 2(b) and 2(c) show maps of 𝜌 and 𝑑𝜌/𝑑𝑛𝑡𝑜𝑡 as a function of 𝑛𝑡𝑜𝑡 and 𝐷⊥. In these figures, parabolic ridges are discernible for both the electron and hole regimes. Similar but more complicated parabolic ridge structures were also observed in AB-stacked 4 [31,33] and 6-layer graphene [34,36]. The dispersion relations in the absence and presence of perpendicular electric fields were numerically calculated to examine the relation between the band structure and the resistance ridge structure in the 5-layer graphene. The calculation was based on the effective mass approximation and the Slonczewski-Weiss-McClure (SWMcC) parameters [54-56] of graphite were used. Screening of induced carriers was taken into account by using the distribution of induced carriers in graphene. We assumed that carriers carrier in each layer decay exponentially with a decay length of λ, which is roughly consistent with the results of the Thomas-Fermi approximation [57]. In the calculation we took λ = 0.45 nm, approximately the same value expected from a self-consistent calculation of the screening length [58], and approximately the same as the experimental value obtained from Landau-level structures in multilayer graphene [17]. Figure 3 (a) shows the dispersion relations of the AB-stacked 5-layer graphene numerically calculated for different values of 𝐷⊥. The dispersion relations for AB-stacked 4 layer graphene are shown in Fig. 3(b) for comparison. For the 5-layer graphene, there are two sets of bilayer- like band and a monolayer-like band, which are complicatedly hybridized near 𝐸 = 0 [35] (more complicated than what is shown in Fig. 1(a)). Applying a perpendicular electric field opens energy gaps., i.e., differences in energy between the bottoms of the bands. The gaps increase with increasing 𝐷⊥; thereby, the dispersion relations look rather simplified. For convenience, we labeled the band as αe―γh, and the bottoms of the bands a, b, c, b', and a'. Bands γe and γh in the 5-layer graphene are monolayer- like bands. The remaining bands are principally bilayer-like bands, as in the AB-stacked 4-layer graphene, but they differ significantly between the 4- and 5- layer cases. In particular, the energy gap between 𝛼𝑒 and 𝛽𝑒 and the one between 𝛼ℎ and 𝛽ℎ are significantly larger in the 4-layer graphene than in the 5-layer graphene. It can be seen that the structures of band 𝛼𝑒 and 𝛼ℎ in the 5-layer graphene are more complicated than those in the 4-layer graphene; this difference would originate from the hybridization with the monolayer-like band in the 5-layer graphene. The difference in the band structure results in particular resistance ridge structures. The characteristic band positions are closely related to the resistance ridges. We have calculated the semi-classical resistivity based on the Boltzmann equation with the constant relaxation-time approximation. (The calculation is similar to the one performed on AB-stacked 4-layer graphene [31]). We took into account possible energy broadening due to scattering. Figure 4(a) compares the experimental and calculated maps of 𝑑𝜌/𝑑𝑛𝑡𝑜𝑡 plotted as a function of 𝑛𝑡𝑜𝑡 and 𝐷⊥ . It can be seen that the calculations approximately reproduced the experimental result. Conspicuous ridge structures are labeled with the characteristic positions of the band structure in Fig 3(a). Ridge c stems from the mini-Dirac cones formed at the charge neutrality point. Ridges b and and b' are for the bottoms of the bilayer-like bands 𝛽𝑒 and 𝛽ℎ. As for positions a and a', which correspond to the bottoms of the monolayer-like bands, structures hardly appeared in the experimental results, possibly because the variation in the conductivity was rather smaller than at the other characteristic positions in the bands. As shown in Fig. 4(b), the structures are barely visible in the simulation with reduced energy broadening. Now we let us discuss the differences between the 5-layer and 4-layer cases. The resistance ridges of the 5-layer graphene, which are parabolic in shape, are qualitatively different from those of the 4-layer graphene. The ridges in the 4-layer case show clear splitting with increasing 𝐷⊥ [31,33]. This is due to formation of an energy gap between the bilayer like bands, as shown in Fig. 3(b). On the other hand, in the 5-layer graphene, the bottom of 𝛽𝑒 almost touches αe, and the bottom of 𝛽ℎ has approximately the same energy as the local bottom of 𝛼ℎ. This qualitatively different band structure results in the5-layer graphene not having any split ridge structures for the bilayer-like bands. Although the 4-layer and 5-layer graphene have significantly different electronic band structures, they showed have similar resistance ridges that appear at 𝑛𝑡𝑜𝑡 = 0 for 𝐷⊥ above ∼ 0.5 × 10−7cm−2As. In both cases, this is because the ridge originates from the formation of mini-Dirac cones [31,36,41] near 𝐸 = 0 for large 𝐷⊥, as can be seen in Fig. 3(a) and 3(b). In the 5-layer case, three sets of mini-Dirac cones are created at different wave numbers in k-space. Among them, the two located at 𝑘𝑥 ≠ 0 (see Figure 3(a)) arise from the bilayer-like band because of trigonal warping. They are both three- fold degenerate in a valley (K or K'). The other set of mini-Dirac cones, which are located at 𝑘𝑥 = 0 , apparently originate from the monolayer-like band. The mini-Dirac cone structure in the 4-layer case is strikingly different. There are large mini-Dirac cones (in positive 𝑘𝑥) and a small mini-Dirac cone-like structures (in negative 𝑘𝑥) which have gaps. The cones and the small cone-like structure are both three-fold degenerate in the K and K' valley. In the both the 4- and 5-layer cases, perpendicular electric field resulted complicated massive bands changing into linear bands near the charge neutrality point, and thereby, the resistance ridges near the 𝑛𝑡𝑜𝑡 = 0 appeared. The simulation with reduced energy broadening reveals the resistance ridges associated with the monolayer band for the bottoms of the monolayer bands 𝛾𝑒 and 𝛾ℎ (Fig. 4(b)). However, they are hardly visible in the experimental data. In sufficiently large perpendicular electric fields, energy gaps are created for the monolayer-like band because of hybridization with bilayer-like bands (Fig. 3(a)). Although the monolayer-like band has a non-zero band mass near the bottoms of the bands, the mass is much smaller than those for the bottoms of the bilayer-like bands. This would make it hard the resistance ridges for the bottoms of 𝛾𝑒 and 𝛾ℎ. (2) AB-stacked 7-layer graphene AB-stacked 7-layer graphene, which has three sets of bilayer-like bands and a monolayer-like band, also shows characteristic resistance ridges for odd numbers of layers. We studied the intrinsic resistance peaks of the 7-layer sample that had a similar structure to that of the 5-layer sample. The mobility at a large carrier density was 𝜇 = 6.9 × 104 cm2/Vs in the electron regime and 5.0 × 104 cm2/Vs in the hole regime. Figure 5(a) shows a map of resistivity as a function of 𝑉𝑏 and 𝑉𝑡, which was measured at 𝑇 = 4.2 K. The ratio of the specific capacitance was estimated to be 𝐶𝑡/𝐶𝑏 = 3.98. 𝐶𝑡 = 446 𝑎F/μm2 and 𝐶𝑏 = 112 𝑎F/μm2 . Figure 5 (b) is a replot as a function of 𝑛𝑡𝑜𝑡 and 𝐷⊥. The resistance ridges are distinct from those of the 4-layer [31,34], 5- layer and the 6-layer cases [33,36]. Although the 6-and the 7-layer graphene have more complicated band structures than those of 4- and 5-layer graphene, they show characteristic differences in the band structure reflecting the even-odd layer number effect. Figure 6(a) shows the numerically calculated dispersion relation of the 7-layer graphene for some values of 𝐷⊥, while Fig. 6 (b) shows those for the 6-layer case for comparison. The SWMcC parameters of graphite, and λ = 0.45 nm were used in the calculation. Bands are labeled 𝛼𝑒, 𝛽𝑒, 𝛾𝑒, 𝛿𝑒, 𝛼ℎ, 𝛽ℎ, 𝛾ℎ, and 𝛿ℎ . Characteristic points in the band diagram are labeled a-d and a'-c'. The dispersions for the 7-layer graphene under a perpendicular electric field are much more complicated than those of the 6-layer graphene because of hybridization of the bilayer- like bands with a monolayer-like band, as was seen earlier for the cases of the 4- and 5- layer graphene. In the 6-layer graphene, the application of a perpendicular electric field opens energy gaps between the bilayer-like bands, and the dispersion relations are nearly flat near the bottoms of each band. On the other hand, no such flat dispersion relations form in the 7-layer graphene. The bottoms 𝛾𝑒 and 𝛽𝑒 (𝛾ℎ and 𝛽ℎ) nearly make contact with the small energy gaps. The structures apparently originate from hybridization with the monolayer-like band. On the other hand, for large E, one can see that bands δe and 𝛿ℎ, which originate from the monolayer-like band, have rather simple shapes. To see the correspondence of the intrinsic resistance ridges to the dispersion relations, we compared the experimental results with the numerically calculated resistivities (Fig. 7).It is clear that the theoretical results approximately explain the experimental results. Resistance ridges appear at the corresponding positions in the band structures. First, let us examine the ridges appearing in the vicinity of 𝑛𝑡𝑜𝑡 = 0. One can recognize the resistance ridge near the charge neutrality condition as in the 4-layer, 5-layer, and 6- layer graphene. Comparing the experimental results with those of the band calculation, it can be seen that mini-Dirac cones are created at points d in the vicinity of the charge neutrality point for large 𝐷⊥. In the AB-stacked 7-layer graphene, the dispersion relations at 𝐷⊥ = 0 show a semi-metallic band structure; the electron and hole bands overlap near 𝐸 = 0. Applying a perpendicular electric field created mini-Dirac cones, from which conspicuous ridges formed. Next, we turn to the other resistance ridges. The bottoms of the bilayer bands b and b' (Fig. 6 (a)) appear as resistance ridges in Fig. 7. Apparently, there are no split ridge structures arising from the bottoms of bilayer-like bands, as in the 5-layer case. Unlike the 5-layer case, conspicuous arising from mini-Dirac cones (indicated by c and c' in Fig. 6(a)) are visible as in the 6-layer case [36], at carrier densities different from charge neutrality. In addition, the experimental results do have clear ridge structures for the monolayer-like band a and a', as in the 5-layer case; the lack should again be due to relatively small carrier density and light band mass. ４ Discussion The even-odd layer-number effect in the band structure is an intrinsic feature of AB- stacked multilayer graphene. This feature can be seen in the Landau level structures: the absence or presence of the Landau levels due to the monolayer-like band [11,13,17,25,28,30-32-35,45,59-60] and the absence or presence of valley splitting at zero perpendicular electric field [25,35]. As for the dispersion relation at zero magnetic field, the low-energy band structures are expected to be rather complicated, and information can be extracted from the Landau level structure indirectly through the band parameters with which the dispersion relation can be calculated. As described in the previous section, the intrinsic resistance ridge structure reflects the specific band structure of graphene, and this allows us to probe the band structure directly in the transport experiments. Here, we address the evolution of the resistance ridge structure in AB-stacked multilayer graphene with increasing layer number. The numerically calculated resistance ridge structures for 4 to 7 layers are summarized in Fig. 8. (The calculation for the 4-layer case is reported in Ref. [31].) It is clear that the resistance ridges (peaks) appear at different positions in the diagram: the ridge structures have a specific pattern depending on the number of layers. One can thus determine the number of layers and stacking by using the diagram. Resistance ridges due to bilayer bands show splitting in graphene with even numbers of layers, while the splitting is absent from graphene with odd numbers of layers. In addition, the resistance ridge due to the monolayer-like band in the graphene with the odd numbers of layers are rather small. On the other hand, the mini-Dirac points form relatively strong peaks compared with the bottoms of the bands. For example, ridge structures at 𝑛𝑡𝑜𝑡 = 0 appear regardless of the number of layers. The 6- and 7-layer graphene show relatively strong peaks at the mini-Dirac points (MDP) at non-zero carrier densities. On the ridges formed at 𝑛𝑡𝑜𝑡 = 0, the resistivity tends to increase with increasing 𝐷⊥, but it saturates (and slightly decreases in some cases) at large 𝐷⊥. The early graphene research reported that bi- and trilayer graphene had different responses to a perpendicular electric field: bilayer graphene becomes insulating because the energy gap opens [47-48,61], while trilayer graphene becomes more metallic [50-52,61] (i.e. its resistivity decreases). However, this sort of behavior does not persist in graphene consisting of more layers, as shown in previous work [31,33,36] and this study. The behavior is consistent with the formation of mini-Dirac cones in the vicinity of the charge neutrality point. AB-stacked 5-7 layer graphenes (and possibly the 4 layer graphene) are semi-metallic near the charge neutrality point in the absence of a perpendicular electric field. Electrons and holes are compensated, so that there would be considerably many number of carriers that contribute to the conductance. The formation of mini-Dirac cones tends to decrease the number of carriers. The absence of insulating behavior can be understood from the minimum conductivity of monolayer graphene at the charge neutrality point. Theory predicts a minimum conductivity of about e2/ℏ at the Dirac point [4,6,62]. Although a Dirac point is difficult to realize in an actual experiment because of inhomogeneity [63-66], may experiments have shown that there is a minimum conductivity, whose value is not universal. Summary and concluding remarks Intrinsic resistance ridge structures of AB-stacked 5- and 7- layer graphene, which appear as a function of carrier density and perpendicular electric field, were studied together with the band structure by using an encapsulated graphene device equipped with top and bottom gate electrodes. We found that the intrinsic resistance peaks (ridges) in multilayer graphene with an odd numbers of layer are strikingly different from the graphene with in an even number of layers: only graphene with an even number of layers show split ridges due to the formation of nearly flat bands. This difference results from hybridization of the bilayer-like band with the monolayer-like band in the graphene with odd number of layers. Thus, these results show that the resistance ridges can be used to probe the electronic band structure of two-dimensional materials. Appendix A Determination of number of layers and stacking The number of layers and their stacking were determined by combined use of atomic force microscopy (AFM) and Raman spectroscopy. In particular, the number of layers and stacking were determined after calibrating the relation between the Raman spectral shape and the number of layers of graphene determined by AFM. The spectral shape of the ABA stacking showed a systematic evolution [17,67-70], that was considerably different from that of ABC stacking [67,69-72]. The details are described in Ref. [17]. We also used the Landau level structures which can be deduced from the Shubnikov-de Haas oscillations in the low-temperature magnetoresistance. The Landau level structures reflects the electronic band structure of graphene directly, meaning that it is one of the most reliable methods to determine the number of layers and stacking. A map of magnetoresistance with respect to the carrier density and magnetic field (Landau fan diagram) reveals graphene's detailed low-energy band structure that is specific to the number of layers and stacking. The number of layers and stacking of the measured samples were verified by referring a list of fan diagrams for AB-stacked graphene with known numbers of layers [17]. B Landau level structure in AB-stacked 5-layer graphene The AB-stacked 5-layer graphene sample showed Shubnikov-de Haas oscillations in the magnetoresistance which was measured at 𝑇 = 4.2 K. Figures 9(a) and 9(b) show maps of the longitudinal resistivity (𝜌𝑥𝑥) and its derivative with respect to the magnetic field (𝑑𝜌𝑥𝑥/𝑑𝐵), plotted as a function of magnetic field B and carrier density 𝑛𝑡𝑜𝑡. Here, 𝑛𝑡𝑜𝑡 was varied by controlling the top and bottom gate voltages so as to satisfy the condition, 𝐷⊥ = 0 . The stripes are Landau levels for particular bands with particular Landau indices. The observed Landau level structure near the charge neutrality point is approximately the same as that in the previous report for AB-stacked 5-layer graphene, which was measured from a sample with a single gate electrode [17]. This confirms that our sample was identified AB-stacked 5-layer graphene, because Landau level structure is the fingerprint of the electronic band structure of graphene. The overall structure of the Landau levels can be approximately explained by a numerical calculation based on the effective mass approximation. Figure 9(c) shows energy eigenvalues calculated for the Slonczewski-Weiss-McClure parameters which are approximately the same as those of graphite, and Figure 9(d) is the calculated density of states. Although refining the SWMcC parameters would give a better fitting to the experiment, energy gaps with ν = −2, 14, 18 are clearly visible in the experimental data. The filling factor for the gaps satisfies the relation 4(N+1/2) with integer N, as in the mono-layer graphene. In addition, the Landau levels for the monolayer-like band are visible in Fig 8(b) (indicated by the red bars). The energy gap with ν = −2 characterizes the AB-stacked 5-layer graphene. It occurs near the charge neutrality point above a few Tesla and appears between the zero-mode Landau levels; no Landau level crossings occur for the larger magnetic field. Similar characteristic energy gap structures appears in AB-stacked multilayer graphene with more layers, and one can identify the number of layers by using the filling factor of the gap. The gap occurs at 𝜈 = 0 in the case of AB-stacked graphene [31,33,35], while it appears at 𝜈 = 4 in AB-stacked 6 layer [36]. To be shown later, in the 7 layer, it appears at 𝜈 = 6. C Landau level structure in AB-stacked 7-layer graphene Figures 10(a) and 10(b) show maps of 𝑅𝑥𝑥 and 𝑑𝑅𝑥𝑥/𝑑𝐵 as a function of 𝑛𝑡𝑜𝑡 and B. Highly complicated beatings of the Shubnikov-de Haas oscillations can be seen. The energy gaps and the Landau level crossing near the charge neutrality point approximately reproduce the fan diagram measured for single-gated graphene samples [17], which confirms that our sample is the AB-stacked 7-layer graphene. Conspicuous energy gaps appear at ν = −6. Figure 10(c) shows the numerically calculated Landau level spectra for the SWMcC parameters of graphite, while Figure 10(d) is a map of the corresponding density of states. The calculation approximately accounts for the overall Landau level structures and positions of the conspicuous energy gaps. In particular, the energy gap at 𝜈 = −6 is visible between the zero-mode Landau levels of the bilayer-like band. D Calculation of dispersion relation and Landau levels The dispersion relations at zero magnetic field were calculated using the Hamiltonian for the effective mass approximation which is based on the tight-binding model [5,35,43,45]. Landau levels were numerically calculated by expanding the wave functions with Landau functions [14,35,59,74-75] and evaluating the eigenvalues of the Hamiltonian. The density of states was calculated by assuming that each Landau level had a carrier density of degeneracy multiple 𝑒𝐵/ℎ [35]. The electrostatic potential due to the perpendicular electric field was calculated by taking the screening of each layer into account. Multilayer graphene is atomically thin, as are other two-dimensional materials, so that an externally applied perpendicular electric field is expected to penetrate the graphene but to be shielded layer by layer [17,23,57-58,76-81]. The internal electric field significantly changes the electrostatic potential for each layer in the graphene and affects the band structure [17,58]. Here, we used the same method as in Refs. [17], where it was assumed that the external electric field diminishes exponentially with the screening length λ, which is a fitting parameter to be experimentally determined. We estimated it to be about 0.43 in our previous work on the Landau level structure in AB-stacked multilayer graphene in which we measured samples with a single gate electrode [17,36]. The resistance ridges observed in the present experiment were best explained for λ ∼ 0.45 nm. Here, we assumed the dielectric constant in the graphene to be ε /ε0 = 2.0. E Calculation of conductivity at zero magnetic field. The Drude conductivity was calculated by using the numerically calculated dispersion relations. The resistivity was then determined by taking the reciprocal of the conductivity. A constant relaxation time was assumed. For a small electric field 𝐸𝑥 applied in the 𝑥 -direction, the solution of the Boltzmann equations is simply approximated at low temperature by shifting the wave number (k ) of all the existing electrons by −𝑒𝐸𝑥𝜏 . Thus, the conductivity is proportional to the sum of the group velocities for all of the filled electronic states. To make a comparison with the experiment, we took energy broadening of the distribution function into account; this would possibly arise for various reasons, e.g, scattering, inhomogeneity, etc. We assumed that the derivative of the distribution function with respect to energy is simply a Gauss function with a standard deviation, 𝛤/√2. The details of the distribution function would not change the important feature of the simulation. Acknowledgements This work was supported by KAHENHI No. 25107003 from MEXT Japan. References [1] A. K. Geim, and K. S. Novoselov, Nat. Mater 6, 183 (2007). [2] S. Das Sarma, S. Adam, E. H. Hwang, and E. Rossi, Rev. Mod. Phys. 83, 407 (2011). [3] A. H. Castro Neto, F. Guinea, N. M. R. Peres, K. S. Novoselov, and A. K. Geim, Rev. Mod. Phys. 81, 109 (2009). [4] 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). [5] P. R. Wallace, Phys. Rev. 71, 622 (1947). [6] N. H. Shon, and T. Ando, J. Phys. Soc. Jpn. 67, 2421 (1998). [7] E. McCann, D. S. L. Abergel, and V. I. Fal'ko, Solid State Commun. 143, 110 (2007). [8] B. Partoens, and F.M. Peeters, Phys. Rev. B 75, 193402 (2007). [9] M. Koshino, and T. Ando, Phys. Rev. B 73, 245403 (2006). [10] E. McCann, Phys. Rev. B 74, 161403(R) (2006). [11] M. Koshino, and T. Ando, Phys. Rev. B 76, 085425 (2007). [12] M. Koshino, and T. Ando, Solid State Commun. 149, 1123 (2009). [13] M. Nakamura, and L. Hirasawa, Phys. Rev. B 77, 045429 (2008). [14] H. K. Min, and A. H. MacDonald, Phys. Rev. B 77, 155416 (2008). [15] M. Koshino, and T. Ando, Solid State Commun. 149, 1123 (2009). [16] S. Latil, and L. Henrard, Phys. Rev. Lett. 97, 036803 (2006). [17] R. Yagi, T. Hirahara, R. Ebisuoka, T. Nakasuga, S. Tajima, K. Wtatanabe and T. Taniguchi, Sci. Rep. 8, 13018 (2018). [18] M. Sprinkle, D. Siegel, Y. Hu, J. Hicks, A. Tejeda, A. Taleb-Ibrahimi, P. LeFevre, F. Bertran, S. Vizzini, H. Enriquez, S. Chiang, P. Soukiassian, C. Berger, W. A. deHeer, A. Lanzara, and E. H. Conrad, Phys. Rev. Lett. 103, 226803 (2009). [19] A. B. Kuzmenko, I. Crassee, D. van Der Marel, P. Blake, and K. S. Novoselov, Phys. Rev. Lett. Phys. Rev. B, 80 (165406). [20] K. F. Mak, C. H. Lui, J. Shan, and T. F. Heinz, Phys. Rev. Lett. 102, 256405 (2009). [21] L. M. Zhang, Z. Q. Li, D. N. Basov, M. M. Fogler, Z. Hao, and M. C. Martin, Phys. Rev. B 78, 235408 (2008). [22] T. Ohta, A. Bostwick, T. Seyller, K. Horn, and E. Rotenberg, Science 313, 951 (2006). [23] T. Ohta, A. Bostwick, J. L. McChesney, T. Seyller, K. Horn, and E. Rotenberg, Phys. Rev. Lett. 98, 206802 (2007). [24] C. Coletti, S. Forti, A. Principi, K. V. Emtsev, A. A. Zakharov, K. M. Daniels, B. K. Daas, M. V. S. Chandrashekhar, T. Ouisse, and D. Chaussende, A. H. MacDonald, M. Polini, U. Starke, Phys. Rev. B 88, 155439 (2013). [25] T. Taychatanapat, K. Watanabe, T. Taniguchi, and P. Jarillo-Herrero, Nat. Phys. 7, 621 (2011). [26] L. C. Campos, T. Taychatanapat, M. Serbyn, K. Surakitbovorn, K. Watanabe, T. Taniguchi, D. A. Abanin, and P. Jarillo-Herrero, Phys. Rev. Lett. 117, 066601 (2016). [27] Y. Asakawa, S. Masubuchi, N. Inoue, S. Morikawa, K. Watanabe, T. Taniguchi, and T. Machida, Phys. Rev. Lett. 119, 186802 (2017). [28] A. Kumar, W. Escoffier, J. M. Poumirol, C. Faugeras, D. P. Arovas, M. M. Fogler, F. Guinea, S. Roche, M. Goiran, and B. Raquet, Phys. Rev. Lett. 107, 126806 (2011). [29] W. Z. Bao, Z. Zhao, H. Zhang, G. Liu, P. Kratz, L. Jing, J. Velasco, D. Smirnov, and C. N. Lau, Phys. Rev. Lett. 105, 246601 (2010). [30] C. H. Ho, S. J. Tsai, R. B. Chen, Y. H. Chiu, and M. F. Lin, Science 11, 4938 (2011). [31] Y. Shi, S. Che, K. Zhou, S. Ge, Z. Pi, T. Espiritu, T. Taniguchi, K. Watanabe, Y. Barlas, R. Lake and C. N. Lau, Phys. Rev. Lett. 120, 096802 (2018). [32] Z. Wu, Y. Han, J. Lin, W. Zhu, M. He, S. Xu, X. Chen, H. Lu, W. Ye, T. Han, Y. Wu, G. Long, J. Shen, R. Huang, L. Wang, Y. He, Y. Cai, R. Lortz, D. Su, N. Wang, Phys. Rev. B 92, 075408 (2015). [33] T. Hirahara, R. Ebisuoka, T. Oka, T. Nakasuga, S. Tajima, K. Watanabe, T. Taniguchi, and R. Yagi, Sci. Rep. 8, 13992 (2018). [34] T. Hirahara, R. Ebisuoka, K. Watanabe, T. Taniguchi, and R. Yagi, J. of Phys. Conf. Ser. 969, 012150 (2018). [35] M. Koshino, and E. McCann, Phys. Rev. B 83, 165443 (2011). [36] T. Nakasuga, S. Tajima, T. Hirahara, R. Ebisuoka, T. Oka, K. Watanabe, T. Taniguchi, and R. Yagi, Phys. Rev. B 99, 085404 (2019). [37] A. A. Avetisyan, B. Partoens, and F. M. Peeters, Phys. Rev. B 80, 195401 (2009). [38] A. A. Avetisyan, B. Partoens, and F. M. Peeters, Phys. Rev. B 79, 035421 (2009). [39] E. V. Castro, K. S. Novoselov, S. V. Morozov, N. M. R. Peres, J. M. B. Lopes dos Santos, J. Nilsson, F. Guinea, A. K. Geim, and A. H. Castro Neto, Phys. Rev. Lett. 99, 216802 (2007). [40] Y. B. Zhang, T. T. Tang, C. Girit, Z. Hao, M. C. Martin, A. Zettl, M. F. Crommie, Y. R. Shen, and F. Wang, Nature 459, 820 (2009). [41] T. Morimoto, and M. Koshino, Phys. Rev. B 87, 085424 (2013). [42] L. Wang, I. Meric, P. Y. Huang, Q. Gao, Y. Gao, H. Tran, T. Taniguchi, K. Watanabe, L. M. Campos, and D. A. Muller, Science 342, 614 (2013). [43] B. Partoens, and F. M. Peeters, Phys. Rev. B 74, 075404 (2006). [44] F. Guinea, A. H. Castro, and N. M. R. Peres, Solid State Commun. 143, 116 (2007). [45] B. Partoens, and F.M. Peeters, Phys. Rev. B 75, 193402 (2007). [46] T. Taychatanapat, and P. Jarillo-Herrero, Phys. Rev. Lett. 105, 166601 (2010). [47] J. Yan, and M.S. Fuhrer, Nano Lett. 10, 4521 (2010). [48] H. Miyazaki, K. Tsukagoshi, A. Kanda, M. Otani, and S. Okada, Nano Lett. 10, 3888 (2010). [49] W.J.Zhu, D. Neumayer, V. Perebeinos, and P. Avouris, Nano Lett. 10, 3572 (2010). [50] M. F. Craciun, S. Russo, M. Yamamoto, J. B. Oostinga, A. F. Morpurgo, and S. Tarucha, Nat. Nanotechnol. 4, 383 (2009). [51] 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, Phys. Rev. B 84, 161408(R) (2011). [52] K. Zou, F. Zhang, C. Capp, A. H. MacDonald, and J. Zhu, Nano Lett. 13, 369 (2013). [53] A. A. Avetisyan, B. Partoens, and F. M. Peeters, Phys. Rev. B 81, 115432 (2010). [54] J. C. Slonczewski, and P. R. Weiss, Phys. Rev. 109, 272 (1958). [55] J. W. McClure, Phys. Rev. 108, 612 (1957). [56] M. S. Dresselhaus, and G. Dresselhaus, Adv. Phys. 30, 139 (1981). [57] F. Guinea, A. H. Castro Neto, and N. M. R. Peres, Phys. Rev. B 73, 245426 (2006). [58] M. Koshino, Phys. Rev. B 81, 125304 (2010). [59] S. J. Yuan, R. Roldan, and M. I. Katsnelson, Phys. Rev. B 84, 125455 (2011). [60] 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, Nat. Phys. 7, 948 (2011). [61] M. Koshino, and E. McCann, Phys. Rev. B 79, 125443 (2009). [62] Y.S.Zheng, and T. Ando, Phys. Rev. B 65, 245420 (2002). [63] S. Adam, E. H. Hwang, V. M. Galitski, and S. Das Sarma, Science 104, 18392 (2007). [64] Y.W. Tan, Y. Zhang, K. Bolotin, Y. Zhao, S. Adam, E.H. Hwang, S. Das Sarma, H.L. Stormer, and P. Kim, Phys. Rev. Lett. 99, 246803 (2007). [65] S. Das Sarma, S. Adam, E. H. Hwang, and E. Rossi, Rev. Mod. Phys. 83, 407 (2011). [66] R. Yagi, S. Fukada, H. Kobara, N. Ogita, and M. Udagawa, Physica 42, 673 (2010). [67] L. M. Malard, M. A. Pimenta, G. Dresselhaus, and M. S. Dresselhaus, Phys. Rep. 473, 51 (2009). [68] A. C. Ferrari, J. C. Meyer, V. Scardaci, C. Casiraghi, M. Lazzeri, F. Mauri, S. Piscanec, D. Jiang, K. S. Novoselov, and S. Roth, A. K. Geim, Phys. Rev. Lett. 97, 187401 (2006). [69] C. H. Lui, Z. Q. Li, Z. Y. Chen, P. V. Klimov, L. E. Brus, and T. F. Heinz, Nano Lett. 11, 164 (2011). [70] T. A. Nguyen, J. U. Lee, D. Yoon, and H. Cheong, Sci. Rep. 4, 4630 (2014). [71] C. X. Cong, T. Yu, K. Sato, J. Z. Shang, R. Saito, G. F. Dresselhaus, and M. S. Dresselhaus, ACS NANO 5, 8760-8768 (2011). [72] Y. F. Hao, Y. Y. Wang, L. Wang, Z. H. Ni, Z. Q. Wang, R. Wang, C. K. Koo, Z. X. Shen, and J. T. L. Thong, SMALL 6, 195 (2010). [73] D. Graf, F. Molitor, K. Ensslin, C. Stampfer, A. Jungen, C. Hierold, and L. Wirtz, Nano Lett. 7, 238 (2007). [74] E. McCann, and V. I. Fal'ko, Phys. Rev. Lett. 96, 086805 (2006). [75] M. Koshino, and E. McCann, Phys. Rev. B 87, 045420 (2013). [76] P. B. Visscher, and L. M. Falicov, Phys. Rev. B 3, 2541 (1971). [77] R. van Gelderen, R. Olsen, and C. M. Smith, Phys. Rev. B 88, 115414 (2013). [78] L. Pietronero, S. Strassler, H. R. Zeller, and M. J. Rice, Phys. Rev. Lett. 41, 763 (1978). [79] D. P. DiVincenzo, and E. J. Mele, Phys. Rev. B 29, 1685 (1984). [80] N. J. Lee, J. W. Yoo, Y. J. Choi, C. J. Kang, D. Y. Jeon, D. C. Kim, S. Seo, and H. J. Chung, Appl. Phys. Lett. 95, 222107 (2009). [81] H. Miyazaki, S. Odaka, T. Sato, S. Tanaka, H. Goto, A. Kanda, K. Tsukagoshi, Y. Ootuka, and Y. Aoyagi, Appl. Phys. Express, 1, 034007 (2008). Figure captions Fig. 1 (a) Simplified dispersion relations forAB-stacked multilayer graphene. Graphene with an odd number of layers consists of bilayer band(s) and a monolayer band. Graphene with an even number of layers consists of bilayer band(s). (b)Optical micrograph of encapsulated graphene sample with top and bottom gate electrodes (top). The bar is 10 μm. (c) Illustration of vertical structure of encapsulated graphene in the effective sample area. G means graphene. TG means the top gate electrode, and BG means the conducting Si substrate. (d) Back gate voltage (𝑉𝑏) dependence of resistivity of AB-stacked 5-layer graphene sample for different top gate voltages (𝑉𝑡). 𝑉𝑡 was varied between -10 and 10 V in 2 V steps. Fig. 2 Top and bottom gate voltage dependence of resistivity in AB-stacked 5-layer graphene. (a) Map of resistivity as a function of 𝑉𝑡 and 𝑉𝑏. 𝑇 = 4.2 K. 𝐵 = 0 T. (b) Map of resistivity as a function of 𝑛𝑡𝑜𝑡 and 𝐷⊥. (c) Similar map for 𝑑𝜌/𝑑𝑛𝑡𝑜𝑡. Fig. 3 Band structure of multilayer graphene in perpendicular electric field. (a) Dispersion relations in AB-stacked 5-layer graphene. 𝐷⊥ was varied from 0 to 0.802 × 10−7 and 4.81 × 10−7 cm−2As. (b) Similar results for AB-stacked 4-layer graphene. Bands are labeled 𝛼𝑒, 𝛽𝑒, 𝛾𝑒, 𝛼ℎ, 𝛽ℎ, and 𝛾ℎ. The characteristic points in the bands are labeled a -- c and a'-b'. The right inset shows the definition of the Slonczewski-Weiss-McClure (SWMcC) parameters. The SWMcC parameters of graphite were used for the calculations (𝛾0 =3.19 eV, 𝛾1 =0.39 eV, γ2 =−0.02 eV, 𝛾3 =0.3 eV, 𝛾4 =0.044 eV, 𝛾5 =0.038 eV, and 𝛥𝑝=0.037 eV). Fig 4. Resistance ridges and characteristic band points in AB-stacked 5-layer graphene. (a) Map of 𝑑𝜌/𝑑𝑛𝑡𝑜𝑡 as a function of 𝑛𝑡𝑜𝑡 and 𝐷⊥. The left panel shows results from the experiment, and the right panel is the numerical calculation with 𝛤 = 3 meV. Resistance ridges b, c and b' correspond to the positions in the dispersion relation in Fig. 3(a). The areas surrounded by the red lines indicate the measured area in the experiment. (b) Similar plot for numerical simulation with 𝛤 = 1 meV. Resistance ridges originating from the monolayer-like band (a and a') are discernible at large values of 𝑛𝑡𝑜𝑡. Fig. 5 Intrinsic resistance ridges for AB-stacked 7-layer graphene (a) Map of 𝜌 as a function of 𝑉𝑏 and 𝑉𝑡. 𝑇 = 4.2 K. 𝐵 = 0 T. (b) Replot as a function of 𝑛𝑡𝑜𝑡 and 𝐷⊥. Fig. 6 Band structure of multilayer graphene in perpendicular electric field. (a) Dispersion relations of AB-stacked 7-layer graphene. From left to right, 𝐷⊥ was varied from 0 to 0.802 × 10−7 and 4.81 × 10−7 cm−2As. (b) Dispersion relations of AB-stacked 6- layer graphene. Bands are labeled 𝛼𝑒 , 𝛽𝑒 , 𝛾𝑒, 𝛿𝑒, 𝛼ℎ, 𝛽ℎ , 𝛾ℎ and 𝛿𝑒 . The characteristic points in the bands are labeled with a -- d and a'-c'. Fig7. Resistance ridges and characteristic band points in AB-stacked 7-layer graphene. Map of 𝑑𝜌/𝑑𝑛𝑡𝑜𝑡 as a function of 𝑛𝑡𝑜𝑡 and 𝐷⊥. The left panel shows experimental results, and the right panel is a calculation with energy broadening 𝛤 = 3 meV. b, c, d, b', and c' correspond to the positions in the dispersion relation. The areas surrounded by the red lines indicate the measured area in the experiment. Fig. 8. Evolution of resistance ridge structure in AB-stacked multilayer graphene. Numerically calculated maps of 𝑑𝜌/𝑑𝑛𝑡𝑜𝑡 (upper panels) and 𝜌 (lower panels) are plotted against 𝑛𝑡𝑜𝑡 and and 𝐷⊥. From left to right, the number of layers are 4, 5, 6 and 7. b stands for the ridge structure due to bilayer-like bands, and m stands for that due to monolayer-like bands. MDP stands for the resistance ridge structure arising from mini-Dirac points. 𝛤=1 meV, and the SWMcC parameter of graphite were used for these calculations. Fig. 9 Landau level structure in AB-stacked 5-layer graphene (a) Map of longitudinal resistivity 𝜌𝑥𝑥 in AB-stacked 5-layer graphene. 𝐷⊥ = 0 cm−2As. 𝑇 = 4.2 K. Numbers show filling factors for some energy gaps. (b) Map of 𝑑𝜌𝑥𝑥/𝑑𝑛𝑡𝑜𝑡. Red bars indicate Landau levels for the monolayer-like band, which appear as a beating of the magnetoresistance oscillations. (c) Numerically calculated energy eigenvalues for AB-stacked 5-layer graphene. Red and black lines show data for K and K' points, respectively. The SWMcC parameters of graphite were used for this calculation. (d) Map of numerically calculated density of states (DOS). Fig. 10. Landau level structure in AB-stacked 7-layer graphene (a) Map of longitudinal resistivity of AB-stacked 7-layer graphene. 𝐷⊥ = 0 cm−2As. 𝑇 = 4.2 K. Numbers show filling factors for some energy gaps.(b) Map of 𝑑𝜌𝑥𝑥/𝑑𝑛𝑡𝑜𝑡 . (c) Numerically calculated energy eigenvalues. Red and black lines show data for K and K' points, respectively. The SWMcC parameters of graphite were used. (d) Map of numerically calculated density of states. Fig 1 (a) y g r e n E (b) Wave vector 3L 4L 5L G (c) TG hBN TG BN G SiO2 BG 6L (d) 1000 ) m h O ( 𝜌 100 7L 𝑉𝑡 (V) 10 -10 -50 0 𝑉𝑏 (V) 50 Fig 2 (a) 10 ) V ( 𝑡 𝑉 -10 -50 𝜌 (Ohm) 2000 0 50 𝑉𝑏 (𝑉) (b) 5 ) s A 2 − m c 7 − 0 1 × ( ⊥ 𝐷 -5 𝜌 (Ohm) (c) 2000 0 5 ) s A 2 − m c 7 − 0 1 × ( ⊥ 𝐷 -5 -5 𝑛𝑡𝑜𝑡 (× 1012 𝑐𝑚−2𝐴𝑠) 5 -5 𝑛𝑡𝑜𝑡 (× 1012 𝑐𝑚−2𝐴𝑠) 5 𝑑𝜌/𝑑𝑛𝑡𝑜𝑡 (arb. units) 1 0 -1 Fig 3 (a) 100 ) V e m ( E 5L -100 -1 (b) 100 ) V e m ( E 4L -100 -1 𝐷⊥ = 0 cm−2 As 0.802 × 10−7 cm−2As 4.81 × 10−7 cm−2As 𝛾𝑒 𝛽𝑒 𝛼e a b c b' a' 𝛼ℎ 𝛽ℎ 𝛾ℎ ℏ𝑘𝑥𝑣0/𝛾1 1 -1 ℏ𝑘𝑥𝑣0/𝛾1 1 -1 ℏ𝑘𝑥𝑣0/𝛾1 1 g1 g2 g3 𝐷⊥ = 0 cm−2 As 0.802 × 10−7 cm−2As 4.81 × 10−7 cm−2As g4 g5 𝛽𝑒 𝛼e ℏ𝑘𝑥𝑣0/𝛾1 1 -1 ℏ𝑘𝑥𝑣0/𝛾1 1 -1 ℏ𝑘𝑥𝑣0/𝛾1 𝛽ℎ 𝛼ℎ 1 g0 a b c b' a' ntot=0にて計算 Fig 4 (a) 5 -5 ) s A 2 − m c 7 − 0 1 × ( ⊥ 𝐷 Experiment b' c -5 b 5 𝑛𝑡𝑜𝑡 (× 1012 𝑐𝑚−2𝐴𝑠) 𝑑𝜌/𝑑𝑛𝑡𝑜𝑡 (arb. units) 1 0 -1 5 ) s A 2 − m c 7 − 0 1 × ( ⊥ 𝐷 -5 (b) 5 ) s A 2 − m c 7 − 0 1 × ( ⊥ 𝐷 -5 Calculation b' c b -5 5 𝑛𝑡𝑜𝑡 (× 1012 𝑐𝑚−2𝐴𝑠) Calculation a' b' c a b -5 5 𝑛𝑡𝑜𝑡 (× 1012 𝑐𝑚−2𝐴𝑠) 𝑑𝜌/𝑑𝑛𝑡𝑜𝑡 (arb. units) 1 0 -1 Fig5 (a) 10 ) 𝑉 ( 𝑡 𝑉 -10 -50 𝜌 (Ω) 400 0 (b) 5 ) A s 2 − m c 7 − 0 1 ( ⊥ 𝐷 -5 𝜌 (Ω) 400 0 𝑉𝑏 (𝑉) 50 -5 𝑛𝑡𝑜𝑡 (1012cm−2) 5 Fig 6 (a) ７L 100 ) V e m ( E 𝐷⊥ = 0 cm−2 As 0.802 × 10−7 cm−2As 4.81 × 10−7 cm−2As 𝛿𝑒 𝛾𝑒 𝛽𝑒 𝛼e a b ｄ b' a' ｃ c' -100 -1 ℏ𝑘𝑥𝑣0/𝛾1 1 -1 ℏ𝑘𝑥𝑣0/𝛾1 1 -1 𝛿ℎ 𝛾ℎ ℏ𝑘𝑥𝑣0/𝛾1 𝛼h 𝛽ℎ 1 (b) 𝐷⊥ = 0 cm−2 As 0.802 × 10−7 cm−2As 4.81 × 10−7 cm−2As 6L 100 ) V e m ( E -100 -1 𝛾𝑒 𝛽𝑒 𝛼e a b ｃｄｃ' b' a' ℏ𝑘𝑥𝑣0/𝛾1 1 -1 ℏ𝑘𝑥𝑣0/𝛾1 1 -1 ℏ𝑘𝑥𝑣0/𝛾1 𝛼h 𝛽ℎ 1 𝛾ℎ a' a Calculation d b' c' ｃ b 𝑑𝜌/𝑑𝑛𝑡𝑜𝑡 (arb. units) 1 0 -1 Fig ７ 5 ) A s 2 − m c 7 − 0 1 ( ⊥ 𝐷 -5 Experiment c' b' d ｃ b 𝑑𝜌/𝑑𝑛𝑡𝑜𝑡 (arb. units) 1 0 -1 5 ) A s 2 − m c 7 − 0 1 ( ⊥ 𝐷 -5 -5 𝑛𝑡𝑜𝑡 (1012cm−2) 5 -5 𝑛𝑡𝑜𝑡 (1012cm−2) 5 e' d Fig 8 4L b MDP b b b b m 5L MDP 6L 7L 𝑑𝜌/𝑑𝑛𝑡𝑜𝑡 (arb. units) MDP MDP MDP b b b m b b m b MDP MDP MDP b 1 m 0 -1 5 ) s A 2 − m c 7 − 0 1 ( ⊥ 𝐷 -5 5 -5 -5 5 -5 5 -5 5 -5 𝜌 5 (arb. units) 1 0 -5 5 -5 5 -5 𝑛𝑡𝑜𝑡 (1012cm−2) 5 -5 5 Fig 9 (a) 7 -22 -14 -2 14 18 (b) 7 𝜌𝑥𝑥 (Ohm) ) T ( B 0 7 ) T ( B (c) -5 𝑛𝑡𝑜𝑡 (× 1012 𝑐𝑚−2) -2 14 18 0 -50 E (meV) 50 10000 1000 100 10 5 ) T ( B 0 -5 (d) 7 K K' ) T ( B 0 -5 𝑑𝑅𝑥𝑥 𝑑𝐵 (arb. units) 1000 100 10 0 DOS (arb. units) 1 0 𝑛𝑡𝑜𝑡 (× 1012 𝑐𝑚−2) 5 -22 -2 14 18 𝑛𝑡𝑜𝑡 (× 1012 𝑐𝑚−2) 5 Fig 10 (a) 7 -6 𝜌𝑥𝑥 (Ohm) (b) 7 ) T ( 𝐵 0 -5 (c) 7 ) T ( 𝐵 0 -50 𝑛𝑡𝑜𝑡 (× 1012 cm−2) 5 -6 1000 100 10 1 ) T ( 𝐵 0 7 ) T ( 𝐵 0 (d) K K' -5 5 𝑛𝑡𝑜𝑡 (× 1012 cm−2) -6 E (meV) 50 -5 5 𝑛𝑡𝑜𝑡 (× 1012 cm−2) 𝑑𝑅𝑥𝑥 𝑑𝐵 (arb. Units) 1000 100 10 0 DOS (arb. units ) 1 0
1605.05739	1	1605	2016-05-18T20:03:59	Interfacial magnetic anisotropy from a 3-dimensional Rashba substrate	[ "cond-mat.mes-hall" ]	We study the magnetic anisotropy which arises at the interface between a thin film ferromagnet and a 3-d Rashba material. The 3-d Rashba material is characterized by the spin-orbit strength $\alpha$ and the direction of broken bulk inversion symmetry $\hat n$. We find an in-plane uniaxial anisotropy in the $\hat{z}\times\hat{n}$ direction, where $\hat z$ is the interface normal. For realistic values of $\alpha$, the uniaxial anisotropy is of a similar order of magnitude as the bulk magnetocrystalline anisotropy. Evaluating the uniaxial anisotropy for a simplified model in 1-d shows that for small band filling, the in-plane easy axis anisotropy scales as $\alpha^4$ and results from a twisted exchange interaction between the spins in the 3-d Rashba material and the ferromagnet. For a ferroelectric 3-d Rashba material, $\hat n$ can be controlled with an electric field, and we propose that the interfacial magnetic anisotropy could provide a mechanism for electrical control of the magnetic orientation.	cond-mat.mes-hall	cond-mat	a Interfacial magnetic anisotropy from a 3-dimensional Rashba substrate Junwen Li1,2, Paul M. Haney1 1. Center for Nanoscale Science and Technology, National Institute of Standards and Technology, Gaithersburg, MD 20899 2. Maryland NanoCenter, University of Maryland, College Park, MD 20742, USA We study the magnetic anisotropy which arises at the interface between a thin film ferromagnet and a 3-d Rashba material. The 3-d Rashba material is characterized by the spin-orbit strength α and the direction of broken bulk inversion symmetry n. We find an in-plane uniaxial anisotropy in the z × n direction, where z is the interface normal. For realistic values of α, the uniaxial anisotropy is of a similar order of magnitude as the bulk magnetocrystalline anisotropy. Evaluating the uniaxial anisotropy for a simplified model in 1-d shows that for small band filling, the in-plane easy axis anisotropy scales as α4 and results from a twisted exchange interaction between the spins in the 3-d Rashba material and the ferromagnet. For a ferroelectric 3-d Rashba material, n can be controlled with an electric field, and we propose that the interfacial magnetic anisotropy could provide a mechanism for electrical control of the magnetic orientation. I. INTRODUCTION Interfacial magnetic anisotropy plays a key role in thin film ferromagnetism. For ultra thin magnetic lay- ers (less than 1 nm thickness), the reduced symmetry at the interface and orbital hybridization between the fer- romagnet and substrate can lead to perpendicular mag- netic anisotropy [1 -- 3]. Perpendicular magnetization in magnetic multilayers can enable current-induced mag- netic switching at lower current densities [4, 5]. Inter- facial magnetic anisotropy is also at the heart of several schemes of electric-field based magnetic switching. In this case an externally applied field can modify the elec- tronic properties of the interface, changing the magnetic anisotropy and leading to efficient switching of the mag- netic layer [6 -- 9]. The combination of symmetry break- ing at the interface and the materials' spin-orbit cou- pling generally leads to an effective Rashba-like interac- tion acting on the orbitals at the interface [10, 11]. The interfacial magnetic anisotropy can be studied in terms of a minimal model containing both ferromagnetism and Rashba spin-orbit coupling [12]. The interfacial magnetic anisotropy direction is a structural property of the sam- ple leading to easy- or hard-axis out-of-plane anisotropy, and isotropic in-plane anisotropy. There has been recent interest in materials with strong spin-orbit coupling which lack structure inversion sym- metry in the bulk. These are known as 3-d Rashba ma- terials, and examples include BiTeI [13] and GeTe [14]. In BiTeI, the structure inversion asymmetry results from the asymmetric stacking of Bi, Te, and I layers, and pho- toemission studies reveal an exceptionally large Rashba parameter α [13]. In GeTe, a polar distortion of the rhombohedral unit cell leads to inversion asymmetry and ferroelectricity [15 -- 17]. Both materials are semiconduc- tors in which the valence and conduction bands are de- scribed by an effective Rashba model with symmetry- breaking direction n, which is determined by the crystal structure. There is interest in finding other ferroelectric materials with strong spin-orbit coupling, motivated by the desire to control the direction of n with an applied electric field [17 -- 22]. In this work, we study the influence of a 3-d (non- magnetic) Rashba material on the magnetic anisotropy of an adjacent ferromagnetic layer. The interface be- tween these materials breaks the symmetry along the z- direction, and the addition of another symmetry breaking direction n enriches the magnetic anisotropy energy land- scape. For a general direction of n, we find a complex de- pendence of the system energy on magnetic orientation. In our model system, we find the out-of-plane anisotropy is much smaller than the demagnetization energy. How- ever an in-plane component of n leads to a uniaxial in- plane magnetic anisotropy which can be on the order of (or larger than) the magnetocrystalline anisotropy of bulk ferromagnetic materials. Control of n (for exam- ple via an electric field in a 3-d Rashba ferroelectric) can therefore modify the magnetization orientation, opening up new routes to magnetic control. FIG. 1: (a) shows the system geometry of a thin film fer- romagnet adjacent to a nonmagnetic material described by a 3-d Rashba model. (b) shows the unit cell of the model system. II. NUMERICAL EVALUATION OF 2-D MODEL We first consider a bilayer system as shown in Fig. 1(a), with unit cell as shown in Fig. 1(b). We take 4 layers of the Rashba material and 2 layers of the ferro- magnetic material with stacking along the z-direction, and assume a periodic square lattice in the x − y plane with lattice constant a. The Hamiltonian of the system is given by H = HTB + HF + HR + HF−R, where: (1) ∆ c† i cj HF = HTB = tXhiji 2 Xiγν 2a Xhijiγν HR = i iγciν (cid:16) M · ~σγν(cid:17) + HTB, c† c† iγcjν [(rij × ~σγν ) · n] + HTB, (3) (2) α HF−R = HTB + α 2a (cid:16)ic† F γcRνσx γν + h.c.(cid:17) . (4) HTB describes nearest-neighbor hopping with amplitude t. HF is the on-site spin-dependent exchange interaction in the ferromagnet. Its magnitude is ∆ and is directed along the magnetization orientation M . HR describes the Rashba layer: spin-orbit coupling and the broken symme- try direction n lead to spin-dependent hopping between sites i and j which is aligned along the rij × n direction, where rij is the direction of the vector connecting sites i and j. α is the Rashba parameter (with units of energy× length). HF −R is the coupling between ferromagnetic and Rashba layers - it includes both spin-independent hopping and spin-dependent hopping. In Eq. 4, the F (R) subscript in the creation and annihilation operators labels the interfacial ferromagnet (Rashba) layer. We find the model results are similar if HF−R includes only spin-independent hopping. The Fermi energy EF is determined by the electron density ρ and temperature T according to: ρ = 1 (2π)2 Z dk f (cid:18) Ek − EF kBT (cid:19) (5) where kB is the Boltzmann constant, and f (x) is the Fermi distribution function: f (x) = (1 + ex)−1. The integral is taken over the two-dimensional Brillouin zone. For a given electron density ρ, Eq. 5 determines the Fermi energy (which generally depends on M ). The total electronic energy is then given by: E( M ) = 1 (2π)2 Z dk Ekf (cid:18) Ek − EF kBT (cid:19) (6) The default parameters we use are ∆ = t, α = 0.4×ta. For t = 1 eV and a = 0.4 nm, this corresponds to a Rashba parameter of 0.16 eV · nm (compared to a value α = 0.38 eV · nm for BiTeI [13]). We let T = 0.1 K and use a minimum of 12002 k-points to evaluate the integrals in Eqs. 5-6. 2 Fig. 2(a) shows the total energy versus magnetic ori- entation for n = y. We observe an out-of-plane mag- netic anisotropy, however its magnitude is much smaller than the demagnetization energy, which is typically on the order of 1000 µJ/m2. In the rest of the paper, we assume that the demagnetization energy leads to easy-plane anisotropy of the ferromagnet, so that the most relevant features of the Rashba substrate-induced anisotropy energy are confined to the x − y plane. The energy versus easy-plane magnetic orientation (parame- terized by the azimuthal angle φ) is shown in Fig. 2(b). The anisotropy is uniaxial and favors orientation in the ±y-directions. This is in contrast to the bulk magne- tocrystalline anisotropy of cubic transition metal ferro- magnets, which has in-plane biaxial anisotropy. As a point of reference for the magnitude of the cal- culated substrate-induced uniaxial anisotropy, we com- pare it to the magnetocrystalline anisotropy EMC for 2-monolayer thick film of Fe, Ni, and Co, for which EMC=(34, 3.5, 318) µJ/m2, respectively [23]. Fig. 2(b) shows that for the material parameters in our model, the Rashba-induced uniaxial anisotropy is larger than the magnetocrystalline anisotropy of Ni. We also note that permalloy, commonly used as a thin film ferromagnet, has a vanishing magnetocrystalline anisotropy [24]. FIG. 2: (a) The system energy as a function of the magnetic orientation for n = y, (c) shows the same for n = (y + z) /√2. (b) shows the system energy versus magnetic orientation within the x − y plane, parameterized by the azimuthal angle φ for n = y (d) shows the same for n = (y + z) /√2. Next we consider Rashba layer with both in-plane and out-of-plane components of n: n = (y + z) /√2. This is motivated in part by the fact that the symmetry is strongly broken along the z-direction by the interface. Our interest is in the influence of an out-of-plane com- ponent n when there is also an in-plane component of n. The resulting E( M ) shown in Fig. 2(c). As before, there is an in-plane uniaxial anisotropy, as shown in Fig. 2(d), with a larger energy barrier as the previous n = y case. Note that the y direction is now a hard axis. In general we find that for an in-plane component of n, the n × z direction can be either a hard or an easy axis, depending on details of the electronic structure. We define the uniaxial in-plane anisotropy energy EA as the difference in energy for M = y and M = x. Fig. 3(a) shows EA as the electron density ρ is varied, and indicates that sign of the anisotropy can change depend- ing on the value of ρ. Fig. 3(b) shows the dependence of EA on α for two values of the electron density ρ. We find that the uniaxial anisotropy energy varies as a power of α which depends on ρ (or equivalently on the band filling). The origin of this dependence is discussed in more detail in the analytic model we develop next. 3 As discussed earlier, we assume the ferromagnet is easy-plane and are therefore interested in the magnetic anisotropy within the plane. In-plane anisotropy in E(k) appears only at 2nd order in α. Taking Mz = 0, we find E(k) takes a simple form in the limits that α ≪ ∆ ≪ 2t and ka ≪ 1. We present the result for the lowest energy band: E (kx) = Ak2 x + Bα2kxMy + C (7) Where A, B, C depend on ∆ and t, whose precise form is not essential for this discussion [25]. Fig. 4(b) shows the numerically computed dispersion for two orientations of the ferromagnet. For M = x, the energy bands are purely quadratic in kx, while for M = y, the energy bands acquire a linear-in-kx component, which is of op- posite sign for the two lowest energy bands. In the case where only the lowest band is occupied, the total energy again depends on the square of the linear-in-kx coeffi- cient, resulting in an in-plane anisotropy energy which is proportional to α4M 2 y . FIG. 3: (a) EA versus dimensionless electron density ρa2. (b) EA versus α for two values of ρ (ρ1a2 = 0.02, ρ2a2 = 2.1), along with fitting to powers of α. For small ρ, the α4 depen- dence can be understood in terms of perturbation theory. III. ANALYTICAL TREATMENT OF 1-D MODEL To gain some insight into the physical origin of the uni- axial in-plane anisotropy, we consider a simplified system of a 1-d chain of atoms extending in the x-direction with 2 sites in the unit cell, as shown in Fig. 4(a), and take n = y. The Hamiltonian is the same as Eqs. 1-4. We compute E(kx) in a perturbation expansion of the spin- orbit parameter α, then determine the total energy as a function of α and M . The lowest order term in α is of the form E(kx) ∝ αkxMz. This can be understood as a simple magnetic exchange interaction between the F and R sites: spin- orbit coupling leads to effective magnetic field on R along the z-direction. This effective magnetic field is propor- tional to αkx and is exchange coupled to the effective magnetic field on the F site (see Fig. 4(a)). The linear- in-kx term in E(kx) shifts the energy bands downward by an amount proportional to the square of the coeffi- cient multiplying kx. The total energy decreases by the same factor. This finally results in a magnetic anisotropy energy which is proportional to α2M 2 z , describing an out- of-plane anisotropy. FIG. 4: (a) cartoon of the unit cell of a 1-d model which extends along the x-direction. The top site is ferromag- netic and the bottom site includes Rashba spin-orbit cou- pling. The arrow along the +z-direction on the R site in- dicates the direction of the Rashba effective magnetic field (we've assumed kx > 0 in the figure). The double arrow along the ±x-direction depicts the spin direction of the spin- dependent hopping between F and R sites. (b) The energy dispersion for the lower two bands for two magnetic orienta- tions (∆ = 1, α = 0.5 for this plot). We can understand the physical origin of the depen- dence of the energy on My: the spin-dependent hopping between F and R sites induces a twisted exchange in- teraction between the spins on these sites, which favors a noncollinear configuration in which both spins lie in the y − z plane [26]. The effective magnetic field on the R-site is always along the z-direction, so the exchange energy therefore differs when the ferromagnet is aligned in the x versus y direction. The twisted exchange interac- tion energy contains a factor proportional to the effective magnetic field on R (which varies as αkx), and a factor proportional to the spin-dependent hopping (which is lin- ear in α) - so the energy varies as finally as α2kxMy. We find that the lowest energy band of the 2-d system of the previous section also contains a linear-in-kx term which varies as α2kxMy, indicating that the physical pic- ture developed for the 1-d system also applies for the 2-d system. In the case where multiple bands are oc- cupied, we're unable to find a closed form solution for the anisotropy energy, and find that it can vary with a power of α that depends on the electron density (and corresponding Fermi level). This is shown in Fig. 3(b) where the uniaxial anisotropy varies as α3 for multiply filled bands (we've observed several different power-law scalings with α for different system parameters). Never- theless the case of a singly occupied band is sufficient to illustrate the physical mechanism underlying the in-plane magnetic anisotropy. IV. CONCLUSION We've examined the influence of a 3-d Rashba mate- rial on the magnetic properties of an adjacent ferromag- netic layer. A uniaxial magnetic anisotropy is developed within the plane of the ferromagnetic layer with easy- axis direction determined by n. Depending on mate- rial parameters, the easy-axis can be parallel or perpen- dicular to n. For large but realistic values of the bulk Rashba parameter of the substrate, the magnitude of the anisotropy indicates that the effect should be observable. For materials in which the direction of the bulk sym- metry breaking is tunable - for example in a ferroelectric 4 3-d Rashba material - the interfacial magnetic anisotropy offers a novel route to controlling the magnetic orienta- tion. The magnitude of the Rashba-induced anisotropy is much less than the demagnetization energy, so its influ- ence is confined to fixing the in-plane component of M . This control would nevertheless be useful in a bilayer geometry in which electrical current flows in plane. In this case the anisotropic magnetoresistance effect yields a resistance which varies as ( J · M )2 [27], which can be utilized to read out the orientation of M . We also note recent works have utilized the reduced crystal symmetry of the substrate to achieve novel directions of current- induced spin-orbit torques [28]. This indicates that the symmetry of the substrate can influence the nonequilib- rium properties of the magnetization dynamics, in addi- tion to modifying the equilibrium magnetic properties, as studied in this work. Acknowledgment J. Li acknowledges support under the Cooperative Re- search Agreement between the University of Maryland and the National Institute of Standards and Technol- ogy Center for Nanoscale Science and Technology, Award 70NANB10H193, through the University of Maryland. [1] P. F. Carcia, A. D. Meinhaldt, and A. Suna, App. Phys. hara, A. Kimura, et al., Nat. Mat. 10, 521 (2011). Lett. 47, 178 (1985). [14] G. Pawley, W. Cochran, R. Cowley, and G. Dolling, Phys. [2] H. Yang, M. Chshiev, B. Dieny, J. Lee, A. Manchon, and Rev. Lett. 17, 753 (1966). K. Shin, Phys. Rev. B 84, 054401 (2011). [3] G. H. O. Daalderop, P. J. Kelly, and M. F. H. Schuur- mans, Phys. Rev. B 50, 9989 (1994). [4] S. Ikeda, K. Miura, H. Yamamoto, K. Mizunuma, H. D. Gan, M. Endo, S. Kanai, J. Hayakawa, F. Matsukura, and H. Ohno, Nat. Mat. 9, 721 (2010). [5] S. Mangin, D. Ravelosona, J. A. Katine, M. J. Carey, B. D. Terris, and E. E. Fullerton, Nat. Mat. 5, 210 (2006). [6] W.-G. Wang, M. Li, S. Hageman, and C. Chien, Nat. Mat. 11, 64 (2012). [7] X. Li, G. Yu, H. Wu, P. Ong, K. Wong, Q. Hu, F. Ebrahimi, P. Upadhyaya, M. Akyol, N. Kioussis, et al., App. Phys. Lett. 107, 142403 (2015). [8] T. Maruyama, Y. Shiota, T. Nozaki, K. Ohta, N. Toda, M. Mizuguchi, A. Tulapurkar, T. Shinjo, M. Shiraishi, S. Mizukami, et al., Nat. Nanotech. 4, 158 (2009). [9] Y. Shiota, T. Nozaki, F. Bonell, S. Murakami, T. Shinjo, and Y. Suzuki, Nat. Mat. 11, 39 (2012). [15] M. Liebmann, C. Rinaldi, D. Di Sante, J. Kellner, C. Pauly, R. N. Wang, J. E. Boschker, A. Giussani, S. Bertoli, M. Cantoni, et al., Adv. Mat. 28, 560 (2016). [16] J. Krempask`y, H. Volfov´a, S. Muff, N. Pilet, G. Landolt, M. Radovi´c, M. Shi, D. Kriegner, V. Hol`y, J. Braun, et al., arXiv preprint arXiv:1503.05004 (2015). [17] A. V. Kolobov, D. J. Kim, A. Giussani, P. Fons, J. Tom- inaga, R. Calarco, and A. Gruverman, APL Mat. 2, 066101 (2014). [18] D. Di Sante, P. Barone, R. Bertacco, and S. Picozzi, Adv. Mat. 25, 509 (2013). [19] A. Narayan, Phys. Rev. B 92, 220101 (2015). [20] A. Stroppa, D. Di Sante, P. Barone, M. Bokdam, G. Kresse, C. Franchini, M.-H. Whangbo, and S. Picozzi, Nat. Comm. 5 (2014). [21] M. Kepenekian, R. Robles, C. Katan, D. Sapori, L. Pedesseau, and J. Even, ACS Nano 9, 11557 (2015). [22] M. Kim, J. Im, A. J. Freeman, J. Ihm, and H. Jin, Proc. [10] J.-H. Park, C. H. Kim, H.-W. Lee, and J. H. Han, Phys. Nat. Ac. Sc. 111, 6900 (2014). Rev. B 87, 041301 (2013). [23] B. D. Cullity and C. D. Graham, Introduction to mag- [11] P. M. Haney, H.-W. Lee, K.-J. Lee, A. Manchon, and netic materials (John Wiley & Sons, 2011). M. D. Stiles, Phys. Rev. B 88, 214417 (2013). [12] S. E. Barnes, J. Ieda, and S. Maekawa, Scient. Rep. 4 (2014). [13] K. Ishizaka, M. Bahramy, H. Murakawa, M. Sakano, T. Shimojima, T. Sonobe, K. Koizumi, S. Shin, H. Miya- [24] L. Yin, D. Wei, N. Lei, L. Zhou, C. Tian, G. Dong, X. Jin, L. Guo, Q. Jia, and R. Wu, Phys. Rev. Lett. 97, 067203 (2006). [25] For completeness we give lowest two the energy bands here, A1,2 = 1 − forms the for α2/64(cid:0)±16 + 8∆ ∓ 16∆2 + 7∆3(cid:1), B1,2 = (∓1 + ∆/2), and C1,2 = 1/16(cid:0)−8α2 + ∆2α2 − 16 ∓ 8∆ − ∆2(cid:1). The α-dependent terms in these factors lead to higher order corrections to the total energy. [26] H. Imamura, P. Bruno, and Y. Utsumi, Phys. Rev. B 69, 121303 (2004). [27] T. McGuire and R. Potter, IEEE Trans. Mag. 11, 1018 (1975). [28] D. MacNeill, G. M. Stiehl, M. H. D. Guimaraes, R. A. Buhrman, J. Park, and D. C. Ralph, arXiv:1605.02712 (2016). 5
1309.3364	1	1309	2013-09-13T05:21:07	Mechanical control of heat conductivity in microscopic models of dielectrics	[ "cond-mat.mes-hall", "physics.comp-ph" ]	We discuss a possibility to control a heat conductivity in simple one-dimensional models of dielectrics by means of external mechanical loads. To illustrate such possibilities we consider first a well-studied chain with degenerate double-well potential of the interparticle interaction. Contrary to previous studies, we consider varying length of the chain with fixed number of particles. Number of possible energetically degenerate ground states strongly depends on the overall length of the chain, or, in other terms, on average length of the link between neighboring particles. These degenerate states correspond to mechanical equilibrium, therefore one can say that the transition between them mimics to some extent a process of plastic deformation. We demonstrate that such modification of the chain length can lead to quite profound (almost five-fold) reduction of the heat conduction coefficient. Even more profound effect is revealed for a model with single-well non-convex potential. It is demonstrated that in certain range of constant external forcing this model becomes "effectively"\ double-well, and has a multitude of possible states of equilibrium for the same value of the external load. Thus, the heat conduction coefficient can be reduced by two orders of magnitude. We suggest a mechanical model of a chain with periodic double-well potential, which allows control over the heat conduction. The models considered may be useful for description of heat transport in biological macromolecules and for control of the heat transport in microsystems.	cond-mat.mes-hall	cond-mat	Mechanical control of heat conductivity in microscopic models of dielectrics A. V. Savin1, ∗ and O. V. Gendelman2, † 1Semenov Institute of Chemical Physics, Russian Academy of Sciences, Moscow 119991, Russia 2 Faculty of Mechanical Engineering, Technion -- Israel Institute of Technology, Haifa 32000, Israel (Dated: June 26, 2021) We discuss a possibility to control a heat conductivity in simple one-dimensional models of dielectrics by means of external mechanical loads. To illustrate such possibilities we consider first a well-studied chain with degenerate double-well potential of the interparticle interaction. Contrary to previous studies, we consider varying length of the chain with fixed number of particles. Number of possible energetically degenerate ground states strongly depends on the overall length of the chain, or, in other terms, on average length of the link between neighboring particles. These degenerate states correspond to mechanical equilibrium, therefore one can say that the transition between them mimics to some extent a process of plastic deformation. We demonstrate that such modification of the chain length can lead to quite profound (almost five-fold) reduction of the heat conduction coefficient. Even more profound effect is revealed for a model with single-well non-convex potential. It is demonstrated that in certain range of constant external forcing this model becomes "effectively" double-well, and has a multitude of possible states of equilibrium for the same value of the external load. Thus, the heat conduction coefficient can be reduced by two orders of magnitude. We suggest a mechanical model of a chain with periodic double-well potential, which allows control over the heat conduction. The models considered may be useful for description of heat transport in biological macromolecules and for control of the heat transport in microsystems. I. INTRODUCTION Heat conduction in low-dimensional systems has attracted a lot of attention and has been a subject of intensive studies [1]. The main objective here is to derive from first principles (on the atomic-molecular level) the Fourier law -- proportion- ality of the heat flux to the temperature gradient J = −κ∇T , where κ is the heat conduction coefficient. To date, there exists quite extensive body of works devoted to the numer- ical modeling of the heat transfer in the one-dimensional chains. Anomalous characteristics of this process are well- known since celebrated work of Fermi, Pasta and Ulam [2]. In integrable systems (harmonic chain, Toda lattice, the chain of rigid disks) the heat flux J does not depend at all on the chain length L, therefore, the thermal conductivity formally diverges. The underlying reason for that is that the energy is transferred by non-interacting quasiparticles and therefore one cannot expect any diffusion effects. Non-integrability of the system is a necessary but not sufficient condition to obtain the convergent heat conduction coefficient. Well-known exam- ples are Fermi-Pasta-Ulam (FPU) chain[3 -- 5], disordered har- monic chain [6 -- 8], diatomic 1D gas of colliding particles [9 -- 11] and the diatomic Toda lattice [12]. In these non-integrable systems also have divergent heat conduction coefficient; the latter diverges as a power function of length: κ ∼ Lα, for L → ∞. The exponent lies in the interval 0 < α < 1. On the other side, the 1D lattice with on-site potential can have finite conductivity. The simulations had demon- strated the convergence of the heat conduction coefficient for Frenkel-Kontorova chain [13, 14], the chain with hyperbolic sine on-site potential [15], the chain with φ4 on-site poten- ∗[email protected] †[email protected] tial [16, 17] and for the chain of hard disks of non-zero size with substrate potential [18]. The essential feature of all these models is existence of an external potential modeling the in- teraction with the surrounding system. These systems are not translationally invariant, and, consequently, the total momen- tum is not conserved. In paper [12] it has been suggested that the presence of an external potential plays a key role to ensure the convergence of the heat conductivity. This hypothesis has been disproved in works [19, 20], where it was shown that the isolated chain of rotators (a chain with a periodic poten- tial interstitial interaction) has a finite thermal conductivity. In recent studies [21, 22] it was demonstrated that the heat con- duction is convergent even in a chain with Morse potential of the nearest-neighbor interaction. It seems that this finding is intimately related to the fact that the "extension" branch of the potential function is finite -- in other terms, a dissociation of the neighboring particles is possible [23]. Thus, the heat flux is scattered on these fluctuating large gaps between the fragments of the chain. In the systems mentioned above, the strong non- homogeneities, which critically effect the heat transfer, are caused by the thermal fluctuations. In current work, we would like to explore somewhat different idea -- to design the inter- action potential and external conditions in a way that the in- homogeneities will appear in controllable manner and with desired density. Thus, it might be possible to control the heat conduction coefficient in wide range by simple variation of the external conditions -- for instance, by stretching the chain. In order to accomplish this goal, we study a chain with a double- well (DW) potential of the nearest-neighbor interactions. We also study certain modification of the DW model, which has only one minimum but can acquire the double-well structure under action of external force. These systems are studied both under nonequilibrium conditions using Langevin thermostats, and within the framework of equilibrium molecular dynamics using Green-Kubo formula [24]. In the first time the thermal conductivity of the chain with double-well potential was considered in [19], with the of help nonequilibrium molecular-dynamics simulation with Nose- Hoover thermostats. It was shown that the use of Nose- Hoover thermostat for the non-equilibrium problems can be misleading [25, 26]. More accurate modeling of the heat transfer using c Langevin thermostat was presented in [27]. It seems that neither of these previous works considered the relationship between the variations of the chain length and the thermal conductivity. However, an important feature of the chain with the DW potential is the existence of a large num- ber of possible ground states of the chain. So, the chain with N particles and periodic boundary conditions has 2N −1 possi- ble ground states with the same energy; the only possible dif- ference between these states is the overall equilibrium length of the chain. We show that the thermal conductivity of the chain depends essentially on its ground state, governed by the length. In each particular simulation, the overall length of the chain is fixed, and all ground states corresponding to this par- ticular value of the length are considered to be equivalent. So, we are going to show that a variation of the chain length brings about significant change of its thermal conductivity. We also show that the same (and even much stronger) effect can be achieved for special design of a single-well nearest-neighbor potential; in this case, the multiplicity of the ground states is achieved by application of a uniform external stretching. Besides, we demonstrate that also the phenomena related to a non-equilibrium heat conduction, like relaxation modes of thermal perturbations, are strongly affected by the number of competing ground states It should be mentioned that the heat conduction coefficient of some models considered in this paper is believed to be di- vergent in the thermodynamic limit. This point is not that sig- nificant here, since we discuss the effect of length, stretching and number of the ground states in a chain with fixed num- ber of particles. Therefore, the heat conduction coefficient is well-defined in all considered cases. II. DESCRIPTION OF THE MODEL Let us consider a general chain with N particles. In a di- mensionless form the Hamiltonian of the chain can be written as H = N Xn=1 1 2 u2 n + N −1 Xn=1 V (un+1 − un), (1) where N is the total number of particles in the chain, un is a coordinate of the n-th particle, dot denotes differentiation with respect to dimensionless time t, V (ρn) is the nearest-neighbor interaction potential, ρn is the length of the n-th link between the neighboring particles. The coordinate of the particle un can not only describe the position of the particles with respect to the the chain axis; it may also correspond to the rotation angle of the n-th monomer around the rotation axis. In this sort of models ρn will denote the relative angle between the (n + 1)-st and the n-th monomer. 2 2 0.5 1 1 1.5 ρ 2 2.5 0.2 ) ρ ( V 0.1 0 FIG. 1: Sketch of double-well potential (2) (curve 1) and single-well potential (8). The straight line connecting the the points (1, V (1)) and (2, V (2)) of graph single-well potential sets its convex hull. We choose the double-well (DW) potential of the interpar- ticle interaction in the following form: V (ρ) = ǫ(ρ − 1)2(ρ − 2)2, (2) where we choose ǫ = 1/2; it leads to V ′′(1) = V ′′(2) = 1. The shape of the potential is presented in Fig. 1. The height of the barrier between the minima of the potential E0 = V (1.5) = 1/25 = 0.0313. To simulate the heat transfer in the chain, we use the stochastic Langevin thermostat. The chain has in general N+ + N + N− particles. We connect N+ particles from one side of the chain to a "hot" Langevin thermostat with tem- perature T+, and N− particles from the other side -- to the Langevin thermostat with temperature T−. The correspond- ing system of equations of motion of the chain can be written as: un = −∂H/∂un − γ un + ξ+ un = −∂H/∂un, n = N+ + 1, ..., N+ + N, un = −∂H/∂un − γ un + ξ− n , n > N+ + N, n , n ≤ N+, (3) where γ = 0.1 is a relaxation coefficient, ξ± n models a white Gaussian noise normalized by the conditions hξ± n (t)i = 0, n (t1)ξ− k (t2)i = 2γT±δnkδ(t2 − t1). hξ+ The system of equations of motion (3) was integrated nu- merically. Considered a chain with fixed ends: u1 ≡ 0, uN ≡ (N − 1)a. Used on-initial condition k (t2)i = 0, hξ± n (t1)ξ± {un(0) = (n − 1)a, un(0) = 0}N n=1, where 1 < a < 2 -- average value of the chain link length. After an initial transient, thermal equilibrium with the ther- mostats was established and a stationary heat flux along the chain appeared. A local temperature is numerically defined as Tn = h u2 nit and the local heat flux -- as Jn = ¯ahjnit, where jn = − unV ′(un − un−1). In numerical simulations we used the following values of temperature T± = (1± 0.1)T x 10−4 n J 7 6 5 4 3 2 1 0 0 0.11 n T 0.1 0.09 0 (a) 40 80 120 160 (b) 40 80 120 160 n FIG. 2: Profile of (a) local heat flux Jn and (b) local temperature Tn in the chain with DW potential (2) for N± = 40, N = 80, T+ = 0.11, T− = 0.09. Linear approximation of the temperature distribution (thin red line) is used for evaluation of the heat conduc- tion coefficient κ(N ). (T = 0.01, 0.03, 0.1), the relaxation coefficient γ = 0.1, the number of units N± = 40, N = 20, 40, 80, ..., 10240. This method of thermalization overcomes the problem of the thermal boundary resistance. The distribution of the heat flow Jn and the temperature in the chain Tn (Fig. 2) clearly demonstrates that inside the "internal" fragment of the chain N+ < n ≤ N+ + N we observe a heat flux independent on the chain cite, as one would expect in an energy-conserving system. Temperature profile also is almost linear. Thus, from this such simulation one can unambiguously define the heat conduction coefficient of the internal chain fragment: κ(N ) = J(N − 1)/(TN++1 − TN++N ). (4) In thermodynamical limit, one can say that the system obeys the Fourier law, if there exists a finite limit ¯κ = lim N→∞ κ(N ). If such limit does exist, one can say that the chain has normal (finite) or convergent heat conduction. In the opposite case, an anomalous heat conduction is observed. Alternative way for evaluation of the heat conduction co- efficient is based on linear response theory, which leads, in 3 (a) (b) κ κ 45 40 35 30 25 20 15 12 9 6 3 3 1 2 102 103 104 3 2 102 1 N 103 104 FIG. 3: Dependence of the heat conduction coefficient κ on the size of the internal chain fragment N for the chain with the DW potential (2) for the average interatomic distance (a) a = 1 and (b) a = 1.5 for temperatures T = 0.01, 0.03, 0.1 (curves 1, 2, 3 respectively). particular, to famous Green-Kubo expression [24]: κc = lim t→∞ lim N→∞ 1 N T 2 Z t 0 c(τ )dτ, (5) where the autocorrelation function of the heat flux in the chain is defined as c(t) = hJs(τ )Js(τ − t)iτ . Here a total heat flux in the chain is defined as Js(t) = Pn jn(t). In order to compute the self-correlation function c(t) we simulate a cyclic chain consisting of N = 104 particles and couple all these particles to the Langevin thermostat with tem- perature T . After initial thermalization, the chain has been detached from the thermostat and Hamiltonian dynamics has been simulated. In order to improve the accuracy, the self- correlation function has been computed by averaging over 104 realizations with independent initial conditions, correspond- ing to the same initial temperature T . The heat conduction turns out to be convergent if the self- correlation function c(τ ) decreases fast enough as τ → ∞. Namely, if the integral in expression (5) converges then the heat conduction may be treated as normal. κ n l 2 1 0 3 N0.29 1 N0.09 N0.29 2 ) t ( c 102 N 103 104 FIG. 4: Dependence of the heat conduction coefficient κ on the size of the internal chain fragment N for the chain with the DW potential (2) for the average interatomic distance a = 1.5 for temperatures T = 0.01, 0.03, 0.1 (curves 1, 2, 3 respectively). The graphs are presented in double logarithmic scale and straight lines correspond to fitting N α, α = 0.09, 0.29, 0.29 for curves 1, 2, 3 respectively. III. CHAIN WITH SYMMETRIC DW POTENTIAL Dependence of the heat conduction coefficient κ on the length of the chain fragment between the Langevin ther- mostats N for the chain with symmetric DW potential (2) is presented in Fig. 3. As one can see from this figure, both for a = 1 and a = 1.5 the heat conduction coefficient gown monotonously with N and demonstrates no trend for conver- gence. For a = 1.5 the heat conduction coefficient grows like N α, with α = 0.09 for the temperature T = 0.01 and α = 0.29 for T = 0.03, 0.1, as it is demonstrated in Fig. 4. Numeric analysis of the heat flux autocorrelation function c(t) as t → ∞ supports the conclusion on divergent heat con- ductivity in the system. For all values of the average link length 1 ≤ a ≤ 2 the function c(t) decreases as t → ∞ according to the power law t−β with exponent β < 1, as it is demonstrated in Fig. 5. According to Green-Kubo formula (5) one arrives to the same conclusion on divergence of the heat conduction coefficient κ(N ) as N → ∞. Two approaches used here are independent and point in the same direction -- one obtains clear indication that the heat conductivity in the chain with DW potential diverges. So, as it was already mentioned in the Introduction, it is reasonable to explore and compare the heat conduction coef- ficient κ(N ) for some fixed chain length. To be specific, we choose N = 640 and consider the temperature dependence of the heat conductivity for fixed number of particles and varying average link length. The results are presented in Fig. 6. For all presented values of the link length the heat conductivity in the case of low temperatures T < 0.01 sharply decreases as the temperature grows. For large temperatures the heat conduc- tivity weakly increases with the temperature. In all cases the minimum is achieved in the temperature interval 0.01 ÷ 0.04, which is close to the height of the potential barrier. The low- 4 10−3 10−4 10−5 10−6 10−7 10−8 t−0.85 t−0.77 t−0.87 t−0.9 3 2 1 6 5 4 t−0.72 t−0.7 101 102 t 103 FIG. 5: Power-law decrease of the autocorrelation function c(t) for the chain with symmetric DW potential (2) for average link length a = 1 and temperatures T = 0.01, 0.03, 0.1 (curves 1, 2, 3 respec- tively) and a = 1.5, T = 0.01, 0.03, 0.1 (curves 4, 5, 6 respectively). The graphs are presented in double logarithmic scale and straight lines correspond to fitting t−β, β = 0.9, 0.87, 0.85, 0.7, 0.72, 0.77 for curves 1, 2,..., 6 respectively. 300 250 200 κ 150 100 50 0 2 1 3 10−2 10−1 T FIG. 6: Dependence of the heat conduction coefficient κ on temper- ature T of the chain containing in general 720 particles (N = 640, N± = 40) with the DW potential (2) for a = 1.5, 1.1, 1.0 (curves 1, 2, 3 respectively). est values of the heat conduction coefficient are obtained for a = 1.5. This result is quite expectable, since namely for this value of the average link length the number of topologically different degenerate ground states achieves a maximum. From Fig. 6 one can see that if the average link length a varies from 1 to 1.5, the heat conduction coefficient decreases a µ 1 0.8 0.6 0.4 0.2 0 10−2 T 10−1 1.3 1.2 a 1.4 1.5 1 1.1 0.25 0.2 N µ 0.15 0.1 5 3 2 102 1 N 103 104 FIG. 7: Temperature dependence of the reduction coefficient µa = κ(a)/κ(1) on the average link length a. monotonously. In order to estimate the efficiency of the heat conductivity reduction, we define the reduction coefficient µa = κ(a, N, T )/κ(1, N, T ), where κ(a, N, T ) denotes the heat conduction coefficient for the average link length a, num- ber of particles N and temperature T . We note that by sym- metry considerations it is enough to consider only the interval 1 ≤ a ≤ 1.5. A dependence of µa on the temperature and the average link length is presented in Fig. 7. The heat conductivity decreases with increase of a for all studied values of the temperature. Maximal efficiency of the reduction is achieved for the tem- perature T = 0.02, which is close to the height of the potential barrier E0 = 0.0313. As one could expect, maximal reduction of the heat conductivity occurs at a = 1.5. However, other aspect of the reduction phenomenon is somewhat unexpected. In Fig. 8 we present the dependence of the reduction efficiency for different temperatures µN = κ(1.5, N, T )/κ(1, N, T ) as a function of the particle number N . It is somewhat surprising to see that this dependence is not monotonous. The chain length for which the most efficient re- duction of the heat conductivity is observed, strongly depends on the temperature. For instance, for T = 0.01 the most effi- cient reduction is observed when N0 = 1280, and for higher temperatures T = 0.03, 0.1 -- when N0 = 160. It is easy to explain why for relatively short chains the re- duction efficiency is higher when the chain gets longer. We believe that the reduction of the heat conductivity occurs due to increase of a number of the degenerate ground states. Quite obviously, this effect becomes more profound as the number of particles grows. The decrease of the reduction efficiency for relatively high N is more difficult to explain. Qualita- tively, one can speculate that, since the heat conduction coef- ficient in the chain with DW potential diverges, for very large N the heat transfer is governed by long-wavelength weakly interacting phonons. Such waves are less sensitive to the de- tails of the chain structure and "feel" only average density of the particles. It is also instructive to compare the process of non- stationary heat conduction in the DW model with different average link length. For this sake, following a methodology developed in [28, 29], we simulate a relaxation of various spa- FIG. 8: Dependence of the reduction coefficient µN = κ(1.5)/κ(1) on the chain length N for temperature values T = 0.01, 0.03, 0.1 (curves 1, 2, 3 respectively). tial modes of a thermal perturbation in the cyclic DW chain with N particles. Thus, the initial temperature distribution is defined according to the following relationship: Tn = T0 + A cos[2π(n − 1)/Z], (6) where T0 is the average temperature, A is the amplitude of the perturbation, and Z is the length of the mode (number of par- ticles). The overall number of particles N has to be multiple of Z in order to ensure the periodic boundary conditions. In order to realize this initial thermal perturbation, each par- ticle in the chain is first attached to a separate Langevin ther- mostat. In other terms, we integrate numerically the following system of equations: un = −∂H/∂un − γ un + ξn, (7) where n = 1, 2, ..., N , γ = 0.1, and the action of the thermo- stat is simulated as white Gaussian noise normalized accord- ing to conditions hξn(t1)ξk(t2)i = 2γTnδknδ(t2 − t1). After the initial heating in accordance with (6), the Langevin thermostat is removed and relaxation of the system to a sta- tionary temperature profile is studied. The results were av- n=1 in eraged over 106 realizations of the initial profile {Tn}N order to reduce the effect of fluctuations. Samples of the non-stationary simulations are presented in Figs. 9 and 10. In both figures, we compare the relaxation of similar thermal perturbation profiles with similar average tem- perature, perturbation amplitude and spatial wavelength. The only difference is the average link length. We compare the two extreme cases a = 1 and a = 1.5. In Fig. 10 one can see that the same thermal perturbation for a = 1 decays in oscillatory manner, whereas for a = 1.5 the decay is smooth. Thus, one can suggest that for the case a = 1 one should take into ac- count phenomena related to hyperbolic corrections to Fourier x 10−3 14 12 10 n n T T 8 6 16 32 n 48 64 360 240 t 120 x 10−3 14 12 10 n T 8 6 16 32 n 48 64 240 160 t 80 (a) (b) 0 0 FIG. 9: Relaxation of initial periodic thermal profile in the periodic chain with potential (2) N = 1024, T0 = 0.01, A = 0.05, Z = 64 for (a) a = 1 and (b) a = 1.5. law (like possibility of the second sound); in the same time, for the case a = 1.5 one should expect primarily diffusive behavior. This conclusion is further confirmed by simulation results presented in Fig. 9. There we can see that for a = 1 the oscillatory (i.e. hyperbolic) behavior is observed for very broad diapason of the wavelengths. For all these modes, the thermal perturbations for a = 1.5 decay in primarily diffusive manner. The reason for this difference, again, is a "perfect" structure of the ground state for a = 1 and large number of the degenerate ground states for a = 1.5. IV. A MODEL WITH NON-CONVEX SINGLE-WELL POTENTIAL The results presented in the previous Section suggest that the increase of the number of the ground states peculiar for the DW chain leads to suppression of the heat conduction. 6 (a) 0.014 T 0.01 0.006 7 0.015 T 0.01 0.005 6 6 5 k 4 5 k 4 3 2 1 0 400 t/t k 200 (b) 3 2 1 0 200 400 t/t k 600 600 FIG. 10: Evolution of the relaxation profile in the periodic chain with double-well potential (2) (N = 2048, T0 = 0.01, A = 0.05) for (a) a = 1 (Z = 23+k, k = 1, 2, ..., 7, tk = 0.15, 0.3, 0.6, 1.2, 2.0, 4.0, 10.0) and (b) a = 1.5 (Z = 23+k, tk = 0.1 · 2k−1, k = 1, 2, ..., 6). Time dependence of the mode maximum T (1+Z/2) [red (gray) lines] and minimum T(1) [blue (black) lines] are depicted. One should expect even stronger suppression effect, when the chain is modified in a way that the potential changes from single-well to double-well. Possibility of such modification in particular chain model under action of an external force, and the consequences for the heat conductivity are discussed here. First of all, we suggest a model of the chain with a non-convex single well potential of the nearest-neighbor interaction: V (ρ) = ǫ(ρ − 1)2(ρ − 2)2 + βρ + c, (8) where ǫ = 1/2, β = 0.15, and (physically insignificant) constant c may be found from a condition min V (ρ) = 0. Under these values of parameters potential function (8) has a single minimum at ρ0 = 0.89 (Fig. 1). Let us consider the case when the chain with potential (8) is elongated by external force F applied to one of its ends, whereas the other end remains fixed. It is easy to see that such external forcing is equivalent to a modification of the interaction potential (8) by addition of negative linear term. This modified potential will have the following form: V ∗(ρ) = ǫ(ρ − 1)2(ρ − 2)2 + βρ − F ρ + c. (9) Thus, the application of the external force can modify the qualitative shape of the potential function. Namely, it is easy to demonstrate that for F < Fmin = β − ǫ/3√3 and for F > Fmax = β + ǫ/3√3 the potential V ∗(ρ) remains single- well. However for Fmin < F < Fmax the effective potential becomes double-well and thus one can expect significant re- duction of the heat conductivity as a result of application of the external force. In our numeric experiments we control the average link length a rather than the value of the external force. However, it is easy to translate between these quantities. Namely, poten- tial function (9) will have two wells if the following condition holds: amin < a < amax, where amin = 3/2 − 1/√3 ≈ 0.9226 and amax = 3/2 + 1/√3 ≈ 2.0774. Creation of the effective double-well potential can be also explained from observation of the single-well potential func- tion (8) depicted in Fig. 1. One can see that this function is not convex (the second derivative is negative for 1 ≤ ρ ≤ 2). Ho- mogeneous extension of such chain, when all the links have the same length ρn ≡ a > ρ0, is energetically favorable only when the average link length belongs to the interval where the potential function has positive second derivative: a ≤ ρ1 = 1 and a ≥ ρ2 = 2. If ρ1 < a < ρ2, it is more favorable to have part of the links with the length ρn ≡ ρ1 = 1, and the rest of the links -- with the length ρn ≡ ρ2 = 2. In this case, the dependence of the average potential energy per particle on the average link length will follow the straight line connecting the points (ρ1, V (ρ1)) and (ρ2, V (ρ2)), as it is demonstrated in Fig. 1 [30, 31]. Thus, as the average link length increases, the numbers of "short" and "long" links vary accordingly, thus giving rise to large number of possible realizations and large variations of the heat conduction coefficient. Heat transfer in this system has been simulated with the help of Langevin thermostats (3) under fixed-ends boundary conditions: u1(t) ≡ 0, uN++N +N− = (N+ + N + N− − 1)a, where again a ≥ ρ0 is the average length of the link. Numeric simulation of the heat conduction demonstrated that for all studied values of the average link length a ≥ ρ0 the heat conduction coefficient diverges: κ(N ) ր ∞ as N ր ∞ -- see. Fig. 11. Equilibrium simulations based on Green-Kubo formula (5) bring about the same conclusion. The dependence of the heat conduction coefficient κ on the average link length a for the chain with the single-well non- convex potential of the nearest-neighbor interaction (8) is pre- sented in Fig. 12. One can observe extremely significant ef- fect of the chain extension on the heat conduction coefficient. This drastic reduction is observed for values of the average link length approximately in the interval 1 < a < 2, in accor- dance with earlier findings on relationship between a number of possible degenerate (or close energetically) ground states and the reduction of the heat conductivity. It is possible to say that these drastic modifications of the heat conduction coefficient occur due to "latent" double-well nature of the non-convex potential of the nearest-neighbor in- teraction. Notably, the most efficient reduction is achieved when the temperature is close to the height ǫ of the "latent" potential barrier. The reduction of the heat conductivity by two orders of magnitude is achieved for T = 0.02. 6 5.5 κ n l 5 4.5 3.5 κ n l 3 2.5 1 κ n l 0.5 0 7 (a) (b) (c) 102 102 102 103 103 103 N FIG. 11: Dependence of heat conduction coefficient κ on the chain length N for the chain with single-well nearest-neighbor potential (8) for different values of the average link length: (a) a = ρ0, (b) a = 2.0, (c) a = 1.5. In all cases the temperature T = 0.01. It is important to mention that the effective non-convex single-well potential is obtained in effective models of some quasi-one-dimensional objects. For instance, it is realized in α-spirals of protein macromolecules, double helix of DNA [30, 31], as well as in a model of intermetallic NiAl crys- talline nanofilms [32]. In these structures the external exten- sion brings a non-homogeneous equilibrium configurations. Then, one should expect strong effect of external mechanic loads on transport properties in systems of this sort. V. CONTROL IN MODELS WITH CONVERGENT HEAT CONDUCTION All simple models considered above are commonly be- lieved to have the divergent heat conductivity in the thermody- namical limit. In this section, we are going to consider a mod- ification of chain of rotators, which allows external control of the heat conduction coefficient. Simple chain of rotators is believed to exhibit convergent heat conductivity [19, 20]. To explain the idea of possible modification, let us consider a mechanical model sketched in Fig. 13 -- a chain of equal par- allel disks of radius r with centers fixed at equal intervals of length a along x axis. The disks can rotate along x axis and the rotation angle of the n-th disk is denoted as φn. Neighbor- ing disks are coupled by harmonic springs of equal stiffness and equilibrium length. Hamiltonian of such model can be simply expressed as follows: H = N Xn=1 1 2 I φ2 n + V (φn+1 − φn), (10) κ 104 103 102 101 100 0.5 0.4 V 0.3 0.2 0.1 0 −1 1 2 1 1.2 1.4 1.6 3 a 1.8 2 2.2 2.4 8 −0.5 0 ∆φ/π 0.5 1 FIG. 12: Dependence of the heat conduction coefficient κ in the chain with single-well potential of the nearest-neighbor interaction (8) on the average link length a for temperatures T = 0.002, 0.02, 0.2 (curves 1, 2, 3 respectively). Note a logarithmic scale for κ. Simulated chain contained in all cases N+ + N + N− particles, N = 640 of them were free and N± = 40 were immersed in the Langevin thermostats with temperatures T± = (1 ± 0.1)T . z y x l n φ n r a n−2 n−1 n n+1 n+2 FIG. 13: Sketch of the modified chain of rotators where I is a moment of inertia of the disk. Potential energy of relative rotation appears due to deformation of the springs, which couple the neighboring disks, and is expressed as V (∆φn) = K(ln − L0)2 K{[a2 + 2r2(1 − cos(∆φn)]1/2 − L0}2, 1 2 = 1 2 FIG. 14: Shape of potential function (11) V (∆φ) for a = 1, r = 1, L0 = 1.2, K = 1. Heights of the potential barriers are ǫ0 = 0.02, ǫ1 = 0.5367. The potential function achieves the minima for ∆φ = ±0.2152π. Characteristic shape of the modified potential function (11) is presented in Fig. 14. Without affecting the generality, one can set I = 1, a = 1 and K = 1. To be specific, we also choose r = 1. Thus, the potential will have two wells for the equilibrium length of the spring in the interval 1 < L0 < √5. The function V (∆φ) is 2π-periodic and has two potential barriers 0 < ǫ0 < ǫ1. If the equilibrium spring length L0 will only slightly exceed the distance a between the disk centers, then one will obtain ǫ0 ≪ ǫ1. For instance, for L0 = 1.2 one obtains the potential minima ∆φ0 = ±0.2152π ≈ 0.6761, and the barriers ǫ0 = 0.02, ǫ1 = 0.5367. We expect that, similarly to the simple chain of dipole rota- tors [19, 20], the chain with Hamiltonian (10) will have con- verging heat conduction coefficient. To simulate the heat conductivity, we attach the chain ends with N± = 40 disks to Langevin thermostats with tem- peratures T± = (1 ± 0.1)T . Then, we simulate numeri- cally the system of equation (3) with potential function ob- tained from (11) under fixed boundary conditions ϕ0 ≡ 0, ϕM ≡ (N+ + N + N− − 1)ϕ0 and with initial conditions (11) {ϕn(0) = (n − 1)ϕ0, ϕn(0) = 0}N++N +N− n=1 , Here ∆φn = φn+1 − φn is a relative rotation angle of the neighboring disks, K and L0 are the stiffness and equilibrium length of the springs respectively, ln is the length of the n-th spring. Potential function (11) will be double-well provided that a < L0 < [a2 + 4r2]1/2. The potential minima in this case correspond to the following values of the relative rotation angle: ∆φ0 = ±2 arcsin{[(L2 0 − a2)/4r2]1/2}. where ϕ0 -- average value of angle between neighboring disks. In other terms, we keep constant relative rotation angle be- tween the ends of the chain (in other terms, we apply constant momentum to the chain) and study the dependence of the heat conduction coefficient on these external conditions. Dependence of the heat conduction coefficient κ on the length of the chain fragment between the Langevin ther- mostats N for the chain with periodic potential (11) is pre- sented in Fig. 15. As one can see, the value of N for which the heat conductivity converges, strongly depends on the av- erage temperature T . 2 1 0 κ n l 1 3 4 2 6 5 102 103 N FIG. 15: Dependence of the heat conduction coefficient κ on the size of the internal chain fragment N for the chain with periodic potential (11) for the average angle between the disks ϕ0 = ∆φ0 and ϕ0 = 0 respectively (curves 1 and 2 for temperature T =0.01, curves 3 and 4 -- for T = 0.02, curves 5 and 6 -- for T = 0.2). For average temperature T = 0.2 comparable to a value of the higher potential barrier ǫ1, the convergence is reliably achieved already for N = 320. We also see that the heat con- duction coefficient almost does not depend on ϕ0. The latter result seems natural, since the temperature is large enough to allow frequent transitions over the higher barrier; then, the initial mutual rotation of the disks is relaxed. If the average temperature will be much lower than the higher potential barrier T ≪ ǫ1, the relaxation mentioned above will require exponentially large time. In this case, the heat conduction still converges, but the convergence requires consideration of essentially larger values of the chain length N . For every chain length, the heat conductivity in the chains with ϕ0 = 0 is significantly higher that in the chains with ϕ0 = ∆φ0. For the average temperature T = 0.01 the differ- ence is 6.5 times, and for T = 0.02 -- 5 times. To check this assumption further, let us consider also the dependence of κ on the initial angle between the disks ϕ0 for the chain length N = 1280. In Fig. 16, one can clearly ob- serve that this dependence is strong enough, provided that the temperature is sufficiently low to prevent the fast relaxation. Physical reasons of this behavior are very similar to those described in the preceding Sections. For ϕ0 ≤ ∆φ0 the chain has energetically degenerate ground states, where part of the neighboring disk pairs have relative equilibrium angle φn+1−φn = ∆φ0, and the other pairs -- φn+1−φn = −∆φ0. A number of possible different ground states grows as ϕ0 tends to zero. That is why the heat conductivity decreases in this limit. This effect becomes more pronounced as the temperature decreases. Still, for extremely small temperatures the heat conductivity will cease to depend on ϕ0, since even the smaller potential barriers will become prohibitive. For exponentially large simulation times the dependence of the heat conductivity on the initial relative rotation of the disks 9 102 κ 101 100 2 1 −0.9 −0.6 −0.3 0 0.3 0.6 0.9 ϕ0 FIG. 16: (a) Dependence of the heat conduction coefficient κ in the chain with periodic potential of the nearest-neighbor interaction (11) on the average interdipole angle ϕ0 for temperatures T = 0.01, 0.02 (curves 1, 2 respectively). is expected to disappear due to thermally activated relaxation over higher potential barriers. In our simulations the times were not large enough to observe this relaxation for lower temperatures. Still, we believe that this problem may be eas- ily overcome. It seems sufficient to act with constant external momentum on the right end of the chain, rather than to fix it. Then, one should investigate a dependence of the heat con- duction coefficient on the value of this external momentum. The idea presented in this Section could be useful for prac- tical design of the systems with controlled heat conductiv- ity. One can use, for instance, the stretched polymer macro- molecules. Their heat conductivity can be modified by appli- cation of the external momentum. Concluding remarks In this paper we demonstrated that one can efficiently con- trol the transport properties of model atomic chains by purely mechanical means. In the case of the chain with double-well interparticle potential, it is enough to change the average in- terparticle distance in order to modify a number of possible degenerate ground states and to reduce or to increase the heat conductivity. The effect is rather pronounced (about five-fold reduction was observed). However, even stronger effect -- re- duction by two orders of magnitude -- was observed in more realistic model with single-well nonconvex interparticle po- tential. In this model the heat conductivity is directly related to applied external strain. Also in this case it seems that the reduction effect is caused by formation of "effective" double- well potential and variation of a number of possible states of mechanical equilibrium. Modification of thermal conductivity in conditions of exter- nal mechanical load can be related to broader field of thermoe- lasticity. It is well-known that elasticity and plasticity in real materials can be strongly coupled with thermodynamic phe- nomena and heat transport. It seems, however, that this pos- sible coupling has not received proper attention in numerous recent studies devoted to microscopic foundations of the heat conductivity. Our results demonstrate that these effects can be rather profound. One can be tempted to say that the extension of the chain with the DW potential exemplifies the effect of plasticity on the heat transport. The chain with the single-well potential requires constant external forcing to reduce the heat conductivity; this phenomenon is primarily elastic. Needless 10 to say, these statements are quite crude and schematic; still we believe that the subject is of considerable fundamental inter- est. Besides, interesting practical implications are straightfor- ward -- one can be interested in simple mechanical means of control over heat transport in micro- and nanosystems. Acknowledgements A. V. S. thanks the Joint Supercomputer Center of the Rus- sian Academy of Sciences for the use of computer facilities. [1] S. Lepri, R. Livi, A. Politi. Thermal conduction in classical low- Phys. Rev. Lett. 92(7), 074301 (2004). dimensional lattices. Physics Reports 377, 1-80 (2003). [2] E. Fermi, J. Pasta, S. Ulam. Studies of non linear problems. Los Alamos Rpt LA.1940 (1955). [3] S. Lepri, L. Roberto, A. Politi. Heat Conduction in Chains of Nonlinear Oscillators. Phys. Rev. Lett. 78, 1896 (1997). [4] S. Lepri, R. Livi, A. Politi. Energy transport in anharmonic lat- tices close and far from equilibrium. Physica D 119, 140 (1998). [5] S. Lepri, R. Livi, A. Politi. On the anomalous thermal con- ductivity of one-dimensional lattices. Evrophys. Lett. 43, 271 (1998). [6] R. Rubin and W. Greer, Abnormal Lattice Thermal Conductiv- ity of a One-Dimensional, Harmonic, Isotopically Disordered Crystal. J. Math. Phys. 12, 1686 (1971). [7] A. Casher and J.L. Lebowitz, Heat Flow in Regular and Disor- dered Harmonic Chains. J. Math. Phys. 12, 1701 (1971). [19] C. Giardina, R. Livi, A. Politi, and M. Vassalli. Finite thermal conductivity in 1D lattices. Phys. Rev. Lett. 84, 2144 (2000). [20] O. V. Gendelman and A. V. Savin. Normal heat conductivity of the one-dimensional lattice with periodic potential of nearest- neighbor interaction. Phys. Rev. Lett. 84, 2381 (2000). [21] Y. Zhong, Y. Zhang, J. Wang and H. Zhao. Normal heat con- duction in one-dimensional momentum conserving lattices with asymmetric interactions. Phys. Rev. E 85, 060102(R) (2012). [22] S. Chen, Y. Zhang, J. Wang, and H. Zhao. Breakdown of the power-law decay prediction of the heat current cor- relation in one-dimensional momentum conserving lattices. arXiv:1204.5933v3 [cond-mat.stat-mech] 15 Oct 2012. [23] A. V. Savin and Y. A. Kosevich Thermal conductiv- interaction. ity of arXiv:submit/0761690 [cond-mat.stat-mech] 17 Jul 2013. the chain with an asymmetric pair [8] A. Dhar, Heat Conduction in the Disordered Harmonic Chain [24] R. Kubo, M. Toda, N. Hashitsume. Statistical Physics II. / Revisited. Phys. Rev. Lett. 86, 5882 (2001). [9] A. Dhar, Heat Conduction in a One-Dimensional Gas of Elas- tically Colliding Particles of Unequal Masses. Phys. Rev. Lett. 86, 3554 (2001). [10] A.V. Savin, G.P. Tsironis, and A.V. Zolotaryuk, Heat Conduc- tion in One-Dimensional Systems with Hard-Point Interparticle Interactions. Phys. Rev. Lett. 88, 154301 (2002). [11] P. Grassberger, W. Nadler, and L. Yang, Heat Conduction and Entropy Production in a One-Dimensional Hard-Particle Gas. Phys. Rev. Lett. 89, 180601 (2002). [12] T. Hatano. Heat conduction in the diatomic Toda lattice revis- ited. Phys. Rev. E 59, R1 (1999). [13] B. Hu, B. Li, and H. Zhao, Heat conduction in one-dimensional chains. Phys. Rev. E 57, 2992 (1998). [14] A. V. Savin and O. V. Gendelman, Heat conduction in one- dimensional lattices with on-site potential. Phys. Rev. E 67, 041205 (2003). [15] G.P. Tsironis, A.R. Bishop, A.V. Savin, and A.V. Zolotaryuk, Dependence of thermal conductivity on discrete breathers in lattices. Phys. Rev. E 60, 6610 (1999). [16] B. Hu, B. Li, and H. Zhao, Heat conduction in one-dimensional nonintegrable systems. Phys. Rev. E 61, 3828 (2000). [17] K. Aoki and D. Kuznezov, Bulk properties of anharmonic chains in strong thermal gradients: non-equilibrium ϕ4 theory. Phys. Lett. A 265, 250 (2000). [18] O.V. Gendelman and A.V. Savin. Heat Conduction in a One- Dimensional Chain of Hard Disks with Substrate Potential. Springer, Ser. Solid State Sci. V. 31 (1991). [25] A. Fillipov, B. Hu, B. Li, and A. Zeltser. Energy transport be- tween two attractors connected by a Fermi-Pasta-Ulam chain. J. Phys. A: Math. Gen. 31, 7719 (1998). [26] F. Legoll, M. Luskin, and R. Moeckel. Non-ergodicity of Nose- Hoover dynamics. Nonlinearity 22, 1673 (2009). [27] D. Roy. Crossover from Fermi-Pasta-Ulam to normal diffusive behavior in heat conduction through open anharmonic lattices. Phys. Rev. E 86, 041102 (2013). [28] O.V.Gendelman and A.V.Savin. Nonstationary heat conduction in one-dimensional chains with conserved momentum Phys Rev. E 81, 011105 (2012) [29] O.V.Gendelman, R.Shvartsman, B.Madar and A.V.Savin. Non- stationary heat conduction in one-dimensional models with sub- strate potential Phys Rev. E 86, 020103(R) (2010) [30] A. V. Savin, I. P. Kikot, M. A. Mazo, and A. V. Onufriev. Two- phase stretching of molecular chains. PNAS February 1, 2013 201218677. [31] A. V. Savin, M. A. Mazo, I. P. Kikot, A. V. Onufriev Two-phase stretching of molecular chains. arXiv:1110.3165v2 [cond-mat.mes-hall] 25 Feb 2013. [32] R. I. Babicheva, K. A. Bukreeva, S. V. Dmitriev, and K. Zhou. Discontinuous elastic strain observed during stretching of NiAl signal crystal nanofilms. Computational Materials Science 80 (2013).
1605.02195	4	1605	2017-02-21T06:56:35	Robust Low-Bias Negative Differential Resistance in Graphene Superlattices	[ "cond-mat.mes-hall" ]	Here, we present a detailed study on low bias current-voltage (I-V) characteristic of graphene superlattice (GSL) resonant tunneling diode (RTD) with heterostructured substrate and series of grounded metallic planes placed over graphene sheet which induce periodically modulated Dirac gap and Fermi velocity barrier. We investigate the effect of GSL parameters on I-V characteristics within the Landauer-Buttiker formalism and adopted transfer matrix method. We show how the engineering these parameters results in multipeak NDR in proposed device. Moreover we provide a novel venue to control the NDR in GSL with Fermi velocity engineering. From this viewpoint we obtain multipeak NDR through miniband align in GSL. Maximum Pick to Valley ratio (PVR) up to 167 obtained for correlation velocity of 1.9 and bias voltages between 70-130 mV. Our finding in low bias regime and high PVR multipeak NDR have a considerable importance in multi-valued memory functional circuit, low power and high-speed nanoelectronic devices application.	cond-mat.mes-hall	cond-mat	Robust Low-Bias Negative Differential Resistance in Graphene Superlattices S. M. Sattari-Esfahlan, J. Fouladi-Oskuei and S. Shojaei * Photonics Department, Research Institute for Applied Physics and Astronomy (RIAPA), University of Tabriz, 51665-163 Tabriz, Iran Abstract In this work, we present a detailed theoretical study on low bias current-voltage (I-V) characteristic of biased planar graphene superlattice (PGSL) provided by heterostructured substrate and series of grounded metallic planes placed over graphene sheet which induce periodically modulated Dirac gap and Fermi velocity barrier, respectively.We investigate the effect of PGSL parameters on I-V characteristic and appearance of multipeak negative differential resistance (NDR) in proposed device within the Landauer-Buttiker formalism and adopted transfer matrix method. Moreover‚ we propose a novel venue to control the NDR in PGSL with Fermi velocity barrier. Different regimes of NDR have been recognized based on PGSL parameters and external bias. From this viewpoint‚ we obtain multipeak NDR through miniband aligning in PGSL. Maximum pick to valley ratio (PVR) up to 167 obtained for ߭௖, Fermi velocity correlation (ratio of Fermi velocity in barrier and well region), of 1.9 at bias voltages between 70-130 mV. While our findings have good agreement with those of experiments can be considered in multi-valued memory‚ functional circuit, low power and high- speed nanoelectronic devices application. Keywords: Negative differential resistance (NDR); Planer graphene superlattices (PGSLs); Fermi velocity barrier; Dirac gap PACS: 72.80.Vp, 68.65.Cd, 73.22.Pr *Corresponding author. E. mail addresses: [email protected], [email protected] (S. Shojaei). Tel: +98 4133392995, Fax: +98 4133347050, Skype: Saeed.shojaei I. Introduction Graphene, an atomically thin two-dimensional material, has attracted a great deal of interest due to its superior electrical and optical properties [1-3] which is become one of the most promising materials for nanoelectronic and nanophotonic devices [4-6]. The charge carriers in monolayer graphene behave like a massless Dirac fermions with linear energy dispersion, can be governed by an effective Dirac equation, which leads to many novel electronic and transport properties [7- 10]. However, due to the lack of energy gap in the pristine graphene electronic band structure, the Klein tunneling prevents Dirac electrons in graphene from being confined by an electrostatic potential, which impedes its use in electronic devices. In order to overcome this limitation, several schemes have been proposed which leads to suppression of Klein tunneling and confinement of the Dirac electrons in graphene. For instance, a energy gap can be induced by doping [7-9], substrate [10-12], strain [13-15], quantum confinement effects [16-19], spatial modulation of Dirac gap and external periodic potentials [20-23].The Fermi velocity is another key concept to study electronic properties of graphene. According to Lorentz invariant theory‚ by controlling the electron-electron interaction‚ one can regulate Fermi velocity in graphene. This implements various procedures such as dielectric screening [24, 25]; carrier concentration [26- 28]; periodic potentials [29]; curvature of the graphene sheet [30] and substrate modulation [31]. Recently, graphene superlattice, graphene under periodic potentials, has been extensively studied both experimentally and theoretically [20-23, 32-38]. These works have considered periodic potentials of a different nature (electric & magnetic) and different profiles. This considerable interest is motivated by discovering novel transport properties of charge carriers through PGSL structures (e.g. extra Dirac points, highly anisotropic Dirac points) which has not been observed in the pristine graphene, and also by promising potential applications in a diverse area of nanoelectronic devices. Beyond the usual linear or saturation performances expected to occur in graphene based devices, an appearance of prominent non-linear effects such as the NDR in the current–voltage characteristics has attracted strong research interest both theoretically and experimentally [39- 51], since it could potentially impact the number of key applications such as high frequency oscillators, reflection amplifiers, memories, multi-level logic devices, and fast switches [52]. NDR at high bias voltages (1–2 V), in narrow nanoribbons have been studied in lots of works [46, 47]. NDR at low bias regime can also be achieved in monolayer and bilayer graphene, and graphene nanoribbon superlattices systems [48-50]. In present study, for a first time, we investigate the NDR features of graphene based planar superlattices composed of periodically modulated Fermi velocity, Dirac gap and electrostatic potential at low bias voltages VSD<200 mV , above mentioned effects are induced by means of velocity barrier, nanostructured substrate and one-dimensional potentials of square barriers, respectively. Transmission properties of Dirac fermion beams tunneling through PGSL are investigated by adopted transfer matrix method (TMM) in details. We have found that the transmission properties of PGSL can be tuned readily by changing the main parameters of the PGSL, i.e. well and barrier widths, energy and angle of the incident electrons, the number of periods of PGSLs, Dirac gap and Fermi velocity barrier. Current-voltage characteristics obtained within the Landauer-Buttiker formalism. We have found that the NDR appears with appropriate structural parameters, which means that NDR could be controlled and tuned by PGSL parameters. More interestingly‚ we investigated the effect of Fermi velocity barrier on I-V characteristics of PGSL resonant tunneling diode (PGSLRTD).We found that PVR in PGSLRTD is enhanced exponentially by increasing Fermi velocity correlation. II. Model & Method A schematic view of our proposed device is shown in Fig. 1. A monolayer of graphene is placed on planar heterostructured substrate composed of two different materials shown by I‚II. This composed substrate‚ opens different energy gaps in different regions of the graphene sheet, denoted by ∆ௐ and ∆஻.W, B refer to well and barrier, respectively. Moreover, according to Lorentz invariant theory‚ Fermi velocity differs in well & barrier regions of graphene placed on different materials of substrate [52]; thus‚ composed substrates can also modifies Fermi velocity in graphene. The Fermi velocity in each region will be denoted by ߭ௐand߭஻. We define ߭௖ = ߭஻/߭ௐis the Fermi velocity correlation between barrier and well region. Series of grounded metallic planes are placed over graphene sheet which induce a periodic velocity barrier. Metallic planes do electron-electron interaction weaker; therefore reduce Fermi velocity in the relevant region [25]. Metal electrodes in left and right create electrostatic potential across the graphene sheet acting as source-drain voltage,ܸ௕ = ܸோ −ܸ௅. The Kronig-Penney (KP) model is applied for investigation of PGSL electrostatic potential profile. In the vicinity of the Dirac points, the PGSL electronic structure can be described by the Dirac-like equation. The effective 2D Dirac Hamiltonian for a PGSL with a position dependent energy gap and Fermi velocity, ߭ி, is written as: H = −iℏቀඥv୊ሺxሻσ୶∂୶ඥv୊ሺxሻ+v୊ሺxሻσ୷∂୷ቁ+Vሺxሻ1෠ +∆ሺxሻσ୸ (1) Where ߪ௜are the Pauli matrices and 1෠ is the 2× 2 unitary matrix, ∆ሺݔሻ is the position dependent graphene energy gap,Vሺxሻis an external position dependent electrostatic potential that is composed of two parts: first, KP potential(VKP) that takes as zero in well region and 400meV in barrier region. Second part indicates the applied external potential bias that is taken as eEx (e is electron charge, E is electric field and x is growth direction of PGSL) ܸሺݔሻ = ቊ ܸ஻ሺݔሻ = ܸ௄௉஻ −݁ܧݔ ݂݋ݎ ܾܽݎݎ݅݁ݎ ܸௐሺݔሻ = ܸ௄௉ௐ −݁ܧݔ ݂݋ݎ ݓ݈݈݁, (2) Where indices W, B refer to well, barrier regions. According to Bloch's theorem, it is straightforward to obtain the electronic dispersion for periodic structure of PGSL by following equation as: ܿ݋ݏሺ݇௫݈ሻ = ܿ݋ݏሺ݇஻݀஻ሻܿ݋ݏሺ݇ௐ݀ௐሻ+ ௞೤మℏమ௩ಳ௩ೈିሺாି௏ಳሺ௫ሻሻሺாି௏ೈሺ௫ሻሻା∆ಳ∆ೈ Where ݇ௐ = ሺሺሺܧ−ܸௐሺݔሻሻଶ −∆ௐଶ ሻ/ ℏଶݒௐଶ −݇௬ଶሻଵ/ଶ‚݇஻ = ሺሺሺܧ −ܸ஻ሺݔሻሻଶ −∆஻ଶሻ/ ℏଶݒ஻ଶ − ݇௬ଶሻଵ/ଶ‚݇௬ = ሺܧݏ݅݊ߠ/ℏߥிሻଵ/ଶ, ߠis incident angle of electron beam that is defined as angle between growth direction of planar PGSL and direction of incidence,݀௪and ݀஻are the well and barrier width, respectively. We chooseܸ௄௉ௐ = 0,ܸ௄௉஻ = 400 meV, ߂஻ ≠ 0, ߂௪ = 0. ݏ݅݊ሺ݇஻݀஻ሻݏ݅݊ ሺ݇ௐ݀ௐሻ(3) ℏమ௩ಳ௩ೈ௞ಳ௞ೈ In order to adopt the well known TMM to the graphene superlattice under bias in this work, we consider m layers containing barrier and well with the potential mentioned in eq. 2 that can be rewritten as: Index m refers to number of layers (m=1,2,3,…N). VRL is difference of potential applied between two left and right end electrodes. ܸ௠ = ܸ௠௄௉ − ܸ݁ோ௅ݔ௠/ܮ. (4) The solution to Dirac equation, ܪ߰ ሺݔ,ݕሻ = ܧ߰ ሺݔ,ݕሻ is two component spinor that ߰௠ሺݔ,ݕሻ = ߰௠ሺݔሻ݁௜௞೤௬. (5) and definingඥߥிሺݔሻ௠߰௠ሺݔሻ = ߶௠ሺݔሻ, the solution to Dirac equation in x direction can be illustrates the two graphene sublattices reads: written as a linear combination of forward/backward plane-waves [55]: (6) ߶௠ଵሺݔ,ݕሻ = ටߥி௠ሺݔሻሺܣ௠݁௜௞೘௫ +ܤ௠݁ି௜௞೘௫ሻ݁௜௞೤௬, ߶௠ଶሺݔ,ݕሻ = ටߥி௠ሺݔሻሺܣ௠݁௜௞೘௫ା௜ఈ೘ +ܤ௠݁ି௜௞೘௫ି௜ఈ೘ሻ݁௜௞೤௬, where Amand Bm are the transmission amplitudes, and ݇௬ = ሺܧݏ݅݊ߠ/ℏߥிሻଵ/ଶ. ߠis incident angle of incidence. ߙ௠ = ݐܽ݊ିଵሺ݇௬/݇௠ሻ. The external potential in our work is linear with x. To of electron beam that is defined as angle between growth direction of planar PGSL and direction construct the transfer matrix for this structure we divided the linear potential to very small potential steps where transfer matrix of each small step in mth region can be written simply through continuity conditions between spinors as[56]: ܯ௝ = ቆ ݁௜௞೘ೕ ௫ ܯ௠ሺݔሻ = ෑܯ௝ሺݔሻ ௝ ݁௜௞೘ೕ ௫ା௜ఈ೘ −݁ି௜௞೘ೕ ௫ି௜ఈ೘ቇ (7) ݁ି௜௞೘ೕ ௫ Where, index j refers to jth division in mth region of PGSL. The transfer matrix of whole system, connects left electrode to right electrode, is obtained as: ݐ = ∏ ܯ௠ିଵሺݔ௠,௠ାଵሻ ܯ௠ାଵሺݔ௠,௠ାଵሻ ௠ (8) From the current density in mth region, ܬ௠ሺݔሻ= ߥி௠ሺݔሻ߮௠றሺݔሻߪ௫߮௠ሺݔሻ, The current flow must be conservative at left and right sides of PGSL that implies ܬ௅ሺݔሻ = ܬோሺݔሻ. From which we ܶ௅ோሺܧ,݇,ܸ௕ ሻ = ቚ௃ೃ௃ಽቚ = ௖௢௦ఏೃ ௖௢௦ఏಽݐଶ. (9) identify the transmission coefficient, TLR[43, 53].: In our model, the potential drop from source to drain follows a linear function. The zero- temperature ballistic net current as a function of bias is computed by the Landauer-Büttiker formalism [57- 58]: ×׬ గ/ଶ ିగ/ଶ ఓಽఓೃ ࣮ሺܧி,ߙ,߭௕ሻܿ݋ݏߙ ݀ߙ ܫ = ܫ଴׬ ݀ܧܧ Where ܫ଴ = 2ܹ݃݁/߭ிℎଶ ‚ g = 4 is the degeneracy of electron states in graphene, W is the sample width that we choose it as ݀஻ and ܸ௕ is applied voltage between the left and right electrode (ܸ௕௜௔௦ in Fig. 1)‚ a linear voltage along the x-direction defined as ܸ݁௕=ߤ௅ −ߤோ and ߤ௅(ߤோ) is the bias- (10) dependent local Fermi energy in the left (right) electrode. III. Results & Discussions Due to significant roles of device dimension and geometry in I-V characteristic, first, we analyze the effect of well and barrier width, ݀ௐ,݀஻, and number of periodicity, N, on NDR. In Fig. 2 we present I-V diagram of device for different values of ݀ௐ and certain value of݀஻. We consider well width bigger than barrier width, and N=10. As it can be seen from Fig. 2, for wider wells, correspondent current peaks values are smaller than that of thinner wells. Also, peaks are shifted to lower biases. This behavior can be attributed by red shifting of resonant energies with increasing well width. In the case of wider wells, lower biases are needed to align the resonant states. We found that transmission oscillations increases with increasing well widths (not shown here) that is because of increasing resonant states. On the other hand, increasing oscillations reduces the area of under transmission curve that subsequently reduces the current value for wider wells(Eq. 4).We found that for barrier width remarkably bigger than and comparable with well width, applying bias aligns resonant modes, so that resonant sequential resonant tunneling occurs and current increases to reach a peak. By increasing bias more, the resonant states are misaligned and strong tunneling suppression appears leads to in current decreasing and pronounced spike in I-V curve. Now, we turn our attention to the impact of N on NDR and discuss on its different distinct regimes in our system. Fig. 3.adepicts the transmission spectrum(t) of PGSL as function of N. From figure, it is observed that by increasing N and subsequently increasing interfaces, Fabry- Perot like interferences are increased and lots of resonant modes are created. Furthermore, resonant modes show up below and above the stop band for all values of N. In the case of thin barriers in Fig. 3.b, for small value of N as 2 current increases monotonically upon resonant modes are aligned and sequential resonant tunneling occurs. Higher external biases misalign theresonant modes and tunneling is completely suppressed results in decreasing current value and NDR appears as a peak (curve of N=10). This trend indicates classical regime. For large number of N (10 and more), resonant modes form superlattice minibands. At low biases the current is dominated by transmission across these minibands, called miniband regime. With increasing bias, resonant modes, (responsible for formation of minibands),are misaligned that breaks up the minibands into off resonant Wannier Stark ladders with suppressed transmission leading to current decreasing. By increasing bias, rungs of ladders from distinct minibands cross showing new resonant peaks in transmission and system encounter Wannier- Stark regimes. Such alternation process from miniband to Wannier-Stark regime is the reason of multi-peak NDR.In the other words, this trend of NDR for larger number of barriers originates from destructive-constructive interferences of Dirac fermions creating several peaks in transmission spectrum and I-V diagram. Hereafter, we consider the case of miniband and Wannier-Stark regime by taking number of barriers as N=10. It is worth to mention that Fabry-Pérot like resonances appear for non-perpendicular incident (i.e. ݇௬ ≠ 0 ሻ while for ݇௬ = 0 Klein tunneling suppresses resonant tunneling and NDR. Fig. 4 illustrates Dirac barrier gap (∆஻) effect on I-V characteristic in low bias regime. Based on our calculations, with increasing Dirac barrier gap, wider stop bands in transmission spectrum causes decreasing of current. On the other hand, increasing Dirac barrier gap inhibits Dirac carrier transport meaning that PVR increases. Also, multi peak trend appears for bigger Dirac barrier gap caused by resonant tunneling enhancement. This finding has a good agreement with that of Song et al and Sollner et al [59, 60]. Appropriate structural parameters to reach maximum NDR obtained as ߭௖ = 1.9, Vb=70-130mV, N=10, dW=40nm and dB=5nm.Considerable value of PVR as 167 obtained for Dirac barrier gap of 150meV that is very good value for optimal performance of RTD based on PGSL that can be realized experimentally. In order to compare with other works done on graphene based NDR, we consider reference 59 where Song et al reported maximum value of PVR as almost 2.They used double barrier RTD based on graphene strips to obtain PVR. The advantage of our model is obtaining very big PVR in ultra-low bias regime by the means of PGSL. In our work variety of parameters have been taken into account to control NDR and obtain high value of PVR that NDR appears for suitable values of them. For example, very high value of barrier width prohibits the tunneling of electrons and current increases with increasing bias voltage linearly leading to prohibition of NDR appearance. Optimization on parameters should be done to find high value of PVR One of key parameters in I-V characteristic is the ratio of Fermi velocity in barrier/well regions (߭௖). It is very important to study the effect of this parameter to control NDR. In Fig.5(a) we report the transmission coefficient spectrum for different values of߭஼. It can be seen that transmission is enhanced for special value of ߭஼ and energies. Closer view of Fig 5.a determines of ߭஼. In contrast, at higher values of ߭஼ oscillation diminishes and energy gap is increased. Fig. 5(b) depicts that big values of current are obtained for smaller߭஼. This finding can be explained that oscillatory behavior in transmission spectrum becomes more pronounced at specific values by, transmission spectrum in Fig. 5(a) where larger area under transmission curve can be observed that leads to bigger current (Eq.4). Two main peaks are observed in low bias regime around 70mV and 130mVthat implies to NDR appearance. Such phenomenon originate from this fact that bigger ߭஼creates higher Fermi velocity barrier so that it is difficult for Dirac electrons to be able to tunnel from them leading to reduction of current. NDR is resulted by resonant tunneling occurring in miniband and Wannier-Stark regime as discussed before. To get closer insight, we focus on the trends of PVRs related to two main peaks versus ߭஼,݀ݓ,ܰ, ∆஻in Fig.6. It is clear that PVR's values are increased with increasing these parameters except than N. Fig. 6.c illustrates that increasing number of periods, N, increases PVR1 but decreases PVR2. Optimum values of structural parameters to enhance the PVR at low biases can be obtained from Fig.6. Our results show that among structural parameters ߭஼, has remarkable effects on the value of PVR1 and PVR2, hence it is very important in characterization of electronic device through cut-off frequency which is fundamental parameter in high frequency PGSLRTD based oscillators and switches. Our obtained high value of PVR at low biases is desired for implementation innovel low power digital logic circuits and multi-valued memory. Finally, in order to verify our results, we compared our findings with those of Antonova et. al[61].They investigated NDR in the films fabricated from partially fluorinated graphene suspension. The formation of Graphene islands (quantum dots) are observed in these films. Various types of negative differential resistance (NDR) and a step-like increase in the current are found for films created from the fluorinated graphene suspension. NDR resulted from the formation of the potential barrier system in the film and graphene quantum dots, was reported in their study. In our study, we model this system by quantum well/barrier systems according to our theoretical framework. I-V characteristic found by our calculation has very good agreement with I-V characteristic measured by them in low biases. IV. Conclusion In summary, we theoretically studied the I-V characteristics of planar graphene superlattice modeled by patterning graphene on nanostructured substrate and top grounded metal planes. To reach this goal, we adopted transfer matrix method to obtain transmission spectrum under external bias. The effects of parameters such as number of periods, N, well and barrier width, ݀ௐ,݀஻ and ratio of Fermi velocity in each region, ߭஼, Dirac barrier and electrostatic barrier have been analyzed in details to enhance peak to valley ratio, PVR, in I-V curve at very low external biases through within the Landauer-Buttiker formalism. Different regimes of NDR were discussed according to parameters used in our model device. Appropriate structural parameters to reach maximum NDR obtained as ߭஼ =1.9, Vb=70-130 mV, N=10, dw= and db=5 nm. Robust value of PVR as 167 obtained for Dirac barrier gap of150 meVat ultra-low biases. Our proposed device paves a novel way in the PGSL based NDR devices such as RTDs, multilevel memory‚ low power and high-speed electronic devices working at low bias. Refrences [1] K. S. Novoselovet al., Nature, vol. 438, no. 7065, pp. 197–200, Nov (2005). [2] S. Baeet al., Nature Nanotechnol, vol. 5, no. 8, pp. 574–578, Jun (2010). [3] A. K. Geim and K. S. Novoselov, Nature Mater. 6, 183(2007). [4] A. H. Castro Neto, F. Guinea, N. M. R. Peres, K. S. Novoselov, and A. K. Geim, Rev. Mod. Phys. 81, 109 (2009). [5] Bonaccorso, F., Sun, Z., Hasan, T. & Ferrari, A. C, Nature Photon. 4, 611–622 (2010). [6] Xia, F.; Wang, H.; Jia, Y. Nature communications, 5, 4458( 2014). [7] J. Gebhardt, R. J. Koch, W. Zhao, O. Hofert, K. Gotterbarm, S. Mammadov, C. Papp, A. Gorling, H.-P. Steinruck, and T. Seyller, Phys. Rev. B 87, 155437 (2013). [8] T. B. Martins, R. H. Miwa, A. J. R. da Silva, and A. Fazzio, Phys. Rev. Lett. 98, 196803 (2007). [9] H. Wang, T. Maiyalagan, and X. Wang, ACS Catalysis 2, 781 (2012). [10] G. Giovannetti, P. A. Khomyakov, G. Brocks, P. J. Kelly, and J. van den Brink, Phys. Rev. B 76, 073103 (2007). [11] V. V. Ilyasov, B. C. Meshi, V. C. Nguyen, I. V. Ershov, and D. C. Nguyen, The Journal of Chemical Physics 141, 014708 (2014). [12] 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,Nat Mater 6, 770 (2007). [13] V.M. Pereira, A.H. Castro Neto, Phys. Rev. Lett. 103 046801 (2009). [14] T. Low, F. Guinea, M.I. Katsnelson, , Phys. Rev. B 83 195436 (2011). [15] F. Guinea , Solid State Commun. 152(15) 1437–1441 (2012). [16] G. Giavaras and F. Nori, Phys. Rev. B 85, 165446 (2012). [17] A. De Martino, L. Dell'Anna, and R. Egger, Phys. Rev. Lett. 98, 066802 (2007). [18] G. Giavaras, P. A. Maksym, and M. Roy, Journal of Physics: Condensed Matter 21, 102201 (2009). [19] Y. W. Son, M. L. Cohen, and S. G. Louie, Phys. Rev. Lett. 97, 216803(2006). [20] N. M. R. Peres, J.Physics: Condensed Matter 21, 095501 (2009). [21] G. Giavaras and F. Nori, Appl. Phys. Lett97, 243106 (2010). [22] M. Barbier, P. Vasilopoulos and F. M. Peeters, Philos. Trans. R. Soc. London Ser. A 368, 5499(2010). [23] Park C-H, Yang L, Son Y-W, Cohen M L and Louie S G Nat. Phys. 4 213 (2008). Park C-H, Yang L, Son Y-W, Cohen M L and Louie S G, Phys. Rev. Lett.101126804(2009). Park C-H, Yang L, Son Y-W, Cohen M L and Louie S G, Phys. Rev. Lett.103046808 (2009). [24] Siegel, D. A. et al. Many-body interactions in quasi-freestanding graphene. Proc. Natl. Acad. Sci. USA 108, 11365–11369 (2011). [25]Raoux, A. et al, Phys. Rev. B 81, 073407(2010). [27] Elias, D. C. et al,.Nat. Phys. 7, 701–704(2011). [28] Bostwick, A. et al, Solid State Commun. 143, 63–71(2007). [29] Li, Z. Q. et al, Nat. Phys. 4, 532–535 (2008). [30] Park, C.-H., Yang, L., Son, Y.-W., Cohen, M. L. & Louie, S. G, Nat. Phys. 4, 213–217 (2008). [31] C. Hwang, D. A. Siegel, S.-K. Mo, W. Regan, A. Ismach, Y. Zhang, A. Zettl, and A. Lanzara, Sci. Rep. 2, 590 (2012). [32] Brey L and Fertig H A, Phys. Rev. Lett. 103046809 (2009). [33] Le V Q, Pham C H and Nguyen V L J. Phys.: Condens. Matter 24345502 (2012). [34] X.X. Guo,D.Liu,Y.X.Li, Appl.Phys.Lett. 98242101 (2011). [35] P.L.Zhao,X.Chen, Appl.Phys.Lett. 99182108 (2011). [36] T.X.Ma,L.Chun,L.G.Wang,H.Q.Lin, Appl.Phys.Lett.100252402 (2012). [37] M. Esmailpour,A.Esmailpour,R.Asgari,E.Elahi,M.R.RahimiTabar, SolidStateCommun. 150655 (2010). [38] J. R. F. Lima, J. Appl. Phys. 117, 084303 (2015).J. R. F. Lima, Physics Letters A 379, 179 (2015). [39] D. Dragoman and M. Dragoman, Appl. Phys. Lett. 90, 143111 (2007); V. Nam Do, ibid. 92, 216101 (2008); V. Nam Do, V. Hung Nguyen, P.Dollfus, and A. Bournel, J. Appl. Phys. 104, 063708 (2008). [40] Z. F. Wang, Q. Li, Q. W. Shi, X. Wang, J. Yang, J. G. Hou, and J. Chen, Appl. Phys. Lett. 92, 133114 (2008). [41] H. Ren, Q.-X. Li, Y. Luo, and J. Yang, Appl. Phys. Lett. 94, 173110 (2009); H. Fang, R.-Z.Wang, S.-Y.Chen, M. Yan, X.-M.Song, and B.Wang, ibid. 98, 082108 (2011). [42] V. Nam Do and P. Dollfus, J. Appl. Phys. 107, 063705 (2010); V. Hung Nguyen, A. Bournel, and P. Dollfus, ibid. 109, 093706 (2011); V. Hung Nguyen, F. Mazzamuto, J. Saint-Martin, A. Bournel, and P. Dollfus, Appl. Phys. Lett. 99, 042105 (2011). [43] G. J. Ferreira, M. N. Leuenberger, D. Loss, and J. C. Egues, Phys. Rev. B 84, 125453 (2011) [44] Y. Wu, D. B. Farmer, W. Zhu, S.-J.Han, C. D. Dimitrakopoulos, A. A. Bol, P. Avouris, and Y.-M.Lin, ACS Nano 6, 2610 (2012). [45] R. Sharma, et al, ACS Nano 9 (1), 620-625 (2015). [46] Z. F. Wang, Q. Li, Q. W. Shi, X. Wang, J. Yang, J. G. Hou, and J. Chen, Appl. Phys. Lett. 92, 133114 (2008). [47] V. N. Do and P. Dollfus, J. Appl. Phys. 107, 063705 (2010). [48] H. Ren, Q-X. li, Y. Luo, and J. Yang, Appl. Phys. Lett. 94, 173110 (2009). [49] K. Habib, F. Zahid, and R. Lake, Appl. Phys. Lett. 98, 192112 (2011). [50] H. Fang, R. Wang, S. Chen, M. Yan, X. Song, and B. Wang, Appl. Phys. Lett. 98, 082108 (2011). [51] Y. Zhao, Z. Wan1, X. Xu, S. R. Patil, U. Hetmaniuk3 & M. P. Anantram. Sci. Rep. 5, 10712; doi: 10.1038/ srep10712 (2015). [52] Mizuta H and Tanoue T 1995, The Physics and Application of Resonant Tunnelling Diodes (Cambridge: Cambridge University Press) [53] M. Born and E. Wolf, Principles of Optics: Electromagnetic Theory of Propagation, Interference and Diffraction of Light (Pergamon Press, Oxford, 1964). [54] Y. M. Blanter and M.Buttiker, Phys. Rep. 336, 1 (2000). [55] Bai, C., Zhang, X, Phys. Rev. B 76, 075430 (2007). [56] A, Harwit, J. S. Harris, Jr. A. Kapitulnik, J. Appl. Phys, 60 (9),3211 (1986) [57] S. Datta, Electronic Transport in Mesoscopic Systems (Cambridge University Press, Cambridge, England, (1997). [58]. A. V. Andreev, PRL. 99, 247204 (2007). [59] Y. Song, H.C. Wu and Y. Guo, Appl. Phys. Lett. 102, 093118 (2013). [60]T. C. L. G. Sollner, W. D. Goodhue, P. E. Tannenwa1d, C. D. Parker, and D. D. Peck, Appl. Phys. Lett. 43, 588 (1983); A. R. Bonnefoi, R. T. Collins, T. C. McGill, R. D. Burnham, and F. A. Ponce, ibid. 46, 285 (1985). [61] I. V. Antonova, I. I. Kurkina, N. A. Nebogatikova, A. I. Komonov, S. A. Smagulova, Nanotechnology 28 074001 (2017) . Figure caption Fig.1. (Color online) (a) Schematic view of Dirac-cone distribution and band-edge profile as Kronig-Penney model in typical PGSL in the presence of electric field. ∆௚ shows gap opened in barrier/well regions, shown in our calculation as ∆௪, ∆௕ ,imposed by periodic substrate(b) Schematic diagram of the device containing top metal plates ( shown with yellow color) that create Fermi velocity barrier. Left drain and source electrodes labeled with left and right electrode respectively. L and W are length and width of device.݀ௐ,݀஻ are well, barrier width ,respectively. Fig. 3 (Color online) (a) Map of Transmission spectrumfor different values of N. Incident angle, Fig. 2. (Color online) I-V characteristic for three values of well width,݀ௐ, for ݀஻ = 5 ݊݉, N=10 and ∆஻= 150 ܸ݉݁. ߠ, is ݌݅/6. (b) I-V characteristics for different values of N. ݀ௐ = 40݊݉ , ∆஻= 150 ܸ݉݁ and ݀஻ = 5 ݊݉. Fig. 4. (Color online) The effect of gap opening in barrier region, ∆஻, on I-V characteristic. N=10,݀ௐ = 40݊݉,݀஻ = 5 ݊݉. correlation velocity, ߭஼ for incident angle ofߠ = ݌݅/6. (b) I-V characteristics for different correlation velocity. ∆஻= 150 ܸ݉݁, N=10, ݀ௐ = 40 ݊݉ and ݀஻ = 5 ݊݉. (a)߭௖, (b) dW, (c)N and (d)∆஻. Constant parameters in each plot are considered as ∆஻= 150 ܸ݉݁, N=10, ݀ௐ = 40 ݊݉ and ݀஻ = 5 ݊݉. Fig. 6: (Color online) Variation of PVR1 and PVR2 for various structural parameters of PGSL Fig. 5. (Color online) (a) Counter plot of transmission spectrum for different values of Fig1: Fig. 2: Fig. 3: (a) Fig. 4: Fig. 5: Fig. 6:
1802.10253	5	1802	2018-09-24T08:26:39	Theory of metal-insulator transitions in graphite under high magnetic field	[ "cond-mat.mes-hall", "cond-mat.dis-nn", "cond-mat.str-el" ]	Graphite under high magnetic field exhibits consecutive metal-insulator (MI) transitions as well as re-entrant insulator-metal (IM) transition in the quasi-quantum limit at low temperature. In this paper, we identify the low-$T$ insulating phases as excitonic insulators with spin nematic orderings. We first point out that graphite under the relevant field regime is in the charge neutrality region, where electron and hole densities compensate each other. Based on this observation, we introduce interacting electron models with electron pocket(s) and hole pocket(s) and enumerate possible umklapp scattering processes allowed under the charge neutrality. Employing effective boson theories for the electron models and renormalization group (RG) analyses for the boson theories, we show that there exist critical interaction strengths above which the umklapp processes become relevant and the system enter excitonic insulator phases with long-range order of spin superconducting phase fields ("spin nematic excitonic insulator"). We argue that, when a pair of electron and hole pockets get smaller in size, a quantum fluctuation of the spin superconducting phase becomes larger and destabilizes the excitonic insulator phases, resulting in the re-entrant IM transitions. We also show that an odd-parity excitonic pairing between the electron and hole pockets reconstruct surface chiral Fermi arc states of electron and hole into a 2-dimensional helical surface state with a gapless Dirac cone. We discuss field- and temperature-dependences of in-plane resistance by surface transports via these surface states.	cond-mat.mes-hall	cond-mat	Theory of metal-insulator transitions in graphite under high magnetic field Zhiming Pan,1, 2 Xiao-Tian Zhang,1, 2 and Ryuichi Shindou1, 2, ∗ 1International Center for Quantum Materials, School of Physics, Peking University, Beijing 100871, China 2Collaborative Innovation Center of Quantum Matter, Beijing 100871, China (Dated: September 25, 2018) Graphite under high magnetic field exhibits consecutive metal-insulator (MI) transitions as well as re-entrant insulator-metal (IM) transition in the quasi-quantum limit at low temperature. In this paper, we identify the low-T insulating phases as excitonic insulators with spin nematic orderings. We first point out that graphite under the relevant field regime is in the charge neutrality region, where electron and hole densities compensate each other. Based on this observation, we introduce interacting electron models with electron pocket(s) and hole pocket(s) and enumerate possible umk- lapp scattering processes allowed under the charge neutrality. Employing effective boson theories for the electron models and renormalization group (RG) analyses for the boson theories, we show that there exist critical interaction strengths above which the umklapp processes become relevant and the system enter excitonic insulator phases with long-range order of spin superconducting phase fields ("spin nematic excitonic insulator"). We argue that, when a pair of electron and hole pockets get smaller in size, a quantum fluctuation of the spin superconducting phase becomes larger and destabilizes the excitonic insulator phases, resulting in the re-entrant IM transitions. We also show that an odd-parity excitonic pairing between the electron and hole pockets reconstruct surface chiral Fermi arc states of electron and hole into a 2-dimensional helical surface state with a gapless Dirac cone. We discuss field- and temperature-dependences of in-plane resistance by surface transports via these surface states. I. INTRODUCTION Graphite under high magnetic field exhibits a metal- insulator transition at low temperature (H ≥ Hc,1 (cid:39) 30 T) [1, 2]. The transition has been often considered as a prototype of one-dimensional Peierls density-wave in- stability associated with the 2kF logarithmic singularity in the Lindhard response function [3 -- 12]. A transition temperature Tc of the density wave ordering is deter- mined by a BCS type gap equation, ln Tc ∝ −1/ρ(0). The density of states at the Fermi level ρ(0) is pro- portional to the magnetic field H, so that Tc increases monotonically in the magnetic field [3 -- 6]. Further ex- periments discovered that graphite shows another metal- insulator transition (H ≥ H0 (cid:39) 53 T) [13 -- 21] as well as an insulator-metal re-entrant transition at higher mag- netic field (H = Hc,2 (cid:39) 75 T) [17 -- 21]. So far, there exist at least two distinct low-temperature insulating phases in graphite under high magnetic field: one insulating phase ranges in Hc,1 < H < H0 and the other ranges in H0 < H < Hc,2. The re-entrant transition at H = Hc,2 indicates a presence of a normal metal phase with pris- tine electron and hole pockets above the transition field, bringing about a skepticism against the density wave sce- narios. Namely, the transition temperature of the den- sity wave phase would increase monotonically in the field, until the electron and hole pockets that would form the Peierls density wave leave the Fermi level [3 -- 6]. Theoretically, stabilities of the Peierls density wave phases against random single-particle backward scatters depend crucially on a commensurability condition of an ∗ [email protected] electron filling [25 -- 28]. From preceding ab-initio band calculations of graphite under high magnetic field [8, 19], a sequence of specific values of the field in a range of 30 T (cid:46) H (cid:46) 50 T satisfy the commensurability condition. Nonetheless, experimental transition temperatures of the two insulating phases do not show any dramatic sensitiv- ities on certain values of the field in the range. Both of the insulating phases range rather broadly in field (over 20 Tesla) [17 -- 21]. In this paper, we explain these two low-T insulating phases in graphite under the high field as manifestation of excitonic insulators with spin nematic orderings. We first argue that graphite under high magnetic field (H (cid:38) 20T) is in the charge neutrality region, where electron den- sity and hole density compensate each other. Based on this observation, we begin with interacting electron mod- els with electron pockets and hole pockets, to enumerate possible umklapp scattering processes allowed under the charge neutrality condition. Using perturbative renor- malization group (RG) analyses on their effective boson theories, we show that the umklapp terms have critical interaction strength above/below which they become rel- evant/irrelevant on the renormalization. Above the criti- cal interaction strength, the umklapp term locks the total displacement field as well as spin superconducting phase field. The former locking causes the insulating behavior along the field direction, while the latter results in a long- range order of spin quadrupole moment. We explain the re-entrant insulator-metal transition in graphite, through a quantum fluctuation of the spin superconducting phase field. We characterize the spin nematic excitonic insula- tor phases by out-of-plane (infrared optical) conductivity as well as in-plane transport property [out-of-plane cur- rent is parallel to the field]. The field and temperature 8 1 0 2 p e S 4 2 ] l l a h - s e m . t a m - d n o c [ 5 v 3 5 2 0 1 . 2 0 8 1 : v i X r a 2 a result, the RPA density correlation function is char- acterized by the Lindhard response function in the one- dimension [3, 4]. The function has the logarithmic singu- larity at kz = 2kF , where 2kF is a distance between the left and right Fermi points in the same energy band [22]. Thereby, the system has a generic instability toward the charge density wave ordering, that breaks the spatially translational symmetry along the field direction [3 -- 12]. Meanwhile, graphite under the relevant field regime has four bands that run across the Fermi level (two elec- tron pockets and two hole pockets; H ≤ H0 (cid:39) 53 T) or two bands (one electron pocket and one hole pocket; H0 ≤ H) [8, 19]. When each of these four (or two) bands would undergo the Peierls density wave (DW) instabil- ities individually, the respective instabilities would ap- pear at different critical temperatures or critical fields. In other words, the graphite transport experiment would observe a step-wise increase of the (out-of-plane) resis- tance Rzz on lowering temperature or on increasing the magnetic field. Nonetheless, the graphite experiment observed a di- rect phase transition from high-T normal metal phase to the low-T insulating phase [16 -- 18, 20, 21]. Around the transition, the resistance along the field direction Rzz continuously increases [16 -- 21] and it increases by 100 times within windows of several Kelvin or Tesla [16 -- 18, 21]. These experimental observations clearly dictate that all the energy bands (pockets) at the Fermi level are gapped out simultaneously at the transition point. Thereby, the key question to be asked here is; what is a "talking-channel" among these four (or two) bands that enables this direct metal-insulator transition ? In this paper, we consider this channel as umklapp scat- tering terms and construct a mean-field theory that ex- plains this direct metal-insulator transition. To be more specific, all the excitonic insulator phases discussed in this paper are stabilized by the umklapp terms that lock a total displacement field along the field direction, a sum of the displacement fields of the four (or two) bands. When the umklapp terms become relevant in the standard RG argument sense, the total displacement field (electric po- larization) is locked, resulting in the electrically insulat- ing behaviour along the field direction. By calculating an optical conductivity along the field direction, we ex- plicitly demonstrate the presence of finite mobility gaps in the excitonic insulator phases. FIG. 1. (color online) Theoretical phase diagram for graphite under high magnetic field. The phase diagram is obtained from the RG equations, Eqs. (42,43,44) for H < H0 and Eqs. (67,68,69) for H0 < H < H1. "SNEI-I" and "SNEI- II" stand for two distinct spin nematic excitonic insulator phases (strong-coupling phase). For H < H0, the electronic state near the Fermi level comprises of two electron pockets (n = 0 LL with ↑ spin and ↓ spin) and two hole pockets (n = −1 LL with ↑ spin and ↓ spin). At H = H0, the outer two pockets (n = 0 LL with ↑ spin and n = −1 LL with ↓ spin) leave the Fermi level. For H0 < H < H1, the elec- tronic state has one electron pocket (n = 0 LL with ↓) and one hole pocket (n = −1 LL with ↑). We choose H0 = 50 T and H1 = 120 T. For a detailed parameter set of the RG equations, see Appendix C4. Our theory may not be able to predict much about a transition between SNEI-I and SNEI- II phases (a shaded area around H = H0); see a discussion in Sec. XB. T = 0 metal-insulator transition at H = Hc,1 and insulator-metal transition at H = Hc,2 are the quantum phase transition with the dynamical exponent z = 1. dependences of the transport properties are consistent with experimental observation in graphite. A. issues to be addressed in this paper 1. direct metal-insulator transition 2. re-entrant insulator-metal transition Under the magnetic field H ((cid:107) z), kinetic energy part of the three-dimensional semimetal takes a form of de- coupled one-dimensional quantum chains (or quantum wires). Namely, the kinetic energy within the xy plane is quenched by the Landau quantization, while the kinetic energy along the field direction remains intact, form- ing one-dimensional momentum-energy dispersion. As The second issue is the re-entrant insulator-metal tran- sition observed at the higher field region in the graphite experiment [14 -- 21], that can hardly be explained by the conventional Peierls DW scenarios. Namely, the RPA density correlation function at finite temperature sug- gests that the transition temperature of the Peierls DW phase increases monotonically in the field, until the elec- tron and/or hole pockets that would form the DW leave kzEEzkzEEzSNEI-ISNEI-IIH0Hc,1Hc,2H10,↑0,↓−1,↑−1,↓0,↓−1,↑204060800.00.20.40.60.8H(T)T(meV)the Fermi level. When they leave the Fermi level, how- ever, the electronic state simply ends up in semiconduc- tor phase rather than metallic phase. Contrary to this, the graphite experiments clearly observe the insulator- metal re-entrant transition in the higher-field region. The low-T electric transport along the field direction above the critical field (H > Hc,2 (cid:39) 75 T) is as metallic as the electric transport in the high-T normal metal phase [17, 18, 21]. The experiment clearly indicates a presence of pristine electron and hole pockets at the Fermi level above the critical field. In this paper, we explain this re-entrant insulator- metal transition as a consequence of quantum spin fluctu- ation enhanced by raising the magnetic field. To be more specific, we first point out that the umklapp terms lock not only the total displacement field but also a spin super- conducting phase field, a difference between a supercon- ducting phase field of an electron/hole pocket with ↑ spin and hole/electron pocket with ↓ spin respectively. The higher magnetic field makes the electron and hole pock- ets to be smaller in size in the kz space. In the presence of the repulsive electron-electron interaction, the smaller pockets make their Luttinger parameters to be smaller than the unit. Smaller Luttinger parameters mean larger quantum fluctuation of superconducting phase field as well as the spin superconducting phase field. Thus, we can naturally argue that, in the presence of such smaller electron and hole pockets, the umklapp terms suffer from the enhanced quantum spin fluctuation, and become ir- relevant in the RG argument sense. When the umklapp terms become irrelevant, the spin superconducting phase field as well as the total displacement field are unlocked, resulting in the re-entrant insulator-metal transition. Im- portantly, the electronic state still possesses electron and hole pockets above the critical field, though their sizes in the kz space might be small. 3. field-dependence of in-plane resistance The third issue to be addressed in this paper is an unusual field-dependence of the electric transport in the directions transverse to the magnetic field [14 -- 21]. Gen- erally, the bulk electric transport perpendicular to the field is quenched in the clean limit at low temperature (T (cid:28) hω0; hω0 is the cyclotron frequency). Nonetheless, the system still has low-T electric transport perpendicu- lar to the field through the so-called surface chiral Fermi arc (SCFA) states [23, 24]. The associated surface re- sistance is inversely proportional to a length of the arc in the kz space. The length is approximately equal to the size of the respective electron (or hole) pocket in the bulk. The size of the pocket generally decreases in the field. Thereby, the surface resistance perpendicular to the field is expected to increase in the field. Contrary to this theory expectation, the in-plane resistance Rxx in the graphite under the field (H (cid:107) z) shows an un- usual field-dependence. The low-T resistance Rxx shows 3 a broad peak around 15 T < H < 30 T [1, 2, 5, 6, 13 -- 21]. From H = 30 T to H = H0 (cid:39) 53 T, Rxx typically re- duces by half [14 -- 21]. Inside the high-field-side insulating phase (H0 < H < Hc,2 (cid:39) 75 T), the low-T in-plane resis- tance Rxx stays nearly constant in the field [17 -- 21]. For Hc,2 < H, Rxx starts increasing in the field again [21]. Field-(nearly) independent and metallic Rxx in the high-field-side insulating phase can be naturally ex- plained by a novel surface reconstruction of the surface chiral Fermi arc (SCFA) states due to the excitonic pair- ing in the bulk. To be more specific, we will show that an odd-parity excitonic pairing between electron and hole pockets in the bulk reconstructs the SCFA state of elec- tron and that of hole into a (2 + 1)-d helical surface state with a gapless Dirac cone. Rxx through such a Dirac-cone surface state is determined by carrier density doped in the surface region, that is typically independent from the magnetic field. Namely, unlike 'decoupled' SCFA states of electron and hole, the reconstructed Dirac-cone sur- face state barely changes its shape as a function of the magnetic field. At the zeroth order approximation, the field only changes a 'depth' of a band inversion between electron and hole pockets, while the shape of the Dirac- cone surface state is mainly determined by the excitonic pairing strength inside the inverted band gap. Thereby, one can naturally expect that the surface resistance due to the reconstructed Dirac-cone surface state is much less field-dependent than that of the decoupled SCFA states of electron and hole. B. structure of the paper The structure of the paper is as follows. In the next section with a help of appendix A, we argue that the graphite under the relevant field regime (20 T < H) is in the charge neutrality region, where electron and hole densities compensate each other. Based on this obser- vation, we enumerate in Sec. III possible umklapp terms that are allowed under the charge neutrality condition in the four pockets model (a model with two electron pock- ets and two hole pockets; H ≤ H0 (cid:39) 53 T). Employing a Hartree-Fock approximation, we construct effective field theories for possible insulating phases that can be stabi- lized by these umklapp terms (Sec. IV). There are three such phases; spin-nematic excitonic insulator, magnetic Mott insulator and plain excitonic insulator phases. Us- ing renormalization group (RG) analyses, we argue typi- cal field-dependences of the respective transition temper- atures of these three phases and conclude that the spin nematic excitonic insulator (SNEI-I) phase could natu- rally fit in the phenomenology of the low-field-side out-of- plane insulating phase (Hc,1 < H < H0) in the graphite experiment (Sec. V). In Sec. VI, we enumerate possible umklapp terms that are allowed under the charge neutral- ity condition in the two-pockets model (one electron and one hole pocket; H0 < H). We construct effective field theories for the possible insulating phases that can be sta- bilized by the umklapp terms. We found two such phases; a phase with two superposed charge density waves and a spin nematic excitonic insulator (SNEI-II) phase. Using the RG analyses, we conclude that the SNEI-II phase can naturally explain the high-field-side out-of-plane insulat- ing phase (H0 < H < Hc,2). In Secs. VII and VIII, we argue field-dependences of the in-plane resistance in the graphite experiment by the surface electric transports. Especially, we show in Sec. VIII that the odd-parity ex- citonic pairing in the two-pockets model reconstructs the surface chiral Fermi arc (SCFA) states of electron and hole into a (2 + 1)-d helical surface state with a gapless Dirac cone. The surface Dirac-cone state could naturally explain field-(nearly) independent and metallic behaviour of the in-plane resistance inside the high-field-side insu- lating phase. After a brief summary in Sec. IX, we give a discussion with complementary viewpoint (Sec. X). II. CHARGE NEUTRALITY REGIME IN GRAPHITE UNDER HIGH MAGNETIC FIELD Low-temperature transport properties of graphite are dominated by four π-orbital bands around zone bound- aries of the first Brillouin zone [32 -- 34]. Graphite is a three-dimensional AB stacking of graphene layers. A unit cell has two graphene layers and it has four inequiv- alent carbon sites. Call them as A, A(cid:48), B and B(cid:48). A and B share the same layer, and so do A(cid:48) and B(cid:48). A comes right above A(cid:48) in the cell. The electronic band structure near the Fermi level of graphite is composed by π orbitals of carbon atoms that are odd under the mirror with respect to the layer, for example, 2pz or- bital [32 -- 34]. π orbitals of A and A(cid:48) carbon atoms hy- bridize rather strongly, forming two π orbital bands at the zone boundaries that have large momentum-energy dispersions along the c-axis (4000 K). Call these π or- bitals as πA and πA(cid:48) respectively. π orbitals of B and B(cid:48) hybridize much weakly, as B and B(cid:48) locate right above the centers of the hexagon in their neighboring layers. These two, which we call πB and πB(cid:48) henceforth, form two degenerate bands at the zone boundaries that have a weaker energy-momentum dispersion along the c-axis (400 K). Under the field along the c-axis, the four bands in the zone boundaries are split into Landau levels (LLs) [8, 19, 34 -- 37]. For H (cid:38) 30 T, the n = 0 LLs with ↑ spin and ↓ spin form two electron pockets around kz = 0, and the n = −1 LLs with ↑ spin and ↓ spin form two hole pockets around kz = π/c0. Here c0 is a lattice constant along the c-axis. According to the band calculation, the outer electron pocket (n = 0 LL with ↑ spin) and the outer hole pocket (n = −1 LL with ↓ spin) leave the Fermi level at H = H0 (cid:39) 53 T. The Hall conductivity measurements in a regime of 20 T (cid:46) H (cid:46) 60 T [18, 29 -- 31] suggest that the number of the electron states and that of the hole states compen- sate each other almost completely. An estimation gives 4 Ne − Nh : Lz/c0 = 10−4 : 1, where Lz is a linear di- mension along the c-axis, Ne and Nh are numbers of the kz points within the electron pockets and hole pockets respectively [kz is a crystal momentum along the c-axis] (see Appendix A for a validity of the estimation). III. FOUR POCKETS MODEL (H < H0) Based on this observation, we consider an electron model with two electron pockets (n = 0 LL with ↑ spin and that with ↓ spins) and two hole pockets (n = −1 LL with ↑ spin and that with ↓ spins) with the charge neutrality condition (Ne = Nh); (cid:88) (cid:88) (cid:88) Hkin = n=0,−1 σ=↑,↓ kz,j En,σ(kz)c † (n,σ),j(kz)c(n,σ),j(kz). (1) FIG. 2. (color online) Schematic picture of electronic states of graphite under high field (H < H0). Solid/dotted lines de- scribe Fermi surfaces of two electron/hole pockets in both bulk and edge regions. Two electron/hole pockets in the bulk region are terminated by electron/hole-type surface chi- ral Fermi arc states in edge regions respectively. Namely, E0,σ(kz, yj)/E1,σ(kz, yj) goes higher/lower in energy, when yj goes from the bulk region to the edge regions (see Appendix A). edgeedgebulkc0π−+kF,0,↑kF,0,↓kF,−1,↑kF,−1,↓kF,0,↓kF,0,↑−−c0πkzyjkF,−1,↓kF,−1,↑−−−2Ly2Ly5 (2) En,σ(kz) = −2γ2 The two electron pockets encompass kz = 0 and two hole pockets are around the zone boundary kz = ±π/c0, (cid:2) cos(kzc0) − cos(kF,n,σc0)(cid:3), (cid:112)c/(eH). Since the kinetic energy with 0 < kF,0,↑ < kF,0,↓ < π/(2c0) < kF,−1,↑ < kF,−1,↓ < π/c0 [4]. An index j (= 1, 2,··· , (LxLy)/(2πl2)) counts degenerate electron states within each LL. l denotes a magnetic length, l ≡ part takes the exactly same one-dimensional momentum- energy dispersion along kz direction for different j, we regard the system as coupled chains and call j as a 'chain index' [38 -- 40]. The charge neutrality condition takes a form of A chirality index τ specifies left mover (τ = −1) or right mover (τ = +1). vF,a is a bare Fermi velocity of each pocket with a ≡ (n, σ). For simplicity, we label (n, σ) = (0,↑), (0,↓), (−1,↑) and (−1,↓) as a = 1, 2, 3 and 4 respectively throughout this paper, e.g. kF,a ≡ kF,n,σ, ca,j(kz) ≡ c(n,σ),j(kz), and ψa,±,j(z) ≡ ψ(n,σ),±,j(z). ψa,±,j(z) is a slowly-varying Fourier transform of those ca,j(kz) around kz (cid:39) ±kF,a; ψa,τ,j(z) ≡ 1 √Lz kz−τ kF,a<Λ ca,j(kz)ei(kz−τ kF,a)z. (5) A short-ranged repulsive interaction is considered; (cid:88) (cid:90) (cid:90) (cid:88) (cid:88) Hint = σ,σ(cid:48) c,d=A,A(cid:48),B,B(cid:48) † σ(cid:48)(r(cid:48) × ψ † σ(r, c)ψ dr dr(cid:48) V (r − r(cid:48)) , d)ψσ(cid:48)(r(cid:48) , d)ψσ(r, c), (6) kF,0,↑ + kF,0,↓ + kF,−1,↑ + kF,−1,↓ = 2π c0 . (3) where For low-temperature electric transports at those tem- perature much below the band width (2γ2 (cid:39) 400 K), the kinetic energy part can be linearized around the Fermi points of each pockets (kz (cid:39) ±kF,n,σ); (cid:90) (cid:88) (cid:88) j a,τ Hkin = τ vF,a dzψ † a,τ,j(z)i∂zψa,τ,j(z) + ··· . (4) V (r) ≡ g (√2π)3l0,xl0,yl0,z − z2 2l2 0,z −− x2 2l2 0,x − y2 2l2 0,y , e (7) r ≡ (x, y, z), σ, σ(cid:48) =↑,↓, and g > 0. l0,µ denotes an interaction length along the µ-direction. ψ† σ(r, c) denotes an electron creation at π-orbital πc (c = A, A(cid:48), B, B(cid:48)) of carbon atom at r with spin σ. The creation field can be expanded in term of single-particle bases of the n = 0 and n = −1 LLs in the Landau gauge;  ψσ(r, A) ψσ(r, A(cid:48)) ψσ(r, B) ψσ(r, B(cid:48))  = (cid:88) j eikj x √Lx (cid:26) γA,σY0,j(y) γA(cid:48),σY0,j(y) γB,σY1,j(y) γB(cid:48),σY1,j(y) (cid:88) τ =±  eiτ kF,0,σzψ(0,σ),τ,j(z) +  0 0 ηB,σY0,j(y) ηB(cid:48),σY0,j(y)  eiτ kF,−1,σzψ(−1,σ),τ,j(z) (cid:27) , (8) where Y0,j(y) ≡ 1(cid:112)√πl − (y−yj )2 2l2 e , (9) (cid:112)√πl3 √2(y − yj) − (y−yj )2 √2l d dyj , e 2l2 (10) Y0,j(y) = Y1,j(y) ≡ with yj ≡ kjl2 and kj ≡ 2πj/Lx. The slowly varying field ψ(n,σ),τ,j(z) ≡ ψa,τ,j(z) was defined in Eq. (5) with a ≡ (n, σ). γc,σ (c = A, A(cid:48), B, B(cid:48)) comprises an eigen- vector of a 4 by 4 SWM (Slonczewski-Weiss-McClure) Hamiltonian at kz = ±kF,0,σ [33 -- 37]. ηc,σ (c = B, B(cid:48)) comprises the eigenvector at kz = ±kF,−1,σ. Lx is a lin- ear dimension of the system size along the x-direction. A substitution of Eqs. (8,9,10) into Eq. (6) and expansion in ψ(n,σ),τ,j(z) ≡ ψa,j(z) lead to scatterings between dif- ferent pockets (inter-pocket scattering) and scatterings within the same pocket (intra-pocket scattering). In this paper, we take into consideration only umklapp scattering terms that are allowed under the charge neu- trality condition (Fig. 3), inter-pocket scattering terms between opposite chiralities (Fig. 4), and intra-pocket scatterings Hf . This is because, in fermionic functional renormalization group analyses [12, 44, 45], these scatter- ings are coupled with one another at the one-loop level; they have larger chances to become relevant upon the renormalization than those scattering terms omitted. Under the charge neutrality condition (Eq. (3)), the interaction allows the following four umklapp terms and their hermitian conjugates; † 3,+,j+m−nψ1,−,mψ2,−,j, † 3,+,j+m−nψ1,−,mψ4,−,j, † 1,+,j+m−nψ3,−,mψ2,−,j, † 1,+,j+m−nψ3,−,mψ4,−,j. † 4,+,nψ † 2,+,nψ † 4,+,nψ † 2,+,nψ  (cid:88) ψ ψ ψ ψ Hu = (11) j,m,n Due to the translational symmetry along x in the Lan- dau gauge, the scattering processes conserve a momen- tum kj ≡ 2πj/Lx that is conjugate to x. In eq. (11), integrals over the spatial coordinate z, and the scatter- ing matrix elements that depend on z and j, m, n = 1, 2,··· , (LxLy)/(2πl2) are omitted for clarity. For ex- ample, the first and fourth terms in Eq. (11) take the following explicit form with their Hermitian conjugates, (1st and 4th terms in Eq. (11)) = 2 V (12) m−n,j−n dz − (z−z(cid:48) )2 2l2 0,z (cid:48) e dz (cid:90) (cid:90) −ikF,3z−ikF,4z(cid:48)−ikF,2z(cid:48)−ikF,1z † 4,+,nψ † 3,+,j+m−nψ1,−,mψ2,−,j ψ † 2,+,nψ † 1,+,j+m−nψ3,−,mψ4,−,j + ψ (cid:17) (cid:27) + h.c. . (12) (cid:88) j,m,n (cid:26) e (cid:16) The matrix element in Eq. (12) are given by a dimen- sionless function f (12)(x, y) as V (12) m,n ≡ g Lx 1 2πl0,zl f (12)(ym/l, yn/l). (13) The function f (12)(x, y) can be calculated by the direct substitution of Eqs. (7,8,9,10) into Eq. (6). g is from Eq. (7). In addition to Hu, we consider the inter-pocket scat- terings between the opposite chirality (Hb) as well as the intra-pocket scatterings (Hf ). They are (cid:88) j,m,n Hb =  † 4,±,nψ † 3,±,nψ † 4,±,nψ † 3,±,nψ † 4,±,nψ † 2,±,nψ † 1,∓,j+m−nψ1,∓,mψ4,±,j, † 2,∓,j+m−nψ2,∓,mψ3,±,j, † 2,∓,j+m−nψ2,∓,mψ4,±,j, † 1,∓,j+m−nψ1,∓,mψ3,±,j, † 3,∓,j+m−nψ3,∓,mψ4,±,j, † 1,∓,j+m−nψ1,∓,mψ2,±,j, (14) ψ ψ ψ ψ ψ ψ and Hf = (cid:88) (cid:88) † a,nψ ψ † a,j+m−nψa,mψa,j, (15) a j,m,n with ψa,n(z) ≡ eikF,azψa,+,n(z) + e−ikF,azψa,−,n(z) (a = 1, 2, 3, 4; see Eq. (B1) for an actual form of Hf ). We do not take into account the inter-pocket scatterings be- tween the same chirality, because, at the one-loop level of the fermionic renormalization group equations [12, 44, 45], they are decoupled from Hu, Hb and Hf , and do not grow up into larger values upon the renormalization. IV. EFFECTIVE BOSON THEORY In this section, we construct effective field theories of possible insulating phases that are stabilized by the umk- lapp scattering terms in Eq. (11). To this end, we first 6 FIG. 3. Schematic pictures of one of the umklapp scatter- ing, Hu,2. As in Fig. 2, the vertical axis denotes the mo- mentum along the field direction (kz), while the horizontal axis denotes the chain index yj = kjl2 with kj ≡ 2πj/Lx (j = 1, 2,··· , LxLy/(2πl2)). The two-particle scatterings with solid/dotted arrows are the exchange processes (m = n) of the first/fourth terms in Eq. (11) with (0,↑) ≡ 1, (0,↓) ≡ 2, (−1,↑) ≡ 3 and (−1,↓) ≡ 4. assume that the low-T insulating phases in the graphite experiment do not break the translational symmetries within the graphene plane [the graphene plane is per- pendicular to the field (z)]. The assumption apparently does not contradict with any experimental observations in the past literatures [1, 2, 5, 6, 13 -- 21]. We thus introduce as 'mean fields' the pairings among electron creation/annihilation operators within the same chain, and treat the inter-chain electron-electron interactions within a Hartree-Fock approximation. To be more spe- cific, we keep only the direct process (Hartree term: j = n) and exchange process (Fock term: m = n) in Eqs. (11,14,15). Within such effective theory framework, the metal-insulator transitions in the graphite experi- ment are described by a phase transition between a phase with the mean fields being zero and a phase with the mean fields being finite. The former phase corresponds to the high-T normal metal phase and the latter corre- sponds to the low-T insulating phases. To do this construction transparently, we first bosonize the slowly-varying fermion field in terms of two phase 0,↑,L 0,↓,L−1,↑,L−1,↓,L 0,↑,R 0,↓,R−1,↑,R−1,↓,RHkin + Hu + Hb + Hf = H0 + Hu,i + Hamiltonian 4(cid:88) (cid:110) uaKaπ i=1 (cid:90) (cid:88) 4(cid:88) m a=1 H0 = 7 4(cid:88) i=1 Hb,i + ··· , (18) [Πa,m(z)]2 [∂zφa,m(z)]2(cid:111) , dz + 2 ua 2πKa (19) FIG. 4. Schematic pictures of one of the inter-pocket scat- terings, Hb,2. They are the exchange processes (m = n) of the first two terms in Eq. (14) with (0,↑) ≡ 1, (0,↓) ≡ 2, (−1,↑) ≡ 3 and (−1,↓) ≡ 4. As in Fig. 2, the vertical axis denotes the momentum along the field direction (kz), while the horizontal axis denotes the chain index yj = kjl2 with kj ≡ 2πj/Lx (j = 1, 2,··· , LxLy/(2πl2)). and πΠa,j(z) ≡ ∂zθa,j(z). Ka and ua are Luttinger pa- rameter and Fermi velocity of a pocket with a = (n, σ) that are renormalized by the intra-pocket forward scat- terings Hf (see appendix B for its details). As in the standard bosonization, the renormalizations are given by two parameters g2,a(> 0) and g4,a(> 0) as, (cid:115)(cid:16) (cid:115) ua vF,a = Ka = (cid:17)2 (cid:16) g2,a (cid:17)2 − 2πvF,a 1 + g4,a 2πvF,a , (20) 2πvF,a + g4,a − g2,a 2πvF,a + g4,a + g2,a . (21) variables defined for each pocket a = (n, σ) and each chain j = 1,··· , (LxLy)/(2πl2) [28, 41]; See also Appendix B for expressions of g2,a and g4,a in terms of g in Eq. (7). The Hartree and Fock terms in the umklapp scatterings of Eq. (11) are bosonized as (cid:40) ψa,+,j(z) ≡ ηa,j√ ψa,−,j(z) ≡ ηa,j√ 2πα e−i(φa,j (z)−θa,j (z)), 2πα e−i(−φa,j (z)−θa,j (z)). (16) Hu,1 = M (1) j−m Here (n, σ) = (0,↑), (0,↓), (−1,↑) and (−1,↓) are abbre- viated as a = 1, 2, 3 and 4 respectively. α is a short-range cutoff for the spatial coordinate z. φa,j(z), θa,j(z) and ∂zθa,j(z) are a displacement field along the field direc- tion (z), superconducting phase field, and current den- sity field along the field respectively. They are associated with the pocket a and the j-th chain. The displacement field and superconducting phase field cannot be simul- taneously definite; they are canonical conjugate to each other; Hu,2 = Hu,3 = [φa,j(z), ∂z(cid:48)θb,m(z (cid:48))] = iδa,bδj,mδ(z − z (cid:48)). (17) Hu,4 = ηa,j and ηa,j in Eq. (16) are Klein factors ensuring the anticommutation relation among fermion fields on dif- ferent chains (j), pockets (a) and chiralities (τ = ±); {ηa,j, ηb,m} = {ηa,j, ηb,m} = δa,bδj,m, and {ηa,j, ηb,m} = 0. Due to the Klein factor, the interaction parts given in Eqs. (11,14,15) cannot be fully bosonized without ap- proximation. To obtain the effective boson theories of the insulating phases, we thus employ the Hartree-Fock approximation for the inter-chain interactions in Eqs. (11,14,15), to keep only direct process (j = n) and exchange process (m = n) in Eqs. (11,14,15). This leads to a fully-bosonized j,m (cid:88) (cid:88) (cid:88) (cid:88) j,m j,m j,m dz dz +,j + Q14 +,j + Q24 (cid:90) (cid:110) σ31,jσ42,m cos(cid:2)Q13 (cid:3)(cid:111) + σ31,jσ42,m cos(cid:2)Q13−,j + Q24−,m (cid:90) (cid:110) σ32,jσ41,m cos(cid:2)Q23 (cid:3)(cid:111) + σ32,jσ41,m cos(cid:2)Q23−,j + Q14−,m (cid:90) (cid:110) σ31,jσ24,m cos(cid:2)Q13 (cid:3)(cid:111) + σ31,jσ24,m cos(cid:2)Q13−,j + Q24 (cid:90) (cid:110) σ34,jσ21,m cos(cid:2)Q34−,j + Q12 (cid:3)(cid:111) + σ34,jσ21,m cos(cid:2)Q34 +,j + Q12−,m +,m dz dz , +,j + Q24−,m (cid:3) (cid:3) (cid:3) (cid:3) +,m , (22) +,m , (23) , (24) +,m (25) M (2) j−m M (3) j−m M (4) j−m where Qab±,j ≡ φa,j + φb,j ± (θa,j − θb,j), (26) with a, b = 1, 2, 3, 4. σab,j and σab,m are Ising variables associated with the Klein factors within the same chain, σab,j ≡ iηa,jηb,j, and σab,m ≡ iηa,mηb,m. The Ising vari- ables take ±1. The Fock term (m = n) of the inter-pocket scatterings, 0,↑,L 0,↓,L−1,↑,L−1,↓,L 0,↑,R 0,↓,R−1,↑,R−1,↓,R 0,↑,L 0,↓,L−1,↑,L−1,↓,L 0,↑,R 0,↓,R−1,↑,R−1,↓,REq. (14), are bosonized as j,m j,m (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) j,m j,m j,m j,m Hb,13 = + Hb,2 = + Hb,4 = + H (13) j−m (13) j−m H H (2) j−m (2) j−m H H (4) j−m (4) j−m H , + + + dz dz dz (cid:90) (cid:110) σ31,jσ31,m cos(cid:2)∆jmQ13 (cid:3) + σ31,jσ31,m cos(cid:2)∆jmQ13−(cid:3)(cid:111) (cid:90) (cid:110) σ42,jσ42,m cos(cid:2)∆jmQ24 (cid:3) + σ42,jσ42,m cos(cid:2)∆jmQ24−(cid:3)(cid:111) (cid:90) (cid:110) (cid:3) σ32,jσ32,m cos(cid:2)∆jmQ23 + σ32,jσ32,m cos(cid:2)∆jmQ23−(cid:3)(cid:111) (cid:90) (cid:110) (cid:3) σ41,jσ41,m cos(cid:2)∆jmQ14 + σ41,jσ41,m cos(cid:2)∆jmQ14−(cid:3)(cid:111) (cid:90) (cid:110) σ34,jσ34,m cos(cid:2)∆jmQ34−(cid:3) (cid:3)(cid:111) + σ34,jσ34,m cos(cid:2)∆jmQ34 (cid:90) (cid:110) σ21,jσ21,m cos(cid:2)∆jmQ12 (cid:3) + σ21,jσ21,m cos(cid:2)∆jmQ12−(cid:3)(cid:111) dz dz dz + + + , , (27) (28) (29) with ∆jmf ≡ fj − fm. The Hartree term (j = n) of the inter-pocket scatterings in Eq. (14) could also renormal- ize the Luttinger parameters and Fermi velocities. For simplicity, however, we consider the renormalizations of Ka and ua only by the intra-pocket scatterings Hf as given in appendix B. Figs. 3 and 4 schematically show the inter-pocket scattering processes that lead to Hu,2 and Hb,2 respectively. The inter-chain interactions in Hu,i and Hb,i range over the magnetic length; M (n) j−m ≡ H (n) j−m ≡ j−m ≡ H (n) g g Lxα2lM(n)((yj − ym)/l), Lxα2lH(n)((yj − ym)/l), ((yj − ym)/l), Lxα2lH (n) g (30) (31) (32) with n = 1, 2, 3, 4 and 13. M(y) and H(y) as well as H(y) are dimensionless functions. For example, Eqs. (22,23,30) are obtained from the direct (j = n) and exchange pro- cesses (m = n) of Eq. (12) respectively with M(1)(x) = −e − 1 8 (kF,1+kF,3−kF,2−kF,4)2l2 0,z f (12)(x, 0) √2ππ2 , (33) M(2)(x) = e − 1 8 (kF,2+kF,3−kF,1−kF,4)2l2 0,z f (12)(0, x) √2ππ2 . (34) 8 i Hu,i +(cid:80) model, H0 +(cid:80) For the repulsive interaction case (g > 0), integrals of Eqs. (31,32) over y ≡ (yj − ym)/l give negative values for any n = 13, 2, 4. In the next section, we will use a perturbative renor- malization group (RG) analyses on the effective boson i Hb,i, where Hu,i and Hb,i are treated perturbatively (appendix C). We show that, at the one-loop level of the perturbative RG equations, Hb,13, Hu,1 and Hu,3 are coupled with one another and stabilize (what we call) a plain excitonic insulator phase. Meanwhile, Hu,2 and Hb,2 stabilize spin nematic exci- tonic insulator phase, Hu,4 and Hb,4 stabilize magnetic Mott insulator phase. V. RENORMALIZATION GROUP ANALYSES A. spin-nematic excitonic insulator (SNEI-I) phase We begin with the spin-nematic excitonic insulator phase stabilized by Hb,2 and Hu,2. The renormaliza- tion group (RG) equations for the inter-chain interaction functions in Hb,2 and Hu,2 take following forms at the one-loop level; M (2) j−m A23 + A14 2 M (2) j−nH (2) = (cid:88) n n−m − 2C14 (cid:88) n M (2) j−nH (2) n−m, (35) dM (2) j−m dlnb − 2C23 dH (2) j−m dlnb 1 − 2 dH (2) j−m dlnb 1 − 2 n (cid:88) n (cid:88) = A23H (2) j−m (cid:0)C14M (2) (cid:0)C23M (2) (2) j−m = A14H j−nM (2) n−m + 4C23H (2) j−nH (2) n−m j−nM (2) n−m + 4C14H (2) j−nH (2) n−m (36) (cid:1), (cid:1). (37) lnb > 0 is a scale change of the RG equations [see ap- pendix C for their derivations]. The temperature T in- creases monotonically on renormalization; dT /dlnb = T . A23, A14 and their linear combination are the scaling di- mensions of Hj−m, H j−m and Mj−m at the tree-loop level; (cid:88) c=a,b (cid:0)Kc + K −1 c (cid:1) coth (cid:16) ucΛ (cid:17) 2T Aab ≡ 2 − 1 2 < 0. (38) a, b = 1, 2, 3, 4 are the pocket indice, where 1 ≡ (0,↑), 2 ≡ (0,↓), 3 ≡ (−1,↑) and 4 ≡ (−1,↓). Λ is a short- range cutoff in the momentum space, Λ = α−1. Cab in Eqs. (35,36,37) is always finite positive definite constant for any a, b = 1, 2, 3, 4 (see Appendix C3). We assume that Cab has no dependence on temperature and mag- netic field. Eqs. (35,36,37) are functional RG equations under which inter-chain interactions change their func- tional forms. To gain a simpler idea of these functional RG equations, we take a sum of the inter-chain inter- actions over their chain indices. The sum reduces the inter-chain coupling functions into coupling constants as follows, m(2) ≡ 2πl2(cid:88) h(2) ≡ 2πl2(cid:88) h(2) ≡ 2πl2(cid:88) j j j M (2) j = H (2) j = (2) j = H g α2 g α2 g α2 (cid:90) (cid:90) (cid:90) M(2)(y) dy, H(2)(y) dy < 0, (39) (40) (2) H (y) dy < 0. (41) As mentioned above, the inequalities in Eqs. (40,41) hold true for the repulsive interaction case. Considering the repulsive interaction case, we assume the negative bare values of h(2) and h(2) in the followings. The RG equations for the coupling constants take forms of dm(2) dlnb = A23 + A14 2 m(2) − 1 πl2 m(2) (cid:0)C23h(2) + C14h(2) (cid:1), dh(2) dlnb dh(2) dlnb = A23h(2) − = A14h(2) − 1 4πl2 1 4πl2 (cid:0)C14m2 (cid:0)C23m2 (2) + 4C23h2 (2) 2 (2) + 4C14h (2) (cid:1), (cid:1). (42) (43) (44) The equations dictate that the umklapp term as well as the inter-chain backward scattering are irrelevant at the tree-loop level, as Aab is negative semi-definite (Eq. (38)). Smaller m(2), h(2) and h(2) are always renormalized into zero ('weak coupling phase'; normal metal phase). Cab is positive definite. Thus, the bare repulsive inter- action g has a critical strength, above which m(2), h(2) and h(2) help one another to grow up into larger values ('strong coupling phase'). The critical strength decreases not only on increasing the magnetic field through a de- pendence of the one-loop terms on the magnetic length l, but also on decreasing the temperature through a de- pendence of Aab on the temperature. This suggests that the strong coupling phase generally appears in low tem- perature side and a transition temperature of the strong coupling phase increases in larger magnetic field (e.g. see a field-dependence of the transition temperature of the SNEI-I phase in Fig. 1 in a region of H < 40 T). The transition temperature can also decrease when the Luttinger parameters Ka (a = 1,··· , 4) deviate largely from the unit. Aab has a global minimum at Ka = Kb = 1 and T = 0. When Ka deviates away from 1, Aab becomes negatively larger and thus the critical strength for g increases; the transition temperature de- creases. Physically speaking, Ka being greater/smaller than the unit means stronger quantum fluctuation of the displacement field/superconducting phase field of the a- th pocket [Eqs. (19,17)]. The enhanced quantum fluctu- ations generally destabilize the strong coupling phase. 9 This observation readily lets us propose a new micro- scopic mechanism for the re-entrant transition from the strong-coupling to weak-coupling phases; the transition induced by raising the magnetic field. The higher mag- netic field generally makes the electron pocket (a) and hole pocket (b) to be smaller in size in the kz space. This makes their bare Fermi velocities, vF,a,vF,b, to be smaller with respect to the electron interaction energy scale. Thus, in the presence of the repulsive interaction, g2,a, g2,b > 0 in Eq. (21), the smaller Fermi velocities make their Luttinger parameters to be smaller than the unit, Ka, Kb < 1. Especially, for H < H0, K1 and K4 are expected to be much smaller than the unit near H = H0, where the electron pocket with a = 1 [(n, σ) = (0,↑)] and the hole pocket with b = 4 [(n, σ) = (−1,↓)] are about to leave the Fermi level. Thus, the transition tempera- ture of the strong coupling phase reduces dramatically −1 −1 near H = H0 through an enhancement of K and K 4 1 in Eq. (38) (e.g. see the field-dependence of Tc of the SNEI-I phase in Fig. 1 in a region of 40 T < H < 50 T). Physically speaking, this reduction is nothing but a con- sequence of the enhanced quantum fluctuation of spin superconducting phase variable. When the bare repulsive interaction is greater than the critical value (strong coupling phase), the umklapp and inter-pocket backward scattering terms grow up into larger values; h(2), h(2) → −∞, m(2) → ±∞. The following argument does not depend on the sign of m(2), so that we set m(2) > 0 henceforth. In the strong coupling regime, Hu,2 and Hb,2 are maximally minimized by (cid:26) 2nπ − Φ− (cid:26) (2n + 1)π − Θ− (2n + 1)π − Φ− (45) , 2nπ − Θ− (46) φ3,m + φ2,m = Φ−, φ4,m + φ1,m = θ3,m − θ2,m = Θ−, θ4,m − θ1,m = with σ32,m = σ41,m = σ32,m = σ41,m = σ. (47) The locking of the total displacement field, φ3,m + φ2,m + φ4,m + φ1,m = 2nπ or (2n + 1)π, dictates that the (cid:80) system is electrically insulating along the field direc- tion. Meanwhile, any electron densities, (cid:104)ρ(r, c)(cid:105) ≡ σ(r, c)ψσ(r, c)(cid:105) with c = A, A(cid:48), B, B(cid:48), do not σ=↑,↓(cid:104)ψ† break the translational symmetry along the field direc- tion (z), because † (n,σ),τ,j(z)ψ(n,σ),τ ,j(z)(cid:105) = 0, † (0,σ),τ,j(z)ψ(−1,σ),τ(cid:48),j(z)(cid:105) = 0, with n = 0,−1, τ, τ(cid:48) = ±, τ = −τ , σ =↑,↓. Due to the charge neutrality condition, the mean electron density (cid:104)ψ (cid:104)ψ 10 metric part of a 2nd-rank spin tensor composed by two spin-1/2 moments [see appendix E]. One spin-1/2 is from the π orbital of A or A(cid:48) carbon atom, while the other spin-1/2 is from the π orbital of B or B(cid:48) carbon atom. The 2nd rank spin tensor is defined as Qcd µν(r) ≡ (cid:104)Sc,µ(r)Sd,ν(r) + Sc,ν(r)Sd,µ(r)(cid:105) − δµν(cid:104)Sc,⊥(r) · Sd,⊥(r)(cid:105), (48) with c = A, A(cid:48), d = B, B(cid:48), µ, ν = x, y, Sc,⊥ ≡ (Sc,x, Sc,y), † σ(cid:48)(r, c)[σµ]σ(cid:48)σ(cid:48)(cid:48) ψσ(cid:48)(cid:48) (r, c) and σ(cid:48), σ(cid:48)(cid:48) =↑,↓. 2Sc,µ(r) ≡ ψ The order of the spin superconducting phase [Eq. (46)] leads to a ferro type as well as density-wave type ordering (cid:17) xy (r) = e2iΘ−(cid:16) of the 2nd rank spin tensor, e.g. u + u cos(∆Kz − 2Φ−) xx (r) + iQAB QAB , where ∆K ≡ kF,3 + kF,2 − kF,4 − kF,1. u is a complex- valued coefficient. Symmetry-wise speaking, the long- range order given in Eq. (46) can be also accompanied by a helical magnetic order whose magnetic moment lies in the xy plane. The helical order has two spatial pitches along the c-axis, (2π)/(kF,3+kF,2) and (2π)/(kF,1+kF,4). Microscopically speaking, however, an amplitude of the magnetic moment is tiny and, if any, it appears only in those spatial regions in the cell where two neighboring π orbitals in the same graphene layer overlap [Appendix E]. On increasing the magnetic field, the outer electron pocket with (n, σ) = (0,↑) and hole pocket with (−1,↓) leave the Fermi level at H = H0. Ab-initio electronic band structure calculations evaluate H0 around 53 T [8]. For H → H0 (H < H0), the bare Fermi velocities of the two pockets vF,1 and vF,4 become smaller. So do the Luttinger parameters of the two pockets K1 and K4 [Eq. (21)]. The reduction of the Luttinger parameters makes the tree-level scaling dimension A14 negatively very large [Eq. (38)]. Thus, according to Eqs. (42,44), m(2) and h(2) are renormalized into smaller values at an early stage of the RG flow for H (cid:46) H0, irrespective of bare values of m(2) and h(2). Meanwhile, A23 as well as vF,2 and vF,3 remain rather constant around H = H0. Thus, according to Eq. (43), h(2) grows up to a larger value and eventually diverges, provided that its bare (ini- tial) value is greater than a critical value (see below for the critical value). Larger h(2) then helps m(2) and h(2) to grow up at a late stage of the RG flow, by way of the one-loop terms in Eqs. (42,44). The argument so far con- cludes that, for H (cid:46) H0, the transition temperature of the strong coupling phase is determined only by Eq. (43) with m(2) = 0; dh(2) dlnb = A23h(2) − C23 πl2 h2 (2). (49) At the zero temperature, Eq. (49) gives the critical value FIG. 5. (color online) Theoretical calculation results of the optical conductivity σzz(ω) in the SNEI-I phase (H = 40, 45, 49.5 T). (Inset) σzz(ω) in the SNEI-II phase (H = 55 T). We use the same parameter sets as in Fig. 1. For its details, see the appendix D. Unlike its appearance in the figures, the delta function at ω = ω∗ is the most prominent in amplitude, while the continuum spectrum is much less significant. The renormalized gap ω∗ is on the order of EintEbw, where Eint is an interaction energy scale, Eint ∼ e2/(l), and Ebw is a band width energy scale (see appendix C4). √ is 2 per two LLs, n = 0 and n = −1 LLs, and per the unit cell along the c-axis. Besides, the insulating phase is associated with particle-hole pairings between n = 0 LL (electron pocket) and n = −1 LL (hole pocket). Thus, we regard this phase as excitonic insulator [18, 20, 42 -- 45] instead of charge density wave phase. (cid:80) j j M (2) a uaKa. ω2 2πuK(cid:80) An insulating property is manifested by the optical conductivity along the c-axis, σzz(ω). In the strong cou- pling phase with large m(2), we may employ a Gaussian approximation for the cosine terms in Hu,2. σzz(ω) is calculated within the linear response theory as σzz(ω) = (e2uK)/(2πl2)δ(ω − ωg), where uK ≡ g ≡ defines a gap for collective particle-hole excitation associated with a fluctuation of the total dis- placement field. An inclusion of a short-ranged dielectric disorder renormalizes the gap into a smaller value ω∗ with a smaller spectral weight for the delta function (see Ap- pendix D). Meanwhile, it adds a continuum spectrum in higher energy region. The continuum spectra compen- sate the reduced spectral weight of the delta function. The observation concludes that the excitonic insulator phase is robust against any small dielectric disorder, pro- vided that the renormalized gap size and the spectral weight of the delta function remains finite (see appendix D). The long-range order of the spin superconducting phases such as θ3 − θ2, θ4 − θ1 in Eq. (46) breaks the U(1) spin-rotational symmetry around the field direc- tion. The breaking of the continuous spin-rotational symmetry is manifested by a long-range ordering of spin quadrupole moment ('spin-nematic excitonic insulator'). The quadrupole moment that exhibits the order is a sym- 020406080ω49.5 45 4001020304050ω(meV)(meV)for h(2) as, h(2),c ≡ = πl2 C23 πl2 C23 (cid:104) (A23)T =0 (cid:88) (cid:0)Kc + K −1 c (cid:1)(cid:105) 2 − c=2,3 Hb,4 and Hu,4 are maximally minimized by φ3,j + φ4,j = Φ−, φ2,j + φ1,j = < 0. (50) θ3,j − θ4,j = Θ−, θ2,j − θ1,j = 11 (cid:26) 2nπ − Φ− (cid:26) (2n + 1)π − Θ− (2n + 1)π − Φ− 2nπ − Θ− (54) , (55) When h(2) < h(2),c < 0, the spin nematic excitonic insu- lator phase always appears below a finite critical temper- ature Tc at H ≤ H0 (Fig. 1). The situation is consistent with the experimental phase diagram of the graphite un- der high field. Meanwhile, RG phase diagrams of the other insulators stabilized by Hu,1, Hu,3 or Hu,4 are not consistent with the graphite experiment. B. magnetic Mott insulator and plain excitonic insulator phases To see this, let us next consider a nature and a RG phase diagram of the magnetic Mott insulator phase sta- bilized by Hu,4 and Hb,4. By exchanging 2 and 4 in Eqs. (42,43,44), we can readily obtain corresponding one- loop RG equations for their coupling constants; dm(4) dlnb = A34 + A12 2 m(4) − 1 πl2 m(4) dh(4) dlnb dh(4) dlnb = A34h(4) − = A12h(4) − 1 4πl2 1 4πl2 (cid:0)C34h(4) + C12h(4) (cid:1), (cid:0)C12m2 (cid:0)C34m2 (4) + 4C34h2 (4) 2 (4) + 4C12h (4) (cid:1), (cid:1). (51) (52) (53) Here, the coupling constants are integrals of the inter- chain coupling functions in Hb,4 and Hu,4; m(4) ≡ 2πl2(cid:88) h(4) ≡ 2πl2(cid:88) h(4) ≡ 2πl2(cid:88) j j j (cid:90) (cid:90) (cid:90) M (4) j = H (4) j = (4) j = H g α2 g α2 g α2 M(4)(y) dy, H(4)(y) dy < 0, (4) H (y) dy < 0. The inequalities hold true for bare values of h(4) and h(4) in the presence of the repulsive interaction g (> 0). The RG equations tell that the bare value of the re- pulsive interaction g has a critical strength above/below which m(4) as well as h(4) and h(4) become rele- vant/irrelevant on the renormalization. In the strong coupling phase with m(4) → ±∞ and h(4), h(4) → −∞, with σ34,m = σ21,m = σ34,m = σ21,m = σ, (56) for m(4) > 0. The locking of the total displacement field results in an electrically insulating behavior along the field direction, while the long-range order of the spin-superconducting phases leads to a long-range helical magnetic order, e.g. (cid:104)SA,x(r)(cid:105) + i(cid:104)SA,y(r)(cid:105) = v (cid:104)SB,x(r)(cid:105) + i(cid:104)SB,y(r)(cid:105) = v (cid:48)(cid:48) + w (cid:48) eiΘ− cos((kF,1 + kF,2)z), (cid:48)(cid:48) eiΘ− cos(cid:0)(kF,1 + kF,2)z(cid:1) eiΘ− cos(cid:0)(kF,3 + kF,4)z(cid:1). As for the charge degree of freedom, the insulating phase does not break the translational symmetry; (cid:104)ρ(r, c)(cid:105) always respects the translational symmetry for c = A, B, A(cid:48), B(cid:48). The phase is stabilized by the pairings with the same LL but between the different spins, so that we call this phase as a magnetic Mott insulator. Unlike the spin-nematic excitonic insulator, a transi- tion temperature of the magnetic Mott insulator goes to zero at a certain critical field below H0. For H → H0 (H < H0), where K1 and K4 become very small, both A34 and A12 in Eqs. (51,52,53) become negatively very large. Accordingly, unlike in the spin nematic excitonic insulator case in the previous section, all of the three coupling constants, m(4), h(4) and h(4), are renormalized to zero for those H sufficiently close to H0 (H < H0). In other words, the transition temperature of the mag- netic Mott insulator always goes to zero at a certain crit- ical field below H0. This is also the case with the plain excitonic insulator stabilized by Hu,1, Hu,3 and Hb,13. These RG phase diagrams are not consistent with the experimental phase diagram of graphite under the high field [13 -- 18, 20]. Besides, the helical magnetic order in the Mott insu- lator is expected to be weak against magnetic disorders. Considering an anisotropy of g-factor in graphite [46], it is natural to assume that the high magnetic field al- lows the system to have single-particle backward scatter- ings between two electron pockets with (n, σ) = (0,↑) and (0,↓), and also that between two hole pockets with (n, σ) = (−1,↑) and (−1,↓). The backward scatter- ings do exist, especially when graphite contains those graphene layers whose normal vectors (c-axis) have non- zero angles with respect to the field direction. Such graphene layers can appear anywhere and randomly VI. TWO POCKETS MODEL (H > H0) 12 For H > H0, both the electron pocket with (n, σ) = (0,↑) and hole pocket with (n, σ) = (−1,↓) leave the Fermi level [8, 19]. The low-energy electronic system for H > H0 comprises only of the electron pocket with (n, σ) = (0,↓) and the hole pocket with (n, σ) = (−1,↑). As before, we call (n, σ) = (0,↓) as a = 2 and (n, σ) = (−1,↑) as a = 3. The charge neutrality condition is given by kF,0,↓ + kF,−1,↑ = π/c0. Under the condition, the in- teraction allows the following umklapp term; (cid:48) u = H † 3,+,nψ † 2,+,j+m−nψ2,−,mψ3,−,j + h.c., ψ (57) (cid:88) j,m,n where the integrals over z and scattering matrix elements are omitted. Other two-particle interaction terms that are linked with the umklapp term at the one-loop level of the fermionic RG equations are inter-pocket and intra- pocket scatterings between different chiralities [45]. They are (cid:48) b = H † 3,±,nψ † 2,∓,j+m−nψ2,∓,mψ3,±,j, (58) ψ † 2,±,nψ † 3,±,nψ † 2,∓,j+m−nψ2,∓,mψ2,±,j, † 3,∓,j+m−nψ3,∓,mψ3,±,j, ψ ψ (59) respectively. To construct effective boson theories of possible insu- lating phases stabilized by H(cid:48) u, we first assume the in- plane (graphene-plane) translational symmetry of the in- sulating phases, consider electron pairing within the same chain, and treat the inter-chain electron-electron interac- tions by the Hartree-Fock approximation. Specifically, we keep only the direct process (Hartree; j = n) and the exchange process (Fock; m = n) in Eqs. (57,58,59), and bosonize them into cosine terms; (cid:88) (cid:40) (cid:88) j,m,n j,m,n and (cid:48) d = H along the c-axis, so that the backward scatterings are generally accompanied by random U(1) phases λj,±(z); (cid:90) (cid:88) dzAj,+(z)(cid:8)eiλj,+(z)ψ (cid:90) dzAj,−(z)(cid:8)eiλj,−(z)ψ † j † 1,+,j(z)ψ2,−,j(z) + h.c.(cid:9) 1,−,j(z)ψ2,+,j(z) + h.c.(cid:9) + ··· (cid:48) imp = H (cid:88) j + (cid:48) imp = H When bosonized, these single-particle backward scat- terings add random U (1) phases into Φ− ± Θ− in Eqs. (54,55) respectively; j dzAj,+(z)σ12,j (cid:90) (cid:88) × cos(cid:2)φ2,j + φ1,j − θ2,j + θ1,j + λ+,j(z)(cid:3) (cid:90) (cid:88) × cos(cid:2)φ2,j + φ1,j + θ2,j − θ1,j + λ−,j(z)(cid:3) + ··· . dzAj,−(z)σ12,j j + Since Φ− and Θ− comprise gapless Goldstone modes in the magnetic Mott insulator, the added random U(1) phases readily kill the long-range orders of Φ− and Θ−, however small the amplitudes Aj,±(z) are [25 -- 28]. Like- wise, the plain excitonic insulator phase stabilized by Hu,1, Hu,3 and Hb,13 is expected to be weak against short-ranged charged disorders. The short-ranged disor- der causes single-particle type backward scatterings be- tween (0,↑) and (−1,↑) pockets and those between (0,↓) and (−1,↓) pockets. From these reasonings as well as in- consistency between their RG phase diagrams and the ex- perimental phase diagram of graphite, we conclude that the magnetic Mott insulator as well as the plain excitonic insulator can hardly explain the graphite experiment co- herently. One may expect that the spin-nematic excitonic insu- lator could also suffer from random single-particle back- ward scatterings between (0,↑) and (−1,↓) pockets or those between (0,↓) and (−1,↑) pockets. Nonetheless, these scatterings unlikely exist in the real system. Or, if any, they are much smaller than the others, because the relativistic spin-orbit interaction is needed for them, and it is extremely small in graphite [36, 46]. Without the spin-orbit interaction, these backward scatterings need both the magnetic scatter and the short-ranged charged scatter on the same spatial point. Microscopically, how- ever, these two types of the scatters are of different origins and they have no correlation at all. From these reason- ings as well as the generic consistency between the RG phase diagram (H < H0 in Fig. 1) and the experimental phase diagram, we conclude that an insulating phase in graphite at H < H0 is the spin-nematic excitonic insula- tor stabilized by the interplay between Hu,2 and Hb,2. j,m (cid:88) (cid:88) (cid:88) j,m j,m O (cid:88) (cid:48) b + H (cid:48) d = H (cid:48) u + H (cid:48) u,1 = H H (cid:48) u,2 = H N (1) j−m N (2) j−m (cid:48) d,1 = H O(1) j−m (cid:90) (cid:90) (cid:88) j,m (cid:88) j,m + + (1) j−m j,m (2) j−m P (cid:48) b,2 = H P (2) j−m (60) (cid:48) d,1 + H (cid:48) u,2 + H (cid:48) u,1 + H +,j + Q23−,m (cid:48) b,2 + ··· , (cid:90) dzσ33,jσ22,m cos(cid:2)2φ3,j + 2φ2,m (cid:3), (cid:90) (cid:3), dzσ23,jσ32,m cos(cid:2)Q23 (cid:90) (cid:3) dzσ33,jσ33,m cos(cid:2)2φ3,j − 2φ3,m (cid:3), (62) dzσ22,jσ22,m cos(cid:2)2φ2,j − 2φ2,m (cid:90) (cid:3) dzσ23,jσ23,m cos(cid:2)Q23−,j − Q23−,m (cid:3). dzσ32,jσ32,m cos(cid:2)Q23 (63) (61) +,m +,j − Q23 range orders individually; 2φ2,j = Φ2, 2φ3,j = Φ3, (cid:26) 2nπ Φ2 + Φ3 = with (n(1) < 0) (2n + 1)π (n(1) > 0) , σ22,j = σ33,j = σ. 13 (64) (65) (66) u,1 and H(cid:48) FIG. 6. (color online) (Left) schematic pictures of two-particle umklapp scatterings that are allowed in the two-pocket model under the charge neutrality condition, H(cid:48) u,2. They are direct (j = n) and exchange (m = n) processes of Eq. (57) respectively. (Middle) two-particle intra-pocket scatterings H(cid:48) (Right) two-particle inter-pocket scatterings H(cid:48) b,2; exchange processes (m = n) of Eq. (58). As in Fig. 2, the vertical axis denotes the momentum along the field direction (kz), while the horizon- tal axis denotes the chain index yj = kjl2 with kj ≡ 2πj/Lx (j = 1, 2,··· , LxLy/(2πl2)). d,1: exchange processes (m = n) of Eq. (59). d,1 and H(cid:48) Here H(cid:48) u,1 is from the Hartree process (j = n) of Eq. (57), u,2, H(cid:48) while H(cid:48) b,2 are from the Fock processes (m = n) of Eq. (57), Eq. (59) and Eq. (58) respectively d and H(cid:48) (Fig. 6). The Hartree processes of H(cid:48) b renormal- ize the Luttinger parameters and Fermi velocities in H0 [Eq. (19)]. Especially, the Hartree term of H(cid:48) d gives rise to positive g2,a (a = 2, 3) in Eqs. (20,21) in the presence of the repulsive electron interaction (g > 0). u,1 + H(cid:48) u,2 + H(cid:48) d,1 + H(cid:48) As in the previous section, we carried out the per- turbative RG analyses on these effective boson models, H0 + H(cid:48) b,2. At the one-loop level of the perturbative RG equations, H(cid:48) d,1 are cou- pled with each other, and so are H(cid:48) b,2. When the bare interaction strength g is greater than critical interaction strength, respective pairs of the cosine terms grow up, to have larger amplitudes and the system enters strong coupling phases. In the following two subsections, we argue that H(cid:48) d,1 stabilize a plain superposed CDW phase, while H(cid:48) b,2 stabilize spin-nematic excitonic insulator phase. u,1 and H(cid:48) u,2 and H(cid:48) u,2 and H(cid:48) u,1 and H(cid:48) A. a superposed CDW phase When H(cid:48) u,1 and H(cid:48) d,1 become relevant, the cosine terms in Eqs. (60,62) are maximally minimized by a charge den- sity wave (CDW) phase, where a displacement field of the electron pocket and that of the hole pocket exhibit long- Such CDW is a plain superposition of a charge density wave of the electron pocket with ↓ spin and π/kF,0,↓ spa- tial pitch and that of the hole pocket with ↑ spin and π/kF,−1,↑ spatial pitch. Since this strong coupling phase is not accompanied by any long-range order of spin su- perconducting phase fields, the transition temperature of the superposed CDW phase increases monotonically in the magnetic field in the presence of the repulsive electron-electron interaction; g2,a=2, g2,b=3 > 0. Such behaviour of the transition temperature is not consistent with the graphite's experimental phase diagram; the ex- periment shows the re-entrant insulator-metal transition at H = Hc,2 (cid:39) 75 T. Besides, the long-range order of the relative displace- ment between the two charge density waves, Φ2 − Φ3, is weak against random charged impurities, unless their spatial pitches are commensurate to the underlying lat- tice constant c0 [25 -- 28]. Namely, the impurity poten- tials induce single-particle backward scatterings within the same electron pocket and/or within the same hole pocket. The impurities appear spatially randomly as a function of the coordinate z. Thus, the scatterings add random U(1) phases into 2φ2,j and 2φ3,j, unless Φ2 and Φ3 in Eq. (64) have finite mass in the CDW phase. When 2kF,0,↓ or 2kF,−1,↑ is incommensurate with respect to 2π/c0, the long-range ordering of Φ2 − Φ3 in Eq. (64) is generally accompanied by a gapless phason excitation. Thereby, even small random charged impurities wipe out the long-range order of the relative phase between the two density waves. Meanwhile, being locked into the dis- crete values by the cosine potential in the umklapp term (H(cid:48) u,1), the total displacement field, Φ2 +Φ3, always has a finite mass in the superposed CDW phase. The locking is therefore robust against the random charged impurities, as far as their amplitudes are small. B. spin nematic excitonic insulator (SNEI-II) phase When H(cid:48) u,2 and H(cid:48) b,2 become relevant, the cosine terms in Eqs. (61,63) are maximally minimized by the excitonic insulator phase with broken U(1) spin rotational symme- try. To see a nature and an RG phase diagram of this strong coupling phase, let us first reduce the inter-chain coupling functions in H(cid:48) b,2 into coupling con- u,2 and H(cid:48) 0,↓,L−1,↑,L 0,↓,R−1,↑,RH(cid:31)u,1H(cid:31)u,2 0,↓,L−1,↑,L 0,↓,R−1,↑,RH(cid:31)d,1 0,↓,L−1,↑,L 0,↓,R−1,↑,RH(cid:31)b,2stants; n(2) ≡ 2πl2(cid:88) p(2) ≡ 2πl2(cid:88) j j 14 (2) j−m. P N (2) j−m, j−m, p(2) ≡ 2πl2(cid:88) P (2) j For the repulsive interaction case (g > 0), bare values of these three coupling constants are negative; the co- sine terms in H(cid:48) b,2 are all from the exchange processes. The one-loop RG equations for these coupling constants take the following forms; u,2 and H(cid:48) dn(2) dlnb dp(2) dlnb dp(2) dlnb = A23n(2) − = A23p(2) − = A23p(2) − (cid:0)n2 (cid:0)n2 C23 πl2 n(2) C23 πl2 C23 πl2 (cid:1), (cid:0)p(2) + p(2) (cid:1), (cid:1). (2) + p2 (2) (2) + p2 (2) (67) (68) (69) Negative semi-definite A23 and positive definite C23 were already defined in Eq. (38) and appendix C3 respectively. Thanks to an inversion symmetry (Qab +,j → −Qab−,j), the coupled equations as well as the bare values of the cou- pling constants are symmetric with respect to an ex- change between p(2) and p(2). This decouples the RG equations into df± dlnb = A23f± ∓ C23 πl2 f 2±, (70) where f± ≡ n(2) ± p(2) = n(2) ± p(2). At the zero tem- perature, A23 and C23 have no dependence of the scale change lnb. Thereby, the equations immediately give out a RG flow diagram as in Fig. 7. The strong and weak (cid:26) coupling phases at T = 0 are defined by n(2) − p(2) > xc (strong coupling phase), n(2) − p(2) < xc (weak coupling phase), with xc ≡ − πl2 C23 A23 > 0. (71) (72) In the strong-coupling side, the cosine terms in the bosonized Hamiltonian are maximally minimized by σ23,j = σ32,j = σ, θ2,j − θ3,j = Θ, 2(φ2,j + φ3,j) = (cid:26) 2nπ (n(2) < 0) (2n + 1)π (n(2) > 0) . (73) (74) (75) The locking of a sum of the two displacement fields leads to an electrically insulating property along the field di- rection. The optical conductivity calculated within the Gaussian approximation shows a gap behavior, σzz(ω) = a=2,3 uaKa and (see also an inset of Fig. 5). The (e2uK)/(2πl2)δ(ω − ωg) with uK = (cid:80) g ≡ 2πuK(cid:80) j N (2) ω2 j FIG. 7. (color online) Renormalization group (RG) flow at T = 0 in the two-dimensional parameter space subtended by n(2) and p(2) = p(2). Weak/strong coupling phases stand for normal metal phase/spin nematic excitonic insulator (SNEI- II) phase respectively. Quantum criticality of the quantum phase transition between these two are controlled by a fixed point named as 'FP1'. The scaling dimension of the relevant parameter at FP1, ν2, is given in Eq. (77). long-range order of the spin superconducting phase θ2−θ3 in Eq. (74) breaks the global U(1) spin rotational sym- metry. The strong coupling phase is accompanied by particle-hole pairing between the electron pocket with ↓ spin and the hole pocket with ↑ spin, so that we name the phase also as spin nematic excitonic insulator phase. Nonetheless, the phase could be symmetrically distinct from the spin nematic excitonic insulator discussed in the previous section, depending on a spatial parity of the excitonic pairing (see also Sec. VIIIA and Sec. XB). We thus distinguish these two by calling them as SNEI-I for H < H0 and SNEI-II for H > H0 respectively. The phase boundary condition, Eqs. (71,72), explains the metal-insulator transition at a lower field regime. For simplicity, we assume that the bare values of n(2) and p(2) = p(2) as well as (C23)T =0 have no H-dependence. For a low H regime, the magnetic length l is large, so is the critical value xc in Eq. (72). Thereby, a given bare value of n(2) − p(2) can be below the critical value xc in lower H regime (weak-coupling phase; normal metal phase). On increasing H, the magnetic length l decreases and so does the critical value xc. Thus, the bare value of n(2) − p(2) exceeds the critical value xc at a certain critical magnetic field (H = H∗ c < H, the system enters the strong coupling phase (SNEI-II phase). From a comparison with the graphite experiment [17, 18, 20], we assume that H∗ c is smaller than H0. In this case, the system at T = 0 undergoes a phase transition from SNEI-I to SNEI-II at H = H0. c ). For H∗ The phase boundary condition Eqs. (71,72) also ex- plains the T = 0 insulator-metal re-entrant transition at a higher field regime. When the field H increases fur- ther, both electron and hole pockets become smaller in size in the kz space. This makes their bare Fermi veloc- -101-10p(2)FP1FP0weak coupling phasestrong coupling phase n (2)−1 3 and K ities vF,2 and vF,3 as well as Luttinger parameters K2 and K3 smaller. The smaller Luttinger parameters can increase the critical value xc through the dependence of −1 A23 on K [Eq. (38)]. To be more precise, 2 suppose that the electron pocket with n = 0 LL with ↓ spin and the hole pocket with n = −1 LL with ↑ spin leave the Fermi level at H = H1. When H gets 'close' to H1 from below (H < H1), the increase of −A23 can overcome the decrease of l2 in Eq. (72), such that xc in- creases again. Namely, for H < H1, l2 is always bounded by (c)/(eH1) from below, while K as well as −A23 have no upper bound in principle. Thus, for some magnetic field Hc,2 with H∗ c < H0 < Hc,2 < H1, xc exceeds the bare value of n(2) − p(2) again and the system falls into the weak coupling phase (normal metal phase) again. From a set of reasonable parameter values used in Fig. 1 (see appendix C4 for a set of parameters used in Fig. 1), we obtain Hc,2 = 82 T and H1 = 120 T. −1 and K 3 −1 2 C. critical natures of the MI and re-entrant IM transitions The re-entrant transition point at H = Hc,2 is a zero- temperature continuous phase transition with dynamical exponent z = 1. Toward this quantum critical point, a correlation length along the field direction ξz diverges as ξz ∝ H − Hc,2 −1/ν2. (76) A critical exponent ν2 is given only by the Luttinger pa- rameters at the critical point (H = Hc,2), (cid:88) (cid:0)Ka + K −1 a (cid:1) ν2 = 1 2 a=2,3 Since z = 1, the correlation length is inversely propor- tional to the gap ωg in the optical conductivity along the field direction, σzz(ω); ωg ∝ (Hc,2 − H)z/ν2 = (Hc,2 − H)1/ν2, (78) for H < Hc,2. By measuring how the gap vanishes toward H = Hc,2 as a function of the field, one can determine the values of the Luttinger parameters at the quantum critical point. By seeing how much the Luttinger param- eters thus determined deviate from 1, one could also test a validity of our theory of the re-entrant insulator-metal transition. The low-H metal-insulator transition between the nor- mal metal and SNEI-I phases is also a quantum critical point. Toward this point, H = Hc,1, the gap ωg in the SNEI-I phase also vanishes, for Hc,1 < H. The critical exponent ν1 is given by the Luttinger parameters at H = Hc,1; ωg ∝ (H − Hc,1)1/ν1, (cid:88) (cid:1) (cid:0)Ka + K −1 a 1 2 a − 2, ν1 = (79) (80) 15 (cid:26) a = 1, 4 a = 2, 3 where the summation in the pocket index a is taken over (A14C23h(2) (cid:29) A23C14h(2)), (A14C23h(2) (cid:28) A23C14h(2)). (81) Meanwhile, the gap in σzz(ω) reaches finite constant val- ues at H = H0±0, when the phase transition from SNEI-I phase to SNEI-II phase is of the first order. This is the case when the spatial parities of the exctionic pairings in the two phases are different from each other (see also Sec. VIIIA and Sec. XB). VII. IN-PLANE RESISTANCE IN THE FOUR POCKETS MODEL (H < H0) Generally, in-plane current operators in the clean limit have finite matrix elements only between neighboring Landau levels. When the temperature is much lower than the cyclotron frequency ω0, the in-plane resistance in- creases on increasing magnetic field H. Contrary to this expectation, the low-temperature in-plane resistance in graphite under high magnetic field shows an unexpected H-dependence [14, 15, 17, 18]. It shows a broad peak around 15 T ∼ 30 T, and then decreases slowly on fur- ther increasing H. From H = 30 T to H = H0 (cid:39) 53 T, the resistance reduces by half or more. Besides, when the system enters the low-field-side out-of-plane insulat- ing phase (Hc,1 < H < H0), the in-plane resistance shows an additional steep increase by 15% to 30% [14, 15, 18]. Unlike the out-of-plane resistivity, the additional increase amount becomes smaller for lower temperature. The H-dependence of the in-plane resistance in 30 T < H < H0 (cid:39) 53 T can be explained by charge trans- ports along the surface chiral Fermi arc (SCFA) states. To see this, notice first that the electron/hole pockets in the bulk are terminated with SCFA states of the electron/hole type around the boundary regions of the system (see Fig. 2 and appendix A). A SCFA state of the electron/hole type is a bundle of Na-number of chi- ral edge modes of electron/hole type respectively, where Na is a number of kz points within the electron/hole pocket (a = 1, 2, 3, 4). Here 'a' denotes the pocket index; 1 ≡ (0,↑), 2 ≡ (0,↓), 3 ≡ (−1,↑), and 4 ≡ (−1,↓). The chiral edge mode enables unidirectional electric current flow along the boundary in a xy plane. The chiral di- rections of the electric current flows of the electron-type and hole-type edge modes are opposite to each other. In the presence of short-ranged charged impurities, the current flow along the electron-type edge mode with σ spin can be scattered into the hole-type edge mode with the same σ spin. In this respect, the SCFA state with (−1, σ) (hole-type) and that with (0, σ) (electron-type) cancel each other by the intra-surface backward scat- terings due to the charged impurities. In the absence − 2. (77) A. H-dependence of Rxx at H < H0 16 with positive Bt and Ct (t = 1, 2), and k ≡ (kz, k). kz and k are conjugate to z and yj ≡ 2πl2j/Lx respectively, (cid:88) (cid:88) k k fj(z) ≡ gj(z) ≡ 1 LzN 1 LzN eikzz+ikyj f (k), eikzz+ikyj g(k). (82) (cid:88) (cid:88) (cid:88) a,τ,b n m (cid:48) = H 1 Lx The gapless modes couple with the SCFA states through a simple density-density interaction, e.g. (cid:90) dz Ae−b × ρa,τ,n(z)(cid:0)ψ (a,τ ;b)(yn, ym) b,m(z)ψb,m(z)(cid:1), (87) † of any backward scatterings between (0, σ) and (−1, σ) ((σ, σ) = (↑,↓), (↓,↑); see the last paragraph in Sec. V for the reasonings of the absence), both (N2 − N4)-number of anticlockwise (electron-type) chiral edge modes with ↓ spin and (N3 − N1)-number of clockwise (hole-type) chi- ral edge modes with ↑ spin individually contribute to the two-terminal conductance within the xy plane; (cid:1) (cid:0)N2 − N4 + N3 − N1 (cid:0)N3 − N1 (cid:1). e2 h 2e2 h Gs = = From the first line to the second line, we used the charge- neutrality condition; N1 + N2 = N3 + N4. Impor- tantly, the in-plane conductance given by Eq. (82) usu- ally increases on increasing H for H < H0. This is because a variation of N1 with respect to H is larger than that of N3; dN1/dH < dN3/dH < 0. For N3 = (Lz/(2c0))(1− H/H1), and N1 = (Lz/(2c0))(1− H/H0), the H-dependence of the resistance due to the surface charge transport is given by Rs = h e2 c0 Lz H0H1 H(H1 − H0) . (83) The resistance is on the order of 1 Ω at H = 30 T [Lz = 50 µm, c0 = 0.67 nm, H0 = 50 T and H1 = 120 T.] The value is on the same order of the experimental value (2 Ω ∼ 3 Ω) [17]. B. T -dependence of Rxx at H < H0 The T -dependence of the in-plane resistance inside the low-field-side insulating phase (Hc,1 < H < H0) can be explained by a coupling between the SCFA states and gapless Goldstone modes associated with the spin ne- matic order in the bulk. The spin-nematic excitonic insu- lator (SNEI-I) phase breaks two global U(1) symmetries. They are the U(1) spin-rotational symmetry around the field direction and a translational symmetry associated with a spatial polarization of the spin (↑ or ↓) and pseu- dospin (n = 0 LL or n = −1 LL) densities. Such SNEI-I phase has two low-energy gapless excita- tions. They are space-time fluctuations of the following two phase variables [Eqs. (45,46)]; fj(z) ≡ − 2Θ−, (84) − 2Φ−. (85) When they vary slowly in z/c0 and yj/l ≡ 2πlj/Lx, their energy dispersions become linear in the momenta; (cid:0)θ4,j(z) − θ1,j(z)(cid:1) (cid:0)φ4,j(z) + φ1,j(z)(cid:1) (cid:0)θ3,j(z) − θ2,j(z)(cid:1) (cid:0)φ3,j(z) + φ2,j(z)(cid:1) gj(z) ≡ − − (cid:88) (cid:1)f (cid:0)B1k2 + C1k2 (cid:88) (cid:1)g (cid:0)B2k2 + C2k2 k z z †(k)f (k) Hsw = 1 2LzN 1 + 2LzN k †(k)g(k), (86) with bulk density operator ρa,τ,n(z) ≡ ψ† a,τ,n(z)ψa,τ,n(z) (yn ≤ Ly/2). a, b = 1, 2, 3, 4 denote the pocket indices, and τ = ± is the chirality index. By definition, the sum- mations over the chain indices n and m in Eq. (87) are restricted within the bulk region and edge region respec- tively; Ly/2 ≤ ym. When bosonized, the density operator in the bulk re- gion is given by a linear combination of the phase vari- ables, 2πρa,τ,n(z) ≡ ∂zφa,n(z) − τ ∂zθa,n(z). Such phase variables generally contain the two low-energy gapless ex- citations with the linear dispersions. Thus, the situation becomes precisely analogous to the electron-phonon in- teraction in metal [47, 48]. The coupling gives the SCFA electrons with finite transport life times [48]. When the temperature is on the order of a band width of the gapless Goldstone modes (but smaller than the transition tem- perature of the SNEI-I phase), the transport life time of the SCFA states is linear in temperature T ; so is the re- sistivity due to the surface charge transport. This can explain the T -dependence of the in-plane resistance in the insulating phases in graphite [14, 15, 18]. VIII. IN-PLANE RESISTANCE IN THE TWO POCKETS MODEL (H > H0) The in-plane resistance of graphite under the high magnetic field stays almost constant in the field inside the high-field-side out-of-plane insulating phase (H0 < H < Hc,2) [17 -- 21]. Above the re-entrant insulator-metal (IM) transition field (Hc,2 < H), the resistance shows the normal behaviour; Rxx increases in the field [21]. In the following, we will argue that the SNEI-II phase in H > H0 can be either topological [49 -- 51] or topologi- cally trivial, depending on the spatial parity of the exci- tonic pairing between electron pocket (n = 0,↓) and hole pocket (n = −1,↑). When the excitonic pairing field is an odd function in the momentum kz, the SNEI-II phase becomes topological and thereby the SCFA state of elec- tron type (n = 0,↓) and the SCFA state of hole type (n = −1,↑) are reconstructed into a helical surface state with a gapless Dirac cone. The electric transport through 17 FIG. 9. (color online) Schematic pictures of (A) side surfaces (grey area) with the two-dimensional helical surface state with a gapless Dirac cone. (B) top surface (grey area) with the two-dimensional Chalker-Coddington network model. with (cid:126)N ≡ (N1(qz), N2(qz), 0). The non-zero bulk winding number reconstructs the SCFA state of the electron type and that of the hole type into a 2-d surface state with a gapless Dirac cone at side surfaces. The side surface is subtended by z ((cid:107) H) and either x or y (Fig. 9(A)). To be concrete, impose the periodic boundary conditions along z and x, put a confining potential along y direction. The mass term M in Eq. (88) depends on the coordinate y. In the vacuum regime, y > Ly/2, the electron/hole pocket goes above/below the Fermi level (Fig. 8(c)). Thereby, Eq. (88) enters a normal 1-dimensional semiconductor regime, M > 2γ2; the winding number takes zero in the vacuum. In the bulk region, y < Ly/2, the gapped mean-field Hamiltonian with the negative n(2) is in the band-inverted regime, M < 2γ2; the winding number takes ±1. Such two topologically distinct 1-dimensional gapped systems are inevitably separated by a 1-d gapless Dirac Hamiltonian, that should come somewhere around y = Ly/2. In other words, the side surface has a 2-d helical surface state that forms a gapless Dirac cone as a function of kz and y ≡ kxl2 (Fig. 9(A)). The reconstructed surface state has the helical veloc- ities not only along the z-direction but also along the x-direction. To see this, notice that the velocity along the x-direction is given by a derivative of the single- particle energy with respect to the spatial coordinate y; vx ≡ l2∂EEI/∂y. Such velocity changes its sign around y = Ly/2, where EEI(qz) forms the 1-d gapless Dirac dis- persion; vx < 0 for y < Ly/2 and vx > 0 for y > Ly/2 (See also Figs. 8(B,C)). Quantitatively, the Dirac cone is highly anisotropic in its velocity within the side surface. Namely, the velocity along the x direction is determined by a work function in the edge region; vx = O(l2∂M/∂y). Conventionally, the work function varies in energy on the order of eV within a length scale of A; ∂M/∂y = O(eV/A). Thus, the velocity along x direction is much faster than that along z direction, the latter of which is given by an energy scale of the band width (2γ2) or the excitonic pairing (∆EI). The 2-d helical surface state in the side surface is con- tinuously connected to a 2-d critical wavefunction sitting on a top (bottom) surface. The top (bottom) surface is subtended by x and y coordinates (Fig. 9(B)). Theoret- FIG. 8. (color online) (a) single-particle electronic states in normal metal phase (two pocket model). The electron pocket (blue curve) is formed by the n = 0 LL with ↓ spin, and the hole pocket (yellow curve) is by the n = −1 LL with ↑ spin. (b) single-particle electronic states with the excitonic pairing. (c) single-particle electronic states in the vacuum region. such Dirac-cone surface state is primarily determined by carrier density doped in the surface region, that has lit- tle field-dependence. Thus, the reconstructed Dirac-cone surface state may provide a simple explanation for the field-(nearly) independent and metallic behaviour of the in-plane resistance observed in the high-field-side out-of- plane insulating phase (H0 < H < Hc,2). A. topological SNEI phase The strong coupling phase discussed in Sec. VI (SNEI- II phase) consists of two topologically distinct phases, depending on the sign of the umklapp term n(2). A mean-field one-dimensional electronic Hamiltonian of the strong coupling phase can be schematically described by the 2 by 2 Pauli matrices as H snei2 mf (qz) = (M − 2γ2 cos(qzc0))σ3 + ∆EI(qzc0)σ1 (cid:9), ≡ EEI(qz)(cid:8)N1(qz)σ3 + N2(qz)σ1 (cid:113)(cid:0)M − 2γ2 cos(qzc0)(cid:1)2 + ∆2 with M < 2γ2, and EEI(qz) ≡ (88) EI(qzc0). (89) The first and second elements of the 2 by 2 matrices cor- respond to the n = 0 LL with ↓ spin and n = −1 LL with ↑ spin respectively (Fig. 8(a)). For clarity, the elec- tron pocket around kz = 0 is shifted by π/c0 in Eq. (88); qz ≡ kz − π . ∆EI(qzc0) stands for an excitonic pairing between the electron and hole pockets (Fig. 8(b)). The pairing is induced by the umklapp H(cid:48) u,2 and inter-pocket scattering terms H(cid:48) b,2. A function form of ∆EI(qzc0) is determined by a value of the total displacement field, such as in Eq. (75). c0 For the negative umklapp term, n(2) < 0, the excitonic pairing field ∆EI(qzc0) is an odd function in qz, while, for the positive case, n(2) > 0, it is even in qz. These two cases represent two topologically distinct phases. In the former/latter case, the following topological winding number defined for the bulk 1-dimensional Hamiltonian Eq. (88) takes ±1/zero respectively; [52 -- 54] (cid:90) π c0 − π c0 Z ≡ (cid:0) (cid:126)N × ∂qz (cid:126)N(cid:1) dqz 2π 3, (90) (a)(b)(c)qzqzqz∆EI+++++---(B)(A)z Hz H18 IX. SUMMARY Graphite under high magnetic field exhibits mysteri- ous metal-insulator (MI) transitions as well as insulator- metal (IM) re-entrant transitions. We discuss these enig- matic electronic phase transitions in terms of perturba- tive RG analyses of effective boson theories. We argue that the two insulating phases in graphite under high field are excitonic insulators with spin nematic order- ings. Similar conclusions were suggested by experimental works both for H < H0 [20] and H > H0 [18]. This pa- per enumerates possible umklapp terms allowed under the charge neutrality condition, clarifies natures of insu- lating states stabilized by each of them, and argues that excitonic insulators with long-range orderings of spin su- perconducting phases can give a possible explanation to the graphite experiments. FIG. 10. (color online) Schematic picture of energetically de- generate SSH end states within the bulk excitonic band gap (blue dotted line). In a generic situation, the degeneracy is lifted by an electrostatic potential (black solid curve). An associated spatial gradient of the end-state eigenenergy with respect to y leads to a chiral electric current along −x direc- tion. ically, the critical wavefunction belongs to the 2D quan- tum Hall universality class, while it is generically off the Fermi level. To see this, impose the open boundary condition along z ((cid:107) H) direction. The non-zero bulk winding number leads to an in-gap end state called as SSH (Su-Schrieffer- Heeger) state within the bulk excitonic gap (left figure of Fig. 10). The end states are localized at the two open boundaries along z direction, top and bottom sur- faces. Due to the Landau degeneracy associated with the in-plane coordinate degree of freedom, each bound- ary has extensive number of such end states. In the clean limit, they are energetically degenerate. In the presence of charged impurities on the surface, the degeneracy is lifted by an electrostatic potential created by the impu- rities (right figure of Fig. 10). The potential depends on x and y, causing a finite spatial gradient of the end- state eigenenergy. The gradient in x or y gives rise to a chiral electric current (one-dimensional chiral mode) along y or −x direction respectively. Such chiral mode encloses a region with higher electrostatic potential. An uneven potential landscape gives rise to a group of chi- ral modes on the surface (Fig. 9(B)), where two spatially proximate (and thus counter-propagating) modes have finite inter-mode hoppings. Electronic states of such sur- face can be described by the Chalker-Coddington net- work (CCN) model. [55, 56] The previous studies on the CCN model [55, 57] conclude that a phase diagram as a function of the chemical potential has two localized regimes and the 2D quantum Hall critical point inter- venes between these two localized regimes. Thus, in-gap surface electronic states sitting on the top (bottom) sur- face are generally localized within the in-plane direction, unless the chemical potential is fine-tuned to the critical point. Based on this, we propose a new mechanism for the re-entrant IM transition. When a pair of electron and hole pockets get smaller in size, strong quantum fluctu- ation of the spin superconducting phase destabilizes the spin-nematic excitonic insulator, causing the re-entrant IM transition. The strength of the quantum fluctuation is quantified by the Luttinger parameters of the electron and hole pockets. We relate the Luttinger parameters with the critical exponent of the T = 0 re-entrant IM transition point. We show that the exponent can be ex- perimentally determined from the infrared optical spec- troscopy. By determining the Luttinger parameters at the transition point, experimentalists can test a validity of our theory for the re-entrant IM transition. We attribute an 'unexpected' field- and temperature- dependences of the in-plane electric transport in graphite under the high field as surface charge transports through surface chiral Fermi arc (SCFA) states and reconstructed Dirac-cone surface state. We first argue that a metallic temperature dependence of the in-plane transport ob- served in the low-field-side insulating phases is due to bulk-edge couplings between the SCFA states and gapless Goldstone modes associated with the spin nematic or- derings. Being gapless excitations, the Goldstone modes in the spin-nematic excitonic insulator phases could be experimentally detected through ultrasound measure- ments [58]. We also argue that the odd-parity excitonic pairing in the bulk reconstructs SCFA states of electron and hole into a (2 + 1)-d helical surface state with a gapless Dirac cone. Based on this finding, we discuss the field- (nearly) independent and metallic behaviour of the in-plane transport inside the high-field-side insulat- ing phase [14, 15, 17, 18]. SSHend stateBulk state(filled)Bulk state(empty)19 in the graphite under the high magnetic field [17, 20]. When the metal phase is characterized by a new free theory instead of the free theory of the decoupled LL phase (a gaussian theory given by Eq. (19)), critical properties of the metal-insulator (MI) and re-entrant insulator-metal (IM) transitions are characterized by a new saddle-point fixed point (schematically denoted by 'FP4' in Fig. 11), rather than by the FP1 that leads to the argument in Sec. VIC. Meanwhile, having a finite charge gap, a fixed point of the excitonic insulator (EIs) is expected to be locally stable against the small per- turbation. Thereby, the primary features of the two EI phases discussed in the paper will not change dramati- cally even in the presence of such perturbations. These features include the finite mobility gaps in σzz(ω) in the two SNEI phases, an overall structure of the H-T phase diagram as well as topological Dirac-cone surface state in the SNEI-II phase and in-plane electric transport due to the surface state. B. excitonic BCS-BEC crossover and nature of a transition between SNEI-I and SNEI-II phases Our theory does not include an effect of an excitonic condensation, as emphasized in Ref. [20]. When H ap- proaches H0 from above (H > H0), electron-hole bound states formed by electron in (n, σ) = (0,↑) LL and hole in (n, σ) = (−1,↓) LL could undergo the Bose- Einstein condensation. Such condensation further assists electron-hole BCS pairings between (n, σ) = (0,↓) and (n, σ) = (−1,↑) LLs, through the umklapp term Hu,2. This leads to a phase with electrically insulating be- haviour along the field direction; the phase is essentially same as the SNEI-I phase discussed in the paper. When the exciton BEC effect is included into our theory, the phase boundary between SNEI-I and SNEI-II phases (say H = Hc,3) will presumably go above H0 (H0 < Hc,3). For H ≥ Hc,3, the long-range phase coherences defined by θ4,j − θ1,j = nπ − Θ− and φ4,j + φ1,j = (m + 1)π − Φ− in Eqs. (45,46) fade away, while the other long-range phase coherences defined by θ3,j − θ2,j = Θ− and φ3,j + φ2,j = Φ− may survive, leading to a phase similar to the spin nematic excitonic insulator phase discussed in Sec. VIB. From this viewpoint, the SNEI-II phase could be regarded as a 'partial ordered phase' derived from the SNEI-I phase. Nonetheless, it can be entirely possible that these two SNEI phases are symmetrically distinct from each other, depending on the spatial parities of the excitonic pairings in the two phases, whose importance was emphasized in Sec. VIIIA. Qualitative natures of the phase transition between these two excitonic insulator phases need further theoretical studies. FIG. 11. (color online) Schematic picture of a possible RG phase diagram in the presence of a relevant perturbation (de- noted by 'X' ) around the decoupled Luttinger liquid (LL) fixed point (denoted by 'FP0'). In the presence of such rele- vant perturbation, the LL fixed point is unstable; the normal metal phase is characterized by a new stable fixed point (de- noted by 'FP3'). A horizontal axis ('Y ') denotes the umklapp and inter-pocket scattering terms that drive the system into the excitonic insulator (EI) phases. 'FP2' represents a stable fixed point charactering the EI phases. The critical proper- ties of the metal-insulator (MI) and re-entrant insulator-metal (IM) transitions are characterized by a new saddle fixed point (denoted by 'FP4') instead of by the FP1. In this schematic picture, we assume that the fixed point for the EI phases ('FP2') is locally stable against the small perturbation X. X. DISCUSSION A. natures of the 'normal' metal phase and criticalities of metal-insulator transitions Our theory regards the 'normal' metallic phase in the graphite experiment as decoupled Luttinger liquid (LL) phase, where we assume that inter-chain electron- electron interactions only renormalize the Luttinger pa- rameters and Fermi velocities as in Eqs. (20,21,B6,B7). Nonetheless, it could be possible that a fixed point of the decoupled LL phase (a gaussian theory given by Eq. (19); schematically denoted by 'FP0' in Fig. 11) is unstable against a certain perturbation associated with the inter- chain interactions (denoted by 'X' in Fig. 11) and, as a result, the 'normal' metal phase is characterized by a new stable fixed point (schematically denoted by 'FP3' in Fig. 11). The stable fixed point could be the Fermi- liquid fixed point [59 -- 64] or the sliding Luttinger-liquid fixed point [65, 66]. One of the experimental evidences that could support our theory's assumption of the de- coupled Luttinger liquid is a T -linear behaviour (or at least non-Fermi-liquid behaviour) in the out-of-plane re- sistivity in the high-T 'normal' metal phase. To our best knowledge, however, no comprehensive experimental studies have been carried out so far for the temperature- dependence of the resistivity in the 'normal' metal phase YXFP2(EIs)FP0(LL)FP3FP4FP1Appendix A: charge neutrality condition Transverse conductivity σxy gives a precise information of electron carrier density ne and hole carrier density nh in any given metal and semimetal under high magnetic field through the following formula; σxyH = ec(ne − nh). (A1) e (> 0) and c are the electron charge, and the speed of light respectively. In the main text, we use the for- mula and evaluate the total number of kz points in the electron/hole pockets Ne/Nh in graphite under the field. With the formula, the previous Hall conductivity mea- surement in the regime of 20 T (cid:46) H (cid:46) 55 T [18, 29 -- 31] gives (Ne − Nh) : Lz/c0 = 10−4 : 1. Using the Kubo formula of the Hall conductivity, Akiba discussed a va- lidity of the formula in the quasi-quantum limit in the graphite [18]. In the following, we employ the Buttiker's theory of the Hall conductivity [67, 68], to demonstrate a validity of the formula in a generic three-dimensional metal and semimetal under high field. Use the Landau gauge and assume that a given three- dimensional system is translational symmetric along x and z directions. Electrons are confined along y direction within y < Ly/2 by a confining potential. A single- particle Hamiltonian comprises of two parts; HT ≡ H0(kz; κ±) + V(kz; κ±, y), (A2) c ). H0 is a bulk Hamil- with κ± ≡ (−i∂y) ± i(−kx + eHy tonian that depends on the coordinate y through κ+ and κ−. V describes an effect of the confining potential; V ≡ 0 when y (cid:28) Ly/2. V depends on y explicitly. HT in Eq. (A2) is already Fourier-transformed with respect to x and z: they are functions of the conjugate momenta kx and kz. In a system with multiple energy bands, HT takes a matrix form. For the spinless graphite case, HT is a four by four matrix; the four bases are from the π orbitals in A, A(cid:48), B and B(cid:48) carbon atoms within the unit cell. Using the k · p expansion, Slonczewski, Weiss and McClure derived H0 around the zone boundary of the first Brillouin zone of the graphite. In the following, we only assume that H0(kz; κ±) as well as V(kz; κ±, y) are given by finite order polynomi- als in κ± and y. Under this assumption, the explicit y- dependence of V can be rewritten into the yc-dependence by use of y ≡ (−i)(l2/2)(κ+ − κ−) + yc and yc ≡ kxl2; (A3) (cid:48) T (kz, yc; κ±). HT (kz; κ±, y) = H Eigenstates of such HT are localized in the y coordinate at y = yc. Eigenvalues depend on kz, yc and the Landau index n; HT φn,kz,yc(y − yc) = En(kz, yc)φn,kz,yc(y − yc). A single-particle velocity operator along x is given by a kx derivative of HT . With kxl2 ≡ yc, an expecta- tion value of the velocity with respect to the eigenstate (A4) 20 is given by a yc-derivative of the eigenvalue. Besides, the eigenstate is uniformly extended along x. Thus, an electric current carried by the eigenstate is given by Jx,n,kz,yc = (−e)l2 Lx ∂En(kz, yc) ∂yc . (A5) The total current density from the n-the Landau level is a sum of Jx,n,kz,yc over all the filled kz and kx ≡ yc/l2 points; (cid:88) (cid:90) π kz c0 − π c0 (cid:88) kx dkz 2π (cid:90) +∞ −∞ jx = 1 LzLy = (−e) Ly Jx,n,kz,yc fT (En(kz, yc)) dyc 2π ∂En ∂yc fT (En(kz, yc)). (A6) fT (E) is a Fermi distribution function. At the zero tem- perature, this reduces to a step function, (cid:26) θ(µ+ − En(kz, yc)) θ(µ− − En(kz, yc)) fT =0(En) = (yc (cid:39) Ly/2), (yc (cid:39) −Ly/2). (A7) µ± are Fermi levels around y = ±Ly/2 respectively. In the presence of a Hall voltage VH in the +y direction, µ+ − µ− = −eVH . In graphite under the high field, the two electron/hole pockets in the bulk region (n = 0/n = −1 LLs with ↑ and ↓ spins) end up with two electron/hole surface chiral Fermi arc (SCFA) states in the boundary region. Namely, En=0/−1,σ(kz, yc) increases/decreases in energy, when yc goes from the bulk region to the boundary region (Fig. 2); (cid:26) En=0,σ(kz, yc) (cid:37) (yc (cid:37)), En=−1,σ(kz, yc) (cid:38) (yc (cid:37)). (A8) Accordingly, the current density induced by the finite Hall voltage comprises of two parts that cancel each other: (cid:16)(cid:90) k1 (cid:16)(cid:90) 2π −k1 −k3 c0 (cid:90) k2 (cid:90) 2π −k2 c0 k4 (cid:17) dkz 2π dkz 2π −k4 dkz 2π dkz 2π + + jx = − e h − (µ+ − µ−) e h Ly (µ− − µ+) Ly k3 (cid:17) . (A9) The first part is from the two electron surface states that subtend chiral arcs from kz = −k1 to kz = k1 and from kz = −k2 to kz = k2 respectively. The other part is from the two hole surface states that subtend chiral arcs from kz = k3 to 2π/c0−k3 and from kz = k4 to kz = 2π/c0−k4 respectively (Fig. 2). To have Eq. (A9), we assume that the hole pocket energies are same in the vacuum, En=−1,σ(kz, yc = −∞) = En=−1,σ(kz, yc = +∞). (A10) Eq. (A9) gives the Hall conductivity as, (cid:16) σxy = = e2 h ec H 1 Lz (ne − nh), N1 + N2 − N3 − N4 (cid:17) (A11) with (N1+N2)/Lz = 2πl2ne and (N3+N4)/Lz = 2πl2nh. From the previous Hall conductivity measurement [18], we typically have ne − nh = 5 × 1015 [cm−3], c0 = 6.7 × 10−10 [m], l = 45 × 10−10 [m], for H = 30 T and ne − nh = −10 × 1015 [cm−3], c0 = 6.7 × 10−10 [m], l = 40 × 10−10 [m], for H = 55 T. This gives out a ratio between Ne − Nh and Lz/c0 as Ne − Nh : Lz/c0 = ±3 × 10−4 : 1 (A12) for 30 T < H < 55 T. From this very small number, we conclude that graphite under this field regime safely satisfies the charge neutrality condition. Appendix B: renormalization of Luttinger parameters and Fermi velocities In the main text, we use the Hartree-Fock approx- imation for the four pockets model or two pockets model, to introduce effective boson Hamiltonians, such as Eqs. (18,19,20,21) with Eqs. (22,23,24,25,27,28,29) or with Eqs. (60,61,62,63). Thereby, the bare kinetic en- ergy part takes a quadratic form in the phase variables, Eq. (19), whose coefficients (Luttinger parameters and Fermi velocities) are further renormalized by intra-pocket forward scattering terms. In the following, we summarize how the intra-pocket forward scattering terms renormal- ize the Luttinger parameters and Fermi velocities. The electron interaction within the same pockets is given by Hf = (cid:88) (cid:88) (cid:90) (cid:90) (cid:48) e dz − (z−z(cid:48))2 0,z V (1),a 2l2 n−m,n−j dz j,m,n a=1,2,3,4 (2,3) † a,n(z)ψ † a,j+m−n(z ψ (cid:48))ψa,m(z (cid:48))ψa,j(z), (B1) with ψa,n(z) ≡ eikF,azψa,+,n(z) + e−ikF,azψa,−,n(z). The matrix element V (1),a n,m (a = 1, 2, 3, 4) is obtained by the substitutions of Eqs. (7,8,9,10) into Eq. (6). In the limit of short interaction length (l0,z (cid:28) l), the matrix element takes a form of f (1),a(cid:16) (cid:17) yn/l, ym/l . (B2) V (1),a n,m ≡ g Lx 1 l0,zl 21 (cid:1) a a + j,m j,m × Hf = dz√2πl0,z j−m,0 − V (1),a 0,j−m (cid:1) −2(kF,al0,z)2(cid:1) Dimensionless functions f (1),a(x, y) decay quickly for x,y (cid:29) 1. With the Hartree-Fock approximation, Hf is bosonized into the followings, (cid:90) (cid:88) (cid:88) (cid:0)V (1),a (cid:0)ρa,+,jρa+,m + ρa,−,jρa,−,m (cid:90) (cid:88) (cid:88) (cid:0)V (1),a dz √2πl0,z (cid:0)ρa,+,jρa−,m + ρa,−,jρa,+,m (cid:90) (cid:88) (cid:88) (cid:0)V (1),a × ηa,+,jηa,−,jηa,−,mηa,+,m cos(cid:2)2(φa,j(z) − φa,m(z))(cid:3) (cid:1) (cid:1) j−m,0 − V (1),a 0,j−me dz√2πl0,z − V (1),a 0,j−m −2(kF,al0,z)2 j−m,0e + ··· , (B3) where ρa,±,j(z) stands for an electron density in the right (+) or left (−) branch in the a-th pocket (a = 1, 2, 3, 4) of the j-th chain (j = 1, 2··· , S † 1 a,±,jψa,±,j = − 2π (cid:0)∂zφa,j ∓ ∂zθa,j ρa,±,j(z) ≡ ψ (cid:1). 2πl2 ); + 2 × j,m a The third term in Eq. (B3) represents a rigidity between two displacement fields in different chains in the same pocket. When the corresponding inter-chain interaction is negative definite, this could result in charge density wave orders with broken translational symmetry along the field direction. An interplay between this inter-chain rigidity term and one of the umklapp term is discussed for the two-pocket model case (see Sec. V). The first two terms in Eq. (B3) lead to the renormal- izations of the Luttinger parameters and Fermi velocities. To quantify them, we employ a gradient expansion with respect to the chain index, ρa,τ,m = ρa,τ,j + (ym − yj)∂yj ρa,τ,j yj ρa,τ,j + ··· , to keep only the leading order. This leads to (ym − yj)2∂2 + 1 2 (cid:88) (cid:88) (cid:26) g2,a + g4,a a j Hf = (2π)2 with g2,a = 2√2πl0,z (cid:90) (cid:114) 2 = g l2 π g4,a = 2√2πl0,z (cid:90) (cid:114) 2 = g l2 π dz (2π)2 + −g2,a + g4,a (cid:90) (cid:1)2(cid:27) (cid:0)∂zθa,j (cid:1)2 (cid:0)∂zφa,j (cid:88) −2(kF,al0,z)2(cid:1) (cid:0)V (1),a dx(cid:0)f (1),a(x, 0) − f (1),a(0, x)e (cid:88) (cid:0)V (1),a dx(cid:0)f (1),a(x, 0) − f (1),a(0, x)(cid:1). m,0 − V (1),a 0,m e m,0 − V (1),a (cid:1) 0,m m m (B4) + ··· , (B5) −2(kF,al0,z)2(cid:1), (B6) (B7) When combined with the bare kinetic energy part; (cid:88) a,j vF,a 2π Hkin = (cid:90) (cid:110)(cid:0)∂zφa,j (cid:1)2 +(cid:0)∂zθa,j (cid:1)2(cid:111) dz , (B8) Eq. (B5) gives out Eq. (19) with Eqs. (20,21). Appendix C: derivation of renormalization group (RG) equations In the main text, we employ one-loop RG equations, Eqs. (35,36,37), and clarify possible insulating phases as well as natures of T = 0 metal-insulator and insulator- metal transition points in graphite under the high field. We solve the RG equations numerically to obtain a finite temperature phase diagram as in Fig. 1. The RG equa- tions are derived perturbatively by use of the standard momentum-shell renormalization method [41]. In the fol- lowing, we briefly summarize how to derive the one-loop RG equations for Hu,2 and Hb,2, Eqs. (35,36,37). We begin with a partition function of the effective field theory; Z = (cid:90) (cid:88) σ··· An action S comprises of a gaussian part S0 and non- gaussian part S1; (cid:88) a,j (cid:110) 1 2π dτ dz + uaKa[∂zθa,j(r)]2 + − 2i∂zθa,j(r)∂τ φa,j(r) [∂zφa,j(r)]2(cid:111) , ua Ka (cid:90) β 0 (cid:90) β 0 S0 = S1 = (cid:90) (cid:110) (cid:111) . (C2) (C3) dτ Hu,2 + Hb,2 + ··· Here a is the pocket index (a = 1, 2, 3, 4). The summa- tion over Ising variables σ··· represent traces over two- dimensional Hilbert spaces subtended by two Klein fac- tors associated with the bosonization. With r = (z, τ ), q = (kz, iωn) and Matsubara frequency ωn = 2nπ/β, the Fourier transforms of φj,a(z, τ ) and θj,a(z, τ ) are given by φj,a(r) = 1 βLz eikzz−iωnτ φj,a(q). (C4) (cid:88) (cid:88) iωn kz<Λ Λ is a cutoff in the momentum space. We decompose the field operators into a slow mode and a fast mode in the momentum space, −S0[φ,θ]−S1[φ,θ]. DφDθe (C1) and 22 with Λ(cid:48) = Λb−1. b (> 1) denotes a scale change. First integrate out the fast mode φ> and θ> in the par- tition function and rescale spatial and temporal length scales as znew = zoldb −1, τnew = τoldb −1, βnew = βoldb −1. (C5) This gives a partition function for the slow mode. The partition function takes essentially the same form as Eqs. (C2,C3), while the interchain interactions in Eq. (C3) are renormalized. The renormalization is calcu- lated with respect to an infinitesimally small scale change ln b ((cid:28) 1). This gives the RG equations for the interac- tions as in Eqs. (35,36,37). We derive the partition function for the slow mode perturbatively in the non-gaussian part S1. We do so up to the second order in S1; Dφ<Dθ<e −S< 0 e Z = Z > 0 where i ), −(cid:104)SU(cid:105)> + O(S3 (cid:1), 1(cid:105)> − (cid:104)S1(cid:105)2 (cid:104)S2 > (cid:0) 1 2 (C6) (C7) (cid:90) (cid:104)SU(cid:105)> = (cid:104)S1(cid:105)> − (cid:90) 1 Z > 0 1 2βLz 1 2βLz (cid:104)···(cid:105)> = S< 0 = S> 0 = −S> 0 , (cid:88) (cid:88) (cid:88) Dφ>Dθ> ··· e (cid:88) (cid:88) (cid:88) kz<Λ(cid:48) ··· , iωn a,j a,j iωn Λ(cid:48)<kz<Λ ··· , 0 = (cid:104)1(cid:105)>. "··· " in the right-hand sides of S</> with Z > is a Fourier transform of the integrand in Eq. (C2). The first term in Eq. (C7) gives a tree-level renormalization to the interchain interactions, while the second term gives a one-loop level renormalization. 0 1. tree-level renormalization (cid:104)S1(cid:105)> in Eq. (C7) gives the tree-level renormalization to the interchain interactions; (cid:68)(cid:90) β (cid:68)(cid:90) β 0 (cid:90) (cid:90) (cid:69) (cid:69) (cid:88) (cid:88) =± j,m Hu,2dτ = 1 2 > d2r M (2) j−mση j τ η 1 2 = > meiM η,< jm (r)e (cid:88) (cid:88) d2r j(cid:54)=m meiH η,< jm (r)e =± − 1 2 H (2) j−mση j ση Hb,2dτ (cid:110) + H (2) j−mτ η j τ η meiH η,< jm (r)e − 1 2 2 η=± − 1 (cid:88) (cid:10)M η,> jm (r)2(cid:11) (cid:88) jm (r)2(cid:11) (cid:10)H η,> (cid:10)H η,> jm (r)2(cid:11) η=± > (cid:111) > > (C8) (C9) φj,a(r) = φ< j,a(r) + φ> 1 j,a(r), φ< j,a(r) = (cid:88) (cid:88) iωn (cid:88) (cid:88) kz<Λ(cid:48) φ> j,a(r) = βLz 1 βLz iωn Λ(cid:48)<kz<Λ eiq·rφj,a(q), 0 eiq·rφj,a(q), As the leading order in the infinitesimally small ln b, we obtain, where and M η jm(r) ≡ Q23 H η jm(r) ≡ Q23 η jm(r) ≡ Q14 η,j(r) + Q14 η,j(r) − Q23 η,j(r) − Q14 η,m(r), η,m(r), η,m(r), H − σ+ j ≡ σ32,j, σ j ≡ σ32,j, − τ + j ≡ σ41,j, τ j ≡ σ41,j. a=1,2,3,4 1 2 1 2 1 2 a=1,4 a=2,3 (cid:88) (cid:88) (cid:88) (cid:0)Ka + (cid:0)Ka + (cid:0)Ka + (cid:88) (cid:0)Ka + (cid:88) (cid:0)Ka + (cid:88) (cid:0)Ka + 1 Ka 1 Ka 1 Ka 1 Ka 1 Ka (cid:1) coth (cid:1) coth (cid:1) coth (cid:1) coth (cid:1) coth (cid:1) coth 1 Ka uaΛ 2T uaΛ 2T a=1,2,3,4 a=2,3 a=1,4 2 2 (cid:105) (cid:105) (cid:104)M η,> jm (r)2(cid:105)> = (cid:104)H η,> jm (r)2(cid:105)> = 2 (cid:104)H η,> jm (r)2(cid:105)> = 2 (cid:104) (cid:104) (cid:104) 1 4 1 2 1 2 2 − 2 − 2 − dM (2) j−m d ln b dH (2) j−m d ln b dH (2) j−m d ln b = = = This leads to the tree-level RG equation as βuaΛ 2 ln b, βuaΛ βuaΛ ln b, ln b, (cid:105) M (2) j−m, uaΛ 2T H (2) j−m, (2) j−m. H 2. one-loop level renormalization (cid:104)S2 1(cid:105)>−(cid:104)S1(cid:105)2 1(cid:105)>,c ≡ (cid:104)S2 > in Eq. (C7) gives the one-loop level renormalization to the interchain interactions. The one-loop renormalization comprises of products between different interactions; S1 = SM + SH + SH + ··· , S2 1 = S2 H + S2 H M + S2 + 2SM SH + 2SM SH + 2SH SH + ··· , where SM , SH and SH are defined as follows, (cid:90) (cid:88) (cid:88) (cid:88) SI ≡ 1 2 d2r with I = M, H, H. The products of two interaction terms take forms of j(cid:54)=m I (2) j−m(··· )η (cid:90) η=± =± meiI η,< j (··· )η (cid:90) jm (r)eiI η,> jm (r), d2r(cid:48)(cid:88) (cid:88) (cid:88) i(cid:54)=j ij (r)ei(cid:48)J η(cid:48) n (cid:104)eiI η m(cid:54)=n ,(cid:48),η,η(cid:48) I (2) i−jJ (2) m−n mn(r(cid:48))(cid:105)>,c (C12) (cid:104)SI SJ(cid:105)>,c = 1 4 d2r (··· )η i (··· )η j (··· )η(cid:48) m(··· )η(cid:48) (C10) (C11) 23 where (cid:104)AB(cid:105)>,c ≡ (cid:104)AB(cid:105)> − (cid:104)A(cid:105)>(cid:104)B(cid:105)>. When i (cid:54)= m, n and j (cid:54)= m, n in Eq. (C12), the right hand side vanishes identically. The terms with i = m and j = n or those with i = n and j = m are negligibly smaller than the others in the larger Lx limit. We thus consider only those terms in Eq. (C12) with i = m, n and j (cid:54)= m, n and/or those terms with i (cid:54)= m, n and j = m, n. The one-loop renormalization in Eq. (C12) generates SM , SH and SH as well as other types of cosine terms. Nonetheless, tree-level scaling dimensions of all the other cosine terms thus generated are negatively much larger than those of SM , SH and SH . Namely, they are much more irrelevant than SM , SH and SH at the tree-level renormalization group flow. Thus, we only keep those terms in Eq. (C12) that generate SM , SH and SH . S2 with = −(cid:48), η = η(cid:48) and i = m (or j = n) generates M ) with = −(cid:48), η = η(cid:48), SH (or SH ) respectively. S2 and i = m or j = n or with = (cid:48), η = η(cid:48), and i = n or j = m generates SH (SH ) respectively. SM SH (SM SH ) with = (cid:48), η = η(cid:48), and i = n (j = n) or with = −(cid:48), η = η(cid:48), and i = m (j = m) generates SM . SH SH does not generate any of SM , SH and SH . In the following, we only demonstrate how S2 With = −(cid:48), η = η(cid:48) and i = m, Eq. (C12) with M generates SH . H (S2 H I = J = M reduces to (r)−M η,< −iM η,> (cid:90) (cid:90) 1 4 1 2 ij ,η j,n d2r d2r d2r(cid:48) d2r(cid:48) i=m (r)e (cid:88) (cid:88) (cid:88) (cid:90) j(cid:54)=n(cid:88) in (r(cid:48)))(cid:10)eiM η,> (cid:90) (cid:88) j(cid:54)=n(cid:88) in (r(cid:48)))(cid:3)(cid:10)M η,> in (r(cid:48)))(cid:3) = η,n (r(cid:48))(cid:3) cos(cid:2)Q23,< η,n (r(cid:48))(cid:3) sin(cid:2)Q23,< i=m j,n ij (r) − M η,< (r) − M η,< η,j (r) − Q14,< η,j (r) − Q14,< (cid:104)S2 M(cid:105)>,c = ei(M η,< ij ij = cos(cid:2)M η,< cos(cid:2)M η,< cos(cid:2)Q14,< − sin(cid:2)Q14,< ij where i−jM (2) i−n >,c n M (2) τ η j τ η in (r(cid:48))(cid:11) in (r(cid:48))(cid:11) τ η j τ η n M (2) ,η (r)M η,> i−jM (2) i−n > (C13) η,i (r(cid:48))(cid:3) η,i (r(cid:48))(cid:3). η,i (r) − Q23,< η,i (r) − Q23,< (C14) The largest part of the contribution comes from r = r(cid:48). In this case, the second term in Eq. (C14) vanishes (see the next subsection for a justification of this approxi- mation). For the first term with j (cid:54)= n, we replace η,i (r(cid:48))] by its normal ordering with cos[Q23,< use of a formula cos Φ =: cos Φ : exp[−(cid:104)Φ2(cid:105)/2] [41, 69]. Within the normal order, we employ a Taylor expansion with respect to small r(cid:48) − r. At the leading order expan- sion, Eq. (C14) becomes η,i (r) − Q23,< cos(cid:2)M η,< cos(cid:2)Q14,< ij in (r(cid:48))(cid:3) η,n (r)(cid:3)e (cid:39) − 1 (r) − M η,< η,j (r) − Q14,< 2(cid:104)(Q23,< η,i (r)−Q23,< η,i (r(cid:48)))2(cid:105)< . (C15) Thereby, we have (cid:104)S2 M(cid:105)>,c = (cid:90) j(cid:54)=n(cid:88) j,n d2r (cid:88) (cid:88) ,η × C23 n cos(cid:2)H jn (r)(cid:3) η,< τ η j τ η where (cid:90) Ccd ln b ≡ − 1 dr(cid:48) 1 e 2 2(cid:104)(Qcd,< η,i (r)−Qcd,< η,i M (2) i−jM (2) i−n ln b, (C16) i (r(cid:48)))2(cid:105)<(cid:104)Qcd,> η,i (r)Qcd,> η,i (r(cid:48))(cid:105)>, (C17) with c, d = 1, 2, 3, 4. Note that the integrand in Eq. (C17) is short-ranged in r − r(cid:48) and Ccd is a positive definite real-valued quantity (see the next subsection). Eq. (C16) (2) j−n in combination with Eqs (C7,C10) dictates that H acquires the following one-loop renormalization, (2) j−n dH d ln b = ··· − C23 2 M (2) i−jM (2) i−n + ··· . (C18) (cid:88) i i−j = M (2) Since M (2) j−i, this is nothing but the first term of the one-loop renormalization in Eq. (37). Similarly, one can show all the other terms of the one-loop renormal- izations in Eqs. (35,36,37). A factor "4" in the second term of the one-loop renormalization in Eq. (37) is due to the four distinct contributions to SH from S2 ; (i) = −(cid:48), η = η(cid:48), i = m, (ii) = −(cid:48), η = η(cid:48), j = n, H (iii) = (cid:48), η = η(cid:48), i = n, (iv) = (cid:48), η = η(cid:48), j = m in Eq. (C12). Likewise, 2SM SH (2SM SH ) has two distinct contributions to SM , giving rise to the first (second) term of the one-loop renormalization in Eq. (35); (i) = (cid:48), η = η(cid:48), i = n (j = n), (ii) = −(cid:48), η = η(cid:48), i = m (j = m) in Eq. (C12). This completes the derivation of Eqs. (35,36,37). 3. evaluation of Ccd F Cab is defined in Eq. (C17). Let us first calculate the > = integrand in Eq. (C17); η,i (r)Qab,> 1 (cid:10)Qab,> (cid:10)(cid:0)Qab,< η,i (r(cid:48))(cid:11) (cid:88) (cid:88) eiq(r−r(cid:48))(cid:10)Qab,> η,i (r(cid:48))(cid:1)2(cid:11) (cid:88) (cid:88) 2(1 − eiq(r−r(cid:48)))(cid:10)Qab,< η,i (r) − Qab,< Λ(cid:48)<kz<Λ η,i (q) (βLz)2 (βLz)2 < = iωn 1 kz<Λ(cid:48) iωn ∗ Qab,> >, η,i (q)(cid:11) η,i (q)(cid:11) Qab,< ∗ η,i (q) η,i where(cid:10)Qab,>/< (cid:110) (cid:88) + η(−1)c(cid:0) c=a,b (cid:104)φ (q)(cid:11) η,i Qab,>/< ∗ >/< = (q) ∗ ∗ c,i(q)φc,i(q)(cid:105)>/< + (cid:104)θ c,i(q)θc,i(q)(cid:105)>/< ∗ ∗ c,i(q)φc,i(q)(cid:105)>/< c,i(q)θc,i(q)(cid:105)>/< + (cid:104)θ (cid:104)φ 24 (cid:1)(cid:111) . (C19) with (−1)a = 1 and (−1)b = −1. We used Fourier transform in Eq. (C4). The gaussian integrals over the fast/slow modes lead to ∗ c,i(q)φc,i(q)(cid:105)>/< = (cid:104)φ ∗ c,i(q)θc,i(q)(cid:105)>/< = ∗ c,i(q)θc,i(q)(cid:105)>/< = − (cid:104)φ (cid:104)θ βLzπucKc , z + ω2 ck2 u2 n βLzπucK−1 c z + ω2 u2 ck2 n βLziπωn , kz(u2 ck2 z + ω2 n) . (cid:48) c=a,b −1 c (cid:88) (cid:1)Mc(r − r(cid:48)) (cid:0)Kc + K 2,c(r − r(cid:48)), η(−1)cF (cid:88) (cid:0)Kc + K 1 < = 2 η(−1)cF2,c(r − r(cid:48)), −1 c c=a,b (C20) (cid:1)F1,c(r − r(cid:48)) Accordingly, we have η,i (r)Qab,> (cid:104)Qab,> (cid:10)(Qab,< η,i (r) − Qab,< + 1 2 η,i (r(cid:48))(cid:105)> = (cid:88) η,i (r(cid:48)))2(cid:11) (cid:88) c=a,b + c=a,b (cid:90) Λ(cid:48)<kz<Λ = cos(Λz)e (cid:90) with Mc(r) ≡ dkz 1 β uceiqr n + u2 ω2 ck2 z (cid:88) (cid:88) iωn −ucΛτ ln b, 1 β dkz iωn kz eiqr n + u2 ck2 z iωn ω2 −ucΛτ ln b, 2(1 − cos(qr))uc ω2 n + u2 ck2 z iωn kz 2 · eiqr ck2 n + u2 ω2 z ≡ 2iθc(r), (cid:48) 2,c(r) ≡ − Λ(cid:48)<kz<Λ 1 β dkz kz<Λ(cid:48) = −i sgn(τ ) sin(Λz)e (cid:90) (cid:88) c )/α2(cid:3), = log(cid:2)(x2 + y2 (cid:90) (cid:88) = 2iArg(cid:2)yc + ix(cid:3) kz<Λ(cid:48) dkz 1 β iωn iωn F1,c(r) ≡ F2,c(r) ≡ <, and yc ≡ ucτ + αsgn(τ ). In the right hand side, Mc(r), F (cid:48) 2,c(r), F1,c(r) and F2,c(r) are evaluated at the zero tem- perature. Substituting these into Eq. (C17), we obtain Cab at T = 0 as, Cab,T =0 = (cid:19)λb e dτ dz c=a,b −∞ −∞ z2 + y2 a (cid:90) ∞ (cid:90) ∞ (cid:18) α2 (cid:19)λa(cid:18) α2 (cid:111) (cid:88) −ucΛτ(cid:110) (cid:90) ∞ (cid:88) (−1)c sin(Λz)sgn(τ ) sin(∆ab(r)) (cid:19)λa(cid:18) α2 (cid:18) α2 λc −ucΛτ (cid:90) ∞ (cid:19)λb λc cos(Λz) cos(∆ab(r)) −∞ −∞ dτ e c=a,b (cid:39) η 2 dz + z2 + y2 b cos(Λz), (C21) z2 + y2 a z2 + y2 b 4 (Ka + K−1 c ≡ (ucτ + α)2, λa ≡ 1 with y2 a ) and ∆ab(r) ≡ θa(r) − θb(r). The integrand in the first line is short- ranged in r, justifying a posteriori the approximations made in Eqs. (C14,C15). Based on the same spirit, we approximate ∆ab(r) by zero, to obtain the second line. Cab is positive definite. One can show this by carrying out the z-integral formally, Cab,T =0 = dτ G(τ )e −ucΛτ , (C22) (cid:90) ∞ λc −∞ c=a,b (cid:88) (cid:90) ∞ (cid:90) ∞ −∞ −∞ and G(τ ) ≡ Fa(ξ; τ ) ≡ 25 Meanwhile, Cab,T =0/Aab,T =0 goes to the zero in the limit of λa + λb → +∞. For simplicity, we assume that Cab,T =0 does not depend on the magnetic field H in the main text. A typical value of Cab,T =0 is evaluated in a simple case with Ka = Kb = 1 and ua = ub = u; Cab,T =0,Ka,b=1,ua,b=u (cid:90) −uΛτ(cid:90) (cid:90) ∞ = dτ e = e −Λα 2α2 u 0 α2 z2 + (uτ + α)2 eiΛz dz e−2Λx x + α = eΛα 2α2 u dx E1(2Λα). (C26) E1(x) is the exponential integral. α is a lattice constant along the z-direction while Λ is a high energy cutoff in the momentum space; Λα = O(1). 4. parameters used in Fig. 1 To obtain theoretical phase diagram at finite temper- ature as in Fig. 1, we solved numerically the RG equa- tions Eqs. (42,43,44) for H < H0 and Eqs.(67,68,69) for H0 < H < H1. Thereby, a set of parameters in the RG equations are chosen in the following way. Cab has an engineering dimension of [length]/[energy]. From Eq. (C26), we set dξFa(ξ; τ )Fb(Λ − ξ; τ ) dξ, (cid:17)λa dz eiξz(cid:16) α2 (cid:17)λa− 1 (cid:16) ξ z2 + y2 a 2 Kλa− 1 2 , 2ya = 2√πα2λa (yaξ) Γ(λa) (C24) (C23) Cab = 2α ΛE , (C27) for any a, b = 1, 2, 3, 4. α is the lattice constant of the graphite along the c-axis, α = c0 = 6.7A. ΛE is a high energy cutoff in the energy scale. We set this to be a band width of the four pockets, ΛE = 40 [meV]. , with the Bessel function Kν(x) and the Gamma function Γ(x). Since λa > 1/2, Fa(ξ, τ ) is positive definite and so is G(τ ). With Eq. (C22), this assures the positive definiteness of Cab,T =0. Cab,T =0 in Eq. (C21) depends on the Luttinger param- eters Ka and Kb. Nonetheless, the dependence is much weaker than that of Aab in Eq. (38). One can see this, by evaluating an upper bound of Cab,T =0, (cid:90) (cid:18) α2 (cid:19)λa+λb(cid:90) λc dz z2 + α2 −ucΛτ dτ e (cid:88) (cid:88) c=a,b c=a,b Cab,T =0 < = αλc ΛE Γ( 1 2 )Γ(λa + λb − 1 2 ) Γ(λa + λb) ≡ Cu. ΛE denotes a finite high-energy cutoff in the energy scale, ΛE = Λ× maxc=a,b(uc). When the Luttinger parameters get much smaller/larger than 1, λa+λb → +∞, the upper bound of Cab,T =0 as well as Aab,T =0 diverge; (cid:16) 1 (cid:17)(cid:0)λa + λb Aab,T =0 → 2(cid:0)λa + λb (cid:1). Cu → α ΛE Γ 2 (cid:1) 1 2 , According to Eqs. (39,40,41), m(2), h(2), h(2), n(2), p(2), and p(2) have the same engineering dimension as g ≡ g/α2, where g represents an interaction strength as in Eq. (7). For initial values of m(2), ... ,p(2) in the RG flow, we set (cid:26) (cid:0)m(2), h(2), h(2) (cid:0)n(2), p(2), p(2) (cid:1) = g (3,−1.25,−1.25), (cid:1) = g (1.1,−1.25,−1.25). (C28) A value of g is set in the following way. We consider that the interaction is from the Coulomb interaction and therefore its typical interaction energy scale is given by Eint = e2 l . (C29) The magnetic length l depends on the magnetic field and the relative permittivity is set to 13 for graphite. We regard that the Coulomb interaction ranges over the mag- netic length in the xy plane, and ranges over the Tohmas- Fermi screening length along the z direction λTF. We thus compare Eint with g/(l2λTF) [see Eq. (7)]. This leads to (C25) g = g α2 = e l l2 α2 λTF. (C30) The screening length along the c-axis is set to λTF = c0/√6. index (j), Aab in the RG equations is given by Eq. (38). ucΛ in Eq. (38) (c = 1, 2, 3, 4) is set to the high-energy cutoff in the energy scale, ΛE = 40 [meV] . For the Luttinger parameters Ka in Eq. (38), we use Eq. (21). The intra- pocket forward scattering strengths in Eq. (21) are set as, g4,a=1 = g4,a=4 = g, g4,a=2 = g4,a=3 = g, g2,a=1 = g2,a=4 = g/1.6, g2,a=2 = g2,a=3 = g/1.1, where g is given in Eq. (C30). The bare Fermi velocity in Eq. (21) vF,a is a kz derivative of the energy dispersion of the four pockets given in Eq. (2); ∂En,σ(kz) vF,a = ∂kz kz=kF,n,σ ≡ −2γ2c0 sin(2πξn,σ) (C31) 26 (cid:88) (cid:88) (cid:90) j a dzφa,j(z). (D1) uaKa dz∂zθa,j(z). (D2) e π Pz = − (cid:88) (cid:88) j a Jz = e π (cid:90) The latter is a sum of the current density fields, The correlation function is calculated with respect to a mean field action for the SNEI phases. For the mean field action, we employ a Gaussian approximation for Hu,2 and H(cid:48) u,2, to replace their cosine terms by proper quadratic terms, (cid:88) (cid:88) j,m 1 2 j,m Hu,2 (cid:39) H (cid:48) u,2 (cid:39) M (2) j−m N (2) j−m dz (cid:90) (cid:110)(cid:0)φ2,j + φ3,j + φ1,m + φ4,m (cid:1)2 (cid:1)2(cid:111) +(cid:0)θ2,j − θ3,j + θ1,m − θ4,m (cid:90) (cid:110)(cid:0)φ2,j + φ3,j + φ2,m + φ3,m (cid:1)2 (cid:1)2(cid:111) +(cid:0)θ2,j − θ3,j − θ2,m + θ3,m (cid:18)(cid:126)φK (cid:88) (cid:1) [M0,K] (cid:0)(cid:126)φ (cid:19) dz . † K † (cid:126)θ K , (D3) (cid:126)θK with K ≡ (kz, k, iωn). The Fourier transform is taken with respect to the spatial coordinate z, imaginary time τ and the chain index j (yj ≡ 2πl2j/Lx); φa,j(z, τ ) ≡ 1 βLzN K eikzz+ikyj−iωnτ φa,K. (D4) (cid:88) In the following, we briefly summarize how to calculate the retarded correlation function with respect to SMF in the SNEI-I phase with/without disorder. For the model with two electron pockets and two hole pockets, the gaussian action is described by a 8 by 8 matrix, (cid:20) AK BK CK DK (cid:21) [M0,K] ≡ . (D5) A 4 by 4 matrix AK is for the displacement fields of the four pockets φa (a = 1, 2, 3, 4), and 4 by 4 matrix DK is for the superconducting phase fields of the four pockets θa (a = 1, 2, 3, 4). They are given by with a = (n, σ); 1 = (0,↑), 2 = (0,↓), 3 = (−1,↑), and 4 = (−1,↓). We set 2γ2 = 40 [meV], and This in combination with H0 in Eq. (19), gives a gaussian ('mean-field') action that takes a form of  ξ0,↑ = 1 ξ0,↓ = 1 ξ−1,↑ = 1 ξ−1,↓ = 1 4 − H 4 − H 4 + H 4 + H 200[T] , 480[T] , 480[T] , 200[T] . SMF = (C32) 1 2βLzN K Eq. (C32) realizes H0 = 50 [T] and H1 = 120 [T]. Appendix D: calculation of optical conductivity σzz(ω) In the main text, we describe how the longitudinal optical conductivity along the field direction behaves in the SNEI phases as well as the metal-insulator transi- tion points at H = Hc,1 and H = Hc,2. According to the linear response theory, the conductivity is given by a retarded correlation function between an electron po- larization operator Pz and current operator Jz. In the bosonization language, the former is a sum of the dis- placement fields over the pocket index (a) and the chain k2 z + 2M (0) 2M (0) 2M (k) 2M (k) u4 πK4 π k2 z + 2M (0) −2M (0) 2M (k) −2M (k) 2M (0) k2 z + 2M (0) 2M (k) 2M (k) −2M (0) z + 2M (0) −2M (k) 2M (k) u2 πK2 2M∗(k) 2M∗(k) k2 z + 2M (0) 2M (0) 2M∗(k) −2M∗(k) π k2 z + 2M (0) −2M (0) u2K2 u4K4 π k2 2M∗(k) 2M∗(k) 2M (0) k2 z + 2M (0) u3 πK3 −2M∗(k) 2M∗(k) −2M (0) z + 2M (0) u3K3 π k2  , 27 (D6) (D7)  , πK1  u1  u1K1 AK ≡ DK ≡ (cid:80) j eikyj .The other 4 by 4 matrices where M (k) ≡ BK and CK connect the four φ fields and the four θ fields, j M (2) BK = CK = ikzωn π 14×4. (D8) 14×4 stands for the 4 by 4 unit matrix. For later convenience, we introduce a new basis with respect to the pocket index;  Φ+ ΦI ΦII ΦIII  ≡  1 1 1 2 1 1 1 1 −1 −1 1 −1 1 −1 1 −1 −1 1   φ1 φ4 φ2 φ3  ≡ T (cid:126)φ, (cid:126)Φ ≡ (cid:126)Θ ≡ T (cid:126)θ. In the right hand side, we omitted the subscript K for the φ, θ, Φ and Θ fields. With the new basis, the gaussian action is given by 1 2βLzN SMF = and [Mc,K] ≡ † K (cid:88) (cid:0)(cid:126)Φ (cid:20) T AK T K † K (cid:126)Θ (cid:1) [Mc,K] (cid:21) BK CK T DK T (cid:18)(cid:126)ΦK (cid:19) (cid:126)ΘK , . (D10) analytic continuation, iωn → ω + iη. This gives the re- tarded correlation function. The real part of the retarded correlation function is nothing but the optical conductiv- ity σzz(ω); σzz(ω) = Re σzz(iωn)iωn=ω+iη , σzz(iωn) = (cid:126)e T zz(iωn) T (cid:126)e+, (D13) with (cid:126)e+ ≡ (1, 1, 1, 1)T . U−1 and Qc the quenched average of Qc 4 matrices, zz(iωn) is zz(iωn)) as well as T are 4 by zz(iωn) (Qc (cid:110) + U−1 T Qc (cid:111)  .  u1K1 (D9) U−1 ≡ u4K4 u2K2 u3K3 (D14) Qc zz(iωn) is a Fourier transform of the imaginary-time time-ordered correlation function between four Φ fields and four Θ fields, Qc zz(iωn) = dτ Qc zz(τ ) eiωnτ , (cid:90) β 0 e2 π2V (cid:90) (cid:88) dz (cid:82) d(cid:126)Φd (cid:126)Θe−SMF ∂zΘα,j(z, τ )Φβ,m(z, 0) jm(τ, z0, z j,m (cid:48))]αβ, (cid:48)[Rc (cid:90) (cid:82) d(cid:126)Φd(cid:126)Θe−SMF dz , (D15) [Qc zz(τ )]αβ ≡ (cid:48))]αβ ≡ We consider that the total displacement field Φ+ couples with a disorder potential through; [Rc jm(τ, z0, z (cid:90) (cid:88) j Himp = dz j(z)Φ2 +,j(z). (D11) Physically, such disorder potential j(z) is nothing but a local fluctuation of the dielectric constant. We take a quenched average over the local fluctuation as (cid:82) dj(z)··· e (cid:82) dj(z) e − 1 gy − 1 gy (cid:82) dz 2 (cid:80) (cid:82) dz 2 (cid:80) j j j (z) j (z) ··· ≡ . (D12) gy stands for a disorder strength associated with spatially (but not temporally) fluctuating dielectric constant. We first calculate an imaginary-time time-ordered cor- relation function between Pz and Jz, and then take an with α, β = +, I, II, III and the chain index j, m = 1,··· , S/(2πl2). the quenched average of Qc With use of a Born approximation [28], we can take zz(iωn), (cid:48)(cid:48)(cid:88) (cid:90) (cid:88) (cid:2)P (iωn)(cid:3)(cid:105)−1(cid:2)M−1 dz m e k (cid:3) −ikzz(cid:48)(cid:48)−ikym(−ikz) c,K ΦΘ, (D16) T 2e2 π2V = (cid:2)M−1 (cid:3) c,K]ΦΦ, [M−1 c,K ΦΦ Qc zz(−iωn) (cid:104) 14×4 − where [M−1 matrices. [M−1 c,K]ΦΦ and [M−1 c,K]ΦΘ, and [P (iωn)] are 4 by 4 c,K]ΦΘ are 4 by 4 blocks of T , BD−1 T . (D17) (D18) an inverse of the 8 by 8 matrix [Mc,K] that connects Φ and Φ and that connects φ and Θ respectively; 4 by 4 matrices A, B, C, D and T in the right hand sides are given by Eqs. (D6,D7,D8,D9). [P (iωn)] is a 4 by 4 diagonal matrix that represents an effect of the disorder, [M−1 [M−1 c,K]ΦΦ ≡ T(cid:0)A − BD−1C(cid:1)−1 c,K]ΦΘ ≡ T(cid:0)A − BD−1C(cid:1)−1  gym(iωn) (cid:88) [P (iωn)] ≡ 0 0  . 0 m(iωn) is a sum of the (Φ+, Φ+)-component of the inverse of the 8 by 8 matrix [Mc,K] over k ≡ (kz, k); c,K]Φ+Φ+. [M−1 (D20) 2 m(iωn) ≡ LzN k Note that m(iωn) is an even function of ωn (see below). One may rewrite Eq. (D16) into Qc zz(−iωn) 2e2 = π2V (cid:104) π2 T T e m dz (cid:90) (cid:88) (cid:0)DA − DT P T(cid:1) + ω2 (cid:104) π2 (cid:48)(cid:48)(cid:88) −ikzz(cid:48)(cid:48)−ikym (−ωn) (cid:105)−1 (cid:0)DA − DT P T(cid:1) + ω2 n14×4 T k k2 z k2 z e2ωn π2l2 T = − n14×4 (cid:105)−1 k=0 T From the first to the second line, we took the sum over the chain index l and the integral over z(cid:48)(cid:48); 1 2πl2 δ2 −ikzz(cid:48)(cid:48)−ikym = (cid:88) (D22) k,0. 1 V dz (cid:48)(cid:48) e (cid:90) m Substituting Eq. (D21) into Eq. (D13), we obtain the imaginary-time optical conductivity as σzz(iωn) = e2ωn π2l2 (cid:126)e T + (cid:104) π2 k2 z (cid:0)DA − DT P T(cid:1) + ω2 n14×4 (cid:105)−1 k=0 U−1 (cid:126)e+. (D23) The k = 0 limit in the integrand is well-defined. To see this, use Taylor expansions of A and D in small k; AK = 2M (0)A0 + k2 DK = 2M (0)D0 + k2  1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1  , D0 ≡ zA1 + O(k), zD1 + O(k),  1 −1 1 −1 −1 1 −1 1 1 −1 1 −1 −1 1 −1 1  , with A0 ≡ and A1 ≡ D1 ≡ 1 π 1 π u4 K4 K1  u1  u1K1  , u2 K2 u3 K3 u4K4 u2K2 u3K3 28  . (D19) Since D0T P = 0 and D0A0 = 0, the integrand in the k = 0 limit takes a finite value; lim kz→0 (cid:0)DKAK − DKT P T(cid:1) (cid:1) = 2M (0)(cid:0)D0A1 + D1A0 (cid:16) lim k→0 1 k2 z = 2M (0)D0A1 + 2M (0) − 1 π − D1T P T gym(iωn) 4 (cid:17) U−1A0. From the second to the last line, we used T P T = gym(iωn) 4 A0 and πD1 = U−1. The imaginary-time optical conductivity is further cal- culated from Eq. (D23) as, σzz(iωn) = 2π2M (0)D0A1 (cid:16) (cid:20) + e2ωn π2l2 (cid:126)e T (cid:20) (cid:16) 4 (cid:17) + π 2M (0) − e2ωn π2l2 (cid:126)e T + π (D21) = (cid:17) gym(iωn) U−1A0 + ω2 n14×4 2M (0) − gym(iωn) U−1(cid:126)e+(cid:126)e T + (cid:21)−1 4 U−1 (cid:126)e+, + ω2 n14×4 ωn n + πuK(cid:0)2M (0) − gym(iωn) (cid:80) 4 ω2 (cid:1) , = e2uK π2l2 (cid:21)−1 U−1 (cid:126)e+ (D24) with uK ≡ line, we used A0D0 = 0, (cid:126)e T From the second to the last line, we used (cid:126)e T uK. a=1,2,3,4 uaKa. From the first to the second + D0 = 0 and A0 = (cid:126)e+(cid:126)e T + . + U−1(cid:126)e+ = In the clean limit (gy = 0), this gives σzz(ω) = after (e2uK)/(2πl2)δ(ω − ωg) with ωg ≡ 2πuK(cid:80) j M (2) the analytic continuation. j The effect of the disorder average is included in m(iωn). To see this effect in σzz(ω), let us take u1 = u4, K1 = K4, u2 = u3 and K2 = K3 for simplicity. With use of M (k) = 0 for k (cid:29) 1/l [28], we obtain the following expression for m(iωn), (cid:16) K1(cid:112) (cid:17) K2(cid:112) + n + ω2 ω2 1 n + ω2 ω2 2 , (D25) m(iωn) = π 2 with ω2 2. After the analytic continuation, we finally obtain the optical 1 ≡ 4πM (0)u1K1 < 4πM (0)u2K2 ≡ ω2 conductivity as follows,  σzz(ω) = e2uK πl2 ω∗ g(cid:48)(ω∗) δ(ω − ω∗) 0 < ω < ω1, ω1 < ω < ω2, e2uK π2l2 e2uK πl2 ωb1(ω) 1(ω)+b2 a2 ωb2(ω) 2(ω)+b2 a2 1(ω) 2(ω) ω1 < ω2 < ω, (D26) g(ω) = −ω2 + ω2 g gy 4 − π2uK 2 (cid:18) K1(cid:112) ω2 1 − ω2 (cid:19) , K2(cid:112) ω2 2 − ω2 + where and and magnetic order in the SNEI phases; 29 † ↑(r, A)ψ↓(r, A)(cid:105) = 0, (cid:17) (cid:88) † ↑(r, B)ψ↓(r, B)(cid:105) (cid:16) (cid:104)SA,+(r)(cid:105) = (cid:104)ψ (cid:104)SB,+(r)(cid:105) = (cid:104)ψ 1 Lx ∗ B,↑ηB,↓e (cid:88) (cid:16) = γ j × τ =± ∗ B,↑γB,↓e + η Y1,j(y)Y0,j(y) × † −iτ (kF,1+kF,4)z(cid:104)ψ (cid:17) 1,τ,jψ4,−τ,j(cid:105) † 3,τ,jψ2,−τ,j(cid:105) −iτ (kF,2+kF,3)z(cid:104)ψ (cid:88) Y1,j(y)Y0,j(y) j dyY1,j(y)Y0,j(y) = 0. (E1) because = 0, lim Lx→∞ 1 Lx (cid:90) = 1 2πl2 K2(cid:112) ω2 2 − ω2 , a1(ω) ≡ −ω2 + ω2 π2uK g − b1(ω) ≡ gy 4 2 π2uK gy 4 2 K1(cid:112) ω2 − ω2 1 a2(ω) ≡ −ω2 + ω2 g, π2uK b2(ω) ≡ gy 4 2 (cid:18) K1(cid:112) ω2 − ω2 1 , (cid:19) . K2(cid:112) ω2 − ω2 2 + Magnetism of the SNEI-I phase is most explicitly man- ifested by a long-range order of a symmetric part of a 2nd rank spin tensor composed of spin- 1 2 moment of A- carbon-site π-orbital electron and that of B-carbon-site. Such 2nd rank spin tensor has two components, QAB +−(r) ≡ (cid:104)SA,+(r)SB,−(r)(cid:105), QAB ++(r) ≡ (cid:104)SA,+(r)SB,+(r)(cid:105). Note that ω = ω∗(< ωg) in Eq. (D26) is one and only one solution of g(ω) = 0 within 0 < ω < ω1. The renor- malized gap ω∗ becomes progressively smaller, when the disorder strength increases. There exists a critical value of the disorder, gy,c ≡ 1 8ω2 gω1ω2 π2uK K1ω2 + K2ω1 . (D27) When gy approaches the critical value, the renormalized gap ω∗ reduces to zero continuously. At gy = gy,c, the system undergoes a quantum phase transition from the SNEI-I phase (gy < gy,c) to a disorder-driven phase (gy > gy,c). To obtain Fig. 5, we use the same parameter sets as in the appendix C4. We set u1 = u4 and u2 = u3 by Eq. (20). We set gy to be smaller than gy,c. Appendix E: magnetism and spin nematicity in SNEI phases SNEI phases introduced in the main text are charac- terized by particle-hole pairings between n = 0 LL with ↑ (↓) spin and n = −1 LL with ↓ (↑) spins. The phases break the U(1) spin rotational symmetry around the field direction. Nonetheless, neither A-carbon site π-orbital electron spin nor B-carbon site electron spins exhibit +−(r) vanishes identically, while ++(r) exhibits both a ferro-type and a density-wave- In the SNEI-I phase, QAB QAB type order; (cid:110) × QAB ++(r) = (cid:104)ψ 1 Lx = (cid:16)(cid:88) j (cid:17) 1 (cid:17) † ↑(r, B)ψ↓(r, B)(cid:105) (cid:16)(cid:88) Y 2 0,m(y) Lx m † ↑(r, A)ψ↓(r, A)ψ Y 2 0,j(y) γ + γ ∗ A,↑γA,↓ηB,↓η −2iΘ− ∗ B,↑e ∗ ∗ A,↑γA,↓ηB,↓η B,↑e ∗ ∗ B,↑ei∆Kze A,↑γA,↓ηB,↓η ∗ ∗ A,↑γA,↓ηB,↓η B,↑e −2iΘ− + γ + γ −i2Φ−−2iΘ− −i∆Kzei2Φ−−2iΘ−(cid:111) , with ∆K ≡ kF,2 + kF,3 − kF,1 − kF,4. Here we used Eqs. (45,46,47) and † 1,+,j(z)ψ4,−,m(z)(cid:105) = δjmiσ41,mei(φ1+φ4)+i(θ4−θ1), (cid:104)ψ † −i(φ1+φ4)+i(θ4−θ1), 1,−,j(z)ψ4,+,m(z)(cid:105) = δjmiσ41,me (cid:104)ψ † −i(φ2+φ3)−i(θ3−θ2), 3,−,m(z)(cid:105) = δjmiσ32,me (cid:104)ψ2,+,j(z)ψ † 3,+,m(z)(cid:105) = δjmiσ32,mei(φ2+φ3)−i(θ3−θ2). (cid:104)ψ2,−,j(z)ψ The spatial inversion symmetry generally allows ∗ A,↑γA,↓ηB,↓η ∗ B,↑ = u γ (E2) This gives Qab ++(r) = (cid:16) u + u cos(cid:0)∆Kz − 2Φ−(cid:1)(cid:17) . e−2iΘ− (π2l2)2 (E3) with Note also that the SNEI phases could be accompa- nied by a long-range ordering of small magnetic moments within the xy plane. Nonetheless, the moment does exist only in those spatial regions in the unit cell where two π- orbitals of A-carbon site and B-carbon site overlap. This statement is suggested by Eq. (E1) and finite expectation values of the following two quantities in the SNEI phases; † ↑(r, A)ψ↓(r, B)(cid:105) = † ↓(r, A)ψ↑(r, B)(cid:105) = (cid:104)ψ (cid:104)ψ −iΘ− cos(cid:0)(kF,1 + kF,4)z + Φ−(cid:1), πl2 eiΘ− cos(cid:0)(kF,2 + kF,3)z − Φ−(cid:1), √2iv πl2 e √2iw 30 v ≡ γ w ≡ γ ∗ A,↑ηB,↓(cid:104)σ14,j(cid:105) (cid:54)= 0, ∗ A,↓ηB,↑(cid:104)σ23,j(cid:105) (cid:54)= 0. ACKNOWLEDGMENTS RS appreciate helpful discussion with Zengwei Zhu, Benoit Fauque, Kamran Behnia, John Singleton, Miguel A. Cazalilla, Kazuto Akiba, Masashi Toku- naga, Toshihito Osada, Gang Chen, and Yoshihiro Iwasa. This work was supported by NBRP of China Grants No. 2014CB920901, No. 2015CB921104, and No. 2017A040215. [1] S. Tanuma, R. Inaba, A. Furukawa, O. Takahashi, Y. Iye and Y. Onuki, in Physics in High Magnetic Fields, edited by S. Chikazumi, and N. Miura, (Springer, Berlin, 1981), p. 316. [2] Y. Iye, P. M. Tedrow, G. Timp, M. Shayegan, M. S. Dres- selhaus, G. Dresselhaus, A. Furukawa, and S. Tanuma, Phys. Rev. B 25, 5478 (1982). [3] H. Fukuyama, Solid State Communications, 26, 783, K. A. Modic, F. F. Balakirev, and N. Harrison, Scientific Reports, 7, 1733 (2017). [21] Z. Zhu, P. Nie, B. Fauque, R. D. McDonald, N. Harrison, and K. Behnia, arXiv:1808.03558. [22] G. Gruner, Density Waves in Solids (Perseus Publishing, Cambridge, Massachusetts 2000). [23] B. I. Halperin, Jpn. J. Appl. Phys. 26, 1913 (1987). [24] L. Balents and M. P. A. Fisher, Phys. Rev. Lett. 76, 2782 (1978). (1996). [4] D. Yoshioka, and H. Fukuyama, J. Phys. Soc. Japan, 50, 275, (1981). [25] Y. Imry, and S. K. Ma, Phys. Rev. Lett. 35, 1399 (1976). [26] L. J. Sham, and B. R. Patton, Phys. Rev. B 13, 3151 [5] Y. Iye, P. M. Berglund, and L. E. McNeil, Solid State (1976). Communications, 52, 975 (1984). [27] H. Fukuyama, and P. A. Lee, Phys. Rev. B 17, 535 [6] Y. Iye, and G. Dresselhaus, Phys. Rev. Lett. 54, 1182 (1978). (1985). [28] . X. T. Zhang, and R. Shindou, Phys. Rev. B 95, 205108 [7] K. Takahashi, and Y. Takada, Physica B 201, 384 (1994). [8] Y. Takada, and H. Goto, J. Phys. Condens. Matter 10, (2017). [29] S. Uji, and J. S. Brooks, and Y. Iye, Physica B 246-247, 11315 (1998). 299 (1998). [9] K. Sugihara, Phys. Rev. B 29, 6722 (1984). [10] Z. Tesanovic, and B. I. Halperin, Phys. Rev. B 36, 4888 (1987). [11] A. H. MacDonald and Garnett W. Bryant, Phys. Rev. Lett. 58, 515 (1987). [12] V. M. Yakovenko, Phys. Rev. B 47, 8851 (1993). [13] H. Ochimizu, T. Kakamasu, S. Takeyama, S. Sasaki, and N. Miura, Phys. Rev. B 46, 1986 (1992). [14] H. Yaguchi and J. Singleton, Phys. Rev. Lett. 81, 5193 [30] Y. Kopelevich, B. Raquet, M. Goiran, W. Escoffier, R. R. daSilva, J. C. Medina Pantoja, I. A. Lukyanchuk, A. Sinchenko, and P. Morceau, Phys. Rev. Lett. 103, 116802 (2009). [31] A. Kumar, J. Poumirol, W. Escoffier, M. Goiran, B. Ra- quet, and J. C. Pivin, J. Phys. Condens. Matter 22, 436004 (2010). [32] P. R. Wallace, Phys. Rev. 71, 622 (1947). [33] J. C. Slonczewski, and P. R. Weiss, Phys. Rev. 99, 636 (1998). (A) (1955). [15] H. Yaguchi and J. Singleton, Physica B 256-258, 621 (1998). [16] H. Yaguchi, J. Singleton, and T. Iwata, Physica B 298, [34] J. W. McClure, Phys. Rev. 108, 612 (1957). [35] M. Inoue, J. Phys. Soc. Japan, 17, 808, (1962). [36] G. Dresselhaus, and M. S. Dresselhaus, Phys. Rev. 140, 546 (2001). 401 (A) (1965). [17] B. Fauque, D. LeBoeuf, B. Vignolle, M. Nardone, C. Proust, and K. Behnia, Phys. Rev. Lett. 110, 266601 (2013). [18] K. Akiba, A. Miyake, H. Yaguchi, A. Matsuo, K. Kindo, and M. Tokunaga, J. Phys. Soc. Japan, 84, 054709 (2015). [37] K. Nakao, J. Phys. Soc. Japan, 40, 761 (1976). [38] C. Biagini, D. L. Maslov, M. Y. Reizer, and L. I. Glaz- man, Euro. Phys. Lett. 55, 383 (2001). [39] S. W. Tsai, D. L. Maslov, and L. I. Glazman, Phys. Rev. B 65, 241102(R) (2002). [40] S. W. Tsai, D. L. Maslov, and L. I. Glazman, Physica B [19] F. Arnold, A. Isidori, E. Kampert, B. Yager, M. Eschrig, 312-313, 586 (2002). and J. Saunders, Phys. Rev. Lett. 119, 136601 (2017). [41] T. Giamarchi, Quantum Physics in One Dimension [20] Z. Zhu, R. D. MacDonald, A. Shekhter, B. J. Ramshaw, (Clarendon, Oxford 2003). 31 [42] E. W. Fenton, Phys. Rev. 170, 816 (1968). [43] D. Jerome, T. M. Rice, and W. Kohn, Phs. Rev. 158, 462 (1967). [44] A. A. Abrikosov, J. Low Temp. Phys. 2, 37 (1970); J. [55] J. T. Chalker and P. D. Coddington, Journal of Physics C: Solid State Physics, 21 2665 (1988). [56] C. M. Ho and J. T. Chalker, Phys. Rev. B 54 8708 (1996). [57] F. Evers, and A. D. Mirlin, Rev. Mod. Phys. 80, 1355 Low Temp. Phys. 3, 10 (1973). (2008). [45] S. A. Brazovskii, Zh. Eksp. Teor. Fiz. 62, 820 (1972) [Sov. Phys. JETP 35, 433 (1972)]; Zh. Eksp. Teor. Fiz. 61, 2401 (1971) [Sov. Phys. JETP 34, 1286 (1972)]. [46] K. Matsubara, T. Tsuzuku, and K. Sugihara, Phys. Rev. B 44, 11845 (1991). [47] A. L. Fetter and J. D. Walecka, Quantum Theory of Many-Particle Systems (Dover Publications, Mineola, New York, 2003). [48] G. Mahan, Many-Particle Physics Third Edition, (Kluwer Academic/Plenum Publishers, New York, 2000). [49] L. Fu, C. L. Kane, and E. J. Mele Phys. Rev. Lett. 98, [58] D. LeBoeuf, C. W. Rishchau, G. Seyfarth, R. Kuchler, M. Berben, S. Wiedmann, W. Tabis, M. Frachet, K. Behnia, and B. Fauque, Nature Communications, 8, 1337 (2017). [59] R. Shankar, Physica A 177, 530 (1991). [60] A. Houghton, and J. B. Marston, Phys. Rev. B 48, 7790 (1993). [61] M. Fabrizio, Phys. Rev. B 48, 15838 (1993). [62] A. H. Castro Neto, and E. Fradkin, Phys. Rev. Lett. 72, 1393 (1994). [63] A. H. Castro Neto, and E. Fradkin, Phys. Rev. B 49, 10877 (1994). 106803 (2007). [64] A. Houghton, H.-J. Kwon, and J. B. Marston, Phys. Rev. [50] Rahul Roy, Phys. Rev. B 79, 15322 (2009). [51] J. E. Moore, and L. Balents, Phys. Rev. B 75, 121306(R) (2007). B 50, 1351 (1994). [65] V. J. Emery, E. Fradkin, S. A. Kivelson, and T. C. Luben- sky, Phys. Rev. Lett. 85, 2160 (2000). [52] A. J. Heeger, S. A. Kivelson, J. R. Schrieffer, and W. P. [66] A. Vishwanath, and D. Carpentier, Phys. Rev. Lett. 86, Su, Rev. Mod. Phys. 60, 781 (1988). 676 (2000). [53] X. G. Wen and A. Zee, Nucl. Phys. B 316, 641 (1989). [54] M. Sato, and S. Fujimoto, Phys. Rev. B 79, 094504 (2009). [67] B. I. Halperin, Phys. Rev. B 25, 2185 (1982). [68] M. Buttiker, Phys. Rev. B 38, 9375 (1988). [69] P. Nozieres, and F. Gallet, J. Phys. (Paris) 48, 353 (1987).
1710.02152	1	1710	2017-10-05T18:00:07	Controlled parity switch of persistent currents in quantum ladders	[ "cond-mat.mes-hall", "quant-ph" ]	We investigate the behavior of persistent currents for a fixed number of noninteracting fermions in a periodic quantum ladder threaded by Aharonov-Bohm and transverse magnetic fluxes $\Phi$ and $\chi$. We show that the coupling between ladder legs provides a way to effectively change the ground-state fermion-number parity, by varying $\chi$. Specifically, we demonstrate that varying $\chi$ by $2\pi$ (one flux quantum) leads to an apparent fermion-number parity switch. We find that persistent currents exhibit a robust $4\pi$ periodicity as a function of $\chi$, despite the fact that $\chi \to \chi + 2\pi$ leads to modifications of order $1/N$ of the energy spectrum, where $N$ is the number of sites in each ladder leg. We show that these parity-switch and $4\pi$ periodicity effects are robust with respect to temperature and disorder, and outline potential physical realizations using cold atomic gases and, for bosonic analogs of the effects, photonic lattices.	cond-mat.mes-hall	cond-mat	Controlled parity switch of persistent currents in quantum ladders Michele Filippone, Charles-Edouard Bardyn, and Thierry Giamarchi Department of Quantum Matter Physics, University of Geneva, 24 Quai Ernest-Ansermet, CH-1211 Geneva, Switzerland We investigate the behavior of persistent currents for a fixed number of noninteracting fermions in a peri- odic quantum ladder threaded by Aharonov-Bohm and transverse magnetic fluxes Φ and χ. We show that the coupling between ladder legs provides a way to effectively change the ground-state fermion-number parity, by varying χ. Specifically, we demonstrate that varying χ by 2π (one flux quantum) leads to an apparent fermion- number parity switch. We find that persistent currents exhibit a robust 4π periodicity as a function of χ, despite the fact that χ → χ + 2π leads to modifications of order 1/N of the energy spectrum, where N is the number of sites in each ladder leg. We show that these parity-switch and 4π periodicity effects are robust with respect to temperature and disorder, and outline potential physical realizations using cold atomic gases and, for bosonic analogs of the effects, photonic lattices. Persistent currents [1, 2] are one of the most distinctive phe- nomena of mesoscopic systems in the presence of magnetic fields [3]. They provide a remarkable macroscopic manifesta- tion of the phase coherence of electrons in metallic rings, and of the nonlocal effects of magnetic fields in quantum physics: in the presence of a Aharonov-Bohm (AB) flux Φ threading a ring, electrons acquire a phase Φ upon looping around the ring, responding with a current that persists at thermal equi- librium, even in the absence of coupling to external reservoirs. The effects of magnetic fields in quantum systems have also been studied in numerous experiments focusing on bulk (and typically topological) properties of fermions or bosons in two- dimensional (2D) lattices under transverse magnetic fields. In the context of cold atomic gases [4], 2D fermionic [5–7] and bosonic [8–10] lattices with large transverse flux have been re- alized via synthetic gauge fields [11, 12]. Analogs have been implemented in photonic lattices [13–18], with recent real- izations of magnetic plaquettes hosting interacting photons in circuit cavity electrodynamics (circuit QED) [19, 20]. The above experimental platforms provide promising sys- tems to investigate the combined effects of AB and transverse magnetic fluxes [21] in the mesoscopic realm. The minimal setup hybridizing between 1D rings with AB flux and 2D lat- tices with a transverse field is a two-leg periodic quantum lad- der thread by both types of fluxes, as illustrated in Fig. 1. Ex- tensions to multi-leg ladders provide a realization of the cel- ebrated Laughlin's thought experiment for quantum-Hall sys- tems in cylinder geometry [21, 22]. In the mesoscopic setting, intriguing questions arise such as how the presence of a trans- verse flux affects the behavior of persistent currents. In this Letter, we show that the combination of AB and transverse magnetic fluxes in a periodic two-leg ladder with a fixed number of noninteracting fermions enables a controlled "parity switch" of persistent currents. Specifically, we demon- strate that changing the transverse flux χ by a single flux quan- tum (i.e., χ → χ ± 2π) modifies the behavior of the persis- tent current induced by the AB flux Φ as if we had, instead, switched the fermion-number parity. We find that the persis- tent current becomes 4π instead of 2π periodic, as a func- tion of χ and in specific regimes, upon introducing the cou- pling (rungs) between ladder legs. We identify the conditions FIG. 1. Top: Schematic setup consisting of a one-dimensional pe- riodic quantum ladder thread by a Aharonov-Bohm flux Φ (yellow) and a "transverse" flux χ (orange). Each lattice site (in red) can host a single spinless fermion, and particles can "hop" between sites along and across ladder legs (typically along the rungs illustrated here). This setup could be realized, e.g., using cold atomic gases or pho- tonic lattices (for bosonic analogs). Bottom: Parity switch revealed in persistent currents: when the parity of χ/(2π) is switched, the be- havior of the persistent current I(Φ) induced by varying Φ changes as if the parity P of the number of fermions had been switched [I(Φ) is measured in units of the Fermi velocity vF over the number N of lattice sites in each ladder leg]. required for such phenomena and discuss experimental plat- forms where they could be observed. The ladder system that we consider is illustrated in Fig. 1: it consists of two tunnel-coupled periodic chains of N sites - the "upper" (+) and "lower" (−) ladder legs - where each site can host a single spinless fermion. In the absence of dis- order and before introducing magnetic fluxes, both legs are described by the same quadratic Hamiltonian Hσ =Xi,j c†i,σ(cid:0)hk(cid:1)ij cj,σ, (1) where ci,σ annihilates a fermion on site i of leg σ (with i = arXiv:1710.02152v1 [cond-mat.mes-hall] 5 Oct 2017 0, . . . , N − 1 and σ = ±), and hk is a σ-independent matrix. The coupling between ladder legs is described by H+− =Xi,j c†i,+ (h⊥)ij cj,− + h.c. , (2) 2 where "h.c." denotes the Hermitian conjugate. The full ladder Hamiltonian H = H++H−+H+− is manifestly number con- serving, and we additionally assume that it is invariant under translations i → i + 1 along the ladder (by a lattice constant a = 1) and under time reversal (such that hk and h⊥ are real matrices [23]). To induce and control persistent currents, we introduce two types of (real or synthetic) magnetic fluxes: (i) an AB flux Φ threading the whole ladder "loop", and (ii) a transverse flux χ threading the area between ladder legs (see Fig. 1). The flux Φ plays the same role as the AB flux threading a single periodic chain of N sites: it generically breaks time-reversal symme- try and induces a persistent current I(Φ) = −h∂ΦHi/(2π) along the ladder [24], where h. . .i denotes the expectation value over the relevant state of the system (the ground state, at zero temperature). In position space, Φ can be described as a twisted boundary condition cN,σ = eiΦc0,σ, which translates as a global shift k → k + Φ/N in the crystal momentum (or quasimomentum) of the ladder. As opposed to Φ, the trans- verse flux χ has no analog in individual or decoupled chains: it does not induce any net current around the ladder but pro- vides, as we demonstrate below, a key level of control over the persistent current induced by Φ. A natural gauge for χ is given by the so-called Landau gauge [25], where χ is described as a phase eiχ/N for "hopping" from site i to i + 1 along the upper leg of the ladder only, which translates as a global momentum shift k → k + χ/N in the upper ladder leg. To maintain some degree of symmetry between ladder legs, we perform the gauge transformation cj,σ = eij(χ/2)/N cj,σ which transfers half of the hopping phase eiχ/N onto the lower leg while imposing the modified twisted boundary condition cN,σ = ei(Φ+χ/2)c0,σ. In this "symmetric gauge", the flux Φ threads "symmetrically" the two "rings" corresponding to in- dividual ladder legs, while χ threads both rings "antisymmet- rically" - inducing momentum shifts k → k + σ(χ/2)/N of opposite sign in opposite legs. In momentum space, the flux- dependent ladder Hamiltonian takes the 2 × 2 matrix form 2N(cid:1) H(k, χ) = c†k(cid:18) hk(cid:0)k + χ where ck ≡ (ck,+, ck,−) is a vector collecting the momentum- space analogs of the operators cj,±. Twisted boundary condi- tions lead to the quantization condition k ∈ {kn = 2πn/N + Φ/N + (χ/2)/N}, with n = 0, . . . , N − 1. In addition, χ is constrained to multiples of 2π (one flux quantum) to ensure that (i) the system Hamiltonian is periodic, and (ii) χ does not induce any current along the legs of the ladder when the inter-leg coupling is set to zero, as physically expected [26]. Shifts Φ → Φ + 2πm with integer m leave the system invari- ant [27]. Although shifts χ → χ+4πm (with integer m) leave 2N(cid:1)(cid:19) ck, h∗ ⊥(k) h⊥(k) hk(cid:0)k − χ (3) FIG. 2. Typical energy spectrum of a two-leg ladder around k = 0, at integer Φ/(2π). The transverse flux χ has two effects: (i) it shifts the two bands corresponding to upper and lower ladder legs (red and blue faint lines) by χ/N with respect to each other - in a similar way as a spin-orbit coupling. The inter-leg coupling h⊥(k) [Eq. (3)] then opens "gaps" (avoided crossings) at the time-reversal-invariant quasimomenta k = 0 and k = π (not shown) where shifted bands cross, leading to hybridized bands (mixed red and blue lines). (ii) The transverse flux additionally controls the allowed quasimomenta kn = 2πn/N +(χ/2)/N (with integer n). Crucially, single-particle eigenstates are only present at k = 0 when the number of transverse flux quanta χ/(2π) is even [filled dots, as opposed to the empty dots showing allowed states at odd χ/(2π)]. As a result, the parity of χ/(2π) controls the parity of the number of states below an arbitrary energy level E (gray horizontal line) in the gap. the set of allowed quasimomenta invariant, shifting χ by 4π also leads to modifications of the Hamiltonian of order 1/N [see Eq. (3)]. We thus expect typical physical properties to be invariant under shifts χ → χ + 4π up to corrections ∼ 1/N. Although fluxes break time-reversal (TR) symmetry, the Hamiltonian matrix in Eq. (3) is invariant under the effective TR symmetry defined by the antiunitary operator Θ = σxK, where K is the complex-conjugation operator and σx is the standard Pauli matrix. Since Θ2 = +1, however, Kramers' theorem does not hold and states ψ(k)i and Θψ(k)i (with the same energy but opposite momenta) need not belong to distinct bands. Similarly, states at the TR-invariant (TRI) mo- menta k = 0 and k = π need not be degenerate, which plays a crucial role in what follows. Note that Θ is only a symmetry of the whole system when it is also a symmetry of the twisted boundary conditions. Accordingly, the system is invariant un- der Θ when Φ = πm and χ = 2πm (with integer m). When ladder legs are decoupled [h⊥(k) = 0], Eq. (3) de- scribes two identical bands hk(k) shifted by χ/(2N ) in oppo- site, σ-dependent directions in momentum space (see Fig. 2). Seeing σ as an effective "spin", χ can thus be interpreted as a (Rashba) spin-orbit coupling [28, 29]. Momentum-shifted bands cross at the TRI points k = 0 and k = π where, cru- cially, states are present or absent depending on the parity of χ/(2π) [recall that kn = 2πn/N + (χ/2)/N, for inte- ger Φ/(2π)]. The coupling h⊥(k) between bands can be seen as a Zeeman field which opens "gaps" (avoided crossings) at the TRI momenta, thereby lifting the twofold degeneracy of states at these points (Fig. 2). Therefore, for integer Φ/(2π) 3 fermions, instead, a single fermion is shared between degen- erate states. The ground state is degenerate, and this degener- acy is lifted as soon as Φ 6= 2πm, leading to a discontinuity in I(Φ). This behavior "switches" (with "odd" ↔ "even" above) when Φ → Φ ± π, leading to the typical "sawtooth" behav- ior illustrated in the lower left panel of Fig. 1. Explicitly, for Φ ∈ (−π, π], one finds , I(Φ)even = vF N sgn(Φ)(cid:18)1 − Φ π (cid:19) . (4) We now show that inserting a full quantum of transverse flux χ leads to a similar "parity switch" as when introducing a half quantum of AB flux Φ, which is one of the main results of this work. The parity switch induced by χ can be under- stood by examining single-particle eigenstates and their occu- pation as a continuous function of χ: as illustrated in Fig. 3, nondegenerate states at TRI momenta turn into twofold de- generate states as the rest of the spectrum when the parity of χ/(2π) is modified. Therefore, when a TRI state lies below the Fermi energy (typically, the k = 0 state), higher-lying, twofold degenerate states at the Fermi energy must switch be- tween single and double occupation as the parity of χ/(2π) switches - irrespective of the (fixed) number of fermions in the system. Up to modifications ∼ 1/N of the spectrum [and, hence, of vF in Eq. (4)], everything happens as if the number of fermions was changed by ±1, i.e., as if the fermion-number parity was switched. This parity-switch effect is exemplified in the right part of Fig. 3: when χ/(2π) is even, the TRI quasi- momentum k = 0 is allowed, and the number of states below EF is odd. In that case, the usual behavior of persistent cur- rents [Eq. (4)] is observed. When χ is odd, instead, k = 0 is forbidden and the number of states below EF is even. In that case, the behavior of I(Φ) changes as if one had switched the fermion-number parity (see also the lower right panel of Fig. 1). We provide in the Supplemental Material [34] a quan- titative analysis of the parity-switch effect in a minimal lattice model [Eq. (3) with nearest-neighbor hoppings]. The above results imply that the persistent current I(Φ) ex- hibits a remarkable 4π periodicity in χ (and a trivial 2π peri- odicity in Φ): although changes χ → χ + 4π lead to O(1/N ) modifications of the bands, the main features of I(Φ) - zeros and discontinuities - are strictly 4π periodic in χ. Indeed, as argued above, such features are solely determined by energy crossings between single-particle eigenstates, and crossings are, here, strictly imposed at integer values of χ/(2π) by the effective TR symmetry Θ (as in Fig. 3). Figure 4 shows the persistent current as a function of Φ and χ, for the minimal lattice model detailed in the Supplemental Material [34]. As expected, the current exhibits a robust 4π periodicity in χ. We now comment on the robustness of parity-switch and 4π periodicity effects with respect to temperature, disorder, and system size. We recall that the existence of these effects stems from the effective TR symmetry Θ present at integer values of χ/(2π). Since temperature does not affect this symmetry, it does not lift the effects. Yet it does affect their visibility: a fi- Φ π vF N I(Φ)odd = − FIG. 3. Left: Typical single-particle energy spectrum (lower band of Fig. 2 in the limit of large inter-leg coupling) as a function of the transverse flux χ. The effective time-reversal symmetry Θ present at integer values of χ/(2π) (see text) enforces a twofold degeneracy of the entire spectrum, except for states at time-reversal-invariant quasi- momenta k = 0 or k = π. Occupied (empty) levels are indicated by filled (empty) dots. Right: Same single-particle eigenstates shown, instead, as a function of quasimomentum k, for χ = 0 and χ = 2π. When χ/(2π) alternates between even and odd parity, degenerate states at the Fermi energy EF alternate between double and single occupation, respectively, leading to the parity-switch effect discussed in the text (half-filled dots indicate states that share a single particle). The spectrum is 4π periodic in χ, up to corrections ∼ 1/N. and χ/(2π), the ladder spectrum is generically twofold degen- erate (with pairs of states at k and −k related by the effective TR symmetry Θ), except at the TRI points k = 0 and k = π where states are unique (see Fig. 3). We now examine the effects of the transverse flux χ on the persistent current I(Φ) = −h∂ΦHi/(2π). We focus, for pedagogical reasons, on small variations of Φ around integer values of Φ/(2π) where the system is invariant under Θ and the parity of χ/(2π) controls the existence of nondegenerate states at TRI momenta (Fig. 3). In that case, as we will demon- strate, changes in the parity of χ/(2π) induce changes in I(Φ) that mimic a "switch" of the fermion-number parity. In general, spectral degeneracies determine the main fea- tures of I(Φ): they lead to a known "parity effect" where I(Φ) is either discontinuous or zero depending on the par- ity of the (fixed) number of fermions in the system [30–33]. Here, the usual parity effect appears at χ = 0 when the Fermi energy (energy of the highest occupied energy level) crosses the lowest energy band two times (such that the lowest band is partially occupied; see Fig. 3). In that case, the persistent current reads I(Φ) = −(1/N )Pn vkn nkn, where vk = ∂kεk is the fermion velocity and nk is the occupation distribution of single-particle eigenstates with energy εk, and the parity effect manifests itself as follows: at Φ = 2πm (integer m), a pair of degenerate states is available at the Fermi energy. For an odd number of fermions, both states are occupied and the ground state is unique; occupied states contribute with oppo- site fermion velocities, and I(Φ) = 0. For an even number of nite temperature T > 0 spreads the occupation distribution of single-particle eigenstates, which decreases the overall ampli- tude of I(Φ). The patterns in Fig. 4 remain visible as long as T is lower than the typical energy separation ∼ EF /N between states at the Fermi energy. Disorder that does not break the Θ symmetry has a similar effect. When it breaks Θ, however, the situation changes: discontinuities of I(Φ) are smoothed out and parity switches need not occur at exact, integer values of χ/(2π) anymore, leading to an approximate (on average) 4π periodicity of I(Φ) in χ. A detailed discussion of disorder effects can be found in the Supplemental Material [34]. Our results can be extended to ladders with L > 2 legs where topological effects can lead, for L (cid:29) 2, to enhanced robustness against disorder. In general, for h⊥ = 0, each leg contributes a single band to the ladder spectrum: each band has a well-defined leg index l = 0, . . . , L − 1 (corresponding to the leg position in the y direction perpendicular to the legs), and neighboring bands (with index difference ∆l = 1) are shifted by (χ/N )/(L−1) with respect to each other, as shown in Fig. 2 for L = 2. The inter-leg coupling h⊥ mixes these bands and opens gaps that decrease exponentially with l1−l2 at crossings between bands l1 and l2 (see Supplemental Mate- rial [34]). Two pictures can then be distinguished depending on the value of χ around which small changes are then made: (i) For χ/N . 2π, the L minima of the L momentum- shifted bands all "fit" within the first Brillouin zone, such that band crossings at the nth lowest energy occur between bands with index difference ∆l = n. Accordingly, the size of the lowest energy gap (n = 1) is of order h⊥, while higher en- ergy gaps are exponentially smaller [34]. In this regime, hy- bridized bands can be seen as "Landau levels", in a similar way as in the coupled-wire construction for quantum Hall sys- tems introduced in Ref. [35]. As we show in the Supplemen- tary Material [34], parity switches induced by integer varia- tions of χ/(2π) are then observed whenever the number ν of occupied hybrized bands (or Landau levels) is an odd integer. This corresponds to the situation where the multi-leg ladder exhibits an odd number ν of chiral edge modes on each side of its cylinder geometry (i.e., around each extremal leg) [35]. The integer ν and the edge modes have a topological nature (see, e.g., Ref. [25]), which suppresses the effects of disorder: the overlap between (counter-propagating) modes on opposite edges can be made exponentially small by increasing L, which suppresses disorder-induced scattering between them. (ii) When χ/N & 2π, instead, some of the bands of individ- ual ladder legs are shifted beyond the first Brillouin zone, and are "backfolded" into it. When χ reaches values of the order of one flux quantum per unit cell, the system enters the con- ventional Harper-Hofstadter quantum Hall regime [36, 37], where gaps are mainly determined by the fraction p/q of flux quantum per unit cell (where q is a prime integer and p can take values from 1 to q). As in the low-flux situation (i) above, topological chiral edge modes typically appear in these gaps, and robust parity switches can be observed when the number ν of occupied edge modes per edge is odd. Disorder-induced scattering is similarly suppressed with increasing L [34]. 4 FIG. 4. Density plots of the persistent current induced by the AB flux Φ as a function of Φ and the transverse flux χ, for even (left) and odd (right) fermion-number parity P . The fermion number is fixed such that the Fermi energy lies in the "gap" at k = 0 (as the energy level E in Fig. 2), and positive (negative) values of the persistent current are shown in blue and red, respectively. Plots correspond to the minimal lattice model detailed in the Supplemental Material [34], with unit hopping strength between nearest-neighboring sites along and across ladder legs. The left and right plots correspond to 61 and 60 particles, respectively, for a system size N = 150. Changing P does not only lead to an apparent relative shift χ → χ ± 2π, as expected, but also to small differences ∼ 1/N that cannot be identified visually here. In view of their multi-leg generalization, the parity-switch and 4π periodicity effects identified in this work can be re- garded as mesoscopic analogs of more conventional Landau quantization effects [38] where, for free fermions in Landau levels, additional quanta of transverse flux χ introduce addi- tional states (or "cyclotron orbitals") per Landau level. While large fluxes (of the order of one flux quantum per unit cell) are typically required to observe such effects [39], the parity- switch effect identified in the present mesoscopic two-leg- ladder setting can be observed at arbitrarily low flux χ. Cold atoms trapped in optical lattices are prime candidates to realize the periodic ladder with AB and transverse fluxes studied in this work: in fact, a theoretical implementation of our setup (Fig. 1) has recently been proposed with cylindrical optical lattices generated by Laguerre-Gauss beams (rotated to induce a synthetic flux Φ) [21]. More broadly, periodic ladders with synthetic fluxes Φ and χ could be realized by taking advantage of internal degrees of freedom to simulate one or both of their spatial dimensions [6, 40, 41]. In particu- lar, one could realize a periodic geometry either in real space (as achieved for bosonic superfluids [42–44]), or along a syn- thetic dimension with periodic boundary conditions imposed by coupling extremal internal states [45, 46]. To observe the parity-switch and 4π periodicity effects in cold-atom experi- ments, one challenge would be to suppress particule-number fluctuations between measurements: when measuring the per- sistent current I(Φ) for different Φ and χ through repeated experiments, parity fluctuations generically lead to a statisti- cal average between the two panels of Fig. 4 [47]. The visibil- ity of the effects reflects the average fermion-number parity, and vice-versa. In particular, the periodicity of I(Φ) in Φ and χ is reduced from 4π to 2π when the average fermion-number parity vanishes (see Supplemental Material [34]). Periodic ladders with synthetic fluxes can also be realized with bosons, e.g., in photonic lattices [18, 48, 49]. Although there is no direct analog of Pauli blocking to suppress photon- number fluctuations, strong interactions can lead to the ana- log of persistent currents [20] and to an effective chemical potential [50]. Regardless, noninteracting photons described by the ladder Hamiltonian in Eq. (3) (with the same matrix but bosonic operators) would exhibit the same approximate 4π periodicity in χ, leading to physical observables with a similar periodicity. In conclusion, we have shown that the possibility to insert transverse-flux quanta χ/(2π) in quantum systems with a lad- der geometry - or, equivalently, a Corbino disk geometry - provides a robust way to perform controlled parity switches revealed in mesoscopic quantities such as persistent currents. The effect is accompanied by a remarkable 4π periodicity of physical observables in χ, up to corrections of order 1/N. An interesting direction for future work will be to examine how these effects are modified by interactions. We thank Dmitry Abanin, Marcello Dalmonte, Ivan Pro- topopov and Luka Trifunovic for useful discussions. We also acknowledge support by the Swiss National Science Founda- tion under Division II. [1] M. Buttiker, Y. Imry, and R. Landauer, Physics Lett. A 96, 365 (1983). [2] L. P. L´evy, G. Dolan, J. Dunsmuir, and H. Bouchiat, Phys. Rev. Lett. 64, 2074 (1990); I. O. Kulik, Low Temp. Phys. 36, 841 (2010); L. Saminadayar, C. Bauerle, and D. Mailly, in Encyclopedia of Nanoscience and Nanotechnology, eds. H.S. Nalwa, Valencia, CA, American Scientific 3, 267 (2004); A. Bleszynski-Jayich, W. Shanks, B. Peaudecerf, E. Ginossar, F. von Oppen, L. Glazman, and J. Harris, Science 326, 272 (2009). [3] E. Akkermans and G. Montambaux, Mesoscopic physics of electrons and photons (Cambridge university press, 2007). [4] I. Bloch, J. Dalibard, and W. Zwerger, Rev. Mod. Phys. 80, 885 (2008). [5] G. Jotzu, M. Messer, R. Desbuquois, M. Lebrat, T. Uehlinger, D. Greif, and T. Esslinger, Nature 515, 237 (2014). [6] M. Mancini, G. Pagano, G. Cappellini, L. Livi, M. Rider, J. Catani, C. Sias, P. Zoller, M. Inguscio, M. Dalmonte, et al., Science 349, 1510 (2015). [7] M. E. Tai, A. Lukin, M. Rispoli, R. Schittko, T. Menke, and D. Borgnia, P. M. Preiss, F. Grusdt, A. M. Kaufman, M. Greiner, arXiv preprint arXiv:1612.05631 (2016). [8] Y.-J. Lin, R. L. Compton, K. Jimenez-Garcia, J. V. Porto, and I. B. Spielman, Nature 462, 628 (2009). [9] H. Miyake, G. A. Siviloglou, C. J. Kennedy, W. C. Burton, and W. Ketterle, Phys. Rev. Lett. 111, 185302 (2013). [10] M. Aidelsburger, M. Atala, M. Lohse, J. T. Barreiro, B. Paredes, and I. Bloch, Phys. Rev. Lett. 111, 185301 (2013). [11] D. Jaksch and P. Zoller, New Journal of Physics 5, 56 (2003). [12] J. Struck, C. Olschlager, M. Weinberg, P. Hauke, J. Simonet, and P. Wind- A. Eckardt, M. Lewenstein, K. Sengstock, passinger, Phys. Rev. Lett. 108, 225304 (2012). 5 [13] F. D. M. Haldane and S. Raghu, Phys. Rev. Lett. 100, 013904 [14] Z. Wang, Y. Chong, J. Joannopoulos, and M. Soljaci´c, Nature [15] M. Hafezi, E. A. Demler, M. D. Lukin, and J. M. Taylor, Nature (2008). 461, 772 (2009). Physics 7, 907. [16] M. Hafezi, S. Mittal, J. Fan, A. Migdall, and J. Taylor, Nature Photonics 7, 1001 (2013). [17] K. Fang, Z. Yu, and S. Fan, Nature photonics 6, 782 (2012). [18] J. Ningyuan, C. Owens, A. Sommer, D. Schuster, and J. Simon, Phys. Rev. X 5, 021031 (2015). [19] J. Koch, A. A. Houck, K. L. Hur, and S. M. Girvin, Phys. Rev. A 82, 043811 (2010). [20] P. Roushan, C. Neill, A. Megrant, Y. Chen, R. Babbush, R. Barends, B. Campbell, Z. Chen, B. Chiaro, A. Dunsworth, et al., Nature Physics 13, 146 (2016). [21] M. Łacki, H. Pichler, A. Sterdyniak, A. Lyras, V. E. Lembessis, O. Al-Dossary, J. C. Budich, and P. Zoller, Phys. Rev. A 93, 013604 (2016). [22] R. B. Laughlin, Phys. Rev. B 23, 5632 (1981). [23] For spinless fermions, the relevant time-reversal operator coin- [24] We set = 1 and assume that fermions have a (real or syn- cides with the complex-conjugation operator K. thetic) charge e = 1. [25] B. A. Bernevig and T. L. Hughes, Topological insulators and topological superconductors (Princeton University Press, 2013). [26] When ladder legs are coupled as in Fig. 1, the flux χ can be de- scribed as a phase eijχ/N for hopping across ladder legs along each rung j = 0, . . . , N − 1. The quantization of χ is then re- quired for periodicity. When ladder legs are decoupled, instead, χ can simply be described as a twisted boundary condition e−iχ (via another gauge transformation), and the quantization of χ is then required both for periodicity and for ensuring that χ does not generate any current along the leg. [27] They correspond to trivial shifts of the Brillouin zone. [28] G. Dresselhaus, Phys. Rev. 100, 580 (1955). [29] A. Haller, M. Rizzi, and M. Burrello, arXiv preprint arXiv:1707.05715 (2016). [30] N. Byers and C. N. Yang, Phys. Rev. Lett. 7, 46 (1961). [31] R. Landauer and M. Buttiker, Phys. Rev. Lett. 54, 2049 (1985). [32] H.-F. Cheung, Y. Gefen, E. K. Riedel, and W.-H. Shih, Phys. Rev. B 37, 6050 (1988). [33] D. Loss and P. Goldbart, Phys. Rev. B 43, 13762 (1991). [34] See Supplemental Material. [35] C. L. Kane, R. Mukhopadhyay, and T. C. Lubensky, Phys. Rev. [36] P. G. Harper, Proceedings of the Physical Society. Section A 68, Lett. 88, 036401 (2002). 874 (1955). [37] D. R. Hofstadter, Phys. Rev. B 14, 2239 (1976). [38] L. Landau, Zeitschrift fur Physik 64, 629 (1930). [39] W. J. de Haas and P. M. van Alphen, Proc. Neth. R. Acad. Sci. , 1106 (1930). [40] L. F. Livi, G. Cappellini, M. Diem, L. Franchi, C. Clivati, M. Frittelli, F. Levi, D. Calonico, J. Catani, M. Inguscio, and L. Fallani, Phys. Rev. Lett. 117, 220401 (2016). [41] S. Kolkowitz, S. Bromley, T. Bothwell, M. Wall, G. Marti, A. Koller, X. Zhang, A. Rey, and J. Ye, Nature 542, 66 (2017). [42] K. C. Wright, R. B. Blakestad, C. J. Lobb, W. D. Phillips, and G. K. Campbell, Phys. Rev. Lett. 110, 025302 (2013). [43] S. Eckel, J. G. Lee, F. Jendrzejewski, N. Murray, C. W. Clark, C. J. Lobb, W. D. Phillips, M. Edwards, and G. K. Campbell, Nature 506, 200 (2014). [44] S. Eckel, F. Jendrzejewski, A. Kumar, C. J. Lobb, and G. K. Campbell, Phys. Rev. X 4, 031052 (2014). [45] A. Celi, P. Massignan, J. Ruseckas, N. Goldman, I. B. Spiel- man, G. Juzeli¯unas, and M. Lewenstein, Phys. Rev. Lett. 112, 043001 (2014). [46] O. Boada, A. Celi, J. Rodrguez-Laguna, J. I. Latorre, and M. Lewenstein, New Journal of Physics 17, 045007 (2015). [47] The same problem appears in solid-state metallic rings [2], where parity fluctuations effectively reduce the periodicity of persistent currents in Φ from 2π to π [51]. [48] W. Hu, J. C. Pillay, K. Wu, M. Pasek, P. P. Shum, and Y. D. 6 Chong, Phys. Rev. X 5, 011012 (2015). [49] S. Mittal, S. Ganeshan, J. Fan, A. Vaezi, and M. Hafezi, Nature Photonics 10, 180 (2016). 174305 (2015). [50] M. Hafezi, P. Adhikari, and J. M. Taylor, Phys. Rev. B 92, [51] H. Bouchiat and G. Montambaux, Journal de Physique 50, 2695 (1989); G. Montambaux, H. Bouchiat, D. Sigeti, and R. Fries- ner, Phys. Rev. B 42, 7647 (1990). (S3) (S4) 2N ) sin(k) sin( χ qsin2(k) sin2( χ 2N ) + τ 2, 2N(cid:17) ±rsin2(k) sin2(cid:16) χ 2N(cid:17) + τ 2(cid:21) ,  1 ± 1 2 uk,± =vuuut Supplementary Material for "Controlled parity switch of persistent currents in quantum ladders" Michele Filippone, Charles-Edouard Bardyn, and Thierry Giamarchi Department of Quantum Matter Physics, University of Geneva, 24 Quai Ernest-Ansermet, CH-1211 Geneva, Switzerland PARITY SWITCH AND 4π PERIODICITY IN AN EXPLICIT MINIMAL LATTICE MODEL In this section, we demonstrate the parity-switch and 4π periodicity effects presented in the main text in an explicit minimal lattice model corresponding to the two-leg ladder illustrated in Fig. 1 of the main text, and described by Eq. (3) thereof, with hoppings tk and t⊥ between nearest-neighboring sites along and across ladder legs, respectively. Explicitly, we start from a description in the standard Landau gauge where the intra-leg Hamiltonian takes the explicit form (S1) (S2) H+− = −t⊥ N−1Xj=0(cid:16)c†j,+cj,− + h.c.(cid:17) . (Φ+ 1+σ 2 χ)c†j+1,σcj,σ + h.c.i , iN N−1Xj=0he tk 2 Hσ = − with periodic boundary conditions, and the inter-leg coupling reads Performing the gauge transformation cj,σ = eij(Φ+χ/2)/N cj,σ (symmetric gauge), and moving to momentum space via the Fourier transformation cj,σ = Pk eikj ck,σ/√N (where k ∈ {kn = 2πn/N + Φ/N + (χ/2)/N} as in the main text), the ladder Hamiltonian H = H+ + H− + H+− takes the same form as in Eq. (3) of the main text, with the replacements hk[k ± χ/(2N )] = −tk cos[k ± χ/(2N )] and h⊥(k) = −t⊥. The momentum-space Hamiltonian can be diagonalized via a straightforward Bogoliubov transformation ck,± = u∓ dk,+ with dk,− ± u± where we have defined τ = t⊥/tk [S1, S2]. The eigenmodes dk,± correspond to two hybridized bands [Fig. 2 in the main text] ±(k) = −tk(cid:20)cos(k) cos(cid:16) χ shown in Fig. S1 for different values of χ and τ = t⊥ (setting tk = 1). We recall that k ∈ {kn = 2πn/N + Φ/N + (χ/2)/N}, where n = 0, . . . , N − 1. We focus on the situation where the Fermi energy EF lies in the "gap" (avoided crossing) opened by τ > 0 at k = 0, as ensured by the condition −τ < EF /tk + cos[χ/(2N )] < τ [S3]. As discussed in the main text, the parity-switch and 4π periodicity effects can be observed in that case. We demonstrate this explicitly in Fig. S1 for a relatively short ladder of N = 10 sites per leg [with Aharonov-Bohm flux Φ = 0 and an integer number of transverse-flux quanta χ/(2π)]. Figure S1 illustrates how the filling of single-particles eigenstates and the corresponding ground-state degeneracy change for different values of the inter-leg coupling τ. When τ is such that the Fermi level enters the upper band, the ground-state degeneracy does not change with modifications of χ by ±2π anymore, and the parity-switch and 4π periodicity effects disappear. We conclude this section by examining the situation where the number of fermions - and, hence, the parity thereof - is not controlled, which is typically the case in cold-atom experiments. For long ladders in which the behavior of the persistent current is typically described by Eq. (4) in the main text, one readily sees that the process of averaging measurements corresponding to distinct particle numbers (with completely random parity) leads to an effective reduction by a half of the periodicity of persistent currents in χ and Φ, as shown in Fig. S2. arXiv:1710.02152v1 [cond-mat.mes-hall] 5 Oct 2017 2 FIG. S1. Controlled parity switch in a short two-leg ladder of N = 10 sites per leg. Top row: Analog of Fig. 4 in the main text (where N = 150). For large τ where the lower band only is occupied, the 4π periodicity of the persistent current is clearly visible in this shorter ladder. Middle row: Energy spectrum at Φ = 0 and χ/(2π) = 2 (even), for different values of the inter-leg coupling τ. The filling of single- particle eigenstates is shown for different fermion numbers Nf (corresponding to distinct parities). In a similar way as in Figs. (2) and (3) of the main text, red filled squares indicate occupied states, while unfilled squares indicate degenerate states that share a single fermion. Note that, for a fixed number of fermions, the ground-state degeneracy changes as soon as τ is reduced enough for the upper band to become occupied. Lower row: Same as middle row, for an odd number of transverse-flux quanta χ/(2π) = 3. As opposed to the case with even χ/(2π), the ground-state degeneracy does not change when τ is decreased such that the upper band becomes occupied. EXTENSION TO MULTI-LEG LADDERS WITH WEAK TRANSVERSE FLUX - CONNECTION TO LANDAU LEVELS In the following two sections, we show how the parity-switch and 4π periodicity effects extend to multi-leg ladders, thereby providing explicit connections between the mesoscopic effects presented in the main text and more conventional quantum Hall effects. We start by focusing on scenarios where the transverse flux is weak, namely, χ/N . 2π, for a ladder with L ≥ 2 legs. In this regime, the parity-switch effect discussed in the main text can be interpreted as a mesoscopic manifestation, in a two-leg ladder, of changes in the number of occupied states per Landau level in larger, multi-leg ladders. For L ≥ 2 under the condition χ/N . 2π, the backfolding of bands into the first Brillouin zone is irrelevant for the low-energy physics. In that case, the behavior illustrated in Fig. (2) of the main text for L = 2 - where bands of individual ladder legs are shifted in momentum space by the transverse flux χ - readily generalizes to multiple bands (see Fig. S3). The situation is analogous to the one considered by Kane et al. in Ref. [S4], where continuous 1D systems with parabolic dispersion are tunnel-coupled to each other. The transverse flux χ shifts all bands by (χ/N )/(L − 1), and the inter-leg coupling h⊥ opens gaps at band crossings, leading to hybridized bands that can be interpreted as Landau levels [S4]. Note that gaps decrease exponentially as one moves towards higher energies where crossings occur between bands corresponding to more distant ladder legs [as h⊥ is the only (nearest-neighbor) direct coupling between legs (or bands)]. 3 ⇒ ⇒ FIG. S2. Average of persistent currents over measurements with different (random) fermion parity. Top: Plots on the left side correspond to Fig. (4) in the main text, while the plot on the right side corresponds to the average between the two, showing the apparent reduction by half of the periodicity on χ and Φ. Bottom: Same as the top row, for weaker coupling τ where the upper band is occupied. In that case, changing the parity of the number of fermions does not lead to a shift of π along the Φ axis anymore. The average of the plots for even and odd fermion-number parities does not lead to a complete reduction by half of the periodicity in χ and Φ of the persistent current anymore. In the usual Landau gauge, and in the limit of decoupled legs, the energy dispersion of individual legs with index l (where l = 0, . . . , L − 1) reads hl(k) = hk[k + l(χ/N )/(L − 1)]. (S5) As in the previous section, one can think of hl(k) as cosine bands with minima located at k = −l(χ/N )/(L− 1). At low energy close to these minima, the situation is thus similar to that of free fermions in the continuum with parabolic energy dispersions centered around the same values of k (see Ref. [S4]). As in the main text, the parity-switch effect can be understood more easily by moving to the symmetric gauge defined by the gauge transformation cj,l = eij(χ/2)/N cj,l. In this gauge, the symmetry of the system under the effective time-reversal symmetry operator Θ = σxK becomes apparent, where, for L > 2, the operator σx generalizes to a mirror symmetry around the center of the ladder system [exchanging leg indices l and (L − 1) − l]. As in the main text, χ imposes the twisted boundary condition cN,l = eiχ/2c0,l (independent of L). Therefore, for L ≥ 2, the presence of states at k = 0 is crucially allowed or forbidden depending on the parity of χ/(2π). As in two-leg ladders, changing χ by 2π generically leads to parity switches, as illustrated in Fig. S3 for L = 4. Specifically, the parity of the number of single-particle eigenstates appearing below a fixed energy changes when shifting χ → χ ± 2π if and only if the Fermi energy lies in a gap between Landau levels and the filling is such that an odd number ν of levels is occupied. The parity-switch and 4π periodicity effects are therefore sensitive to the parity of the number of occupied Landau levels (see Fig. S3). We emphasize that, in the limit where L is larger than the typical correlation length of the system (controlled by the gap ∼ 2h⊥), the integer ν coincides with the topological number of chiral edge states appearing at the edges of the ladder (around l = 0 and l = L− 1). These states can already be seen for small L = 4 in Fig. S3: for ν = 1, for example, two counter-propagating modes (at k and −k) are found in the gap, exponentially located at l = 0 and l = 3, respectively. We remark that the above picture holds provided that the transverse flux threads the lateral surface of the cylindric ladder system uniformly - at least before inserting or removing a small number of flux quanta to observe parity switches. The small changes χ → χ ± 2π ≡ χ + ∆χ required for parity switching, in contrast, need not be made in a completely uniform way. Additional flux quanta must only be inserted in a way that preserves: (i) translation invariance in the x direction along ladder legs, and (ii) the effective time-reversal symmetry Θ involving a mirror symmetry about the center of the ladder, in the y direction perpendicular to ladder legs. Condition (i) is satisfied provided that ∆χ is uniform in the x direction. Condition (ii), in contrast, 4 FIG. S3. Parity-switch effect for weak overall transverse flux χ/N . 2π. The schematic band structure depicted here corresponds to the low-energy spectrum of a ladder with L = 4 legs. This figure is the direct extension to a multi-leg ladder of Fig. (2) in the main text. In the symmetric gauge (see text), bands corresponding to individual ladder legs with index l always cross at k = 0 where single-particle eigenstates are present or not depending on the parity of χ/(2π). In the general case L ≥ 2, a similar parity-switch effect as discussed in the main text can be observed whenever an odd number ν of hybridized bands (or "Landau levels") is occupied. does not require ∆χ to be uniform in the y direction - it only requires the flux to be symmetric about the ladder center in the y direction. This has the following consequence for the observation of the parity-switch effect: in ladders with an odd number of legs L and, hence, an even number L − 1 of unit cells in the y direction, the insertion of a single flux quantum must be done uniformly in the y direction to preserve the symmetry Θ. For L even, instead, a single flux quantum can be inserted through the surface between ladder legs L/2 − 1 and L/2 without breaking Θ. Finally, we remark that χ must be modified by L − 1 quanta (one quantum per unit cell in the y direction) if one wants to ensure that the transverse flux does not induce any current along ladder legs in the limit where the inter-leg coupling vanishes. In that case, parity switches can only be observed in ladders with an even number of legs, where L − 1 is odd. EXTENSION TO MULTI-LEG LADDERS WITH LARGE TRANSVERSE FLUX - CONNECTION TO THE HARPER-HOFSTADTER MODEL ON A CYLINDER We now consider extensions of the two-leg ladder model discussed in the main text to multi-leg ladders with the same trans- verse flux in the case of large χ/N & 2π, namely, for transverse fluxes of the order of one flux quantum per plaquette. To inves- tigate this regime, we first notice that our model coincides, for multiple legs, with the standard Harper-Hofstadter model [S5, S6] with transverse flux χ, on a cylinder threaded by a Aharonov-Bohm flux Φ. Large fluxes χ generically induce the opening of topological gaps crossed by chiral edge states [S7], in which case the parity-switch and 4π periodicity effects investigated in the main text can exhibit an enhanced robustness against disorder. The same is true in the low-flux regime examined in the previous section. Here, however, topological gaps are not only controlled by the inter-leg coupling h⊥, and may be sizeable all across the energy spectrum. As we demonstrate below, all results presented in the main text are directly applicable to cases where the Fermi energy EF lies in a topological gap crossed, as in the low-flux regime, by an odd number ν of pairs of counter-propagating edge modes (where each mode crosses EF exactly once, as generically expected). Counter-propagating states at EF not only correspond to opposite quasimomenta k and −k, but are also located on opposite edges of the multi-leg ladder, leading to a cru- cial suppression of disorder-induced scattering between them (exponential suppression with increasing number of ladder legs, or increasing "bulk" size). The direct extension of the two-leg model defined by Eqs. (S1) and (S2) to multiple legs leads, in the standard Landau gauge, 5 FIG. S4. Left: Energy spectrum of a multi-leg ladder described by Eq. (S6) (Harper-Hofstadter Hamiltonian) for L = 48 legs of N = 60L sites each, with transverse flux 2π/3 per unit cell and hopping amplitudes t⊥/tk = 1/2. In this regime where a macroscopic transverse flux threads the system, the ladder exhibits topological gaps crossed by counter-propagating edge modes (thick colored lines). For the chosen flux, the energy dispersion of the edge modes exactly coincides with the band structure of a two-leg ladder with the same flux F per unit cell (and the same couplings t⊥, tk). Right: Zoom on the edge states alone in the region delimited by the two vertical dashed lines in the left plot. The smaller black dots correspond to the gapped bands in the left panel. The lower plot is the same as the upper one, except for an additional flux 2π/[(L − 1)N ] per unit cell - leading to the disappearance of states at k = 0 and to the parity-switch effect discussed in the main text. to the following Harper-Hofstadter model in cylinder geometry: HH-H = − tk 2 N−1Xx=0 qM−1Xy=0 heiy χ (L−1)N c†x+1,ycx,y + h.c.i − t⊥ N−1Xx=0 qM−2Xy=0 hc†x,ycx,y+1 + h.c.i, (S6) where χ = 2π(L − 1)N p/q is the (uniform) transverse flux threading the system (where q is a prime number and p can take values from 1 to q), x indexes positions along ladder legs, and y indexes ladder legs for a total of L = qM legs, with integer M. As in the main text, we consider periodic boundary conditions cx+N,y = cx,y, leading to the aforementioned cylinder geometry. The multi-leg ladder model defined by Eq. (S6) supports a variety of topological phases well suited for the observation of robust parity-switch and 4π periodicity effects. For concreteness, we focus on the special case of a transverse flux with p/q = 1/3, for which the multi-leg ladder features topological edge states whose energy dispersion exactly coincide with the band structure of the two-leg model examined in the main text (see Fig. S4 and discussion below). We start by demonstrating this remarkable correspondence: since q = 3, the magnetic unit cell (smallest cell containing an integer number of flux quanta) consists of 3 regular unit cells, and the spectrum of the system, accordingly, consists of 3 subbands separated by q − 1 = 2 gaps available for topological edge states. This shows that q = 3 is a necessary condition for the desired correspondence: the two bands of the two-leg ladder can only correspond to topological edge states in the multi-leg ladder if the latter exhibits exactly two topological gaps. To establish the full correspondence explicitly, one must choose a gauge in which the crystal momentum k in the x direction along ladder legs is preserved as in the two-leg ladder, i.e., one must use a gauge in which the magnetic unit cell is fully oriented along the y direction perpendicular to ladder legs, such that the system is invariant under usual translations (by one unit cell) in the x direction and invariant under magnetic translations (by 3 unit cells) in the y direction. In that case, the Schrodinger equation corresponding to Eq. (S6) can be expressed as ψx,y = − tk 2 eiy χ (L−1)N ψx+1,y − tk 2 e−iy χ (L−1)N ψx−1,y − t⊥ψx,y+1 − t⊥ψx,y−1, (S7) where ψx,y denotes the single-particle wavefunction on site (x, y). Equation (S7) is valid in the bulk, with straightforward modifications at edges corresponding to open boundary conditions in the y direction. Our goal is to find edge solutions that map to the modes of the two-leg ladder. Taking advantage of translation invariance in the bulk, we look for solutions of the (Bloch) form ψx,y = eikxeikyyuy, where uy is a periodic mode function satisfying uy+q = uq, k ≡ kx = 2πn/N with integer n is the 6 FIG. S5. Left: Ground-state energy EGS as a function of Φ for the multi-leg ladder described by Eq. (S6) (Harper-Hofstadter Hamiltonian) with on-site disorder as described by Eq. (S10). The zero of energy is set as the minimum of EGS in the absence of disorder. As in Fig. S4, we consider a system with transverse flux 2π/3 per unit cell and hopping amplitudes t⊥/tk = 1/2. The number of sites per ladder leg is fixed (N = 780), and we examine cases corresponding to a different number of legs L. In the clean case (black solid line), and for an odd number Nf of fermions with Fermi energy in the lower topological gap [we choose Nf = (L − 2)N/3 + 3N/4], the energy is continuous and minimum at Φ = 0, and the same for all L. When disorder is added (dashed-dotted curves), EGS is slightly shifted in a random direction along the Φ axis and discontinuities in ∂ΦEGS are smoothed out at small values of L where disorder-induced scattering between edge states at the Fermi energy is not entirely suppressed. The solid-dotted lines correspond to the same disorder realization with χ shifted by 2π(L − 1) (odd number of transverse-flux quanta), showing the robustness of the parity-switch effect. Right: Average difference ∆ between maximum and minimum of EGS as a function of Φ for increasing number of legs L, demonstrating the (exponential) increase in the robustness of ∆ (and, hence, the enhanced robustness of the parity switch of persistent currents) with increasing L. crystal momentum in the x direction, and ky is the analog of the crystal momentum in the y direction [which would take values ky = 2πm/(qM ) with integer m if the system was periodic in the y direction]. Plugging this ansatz into Eq. (S7), we obtain uy = −tk cos(cid:20)k + y χ (L − 1)N(cid:21) uy − t⊥eiky uy+1 − t⊥e−iky uy−1, (S8) which looks very similar to the Schrodinger equation of the two-leg ladder. To make the similarity even more apparent, we denote y ≡ (m, s), where m = 0, . . . , M − 1 indexes magnetic unit cells and s = 0, . . . , q − 1 indexes sites within the latter (or, equivalently, subbands). We then focus on the mode function u0y ≡ u0m,s = eikysuy, for which Eq. (S8) reduces to (S9) u0y = −tk cos(cid:20)k + y χ (L − 1)N(cid:21) u0y − t⊥eikyδs,q u0y+1 − t⊥e−ikyδs,0u0y−1. Although the "crystal momentum" ky is not a good quantum number due to edges in the y direction, exponentially decaying edge solutions can be found by making the replacement ky → iξ, where ξ is the corresponding localization length. By doing so, plane-wave propagation factors e±iky become exponential-decay envelope factors e±ξ, and Eq. (S9) reduces to the Schrodinger equation of the two-leg ladder: within a magnetic unit cell (i.e., for fixed m), the mode functions u0y ≡ u0m,s satisfy the same Schrodinger equation as the single-particle wavefunctions of a two-leg ladder with the same flux 2πp/q per unit cell. The edge solutions u0y of the multi-leg ladder correspond to copies of the states of the two-leg ladder translated by q sites in the y direction, with an exponentially decaying envelope ∝ e−ξ. More importantly, the energy dispersion of these edge modes coincides with the band structure of the two-leg ladder, as mentioned in the main text. The spectrum of the multi-leg ladder with transverse flux 2πp/q = 2π/3 per unit cell is shown in Fig. S4: the system is in a topological phase [S7] with q = 3 subbands and q − 1 = 2 topological gaps induced by the macroscopic transverse flux χ. Each gap is crossed by a pair of counter-propagating edge modes located on opposite edges of the cylinder. As expected, the energy dispersion of these topological modes coincides with the band structure of a two-leg ladder with the same flux per unit cell (Eqs. S1 and S2 with χ = 2πN/3). As argued above and in the main text, the parity-switch effect can also be observed in that case, induced by the disappearance/appearance of states at the time-reversal invariant quasimomentum k = 0 as χ is varied by ±2π: for an arbitrary energy level E set in one of the two topological gaps, the parity of the number of single-particle eigenstates below E switches every time χ is modified by ±2π. As in the low-flux regime discussed in the previous section, 7 transverse-flux quanta should be inserted in a uniform way or, more broadly, in a way that preserves translation invariance in the x direction along ladder legs, and the effective time-reversal symmetry Θ. To demonstrate that topology enhances the robustness of the parity-switch effect against disorder, we solve numerically the Harper-Hofstadter model defined by (S6) in the presence of local (on-site) disorder of the form Hdisorder =Xx,y εx,yc†x,ycx,y, (S10) with on-site energies εx,y uniformly distributed in the window [−W, W ] (where W can be regarded as the disorder "strength"). We examine the Φ dependence of the ground-state energy EGS of the system for an odd fixed number of fermions with Fermi energy in the lower topological gap. For "clean" systems (W = 0), the contribution to EGS of fermions in the bulk ("valence" band) is the same irrespective of the number of ladder legs (chosen as L = 3M + 2 with integer M, such that the system consists of an integer number of magnetic unit cells in the y direction). The derivative ∂ΦEGS is proportional to the persistent current along the periodic x direction of the cylinder, and completely filled bands do not contribute to this current. The left panel of Fig. S5 shows EGS as a function of Φ in the clean case and for individual realizations of the disorder potential, respectively, for an increasing number of legs L. For single realizations of the disorder, the energy generically does not exhibit a minimum at Φ = 0 anymore, which corresponds to the existence of a finite persistent current in the absence of any Aharonov-Bohm flux. This can be understood by noticing that the nonzero transverse flux χ induces chiral currents (as can be seen from the existence of counter-propagating edge modes), and that disorder generically favors a specific chirality by breaking the "mirror" symmetry between the two edges of the cylinder (the analog of the time-reversal symmetry Θ defined in the main text for a two-leg ladder). For L = 2, as expected, discontinuities in ∂ΦEGS are generally "smoothed out" by disorder as a result of Anderson localization [S8, S9]. More importantly, however, the effect of disorder is clearly suppressed with increasing number of ladder legs L (i.e., with increasing cylinder width). This suppression is a direct consequence of the topological nature of the two edge states at the Fermi energy: since the latter are located on opposite edges of the cylinder, disorder-induced scattering between them is strongly suppressed (exponentially with L). The right panel of Fig. S5 shows the average difference ∆ between the minimum and maximum of EGS as a function of Φ: as expected, ∆ becomes more stable against disorder as the number of legs L increases. We have also verified that the parity-switch effect, corresponding to an effective shift Φ → Φ ± π induced by χ → χ ± 2π, is increasingly robust against disorder for increasing L (solid-dotted lines in the left panel of Fig. S5). [S1] B. N. Narozhny, S. T. Carr, and A. A. Nersesyan, Phys. Rev. B 71, 161101 (2005); S. T. Carr, B. N. Narozhny, and A. A. Nersesyan, [S2] M. E. Tai, A. Lukin, M. Rispoli, R. Schittko, T. Menke, D. Borgnia, P. M. Preiss, F. Grusdt, A. M. Kaufman, and M. Greiner, arXiv Phys. Rev. B 73, 195114 (2006). preprint arXiv:1612.05631 (2016). [S3] To ensure that the Fermi energy crosses the lower band, one must also have EF /tk < −(sin2[χ/(2N )] + τ 2)1/2. [S4] C. L. Kane, R. Mukhopadhyay, and T. C. Lubensky, Phys. Rev. Lett. 88, 036401 (2002). [S5] P. G. Harper, Proceedings of the Physical Society. Section A 68, 874 (1955). [S6] D. R. Hofstadter, Phys. Rev. B 14, 2239 (1976). [S7] B. A. Bernevig and T. L. Hughes, Topological insulators and topological superconductors (Princeton University Press, 2013). [S8] H.-F. Cheung, Y. Gefen, E. K. Riedel, and W.-H. Shih, Phys. Rev. B 37, 6050 (1988). [S9] G. Bouzerar, D. Poilblanc, and G. Montambaux, Phys. Rev. B 49, 8258 (1994); M. Filippone, P. W. Brouwer, J. Eisert, and F. von Oppen, Phys. Rev. B 94, 201112 (2016).
1610.00152	1	1610	2016-10-01T15:44:36	An ultrafast polarised single photon source at 220 K	[ "cond-mat.mes-hall" ]	A crucial requirement for the realisation of efficient and scalable on-chip quantum communication is an ultrafast polarised single photon source operating beyond the Peltier cooling barrier of 200 K. While a few systems based on different materials and device structures have achieved single photon generation above this threshold, there has been no report of single quantum emitters with deterministic polarisation properties at the same high temperature conditions. Here, we report the first device to simultaneously achieve single photon emission with a g(2)(0) of only 0.21, a high polarisation degree of 0.80, a fixed polarisation axis determined by the underlying crystallography, and a GHz repetition rate with a radiative lifetime of 357 ps at 220 K. The temperature insensitivity of these properties, together with the simple planar growth method, and absence of complex device geometries, makes this system an excellent candidate for on-chip applications in integrated systems.	cond-mat.mes-hall	cond-mat	An ultrafast polarised single photon source at 220 K Tong Wang1†, Tim J. Puchtler1†, Tongtong Zhu2, John C. Jarman2, Luke P. Nuttall1, Rachel A. Oliver2, and Robert A. Taylor1 ABSTRACT: A crucial requirement for the realisation of efficient and scalable on-chip quantum communication is an ultrafast polarised single photon source operating beyond the Peltier cooling barrier of 200 K. While a few systems based on different materials and device structures have achieved single photon generation above this threshold, there has been no report of single quantum emitters with deterministic polarisation properties at the same high temperature conditions. Here, we report the first device to simultaneously achieve single photon emission with a g(2)(0) of only 0.21, a high polarisation degree of 0.80, a fixed polarisation axis determined by the underlying crystallography, and a GHz repetition rate with a radiative lifetime of 357 ps at 220 K. The temperature insensitivity of these properties, together with the simple planar growth method, and absence of complex device geometries, makes this system an excellent candidate for on-chip applications in integrated systems. 1Department of Physics, University of Oxford, Parks Road, Oxford, OX1 3PU, UK. 2Department of Materials Science and Metallurgy, University of Cambridge, 27 Charles Babbage Road, Cambridge, CB3 0FS, UK. †These authors contributed equally to this work. email: [email protected]; [email protected] Exploiting the quantum nature of light-matter interactions often relies on the presence of extreme conditions that cannot be applied outside the laboratory setting. In the case of polarised single photon sources1, most of today's semiconductor systems require cryogenic temperatures below 20 K for the observation of sub-Poissonian photon statistics2–9 and linearly polarised emission10– 15. In order to move from proof-of-concept polarised single photon emitters towards integration into scalable electronic platforms, these non-classical semiconductor light sources need to break this temperature barrier and operate above the thermoelectric cooling threshold, typically around 200 K. Such devices would then be the key building-block for truly secure on-chip quantum communication16–18 based on polarisation-reliant protocols, such as BB8419, and act as qubit sources for linear optical quantum computing20. Although several systems have demonstrated unpolarised single photon emission above this temperature barrier21–27, with some even reaching operation temperatures beyond ambient conditions28, there have been no direct reports of single photon emission with deterministic polarisation properties above the 200 K threshold. Polarisation control alone has proven difficult to achieve, and has only been demonstrated at < 10 K via complicated device geometry manipulation10 or the use of horizontally aligned nanowires13, making electrical contacting laborious and integration of multiple devices or batch production inherently difficult. Similarly, high temperature single photon emission, regardless of polarisation control, is challenging to achieve. In order to produce an increased degree of quantum confinement to combat the thermal activation of non-radiative exciton decay pathways, it is often necessary to use both wide bandgap materials, such as III-nitrides21–23,28, and structures with complex 3D growth routines, such as nanowire (NW)-quantum dots (QDs)23,25,28, which again pose challenges for the development of electrically driven devices. Moreover, although the use of nitrides allows large band-offsets29, the commonly used (0001) polar plane introduces the undesirable quantum- confined Stark effect (QCSE), thereby significantly reducing the exciton oscillator strength and decreasing the repetition rate and quantum efficiency. These complications and trade-offs in concurrently realising polarisation control, single photon emission, high temperature operation, and fast repetition rates, and the resulting complex device geometries, have significantly inhibited the incorporation of single photon emitters into realistically processable on-chip devices for applications in photonics. In this work, we report the first device to simultaneously achieve on-demand single photon generation, a high degree of optical linear polarization, an identifiable predefined polarisation axis, and an ultrafast repetition rate at 220 K (−53 ºC). The strong light emission in the blue spectral region also enables use of the most efficient available detectors, at a temperature that can be reached easily by commercial Peltier coolers. Our sample consists of non-polar (11-20) a- plane InGaN QDs grown by metal-organic vapour phase epitaxy (MOVPE) embedded within a p-i-n doped GaN matrix, allowing future fabrication of electrically pumped devices. Nanopillar structures, such as that shown in Fig. 1a, were post processed for increased photon extraction efficiencies. The ability to grow InGaN/GaN QDs on the a-plane not only minimises the unwanted QCSE, increasing the exciton oscillator strength and radiative decay rate, enhancing quantum efficiency, and reducing coupling to phonon-assisted non-radiative exciton decay routes30–32, but also causes highly polarised emission aligned to the [1-100] crystal axis33. Moreover, this high temperature ultrafast polarised single photon source requires only simple planar epitaxial growth, allowing devices to be fabricated easily using simple lithographic techniques. Figure 1 Temperature-resolved µ-PL of nanopillar-enhanced a-plane InGaN QDs. a, Scanning electron microscope (SEM) image and schematic of a representative nanopillar structure in which QDs are embedded. b, Temperature dependent µ-PL spectra at 4.7 K, and from 40 to 250 K at 30 K intervals, with peak intensities normalized to the emission at 4.7 K. c, Integrated intensity from QD spectra between 4.7 and 250 K, at 5 K intervals. The plotted line shows an Arrhenius- type fit which well describes the decrease in emission intensity due to carrier escape34. The optical properties of the QDs were assessed by micro-photoluminescence (μ-PL) measurements under 800 nm excitation (see Methods). The wavelength of 800 nm was chosen to allow 2-photon excitation of the sample, as QDs have a larger relative absorption cross-section for multi-photon processes than quantum wells (QW), and hence the QD:background ratio is improved35. An example of the typical emission features can be seen in Fig. 1b, exhibiting a single sharp peak at ~ 2.58 eV with a full width at half-maximum (FWHM) of 722 ± 65 µeV, attributed to the QD exciton and a low-intensity background caused by the fragmented QW underlying the QDs arising from the growth method30. As expected for a QD-bound exciton36, the QD emission both redshifts and broadens in linewidth with increasing temperature, as evident in Fig. 1b, due to the well-known bandgap shrinkage and exciton-phonon coupling respectively. At 220 K, the emission energy has redshifted to 2.54 eV, and its FWHM increased to 19.0 ± 0.4 meV. Furthermore, as thermally-assisted carrier escape becomes prominent at elevated temperatures, the emission intensity of the QD also decreases. As shown in Fig. 1c, the decrease in integrated intensity of the QD emission agrees very well with a standard single-channel semiconductor quenching model34. It is worth noting that the intensity decrease is slow, so that by 220 K, ~ 11% of the original intensity remains for the same excitation power. The strong QD emission at 220 K not only demonstrates operation well beyond the Peltier cooling barrier, but also allows us to study the QD's optical properties accurately. In order to demonstrate the emission of single photons, autocorrelation experiments were performed using a Hanbury Brown and Twiss (HBT) method under pulsed excitation. For the studied QD, histograms of the variation in the number of coincidences with the delay time, τ, are proportional to the second order autocorrelation function 𝑔(2)(𝜏) = 〈𝑛(𝑡)𝑛(𝑡 + 𝜏)〉 〈𝑛(𝑡)〉〈𝑛(𝑡 + 𝜏)〉 (1) where 𝑛(𝑡) is the number of photons recorded by the detectors. Hence, the presence of an 𝑛 = 1 Fock state for the verification of single photon emission requires the 𝑔(2)(0) < 0.5 condition37. Figure 2 Experimental evidence of ultrafast polarised single photon emission at both 4.7 (blue) and 220 K (red). a,b, Photon autocorrelation data, with raw and background-corrected g(2)(0) values demonstrating single photon emission. c,d, Time-resolved PL intensity plots of filtered QD signals from which the exciton decay constant is extracted. For greater accuracy, the fitted curves are double exponential functions convoluted with the near-Gaussian instrument response function. A slower, significantly weaker component becomes dominant after ~ 2 ns. e,f, Emission intensity variation with polariser angle. A Malus' law type sinusoidal fitting has been used to demonstrate that the emission is polarised, with the same polarisation axis at both low and high temperatures. At 4.7 and 220 K, we measure raw g(2)(0) values of 0.37 and 0.47 respectively (Fig 2a and 2b), thus confirming the single photon nature of the emission. This is the first direct proof of single photon emission from an a-plane InGaN QD. Moreover, unlike other non-nitride single photon systems such as Ref. 24, the g(2)(0) values are very insensitive to temperature changes, and only increased slightly at 220 K. This could be attributed to the high exciton binding energies and large band offsets in III-nitride systems. For this reason, we have even observed ultrafast anti- bunching behaviour of this QD at 250 K (see Supplementary Information). Non-zero raw g(2)(0) values are expected given the presence of spectrally overlapping background QW emission, as the QDs were formed on top of fragmented QWs during growth. Since single photon emission only comes from the QD; light of the selected wavelength from any other source present will increase the g(2)(0) values. To understand the g(2)(0) values expected from the QD alone, a commonly used background reduction calculation is performed. This background reduction can be accounted for using ρ, the ratio of QD intensity to the total intensity recorded by the PMTs, and the correction formula38 (2) (0) − 1 𝑔expt (2) (0) − 1 𝑔𝑐𝑜𝑟𝑟 = 𝜌2. (2) The 𝜌 was measured to be 85% at 4.7 K, and 82% at 220 K, yielding 𝑔corr (2) (0) values of 0.13 and 0.21 respectively (Fig. 2a and 2b). These values are much closer to 0, indicating that the QD is behaving as a pure SPS merely in the presence of a weak QW background. As we use non- resonant excitation, rapid repopulation and re-emission39 from the QDs could be reasons for the occasional emission of more than one photon. The finite response time of the PMTs and the fast exciton radiative lifetime could also contribute to the non-zero 𝑔corr (2) (0) values as explained below. The characteristic radiative lifetime of the QD has been assessed by performing time-resolved μ- PL. The resulting decay plots at 4.7 and 220 K can be seen in Fig. 2c and 2d. Fitting of this lifetime data has been performed using a modified Gaussian function 𝑓(𝑡) = 𝑓0 + (𝑓Gauss ⊗ 𝑓bi-exp)(𝑡) (3) where the instrument response function of the PMT detector (measured as a Gaussian with FWHM of 130 ps), 𝑓Gauss, is convoluted with a bi-exponential decay 𝑓bi-exp. The fitting gives a two-component exponential, with one intense fast component (~ 400 ps) attributed to the QD and another significantly less intense slow component (~ 4.0 ns). The fast component has a decay constant of 480 ± 20 ps at 4.7 K, decreasing to 357 ± 20 ps at 220 K. Such a decrease in decay time with increasing temperature is seen in CdSe QDs40 , although arsenide QDs typically show uncertain temperature dependence of radiative lifetimes41. The closeness of these radiative lifetimes to the finite detector response time of 130 ps not only requires a modified Gaussian function for analysis accuracy, but also contributes to the non-zero (2) (0) values obtained in the previous section. There is a significant probability that the 𝑔corr emission happened before 130 ps, during which time the PMTs could not distinguish the two events, thus adding to the histograms at 𝜏 = 0. This is also at least part of the reason that the (2) (0) at 220 K is higher than that at 4.7 K, given the even faster radiative lifetime at higher 𝑔corr temperatures. These decay times are significantly faster than those reported in c-plane InGaN QDs, which are typically a few ns, supporting the assertion that the use of the a-plane orientation for these QDs reduces the internal field of the QDs, reducing the QCSE and allows higher oscillator strengths and faster repetition rates30–32. Furthermore, the lifetimes remain short and relatively temperature insensitive across the studied temperature range, promising reliable GHz repetition under Peltier cooled conditions. The polarisation properties of the ultrafast single photon emission have been analysed by polarisation-resolved µ-PL. Our previous investigations33 have shown that the emission from an a-plane InGaN QD not only has a statistically high average degree of optical linear polarisation (DOLP) of 0.90 ± 0.08, but also possesses an intrinsic axis of polarisation along the crystal m- direction. PL intensities of the studied QD at different polariser angles at both 4.7 and 220 K have been recorded and analysed. As shown in Fig. 2e and 2f, the sinusoidal fittings in accordance with Malus' Law show that the emission is polarised at both temperatures. The maximum and minimum intensities recorded, 𝐼max and 𝐼min, were used to calculate the polarisation degree with the formula DOLP = 𝐼max − 𝐼min 𝐼max + 𝐼min . (4) DOLP values of 0.83 ± 0.01 and 0.80 ± 0.13 have been obtained for 4.7 and 220 K respectively, indicating highly polarised single photon emission at both low and high temperatures. The use of a-plane QDs breaks the nitride wurtzite symmetry and lowers it to orthorhombic, leading to hole state splitting and band mixing effects42. The resultant exciton hole ground state has a much higher contribution from the state associated with emission along the m-direction than the c- direction, yielding not only high DOLPs, but also a deterministic polarisation axis33,43. Indeed, the axis of polarization for all QDs observed lies along the m-direction of the sample. As such, the direction of polarisation for these polarised single photon emitters is predefined by the material crystallography, a simpler approach than attempting to use strain engineering for polarisation control10. After sample preparation, striations are visible along the sample surface, arising from the growth process, and are aligned perpendicular to the m-direction, making the identification of the polarisation axis straightforward in our experiments. Furthermore, the DOLP at 220 K indicates that the QD-bound exciton transition has similarly high 𝑚⟩-like characteristics at elevated temperatures, thereby confirming the ability of a-plane InGaN QDs to operate as on-chip polarised single photon sources. Figure 3 Comparison with other high temperature solid-state single photon sources. Representative single photon devices fabricated with various materials and device structures. A few reports demonstrate high temperature single photon emission, but without directly showing polarised photon emission at such elevated temperatures. In these cases, such as Ref 13, two different markers have been used indicating single photon and polarised single photon emission separately, based on the results reported. For applications of on-chip polarised single photon sources, a device must be operating in the top right section of the figure. Apart from the added advantages of a simple growth routine, planar structure, and ultrafast repetition rate, the two most important features for future device integration are simultaneous high temperature single photon emission and polarisation control. A review of state-of-the-art single photon emitters, to the best of our knowledge, is shown in Fig. 3, with the top right-hand section indicating the necessary condition for optimal on-chip single photon applications. Although many single photon sources, such as the mature arsenide cavity-QD systems, have been able to produce highly indistinguishable near perfect single photon emission2–9, their operations are limited to low temperatures, thereby precluding the possibility of thermoelectrically cooled on-chip operation on an electronic platform. Single photon emission at > 200 K is a challenge that only a few systems have achieved, as indicated in the top left-hand section of Fig. 3. However, most of the systems rely on NW-QD structures, SiC defect states, and diamond nitrogen vacancy centre engineering, which offer little prospect of electrically pumped devices. Furthermore, many of these devices have demonstrated either low temperature polarised single photon emission, or high temperature single photon emission, separately, which does not prove that they can be achieved simultaneously. The single photon emitters shown in the upper left-hand section, albeit reaching even beyond room temperature operation, lack the ability to generate polarised photons under the same conditions, unlike our a-plane InGaN/GaN QDs. In summary, we have demonstrated simultaneous single photon generation and polarised light emission with a predefined polarisation axis, an ultrafast GHz repetition rate, and operation at 220 K, well above the thermoelectric cooling barrier for device integration into electronic systems. The radiative lifetime, polarisation degree and orientation, and g(2)(0) are all temperature insensitive. This breakthrough brings us closer to the realisation of on-chip polarised single photon sources. Method Sample Preparation Non-polar (11-20) InGaN QDs were grown by a modified droplet epitaxy method as described previously30. InGaN QDs were positioned in the centre of a 50 nm thick intrinsic GaN layer, which was clad by 600 nm of n-doped GaN and 200 nm of p-doped GaN. Based on the analysis of atomic force microscopy (AFM) images of an uncapped InGaN QD sample, the QDs have an average height of ~ 7 nm, an average diameter of ~ 35 nm, and a density of approximately 1 × 109 cm-2. To isolate individual QDs, we have processed the as-grown wafer into nanopillar structures by drop-casting of silica nanospheres onto the wafer as an etch mask, followed by dry etching to a depth of ~ 350 nm. The residual silica nanosphere etch mask was then removed by ultra-sonication and a buffered-oxide etch. Details of the p-i-n sample growth and nanopillars processing conditions can be found in Ref 33. Such nanopillar structures provide better collimation and directionality for the emitted photon, thus higher photon extraction efficiency. Optical Characterization: Micro-photoluminescence experiments were performed with the sample mounted on a nano- positioning system (~ nm precision Attocube positioners) contained in a closed-cycle cryostat (AttoDRY 800), varying the sample temperature between 4.7 and 300 K. Excitation was provided using a mode-locked Ti:Sapphire laser operating at a wavelength of 800 nm (pulse duration of ~ 1 ps, repetition rate ~ 76 MHz). The excitation laser was focussed onto the sample through an objective lens (100×, 0.5 N.A.), with sample emission collected back through the same objective. Spectra are assessed using a 0.5 m focal length spectrometer (1200 l/mm grating). HBT experiments were performed by spectrally isolating the QD using a pair of tuneable bandpass filters and passing the filtered signal through a 50:50 beam-splitter connected to two PMTs. Signals from these PMTs are time- tagged using a time-correlated single-photon counting module with time resolution 25 ps. The PMTs define start and stop times individually. Similarly, lifetime data is collected by passing the spectrally filtered emission from the sample to a single PMT and using a fast-photomultiplier at the excitation laser as the start timer of the TCSPC module. Polarization measurements were performed by introducing a linear polariser and half-wave plate into the optical collection arm of the μ-PL system, with the transmission axis (0º marking) aligned to the PL component parallel to the [1-100] m-axis direction of the sample. The half- wave plate was used in order to maintain the same polarisation axis for light entering the spectrograph, in order to negate any polarisation dependent effects in the detection system itself. 1. Lounis, B. & Orrit, M. Single-photon sources. Rep. Prog. Phys. 68, 1129–1179 (2005). 2. Somaschi, N. et al. Near-optimal single-photon sources in the solid state. Nature Photon. 10, 340–345 (2016). 3. Madsen, K. H. et al. Efficient out-coupling of high-purity single photons from a coherent quantum dot in a photonic-crystal cavity. Phys. Rev. B 90, 155303 (2014). 4. Gazzano, O. et al. Bright solid-state sources of indistinguishable single photons. Nature Commun. 4, 1425 (2013). 5. Claudon, J. et al. A highly efficient single-photon source based on a quantum dot in a photonic nanowire. Nature Photon. 4, 174–177 (2010). 6. Reimer, M. E. et al. Bright single-photon sources in bottom-up tailored nanowires. Nature Commun. 3, 737 (2012). 7. Ding, X. et al. On-Demand Single Photons with High Extraction Efficiency and Near-Unity Indistinguishability from a Resonantly Driven Quantum Dot in a Micropillar. Phys. Rev. Lett. 116, 020401 (2016). 8. He, Y.-M. et al. On-demand semiconductor single-photon source with near-unity indistinguishability. Nature Nanotech. 8, 213–217 (2013). 9. Wei, Y.-J. et al. Deterministic and robust generation of single photons from a single quantum dot with 99.5% indistinguishability using adiabatic rapid passage. Nano Lett. 14, 6515–6519 (2014). 10. Lundskog, A. et al. Direct generation of linearly polarized photon emission with designated orientations from site-controlled InGaN quantum dots. Light: Science & Applications 3, e139 (2014). 11. Holmes, M. J., Kako, S., Choi, K., Arita, M. & Arakawa, Y. Linearly polarized single photons from small site-controlled GaN nanowire quantum dots. Proc. SPIE 9748, 9748E (2016). 12. Zhang, L., Teng, C.-H., Ku, P.-C. & Deng, H. Site-controlled InGaN/GaN single-photon- emitting diode. Appl. Phys. Lett. 108, 153102 (2016). 13. Deshpande, S., Heo, J., Das, A. & Bhattacharya, P. Electrically driven polarized single- photon emission from an InGaN quantum dot in a GaN nanowire. Nature Commun. 4, 1675 (2013). 14. Amloy, S. et al. Polarization-resolved fine-structure splitting of zero-dimensional InxGa1-xN excitons. Phys. Rev. B 83, 201307 (2011). 15. Jemsson, T., Machhadani, H., Holtz, P.-O. & Karlsson, K. F. Polarized single photon emission and photon bunching from an InGaN quantum dot on a GaN micropyramid. Nanotechnology 26, 065702 (2015). 16. Muller, A., Breguet, J. & Gisin, N. Experimental demonstration of quantum cryptography using polarized photons in optical fibre over more than 1 km. Europhys. Lett. 23, 383–388 (1993). 17. Kurtsiefer, C. et al. Quantum cryptography: A step towards global key distribution. Nature 419, 450 (2002). 18. Peng, C.-Z. et al. Experimental free-space distribution of entangled photon pairs over 13 km: Towards satellite-based global quantum communication. Phys. Rev. Lett. 94, 150501 (2005). 19. Bennett, C. H. & Brassard, G. Quantum cryptography: public key distribution and coin tossing. Proceedings of IEEE International Conference on Computers, Systems, and Signal Processing, 175–179 (1984). 20. Loss, D. & DiVincenzo, D. P. Quantum computation with quantum dots. Phys. Rev. A 57, 120 (1998). 21. Kako, S., Santori, C., Hoshino, K, Götzinger, S., Yamamoto, Y. & Arakawa, Y. A gallium nitride single-photon source operating at 200K. Nature Mater. 5, 887 (2006). 22. Deshpande, S., Frost, T., Hazari, A. & Bhattacharya, P. Electrically pumped single-photon emission at room temperature from a single InGaN/GaN quantum dot. Appl. Phys. Lett. 105, 141109 (2014). 23. Deshpande, S., Das, A. & Bhattacharya, P. Blue single photon emission up to 200K from an InGaN quantum dot in AlGaN nanowire. Appl. Phys. Lett. 102, 161114 (2013). 24. Fedorych, O., Kruse, C., Ruban, A., Hommel, D., Bacher, G. & Kümmell, T. Room temperature single photon emission from an epitaxially grown quantum dot. Appl. Phys. Lett. 100, 061114 (2012). 25. Bounouar, S. et al. Ultrafast room temperature single-photon source from nanowire-quantum dots. Nano Lett. 12, 2977−2981 (2012). 26. Mizuochi, N. et al. Electrically driven single-photon source at room temperature in diamond. Nature Photon. 6, 299–303 (2012). 27. Castelletto, S. et al. A silicon carbide room-temperature single-photon source. Nature Mater. 13, 151–156 (2014). 28. Holmes, M. J., Kako, S., Choi, K., Arita, M. & Arakawa, Y. Single photons from a hot solid- state emitter at 350 K. ACS Photonics 3, 543–546 (2016). 29. Jarjour, A. F., Oliver, R. A., Tahraoui, A., Kappers, E. J., Colin J. Humphreys, C. J. & Taylor, R. A. Control of the oscillator strength of the exciton in a single InGaN-GaN quantum dot. Phys. Rev. Lett. 99, 197403 (2007). 30. Zhu, T. et al. Non-polar (11-20) InGaN quantum dots with short exciton lifetimes grown by metalorganic vapor phase epitaxy. Appl. Phys. Lett. 102, 251905 (2013). 31. Griffiths, J. T. et al. Growth of non-polar (11-20) InGaN quantum dots by metal organic vapour phase epitaxy using a two temperature method. APL Materials 2, 126101 (2014). 32. Reid, B. P. L. et al. High temperature stability in non-polar (11-20) InGaN quantum dots: Exciton and biexciton dynamics. Phys. Status Solidi C 11, 702–705 (2014). 33. Wang, T. et al. Theoretical and experimental analyses of strongly polarized photon emission from non-polar InGaN quantum dots. Under review. Preprint at arXiv:1609.06885. 34. Gačević, Ž. et al. Internal quantum efficiency of III-nitride quantum dot superlattices grown by plasma-assisted molecular-beam epitaxy. J. Appl. Phys. 109, 103501 (2011). 35. Jarjour, A. et al. Two-photon absorption from single InGaN/GaN quantum dots. Physica E 32, 119–122 (2006). 36. Jagtap, A. M., Khatei, J. & Rao K. S. R. K. Exciton–phonon scattering and nonradiative relaxation of excited carriers in hydrothermally synthesized CdTe quantum dots. Phys. Chem. Chem. Phys. 17, 27579 (2015). 37. Buckley, S., Rivoire K. & Vučković, J. Engineered quantum dot single-photon sources. Rep. Prog. Phys. 75, 126503 (2012). 38. Brouri, R., Beveratos, A., Poizat, J.-P. & Grangier, P. Photon antibunching in the fluorescence of individual color centers in diamond. Optics Lett. 25, 1294–1296 (2000). 39. Holmes, M. J., Choi, K., Kako, S., Arita, M. & Arakawa, Y. Room-temperature triggered single photon emission from a III-nitride site-controlled nanowire quantum dot. Nano Lett. 14, 982–986 (2014). 40. Donegá, C. D. M., Bode, M. & Meijerink, A. Size- and temperature-dependence of exciton lifetimes in CdSe quantum dots. Phys. Rev. B 74, 085320 (2006). 41. Kong, L., Feng, Z. C., Wu, Z. & Lu, W. Temperature dependent and time-resolved photoluminescence studies of InAs self-assembled quantum dots with InGaAs strain reducing layer structure. J. Appl. Phys. 106, 013512 (2009). 42. Badcock, T. J. et al. Optical polarization anisotropy of a-plane GaN/AlGaN multiple quantum well structures grown on r-plane sapphire substrates. J. Appl. Phys. 105, 123112 (2009). 43. Patra, S. K., Marquardt, O. & Schulz, S. Polar, semi- and non-polar nitride-based quantum dots: influence of substrate orientation and material parameter sets on electronic and optical properties. Opt Quant Electron 48, 151 (2016). 44. Stevens, M. J., Hadfield, R. H., Schwal, R. E., Nam, S. W. & Mirin R. P. IEEE J. Sel. Top. Quantum Electron. 12, 1255–1268 (2006). 45. Bommer, M. et al. Triggered single-photon emission in the red spectral range from optically excited InP/(Al,Ga)InP quantum dots embedded in micropillars up to 100 K. J. Appl. Phys. 110, 063108 (2011). 46. Kako, S., Holmes, M., Sergent, S., Bürger, M., As, D. J. & Arakawa, Y. Single-photon emission from cubic GaN quantum dots. Appl. Phys. Lett. 104, 011101 (2014). Supplementary Information: An ultrafast polarised single-photon source at 220 K Tong Wang1,, Tim J. Puchtler1,, Tongtong Zhu2, John C. Jarman2, Luke P. Nuttall1, Rachel A. Oliver2, and Robert A. Taylor1 1Department of Physics, University of Oxford, Parks Road, Oxford, OX1 3PU, UK. 2Department of Materials Science and Metallurgy, University of Cambridge, 27 Charles Babbage Road, Cambridge, CB3 0FS, UK. [email protected] *[email protected] QD PERFORMANCE AT 250 K PL Characteristics and QD:Background Estimation The strong confinement and exciton binding energy of nitrides allow us to observe QD emission at temperatures higher than 220 K. Figure S1 displays the µ-PL spectrum of the studied dot at 250 K, a temperature even closer to ambient conditions, and easier to reach for on-chip thermoelectric cooling. This is the highest operation temperature of a-plane InGaN QDs ever reported. Despite significant carrier recombination through non-radiative pathways, the sharp emission feature of the QD emerges from the QW background with a peak intensity of more than 50 counts/s on our µ-PL system. Under the same excitation power, the integrated intensity of the QD remains ~ 6% of that at 4.7 K, which again shows the relatively slow thermal quenching of emission. At this temperature, the QD energy has redshifted to ~ 2.53 eV. Phonon-assisted homogeneous broadening has also increased the QD linewidth to 24.8 ± 0.5 meV. While a Voigt function that combines a Gaussian and a Lorentzian might provide a more accurate fitting, a single Lorentzian profile, as shown in Fig. S1, does indeed model the experimental data well enough. The underlying QW emission has been fitted to a Gaussian with a much wider linewidth of ~ 80 meV. As an example of the background estimation method used in the main manuscript, the integrated intensity of the QD and the background QW, within the spectral window indicated in Fig. S1, have been calculated. The ratio of the QD intensity to the total intensity, ρ, can hence be estimated accordingly. In the 250 K case, the ρ has been estimated to be ~ 57%. However, the difficulty in ascertaining the exact spectral coverage to be included in the intensity integration is a limitation of this estimation method. The band pass filters used in the spectral selection do not have perfect boxcar-like transmission profiles, adding uncertainty to the background estimation. Figure S1 µ-PL of the studied QD at 250 K. The sharp emission feature of the QD has been fitted to a Lorentzian, and the background QD emission to a Gaussian. The dotted line shows the spectral window with which the integrated intensities of the QD and QW emission have been estimated. Photon Antibunching The emission of the studied QD at 250 K has been passed to the HBT setup of the optical system, and the photon autocorrelation data recorded in Fig. S2. Despite the large amount of phonon- induced linewidth broadening and ~ 43% of background QW emission, the studied QD shows antibunching behaviour with a raw g(2)(0) value of 0.73. Although this uncorrected result is more than 0.5, and thus not sufficient to directly provide evidence for single photon generation, the g(2)(0) value can be corrected with equation (2) in the main manuscript text. With a ρ of 57% as estimated earlier, the corrected g(2)(0) becomes 0.17, thereby indicating that the QD is indeed emitting as a single photon source at 250 K in the presence of underlying background QW emission. However, the uncorrected g(2)(0) should still be treated as the main performance indicator of the QD, and background reduction techniques would be needed for cleaner single photon emission at such elevated temperatures. 250 K is hence the highest temperature at which photon antibunching has been observed for a-plane InGaN/GaN QDs. Figure S2 Autocorrelation data from HBT experiments of the studied QD at 250 K. The value in the bracket is the background-corrected g(2)(0) value calculated using the method described in the main manuscript. The raw data provides direct evidence for antibunching behaviour, while the corrected g(2)(0) shows that the QD is emitting as a single photon source in the presence of overlapping QW emission. Radiative Lifetime Time-resolved µ-PL has also been performed on the same QD to measure its radiative lifetime at 250 K. The fitting in Fig. S3 is performed using a modified Gaussian function, convoluted with an exponential decay, as described in the main manuscript. The decay constant has been found to be 191 ± 8 ps, indicating similarly strong electron and hole wavefunction overlap even at 250 K, and thus ultrashort radiative lifetime. The presence of a slower component after 0.6 ns, similar to that at 4.7 and 220 K, is still significantly weaker (< 10% intensity). Hence, the fast component is attributed to the QD emission, and agrees with expected radiative lifetimes of a-plane InGaN/GaN QDs. As such, we have shown strong, ultrafast, and antibunched photon generation from the studied QD at 250 K. Figure S3 Radiative lifetime measurements with time-resolved µ-PL at 250 K. TCSPC data has been fitted with a Gaussian function convoluted with an exponential. The lifetime calculation shows ultrafast antibunched photon emission at 250 K.
1102.4113	2	1102	2011-08-10T09:07:10	Multi-resonance of energy transport and absence of heat pump in a force-driven lattice	[ "cond-mat.mes-hall", "cond-mat.stat-mech" ]	Energy transport control in low dimensional nano-scale systems has attracted much attention in recent years. In this paper, we investigate the energy transport properties of Frenkel-Kontorova lattice subject to a periodic driving force, in particular, the resonance behavior of the energy current by varying the external driving frequency. It is discovered that in certain parameter ranges, multiple resonance peaks, instead of a single resonance, emerge. By comparing the nonlinear lattice model with a harmonic chain, we unravel the underlying physical mechanism for such resonance phenomenon. Other parameter dependencies of the resonance behavior are examined as well. Finally, we demonstrate that heat pumping is actually absent in this force-driven model.	cond-mat.mes-hall	cond-mat	Multi-resonance of energy transport and absence of heat pump in a force-driven lattice Song Zhang1, Jie Ren2,1,∗ and Baowen Li1,2† 1Department of Physics and Centre for Computational Science and Engineering, National University of Singapore, Singapore 117546, Republic of Singapore 2NUS Graduate School for Integrative Sciences and Engineering, Singapore 117456, Republic of Singapore (Dated: October 10, 2018) Energy transport control in low dimensional nano-scale systems has attracted much attention in recent years. In this paper, we investigate the energy transport properties of Frenkel-Kontorova lattice subject to a periodic driving force, in particular, the resonance behavior of the energy cur- rent by varying the external driving frequency. It is discovered that in certain parameter ranges, multiple resonance peaks, instead of a single resonance, emerge. By comparing the nonlinear lattice model with a harmonic chain, we unravel the underlying physical mechanism for such resonance phe- nomenon. Other parameter dependencies of the resonance behavior are examined as well. Finally, we demonstrate that heat pumping is actually absent in this force-driven model. PACS numbers: 05.60.-k, 44.10.+i, 66.70.-f, 07.20.Pe I. INTRODUCTION well [13]. In the last decade much attention has been paid to heat conduction on the nanoscale. Numerous studies in the quest of manipulating heat have brought substan- tial progresses in the area of Phononics, the science and engineering of phonons [1]. Theoretical models of var- ious thermal devices such as thermal rectifier [2], ther- mal transistor [3], thermal logic gates [4] and thermal memory [5] have been proposed to control phonon-based thermal transport. In addition, some experimental works have been carried out. For instance, solid-state thermal diodes based on asymmetric nanotube [6], asymmetric cobalt oxides [7], the nanotube phonon waveguide [8] and solid thermal memory[9] have been realized experimen- tally. In the works mentioned above, heat always flows from high temperature region to low temperature one. This is in accordance with the second law of thermody- namics. However, just like its macroscopic counterpart, a heat pump which directs heat against thermal bias by using external force exists on the molecular level. Recent stud- ies have suggested several such kind of models based on different mechanisms. Li et al. proposed a heat ratchet to direct heat flux from one bath to another in a non- linear lattice, which periodically adjusts two baths' tem- peratures while the average remains equal [10]. Later, Ren and Li demonstrated that heat energy can be rec- tified between two baths of equal temperatures at any instant and the correlation between the baths can even direct energy current against thermal bias without exter- nal modulations [11]. Beyond those classical models, a quantum heat pump consisting of a molecule connected to two phonon baths was proposed recently [12], and the Berry phase effect induced heat pump was unraveled as Nevertheless, some contradiction exists in the force- driven heat pump. Marathe et al showed that under peri- odic force driving, two coupled oscillators connected with thermal baths fail to function as a heat pump [14]. Later Ai et al claimed the heat pumping appeared in Frenkel- Kontorova (FK) chain under the influence of a periodic driving force [15]. In addition, they observed a ther- mal resonance phenomenon that the heat flux attained a maximum value at a particular driving frequency, which is similar to the previously found resonance induced by driving bath temperatures [11]. However, other than the temperature driven case, a clear physical mechanism of the driving-force-induced resonance is still unavailable. In this paper, we investigate the force-driven FK chain as a typical nonlinear lattice model and analyze ther- mal properties which are frequency dependent. We dis- cover that multiple resonance peaks, instead of a sin- gle peak, emerge in certain parameter ranges. The ori- gin of this phenomenon in fact relies on the eigenfre- quencies of this system. The paper is organized as fol- lows. First of all, we will introduce the FK model and briefly show the crossover from a single resonance to mul- tiple resonances. To look into the underlying mecha- nism of the multi-resonance phenomenon, we invoke a harmonic model through which analytic expressions of various quantities are obtained. Then the resonance be- havior of the FK model is discussed in detail. We will show that as far as the resonance property is concerned, there is much similarity between the FK model and the harmonic model. The effect of different system param- eters on the resonance phenomenon is explored as well. In the end, we shall briefly demonstrate that the force- driven model fails to perform as a heat pump. II. MODEL AND RESULTS ∗Electronic address: [email protected] †Electronic address: [email protected] We start with a FK chain coupled to Langevin heat baths at two ends. A periodic force is applied to the left TL TR f(t)=A0sin(Ωt) FIG. 1: (Color online) Schematic of a one-dimensional FK lattice coupled to two Langevin baths. The left end is at- tached to a periodic driving force f (t). end (See Fig. 1). The Hamiltonian of the system is given as: N H = + p2 i 2m 1 Xi 2 − δiαxif (t), k(xi − xi+1)2 − V (2π)2 cos( 2π a xi) (1) where f (t) = A0 sin(Ωt) is the driving force with am- plitude A0 and frequency Ω imposed on particle α. xi denotes the displacement of the ith particle and pi the corresponding momentum. k is the spring constant, V the strength of the on-site potential and a = 1 the spac- ing between adjacent particles. We set α = 1 in follows unless otherwise stated. Under fixed boundary condition, the equation of motion (EOM) of the coupled particles (i = 2,··· N − 1) reads: mx1 = − mxi = − mxN = − V 2π − γ x1 + ηL(t), V 2π V 2π − γ xN + ηR(t). sin(2πx1) + k(x2 − 2x1) + f (t) sin(2πxi) + k(xi+1 + xi−1 − 2x1), sin(2πxN ) + k(xN −1 − 2xN ) (2a) (2b) (2c) The two noise terms are white noise with hηi(t)ηj (t′)i = 2γkBTiδijδ(t − t′), (i, j = L, R), where γ is the friction coefficient, kB is the Boltzmann's constant and TL(R) de- notes the temperature of L(R) reservoir. Without loss of generality, we set m = 1, kB = 1 and integrate the EOM by the symplectic Velocity Verlet algorithm. The simu- lation are performed long enough (of order 108) with a time step of 0.005 to allow the system to reach a non- equilibrium steady state. The local energy current is defined as Ii(t) = h−k xi(xi − xi−1)i, where h···i de- notes the ensemble average. After the transient time, the transport approaches an oscillatory steady state and the periodic average of local current is equivalent at each site i, reading Ji = Ω 2π Z 2π/Ω 0 Ii(t)dt. (3) There is no doubt that the thermal resonance exists for the energy current as a function of the driving frequency. 0.8 0.6 0.4 J 0.2 0.0 -0.2 0.0 A0=3.2 A0=1.0 0.4 0.8 1.2 1.6 2.0 2 0.05 0.04 0.03 0.02 0.01 0.00 -0.01 FIG. 2: (Color online) Energy current J vs the driving fre- quency Ω in the force-driven FK model. Parameters are TL = 0.4, TR = 0.6, k = 1, V = 5, γ = 1, and N = 64. J denotes Ji (i > α = 1). The left arrow means the upper curve corresponds to the left scale while the right arrow means the lower curve corresponds to the right scale. At large Ω, the response of the system cannot catch up with the fast driving as if there is no driving. At small Ω, the system reduces to a quasi-static counterpart as if there is no driving as well. Therefore, at these two regions, J < 0 since TL < TR. However, surprisingly, we observe that multiple reso- nance peaks emerge when the amplitude A0 is increased, as shown in Fig. 2. To analyze this multi-resonance phe- nomenon further, we invoke the harmonic chain model, of which clear analytic results are possible. A. Analytic results for force-driven harmonic lattice Harmonic potential is a second order approximation of realistic potential around its minimum. Under har- monic approximation, phonons will not interact with each other. Starting from this noninteracting picture, interactions within phonon gas can be introduced, by in- cluding higher order terms in the potential expansion. Compared to non-linear models such as the FK lattice, a harmonic model has the advantage of having an exact so- lution and it gives many satisfactory explanations about thermal properties of the crystal. The linear harmonic model considered in this section is obtained by replacing the nonlinear on-site potential in Eq. (1) by 1 i , where ko is the force constant of the on-site harmonic potential. As we shall see, under certain conditions the effect of the driving force can be separated from that of the thermal reservoirs. The EOM in a compact matrix form is: 2 kox2 M X = −ΦX − Γ X + η(t) + F (t), (4) where M is the mass matrix set as identity matrix and Φ the real symmetric force matrix, with Φij = (2k + ko)δij − kδi,j+1 − kδi,j−1. Γij = γδij(δi1 + δiN ) denotes the coupling to reservoirs. X = (x1, x2,··· , xN )T is the vector of displacements and η(t) = (ηL, 0,··· , 0, ηR)T depicts the thermal noise in reservoirs. F (t) is a column vector with elements fi(t) = δiαA0 sin(Ωt), denoting the driving force which acts on the αth particle. Defining the Fourier transform of quantity A and its inverse: A(ω) = 1 2π Z ∞ −∞ eiωtA(t)dt, A(t) = Z ∞ −∞ e−iωt A(ω)dw, and applying them to Eq. (4), we obtain X(t) = X s(t) + X d(t), (5) X s(t) = Z ∞ X d(t) = Z ∞ −∞ −∞ dω e−iωtG(ω)η(ω), dω e−iωtG(ω) F (ω), where G(ω) = (Φ−ω2M−iωΓ)−1 is the phonon Green's function. It is clear that the displacement vector X(t) is a superposition of two contributions: X s, as the ef- fect of the stochastic heat bath, and X d, as the con- sequence of the driving force. For our periodic driving case, it is easy to get X d(t) = −Im[G(Ω) f e−iΩt], with fi = δiαA0. Similarly, the velocity has the same decom- position X = X s + X d, with X d = Im[iΩG(Ω) f e−iΩt]. Straightforwardly, following the definition, the local en- ergy current can be similarly expanded as well: Ii(t) = I s I s i (t) = −kh xs I d i (t) = −k xd i (t) + I d i (xs i (xd i (t) i − xs i − xd i−1)i i−1). (6) i (t). With Ii(t) written in this form, it is obviously that the contributions of noise term and driving force to the en- ergy current are independent from each other. We should note this independence comes from the fact that the driv- ing force is not statistically correlated to the heat baths. Since in the present case, the driving force is determin- istic, we have dropped the notation of ensemble average h···i in I d I s i (t) is just the conventional steady-state heat flux, which is proportional to the temperature difference and is given by I s [16]. Thus, the contribution of stochastic heat bath to the oscillatory steady state energy current is just J s i = Ω/(2π)R 2π/Ω i . In all the following discussions and simulations, we will set ∆T = TL − TR = 0 un- less otherwise stated, since a finite temperature differ- ence mainly causes an additive shift of the energy current curve. R dω ω2G1N (ω)2 i = γ 2(TL−TR) I s i dt = I s π 0 To obtain the expression of J d i , we just need to calcu- late the product of displacement and velocity contributed from the driving force by applying Eq. (5). Denoting R(Ω) ≡ Re[G(Ω)] and I(Ω) ≡ Im[G(Ω)], we have 0Ω[RiαIjα sin2(Ωt) − IiαRjα cos2(Ωt) j = A2 1 1 2RiαRjα sin(2Ωt) − 2IiαIjα sin(2Ωt)]. i xd xd (7) + 3 Since the driving force is periodic such that the steady state energy current is oscillatory, we take the periodic average of the physical quantities, and the final expres- sion of driving-force-contributed energy current is: J d i = = kA2 0Ω 2 kA2 0Ω 2 [Ri−1α(Ω)Iiα(Ω) − Ii−1α(Ω)Riα(Ω)] Im[G∗ i−1α(Ω)Giα(Ω)]. (8) Therefore, when T1 = T2, Ji = J s i . In fact, J d i has two values (i ≦ α and i > α) and is constant on each side of the driven site α. A direct proof is detailed in the Appendix. i = J d i + J d From the expression of Eq. (8), we should be aware that the energy flux possesses the denominator D(Ω) ≡ det[Z(Ω)]2 with Z(Ω) ≡ Φ − Ω2M − iΩΓ. Therefore, when D(Ω) approaches its minimums under certain driv- ing frequencies, the energy flux will exhibit its maximum values. As a consequence, resonance emerges. Let us denote PN (Ω) = det(ΦN − Ω2M) to be the character- istic polynomial of the N × N force matrix ΦN with N particles. It can be shown that, det[Z(Ω)] = PN (Ω) − γ2Ω2PN −2(Ω) − 2iγΩPN −1(Ω), (9) where PN −1(Ω) is the characteristic polynomial of the (N−1)×(N−1) force matrix ΦN −1 with the first row and column or the last row and column taken out from ΦN . PN −2(Ω) is the characteristic polynomial of the (N −2)× (N − 2) force matrix ΦN −2 with both the first and last rows and columns taken out from ΦN . Therefore, the denominator D(Ω) is given by N −1(Ω). D(Ω) = [PN (Ω) − γ2Ω2PN −2(Ω)]2 + 4γ2Ω2P 2 (10) Apparently, resonant frequencies correspond to those val- ues of Ω which make D(Ω) locally minimized. It is dif- ficult to get the explicit exact solutions for those locally minimized solutions. Nevertheless, much simpler and in- spiring results can still be obtained if we consider the limiting case with either small friction coefficient or suf- ficiently large friction. For small friction, we can just set γ = 0 in Eq. (10) and obtain D(Ω) = P 2 N (Ω). Then, the resonant frequencies of the energy current are just the N eigenfrequencies of the force matrix ΦN . For large friction, by keeping only the highest order terms of γ in Eq. (10), we get D(Ω) ∼ P 2 N −2(Ω). In this limiting case, the resonant frequencies correspond to the N − 2 eigen- frequencies of ΦN −2. When the friction is in between the two cases, we have to minimize D(Ω) to get those resonant Ω, which should be a gradual shift between the eigenfrequencies of ΦN −2 and ΦN . Using Eq. (8), we calculate energy current vs driven frequencies, as shown in Fig. 3. It is not surprising to observe the multiple-peak resonance behavior in the harmonic case, since whenever the driven frequency ap- proaches one of the system's eigenfrequencies, resonance 4 1.2 0.9 J 0.6 0.3 0.0 0.0 0.4 0.8 1.2 1.6 2.0 1.3 1.1 J 0.9 0.7 1.10 Harmonic chain FK chain 1.15 1.20 1.25 FIG. 3: (Color online) Steady state energy current versus driving frequency in the harmonic lattice. k = 1, ko = 0, A0 = 3.2, γ = 1, ∆T = TL − TR = 0, N = 32 and α = 1. FIG. 4: (Color online) Comparison of the FK case and har- monic case with force amplitude A0 = 3.2 in a randomly se- lected frequency region. Other parameters are k = 1, ko = 0, V = 5, γ = 1, ∆T = 0, N = 64 and α = 1. will occur. The number of the eigenfrequencies is mainly determined by the system size. We can also define other quantities such as the rate of heat released from the two heat baths and rate of work done by the driving force. Their definitions and analyti- cal expressions are given below: qL ≡ h x1(ηL − γ x1)i = h xs where qL stands for the rate of heat released from the left bath. The first term is contributed by heat baths and is zero when ∆T = 0, while the second term comes from the driving force. Therefore the periodic average at ∆T = 0 gives 1(ηL − γ xs 1)i − γ( xd 1)2, (11) Ω 2π Z 2π/Ω 0 QL = L + Qd dt qL = Qs G1α(Ω)2. (12) Similarly, the expression for QR, the average rate of heat released from the right bath, is given as follows L = − 0Ω2 γA2 2 QR = Qs R + Qd R = − 0Ω2 γA2 2 GN α2. (13) The rate of work done by the driving force is defined as w(t) ≡ hf (t) xαi = f (t) xd α (14) where f (t) = A0 sin(Ωt). Its periodical average gives W = Ω 2π Z 2π/Ω 0 dt w(t) = A2 0Ω 2 Iαα(Ω). (15) W , QL and QR also show multi- Just like the energy flux, resonance behavior [17]. By the continuity equation, the energy current flowing out and into a particle must can- cel each other when the system reaches its steady state, since the local energy density does not vary with time at steady state. Therefore, we immediately obtain the relation W + QL = Ji = − QR (i > α), which obeys the energy conservation. For the harmonic model, a direct proof is given in the Appendix. We should also be aware that this relation also holds in the FK model and this is verified by numerical results. B. Multiple resonances of energy transport in force-driven FK model The main qualitative difference between the multiple- resonance curves in Fig. 2 and 3 lies in the fact that the height of the peak for the FK case is smaller than that for the harmonic case, due to the nonlinear on-site po- tential. This is more obvious for low and high frequency regime. More importantly, we notice that the positions of multiple peaks of the two models seem to be close to each other, as illustrated in Fig. 4. This suggests that the FK model and harmonic model share a similar resonance mechanism. The harmonic model discussed in Sec. II A is able to shed lights on the multi-resonance mechanism in a force-driven nonlinear lattice. It is worth- while to point out that when sampling frequency is not dense enough, one may get a single peak, which actually corresponds to the lower envelope of the multi-resonance curve. When A0 decreases, the system seems to have a single peak even though we increase the density of the sampling points, as shown in Fig. 2. The mechanism for the single peak should still be explained by the fact that the driving frequency is resonant with the system's char- acteristic frequencies. However, the non-linear potential smooths out the peaks of multi-resonances. In this way, we can only observe the lower envelope so that only one peak is observable. Thus the single-resonance curve does not completely reflect the intrinsic transport property of the FK model. Let us examine how the parameters of the system will affect the multi-resonance curve. We have already seen the case of the appearance of multiple peaks when A0 in- creases, as shown in Fig. 2. When A0 becomes large, the kinetic energy of the particle gained from driving force will be much larger than the height of the on-site po- tential. In this way, the mask effect of nonlinear on- site potential becomes negligible and the FK model ef- fectively reduces to a harmonic model without on-site potential. This is also demonstrated in Fig. 5 with 12 9 J 6 3 0 FK Harmonic 0.33 0.35 1.31 1.33 1.93 1.94 FIG. 5: (Color online) Heat current vs driven frequency for large force amplitude A0 = 10. (a) Low frequency regime. (b) Moderate frequency regime. (c) High frequency regime. Other parameters are the same as in Fig. 4. A0 = 10. We arbitrarily choose three typical regimes of driving frequency: small, moderate and large Ω. In all three regimes, the FK model is very close to the har- monic one. Thus, as long as A0 becomes large enough, the mask effect of nonlinearity will fade away and the intrinsic multiple resonant peaks will emerge. We are also interested in the effect of temperature. Considering that T (defined as T ≡ (TL + TR)/2) will take the role of thermal excitation which mainly affects the kinetic energy of particles, we expect a large value of T will have a similar effect with A0 on the resonance behaviors. As shown in Fig. 2 with T = 0.5 and A0 = 1, there is no multiple resonances behavior. While for the increased temperature case T = 2.0, there is clearly a peak in a small Ω range, as shown in Fig. 6(a), which indicates the multiple peaks resonance behavior in the whole frequency range. At low temperature with large amplitude, this is also the case, as illustrated in Fig. 6(b). When T and A0 are both small, compared to the nonlin- ear on-site potential V , particles are confined near their equilibrium positions, within the valley of non-linear on- site potential. By Taylor expansion of the potential to the second order, we see that the non-linear potential reduces to a harmonic one, so that the harmonic ap- proximation works. In this case, the lower bound of the phonon band will be raised by √V . The phonon band in this case is therefore moved from 0 < Ω < √4k to √V < Ω < √V + 4k [18]. (Accordingly, the phonon band of the harmonic model is shifted by the on-site po- tential ko to √ko < Ω < √ko + 4k). Energy transport is forbidden outside the frequency region. It is thus ex- pected to see a shift of the energy flux curve to the right while the multi-resonance are still observed, as shown in Fig. 6(c). The above discussion reveals that there are mainly three dynamic regimes: (1) When kBT + A0a/(2π) ≫ V /(2π)2 (a is the lat- tice constant as defined previously), the FK model will approach a harmonic model without an on-site potential. 5 FK Harmonic 1.27 1.28 1.29 1.30 1.27 1.28 1.29 1.30 (a) 1.26 (b) 1.26 (c) 0.075 0.072 0.069 5.2 4.8 4.4 J 0.0014 0.0012 0.0010 5.90 5.95 6.00 6.05 6.10 FIG. 6: (Color online) Temperature effect on comparison of the FK case and harmonic case. (a) T = 2.0, A0 = 1. (b) T = 0.01, A0 = 6.4. (c) T = 0.01, A0 = 0.1. Other parameters are the same as in Fig. 4, except in (c): k = 10, ko = 15 and V = 15. Multi-resonances are observable. (2) For the opposite case, in which kBT + A0a/(2π) ≪ V /(2π)2, the FK model reduces to a harmonic model with pure harmonic on-site potential of strength ko = V . Multi-resonances are still observable. (3) When T + A0a/(2π) is comparable with V /(2π)2, the multiple peaks will be smoothed out by the effect of nonlinear on-site potentials and only single resonant peak is observable. In the intermediate regimes, there is the crossover from single peak to multiple peaks, of which the resonant magnitudes are smaller compared with those in the harmonic model. In addition, the heat released from the baths and the work done by the force in FK model also show parameter- dependent multiple resonances [17], which can be ex- plained by the same mechanism. III. ABSENCE OF HEAT PUMPING IN A FORCE-DRIVEN LATTICE In all simulations so far, we only consider the energy flux flowing into the right reservoir as a whole. In fact it is a sum of two parts: the rate of heat released from the left reservoir into the system and the rate of work done by f (t), i.e, J = − QR = QL + W . Now let's con- sider a typical result of energy transport in a force-driven FK lattice with TL < TR, where these two contributions are separated, as shown in Fig. 7. Notably, though J QL is is positive (from L to R) in the resonance region, negative in the full range. This indicates heat will always flow into the left reservoir from the chain, whatever the driving frequency is. The corresponding energy flow di- agram is depicted in Fig. 8(a). This actually shows that -QL W J=W+QL 0.3 0.2 0.1 0.0 0.0 0.5 1.0 1.5 2.0 FIG. 7: (Color online) − QL, W , J vs the driving frequency in force-driven FK model. Parameters are the same as those for lower single peak curve with A0 = 1, TL < TR in Fig. 2. QL is negative at all frequencies, showing a negative result for pumping effect. FIG. 8: (Color online) (a) A typical energy flow diagram in force-driven FK model with TL < TR. (b) The situation when a model performs as a heat pump with TL < TR. The arrow denotes the direction of the energy flow. the FK system fails to function as a heat pump since nor- mally, a pump should have an energy flow diagram as in Fig. 8(b), where the heat is absorbed from the low tem- perature bath and released to the high temperature one. We have tested a broad range of parameters and even in the regimes with multiple resonances. In all cases, the energy never flows as in Fig. 8(b). Therefore we conclude that there is no heat pumping action in a force-driven FK lattice. This finding is consistent with Ref. [14], wherein it is found that two coupled oscillators of either harmonic or FPU-like interaction would not act as a heat pump un- der external driving forces. Ref. [15] gets the opposite QL conclusion, because they overlook the direction of and claim the model can act as a heat pump whenever J flows to the right bath of a higher temperature. To understand the underling mechanism for the ab- sence of heat pumping in a force-driven lattice system, we may consider Eq. (11). The first stochastic contribution 6 is always negative since TL < TR. And the second driv- ing contribution, which does not depend on the temper- ature, always contributes a negative value as well. Thus, QL will still be negative, which indicates the low tem- perature bath always absorbs rather than releases heat, even we may change the position of the driving force. The situation just corresponds the energy flow diagram in Fig. 8(a). Clearly as long as the effects of the driving force and the baths are independent, Eq. (11) holds for harmonic chain. Thus for a harmonic system, in order to get a heat pump, a necessary condition is that the driving force is statistically correlated to the bath, such that these two contributions could be synergetic. While in the FK model, effects from the deterministic driving force and stochastic baths are not separable as in Eq. (11) because of the nonlinear on-site potential. However, the non- linearity can not yet make the two contributions syner- getic. And still, our numerical simulations show a nega- tive result for a force-driven FK lattice acting as a heat pump. We thus speculate that for both the force-driven harmonic and nonlinear lattice, the correlation may be the key ingredient for the presence of heat pump ac- tion, similar to the entanglement in quantum thermal baths [19], the off-equilibrium nonthermal reservoirs [20] and the bath noise correlation in temperature-driven case [11], for "low temperature to high temperature" thermal transports. IV. CONCLUSION To summarize, we have studied the energy trans- port control in a force-driven one dimensional nonlin- ear lattice -- the Frenkel-Kontorova model. We have found multiple thermal resonances as a function of the driven frequency. Although not exactly the same, the resonant frequencies are closely related to the eigenfrequencies of the force matrix, and the number of resonant peaks is bounded above by the system size N . Moreover, since the onsite nonlinear potential tends to decrease the mag- nitude of energy current and smoothes out the multiple peaks, the multi-resonance phenomenon is only observ- able in certain parameter ranges and the crossover from multiple resonances to single resonance will occur. Fi- nally, by following a rigorous definition of heat pump, we clarify a previous contradiction and conclude that heat pumping effect is absent in force-driven lattices. In this paper, we focus on a one-dimensional chain. However the analysis and results can be generalized to high dimensional system with arbitrary topology, like polymer networks, proteins [21], and the three dimen- sional random elastic network [22], of which the eigen- spectra are much more complicated and non-trivial. The investigation of the resonances in such systems will give us more flexible methods for mechanical control of energy transport in real applications. V. ACKNOWLEDGEMENT This work has been supported in part by an NUS grant, R-144-000-285-646. This work results from Mr. Zhang Song's "UROPS" (Undergraduate Research Opportuni- ties) project under supervision of Jie Ren and Baowen Li. APPENDIX Intuitively, from the energy conservation point of view, the following relations should hold − QR = Jα+1 = ··· = JN , QL = J1 = ··· = Jα, W = − QL − QR. (16) (17) (18) However, it is highly nontrivial to give direct proofs for arbitrary nonlinear force-driven lattices. In the follow- ing, we will provide a direct proof of the above equations in force-driven harmonic chains by using the analytical forms of W and Ji. QL,R, Let us first define the dimensionless matrix Z(Ω) ≡ Z(Ω)/k, such that Zij = δi,j (c− ibδi,1− ibδi,N )− δi,j+1 − δi,j−1, where b = γΩ/k and c = 2 − mΩ2/k are two dimensionless parameters. Here we set ko = 0 without loss of generality. Considering G = Z−1 = Z−1/k, we need a formula for the inversion of the matrix Z, which is given in its recurrence form [23]: (Z−1)ij = (Z−1)ij = θi−1φj+1 θN θj−1φi+1 θN if i ≤ j, if i > j, (19) (20) where θi and φi are two sequences, defined by the recur- rence relations: θi = Ziiθi−1 − θi−2, φi = Ziiφi+1 − φi+2, (21) (22) with the initial condition θ0 = 1, θ1 = Z11, φN +1 = 1 and φN = ZN N . Note θN is actually the determinant of the matrix Z, which will be denoted as d in follows. Now, to prove Eq. (16), let us recall the expression for QR and Ji (see Eq. (8) and (13)) and use Eq. (19) and (20). We then get (set A0 = 1 for clearness): − QR = Ji = = = γΩ2 2 GN α2 = kΩ Im[G∗ 2 i−1αGiα] γΩ2 2d2k2 θα−12, (23) Ω 2kd2 Im[(θα−1φi)∗(θα−1φi+1)] i φi+1] ΩIm[φ∗ 2kd2 θα−12 (24) 7 where i > α. Hence, to prove the above two equa- tions are equivalent with each other, we need to show Im[φ∗ i φi+1] = γΩ/k for all particles to the right of the driving force. We do induction on the index l. For l = N , the relation is clearly true, which can be taken as the base case. Then suppose the relation is true for all l ≥ i, we need to show it is also true for l = i − 1 case. Use the recurrence equation Eq. (21), we have Im[φ∗ i−1φi] = Im[(cφ∗ i φi+1] = γΩ/k. Thus, we have finished the proof that Eq. (16) holds for all particles to the right of the force. A similar proof can be given to show Eq. (17) holds for all particles to the left of the driving force. i+1)φi] = −Im[φ∗ i+1φi] = Im[φ∗ i −φ∗ Before proceeding to prove Eq. (18), we need more notations. Denote the n × n matrix M(n) with elements M (n) ij = cδi,j − δi,j+1 − δi,j−1 and its determinant Kn = det[M(n)], with boundary K−1 = 0 and K0 = 1. The explicit expression of the determinant is given by Kn−1 = (yn 1 − yn 2 )/(y1 − y2), where y1, y2 is the solution of the quadratic equation y2 − cy + 1 = 0. We observe that θi is in fact the determinant of the sub-matrix in Z starting from the 1st row and column to the ith row and column. Similarly φi is the determinant of the sub-matrix in Z starting from the ith row and column to the N th row and column. Now we may expand θi, φi and the determinant of Z in terms of Kn d = KN − b2KN −2 − 2ibKN −1, θi = Ki − ibKi−1, φi = KN −i+1 − ibKN −i. (25) (26) (27) Now we are ready to prove Eq. (18). Let us assume W , Eq. (18) holds and substitute in the expressions of QL and QR (see Eq. (12), (13) and (15)). Then Eq. (18) becomes Iαα = γΩ(G1α2 + GN α2). With b = γΩ/k and G = Z−1/k, also using Eq. (19)∼ (22), we get Im(θα−1φα+1d∗) = b(θα−12 + φα+12). (28) Express the left hand side (LHS) and right hand side (RHS) of Eq. (28) in terms of Kn (see Eq.(25)∼(27)), and after some algebraic work, it can be shown LHS =b3(Kα−2KN −2KN −α + Kα−1KN −2KN −α−1 − 2Kα−2KN −1KN −α−1) + b(2Kα−1KN −1KN −α − Kα−2KN KN −α − Kα−1KN KN −α−1), (29) RHS =b3(K 2 α−2 + K 2 N −α−1) + b(K 2 α−1 + K 2 N −α), (30) Now in order to show Eq. (18) holds, it is sufficient to prove LHS=RHS to complete the proof. Denote the coef- ficient of b3 (b) in LHS by L3 (L1) and that of RHS by R3 (R1). By using the formula Kn−1 = (yn 2 )/(y1 − y2) and the fact y1y2 = 1, we get − y2α−1 ) − (y2N −2α−1 (y1 − y2)3 [y2α−1 − (y2α−3 − y2α−3 − 4(y1 − y2)] = R3. − y2N −2α−1 − y2N −2α+1 + y2N −2α+1 1 − yn L3 = (31) 1 1 ) 1 2 2 1 2 1 2 Note that by a relabeling of α → α + 1 and N → N + 2 in L3 and R3, and utilizing the relation Kn = cKn−1 − Kn−2, we can show L3 → L1 and R3 → R1. Therefore, LHS=RHS and the proof is completed. 8 [1] L. Wang and B. Li, Phys. World 21, 27 (2008). [2] M. Terraneo, M. Peyrard, and G. Casati, Phys. Rev. Lett. 88, 094302 (2002); B. Li, L. Wang, and G. Casati, Phys. Rev. Lett. 93, 184301 (2004). D. Segal, Phys. Rev. Lett. 101, 260601 (2008). [13] J. Ren, P. Hanggi, and B. Li, Phys. Rev. Lett. 104, 170601 (2010). [14] R. Marathe, A. M. Jayannavar, and A. Dhar, Phys. Rev. [3] B. Li, L. Wang, and G. Casati, Appl. Phys. Lett. 88, E 75, 030103(R) (2007). 143501 (2006). [15] B. Q. Ai, D. He, and B. Hu, Phys. Rev. E 81 031124 [4] L. Wang and B. Li, Phys. Rev. Lett. 99, 177208 (2007). [5] L. Wang and B. Li, Phys. Rev. Lett. 101, 267203 (2008). [6] C. W. Chang, D. Okawa, A. Majumdar, and A. Zettl, Science 314, 1121 (2006). (2010). [16] A. Dhar, Adv. Phys. 57, 457 (2008). [17] Song Zhang's "UROPS" project report, Faculty of Sci- ence, National University of Singapore, 2010. [7] W. Kobayashi et al Appl. Phys. Lett. 95, 171905 (2009); [18] B. Li, J.H. Lan, and L. Wang, Phy. Rev. Lett. 95, 104302 D Sawaki et al, Appl. Phys. Lett. 98, 081915 (2011) (2005). [8] C. W. Chang, D. Okawa, H. Garcia, A. Majumdar, and [19] A. E. Allahverdyan and T. M. Nieuwenhuizen, Phys. Rev. A. Zettl, Phys. Rev. Lett. 99, 045901 (2007). Lett. 85, 1799 (2000). [9] R.-G. Xie, C.-T. Bui, B. Varghese, M.-G. Xia, Q.-X. Zhang, C.-H. Sow, B. Li, and J. T. L. Thong, Adv. Funct. Mat 21, 1602 (2011). [10] N. Li, P. Hanggi, and B. Li, Europhys. Lett. 84, 40009 (2008); N. Li, F. Zhan, P. Hanggi, and B. Li, Phys. Rev. E 80, 011125 (2009). [11] J. Ren and B. Li, Phys. Rev. E 81, 021111 (2010). [12] D. Segal and A. Nitzan, Phys. Rev. E 73, 026109 (2006); [20] S. Das, O. Narayan, and S. Ramaswamy, Phys. Rev. E 66, 050103(R) (2002). [21] J. Ren and B. Li, Phys. Rev. E 79 051922 (2009). [22] H. Hu, A. Strybulevych, J. H. Page, S. E. Skipetrov, and B. A. van Tiggelen, Nature Physics 4, 945 (2008). [23] C. M. da Fonseca, J. Comp. Appl. Math. 200, 283 (2007).
1504.02574	2	1504	2015-06-09T10:26:24	Aharonov-Bohm effect for excitons in a semiconductor quantum ring dressed by circularly polarized light	[ "cond-mat.mes-hall" ]	We show theoretically that the strong coupling of circularly polarized photons to an exciton in ring-like semiconductor nanostructures results in physical nonequivalence of clockwise and counterclockwise exciton rotations in the ring. As a consequence, the stationary energy splitting of exciton states corresponding to these mutually opposite rotations appears. This excitonic Aharonov-Bohm effect depends on the intensity and frequency of the circularly polarized field and can be detected in state-of-the-art optical experiments.	cond-mat.mes-hall	cond-mat	Aharonov-Bohm effect for excitons in a semiconductor quantum ring dressed by circularly polarized light O. V. Kibis,1, 2, ∗ H. Sigurdsson,2 and I. A. Shelykh2, 3, 4 1Department of Applied and Theoretical Physics, Novosibirsk State Technical University, Karl Marx Avenue 20, Novosibirsk 630073, Russia 2Division of Physics and Applied Physics, Nanyang Technological University 637371, Singapore 3Science Institute, University of Iceland, Dunhagi-3, IS-107, Reykjavik, Iceland 4ITMO University, St. Petersburg 197101, Russia We show theoretically that the strong coupling of circularly polarized photons to an exciton in ring-like semiconductor nanostructures results in physical nonequivalence of clockwise and counter- clockwise exciton rotations in the ring. As a consequence, the stationary energy splitting of exciton states corresponding to these mutually opposite rotations appears. This excitonic Aharonov-Bohm effect depends on the intensity and frequency of the circularly polarized field and can be detected in state-of-the-art optical experiments. PACS numbers: 73.21.-b,71.35.-y,42.50.Nn,78.67.-n I. INTRODUCTION Progress in semiconductor nanotechnologies has led to developments in the fabrication of various mesoscopic ob- jects, including quantum rings. The fundamental physi- cal interest attracted by these systems arises from a wide variety of purely quantum-mechanical effects which can be observed in ring-like nanostructures. One of them is the Aharonov-Bohm (AB) effect arisen from the direct influence of the vector potential on the phase of the elec- tron wave function.1,2 In ring-like nanostructures pierced by a magnetic flux, this effect results in the energy split- ting of electron states corresponding to mutually oppo- site directions of electronic rotation in the ring.3 As a consequence, magnetic-flux-dependent oscillations of the conductance of the ring appear.4 -- 9 Since the AB effect takes place for both a single electron and many-particle quantum states,10 it can be observed for elementary ex- citations in semiconductor nanostructures as well. The simplest of them is an exciton which is a bound quantum state of a negative charged electron in the conduction band and a positive charged hole in the valence band. Manifestations of various excitonic effects in semiconduc- tor ring-like structures, including the AB effect induced by a magnetic field, have attracted great attention of both theorists11 -- 22 and experimentalists.23 -- 27 Fundamentally, the AB effect is caused by the bro- ken time-reversal symmetry in an electron system sub- jected to a magnetic flux. Namely, the flux breaks the equivalence of clockwise and counterclockwise elec- tron rotation inside a ring-like structure, which results in the flux-controlled interference of the electron waves corresponding to these rotations. The similar broken equivalence of electron motion for mutually opposite di- rections caused by a magnetic field can take place in various nanostructures, including quantum wells,28 car- bon nanotubes29 and hybrid semiconductor/ferromagnet nanostructures.30 However, the time-reversal symmetry can be broken not only by a magnetic flux but also by a circularly polarized electromagnetic field. Indeed, the field breaks the symmetry since the time reversal turns clockwise polarized photons into counterclockwise polar- ized ones and vice versa. In quantum rings, the strong electron coupling to circularly polarized photons results in the magnetic-flux-like splitting of electron energy lev- els corresponding to mutually opposite electronic rota- tion in the ring31 and oscillations of the ring conduc- tance as a function of the intensity and frequency of the irradiation.32 This phenomenon can be described in terms of a stationary artificial U (1) gauge field generated by the strong coupling between an electron and circu- larly polarized photons.32 Therefore, various stationary phenomena similar to the AB effect can take place in ring-like electronic systems interacting with a circularly polarized electromagnetic field. As a consequence, the new class of quantum optical phenomena in semiconduc- tor nanostructures appears. Although a theory of these AB-like phenomena in quantum rings has been elabo- rated for a single electron,31,32 the optically-induced AB effect for excitons still awaits detailed analysis. The given article is aimed to fill this gap in the theory, which lies at the border between quantum optics and physics of semi- conductor nanostructures. The paper is organized as follows. In Section II, the exciton-photon Hamiltonian is analyzed and solutions of the exciton-photon Schrodinger problem are found. In Section III, the energy spectrum of the dressed excitons is discussed and experimental sets to detect the effect are proposed. In Section IV, conclusion is presented. II. MODEL An electron-hole pair in an one-dimensional quantum ring (see Fig. 1) can be described by the Hamiltonian H0 = − 2 2mhR2 ∂2 h − ∂ϕ2 2 2meR2 ∂2 ∂ϕ2 e + V (ϕe − ϕh), (1) where R is the radius of the ring, me,h is the effective masse of an electron (hole) in the ring, V (ϕe − ϕh) is the potential energy of hole-electron interaction, and ϕe,h is the azimuthal angle of the electron (hole) in the ring. Introducing the new variables, ϕ = meϕe + mhϕh me + mh , θ = ϕe − ϕh, the Hamiltonian (1) can be rewritten as H0 = − 2 2M R2 ∂2 ∂ϕ2 − 2 2µR2 ∂2 ∂θ2 + V (θ), (2) where M = me + mh is the exciton mass, and µ = memh/M is the reduced exciton mass. The eigenfunc- tions of the stationary Schrodinger equation with the Hamiltonian (2) have the form ψnm(ϕ, θ) = χn(θ) eimϕ √2π , (3) where the function χ(θ) meets the Schrodinger equation 2 2µR2 − ∂2χn(θ) ∂θ2 + V (θ)χn(θ) = εnχn(θ), (4) m = 0,±1,±2, ... is the exciton angular momentum along the ring axis, n = 0, 1, 2, ... is the principal quantum num- ber of the exciton, and εn is the exciton binding energy. Correspondingly, the full energy of exciton reads as εn,m = εn + 2m2 2M R2 , (5) where the second term is the kinetic energy of rotational motion of exciton in the ring. Let the ring be subjected to a circularly polarized elec- tromagnetic wave with the frequency ω, which propa- gates along the ring axis (see Fig. 1). Then the full Hamiltonian of the exciton-photon system, including both the field energy, ωa†a, and the exciton Hamilto- nian, H0, is H = ωa†a + H0 + U , (6) where a and a† are the operators of photon annihilation and creation, respectively, written in the Schrodinger rep- resentation (the representation of occupation numbers), and U is the operator of exciton-photon interaction. Gen- eralizing the operator of electron-photon interaction in a quantum ring31 for the considered case of electron-hole pair, we can write this operator as iqeR 2 r ω U = ǫ0V0(cid:2)(e−iϕe − e−iϕh )a† + (eiϕh − eiϕe)a(cid:3) , (7) where qe is the electron charge, V0 is the quantization volume, and ǫ0 is the vacuum permittivity. To de- scribe the exciton-photon system, let us use the notation n, m, Ni which indicates that the electromagnetic field 2 Quantum ring FIG. 1: (Color online) Sketch of an exciton-field system in a quantum ring under consideration. The exciton coupling to the circularly polarized electromagnetic field results in phys- ical nonequivalence of exciton states corresponding to clock- wise and counterclockwise rotations of the exciton as a whole along the ring (shown by the arrows). These exciton states are described by mutually opposite angular momenta m and −m along the ring axis. is in a quantum state with the photon occupation num- ber N = 1, 2, 3, ... , and the exciton is in a quantum state with the wave function (3). The electron-photon states n, m, Ni are true eigenstates of the Hamiltonian H(0) R = ωa†a + H0, which describes the non-interacting exciton-photon sys- tem. Correspondingly, their energy spectrum is ε(0) n,m,N = N ω + εn,m. In order to find the energy spectrum of the full electron- photon Hamiltonian (6), let us use the conventional per- turbation theory, considering the term (7) as a pertur- bation with the matrix elements hn′, m′, N ′ Un, m, Ni. Taking into account in Eq. (7) that ϕe = ϕ + mhθ/M and ϕh = ϕ − meθ/M , these matrix elements read as hn′, m′, N ′ Un, m, Ni = eRr ω ×hIn′n√N + 1δm,m′+1δN,N ′−1 √N δm,m′−1δN,N ′+1i , −I ∗ ǫ0V0 n′n (8) where In′n =Z π −π n′ (θ)χn(θ)e−i(mh−me)θ/2M sin(θ/2)dθ. χ∗ Performing trivial calculations within the second order of the perturbation theory, we can derive eigenenergies of the exciton-photon Hamiltonian (6), εn,m,N = ε(0) n,m,N + Xn′ "hn′, m + 1, N − 1 Un, m, Ni2 # . + hn′, m − 1, N + 1 Un, m, Ni2 ε(0) n,m,N − ε(0) n′,m+1,N −1 n′,m−1,N +1 n,m,N − ε(0) ε(0) 3 where εR = 2/2M R2 is the characteristic energy of exciton rotation. In order to calculate the integral in Eq. (11), we have to solve the Schrodinger equation (4) and find the wave function χ0(θ). Approximating the electron-hole interaction potential V (θ) in Eq. (4) by the delta-function12 and assuming the characteristic exciton size, a = /√8µε0, to be much less than the ring length 2πR, we can write the splitting (11) in the final form (9) Since Eq. (9) is derived within the second order of the perturbation theory, it describes the problem correctly if the energy differences in denominators of all terms lie far from zero. In what follows, we have to keep in mind that all parameters of the problem must lie far from these resonant points. The energy spectrum of exciton-photon system (9) can the first term is the field energy. Following the con- ventional terminology of quantum optics,33,34 the sec- be written formally as εn,m,N = N ω +eεn,m,N , where ond term, eεn,m,N , is the energy spectrum of the exciton dressed by the circularly polarized field (dressing field). Restricting our analysis by the most interesting case of classically strong dressing field (N ≫ 1), we arrive from Eq. (9) to the sought energy spectrum of dressed exciton, eεn,m = Xn′ (cid:20) + (qeE0R)2Inn′2 εn,m − εn′,m+1 + ω (qeE0R)2Inn′2 εn,m − εn′,m−1 − ω(cid:21) , (10) where E0 = pN ω/ǫ0V0 is the classical amplitude of electric field of the electromagnetic wave. It is appar- ent that dressed exciton states with mutually opposite angular momenta, m and −m, have different energies (10). Physically, this should be treated as a field-induced nonequivalence of clockwise and counterclockwise exciton rotations in the ring. As a consequence, the excitonic Aharonov-Bohm effect induced by the circularly polar- ized field appears. In order to simplify the calculation of the field-induced splitting, ∆eεn,m =eεn,m −eεn,−m, we will restrict our consideration to the case of the ground exciton state with n = 0. Let us assume that the char- acteristic binding energy of exciton, q2 e /4πǫ0R2, is much more than both the characteristic energy of rotational ex- citon motion, 2m/2M R2, and the photon energy ω. Then we can neglect the field-induced mixing of exciton states with n′ 6= 0 in Eq. (10). As a result, we arrive from Eq. (10) to the field-induced splitting of exciton states with mutually opposite angular momenta, ∆eε0,m =(cid:12)(cid:12)(cid:12)(cid:12)Z π ×(cid:20) ε2 −π χ0(θ)2 sin(cid:18) mh − me 2M 2 2(cid:19) dθ(cid:12)(cid:12)(cid:12)(cid:12) θ(cid:19) sin(cid:18) θ R(1 + 2m)2 − (ω)2(cid:21) , 2ω(qeE0R)2 ε2 2ω(qeE0R)2 R(1 − 2m)2 − (ω)2 − (11) ∆eε0,m = R(1 − 2m)2 − (ω)2 − 1 ε2 ω (eE0a)2 2 (cid:18) mh − me M (cid:19)2 R(1 + 2m)2 − (ω)2(cid:21) .(12) ε2 1 ×(cid:20) It should be stressed that the simplest delta-function model12 leads to reasonable results. This follows for- mally from the fact that the final expression (12) does not depend on model parameters: It depends only on the exciton binding energy ε0 which should be treated as a phenomenological parameter. We checked that numer- ical calculation using the Coulomb potential gives very similar results to those obtained analytically for the case of the delta potential if the binding energy ε0 is kept the same. Let us estimate the main limitation of the model one- dimensional Hamiltonian (1) which neglects the exciton motion in the radial direction. It can be important since the radial motion weakens the AB effect in wide rings.21 Let a ring with the radius R has the width ∆R. It fol- lows from the numerical calculations that amplitudes of the AB oscillations for the case of R/∆R > 5 and for the case of ideal one-dimensional ring (∆R → 0) are almost identical.21 Therefore, the one-dimensional Hamiltonian (1) correctly describes the solved AB problem for typical semiconductor rings with radius R in the tens of nanome- ters and width ∆R in the nanometer range. III. RESULTS AND DISCUSSION The field-induced splitting (11) -- (12) vanishes if the electron mass is equal to the hole mass, me = mh. Phys- ically, this can be explained in terms of an artificial U (1) gauge field produced by the coupling of a charged parti- cle to circularly polarized photons.32 Since the artificial field32 depends on a particle mass, it interacts differently with an electron and a hole in the case of me 6= mh. As a consequence, the splitting (11) -- (12) is nonzero in the case of me 6= mh, though an exciton is electrically neutral as a whole. In the case of me = mh, the artifi- cial gauge field interacts equally with both electron and hole. However, signs of the interaction are different for the electron and the hole since electrical charges of elec- tron and hole are opposite. Therefore, the interaction of the artificial gauge field with an exciton is zero in the case of me = mh. The splitting (12) for exciton states with the angular momenta m = 1 and m = −1 in a GaAs quantum ring is 20 15 10 5 ) V e µ ( ε ∆ , g n i t t i l p s y g r e n E I0 = 50 W·cm−2 I0 = 100 W·cm−2 I0 = 150 W·cm−2 5 10 15 Exciton binding energy, ε0 (meV) FIG. 2: (Color online) The energy splitting of the exciton states with angular momenta m = 1 and m = −1 in a GaAs ring with the radius R = 9.6 nm as a function of the exciton binding energy ε0 for a circularly polarized dressing field with the frequency ω = 1050 GHz and different intensities I0. FIG. 3: (Color online) The energy splitting of exciton states with angular momenta m = 1 and m = −1 in a GaAs ring with the radius R = 9.6 nm as a function of the field intensity I0 and the field frequency ω for different binding energies of the exciton: (a) ε0 = 2 meV; (b) ε0 = 4 meV; (c) ε0 = 6 meV; (d) ε0 = 8 meV. The physically relevant areas of the field parameters, which correspond to applicability of the basic expressions derived within the perturbation theory, lie below of the dashed lines. 4 presented graphically in Figs. (2) -- (3) for various intensi- ties of the dressing field, I0 = ǫ0E2 0 c. The used effective masses of electron and holes in GaAs, me/m0 = 0.063 and mh/m0 = 0.51, are taken from Ref. 35, where m0 is the mass of electron in vacuum. In Fig. (2), the split- ting ∆ε = eε0,1 −eε0,−1 is plotted as a function of the exciton binding energy, ε0, which depends on the con- finement potential of a quantum ring.13 It is apparent that the splitting decreases with increasing the binding energy. Physically, this is a consequence of decreasing the exciton size, a. Indeed, an exciton with a very small size looks like an electrically neutral particle from viewpoint of the dressing electromagnetic field. As a consequence, the splitting (12) is small for small excitons. It follows from Figs. (2) -- (3) that the typical splitting is of µeV scale for stationary irradiation intensities of tens W/cm2. This splitting is comparable to the Lamb shift in atoms and can be detected experimentally by optical methods. It order to increase the splitting, the irradia- tion intensity I0 should also be increased. However, the increasing of stationary irradiation can fluidize a semi- conductor ring. To avoid the fluidizing, it is reasonable to use narrow pulses of a strong dressing field which splits exciton states and narrow pulses of a weak probing field which detects the splitting. This well-known pump-and- probe methodology is elaborated long ago and commonly used to observe quantum optics effects -- particularly, modifications of energy spectrum of dressed electrons arisen from the optical Stark effect -- in semiconductor structures (see, e.g., Refs. 36 -- 38). Within this approach, giant dressing fields (up to GW/cm2) can be applied to semiconductor structures. As a consequence, the split- ting (12) can be of meV scale in state-of-the-art optical experiments. It should be stressed that the discussed effect is qual- itatively different to those arisen from absorption of circularly polarized light in quantum rings (see, e.g., Refs. 39 -- 41). Namely, the absorption of photons with non-zero angular momentum by electrons leads to the transfer of angular momentum from light to electrons in a ring. Correspondingly, photoinduced currents in the ring appear.39 -- 41 Since this effect is caused by light ab- sorption, it can be described within the classical elec- trodynamics of ring-shaped conductors. In contrast, we consider the Aharonov-Bohm effect induced by light in the regime of electromagnetic dressing, when absorption of real photons is absent. To be more specific, the dis- cussed AB effect arises from light-induced changing phase of electron wave function, which results in the appearance of the artificial gauge field32 and shifts exciton energy levels in the ring. Evidently, this purely quantum phe- nomenon cannot be described within classical physics. IV. CONCLUSION Summarizing the aforesaid, we predict a new quantum- optical phenomenon in semiconductor ring-like nanos- tructures. Namely, a high-frequency circularly polarized electromagnetic wave splits the energy levels of excitons in a semiconductor quantum ring. This effect should be treated as an optically-induced Aharonov-Bohm effect for excitons and can be observed in quantum rings with us- ing modern experimental technics. It should be noted that, besides semiconductor quantum rings, perspective objects for observing the discussed effect are such ring- like semiconductor structures as carbon nanotubes. Acknowledgments 5 The work was partially supported by FP7 IRSES projects POLATER and QOCaN, FP7 ITN project NOTEDEV, Rannis project BOFEHYSS, RFBR project 14-02-00033, and the Russian Ministry of Education and Science. ∗ Electronic address: [email protected] 1 Y. Aharonov and D. Bohm, Phys. Rev. 115, 485 (1959). 2 R. G. Chambers, Phys. Rev. Lett. 5, 3 (1960). 3 M. Buttiker, Y. Impry, and R. Landauer, Phys. Lett. A 96, 365 (1983). 4 Y. Gefen, Y. Imry and M. Ya. Azbel, Phys. Rev. Lett. 52, 129 (1984). 5 M. Buttiker, Y. Imry and M. Ya. Azbel, Phys. Rev. A 30, 1982 (1984). 6 R. A. Webb, S. Washburn, C. P. Umbach, and R. B. Lai- bowitz, Phys. Rev. Lett. 54, 2696 (1985). 7 G. Timp, A. M. Chang, J. E. Cunningham, T. Y. Chang, P. Mankiewich, R. Behringer, and R. E. Howard, Phys. Rev. Lett. 58, 2814 (1987). 8 B. J. van Wees, L. P. Kouwenhoven, C. J. P. M. Harmans, J. G. Williamson, C. E. Timmering, M. E. I. Broekaart, C. T. Foxon, and J. J. Harris, Phys. Rev. Lett. 62, 2523 (1989). 9 I. A. Shelykh, N. T. Bagraev, N. G. Galkin and L. E. Klyachkin, Phys. Rev. B 71, 113311 (2005). (2000). 24 M. Bayer, M. Korkusinski, P. Hawrylak, T. Gutbrod, M. Michel, and A. Forchel, Phys. Rev. Lett. 90, 186801 (2003). 25 E. Ribeiro, A. O. Govorov, W. Carvalho, and G. Medeiros- Ribeiro, Phys. Rev. Lett. 92, 126402 (2004). 26 D. Haft, C. Shulhauser, A. O. Govorov, R. J. Warburton, K. Karrai, J. M. Garcia, W. Schoenfeld, and P. M. Petroff, Physica E 13, 165 (2002). 27 F. Ding, N. Akopian, B. Li, U. Perinetti, A. Govorov, F. M. Peeters, C. C. Bof Bufon, C. Deneke, Y. H. Chen, A. Rastelli, O. G. Schmidt, and V. Zwiller, Phys. Rev. B 82, 075309 (2010). 28 O. V. Kibis, JETP Lett. 66, 588 (1997). 29 D. A. Romanov, O. V. Kibis, Phys. Lett. A 178, 335 (1993). 30 O. V. Kibis, Physica E (Amsterdam) 13, 699 (2002). 31 O. V. Kibis, Phys. Rev. Lett. 107, 106802 (2011). 32 H. Sigurdsson, O. V. Kibis, I. A. Shelykh, Phys. Rev. B. 90, 235413 (2014). 33 M. O. Scully and M. S. Zubairy, Quantum Optics (Univer- 10 B. Sutherland and B. S. Shastry, Phys. Rev. Lett. 65, 1833 sity Press, Cambridge, 2001). (1990). 11 A. V. Chaplik, JETP Lett. 62, 900 (1995). 12 R. A. Romer and M. E. Raikh, Phys. Rev. B 62, 7045 (2000). 13 J. Song ans S. E. Ulloa, Phys. Rev. B 63, 125302 (2001). 14 Hui Hu, Jia-Lin Zhu Dai-Jun Li and Jia-Jiong Xiong, Phys. Rev. B 63, 195307 (2001). 34 C. Cohen-Tannoudji, J. Dupont-Roc, G. Grynberg, Atom- Photon Interactions: Basic Processes and Applications (Wiley, Weinheim, 2004). 35 M. Levinshtein, S. Rumyantsev, and M. Shur, Handbook Series on Semiconductor Parameters (World Scientific, Sin- gapore, 1999), Chap. 3. 36 M. Joffre, D. Hulin, A. Migus, A. Antonetti, J. Mod. Op- 15 A. O. Govorov, S. E. Ulloa, K. Karrai, and R. J. Warbur- tics 35 1951 (1988). ton, Phys. Rev. B 66, 081309(R) (2002). 16 L. G. G. V. Dias da Silva, S. E. Ulloa, and A. O. Govorov, Phys. Rev. B 70, 155318 (2004). 17 K. Moulopoulos and M. Constantinou, Phys. Rev. B 70, 235327 (2004). 18 V. M. Kovalev, A. V. Chaplik, EPL 77, 47003 (2007). 19 M. Willatzen and B. Lassen, J. Phys. C 21, 205302 (2009). 20 M. Bagheri, J. Appl. Phys. 107, 114305 (2010). 21 C. Gonz´alez-Santander, F. Dom´ınguez-Adame and R. A. Romer, Phys. Rev. B 84, 235103 (2011). 22 Bin Li and F. M. Peeters, Phys. Rev. B 83, 115448 (2011). 23 A. Lorke, R. J. Luyken, A. O. Govorov, J. P. Kotthaus, J. M. Garcia, and P. M. Petroff, Phys. Rev. Lett. 84, 2223 37 M. Joffre, D. Hulin, A. Migus, A. Antonetti, C. Benoit a la Guillaume, N. Peyghambarian, M. Lindberg, S. W. Koch, Optics Lett. 13, 276 (1988). 38 S. G. Lee, P. A. Harten, J. P. Sokoloff, R. Jin, B. Fluegel, K. E. Meissner, C. L. Chuang, R. Binder, S. W. Koch, G. Khitrova, H. M. Gibbs, N. Peyghambarian, J. N. Polky, G. A. Pubanz, Phys. Rev. B 43 1719 (1991). 39 A. Matos-Abiague and J. Berakdar, Phys. Rev. Lett. 94, 166801 (2005). 40 Y. V. Pershin and C. Piermarocchi, Phys. Rev. B 72, 245331 (2005). 41 K. Nobusada, K. Yabana, Phys. Rev. A 75, 032518 (2007).
1204.2641	1	1204	2012-04-12T07:37:04	Local Density of States and Interface Effects in Semimetallic ErAs Nanoparticles Embedded in GaAs	[ "cond-mat.mes-hall", "cond-mat.mtrl-sci" ]	The atomic and electronic structures of ErAs nanoparticles embedded within a GaAs matrix are examined via cross-sectional scanning tunneling microscopy and spectroscopy (XSTM/XSTS). The local density of states (LDOS) exhibits a finite minimum at the Fermi level demonstrating that the nanoparticles remain semimetallic despite the predictions of previous models of quantum confinement in ErAs. We also use XSTS to measure changes in the LDOS across the ErAs/GaAs interface and propose that the interface atomic structure results in electronic states that prevent the opening of a band gap.	cond-mat.mes-hall	cond-mat	Local Density of States and Interface Effects in Semimetallic ErAs Nanoparticles Embedded in GaAs Jason K. Kawasaki,1 Rainer Timm,2 Kris T. Delaney,3 Edvin Lundgren,2 Anders Mikkelsen,2 and Chris J. Palmstrøm1,4 1Materials Department, University of California, Santa Barbara, CA 93106 2Nanometer Structure Consortium (nmC@LU), Department of Physics, Lund University, Sweden 3Materials Research Laboratory, University of California, Santa Barbara, California 93106, USA 4Department of Electrical and Computer Engineering, University of California, Santa Barbara, California 93106, USA ABSTRACT The atomic and electronic structures of ErAs nanoparticles embedded within a GaAs matrix are examined via cross-sectional scanning tunneling microscopy and spectroscopy (XSTM/XSTS). The local density of states (LDOS) exhibits a finite minimum at the Fermi level demonstrating that the nanoparticles remain semimetallic despite the predictions of previous models of quantum confinement in ErAs. We also use XSTS to measure changes in the LDOS across the ErAs/GaAs interface and propose that the interface atomic structure results in electronic states that prevent the opening of a band gap. 1 The electronic properties of low dimensional semimetals and semiconductors are of great importance for a wide range of topics, from applications in nanostructured thermoelectric materials [1] to fundamental studies of topological insulators [2]. In these materials, quantum size effects are expected to produce significant changes from the bulk electronic band structure. For example, bulk HgTe is a semimetal with a band overlap of 150 meV [3]. But when its dimensions are confined to near the 2D limit, HgTe quantum become topological insulators, characterized by a topological Z2 invariant [2,4]. When confined even further, extremely thin HgTe 2D quantum wells [4] and 0D nanoparticles [3] undergo a quantum confinement-induced semimetal-to-semiconductor transition. ErAs is another technologically important semimetal, as it has been shown to grow epitaxially on III-As semiconductors with the As-sublattice remaining continuous across the interface [5,6]. Bulk ErAs has rocksalt crystal structure and is a semimetal with valence band maximum at Γ and conduction band minimum at X. However given its relatively large Γ -X band overlap of Δ = 700 meV [7] (compare to 150 meV for HgTe [3,4]), the role of quantum confinement in determining its electronic band structure is much less certain. Indeed, for ultrathin ErAs films embedded in GaAs, simple effective mass models predicted that quantum confinement would open a band gap for films of thickness 1.73 nm (6 monolayers, ML) or less [8]. However, magnetotransport resolved photoemission spectroscopy [8] and angle measurements [9] have shown that such films remain semimetallic for thicknesses as low as 0.86 nm (3 ML). For ErAs nanoparticles the confinement effects are expected to be even stronger, resulting from the reduced dimensionality of the nearly 0D nanoparticles. ErAs nanoparticles embedded within GaAs exhibit optical absorption peaks in the near infrared region [10], and one interpretation is that the absorption results from transitions across a confinement-induced band gap [7]. Based on this interpretation, Scarpulla et al. proposed a simple hard-walled finite- potential model that predicted a gap opening for embedded ErAs nanoparticles with diameters of approximately 3 nm [7]. However, given the failures of effective mass model for ErAs thin films [8,9], this hard-walled finite potential model has remained controversial. An alternative explanation is that the nanoparticles remain semimetallic, with the absorption resulting from excitation of surface plasmon resonances [10]. A direct measurement of the electronic structure of embedded ErAs nanoparticles is still needed in order to determine the validity of the models. In this Letter, we report the first direct measurements of the electronic structure of ErAs nanoparticles embedded within a semiconducting GaAs matrix. We employ cross-sectional scanning tunneling microscopy (XSTM) and spectroscopy (XSTS). The embedded ErAs nanoparticle samples were grown by molecular beam epitaxy using (001) n-type GaAs substrates. The sample structure consists of four layers of varying coverage of ErAs (0.125, 0.25, 0.5, and 1.0 ML) separated by 125 nm n-type GaAs spacers. All layers were grown at a substrate temperature of 540C with a constant Si doping of roughly 5 x 1018 cm-3. Further growth details are described elsewhere [11]. After growth, the samples were cleaved in ultrahigh vacuum to expose a clean {110} surface [11] and analyzed at room temperature in an Omicron variable temperature scanning tunneling microscope. XSTS was performed by interrupting the feedback and simultaneously measuring the tunneling current (I) and the differential conductance (dI/dV) as a function of voltage (V) at specified points on the {110} surface. The conductance was measured using a lock-in amplifier 2 with a 30 mV, 1.3 kHz modulation on the tip-sample bias. In order to amplify the conductance signal and gain a greater dynamic range, spectroscopy measurements were performed in variable gap mode [11,12]. To remove the tip-sample distance dependence, dI/dV was normalized by the absolute conductance I/V, which we have broadened by convolution with an exponential function in order to avoid divergence at the band gap [12]. After normalization, the quantity is proportional to the local density of states (LDOS), where the sample voltage corresponds to energy, in eV, referenced to the Fermi level [12]. Figures 1(a)-1(c) show representative filled states XSTM images of the ErAs nanoparticles in the low coverage limit of less than 0.5 ML. The vertical lines are As atomic rows on the GaAs {110} surface. Since the {110} is not the rocksalt ErAs cleavage plane, the particles tend not to cleave [11]. Instead, the particles remain stuck in one of the cleavage surfaces and are pulled out of the other. This results in protruding particles [Fig. 1(a)] or holes due to missing particles [Fig. 1(b)] in the cross-sectional STM images. The corresponding height profiles are shown in Fig. 1(d). A histogram of particle lengths for the protruding and pulled-out particles is shown in Fig. 1(e). The particles appear nearly spherical, with average lengths of roughly 2.4 and 2.3 nm along the and [001] directions, respectively. The 2.4 nm length along is consistent with Kadow et al. [13], who measure a 2 nm diameter in the (001) plane for particles grown at a similar temperature. Thus the particles are clearly within the sub-3nm regime where hard-walled potential models predict a band gap [7]. A buried ErAs nanoparticle is shown in Fig. 1(c), with the corresponding height profile in Fig. 1(d). Here we see a smooth profile 0.07 nm in height overlaid on the atomic corrugation. This profile is Gaussian in shape with a standard deviation of σ = 4.1 nm and full width at half maximum of 4.8 nm. The apparent height further reduces from 0.07 nm to 0.05 nm when the sample bias is changed from -1.8 to -2.0 V. The small apparent height (less than one atomic step) and strong bias dependence suggest that this profile results from an electronic rather than a topographical feature. It is interpreted to be a buried ErAs particle whose electronic states induce electronic changes in the surrounding GaAs matrix, such as band bending or introduction of localized states into the GaAs band gap. XSTS measurements were performed in order to further explore the electronic structure of the embedded ErAs nanoparticles. Figure 2(a) shows normalized dI/dV spectra for the GaAs matrix and protruding ErAs nanoparticles. Both curves are averaged over at least 20 individual spectra. In the GaAs spectra a clear band gap extending from -1 to 0.8 V is observed. Because of tip-induced band bending the measured band gap of 1.8 eV is larger than the true band gap of 1.4 eV, consistent with previous STS studies [14,15]. Additionally, despite the heavy n-type doping (5 x 1018 cm-3 Si) the GaAs Fermi level is pinned near midgap, which is often observed for metal-GaAs interfaces [16,17] and for cleaved surfaces due to atomic steps [18]. The ErAs nanoparticle dI/dV shows no evidence of a band gap. Instead, dI/dV (LDOS) exhibits a sharp but finite minimum at the Fermi level, indicating that the nanoparticles are semimetallic. This curve is qualitatively similar to density functional theory (DFT) calculations for the bulk ErAs density of states [19]. Additionally, spectra measured directly over buried particles [Fig. 1(c)] are nearly identical to spectra measured over protruding particles [Fig. 1(a)]. Thus the observed semimetallic behavior is not induced by cleavage defects or the vacuum interface, but is instead a feature of the particles themselves. These measurements suggest that the observed near-IR optical absorption is probably not due to optically driven electron-hole excitations, but instead results from the excitation of surface plasmons. This lies in direct 3 contrast with the simple hard-walled potential model, which predicts that 2.3 nm spherical particles should have a band gap on the order of 0.5 eV [7]. The local electronic features across the interface between ErAs and GaAs may also influence this behavior. Fig. 2(b) shows a series of individual normalized dI/dV spectra starting at a point directly on top of a nanoparticle and moving in steps of 1.3 nm along the direction into the GaAs matrix. Directly on top of the ErAs particle (0 nm) and near the particle edge (1.3 nm) the spectra retain the finite minimum at the Fermi level, consistent with semimetallic behavior. In both curves there is clear evidence of an extra state, not derived from bulk GaAs or ErAs, at 0.2 eV, indicated by an arrow. Moving across the ErAs/GaAs interface to a distance of 2.6 nm, which is roughly 1.4 nm into the GaAs matrix, the state at 0.2 eV begins to decay and the minimum at the Fermi level broadens; however there are still states within the GaAs band gap close to the particle. These states continue to decay and the bulk GaAs DOS is recovered near a distance of 3.9 nm from the particle center. This 3.9 nm decay radius is in good agreement with the σ = 4.1 nm radius of electronic contrast for the buried particle observed by XSTM [Fig. 1(c)]. These states within the band gap, and, in particular, the state at 0.2 eV that decays with distance into the GaAs matrix, may result from interface states. Note that the state at 0.2 eV does not appear in DFT calculations for bulk ErAs [19] or in photoemission spectra of continuous ErAs films [20]. But for ErAs/GaAs interfaces, DFT calculations predict the existence of interface states for both (001) [17,21] and (110) planar interfaces [21] at positions within the GaAs band gap. These states arise from differences in bonding and coordination across the ErAs (rocksalt) / GaAs (zincblende) interface, and they peak at the interface and decay into the GaAs matrix, just as observed in our XSTS measurements. Here the decay occurs primarily into the GaAs side because in the case of a semimetal/semiconductor interface , the interface states correspond to extended states from the semimetal ErAs side [21]. These interface states may be responsible for preventing the opening of a band gap. For ErAs thin film superlattices on (001) GaAs, DFT calculations by Said et al. show that ErAs/GaAs interface states persist even with reduced ErAs film thickness, and their positions at and near the Fermi level prevent a gap from opening [22]. Additionally, tight binding calculations for GdAs/GaAs superlattices by Xia et al. [23] identify a heavy hole interface band along the Γ-X dispersion that curves up and turns into a conduction band. This partially filled interface band prevents GdAs/GaAs superlattices from turning into a semiconductor, and Xia et al. argue that the same may be true for ErAs/GaAs planar superlattices. Similar mechanisms may prevent ErAs nanoparticles from opening a band gap; however for the case of embedded nanoparticles, the interfaces are more complicated than the simple (001) and (110) planar interfaces. A potential effect of the observed interface states is to effectively reduce the size of the confining potential over some length scale into the GaAs matrix. Following Scarpulla et al. [7], we begin modeling the confinement using a spherically symmetric step potential whose height is given by the energy differences in the band extrema for GaAs and ErAs (Fig. 3 inset). The potential height for holes is U0,h = ΓVB,ErAs - ΓVB,GaAs= 1.03 eV and for electrons is U0,e = XCB,GaAs - XCB,ErAs = 1.47 eV. Note we used the room temperature band gap for GaAs, whereas Scarpulla et al. used the 0 K band gap. Our effective masses were m* h/m0 = 0.5 (0.235) and m* e,X/m0 = 0.32 (0.25) for GaAs (ErAs) [7]. We next apply two modifications to the finite-step potential model to include (1) the effects of interface states and (2) many-body effects (Fig. 3 insert). In the first modification we model an interface state as an intermediate step in the confinement potential with energy E int and spatial 4 extent dint. From XSTS measurements this state is located at approximately Eint = 0.2 eV above the Fermi level, and from DFT [22] and XSTS we find that the state is highly localized at the interface with width on the order of dint = aGaAs (lattice constant of GaAs, 5.65 Å). The resulting interface step potential has the form Ustep(r) = 0 for r < a, Ustep(r) = U0,int for a < r < a + dint, and Ustep(r) = U0,e/h for r > a + dint, where a is the radius of the spherical ErAs nanoparticle. To model many-body effects at the interface we note that the semimetallic nature of bulk ErAs, and the potential presence of surface plasmons at the ErAs/GaAs interface, motivate a Thomas-Fermi-like screening of the confining potential of the form Uscreen(r)= 1 for r < a and Uscreen(r)=-exp[-keff(r-a)]+1 for r > a, where keff is the effective screening wave number. For an electron density of 5 x 1018 cm-3 the Thomas-Fermi wave number is 3.57 nm-1, and we use this to guide the order of magnitude of the screening wave vector: keff = 1 nm-1. The total confinement potential is given by Utotal,e/h(r) = Ustep,e/h(r)Uscreen(r), where we have adjusted U0,int such that after multiplying by the screening, Utotal,e/h(a+dint)=Eint (Fig. 3 insert). We next solve the Schrödinger equation in spherical coordinates to find the band shifts of occupied electron and hole states subject to this confining potential. The confinement-induced ErAs band gap is given by Eg(a)=Ee(a)+Eh(a)-∆. The results for the modified model with interface states and screening are shown in Fig. 3. We find that compared to the simple hard-walled step-potential model, the presence of features associated to interface states and metallic screening provides a strong modification to the predicted confinement-induced gap opening. With these effects, at 2.3 nm diameter the particles are predicted to remain semimetallic, consistent with our XSTS measurements. Furthermore, when solved for a 2D thin film, the interface and screening model predicts that ErAs films should remain semimetallic down to a critical thickness of 0.15 nm. This 0.15 nm thickness is much less than the 1 ML (0.287 nm) physical limit, indicating that ErAs thin films will in fact never become semiconducting, consistent with previous experimental [8,9] and DFT [22,23] work on ErAs thin films. Our analysis clearly demonstrates that the choice of the form of the confining potential has a strong effect on the predictions of simple one-electron confinement models. Our results also highlight the importance of including physically motivated features of the interface electronic structure in modeling the subtle effects of quantum confinement, especially in systems where differences in bonding and crystal structure across the interface lead to highly localized interface states. However, we caution that the results of such simple models are strongly dependent on the choice of parameters. For example, a choice of keff = 0.5 nm-1 instead of 1 nm-1 with the same values of Eint and dint yields a band gap opening at 1.5 nm diameter instead of roughly 2.2 nm. Thus while these modifications may capture more of the complex interfacial physics, they also motivate future theoretical work of fully atomistic and parameter-free calculations to provide a truly quantitative understanding of the effects of quantum confinement in ErAs/GaAs. In conclusion, we have examined the atomic and electronic structures of ErAs nanoparticles embedded within GaAs (001) via XSTM/XSTS. Tunneling spectroscopy shows that the LDOS of the ErAs particles has a sharp but finite minimum at the Fermi level, demonstrating that the particles are semimetallic. The data strongly suggest that previously observed optical absorption is due to surface plasmon resonances and that the simple hard-walled potential model does not provide an accurate description of quantum confinement for embedded ErAs nanoparticles. Tunneling spectroscopy also shows a state at 0.2 eV above the Fermi level that decays with distance across the ErAs/GaAs interface, and we attribute this to an interface state. We have shown that small changes to the model potential, motivated by the presence of interface states 5 and metallic screening, strongly modify the predictions of the model and provide agreement with measurements, demonstrating the importance of considering the atomistic and electronic structure of the interface itself. We thank A. C. Gossard, C. G. Van de Walle, and J. Buschbeck for helpful discussions. This work was supported by NSF through the IMI Program (No. DMR 0843934) and the UCSB MRL (No. DMR 05-20415), ARO (Award No. W911NF-07-1-0547), AFOSR (No. FA9550-10-1- 0119), Swedish Research Council (VR), Swedish Foundation for Strategic Research (SSF), Crafoord Foundation, Knut and Alice Wallenberg Foundation, and European Research Council under European Union's Seventh Framework Programme (FP7/2007-2013)-ERC Grant agreement No. 259141. J. K. K. acknowledges funding from NDSEG. R. T. acknowledges support from the European Commission under a Marie Curie Intra-European Fellowship. [1] W. Kim, et al., Phys. Rev. Lett. 96, 045901 (2006). [2] M. Z. Hasan and C. L. Kane. Rev. Mod. Phys. 82 3045 (2010). [3] A. Rogach, et al., Adv. Mater. 11, 552 (1999). [4] M. Konig, et al., Science 318 766 (2007). [5] [6] C. J. Palmstrøm, N. Tabatabaie, and S. J. Allen, Appl. Phys. Lett. 53, 2608 (1988). N. G. Stoffel, C. J. Palmstrøm, and B. J. Wilkens. Nucl. Instr. Meth. Phys. Res. B. 56/57 792 (1991). [7] M. A. Scarpulla, et al., Appl. Phys. Lett. 92, 173116 (2008). [8] [9] S. J. Allen, et al., Surf. Sci. 228, 13 (1990). L. Ilver, et al., Phys. Rev. Lett. 77, 4946 (1996). [10] E. R. Brown, et al., Phys. Rev. Lett. 90, 077403 (2003). [11] J. K. Kawasaki, et al., J. Vac. Sci. Tech. B. 29, 03C104 (2011). [12] R. M. Feenstra, Phys. Rev. B. 50, 4561 (1994). [13] C. Kadow, et al., J. Vac. Sci. Tech. B. 18, 2197 (2000). [14] R M Feenstra, et al., Nanotechnology 18, 044015 (2007). [15] R. Timm, et al., Nano Lett. 10, 3972 (2010). [16] R.M. Feenstra and P. Mårtensson, Phys. Rev. Lett. 61, 447 (1988). [17] W. R. L. Lambrecht and A. G. Petukhov, Solid State 108, 361 (1998). [18] P. Chiaradia, M. Fanfoni, and C. Goletti, Phys. Rev. B. 52 10721 (1995). [19] K. T. Delaney, N. A. Spaldin, and C. G. Van de Walle, Phys. Rev. B. 77, 235117 (2008). [20] T. Komesu, et al., Phys. Rev. B. 67 035104 (2003). [21] K. T. Delaney, N. A. Spaldin, and C. G. Van de Walle, Phys. Rev. B 81, 165312 (2010). [22] M. Said, et al., Solid State Comm. 100, 477 (1996). [23] J. B. Xia, S. F. Ren, and Y. C. Chang, Phys. Rev. B. 43, 1692 (1991). 6 Figure 1. (color online) Filled states XSTM images of (a) protruding, (b) pulled-out, and (c) buried ErAs nanoparticles grown on (001) GaAs. (d) Height profiles of the three particle sites. (e) Histogram of the protruding and pulled-out particle lengths along [001] and . 7 Figure 2. (color online) (a) Averaged differential conductance curves for protruding ErAs nanoparticles and the GaAs matrix. (b) Individual differential conductance spectra at varying points directly on top of an ErAs particle (0 nm) and moving in steps of 1.3 nm into the GaAs matrix (3.9 nm). 8 Figure 3. (color online) Calculated energy gap versus ErAs particle diameter. Insert shows schematic of the modified confinement potential model. 9
1506.03967	1	1506	2015-06-12T10:13:47	Spatial-temporal redistribution of point defects in three-layer stressed nanoheterosystems within the framework of self-assembled deformation-diffusion model	[ "cond-mat.mes-hall" ]	The model of spatial-temporal distribution of point defects in a three-layer stressed nanoheterosystem GaAs/In$_x$Ga$_{1 - x}$As/GaAs considering the self-assembled deformation-diffusion interaction is constructed. Within the framework of this model, the profile of spatial-temporal distribution of vacancies (interstitial atoms) in the stressed nanoheterosystem GaAs/In$_x$Ga$_{1 - x}$As/GaAs is calculated. It is shown that in the case of a stationary state ($t > 5\tau _d^{(2)}$), the concentration of vacancies in the inhomogeneously compressed interlayer is smaller relative to the initial average value $N_{d0}^{(2)}$ by 16%	cond-mat.mes-hall	cond-mat	CondensedMatterPhysics,2015,Vol.18,No2,23602:1 -- 12 DOI:10.5488/CMP.18.23602 http://www.icmp.lviv.ua/journal Spatial-temporalredistributionofpointdefectsin three-layerstressednanoheterosystemswithinthe frameworkofself-assembleddeformation-diffusion model R.M.Peleshchak,N.Ya.Kulyk∗,M.V.Doroshenko DrohobychIvanFrankoStatePedagogicalUniversity,24FrankoSt.,82100Drohobych,Ukraine ReceivedNovember20,2014,innalformFebruary10,2015 The model of spatial-temporal distribution of point defects in a three-layer stressed nanoheterosystem GaAs/InxGa1−xAs/GaAsconsideringtheself-assembleddeformation-diffusioninteractionisconstructed.Within the framework of this model, the prole of spatial-temporal distribution of vacancies (interstitial atoms) in the stressed nanoheterosystem GaAs/InxGa1−xAs/GaAs is calculated. It is shown that in the case of a stationary ),theconcentrationofvacanciesintheinhomogeneouslycompressedinterlayerissmallerrela- state(t > 5τ(2) tivetotheinitialaveragevalue N (2) Keywords:spatial-temporaldistribution,vacancies,interstitialatoms PACS:67.80.Mg,68.55.Ln by16%. d d0 1. Introduction Intensive development of nanotechnologies has provided an opportunity to create nanoelectronic devicesonthebasisofstressednanoheterosystemsGaAs/InxGa1−xAs/GaAs(ZnTe/Zn1−xCdxTe/ZnTe).The active region of such structures are layers InxGa1−xAs, Zn1−xCdxTe, in which the electron-hole gas is localizedbeingboundedontwosidesofthepotentialbarriersGaAs(ZnTe). Itisknownthatopticalandelectricpropertiesofsuchdevicesdependsignicantlyonboththelattice deformationofthecontactingsystemsandthespatialdistributionofpointdefects. Such defects can penetrate from the surface or arise in the process of epitaxial growth. Besides, dif- fusionprocessesplayanimportantroleinthetechnologyoffabricatingoptoelectronicdevices.Theyare related with the redistribution of impurities in a semiconductor structure caused by both the ordinary gradientconcentrationofdefectsandthegradientofdeformationtensor. Theinteractionofdefectswiththedeformationeld,createdbyboththemismatchofthecrystallat- tice of the contacting materials and the point defects, causes a spatial redistribution of the latter. It can lead both to accumulation and to a decrease of the number of defects in the active region (InxGa1−xAs, Zn1−xCdxTe) of the operating element depending on the character of the deformation created both by themismatchbetweenparametersofcontactingcrystallattices(ε0 = 7%(4%)forGaAs/InxGa1−xAs/GaAs (ZnTe/Zn1−xCdxTe/ZnTe), respectively [1, 2]), and by the action of defects. In particular, it is known that thegalliumarsenidegrownusingthemethodofthemolecular-beamepitaxyatlowtemperaturecontains an excess of arsenic [3, 4]. Introduction of excessive arsenic causes a tetragonal distortion of the lattice material GaAs and the generation of point defects in it: interstitial atoms (As), vacancies (Ga) and anti- structuraldefects(AsGa),which,inturn,leadstotheirspatialredistribution.Thelatticedeformationand concentration of point defects generated under the action of the gradient of deformation tensor depend ∗E-mail:[email protected] ©R.M.Peleshchak,N.Ya.Kulyk,M.V.Doroshenko,2015 23602-1 5 1 0 2 n u J 2 1 ] l l a h - s e m . t a m - d n o c [ 1 v 7 6 9 3 0 . 6 0 5 1 : v i X r a R.M.Peleshchak,N.Ya.Kulyk,M.V.Doroshenko onthemismatchbetweenthelatticeparametersofcontactinglayersofheterostructure,thetemperature ofgrowth,molecularuxesGaandAs,concentrationandchemicalnatureofthedopedimpurities. The strain caused by the mismatch between the lattice of the epitaxial layer and the substrate can be elastic when the thickness of the layer does not exceed a dened critical value [5]. Otherwise, mis- match dislocations are formed accompanied by a sharp worsening of both the optical and the electric characteristics of devices. However, in the layers InxGa1−xAs with the mismatch less than critical there isasignicantdeclineofthemobilityandtheintensityofphotoluminescenceatcertainterms[5],which isrelatedtotheincreasednumberofpointdefectsandacorrespondingincreaseofthediffusionbarrier totheatomsofthethirdgroup. In experimental work [6], it is shown that in a heterostructure GaAs/InxGa1−xAs, the stressed quantum-size heterolayers hamper the diffusion of hydrogen and defects into the bulk of the material which leads to a substantial difference of their spatial distribution in a heterostructure and homoge- neous layers. Theoretical research of the stationary distribution of defects within the framework of the self-assembleddeformation-diffusionmodelhasbeenconsideredinthework[7]. Therefore, in order to create devices with prescribed physical properties, it is necessary to construct a spatial-temporal deformation-diffusion model that describes the self-assembled deformation-diffusion processesinstressednanoheterostructureshavingtheirownpointdefectsandimpurities. The aim of this work is to construct a spatial-temporal deformation-diffusion model and calculate the spatial-temporal prole distribution of point defects (interstitial atoms and vacancies) in three-layer stressednanoheterosystemsGaAs/InxGa1−xAs/GaAs(ZnTe/Zn1−xCdxTe/ZnTe). 2. The model of spatial-temporal redistribution of defects in a three- layerstressednanoheterosystem Let us consider stressed nanoheterosystems GaAs/InxGa1−xAs/GaAs (ZnTe/Zn1−xCdxTe/ZnTe) having interlayersInAs(CdTe)ofthethickness 2a,thatincludethreelayers(gure1),where N (1) are theinitialaveragedefectconcentrations,respectively,and D1, D2, D3 arediffusioncoecients.Suppose that the external layers GaAs (ZnTe) are of the thickness that considerably exceeds the width of the in- terlayer of the heterostructure (2a/L (cid:191) 1), so deformation of these layers can be neglected [εi (z) = 0, i = 1,3]. Themechanicaldeformationthatoccursduetothemismatchbetweenthelatticeparametersofcon- tactingmaterialsofaheterosystemisapproximatedbythefunction[7]: , N (2) , N (3) d0 d0 d0 (cid:40) (1) i = 2, i = 1,3, where ε0 = εxx + εy y + εzz < 0istherelativechangeoftheelementarycellvolumeofthegrownlayeron heteroboundaries z = a; εy y = εzz = (ai+1 − ai )/ai, εxx = −(2C (2) 11 )εy y, where i = 1,3 corresponds εi (z) = 12 /C (2) ε0 0, z2 a2 , Figure1.ThemodelofstressednanoheterostructuresGaAs/InxGa1−xAs/GaAs(ZnTe/Zn1−xCdxTe/ZnTe). 23602-2 Spatial-temporalredistributionofpointdefects... to the layers GaAs (ZnTe), i = 2 corresponds to the interlayers InxGa1−xAs (Zn1−xCdxTe); ai are the crys- tal lattice parameters of the contacting materials GaAs (ZnTe) and InAs (CdTe) of the heterostructure, respectively; C (2) and C (2) Epitaxial growth on a substrate with the mismatch between the lattice parameters takes place si- multaneously with the diffusion process, which is caused by both the concentration gradient of point defects [gradN (i ) dl (z, t)] and the gradient parameter of deformation [gradUi (z, t)]. The latter induces an additional diffusion ux of defects, which is opposite to the ordinary gradient concentration ux of de- fects.Therefore,inthebasisofthismodelitisnecessarytoputaself-assembledsystemofnon-stationary equationsfortheparameterdeformation Ui (z, t)andconcentrationofimpurities N (i ) d (z, t)inastressed heterosystem,theredistributionofwhichisperformedsimilarlytoanordinarydiffusionux, aretheelasticconstantsofthematerialInxGa1−xAs(Zn1−xCdxTe). 11 12 thus,bythedeformationcomponentoftheux J (i )dif (z, t) = −Di ∂N (i ) d (z, t) ∂z , J (i )def (z, t) = −Di θ(i ) d kBT N (i ) d (z, t) ∂Ui (z, t) , ∂z d 11 = K (i )∆Ω(i ) is the mechanical deformation potential, K (i ) = (C (i ) + 2C (i ) Let the point defects be distributed with the initial average concentration N (i ) where θ(i ) 12 )/3 is the module of uniformcompressionofthe i-thmaterial, ∆Ω(i ) isthevariationofelementarycellvolumeatthepresence ofadefectinthe i-thlayer. in the i-th layer in a particularheterosystem.Asaresultoftheirself-assembledinteractionthroughthedeformationeld,cre- atedbyboththemismatchbetweenlatticeparametersofcontactingmaterialsofaheterosystemandthe presence of defects, there is a variation of the concentration prole of point defects and of the character ofdeformation. Themechanicalstressinepitaxiallayerscreatedbyboththepointdefectsandthemismatchbetween latticeparametersofcontactingmaterialsisdescribedbytheexpression: (2) d (z, t)− ρi c2 where ρi ,ci arethedensityofthe i-thmediumandthelongitudinalspeedofthesound,respectively. Thewaveequationforthedeformationparameter Ui (z, t)isoftheform: i Ui (z, t)− θ(i ) σi (z, t) = ρi c2 d N (i ) εi (z), d0 i Takingintoaccount(2),equation(3)fortherenormalizeddeformation, Ui (z, t)looksasfollows: ρi ∂2Ui ∂t 2 = ∂2σi ∂z2 . (3) (4) − θ(i ) d ρi c2 i (cid:183) ∂2Ui (z, t) = ∂2Ui (z, t) ∂t 2 ∂z2 1 c2 i ∂2N (i ) d (z, t) ∂z2 − ∂2εi (z) ∂z2 . ∂z N (i ) ∂ ∂z − Di = Di d (z, t) ∂Ui (z, t) θ(i ) d kBT ∂N (i ) d (z, t) ∂t ∂2N (i ) d (z, t) ∂z2 Theequationforthedefectconcentration(interstitialatomsandvacancies)isoftheform[7]: (cid:184)+G(i ) (5) where Di is the diffusion coecient of point defects in the i-th layer, G(i ) is the generation rate of the is the lifetime of the defects in the i-th layer that is determined by the frequency and the defects, τ(i ) ∼ 1 µs) that arise in the amplitude of mechanical uctuations in the megahertz range (ω (cid:202) 106 Hz, τ(i ) processoftheformationofheteroboundariesinstressednanoheterostructuresandintheprocessofthe occurrenceofdefects(acousticemission)[8]. As a result, a self-assembled system of equations (4), (5) is received for determination of the spatial- temporaldistributionoftheconcentrationofdefects N (i ) d (z, t)andthedeformationparameter Ui (z, t)in thedifferentregionsofthethree-layernanoheterostructure. 23602-3 − N (i ) d (z, t) τ(i ) d d d d d , R.M.Peleshchak,N.Ya.Kulyk,M.V.Doroshenko Thedefectconcentrationcanbewrittenintheform: d (z, t) = N (i ) N (i ) d0 + N (i ) dl (z, t), (cid:191) N (i ) , the equation of diffusion Taking into account the presentation (6) in the approximation N (i ) dl (z, t)isthespatiallyinhomogeneouscomponentofthedefectconcentration. where N (i ) iswrittenasfollows: (7) ∂N (i ) dl (z, t) ∂t is the generation rate of point defects under the effect of mechani- where G − N (i ) cal uctuations (ω ∼ 106 Hz) that arise in the process of the formation of heteroboundaries in stressed nanoheterostructures. A further solution of self-assembled systems of equations (4), (5) will be searched in the approxima- tion − N (i ) dl (z, t) τ(i ) d ∂2N (i ) dl (z, t) ∂z2 d0(z, t)/τ(i ) − Di N (i ) ∂2Ui (z, t) = G(i ) θ(i ) d kBT = Di +G (cid:48)(i ) d (cid:48)(i ) d ∂z2 d0 d0 dl d d , (6) (cid:179)  (cid:180)2 2 L(i ) d 2Di ∂2Ui (z, t) ∂t 2 (cid:191) ε0, namely ∂t 2 ∂2Ui (z, t) In approximation (8), from equation (4), there will be found ∂2Ui (z,t) (8) ≈ 0, where Ldi isthediffusionlengthofthedefectinthe i-thlayer. and it will be put into the equa- tion(7).Asaresult,differentialequationfordeterminationofthespatial-temporaldistributionofdefects inthestressedheterosystemisreceived 1− N (i ) d0 N (i ) dc (9) is the critical defect concentration, which being exceeded results in the self- In addition, the conditions of the equality of concentration of impurities and their uxes must be where N (i ) organizationofthedefects[9]. satisedontheboundarylayeroftheheterostructureshowningure1: − N (i ) dl (z, t) τ(i ) d (cid:33)(cid:35) ∂2N (i ) ∂N (i ) dl (z, t) ∂t dl (z, t) ∂z2 = kBT ρc2 N (i ) d0 ∆Ω ∂2εi (z) i /θ(i ) d − Di +G N (i ) dc (cid:48)(i ) d (cid:34) ∂z2 (cid:195) Di = ∂z2 dc , where Ji (z, t) = − ∂ ∂z dl (a, t), (cid:110) = 0, Di N (i ) ∂N (1) dl (z, t) ∂z dl (z, t)[1− εi (z)] ∂N (3) dl (z, t) ∂z dl (a, t) = N (3) N (2) J2 (a, t) = J3 (a, t), (cid:175)(cid:175)(cid:175)(cid:175)z→∞ (cid:175)(cid:175)(cid:175)(cid:175)z→−∞ = 0, dl (−a, t), dl (−a, t) = N (2) N (1) (cid:111).Attheprimarymomentoftime J1 (−a, t) = J2 (−a, t), dl (z,0) = 0. N (i ) Enteringthefollowingdimensionlessvariables =(cid:113) dl (z, t) = Yi (z, t)G N (i ) N (i ) d0 τ(i ) d , (cid:48)(i ) = G d θ = t τ(2) d Di τ(i ) d , τ(i ) d , (cid:48)(i ) d L(i ) d (cid:95) , , , L = L a (cid:95)z = z L 23602-4 (10) (11) Spatial-temporalredistributionofpointdefects... equations(9)taketheform: ∂Y1((cid:95)z, θ) ∂Y2((cid:95)z, θ) ∂Y3((cid:95)z, θ) N (2) d0 N (2) dc , ∂θ d L = 2ε0 2 ∂(cid:95)z − λ1 = D1 D2 ∆Ω(2)a2G(2) d β = D2 (cid:195) L(2) (cid:33)2(cid:195) (cid:33) ∂2Y1((cid:95)z, θ) 1− N (1) d0 (cid:33)2(cid:195) (cid:195) L(2) (cid:33) ∂2Y2((cid:95)z, θ) N (1) dc 1− N (2) d0 (cid:195) L(2) (cid:33)2(cid:195) (cid:33) ∂2Y3((cid:95)z, θ) N (2) dc 1− N (3) d0 (cid:180)2,andboundaryconditions(10)canbewritten: (cid:179) (cid:180)2, λ2 = 1, λ3 = D3 N (3) dc (cid:175)(cid:175)(cid:175)(cid:95) (cid:105) (cid:105) Y2((cid:95)z, θ)− 1 (cid:105) (cid:104) Y1((cid:95)z, θ)− 1 (cid:104) (cid:104) Y3((cid:95)z, θ)− 1 + β− λ2 = D3 D2 L(2) d /L(3) ∂Y1((cid:95)z, θ) − λ3 = 0, 2 ∂(cid:95)z ∂(cid:95)z d L d L ∂θ ∂θ D2 d 2 , , , ∂(cid:95)z z →−(cid:95) L where λ1 = D1 D2 (cid:179) L(2) d /L(1) d (12) (13) (14) (15) (16) 23602-5 D1 D2 d L (cid:33)2(cid:195) (cid:195) L(2) −1+ N (1) d0 (cid:33)2(cid:195) (cid:195) L(2) N (1) dc −1+ N (2) d0 N (2) dc d L τ(2) d τ(1) d τ(3) d τ(2) d Y1(−(cid:95)a, θ) = G(2) d G(1) d Y2((cid:95)a, θ) = G(3) d (cid:195) L(2) (cid:33) ∂Y1(−(cid:95)a, θ) G(2) d (cid:33) ∂Y2((cid:95)a, θ) ∂(cid:95)z = d L + β(cid:95)a = D3 D2 ∂(cid:95)z (cid:175)(cid:175)(cid:175)(cid:95) ∂Y3((cid:95)z, θ) ∂(cid:95)z = 0. z →(cid:95) L Y2(−(cid:95)a, θ), Y3((cid:95)a, θ), (cid:33)2(cid:195) −1+ N (2) d0 (cid:33)2(cid:195) (cid:195) L(2) N (2) dc −1+ N (3) d0 N (3) dc (cid:33) ∂Y2(−(cid:95)a, θ) (cid:33) ∂Y3((cid:95)a, θ) ∂(cid:95)z ∂(cid:95)z d L , − β(cid:95)a, As seen from equation (12), parameter β describes the nature of the deformation effect caused both by the action of the stressed heteroboundary and the action of point defects of the type of compression or tension centers. This parameter can take both the positive values β > 0 (ε0 > 0, ∆Ω(2) > 0; ε0 < 0, ∆Ω(2) < 0)andthenegativevalues β < 0(ε0 > 0, ∆Ω(2) < 0; ε0 < 0, ∆Ω(2) > 0). Thesolutionofequations(13)withboundaryconditions(14)issearchedintheform: −λi θ Zi ((cid:95)z, θ), Yi ((cid:95)z, θ) = e where Zi ((cid:95)z, θ)satisfythefollowingequations: (cid:33) ∂2Z1((cid:95)z, θ) (cid:195) L(2) (cid:33)2(cid:195) 1− N (1) d0 (cid:33)2(cid:195) (cid:195) L(2) (cid:33) ∂2Z2((cid:95)z, θ) N 1 dc 1− N (2) d0 (cid:33) ∂2Z3((cid:95)z, θ) (cid:33)2(cid:195) (cid:195) L(2) N (2) dc 1− N (3) d0 N (3) dc = D3 D2 = D1 D2 ∂Z2((cid:95)z, θ) ∂Z3((cid:95)z, θ) ∂Z1((cid:95)z, θ) 2 ∂(cid:95)z 2 ∂(cid:95)z ∂(cid:95)z = d L d L d L ∂θ ∂θ ∂θ 2 + λ1eλ1θ, + eθ(cid:161)β+ 1(cid:162), + λ3eλ3θ, withboundaryconditions: e (cid:195) −λ1θ e (cid:195) (cid:33)2 (cid:195) L(2) (cid:33)2 (cid:195) L(2) d L D1 D2 d L −θ e −1+ N (2) d0 N (2) dc −1+ N (1) d0 N (1) dc (cid:175)(cid:175)(cid:175)(cid:95) ∂Z1((cid:95)z, θ) e ∂(cid:95)z z →−(cid:95) L −λ1θ Z1(−(cid:95)a, θ) = G(2) τ(2) d d G(1) τ(1) d d −θ Z2((cid:95)a, θ) = G(3) τ(3) d d (cid:33) ∂Z1(−(cid:95)a, θ) τ(2) G(2) d d (cid:33) ∂Z2((cid:95)a, t) ∂(cid:95)z = e + β(cid:95)a = D3 D2 ∂(cid:95)z (cid:175)(cid:175)(cid:175)(cid:95) ∂Z3((cid:95)z, θ) ∂(cid:95)z = 0. z →(cid:95) L R.M.Peleshchak,N.Ya.Kulyk,M.V.Doroshenko = 0, −θ Z2(−(cid:95)a, θ), e (cid:195) −λ3θ Z3((cid:95)a, θ), (cid:33)2 (cid:195) L(2) (cid:195) L(2) (cid:33)2 −θ d L e (cid:33) ∂Z2(−(cid:95)a, θ) (cid:33) ∂Z3((cid:95)a, t) ∂(cid:95)z − β(cid:95)a, , ∂(cid:95)z −1+ N (2) d0 (cid:195) N (2) dc −1+ N (3) d0 N (3) dc −λ3θ e d L (17) D1 D2 = 1, = 1, τ(2) d τ(3) d τ(2) d τ(1) d G(2) d G(3) d G(2) d G(1) d λ1 = λ2 = λ3 = 1, Solutionsofequations(16)withboundaryconditions(17)arepresentedintheappendixat (cid:33)2 (cid:195) L(2) (cid:33)2 (cid:195) L(2) = 1, 3. Analysisofthenumericalresultsanddiscussion In gures 2 and 3) there is shown the spatial-temporal redistribution of vacancies Y ((cid:95)z, θ) (gure 2) and interstitial atoms (gure 3) in a three-layer stressed nanoheterosystem GaAs/InxGa1−xAs/GaAs un- der the effect of the deformation caused by both the mismatch between parameters of the contacting lattices (ε0 = ∆a/a = 7%) and by the action of a point defect. Calculations were carried out for the fol- lowing values of the parameters: ε0 = 0.07; a = 0.05L, 0.1L (L -- the thickness of nanoheterostructure); = 1.223Mbar; d0 /N (1) N (1) = 0.571Mbar; T = 300K; θ(i ) C (1) 12 = 0.833Mbar; C (2) = 1µs. −5 cm2/s; τ(2) = 0.6; C (2) d0 /N (3) = 5eV[9]; Di = 10 = 0.453Mbar; C (1) = 0.5; N (2) = 0.8; N (3) d0 /N (2) = 1. D3 D2 d a d a dc dc dc 11 12 11 d d Figure 2. (Color online) Prole of the spatial-temporal distribution of the vacancies concentration in a three-layer stressed nanoheterosystem having inhomogeneous-compressed interlayer N (2) = 0.5; N (2) d0 /N (2) 23602-6 = 0.6; β = 10.2[formula(12)]. = 0.8; N (2) d0 /N (2) d0 /N (2) dc dc dc Spatial-temporalredistributionofpointdefects... Figure 3. (Color online) Prole of the spatial-temporal distribution of the concentration of intersti- tial atoms in a three-layer stressed nanoheterosystem having inhomogeneous-compressed interlayer d0 /N (2) N (2) = 0.6; β = −10.2[formula(12)]. = 0.8; N (2) = 0.5; N (2) d0 /N (2) d0 /N (2) dc dc dc As shown in gures 2 and 3, the prole of the spatial-temporal distribution of the defect concentra- tionofthetypeofcompression(vacancies,gure2)ortension(gure3)centersinathree-layerstressed nanoheterosystem is of a nonmonotonous character. If an internal epitaxial layer undergoes an inho- mogeneous compression deformation (ε0 < 0) due to the mismatch between the lattice parameter of the contacting epitaxial layers, then a decrease (an increase) of the concentration of vacancies (interstitial atoms)intheinterlayerofthethree-layernanoheterostructureswillbeobserved. If the epitaxial layer undergoes the tension deformation due to a mismatch between the lattice pa- rameter of the epitaxial layer and the substrate (as > a0, ε0 > 0, where as is the lattice parameter of the substrate; a0 isthelatticeparameterofthestackablelayer),theoppositeeffectwillbeobserved:nearthe heteroboundary there will be accumulation of vacancies and a decrease of the concentration of intersti- tialatoms.This,inturn,willleadtoadecreaseofthetensiondeformationintheepitaxiallayernearthe heteroboundary. Theeffectofimpoverishment(enrichment)intheinterlayerofvacancies(interstitialatoms)hasbeen observed in experimental works [6, 10] after the growth (decline) of the intensity of photoluminescence instressednanoheterostructures. Figure 4. The cut of the spatial-temporal distribution of the concentration of vacancies along the growth axis at different times: 1 -- at the moment t = τ(2) = 0.8; = 0.6; β = 10.2[formula(12)]; a = 0.05L. N (3) d0 /N (3) ; 2 -- t = 5τ(2) = 0.5; N (2) ; N (1) d0 /N (2) d0 /N (1) dc dc dc d d 23602-7 R.M.Peleshchak,N.Ya.Kulyk,M.V.Doroshenko dc d d d d d d0 d0 /N (2) dc ; 5τ(2) d (τ(2) d , 5τ(2) d = 0.5; N (2) ; N (1) d0 /N (1) dc ;2 -- t = 5τ(2) If there are no defects (N (i ) = 0.6; β = 10.2[formula(12)]; a = 0.1L. , respectively. Starting from the time t > 5τ(2) As seen from gures 4 and 5, during the time interval 0 (cid:201) t (cid:201) 5τ(2) Figure 5. The cut of the spatial-temporal distribution of the concentration of interstitial atoms along the growthaxisatdifferenttimes:1 -- atthemoment t = τ(2) = 0.8; N (3) d0 /N (3) In gures 4 and 5, numerical calculations of the cut of the spatial-temporal distribution of the con- centration of vacancies along the growth axis (OX) of heterosystem at different times are presented: t = 0; τ(2) is the average time of nding the defect in one of the equilibrium positions in the interlayer nanoheterosystem, namely a settled life) and for the different thicknesses of the interlayer of ananoheterostructure(a = 0.05L,gure4and a = 0.1L,gure5). there occurs a spatial-temporal redistribution of the defects, so that in the inhomogeneous-compressed interlayer [gure 1, region 2, formula (1)] they become smaller relative to their initial average value N (2) by ≈ 13.7%,16% at different times τ(2) , there practically establishes a stationary stateofthedistributionofthedefectsinathree-layerstressednanoheterosystem.Thus,thedeformation eld of the interlayer (−a (cid:201) z (cid:201) a) clears away the workspace from the defects which nally makes the materialoftheworkspacehavingagreaterintensityofphotoluminescence[10].Inexterlayers(gure1, regions1,3)oftheheterosystem,theconcentrationofthedefectsasymmetricallymonotonouslyincreases fromtheboundaryofthecontactingmaterialsandbecomeslargerthantheiraveragevalue N (1) , N (3) . = 0) in the contacting materials, inhomogeneous deformation is created only due to a mismatch between the lattice parameters of contacting materials [εi (z) = ε0 a2 ,i = 2], and in the absence of the mismatch between the lattice parameters (ε0 = 0), the deformation [U (i ) = d /K (i ))N (i ) 4. Conclusions • It has been established that the concentration prole of point defects at N (i ) is of non- monotonous character with a minimum in the middle of the interlayer InxGa1−xAs which is im- poverished by the point defect of the type of compression centers when the interlayer of the het- erostructureGaAs/InxGa1−xAs/GaAsundergoesaninhomogeneouscompression,whileinthecase ofinhomogeneoustensiontheoppositeeffecttakesplace.Thus,thedeformationeldandthenum- berofthedefectsintheworkspaceofnanooptoelectronicdevicescanbecontrolled. • Itisshownthatiftheratioofthethicknessofthemiddlelayertothethicknessesoftheexternallay- ersofananoheterosystemGaAs/InxGa1−xAs/GaAsis a/L = 0.05,theratiooftheinitialaveragecon- centrationsofthevacanciesinthelayersofananoheterosystemis N (1) = 0.8, = 0.6 and the value of the deformation parameter is β = 10.2, then the established con- N (3) d0 /N (3) centrationofvacanciesinthemiddlelayer Y ((cid:95)z,5τ(2) d )islessthantheinitialaverageconcentration ]iscausedonlybythepointdefects. = 0.5, N (2) d0 /N (1) dc < N (i ) dc d0 d0 /N (2) dc (θ(i ) d d0 d0 z2 d d0 dc 23602-8 Spatial-temporalredistributionofpointdefects... by16%.Iftheratio a/L = 0.1,thentheestablishedconcentrationofthevacanciesinthemid- N (2) dle layer is larger than the initial average concentration N (2) . Such a reduction of the established d0 concentration of the vacancies in the workspace of a nanoheterosystem is correlated with the ex- perimentalresultsofthework[10]. d0 Appendix Tondthesolutionofdifferentialequations(16)withboundaryconditions(17),theintegralLaplace transformationisused: Then,thedifferentialequation(16)andboundaryconditions(17)takeuptheform: Xi ((cid:95)z, p) = Zi ((cid:95)z, θ)e −pθdθ. ∞(cid:90) 0 (cid:161) z, p(cid:162)+ 1 (cid:161) z, p(cid:162)− p X1 (cid:161) z, p(cid:162)+ β+ 1 (cid:161) z, p(cid:162)− p X2 p − 1 (cid:161) z, p(cid:162)+ 1 (cid:161) z, p(cid:162)− p X3 p − 1 p − 1 (cid:180) = X2 (cid:179)−(cid:95)a, p (cid:179)−(cid:95)a, p (cid:180) (cid:180) = −H2X (cid:179)−(cid:95)a, p (cid:179)−(cid:95)a, p (cid:179)(cid:95)a, p (cid:180) = X3 (cid:179)(cid:95)a, p (cid:180) (cid:180) (cid:179)(cid:95)a, p (cid:180)+ β(cid:95)a = −H3X X2 (cid:48) 2 , , (cid:48) 3 p − 1 , (cid:48)(cid:48) 1 (cid:48)(cid:48) 2 (cid:48)(cid:48) 3 H1X H2X H3X X1 (cid:48) −H1X 1 (cid:179)(cid:95)a, p = 0, = 0, = 0. (cid:180)− β(cid:95)a p − 1 , −H2X (cid:48) 2 (18) (19) (20) (21) (22) 23602-9 where Hi = 1− N (i ) . Analyticalsolutionsofdifferentialequations(19)ineachlayerareasfollows: d0/N (i ) dc Integrationconstants C1, C2, C3, C4 aredeterminedfromthefollowingsystemofalgebraicequations: X1((cid:95)z, p) = C1exp X2((cid:95)z, p) = C2exp X3((cid:95)z, p) = C4exp C1exp (cid:112)H1exp (cid:112)H2exp (cid:181)−(cid:114) p (cid:181)(cid:114) p H1 H2 −C1 −C2 C2exp z 1 1 z z z H2 H1 H2 H2 (cid:182)+ β+ 1 p(cid:161)p − 1(cid:162) , (cid:182)+ (cid:181)(cid:114) p p(cid:161)p − 1(cid:162) , (cid:182)+C3exp (cid:181)(cid:114) p (cid:181)(cid:114) p (cid:182)+ (cid:181)(cid:114) p p(cid:161)p − 1(cid:162) , (cid:181)(cid:114) p (cid:181)−(cid:114) p (cid:182)−C2exp (cid:181)−(cid:114) p (cid:182)−C3exp (cid:181)(cid:114) p (cid:182)−C3 (cid:181)−(cid:114) p (cid:182)+C2 (cid:112)H2exp (cid:112)H2exp (cid:181)−(cid:114) p (cid:182)−C4exp (cid:181)−(cid:114) p (cid:182)+C3exp (cid:181)(cid:114) p (cid:114) p (cid:182)+C3 (cid:181)− (cid:181)−(cid:114) p (cid:182)+C4 (cid:112)H2exp (cid:112)H3exp H2 H1 H2 H2 H2 (cid:95)a (cid:95)a (cid:95)a (cid:95)a (cid:95)a (cid:95)a (cid:95)a (cid:95)a H2 −(cid:95)a (cid:201) (cid:95)z (cid:201) −(cid:95) l , −(cid:95)a (cid:201) (cid:95)z (cid:201) (cid:95)a , (cid:95)a (cid:201) (cid:95)z (cid:201) −(cid:95) l . β (cid:182) = p(cid:161)p − 1(cid:162) , (cid:182) = p(cid:161)p − 1(cid:162) , β(cid:95)a(cid:112) (cid:182) = − p(cid:161)p − 1(cid:162) , (cid:182) = p(cid:161)p − 1(cid:162) . β(cid:95)a(cid:112) β (cid:95)a (cid:95)a (cid:95)a (cid:95)a H2 H2 H3 H3 R.M.Peleshchak,N.Ya.Kulyk,M.V.Doroshenko By carrying out the inverse Laplace transformation we obtain the spatial-temporal redistribution of thepointdefectsintherst,secondandthirdstressedlayers,respectively: Y1((cid:95)z, θ) = e Y2((cid:95)z, θ) = Z2((cid:95)z, θ) Y3((cid:95)z, θ) = e (cid:179) −θ Z1((cid:95)z, θ)− e 1− e −θ Z3((cid:95)z, θ)− e −θ + 1, −θ(cid:180)(cid:161)1+ β(cid:162), −θ + 1, wherefunctions Z1((cid:95)z, θ), Z2((cid:95)z, θ)and Z3((cid:95)z, θ)arethesolutionsofdifferentialequations(16)withbound- aryconditions(17),respectively: (cid:34) (cid:180)(cid:35) Z1((cid:95)z, θ) = 1 eθ − 1+ eθ l2i−1r2i−1θk Φ1((cid:95)z, θ,k)+ ∞(cid:88) × ∞(cid:88) b (cid:182) ∞(cid:88) m=0 i=1 (−1)m am bm+1 (−1)m am bm+1 ∞(cid:88) l3 + 2(cid:88) Erf(cid:179)(cid:112) l2r2θk Φ2((cid:95)z, θ,k), where (cid:95)a(cid:112) H2 (24) m=0 k=1 k=1 (cid:182) (cid:181) θ Φ1((cid:95)z, θ,k) = 1F1 (23) (25) (26) (27) (28) a =(cid:179)(cid:112)H1 −(cid:112) r1 = − (cid:95)z + (cid:95)a(cid:112) H1 (cid:182) , 2 2 2 1 2 1 2 2 , 1 2 2 4θ 2 4θ H2 (cid:112) (cid:112) 1F1 (cid:112) − (−1)i (cid:181)−k, 1 2 ,− r 2 − k, 3 2 ,− r 2 θΓ(cid:161)k + 1 (cid:162) 2i−1 2i−1 + 1F1 4θ 4θ (cid:181) (cid:182) (cid:181) r2i−1Γ(k + 1) 2 ,− r 2 − k, 1 2 ,− r 2 1− k, 3 θΓ(cid:161)k + 1 (cid:162) θr2i−1Γ(cid:161)k + 1 (cid:162) + 1F1 2i−1 2i−1 4θ 4θ (cid:182) (cid:181)−k, 1 (cid:181) (cid:182) 2 ,− r 2 2 ,− r 2 − k, 3 θΓ(cid:161)k + 1 (cid:162) + 1F1 r2Γ(k + 1) (cid:180)(cid:179)(cid:112) b =(cid:179)(cid:112)H1 +(cid:112) (cid:180) H2 −(cid:112)H3 (cid:179)(cid:112)H3 −(cid:112)H2 (cid:180) (cid:180) (cid:179)(cid:112)H3 +(cid:112) + 2(cid:95)a (2m + 1) r2 = − (cid:95)z + (cid:95)a(cid:112) (cid:104) (eθ − 1)(p3 + p4)+ (p5 + p6)eθErf( Z2((cid:95)z, θ) = 1 ∞(cid:88) 4(cid:88) b (cid:181) Φ2((cid:95)z, θ,k) = − 1F1 (cid:180)(cid:179)(cid:112) l1 = β(cid:112)H2 l3 = β(cid:112) + 4(cid:95)a (m + 1) H2 (cid:95)a(cid:112) l2 = − l1 H2 (cid:95)a(cid:112) l4 = l3 H2 (−1)m am bm+1 (cid:182) pi ei θk Φ3((cid:95)z, θ,k), r3 = − (cid:95)z + (cid:95)a(cid:112) (cid:105) (cid:112) θ) ∞(cid:88) m=0 k=1 i=1 (cid:112) (cid:112) H1 H2 H2 H2 H1 H2 + , , , , , , , H2 +(cid:112)H3 (cid:180) , Φ3((cid:95)z, θ,k) = − 1F1 (cid:112) i 4θ (cid:182) (cid:181)−k, 1 2 ,− e2 (cid:181) ei Γ(k + 1) 2 ,− e2 1− k, 3 θΓ(cid:161)k + 1 (cid:162) + 1F1 (cid:182) (cid:162) + 1F1 i 4θ i 4θ 2 (cid:182) , 1 2 i 4θ 2 ,− e2 − k, 3 θΓ(cid:161)k + 1 (cid:162) (cid:181) 2 ,− e2 − k, 1 θei Γ(cid:161)k + 1 1 2 2 1F1 (cid:112) 2 + (cid:95)a(cid:112) H2 + 4(cid:95)am(cid:112) H2 . where 23602-10 where (29) (30) (31) Spatial-temporalredistributionofpointdefects... (cid:180) , (cid:95)a(cid:112) p3 = − p2 H2 (cid:179)(cid:112)H2 −(cid:112)H3 (cid:180) , , , , , H2 (cid:112) (cid:180) p2 = β(cid:95)a (cid:179)(cid:112)H2 −(cid:112)H3 p1 = β(cid:112)H2 p4 = −β(cid:112)H3 (cid:179)(cid:112)H3 −(cid:112)H2 (cid:179)(cid:112)H1 +(cid:112)H2 (cid:180) (cid:95)a(cid:112) p5 = − p4 H3 + 4(cid:95)a (m + 1) e1 = − (cid:95)z + (cid:95)a(cid:112) e2 = − (cid:95)z + (cid:95)a(cid:112) e3 = (cid:95)z − (cid:95)a(cid:112) + 4(cid:95)a (m + 1) e4 = (cid:95)z − (cid:95)a(cid:112) (cid:34) (cid:35) l3 + 2(cid:88) Z3((cid:95)z, θ) = 1 l2i−1r2i−1θk Φ4((cid:95)z, θ,k)+ ∞(cid:88) × ∞(cid:88) b (cid:182) p6 = β(cid:112)H1 + 2(cid:95)a (2m + 1) + 2(cid:95)a (2m + 1) ∞(cid:88) m=0 i=1 (−1)m am bm+1 (−1)m am bm+1 ∞(cid:88) eθ − 1+ eθ (cid:95)a(cid:112) H2 (cid:112) θ) Erf( m=0 k=1 k=1 (cid:181) (cid:112) (cid:112) (cid:112) H2 H2 H2 H2 H2 H2 H2 , . , , Φ4((cid:95)z, θ,k) = − 1F1 l2r2θk Φ5((cid:95)z, θ,k), (cid:182) , 2 2 1 2 1 2 (cid:112) (cid:182) 1F1 (cid:112) (cid:181)−k, 1 2 ,− r 2 − k, 3 2 ,− r 2 θΓ(cid:161)k + 1 (cid:162) 2i−1 2i−1 + 1F1 4θ 4θ (cid:181) (cid:182) (cid:181) r2i−1Γ(k + 1) 2 ,− r 2 1− k, 3 2 ,− r 2 − k, 1 θΓ(cid:161)k + 1 (cid:162) θr2i−1Γ(cid:161)k + 1 (cid:162) + 1F1 2i−1 2i−1 4θ 4θ (cid:182) (cid:181)−k, 1 (cid:182) (cid:181) 2 ,− r 2 2 ,− r 2 θΓ(cid:161)k + 1 (cid:162) r2Γ(k + 1) (cid:180) (cid:179)(cid:112)H1 −(cid:112)H2 (cid:180) (cid:179)(cid:112)H2 +(cid:112)H1 + 2(cid:95)a (2m + 1) r2 = (cid:95)z − (cid:95)a(cid:112) (cid:95)a(cid:112) l2 = − l1 H2 (cid:95)a(cid:112) l4 = l3 H2 r3 = (cid:95)z − (cid:95)a(cid:112) + 1F1 − k, 3 (cid:112) 2 4θ 2 4θ 1 2 2 2 , , , , , , (cid:112) + 4(cid:95)am(cid:112) H2 . H1 + (cid:95)a(cid:112) H2 Φ5((cid:95)z, θ,k) = − 1F1 l1 = β(cid:112)H2 l3 = β(cid:112)H2 r1 = (cid:95)z − (cid:95)a(cid:112) H1 + 4(cid:95)a (m + 1) (cid:112) , H2 H1 H2 References 1. LedentsovN.N.,UstinovV.M.,ShchukinV.A.,Kop'evP.S.,AlferovZh.I.,BimbergD.,Fiz.Tekh.Poluprovodn.,1998, 32,No.4,385[Semiconductors,1998,32,No.4,343;doi:10.1134/1.1187396]. 2. KozlovskiiV.I.,LitvinovV.G.,Sadof'evYu.G.,Fiz.Tekh.Poluprovodn.,2000,34,No.8,998[Semiconductors,2000, 34,No.8,960;doi:10.1134/1.1188108]. 3. Vilisova M.D., Kunitsyn A.E., Lavrent'eva L.G., Preobrazhenskii V.V., Putyato M.A., Semyagin B.R., Toropov S.E., Chaldyshev V.V., Fiz. Tekh. Poluprovodn., 2002, 36, No. 9, 1025 [Semiconductors, 2002, 36, No. 9, 953; doi:10.1134/1.1507270]. 4. Vilisova M.D., Ivonin I.V., Lavrentieva L.G., Subach S.V., Yakubenya M.P., Preobrazhenskii V.V., Putyato M.A., Semyagin B.R., Bert N.A., Musikhin Yu.G., Chaldyshev V.V., Fiz. Tekh. Poluprovodn., 1999, 33, No. 8, 900 [Semi- conductors,1999,33,No.8,824;doi:10.1134/1.1187790]. 5. Zhukov A.E., Egorov A.Yu., Ustinov V.M., Tsatsul'nikov A.F., Maksimov M.V., Faleev N.N., Kop'ev P.S., Fiz. Tekh. Poluprovodn.,1997,31,No.1,19[Semiconductors,1997,31,No.1,15;doi:10.1134/1.1187044]. 6. Karpovich I.A., Anshon A.V., Filatov D.O., Fiz. Tekh. Poluprovodn., 1998, 32, No. 9, 1089 [Semiconductors, 1998, 32,No.9,975;doi:10.1134/1.1187528]. 23602-11 R.M.Peleshchak,N.Ya.Kulyk,M.V.Doroshenko 7. PeleschakR.M.,KusykO.V.,Ukr.J.Phys.,2007,52,No.7,689. 8. FractureMechanicsandStrengthofMaterials.ReferenceGuidein4volumes,vol.3,PanasyukV.V.(Ed.),Naukova 9. Emel'yanovV.I.,PaninI.M.,LaserPhys.,1996,6,No.5,971. 10. ChenY.C.,SinghJ.,BhattacharyaP.K.,J.Appl.Phys.,1993,74,No.6,3800;doi:10.1063/1.354472. Dumka,Kiev,1988,p.463(inRussian). Просторово-часовийперерозподiлточковихдефектiву тришаровихнапруженихнаногетеросистемахумежах самоузгодженоїдеформацiйно-дифузiйноїмоделi Р.М.Пелещак,Н.Я.Кулик,М.В.Дорошенко Дрогобицькийдержавнийпедагогiчнийунiверситетiм.IванаФранка, вул.I.Франка,24,82100Дрогобич,Україна Побудованомодельпросторово-часовогорозподiлуточковихдефектiвутришаровiйнапруженiйнаноге- теросистемi GaAs/InxGa1−xAs/GaAs з врахуванням самоузгодженої деформацiйно-дифузiйної взаємодiї. У межах цiєї моделi розраховано профiль просторово-часового розподiлу вакансiй (мiжвузлових атомiв) унапруженiйнаногетеросистемiGaAs/InxGa1−xAs/GaAsтапоказано,щоувипадкувстановленнястацiо- нарногостану(t > 5τ(2) ),концентрацiявакансiйвнеоднорiдностиснутомувнутрiшньомушарiєменшою вiдносновихiдногосередньогозначення N (2) Ключовiслова:просторово-часовийперерозподiл,вакансiї,мiжвузловiатоми на16%. d d0 23602-12
1805.11283	1	1805	2018-05-29T07:42:23	Switching mechanism of CO2 by alkaline earth atoms decorated on g-B4N3 nanosheet	[ "cond-mat.mes-hall" ]	The adsorption and desorption of carbon dioxide (CO2) molecule by alkaline earth metal (AEM) (Mg+2, Ca+2, Sr+2 and Ba+2) functionalized on graphitic boron nitride (g-B4N3) nanosheet have been analyzed by using density functional theory (DFT) approach includes long-range correlation (DFT+D). This method has been implemented in such a way to understand the switchable or capture/release mechanism of CO2 molecule by computing the electron mobility, electronic properties, charge accumulation, charge transfer (e-) and adsorption energy (Kcal/mol). The g-B4N3 nanosheet yields high carrier mobility (8020 cm2 V-1s-1) at 300 K. The positive alkaline earth adatoms on the nanosheet of g-B4N3 has been provided external energy to do the capture/release process of greenhouse gas CO2. Here, Mg positive ion work as adatom which confirms physisorption while others show chemisorption behaviors. Therefore, due to the weak absorption of CO2, it makes possible to discharge from the g-B4N3 nanosheet and shows instantaneous switching mechanism. Briefly, the negatively charged g-B4N3 nanosheets are highly sensitive for CO2.	cond-mat.mes-hall	cond-mat	Switching mechanism of CO2 by alkaline earth atoms decorated on g-B4N3 nanosheet Shivam Kansara1, Sanjeev K. Gupta2,, Yogesh Sonvane1, and Anurag Srivastava3 1Advanced Materials Lab, Department of Applied Physics, S.V. National Institute of Technology, Surat 395007, India 2Computational Materials and Nanoscience Group, Department of Physics, St. Xavier's College, Ahmedabad 380009, India 3Advanced Materials Research Group, CNT Lab, ABV-Indian Institute of Information Technology and Management, Gwalior 474010, India Abstract The adsorption and desorption of carbon dioxide (CO2) molecule by alkaline earth metal (AEM) (Mg+2, Ca+2, Sr+2 and Ba+2) functionalized on graphitic boron nitride (g-B4N3) nanosheet have been analyzed by using density functional theory (DFT) approach includes long-range correlation (DFT+D). This method has been implemented in such a way to understand the switchable or capture/release mechanism of CO2 molecule by computing the electron mobility, electronic properties, charge accumulation, charge transfer (e-) and adsorption energy (Kcal/mol). The g-B4N3 nanosheet yields high carrier mobility (8020 cm2 V-1s-1) at 300 K. The positive alkaline earth adatoms on the nanosheet of g-B4N3 has been provided external energy to do the capture/release process of greenhouse gas CO2. Here, Mg positive ion work as adatom which confirms physisorption while others show chemisorption behaviors. Therefore, due to the weak absorption of CO2, it makes possible to discharge from the g-B4N3 nanosheet and shows instantaneous switching mechanism. Briefly, the negatively charged g-B4N3 nanosheets are highly sensitive for CO2. Keywords: CO2 switching; g-B4N3 nanosheet; alkaline earth adatoms; Charge accumulation; Density functional calculations *Corresponding author(s): [email protected] (Dr Yogesh Sonvane) [email protected] (Dr Sanjeev K. Gupta) 1 Introduction In the last few years, the greenhouse emission has become a great concern for the whole scientific community as it affected the climate badly. As from the results, various methods have been proposed to separate, capture, store and detect greenhouse gases [1–3], with a significant contribution due to CO2 in global warming [4,5]. Here, the key issues are predicting / proposing new materials as efficient to CO2, capturing, storing and/or modifying materials [6,7]. Though in industries, the most common adsorbent being utilized for CO2 capturing is aqueous amine solutions or chilled ammonia [8,9], the method suffers from low productivity, equipment corrosion, and solvent loss. To make this process a cost-effective and robust, the choice of adsorption material is critical. Therefore, various solid materials have been attempted as a CO2 adsorbent, like metal-organic frameworks (MOFs) [10–12], porous graphene [13,14], carbon nanotube [15], silicon carbide (SiC) nanotube [16], graphene nanosheet [17] and aluminium nitride (AlN) nanosheet [18]. Recently, hexagonal boron-nitride (h-BN) monolayer, a structural analogue of graphene, has attracted enough attention of the scientific community. The periodic group III- V compounds, with high thermal, chemical stability, and high conductivity are suitable for different applications in electronics, optoelectronics and catalysis applications [19,20]. Due to strong covalent forces, all the atoms of each layer of BN are tightly bound together and removal of one N-atom creates defect or vacancies. It is a well-known fact in the multilayer system, where weak van der Waals (vdW) forces are responsible to hold them together as graphite. This has been the key motivation for choosing g-B4N3 nanosheet as a CO2 adsorbent. Experimentally, h-BN films drawn from bulk BN crystal using micro-mechanical cleavage and a dielectric layer [21] with intrinsic iconicity in B-N bonds and high surface areas of h-BN nanosheet, as well as its tubular analogues, suggested for the gas separation or as a gas storage [22–26]. There are number of experimental investigation for CO2 adsorption using zeolites and mesoporous adsorbents in the literature. A previous study of adsorbents for the CO2 removal has been commonly used such as 13X zeolites [27,28], metal-organic frameworks (MOFs) [29], activated carbons [30], mesoporous silica [31,32] and surface-functionalized silica [33]. Meanwhile recently, electro-catalytically switchable CO2 capture scheme has been proposed for a controlled high selective and reversible h-BN nanosheet [34,35]. Further, the CO2 molecules are weakly adsorbed (i.e. physisorbed) on neutral h-BN nanosheet. In earlier studies [34–36], the CO2 capture/release has been reported spontaneously and it may 2 be merely controlled. h-BN is a wide-gap semiconductor with band gap around 5.8 eV [36]. Generally, the metal atoms should not aggregate to form clusters [37] thus in our study, metal atoms (Mg, Ca, Sr, Ba) have supportive substrate of g-B4N3. By adding positive ions from group II elements on g-B4N3 nanosheet, except beryllium (Be) and radium (Ra), confirms the CO2 adsorption. Further, the chemically adsorbed CO2 has been released and electrons are ejected. Due to the absence of the experimental and theoretical data of g-B4N3 nanosheet, the texture properties are also unknown. Meanwhile, the graphitic carbon nitride nanosheet has been successfully synthesized [38–42] and described its synthesis, functionalization, and applications process [43,44] so g-B4N3 nanosheet can be synthesized as the same. Here, we show that an electro-catalytically switchable mechanism of CO2 capture is indeed possible on a conductive g-B4N3 nanosheet; there is no reported work with same structural configuration. So we have compared our work with g-C4N3 nanosheet [45]. The present work also shows that the greenhouse gas CO2 capture/release is spontaneous and improved with positive alkaline earth adatoms on the nanosheet of g-B4N3. These cycles are merely controlled and reversed without any switching on/off the charging voltage. In addition, these negatively charged g-B4N3 nanosheets with alkaline earth adatoms are good for the switching CO2 from mixtures of gases in the atmosphere. This manuscript is divided into 4 sections. In upcoming section 2, the methodology has been discussed, followed by results and discussions in section 3 and conclusions in section 4. Methodology The present computational analysis of CO2 capturing has been analyzed using Quantum Espresso (QE) code [46]. The generalized gradient approximation (GGA) in Perdew-Burke-Ernzerhof (PBE) [47] as well as HSE06 [48] type functionals have been customized for the treatment of the exchange-correlation. As the standard PBE functional is inefficient to define accurately weak correlation effects, the long-range interaction approach of DFT+D [49,50] has been used in the calculations. In all the computations, the substrate and adatoms have been assumed fixed, whereas CO2 molecule relaxed. Further, the system is represented as the three-layer system. Where, the vacuum has been approximated as 20 Å from the upper surface as CO2 of the system, which is large enough to avoid the interaction with the periodic images. In this case, 2×2 supercells with 8×8×1 Monkhorst-Pack k-point mesh [51] have been taken. The total energy includes the Kohn-sham energy, weak interaction as vdW and EDFT are described as 3 (2)dispE is given by, (1) (2) Where represents the number of present atoms, shows the dispersion coefficient, signify the distance between atoms for atom pair ij, and S6 as a scaling factor, subject to the exchange correlation (XC)-functional. A damping factor is utilized to inhibit singularities at few distances. To investigate the adsorption on adsorbent, the adsorption energy ( ) of CO2 molecule on g-B4N3 can be calculated as below [45] (3) Where , and are the total energy of a complex system, the energy of nanosheet with adatoms, and energy of an adsorbed CO2 molecule on g- B4N3, respectively. Results and discussions a. Structural and electronic properties Fig. 1(a) and (b) shows the lowest energy configuration of structure and electron mobility of isolated g-B4N3 nanosheet. The unit cell of g-B4N3 nanosheet consists of seven atoms, where, four atoms are of B and three atoms of N. The average bond length between B-B, B-N and B- B is 1.43 Å with the buckled angle of 113º. In case of the static structure of g-B4N3, the value of electron mobility is 8020 cm2/V·s for 300 K and computed using ab initio model for mobility and Seebeck coefficient using Boltzmann transport (MoBT) equation [52]. High electron mobility is a prerequisite for injecting extra electrons into electro-catalytically switchable CO2 capture materials. This readily facilitates the electron injection/release for electrocatalytically switchable CO2 capture. The value of electron mobility of planer h-BN monolayer is between the range of 9200-28800 cm2/V·s [53], which is slightly larger than presenting monolayer. 4 (2)DFTDKSDFTdispEEE(2)dispE61(2)6611()NNatatijdmpijdispijiijCEsfRRatN6ijCijRdmpfadsE()432()complexBNadatomsCOadsEEEEncomplexE43()BNadatomsE2COE Figure 1: Top (upper) and side (lower) display (a) nanosheet of g-B4N3 where the defect has been created by removing one nitrogen atom from unit cell. The blue and yellow balls represent B and N atoms, respectively, and the unit cell of g-B4N3 is represented by a dark red line. The calculated electron mobility (b) of nanosheet of g-B4N3 with varying temperature and the red line represents the mobility at 300K temperature. Table 1: Calculated formation energy using electronic energy and its comparison with earlier reports for different 2D material per atom. Reported work [39,40] Present work Structure graphene silicene Si2BN BN g-B4N3 Formation energy (eV) -84.95 -38.20 -50.38 -81.70 -46.76 At 50 K, the electron mobility is at its highest peak  3635 ×102 cm2/V·S. The electron mobility can be defined by the equation (4); (4) Where, is the drift velocity; and E is electron mobility and magnitude of the electric field, respectively. Table 1 represents the formation energy for the different system at 0 K. These energies were calculated for the unit cell and the formation energy of g-B4N3 has a good comparison to the Si2BN [54]. The previously reported works [55,56] were also calculated the formation energy using chemical potential instead of electronic energy of defected h-BN nanosheet. The high formation energy of g-B4N3 confirms the stability of the structure. The formation energy ( ) using chemical potential of doped AEM sheets have been computed through the following equation (5); 5 dEdfE (5) In equation 5, and are the total energies of considered defect sheet and pure sheet of boron nitride (BN). , and are chemical potentials of boron, nitrogen and alkaline earth atoms, respectively. n represents the number of atoms present in the sheet. The values of the formation energy are shown in the ESI Table S2. The band structure and projected density of states (PDOS) have been computed for the nanosheet of g-B4N3 and is shown in Fig. 2(a) and (b) using the PBE functional. The electronic band profile has been calculated with respect to the highest symmetries in the first Brillouin zone for an energy range of -6 to 6 eV. In the band curve Γ- M- K plane of the Brillouin zone (BZ) bands split, due to the defect introduced by replacing one atom of N from the unit cell. It has been assumed that by replacing one N atom from the h-BN nanosheet, the Fermi level gets shifted downward to the valence band. From band structure Fig. 2 (a), the nanosheet of g-B4N3 is found metallic. The calculated band structure and PDOS using HSE06 functional are shown in Electronic Supplementary Part (ESI) as Fig. (S1). Figure 2: The calculated (a) electronic band curves of nanosheet of g-B4N3. The blue and black lines show the conduction and valence band, respectively. (b) Density of states of a nanosheet of g-B4N3. The red dotted line represents the Fermi level. To understand the electronic coupling and orbital contributions, the PDOS profile has also been computed, shows the contribution of subshell from the last orbital (p) of B and N atoms, in the states at Fermi level. The 2px subshell and outermost 2pz subshell of B and N atom are much contributed at the Fermi level where N atom is more dominant to the B atom. For the g-B4N3 nanosheet, the total PDOS shows the mixed character of the B and N atoms of 6 fdefectpristineBBNNddEEEnnndefectEpristineEBNdtheir conduction band (CB) and valence band (VB) channels. VB is more dominated by B-2p orbitals in comparison to small contribution from N-2p orbitals. Furthermore, B-2p orbitals dominate at the CB, while all the details regarding adsorption energy and bader charge have been illustrated in Fig. 4. Figure 3 (a-d) shows the partial density of states of the AEM doped BN sheets. The p orbital of B and N are highly contributed near and at the Fermi level in Figure 3. As we can see that as increase the number of electron of the systems, s and p orbital of alkaline metal and CO2 molecule is going to be more dominate near the Fermi level, respectively. This hybridization affects the bonding of the CO2 molecule to the sheet at the lowest distance. The grey shaded region represents the total density of states. Figure 3: Partially density of states of group II atoms doped BN sheets. The projections of B with N and C with O atoms are considered together for the p orbitals, respectively. The dotted dashed lines represent the Fermi level. The adsorption energies and charge transfer of CO2 have been calculated on neutral and negatively charged g-B4N3 and also by adsorbing positive ions from alkaline earth elements. Fig. 4 depicts the comparative adsorption energies of CO2 on g-B4N3 with alkaline earth adatoms and the charge transfer with respect to the bond length between the oxygen 7 atoms of CO2 to g-B4N3. In contrast, CO2 is weakly adsorbed on the Sr+2 atom with smallest adsorption energy. It has been noticed that reaction enhances from -0.35 kcal/mol to 9.66 kcal/mol in the case of Ba and Mg adatoms, the processes go from chemisorption to physisorption, respectively. CO2 is tightly chemisorbed on g-B4N3 for the Ca+2 adatom with large adsorption energy as -13.37 kcal/mol. The total charge transfer for Ca+2 adatom is 0.46 e. The details of adsorption and Bader charge transfer for the different adatoms are shown in Fig. 3. The adsorption energy of adatoms as Mg+2, Ca+2, Sr+2 and Ba+2 with the substrate of g- B4N3 nanosheet is shown in Electronic Supplementary Part (ESI) as Table (S1). Figure 4: Comparison between the (a) adsorption energy (Kcal/mol) (ΔEDFT) (b) Charge transfer (e-) calculated with respect to the bond length (Å) using DFT and energies estimated using universal scaling relations (ΔEpredicted). b. Adsorption of CO2 molecule on the nanosheet and charge accumulation To visualize the charge redistribution on CO2 molecule, the charge density difference or charge accumulation has been computed using following equation (6); (6) Where , , ,and are the total charge density of complex, density of g-B4N3, density of extra injected positive ions as adatoms and density of molecules, respectively. Figs. 5(a-e) addresses the adsorption process with different adatoms interacting with a negative negatively charged nanosheet g-B4N3, where distance and angle of CO2 molecule have been changed due to the effect of the system. The charge transfer of total five systems 8 ..432432gBNadatomCOgBNadatomCO432(.)gBNadatomCO43.gBNadatom2COsuch as direct CO2 molecule adsorption on nanosheet g-B4N3 along with different alkaline adatoms as Mg+2, Ca+2, Sr+2, Ba+2 atoms have been analyzed. From the Figs. 5(a-e), we can see that the charge transfer regions are between two interfaces of CO2 molecule and nanosheet. This process is analyzed by surface hybridization of the top layer to the bottom layer, which is indicating the effect of weak vdW interlayer interaction and covalent forces. The interaction between a CO2 molecule and alkaline earth adatoms exerts a driving force, these plays a basic role in the electron transfer from the molecule to the adatoms region, while holes move in the converse direction. The charge deficient region between the O=C=O concludes the loss of charge, which decreases from double bond to the O=C-O bond formation. So, the Bader charge analyzer helps to calculate the electron transfer. Figure 5: Top and side views of CO2 sensing and charge accumulation on a 2×2 g-B4N3 nanosheet. Atom color code: blue and yellow atomic colors are represented as green and silver in accumulation, respectively. Fig. 5 depicts the process of direct CO2 molecule adsorption on g-B4N3 nanosheet and noticed that the distance of CO2 molecule and angle have been changed as compared to initial position. Due to the strong repulsive force arising from electron density as well as a rapid increase of ground energy, the molecular relative distance has been increased. The yellow and blue colors are in the figures representing the positive and negative value of iso-surface, respectively. In all the figures, the yellow surfaces are bound to state, where the plotted amount is more prominent than a (positive) threshold, and the blue surfaces bound regions where the plotted amount is not much as an alternate (negative) threshold. The blue cloud means losing charges and yellow cloud represents the gaining of charges, respectively. At the initial structure, there is a high electron density cover the nanosheet and mixed electron density of both while after moving far from CO2 gas from the substrate the electron density also get separated as shown in Fig. 5. 9 Figure 6: Top and side views of switching of CO2 and charge accumulation on a 2×2 g- B4N3+ Mg as adatoms on nanosheet. Atom color code: blue, yellow and magenta colors as green, silver, and orange in accumulation, respectively. Fig. 6 shows that the Mg2+ ion makes high interaction among the molecules. From the initial state to final state, g-B4N3 nanosheet plays a significant role to make a possible switching of CO2. The external injected positive ion as Mg2+ strongly interacts with the electron enriched O atom of CO2 molecule. Bader charge analysis disclosed noteworthy charge transfer (0.42 e) from CO2 molecule to the substrate at the time of bonding. The transferred charges were non-uniformly distributed during the whole process. The reason for this interaction is the quantum mechanical change of momentary dipoles, which begins from the shared electrons in the adsorbate and the surface. However when an atom comes closer to the surface, kinetic energy of the electrons increases, as an effect of overlapping of wave functions of the atom. This is responsible for the high repulsion potential [57]. As well as also introduced the pressure dependent adsorption process in the Table S3. In this case, the adsorption energy is positive so for an easy physisorption process, the relation between the adsorbate concentration and the gas molecule can be represented by the Langmuir adsorption isotherm equation (7): (7) 10 1bpbpWhere, is surface coverage, b and p are the temperature constants. Fig. 7 shows a CO2 molecule adsorbed on a Ca2+ adatom. In this case, the adsorption energy [ ] increases from that for the single support of g-B4N3 nanosheet. Therefore, the supported Ca2+ adatom would prefer to be adsorbed on O atoms from the CO2 molecule. The bader charge analysis data indicate that the complex system and CO2 molecule have 0.460e and 0.125e charges at the process of adsorption. And also carry a little negative charge on the adatoms. The accumulation and depletion charges are raised on the system at the time of switching, while the depletion process begins on behalf of accumulation. To understand the bonding property of CO2 molecule on adatom, the two simple processes such as bonding and after repel CO2 molecule like switching process is also analyzed here. The charge redistribution form corresponds to each new system, so the visual diagram could be more clearly expressed by the charge distribution. From the strong charge distribution on the accumulation or depletion recognized the strength of the bonding and distributed charge cloud determined the orbit bonding [58]. In all figures of charge transfer, blue and red color represents the charge accumulation and depletion, respectively. Here, the adsorption energy is chemisorption so that it should be chemical bonding. 11 adsEFigure 7: Top and side views of switching of CO2 and charge accumulation on a 2×2 g- B4N3+ Ca as adatoms on nanosheet. Atom color code: blue, yellow and green colors as green, silver, and black in accumulation, respectively. The adsorption of CO2 molecule on g-B4N3 nanosheet supported with extra adatom as a catalyst Sr2+, there are three cases displayed in this case, such as initial, lowest distance and repelled position of CO2 molecule (See Fig. 8). The adsorption of the CO2 molecule in the complex system or the distance of the molecule from the substrate is 3.10 Å, 2.50 Å and it repels at 2.88 Å at switching process. It is weakly chemically adsorbed on the surface with the adsorption energy of -0.99 kcal/mol. Therefore, the CO2 is adsorbed on the supported catalyst as adatom Sr2+ on the g-B4N3 support. Figure 8: Top and side views of switching of CO2 and charge accumulation on a 2×2 g- B4N3+ Sr as adatoms on nanosheet. Atom color code: blue, yellow and magenta colors as green, silver, and orange in accumulation, respectively. 12 Figure 9: Top and side view of switching of CO2 and charge accumulation on g-B4N3+ Ba as adatoms on nanosheet. Atom color code: blue, yellow and sky blue colors as green, silver, and green in accumulation, respectively. To further explain the charge transfer phenomenon between gas adsorbates and substrate, Fig. 9 shows the iso-surface plot of electron charge density gas molecules. Here, the charge difference is calculated in three ways by subtracting the electron densities of non- interacting components (gas) from the charge density of the adatoms and initial system with the same atomic positions of the systems. It is also seen that there is a charge accumulation on the substrate for CO2 adsorbate characteristic. In this case, CO2 acts as an electron acceptor from the Bader charge calculation. The CO2 molecule accepts the electron from the adatom as a catalyst and shows the charge depletion region between the inner surface of the molecule and the substrate. The adsorption energy of this system is -0.93 Kcal/mol, and the distance between molecules to the substrate is 3.30 Å and after making bond, it is 2.63 Å with -1.62 Kcal/mol adsorption energy. The small value of adsorption energy with respect to atomic distance indicates a weak interaction. The charge transfer between CO2 to the substrate has been obtained from bader charge analysis (Fig. 4). The adsorption of CO2 on 13 Ba2+, the calculated charges on the CO2 and complex system are 0.13 e and 0.512 e, respectively. Meanwhile, very dense charges have been transferred from the upper two layers as CO2 and Ba2+. Our reported work is being used for the instant switching mechanism of the CO2 molecule due to the small adsorption value with compared with other works [31] and shown in Table 2. Table 2: The comparisons of adsorption energy with previous works have been compared. Work Systems Supportive Adsorption Energy (Kcal/mol) Our Graphitic BN nanosheet Ref. [35] BN nanosheet BN nanotube Ref. [59] C2N sheet Hybrid BN/BN/G sheet Hybrid BN/G sheet Ref. [44] Conclusions Mg Ca Sr Ba 0 e- 1 e- 0 e- 1 e- - 0 e- 4 e- 0 e- 4 e- 9.23 -13.37 -0.99 -1.62 -49.57 -13.87 -69.21 -11.91 -8.05 -5.30 -59.49 90.16 -39.43 In summary, we have used DFT-D to investigate the adsorption and desorption of CO2 molecule on g-B4N3. The structure of g-B4N3 nanosheet has high electron mobility 8020 cm2/V·S at 300 K and the formation energy -6.67 eV/atom, which confirms the stability of the system. As well as, we also focused on the stability of the alkaline doped nanosheet. The calculations show that the switching process of greenhouse CO2 is varied by using different alkaline adatoms on the surface of g-B4N3 nanosheet. Furthermore, CO2 spontaneously releases electrons from the absorbent without any reaction barrier. Our results suggest that alkaline earth adatoms on g-B4N3 nanosheet are good adsorbents for CO2 and they can be used to switch CO2 from the atmosphere. Mg+2 adatom takes physisorption reaction (9.23 kcal/mol to 9.66 kcal/mol) while expecting all behave as chemisorption 14 reaction. The direct CO2 molecule has been repelled from the substrate of g-B4N3. The chemisorption reaction has been enhanced from -0.93 kcal/mol to -13.37 kcal/mol for Ba+2 to Ca+2 adatoms. Due to the lower adsorption values of CO2 to the substrate, CO2 molecule makes easy to discharge from the g-B4N3 nanosheet without any external source and shows instant switching mechanism. This work is interesting one for computational predictions, which might take place an experiment possible high-selectivity mechanism of CO2 switching on g-B4N3 material with the ideal reversible process. ACKNOWLEDGEMENTS Y.A.S & S. K. G. acknowledges the use of high performance computing clusters at IUAC, New Delhi and YUVA, PARAM II, Pune to obtain the partial results presented in this paper. 15 References [1] X.-T. Fan, M. Jia, M.-H. Lee, J. Cheng, J. Energy Chem. 26 (2017) 724–729. [2] H. Ran, J. Wang, A.A. Abdeltawab, X. Chen, G. Yu, Y. Yu, J. Energy Chem. 26 (2017) 909–918. [3] Q. Liu, Y. Shi, S. Zheng, L. Ning, Q. Ye, M. Tao, Y. He, J. Energy Chem. 23 (2014) 111–118. [4] K.C. Kemp, V. Chandra, M. Saleh, K.S. Kim, Nanotechnology 24 (2013) 235703. [5] H. Xu, S. Ouyang, L. Liu, D. Wang, T. Kako, J. Ye, Nanotechnology 25 (2014) 165402. [6] S. Chu, A. Majumdar, Nature 488 (2012) 294–303. [7] J. Shah, S. Kansara, S.K. Gupta, Y. Sonvane, Phys. Lett. A 381 (2017) 3084–3088. [8] S. Zeng, P. Kar, U.K. Thakur, K. Shankar, Nanotechnology 29 (2018) 052001. [9] C. Zhang, K.-W. Jun, G. Kwak, Y.-J. Lee, H.-G. Park, J. CO2 Util. 16 (2016) 1–7. [10] J. Liu, P.K. Thallapally, B.P. McGrail, D.R. Brown, J. Liu, Chem. Soc. Rev. 41 (2012) 2308–2322. [11] A.L. Dzubak, L.-C. Lin, J. Kim, J.A. Swisher, R. Poloni, S.N. Maximoff, B. Smit, L. Gagliardi, Nat. Chem. 4 (2012) 810–816. [12] N. Guo, K.M. Yam, X. Wang, C. Zhang, Nanotechnology 29 (2018) 105707. [13] M. Shan, Q. Xue, N. Jing, C. Ling, T. Zhang, Z. Yan, J. Zheng, Nanoscale 4 (2012) 5477–5482. [14] J. Schrier, ACS Appl. Mater. Interfaces 4 (2012) 3745–3752. [15] H. Liu, V.R. Cooper, S. Dai, D. Jiang, J. Phys. Chem. Lett. 3 (2012) 3343–3347. [16] J. Zhao, Y. Ding, J. Chem. Theory Comput. 5 (2009) 1099–1105. [17] A.K. Mishra, S. Ramaprabhu, J. Mater. Chem. 22 (2012) 3708–3712. [18] Y. Jiao, A. Du, Z. Zhu, V. Rudolph, S.C. Smith, J. Phys. Chem. C 114 (2010) 7846– 7849. [19] Y. Kubota, K. Watanabe, O. Tsuda, T. Taniguchi, Science 317 (2007) 932–934. [20] T. Ma, Q. Fan, H. Tao, Z. Han, M. Jia, Y. Gao, W. Ma, Z. Sun, Nanotechnology 28 (2017) 472001. [21] X. Li, X. Wu, X.C. Zeng, J. Yang, ACS Nano 6 (2012) 4104–4112. [22] L. Song, L. Ci, H. Lu, P.B. Sorokin, C. Jin, J. Ni, A.G. Kvashnin, D.G. Kvashnin, J. Lou, B.I. Yakobson, P.M. Ajayan, Nano Lett. 10 (2010) 3209–3215. [23] H. Choi, Y.C. Park, Y.-H. Kim, Y.S. Lee, J. Am. Chem. Soc. 133 (2011) 2084–2087. [24] J. Han, S. Lee, K. Choi, J. Kim, D. Ha, C.-G. Lee, B. An, S.-H. Lee, H. Mizuseki, J.-W. Choi, S. Kang, J. Hazard. Mater. 302 (2016) 375–385. [25] X. Wang, Y. Huang, Z. Pan, Y. Wang, C. Liu, J. Hazard. Mater. 295 (2015) 43–54. [26] S. Kansara, S.K. Gupta, Y. Sonvane, AIP Conf. Proc. 1961 (2018) 030052. [27] Y. Belmabkhout, R. Serna-Guerrero, A. Sayari, Chem. Eng. Sci. 64 (2009) 3721–3728. [28] P.J.E. Harlick, A. Sayari, Ind. Eng. Chem. Res. 46 (2007) 446–458. [29] L. Hamon, E. Jolimaître, G.D. Pirngruber, (2010). [30] X. Wang, J. Qiu, P. Ning, X. Ren, Z. Li, Z. Yin, W. Chen, W. Liu, J. Hazard. Mater. 229–230 (2012) 128–136. [31] E. Beňová, V. Zeleňák, D. Halamová, M. Almáši, V. Petrul'ová, M. Psotka, A. Zeleňáková, M. Bačkor, V. Hornebecq, J. Mater. Chem. B 5 (2017) 817–825. [32] Q. Lang, W. Hu, P. Zhou, T. Huang, S. Zhong, L. Yang, J. Chen, S. Bai, Nanotechnology 28 (2017) 484003. [33] M.I.M. Alzeer, K.J.D. MacKenzie, R.A. Keyzers, Microporous Mesoporous Mater. 241 (2017) 316–325. [34] H. Guo, W. Zhang, N. Lu, Z. Zhuo, X.C. Zeng, X. Wu, J. Yang, J. Phys. Chem. C 119 (2015) 6912–6917. 16 [35] Q. Sun, Z. Li, D.J. Searles, Y. Chen, G. (Max) Lu, A. Du, J. Am. Chem. Soc. 135 (2013) 8246–8253. [36] K. Watanabe, T. Taniguchi, H. Kanda, Nat. Mater. 3 (2004) 404–409. [37] Q. Sun, Q. Wang, P. Jena, Y. Kawazoe, J. Am. Chem. Soc. 127 (2005) 14582–14583. [38] R.B.N. Baig, S. Verma, M.N. Nadagouda, R.S. Varma, Sci. Rep. 6 (2016) 39387. [39] R.C. Dante, P. Martín-Ramos, A. Correa-Guimaraes, J. Martín-Gil, Mater. Chem. Phys. 130 (2011) 1094–1102. [40] M. Zarei, H. Ahmadzadeh, E.K. Goharshadi, A. Farzaneh, Anal. Chim. Acta 887 (2015) 245–252. [41] P. Praus, L. Svoboda, M. Ritz, I. Troppová, M. Šihor, K. Kočí, Mater. Chem. Phys. 193 (2017) 438–446. [42] L. Wang, C. Wang, X. Hu, H. Xue, H. Pang, Chem. – Asian J. 11 (2016) 3305–3328. [43] J. Zhang, Y. Chen, X. Wang, Energy Environ. Sci. 8 (2015) 3092–3108. [44] X. Tan, L. Kou, S.C. Smith, ChemSusChem 8 (2015) 2987–2993. [45] X. Tan, L. Kou, H.A. Tahini, S.C. Smith, Sci. Rep. 5 (2015) 17636. [46] P. Giannozzi, S. Baroni, N. Bonini, M. Calandra, R. Car, C. Cavazzoni, Davide Ceresoli, G.L. Chiarotti, M. Cococcioni, I. Dabo, A.D. Corso, S. de Gironcoli, S. Fabris, G. Fratesi, R. Gebauer, U. Gerstmann, C. Gougoussis, Anton Kokalj, M. Lazzeri, L. Martin-Samos, N. Marzari, F. Mauri, R. Mazzarello, Stefano Paolini, A. Pasquarello, L. Paulatto, C. Sbraccia, S. Scandolo, G. Sclauzero, A.P. Seitsonen, A. Smogunov, P. Umari, R.M. Wentzcovitch, J. Phys. Condens. Matter 21 (2009) 395502. [47] J.P. Perdew, K. Burke, M. Ernzerhof, Phys. Rev. Lett. 77 (1996) 3865–3868. [48] J. Heyd, G.E. Scuseria, M. Ernzerhof, J. Chem. Phys. 118 (2003) 8207–8215. [49] V. Barone, M. Casarin, D. Forrer, M. Pavone, M. Sambi, A. Vittadini, J. Comput. Chem. 30 (2009) 934–939. [50] S. Grimme, J. Comput. Chem. 27 (2006) 1787–1799. [51] H.J. Monkhorst, J.D. Pack, Phys. Rev. B 13 (1976) 5188–5192. [52] A. Faghaninia, J.W. Ager, C.S. Lo, Phys. Rev. B 91 (2015) 235123. [53] W. Gannett, W. Regan, K. Watanabe, T. Taniguchi, M.F. Crommie, A. Zettl, Appl. Phys. Lett. 98 (2011) 242105. [54] A.N. Andriotis, E. Richter, M. Menon, Phys. Rev. B 93 (2016) 081413. [55] J. Zhu, J. Zhang, S. Xu, Y. Hao, Appl. Surf. Sci. 402 (2017) 175–181. [56] S.K. Gupta, H. He, D. Banyai, M. Si, R. Pandey, S.P. Karna, Nanoscale 6 (2014) 5526– 5531. [57] E. Zaremba, W. Kohn, Phys. Rev. B 15 (1977) 1769–1781. [58] M. Zhang, Y. Yu, Y. Zhang, Appl. Surf. Sci. 280 (2013) 15–24. [59] G. Qin, A. Du, Q. Sun, Energy Technol. 6 (2018) 205–212. 17 Electronic Supplementary Information (ESI) Electronic properties: The calculated electronic properties for the sheet of g-B4N3 using HSE06 functional has been described in Fig. (S1). Figure S1. The calculated electronic band curve of (a) sheet of g-B4N3. The red dashed line represents the Fermi level. The blue and black lines in (a) show the conduction and valence band, respectively. (b) Density of states of a sheet of g-B4N3. The red dotted line represents the Fermi level. 18 Adsorption Energy: Table S1. The adsorption energy of adatoms is as Mg+2, Ca+2, Sr+2 and Ba+2 with the substrate of g-B4N3. Adatoms Adsorption Energy (Kcal/mol) Mg Ca Sr Ba with g-B4N3 -111.613 -140.439 -144.128 -135.596 19 Table S2. Represent the formation energy (eV) and comparison of adsorption energy of the C atom and the CO2 molecule with the substrate of g-B4N3 along different adatoms (Mg, Ca, Sr, Ba). Systems / Formation energy C- Atom CO2- Molecule Adsorption Sites ( ) (eV) (Kcal/mol) (Kcal/mol) g-B4N3+Mg+CO2 g-B4N3+Ca+CO2 g-B4N3+Sr+CO2 g-B4N3+Ba+CO2 -38.46 -35.53 -40.47 -40.46 -29.37 -35.28 -35.97 -33.89 9.23 -13.37 -0.99 -1.62 20 fE
1706.06313	1	1706	2017-06-20T08:34:22	Imaging of optically active defects with nanometer resolution	[ "cond-mat.mes-hall" ]	Point defects significantly influence the optical and electrical properties of solid-state materials due to their interactions with charge carriers, which reduce the band-to-band optical transition energy. There has been a demand for developing direct optical imaging methods that would allow in-situ characterization of individual defects with nanometer resolution. Here, we demonstrate the localization and quantitative counting of individual optically active defects in monolayer hexagonal boron nitride using single molecule localization microscopy. By exploiting the blinking behavior of defect emitters to temporally isolate multiple emitters within one diffraction limited region, we could resolve two defect emitters with a point-to-point distance down to ten nanometers. The results and conclusion presented in this work add unprecedented dimensions towards future applications of defects in quantum information processing and biological imaging.	cond-mat.mes-hall	cond-mat	Imaging of optically active defects with nanometer resolution Jiandong Feng1, Hendrik Deschout1, Sabina Caneva2, Stephan Hofmann2, Ivor Lončarić3, Predrag Lazić3 and Aleksandra Radenovic1 1Laboratory of Nanoscale Biology, Institute of Bioengineering, School of Engineering, EPFL, 2Department of Engineering, University of Cambridge, JJ Thomson Avenue, CB3 0FA 1015 Lausanne, Switzerland Cambridge, United Kingdom 3Institut Ruđer Bošković, Bijenička 54, 10000 Zagreb, Croatia Correspondence should be addressed [email protected] and [email protected] Abstract Point defects significantly influence the optical and electrical properties of solid-state materials due to their interactions with charge carriers, which reduce the band-to-band optical transition energy. There has been a demand for developing direct optical imaging methods that would allow in-situ characterization of individual defects with nanometer resolution. Here, we demonstrate the localization and quantitative counting of individual optically active defects in monolayer hexagonal boron nitride using single molecule localization microscopy. By exploiting the blinking behavior of defect emitters to temporally isolate multiple emitters within one diffraction limited region, we could resolve two defect emitters with a point-to-point distance down to ten nanometers. The results and conclusion presented in this work add unprecedented dimensions towards future applications of defects in quantum information processing and biological imaging. . INTRODUCTION Vacancy defects are of crucial importance in determining both transport and optical properties of semiconductors. Through interaction with excitons, defects can be optically active at energies lower than the band-to-band optical transition energy, as revealed in their photoluminescence experiments(1). Defects in diamond(2) and two-dimensional materials(3) have been demonstrated to be able to serve as single photon emitters, which are essential for quantum-information processing(4). Due to their extraordinary quantum mechanical properties in sensing the local electromagnetic field and temperature, defects are also considered as promising candidates for biological imaging at the nanoscale(5). Mapping and localizing individual defects at high spatial resolution will, therefore, add unprecedented dimensions to both applications. Scanning tunneling microscope (STM) has long been used to directly visualize single defects and their electronic structures. With STM, manipulation of individual point defects can be realized via tip-bias-induced electrochemical reactions(6). Advance in aberration- corrected transmission electron microscope (TEM) greatly improved the resolution in characterizing individual defects(7, 8). Nevertheless, both methods require special sample preparations and are still not optimal for fast in-situ operations. Therefore, development of direct optical technology that images individual defects would match the emerging requirements in both quantum information processing and bioimaging. However, the optical detection and control are hampered by the diffraction limit, which can accumulate multiple defect emitters for defective materials. Super-resolution microscopy techniques can in principle address these problems. Stimulated emission depletion (STED) microscopy has demonstrated a capability of imaging nitrogen–vacancy (N-V) centers in nanodiamond with a resolution of a few nanometers(9, 10). Single-molecule localization microscopy (SMLM), based on localizing sparse sets of switchable fluorescent molecules(11, 12), has also been used for wide-field parallel imaging of N-V centers in diamond(13, 14). Recently, 2D materials have been shown to host defects that serve as single photon emitters at cryogenic temperatures for transition metal dichalcogenides(15-18) and room temperature for hexagonal boron nitride (h-BN)(19). In quantum-information processing, optically active defects in 2D materials are attractive since inherently they don't require total internal reflection. In addition, optically active defects have high light extraction efficiency, and electrical connections can be easily integrated(3). Therefore, imaging them with super-resolution techniques is of high significance for the realization of reading large- scale networks. The operation at room temperature, together with the high brightness and contrast and defective nature of h-BN makes it the most suitable 2D candidate for room- temperature quantum computing(3). Here we demonstrate the localization and spatial mapping of individual defect emitters in monolayer h-BN by SMLM. This approach allows us to resolve the most precisely localized defect emitters at separations around 10 nm. Monolayer h-BN comprises alternate boron and nitrogen atoms in a two-dimensional honeycomb arrangement. In order to clarify the structure of atomic defects in h-BN, we first imaged the h-BN samples by aberration-corrected TEM, as shown in Fig. 1A. The uniform orientation of all observed defects excludes the possibility of coexistence between boron and nitrogen monovacancies due to the threefold symmetry of h-BN(8). Boron monovacancies, further confirmed by the line profile of the image contrast (Fig. 1B), is found to be dominant for the h-BN materials(8). Because of the electron knock-on effect during TEM irradiation coupled with the beam-induced etching with residual water or oxygen present in the system, more defects can be directly introduced and enlarged to triangular pores(20), as shown in Fig. 1C. Photoluminescence experiments on h-BN specimens that were not subjected to electron radiation TEM suggest that the boron monovacancies are intrinsic (Fig. 1D), which may form during the growth but as well during the electrochemical transfer process(21). Two kinds of emissions are observed in our experiments: the weak homogeneous fluorescence that follows the entire triangular shaped h-BN monolayer(21) with an emission peak centered at 580 nm and the ultra-bright individual emitters with emission peak centered at 620 nm (Fig. 1F). As defect-free monolayer h-BN has a large bandgap (around 5-6 eV), we speculate that two candidates for the observed emission could be either boron or nitrogen monovacancies. Due to the uncertainty in calculating their electronic structure using the density functional calculations(22, 23), both are reasonable candidates for the observed spectrum. Possible causes for this uncertainty are discussed in supporting information. However, according to the TEM results (Fig. 1a) and considering the relative density, in monolayer h-BN we have detected mostly boron vacancies that may be responsible for an observed optical activity. We infer that the observed weak homogeneous emission is due to the high density of boron monovacancies with neutral charges VB0 in h-BN, and the ultra-bright emitters are negatively charged boron monovacancies VB−, as proposed in Fig. 1E. As shown in the emission spectrum Fig.1F, fluorescence emission above 600 nm is mainly coming from VB−. We could not avoid the averaging effect in our photoluminescence experiments due to difficulties in isolating individual defects. Better spectroscopic results could be achieved with isolated individual emitters under low-temperature conditions(23). Other candidates for the observed defect fluorescence can be boron-nitrogen vacancy(19) or three boron plus one nitrogen vacancy. The unique and uniform defect structure of h-BN (Fig. 1a) is crucial for the scaling up of a parallel quantum sensor array via selective defect patterning. Photoswitching was observed after exposure to green light illumination (561 nm). One of the possible explanations for this switching behavior can refer to photo-induced ionization and recombination of VB between its neutral state and negative state(24). We then verified the wavelength dependence (405 nm, 488 nm, and 561 nm) of defect charge state dynamics and found maximum switching rates at 561 nm. No switching event was observed at 405 nm. The charge conversion process can be understood as VB0 / VB-1 is first excited and then an electron is captured from conduction band for ionization (VB0 to VB-1) or from valence band for recombination (VB-1 to VB0), consistent with the evidence reported for both optical excitation of defect transitions in h-BN(25) and STM measurements on the excited h-BN defects(26). This photoswitching behavior sets the playground for optical super-resolution reconstruction, which is based on the temporal isolation of densely packed multiple emitters within one diffraction limited region. SMLM trades temporal resolution for spatial resolution, and therefore it requires an ideal imaging sample to be spatially fixed and temporally static(11). Both requirements are fulfilled by the solid-state defective h-BN samples. In addition, in this case, no labeling is involved, and the strong emitters are extremely bright and do not bleach. Direct imaging of the sample itself makes it an ideal test case for achieving the fundamental limit of SMLM resolutions(11). Initial image frames typically consisted of dense weak fluorescence background presumably dominated by the much larger population of boron monovacancies still in the inactivated state VB0. After initial activation, a lot of asynchronous blinking events between two states were observed. Due to the brightness of solid-state emitters, we achieved a high localization precision (11) from strong emitters. For example, a localization precision of 3.5 nm is obtained for the emitters shown in Fig. 2C. Giving an average localization precision is challenging due to the merged broad distribution of weak and bright emitters. Reconstruction by summing up all temporally isolated Gaussian location distributions of individual emitters defined by their coordinates and localization precision yields Fig. 2 A and B. In contrast, a diffraction limited image is produced by summing all image stacks to overlap their signals (SI Fig. 2). The ability to distinguish two closest emitters also imposes stringent requirements on stage drift correction. Precise drift correction can be achieved with an active feedback- controlled system(27, 28) or post-imaging processing method using fiducial markers(11). We used a large number of fiducial markers to achieve an accurate drift correction. In a representative experiment, trajectories of 27 fiducial markers are used to calculate the averaged drift trajectory and remaining drift error after correction. To demonstrate the imaging capability of solid-state SMLM for individual defects, we first employed Fourier ring correlation to estimate the resolution(29), and we obtained 46 nm (SI Fig. 3). However, in this case, individual emitters, as well as their distinguishability are more relevant instead of revealing the structure itself. As shown in Fig. 2C, high localization precision, and accurate drift correction allowed distinguishing two neighboring emitters with a separation of 10.7 nm. We would like to note that the emitters exhibit a broad distribution of their photon counts (SI Fig. 4), thus this resolving capability is dictated by the photophysics and density of strong emitters. Direct optical imaging also allows for in-situ characterization of defects under ambient conditions, for example, probing their chemistry and dynamics in different pH environments (SI Fig.5), surface depositions (SI Fig. 6) and isotopic solvents (SI Fig.7). Among these conditions, the fluorescence signals were found to be sensitive to acidity of the environment. Compared to TEM and STM, for in-situ operation under ambient conditions, SMLM introduces minimum destruction to the sample. The same argument also holds for the high power density used for the depletion laser beam in STED(9). In addition, standard SMLM has a relatively large field of view, adequate for the imaging of single-crystal 2D material grains. Recently, novel SMLM modalities (30, 31) have extended the field of view considerably. The spatial resolution of our results can still be improved using better drift correction methods(28) while a better temporal resolution can be achieved by a spatio- temporal cumulant analysis of the image sequence in super-resolution optical fluctuation imaging (SOFI)(32), shown in SI Fig. 8. SOFI would be particularly interesting for revealing the dynamical processes on defects at the relevant time scales. Although we are imaging the structure itself without labeling, solid-state SMLM can also provide additional information for revealing chemical contrast of defect types owning to the imaging capability with multiple color channels, as shown in the composited Fig. 3A. The dual color imaging led to a preferred spatial distribution of strong emitters near the edge under red channel (Fig. 3B and D, laser 561 nm) while no spatial preference was found for green channel (Fig. 3C and E, laser 488 nm). This difference could result either from the difference of charge transition rates or defect number and types (see the discussion of possible defect types in supporting information), which was natively formed during growth or transfer(21). A further determination would require the combination of SMLM with spectroscopic approaches(33) like photoluminescence, deep-level transient spectroscopy, or electron energy-loss spectroscopy. We foresee that such correlative techniques will provide further information on the electronic structures of individual defects. In addition to chemical contrast, SMLM also offers quantitative information on the number of defects, which can be estimated based on their localizations (Fig. 4A and Fig. 4B-E). We also merged the localizations accounting for the localization error(34). The counting results from strict correction and total localizations offered both the bottom and upper limit of the defect density (Fig. 4F-I). Further improvement on this estimation can be made if the photophysics of defect emitters would be clearer. The obtained defect density (3268 μm-2 -7246 μm-2 for dense region Fig. 4B and 514 μm-2 - 606 μm-2 for sparse region Fig. 4C) of h-BN is comparable to previous results in ion transport(35) while quantitative SMLM avoids the averaging effect in ion transport measurements. We also used balanced SOFI (bSOFI) to obtain a density map of the defects as bSOFI (SI Fig. 8) was demonstrated to be more accurate than PALM for high-density emitters(36). This method provides a quantitative single molecule approach for estimating defect density in semiconductors on large areas. A further interest to look at the quantum emitters is their spin properties, which can be optically read out when subject to electromagnetic fields, as shown in the measurements with nuclear quadrupole resonance spectroscopy(37). On the other hand, our method also provides a new approach to probe the properties of 2D materials since defects have been proven to perform a crucial role in determining their electrical transport and optical properties. To conclude, we have demonstrated the solid-state SMLM technique for imaging and counting quantum emitters in h-BN with the capability of distinguishing two closest emitters at a few nanometers. The achieved high localization precision may find future applications in semiconductor thin film characterization, quantum information processing, and biological imaging. Author Contributions J.F conceived the idea and interpreted the results. J.F wrote the manuscript with inputs from A.R. J.F, H.D, and A.R designed experiments. J.F and H.D performed experiments. H.D processed all images. S.C and S.H performed materials growth. I.L. and P.L performed DFT calculations. All authors agreed with the final version of the manuscript. The authors declare no competing financial interest. Acknowledgment This work was financially supported by Swiss National Science Foundation SNSF (200021 153653). We thank the Centre Interdisciplinaire de Microscopie Electronique (CIME) at the École Polytechnique fédérale de Lausanne (EPFL) for access to electron microscopes, Ke Liu, Davide Deiana, and Duncan T. L. Alexander for TEM imaging. Device fabrication was partially carried out at the EPFL Center for Micro/Nanotechnology (CMi). Special thanks to Martina Lihter for hBN transfer and Tomas Lukes for bSOFI image analysis which results in SI Fig. 8. The work performed in Cambridge was supported by the EPSRC Cambridge NanoDTC, EP/L015978/1. I. L. and P. L. were supported by the Unity Through Knowledge Fund, Contract No. 22/15 and H2020 CSA Twinning Project No. 692194, RBI-T-WINNING. References 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. in S. Tongay et al., Defects activated photoluminescence two-dimensional semiconductors: interplay between bound, charged, and free excitons. Scientific reports 3, 2657 (2013). C. Kurtsiefer, S. Mayer, P. Zarda, H. Weinfurter, Stable solid-state source of single photons. Physical review letters 85, 290 (2000). I. Aharonovich, D. Englund, M. Toth, Solid-state single-photon emitters. Nature Photonics 10, 631 (2016). J.-W. Pan et al., Multiphoton entanglement and interferometry. Reviews of Modern Physics 84, 777 (2012). R. Schirhagl, K. Chang, M. Loretz, C. L. Degen, Nitrogen-vacancy centers in diamond: nanoscale sensors for physics and biology. Annual review of physical chemistry 65, 83 (2014). D. Wong et al., Characterization and manipulation of individual defects in insulating hexagonal boron nitride using scanning tunnelling microscopy. Nature nanotechnology 10, 949 (2015). A. Hashimoto, K. Suenaga, A. Gloter, K. Urita, S. Iijima, Direct evidence for atomic defects in graphene layers. Nature 430, 870 (2004). C. Jin, F. Lin, K. Suenaga, S. Iijima, Fabrication of a freestanding boron nitride single layer and its defect assignments. Physical review letters 102, 195505 (2009). E. Rittweger, K. Y. Han, S. E. Irvine, C. Eggeling, S. W. Hell, STED microscopy reveals crystal colour centres with nanometric resolution. Nature Photonics 3, 144 (2009). S. Arroyo-Camejo et al., Stimulated emission depletion microscopy resolves individual nitrogen vacancy centers in diamond nanocrystals. ACS nano 7, 10912 (2013). E. Betzig et al., Imaging intracellular fluorescent proteins at nanometer resolution. Science 313, 1642 (2006). 12. M. J. Rust, M. Bates, X. Zhuang, Sub-diffraction-limit imaging by stochastic optical reconstruction microscopy (STORM). Nature methods 3, 793 (2006). 13. M. Gu, Y. Cao, S. Castelletto, B. Kouskousis, X. Li, Super-resolving single nitrogen vacancy centers within single nanodiamonds using a localization microscope. Optics express 21, 17639 (2013). 14. M. Pfender, N. Aslam, G. Waldherr, P. Neumann, J. Wrachtrup, Single-spin stochastic optical reconstruction microscopy. Proceedings of the National Academy of Sciences 111, 14669 (2014). Y.-M. He et al., Single quantum emitters in monolayer semiconductors. Nature nanotechnology 10, 497 (2015). C. Chakraborty, L. Kinnischtzke, K. M. Goodfellow, R. Beams, A. N. Vamivakas, Voltage-controlled quantum light from an atomically thin semiconductor. Nature nanotechnology 10, 507 (2015). 17. M. Koperski et al., Single photon emitters in exfoliated WSe2 structures. Nature 15. 16. 18. 19. nanotechnology 10, 503 (2015). A. Srivastava et al., Optically active quantum dots in monolayer WSe2. Nature nanotechnology 10, 491 (2015). T. T. Tran, K. Bray, M. J. Ford, M. Toth, I. Aharonovich, Quantum emission from hexagonal boron nitride monolayers. Nature nanotechnology 11, 37 (2016). 20. 21. 22. 23. 24. 25. 26. 27. 28. 29. 30. 31. 32. 33. 34. J. C. Meyer, A. Chuvilin, G. Algara-Siller, J. Biskupek, U. Kaiser, Selective sputtering and atomic resolution imaging of atomically thin boron nitride membranes. Nano letters 9, 2683 (2009). S. Caneva et al., Nucleation control for large, single crystalline domains of monolayer hexagonal boron nitride via Si-doped Fe catalysts. Nano letters 15, 1867 (2015). C. Attaccalite, M. Bockstedte, A. Marini, A. Rubio, L. Wirtz, Coupling of excitons and defect states in boron-nitride nanostructures. Physical Review B 83, 144115 (2011). X. Li et al., Non-Magnetic Quantum Emitters in Boron Nitride with Ultra-Narrow and Sideband-Free Emission Spectra. ACS nano, (2017). N. Aslam, G. Waldherr, P. Neumann, F. Jelezko, J. Wrachtrup, Photo-induced ionization dynamics of the nitrogen vacancy defect in diamond investigated by single- shot charge state detection. New Journal of Physics 15, 013064 (2013). L. Ju et al., Photoinduced doping in heterostructures of graphene and boron nitride. Nature nanotechnology 9, 348 (2014). J. Velasco Jr et al., Nanoscale control of rewriteable doping patterns in pristine graphene/boron nitride heterostructures. Nano letters 16, 1620 (2016). A. R. Carter et al., Stabilization of an optical microscope to 0.1 nm in three dimensions. Applied Optics 46, 421 (2007). A. Pertsinidis, Y. Zhang, S. Chu, Subnanometre single-molecule localization, registration and distance measurements. Nature 466, 647 (2010). R. P. Nieuwenhuizen et al., Measuring image resolution in optical nanoscopy. Nature methods 10, 557 (2013). R. Diekmann et al., Chip-based wide field-of-view nanoscopy. Nature Photonics 11, 322 (May, 2017). K. M. Douglass, C. Sieben, A. Archetti, A. Lambert, S. Manley, Super-resolution imaging of multiple cells by optimized flat-field epi-illumination. Nature Photonics 10, 705 (Nov, 2016). T. Dertinger, R. Colyer, G. Iyer, S. Weiss, J. Enderlein, Fast, background-free, 3D super-resolution optical fluctuation imaging (SOFI). Proceedings of the National Academy of Sciences 106, 22287 (2009). Z. Zhang, S. J. Kenny, M. Hauser, W. Li, K. Xu, Ultrahigh-throughput single-molecule spectroscopy and spectrally resolved super-resolution microscopy. Nature methods 12, 935 (2015). P. Annibale, S. Vanni, M. Scarselli, U. Rothlisberger, A. Radenovic, Quantitative photo activated localization microscopy: unraveling the effects of photoblinking. PloS one 6, e22678 (2011). 35. M. I. Walker et al., Extrinsic Cation Selectivity of 2D Membranes. ACS nano, (2017). S. Geissbuehler et al., Mapping molecular statistics with balanced super-resolution 36. optical fluctuation imaging (bSOFI). Optical Nanoscopy 1, 4 (2012). I. Lovchinsky et al., Magnetic resonance spectroscopy of an atomically thin material using a single-spin qubit. Science 355, 503 (2017). 37. Figures Figure 1. Optically active defects in h-BN. A, High-resolution TEM imaging of vacancy defects in h-BN. Dominant defects are found to be point defects. B, Image contrast of the line in A suggests the defect type to be boron monovacancy. C, High-resolution TEM imaging of h-BN edges. D, Schematics of in-situ characterization of defects in h-BN using SMLM (More details are shown in SI methods). E, Proposed structure of boron monovacancy with neutral and negative charge. F, Emission spectrum of h-BN defects. Figure 2. Super-resolution imaging of optically active defects in h-BN. A, Reconstructed SMLM image of defects in h-BN from an image sequence of 20, 000 frames. The imaging condition: DI water and 561 nm laser. Display pixel size: 5 nm. B, Selected region in A. Display pixel size: 1 nm. C, With best drift correction achieved with fiducial markers shown in SI Fig. 3, SMLM allows distinguishing two close emitters. D, Distance profile shown in C shows a resolving capability of 10.7 nm. The images were rendered as probability maps. Figure 3. Dual color imaging reveals chemical contrast. A, Composited dual color super- resolution image of quantum emitters in h-BN from image sequence of 20, 000 frames. The imaging condition: DI water. Green channel: 488 nm laser. Red channel: 561 nm laser. B, C, D, E, Selected regions in A. The images were rendered as probability maps. Display pixel size: 5 nm. Figure 4. Counting defects in h-BN using quantitative SMLM. A, Super-resolution imaging and mapping of defect emitters in h-BN with color marked region of interest (ROI). B, C, D, E, localizations in color marked ROI in A. ROI size: 600 nm × 600 nm. F, G, H, I, Estimation of defect number in B, C, D, E as a function of image frame numbers using total number of localizations(original) and blink correction, respectively. The obtained density in B: 3268 μm-2 (blink corrected) to 7246 μm-2 (original). C: 514 μm-2 (blink corrected) to 606 μm-2 (original). D: 2849 μm-2 (blink corrected) to 5208 μm-2 (original). E: 2252 μm-2 (blink corrected) to 3247 μm-2 (original). Imaging condition: DI water and 561 nm laser. Imaging optically active defects with nanometer resolution Jiandong Feng1, Hendrik Deschout1, Sabina Caneva2, Stephan Hofmann2, Ivor Lončarić3, Predrag Lazić3 and Aleksandra Radenovic1 1Laboratory of Nanoscale Biology, Institute of Bioengineering, School of Engineering, EPFL, 1015 Lausanne, Switzerland 2Department of Engineering, University of Cambridge, JJ Thomson Avenue, CB3 0FA Cambridge, United Kingdom 3Institut Ruđer Bošković, Bijenička 54, 10000 Zagreb, Croatia Correspondence should be addressed [email protected] [email protected] Supporting Information SI Text 1. Sample preparation ................................................................................................................ 2 CVD growth ................................................................................................................. 2 1.1 Transfer ...................................................................................................................... 2 1.2 2 Imaging conditions .......................................................................................................... 2 2.1 Microscope setup ........................................................................................................ 2 2.2 Imaging ....................................................................................................................... 2 Localization procedure ................................................................................................ 3 2.3 Counting ..................................................................................................................... 3 2.4 3. Defect electronic band diagrams ............................................................................................ 4 4. References ............................................................................................................................. 9 5. SI Figure Captions .................................................................................................................. 10 1 SI Text 1. Sample preparation 1.1 CVD growth The h-BN samples are grown under same conditions described elsewhere (1). Briefly, as-received Fe foil (100 μm thick, Alfa Aesar, 99.99% purity) is loaded in a customized CVD reactor (base pressure 1x10-6 mbar). The foils are exposed to 4 mbar of NH3 during heating to 900 °C. Subsequently, the NH3 is removed and borazine (HBNH)3 is introduced into the chamber through a leak valve from a liquid reservoir. After growth (45-90 s borazine exposure at 6x10-4 mbar and 900 °C) the heater is switched off and the foils are cooled in a vacuum. 1.2 Transfer The h-BN domains are transferred onto SiNx membranes using the electrochemical bubbling method (2). A polymer support layer is deposited onto the h-BN/Fe by spin coating poly(methyl methacrylate) (PMMA) at 5000 rpm for 40 s. The samples are placed in a NaOH bath (1M) and are contacted with a Pt wire that acts as the cathode, while another Pt wire (anode) is immersed in the electrolyte. During electrolysis, H2 bubbles are generated at the h-BN/Fe interface, lifting the h-BN/PMMA film from the substrate. The film is subsequently rinsed in DI water and scooped with the target substrate, where it is allowed to dry. The PMMA is removed by immersing the sample in acetone overnight, followed by a rinse in IPA. 2. Imaging conditions 2.1 Microscope setup Imaging was carried out on a custom-built microscope that was described previously (3, 4). Briefly, two laser sources were used for excitation: a 100mW 488 nm laser beam (Sapphire, Coherent) and a 100mW 561 nm laser beam (Excelsior, Spectra Physics). The laser beams were combined using a dichroic mirror (T495lpxr, Chroma) and sent through an acousto-optic tunable filter (AOTFnC-VIS-TN, AA Opto- Electronic). Both laser beams were focused into the back focal plane of an objective (UApo N ×100, Olympus) with a numerical aperture of 1.49 mounted on an inverted optical microscope (IX71, Olympus). Excitation and fluorescence light were separated by a filter cube containing a dichroic mirror (493/574 nm BrightLine, Semrock) and an emission filter (405/488/568 nm StopLine, Semrock). The fluorescence light was detected by an EMCCD camera (iXon DU-897, Andor) with a back-projected pixel size of 105 nm. An optical system (DV2, Photometrics) equipped with a dichroic mirror (T565lpxr, Chroma) was used to split the fluorescence light into a green and red color channel that were each sent to a separate half of the camera chip. 2.2 Imaging Before imaging, 100 nm gold nanospheres (C-AU-0.100, Corpuscular) were added to the chip for lateral drift monitoring. The addition of fiducial markers was done by incubating the chip at least 15 minutes with a droplet of the gold nanospheres diluted up to 10 times in water. After removing the droplet, the 2 chip was baked at 120 °C for at least 10 minutes and rinsed with water. The chip was placed upside down on a 25 mm diameter round cover slip (#1.5 Micro Coverglass, Electron Microscopy Sciences) that was cleaned with an oxygen plasma for 5 minutes. Imaging was performed in water at room temperature. Excitation in the red channel was done at 561 nm with ~20 to ~30 mW power (as measured in the back focal plane of the objective). Excitation in the green channel was done at 488 nm with ~10 mW power. The gain of the EMCCD camera was set at 100 and the exposure time to 50 ms. For each experiment, at least 20,000 camera frames were recorded. Dual-colour imaging was performed by first acquiring at least 10,000 camera frames in the red channel, and subsequently recording at least 10,000 camera frames in the green channel. Gold nanospheres were imaged to co-register the two color channels a posteriori. Axial drift correction was ensured by a nanometer positioning stage (Nano-Drive, Mad City Labs) driven by an optical feedback system (5) 2.3 Localization procedure The camera frames were analyzed by a custom written algorithm (Matlab, The Mathworks) that was adapted from an algorithm that was described elsewhere (3, 4). First, a Gaussian filter and subsequently the Gaussian curvature filter (5) was applied to each frame separately. This yields an image of the uneven background consisting of triangular structures that correspond to fluorescence from the bulk hBN. This background image was subtracted from the original frame (after applying the same Gaussian filter), resulting in an image of the emitters on a uniform background. Only peaks with an intensity of at least 4 times the background were considered to be emitters. These peaks were fitted by maximum likelihood estimation of a 2D Gaussian point spread function (PSF) (6) and the localization precision was obtained from the Cramér-Rao lower bound (7). Drift was corrected in each frame by subtracting the average position of the gold nanospheres from the emitter positions in that frame. Co-registration of the two color channels was done using a second order polynomial transformation that was derived from the localizations of the gold nanospheres visible in both color channels. The SMLM images were generated as a probability map by plotting a 2D Gaussian PSF centered on each fitted position with a standard deviation equal to the corresponding localization precision. Only positions with a localization precision below 30 nm were plotted. 2.4 Counting Determining the number of emitters by simply counting the localizations yields an overestimation in case there is "blinking," i.e. the same emitter reappears once or several times. We corrected this possible error by borrowing a method that was developed in the context of counting photo-activatable fluorescent proteins(8). The idea is that two different localizations are considered to originate from the same blinking event if they are close enough in space and time. Merging these localizations based on a suitable spatial and temporal threshold, therefore, results in a corrected number of localizations. To account for the localization precision, the spatial threshold was calculated from the Hellinger distance, which was taken equal to 0.9 (4). The temporal threshold was varied between multiples of the camera exposure time, i.e. the first five multiples (8), yielding five different corrected localization numbers. A semi-empirical model was fitted to these values as a function of the temporal threshold, resulting in a final corrected localization number (8). The model assumes that the emitters go from an off-state to an on-state, and subsequently they either reversibly go to a dark state or irreversibly to a photobleached state. 3 3. Defect electronic band diagrams Electronic structure of vacancies in monolayer h-BN has previously been studied by Attaccalite et al.(9), Li et al.(10), and Tran et al.(11). Attaccalite et al. used both simple and rather inaccurate local density approximation (LDA) to density functional theory (DFT) and complex but usually very accurate the so-called GW/BSE theory. Li et al. and Tran et al. used generalized gradient approximation (GGA) to DFT, which in this case gives similar results to LDA. From these studies at LDA or GGA level, it is known that several type of defects have energy levels that can result in optically active transitions with energies of transition of around 2 eV that are observed in our and previous experiments. This is clearly shown in Fig. 4 of Ref. (11) for N monovacancy and B vacancy with the N atom shifted over into the empty site (NBVN), and also in Fig. 6 of Ref. (10) for B monovacancy, N monovacancy, divacancy and NBVN. However, it is well known that LDA and GGA have a tendency to underestimate band gaps of insulators, and therefore their predictions of optical properties of insulators are rather inaccurate(12). Additionally, DFT neglects excitonic effects. Band gap problem of LDA/GGA can be corrected by the so-called GW method, and excitonic effects can be included by solving the Bethe-Salpeter equation (BSE). Attaccalite et al. performed such calculation for B monovacancy, N monovacancy, and divacancy. Not taking into account excitonic effects, the lowest optically relevant transitions in GW for B monovacancy was obtained at 3.5 eV and for N monovacancy at 4 eV. When excitonic effects were included (GW/BSE), Attaccalite et al. predict that the lowest optical peak for B monovacancy should be at around 3.3 eV and for N monovacancy at 2.7 eV. The GW/BSE theory is probably the best method for solid state optical spectroscopy simulations, and therefore it is curious that it seems that this theory does not predict any defect related optical peak that would be in agreement with experiments. One could conclude that either neutral monovacancies are not responsible for peaks around 2 eV, or GW/BSE fails significantly for h-BN defects, or the theoretically studied structure of the system does not represent well the experimental conditions. To enlarge and deepen the search for defects observed in experiments we perform calculations for some of the observed defects in TEM images. Instead of using computationally very demanding GW/BSE, we use hybrid DFT functional that is known to largely correct the band gap problem of LDA/GGA(12). Our TEM images show the uniform orientation of all defects. The most frequent defect is B monovacancy, but several larger vacancies are also visible. Due to these experimental facts, we have performed DFT calculations of B monovacancy and two larger vacancies with the same orientation, 3B+N and 6B+3N. Structures of these defects are shown in Fig. S9. 4 Fig. S9. PBE relaxed structure for a) B monovacancy, b) 3B+N vacancy, and c) 6B+3N vacancy. All our calculations are performed using the plane-wave basis set VASP code(13). We use 10 × 10 BN unit cell to avoid interaction between defects and Γ point sampling of reciprocal space. For each defect, we first use commonly employed PBE exchange-correlation functional(14) to relax structures and then we perform hybrid DFT calculation using the so-called HSE06 functional(15, 16). As noted above, it is well-known that ordinary PBE functional usually underestimates band gaps and that it is prone to delocalization errors. As we want to simulate optical properties of defect levels in monolayer BN, both of these errors can severely limit the applicability of PBE to simulate this system. Hybrid functionals that include a portion of exact exchange usually give much more accurate band gaps and delocalization errors are smaller. Our DFT results are shown as band diagrams in Fig. S10 for B monovacancy, Fig. S11 for 3B+N vacancy, and Fig. S12 for 6B+3N vacancy. Additionally, as we connect photo switching of strong emitters, observed in experiments, with charged defects, in Fig. S10 we also show the calculation for charged B monovacancy. Our PBE results are consistent with previous studies(9-11). We observe large differences between PBE and HSE06 results for all three vacancies. Differences in band gaps are clearly visible both for bulk bands (bulk band gap is around 4.7 eV in PBE calculation and around 6.2 eV in HSE06 calculation) and defect levels. Moreover, there are some vacancy related levels inside the band gap that appear in the HSE06 calculations and not in the PBE calculations and vice versa. Due to the reasons stated above, we believe that HSE06 results should be much closer to reality and, therefore, in the following, we base our conclusions on them. To explain the experimental results, we should focus on defect levels that lead to transition energies of around 2 eV. Note that experimentally we measure spectra only in narrow energy range (1.65 eV to 2.75 eV) and, therefore, there might be additional, possibly stronger, optical peaks outside this range that are not seen in experiments. Fig. S10. Band diagrams of neutral (PBE and HSE06) and charged (HSE06) B monovacancy. Occupied states are in orange and unoccupied states are in blue. 5 Fig. S11. Band diagrams of neutral 3B+N vacancy. Occupied states are in orange and unoccupied states are in blue. Fig. S12. Band diagrams of neutral 6B+3N vacancy. Occupied states are in orange and unoccupied states are in blue. 6 In addition to available transition energies, photoexcitation and fluorescence spectra also depend on transition strength between two states that is given by dipole transition matrix elements. Therefore, in Fig. S13 we plot spectra that show both transition energies and transition strengths. • B monovacancy: B monovacancy is the most common defect in our TEM images and therefore Results for each vacancy are listed below: it is the first candidate for the homogeneous fluorescence. In agreement with previous studies, PBE results show three unoccupied defect levels at 0.66-1.48 eV above the valence band (left panel of Fig. S10). The highest one is responsible for optical transitions to states within valence band with the energy of around 2.3 eV as shown in the left panel of Fig. S13. Such peak in optical spectra would fit experiments. On the other hand, hybrid functional DFT results (middle panel in Fig. S10) also show three unoccupied defect levels within bulk band gap but with higher energies of 2.26 eV to 3.41 eV above the valence band. These levels have transition energies toward the top of the valence band that would also be in accordance with the experimental results, especially considering that excitonic effects would lower these transition energies. However, as can be seen in the right panel of Fig. S13, transition strengths for these transitions are very low and we obtain the same conclusions that were obtained in Ref. (9) with the GW/BSE theory. Namely, there are no significant optical transitions below 3.5 eV, far higher than observed in experiments. These results are puzzling and should be resolved by future measurements and calculations. From the one side, experimentally the most likely candidate for homogenuous emission is supported by PBE calculations that are typically rather inaccurate. From the other side, state of the art theoretical calculations like hybrid functional DFT or GW/BSE do not support it. Before concluding that B monovacancy is not responsible for homogeneous emission, further research is needed keeping in mind that even though GW and HSE are robust methods, it is possible that for some (still unknown) reasons they fail in predicting optical properties of this defect. Additionally, it should be explored whether defect-defect, defect-substrate, and defect-adsorbate interactions that are present in experimental conditions influence the electronic structure of the defect. From an experimental point of view, it would be helpful to have additional measurements in the full range of energies. • Charged B monovacancy: Results for negatively charged B monovacancy (right panel in Fig. S10) show several additional occupied and unoccupied defect levels within the bulk band gap. The transition from the highest occupied to the lowest unoccupied defect level corresponds to the energy of 1.78 eV. Right panel of Fig. S13 shows a clear peak at this transition energy, which demonstrates substantial transition strength. As this transition is due to the two defect levels within band gap it would be consistent with a single emitter. However, this transition energy is still significantly lower than in experiments, and the difference is larger than typical hybrid functional errors(12). It should be said that we did not additionally relax the charged defect compared to the neutral defect and therefore the forces on N atoms around the defect are rather large (~1 eV/Å). This could have an impact on the position of the defect levels. As can be seen in Fig. S13, there is an additional smaller absorption peak at around 2.8 eV that is also due to the transition between two defect levels. Since our results do not include excitonic effects that can easily shift transition energies down for 0.4 eV (see Ref. (9)), this peak would be consistent with experiments. All in all, it seems that charged B monovacancy could be responsible for ultra- bright emitters. 7 • 3B+N vacancy: HSE band diagram of 3B+N vacancy show unoccupied defect level at 1.0 eV above valence band and double degenerate level at 1.5 eV above valence band. As can be seen in right panel of Fig. S13 transition strengths to the top of the valence band are small and they become large only for transition energies larger than 2.5 eV. Therefore, hybrid functional DFT without considering excitonic effects (that would shift peak to lower energies) predict that 3B+N vacancy have a peak in optical spectra at around 2.8 eV. Due to non-zero dipole transition matrix elements at lower transition energies (toward the top of the valence band), fluorescence is expected to proceed also on lower transition energies. We conclude that 3B+N vacancy would be a good candidate for defect showing homogeneous fluorescence if its density would be high enough. • 6B+3N vacancy: For this vacancy, there is a transition with the energy of 1.2 eV which is clearly visible in the right panel of Fig. S10. In the spectra there is also a peak around 1.5 eV which corresponds to transitions from the defect level at 1.41 eV above valence band to the levels inside the valence band. Some small peaks also exist at transition energies in range 2-2.5 eV that could also be relevant considering the experiments, but the density of this type of vacancy is probably too low. Fig. S13. Simulated optical spectra in independent particle approximation calculated with PBE (left panel) and HSE06 (right panel) functional. 8 4. References 1. S. Caneva et al., Controlling Catalyst Bulk Reservoir Effects for Monolayer Hexagonal Boron Nitride CVD. Nano Letters 16, 1250 (Feb, 2016). L. B. Gao et al., Repeated growth and bubbling transfer of graphene with millimetre-size single-crystal grains using platinum. Nat Commun 3, (Feb, 2012). E. Betzig et al., Imaging intracellular fluorescent proteins at nanometer resolution. Science 313, 1642 (Sep 15, 2006). H. Deschout et al., Complementarity of PALM and SOFI for super-resolution live-cell imaging of focal adhesions. Nat Commun 7, 13693 (Dec 19, 2016). Y. Gong, I. F. Sbalzarini, in Image Processing (ICIP), 2013 20th IEEE International Conference on. (IEEE, 2013), pp. 534-538. K. I. Mortensen, L. S. Churchman, J. A. Spudich, H. Flyvbjerg, Optimized localization analysis for single-molecule tracking and super-resolution microscopy. Nat Methods 7, 377 (May, 2010). R. J. Ober, S. Ram, E. S. Ward, Localization accuracy in single-molecule microscopy. Biophys J 86, 1185 (Feb, 2004). P. Annibale, S. Vanni, M. Scarselli, U. Rothlisberger, A. Radenovic, Quantitative photo activated localization microscopy: unraveling the effects of photoblinking. PloS one 6, e22678 (2011). C. Attaccalite, M. Bockstedte, A. Marini, A. Rubio, L. Wirtz, Coupling of excitons and defect states in boron-nitride nanostructures. Physical Review B 83, 144115 (2011). X. Li et al., Non-Magnetic Quantum Emitters in Boron Nitride with Ultra-Narrow and Sideband- Free Emission Spectra. ACS nano, (2017). T. T. Tran, K. Bray, M. J. Ford, M. Toth, I. Aharonovich, Quantum emission from hexagonal boron nitride monolayers. Nature nanotechnology, (2015). J. M. Crowley, J. Tahir-Kheli, W. A. Goddard III, Resolution of the Band Gap Prediction Problem for Materials Design. The journal of physical chemistry letters 7, 1198 (2016). G. Kresse, J. Furthmüller, Efficient iterative schemes for ab initio total-energy calculations using a plane-wave basis set. Physical Review B 54, 11169 (1996). J. P. Perdew, K. Burke, M. Ernzerhof, Generalized gradient approximation made simple. Physical review letters 77, 3865 (1996). J. Heyd, G. E. Scuseria, M. Ernzerhof, Hybrid functionals based on a screened Coulomb potential. The Journal of Chemical Physics 118, 8207 (2003). J. Paier et al., Screened hybrid density functionals applied to solids. The Journal of Chemical Physics 124, 154709 (2006). 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 9 5. SI Figure Captions Fig. S1.Generated diffraction-limited image by summing all 20,000 frames, corresponding to the SMLM image shown in Fig. 2a. Fig. S2.FRC resolution analysis. A. Generated SMLM image using even number frames. B. Generated SMLM image using odd number frames. Total number of frames: 20,000. Image conditions: DI water, 561 nm laser. C. FRC curve indicates the decay of the correlation with spatial frequency. The image resolution (46 nm) is obtained using the inverse of the spatial frequency where FRC curve drops below the threshold 1/7. Fig. S3.Drift correction using fiducial markers. A. Corrected position of 30 fiducial markers (100 nm gold nanoparticles- from which 27 fiducial markers are used for drift correction). B. Drift trajectories of all 27 fiducial markers over 20,000 frames. C and D. x,y stage drift as a function of frame numbers. Fig. S4.Log-scale histogram of photon counts for the results shown in Fig. 2A. Insets show zoomed region of high photon counts. Fig. S5.SMLM imaging of defects in h-BN under basic and acidic conditions. A. SMLM image acquired under basic environment (pH 11, water).. B and C, Selected ROI in A and corresponding counting results D and E, respectively. The average defect density is from 1018 μm-2 (blink corrected) to 1648 μm-2 (total localizations). F. SMLM image acquired under acidic environment (pH 3, water). G and H. Selected ROI in F and corresponding counting results I and J, respectively. The average defect density is from 198 μm-2 (blink corrected) to 307 μm-2 (total localizations). Fig. S6.SMLM imaging of defects in h-BN after surface deposition. A. SMLM image of h-BN sample where defects are sealed by depositing 2 nm Al2O3 with atomic layer deposition. B and C, Selected ROI in A and corresponding counting results D and E, respectively. The average defect density is from 201 μm-2 (blink corrected) to 253 μm-2 (total localizations). Fig. S7.SMLM imaging of defects in h-BN under isotopic solvent conditions. A. SMLM image of defects in h-BN in D2O solvent. B and C, Selected ROI in A and corresponding counting results D and E, respectively. The average defect density is from 2186 μm-2 (blink corrected) to 4132 μm-2 (total localizations). Fig. S8. Balanced super-resolution optical fluctuation (bSOFI) images obtained from the same raw image sequences as in Fig 2a. 10 Fig. S1 3 (cid:80)m Fig. S2 A B C n o i t l a e r r o C Even Odd 1.0 0.8 0.6 0.4 0.2 0.0 0.0 0.1 FRC Smoothed FRC Threshold 0.2 0.3 Spatial frequency 0.4 0.5 Fig. S3 (cid:20)(cid:25) (cid:20)(cid:23) (cid:20)(cid:21) (cid:20)(cid:19) (cid:27) (cid:25) (cid:23) (cid:21) A ) m (cid:80) ( n o i t s o p y C ) m n ( n o i t s o p x (cid:21)(cid:26) (cid:21)(cid:27) (cid:21)(cid:28) (cid:22)(cid:19) (cid:21)(cid:25) (cid:21)(cid:24) 9 (cid:27) (cid:25) 7 1 3 (cid:21) (cid:23) 5 (cid:21)(cid:23) (cid:20)(cid:27) (cid:20)(cid:19) (cid:20)(cid:21) (cid:21)(cid:22) (cid:21)(cid:20)(cid:21)(cid:21) (cid:21)(cid:19) 19 11 13 (cid:20)(cid:23) 15 17 (cid:20)(cid:25) (cid:20)(cid:21) (cid:20)(cid:19) (cid:27) (cid:20)(cid:23) x position ((cid:80)m) (cid:20)(cid:25) (cid:20)(cid:27) (cid:21)(cid:19) (cid:19) (cid:21) (cid:23) (cid:25) (cid:21)(cid:19)(cid:19) (cid:20)(cid:24)(cid:19) (cid:20)(cid:19)(cid:19) (cid:24)(cid:19) (cid:19) (cid:237)(cid:24)(cid:19) (cid:237)(cid:20)(cid:19)(cid:19) (cid:237)(cid:20)(cid:24)(cid:19) (cid:237)(cid:21)(cid:19)(cid:19) (cid:19) 5 (cid:20)(cid:19) 15 (cid:21)(cid:19) (cid:49)(cid:88)(cid:80)(cid:69)(cid:72)(cid:85)(cid:3)(cid:82)(cid:73)(cid:3)(cid:73)(cid:85)(cid:68)(cid:80)(cid:72)(cid:86)(cid:3)(cid:91)(cid:20)(cid:19)3 B ) m n ( n o i t s o p y D (cid:22)(cid:19)(cid:19) (cid:21)(cid:19)(cid:19) (cid:20)(cid:19)(cid:19) (cid:19) (cid:16)(cid:20)(cid:19)(cid:19) (cid:16)(cid:21)(cid:19)(cid:19) (cid:16)(cid:22)(cid:19)(cid:19) (cid:16)(cid:23)(cid:19)(cid:19) (cid:16)(cid:22)(cid:19)(cid:19) (cid:16)(cid:21)(cid:19)(cid:19) (cid:16)(cid:20)(cid:19)(cid:19) (cid:19) (cid:20)(cid:19)(cid:19) (cid:21)(cid:19)(cid:19) (cid:22)(cid:19)(cid:19) x position (nm) ) m n ( n o i t s o p y (cid:20)(cid:21)(cid:19) (cid:20)(cid:19)(cid:19) (cid:27)(cid:19) (cid:25)(cid:19) (cid:23)(cid:19) (cid:21)(cid:19) (cid:19) (cid:237)(cid:21)(cid:19) (cid:237)(cid:23)(cid:19) (cid:237)(cid:25)(cid:19) (cid:237)(cid:27)(cid:19) (cid:19) 5 (cid:20)(cid:19) 15 (cid:21)(cid:19) (cid:49)(cid:88)(cid:80)(cid:69)(cid:72)(cid:85)(cid:3)(cid:82)(cid:73)(cid:3)(cid:73)(cid:85)(cid:68)(cid:80)(cid:72)(cid:86)(cid:3)(cid:91)(cid:20)(cid:19)3 Fig. S4 s n o i t a z i l o c o l f o r e b m u N 4500 4000 3500 3000 2500 2000 1500 1000 500 0 1 s n o i t a z i l o c o l f o r e b m u N 80 60 40 20 0 3.5 3.9 3.7 Photon counts (log scale) 4.3 4.1 0.5 2 2.5 3 3.5 4 4.5 Photon counts (log scale) Fig. S5 A F B 600 ) m n ( y 0 0 D 3000 s t n u o c t c e f e D 0 0 0 0 G 600 ) m n ( y 0 0 I 700 s t n u o c t c e f e D 0 0 0 0 C 200 400 x(nm) 0 600 E 1400 200 400 x(nm) 600 original original blink corrected blink corrected original original blink corrected blink corrected 0 0 0 0 20 20 Number of frames x103 20 20 H 200 400 x(nm) 0 600 J 500 200 400 x(nm) 600 original original blink corrected blink corrected original original blink corrected blink corrected 0 0 0 0 20 20 Number of frames x103 20 20 Fig. S6 A B 600 ) m n ( y 0 0 D 600 s t n u o c t c e f e D 0 0 0 0 C 200 400 x(nm) 0 600 E 100 200 400 x(nm) 600 original original blink corrected blink corrected original original blink corrected blink corrected 0 0 0 0 20 20 Number of frames x103 20 20 Fig. S7 A B 600 ) m n ( y 0 0 D 1400 s t n u o c t c e f e D 0 0 0 0 C 200 400 x(nm) 0 600 E 450 200 400 x(nm) 600 original original blink corrected blink corrected original original blink corrected blink corrected 0 0 0 0 20 20 Number of frames x103 20 20 Fig. S8 5 (cid:80)m
1607.03813	2	1607	2017-05-22T12:50:15	Measuring topological invariants from generalized edge states in polaritonic quasicrystals	[ "cond-mat.mes-hall", "cond-mat.quant-gas", "physics.optics" ]	We investigate the topological properties of Fibonacci quasicrystals using cavity polaritons. Composite structures made of the concatenation of two Fibonacci sequences allow investigating generalized edge states forming in the gaps of the fractal energy spectrum. We employ these generalized edge states to determine the topological invariants of the quasicrystal. When varying a structural degree of freedom (phason) of the Fibonacci sequence, the edge states spectrally traverse the gaps, while their spatial symmetry switches: the periodicity of this spectral and spatial evolution yields direct measurements of the gap topological numbers. The topological invariants that we determine coincide with those assigned by the gap-labeling theorem, illustrating the direct connection between the fractal and topological properties of Fibonacci quasicrystals.	cond-mat.mes-hall	cond-mat	a Measuring topological invariants from generalized edge states in polaritonic quasicrystals Florent Baboux∗1,2, Eli Levy∗3,4, Aristide Lemaıtre1, Carmen G´omez1, Elisabeth Galopin1, Luc Le Gratiet1, Isabelle Sagnes1, Alberto Amo1, Jacqueline Bloch1, Eric Akkermans3 ∗These authors contributed equally to this work. 1Centre de Nanosciences et de Nanotechnologies, CNRS, Univ. Paris-Sud, Universit´e Paris-Saclay, C2N Marcoussis, 91460 Marcoussis, France 2Laboratoire Mat´eriaux et Ph´enom`enes Quantiques, Sorbonne Paris Cit´e, Universit´e Paris Diderot, CNRS UMR 7162, 75013 Paris, France 3Department of Physics, Technion Israel Institute of Technology, Haifa 32000, Israel and 4Rafael Ltd., P.O. Box 2250, Haifa 32100, Israel We investigate the topological properties of Fibonacci quasicrystals using cavity polaritons. Com- posite structures made of the concatenation of two Fibonacci sequences allow investigating gener- alized edge states forming in the gaps of the fractal energy spectrum. We employ these generalized edge states to determine the topological invariants of the quasicrystal. When varying a structural degree of freedom (phason) of the Fibonacci sequence, the edge states spectrally traverse the gaps, while their spatial symmetry switches: the periodicity of this spectral and spatial evolution yields direct measurements of the gap topological numbers. The topological invariants that we determine coincide with those assigned by the gap-labeling theorem, illustrating the direct connection between the fractal and topological properties of Fibonacci quasicrystals. PACS numbers: 03.65.Vf, 61.44.Br, 71.36.+c, 78.67.-n, 78.67.Pt tant advances on the topological properties of quasicrystals [19–24, 28–30] the topological invariants have not yet been directly measured as winding numbers. Topology has long been recognized as a powerful tool both in mathematics and in physics. It allows identifying families of structures which cannot be related by continu- ous deformations and are characterized by integer numbers called topological invariants. A physical example where topological features are particularly useful is provided by quantum anomalies, i.e. classical symmetries broken at the quantum level [1], such as the chiral anomaly recently ob- served in condensed matter [2]. From a general viewpoint, wave or quantum systems possessing a gapped energy spec- trum, such as band insulators, superconductors, or 2D conductors in a magnetic field, can be assigned topolog- ical invariants generally called Chern numbers [3]. These numbers control a variety of physical phenomena: for in- stance in the integer quantum Hall effect, they determine the value of the Hall conductance as a function of magnetic field [4, 5]. Such topological features related to Chern num- bers have been explored in crystals [6] and more recently in various artificial periodic lattices for cold atoms [7–9], acoustic waves [10] or photons [11–16]. Quasicrystals – nonperiodic structures with long range order – [17, 18] are another important class of systems exhibiting topological effects [19–24]. In particular the topological edge states [25–27] of quasicrystals have been recently investigated in photonic systems [22, 28–30] and exploited to implement topological pumping, a key concept of topology [22]. A paradigmatic example of quasicrystal is given by the 1D Fibonacci chain. It presents a fractal energy spectrum which consists of an infinite number of gaps [31]. A rather surprising and fascinating property is that each of theses gaps can also be assigned a topologi- cal number analogous to the aforementioned Chern num- bers [32]: this constitutes the so-called gap-labeling theo- rem [33]. These integers can take N distinct values, N be- ing the number of letters in the chain [34]. Despite impor- # » # » √ In Eq. (1), σ = (1 + The physical origin of topological numbers in a Fi- bonacci quasicrystal can be related to its structural prop- erties [35]. To understand this, let us introduce a general method to generate a Fibonacci sequence: it is based on the characteristic function χj = sign(cid:2)cos(cid:0)2π j σ−1 + φ(cid:1) − cos(cid:0)π σ−1(cid:1)(cid:3) , (1) proposed in [36], which takes two possible values ±1, re- spectively identified with two letters A and B representing e.g. two different values of a potential energy. A Fibonacci FN (φ) ≡ [χ1χ2 ···χN ] formed sequence of size N is a word by A and B letters. 5)/2 is the golden mean and φ is a structural degree of freedom called phason, that can be continuously varied between 0 and 2π. The role of the phason has been experimentally in- vestigated in the frame of the Harper tight-binding model [22, 28, 36]. In the case of the Fibonacci model φ allows se- F ∞. lecting distinct finite segments along the infinite chain Sweeping φ over a 2π–period induces a series of N inde- pendent local structural changes in the Fibonacci sequence # » FN . Each change corresponds to the exchange of two let- ters (AB ↔ BA) at a given location of the sequence (see vertical arrows in Fig. 1c). Importantly, for two particular values of φ within a period, the Fibonacci sequence FN (φ) becomes palindromic, i.e. it coincides with its mirror sym- In between these two metric values, φ drives a π–periodic symmetry cycle along the 1D structure. When concatenating FN (φ), gener- alized edge states appear at the interface of the two mirror sequences, with properties tightly linked to φ. Since topo- logical invariants can always be written as winding num- bers, it was predicted [35] that the Fibonacci topological FN (φ) = [χN χN−1 ···χ1]. # » FN (φ) with # » #» # » numbers are measurable by counting how many times edge states traverse the gap while scanning the phason degree of freedom. of the energy of the edge states when increasing φ reflects directly the topological properties of the gap in which they appear. 2 To fabricate these structures, we process a planar mi- crocavity (of nominal Q factor 70000) grown by molecular beam epitaxy. The cavity consists in a λ/2 Ga0.05Al0.95As layer surrounded by two Ga0.8Al0.2As/Ga0.05Al0.95As Bragg mirrors with 28 and 40 pairs in the top and bot- tom mirrors respectively. Twelve GaAs quantum wells of width 7 nm are inserted in the structure, resulting in a 15 meV Rabi splitting. Quasi-1D cavities (wires) are realized using electron beam lithography and dry etching. The lat- eral width of these wires is modulated quasi-periodically, as shown in the Scanning Electron Microscopy (SEM) image of Fig. 1b. The modulation consists in two wire sections A and B of same length a = 1 µm, but different widths wA and wB. The width modulation induces an effective 1D potential for the longitudinal motion of polaritons (Fig. 1a), that follows the desired Fibonacci sequence. We chose N = 55 letters for the Fibonacci sequences and thus 110 letters for the concatenated structures. We have fabricated on a single sample the N = 55 concatenated structures cor- responding to all possible values of φ producing a struc- tural change in the sequence. Figure 1c shows a subset of three fabricated structures; the position of the interface FN is indicated by a between the mirror sequences vertical line. The exciton-photon detuning is of the order of −20 meV for all experiments. FN and # » # » 1d). To study the polariton modes in these quasiperiodic structures, we perform low temperature (10 K) micro- photoluminescence experiments (see Fig. Single structures are excited non-resonantly at low power, using a CW monomode laser at 740 nm. The excitation spot covers a 80 µm-long region centered on the interface. The emission is collected with a 0.5 numerical aperture objec- tive and focused on the entrance slit of a spectrometer coupled to a CCD camera. Imaging the sample surface or the Fourier plane of the collection objective allows study- ing the polariton modes either in real or momentum space. # » # » FN and Figure 2 shows the photoluminescence spectrum in real space [Fig. 2(a)] and in momentum space [Fig. 2(b)] of a Fibonacci structure with φ = 0.6 π, wA = 3.5 µm and wB = 2.2 µm. The energy spectrum shows an alternation of minibands and gaps. The emission in real space allows identifying two types of modes in the spectrum: modes de- localized over the whole structure, and modes localized at the interface between the FN sequences (x = 0). The delocalized modes form minibands, as can be seen in the momentum space emission: these modes are bulk modes forming a fractal energy spectrum characteristic of the Fibonacci quasicrystal [43]. We can identify the main gaps of the spectrum by applying the gap-labeling theorem a (p + qσ−1) for the wavevector [33], which predicts k = π position of the gaps [44]. Here, p and q are integers, with q being the gap topological number [33]. From the momen- tum space spectrum of Fig. 2b we extract [p, q] = [−1, 2] for the lower main energy gap, and [p, q] = [1,−1] for the higher main energy gap. (a) Nominal potential energy corresponding to a Figure 1. laterally modulated 1D cavity. (b) SEM image of a portion of a 1D cavity reproducing the Fibonacci sequence (top view). The letters A and B correspond to two different widths of the cavity. (c) SEM image showing the full view of 3 fabricated Fibonacci structures, corresponding to 3 different values of the # » FN phason φ. Each structure consists of the concatenation # » of a Fibonacci sequence FN . Ver- tical white arrows indicate the position of local changes in the sequence introduced when scanning φ. (d) Schematics of the experimental setup. # » FN and its mirror symmetric # » FN In this article we present a direct measurement of the topological invariants of a Fibonacci quasicrystal as spec- tral winding numbers. As an important consequence, we also relate these invariants to the observed change of spa- tial symmetry of generalized edge states, a robust feature largely insensitive to intrinsic disorder and other imper- fections. We employ cavity polaritons, quasiparticles aris- ing from the strong coupling between excitons confined in quantum wells and photons confined in a semiconduc- tor microcavity [37]. This photonic system allows emulat- ing a variety of Hamiltonians [38–41] and characterizing the associated eigenstates both in the spectral and spatial domain [40]. Here, we harness these features to emulate generalized edge states forming in the gaps of the fractal Fibonacci spectrum. # » # » # » FN To explore the edge states, we design concatenated structures FN made of the juxtaposition of a given Fi- bonacci sequence and its mirror symmetric (Fig. 1c). The interface defines a Fabry-Perot cavity of zero geometric length but finite round trip phase θcav due to the reflex- FN (see Supplemental Material [42]). ion between Thus the edge states will appear at energies Egap(φ), im- plicitly determined by the resonance condition FN and # » θcav(Egap, φ) = 2πm , (2) with m ∈ Z. As detailed in [35], θcav is periodic in φ with a period πq , where q is an integer, the topological invariant of the considered gap. Therefore, the evolution (b)Interface𝜙(c)𝜙1𝜙2𝜙3AAAABABAABBABAABPotential10 µm2 µm(a)AB↔BAAAAABABAABBABAABFibonaccistructuresmicroscope objectiveExcitationCCDDetectionLaserMicrocavity(d)3 where (cid:101)δ ≡ Egap−E− is the spectral position of the edge state within the gap, with ∆g being the gap width and E− the energy of the gap lower boundary [45]. ∆g The direction and periodicity of the observed traverse is different for the two edge states we consider: the lower energy state traverses 4 times upwards (winding number W = +4), while the higher energy state traverses 2 times downwards (W = −2), when φ spans a full period [0, 2π]. This winding W = 2q of the edge states allows for a di- rect determination of the gap topological numbers. We deduce q = +2 for the lower energy state and q = −1 for the higher energy state. These values obtained from the winding of the edge states are fully consistent with those previously determined from the bulk band structure (gap- labeling theorem), illustrating the existence of a bulk-edge correspondence in the quasiperiodic system. We now show that the topological invariants of the qua- sicrystal are not only measurable as winding numbers of the edge states, but they can also be directly retrieved from the spatial symmetry of the corresponding wavefunc- tions. Figure 4 (left column) shows the measured pro- file of the q = +2 and q = −1 edge states for values of φ taken in the 4 successive quadrants: [0, π/2], [π/2, π], [π, 3π/2] and [3π/2, 2π]. The states either show a max- imum or a minimum intensity at the interface (x = 0), corresponding to either a node or an antinode of the wave function. In the frame of a Fabry-Perot model [46] we will denote these two cases as symmetric (S) and anti- symmetric (AS) respectively. Because of experimental im- perfections the states are not perfectly S or AS but we shall employ this convenient terminology. We observe that the q = +2 state (red) switches symmetry in each quadrant, while the q = −1 state (blue) keeps the same symmetry in the first two quadrants before switching to the oppo- site symmetry in the last two quadrants. The symmetry index (S or AS) is reported in Fig. 3b for all values of φ, and compared to theory in Fig. 3d. Comparing this behavior to the spectral features reported in Fig. 3a-c, we observe that symmetry flips are exactly synchronized with Figure 2. (a)-(b) Energy-resolved emission of a Fibonacci structure in real space (a) and in momentum space (b), for a given value of the phason (φ = 0.62π). Edge states are visible in the two lowest main energy gaps, characterized by q = +2 and q = −1. # » These states are localized at the interface (x = 0) between the FN Fibonacci sequences. (c) Spatial profile of the edge state of gap q = +2 measured in a series of structures of same A-letter width wA = 3.5 µm but various B-letter width wB, yielding different contrasts for the Fibonacci potential. # » FN and # » # » FN and In addition to the bulk states, we observe states that FN . One are localized at the interface between of them (at energy ∼ 1596.5 meV) lies below the bulk band structure and is thus topologically trivial [42]. Two other localized states (encircled) lie within the widest spec- tral gaps: these are the expected topological edge states. Their spatial localization around the interface depends on the contrast of the Fibonacci quasiperiodic potential. Fig- ure 2c shows the spatial profile (squared modulus of the wavefunction, ψ(x)2) of the edge state of the gap q = +2, measured for a series of structures of same A-letter width wA = 3.5 µm, but various B-letter widths wB. As wB de- creases, the potential contrast (amplitude of the steps in Fig. 1a) increases: this leads to wider spectral gaps, and thus to a stronger spatial localization of the wavefunction. To explore the topological properties of the Fibonacci sequences by means of these edge states, we will now mon- itor their evolution when varying the phason degree of free- dom φ. We investigate a full set of N = 55 structures with wA = 3.5 µm and wB = 2.4 µm. For each structure, we perform spectroscopic measurements similar to Fig. 2a-b, and extract the energy of the two edge states with respect to the lowest bulk energy mode (bottom of the parabola, energy E0). The results are plotted in Fig. 3a, where the gap boundaries are indicated by the horizontal lines. Numerical calculations based on a scattering matrix ap- proach [35] are presented for comparison in Fig. 3c. Note that when φ = 0 or π, no edge state is observed. Indeed, for these particular values of φ, the FN sequence is palin- dromic and FN effectively reduces to a single Fibonacci sequence of size 2N . Hence there is no interface cavity and thus no edge state. FN # » # » # » As clearly seen in Fig. 3a, while scanning the phason φ the states perform piecewise spectral traverses inside the gaps. The number and the direction of the traverses yield a direct determination of the winding number [35]: W = 1 2π dθcav dφ dφ = 1 2π dφ = 2q (3) (cid:90) 2π 0 (cid:90) 2π d(cid:101)δ dφ 0 (a)(b)k (µm-1)x (µm)Energy(meV)(c)EdgestateInterface01-40-200204015961597159815991600-2-1012-40-2002040wB=3 µm2.3 µm2.4 µm2.5 µm2.6 µm2.9 µm2.8 µmIntensity (a.u.)x (µm)wB=2.2 µm4 Figure 3. (a) Measured energy of the edge states of gaps q = +2 et q = −1 as a function of the phason φ. E0 denotes the energy of the lowest bulk mode, and the solid lines indi- cate the gap boundaries. (b) Corresponding spatial symmetry of the edge states. When scanning φ, the wavefunctions evolve from symmetric (S) to anti-symmetric (AS) with respect to the interface (x = 0). (c) Relative spectral position within the gaps of the two considered edge states, obtained from scatter- ing matrix calculations. (d) Calculated symmetry of the two considered edge states. the spectral traverse of the states: they occur in between two successive traverses. Their periodicity thus allows de- termining the absolute value of the topological numbers. In addition, we observe that the sign of q is reflected in the direction of the symmetry flips within a period: the wavefunction of the q = −1 state switches from AS to S while that of the q = +2 state switches from S to AS. These experimental observations can be fully accounted for by a Fabry-Perot interpretation. The resonance condi- tion of Eq. (2), besides giving an accurate prediction for the gap states energies as shown in Figs. 3a and 3c, also encodes information on the mode symmetry. Indeed, for a fixed cavity length the occurrence of a node or antinode at the center is fully determined by the parity of the integer m. The spatial symmetry of the topological states thus switches at each spectral traverse [42]. This can be seen in the right column of Fig. 4, showing the calculated spatial profile of the two gap states under study, in all four quad- rants. These features demonstrate that, in contrast to pre- vious studies using normal edges (interface with vacuum) [22, 29], the use of generalized edges yields an additional degree of freedom (the symmetry index) that can be used to directly measure the topological invariants of the qua- sicrystal. This novel method is independent from the one based on spectral windings. It should thus prove useful in other physical platforms where the spectral degrees of freedom are more difficult to access. Figure 4. Measured (left) and calculated (right) spatial profile of the q = +2 and q = −1 edge states for values of φ taken in 4 different quadrants: [0, π/2], [π/2, π], [π, 3π/2] and [3π/2, 2π]. The mode spatial structure switches from symmetric (S) to antisymmetric (AS) with respect to the interface (x = 0), with a direction and periodicity yielding a direct measurement of the topological invariants of the quasicrystal. The blue lines are shifted vertically to improve their visibility. In summary, we have investigated the topological prop- erties of 1D Fibonacci polaritonic quasicrystals. We image generalized edge states forming in the gaps of the fractal energy spectrum. The behavior of these edge states upon varying a structural degree of freedom (phason) allow a di- rect determination of the topological invariants of the qua- sicrystal. The latter coincide with the bulk invariants as- signed by the gap-labeling theorem, illustrating the direct connection between the fractal and topological properties of Fibonacci quasicrystals. This work demonstrates a novel approach to determine topological features of quasicrystals by means of a direct measurement of winding numbers and of the corresponding symmetry flips of the wavefunctions. Taking advantage of the matter part of polaritons, it could be extended to probe the interplay of topology and interac- tions in quasicrystals [47] and potentially realize strongly correlated topological phases [48]. Furthermore, the build- ing of cavities with topological mirrors could allow the in- vestigation of the topological Casimir effect [49] in a well- controlled platform. Acknowledgments: This work was supported by the Israel Science Foundation Grant No. 924/09, by the Agence Nationale de la Recherche projects Quandyde (Grant No. ANR-11-BS10-001) and Quantum Fluids of 0.00.51.01.52.00.00.51.01.50.00.51.01.52.0(b)Energy - E0 (meV)Phason units of)(a)SASS AS Phason units of)ExperimentTheoryExperimentTheoryMode symmetryMode energygapgap0.00.51.01.52.00.00.51.01.52.0(d)Phason units of)(c) Phason units of)x (µm)-10-50510-10-50510Theoryx (µm)x (µm)-10-50510-10-50510ASASSSSASSExperimentx (µm)Intensity (a.u.)ASLight (Grant No. ANR-16-CE30-0021), the French RE- NATECH network, the European Research Council grant Honeypol and the EU-FET Proactive grant AQuS (Project No. 640800). [1] D. J. Gross and R. Jackiw, Phys. Rev. D 6, 477 (1972). [2] J. Xiong, S. K. Kushwaha, T. Liang, J. W. Krizan, M. Hirschberger, W. Wang, R. Cava, and N. Ong, Sci- ence 350, 413 (2015). [3] S. S. Chern, Complex Manifolds Without Potential Theory (Springer-Verlag, New York-Heidelberg, 1979). [4] K. v. Klitzing, G. Dorda, and M. Pepper, Phys. Rev. Lett. 45, 494 (1980). [5] J. Bellissard, A. van Elst, and H. Schulz-Baldes, Journal of Mathematical Physics 35, 5373 (1994). [6] M. Z. Hasan and C. L. Kane, Rev. Mod. Phys. 82, 3045 (2010). [7] N. Goldman, J. Dalibard, A. Dauphin, F. Gerbier, M. Lewenstein, P. Zoller, and I. B. Spielman, Proceedings of the National Academy of Sciences 110, 6736 (2013). [8] M. Atala, M. Aidelsburger, J. T. Barreiro, D. Abanin, T. Kitagawa, E. Demler, and I. Bloch, Nature Physics 9, 795 (2013). [9] M. Aidelsburger, M. Lohse, C. Schweizer, M. Atala, J. T. and Barreiro, S. Nascimbene, N. Cooper, N. Goldman, Nature Physics 11, 162 (2015). I. Bloch, [10] M. Xiao, G. Ma, Z. Yang, P. Sheng, Z. Zhang, and C. T. Chan, Nature Physics 11, 240 (2015). [11] M. C. Rechtsman, J. M. Zeuner, Y. Plotnik, Y. Lumer, D. Podolsky, F. Dreisow, S. Nolte, M. Segev, and A. Sza- meit, Nature 496, 196 (2013). [12] M. Hafezi, Phys. Rev. Lett. 112, 210405 (2014). [13] W. Hu, J. C. Pillay, K. Wu, M. Pasek, P. P. Shum, and Y. D. Chong, Phys. Rev. X 5, 011012 (2015). [14] C. Poli, M. Bellec, U. Kuhl, F. Mortessagne, and H. Schomerus, Nature communications 6 (2015). [15] S. A. Skirlo, L. Lu, Y. Igarashi, Q. Yan, J. Joannopoulos, and M. Soljaci´c, Phys. Rev. Lett. 115, 253901 (2015). [16] S. Mittal, S. Ganeshan, J. Fan, A. Vaezi, and M. Hafezi, Nature Photonics (2016). [17] D. Shechtman, I. Blech, D. Gratias, and J. W. Cahn, Phys. Rev. Lett. 53, 1951 (1984). [18] D. Levine and P. J. Steinhardt, Phys. Rev. Lett. 53, 2477 (1984). [19] B. Simon, Advances in Applied Mathematics 3, 463 (1982). [20] H. Kunz, Phys. Rev. Lett. 57, 1095 (1986). [21] L.-J. Lang, X. Cai, and S. Chen, Phys. Rev. Lett. 108, 220401 (2012). [22] Y. E. Kraus, Y. Lahini, Z. Ringel, M. Verbin, and O. Zil- berberg, Phys. Rev. Lett. 109, 106402 (2012). [23] Y. E. Kraus, Z. Ringel, and O. Zilberberg, Phys. Rev. Lett. 111, 226401 (2013). [24] S. Ganeshan, K. Sun, and S. Das Sarma, Phys. Rev. Lett. 110, 180403 (2013). [25] E. S. Zijlstra, A. Fasolino, and T. Janssen, Phys. Rev. B 59, 302 (1999). [26] Y. El Hassouani, H. Aynaou, E. H. El Boudouti, B. Djafari- Rouhani, A. Akjouj, and V. R. Velasco, Phys. Rev. B 74, 035314 (2006). [27] X.-N. Pang, J.-W. Dong, and H.-Z. Wang, J. Opt. Soc. Am. B 27, 2009 (2010). [28] M. Verbin, O. Zilberberg, Y. E. Kraus, Y. Lahini, and Y. Silberberg, Phys. Rev. Lett. 110, 076403 (2013). 5 [29] M. Verbin, O. Zilberberg, Y. Lahini, Y. E. Kraus, and Y. Silberberg, Phys. Rev. B 91, 064201 (2015). [30] P. Vignolo, M. Bellec, J. Bohm, A. Camara, J.-M. Gam- baudo, U. Kuhl, and F. Mortessagne, Phys. Rev. B 93, 075141 (2016). [31] D. Damanik, M. Embree, A. Gorodetski, and S. Tcherem- chantsev, Communications in Mathematical Physics 280, 499 (2008). [32] Note that strictly speaking Chern numbers are invariants describing the topology of smooth Riemannian manifolds. In contrast, the Fibonacci chain and quasicrystals in gen- eral, cannot be ascribed such a smooth manifold but rather discrete graphs [50] whose topological numbers are termed differently (K-theory invariants). For simplicity we shall pursue using indifferently the denomination Chern num- bers or topological numbers. [33] J. Bellissard, A. Bovier, and J.-M. Ghez, Reviews in Math- ematical Physics 4, 1 (1992). [34] A. Dareau, E. Levy, M. Bosch Aguilera, R. Bouganne, J. Beugnon, F. Gerbier, E. Akkermans, arXiv:1607.00901 (2016). and [35] E. Levy, A. Barak, A. Fisher, and E. Akkermans, arXiv:1509.04028 (2015). [36] Y. E. Kraus and O. Zilberberg, Phys. Rev. Lett. 109, 116404 (2012). [37] C. Weisbuch, M. Nishioka, A. Ishikawa, and Y. Arakawa, Phys. Rev. Lett. 69, 3314 (1992). [38] M. Bayer, T. Gutbrod, A. Forchel, T. L. Reinecke, P. A. Knipp, R. Werner, and J. P. Reithmaier, Phys. Rev. Lett. 83, 5374 (1999). [39] N. Y. Kim, K. Kusudo, C. Wu, N. Masumoto, A. Loffler, S. Hofling, N. Kumada, L. Worschech, A. Forchel, and Y. Yamamoto, Nature Physics 7, 681 (2011). [40] I. Carusotto and C. Ciuti, Reviews of Modern Physics 85, 299 (2013). [41] F. Baboux, L. Ge, T. Jacqmin, M. Biondi, E. Galopin, A. Lemaıtre, L. Le Gratiet, I. Sagnes, S. Schmidt, H. E. Tureci, A. Amo, and J. Bloch, Phys. Rev. Lett. 116, 066402 (2016). [42] See Supplemental Material for details on the Fabry-Perot interpretation and additional data. [43] D. Tanese, E. Gurevich, F. Baboux, T. Jacqmin, A. Lemaıtre, E. Galopin, I. Sagnes, A. Amo, J. Bloch, and E. Akkermans, Phys. Rev. Lett. 112, 146404 (2014). [44] This property, exact in the limit of infinite chains, consti- tutes an excellent approximation for the considered struc- ture length, as shown in Ref. [43]. [45] Since the quasicrystal is an effectively discrete manifold, the phase θcav defined in Eq. (2) can take only a finite set of discrete values equal to the Fibonacci number FN . Thus, strictly speaking the winding number W should be written as a discrete sum. As a convenient notation which results from the use of a continuous variable φ in Eq. (1), we replace this discrete sum by an integral. [46] E. Levy and E. Akkermans, Eur. Phys. J. Special Topics 226, 1563 (2017). [47] B. Freedman, G. Bartal, M. Segev, R. Lifshitz, D. N. Christodoulides, and J. W. Fleischer, Nature 440, 1166 (2006). [48] X.-G. Wen, Advances in Physics 44, 405 (1995). [49] M. Bordag, G. L. Klimchitskaya, U. Mohideen, and V. M. Mostepanenko, Advances in the Casimir effect (OUP Ox- ford, 2009). [50] J. E. Anderson and I. F. Putnam, Ergodic Theory and Dynamical Systems 18, 509 (1998). Supplemental Material - Measuring topological invariants from generalized edge states in polaritonic quasicrystals FABRY-PEROT MODEL OF THE TOPOLOGICAL EDGE STATES 6 The effective Fabry-Perot model derived here is for a true 1D system, and is given in the language of the scat- tering theory, as originally proposed in Ref. [35]. The basic idea is to regard the interface between the Fibonacci se- FN (φ) as a virtual Fabry-Perot cavity, quences FN (φ) and correspondingly consider the chains as mirrors. In this perspective, the generalized edge states localized at the interface between the two Fibonacci chains are interpreted as Fabry-Perot resonances (see Fig. S1). FN (φ) and FN (φ) and # » # » # » # » # » FN Figure S1. Appearance of gap states in the Fibonacci based # » structure FN with N = 55. (a) Density of states spectrum DOS(k) with interface states appearing in the gaps (here at normalized wavevectors k = 0.39 and k = 0.61). (b) Spatial # » arrangement of the structure FN indicated by yellow and magenta bars (corresponding to letters A and B), with a rep- resentation of the spatial profile of the gap state at k = 0.39. # » FN # » # » Qualitatively speaking, the Fibonacci chains FN (φ) or FN (φ) which have a gapped spectrum are mirrors for spe- cific frequencies, due to the high reflectance values at the spectral gaps. These mirrors are not standard, since they provide a frequency-dependent phase shift upon reflection due to multiple reflections. This phase shift, as we will show now, allows to treat an interface between two such structures as a virtual (not geometric) cavity length. The standard Fabry-Perot resonance condition for a cav- ity of length L, given by 2L/λm = m, m∈Z, (S1) ({λm} being a discrete set of resonant wavelengths), is a basic constructive interference condition. Therefore, it may always be written in terms of the winding of a phase: the cavity phase θcav, representing the total round-trip phase inside the cavity and defined by θcav (k, L) ≡ 4πL λ (k) (S2) , where λ(k) = 2π/k is the wavelength. This gives a reso- nance condition equivalent to Eq. (S1), namely θcav(km) = 2πm , m∈Z. (S3) Here we follow the very same argument, but in the re- FN , which verse order. We begin with the structure FN # » # » (a) and (c): The Fibonacci chain Figure S2. Different boundary conditions for the Fibonacci # » chain. FN with N = 55 bounded from the right (a) by a metallic mirror (orange line), or (c) by a continuum with refractive index smaller than that of the chain (blue line). The presence of a node or antinode of the gap states at the boundary is here rigidly imposed by the # » FN , based boundary conditions. (b) The unfolded structure on the same structure, hosts generalized edge states which can possess either a node or an antinode at the interface (dotted line). # » FN # » # » FN and is a cavity of zero (geometrical) length. The cavity phase is however non-zero, as FN both yield a non-zero phase upon reflection. If we define the reflected phase shift for the left boundary of θ left (k), then the cavity phase for θ left (k). Now, similarly to Eq. (S2), a virtual cavity length is defined as FN is 2 FN as FN # » # » # » #» #» L (k, θcav) ≡ λ(k) 4π θcav, (S4) with resonant interface states occurring at gap frequencies satisfying the Fabry-Perot condition 2L(km)/λ(km) = m, m∈Z. (S5) The model then predicts that for every value of λ(km) lying within a spectral gap, a new interface state will exist. This reasoning is utilized in Eqs. (2) and (3) of the arti- cle directly through θcav, to define the topological winding number of the generalized edge state when the phason φ is scanned. In that case, the scanning of φ monotonically drives the cavity phase (at gap frequencies) such that the resonant states monotonically change frequency with φ. This reasoning fully supports the experimental observa- tion of the topological winding of generalized edge states as a function of φ in momentum space, as evidenced in Fig. 3a,b of the letter. We now describe a Fabry-Perot prop- erty in real space that helps to further clarify the interplay between the topological and the Fabry-Perot properties of the generalized edge states. FN The structure FN is described in the article as a host for generalized edge states, in contrast to previous stud- ies considering the interface with the vacuum. To clarify this, we consider the FN chain bounded from the right by a perfect mirror. Waves traveling through the structure # » # » # » NF(a)(b)NFDOS…(a)(b)(c)AB………INFNFNFNF7 mismatched boundary conditions (S3a,c) taken together. Secondly, for a given gap, generalized edge states tra- verse the gaps as a function of φ due to a monotonic change in the cavity phase. This means that when an edge state merges with one band-edge and a new state bifurcates from the other band-edge, then the parity of m flips and so does the spatial symmetry of the edge state. This is the theo- retical argument supporting the experimental observation reported in Fig. 4 of the article. These features demon- strate that the use of generalized edges yields an additional degree of freedom (the symmetry index) that can be used to directly measure the topological invariants of the qua- sicrystal. TOPOLOGICALLY TRIVIAL MODES OF THE POLARITONIC QUASICRYSTALS In this section we show that our polaritonic structures also hosts topologically trivial modes, that can be clearly distinguished from the topological modes described above. An example is given in Fig. 2a,b of the article. Among the # » FN states that are localized at the interface between the FN sequences, one of them (at energy ∼ 1596.5 meV) and lies below the bulk band structure and is thus topologically trivial. # » Figure S4. (a) Measured energy of the interface states (for the structure considered in Figs. 2,3,4 of the main text) as a function of the phason φ. The topologically trivial mode is indicated with black circles. E0 denotes the energy of the lowest bulk mode, and the solid lines indicate the gap bound- aries. (b) Calculation of the whole energy spectrum from the 2D Schrodinger equation, as a function of φ. This mode only appears for a particular quadrant of φ (between π/2 and π), for which the sequence of letters at the interface is AAAA. Since in our samples, the letter A corresponds to a lower potential value, this AAAA se- quence forms a spatially extended potential well, and the mode that is seen is a bound mode of this well. To verify that this mode is topologically trivial, we can monitor its spectral evolution as a function of φ, both in the experi- ment and the simulation (2D Schrodinger calculation) as shown in Fig. S4. We observe that the mode only ap- pears in the quadrant [π/2, π] and keeps a constant energy within the experimental error: it does not perform any spectral traverse as the phason is scanned, which confirms its topologically trivial nature. Figure S3. Real space profile of selected gap states for the # » structures depicted in Fig. S2: (a) The FN chain with metallic # » reflective boundary, (b) the unfolded structure FN and (c) the # » FN chain with a refractive-index mismatch boundary. # » FN # » # » # » and towards the mirror plane are reflected back into the chain, experiencing the quasiperiodic modulation in the reverse order ( FN ). An equivalent version of this setup consists in removing the mirror and unfolding the chain with respect to the mirror plane, resulting in the structure considered in the article. The equivalence between these two boundary conditions is however not total. In partic- ular, the structure FN may host interface states which are both spatially symmetric and anti-symmetric with re- spect to the interface, namely with a node or an anti-node at the interface, as seen in Fig. S2b. In contrast to that, a single FN chain bounded by a perfect mirror can host only one type of states: with a node on the mirror in the case of a metallic mirror, as shown in Fig. S2a, or with an antinode in the case of a mismatched mirror (continuum with refractive index smaller than that of the chain), as shown in Fig. S2c. FN # » The fact that the generalized-edge scheme allows all pos- sible interface states yields an additional degree of freedom to probe the topological content of the states. Indeed, topological information may be extracted from the real space properties of the edge states through the parity of the Fabry-Perot integer, m in Eq. (S5). In the usual Fabry- Perot picture for phase-conserving perfect mirrors, odd and even values of m alternate between anti-symmetric and symmetric states with respect to the mid-cavity coordinate (with a node and an antinode at mid-cavity) respectively. This result is also true for our Fabry-Perot cavity with a φ dependent virtual length. This leads to two possible predictions. Firstly, for a given value of φ, edge states residing in the various gaps have different spatial symmetries com- pletely predictable by the Fabry-Perot model through the parity of m. For instance, Fig. S3 shows that modes with m = 2 (red) are symmetric with respect to the in- terface, while modes with m = 3 (blue) are antisymmet- ric. Again, we here see that the general-edge scheme (S3b) hosts the union of all edge states of the metallic and index- (a)(b)(c) 00.510.00.51.01.52.00.00.51.01.5q=-1q=+2Energy - E0 (meV)Phason units of)Phason(unitsof )00.511.52ExperimentSimulation(a)(b)01
1512.04793	2	1512	2015-12-18T08:29:32	Phosphorene confined systems in magnetic field, quantum transport, and superradiance in the quasi-flat band	[ "cond-mat.mes-hall" ]	Spectral and transport properties of electrons in confined phosphorene systems are investigated in a five hopping parameter tight-binding model, using analytical and numerical techniques. The main emphasis is on the properties of the topological edge states accommodated by the quasi-flat band that characterizes the phosphorene energy spectrum. We show, in the particular case of phosphorene, how the breaking of the bipartite lattice structure gives rise to the electron-hole asymmetry of the energy spectrum. The properties of the topological edge states in the zig-zag nanoribbons are analyzed under different aspects: degeneracy, localization, extension in the Brillouin zone, dispersion of the quasi-flat band in magnetic field. The finite-size phosphorene plaquette exhibits a Hofstadter-type spectrum made up of two unequal butterflies separated by a gap, where a quasi-flat band composed of zig-zag edge states is located. The transport properties are investigated by simulating a four-lead Hall device (importantly, all leads are attached on the same zig-zag side), and using the Landauer-Buttiker formalism. We find out that the chiral edge states due to the magnetic field yield quantum Hall plateaus, but the topological edge states in the gap do not support the quantum Hall effect and prove a dissipative behavior. By calculating the complex eigenenergies of the non-Hermitian effective Hamiltonian that describes the open system (plaquette+leads), we prove the superradiance effect in the energy range of the quasi-flat band, with consequences for the density of states and electron transmission properties.	cond-mat.mes-hall	cond-mat	Phosphorene confined systems in magnetic field, quantum transport, and superradiance in the quasi-flat band B. Ostahie1,2, and A. Aldea1 1 National Institute of Materials Physics, 77125 Bucharest-Magurele, Romania 2 Faculty of Physics, University of Bucharest, Romania (Dated: September 17, 2018) Abstract Spectral and transport properties of electrons in confined phosphorene systems are investigated in a five hopping parameter tight-binding model, using analytical and numerical techniques. The main emphasis is on the properties of the topological edge states accommodated by the quasi-flat band that characterizes the phosphorene energy spectrum. We show, in the particular case of phosphorene, how the breaking of the bipartite lattice struc- ture gives rise to the electron-hole asymmetry of the energy spectrum. The properties of the topological edge states in the zig-zag nanoribbons are analyzed under different aspects: degener- acy, localization, extension in the Brillouin zone, dispersion of the quasi-flat band in magnetic field. The finite-size phosphorene plaquette exhibits a Hofstadter-type spectrum made up of two unequal butterflies separated by a gap, where a quasi-flat band composed of zig-zag edge states is located. The transport properties are investigated by simulating a four-lead Hall device (importantly, all leads are attached on the same zig-zag side), and using the Landauer-Buttiker formalism. We find out that the chiral edge states due to the magnetic field yield quantum Hall plateaus, but the topological edge states in the gap do not support the quantum Hall effect and prove a dissipative behavior. By calculating the complex eigenenergies of the non-Hermitian effective Hamiltonian that describes the open system (plaquette+leads), we prove the superradiance effect in the energy range of the quasi-flat band, with consequences for the density of states and electron transmission properties. PACS numbers: 73.20.At, 73.63.-b, 73.43.-f 5 1 0 2 c e D 8 1 ] l l a h - s e m . t a m - d n o c [ 2 v 3 9 7 4 0 . 2 1 5 1 : v i X r a 1 I. INTRODUCTION The very recent revival of the black phosporene physics comes from the technical pos- sibility to obtain monolayers, known as phosphorene, with specific topological properties. Phosphorene is a quasi-2D structure organized as a puckered hexagonal lattice, the top and side views being shown in Fig.1a and Fig.1b, respectively. One may think that, due to the structural similarity, the electron properties of phosphorene are resembling those of graphene. However, in contradistinction to graphene, the phosphorene is an anisotropic direct gap semiconductor, much more attractive for electronic devices. Beside the monolay- ered structure, multilayers of black phosporous are also studied, mainly in order to control the band gap, in the perspective of a potential application for field-effect transistors. The phosphorene ribbon geometry (especially, with zig-zag edges) is also conceptually interesting since, instead of the semi-metallic spectrum of graphene, distinguished by a flat band at E = 0, and extending between the points K and K' in the Brillouin zone (BZ), the phosphorene shows well-separated valence and conduction bands and a quasi-flat band in the middle of the gap, composed of edge states that exists at any momentum k ∈ BZ. In the tight-binding model, the phosphorene lattice is described by five hopping integrals t1, t2, .., t5 [1], which induce the significant differences in the electron spectrum that are noticed when compared to graphene. The model points out also the anisotropy of the energy spectrum: both the top of the valence band and the bottom of the conduction band look quadratically as function of ky , but nearly linear as function of kx (Dirac-like) (see Fig.2), situation which is described in terms of hybrid Dirac spectrum [2, 3]. The hopping integral t4 plays a distinctive role as it connects sites of the same kind on the hexagonal lattice, breaking the bipartitism of the lattice, and, as a consequence, the electron-hole symmetry of the energy spectrum is also broken [4]. As an additional effect due to t4, we shall see in Sec.II that the edge states, organized in a perfect flat band at t4 = 0, undergoes dispersion in the case of nonvanishing t4. The properties of the macroscopically degenerate flat (quasi-flat) bands composed of edge states in confined systems (ribbon or finite-size plaquette) attract much attention nowadays, and phosphorene presents a serious advantage coming from the existence of a gap that protects the quasi-flat band, such that its properties can be evidenced in a cleaner way. The study of the spectral and transport properties in the magnetic field, and the response of 2 the quasi-flat band to the invasive contacts of a Hall device, identified as a superradiant phenomenon, are a topic of our paper. Similar to graphene, the confined phosphorene exhibits two types of edge states: i) the chiral edge states generated by a strong perpendicular magnetic field B, and supporting the quantum Hall effect (QHE), and ii) the edge states typical to the zig-zag boundaries in the hexagonal lattice, which exists even in the absence of the magnetic field. The last ones, which will be called topological edge states [5], are non-chiral and remain like that even at B (cid:54)= 0. Obviously, they do not show QHE, but show longitudinal conductance, i.e., they have a dissipative character. The transport calculations assume the knowledge of the full Hamiltonian of the open system consisting of the finite-size system of interest (namely, the phosphorene plaquette) and the semi-infinite leads. Technically speaking, one uses actually an effective Hamiltonian obtained by formal elimination of the degree of freedom of the leads, however, as the price to be paid, the result is a non-Hermitian Hamiltonian with complex eigenvalues. The method of non-Hermitian Hamiltonian has been used for the calculation of transport properties of the quantum dots in the Landauer-Buttiker formalism (see for instance [6]), but also in the localization-delocalization problem in the 1D non-Hermitian Anderson model [7 -- 9]. In these two different problems, the non-hermicity arises from different sources, however we do not enter here such peculiar aspects. In the phosphorene confined system, the complex eigenvalues of the effective Hamiltonian in the energy range of the quasi-flat band, corroborated by the calculation of the electron transmission, density of states and local density of states make evident a specific superradiant behavior of the topological edge states. (We remind that the superradiance consists in the segregation of eigenenergies and overlapping of some resonances, the process being controlled by the lead-system coupling [10].) For instance, the density of states of the quasi-flat band exhibits a miniband structure, each miniband behaving as a 1D-conducting channel with the conductance G = e2/h. These aspects are discussed in Sec.IV. Some quantum transport aspects in phosphorene were very recently revealed. The quan- tum Hall effect and spin splitting of the Landau levels (LL) were observed in [11, 12], and also Shubnikov-deHaas oscillations of the longitudinal resistance were found in [13]. The transport anisotropy, reflecting the structural one, was shown experimentally by measuring the angle dependence of the drain current [14] or by the non-local response [15]. The strain- 3 induced modifications of the phosphorene band structure were studied in [16, 17]. The field effect transistor is also the topic of [18, 19]. The paper is organized as follows. Sec.II presents the tight-binding Hamiltonian, Peierls phases in magnetic field and discusses the question of electron-hole symmetry breaking. Sec.III is devoted to the study of the phosphorene ribbon in the magnetic field, as an extension of Ezawa analysis at B = 0. Sec.IV deals with the spectral and transport properties of the phosphorene mesoscopic plaquette, showing the specific features of the quantum Hall effect in the bands, and the properties resulting from the superradiance effect in the quasi-flat band. The summary and conclusions can be found in the last section. II. THE TIGHT-BINDING MODEL AND ELECTRON-HOLE SYMMETRY BREAKING IN PHOSPHORENE Similar to graphene, the unit cell contains two atoms called A and B, however the phos- phorene tight-binding Hamiltonian is more complicated as it contains five hopping integrals to nearest and next-nearest neighbors. In order to write down the Hamiltonian we define the creation and annihilation operators a† nm, anm, where n and m are cell indexes along the Ox- and Oy-axis, respectively. In the presence of a perpendicular magnetic field, which will be described by the vector potential (cid:126)A = (−By, 0, 0), some of the hopping inte- grals acquire the Peierls phase expressed by the circulation of the vector potential along the nm, anm, b† trajectory connecting the two end points : (cid:90) B A (cid:90) xB xA φAB = 2π Φ0 (cid:126)Ad(cid:126)l = −2πB Φ0 y(x)dx. (1) Much attention should be paid to the calculation of the phases since the angle β describing the deviation from the perfect flat 2D lattice (see Fig.1b) enter also the calculation. Since the hopping integral t4 plays the special role mentioned in the introduction, we separate the terms proportional to this parameter, and the spinless tight-binding Hamiltonian of the 4 phosphorene lattice under perpendicular magnetic field will be written as follows: H 0 = H = H 0 + H 4, Eaa† nmanm + Ebb† † nm+2 + e−iφ3a nm + t3 (cid:88) (cid:0)eiφ3a (cid:88) (cid:0)eiφ4B b (cid:0)eiφ4Aa t4 H 4 = nm + t4 † nm+1 + e−iφ4B b † n−1m+1 † nm+1 + e−iφ4Aa † n−1m+1 (cid:1)bnm + t2a nmbnm + t1 † n−1m+2)bnm + t5a † n+1m + e−iφ1a† † n+1m−1bnm + H.c. , nm (cid:0)eiφ1a (cid:1)bnm (cid:1)anm + H.c., † nm+1bnm (2) where Ea and Eb are the atomic energies at the sites A and B, respectively, and , according to [1], t1 = −1.220eV, t2 = 3, 665eV, t3 = −0.205eV, t4 = −0.105eV, t5 = −0.055eV . Since the hopping parameters are given in electron-volts, all the quantities in the paper having the dimension of energy will be measured in the same units. FIG. 1: (Color online) (a) Schematic representation of phosphorene lattice with two types of edges, zig-zag (along the Ox direction) and armchair (along the Oy direction); t1, t2, t3, t4, t5 are the hopping amplitudes that connect the lattice sites; A (red) and B (blue) index the two types of atoms, and the dashed blue lines represent the unit cell with a1 and a2 as unit vectors. (b) The projection of the lattice on the yz-plane. The number of lattice sites is 7 × 4. In the chosen gauge of the vector potential, only three hopping integrals acquire a Peierls phase in magnetic field, namely t1, t3 and t4. For instance, φ1(m) in Eq.(1) is the Peierls phase corresponding to the hopping from the site B in the cell (n,m) to the site A in the 5 t1t2t3t4t5ABABβ=103.69oa1a2OyOxOyOz(a)(b)next cell (n+1,m), and equals: (cid:90) An+1,m Bnm y(x)dx = −2π Φ Φ0 1 6 (cid:0)(m − 1)(1 + 2sinβ) − 1 2 (cid:1). φ1(m) = 2π B Φ0 Similarly, the other phases in Eq.(2) are [20], φ3(m) = −2π φ4A(m) = −2π φ4B(m) = −2π Φ Φ0 Φ Φ0 Φ Φ0 1 6 1 6 1 6 (cid:1) (cid:0)m(1 + 2sinβ) − 1 (cid:0)(m − 1 (m − 1 2 )(1 + 2sinβ) )(1 + 2sinβ) + 1(cid:1), 2 2 where Φ/Φ0 is the magnetic flux through the hexagonal cell measured in quantum flux units, and β is the angle shown in Fig.1b. One notices that the phases φ4A and φ4B acquired by t4 along the A-A and B-B link, respectively, are different. The spectral properties of the Hamiltonian (2) can be studied under different bound- ary conditions describing different geometries as the infinite sheet, the ribbon or the finite plaquette. The phoshorene infinite sheet can be simulated assuming periodic boundary con- ditions along the both directions Ox and Oy. Let us consider first the case B = 0, and use the Fourier transform of the creation and annihilation operators: anm = a(cid:126)kei(cid:126)k (cid:126)Rnm, bnm = b(cid:126)kei(cid:126)k (cid:126)Rnm, (cid:126)Rnm = n(cid:126)a1 + m(cid:126)a2, (3) (cid:126)k (cid:126)k (cid:88) (cid:88) (cid:88) which helps in writing the Hamiltonian as a 2×2 matrix in the momentum space (cid:126)k = (kx, ky): (cid:88) (cid:16) H 0 (cid:126)k + H 4 (cid:126)k = (cid:126)k (cid:126)k † a (cid:126)k † b (cid:126)k (cid:17) T 4((cid:126)k) T 0((cid:126)k) T 0∗((cid:126)k) T 4((cid:126)k)  a(cid:126)k  , b(cid:126)k H = with T 0((cid:126)k) = t1(1 + e−i(cid:126)k(cid:126)a1) + t2e−i(cid:126)k(cid:126)a2 + t3(e−i2(cid:126)k (cid:126)a2 + ei(cid:126)k (cid:126)a1−i2(cid:126)k (cid:126)a2) + t5e−i(cid:126)k (cid:126)a1+i(cid:126)k (cid:126)a2 T 4((cid:126)k) = 2t4(cos(cid:126)k (cid:126)a2 + cos(cid:126)k( (cid:126)a1 − (cid:126)a2)). (4) (5) In approaching the question of spectrum symmetries, we remind that the electron-hole symmetry of an energy spectrum holds if there exists an operator P that anticommutes with the Hamiltonian, {H,P}+ = 0. Indeed, it is quite straightforward to see that, if E is an eigenvalue, HΨE = EΨE, then the energy −E belongs also to the spectrum, the 6 corresponding eigenfunction being Ψ−E = PΨE. For our specific problem of phosphorene, let us consider the operator P = † a (cid:126)k a(cid:126)k − b † (cid:126)k b(cid:126)k . (6) (cid:88) (cid:126)k Obviously, this operator anticommutes with H 0, attesting that the energy spectrum of H 0 k} ∈ Sp. However, P does not anticommute with the is electron-hole symmetric, {E0 total Hamiltonian H 0 + H 4, the result being proportional to t4. One concludes that the k,−E0 phosphorene spectrum is electron-hole symmetric if t4 = 0, i.e, when one forgets about the hopping to the next-nearest neighbors, but it is not necessarily symmetric otherwise. The energy spectrum of the Hamiltonian (4) can be obtained analytically from the char- acteristic equation (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)E − T 4((cid:126)k) T 0((cid:126)k) T 0∗((cid:126)k) E − T 4((cid:126)k) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) = 0, (7) (8) resulting a two-band spectrum of semiconducting type, with the eigenvalues E±((cid:126)k) = T 4((cid:126)k) ± T 0((cid:126)k). The above equation confirms that for t4 = 0 the spectrum becomes symmetric E = ±T 0((cid:126)k), with a direct gap at the Γ point equal to ∆ = 2T 0(0) = 2(2t1 + t2 + 2t3 + t5), which is Ezawa's result [5]. On the other hand, Eq.(8) proves that a nonvanishing t4 shifts the whole spectrum with 4t4, such that the electron-hole symmetry around E = 0 is lost. One concludes that the spectral asymmetry in phosphorene is the consequence of the hopping parameter t4, which connects sites of the same type and violates in this way the bipartitism of the lattice. The eigenvalues Eq.(8) are displayed in Fig.2, where three aspects have to be noticed: the presence of the gap, the strong anisotropy, and the electron-hole asymmetry of the bands. As explained in [2], the first two properties occur in the hexagonal-type lattice as soon as t1 (cid:54)= t2, even neglecting the other hopping parameters in the Hamiltonian. The Hofstadter spectrum generated by a perpendicular magnetic field is also very specific, consisting of two unequal butterflies separated by a gap. There is no agreement yet on the field dependence of the Landau levels, as in [11, 21] the dependence is linear, while in [22] the dependence is ∼ B2/3. spectrum of a finite (mesoscopic) plaquette, which exhibits a supplementary band in the In Fig.3 we show the numerically calculated Hofstadter middle accommodating the edge states [23]. The narrow width of the band, and the weak 7 FIG. 2: (Color online) The energy spectrum of the phosphorene lattice with periodic boundary conditions. The anisotropy of the spectrum around the Γ point can be observed: (a) the energy dispersion along the kx-axis shows the Dirac-like behavior, and (b) energy dispersion along the ky-axis shows the Schrodinger-like behavior. dependence on the magnetic field should be noticed. One has to observe that the spectrum misses the known periodicity with the magnetic flux E(Φ + Φ0) = E(Φ), which is met in the case of the 2DEG in perpendicular magnetic field. This comes from the presence of three different Peierls phases in the Hamiltonian (2). The case of the phosphorene finite plaquette will be discussed in Sec.III. The Hofstadter-type spectrum in Fig.3 suggests new physical properties of the edge states, and stimulates a more extensive study of the phosphorene mesoscopic systems. They are simulated in the tight-binding model by imposing vanishing boundary conditions for the wave function all along the perimeter (the case of the finite-size plaquette) or only along two parallel zig-zag edges (the ribbon case). The phosphorene ribbon in the absence of the magnetic field is discussed in [5, 24, 25], where it is shown that the zig-zag edges induce eigenvalues in the middle of the gap. The band is perfectly flat (i.e., independent of the momentum k) if t4 = 0, and get a slight dispersion otherwise. At a given k, there are two quasi-degenerate states which become perfectly degenerate in the limit of wide ribbons (similar to the case of graphene). The next section is devoted to the spectral properties of phosphorene ribbon in the presence of the magnetic field, in which case some new aspects of interest can be proved even analytically. 8 kxE(k)E(k)kyFIG. 3: (Color online) The Hofstadter spectrum of the finite phosphorene lattice. The quasi-flat band which accommodates topological edge states can be noticed in the gap. The number of lattice sites is 21 × 20, and the magnetic flux is measured in flux quanta h/e. III. SPECTRAL PROPERTIES OF THE PHOSPHORENE RIBBON IN MAG- NETIC FIELD Let us consider the Hamiltonian (2) and impose two edges parallel to the zig-zag chains at m=1 and m=M, but keeping periodic boundary conditions along the x-direction. The Fourier transform along the x-direction gives rise to the following Hamiltonian for the ribbon geometry (where k stands for kx): k (cid:88) M(cid:88) M−1(cid:88) M−1(cid:88) m=1 m=1 H = H 0(k) = + t2 H 4(k) = t4 H 0(k) + H 4(k), (cid:0)ei(φ1−k) + e−iφ1(cid:1)a † kmbkm † kmakm + Ebb † kmbkm + t1 Eaa a † km+1bkm + t3 M−2(cid:88) (cid:0)eiφ4B + e−i(φ4B−k)(cid:1)b M(cid:88) (cid:0)eiφ3 + e−i(φ3−k)(cid:1)a km+1bkm +(cid:0)eiφ4A + e−i(φ4A−k)(cid:1)a † km+2bkm + t5 m=1 m=1 † e−ika † km−1bkm + H.c. † km+1akm + H.c. . (9) m=1 In the case of vanishing magnetic field B = 0, the energy spectrum of the above Hamil- tonian is described by Ezawa [5]. The numerical calculation takes into account all the five hopping integrals, but the analytical one considers t3 = t5 = 0, while the parameter t4 is con- sidered perturbatively. The existence of a quasi-flat band in the gap, whose dispersion comes 9 M−1(cid:88) m=1 † km+1bkm + H.c. . t2a † kmbkm + H 0(k) = t1 M(cid:88) m=1 (cid:0)ei(φ1−k) + e−iφ1(cid:1)a M(cid:88) Ψ0(k)(cid:105) = (ξA (10) (11) from t4, is proved (see Eq.(22)in [5]). We reobtain this result, which is shown in Fig.5(left), in order to be compared with the case of nonvanishing magnetic field in Fig.5.(right) A. The quasi-flat band in magnetic field The aim of this subsection is to elucidate the effect of the perpendicular magnetic field on the spectral properties of the edge states in the ribbon geometry. The formation of the quasi-flat band in the middle of the gap and the interesting degeneracy lifting due to the magnetic field are put forward both numerically and analytically. Let us consider the atomic energies Ea = Eb = 0, and the hopping integrals t3 = t5 = 0 (as in Ref.5), but keep B (cid:54)= 0. Then, H 0(k) becomes: For any momentum k, we look for the eigenfunctions of H 0(k) as † km + ξB kmb km) 0(cid:105), † kma and, from H 0Ψ0(k)(cid:105) = E0(k)Ψ0(k)(cid:105), the equations satisfied by the coefficients ξA,B identified easily as: km can be m=1 t1(eiφ1 + e−i(φ1−k))ξA t1(e−iφ1 + ei(φ1−k))ξB km + t2ξA km+1 = E0(k)ξB km km + t2ξB km−1 = E0(k)ξA km , (12) with m = 1, .., M , and the ribbon-type boundary conditions ξB k,0 = 0, ξA k,M +1 = 0. We approach the study of the edge states in a way similar to the graphene case [26], i.e. assume the existence of a perfectly flat band in the middle of the spectrum E0(k) = 0, km. With the notation t1(m) = t1(e−iφ1 + ei(φ1−k)), and examine the properties of ξA km and ξB Eqs.(12) provide ξA k,m = ξA ∗ k,1(−t 1/t2)m−1, k,M (−t1/t2)M−m, ξB k,m = ξB (13) k,M can be obtained from the normalization condition. Since t1/t2 < 1, it k,m reaches its maximum value at the edge m = 1 and the minimum at the where ξA k,1 and ξB is obvious that ξA 10 A(k)(cid:105) =(cid:80) B(k)(cid:105) = (cid:80) k,m behaves oppositely. This means that Ψ0 km0(cid:105) † other edge m = M , while ξB m ξA kma describes an edge state localized near the edge m = 1, while Ψ0 km0(cid:105) is † kmb localized at the other edge m = M ; the two functions are obviously orthogonal. It is important to underline that, for a finite width ribbon, {ΨA(cid:105),ΨB(cid:105)} are only approximate eigenfunctions of H 0 (corresponding to the approximate eigenvalue E0(k) = 0 ); this is evident from the fact that the matrix element (cid:104)ΨAH 0(k)ΨB(cid:105) (cid:54)= 0 at any finite M . Indeed, using Eq.(13), a straightforward calculation yields: m ξB (cid:104)ΨAH 0(k)ΨB(cid:105) = −t2(−t1/t2)M ξ∗A k1 ξB kM , k ∈ (0, 2π]. (14) The above result indicates that the actual eigenfunctions describing the edge states in the finite ribbon consist of a superposition of the functions ΨA(cid:105) and ΨB(cid:105), the eigenvalues being E0±(k) = ±t2t1/t2M ξA∗ kM. Taking again into account the convergence condition t1/t2 < 1 (which in the case of phosphoene is ensured at any k), one notices that the +(k)− E0−(k) vanishes exponentially in the limit M → ∞. Only in this limit splitting δk = E0 E0(k) = 0 becomes a double-degenerate non-dispersive (flat) band, similar to the situation k1 ξB in graphene, but with the notable contradistinction that, for phosphporene, this property is true for any momentum k [27]. We checked also numerically the energy splitting δ as function of the ribbon width, and the exponential decay with increasing M is obvious in Fig.4. FIG. 4: The numerically calculated energy splitting, at the Γ point k = 0 and vanishing magnetic field B = 0, as function of the ribbon width M . The calculation takes into account all the five hopping parameters in the Hamiltonian (9). In what follows, we turn our attention to the contribution to the spectrum coming from the Hamiltonian H 4(k) in the presence of the magnetic field B. As already mentioned, the 11 FIG. 5: (Color online) The low energy spectrum of the phosphorene zig-zag nanoribbon. The degeneracy lifting of the quasi-flat band induced by the magnetic field can be noticed by comparing the two panels : in the left panel the magnetic flux is zero, while in the right panel Φ = 0.01Φ0. The red lines represent the numerically calculated spectrum for the width M = 71, the black lines represent the analytical results Eq. (17) and Eq.(18). case B = 0 was studied perturbatively in [5], where one proves that t4 (cid:54)= 0 generates the dispersion of the band, which thus becomes quasi-flat. In our calculation, we shall consider a large M and neglect the splitting δ (which is anyhow much smaller than the band dispersion). Then, the eigenvalues in the presence of the hopping t4 and of the magnetic field B (cid:54)= 0 will be given by the equation: (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)E− < ΨAH 4ΨA > < ΨAH 4ΨB > < ΨBH 4ΨA > E− < ΨBH 4ΨB > (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) = 0. (15) (16) (17) Since < ΨAH 4ΨB >= 0 the result reads: E1(k) =< ΨA(k)H 4ΨA(k) >= t4A(m)ξA∗ k,m+1ξA k,m + c.c. M(cid:88) M(cid:88) m=1 E2(k) =< ΨB(k)H 4ΨB(k) >= t4B(m)ξB∗ (cid:0)eiφ4A(m) + e−i(φ4A(m)−k)(cid:1), and a similar one for t4B(m). k,m + c.c., m=1 k,m+1ξB with the notation t4A(m) = t4 For zero magnetic flux, in the limit M → ∞, Eq.(16) yields Ezawa's result: E1(k) = E2(k) = −4 t4t1 t2 (1 + cosk), 12 (a)(b)saying that the levels remain degenerate but depend on k, such that they get a dispersion equal to 8t4t1/t2. However, the interesting case occurs at Φ (cid:54)= 0 when the degeneracy is lifted. The exact summation in Eq.(16) is difficult, so we approximate it by taking advantage of the strong localization of the coefficients ξA k1 and ξB kM at the edges m = 1 and m = M , respectively. Using also Eq.(13) one obtains: E1(k, Φ) ∼= t4A(1, Φ)(cid:0) − t1(1) (cid:1)ξA E2(k, Φ) ∼= t4B(M − 1, Φ)(cid:0) − t1(M ) t2 ∗ k,12 + c.c. (cid:1)ξB k,M2 + c.c. , t2 (18) with E1(k, Φ) (cid:54)= E2(k, Φ), indicating the degeneracy lifting due to the magnetic field. Figure 5 compares the quasi-flat spectrum in the absence (left panel) and presence (right panel) of the magnetic field applied on the ribbon. While the hopping t4 generates the dispersion, the magnetic field gives rise to the degeneracy lifting. The red lines represent the numerical result, who considers all hopping parameters t1, .., t5, while the black lines represent the analytical formulas Eq.(17) and Eq.(18) [28] calculated with t3 = t5 = 0. The fit being very good, one concludes that the hopping parameters t3 and t5 have negligible influence on the spectrum, at least in the energy range of the quasi-flat band. IV. QUANTUM TRANSPORT AND SUPERRADIANCE IN PHOSPHORENE MESOSCOPIC PLAQUETTE In order to investigate the transport properties in strong perpendicular magnetic field, we simulate the electronic Hall device by attaching four leads to a finite phosphorene plaquette (two leads for injecting/collecting the current, and two voltage probes), all the leads being contacted on the same zig-zag edge of the plaquette. The choice of such a lead configuration is essential, since it is the only one that can read out the current carried by the topological states located close to, and along the zig-zag edge. The electron transmission coefficients between different leads, the longitudinal, and the transverse resistance will be calculated as function of a gate potential Vgate (at a given Fermi energy in the leads) in the Landauer- Buttiker formalism in terms of Green functions. Depending on the position of the gate potential, different types of states become active in the transport process. Fig.3 shows the Hofstadter-type spectrum of the phosphorene 13 plaquette composed of two unequal butterflies, corresponding to the conductance and valence band. As usual, the Landau levels accommodate bulk states, while the gaps that separate consecutive LL are filled with chiral edge states, induced by the quantizing magnetic field, and running all around the plaquette perimeter. One may observe in Fig.3 that the chirality dE/dΦ of the edge states is opposite in the two bands, fact that causes the different sign of the quantum Hall effect in the corresponding energy ranges. FIG. 6: Ψ2 for a pair of quasi-degenerate edge states of a finite- size plaquette in perpendicular magnetic field. The number of lattice sites is 21 × 20, and the magnetic flux is Φ = 0.1Φ0. One has to remark the presence in the semiconducting gap of a narrow, practically dis- persionless band that accommodates also edge states, but of topological origin. They lie along the zig-zag edges only, exist also at B = 0, being the analogous of the edge states in the zig-zag ribbon discussed in the previous Section. It is important to underline that they do not get closed even if the magnetic field is applied, looking as in Fig.6. Figure 6 shows the superposition of two quasi-degenerate edge states located near the two (left and right) zig-zag boundaries. Any perturbation (as a small staggering EA (cid:54)= EB, impurity disorder or coupling to leads) lifts the superposition, and the wave functions become localized either on the left or right edge. Two other striking features of the topological edge states on the plaquette will be proved here: i) the dissipative character, and ii) the splitting of the density of states, and the formation of minibands if the finite system is opened by attaching contacts; this behavior can be interpreted as a superradiance effect. 14 A. The effective Hamiltonian and transport formalism In order to calculate the transport quantities (namely, the longitudinal and transverse resistance) one needs to attach four leads to the finite-size plaquette. Then, the Hamiltonian of the entire system reads: H = H S + H L + τ H LS, (19) where the first term is the Hamiltonian (2) of the phosphorene plaquette, the second term represents all the four semi-infinite leads (also in the tight-binding description), and the last one describes the coupling between the plaquette and the leads. The longitudinal and Hall resistances will be calculated as function of a gate potential Vgate applied on the plaquette, similar to the experimental measurement, where Vgate is simulated by a diagonal term in the Hamiltonian H S. A powerful tool to deal with such an open system is the formalism of the effective Hamil- tonian, which is obtained by removing the degree of freedom of the leads, with the price of losing the Hermicity: H S ef f = H S + τ 2 tL e−ik(cid:88) α α(cid:105)(cid:104)α , (20) where tL is the hopping parameter of the tight-binding model for the leads, k parametrizes the energy in the leads, E = 2tLcosk, and {α(cid:105)} are those localized states that correspond to the sites on the plaquette where the leads are sticked to the sample [29]. The difference between Hamiltonians Eq.(19) and Eq.(20) is just formal, and they are completely equivalent. The deduction of the effective Hamiltonian can be found for instance in Ref.6. After constructing the matrix of the effective Hamiltonian in the representation of the localized functions {nm(cid:105)}, one may calculate immediately the Green function G(E) = (E − H S ef f )−1, which enters the Landauer-Buttiker formula for the transmission coefficients: Tαβ = 4τ 4Gαβ2ImgL α ImgL β , α (cid:54)= β , (21) where α and β (α, β = 1, .., 4) are lead indexes, and Im gL α represents the density of states in the lead α. The transmission coefficients Tαβ being known, the conductance matrix voltages Vβ, Iα = (cid:80) diagonal conductance gαα results from the current conservation rule (cid:80) gαβ, which connects all the currents Iα passing through the system to the corresponding β gαβVβ, can be obtained as gαβ = (e2/h)Tαβ for α (cid:54)= β, while the α gαβ = 0. Finally, 15 FIG. 7: The sketch of a four-lead Hall device. All leads are connected to the same zig-zag edge. the Landauer-Buttiker formalism provides the following expressions for the quantities of interest (longitudinal and Hall resistance), which are to be calculated numerically: RL = R14,23 = (g24g31 − g21g34)/D, RH = (R13,24 − R24,13)/2 = (g23g41 − g21g43 − g32g14 + g12g34)/2D, (22) where D is any 3 × 3 subdeterminant of the conductance matrix. Since the experimental curves show the conductance instead of the resistance, we shall do the same, and show in Fig.8 the the Hall and longitudinal conductance calculated as GH = RH/(R2 GL = RL/(R2 H + R2 L). H + R2 L) and B. Quantum Hall effect in phosphorene In what concerns the Hall conductance in the quantum regime, there are significant new aspects in comparison with the graphene. First, one has to notice the large plateau GH = 0 that corresponds to the central gap. Next, one notice the lack of the valley degeneracy in the low energy range, such that the quantum Hall plateaus are the conventional (spinless) plateaus n = ±1,±2, .. in units e2/h, the same as for the two-dimensional electron gas (2DEG) subject to a perpendicular magnetic field. As a specific feature, one may notice in Fig.8 that the lengths of the plateaus in the positive and negative regions are slightly different, as a manifestation of the spectral asymmetry discussed in Sec.II. The quantum plateaus are supported obviously by the chiral edge states existing in the Hofstadter spectrum of the finite-size plaquette, however one should not forget that the central gap contains also topological edge states bunched in the quasi-flat band. The value GH = 0 everywhere in the gap confirms that these edge states are non-chiral, and do not support the QHE. Recall that, in terms of transmission coefficients, the chirality means 16 1243zig−zag edgeFIG. 8: (Color online) Numerically calculated Hall (red line) and longitudinal (black line) conduc- tance in the quantum Hall regime as a function of the gate potential Vgate, for a slightly disordered sample. The longitudinal conductance shows a series of peaks in the range of the quasi-flat band. The number of lattice sites is 107 × 40, the magnetic flux is φ/φ0 = 0.1, and the Fermi energy in the leads is EF = 0. Tα,α+1 = integer, while Tα,α+1 = 0 (for any lead α and given direction of the magnetic field). On the other hand, in the spectral range occupied by the quasi-flat band, we find the symmetry Tα,α+1 = Tα,α+1, which denotes the lack of chirality. This property was observed numerically using Eq.(21), and occurs no matter the presence or absence of the magnetic field. The longitudinal conductance GL exhibits the non-dissipative behavior in the range of the quantum plateaus, as it should, but striking non-trivial properties are proved in the range [-0.3,0] covered by the quasi-flat band, where the longitudinal conductance is non-vanishing and shows a sequence of peaks. While the dissipative character of the non-chiral edges states was also met in the context of the zero-energy Landau level in graphene [30], we think that the peaked structure of the longitudinal conductance is specific to phosphorene. It is already known that that the flat bands are sensitive to disorder due to their degener- acy [31, 32], and one may expect that the GL peaks are also affected by the disorder existing in the system, which is unavoidable experimentally. Indeed, the unitary limit GL = 1e2/h is reached only in the clean systems (this case is shown in Fig.9a), but any small amount of disorder allows for the backscattering and slightly lowers the values of the peaks below 17 the unitary limit. This is the case in Fig.8, where small Anderson (diagonal) disorder was introduced in the numerical calculation. In what follows we shall pay closer attention to properties of the non-chiral edge states in the quasi-flat band. C. Spectral and transport properties of the quasi-flat band in open system It is obvious that the second term of the effective Hamiltonian Eq.(20) produces shifts of the real eigenvalues of H S, but adds also an imaginary part, meaning the level broadening due to the coupling to the leads. As long as the coupling τ is very small (i.e., we are in the resonant tunneling regime, which was studied for the nanoribbon system in [33]), all the eigenvalues of H S should be practically recovered. However, with increasing coupling, the level broadening Γ increases too, and the merging of neighboring levels occurs. Consequently, the shape of the density of states changes significantly. When Γ ∼ ∆ (∆ = mean interlevel distance) one enters the regime known as superradiative [34]. As we already mentioned the superradiance phenomenon consists in the overlapping and segregation of eigenenergies occurring in open systems under the control of the coupling between the finite system and the infinite reservoir. One may expect that the energy spectrum is not everywhere equally sensitive to this effect, and one may speculate that the energy range occupied by the states located near edges (where the leads are attached) is mostly affected. We assume that the superradiance is the mechanism that gives rise to the miniband structure of the quasi-flat band shown in Fig.9a, where the density of states (DOS = − 1 eigenvalues of the effective Hamiltonian (20). π T rG) exhibits eleven peaks. The confirmation comes from the calculation of the complex In Fig.9b we show the real and imaginary part of the eigenvalues in the energy range of the quasi-flat band, and find the presence of eleven energies with large imaginary part, which perfectly correspond to the positions of the minibands in the density of states. One has to observe in Fig.9b also the multitude of eigenstates with vanishing imaginary part (ImE = 0). They correspond to the edge states localized along the edge opposite to that one where the leads are connected. In other words, the process of overlapping and segregation affects only those edge states that are in the immediate vicinity of the leads, the other ones remaining unchanged. As a next step, we draw the attention to the fact, visible in Fig.9a, that not all DOS 18 FIG. 9: (Color online) (a) The peaked structure of the transmission coefficient T12 (red line) and the density of states (black line) of the phosphorene plaquette in the energy range of the quasi-flat band for a clean system. Note that not all the DOS peaks are transmitting, and also that the unitary limit of the transmission is reached. (b) ImE vs ReE for the eigenvalues of the effective Hamiltonian (20) corresponding to the quasi-flat band. Note that the eigenvalues with ImE (cid:54)= 0 correspond to the miniband structure in the panel (a). The number of lattice sites is 107 × 40, the magnetic flux is Φ = 0.1Φ0, the Fermi energy in the leads is EF = 0, and τ = 2eV . peaks support the electron transmission. This apparently surprising result has a simple explanation in terms of the charge distribution on the plaquette, described by the local density of states. The local density of states, calculated at each site i on the plaquette as the imaginary part of the Green function LDOSi(E) = − 1 π ImGii(E), shows that, in the case of the conducting minibands, the states are located between the contacts, but, for the non-conducting ones, the states are positioned outside the contacts. The two situations are displayed in Fig.10, The same figure tells furthermore that the states accommodated by 19 0.00-0.05-0.10-0.15-0.20-0.25-0.300.00.20.40.60.81.0 T12Vgate(a)05001000150020002500300035004000Dos (arb. units)0.00-0.05-0.10-0.15-0.20-0.25-0.30-0.0025-0.0020-0.0015-0.0010-0.00050.0000 Im ERe E(b)the minibands do not get closed around the whole perimeter of the plaquette, even at such strong magnetic fields that generates chiral edge states in the bands. FIG. 10: (Color online): The local density of states (LDOS) in the presence of the leads: (a) the density of states is localized in-between the leads and contribute to transmission (corresponding to the peak at Vgate = −0.0978eV in Fig.9a), and (b) the density of states is localized outside the leads and corresponds to the non-transmitting DOS peak at Vgate = −0.10735eV in Fig.9a. The parameters are the same as in Fig.9. In what concerns the transport properties, besides the lack of chirality mentioned above, we find that the transmission exhibits a peaked structure and reaches the unitary limit Tαα+1 = Tα+1α = 1 in the middle of the conducting minibands. This behavior of the transmission coefficients is proved by numerical investigation using Eq.(21), and it is shown in Fig.9a. (Of course, the unitary limit, telling that each miniband behaves as a perfect one- dimensional channel, is reached only for disorder-free systems.) All the other transmission coefficients vanish, so that the whole 4 × 4 conductance matrix reads as: (23)  0 g = e2 h 0  , 1 −1 0 1 −2 1 0 0 1 −2 1 1 −1 0 and allows for the calculation of the Hall (RH) and longitudinal (RL) resistances. Indeed, by the use of Eq.(22), one obtains the results already known from the numerical calculation. For the Hall resistance one gets RH = 0, which is the outcome of the lack of chirality, however, a non-trivial result is obtained for the longitudinal response, for which the above matrix yields RL = 1h/e2, indicating a dissipative character of the electron transport in minibands. It is to underline that this distinctive property of the quasi-flat band occurs even in the 20 LDosLDos(b)(a)presence of a strong magnetic field, which, otherwise, is able to generate in the other bands the specific QHE behavior, i.e., quantized non-zero values of RH, and non-dissipative RL. V. SUMMARY AND CONCLUSIONS In this paper we have studied spectral and transport properties of phosphorene, paying special attention to confined systems (zig-zag ribbon and mesoscopic plaquette) subject to a magnetic field, with main focus on the topological edge states organized in the quasi-flat band. Our results are the following: We approach analytically the question of electron-hole symmetry breaking, and demon- strate the role played in this respect by the hopping integral t4, the only parameter in the tight-binding model that violates the bipartitism of the lattice. The Hofstadter-type spectrum of the phosphorene plaquette misses the usual periodicity E(Φ) = E(Φ + Φ0) because three different Peierls phases (depending also on the quasi-2D lattice angle β) are assigned to different hopping terms. The Hofstadter spectrum comprises edge states of chiral and topological origin. The chiral states fill the gaps between the Landau levels and extend all around the perimeter. The other ones extend along the zig-zag edges only, and remain like that even in strong magnetic field. The topological states are bunched in a quasi-flat band located in the middle of the gap. For the zig-zag ribbon, since t1/t2 < 1, we prove that the topological edge states occur at any momentum k in the Brillouin zone, contrary to the graphene case. We analytically show that the degeneracy of a pair of edge states, located at opposite edges of the ribbon, occurs only in the limit of the infinite wide ribbon (M → ∞), and we prove also that the degeneracy is lifted by the magnetic field. The quantum transport in the mesoscopic plaquette is treated numerically. The Hall device may use different lead configuration, however the configuration 'all leads on the same edge' (Fig.7) is that one that evidence better the features of the edge states. We suggest such a configuration for an eventual experimental study of the topological edge states. Specific to phosphorene, the Hall conductance shows a zero plateau in the gap, indicating the non- chiral behavior of the quasi-flat band, but a non-zero longitudinal conductance, indicating the dissipative character. The multiple-peak aspect of GL reflects the miniband structure of the density of states in the presence of the leads. 21 For the energy range occupied by the quasi-flat band, we calculate the transmission coefficient between consecutive leads, the DOS of the plaquette when connected to the leads, and the complex eigenenergies of the non-Hermitian effective Hamiltonian. The ensemble of these quantities (shown in Fig.9), which are controlled by the dot-lead coupling parameter τ , certifies the manifestation of the superradiance phenomenon in phosphorene. We underline that not all the minibands in the DOS are conducting (Fig.9a), the issue being explained by Fig.10. We mention the features of the electron transmission, which besides the lack of chirality, exhibits unitary peaks in the case of the clean system, proving a one-channel- type transport. In what concerns the disorder effects, it is clear that the different quantum states respond differently to disorder. While the chiral states are robust, one expects the topological edge states be sensitive due to their quasi-degeneracy. We think that the localization, level spacing analysis, and the electron transmission are topics of interest, which are however beyond the aim of this paper. In conclusion, phosphorene is a 'beyond' graphene material which, besides potential ap- plications as a semiconductor, shows several interesting conceptual properties, mainly con- cerning the topological edge states accommodated in the quasi-flat band. VI. ACKNOWLEDGMENTS We are grateful to Antonio Castro Neto for illuminating discussions on the phosphorene issue, and to Marian Nita for helpful comments. We acknowledge financial support from PNII-ID-PCE Research Programme (grant no 0091/2011) and Romanian Core Research Programme. [1] A. N. Rudenko and M. I. Katsnelson, Phys. Rev. B 89, 201408(R) (2014). [2] G. Montambaux, F. Piechon, J.-N. Fuchs, and M. O. Goerbig, Phys. Rev. B 80, 153412 (2009). [3] M.O. Goerbig, Rev. Mod.Phys. 83, 1193 (2011). [4] A.Mielke, Phys.Lett. A 174, 443 (1993). [5] M. Ezawa, New J. Phys. 16,115004 (2014). [6] V. Moldoveanu, A. Aldea, A. Manolescu, and M. Nita, Phys. Rev. B 63, 045301 (2000). 22 [7] N. Hatano and D. R. Nelson, Phys.Rev.Lett. 77, 570 (1996). [8] I. Ya. Goldsheid and B. A. Khoruzhenko, Phys. Rev. Lett. 80, 2897 (1998). [9] G. L. Celardo and L. Kaplan, Phys. Rev. B 79, 155108 (2009). [10] A. I. Nesterov, F. Aceves de la Cruz, V. A. Luchnikov, and G. Berman, Phys. Lett. A 379, 2951 (2015). [11] X. Y. Zhou, R. Zhang, J. P. Sun, Y. L. Zou, D. Zhang, W. K. Lou, F. Cheng, G. H. Zhou, F. Zhai, and K. Chang, Scientific Reports 5,12295 (2015). [12] L. Li, F. Yang, G. J. Ye, Z. Zhang, Z. Zhu, W. K. Lou, L. Li, K. Watanabe, T. Taniguchi, K. Chang, Y. Wang, X.H. Chen, and Y. Zhang, arXiv:1504.07155 [cond-mat.mes-hall]. [13] L. Li, G. J. Ye, V. Tran, R. Fei, G. Chen, H. Wang, J. Wang, T. Taniguchi, Li Yang, X. H. Chen, and Y. Zhang, Nature Nanotechnol. 10, 608 (2015). [14] H. Liu, A. T. Neal, Z. Zhu, Z. Luo, X. Xu, D. Tomnek, and P. D. Ye, ACS Nano 8, 4033 (2014). [15] A. Mishchenko, Y. Cao, G. Yu, C. R. Woods, R. V. Gorbachev, K. S. Novoselov, A. K. Geim, and L. Levitov, Nano Lett. 15, 6991 (2015). [16] A. S. Rodin, A. Carvalho, and A. H. Castro Neto, Phys. Rev. Lett. 112, 176801 (2014). [17] R. Fei and L. Yang, Nano Lett. 14, 2884 (2014). [18] Q. Wu, L. Shen, M. Yang, Y. Cai, Z. Huang, and Y. P. Feng, Phys. Rev. B 92, 035436 (2015). [19] L. Li, Y. Yu, G. J. Ye, Q. Ge, X. Ou, H. Wu, D. Feng, X. H. Chen, and Y. Zhang, Nature Nanotechnol. 9 , 372 (2014). [20] The particular expression of the phases depend on the choice of the origin; in our case they correspond to the choice in Fig.1 [21] J. M. Pereira Jr. and M. I. Katsnelson, Phys. Rev. B 92, 075437 (2015). [22] Motohiko Ezawa, Journal of Physics: Conference Series 603, 012006 (2015). [23] R. Zhang, X.Y. Zhou, D. Zhang, W. K. Lou, F. Zhai, K. Chang, arXiv:1507.03808 [cond- mat.mes-hall] [24] E. Taghizadeh Sisakht, M. H. Zare, and F. Fazileh, Phys. Rev. B 91,085409 (2015). [25] A.Carvalho, A. S. Rodin and A.H. Castro Neto, EPL, 108, 47005 (2014). [26] K. Wakabayashi, M. Fujita, H. Ajiki, and M. Sigrist, Phys. Rev. B 59, 8271 (1999). [27] We remind that in the graphene ribbon, since t1 = t2, the convergence, and, consequently, the flat band occur only for k comprised between the points K and K(cid:48) in the Brillouin zone [26] . 23 [28] In Fig.5(right), the black lines represent Eq.(18), where ξA kM = 1 has been considered. [29] One has to keep in mind the technical detail that each lead α = 1, ., 4 is composed of sev- k1 = ξB eral one-dimensional independent channels, and that in Eq.(20) the summation over all the channels is implicit. The same convention is assumed also in Eq.(21). [30] D. A. Abanin, K. S. Novoselov, U. Zeitler, P. A. Lee, A. K. Geim, and L. S. Levitov, Phys. Rev. Lett. 98, 196806 (2007). [31] M. Nita, B. Ostahie, and A. Aldea, Phys. Rev. B 87, 125428 (2013). [32] D. Leykam, S. Flach, O. Bahat-Treidel, and A. S. Desyatnikov, Phys. Rev. B 88, 224203 (2013). [33] J. Paez, D. A. Bahamon, Ana L. C. Pereira, and P. A. Schulz, arXiv:1506.03093 [cond- mat.mes-hall]. [34] A. Ziletti, F. Borgonovi, G. L. Celardo, F. M. Izrailev, L. Kaplan, and V. G. Zelevinsky, Phys. Rev. B 85, 052201 (2012). 24
1207.7282	2	1207	2012-08-06T17:33:24	Numerical studies of the fractional quantum Hall effect in systems with tunable interactions	[ "cond-mat.mes-hall", "cond-mat.str-el" ]	The discovery of the fractional quantum Hall effect in GaAs-based semiconductor devices has lead to new advances in condensed matter physics, in particular the possibility for exotic, topological phases of matter that possess fractional, and even non-Abelian, statistics of quasiparticles. One of the main limitations of the experimental systems based on GaAs has been the lack of tunability of the effective interactions between two-dimensional electrons, which made it difficult to stabilize some of the more fragile states, or induce phase transitions in a controlled manner. Here we review the recent studies that have explored the effects of tunability of the interactions offered by alternative two-dimensional systems, characterized by non-trivial Berry phases and including graphene, bilayer graphene and topological insulators. The tunability in these systems is achieved via external fields that change the mass gap, or by screening via dielectric plate in the vicinity of the device. Our study points to a number of different ways to manipulate the effective interactions, and engineer phase transitions between quantum Hall liquids and compressible states in a controlled manner.	cond-mat.mes-hall	cond-mat	2 Numerical studies of the fractional quantum Hall 1 0 2 effect in systems with tunable interactions g u A 6 ] l l a h - s e m . t a m - d n o c [ 2 v 2 8 2 7 . 7 0 2 1 : v i X r a Z. Papi´c1, D. A. Abanin2, Y. Barlas3, and R. N. Bhatt1,2 1 Department of Electrical Engineering, Princeton University, Princeton, NJ 08544, USA 2 Princeton Center for Theoretical Science, Princeton University, Princeton, NJ 08544, USA 3 National High Magnetic Field Laboratory and Department of Physics, Florida State University, FL 32306, USA Abstract. The discovery of the fractional quantum Hall effect in GaAs-based semiconductor devices has lead to new advances in condensed matter physics, in particular the possibility for exotic, topological phases of matter that possess fractional, and even non-Abelian, statistics of quasiparticles. One of the main limitations of the experimental systems based on GaAs has been the lack of tunability of the effective interactions between two-dimensional electrons, which made it difficult to stabilize some of the more fragile states, or induce phase transitions in a controlled manner. Here we review the recent studies that have explored the effects of tunability of the interactions offered by alternative two-dimensional systems, characterized by non-trivial Berry phases and including graphene, bilayer graphene and topological insulators. The tunability in these systems is achieved via external fields that change the mass gap, or by screening via dielectric plate in the vicinity of the device. Our study points to a number of different ways to manipulate the effective interactions, and engineer phase transitions between quantum Hall liquids and compressible states in a controlled manner. 1. Introduction Strongly correlated phases of electrons confined to move in the plane, subject to a perpendicular magnetic field, have attracted significant attention since the discovery of fractionally quantized Hall conductivity [1]. The profound role of topology in this extreme quantum limit leads to the presence of quasiparticles that carry a fraction of electron charge [2], and fractional (Abelian, or possibly non-Abelian) statistics [2, 3, 4, 5, 6, 7]. The prospect of excitations possessing non-Abelian statistics has motivated different schemes for topological quantum computation [8] based on these systems. These remarkable phenomena occur in the fractional quantum Hall (FQH) regime, when the number of electrons, Ne, is comparable to the number of magnetic flux quanta NΦ through the two-dimensional electron system (2DES). Correlated FQH liquid states appear at certain partial filling ν = Ne/NΦ of the "active" Landau level (LL). In traditional semiconductor heterostructures, the physics of a partially-filled n = 1 LL differs significantly from that of n = 0 LL, due to the node in the single-particle wavefunction [9]. As a consequence, the hierarchy/composite fermion states [10, 11], ubiquitous in the lowest Landau level (LLL), are significantly weakened in n = 1 LL, and some of the more exotic states, such as the Read-Rezayi (RR) parafermion states [12], are likely to be favored. A number of studies have focused on the simplest, k = 2 non-Abelian member of the RR sequence -- the Moore-Read (MR) "Pfaffian" state [5], believed to describe the FQH plateau at ν = 5/2 [13]. Quasiparticles of the MR and higher RR states obey the non-Abelian statistics [5] which is of interest for topological quantum computation [8]. One of the main disadvantages of the GaAs-based devices is that their 2DES is buried inside a larger, three-dimensional structure. This unfortunate fact fixes the effective interactions at values that are often not optimal for some of the most interesting FQH states, including the RR series. For example, the MR state is found to lie very close to the boundary with a compressible phase [14, 15]. Another problem stems from strong dielectric screening and finite well-width [16] in GaAs, which weaken the electron-electron interactions and make FQH states fragile. This has been a major obstacle in the studies of the non-Abelian states, which could only be observed in ultra-high-mobility samples [13]. Thus, it is desirable to find an alternative high-mobility 2DES with strong effective Coulomb interactions that are adjustable in a broad range. Recently, a new class of such high-mobility 2DES which host chiral excitations with non- trivial Berry phases, has been discovered. These "chiral" materials include graphene and bilayer graphene [17], and, more recently, topological insulators [18], as well as certain quantum wells [19]. The chiral nature of the quasiparticles gives rise to new electronic properties, including an unusual LL sequence, the anomalous Hall effect, and suppression of weak localization [17]. When these chiral materials are subject to a perpendicular magnetic field, the kinetic energy quenches into discrete Landau levels, similar to the usual semiconductors with non-chiral carriers. However, due to the chiral band structure [20] and the fact that the surface of these materials is exposed [21], they offer new possibilities to tune the effective interactions and explore strongly correlated phases. In this paper we review two practical ways [21, 20] of tuning the effective interactions in chiral 2DES, and the effect this has on the FQH states. One of the attractive features of the chiral 2DESs is that they allow for a more robust realization of certain FQH states in several LLs, as opposed to a single LL in GaAs. Additional insights can be obtained by driving transitions between the incompressible (FQH) states and the compressible, stripe and bubble, phases. Such transitions can be implemented in chiral and massive 2DESs by varying the external field. Overall, the tunability of the chiral 2DES allows one to explore a larger region of the effective interactions than has been achieved in GaAs. 2. Model We begin this Section by providing a brief overview of the one-body problem in a magnetic field, as well as the summary of the standard recipe of exact diagonalization applied to quantum Hall systems. Consider a single electron moving in a plane, subject to a perpendicular magnetic field B z = ∇ × A(r). The corresponding Hamiltonian can be written in a "covariant" notation as K = 1 2m gabπaπb, (1) where πa = pa − e c Aa(r) (a = x, y) represents the dynamical momentum, and g is the band mass tensor. In the usual isotropic case (g = I), we can obtain the single-particle energies (Landau levels) by choosing, for example, the symmetric gauge Ax = By/2, Ay = −Bx/2. In this case, the dynamical momenta become πx = −i ∂ x, in terms of the magnetic length ℓB = p/eB. Hamiltonian can be transformed into a diagonal form 2(cid:1) with the help of "ladder" operators a ∝ πx + iπy, a† ∝ πx − iπy. y and πy = −i ∂ 2 (cid:0)a†a + 1 ∂x + 2ℓ2 ∂y − 2ℓ2 B (2) (3) K = ωc B However, for each value of a†a, there remains a residual degeneracy equal to the number of magnetic flux quanta, NΦ. This degeneracy is resolved by another pair of operators b, b† that commute with a, a† and depend on the guiding center coordinates of the electron, Ra = ra − ǫab B. Operators b† create the (unnormalized) single particle eigenstates of the πbℓ2 lowest, n = 0 Landau level (LLL), φl(z) ∝ zle−z ∗z/4ℓ2 B , (4) with z = x + iy being the complex coordinate of an electron in the plane (and z∗ denoting its complex-conjugate). The quantum number l is an eigenvalue of the angular momentum, and the single particle states are localized along concentric rings around the origin. Single-particle states in higher LLs are constructed by the repeated action of a†, thus completing the solution of the one-body problem in a rotationally-invariant case. 2.1. Many-body problem: Exact diagonalization An interacting problem of a finite number of electrons can be studied numerically, using exact diagonalization which has been remarkably successful in unravelling many of the essential physical properties of FQH systems, for systems as small as 10 electrons [9, 10]. A crucial approximation that makes exact diagonalization practical is to neglect the excitations between LLs and restrict the Hilbert space to a single, "active" LL, which has a degeneracy of NΦ. This approximation is physically justified in high magnetic fields where such excitations indeed become very costly. One can then construct a many-body Hilbert space, consisting of Slater determinants l1, l2, . . .i built from the states (4), using a suitable choice of boundary condition. In the study of the FQH effect, two kinds of surfaces are available that preserve the translational invariance of an infinite 2DES: sphere [10] and torus [22]. The two choices of boundary conditions illustrate the specific features of a many-body FQH state under investigation: on a sphere, the FQH state couples to the curvature of the manifold, which is characterized by the topological number called shift [23]. The shift produces a small offset between NΦ = N/ν and the magnetic monopole, whose strength is denoted by 2S, placed in the center of a sphere. As a consequence, different FQH states "live" in different Hilbert spaces and in principle can be directly compared only after extrapolation to the thermodynamic limit. On the other hand, the flat surface of a torus leads to a unique definition of NΦ for given N and ν, such that different candidate states describing the same filling ν are all realized in the same Hilbert space. The caveats of this geometry are the additional geometric parameters, the angle and aspect ratio of the torus; the Hamiltonian of a finite-size system depends on these parameters and their specific values favor one FQH phase over the others. Thus, an analysis is slightly more involved but the gain is that the ground-state degeneracy can be used to identify topologically-ordered states. In contrast, on the sphere FQH states always appear as non-degenerate, zero angular momentum ground states. Once the Hilbert space is defined and properly adapted to the symmetry group/boundary condition, the remaining task is to represent the Hamiltonian and diagonalize the corresponding matrix using a variant of Lanczos algorithm [24] (or Jacobi/Householder routine [25] for very small systems). The advantage of symmetry, in particular rotational invariance, is that Wigner- Eckart theorem applies, and leads to the definition of the so-called Haldane pseudopotentials [9]. Any two-body interaction, such as Coulomb potential projected to a single LL, can be written in m2 cm4cm3 . Furthermore, assuming the usual second-quantized form H = P{mi} Vm1m2m3m4 c† rotational invariance, the matrix element factorizes into m1c† Vm1m2m3m4 = 2S XL=0 VL L XM =−L C SSL m1m2M C SSL m3m4M , (5) into a product of a geometrical factor involving the Clebsch-Gordan coefficients C J1J2J i.e. m1m2m, and the Haldane pseudopotential VL. The finite set of Haldane pseudopotentials contains all the information about the interaction projected to a single LL and thus determines all the physical properties of a FQH system. Haldane pseudopotentials can be conveniently evaluated from the Fourier transform of the interaction, V (q). 2.2. Modification of the interaction due to the spinor nature of the wavefunction in massive Dirac materials Once the Hilbert space is defined and symmetry-adapted, the Coulomb Hamiltonian can be represented in the matrix form. In the LLL, the interaction is simply given by V (q) = 1/q, which is the Fourier transform of the Coulomb potential in two dimensions. However, if we consider the case when the chemical potential is such that LLs from 1 through n− 1 are completely filled and "inert", while the nth LL is partially filled and "active", the effective interaction becomes slightly modified and assumes the form 1/q × Fn(q)2, where the form factor [9] Fn(q) = Ln(q2ℓ2 B/2) can be evaluated using the algebra of operators a, a†, Eq. (2) -- see, e.g., Ref. [11] (Ln is the nth Laguerre polynomial). Similarly, if the 2DES has a finite extent in the z-direction, the interaction becomes modified via an appropriate form-factor that results from integrating out the z-component of the wavefunction. The modification of the effective interaction may also result from the spinor nature of the wavefunction [26]. Such wavefunctions arise naturally in a family of 2D materials that have non- trivial Berry phases. One such material is monolayer graphene, a high-mobility atomically thick 2DES [17], where recently several FQH states of the type ν = m/3 have been discovered [27]. A closely related material, bilayer graphene [17] has similarly high mobility, and exhibits interaction-induced quantum Hall states at integer filling factors at low magnetic fields [28]. Graphene and its bilayer are characterized by Berry phase π (graphene-like) and 2π (bilayer- graphene-like with an energy gap), respectively. More explicitly, we refer to the Dirac materials as those described by a family of 2× 2 Hamiltonians with Berry phase π and 2π [20]. The case of π-carriers is realized in graphene, topological insulators, special quantum wells [19]; the case of 2π-carriers occurs in bilayer graphene, where the energy gap can be controlled in a wide range by a perpendicular electric field [17]. Details on how the effective 2 × 2 Hamiltonian can be derived from a more complete tight-binding model can be found in the review [29]. the single-particle wavefunctions are given by spinors [20] ψn = (cos θnφn−1, sin θnφn), where φn is the wave function of the nth non-relativistic LL (n = ±1,±2, . . .), and parameter θ depends on the ratio ∆/(v0/ℓB), where ∆ is the mass gap, and v0 the Fermi velocity. We use the notation (σπ, n) to denote the nth LL for σπ carriers (σ = 1, 2). As a consequence of the spinor wavefunction, the effective form factor [9] F π n (q) that describes the interaction projected to a (π, n) LL is given by For π carriers, F π n (q) = cos2 θLn−1(cid:18) q2ℓ2 2 (cid:19) + sin2 θLn(cid:18) q2ℓ2 2 (cid:19) , B B (6) where for simplicity we omitted the index of θ. The form-factor is a mixture of the (n − 1)th and nth LL form-factors in a non-relativistic 2DES with parabolic dispersion. At θ = π/4, the above equation reduces to the form-factor of graphene [26], however, by varying ∆/(v0/ℓB) in the experimentally accessible range [20], one can realize any value of θ ∈ (0; π/2). Similarly, for carriers with Berry phase 2π, the single-particle wavefunctions are ψn = (cos θnφn−1, sin θnφn+1), and the form-factor is a mixture of standard (n−1)th and (n+1)th form-factors, F 2π n (q) = cos2 θLn−1(cid:18) q2ℓ2 2 (cid:19) + sin2 θLn+1(cid:18) q2ℓ2 2 (cid:19) . B B (7) (a) V(q) V q 2.0 1.5 1.0 0.5 0 ΑΑ0 -0.5 ΑΑΑ0.5 0.5 ΑΑ0.5 (b) V(q) V q 1.0 0.8 0.6 0.4 0.2 0.2 cos 2ΘΘ0.2 0.5 cos 2ΘΘ0.5 0.8 cos 2ΘΘ0.8 2 1 1 2 q x q x 4 2 2 q x Figure 1. (Color online) Two ways of modifying the effective interaction V (q) in the FQH regime: (a) via dielectric screening, for several values of α given in the inset (using Eq.(8)) for d = ℓB, (b) via chiral band structure, for several values of cos2 θ given in the inset (in (π, 1) LL). Because the interaction V (q) is isotropic, it is plotted as a function of qx only. The comparison of the corresponding Haldane pseudopotentials for cases (a) and (b) can be found in Ref. [30]. The tunable form of the effective interactions Eqs.(6,7) provides a way to engineer transitions between strongly correlated phases in situ by changing the field, as illustrated in Fig. 1(b). 2.3. Modification of the effective interaction due to dielectric screening Up to this point, we have discussed the change in the effective interaction V (q)F (q)2 that resulted from a modification of the single-particle wavefunctions, and therefore F (q). The method, proposed in Ref. [21], allows one to directly change V (q). Chiral materials, such as graphene, are often exposed to the environment, which allows for dielectric material to be deposited on top of them. We consider a setup where graphene sample is situated in a dielectric medium with permittivity ǫ1, and a semi-infinite dielectric plate with permittivity ǫ2 6= ǫ1 is placed at a distance d/2 away from the graphene sheet. The effective interactions between electrons in graphene change due to the surface charges induced at the boundary between dielectrics [21]: V (r) = e2 ǫ1r + α e2 ǫ1√r2 + d2 , where α = ǫ1 − ǫ2 ǫ1 + ǫ2 . (8) The ratio d/ℓB controls the effective interactions within a partially filled LL. However, the overall energy scale is also modified and this has an impact on the magnitude of the excitation gap. The gap should be multiplied by a factor ǫGaAs/ǫ1 if comparison is to be made with GaAs 2DES. Again, an important advantage of this setup is that the interactions can be tuned in situ by varying the magnetic field B, which modifies the ratio d/ℓB. The consequences for the many-body system are illustrated in Fig. 1(a). 3. Results We now present some exact-diagonalization results for the interaction models introduced in Sec.2. The focus is on the physical features resulting from the variation of the effective interaction; as stated in the introduction, we follow the traditional assumption of neglecting the excitations between different LLs, which is expected to be quite reasonable in GaAs, but perhaps less so in chiral materials. Much of previous theoretical work on chiral materials has been restricted to graphene, exploring the consequences of the four-fold LL degeneracy (valley and spin) that leads to new SU(2) and SU(4)-symmetric fractional states [31]. Instead, we focus on the high-field limit, neglecting the multicomponent degrees of freedom, and examine the effects originating from the interplay of the Coulomb interaction and band structure. We consider a family of (a) E ∆ 0.1 0.08 0.06 0.04 0.02 0 0 (b) SL SL 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 -0.1 0 0.2 0.2 0.4 0.6 0.8 1 2 cos θ 0.4 2 cos θ 1 0.8 0.6 0.4 0.2 0 -0.2 8 7 6 10 9 L 5 4 0.6 0.8 3 2 1 1 Figure 2. (Color online) Phase transition between the Laughlin and bubble phase in (2π, 1) LL. (a) Low-lying energy spectrum on the torus as a function of θ. (b) Projected structure factor SL as a function of angular momentum L on the sphere and parameter θ. Large ground-state degeneracy for small cos2 θ, as well as the sharp peak in SL, correspond to the bubble phase. band structures introduced in Ref. [20], which describe a number of high-mobility materials, including graphene with massive carriers (mass is generated either spontaneously, or as a result of sublattice symmetry breaking [17]), topological insulators [18], bilayer-graphene [17], trilayer graphene [32], and similar materials. Pristine graphene, which hosts massless Dirac-like fermions, is contained in this model as a particular case. Note that we explore fairly large variations of the effective interactions among fully polarized electrons; more subtle effects due to LL mixing [33] and multicomponent degrees of freedom are left for future study. We show the results for both spherical and torus geometry. A number of useful insights can be inferred from the study of the energy spectrum; in addition, we use overlap calculations to compare an exact, many-body ground state Ψexact, with a numerical representation of a trial wavefunction, Ψtrial. The overlap is defined as a scalar product between two normalized vectors, O = hΨtrialΨexacti. If O is consistently close to unity for a number of system sizes considered, we consider the trial wavefunction to be a faithful description of a FQH phase. From the knowledge of a ground-state wavefunction, we also evaluate the (projected) static structure factor [35]. Sharp peaks in the structure factor indicate the onset of compressible phases [34, 36]. The simplest example of the phase transition can be realized between the Laughlin state and a bubble phase in (2π, 1) LL of the chiral materials. In Fig. 2(a) we show the low-lying part of the energy spectrum obtained by exact diagonalization in the torus geometry, as a function of parameter θ introduced in Sec. 2. For values cos2 θ ≥ 0.2, the ground state of the system is non-degenerate and has a large overlap with the Laughlin wavefunction. When cos2 θ is less than 0.2, the effective interaction resembles that of n = 2 non-relativistic LL, which favors the bubble phase at filling ν = 1/3. This is manifested by a large ground-state degeneracy developing in the spectrum [36]. Similarly, the evolution of the projected structure factor SL on the sphere as a function of θ develops a discontinuity around cos2 θ ≈ 0.2 and a sharp peak below that value, signaling the compressibility of the ground state. Previously [20], it was emphasized that the tunability of the interactions in chiral materials can lead to the more stable non-Abelian states. One such example of considerable experimental interest is the Moore-Read Pfaffian state [5]. In Fig. 3 we show the energy spectrum for N = 12 electrons on a torus in (π, 1) LL. At cos2 θ = 0, the interaction reduces to n = 1 non-relativistic Coulomb interaction, which leads to a fragile Pfaffian state. As shown in the inset of Fig. 3, the overlap of the exact ground state with the Moore-Read wavefunction slightly increases as cos2 θ is increased from zero. However, due to the particle-hole symmetry which is present on the torus at filling ν = 1/2, a better indicator is to compare the exact ground state with a particle-hole symmetrized version of the Moore-Read wavefunction [15]. Indeed, in this case the overlap rises to nearly unity right before the transition at cos2 θ ≈ 0.14. At this point, the p a l r e v O 1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 MR Sym MR 0 0.05 0.1 cos2 θ 0.15 0.2 -2.4 -2.42 -2.44 -2.46 -2.48 -2.5 y g r e n E -2.52 0 0.05 0.1 cos2 θ 0.15 0.2 (Color online) Enhanced Pfaffian state and the transition to the CFL. Energy Figure 3. spectrum of N = 12 electrons on the torus in (π, 1) LL is plotted as a function of θ. Inset shows the ground-state overlap with the Moore-Read wavefunction and its particle-hole symmetrized version. level crossing occurs, and a composite Fermi liquid (CFL) state becomes the ground state of the system. We emphasize that although the level crossing would suggest a first-order transition, we cannot reliably rule out a possibility for the second-order transition in the thermodynamic limit. Note that, in addition to the considerable increase in the overlap with the Moore-Read state, also the excitation gap increases towards the transition to the CFL state, adding to the overall stability of the Pfaffian. As shown in Ref. [20], Pfaffian correlations can also be realized in (π, 2) LL, and other states such as k = 3 RR state can be realized in several LLs [30]. Phase transitions can also be studied without reference to a particular wavefunction by evaluating the excitation gap of the system, Fig. 4. To be more general, we consider a model that involves a superposition of n = 0, n = 1 and n = 2 LL non-relativistic form factors [20]: Fn(q) = cos2 θLn(cid:18) q2ℓ2 2 (cid:19) + sin2 θ cos2 φLn+1(cid:18) q2ℓ2 2 (cid:19) + sin2 θ sin2 φLn+2(cid:18) q2ℓ2 B B B 2 (cid:19) . (9) This model generalizes the case of (π, 1), (π, 2) and (2π, 1) LLs studied earlier, and naturally arises in systems such as trilayer graphene [32]. In Fig. 4 we plot the charge gap at ν = 1/3 and ν = 1/2 as a function of θ and φ. Because the calculation is performed in the spherical geometry, we fix the shift to be −3 (corresponding to the Laughin state at ν = 1/3 and the Moore-Read state at ν = 1/2). Along certain lines, indicated by arrows, the phase diagram reduces to one of the cases mentioned above. Note that proper finite-size scaling needs to be performed in order to get the correct values for the gap; however, it was previously found [21, 20] that this rigorous analysis produces values that are roughly in agreement with the ones shown in Fig. 4. At ν = 1/3 the Laughlin state is realized for a wide range of θ and φ. The maximum of the gap occurs for the effective interaction that is a mixture of n = 0 and n = 1 non-relativistic form-factors. For cos2 θ ≤ 0.2 the excitation gap is significantly reduced and a bubble phase is realized. The point of transition to the bubble phase is insensitive to the value of φ, unless cos2 φ ≈ 1. On the other hand, at ν = 1/2 the behavior of the gap suggests the existence of three phases. On the right side (large cos2 θ), the gaps are close to zero -- this is the region corresponding to the CFL phase. In this parameter regime, the ground state of the system also has a different shift from the one chosen in Fig. 4. On the left side, one can discern two distinct branches -- the middle one corresponds to the Moore-Read state because the overlap with the Pfaffian wavefunction is high in this region [20]. Note that this region is fairly large and encompasses the effective interactions which are quite different from the n = 1 non-relativistic Figure 4. (Color online) Charge gap (in units of e2/ǫℓB) at ν = 1/3 and ν = 1/2 as a function of the band-structure parameters θ and φ. The calculation is performed in the spherical geometry, for N = 10 particles at ν = 1/3 and N = 14 particles at ν = 1/2, with a fixed shift of −3. Coulomb interaction, known to give rise to the Pfaffian state. On the far left there is a stripe phase; the excitation gap is seemingly large in this region, however this is only an artefact of the spherical geometry that cannot adequately accommodate the broken-symmetry state such as the stripe. We expect this gap to be significantly reduced as one approaches the thermodynamic limit. Note that the gaps of the incompressible states can be further increased using a dielectric setup proposed in Ref. [21]. 4. Conclusion In this paper we have discussed two practical ways of tuning the effective interactions in fractional quantum Hall systems. We emphasized an important role of chiral 2DESs where the interactions can be varied in a broader range than in GaAs, and more robust non-Abelian states are likely to be found. Moreover, these non-Abelian states are expected to exist in the new regimes of the effective interactions that are significantly different from the ones in GaAs. Finally, their excitation gaps can be tuned, and quantum phase transitions can be engineered in situ by varying the magnetic field. 5. Acknowledgements This work was supported by DOE grant DE-SC0002140. Y.B. was supported by the State of Florida. References [1] D. C. Tsui, H. L. Stormer, and A. C. Gossard, Phys. Rev. Lett. 48, 1559 (1982). [2] R. B. Laughlin, Phys. Rev. Lett. 50, 1395 (1983). [3] B. I. Halperin, Phys. Rev. Lett. 52, 1583 (1984). [4] D. Arovas, J. R. Schrieffer, and Frank Wilczek, Phys. Rev. Lett. 53, 722 (1984). [5] G. Moore and N. Read, Nucl. Phys. B 360, 362 (1991). [6] C. Nayak and F. Wilczek, Nucl. Phys. B 479, 529 (1996). [7] P. Bonderson, V. Gurarie, and C. Nayak, Phys. Rev. B 83, 075303 (2011). [8] C. Nayak et al., Rev. Mod. Phys. 80, 1083 (2008). [9] The Quantum Hall Effect, 2nd ed., edited by R. E. Prange and S. M. Girvin, Springer-Verlag, New York, 1990. [10] F. D. M. Haldane, Phys. Rev. Lett. 51, 605 (1983). [11] J. K. Jain, Composite fermions, (Cambridge University Press, 2007). [12] N. Read and E. Rezayi, Phys. Rev. B 59, 8084 (1999). [13] R. Willett et al., Phys. Rev. Lett. 59, 1776 (1987). [14] R. H. Morf, Phys. Rev. Lett. 80, 1505 (1998). [15] E. H. Rezayi and F. D. M. Haldane, Phys. Rev. Lett. 84, 4685 (2000). [16] F. C. Zhang and S. Das Sarma, Phys. Rev. B 33, 2903 (1986). [17] A. H. Castro Neto et al., Rev. Mod. Phys. 81, 109 (2009). [18] M. Z. Hasan and C. L. Kane, Rev. Mod. Phys. 82, 3045 (2010). [19] B. A. Volkov and O. A. Pankratov, Pisma Zh. Eksp. Teor. Fiz. 42, 145; JETP Lett. 42, 178 (1985); B. Buttner et al., Nature Phys. 7, 418 (2011). [20] Z. Papi´c, D. A. Abanin, Y. Barlas, and R. N. Bhatt, arXiv:1108.1339. [21] Z. Papi´c, R. Thomale, and D. A. Abanin, Phys. Rev. Lett. 107, 176602 (2011). [22] D. Yoshioka, B. I. Halperin, and P. A. Lee, Phys. Rev. Lett. 50, 1219 (1983); F. D. M. Haldane in Ref. [9]. [23] X. G. Wen and A. Zee, Phys. Rev. Lett. 69, 953 (1992). [24] J. Cullum and R. A. Willoughby, J. Comp. Phys. 44, 329 (1981). [25] W. H. Press, S. A. Teukolsky, W. T. Vetterling, and B. P. Flannery, Numerical recipes, 2nd ed. (Cambridge University Press, Cambridge, 1992). [26] K. Nomura and A. H. MacDonald, Phys. Rev. Lett. 96, 256602 (2006); M. O. Goerbig, R. Moessner, and B. Dou¸cot, Phys. Rev. B 74, 161407 (2006). [27] X. Du et al., Nature 462, 192 (2009); K. Bolotin et al., Nature 462, 196 (2009); C. R. Dean et al., arXiv:1010.1179. [28] B. E. Feldman, J. Martin, A. Yacoby, Nature Phys. 5, 889 (2009). [29] Y. Barlas, K. Yang, and A. H. MacDonald, arXiv:1110.1069 [30] D. A. Abanin, Z. Papi´c, Y. Barlas, and R. N. Bhatt (submitted to New Journal of Physics) [31] V. M. Apalkov and T. Chakraborty, Phys. Rev. Lett. 97, 126801 (2006); C. Toke and J. Jain, Phys. Rev. B 75, 245440 (2007); Z. Papi´c, M. O. Goerbig, and N. Regnault, Phys. Rev. Lett. 105, 176802 (2010). [32] W. Bao et al., arXiv:1103.6088 (2011); T. Taychatanapat et al., arXiv:1104.0438 (2011); A. Kumar et al., arXiv:1104.1020 (2011); C. H. Lui et al., arXiv:1105.4658 (2011). [33] W. Bishara and C. Nayak, Phys. Rev. B 80, 121302 (2009); A. Wojs, C. Toke, and J. K. Jain, Phys. Rev. Lett. 105, 096802 (2010); E. H. Rezayi and S. H. Simon, Phys. Rev. Lett. 106, 116801 (2011). [34] A. A. Koulakov, M. M. Fogler, and B. I. Shklovskii, Phys. Rev. Lett. 76, 499 (1996); R. Moessner and J. T. Chalker, Phys. Rev. B 54, 5006 (1996). [35] S. He, S. H. Simon and B. I. Halperin, Phys. Rev. B 50, 1823 (1994). [36] E. H. Rezayi, F. D. M. Haldane, and K. Yang, Phys. Rev. Lett. 83, 1219 (1999); F. D. M. Haldane, E. H. Rezayi, and K. Yang, Phys. Rev. Lett. 85, 5396 (2000).
1104.4215	2	1104	2012-02-27T05:34:54	Spin Damping Monopole	[ "cond-mat.mes-hall" ]	We present theoretical evidence that a magnetic monopole emerges in dynamic magnetic systems in the presence of the spin-orbit interaction. The monopole field is expressed in terms of spin damping associated with magnetization dynamics. We demonstrate that the observation of this spin damping monopole is accomplished electrically using Ampere's law for monopole current. Our discovery suggests the integration of monopoles into electronics, namely, monopolotronics.	cond-mat.mes-hall	cond-mat	Typeset with jpsj2.cls <ver.1.2.2> Letter Spin Damping Monopole Akihito Takeuchi ∗ and Gen Tatara Department of Physics, Tokyo Metropolitan University, Hachioji, Tokyo 192-0397, Japan We present theoretical evidence that a magnetic monopole emerges in dynamic magnetic systems in the presence of the spin-orbit interaction. The monopole field is expressed in terms of spin damping associated with magnetization dynamics. We demonstrate that the observation of this spin damping monopole is accomplished electrically using Amp`ere's law for monopole current. Our discovery suggests the integration of monopoles into electronics, namely, monopolotronics. KEYWORDS: magnetic monopole, spin damping, inverse spin Hall effect, Maxwell's equations The magnetic monopole predicted by Dirac in 19311 is a unique particle that arises from singularity.2 In high-energy physics, a monopole emerges if the electromagnetic interaction in the world, described by a U(1) algebra, is a result of the symmetry breaking of a unified force having a higher symmetry of SU(5).3, 4 Intensive effort has been exerted to find a grand unified theory (GUT) monopole by waiting for a monopole created in the early universe to go through superconducting detectors5 and by detecting its ionization;6 however, no evidence has been found thus far. The energy needed to create a GUT monopole is about 1017 GeV, and so creating one in an accelerator on earth is impossible. Since a monopole is a consequence of symmetry breaking, it exists in solids too. Symmetry breaking in solids occurs at an energy much lower than the GUT energy, typically below 1 eV, and therefore experiments on this are feasible. The most well-known monopole in solids is the hedgehog monopole (HHM), which arises in magnetic materials;7 it is the SU(2) counterpart of the GUT monopole. The key interaction for the HHM is the coupling of the conduction electron to the magnetization represented by the vector M (r, t), which depends on the space coordinate r and the time t. The electronic spin, represented by a vector σ, is polarized by M owing to the coupling Hsd = −JM · σ, (1) where J is the coupling constant. Since M is an external field for electrons, the SU(2) symme- try of the electronic spin is broken. By diagonalizing Hsd and choosing the spin quantization direction to be along the z-axis, the electrons are described as spin-polarized and interacting with a SU(2) gauge field,7 Aa µ (µ = t, x, y, z and a = x, y, z are the indices in the coordi- ∗E-mail: [email protected] 1/9 J. Phys. Soc. Jpn. Letter µAc µAy ν − Ay µAx µν = ∂µAa ν − ∂νAa nate space and spin space, respectively). In the adiabatic limit, i.e., when J is large, only the z component of the gauge field, Az µ, survives and it acts as a U(1) gauge field. The HHM originates from the deviation from the perfect U(1) symmetry, i.e., from the perpendicular components Ax F a µ. In fact, the electrons feel the SU(2) gauge fields whose strength is µ and Ay µ − (2e/)ǫabcAb µν = ∂µAz ν, where e is the electric charge, is the Planck constant divided by 2π, and ǫabc represents the antisymmetric tensor. When this field is projected onto the U(1) space, we obtain the electromagnetic field tensor as F z µ + Φµν, where Φµν ≡ −(2e/)(Ax ν ). The anomalous field strength Φµν , which in terms of n ≡ M /M reads Φµν = −(/2e)n · (∂µn × ∂ν n), represents the HHM. In fact, the σρ = 0, whose components read ∇ × E + B = −jh, and field strength satisfies (1/2)ǫµνσρ∂νF z ∇·B = ρh, where the electric and magnetic fields are Ei ≡ −∇iAz i = −(/2e)n·( n×∇in) and Bi ≡ ǫijk∇jAz k = (/4e)ǫijkn · (∇j n × ∇kn), respectively. The monopole current (jh) and its density (ρh) for the HHM are given as jh,i = −(3/4e)ǫijk n · (∇j n × ∇kn) and ρh = (/4e)ǫijk∇in · (∇jn × ∇kn), respectively. Although the HHM is mathematically al- lowed, experimental realization has not been achieved thus far. In fact, as seen from the ν − ∂ν Az t − Az expression for jh and ρh, the HHM disappears in common magnets where the length of the local magnetization, n, is constant. In addition, the boundary condition at infinity for the HHM would not be easy to realize experimentally. In this paper, we search for a different monopole in magnets, which exists in conventional ferromagnets where the local magnetization length is constant. Such a monopole current creates the rotational electric field via Amp`ere's law. This means that the monopole current is an anomalous angular momentum source that induces the rotational motion of electric charges. To realize such a monopole in magnetic systems, a coupling between spin and electron orbital motion is, therefore, essential. Such a coupling is known to emerge from a relativistic effect, namely, the spin-orbit interaction. The spin-orbit interaction exists in all elements including magnetic ones and is particularly strong in heavy elements such as platinum and gold. Our aim in this study is to prove the existence of the above monopole theoretically. Since the spin-orbit interaction explicitly breaks the SU(2) invariance, we cannot follow the derivation of the HHM shown above. Instead, we will directly calculate the effective electric and magnetic fields for the electron spin based on a nonequilibrium Green's function formalism, and derive Maxwell's equations they satisfy. We consider two types of spin-orbit interaction. The first is that from a uniform field, ER, namely the Rashba interaction.8 Such a field is realized at interfaces and surfaces.9 The second is that from a random potential, vi, induced by heavy impurities. The spin-orbit interaction thus reads Hso = − 1 (λRER − λi∇vi) · (p × σ), (2) where p is the electron momentum and λ is the coupling constant (the subscripts R and i 2/9 J. Phys. Soc. Jpn. Letter characterize Rashba and impurity-induced spin-orbit intereactions, respectively). The inter- action with the magnetization is described by Hsd. The Hamiltonian of the present system is, therefore, given as H = (p2/2m) + vi + Hsd + Hso, where m is the electron mass. The magneti- zation we consider in Hsd is dynamic. Dynamic magnetization, when coupled to the spin-orbit interaction, generates an electric charge flow.10 -- 12 The pumped electric current j is calculated by evaluating a quantum field theoretical expectation value of the electron velocity operator v = −(i/m)∇ + (1/)(λRER − λi∇vi) × σ. By using field operators for electrons, c† and c, the electric current thus reads j = −etrhc† vci, where tr denotes the trace over spin indices and the bracket represents the expectation value. It is written in terms of the lesser component of the nonequilibrium Green's function,13 defined as G< s′ (r′, t′)cs(r, t)i (s and s′ are spin indices), as ss′(r, t; r′, t′) ≡ (i/)hc† j(r, t) =e tr({ 2 2m (∇r − ∇r′) + i[λRER − λi∇vi(r)] × σ}G<(r, t; r′, t))r′=r. (3) This quantum field theoretical expectation value is evaluated by solving the Dyson's equation for the nonequilibrium Green's function defined on the Keldysh contour (C), Gss′(r, t; r′, t′) =δs,s′gs(r, t; r′, t′) dt′′gs(r, t; r′′, t′′) +Z d3r′′ZC × (δs,s′′vi(r′′) − JM (r′′, t′′) · σss′′ + i{[λRER − λi∇vi(r′′)] × ∇r′′} · σss′′) × Gs′′s′(r′′, t′′; r′, t′), (4) where Gss′(r, t; r′, t′) ≡ −(i/)hTC[cs(r, t)c† g denotes the free Green's function. s′(r′, t′)]i (TC is the path-ordering operator) and In the calculation, the impurities are approximated as random point scatterers, and aver- aging is carried out as hvi(r)vi(r′)ii = niu2 i δ3(r − r′) (ni and ui are the impurity concentration and the strength of scattering, respectively).14 The impurities give rise to an elastic lifetime for the electron, τ , which is calculated in metals as τ = /2πniu2 i ν (ν is the density of states per volume).15 The Dyson's equation is solved by treating λ and J perturbatively to the first and second orders, respectively. We consider a sufficiently slow dynamics of magnetization, namely, Ωτ ≪ 1 (Ω is the frequency of magnetization dynamics), and assume that the mag- netization structure varies smoothly in the space compared with the electron mean free path ℓ, i.e., qℓ ≪ 1 (q is the wave number of magnetization profile). The leading contribution in 3/9 J. Phys. Soc. Jpn. this case turns out to be16, 17 Letter j(r, t) = eJ 2 V Xk,k′,q1,q2 Xω,Ω1,Ω2 e−i(q1+q2)·r+i(Ω1+Ω2)tΩ1 × + (Mq1,Ω1 × Mq2,Ω2) × [ dfω dω 42λi 3πντ 2 (q1 + q2)εkgr k,ω2gr k′,ω4] iλRτ ERgr k,ω2 − D∇ρ(r, t), (5) where V is the system volume, D ≡ 2εFτ /3m is the electron diffusion constant (εF represents the Fermi energy), fω is the Fermi distribution function, gr is the retarded Green's function k,ω = [ω − εk + (i/2τ )]−1, and εk = 2k2/2m. The last term is the diffusive defined as gr contribution arising from vertex corrections, where the electric charge density ρ is19 ρ(r, t) = 4eνλRJ 2τ 3 2V ∇ ·Z d3r′Z dt′ e−iq·(r−r′)+iΩ(t−t′) × Xq XΩ × {ER × [M (r′, t′) × M (r′, t′)]}. Dq2τ + iΩτ (6) Summing over the wave vectors and frequencies in eq. (5), the electric current is obtained as j = − 16eνλiJ 2εFτ 2 32 ∇ × (M × M ) 4eνλRJ 2τ 2 2 − ER × (M × M ) − D∇ρ. (7) This result is rewritten using the effective electric and magnetic fields, Es and Bs, respectively, as (µ and σc are the magnetic permeability and electric conductivity, respectively) j = 1 µ ∇ × Bs + σcEs − D∇ρ, where the effective fields are defined as20, 21 Es ≡ −αRER × N , Bs ≡ −βiN . (8) (9) Here, N ≡ M × M is a vector representing the spin damping torque (inset in Fig. 1).22 The coefficients αR and βi are αR ≡ 4eνλRJ 2τ 2/σc2 and βi ≡ 16eνµλiJ 2εFτ 2/32, respectively. The effective fields calculated here are those acting on the electronic spin in the same manner as the effective fields from the HHM. Clearly, the fields [eq. (9)] do not satisfy Faraday's law or Gauss's law of conventional electromagnetism, but those with monopole contribution: ∇ × Es + Bs = −jm, ∇ · Bs = ρm, 4/9 (10) J. Phys. Soc. Jpn. Letter + + + j z y x ρm M j m E R - - - z d j m M θ x M N N Fig. 1. (Color online). Schematic illustration of monopole pumping and detection in thin ferromag- netic film attached to a nonmagnetic layer. Magnetization (M ) precession is induced by applying an oscillating magnetic field. The Rashba field ER exists at the interface and creates the monopole current jm near the interface. The width of the monopole current distribution, d, is comparable to the decay length of the magnetization at the interface. The monopole current induces the electric current j via Amp`ere's law at the interface. The impurity spin-orbit interaction directly induces positive (+) and negative (−) monopole charge distributions, ρm, at the two edges. This monopole distribution generates an electric current again perpendicular to the monopole current, as is seen from eqs. (8) - (10). Inset: depiction of the spin damping vector N ≡ M × M arising from the precession of magnetization. The component of the spin damping vector perpendicular to the precession axis vanishes when time-averaged, leaving N along the axis as the DC component. where the monopole current and monopole density, respectively read jm = αR∇ × (ER × N ) + βi N , and ρm = −βi∇ · N . (11) (12) Equations (10) - (12) are the central results of this paper. We have thus proved that a monopole exists when spin damping occurs, namely, a spin damping monopole. The spin damping monopole is a composite object made from a magnetization configuration in the same manner as the HHM. It satisfies the conservation law ρm + ∇ · jm = 0. Our result obtained in a disordered system is a general one and can be extended to a clean system. In fact, the same monopole exists in a clean case, but only the coefficients αR and βi appearing in the monopole density and current are changed. The same applies to the HHM; when we take into account the third-order contribution in J, our analysis correctly reproduces the HHM, which was discussed in a clean limit only. Our approach is, therefore, a novel method of identifying monopoles. The spin damping monopole is unique since it does not require a particular non-coplanar spin structure like a hedgehog, and so it exists quite generally in magnetic systems. The simplest candidate for creating the monopole would be a thin ferromagnetic film put on a 5/9 J. Phys. Soc. Jpn. Letter nonmagnetic material, as shown in Fig. 1. We choose the z-axis perpendicular to the film. A Rashba-type spin-orbit field would then arise at the interface along the z-direction.9 We excite the precession of the uniform magnetization by applying the alternating magnetic field in the yz-plane in the presence of a static field along the x-axis (ferromagnetic reso- nance22). The precession results in a spin damping vector with a finite time average, N , along x-direction (inset in Fig. 1). In the present case with uniform magnetization, spatial derivatives in eqs. (11) and (12) arise at the interface and edges, where the magnetization vanishes. The Rashba interaction contributes to the DC monopole current at the interface as jR m,x = −αRER(∂N /∂z) ≃ −(αR/d)ERN , where d is the spatial scale of the magnetiza- tion decay at the interface. The monopole current driven by random spin-orbit impurities, on the other hand, vanishes when time-averaged. The total DC monopole current thus reads jm = −ex(αR/d)ERN (ex represents the unit vector along the x-direction). This monopole current at the interface generates an electromotive force along the y-direction via Amp`ere's law for the monopole. The monopole density induced by the random spin-orbit interaction arises at the edge of the ferromagnetic film since ∇ · N ≃ ∂Nx/∂x is finite there. The induced monopole density at the two edges is ρm = ∓(βi/d)N , where the sign is positive on one side of the edge and negative on the other side. The monopole then produces a magnetic field along the x-direction as Bs = −exβiN . This field creates an electric current in the y-direction via the conventional Amp`ere's law. The total electric current density generated by the spin damping monopole [eq. (8)] thus reduces to j = −ey[σcαRER + (βi/µd)]N . Ω and an angle θ, (ek2 the monopole-induced current density is estimated as When spin damping arises from the magnetization precession with a frequency ¯j = FΩ/π2)(Jτ /)2 sin θ[(∆R/εF) + (4/3kFd)(∆i/vi)], where ∆R and ∆i are the energy of the Rashba and impurity spin-orbit couplings, respectively (kF is the Fermi wave vector). In dis- ordered ferromagnets, J/εF ∼ 0.1, εFτ / ∼ 10 and k−1 F ∼ 2 A. The Rashba interaction can be enhanced on surfaces and at interfaces, resulting in ∆R/εF ∼ 0.1 (∆i/εF is gener- ally smaller).23 When θ = 30◦ and Ω = 1 GHz, the electric current density is thus 2 × 107 A/m2 which is sufficiently large for experimental detection. In addition to DC, there is an AC component in eq. (8), which would be accessible by time-resolved measurement. We note that the electric current estimated here is an initial current that arises when the pumping of monopoles starts. When the monopole current is pumped steadily, monopole accumulation grows at the edges of the system, inducing a diffusive current. The steady monopole distribution is then determined by the balance of this backward diffusion and the pumped monopole current. Direct evidence of the spin damping monopole is given by detecting the electric current discussed above. Surprisingly, the signal from the spin damping monopole might have already been detected. In fact, the electric voltage due to magnetization precession has been observed 6/9 J. Phys. Soc. Jpn. Letter in a junction of a ferromagnet on a Pt film in a pioneering work by Saitoh et al..24 The mechanism of voltage generation has been explained by the inverse spin Hall effect. According to the inverse spin Hall scenario, magnetization precession generates a spin current via the spin pumping effect,25 and the spin current js is converted into an electric current by the spin-orbit interaction (the inverse of the spin Hall effect). This explanation assumes that the conversion mechanism of ji = ǫijkjk s,j where k is the index representing the spin polarization of the spin current.26 A recent theoretical study has revealed, however, that the conversion formula is not universal; it does not apply to the slowly varying magnetization configuration or in the presence of disorder.11 Rather, the conversion formula is an approximated one connecting a physical observable (electric current) to a spin current whose definition depends on the specific system considered. Our result obtained in the present paper suggests a different scenario that is universal owing to the symmetry of Maxwell's equations. For the experimental confirmation of the spin damping monopole, of crucial importance is the separation of the monopole signal from the inverse spin Hall signal driven by a spin current. This is accomplished by applying an electric field (Es) perpendicular to the junction of Fig. 1. The monopole contribution then leads to a transverse electric current as a result of the Hall effect of the monopole,27 while the contribution of the spin current is not affected. In another experiment, the strongest evidence of the spin damping monopole is given by observing a magnetic field (Bs) produced by monopoles via Gauss's law by electron holography. We have shown analytically that a magnetic monopole is a common object in dynamic magnetic systems with damping. The control of spin damping monopoles is as feasible as that of electrons, for both are governed symmetrically by Maxwell's equations of electromagnetism. We here propose the monopolotronics, i.e., the control of monopoles, as a novel concept of realizing spintronic devices. Acknowledgements The authors are grateful to N. Kitazawa, R. Matsumoto, S. Murakami, and N. Nagaosa for discussion. This work was supported by a Grant-in-Aid for Scientific Research in Priority Areas (Grant No. 1948027) from the Ministry of Education, Culture, Sports, Science and Technology, a Grant-in-Aid for Scientific Research (B) (Grant No. 22340104) from Japan Society for the Promotion of Science. One of the authors (A.T.) is financially supported by the Japan Society for the Promotion of Science for Young Scientists. 7/9 Letter J. Phys. Soc. Jpn. References 1) P. A. M. Dirac: Proc. R. Soc. London 133 (1931) 60. 2) J. D. Jackson: Classical Electrodynamics (Wiley, New York, 1998). 3) G. 't Hooft: Nucl. Phys. B79 (1974) 276. 4) A. M. Polyakov: JETP Lett. 20 (1974) 194. 5) B. Cabrera: Phys. Rev. Lett. 48 (1982) 1378. 6) M. Ambrosio et al.: Eur. Phys. J. C 25 (2002) 511. 7) G. E. Volovik: J. Phys. C 20 (1987) L83. 8) E. I. Rashba: Sov. Phys. Solid State 2 (1960) 1109. 9) L. Meier, G. Salis, I. Shorubalko, E. Gini, S. Schon, and K. Ensslin: Nat. Phys. 3 (2007) 650. 10) K. Hosono, A. Takeuchi, and G. Tatara: J. Phys. Soc. Jpn. 79 (2010) 014708. 11) A. Takeuchi, K. Hosono, and G. Tatara: Phys. Rev. B 81 (2010) 144405. 12) K. M. D. Hals, A. Brataas, and Y. Tserkovnyak: Eur. Phys. Lett. 90 (2010) 47002. 13) H. Haug and A.-P. Jauho: Quantum Kinetics in Transport and Optics of Semiconductors (Springer, New York, 2007). 14) J. Rammer: Quantum Transport Theory (Westview Press, Boulder, CO, 2004). 15) Small corrections on the order of /εFτ are neglected.14 This corresponds to the Born approxi- mation. 16) Equation (5) is obtained by evaluating three and six distinct self-energy types of Feynman dia- grams for the the Rashba and impurity-induced spin-orbit interactions, respectively, and vertex corrections including infinite summation of the impurity scattering.14 17) The contribution linear in λi arises from the averaging of the correlation of the impurity potentials in Hso and in the bare scattering term vi, as in the calculation of the anomalous Hall effect.18 18) A. Cr´epieux and P. Bruno: Phys. Rev. B 64 (2001) 014416. 19) The factor of (Dq2τ + iΩτ )−1 represents the effect of electron diffusion.14 20) In eq. (8), electron diffusion (D∇ρ) induces the effective electric and magnetic polarizations Ps and Ms, respectively. By using the relation (−D∇2 + ∂t)ρ = −σc∇ · Es, the electric current [eq. (8)] is described as the rotation of the magnetic field and the time derivative of the electric field, j = ∇ × Hs − Ds, where Hs and Ds are the fields defined as Hs ≡ (1/µ)Bs − Ms and Ds ≡ εEs + Ps (ε = −σcτ being the permittivity), respectively. Thus, Es and Bs are identified with the electric and magnetic fields, respectively. 21) Equations (8) and (9) apparently contain an arbitrariness, Es → Es + (1/σc)∇ × C and Bs → Bs − µC, where C is an arbitrary vector field. However, such a transform in Maxwell's equations with monopoles is generally not allowed because of the gauge invariance in the original space with a higher symmetry. In fact, in the case of HHM, the SU(2) gauge invariance forbids this arbitrariness.7 We expect the same argument applies to the present spin damping monopole. 22) S. Chikazumi: Physics of Ferromagnetism (Oxford University Press, New York, 1997). 23) C. R. Ast, J. Henk, A. Ernst, L. Moreschini, M. C. Falub, D. Pacil´e, P. Bruno, K. Kern, and M. Grioni: Phys. Rev. Lett. 98 (2007) 186807. 24) E. Saitoh, M. Ueda, H. Miyajima, and G. Tatara: Appl. Phys. Lett. 88 (2006) 182509. 25) Y. Tserkovnyak, A. Brataas, and G. E. W. Bauer: Phys. Rev. Lett. 88 (2002) 117601. 26) J. E. Hirsch: Phys. Rev. Lett. 83 (1999) 1834. 8/9 J. Phys. Soc. Jpn. Letter 27) The Lorentz force from an electric field acts on a magnetic monopole (see § 6.8 in ref. 2). 9/9
1302.2012	3	1302	2013-10-29T16:26:55	Abrupt barrier contribution to the electron spin splitting in asymmetric coupled double quantum wells	[ "cond-mat.mes-hall" ]	We have studied the behavior of the electronic energy spin-splitting of InGaAs-InAlAs based double quantum wells (narrow gap structures) under in-plane magnetic and transverse electric fields. We have developed an improved 8x8 version of the Transfer Matrix Approach that consider contributions from abrupt interfaces and external fields when tunneling through central barrier exists. We have included the Land\'e g-factor dependence on the external applied field. Also, we have calculated electron density of states and photoluminescence excitation. Variations of the electron spin-splitting energy lead to marked peculiarities in the density of states. Because the density of states is directly related to photoluminescence excitation, these peculiarities are observable by this technique.	cond-mat.mes-hall	cond-mat	Abrupt barrier contribution to the electron spin splitting in asymmetric coupled double quantum wells. A. Hern´andez-Cabrera∗ and P. Aceituno† Dpto. F´ısica B´asica, Universidad de La Laguna, La Laguna, 38206-Tenerife, Spain, and Instituto Universitario de Estudios Avanzados (IUdEA) en F´ısica At´omica, Molecular y Fot´onica, Universidad de La Laguna, La Laguna, 38206 Tenerife, Spain (Dated: August 15, 2018) We have studied the behavior of the electronic energy spin-splitting of InGaAs − InAlAs based double quantum wells (narrow gap structures) under in-plane mag- netic and transverse electric fields. We have developed an improved 8 × 8 version of the Transfer Matrix Approach that consider contributions from abrupt interfaces and external fields when tunneling through central barrier exists. We have included the Land´e g-factor dependence on the external applied field. Also, we have calcu- lated electron density of states and photoluminescence excitation. Variations of the electron spin-splitting energy lead to marked peculiarities in the density of states. Because the density of states is directly related to photoluminescence excitation, these peculiarities are observable by this technique. PACS numbers: 72.25.-b, 73.21.-b ∗Electronic address: [email protected] †Electronic address: [email protected] 1. Introduction 2 In the last decade a great interest has arisen for the so-called spintronic, or spin-based electronics. The reason is that the spintronic, due to its low power consumption, promises to be a good alternative to the traditional electronics, based on the charge transport [1]. On the other hand, the spin transport is not very dissipative, with low energy losses over long distances, although the relaxation of the spin polarization may exist. Spintronics is a catchall term that refers to the potential use of the spin rather than the charge in electronic devices. For it to be useful in practice, spin up and spin down electronic states (or hole states) of any material must be separated in energy. Also, the material should be electrically polarized as in conventional electronics, which means that carriers, both negative (electrons) and positive (holes), must be able to conduct. A key point is the choice of the material. For example, narrow gap semi- conductors with strong spin-orbit coupling. GaAs-based materials seem to be the suitable candidates[2] due to the long life of the magnetic spin state of photoexcited electrons, which behaves coherently. It is also important that the materials have a good lattice matching to avoid defects accumulation at the interface and internal strains, which would worsen the transport of the polarized spin. There exist techniques to relax this strain, as to place step graded buffers with a progressive variation of the In concentration between the substrate and the active device. These layers mainly absorb strain and defects. Another important point is a large Land´e factor for having a signifi- cant splitting of spin states by applying small magnetic fields. In general, InxGa1−xAs − InyAl1−yAs structure seems to be one of the most appropriate for spintronic purposes. This heterostructure offers the possibility of manipu- late the gap width, the Land´e factor and the interface contributions by varying 3 x and y concentrations. Besides, it presents a remarkable spin-splitting energy when a weak magnetic fields is applied. This peculiarity makes this material suitable for high temperature spin-valves devices. In this work we will focus in the spin-splitting changes in asymmetric cou- pled double quantum wells ACQW, caused by abrupt interfaces, when an in- plane magnetic field is applied. For this purpose we will base on an extended version of the 8 × 8 Kane formalism with nonsymmetric boundary conditions to calculate the band structure and dispersion laws[3]. Although electronic dispersion laws (and the corresponding spin-orbit splitting) in quantum het- erostructures have been widely studied in the last decades[4], many question remain open. One of them is the influence of heterojunctions on the electron spin tunneling and thus, their contribution to the polarization of this spin. In theory, effects from compositional parameters or interface contributions to the spin splitting should be analyzed through the density of states. And the modifications of the density of states can be directly observed using photolu- minescence excitation technique (PLE)[5]. In bulk materials spin-orbit interaction is caused either by a soft potential [6] and by cubic [7] and linear [8] spin-dependent contributions to the effec- tive Hamiltonian. However, in the two-dimensional (2D) case, we can reduce the cubic contribution to a linear one after the squared momentum substitu- tion by its quantized value due to confinement [9]. Moreover, it is necessary to consider the additional spin-orbit splitting caused by the interaction with abrupt heterojunction potentials (see Ref. [1, 3, 10, 11]). This contribution is absolutely different to the contributions mentioned above. The effect of an in-plane magnetic field on the energy spectrum in nonsym- metric heterostructures results in the Pauli contribution to the electron Hamil- tonian. Several peculiarities for transport phenomena in heterostructures have been also discussed [12 -- 15]. However, these discussions only consider the mix- 4 ing between Pauli contribution and effective 2D spin-orbit interaction. As mentioned above, we cannot forget the contribution of the abrupt barriers in asymmetric structures (e.g. ACQW) to the spin splitting, as we will emphasize in this work. Besides, we have included possible changes of the Land´e factor. Although the Land´e g-factor depends on the applied fields [16, 17], this depen- dence is negligible for in-plane magnetic fields and low electron density. The model also applies to narrow-gap heterostructures whenever the slow poten- tial generated by doping or external transverse electric fields can be described self-consistently. 2. Eigenstate problem We will center our attention in the Hamiltonian describing the electronic behavior in the conduction band, considering that any possible strain is already included in the structure through the gap, the conduction and the valence well potentials[18]. We will include effects of electric and magnetic fields as well as the interfaces contribution. Based on the assumption that the gap energy εg is smaller than the energy distance between the valence band (v-band) and the spin-split band extrema, the electronic states in these narrow-gap heterostructures can be described by the three-band Kane matrix Hamiltonian ε(z) + (v · P), P = p − e c A, (1) where the generalized kinetic momentum, P, contains the vector potential A = (Hz, 0, 0), HkOY is an in-plane magnetic field and p = (p, pz) is written in the p, z-representation through the 2D momentum p. We have also introduced the diagonal energy matrix ε(z) whose elements fix the positions of the band extrema and the interband velocity matrix v. From now on it is necessary to introduce a new index µ = w, b to de- note narrow-gap regions (wells) and wide-gap regions (central barrier and lat- eral sides), respectively. In the parabolic approximation we can write the 5 Schrodinger equation for ACQW in the form[3]: (cid:18)εµ p + p2 z 2mµ + εµ c (z) +cW µ(z)(cid:19) Ψµ(p,z) = EΨµ(p,z), where the isotropic kinetic energy is given by εµ p = x + p2 p2 y 2mµ , (2) (3) which includes the effective mass mµ. Parabolic approximation is justified because energy values under consideration are smaller than the gap energy εg in the narrow region. The εµ c (z) energy is εw c (z) = U(z) in the wells, and εb c(z) = ∆Ec + U(z) in the barriers, where ∆Ec is the band offset for c conduction band. Whenever energy values are less than ∆Ec underbarrier penetration (and tunneling) is permitted and described by the boundary conditions. The potential U(z), for an uniform transverse electric field, is U(z) ≃ eF⊥z. Band diagram for ACQW is shown in Fig. 1. The magnetic energycW µ(z), for not very strong magnetic fields, is described by cW µ(z) = −V µ(z) [σ × p]z + gµ(z) 2 µBH σy, (4) where µB ≡ e¯h/(mec) is the Bohr magneton and σ is the Pauli matrix. Finally, the characteristic spin velocity V µ(z), and the effective Land´e factor gµ(z) are V µ(z) = ¯h dεµ c (z)/dz 4mµ εg , gµ(z) = me 2mµ(cid:20)1 + z dεµ c (z)/dz εg (cid:21) . (5) The potential of interfaces determines a part of the spin dependent contri- butions through the parameter χ. Actually, χ takes into account the spin-orbit coupling due to the abrupt potential of the heterojunction at each interface, as[3, 11] χ = 2 ¯hZ δ −δ mµV µ(z) dz, (6) 6 where the integral is taken over the width of the abrupt interface 2δ. Contri- bution of χ to the energy dispersion relations will appear through the third kind boundary conditions after a first integration of the Schrodinger equation (2). We will detail this point in Appendix A. In the present case, where U(z) is linear with z, we get dεµ c (z)/dz = dU(z)/dz = eF⊥, which does not depend on z. Thus, the external applied electric field F⊥ shapes the spin velocity V µ(z) = vµ and the g-factor, gµ(z) = gµ. We can take the characteristic spin velocity for each layer as vµ = eF⊥¯h 4mµεg , and the abrupt interface parameter as χ = 2eF⊥δ + ∆Ec 2εg ∆Ec 2εg ≈ (7) (8) Lastly, we introduce the Pauli splitting energy wH, caused by the magnetic field, as wµ H = (gµ/2)µBH. Thus, Eq. (4) becomes which has lost the z dependence. cW µ = ¯vµ [σ × p]z + wµ H σy, (9) Because cW µ and εµ p do not depend on z, we can factorize fundamental solutions of Eq. (2), Ψµ(p,z), as products of ψµσ(p) and ϕµσ(z) functions, with σ = ±1. The σ value refers to the two possible spin orientations. For an ACQW under a transverse electric field F⊥, the p-dependent spinors ψµσ(p) can be obtained from in the form p +cW µ(cid:17) ψµσ(p) = εµ (cid:16)εµ σpψµσ(p), ψµ+(p) = ψµ−(p) = 1 √2 1 √2 (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) 1 (vµp+ + wµ H)/iwµ p H)/iwµ p (vµp− + wµ 1 , , (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) (10) (11) where p+ = px + ipy, p− = px − ipy and wµ p =q(¯vµpx + wµ H)2 + (vµpy)2 and the energy quasi-paraboloids are εµ pσ = εµ p + σwµ p, 7 (12) (13) Now, z-dependent functions ϕµσ(z) are obtained from the second order differential equation (cid:18) p2 z 2mµ + eF⊥z(cid:19) ϕµσ(z) =(cid:0)E − εµ pσ(cid:1) ϕµσ(z). (14) For an ACQW under a transversal electric field F⊥, eigenstate functions of Eq. (14) are the known linear combination of the Airy Ai- and Bi-functions. Thus, the general solution of Eq. (2) can be written as Ψµ(p,z) = ψµ+(p)(cid:2)aµ+Aµ+ +ψµ−(p)(cid:2)aµ−Aµ−i i i (z) + bµ+Bµ+ (z) + bµ−Bµ−i (z)(cid:3) + (z)(cid:3) , (15) where aµσ, bµσ are four constants by region to be solved through the interface conditions of continuity of the wave functions and current. Because there are two wells, and three barriers (Fig. 1), there are four interfaces and four boundary conditions by interface. Next, to simplify the calculation of the wave functions and the dispersion relations of the electronic levels through the boundary conditions[3, 10], we create two auxiliary parameters: a length lµ ⊥ and an energy εµ ⊥ =(cid:18) ¯h2 2mµeF⊥(cid:19)1/3 lµ ⊥ , εµ ⊥ = ¯h2 2mµ (lµ ⊥ )2 , (16) and a set of momentum dependent functions 8 ρµ p− = p−¯vµ + wµ H iwµ p , with ρb p− = ρw p− = ρp− p+¯vµ + wµ H ρµ p+ = , with ρb p+ = ρw p+ = ρp+ ρp+, p f1p = iχ f2p = iχ iwµ p− ¯h p− ¯h f3p = −iχ f4p = −iχ , p+ ¯h p+ ¯h , . ρp− (17) Next, using preliminary quasi parabolic dispersion relations εµ pσ, we construct the Airy function arguments ξµ pσ = + z lµ ⊥ pσ − E + δµ εµ b ∆Ec εµ ⊥ , (18) where δb µ acts as a Kronecker function: δb µ = 1 when µ = b, and δb µ = 0 when µ = w. We use the transfer matrix method to obtain wave functions and energy dispersion relations of the electronic levels. In the present case, we have im- proved the standard method by using 4 × 4 matrices at each interface, whose elements are (see Appendix A): p−), ), p−), ), p+), p+), Ai(ξµ 12(z, E, p) = Bi(ξµ 24(z, E, p) = Bi(ξµ p− Ai′(ξµ 31(z, E, p) = 13(z, E, p) = ρp− Bi(ξµ 14(z, E, p) = ρp− p− 21(z, E, p) = ρp+Ai(ξµ p+), 22(z, E, p) = ρp+Bi(ξµ 23(z, E, p) = Ai(ξµ p+(cid:1) , 11(z, E, p) = Ai(cid:0)ξµ Mµ Mµ Mµ Mµ Mµ Mµ Mµ Mµ Mµ Mµ Mµ Mµ Mµ Mµ Mµ Mµ σp) means dAi(ξµ me mµ me mµ me mµ me mµ me mµ me mµ me mµ me mµ Ai′(ξµ p− Bi′(ξµ p− 33(z, E, p) = 34(z, E, p) = 43(z, E, p) = 32(z, E, p) = 42(z, E, p) = 41(z, E, p) = 44(z, E, p) = 9 (19) p+) + δw Bi′(ξµ p+) + δw p+(cid:1) , µ f1pAi(cid:0)ξµ µ f1pBi(cid:0)ξµ p+(cid:1) , ) + δw Ai′(ξµ ρp− p− Bi′(ξµ ρp− p− ρp+Ai′(ξµ p+) + δw ) + δw µ f2pAi(ξµ p− µ f2pBi(ξµ p− ), ), p+(cid:1) , µ f3pAi(cid:0)ξµ µ f3pBi(cid:0)ξµ p+(cid:1) , µ f4pAi(cid:0)ξµ p−(cid:1) , p−(cid:1) , µ f4pBi(cid:0)ξµ ) + δw ) + δw ρp+Bi′(ξµ p+) + δw where Ai′(ξµ Bi′(ξµ σp). σp)/dz = (1/lµ ⊥ ) dAi(ξµ σp)/dξµ σp, and the same for ij(z, E, p). Finally, electronic levels for each 2D momentum p = (px, py) Now we are ready to generate transfer matrices, fM µ(z, E, p) which elements are Mµ are obtained from eS44 (E, p) = 0, where eS (E, p) = hfM b(L0, E, p)i−1 · fM w(L0, E, p) ·hfM w(L1, E, p)i−1 fM b(L1, E, p) ·hfM b(L2, E, p)i−1 · fM w(L2, E, p) · hfM w(L3, E, p)i−1 ·hfM b(L3, E, p)i . In the above matrix product, z = Li denotes interfaces position in the growth 10 · (20) direction, starting from the left side. Calculations give us two spin up Ek+(p) and two spin down Ek−(p) paraboloids, where k = 1, 2 corresponds to the deepest coupled levels of the ACQW. Once obtained coefficients aµσ, bµσ (Eq. 15) we normalize wave func- tions for each momentum p. The scheme of Fig. 1 includes the two resonant energy levels and the respective wave functions for p = 0. Although there are four levels only two are observable in this figure. This is because spin sublevel splitting is much smaller than electronic level energy distance and differences between spin down and spin up wave functions are not visible at p = 0. Finally, the density of states can be obtained by using the well-known ex- pression ρ(ε) =Xk,σ Z dp (2π¯h)2 δ(ε − Ekσ(p)). (21) Peculiarities of ρ(ε) can be analyzed experimentally through the photolumi- nescence excitation (PLE) intensity for the case of near-edge absorption, IP LE, because both quantities are related by[11] IP LE ∼ Xλcλv e · vcv2δ(ελc − ελv − ¯hω) ∼ ρ(¯h∆ω) for very low temperature, or IP LE ∼ Xλcλv e · vcv2G(ελc − ελv − ¯hω), (22) (23) 11 when including electron-phonon scattering. Gaussian function is G(x) = 1 γ√2π exp(cid:20)−(cid:16) x−x0 In the above expression, the γ√2(cid:17)2(cid:21). The Gaussian halfwidth γ is related to the scattering and relaxation processes and, thus, to the temperature[19]. Expressions (23, 24) are valid provided the interband velocity vcv does not depend on in-plane momentum. Here e is the light polarization vector, ∆ω = ω − εg/¯h, and ελc, ελv are the conduction and valence band levels, respectively. 3. Results Let's start this section with numerical results for InxGa1−xAs−InyAl1−yAs- based ACQWs, with x = 0.53 and y = 0.52. We have chosen this particular structure because we have reliable data for basic parameters[20, 21]. We have considered two InGaAs wells of 70 and 100 A wide separated by a 20 A InAlAs barrier. We have also applied an electric field of 30 kV/cm, which corresponds to a spin velocity vw = 2.6×105 cm/s for the InGaAs QWs, and vb = 1.4×105 cm/s for the InAlAs barriers, with a transition spin velocity region across de abrupt interface. This electric field is slightly higher than needed to achieve resonance between the deepest levels of both wells (28 kV/cm). To calculate interface contributions we have used a typical abrupt interface size of δ ∼ 3 A for InGaAs − InAlAs[22]. We have also applied in-plane magnetic field of 0.01 T, small enough to allow the anticrossing close to the bottom of energy dispersion relations (zero slope points). Figures 2(a−d) show normalized squared wave function for py = 0, versus z and the dimensionless momentum px/p0, where p0 = mw vw. Upper panels (a, b) correspond to the first deepest level for the two different spin orientations. As expected for an electric field beyond the resonance, charge density is mainly located in the left narrow QW. Consequently, the opposed happens for the second resonant level as can be seen in the lower panels (c, d) Analyzing wave 12 functions versus momentum px and spin orientations by comparing panels (a) and (b), a particular behavior occurs. While the charge distribution coincide for both down and up spins at the zone center (px = 0), there is a tunneling charge transfer between wells for increasing px. For spin down case [panel (a)] charge enhances tunneling from left narrow well to the right wide one. Tunneling shows opposite behavior for spin up electrons [panel (b)] and charge goes from the right to the left well. Lower panels (c, d) correspond to the higher resonant level, mainly at the right well. It might seem the behavior is the opposite to the previous one because now, spin down case [panel (c)] shows a tunneling increase from right to left QW as px increases and, conversely, for spin up electrons [panel (d)]. However, considering the relative charge concentration between wells, we can realize there is a similar behavior for both resonant levels. For spin down electrons there is a charge shift from well with higher concentration to the other well [panels (a), (c)] and conversely for the spin up electrons [panels (b), (d)]. The reason for this charge shift lies in the magnetic energy term wµ p, which induces a breaking of the px momentum symmetry. Because of wµ p is an essential part of the argument of the Airy functions, the behavior of the wave functions is significantly affected. Fig. 3 shows the near parabolic dispersion relations of the two coupled levels and their corresponding spin down and up sublevels, for the electric and mag- netic fields under consideration. It can be seen the py paraboloids symmetry according to Eqs. (12) and (13). Although there is a little difference between spin paraboloids, due to the large energy difference between the resonant levels of both wells (∼12 meV) and the splitting of the spin sublevels (∼0.01 meV at py = 0) it is not possible to distinguish minima behavior. Thus, we have enlarged in Fig. 4 the bottom of the pair of paraboloids (spin down and spin up) for the ground level. As expected, both paraboloids shift in opposite px directions resulting in a sublevel anticrossing. This displacement is due not 13 only to the magnetic field but also to the electric one (12). Considering a py = 0 section of the former figure we can get a more accurate 2D representation (Fig. 5) of the anticrossing, minima px position, and energy splitting. The inset displays anticrossing area enlargement. In this kind of anticrossing the slope of εµ pσ varies without changing the sign. However, as we will see below for the low magnetic fields under study, we work in the region where Van Hove singularities remain in the density of states. In order to have a more detailed overview of the anticrossing region we have also included in Fig. 6 the contour plot around anticrossing for different constant energy values. Next, we analyze the density of states ρ (ε). This function is related to several spin and interwell tunneling properties. Also, it is proportional to the photoluminescence excitation (PLE) intensity, one of the most used techniques to get information of quantum structures [23]. As mentioned before, we have found remains of the Van Hove singularities for fields under consideration. So, we have used magnetic field intensities varying from 0 to 0.1 T to analyze singularities behavior. The shape of ρ (ε) is shown in Fig. 7. Note that, when H = 0 T, energy paraboloids shift a certain amount ∓wµ p = ±¯vµp because of the ¯vµ dependence on the electric field. In this case, because both paraboloids bottom are at the same energy, we have clear ε−1/2 type singularities in each subband. As expected, these peaks disappear gradually by growing magnetic field because of the different vertical paraboloids shift, which leads to a greater slope at the anticrossing point. In turn, interfaces contribute with a manifested delay in the quenching of the ρ (ε) singularities, as well as an additional broad- ening of these peaks. That is, although the singularities should only appear at zero magnetic field, they still remain at the band anticrossing position for low magnetic fields, as shown in Fig. 7. Another significant feature is that the ρ (ε) anticrossing peak is softened when increasing barrier height, disappearing for lower magnetic fields than used in this work[3] 14 Finally, Fig. 8 shows PLE spectra for different Gaussian halfwidth γ at a fixed H = 0.1 T. For the magnetic field used before (H = 0.01 T) the two ad- jacent peaks, corresponding to each resonant pair of states, overlap. Thus, we have used a magnetic field ten times higher because this field allows us to tell the peaks apart for small γ values. Evolution of the first two peaks with γ is depicted in Fig. 9. As can be seen, PLE peaks corresponding to the two differ- ent spin transitions are still distinguishable for γ values beyond 1 meV. These results show the same general behavior than available experimental data[24]. 4. Conclusions In this paper we have analyzed the electron spin behavior in narrow-gap ACQWs under transverse electric and in-plane magnetic fields, including the role of abrupt interfaces. We used the Kane model with nonsymmetric bound- ary conditions, caused by the two different wells width, to solve the eigenvalue problem. Based in this model and the transfer matrix approach we have per- formed a useful tool to tackle any layered structure with abrupt interfaces and subjected to different perturbations. To do that we have implemented 8 × 8 matrices for the boundary conditions to describe near-parabolic disper- sion relations. The model allows us the study of the spin peculiarities of levels anticrossing. Because interface contributions oppose intrinsic spin-orbit effect, mecha- nisms that mix the Pauli contribution with the two kinds of spin-orbit con- tributions (from a low magnetic field and from heterojunctions) are different. As a result, numerical calculations for InGaAs − InAlAs structures lead to magnetoinduced variations of the energy dispersion relations, under very low in-plane magnetic fields. This effect is particularly appreciable at anticross- ings. As dispersion relations are used to obtain the density of states, this function is also affected by the abrupt interfaces: a new kind of ρ (ε) sin- 15 gularity, which does not disappear when increasing magnetic fields, appears. Furthermore, there is an energy broadening of the peaks. Finally, we have calculated PLE intensity because, as mentioned before, mid-infrared PLE spectroscopy is a suitable technique to find characteristics of energy spectrum[23]. This technique provides direct information of the energy spectrum when interband transitions are modified by in-plane magnetic fields. Since, for our purposes, a thorough analysis of the band structure little contributes to describe eigenstates, the main conclusions of this work remain valid. We have also used along the work the assumption that the confined potential is well described. Thus, the effective transverse field F⊥ provides correct estimations both for magnetoinduced changes and for the character of the dispersion laws. However, more detailed numerical calculations are needed to describe the kinetic behavior. We will return to this point in a forthcoming work where we will analyze the spin dynamics in ACQWs. In summary, we have shown the importance of the contribution from the interfaces, in narrow gap structures, to the spin polarization through the electronic dispersion re- lations, and how this contribution can be detected by PLE. The model can be extended to the analysis of the negative magnetoresistance[25] as well as peculiarities of the spin current through ADQW[26] bpz mw Ψw(p,z)(cid:12)(cid:12)(cid:12)(cid:12)z=L − bpz together with the wave function continuity at interfaces mb Ψb(p,z)(cid:12)(cid:12)(cid:12)(cid:12)z=L − iχ [bσ × p]z Ψw(p,z)z=L = 0, Ψw(p,z)z=L − Ψb(p,z)(cid:12)(cid:12)z=L = 0. (A1) (A2) Ψb(p,z)(cid:12)(cid:12)(cid:12)(cid:12)z=L + χ ¯h   p− 0 −p+ 0   Ψw(p,z)z=L = 0, (A3) which lead to a set of four equations we can write as Appendix A: Boundary conditions After a first integration of the Schrodinger equation over an heterojunction 16 at z = L, we get the third class boundary conditions: After substitution of Ψµ(p,z) by ψµσ(p)ϕµσ(z) in (A3) we obtain: p+) + ρb p+) + ρw −i mw ∂ ∂z Ψw(p,z)(cid:12)(cid:12)(cid:12)(cid:12)z=L + i mb ∂ ∂z p+) + bw+Bi(ξw p+) + bb+Bi(ξb p+) + bb+Bi(ξb p+) + bw+Bi(ξw Ψw(p,z)z=L − Ψb(p,z)(cid:12)(cid:12)z=L = 0. (cid:8)ab+Ai(ξb = (cid:8)aw+Ai(ξw (cid:8)ρb p+(cid:2)ab+Ai(ξb = (cid:8)ρw p+(cid:2)aw+Ai(ξw (cid:18)me mb (cid:8)ab+Ai′(ξb = (cid:18) me mw (cid:8)aw+Ai′(ξw ¯h (cid:8)ρw p+p−(cid:2)aw+Ai(ξw (cid:18)me mb (cid:8)ρb p+(cid:2)ab+Ai′(ξb = (cid:18) me mw (cid:8)ρw p+(cid:2)aw+Ai′(ξw ¯h (cid:8)p+(cid:2)aw+Ai(ξw − p+) + bb+Bi′(ξb iχ iχ + p+) + bw+Bi′(ξw p+) + bb+Bi′(ξb p+) + bw+Bi′(ξw p+) + bw+Bi(ξw p+) + bw+Bi(ξw p+) + ρb ) + bb−Bi′(ξb p− ) + bb−Bi(ξb p− ) + bw−Bi(ξw p− p− ) + bb−Bi(ξb p− ) + bw−Bi(ξw p− p− p− p− p− p−(cid:2)ab−Ai(ξb p−(cid:2)aw−Ai(ξw p+)(cid:3) + ab−Ai(ξb p+)(cid:3) + aw−Ai(ξw p−(cid:2)ab−Ai′(ξb p−(cid:2)aw−Ai′(ξw p+)(cid:3) + p−(cid:2)aw+Ai(ξw p+)(cid:3) + ab−Ai′(ξb p+)(cid:3) + aw−Ai′(ξw p+)(cid:3) + ρw p+(cid:2)aw+Ai(ξw p+) + ρw p− p− p− p− , , )(cid:3)(cid:9)(cid:12)(cid:12)z=L )(cid:3)(cid:9)(cid:12)(cid:12)z=L )(cid:9)(cid:12)(cid:12)z=L )(cid:9)(cid:12)(cid:12)z=L )(cid:3)(cid:9)(cid:19)(cid:12)(cid:12)(cid:12)(cid:12)z=L )(cid:3)(cid:9) p+)(cid:3)(cid:9)(cid:19)(cid:12)(cid:12)(cid:12)(cid:12)z=L )(cid:9)(cid:19)(cid:12)(cid:12)(cid:12)(cid:12)z=L )(cid:9) p+)(cid:3)(cid:9)(cid:19)(cid:12)(cid:12)(cid:12)(cid:12)z=L ) + bw−Bi′(ξw p− ) + bw−Bi′(ξw p− ) + bb−Bi′(ξb p− p+) + bw+Bi(ξw p+) + bw+Bi(ξw , . (A4) The matrix form of the above equations for coefficients aµσ, bµσ leads to ex- pression (20). 17 18 [1] B.A. Bernevig and S. Zhang, IBM J. Res. and Dev. 50, 1 (2006). [2] J.M. Kikkawa and D.D. Awschalom, Phys. Rev. Lett. 80, 4313 (1998). [3] A. Hern´andez-Cabrera, P. Aceituno, and F. T. Vasko. Journal of Luminescence 128, 862 (2008). A. Hern´andez-Cabrera, P. Aceituno, and F. T. Vasko, Phys. Rev. B 74, 035330 (2006). [4] W. Zawadzki and P. Pfeffer, Semicond. Sci. Technol. 19 R1 (2004). [5] S. Nomura and Y. Aoyagi, Surface Sci. 529 171 (2003); F. Giorgis, F. Giuliani, C. F. Pirri, A. Tagliaferro, and E. Tresso, Appl. Phys. Lett. 72 (1998). [6] V. F. Gantmakher and I. B. Levinson, Carrier Scattering in Metals and Semi- conductors (North-Holland, Amsterdam, 1987). [7] G. Dresselhaus, Phys. Rev. 100, 580 (1955). [8] E.I. Rashba, Sov Phys. Sol. State 2, 1109 (1960); E.I. Rashba and V.I. Sheka, Sov. Phys. Solid State 3 1718 (1961). [9] M.I. Dyakonov, and V.Y. Kachorovskii, Sov. Phys. Semicond. 20 110 (1986). [10] F.T. Vasko. JETP Lett.30 360 (1979). [11] F.T. Vasko and A.V. Kuznetsov, Electronic States and Optical Transitions in Semiconductor Heterostructures (Springer, New York, 1999). [12] P. Pfeffer and W. Zawadzki, Phys. Rev. B 59 R5312, 1999; Y. Lin, T. Koga, and J. Nitta, Phys. Rev. B 71, 045328 (2005). [13] B. Das, D.C. Miller, S. Datta, R. Reifenberger, W.P. Hong, P.K. Bhattacharya, J. Singh, and M. Jaffe, Phys. Rev.B 39, R1411 (1989). [14] T. Kita, Y. Sato, S. Gozu, and S. Yamada, Physica B 298, 65 (2001). [15] A. Hern´andez-Cabrera, P. Aceituno, and F.T. Vasko, Phys. Rev. B 60, 5698 (1999). [16] P.A. Shields, R.J. Nicholas, K. Takashina, N. Grandjean, and J. Massies, Phys. Rev. B 65, 195320 (2002). 19 [17] A.A. Kiselev, E.L. Ivchenko, and U. Rossler, Phys. Rev. B 58, 16353 (1998). [18] I. Vurgaftman, J.R. Meyer, and L.R. Ram-Mohan, J. Appl. Phys. 89, 5815 (2001). [19] A. L. Ivanov, P. B. Littlewood, and H. Haug, Phys. Rev. B 59, 5032 (1999). [20] Following Refs. 12-14 and I. Vurgaftman, J.R. Meyer, and L.R. Ram-Mohan, J. Appl. Phys. 89, 5815 (2001), we use the data for In0.53Ga0.47As/In0.52Al0.48As structure: m = 0.041me, mb = 0.0754me, where me is the electron mass, factor g = 12.2, permittivity ǫ = 13.9, ∆Ec = 498 meV, ∆Ev = 197 meV, and εg = 813 meV. [21] A. Stroppa and M. Peressi, Phys. Rev. B 71, 205303 (2005). [22] P. Velling, M. Agethen, W. Prost, F.J. Tegude, J. Crystal Growth 221, 722 (2000). C.D. Bessire, M.T. Bjork, H. Schmid, A. Schenk, K. B. Reuter, and H. Riel, Nano Lett 11, 4195 (2011). [23] D. M. Graham, P. Dawson, M. J. Godfrey, M. J. Kappers and C. J. Humphreys, Appl. Phys. Lett. 89, 211901 (2006); D.R. Hang, C.F. Huang, W.K. Hung, Y.H. Chang, J.C. Chen, H.C. Yang, Y.F. Chen, D.K. Shih, T.Y. Chu, and H.H. Lin, Semicond. Sci. and Tech. 17, 999 (2002); M. Poggio, G. M. Steeves, R. C. Myers, N. P. Stern, A. C. Gossard, and D. D. Awschalom, Phys. Rev. B 70, 121305(R) (2004). [24] K. F. Karlsson, H. Weman, M.-A. Dupertuis, K. Leifer, A. Rudra, and E. Kapon, Phys. Rev. B 70, 045302 (2004); C.I. Shih, C.H. Lin, S.C. Lin, T.C. Lin, K.W. Sun, O. Voskoboynikov, C.P. Lee and Y.W. Suen, Nanoscale Res. Lett. 6:409 (2011). [25] M. V. Yakunin, G. A. Al'shanski, Yu. G. Arapov, V. N. Neverov, G. I. Kharus, N. G. Shelushinina, B. N. Zvonkov, E. A. Uskova, A. de Visser, and L. Pono- marenko, Semiconductors 39, 107 (2005). [26] P. S. Alekseev, V. M. Chistyakov, and I. N. Yassievich, Semiconductors 40, 1402 (2006); G. Goldoni and A. Fasolino, Surface Science 305, 333 (1994). 20 600 dl dr ) 0 = p , z ( 2 , ) V e m ( c 300 0 Ec eF z 0 db 100 z (Å) 200 FIG. 1: (Online color) Conduction band diagram for undoped asymmetric double quantum well. Horizontal thin lines show electron energy levels and thin curves correspond to squared wave functions close to the resonance. 21 2 1 0.015 0.010 0.005 0.000 -50 0 /p 0 p x (a) (b) 2 1 0.015 0.010 0.005 0.000 -50 0 /p0 50 px 100 0 100 z(Å) 200 0 100 z(Å) 200 50 100 (c) 0.015 0.010 0.005 0.000 -50 p x/ p 0 0 50 100 200 0 100 ( Å ) z 2 2 (d) 0.015 0.010 0.005 0.000 -50 0 50 100 p x/ p 0 200 0 100 ( Å ) z FIG. 2: (Online color) Squared wave functions vs normalized px/p0 momentum for the first spin resonant levels. 22 100 95 ) V e m ( 90 k E 85 -100 -50 0 p x/p 0 0 100 50 p y/p 0 50 100 -50 -100 FIG. 3: (Online color). Near parabolic dispersion relations of the two coupled levels close to the resonance, and their corresponding spin down (black) and spin up sublevels. 23 ) V e m ( E 83.65 83.64 83.63 83.62 83.61 83.60 -5 0 p x/ p 0 5 5 0 y/ p p 0 -5 FIG. 4: (Online color) Bottom of the two deepest paraboloids just after resonance and detail of spin down (black) and spin up anticrossing. Levels are mainly located in the narrow (left) QW. 24 83.70 ) V e m ( 83.65 1 E 83.60 -10 -5 0 5 10 px/p0 FIG. 5: (Online color) Two dimensional dispersion relations for py/p0 = 0. Inset shows a magnified image of the spin down and spin up anticrossing. Solid line; spin down; dashed line; spin up. 25 0 p / y p (a) 4 2 0 -2 -4 4 (c) 2 0 -2 -4 (b) (d) -4 -2 0 2 4 -4 -2 0 2 4 px/p0 FIG. 6: (Online color). Contour plot of the deepest energy levels around anticross- ing. (a) E1σ = 83.610 meV; (b) E1σ = 83.611 meV; (c) E1σ = 83.612 meV; and (d) E1σ = 83.614 meV. Outer line: spin down; inner line: spin up. 26 3 D 2 2 1 0 82 H=0 T H=0.01 T H=0.05 T H=0.1 T 84 94 (meV) 96 FIG. 7: (Online color) Density of states for the same electric field of previous figures and different magnetic fields. 27 =0.2 meV =0.4 meV =1 meV ) s t i n u . b r a ( E L P I 80 90 g meV 100 FIG. 8: (Online color) Photoluminescence excitation intensity for H = 0.1 T and different γ values. 28 0.04 0.03 0.02 ) s t i n u . b r a ( E L P I 0.01 ( m 0.2 0.4 0.6 0.8 1.0 e V ) 83 82 86 85 ( m e V ) 84 g - FIG. 9: (Online color) Photoluminescence excitation vs γ corresponding to the two first transitions. Parameters are the same as in Fig. 8.
1106.1794	1	1106	2011-06-09T12:57:17	Transition between SU(4) and SU(2) Kondo effect	[ "cond-mat.mes-hall", "cond-mat.str-el" ]	Motivated by experiments in nanoscopic systems, we study a generalized Anderson, which consists of two spin degenerate doublets hybridized to a singlet by promotion of an electron to two conduction bands, as a function of the energy separation $\delta$ between both doublets. For $\delta$=0 or very large, the model is equivalent to a one-level SU(N) Anderson model, with N=4 and 2 respectively. We study the evolution of the spectral density for both doublets ($\rho_{1 \sigma}(\omega)$ and $\rho_{2 \sigma}(\omega)$) and their width in the Kondo limit as $\delta$ is varied, using the non-crossing approximation (NCA). As $\delta$ increases, the peak at the Fermi energy in the spectral density (Kondo peak) splits and the density of the doublet of higher energy $\rho_{2 \sigma}(\omega)$ shifts above the Ferrmi energy. The Kondo temperature $T_K$ (determined by the half width at half maximum of the Kondo peak in density of the doublet of lower energy $\rho_{1 \sigma}(\omega)$) decreases dramatically. The variation of $T_K$ with $\delta$ is reproduced by a simple variational calculation.	cond-mat.mes-hall	cond-mat	Transition between SU(4) and SU(2) Kondo effect L. Tosi a P. Roura-Bas b A. A. Aligia a aCentro At´omico Bariloche and Instituto Balseiro, Comisi´on Nacional de Energ´ıa At´omica, 8400 Bariloche, Argentina bDpto de F´ısica, Centro At´omico Constituyentes, Comisi´on Nacional de Energ´ıa At´omica, Buenos Aires, Argentina Abstract Motivated by experiments in nanoscopic systems, we study a generalized Anderson, which consists of two spin degenerate doublets hybridized to a singlet by promotion of an electron to two conduction bands, as a function of the energy separation δ between both doublets. For δ=0 or very large, the model is equivalent to a one-level SU(N) Anderson model, with N=4 and 2 respectively. We study the evolution of the spectral density for both doublets (ρ1σ(ω) and ρ2σ(ω)) and their width in the Kondo limit as δ is varied, using the non-crossing approximation (NCA). As δ increases, the peak at the Fermi energy in the spectral density (Kondo peak) splits and the density of the doublet of higher energy ρ2σ(ω) shifts above the Ferrmi energy. The Kondo temperature TK (determined by the half width at half maximum of the Kondo peak in density of the doublet of lower energy ρ1σ(ω)) decreases dramatically. The variation of TK with δ is reproduced by a simple variational calculation. Key words: Anderson model, spectral density, Kondo temperature, non-crossing approximation PACS: 75.20.Hr, 75.10.Jm 1 Introduction The Kondo effect, found originally for systems with magnetic impurities in metals is now present in a variety of nanoscopic systems, including semicon- ducting quantum dots [1], magnetic adatoms on surfaces [1,2,3] and carbon nanotubes [4]. In the latter, in addition to the spin Kramers degeneracy, there ∗ Tel: 54-11 67727102; FAX: 54-11 67727121; [email protected] Preprint submitted to Physica B 15 October 2018 is in addition orbital degeneracy due to the "pseudospin" degree of freedom related with the particular band structure of graphene. This leads to an SU(4) Kondo effect which has been observed experimentally [5,6] and also discussed theoretically [7,8,9]. In particular, Lim et al. have studied the spectral density when the SU(4) symmetry is reduced to SU(2), mainly by a change in the tunneling matrix elements [7,9]. Our main motivation in the problem arises from interference phenomena in systems of quantum dots [10,11,12] or molecules [11,13,14]. For example de- pressions in the integrated conductance through a ring described by the Hub- bard or t − J model, pierced by an Aharonov-Bohm magnetic flux, related with spin-charge separation [10,11], are due to a partial destructive interfer- ence when the energy of two doublets cross [11]. Similar interference effects were predicted in molecular transistors [13,14]. The effective Hamiltonian near the crossing is discussed in the next section. To our knowlege, the conductance in these systems has so far been calculated using approximate expressions or a slave-boson formalism [12], which is valid only for very low temperatures and applied bias voltages. This work is a step towards a more quantitative theory to describe the transport through similar systems, treating the effective Hamiltonian within the non-crossing approxi- mation (NCA) [15,16]. Work is in progress to deal with the non-equilibrium situation, which is necessary within our formalism to calculate the current. In this paper, we report on our study of the spectral density of the model as a function of the splitting δ between both doublets in the Kondo regime. We also calculate the dependence of the Kondo temperature TK with δ and analyze the validity of the Friedel sum rule [17]. 2 Model We start from a model in which two doublets of an interacting system are hy- bridized with a singlet by promotion of an electron to two conducting leads. This is the low-energy effective Hamiltonian for several systems with par- tial destructive interference, such as Aharonov-Bohm rings [11] or aromatic molecules [13,14] connected to conducting leads. The Hamiltonian can be writ- ten as [11] H = Es0ih0 + X iσ Eiiσihiσ + X (Viνiσih0cνkσ + H.c.), νkσ + X iνkσ ǫνkc† νkσcνkσ (1) 2 where the singlet 0i and the two doublets iσi (i = 1, 2; σ =↑ or ↓) denote the localized states (representing for example the low-energy states of an isolated molecule), c† νkσ create conduction states in the left (ν = L) or right (ν = R) lead, and Viν describe the four hopping elements between the two leads and both doublets, assumed independent of k. Changing the phase of the conduction states and the relative phase between both doublets, three among the four Viν can be made real and positive. The phase φ of the remaining hopping V eiφ, depends on the particular system and its symmetry. For example in molecules with rotational symmetry φ = (K1 − K2)l, where l is the distance between the sites connected to the left and right leads, and Ki is the wave vector of the state iσi, which can be modified with an applied magnetic flux [11]. In absence of magnetic flux and when the relevant states have wave vector Ki = ±π/2, as in rings with a number of atoms multiple of four, φ = π and there is complete destructive interference in transport [11,14]. In the following we will assume this case, with states 1 and 2 related by symmetry implying V1ν = V2ν. We further assume symmetric leads, ǫLk = ǫRk = ǫk, ViL = ViR. Then, without loss of generality we can take V1L = V1R = V2L = V > 0, V2R = −V , and E2 ≥ E1. For these parameters, changing basis c† 1kσ = (c† Lkσ − Rkσ)/√2, the Hamiltonian takes the form of an SU(4) Anderson model with c† a "field" δ = E2 − E1 and on-site hybridization V ′ = √2V Rkσ)/√2, c† Lkσ + c† 2kσ = (c† H = Es0ih0 + X iσ Eiiσihiσ + X ikσ ǫkc† ikσcikσ +V ′ X ikσ (iσih0cikσ + H.c.). (2) Interchanging the doublet index (1 or 2) with the spin index σ, one realizes that this model also describes transport trough a carbon nanotube with electron density depleated at two points (so as to created an SU(4) quantum dot in the middle) under a real applied magnetic field. 3 Spectral density In Fig. 1 we present numerical results for the spectral density ρiσ of the SU(4) Anderson model (E1 = E2) in the Kondo regime ǫF − Ei ≫ ∆, where the hy- bridization function ∆ = π Pk(V ′)2δ(ω − ǫk), assumed independent of energy. We set Γ = 2∆ = 1 as the unit of energy. We also assume a conduction band 3 0.6ρ1σ 0.6 0.5 0.4 0.4 0.2 0 0.3 0.2 0.1 0 T=0.4 T=0.03 T=0.01 T=0.005 T=0.001 0 0.01 0.02 0.03 0.04 -5 -4 -3 -2 ω -1 0 1 Fig. 1. Spectral density as a function of energy for different temperatures. The inset shows a detail near the Fermi energy. Parameters are ∆ = 0.5. D = 10, E1 = E2 = −4. The lowest temperature is T = 10−3 = 0.076TK . 0.6 ρ1σ 0.4 0.2 0 0.6 0.4 0.2 0 0.6 0.4 0.2 0 δ=0 δ=0.007 ρ2σ δ=0.12 -0.1 ω 0 0.1 Fig. 2. Spectral densities for levels 1 (full line) and 2 (dashed line) as a function of energy for different values of E2 = E1 + δ and T = 10−3. Other parameters as in Fig. 1. Dot-dashed line corresponds to ρ2σ for δ = 0.015, 0.3 and 0.6. symmetric around the Fermi level ǫF = 0, of half width D = 10. The spectral density shows a broad charge transfer peak near E1. For tem- peratures below a characteristic energy scale TK (defined below), ρiσ develops a narrow peak around the Fermi level. In contrast to the better known one- level SU(2) case, this peak is displaced towards positive energies and is much broader, as discussed in Section 5. 4 The evolution of the spectral densities as E2 is displaced to larger energies, breaking the SU(4) symmetry is shown in Fig. 2. The peak near the Fermi energy of ρ2σ(ω) is displaced towards positive energies (near δ = E2 − E1). In contrast, the corresponding peak in ρ1σ(ω) narrows significantly and displaces towards the Fermi energy. This implies that the Kondo temperature TK defined as the half width at half maximum of this peak, also decreases strongly. The evolution of TK with δ is discussed in the Section 5. In addition ρ1σ develops a broad peak near energy −δ which becomes visible when δ becomes greater than TK. 4 Friedel sum rule The Anderson model studied has a Fermi liquid ground state which satisfies well known relationships at zero temperature. One of them is the Friedel sum rule which relates the spectral density at the Fermi level for each "pseudospin" channel with the occupation of that channel [17]. For the simplest case of a constant density of conduction states, this rule reads ρiσ(ǫF ) = 1 π∆ sin2(πniσ), (3) where niσ = hiσihiσi. This is an exact relationship for a Fermi liquid, which is not necessarily sat- isfied by approximations. In particular, it is known that at very low temper- atures, the NCA has a tendency to develop spurious spikes in ρiσ(ω) at the Fermi energy, while thermodynamic properties, such as expectation values are accurately reproduced [15,16]. In Fig. 3, we compare both members of Eq. (3) for the lowest lying doublet, at a temperature T = 0.1TK [18] low enough so that no further increase in ρiσ(ǫF ) takes place as the temperature is lowered (according to physical expectations and Eq. (3)), but high enough to prevent the presence of spurious spikes. The disagreement lies below 20 %. The agreement improves as the parameters are moved deeper in the Kondo regime ǫF − E1 ≫ ∆. Thus, while the spectral density at zero temperature is not well represented by the NCA results at T = 0, one can take the values at T = 0.1TK as a reasonable description of the correct T = 0 ones. This statement is supported by the comparison of results obtained by NCA and numerical renormalization group for the one- level SU(2) case [19]. 5 sin2(π<n1σ>/2) ρ1σ(0)π∆ 1 0.8 0.6 0.4 0.2 0 1e-05 0.0001 0.001 0.01 0.1 δ 1 Fig. 3. Squares: rescaled spectral density of the lowest lying level at the Fermi energy as a function of δ. Circles: corresponding (more accurate) result given by Eq. (3). Lines are guides to the eye. Parameters are T = 0.1TK and the rest as in Fig. 1. 5 Kondo temperature From the half width at half maximum of the peak nearest to the Fermi energy of the spectral density of the lowest level (ρ1σ(ω)), we have calculated the Kondo temperature of the system TK for several values of δ. This requires to solve the self-consistent NCA equations up to low enough temperatures (about 0.1 TK as discussed above) so that the height of the peak does not increase significantly with further lowering of the temperature [18]. Fortunately, the result is not very sensitive to the ratio T /TK. The results are shown in Fig. 4 and compared with Eq. (5) obtained from a variational calculation as explained below. We see that except for an overall multiplicative factor, the agreement is quite good, in spite of the fact that TK changes by nearly two orders of magnitude. To provide an independent estimate of TK, we have calculated the stabilization energy of the following variational wave function ψi = αsi + X ikσ βik(iσih0cikσ)si, (4) where si is the many-body singlet state with the filled Fermi sea of conduction electrons and the state 0i at the localized site, while α and βik are variational parameters. From the resulting optimized energy E, we can define the stabi- lization energy as T ∗ K = E1 − E. The Kondo energy scale defined in this way becomes K = n(D + δ)D exp [πE1/(2∆)] + δ2/4o1/2 T ∗ − δ/2. (5) 6 0.1 TK 0.01 0.001 0.0001 1e-05 1e-06 0.0001 0.01 δ=E2-E1 1 Fig. 4. Squares: Kondo energy scale determined by the width of the peak in the spectral density near the Fermi energy as a function of the splitting δ. The temper- atures used were T = 10−3, T = 10−4, and T = 10−5 depending on δ Dashed line: corresponding variational result Eq. (5) multiplied by a factor 0.606. This expression interpolates between the SU(4) result (T ∗ for δ = 0) and the SU(2) one for one doublet only (T ∗ for δ → +∞). K = D exp [πE1/(4∆)] K = D exp [πE1/(2∆)] 6 Summary and discussion We have studied an impurity Anderson model containing two doublets, which interpolates between the cases for one level with SU(4) and SU(2) symme- try, and is of interest for several nanoscopic systems, using the non-crossing approximation (NCA). We have shown that the NCA provides reasonable results for the equilibrium spectral density. The values of the spectral density for both doublets agree within 20% with the predictions of the Friedel sum rule, in spite of the fact that it is not expected to satisfy Fermi liquid relationships at zero temperature. In addition, the Kondo temperature scale TK obtained from the width of the peak in spectral density near the Fermi energy agrees very well with the stabilization energy of a variational calculation, in spite of the change in sev- eral orders of magnitude of TK when the splitting between both doublets is changed. The approach seems promising for studying transport properties within a non- equilibrium formalism. Work in this direction is in progress. 7 7 Acknowledgments One of us (A. A. A.) is partially supported by CONICET, Argentina. This work was partially supported by PIP No 11220080101821 of CONICET, and PICT Nos 2006/483 and R1776 of the ANPCyT. References [1] L. P. Kouwenhoven and L. I. Glazman, Phys. World 14, 33 (2001). [2] A. A. Aligia and A. M. Lobos, J. Phys.: Condens. Matter 17, S1095-S1122 (2005). [3] A. A. Aligia, Phys. Rev. Lett. 96, 096804 (2006). [4] J. Nygard, D. Cobden, and P. Lindelhof, Nature (London) 408, 342 (2000). [5] P. Jarillo-Herrero, J. Kong, H. S. J. van der Zant, C. Dekker, L. P. Kouwenhoven, and S. De Franceschi, Nature (London) 434, 484 (2005). [6] A. Makarovski, J. Liu, and G. Finkelstein, Phys. Rev. Lett. 99, 066801 (2007). [7] J. S. Lim, M. S. Choi, M. Y. Choi, R. Lopez, and R. Aguado, Phys. Rev. B 74, 205119 (2006). [8] C. A. Busser and G. B. Martins, Phys. Rev. B 75, 045406 (2007). [9] F. B. Anders, D. E. Logan, M. R. Galpin, and G. Finkelstein, Phys. Rev. Lett. 100, 086809 (2008). [10] K. Hallberg, A. A. Aligia, A. P. Kampf, and B. Normand, Phys. Rev. Lett. 93, 067203 (2004). [11] J. Rinc´on, A. A. Aligia, and K. Hallberg, Phys. Rev. B 79, 035112 (2009). [12] A. M. Lobos and A. A. Aligia, Phys. Rev. Lett. 100, 016803 (2008). [13] G. Begemann, D. Darau, A. Donarini, and M, Grifoni, Phys. Rev. B 77, 201406(R) (2008). [14] J. Rinc´on, K. Hallberg, A. A. Aligia, and S. Ramasesha, Phys. Rev. Lett. 103, 266807 (2009). [15] N. S. Wingreen and Y. Meir, Phys. Rev. B 49, 11040 (1994) [16] P. Roura Bas and A. A. Aligia, J. Phys. Cond. Matt. 22, 025602 (2010). [17] D.C. Langreth, Phys. Rev. 150, 516 (1966). [18] Some technical details that allowed us to reach such low temperatures are briefly explained elsewhere [16]. [19] T. A. Costi, J. Kroha, and P. Wolfle, Phys. Rev. B 53, 1850 (1996) 8
1204.1942	1	1204	2012-04-09T18:05:17	Frequency and phase noise of ultra-high Q silicon nitride nanomechanical resonators	[ "cond-mat.mes-hall" ]	We describe the measurement and modeling of amplitude noise and phase noise in ultra-high Q nanomechanical resonators made from stoichiometric silicon nitride. With quality factors exceeding 2 million, the resonators' noise performance is studied with high precision. We find that the amplitude noise can be well described by the thermomechanical model, however, the resonators exhibit sizable extra phase noise due to their intrinsic frequency fluctuations. We develop a method to extract the resonator frequency fluctuation of a driven resonator and obtain a noise spectrum with dependence, which could be attributed to defect motion with broadly distributed relaxation times.	cond-mat.mes-hall	cond-mat	Frequency and phase noise of ultra-high Q silicon nitride nanomechanical resonators King Y. Fong, Wolfram H. P. Pernice, and Hong X. Tang† Departments of Electrical Engineering, Yale University, New Haven, CT 06511, USA Abstract We describe the measurement and modeling of amplitude noise and phase noise in ultra-high Q nanomechanical resonators made from stoichiometric silicon nitride. With quality factors exceeding 2 million, the resonators’ noise performance is studied with high precision. We find that the amplitude noise can be well described by the thermomechanical model, however, the resonators exhibit sizable extra phase noise due to their intrinsic frequency fluctuations. We develop a method to extract the resonator frequency fluctuation of a driven resonator and obtain a noise spectrum with Bk T f dependence, which could be attributed to defect motion with / broadly distributed relaxation times. PACS numbers: 85.85.+j, 62.25.Fg, 05.40.-a, 63.50.Lm Present address: Institute of Nanotechnology, Karlsruhe Institute of Technology, 76133 Karlsruhe, Germany † Corresponding author: [email protected] 1 Nanoelectromechanical systems (NEMS) have shown great promise for both fundamental science and applications, by introducing new tools for studying quantum physics [1-3] and enabling high performance devices for sensing and oscillator applications [4-6]. Recently, nanomechanical resonators made of high stress stoichiometric silicon nitride have provided a solution to one of the most challenging quests in NEMS research, i.e., achieving very high mechanical quality factors (Q factors) [7-9]. For a fixed resonance frequency, high Q is equivalent to a low dissipation rate, which implies weaker coupling to the thermal bath, higher sensitivity to external signals, and lower minimum operating power [10]. To enable real world applications based on such devices, understanding the underlying dissipation and noise mechanism is crucial. Such an understanding is also a key for implementing quantum measurements of phonon shot noise as recently proposed [11]. Several models have been developed to explain the damping mechanism [12-14]. Yet, the noise characteristics such as the amplitude/phase noise and frequency fluctuation have not been addressed so far. In NEMS, resonator frequency fluctuations can arise due to various processes, such as temperature fluctuations [15], adsorption-desorption of molecules on the device surface [16,17], molecule diffusion along the resonator [18]. These frequency fluctuations are usually accompanied by thermal motion of the resonator, which has to be separated to reveal true resonator frequency fluctuations as recently suggested [19]. Here, we experimentally study the frequency fluctuation of ultra high Q nanomechanical resonators made of stoichiometric silicon nitride. We derive and validate a mathematical model to account for the observed frequency fluctuations. Our results show that high Q devices are actually more susceptible to frequency flickering. By measuring the phase noise of the device, a resonator frequency noise spectrum with Bk T f dependence is extracted. We propose that this fluctuation is due to defect motion / 2 with a broad spectrum of thermally-activated relaxation times. Our results give insight into the noise mechanism of high Q silicon nitride resonators and are important for NEMS based applications, such as sensors and low phase noise optomechanical oscillators. We fabricate nanomechanical resonators from LPCVD stoichiometric silicon nitride. Fig. 1(a) shows the SEM image of a fabricated device, which contains a doubly-clamped beam resonator with dimensions 380μm 0.33μm 0.2μm   , an optical readout circuit with grating coupler input/output ports as well as an on-chip interferometer. (Fabrication details can be found in Refs. [9,20]). The devices are mounted in a helium cryostat and the temperature of the sample stage is regulated from 5K to room temperature with milli-Kelvin precision. The pressure is kept below 10-6 Torr for all the measurements carried out. The sample stage is equipped with a piezodisk actuator for mechanical actuation. The mechanical quality factor is measured using a ring-down method [9] and is plotted against temperature in Fig. 1(b). At room-temperature we measure a quality factor of the fundamental in-plane flexural mode of roughly 400,000, which increases up to 2.2 million when the device is cooled down to 5K. The observed high Q of tensile stressed silicon nitride nanostring can be attributed to the increased elastic energy stored in the tension [12,14]. For devices with cross-sectional dimensions similar to the device used here, it was suggested that the Q is limited by internal material damping [12]. It has also been shown that the quality factor of silicon nitride nanostring resonators is not sensitive to pressure change once the pressure is reduced below milli-Torr levels [7,9]. Therefore, we can safely neglect the effect of pressure change on the Q during the cool down. Under tensile stress, a beam with length of L has a mechanical fundamental mode profile of z t ( , )   z L x t ( ) sin( /  ) , where the displacement ( )x t is defined as the displacement of the middle of the beam. Under such normalization condition, we define the effective mass m as half 3 of the total mass so that mx t ( ) / 2 2 correctly expresses the total kinetic energy of the resonator. (Details of definition of effective mass can be found, for example, in section 3.1 of Ref [3].) Because of the sensitive optical interferometric read-out, thermomechanical noise is resolved at all temperatures concerned in this work and thus allows us to calibrate the displacement of the resonator. Meanwhile, the optical power is kept minimal at around 10μW in order not to perturb the thermal equilibrium or induce any optomechanical backaction. Device displacement at different driving intensity is plotted in Fig. 1(c). At large drive the device response displays Duffing nonlinearity due to stiffening effects in the displaced beam. If the nonlinear equation of motion  x   x x 2 2 2      r r K x 3 3  F cos(  d t m ) / and harmonic motion x A  t  cos( )  d are assumed, where ωr is the resonance frequency, / 2r Q   is the damping rate, F is the external driving force, ωd is the driving frequency and A is the amplitude of the displacement, the nonlinear coefficient K3 can be found by fitting a cubic polynomial to the graph 2F vs 2A . For the present case, we obtain K 3  6.9 10 nm 9   2  , which agrees with the theoretical value K 3  L E / 4 2 2   expected for a tensile stressed beam experiencing additional stress due to large strain [21]. Here, E and  are the Young’s modulus and the stress, respectively. To separate and quantify the amplitude and phase noise, we used a lock-in amplifier (Zurich Instruments HF2LI [22]) to measure the two quadrature components of the device in response to a sinusoidal drive. The lock-in bandwidth is chosen such that within the bandwidth LI the noise power is dominated by thermomechanical noise, while r Q /     LI so that most of the thermal noise power is captured. Fig. 2(a) shows the measured two quadrature components in the complex plane in the presence (lower datasets) and absence (upper datasets) of the drive. The data acquisition time for each dataset is kept at 100s, which is much longer than the device 4 ringdown time. The black solid line traces out the polar plot of the driven response when the driving frequency is swept across the resonance. To ensure linear operation, i.e., 2 K A 3  1 / Q , the device is driven to an amplitude of A ~ 4.5nm. In the absence of the drive, the fluctuation of the two quadrature components is due to thermomechanical noise. As expected, the noise is random in phase and therefore has a circularly symmetric distribution [23], as illustrated for clarity in a zoom-in in Fig. 2(b). Fig. 2(c) shows that the fluctuation is Gaussian distributed and has a standard deviation  agrees with the expected value x 2   kT / m 2 r . Here 2x is the ensemble average of the squared displacement fluctuation. In the driven case we would expect to see the same fluctuation profile if thermomechanical noise was the dominant noise source. However, this is apparently not the case. While the amount of amplitude noise is similar to that of the thermomechanical noise, there are large fluctuations along the resonance circle. Fig. 2(d) shows the measured two quadrature components under different driving intensity. It is clear that the phase angle enclosed by the extra fluctuation remains constant when the drive is increased. Therefore, this extra phase noise is independent of the driving intensity. To confirm that the two quadrature components move indeed along the resonance circle, we show in Fig. 2(e) a dataset with a longer acquisition time of 2000s. The results therefore suggest that it is the resonator frequency that is fluctuating. Here, we point out that frequency fluctuation induced by amplitude noise via mechanical nonlinear effects cannot account for the observed phenomenon. Such effect can be estimated by A A rK     3 , which for our case has a value of 2 10 Hz 4     at 5K. This value is orders of magnitude smaller than the observed fluctuation. Besides, the corresponding phase fluctuation is expected to be proportional to the amplitude, which is contrary to our observation. 5 To understand the observed phenomenon, we theoretically analyze a driven harmonic resonator whose resonance frequency r t fluctuates in time. Its displacement ( )x t ( ) follows a stochastic differential equation   x t x t ( ) 2 ( )     r 2 t ( ) x t ( )  F cos(  d t m f ) /  th t m ( ) / , (1) where F t is the external driving force, and  is the damping rate. cos( )d f th )(t is the corresponding thermal fluctuation force due to coupling to the external bath. It is a white force noise assumed to have a correlation time much shorter than any time scale concerned, i.e., f th t ( ) f th t   ( ( ) )   . The Duffing nonlinear term is assumed to be negligible, as justified from above discussions. The displacement ( )x t can be expressed in terms of amplitude and phase as x t ( )  A t ( ) cos  t  t ( )  d  . Here,  is defined as the phase relative to the drive. Before we proceed, we would like to emphasize that there are two distinct frequency concepts, namely the resonator frequency r and the instantaneous frequency i . The former is the natural frequency at which the resonator would oscillate in the absence of drive while the latter is the apparent oscillation frequency which is defined as d /      i d dt . Note that a driven passive resonator should always follow the driving frequency d , with a shift of /d dt . A change of resonator frequency r indirectly affects the instantaneous frequency i by altering r t the phase , ) (   . For a resonator with low damping rate r t ( )   driven near resonance, i.e., t t ( ) ( )      d r and t ( )  d , it can be shown that the slowly varying complex displacement u t ( )  A t ( ) exp  i t ( )  satisfies the equation  u t ( )   i t u t ( ) ( )          F f th  i m t ( ) 2  d , where 6  f th t ( )  2 f t ( ) exp( i t  d ) th integral form [19] as [24]. Using an appropriate integrating factor, ( )u t can be expressed in u t ( )  t   dt ' exp    t (    t ')  t  i dt t ' t " ( ")       F f  th  i m t ( ') 2  d . (2) We assume that in the time span concerned the phase accumulation due to the resonator frequency fluctuation is small, i.e., t '  t dt t  " ( ") 1 . This condition can always be satisfied if the chosen time span is short enough. Upon integration by parts, the amplitude and phase of the displacement, given by A t ( )  u t Im{ ( )} and t u t ( ) Re{ ( )} (   F m / 2 / 2 )    d , can be re- written as   A t  t   dt ' exp  t (    t ')   F f   R t ( ')     t ( ')   g t ( ') I     2  d m t ( )   t   dt ' exp  t (    t ')   t ( ')       f I t ( ') F t ( ')    g t ( ') R F      2 , (3a) (3b) where Rf , If and Rg , Ig are the real and imaginary parts of thf and  g t ( ') th  t '   dt " exp   t ( '    t ")   f th t ( ") , respectively. The first term of Eq. (3a) gives the steady state driven amplitude A 0  F m / 2 d . In both expressions for ( )A t ( )t , the second terms and represent the contribution from thermal fluctuation and the third terms represent the mixing between thermal and resonator frequency fluctuation. The first term of Eq. (3b) is the phase noise due to resonator frequency fluctuations. Note that this is the only term that is independent of the driving force while all the other terms are inversely proportional to the drive. Therefore this term will eventually dominate if the drive is sufficiently large. Our experimental data shown in Fig. 2(d) confirms that the first term 7 dominates in the present case. Thus, the expression for the phase can be simplified as t ( )   t   exp  t (    t ')  t dt ( ')  '  / 2  . It is then straightforward to show that the transfer- ( )t is given by ( )t and function between the spectral densities of S S ( )     ) ( 2 2      . (4) Note that here we did not assume any physical model of the frequency noise nor make any ( )t . Our result is thus generally valid as long as assumption about the correlation time scale of the condition t '  t dt t  " ( ") 1 is satisfied. This condition can always be guaranteed by considering a short enough measurement time. One important implication of Eq. (4) is that the area under the transfer function   0 d Q / ) / ( 2 2        0 is proportional to Q, which means that the amount of resonator frequency noise transferred into phase noise is larger for devices with higher Q. It can be intuitively understood as the result of steeper slope in the phase-frequency curve. Another way of looking at this is to consider the instantaneous frequency spectral density fS S ( ( ( ) ) 2 2 2 2             )   S  ( )  . It can be seen that the transfer function is a high pass filter: the higher the Q, the lower the roll-off frequency and more noise is allowed to pass. This is essentially the reason why we are able to quantify the frequency noise in our ultrahigh Q devices. It is generally believed that high Q always has positive impact on device performance. Here we show however that an increase in device performance based on higher Q is accompanied by higher susceptibility to the resonator’s inherent frequency noise that exists. This intrinsic resonator frequency noise imposes a detection limit for applications that rely on the measurement of resonance shift, such as mass and force gradient sensing. The minimum 8 resolvable frequency change min can be expressed as the change of frequency that produces the same amount of phase shift as the total phase noise within the measurement bandwidth 1/ 2  imposed by thermomechanical noise [25], min does not depend on the driven oscillation . Contrary to the frequency detection limit  d S ( ) 2 2        r Q / 2    min  or bw ,   bw  0  amplitude. Therefore, the dynamic range of the device is no longer a determinant factor for the frequency sensitivity when the phase noise of the device is dominated by the intrinsic resonator frequency noise. To verify and apply the results obtained above, we measure the two quadrature components and compute the amplitude and phase spectral densities by periodogram using the Hanning window function. The measurement times for 5K, 78K and 296K are chosen to be 125s, 25s, and 8s, respectively, such that the condition t '  t dt t  " ( ") 1 is meet. The obtained spectral densities for the amplitude and phase are plotted in Fig. 3(a) and 3(b), respectively. The amplitude and phase noise background of the drive source of our lock-in amplifier [22] is well below the device noise and therefore plays negligible role in the measured spectrum. The black solid lines show the noise contribution due to thermal fluctuation (the second terms in Eq. (3a) and (3b)). The amplitude noise matches well with the thermal term but the phase noise is significantly higher than that, which agrees with the observation from Fig. 2(a). Supported by the above analysis, we attribute this extra phase noise to resonator frequency fluctuation, whose spectral density is computed by Eq. (4) and plotted in Fig. 3(c). A 1/f noise clearly dominates from the lowest frequency of 0.008Hz up to 5Hz. By denoting S f (  )  S T ( ) / 0 f , we can extract S T from the 0 ( ) resonator frequency noise spectrum. In Fig. 4 we plot S T against temperature in a log-log 0 ( ) 9 scale. A power law fit aTn gives an exponent of 0.94 ± 0.10, which strongly suggests that S T 0 ( ) is linearly proportional to temperature T. Therefore, the resonator frequency noise spectrum has an overall Bk T f dependence. / Previously, several processes have been identified to be the sources of resonator frequency fluctuation in nanomechanical systems, such as temperature fluctuation [15], adsorption- desorption of molecules on the device surface [16,17], and molecule diffusion along the resonator [18]. However, these known processes do not give rise to the Bk T f noise spectrum / we observed. We propose that this Bk T f dependent noise spectrum can be understood within / the Dutta-Dimon-Horn model [26], which applies generally to fluctuations with a broad spectrum of thermally activated relaxation times. We suggest that in the present case the fluctuations originate from defect motions within the material [15]. For crystalline solids, local elastic distortion due to a point defect can be considered as an elastic dipole [27]. Depending on its orientation, the dipole produces a distortion of the local strain under a stress field and hence locally modifies the Young’s modulus. The idea of elastic dipoles can also be applied to model the inherent structural disorder in amorphous solids [28]. Following Ref. [15], we consider a defect state with two possible dipole orientations separated by an energy barrier E and a thermally activated reorientation time 0 exp(   )BE k T / . The idea is conceptually similar to the well known two-level states (TLS) model, which has successfully explained many thermal properties of amorphous solids at low-temperature [29] as well as damping of nanomechanical resonators [30]. While the TLS model describes the situation in which atoms with two energy minima are allowed to tunnel through the energy barrier quantum mechanically, here we ignore such tunneling effects and assume that the elastic dipoles reorient through thermal activation. For a tensile-stressed beam, it can be shown that such fluctuation in elasticity leads to a resonator 10 frequency noise spectrum of S     ( (1 ) 2 2 2   ) , where 2 is the variance of the frequency fluctuation. Following the Dutta-Dimon-Horn model [26] and assuming the distribution of activation energy )D E to be relatively flat, i.e, ( ( Bk T d D E dE n n ) / ( ) n  D E ( ) , the spectra for different activation energy can be integrated and give S ( ,   T )  Bk T    D E  , (5) where  E   k T  ln( B 0 ) . The result reproduces the Bk T f / dependence we observe experimentally. If we assume 0 to be 10-12s (on the order of the inverse phonon frequency [26]), the activation energy can be estimated to have a span in range of 0.01-0.8eV, which is consistent with the typical energy scale of TLS and defect motions [26,29]. Here we would like to emphasize that the defect states model is phenomenological. Its microscopic basis remains to be verified by separate experimental approaches, for example material characterization. In conclusion, we have studied the noise characteristics of a high Q nanomechanical resonator made of stoichiometric silicon nitride. The amplitude noise can be explained by the thermal motion of mechanical resonator, while an extra phase noise is observed. We develop a method to extract the resonator frequency fluctuation and obtain a noise spectrum with Bk T f dependence. / We propose that this frequency fluctuation is due to defect states with a broad spectrum of thermally-activated elastic dipole reorientation time. On the other hand, our theory also suggests that high Q devices are more susceptible to the resonator’s inherent frequency fluctuation. Our result is important for understanding the noise mechanism of the high Q silicon nitride nanomechanical resonators, as well as for applying this high performance device to sensing and oscillator applications. 11 Acknowledgements. We acknowledge funding from DARPA/MTO ORCHID program through a grant from the Air Force Office of Scientific Research (AFOSR). H.X.T acknowledges support from a Packard Fellowship and a CAREER award from the National Science Foundation. 12 References 1. A. D. O’Connell et al., Nature 464, 697 (2010). 2. J. D. Teufel, D. Li, M. S. Allman, K. Cicak, A. J. Sirois, J. D. Whittaker, and R. W. Simmonds, Nature 471, 204 (2011). 3. M. Poot and H. S. J. van der Zant, Phys. Reports 511, 273 (2012). 4. M. Li, H. X. Tang, and M. L. Roukes, Nature Nanotech. 2, 114 (2007). 5. A. K. Naik, M. S. Hanay, W. K. Hiebert, X. L. Feng, and M. L. Roukes, Nature Nanotech. 4, 445 (2009). 6. X. L. Feng, C. J. White, A. Hajimiri, and M. L. Roukes, Nature Nanotech. 3, 342 (2008). 7. S. S. Verbridge, H. G. Craighead, and J. M. Parpia, Appl. Phys. Lett. 92, 013112 (2008). 8. S. S. Verbridge, J. M. Parpia, R. B. Reichenbach, L. M. Bellan, and H. G. Craighead, J. Appl. Phys. 99, 124304 (2006). 9. K. Y. Fong, W. H. P. Pernice, M. Li, and H. X. Tang, Appl. Phys. Lett. 97, 073112 (2010). 10. K. L. Ekinci and M. L. Roukes, Rev. Sci. Inst. 76, 061101 (2005). 11. A. A. Clerk, F. Marquardt, and J. G. E. Harris, Phys. Rev. Lett. 104, 213603 (2010). 12. Q. P. Unterreithmeier, T. Faust, and J. P. Kotthaus, Phys. Rev. Lett. 105, 027205 (2010). 13. I. Wilson-Rae, R. A. Barton, S. S. Verbridge, D. R. Southworth, B. Ilic, H. G. Craighead, and J. M. Parpia, Phys. Rev. Lett. 106, 047205 (2011). 14. S. Schmid, K. D. Jensen, K. H. Nielsen, and A. Boisen, Phys. Rev. B 84, 165307 (2011). 15. A. N. Cleland and M. L. Roukes, J. Appl. Phys. 92, 2758 (2002). 13 16. Y. K. Yong and J. R. Vig, IEEE Trans. Ultrason. Ferroelectr. Freq. Control 36, 452 (1989). 17. Y. T. Yang, C. Callegari, X. L. Feng, and M. L. Roukes, Nano Lett. 11, 1753 (2011). 18. J. Atalaya, A. Isacsson, and M. I. Dykman, Phys. Rev. B 83, 045419 (2011). 19. Z. A. Maizelis, M. L. Roukes, and M. I. Dykman, Phys. Rev. B 84, 144301 (2011). 20. M. Li, W. H. P. Pernice, C. Xiong, T. Baehr-Jones, M. Hochberg, and H. X. Tang, Nature 456, 480 (2008). 21. Q. P. Unterreithmeier, S. Manus, and J. P. Kotthaus, Appl. Phys. Lett. 94, 263104 (2009). 22. http://www.zhinst.com; Phase noise of the drive source is measured to be 6 × 10-7 dBc/Hz at 1Hz offset from the carrier. 23. D. Rugar and P. Grütter, Phys. Rev. Lett. 67, 699 (1991). 24. Equation for ( )u t can be derived by substituting x  u i t exp(  )  * u exp(  i t  )   2 into   Eq. (1) and assuming   d , u  , and the rotating wave approximation. u d 25. K. L. Ekinci, Y. T. Yang, and M. L. Roukes, J. Appl. Phys. 95, 2682 (2004). 26. P. Dutta and P. M. Horn, Rev. Mod. Phys. 53, 497 (1981). 27. A. S. Norwick and B. S. Berry, Anelastic relaxation in crystalline solids (Academic Press, New York, 1972). 28. E. R. Grannan, M. Randeria, and J. P. Sethna, Phys. Rev. B 41, 7784 (1990); Phys. Rev. B 41, 7799 (1990). 29. W. A. Phillips, Rep. Prog. Phys. 50, 1657 (1987). 30. L. G. Remus, M. P. Blencowe, and Y. Tanaka, Phys. Rev. B 80, 174103 (2009). 14 15 Fig. 1. (a) SEM image of the fabricated device. (b) The measured mechanical quality factor plotted against temperature. (c) The driven response of the device at different driving intensity at 5K. 16 Fig. 2. (color online) (a) Two quadrature components of the device response with (lower datasets) and without drive (upper datasets) at different temperatures. The black line shows the driven response when driving frequency is swept across the resonance. (b) Zoom-in of (a) showing two quadrature components due to thermomechanical noise. (c) Normalized distribution of in-phase components shown in (b). Black lines show the Gaussian fit to the distributions. (d) Two quadrature components of the device response at different driving intensity at T=5K. Black lines show the driven response when driving frequency is swept across the resonance. (e) Two quadrature components of the device response taken in 2000s at T=296K. 17 Fig. 3. (color online) Power spectral density of amplitude (a), phase (b), and resonator frequency fluctuation (c) plotted against frequency. Black solid lines show the contribution from the thermal noise given by the second terms of Eq. (3). Black dashed line shows the 1/f trend. 18 Fig. 4. Temperature dependence of S T . The black solid line is the power law fit aTn to the 0 ( ) data with an exponent of 0.94 ± 0.10. 19
1102.1828	2	1102	2011-04-18T15:13:10	Coherent properties of nano-electromechanical systems	[ "cond-mat.mes-hall" ]	We study the properties of a nano-electromechanical system in the coherent regime, where the electronic and vibrational time scales are of the same order. Employing a master equation approach, we obtain the stationary reduced density matrix retaining the coherences between vibrational states. Depending on the system parameters, two regimes are identified, characterized by either ($i$) an {\em effective} thermal state with a temperature {\em lower} than that of the environment or ($ii$) strong coherent effects. A marked cooling of the vibrational degree of freedom is observed with a suppression of the vibron Fano factor down to sub-Poissonian values and a reduction of the position and momentum quadratures.	cond-mat.mes-hall	cond-mat	Coherent properties of nano-electromechanical systems 1 Dipartimento di Fisica & CNR-SPIN, Universit`a di Genova,Via Dodecaneso 33, 16146, Genova, Italy. G. Piovano1, F. Cavaliere1, E. Paladino2, and M. Sassetti1 2 Dipartimento di Fisica e Astronomia, Universit`a di Catania & CNR IMM MATIS Catania, C/O Viale A. Doria 6, Ed. 10, 95125 Catania, Italy. (Dated: October 29, 2018) We study the properties of a nano-electromechanical system in the coherent regime, where the electronic and vibrational time scales are of the same order. Employing a master equation approach, we obtain the stationary reduced density matrix retaining the coherences between vibrational states. Depending on the system parameters, two regimes are identified, characterized by either (i) an effective thermal state with a temperature lower than that of the environment or (ii) strong coherent effects. A marked cooling of the vibrational degree of freedom is observed with a suppression of the vibron Fano factor down to sub-Poissonian values and a reduction of the position and momentum quadratures. PACS numbers: 85.85.+j, 73.63.-b I. INTRODUCTION Nano-electromechanical systems1 (NEMS) represent an intriguing class of devices composed of a nano- mechanical resonator coupled to an electronic nanode- vice. Several examples of NEMS have been realized, ranging from single oscillating molecules,2 to suspended carbon nanotubes3 -- 6 and suspended nano-cantilevers or nano-beams.7,8 In all these devices, the coupling between mechanical and electronic degrees of freedom gives rise to peculiar transport phenomena like Franck-Condon blockade,3,9 -- 11 negative differential conductance5,12 -- 16 and remarkable noise characteristics.9,17 Along with their outstanding electronic properties, also mechanical ones are of extreme interest. Indeed, owing to the extreme sensitivity of the vibrating part to the spatial motion, NEMS have been proposed as novel detectors in scanning microscopes18 or as ultra-sensitive nano-scales, being able to detect the mass of even few molecules adhering to them.19,20 In order to successfully employ a NEMS as a precision position detector it is important to reduce its thermal fluctuations, eventually attaining the ultimate goal of cooling it down to its quantum ground state.21 Also, ultra-sensitive NEMS position detectors based on pecu- liar quantum states such as position-squeezed states22 have been proposed and experimentally realized.23 A great variety of NEMS setups have been investigated theoretically. Typically, the electronic part is composed of a semiconducting,24 -- 29 normal30 -- 32 or supercon- ducting33 -- 36 single quantum dot.37 Double-dot setups have been studied as well.27,38 Different models for the coupling between electrons and vibrons have been considered, ranging from the simple Anderson-Holstein (AH) model9,24,26 -- 29,39 -- 42 to microscopic models tai- lored for specific systems, such as suspended carbon nanotubes.43 -- 45 Also the influence of external dissipative baths29,46,47 or radiation fields38,48,49 has been analyzed. In the most complex configurations, interesting physical effects have been predicted. For instance, for a nano- mechanical resonator coupled to a microwave cavity,48 -- 50 to field driven quantum dot51 or in configurations with double quantum dots,38,52 phonon cooling has been found. With radio frequency quantum dots53 and a elec- tromagnetic cavity54 squeezing of the vibron position and momentum quadratures has been theoretically predicted. Even the simple AH model for a single electronic level coupled to an undamped vibrational mode exhibits a rich physics, part of which is still unexplored. Roughly speak- ing, two regimes have been considered so far, according to the ratio γ = Γ0 ω0 (1) between the vibron frequency, ω0, and the average dot- leads electron tunneling rate, Γ0. For fast vibrations, γ (cid:28) 1, every electron tunneling event occurs over many oscillator periods. Then the electrons are not sensitive to the position of the oscillator, but only to its energy.41,55 Consequently, the oscillator den- sity matrix becomes close to diagonal in the basis of the energy eigenstates.55 Many groups focused their atten- tion on this regime, employing rate equations to study transport phenomena as the Franck-Condon blockade, super-Poissonian shot noise 9,10,12,24 and even peculiar effects such as sub-Poissonian out of equilibrium vibron distributions.28,29,33,56,57 In the opposite regime of slow vibrations γ (cid:29) 1, electrons are extremely sensitive to the position of the oscillator, which can be treated in a semi-classical approximation.26,30,31,58 Indeed, Mozyrsky et al. have shown25 that it is the onset of a semi-classical Langevin dynamics. In this regime the electronic properties of the system have been especially investigated, in particular the current and shot noise25 in both limits of weak59 and strong electron-vibron coupling.41 In the latter case, bi-stability and switching have been addressed.46,47,60 Recently, the classical phase space of the vibron has also been studied.27 2 Less attention has been devoted so far to the coherent regime, where the off-diagonal elements of the system density matrix in the energy representation play a rel- In this regime, which starts around γ (cid:38) 1 evant role. (for a more precise discussion, see Sec. III A), the com- petition between the vibron and the electron time scales gives rise to a tough theoretical problem. Most of the results obtained so far, concerning bi-stability and phase space analysis, have been obtained stretching somehow the validity range of semi-classical approaches,25,27,59 in the limit of very low temperatures, τ (cid:28) 1, where τ = kBT ω0 , (2) here kB is the Boltzmann constant and T the environ- ment temperature. The interplay of electron and vibron time scales is expected to strongly influence the dynam- ics of the vibron. For instance, for a NEMS based on a metallic dot in the weak coupling regime, the damping effect of tunneling electrons on the vibron dynamics is maximal when ω0 ≈ Γ0.39 Similar mechanisms could play a role also in the simpler model of a single level quantum dot. Motivated by these considerations, in this article we investigate the vibronic properties in the coherent regime γ (cid:38) 1. Here, because of the off-diagonal structure of the system density matrix, a simple rate equation is no longer justified.61 We derive a generalized master equa- tion62 -- 66 in the sequential tunneling regime, in which all off-diagonal elements of the reduced density matrix in the energy eigenbasis are retained. In the limit of high temperatures considered here, τ > 1, a fairly large number of basis states have to be included. This fact leads to a serious numerical challenge. Our calculation extends up to τ ≤ τmax, where τmax ≈ 10. We remark that, as a difference with previous studies,59 our ap- proach is not restricted to small electron-vibron coupling. Here is a summary of our findings. With the ex- ception of a region γ → τ , in the stationary regime the vibron state can be approximately described in terms of an effective thermal distribution. In the coherent regime, the effective temperature is lower than the environmen- tal temperature. Here, the role of coherences is crucial despite subtle. Non-vanishing off-diagonal elements of the vibron density matrix in the eigenbasis, despite being very small compared with diagonal elements, originate the peculiar effective thermal re-distributions of the vibron occupation probabilities. When γ (cid:39) τ , the system exhibits deviations from the above effective thermal state. The off-diagonal elements are larger, leading to a marked suppression of the vibron fluctuations even below the Poissonian value. The main results of our paper concern the stationary vibron properties in the coherent regime and can be summarized as follows: (i) a cooling of the vibrational mode with respect to the temperature of the electronic environment; (ii) a strong suppression of the vibron Fano factor eventually reaching sub-Poissonian values; (iii) a reduction of the variances of the vibron position and momentum quadratures. We remark that the reported cooling phenomenon is a direct consequence of the NEMS entering the coherent It is not "induced" by any external drive regime. or dynamics, like connecting the system to several reservoirs at different temperatures.40 All the above effects are more pronounced when the electron-vibron coupling strength is increased and none of them comes out treating the system with a simple rate equation involving the diagonal matrix elements only. The paper is structured as follows. In Sec. II we describe the Anderson-Holstein model and the derivation of the generalized master equation in the stationary regime. In Sec. III we illustrate the coherence effects on the vibron behavior and on the electronic degree of freedom. The large discrepancy with respect to the results obtained by means of a simple rate equation is highlighted. Conclu- sions are drawn in Sec IV. II. MODEL AND METHODS A. Anderson-Holstein Model In the AH model,10,11,24 the hamiltonian H = Hdot + Hosc + Hint , (3) describes a ultra-small quantum dot (Hdot) coupled to an harmonic oscillator (Hosc) via the coupling term Hint. The quantum dot is modeled67 as a spin degenerate single level with the average level spacing of the order of the charging energy EC (from now on, = 1) Hdot = n + EC n(n − 1). (4) Here, n = (cid:80) σ=±1 nσ is the occupation number of the level, with nσ = d† σdσ the occupation of spin σ/2 with σ = ±1 and dσ, d† σ are the fermionic dot operators. We assume that EC is the largest energy scale of the problem and we consider only single excess occupancy on the dot, n = 0, 1. The energy = ξ + 2EC(1/2 − ng) includes the energy of the lowest unoccupied single-particle level ξ and a term connected to ng = CgVg/e, the charge induced by the gate voltage Vg with gate capacitance Cg (−e is the electron charge).29 The vibron is described as an harmonic oscillator with mass m and frequency ω0. In terms of the boson opera- tors b, b† it is modeled as (cid:0)b†b + 1/2(cid:1) . Hosc = ω0 (5) The dot and the oscillator are coupled via a term bi-linear in the oscillator position x and in the effective charge number on the dot, n − ng,12,13 x (cid:96)0 (n − ng) , Hint = 2λω0 √ (6) where λ is the adimensional coupling parameter and (cid:96)0 = 1√ mω0 (7) (cid:88) (cid:88) k,σ=±1 is the characteristic length of the harmonic oscillator. The dot is coupled also to the external left (L) and right (R) leads of non interacting electrons Hleads = † εk c α,k,σcα,k,σ , (8) α=L,R † where cα,k,σ and c α,k,σ are the fermionic operators. The leads are assumed in equilibrium with respect to their electrochemical potential µL,R = µ0 ± eV /2, where V is a symmetrically applied bias voltage, and µ0 is the reference chemical potential. The bias V forces electrons to flow from the left to the right lead through the dot via the tunneling hamiltonian Ht = t0 † α,k,σdσ + h.c. , c (9) (cid:88) (cid:88) α=L,R k,σ=±1 where t0 is the tunneling amplitude through both the left and right barriers. Equation (3) can be diagonalized by the Lang -- Firsov po- laron transformation,13 with generator U = exp [η (b† − b)] and η = λ (n − ng). (10) This procedure imposes no restriction on the possible val- ues of λ. The transformed operators in the polaron frame ¯O = UOU† are ¯b = b − η and ¯dσ = dσ exp(cid:2)λ(b − b†)(cid:3), while n is invariant. The diagonal Hamiltonian, ex- pressed in terms of the original operators reads ¯H = ¯ n + ω0(b†b + 1/2) , (11) with renormalized level position ¯ = ξ + (EC − λ2ω0)(1− 2ng). In the following we choose µ0 = ξ setting the resonance between the n = 0, 1 states at ng = 1/2. The eigenstates of Eq. (11) will be denoted as n, l(cid:105), where n is the dot occupation number and l represents the vibron number. The transformed tunneling Hamiltonian is ¯Ht = t0 † eλ(b−b†)c α,k,σdσ + h.c. . (12) (cid:88) (cid:88) α=L,R k,σ=±1 B. Master Equation The dynamics of the dot and the oscillator is described by the reduced density matrix ¯ρ(t), defined as the trace 3 over the leads of the total density matrix ¯ρtot(t), in the polaron frame ¯ρ(t) = Trleads{¯ρtot(t)} . (13) We consider the sequential, weak tunneling regime, treat- ing ¯Ht in Eq. (12) to the lowest order. This approxi- mation is valid for not too low temperatures T , Γ0 = 2πt02ν < kBT , where ν is the leads density of states. We further perform the Born approximation,68 assum- ing that the system and the leads are independent be- fore Ht is switched on. This amounts to take the fac- torized form at t = 0, ¯ρtot(0) = ¯ρ(0) ⊗ ρl(0), where ρl(0) = ρL(0) ⊗ ρR(0) and ρL/R(0) are the initial equi- librium density matrices of the left and the right lead respectively. In the interaction picture with respect to ¯Ht, any op- erator, A, is transformed as AI = ei( ¯H+Hleads)tAe−i( ¯H+Hleads)t. The reduced density matrix evolves according to ¯ρI(t)=−(cid:88) (cid:90) t dt(cid:48)(cid:110) † I,σ(t(cid:48))¯ρI(t(cid:48))]K +(t − t(cid:48)) [QI,σ(t), Q 0 σ=±1 † I,σ(t(cid:48))]K−(t(cid:48) − t) − [QI,σ(t), ¯ρI(t(cid:48))Q (cid:111) † I,σ(t), QI,σ(t(cid:48))¯ρI(t(cid:48))]K−(t − t(cid:48)) + [Q † I,σ(t), ¯ρI(t(cid:48))QI,σ(t(cid:48))]K +(t(cid:48) − t) − [Q (14) . Here, QI,σ(t) = eλ(bI(t)−b † I (t))dI,σ(t) , and K±(t) = K± functions α (t) = t02(cid:88) α (t) = t02(cid:88) L (t) + K± (cid:110) (cid:110) Trleads Trleads K− K + k R (t) are the leads correlation † α,k,σ(t)cα,k,σ(0)ρl c cα,k,σ(t)c † α,k,σ(0)ρl , . (15) (cid:111) (cid:111) k In Eq. (14) we performed the large reservoirs approxima- tion68 ¯ρI,tot(t(cid:48)) = ¯ρI(t(cid:48)) · ρl (16) assuming that tunneling events have a negligible effect on the leads, which remain in the thermal equilibrium state, denoted as ρl. In the weak tunneling regime (Γ0 < kBT ) one can replace ¯ρI(t(cid:48)) ≈ ¯ρI(t) using the standard Markov approximation, and extend the integration limit to ∞. Eq. (14) can be projected on the eingenstates of the hamiltonian (11), obtaining an infinite set of coupled equations for the density matrix elements (cid:104)n, l¯ρ(t)n(cid:48), l(cid:48)(cid:105) (where n, n(cid:48) ∈ {0, 1} and l, l(cid:48) ≥ 0). It can be easily shown that diagonal and off-diagonal elements in the electron number decouple and in the stationary regime the latter is a Fermi function, f− α (E) and (cid:18) 2π (cid:19) α (E) = 1 − f + (cid:20) 1 + Re ψ βωc iβ 2π + 2 (¯ + ωqq(cid:48) − µα) (cid:21) ∆α(ωqq(cid:48)) = log the depolarization shift. The stationary system proper- ties are obtained from the solution of the steady-state qq(cid:48)pp(cid:48) + iω0 (p(cid:48) − p) δpqδp(cid:48)q(cid:48) ¯ρn(cid:48) pp(cid:48) = 0 , (24) (cid:105) GME(cid:88) (cid:104)Rnn(cid:48) pp(cid:48)n(cid:48) (18) (19) written here in the Schrodinger representation, for the stationary values of the reduced density matrix elements ¯ρn pp(cid:48) = lim t→∞ ¯ρn pp(cid:48)(t) . (25) Note that the GME (24) takes into account both all off- diagonal (coherences) and diagonal terms in the vibron number. The steady state current calculated e.g. on the right tunnel barrier, reads (cid:88) (cid:8)−2(cid:2)C + +(cid:2)C− pqq(cid:48) R (ωqp) + C− R (ωpq(cid:48)) + C + R (ωpq)∗(cid:3) XqpXq(cid:48)p ¯ρ0 qq(cid:48)(cid:9) . (26) R (ωq(cid:48)p)∗(cid:3) XpqXpq(cid:48) ¯ρ1 qq(cid:48) I = eΓ0 (λ2) [sgn(q(cid:48) − q)λ] q(cid:48)−q (20) , Note that in the steady-state the current is independent of the barrier index. 4 tend to zero. In fact, the coupling of the leads and the dot charge leads to a rapid decay of superpositions of dot states with different charges.68,69 Since we are interested in the stationary properties, we focus on the density ma- trix elements diagonal in the electron level occupation qq(cid:48)(t) = (cid:104)n, q¯ρ(t)n, q(cid:48)(cid:105). They obey the follow- number, ¯ρn (cid:88) ing generalized master equation (GME) Rnn(cid:48) qq(cid:48)pp(cid:48)eiω0(q−q(cid:48)−p+p(cid:48))t ¯ρn(cid:48) ¯ρn I,qq(cid:48)(t) = I,pp(cid:48)(t) , (17) pp(cid:48)n(cid:48) l R01 R11 where Rnn(cid:48) qq(cid:48)pp(cid:48) are the Redfield tensor elements61 qq(cid:48)pp(cid:48) =XqpXq(cid:48)p(cid:48)(cid:2)C−(ωpq)∗ + C−(ωp(cid:48)q(cid:48))(cid:3) , qq(cid:48)pp(cid:48) = −(cid:88) qq(cid:48)pp(cid:48) =2XpqXp(cid:48)q(cid:48)(cid:2)C +(ωqp) + C +(ωq(cid:48)p(cid:48))∗(cid:3) , (cid:2)Xlq(cid:48)Xlp(cid:48)C−(ωp(cid:48)l)δpq +XlqXlpC−(ωpl)∗δp(cid:48)q(cid:48)(cid:3) , (cid:88) (cid:2)Xq(cid:48)lXp(cid:48)lC +(ωlp(cid:48))∗δpq +XqlXplC +(ωlp)δp(cid:48)q(cid:48)(cid:3) . (cid:115) R10 R00 qq(cid:48)pp(cid:48) = − 2 Here ωqq(cid:48) ≡ ω0(q − q(cid:48)) and Xqq(cid:48) = (cid:104)qeλ(b−b†)q(cid:48)(cid:105) l = e− λ2 2 Lq(cid:48)−q q< q<! q>! are the generalized Franck-Condon factors9,29, with q< = min{q, q(cid:48)}, q> = max{q, q(cid:48)} and Ln q (x) the generalized Laguerre polynomials.70 The factors 2 in Eqs. (19) are due to the spin degeneracy. In Eqs. (18),(19) the gener- alized tunneling rates C±(ωqq(cid:48)) = C± R (ωqq(cid:48)) have also been introduced L (ωqq(cid:48)) + C± C± α (ωqq(cid:48)) = dθK± α (±θ)e−i(¯+ωqq(cid:48) )θ . (21) (cid:90) ∞ 0 Exploiting the identity (cid:90) ∞ dθ exp (iΩθ) = πδ(Ω) + iP.V.(1/Ω) , 0 and the explicit form of the leads correlation function71 α (t) = − K± (cid:88) Γ0 2 α iΓ0e±iµαt β (t − i ωc 2β sinh [ π , )] α (ωqq(cid:48)) ± i∆α(ωqq(cid:48))(cid:3) . (cid:2)f± C±(ωqq(cid:48)) = with ωc the cut-off energy and β = 1/(kBT ), one has Here f + α (E) = 1 1 + eβ(¯+E−µα) C. Rotating Wave Approximation In the regime of fast vibrational motion, γ (cid:28) 1, the contribution of fast oscillatory terms to the solution of Eq. (17) is negligible. It is then possible to perform the rotating wave approximation (RWA), neglecting the os- cillating contributions and including only the dominant secular terms.61 They are those which connect density matrix elements with p−p(cid:48) = q−q(cid:48). This implies that the diagonal elements ¯ρn qq are decoupled from the off-diagonal pp(cid:48) with p(cid:48) (cid:54)= p. In addition, the non-diagonal ele- ones ¯ρn ments vanish in the stationary regime.10 Hence station- ary properties are fully described by the diagonal occu- pation probabilities ¯Pnq = ¯ρn qq and Eq. (24) reduces to a standard rate equation (cid:88) ∞(cid:88) n(cid:48)(cid:54)=n p=0 qp − (cid:88) Γnn(cid:48) ∞(cid:88) n(cid:48)(cid:54)=n p=0 zn ¯Pnq zn(cid:48) ¯Pn(cid:48)pΓn(cid:48)n pq = 0 . (27) (22) (23) The coefficients zn stem from the spin degeneracy: z0 = 2, z1 = 1. The tunneling rates Γnn(cid:48) R,pq for the transition n, p(cid:105) → n(cid:48), q(cid:105), are L,pq + Γnn(cid:48) pq = Γnn(cid:48) pqRe(cid:8)C + pqRe(cid:8)C− α (ωpq)(cid:9) = Γ0X 2 α (ωpq)(cid:9) = Γ0X 2 Γnn+1 α,qp = 2X 2 Γn+1n α,pq = 2X 2 pqf + pqf− α (ωpq); α (ωpq). (28) 5 FIG. 2: (Color online) (a) Stability diagram in the (V, ng) plane: gray shaded regions denotes Coulomb Blockade regime. Blue (solid) lines mark the onset of transitions between ground states, red (dashed) lines signal transitions involving excited states of the vibron. (b) Different regimes as a func- tion of γ = Γ0/ω0 and τ = kBT /ω0. The blue shaded region denotes the "diabatic" regime of fast vibrations, the red one the semi-classical regime with ω0 (cid:28) Γ0. The region above the line γ = τ marks the region accessible by our methods and the yellow shaded area indicates the accessible regime where coherences among vibron states become important. III. RESULTS In the present Section, we investigate the stationary equilibrium properties of the system as a function of the parameters γ, τ defined in Eqs.(1), (2). Here, γ distin- guishes between fast, slow or coherent regimes, and τ is the reduced temperature. We focus mainly on the vi- bronic properties. Relevant quantities of our interest are the vibron Fano factor and the position and momentum quadratures. Their behavior can be understood by first analyzing the structure of the oscillator density matrix in the vibron eigenbasis. In the final part of this Section, the electronic current and the average electronic occupa- tion of the dot will be briefly addressed. A. Parameters regimes Figure 2(a) represents the stability diagram of the system as a function of the bias voltage V and of ng. In the shaded (gray) regions the system is in the Coulomb blockade regime with the dot empty (full) if ng < 1/2 (ng > 1/2). The blue (solid) lines mark transitions between the states 0, q(cid:105) and 1, q(cid:105) (n and q are the occu- pation number of the electronic or vibronic states n, q(cid:105)). Red (dashed) lines indicate the activation threshold for transport channels with transitions n, q(cid:105) → n(cid:48), q(cid:48)(cid:105) where in general q, q(cid:48) (cid:54)= 0. The boundaries of different transport regions are thermally broadened. In the following analysis we will focus at ng = 1/2 where the n = 0 and n = 1 charge states are on resonance (dashed-dotted line in Fig. 2(a)). Qualitatively similar results are observed also off-resonance at ng (cid:54)= 1/2. FIG. 1: (Color online) Vibron occupation probabilities ¯Pq as a function of the vibron number q, calculated numerically solv- ing the RWA rate equation in Eq. (27) (red empty squares) and the GME in Eq. (24). Here γ = 3 (cyan empty tri- angles), γ = 1 (purple empty circles), γ = 10−1 (blue cir- cles), γ = 10−3 (green squares). (a) λ = 0.5 and eV = ω0. Inset: λ = 2 (b) Main panel: λ = 0.5 and eV = 20 ω0. and eV = 20 ω0. In all panels other parameters are τ = 6, ng = 1/2 and ωc = 106 ω0. Note that Eq. (27) does not depend on γ. Within the RWA, the steady-state current Eq. (26) is (cid:88) (cid:104)−2 ¯P0qΓ0,1 IRWA = eΓ0 (cid:105) R,qp + ¯P1qΓ1,0 R,qp . (29) p,q n ¯ρn of the reduced density matrix, ¯Pq =(cid:80) Results presented in the following Section are obtained by numerical solution of the GME Eq. (24) truncating the size of the vibron Hilbert space until convergence is reached (see next Section for details). In the limit γ (cid:28) 1, this approach accurately reproduces the well- known solution of the RWA rate equation.9,24,28,39,40 This is illustrated in Fig. 1, where the diagonal elements qq, obtained by solving the GME for different values of γ are shown. The solutions of the GME converge to those in the RWA for γ → 0 (red squares), both for small and large voltages (panels (a) - (b)) and even in the strong electron-vibron coupling regime (panel c). This convergence has been systematically observed in the whole range of parameters and constitutes a validation of our numerical procedure. On the other hand, with increasing γ, deviations from the RWA are obtained. They are originated from the coherent off-diagonal elements of ¯ρqq(cid:48). These deviations represent the central part of our work and relevant conse- quences will be discussed in details in the following Sec- tion. 6 In Fig. 2(b) we report a sketch of the system behavior in the (γ, τ ) plane. We distinguish a "diabatic" regime of fast vibrations for γ (cid:28) 1 (blue shaded region), where the RWA is valid,9,24,28,39,40 and an "adiabatic" regime of slow vibrations γ (cid:29) 1 where semiclassical methods have been applied25 -- 27,41,47 with approaches confined to low temperatures τ (cid:28) 1. The regime in between these two regions is, on the other hand, unexplored. Our GME method fills in precisely this gap. Indeed, we will show that, increasing γ from the diabatic regime, marked deviations from the solution of the RWA rate equations appear. Here, coherences represented by the off-diagonal elements of the reduced density matrix are relevant. The constraints on our GME method are: (i) the sequential tunneling approximation which implies 0 < γ < τ (the area above the red line in Fig. 2); (ii) a restriction on the temperature τ < τmax ≈ 10, due to the rapid increase with τ of the number of vibron states needed for the convergence of our numerical calculations. Because of these constraints, the transition towards the semiclassical limit cannot be explored. We solve numerically the GME in the steady state, in- creasing the size of the vibron Hilbert space including up to 150 oscillator states per charge state in the ba- sis, until convergence is reached. The limitation on the number of oscillator states employed in the calculation is essentially due to computer memory requirements and by the decreasing rate of convergence. At high temperature τ ≈ 10, already more than 100 oscillator basis states are required for convergence. B. Overview of the results Before entering our analysis, we summarise in Fig. 3 the stationary vibronic characteristics we obtain in the various parameter regimes. We address the temperature regime 1 (cid:46) τ (cid:46) τmax and 0 < γ (cid:46) τ . We will show that in a large parameters range (shaded area) the vibron density matrix (cid:88) ¯ρqq(cid:48) = ¯ρn qq(cid:48) (30) FIG. 3: (Color online) Schematic overview of the results in the (γ, τ ) plane, see text for details on the notation. τeff → τ , as reported in Ref. 28. With increasing γ the off-diagonal elements of ¯ρqq(cid:48) in- crease. The onset of this coherent regime is marked by the condition γ (cid:38) ¯γ(λ), with the latter a λ-dependent threshold value. For typical values λ ≈ 1 we find 0.1 ≤ ¯γ(λ) ≤ 1. When ¯γ(λ) (cid:46) γ < τ the coherences are much smaller than the diagonal elements of the re- duced density matrix. In this weakly coherent regime the vibron density matrix can still be fitted with a thermal distribution, but at a lower effective temperature, even- tually reaching the cooling regime where τeff < τ . The cooling is always accompanied by a reduction both of the fluctuations of the vibronic population and of the vari- ance of position and momentum quadratures. A completely different system behavior takes place when γ → τ , (yellow-green area delimited by the dashed and continuous lines in Fig. 3). In this strongly coher- ent regime off-diagonal terms of the density matrix are paramount and the diagonal part of the density matrix deviates from the simple thermal distribution. Here, de- spite of the high temperature (τ > 1), a non-classical system behavior comes out. Suppression of the vibron populations fluctuations and of the variances of the po- sition and momentum quadratures persists also in this regime. n=0,1 C. Density matrix and cooling (cid:16) (cid:17) is well approximated by an effective thermal distribution with temperature τeff 1 − e−1/τeff e−q/τeff δq,q(cid:48) . ¯ρ(th) qq(cid:48) (τeff ) = (31) For γ (cid:28) 1 the off-diagonal elements of the reduced den- sity matrix are negligible and a fit of ¯ρqq(cid:48) to Eq. (31) leads to τeff ≥ τ . This "heating" phenomenon is due to the finite voltage, and for V → 0 the system attains a thermal equilibrium distribution in the polaron frame, We start our analysis investigating the structure of the oscillator density matrix in the vibron eigenbasis in the regimes indicated in Fig. 3. We first concentrate on the density matrix ¯ρqq(cid:48) when it is well described by the effective thermal distribution, Eq. (31). In Figs. 4(a) - (b) we report the density plots of ¯ρqq(cid:48) in the diabatic and in the coherent regime re- spectively. For γ (cid:28) 1 - panel (a) - the density matrix is strongly peaked around the diagonal, q ≈ q(cid:48). The cor- 7 FIG. 4: (Color online) Density plot of ¯ρqq(cid:48) as a function of q, q(cid:48). Panel (a): γ = 0.01 and τ = 9; Panel (b): γ = 2 and τ = 9. Panel (c): Occupation probabilities ¯Pq = ¯ρqq for τ = 9 and γ = 0.01 (squares) or γ = 2 (dots). Lines are fit to a thermal distribution with effective temperature τeff . In panel (d) γ = 1.2 and τ = 3 (main), and τ = 9 (inset). In all panels, ng = 1/2, eV = ω0, λ = 2 and ωc = 106 ω0. responding occupation probability distribution ¯Pq = ¯ρqq extends over more than fifteen states Fig. 4(c) (squares). We then perform a numerical fit of ¯ρqq(cid:48) on a thermal distribution in Eq. (31). The fit leads to an effective temperature τef f with an error ∆τ , signaling the depar- ture from the approximatively diagonal thermal density matrix. In considered diabatic regime, γ (cid:28) 1, τeff ≈ τ with a very small relative error δτ = ∆τ /τeff < 10−4. Increasing γ, the coherent regime is entered, with a vibron occupation probability considerably altered, see Fig. 4(b). We observe two main modifications. First, off- diagonal elements are larger than in the diabatic regime, even though they remain rather small in comparison with the diagonal ones. Second, the probability distribution along the diagonal gets narrower. Remarkably, the occu- pation probabilities ¯Pq are still approximated by a quasi- thermal state but with an effective temperature τeff lower than the environmental one, see Fig. 4(c) circles. The rel- ative error is still rather small, δτ ≈ 6· 10−3. We remark that this result is a consequence of the non-vanishing vi- bronic coherences, in fact in the RWA the density matrix would be strictly diagonal, as a difference with Fig. 4(b). The crossover from heating to cooling with increasing γ and the accuracy of the thermal approximation mea- FIG. 5: (Color online) Solid lines: effective temperature τeff as a function of γ, extracted from a numerical fit of ¯ρqq(cid:48) on Eq. (31), (see text). Panel (a): τ = 9 and eV = 20 ω0. In the inset τeff as a function of λ for γ = 6. Panel (b): τ = 3 and eV = 6 ω0. In all panels, dashed lines around the solid line delimit the absolute error on the effective temperature, ∆τ . Red (blue) gradient areas signal heating (cooling). Other parameters are ng = 1/2 and ωc = 106 ω0. sured by the error of the fitting procedure are analyzed in Fig. 5(a) - (b). The absolute error is represented by the shaded area around the continuous line limited by dashed lines. Two different regimes are clearly identified. In the diabatic regime, the effective temperature is larger than that of the environment (red shaded region) and the system exhibits heating. The values of τeff ≥ τ are due to the considered high voltage bias and indeed for V → 0 one obtains τeff → τ . As γ is increased, cooling occurs (blue shaded region) and τeff drops markedly below the electronic tempera- ture τ . Even though τeff increases for increasing V (not shown) this cooling effect survives up to eV ≈ 20 ω0. It can be seen that the error grows with increasing γ, signaling the rise of the off-diagonal terms of the density matrix. We can then safely identify the cooling phenomenon provided γ does not approach τ . In the inset of Fig 5(a) the typical behavior of τeff as a function of λ is shown, with an error on τeff which increases increasing λ. We remark that the cooling phenomenon is entirely due to the coherent dynamics of the vibron-electronic system and is not induced by any ad-hoc mechanism acting on the system. The description in terms of an effective thermal distri- bution ceases to be valid when γ → τ , as shown by the increasing errors in Fig. 5. To investigate this regime, we consider the case τ = 3, γ = 1.2 in Fig. 4(d). Here, tion22 W (x, p) = (cid:90) ∞ −∞ 1 π dy e2ipy(cid:104)x − yU† ¯ρUx + y(cid:105) , (32) the quantum analogue of the Liouville density in classi- cal phase space with U defined in Eq. (10). The Wigner function allows the detection of non-classical features sig- naled by regions in the (x, p) plane where W (x, p) < 0.22 8 although an effective temperature τeff ≈ 0.8 < τ can be formally extracted with the fitting procedure, the rela- tive error becomes large δτ ≈ 0.11. The main source of error are the rather large off-diagonal matrix elements. This is a general trend which we always observed when γ approaches τ . As an illustration, we report in the inset of Fig. 4(c) the density matrix for the same parameters of the main panel but at higher temperature (τ (cid:29) γ). In this case off-diagonal elements are strongly suppressed and an effective thermal description is then appropriate. We conclude this paragraph commenting on the depen- dence of the above results on the electron-vibron coupling strength, λ, which is not limited in our approach. In Fig- ure 6 (a) - (d) we report the density plot of ¯ρqq(cid:48) at fixed τ and for increasing values of λ. Increasing the dot-vibron coupling induces an hybridization of the electronic and mechanical degrees of freedom which manifests itself also in the off-diagonal elements of the vibron density matrix between Fock states. Indeed, with increasing λ a "de- localization" phenomenon is induced, the density matrix spreading away from the diagonal. FIG. 7: (Color online) Panels (a) - (c) density plot of the Wigner function W (x, p) as a function of x, p. In panels (a) and (b) τ = 9, and in (a) γ = 0.01, in (b) γ = 2. Note the different color scale in the two panels. (c) Density plot of W (x, p) for τ = 3 and γ = 1.15. (d) Probability density of the oscillator position P (x). In all panels x and p are expressed in unity of (cid:96)0 and (cid:96) respectively - see Eq. (7). In addition ng = 1/2, eV = ω0, λ = 2 and ωc = 106 ω0. −1 0 Figures 7(a) - (b) show W (x, p) for the parameters of Figs. 4(a) - (b), in the effective thermal regime. The Wigner functions are positive and can be regarded as the probability density of the oscillator states in the (x, p) phase space. Consistently, their shape closely resembles that of a thermal state,22 but with a decreased width τeff . The shrinkage of W (x, p) observed when entering the coherent regime can be traced back to the role of co- herences which mediate the redistribution of the ¯Pq and the ensuing reduction of τeff . Figure 7(c) shows W (x, p) for γ = 1.2 and τ = 3 in the strongly coherent regime, when the system can no longer be described by a thermal distribution (as in Fig. 4(d)). In this case, the width of W (x, p) is still reduced. As FIG. 6: (Color online) Density plot of ¯ρqq(cid:48) as a function of q, q(cid:48) for fixed γ, τ and different values of the electron-vibron coupling strength. Panel (a): λ = 1; Panel (b): λ = 1.5; Panel (c): λ = 2; Panel (d): λ = 2.5. In all panels, ng = 1/2, eV = ω0, γ = 3, τ = 6 and ωc = 106 ω0. D. Wigner function and quadratures Further insight on the system behavior is obtained from the Wigner quasi-probability distribution func- a difference with previous case, here we observe that W (x, p) becomes negative in an approximately circular area around the main peak (white dashed circle). This fact signals a non-classical behavior of the system entirely due to the non-diagonal structure of the reduced density matrix. Negative values of W (x, p) are a trademark of the regime γ/τ ≈ 1. This quantum behavior, naively un- expected at the considered high temperatures τ (cid:38) 1, is a relevant result whose implications on the vibronic fluc- tuations will be discussed in the next subsection. Also noteworthy is the elongated shape of W (x, p) along the x direction, Fig. 7(c). This is reflected in the position probability distribution P (x) = dp W (x, p) (33) (cid:90) ∞ −∞ shown in Figure 7(d). While in the effective-thermal regime P (x) exhibits a single peak, in the regime of strong coherences it shows a peculiar shoulder structure. In this latter regime, for relatively slow tunneling events γ ∼ 1, the oscillator adjusts itself to the two equilibrium positions corresponding to zero or one extra electron on the dot, x0 = λng and x1 = λ(ng − 1) (in units of (cid:96)0, see Eq. (7)). When the width of the two probability distri- butions for the uncharged and charged states (∝ τeff ) is of the order of the separation between the rest positions ∝ λ, the shoulder is visible. The effect is therefore more pronounced the larger is the electron-vibron coupling λ and eventually develops into a double-peaked shape. The width of W (x, p) in the x and the p directions is strictly related to the variance of the vibron position and momentum (cid:90) ∞ (cid:90) ∞ (cid:104)xµ(cid:105) = dx −∞ −∞ dp xµW (x, p) , (34) 9 FIG. 8: (Color online) Variance of the position X (a) and of the momentum P (b) as a function of γ. In all panels τ = 9, eV = 20 ω0, ng = 1/2 and ωc = 106 ω0. Tr{UOU† ¯ρ}. This quantity (not be confused with the electronic Fano factor) brings information about the statistics of the vibronic mode and also on electronic properties, like charge fluctuations. This is due to the involved polaron transformation and it can be explicitly seen introducing the hybrid average (cid:104)O(cid:105) ¯ρ = Tr{O ¯ρ}. In terms of (cid:104)·(cid:105) ¯ρ one has √ (cid:104)N(cid:105) =(cid:104)N(cid:105) ¯ρ − 2(cid:104)ηX(cid:105) + (cid:104)η2(cid:105) Var(N ) =V(N ) + V(η2) + 2V(ηX ) + 2C(η2, N ) 2(cid:2)C(ηX , N ) + C(N, ηX ) + 2C(η2, ηX )(cid:3) , √ − (36) and corresponding expressions for p. In Fig 8 (a) and (b) are reported the variance of the position, X = x/(cid:96)0 = √ √ 2, and momentum P = p · (cid:96)0 = i(b† − b)/ (b† + b)/ 2 satisfying always Var(X )Var(P) ≥ 1/2 - with (cid:96)0 defined in Eq. (7). As expected, in the coherent regime both vari- ances are strongly suppressed with respect to the large values attained in the diabatic regime, both for high and low (not shown) voltages. This fact denotes a tendency towards squeezing.22 E. Vibronic Fluctuations The out-of-equilibrium vibron behavior is conveniently discussed introducing the vibron Fano factor28,63 Fv = Var(N ) (cid:104)N(cid:105) , (35) where N ≡ b†b. Here, Fv is the ratio between the variance of the vibron occupation number Var(N ) = (cid:104)N 2(cid:105)−(cid:104)N(cid:105)2 and its average (cid:104)N(cid:105), where (cid:104)O(cid:105) = Tr{Oρ} = ¯ρ and C(A,B) ≡ where V(O) ≡ (cid:104)O2(cid:105) ¯ρ − (cid:104)O(cid:105)2 (cid:104)AB(cid:105) ¯ρ − (cid:104)A(cid:105) ¯ρ(cid:104)B(cid:105) ¯ρ, with adimensional oscillator po- sition X . The quantities (cid:104)N(cid:105) and Var(N ) depend explicitly both on the charge (being η = λ(n − ng)) In Eq. (36) all terms of the and on its fluctuations. form (cid:104)·X(cid:105) and C(·, ·X ) depend on the coherences of the density matrix. In the RWA these terms simplify but keep the η dependence. We remark that, beside the explicit dependence of Var(N ) and (cid:104)N(cid:105) on (cid:104)n(cid:105), fluctuations of the electronic charge intrinsically affect Fv, since the contribution of electronic and vibronic degrees of freedom cannot be factorized in ¯ρn Fv belongs to the class of "bosonic" Fano factors, originally introduced to characterize boson distributions and often used to study photon populations in the context of quantum optics.22 Typically, for bosons the Fano factor is super-Poissonian (Fv > 1). This is, for example, the situation for light from a classical radiation field.22 Deviations from the super-Poissonian regime signal non-classical correlations, which have been observed for instance in experiments with micromaser.72 qq(cid:48). 10 From an experimental point of view, the situation for vibron is more complicated than in the case of photons, although many proposals have been made to observe quantities like the average vibron number.73 -- 75 Recent experiments aiming at characterizing the vibron distributions have appeared.76 A few recent theoretical works investigated the vibron distributions predicting sub-Poissonian vibron Fano factor in the diabatic regime γ (cid:28) 1, for low temperatures τ (cid:28) 1 and in selected parameter ranges.28,29 The behavior of the vibron Fano factor as a function quantum behaviors show up in the steady state regime. The vibron Fano factor also depends on the voltage and FIG. 10: (Color online) Main: plot of Fv as a function of V (units ω0/e) for different values of λ. Inset: dependence of Fv on the electron-vibron strength λ, for eV = 15 ω0. Other parameters are γ = 6, ng = 1/2, τ = 9 and ωc = 106 ω0. on electron-vibron coupling. Fv increases with increasing bias voltage, assuming super-Poissonian values even in the coherent regime, Fig. 10 (main panel). This behavior can be attributed to the increase of scattering between oscillator states induced by the current flow across the system. In the inset, the Fano factor is plotted as a function of λ: Fv decreases with increasing coupling strengths. This trend has been found both in the low (eV < kBT ) and in the high (eV > kBT ) voltage regimes - see also Figs. 9(a) - (b). F. Charge degree of freedom We conclude the survey of our results briefly com- menting on the electronic properties. Figure 11(a) shows the I(V ) characteristics for different values of γ. In the regime τ > 1, the vibron sideband can not be resolved and the current increases monotonically. Figure 11(b) shows the ratio between the current I obtained from the GME and the current in the RWA, IRWA, as a function of γ and fixed voltage V . Increasing γ, the current increases with respect to the RWA limit. This qualitative trend oc- curs in all the parameter regimes explored. The increase of the current is more prominent for larger values of λ. This is a consequence of the delocalization of the vibron density matrix occurring with increasing electron-vibron coupling. The average occupation of the electronic dot level (cid:104)n(cid:105) behaves similarly to the current (Fig. 11 inset). For the case of only n = 0 and n = 1 electron states the variance of the electron occupation number reads Var(n) = (cid:104)n2(cid:105) − (cid:104)n(cid:105)2 = (cid:104)n(cid:105) (1 − (cid:104)n(cid:105)) . (37) The increase of (cid:104)n(cid:105) > 1/2 signals a suppression of the fluctuations of the electronic level population analo- FIG. 9: (Color online) Vibron Fano factor Fv as a function of γ for different values of the electron-vibron coupling strength λ and temperature τ . In panel (a) τ = 9 and eV = ω0, in the inset eV = 20 ω0. In panel (b) τ = 3 and eV = ω0. Solid lines are the results obtained with the GME, dashed lines are obtained within the RWA. In all panels, ng = 1/2 and ωc = 106 ω0. of γ for weak and strong λ is reported in Fig. 9. In the regimes where the vibron can be approximated by an effective thermal distribution, Fv displays a super-Poissonian behavior. This occurs in the diabatic regime, both at low and high voltages (eV smaller or larger than kBT ), and in the weakly coherent regime. In this case however the vibron Fano factor is considerably reduced. These behaviors qualitatively do not depend on the electron-vibron coupling λ. In the diabatic regime the RWA applies with no dependence on γ. On the other hand, entering the strongly coherent regime (γ ≈ 1 and τ = 3 in Fig. 9(b)), non-classical sub-Poissonian vibron Fano factors are obtained. This is in correspondence with the negative values of the Wigner function observed in this parameter regime. Both fea- tures result from the large vibronic coherences. Similar conclusions were drawn from the frequency-dependent shot noise of a NEMS.62 Here, however, these peculiar 11 range. extends from the very fast vibrations ω0 (cid:29) Γ0 to the slow, coherent regime where the off-diagonal elements of the reduced density matrix between energy eigenstates are paramount. In the coherent regime, two peculiar behaviors have been found. For intermediate frequencies, ω0 ≈ Γ0, the system can be described in terms of an effective thermal distribution with a temperature lower than that of the environment. The cooling phenomenon is accompanied by a decrease of position and momentum quadratures. For still slower oscillations, a strongly coherent regime is entered characterized by non classical behavior. A benchmark of this regime is a marked suppression of the vibron Fano factor, which can even attain sub-Poissonian values. This work is one of the first steps towards the understand- ing of dynamical properties of nano-electromechanical systems in the coherent regime and represents a rather tough numerical challenge, due to the slow convergence of the master equation solution and the need of a large number of basis states. Future investigations are certainly in order, to explore the regime of very strong electron-vibron coupling and the crossover towards the semi-classical regime. Further interesting issues to be investigated are higher order electronic properties, like the current fluctuations and their connection with the fluctuations of the mechanical part. Acknowledgments. The authors acknowledge stimulating discussions with A. Nocera. Financial support by CNR- SPIN via both the Seed Project PLASE001 and the "Pro- getto giovani", and by the EU-FP7 via ITN-2008-234970 NANOCTM is also gratefully acknowledged. FIG. 11: (Color online) (a) Current I (units eΓ0) as a function of V (units ω0/e) for different values of γ. (b) Ratio between the current obtained with the GME (I) and the current in the RWA approximation (IRWA) as a function of γ, at eV = 15 ω0. Inset: average occupation (cid:104)n(cid:105) of the electronic level. Other parameters: ng = 1/2, τ = 9, λ = 2 and ωc = 106 ω0. gously to the fluctuations of the vibronic part. IV. CONCLUSIONS In the present article we derived a generalized master equation to explore the steady-state properties of a nano-electromechanical system in a wide parameters 1 M. L. Roukes, Nanoelectromechanical Systems, in Tech- nical Digest of the 2000 Solid State Sensor and Actuator Workshop, Transducers Research Foundation, Cleveland, OH (2000). 2 H. Park, J. Park, A. K. L. Lim, E. H. Anderson, A. P. Allvisatos, and P. McEuen, Nature 407, 57 (2000). 3 R. Leturcq, C. Stampfer, K. Inderbitzin, L. Durrer, C. Hierold, E. Mariani, M. G. Schultz, F. von Oppen, and K. Ensslin, Nature Phys. 5, 327 (2009). 4 B. J. Leroy, S. G. Lemay, J. Kong, and C. Dekker, Nature 432, 371 (2004). 5 S. Sapmaz, P. Jarillo-Herrero, Y. M. Blanter, C. Dekker, and H. S. J. van der Zant, Phys. Rev. Lett. 96, 026801 (2006). 6 B. Lassagne, Y. Tarakanov, J. Kiranet, D. Garcia-Sanchez, and A. Bachtold, Science 325, 1107 (2009). 7 R. G. Knobel and A. N. Cleland, Nature 424, 291 (2003). 8 E. M. Weig, R. H. Blick, T. Brandes, J. Kirschbaum, W. Wegscheider, M. Bichler, and J. P. Kotthaus, Phys. Rev. Lett. 92, 046804 (2004). 9 J. Koch, F. von Oppen, and A. V. Andreev, Phys. Rev. B 74, 205438 (2006). 245302 (2004). 11 S. Braig and K. Flensberg, Phys. Rev. B 68, 205324 (2003). 12 X. Y. Shen, B. Dong, X. L. Lei, and N. J. M. Horing, Phys. Rev. B 76, 115308 (2007). 13 A. Zazunov, D. Feinberg, and T. Martin, Phys. Rev. B 73, 115405 (2006). 14 F. Cavaliere, A. Braggio, J. T. Stockburger, M. Sassetti, and B. Kramer, Phys. Rev. Lett. 93, 036803 (2004). 15 A. Braggio, M. Sassetti, and B. Kramer, Phys. Rev. Lett. 87, 146802 (2001). 16 F. Cavaliere, A. Braggio, M. Sassetti, and B. Kramer, Phys. Rev. B 70, 125323 (2004). 17 F. Haupt, F. Cavaliere, R. Fazio, and M. Sassetti Phys. Rev. B 74, 205328 (2006). 18 J. A. Sidles, J. L. Garbini, K. J. Bruland, D. Rugar, O. Zuger, S. Hoen, and C. S. Yannoni, Rev. Mod. Phys. 67, 249 (1995). 19 B. Ilic, H. G. Craighead, S. Krylov, W. Senaratne, C. Ober, and P. Neuzil, J. Appl. Phys. 95, 3694 (2004). 20 A. K. Naik, M. S. Hanay, W. K. Hiebert, X. L. Feng, and M. L. Roukes, Nature Nanotech. 4, 445 (2009). 21 F. Giazotto, T. Heikkila, A. Luukanen, A. M. Savin, and 10 A. Mitra, I. Aleiner, and A. J. Millis, Phys. Rev B 69, J. P. Pekola, Rev. Mod. Phys. 78, 217 (2006). 12 22 L. Mandel and E. Wolf, Optical Coherence and Quantum (2009). Optics, N. Y. : Cambridge University Press, 1995. 23 J. B. Hertzberg, T. Rocheleau, T. Ndukum, M. Savva, A. A. Clerk, and K. C. Schwab, Nature Phys. 6, 213 (2010). 24 J. Koch and F. von Oppen, Phys. Rev. Lett. 94, 206804 51 P. Rabl, Phys. Rev. B 82, 165320 (2010). 52 S. Zippilli, A. Bachtold, and G. Morigi, Phys. Rev. B 81, 205408 (2010). 53 R. Ruskov, K. Schwab, and A. N. Korotkov, Phys. Rev. B (2005). 71, 235407 (2005). 25 D. Mozyrsky, M. B. Hastings, and I. Martin, Phys. Rev. B 54 A. A. Clerk, F. Marquardt, and K. Jacobs, New J. Phys. 73, 035104 (2006). 10, 095010 (2008). 26 A. Nocera, C. A. Perroni, V. Marigliano Ramaglia, and V. 55 R. El Boubsi, O. Usmani, and Y. M. Blanter, New J. Phys. Cataudella, Phys. Rev. B 83, 115420 (2011). 10, 095011 (2008). 27 R. Hussein, A. Metelmann, P. Zedler, and T. Brandes, 56 R. Sanchez, G. Platero, and T. Brandes, Phys. Rev. Lett Phys. Rev. B 82, 165406 (2010). 98, 146805 (2007). 28 M. Merlo, F. Haupt, F. Cavaliere, and M. Sassetti, New. 57 R. Sanchez, G. Platero, and T. Brandes, Phys. Rev. B 78, J. Phys. 10, 023008 (2008). 125308 (2008). 29 F. Cavaliere, G. Piovano, E. Paladino, and M. Sassetti, 58 O. Usmani, Y. M. Blanter, and Y. V. Nazarov, Phys. Rev. New J. Phys 10, 115004 (2008). B 75, 195312 (2007). 30 A. D. Armour, M. P. Blencowe, and Y. Zhang, Phys. Rev. 59 D. Mozyrsky, I. Martin, and M. B. Hastings, Phys. Rev. B 69, 125313 (2004). 31 A. D. Armour, Phys. Rev. B 70, 165315 (2004). 32 C. B. Doiron, W. Belzig, and C. Bruder, Phys. Rev. B 74, 205336 (2006). Lett. 92, 018303 (2004). 60 M. Galperin, A. Nitzan, and M. A. Ratner, J. Phys.: Con- dens. Matter 20, 374107 (2008). 61 K. Blum, Density Matrix Theory and Application, Plenum 33 T. J. Harvey, D. A. Rodrigues, and A. D. Armour, Phys. Press, New York (1996). Rev. B 81, 104514 (2010). 62 H. Huebener and T. Brandes, Phys. Rev. Lett. 99, 247206 34 T. J. Harvey, D. A. Rodrigues, and A. D. Armour, Phys. (2007). Rev. B 78, 024513 (2008). 63 D.A. Rodrigues, J. Imbers, and A. D. Armour, Phys. Rev. 35 D. A. Rodrigues, J. imbers, T. J. Harvey, and A. D. Ar- Lett. 98, 067204 (2007). mour, New J. Phys. 9, 84 (2007). 36 A. A. Clerk and S. Bennett, New J. Phys. 7, 238 (2005). 37 L. P. Kouwenhoven, C. M. Marcus, P. L. McEuen, S. Tarucha, R. M. Westervelt, N. S. Wingreen Electron trans- port in quantum dots, edited by L. L. Sohn, L. P. Kouwen- hoven, and G. Schon, Kluwer Series E345, 1997. 38 S. H. Ouyang, J. Q. You, and F. Nori, Phys. Rev. B 79, 075304 (2009). 39 D. A. Rodrigues and A. D. Armour, New J. Phys. 7, 251 (2005). 40 F. Pistolesi, J. of Low Temp. Phys. 154, 199 (2009). 41 F. Pistolesi, Ya. M. Blanter, and I. Martin, Phys. Rev. B 78, 085127 (2008). 42 M. G. Schultz, Phys. Rev. B 82, 195322 (2010). 43 W. Izumida and M. Grifoni, New J. Phys. 7, 244 (2005). 44 K. Flensberg, New J. Phys 8, 5 (2006). 45 F. Cavaliere, E. Mariani, R. Leturcq, C. Stampfer, and M. Sassetti, Phys. Rev. B 81, 201303(R) (2010). 64 T. Novotn´y, A. Donarini, C. Flindt, and A.-P. Jauho, Phys. Rev. Lett. 92, 248302 (2004). 65 T. Novotn´y, A. Donarini, and A.-P. Jauho, Phys. Rev. Lett. 90, 256801 (2003). 66 M. G. Schultz, Phys. Rev. B 82, 155408 (2010). 67 J. Konig, J. Schmid, H. Schoeller, and G. Schon, Phys. Rev. B 54, 16820 (1996). 68 C. Timm, Phys. Rev. B 77, 195416 (2008). 69 S.A. Gurvitz and Y.S. Prager, Phys. Rev. B 53, 15932 (1996). 70 M. Abramovitz, I. Stegun, Handbook of Mathematical Functions with Formulas, Graphs, and Mathematical Ta- bles, Dover, 1964. 71 H. Schoeller and J. Konig, Phys. Rev. Lett. 84, 3686 (2000). 72 G. Rempe, F. Schmidt-Kaler, and H. Walther, Phys. Rev. Lett. 64, 2783 (1990). 73 D. H. Santamore, A. C. Doherty, and M. C. Cross, Phys. 46 M. Galperin, A. Nitzan, and M. A. Ratner, Phys. Rev. B Rev. B 70, 144301 (2004). 73 045314 (2006). 74 E. Buks, E. Segev, S. Zaitsev, B. Abdo, and M. P. 47 M. Galperin, M. A. Ratner, and A. Nitzan, Nano Lett. 5 Blencowe, Eur. Phys. Lett. 81, 10001 (2008). 125 (2005). 75 M. J. Woolley, A. C. Doherty, and G. J. Milburn, Phys. 48 I. Wilson-Rae, N. Nooshi, W. Zwerger, and T. J. Kippen- Rev. B 82, 094511 (2010). berg, Phys. Rev. Lett. 99, 093901 (2007). 49 J. D. Teufel, J. W. Harlow, C. A. Regal, and K. W. Lehn- ert, Phys. Rev. Lett. 101, 197203 (2008). 50 J. D. Teufel, T. Donner, M. A. Castellanos-Beltran, J. W. Harlow, and K. W. Lehnert, Nature Nanotech. 4, 820 76 A. D. O'Connell, M. Hofheinz, M. Ansmann, R. C. Bial- czak, M. Lenander, Erik Lucero, M. Neeley, D. Sank, H. Wang, M. Weides, J. Wenner, John M. Martinis, and A. N. Cleland, Nature 464, 697 (2010).
1707.09282	2	1707	2018-01-16T16:15:15	Electrostatically induced quantum point contact in bilayer graphene	[ "cond-mat.mes-hall" ]	We report the fabrication of electrostatically defined nanostructures in encapsulated bilayer graphene, with leakage resistances below depletion gates as high as $R \sim 10~$G$\Omega$. This exceeds previously reported values of $R =~$10 - 100 k$\Omega$.\cite{Zou2010,Yan2010,Zhu2016a} We attribute this improvement to the use of a graphite back gate. We realize two split gate devices which define an electronic channel on the scale of the Fermi-wavelength. A channel gate covering the gap between the split gates varies the charge carrier density in the channel. We observe device-dependent conductance quantization of $\Delta G = 2~e^2/h$ and $\Delta G = 4~e^2/h$. In quantizing magnetic fields normal to the sample plane, we recover the four- fold Landau level degeneracy of bilayer graphene. Unexpected mode crossings appear at the crossover between zero magnetic field and the quantum Hall regime.	cond-mat.mes-hall	cond-mat	Electrostatically induced quantum point contact in bilayer graphene Hiske Overweg,∗ Hannah Eggimann, Marius Eich, Riccardo Pisoni, Yongjin Lee, Peter Rickhaus, Thomas Ihn, and Klaus Ensslin Solid State Physics Laboratory, ETH Zurich, CH-8093 Zurich, Switzerland National Graphene Institute, University of Manchester, Manchester M13 9PL, UK† Xi Chen, Sergey Slizovskiy, and Vladimir Fal'ko National Institute for Material Science, 1-1 Namiki, Tsukuba 305-0044, Japan Kenji Watanabe and Takashi Taniguchi We report the fabrication of electrostatically defined nanostructures in encapsulated bilayer graphene, with leakage resistances below depletion gates as high as R ∼ 10 GΩ. This exceeds previously reported values of R = 10 - 100 kΩ.1–3 We attribute this improvement to the use of a graphite back gate. We realize two split gate devices which define an electronic channel on the scale of the Fermi-wavelength. A channel gate covering the gap between the split gates varies the charge carrier density in the channel. We observe device-dependent conductance quantization of ∆G = 2 e2/h and ∆G = 4 e2/h. In quantizing magnetic fields normal to the sample plane, we recover the four- fold Landau level degeneracy of bilayer graphene. Unexpected mode crossings appear at the crossover between zero magnetic field and the quantum Hall regime. Nanostructures in graphene offer unique perspectives in terms of confinement strength, device geometry and possible spin coherence. In single layer graphene the for- mation of tunnel barriers, a fundamental building block of any nanostructure, has been demonstrated by many experiments in which narrow channels were defined by dry etching. These experiments suffer from randomly positioned localized states along the sample edges.4,5 As a consequence, the barrier transmission cannot be tuned monotonically by electrostatic gates.4,5. Bilayer graphene offers a promising alternative since a vertical electric field opens a band gap, which allows for deple- tion of the system. Several research groups used this property to define one-dimensional channels or quantum dots6–8, where the carriers are guided via a split gate structure with depleted graphene regions below the bi- ased split gates. For the experiments published so far, the minimal conductance achievable in such geometries is limited by leakage currents below the split gates, pre- sumably caused by hopping transport or a small energy gap. For tunnel barriers to be useful for high-quality quantum devices, the tunnelling resistance should exceed the resistance quantum h/e2 by far9. In this work we present two ultra-clean bilayer graphene samples encapsulated in hexagonal boron ni- tride (hBN) with a homogeneous top gate stripe crossing the current path in combination with a global graphite back gate. When depleting the region below the top gate, we measure resistances up to 105 × h/e2. In a next step, a split gate geometry was added to the devices, which was then covered by another insulating layer and a gate on top of the channel. In GaAs, similar QPC gate geome- tries have been studied10. This gate combination allows us to define an electron channel with resistances exceed- ing 1000× h/e2 when depleted. The combination of top gates and back gate is essential to separately tune the gap and the position of the Fermi level in the regions underneath the split gates as well as the carrier density in the channel. When the channel gate voltage is in- creased above the depletion voltage, the electron channel is opened and the conductance displays plateaus. For sample A the plateaus occur at conductance values 8, 10, 12, ..., 18 e2/h and for sample B at 4, 8, 12 e2/h. With increasing magnetic field perpendicular to the two- dimensional layer, we observe mode mixing and mode crossing evolving into the expected Landau level spec- trum for high magnetic fields. Sample A, drawn schematically in Fig. 1a, consists of a stack of bilayer graphene encapsulated in hexago- nal boron nitride on top of a graphite back gate. The stack was assembled using the van der Waals pick-up technique11 and was deposited on a Si/SiO2 substrate chip. The probed graphene area is delimited by the two Ohmic contacts and the natural edges of the graphene flake (dashed blue lines in Fig. 1b,c). On top of the de- vice, a 1 µm wide top gate (TG) and two 300 nm wide split gates (SG), separated by 100 nm, were evaporated (see atomic force microscopy image in Fig. 1b). Atomic layer deposition was performed to add a dielectric layer (Al2O3, 60 nm). Finally, another 200 nm wide gate, re- ferred to as channel gate (CH), was evaporated onto the channel defined by the split gates (see Fig. 1c). Sam- ple B was produced in the same way but has a thinner Al2O3 layer (20 nm), a smaller channel width (80 nm) and a narrower channel gate (60 nm). In sample B, two separate pairs of contacts are used to probe either the graphene region with top gate, or the graphene region with split gate geometry. More details about the sample fabrication and geometry can be found in the Supple- mental Material. Unless stated otherwise the measurements were per- formed at T = 1.7 K. An AC bias voltage of 50 µV 8 1 0 2 n a J 6 1 ] l l a h - s e m . t a m - d n o c [ 2 v 2 8 2 9 0 . 7 0 7 1 : v i X r a 2 maximum corresponds to the charge neutrality point of the outer regions of the sample which are not affected by the top gate voltage. The diagonal resistance maximum is the charge neutrality point of the sample region under- neath the top gate. The displacement field D increases in the direction of the arrow in Fig. 2a. It opens a band gap and hence increases the resistance at charge neutrality by several orders of magnitude. The global resistance maxi- mum, indicated by a black dot in Fig. 2a, coincides with the point of highest displacement field D = 0.7 V/nm. The red dots in Fig. 2c show the evolution of the resis- tance maximum as a function of temperature (the cor- responding configuration is sketched in Fig. 2d). Down to T = 20 K, the resistance follows an Arrhenius law (R ∼ exp(∆/(2kBT )) with a gap size ∆ = 55 meV. Below T = 20 K the resistance shows sub-exponential behavior, presumably because of hopping transport via mid-gap states. The resistance keeps increasing nonetheless. In this highly resistive regime, the resistance has been de- termined from the slope of I − V traces with a DC bias voltage range of VDC = ±10 mV. At T = 5 K the resis- tance is R ∼ 10 GΩ, which is the maximum resistance measurable in our set-up. For sample B, we measure a maximum resistance of R = 10 MΩ. In a third sample with a graphite back gate and a uniform top gate, we also measured resistances on the order of R ∼ 10 GΩ. Resistances on the order of gigaohms are rarely ob- served in bilayer graphene12,13. In most samples, a satu- ration of the resistance occurs in the megaohm range14–16 or below1–3. Zibrov et al.17 already pointed out that the use of graphite gates can significantly reduce sample dis- order. A high device quality with graphite gates was also reported in Ref. 18. The high resistance achieved in our three samples with a graphite back gate might be due to reduced disorder achieved by a better screening of charged impurities in the Si substrate, in the boron nitride and in the graphene itself, which leads to a re- duction of the number of mid-gap states. The differ- ent stray field pattern arising from a close by back gate might also play a role, as it modifies the doping profile across the sample. The lower resistance maximum mea- sured in sample B compared to the other samples can be explained by the fact that the graphene region below the top gate showed some bubbles in the AFM image. The lower quality of the graphene in this region can lead to more mid-gap states. In any case, the resistance of R = 10 MΩ measured in this region is still significantly higher than the resistance quantum. Figure 2b shows the resistance of sample A as a func- tion of split gate voltage (VSG) and back gate voltage (VBG), with a grounded top gate and channel gate. Lines of enhanced resistance follow the same pattern as in Fig. 2a. In contrast to Fig. 2a, the resistance along the displacement field axis does not increase beyond about R = 5 kΩ (note the different color scales of Figs. 2)a and 2b). This is because charge carriers can flow through the channel between the split gates. The channel can be depleted, however, by applying a FIG. 1. Sample layout. (a) Schematic of sample A. A bilayer graphene flake is encapsulated in hexagonal boron nitride. It is contacted by a source (S) and drain (D) contact and has a graphite back gate (BG) below, a top gate (TG), two split gates (SG) and a channel gate (CH) on top. The channel gate is separated from the split gates by a dielectric layer of Al2O3. (b) Atomic force microscopy image of the sample prior to deposition of the channel gate. The position of the graphene flake is indicated by blue dashed lines. (c) Atomic force microscopy image of the sample with the channel gate. (d) Model of the band structure along the y-direction with the electrostatic potential indicated by the blue line. The Fermi level under the split gates lies in the band gap. The channel gate induces a finite carrier density in the channel. was applied and the current I was measured using low- frequency lock-in techniques. In order to illustrate the basic idea of electrostatic con- finement in bilayer graphene, we take a look at Fig. 1d. It shows a schematic of the E(k) dispersion relation at three different points across the quantum point contact (QPC), indicated in Fig. 1b. When the Fermi level under the split gates lies in the gap (I., III.) and the Fermi level in the channel lies in the conduction band (II.), charge carriers can only flow through the narrow channel. A fi- nite element simulation of the electrostatic potential can be found in the Supplemental Material. To demonstrate experimentally that a band gap opens, we first look into the combined effect of the top gate (TG) and the back gate (BG), whilst keeping the split gate and the channel gate grounded. Figure 2a shows the resis- tance of sample A as a function of top gate voltage VTG and back gate voltage VBG. The horizontal resistance Al2O3hBNhBNgraphitebackgateBG↑bilayergraphene−→x−→ySTGDSGSG2µmI.II.III.STGDSGCH2µm←−y0200h(nm)E(a.u.)EFVSGVSGVCHI.III.II.CH−→kx(a)(b)(c)(d)3 FIG. 2. Sample characterization of sample A. (a) Two- terminal resistance R as a function of top gate voltage VTG and back gate voltage VBG. The split gates and the channel gate were grounded. The diagonal line corresponds to charge neutrality underneath the top gate. Along this line the dis- placement field D increases, which results in an increase of resistance. (b) Two-terminal resistance R as a function of split gate voltage VSG and back gate voltage VBG. The chan- nel gate and the top gate were grounded. In contrast to (a), the resistance does not increase with increasing displacement field, because charge carriers can flow through the channel. (c) Resistance R as a function of temperature T for the resis- tance maximum induced by the top gate and back gate (black dot in (a), schematic in (d)) and the resistance maximum in- duced by the combination of the split gates, the back gate (black dot in (b)) and the channel gate at VCH = −12 V, schematic in (e)). Gap sizes of ∆ = 55 meV and ∆ = 47 meV were extracted from the high temperature behavior. channel gate voltage VCH = −12 V. The blue triangles in Fig. 2c show the resistance as a function of temperature for (VSG,VBG) = (−3.9, 4) V (black dot in Fig. 2b) and VCH = −12 V, which gives the highest resistance achiev- able at T = 1.7 K using the split gates and the channel gate (configuration in Fig. 2e). In the high temperature regime a gap energy of ∆ = 47 meV can be extracted. The resistance deviates from the activated behavior be- low T ∼ 50 K and goes up to R = 50 MΩ at T = 1.7 K, which is three orders of magnitude higher than the re- sistance quantum h/e2. In sample B the maximal resis- tance achieved with the split gates and the channel gate is R = 20 MΩ at T = 1.7 K. These results are in contrast with previous works on bilayer graphene QPCs, which showed a minimal conductance above G = e2/h.6,7 They show that it is not only possible to achieve high resis- tances with a rather wide uniform gate, but also with a combination of three narrower gates. The band gap underneath the split gates is sufficient to suppress con- ductance when the Fermi energy is in the gap. We will therefore focus below on the conductance of the channel. We vary the channel gate voltage VCH in the regime where conductance under the split gates is maximally suppressed. For sample A the back gate voltage could not be increased above VBG = 4 V because of the on- set of gate leakage, most likely due to the thin hBN layer between the back gate and the contacts. Suppres- sion of conductance under the gates was only reached at (VSG,VBG) = (−3.9, 4) V (see black dot in Fig. 2b). The conductance G at this operating point as a func- tion of channel gate voltage VCH is shown in Fig. 3a. A series resistance of RS = 150 Ω was subtracted, which was determined by measuring the resistance at (VSG,VBG) = (−0.4, 4) V (see white dot in Fig. 2b). This point corresponds to uniform doping throughout the sample. The conductance shows plateaus at G = 8, 10, 12, 14, 16, 18 e2/h. No plateaus are discernable below G = 8 e2/h. To our knowledge this is the largest number of conductance plateaus observed in bi- layer graphene to date. At VCH = −12 V the channel is depleted, reaching a resistance of R = 50 MΩ. Sample B has a larger back gate voltage range with gate leakage smaller than 0.1 nA. Figure 3b shows its conductance as a function of channel gate voltage for a set of back gate - split gate voltage pairs. Under the split gates, increasing voltage differences VBG − VSG cor- respond to an increasing displacement field D along the charge neutrality line (cf. Fig. 2b). For each curve, a series resistance equal to the resistance measured at uni- form doping at the corresponding back gate voltage was subtracted. Throughout the whole range, plateaus can be observed slightly below G = 4, 12 e2/h (see blue ar- rows). For VBG < 6 V a plateau occurs slightly below 8 e2/h as well (dashed blue arrow). In the range above G = 12 e2/h small oscillations are observed which cannot be identified to be quantized conductance plateaus. Sample B could also be depleted completely at VBG = −8 V, when the entire sample is p-doped. The conduc- -6-4-20246VTG(V)-4-2024VBG(V)←DT=1.7KSDTG102103104105106107R(Ω)-6-4-20246VSG(V)-4-2024VBG(V)←DT=1.7KSDSGSG1234R(kΩ)0.00.10.20.30.40.51/T(1/K)1021041061081010R(Ω)∆=55meV∆=47meVSDTGSDSGSGCH→(a)(b)(c)(d)(e)tance as a function of channel gate voltage in this setting shows several smaller kinks, but no quantized conduc- tance plateaus. In sample A it was not possible to deplete the channel in the p-doped regime, most likely because of the limited back gate voltage range. Our results fit well into the landscape of experiments on single- and bilayer graphene QPCs published previ- ously, where lifted degeneracies were observed in some but not all samples. Theoretically, for pristine bilayer graphene, steps of ∆G = 4 e2/h are expected because of spin- and valley-degeneracy, as observed in sample B. However, the observed step size of ∆G = 2 e2/h in sam- ple A, witnessing a lifted degeneracy, is in agreement with other experimental works on bilayer graphene.6,7 In monolayer graphene, conductance quantization with steps of ∆G = 2 e2/h was observed in both limits of low19,20 and high21 mode number. However, Kim et al.22 reported conductance quantization with a step size ∆G = 4 e2/h in an electrostatically induced channel in monolayer graphene. Zimmermann et al.23 studied a QPC in single layer graphene in the quantum Hall regime where a step size of ∆G = 1 e2/h is observed. We speculate that the difference in the observed de- generacies in samples A and B is caused by the residual disorder in these devices. In the quantum Hall regime all degeneracies in the lowest Landau level are lifted in our samples (see below), which demonstrates the good sample quality. Yet at zero magnetic field, the lack of perfect flatness of plateaus, the deviations from the ex- pected plateau values, the occasionally missing plateaus, and the absence of plateaus in a p-doped channel indi- cate that a further increase in device quality, currently out of reach, would lead to improvements. Beyond that, strain effects could modify the potential landscape. In GaAs heterostructures, it is well known that a difference in thermal expansion coefficient between the metal gates evaporated on top of the semiconductor wafer and the semiconductor material itself can lead to a strain-induced change of the potential of 5-10%.24 In our case, the hBN layer separating the metal gate from the graphene layer is comparatively thinner, and one can imagine that strain effects could also lead to modifications of the potential, in addition to the electrostatic definition of the QPC. While further improvements in device quality will lead to better reproducibility among different devices and allow for in- vestigating more subtle interaction effects, such as spon- taneous spin polarization25, at present the microscopic origin of the lifted degeneracy and the missing plateaus at low mode numbers in device A remains unknown. A magnetic field has the potential to give further in- sights into degeneracy lifting in QPCs. Figure 4a(b) shows the conductance of sample A(B) as a function of VCH for selected magnetic field strengths. For these measurements, the density in the bulk of the sample is considerably higher than the density in the channel. The conductance is therefore governed by the filling factor of the channel (see Supplemental Material). In a magnetic field of B = 7 T we observe that the four-fold degener- 4 acy of the lowest Landau level is completely lifted in both samples, demonstrating the high quality of the samples.26 Sample A shows a step size of ∆G = 4 e2/h at interme- diate magnetic fields (see curve at B = 2.5 T). This is surprising, since the step size at B = 0 T (Fig. 3a) and B = 1.6 T (see arrows in Fig. 4a) is only ∆G = 2 e2/h for this sample. In sample B, no clear quantization of the levels is observed at intermediate magnetic fields. The transconductance as a function of VCH and B, shown in Fig. 4c,d, provides a more complete picture. Transitions between quantized modes are seen as dark lines. In sample A, the transitions between the plateaus are more pronounced than in sample B. In both samples, these lines start out vertically at low magnetic fields, and bend over between B = 1 T and B = 2 T towards more positive gate voltages, ending up as straight lines with finite slope at high fields. This behavior is reminiscent of the magnetoconductance of high quality QPCs, for example in GaAs, where the low magnetic field conduc- tance is confinement dominated, whereas the high mag- netic field conductance is determined by edge channels formed in crossed electric and magnetic fields. The effect is known as the magnetic depopulation of magnetoelec- tric subbands.27 Also in our samples, filling factors can be assigned to the light regions between the lines as in- dicated in the figure. However, when the magnetic field is decreased towards the confinement dominated regime, the mode structure appears to be much more complicated than in GaAs. Lacking a detailed theory we propose a heuristic model which describes the level transitions of sample A as a function of magnetic field. In analogy to magnetic de- population in GaAs 2DEGs28, we assume that the energy separation of the modes in the channel is given by (cid:113) (cid:112) EN = Ω N (N − 1), Ω = ω2 0 + ω2 c (1) where ω0 is the frequency of the electrostatic confine- ment potential in the absence of a magnetic field, and ωc is the cyclotron frequency, given by ωc = eB/m∗. Assuming a linear conversion from gate voltage VCH to energy E = αe(VCH − V ), it is impossible to fit a mode spectrum as that described by Eq. 1 to all the levels ob- served in Fig. 3d using α, V and ω0 as free fitting pa- rameters. Yet by extending the model with a second set of parameters α(cid:48), V (cid:48) and ω(cid:48) 0 it is possible to cap- ture the trends of the level crossings in the low mag- netic field regime. This is demonstrated by the dashed orange and blue lines in Fig. 3d. The employed param- eters are ω0 = 7.5 meV, α = 1.75 × 10−3, V = 13.5 V, 0 = 5 meV, α(cid:48) = 1.4 × 10−3 and V (cid:48) = 17 V. We ω(cid:48) want to stress that the model is purely heuristic. It was designed to capture the dominant features of the experi- ment. The two different frequencies could imply that the two valley/spin split modes may have different effective masses. The difference between V and V (cid:48) indicates an energy offset between the two sets of levels. The model captures the main features of the data, except for the 5 FIG. 3. (a) Conductance G of the induced channel in sample A as a function of VCH at B = 0 T for the gate voltage configuration at the black dot in Fig. 2b. The conductance shows a number of steps of ∆G = 2 e2/h. (b) Conductance G of the channel in sample B as a function of channel gate voltage VCH at B = 0 T for several combinations of back gate and split gate voltage. The conductance shows plateaus slightly below G = 4, 8, 12 e2/h. part where VCH < −10 V (where the conductance at B = 0 T already deviates from the expected pattern), and the features marked by yellow crosses in Fig. 3d. The parameters ω0 and ω(cid:48) 0 are similar to the curva- ture of the harmonic potential calculated in a COMSOL simulation of the electrostatic potential of the device (see Supplemental Material). The parameters α and α(cid:48) are in rough agreement with the slope of the finite bias dia- mond boundaries, which yield a lever arm of 2.6 × 10−3 (see blue dashed lines in Fig. 5b.) The data suggest that around B = 4 T, the spin and valley splittings are too small to be resolved. The only relevant energy spacing is the Landau level spacing ELL (see inset of Fig. 3a). Lowering the magnetic field, the rel- ative influence of the electrostatic potential compared to the magnetic confinement grows, which lifts a degeneracy (the blue and orange dashed lines move apart). The black curly bracket in Fig. 3d indicates the energy range of the lifted degeneracy at B = 0 T (E1 in Fig. 3a), which seems to have grown larger than the mode spacing indicated by the blue curly bracket (E2 in Fig. 3a). The remaining twofold degeneracy implies that the energy scale E3 = 0. Although the model suggests a degeneracy lifting larger than the mode spacing of the QPC, we currently do not know which mechanism could be responsible for such a drastic effect. Another aspect which may contribute to the crossing mode pattern is the fact that the channel gate voltage changes the displacement field D inside the channel. Bi- layer graphene exhibits a valley splitting of the Landau levels which depends on the displacement field29–31. In the devices presented here, the charge carrier density and the displacement field in the channel cannot be varied in- dependently, complicating a systematic study of the ef- fect of the displacement field. Finite bias measurements were performed to extract subband energy spacings. Figure 5a shows the transcon- ductance dG/dVCH as a function of source drain bias measured at T = 1.7 K and B = 0 T. Minima in the transconductance are observed at the positions of the plateaus in Fig. 3a (see crosses), but there is no sim- ple diamond pattern. The energy spacing seems to be on the order of ∆E ≈ 1 meV. In sample B, features with a similar energy spacing are observed in finite bias mea- surements. In a finite magnetic field, a more pronounced diamond pattern is recovered. This can be seen in the transconductance measurement of sample A in Fig. 5b, recorded at T = 0.13 K and B = 1.6 T. The centers of the diamonds correspond well to the conductance plateaus in Fig. 3d, even though these measurements were performed during different cooldowns. Compared to the level spac- ing at B = 0 T, the level spacing ∆E ≈ 7 meV at B = 1.6 T shows a better agreement with the level spac- ing extracted from the heuristic model, which predicts a level spacing of ∆E = 7.5 meV at B = 1.6 T for the orange line set in Fig. 4c. In conclusion, we have demonstrated that bilayer graphene samples with graphite back gate show resis- tances of at least R = 10 MΩ in high displacement fields. This is a significant improvement compared to devices without a graphite back gate, where the resistance usu- ally saturates in the regime of R = 10 − 100 kΩ.1–3 We exploit this result to electrostatically define QPCs in bi- layer graphene and observe the following: 1. In both samples, the channels can be fully depleted by gating. 2. Both samples show quantized conductance though with different values for degeneracies. -12-8-40VCH(V)048121620G(e2/h)sampleAVBG=4.0VVSG=-3.9V-8-404VCH(V)sampleB←D←⇠←VBGVSG5.0V-2.8V5.5V-3.0V6.0V-3.1V6.5V-3.3V7.0V-3.5V7.5V-3.7V8.0V-3.9V(a)(b)6 FIG. 4. (a) Conductance of sample A for several magnetic field strengths. The plateaus at 10, 14 and 18 e2/h are still present at B = 1.6 T (see arrows), but disappear in higher magnetic fields. At B = 7 T (red line) plateaus are present at 1, 2 ,3, 4 e2/h. (b) Conductance of sample B for several magnetic field strengths. (c) Transconductance of sample A as a function of channel gate voltage VCH and magnetic field B. The blue and orange dashed lines both follow the model described by Eq. 1. (d) Transconductance of sample B as a function of channel gate voltage VCH and magnetic field B. The transitions between modes are less pronounced than in sample A. Horizontal dashed lines correspond to the line cuts in (a),(b). 3. Both samples show the expected quantum Hall plateaus with 4 fold degeneracies at high fields and com- plete lifting of degeneracies for the lowest Landau levels. one-dimensional nanostructures in bilayer graphene by electrostatic gating paves the way towards controllable quantum dots in bilayer graphene. 4. Both samples show an intricate crossover regime between zero magnetic field and quantum Hall regime where level crossings and avoided crossings occur. The different step sizes of ∆G = 2 e2/h and ∆G = 4 e2/h in the two samples might be due to a different disorder potential or different strain patterns. Several factors, such as the reduced transmission of the modes of sample B and the absence of conductance quantiza- tion in the p-doped regime, indicate that mesoscopic de- tails of the samples play an important role. Realizing ACKNOWLEDGEMENTS We thank Anastasia Varlet, Leonid Levitov, Kostya Novoselov and Mansour Shayegan for fruitful discus- sions. We acknowledge financial support from the Euro- pean Graphene Flagship and the Swiss National Science Foundation via NCCR Quantum Science and Technol- ogy. Growth of hexagonal boron nitride crystals was sup- ported by the Elemental Strategy Initiative conducted by the MEXT, Japan and JSPS KAKENHI Grant Numbers JP15K21722. ∗ [email protected] † On leave of absence from NRC "Kurchatov Institute" PNPI, Russia 048121620G(e2/h)→→→sampleAVBG=4.0VVSG=-3.9VsampleBVBG=5.0VVSG=-2.8VB=1.6TB=2.5TB=7.0T-12012VCH(V)0246B(T)ν=4ν=8ν=12-404VCH(V)ν=4ν=80dG/dVCH(a.u.)B∼0TB∼4TE2E3E1ELL(a)(b)(c)(d)7 (1997) pp. 105–214, arXiv:9612126 [cond-mat]. 10 Y. J. Um, Y. H. Oh, M. Seo, S. Lee, Y. Chung, N. Kim, V. Umansky, and D. Mahalu, Applied Physics Letters 100 (2012), 10.1063/1.4710522. 11 L. Wang, I. Meric, P. Y. Huang, Q. Gao, Y. Gao, H. Tran, T. Taniguchi, K. Watanabe, L. M. Campos, D. a. Muller, J. Guo, P. Kim, J. Hone, K. L. Shepard, and C. R. Dean, Science 342, 614 (2013). 12 J. Li, K. Wang, K. J. McFaul, Z. Zern, Y. Ren, K. Watan- abe, T. Taniguchi, Z. Qiao, and J. Zhu, Nature Nanotech- nology 11, 1060 (2016). 13 M. Sui, G. Chen, L. Ma, W.-Y. Shan, D. Tian, K. Watan- abe, T. Taniguchi, X. Jin, W. Yao, D. Xiao, and Y. Zhang, Nature Physics 11, 1 (2015), arXiv:1501.04685. 14 R. T. Weitz, M. T. Allen, B. E. Feldman, J. Martin, and A. Yacoby, Science 330, 812 (2010), arXiv:1010.0989. 15 T. Taychatanapat and P. Jarillo-Herrero, Physical Review Letters 105, 1 (2010), arXiv:/arxiv.org/abs/1009.0714 [http:]. 16 F. Xia, D. B. Farmer, Y. M. Lin, and P. Avouris, Nano Letters 10, 715 (2010), arXiv:1001.3915. 17 A. A. Zibrov, C. Kometter, H. Zhou, E. M. Spanton, and A. F. T. Taniguchi, K. Watanabe, M. P. Zaletel, Young, Nature 549, 360 (2017). 18 J. I. A. Li, C. Tan, S. Chen, Y. Zeng, T. Taniguchi, K. Watanabe, J. Hone, and C. R. Dean, Science 2521, 1 (2017), arXiv:1705.07846. 19 N. Tombros, A. Veligura, J. Junesch, M. H. D. Guimaraes, and B. J. van Wees, I. J. V. Marun, H. T. Jonkman, Nature Physics 7, 697 (2011), arXiv:1102.0434. 20 S. Somanchi, B. Terr´es, J. Peiro, M. Staggenborg, K. Watanabe, T. Taniguchi, B. Beschoten, and C. Stampfer, Annalen der Physik 1700082, 1 (2017), arXiv:1706.09781. 21 B. Terr´es, L. A. Chizhova, F. Libisch, J. Peiro, D. Jorger, S. Engels, A. Girschik, K. Watanabe, T. Taniguchi, S. V. Rotkin, J. Burgdorfer, and C. Stampfer, Nature Commu- nications 7, 1 (2016), arXiv:1603.00844. 22 M. Kim, J.-H. Choi, S.-H. Lee, K. Watanabe, T. Taniguchi, (2016), and H.-J. Lee, Nature Physics S.-H. Jhi, 10.1038/nphys3804. 23 K. Zimmermann, A. Jordan, F. Gay, K. Watanabe, T. Taniguchi, Z. Han, V. Bouchiat, H. Sellier, and B. Sac´ep´e, Nature Communications 8, 14983 (2017), arXiv:1605.08673. 24 P. D. Ye, D. Weiss, R. R. Gerhardts, K. von Klitzing, K. Eberl, H. Nickel, and C. T. Foxon, Semiconductor Sci- ence and Technology 10, 715 (1995). 25 E. V. Castro, N. M. R. Peres, T. Stauber, and N. A. P. Silva, Physical Review Letters 100, 186803 (2008), arXiv:0711.0758. 26 A. Varlet, D. Bischoff, P. Simonet, K. Watanabe, T. Taniguchi, T. Ihn, K. Ensslin, M. Mucha-Kruczy´nski, and V. I. Fal'ko, Physical Review Letters 113, 116602 (2014). 27 C. Beenakker and H. V. Houten, Solid state physics 44 (1991), arXiv:0412664v1 [arXiv:cond-mat]. 28 M. Buttiker, Physical Review B 41, 7906 (1990). 29 L. M. Zhang, M. M. Fogler, and D. P. Arovas, Phys- ical Review B 84 (2011), 10.1103/PhysRevB.84.075451, arXiv:1008.1418. 30 K. Lee, B. Fallahazad, J. Xue, D. C. Dillen, K. Kim, T. Taniguchi, K. Watanabe, and E. Tutuc, Science 345, 58 (2014), arXiv:1401.0659. (a) Transconductance dG/dVCH of sample A as a FIG. 5. function of channel gate voltage VCH and source drain bias VSD for the gate configuration of the black dot in Fig.2b. The measurement was performed at T = 1.7 K and B = 0 T. At the positions of the white crosses, corresponding to the plateaus marked by crosses in Fig. 3a, minima in transconduc- tance can be observed. (b) Transconductance at B = 1.6 T. The white crosses indicate the positions of the plateaus in Fig. 4a and coincide with the diamond shaped features ob- served here around VSD = 0 mV, even though the two mea- surements were recorded during different cooldowns. 1 K. Zou and J. Zhu, Physical Review B 82, 1 (2010), arXiv:1008.0984. 2 J. Yan and M. S. Fuhrer, Nano Letters 10, 4521 (2010), arXiv:1010.1036. 3 M. J. Zhu, A. V. Kretinin, M. D. Thompson, D. A. Ban- durin, S. Hu, G. L. Yu, J. Birkbeck, A. Mishchenko, I. J. Vera-Marun, K. Watanabe, T. Taniguchi, M. Polini, J. R. Prance, K. S. Novoselov, A. K. Geim, and M. Ben Shalom, Nature Communications 8, 1 (2017), arXiv:1612.05902. 4 D. Bischoff, P. Simonet, A. Varlet, H. C. Overweg, M. Eich, and K. Ensslin, Physica Status Solidi - Rapid T. Ihn, Research Letters 10, 68 (2016). 5 D. Bischoff, F. Libisch, J. Burgdorfer, T. Ihn, and K. En- sslin, Physical Review B 90, 115405 (2014). 6 M. T. Allen, J. Martin, and A. Yacoby, Nature Commu- nications 3, 934 (2012). 7 A. S. M. Goossens, S. C. M. Driessen, T. a. Baart, K. Watanabe, T. Taniguchi, and L. M. K. Vandersypen, Nano Letters 12, 4656 (2012). 8 S. Droscher, C. Barraud, K. Watanabe, T. Taniguchi, T. Ihn, and K. Ensslin, New Journal of Physics 14, 103007 (2012). 9 N. W. L.P. Kouwenhoven, C. M. Marcus, P.L. McEuen, S. Tarucha, R. M. Westervelt, in Mesoscopic Electron Trans- port, Volume 345 of Nato ASI Series E: Applied Sciences -12-8-40VCH(V)-202VSD(mV)T=1.7K,B=0T-8-404812VCH(V)-12-60612VSD(mV)T=0.13K,B=1.6T0123dG/dVCH(e2/hV−1)(a)(b)31 B. M. Hunt, J. I. A. Li, A. A. Zibrov, L. Wang, T. Taniguchi, K. Watanabe, J. Hone, C. R. Dean, M. Za- letel, R. C. Ashoori, and A. F. Young, Nature Communi- cations 8, 1 (2017), arXiv:1607.06461. 8 SupportingInformationforElectrostaticallyinducedquantumpointcontactsinbilayergrapheneHiskeOverweg,1,∗HannahEggimann,1XiChen,2SergeySlizovskiy,2MariusEich,1RiccardoPisoni,1YongjinLee,1PeterRickhaus,1KenjiWatanabe,3TakashiTaniguchi,3VladimirFal'ko,2ThomasIhn,andKlausEnsslin1SolidStatePhysicsLaboratory,ETHZrich,CH-8093Zrich,Switzerland2NationalGrapheneInstitute,UniversityofManchester,ManchesterM139PL,UK†3NationalInstituteforMaterialScience,1-1Namiki,Tsukuba305-0044,JapansampleAsampleBgraphitethickness(nm)2815bottomBNthickness(nm)3853topBNthickness(nm)3525Al2O3thickness(nm)6030SGheight(nm)6020channelwidth(nm)10080channelgatewidth(nm)20060mobility(cm2/Vs)8×1046×104TABLEI.CharacteristicsofsamplesAandBFABRICATIONANDCHARACTERIZATIONThegeometryandmobilityofbothsamplescanbefoundinTableI.Fortheetchingofthecontacts,weusearecipeadaptedfromRef.1.Weuseareactiveionetcher(OxfordInstru-mentsRIE80Plus),withamixtureofCHF3gas(40sccm)andO2(4sccm).WithanRFpowerof60W,theobtainedetchrateofhBNis45nm/min.Wecarefullychooseanetch-ingtimeforeachindividualsampletomakesurethatthehBNisetchedsufficientlyforthecontactstoreachthegraphenelayer,butnottoofar,sincethiswouldleadtoashortbetweenthecontactsandthegraphitebackgate.ThedepositionofAl2O3wasdoneinanatomiclayerde-positionsystem(PicosunSunaleR-150B)atatemperatureof150oCwithtrimethylaluminum(TMA)andwaterasprecur-sorgases.ForsampleAweperformedtemperaturedependentmea-surementsoftheresistancemaximumfor(VTG,VBG)=-3.9,4V,correspondingtoadisplacementfieldof0.7V/nm,and(VTG,VBG)=-1.7,2V,correspondingtoadisplacementfieldof0.4V/nm.Therespectivegapsizeswere55meVand16meV.Anindirectmeasureofthegapsizeistheresistanceatchargeneutrality.FigureS1ashowstheresistanceofsam-pleAalongseverallinecutsinFig.2a.Theresistanceatthechargeneutralitypointincreasesbyordersofmagnitudewhenabandgapisopened.ThemeasurementwasperformedwithabiasvoltageofV=50µV.Thedataforthehighestdisplace-mentfield(VBG=4V)hasbeenomitted,becausethehighresistancepeakcannotbereliablymeasuredwithasmallbias-4-2024VTG(V)103104105106107R(Ω)(a)VBG-3.0V-2.2V-1.5V-0.8V0.0V0.8V1.5V2.2V3.0V-4-2024VTG(V)104105106107R(Ω)(b)VBG-6.0V-4.5V-3.0V-1.5V0.0V1.5V3.0V4.5V6.0VFIG.S1.(a)ResistanceofsampleAasafunctionoftopgatevoltageforseveralbackgatevoltages.Theresistanceatthechargeneutralitypointincreasesbyordersofmagnitudewhenabandgapisopened.(b)SameforsampleBvoltage.SimilardataforsampleBisshowninFig.S1b.SIMULATIONOFTHEELECTROSTATICPOTENTIALTogetmoreinsightintheelectrostaticpotentialofthequan-tumpointcontact,weuseafiniteelementsimulator(COM-SOL).Thechargecarriersheetdensityn(~r)andthepotentialV(~r)arecalculatedself-consistentlyusingPoisson'sequationandtheThomas-Fermiapproximation.ThedensityofstatesisapproximatedbyD(E)=m∗/(π¯h2)θ(E)withm∗=0.034meandθtheHeavysidestepfunction,whichlimitsthemodeltotransportintheconductionband.WedonottaketheMexi-2y→CHSGSGAl2O3↑bilayergrapheneBGhBNFIG.S2.CrosssectionofthesampleusedforComsolsimulation.Thechannelgatehasanellipticextensionabovethechannel.canhatshapeofthebilayergraphenebandstructureandthepositiondependentbandgapintoaccount.Thequantizationofstatesinsidetheone-dimensionalchannel,duetothelateralconfinement,isalsoneglected.GeometryThegeometryconsideredforthesimulationisthesameasforsampleA(seemaintextandtableI).FromAFMimagesofsampleAitisapparentthatthechannelgatedropspar-tiallyintotheopeningbetweenthesplitgates.Wethereforemodelledthechannelgatewithanellipticalextensionabovethechannelregion(seeFig.S2).Thewidthofthisexten-sionwas150nmandthedepth30nm,inagreementwiththeAFMimages.Whenomittingtheextensionfromthesimula-tion,thechannelgetsdepletedaroundVCH=−25VinsteadoftheexperimentallyobservedVCH=−12V.Withtheex-tensionthechannelgetsdepletedclosetotheexperimentallyobservedvalue.DepletionbelowsplitgatesFigureS3showsthecalculatedelectrostaticpotentialalongalinecut(blacklineininset)underthesplitgateforVBG=4Vandvarioussplitgatevoltages.AtVSG=0V,thepotentialisconstantthroughoutthestructureasexpected.AtVSG=−3.95V,theregionunderthesplitgatesisdepleted.Thisvoltageisinagreementwiththeexperimentallydetermineddepletionvoltage.PotentialinsidechannelInFig.S4athepotentialinsidethechannelisshown,forthecaseinwhichtransportunderthesplitgatesissuppressed(VBG=4V,VSG=−3.95V).ForVCH=0V,thepotentialcanbeapproximatedbyaharmonicpotentialV(y)=12m∗ω20y20.00.20.40.60.81.0x(µm)-4004080120E(meV)VBG=4VEFVSG0V-1V-2V-3V-3.95VFIG.S3.Calculatedelectrostaticpotentialalongalinecut(blacklineininset)underthesplitgateforVBG=4Vandvarioussplitgatevoltages.Thedottedlinesindicatetheextentofthesplitgates.AtVSG=0V,thepotentialisconstantthroughoutthestructure.AtVSG=−3.95V,theregionunderthesplitgatesisdepleted.withanenergylevelseparationof¯hω0=8.4meV.ForthegatevoltagerangeofVCH=−10V-12Vtheenergylevelseparationchangesaccordingto¯hω0(VCH)=8.4meV+αeVCHwithα=0.33×10−3.TheincreasedmodespacingwithhigherchannelgatevoltagecanalsobeobservedinFig.3:theconductanceriseslesssteeplyforhigherchannelgatevoltage.FigureS4bshowsthepotentialacrosstheQPC.FortherangeofVCH=−12V-8Vthepositivecurvatureiny-directionandthenegativecurvatureinx-directionleadtoaconventionalsaddlepointpotential.Similarresultswereob-tainedforsimulationsofsampleB.ROLEOFTHEBULKINMAGNETOTRANSPORTThedensityinthebulkofthesampleishigherthanthedensityinsidetheconstrictionfortheentirerangeofFig.4.Onlytheedgemodesthatexistinboththebulkandthechan-nelcontributetotransportandtheconductanceisgivenbyG=νCHe2/h.Wecanseeamodestinfluenceofthebulkofthedevicewheneverthebulkisatatransitionbetweenintegerfillingfactors.FigureS5showsthederivativeoftheconduc-tancewithrespecttomagneticfieldasafunctionofVCHandBforthesamegatevoltagesettingsasFig.4.Thehorizon-tallines,whichoccurina1/Bperiodicfashion,correspondtoLandauleveltransitionsinthebulkofthesample.Thefill-ingfactorsofthebulkareindicatedonthey-axis.Becauseofthehighchargecarrierdensity,nobrokendegeneraciesareobserveduptoB=8T.30.30.350.40.450.5y(µm)-4004080120E(meV)VBG=4VVSG=-3.95V0.00.20.40.60.81.0x(µm)EFVCH-12V-10V-8V-6V-4V-2V0V2V4V6V8V10V12V(a)(b)FIG.S4.PotentiallandscapeforVBG=4V,VSG=−3.95Vandvariouschannelgatevoltages.(a)Electrostaticpotentialacrossthechannel,whichcanbeapproximatedbyaharmonicpotential(dottedblackline).ThechannelgetsdepletedclosetotheexperimentaldepletionvoltageofVCH=−12V.(b)Electrostaticpotentialalongthechannel.FortherangeofVCH=−12V-8Vaconventionalsaddlepointpotentialisobserved.-12-60612VCH(V)0246B(T)1620242832νbulkdG/dB(a.u.)FIG.S5.DerivativeoftheconductancewithrespecttomagneticfieldforthesamegatevoltagesettingsasFig.4c.1/Bperiodichorizontallinesareobserved,whichcorrespondtoLandauleveltransitionsinthebulkofthesample.∗[email protected]†OnleaveofabsencefromNRC"KurchatovInstitute"PNPI,Rus-sia[1]L.Wang,I.Meric,P.Y.Huang,Q.Gao,Y.Gao,H.Tran,T.Taniguchi,K.Watanabe,L.M.Campos,D.a.Muller,J.Guo,P.Kim,J.Hone,K.L.Shepard,andC.R.Dean,Science342,614(2013).
1308.0554	2	1308	2013-11-12T15:16:20	Topological Boundary Modes in Isostatic Lattices	[ "cond-mat.mes-hall", "cond-mat.soft" ]	Frames, or lattices consisting of mass points connected by rigid bonds or central force springs, are important model constructs that have applications in such diverse fields as structural engineering, architecture, and materials science. The difference between the number of bonds and the number of degrees of freedom of these lattices determines the number of their zero-frequency "floppy modes". When these are balanced, the system is on the verge of mechanical instability and is termed isostatic. It has recently been shown that certain extended isostatic lattices exhibit floppy modes localized at their boundary. These boundary modes are insensitive to local perturbations, and appear to have a topological origin, reminiscent of the protected electronic boundary modes that occur in the quantum Hall effect and in topological insulators. In this paper we establish the connection between the topological mechanical modes and the topological band theory of electronic systems, and we predict the existence of new topological bulk mechanical phases with distinct boundary modes. We introduce model systems in one and two dimensions that exemplify this phenomenon.	cond-mat.mes-hall	cond-mat	Topological Boundary Modes in Isostatic Lattices C. L. Kane and T. C. Lubensky Dept. of Physics and Astronomy, University of Pennsylvania, Philadelphia, PA 19104 Frames, or lattices consisting of mass points connected by rigid bonds or central-force springs, are important model constructs that have applications in such diverse fields as structural engineering, architecture, and ma- terials science. The difference between the number of bonds and the number of degrees of freedom of these lattices determines the number of their zero-frequency “floppy modes”. When these are balanced, the system is on the verge of mechanical instability and is termed isostatic. It has recently been shown that certain extended isostatic lattices exhibit floppy modes localized at their boundary. These boundary modes are insensitive to local perturbations, and appear to have a topological origin, reminiscent of the protected electronic boundary modes that occur in the quantum Hall effect and in topological insulators. In this paper we establish the connection be- tween the topological mechanical modes and the topological band theory of electronic systems, and we predict the existence of new topological bulk mechanical phases with distinct boundary modes. We introduce model systems in one and two dimensions that exemplify this phenomenon. Isostatic lattices provide a useful reference point for un- derstanding the properties of a wide range of systems on the verge of mechanical instability, including network glasses1,2 , randomly diluted lattices near the rigidity perco- lation threshold3,4 , randomly packed particles near their jam- ming threshold5–10 , and biopolymer networks11–14 . There are many periodic lattices, including the square and kagome lat- tices in d = 2 dimensions and the cubic and pyrochlore lattices in d = 3, that are locally isostatic with coordina- tion number z = 2d for every site under periodic bound- ary conditions. These lattices, which are the subject of this paper, have a surprisingly rich range of elastic responses and phonon structures15–19 that exhibit different behaviors as bending forces or additional bonds are added. The analysis of such systems dates to an 1864 paper by James Clerk Maxwell20 that argued that a lattice with Ns mass points and Nb bonds has N0 = dNs − Nb zero modes. Maxwell’s count is incomplete, though, because N0 can ex- ceed dNs − Nb if there are Nss states of self-stress, where springs can be under tension or compression with no net forces on the masses. This occurs, for example, when masses are connected by straight lines of bonds under periodic bound- ary conditions. A more general Maxwell relation21 , ν ≡ N0 − Nss = dNs − Nb , is valid for infinitesimal distortions. In a locally isostatic system with periodic boundary con- ditions, N0 = Nss . The square and kagome lattices have one state of self-stress per straight line of bonds and associ- ated zero modes along lines in momentum space. Cutting a √ √ section of N sites from these lattices removes states of self- N ) bonds and necessarily leads to O( stress and O( N ) zero modes, which are essentially identical to the bulk zero modes. Recently Sun et al.22 studied a twisted kagome lattice in which states of self-stress are removed by rotating adjacent site-sharing triangles in opposite directions. This simple mod- ification converts lines of zero modes in the untwisted lattice to gapped modes of nonzero frequency (except for q = 0) and localizes the required zero modes in the cut lattice to its surfaces. These boundary zero modes are robust and insensitive to local perturbations. Boundary modes also occur in electronic (1) systems, such as the quantum Hall effect23,24 and topologi- cal insulators25–30 . In this paper we establish the connection between these two phenomena. Our analysis allows us to pre- dict the existence of new topologically distinct bulk mechani- cal phases and to characterize the protected modes that occur on their boundary. We introduce a 1D model that illustrates this phenomenon in its simplest form and maps directly to the Su-Schrieffer-Heeger (SSH) model31 for the electronic exci- tations of polyacetylene ((CH2 )n ), a linear polymer with al- ternating single and double bonds between carbon atoms as show in figure 1 that has toplogically protected electron states at free ends and at interfaces. We then prove an index theo- rem that generalizes equation (1) and relates the local count of zero modes on the boundary to the topological structure of the bulk. We introduce a deformed version of the kagome lattice model that exhibits distinct topological phases. The predic- tions of an index theorem are verified explicitly by solving for the boundary modes in this model. Finally, we show that some of the distinctive features of the topological phases can be understood within a continuum elastic theory. Mechanical Modes and Topological Band Theory A mechanical system of masses M connected by springs with spring constant K is characterized by its equilibrium matrix21 Q, which relates forces Fi = QimTm to spring ten- sions Tm . i labels the dNs force components on the Ns sites, and m labels the Nb bonds. Equivalently, em = QT miui relates bond extensions em to site displacements ui . The squared nor- mal mode frequencies ω2 n are eigenvalues of the dynamical matrix D = QQT , where we set K/M to unity. Displace- ments ui that do not lead to stretched bonds satisfy QT ui = 0 and define the null space of QT , or equivalently its kernel ker QT . The dimension of this null space N0 = dim ker QT gives the number of independent zero modes. Similarly, the null space of Q gives the Nss = dim ker Q states of self- stress. The global counts of these two kinds of zero modes are related by the rank-nullity theorem21 , which may be ex- pressed as an index theorem32 . The index of Q, defined as ν = dim ker QT − dim ker Q, is equal to the difference be- tween the number of rows and columns of Q, and is given by equation (1). At first sight, the mechanical problem and the quantum electronic problem appear different. Newton’s laws are second-order equations in time, while the Schrodinger equa- tion is first order. The eigenvalues of D are positive defi- nite, while an electronic Hamiltonian has positive and nega- tive eigenvalues for the conduction and valence bands. To un- cover the connection between the two problems we draw our inspiration from Dirac, who famously took the “square root” of the Klein Gordon equation33 . To take the square root of the dynamical matrix, note that D = QQT has a supersymmet- ric partner D = QT Q with the same non-zero eigenvalues. Combining D and D gives a matrix that is a perfect square, (cid:18) 0 Q (cid:19) (cid:18) QQT (cid:19) 0 QT 0 0 QT Q ; H2 = H = . (2) The spectrum of H is identical to that of D , except for the zero modes. Unlike D , the zero modes of H include both the zero modes of QT and Q, which are eigenstates of τ z = diag(1dNs , −1Nb ) distinguished by their eigenvalue, ±1. (While the concept of supersymmetry is not essential for deriving this result, there is an interesting connection between the analysis leading to equation (2) and the theory of super- symmetric quantum mechanics34,35 .) Viewed as a Hamiltonian, H can be analyzed in the framework of topological band theory29 . It has an intrin- sic “particle-hole” symmetry, {H, τ z } = 0, that guarantees eigenstates come in ±E pairs. Since Qim is real, H = H∗ has “time-reversal” symmetry. These define symmetry class BDI36 . In one-dimension, gapped Hamiltonians in this class, such as the SSH model, are characterized by an integer topo- logical invariant n ∈ Z that is a property of the Bloch Hamil- tonian H(k) (or equivalently Q(k)) defined at each wavenum- ber k in the Brillouin zone (BZ). A mapping of H(k) to the complex plane is provided by det Q(k) = det Q(k)eiφ(k) . If bulk modes are all gapped then det Q(k) is nonzero and Q(k) ∈ GLp , where p is the dimension of Q(k). Q(k) is then classified by the integer winding number of φ(k) around the BZ: (φ(k + G) − φ(k) = 2πn, where G is a recipro- cal lattice vector), which defines an element of the homotopy group π1 (GLp ) = Z. A consequence is that a domain wall across which n changes is associated with topologically pro- tected zero modes31,37,38 . Below, we present an index theorem that unifies this bulk-boundary correspondence with equation (1) and shows how it can be applied to d-dimensional lattices, which form the analog of weak topological insulators28 . Topological edge modes have been previously predicted in 2D photonic39,40 and phononic41 systems. These differ from the present theory because they occur in systems with bandgaps at finite frequency and broken time-reversal sym- metry (symmetry class A). Localized end modes were found in a time-reversal invariant 1D model (class AI)42 . However, the presence of those finite frequency modes is not topologi- cally guaranteed. 2 FIG. 1: One-dimensional SSH and isostatic lattice models. (a) and (b) depict the SSH model of polyacetalene, with A and B sublat- tices indicated in blue and red. (a) describes the gapless state with all bonds identical, while (b) describes the gapped AB dimerized state, with double (single) bonds on the AB (BA) bonds. The BA dimer- ized state with single and double bonds interchanged is not shown. (c) and (d) show the 1D isostatic lattice model in which masses, rep- resented the larger blue dots are connected by springs in red and are constrained to rotate about fixed pivot points represented by small black dots. (c) is the gapless high-symmetry state with ¯θ = 0, and (d) is the gapped lower-symmetry phase with ¯θ > 0. (c) and (d) are equivalent to (a) and (b) if we identify the masses (springs) with the A (B) sublattice sites. (e) shows a domain wall in polyacetalene connecting the AB and BA dimerized states. There is a topologi- cally protected zero-energy state associated with the A sublattice at the defect. (f) shows the equivalent state for the isostatic model with a topologically protected zero-frequency mode at the domain wall. (g) shows domain wall connecting the BA and AB dimerized states, which has a zero energy state associated with the B sublattice. (h) shows the equivalent isostatic state with a state of self-stress at the domain wall. One-Dimensional Model Before discussing the index theorem we introduce a simple 1D elastic model, equivalent to the SSH model, that illustrates the topological modes in their simplest setting. Consider a 1D system of springs connecting masses constrained to rotate at a radius R about fixed pivot points. In Fig. 1c the spring lengths are set so that the equilibrium configuration is (cid:104)θi (cid:105) = 0. Fig. 1d shows a configuration with shorter springs with (cid:104)θi (cid:105) = (cid:112) ¯θ . Expanding in deviations δθi about ¯θ , the extension of spring mi = q1 ( ¯θ)δm,i + q2 ( ¯θ)δm,i+1 m is δ(cid:96)m = QT mi δθi , with QT and q1(2) = r cos ¯θ(r sin ¯θ±1)/ 4r2 cos2 ¯θ + 1. The normal mode dispersion is ω(k) = Q(k), where Q(k) = q1 + q2 eik . When ¯θ = 0, q1 = −q2 , and there are gapless bulk modes near k = 0. For a finite system with N sites and N − 1 springs, there are no states of self stress and only a single ex- tended zero mode, as required by equation (1). For ¯θ (cid:54)= 0 the bulk spectrum has a gap. In this case, the zero mode re- quired by equation (1) is localized at one end or the other, depending on the sign of ¯θ . The ¯θ > 0 and ¯θ < 0 phases δθ1δθ3δθ2δθ4θcdabfhfmfmfmssθ = + θcθ = - θcθ = - θcθ = + θcl1l2l3egare topologically distinct in the sense that it is impossible to tune between the two phases without passing through a tran- sition where the gap vanishes. The topological distinction is captured by the winding number of the phase of Q(k), which is 1 (0) for q1 < (>)q2 . Viewed as a quantum Hamiltonian, equation (2) for this to the SSH model31 , as indicated in model is identical Fig. 1(a,b). The sites and the bonds correspond, respectively, to the A and B sublattices of the SSH model. For ¯θ = 0 the bonds in the SSH model are the same (Fig. 1a), while for ¯θ (cid:54)= 0 they are dimerized (Fig. 1b). The two topological phases correspond to the two dimerization patterns for polyac- etalene. As is well known for the SSH model31,37 , an interface between the two dimerizations binds a zero mode, as indicated in Fig. 1(e,g). This is most easily seen for ¯θ = ±θc where the springs are colinear with the bars, so that q1 or q2 = 0. Fig. 1f shows a domain wall between +θc and −θc , in which the cen- ter two sites share a localized floppy mode. Fig. 1h shows an interface between −θc and +θc with a state of self-stress lo- calized to the middle three bonds, in addition to floppy modes localized at either end. As long as there is a bulk gap, the zero modes cannot disappear when ¯θ deviates from ±θc . The zero modes remain exponentially localized, with a localiza- tion length that diverges when ¯θ → 0. Index Theorem There appear to be two distinct origins for zero modes. In equation (1) they arise because of a mismatch between the number of sites and bonds, while at a domain wall they arise in a location where there is no local mismatch. To unify them, we generalize the index theorem so that it determines the zero- mode count ν S in a subsystem S of a larger system. This is well defined provided the boundary of S is deep in a gapped phase where zero modes are absent. We will show there are two contributions, (3) L + ν S ν S = ν S T , T is a L is a local count of sites and bonds in S and ν S where ν S topological count, which depends on the topological structure of the gapped phases on the boundary of S . To prove equation (3) and to derive formulas for ν S T and ν S L , we adapt a local version of the index theorem originally intro- duced by Callias43–46 to allow for the possibility of non-zero L . The details of the proof are given in the supplementary ν S material. Here we will focus on the results. Consider a d- dimensional system described by a Hamiltonian matrix Hαβ , where α labels a site or a bond centered on rα . The count of (cid:20) (cid:21) zero modes in S may be written τ z ρS (r) i H + i where rαβ = δαβ rα . The region S is defined by the support function ρS (r) = 1 for r ∈ S and 0 otherwise. It is useful to consider ρS (r) to vary smoothly between 1 and 0 on the boundary ∂S . Expanding the trace in terms of eigenstates of H shows that only zero modes with support in S contribute. ν S = lim →0 (4) Tr , 3 (5) (6) ni = 1 2π 1 2π i RT · n, In the supplementary material we show that equation (4) L = Tr (cid:2)ρS (r)τ z (cid:3) can be rewritten as equation (3) with ν S (cid:90) and dd−1S ν S T = pointing normal n. RT = (cid:80) Vcell ∂S where the integral is over the boundary of S with inward i niai is a Bravais lattice vector characterizing the periodic crystal in the boundary region that (cid:73) (cid:73) can be written in terms of primitive vectors ai and integers dk·∇kφ(k). dk·Tr[Q(k)−1∇kQ(k)]. = Ci Ci (7) Here Ci is a cycle of the BZ connecting k and k + bi , where bi is a primitive reciprocal vector satisfying ai · bj = 2πδij . ni are winding numbers of the phase of det Q(k) around the cycles of the BZ, where Q(k) is the equilibrium matrix in a Bloch basis. To apply equations (6) and (7), it is important that the wind- ing number be independent of path. This is the case if there is a gap in the spectrum. We will also apply this when the gap vanishes for acoustic modes at k = 0. This is okay because the acoustic mode is not topological in the sense that it can be gapped by a weak translational-symmetry-breaking pertur- bation. This means the winding number is independent of k even near k = 0. It is possible, however, that there can be topologically protected gapless points. These would be point zeros around which the phase of det Q(k) advances by 2π . These lead to topologically protected bulk modes that form the analog of a “Dirac semimetal” in electronic systems like graphene. These singularities could be of interest, but they do not occur in the model we study below. A second caveat for equation (7) is that, in general, the winding number is not gauge invariant and depends on how the sites and bonds are assigned to unit cells. In the supple- mentary material we show that for an isostatic lattice with cell that satisfy r0 = d (cid:80) i di − (cid:80) uniform coordination it is possible to adopt a “standard unit fined using Bloch basis states k, a = i, m(cid:105) ∝ (cid:80) cell” with basis vectors di(m) for the ns sites (dns bonds) per m dm = 0. Q(k) is de- R exp ik · (R + da )R + da (cid:105), where R is a Bravais lattice vector. In this gauge, RT is uniquely defined and the zero-mode count is given by equations (3) and (5)-(7). To determine the number of zero modes per unit cell on an edge indexed by a reciprocal lattice vector G, consider a cylinder with axis parallel to G and define the region S to be the points nearest to one end of the cylinder (See Supplemen- tary Fig. 1). ν S T is determined by evaluating equation (6), with the aid of equation (7) on ∂S deep in the bulk of the cylinder. It follows that νT ≡ ν S T /Ncell = G · RT /2π . The local count, ν S L , depends on the details of the termination at the surface and can be determined by evaluating the macro- scopic “surface charge” that arises when charges +d (−1) are (8) 4 dp−1 relative to the midpoint of the line along ap that con- nects its neighbors at dp+1 ± ap∓1 (with p defined mod 3), as indicated in Fig. 2a. sp are specified by 6 parameters with 2 constraints. A symmetrical representation is to take sp = xp (ap−1 − ap+1 ) + ypap and to use independent vari- ables (x1 , x2 , x3 ; z ) with z = y1 + y2 + y3 . The constraints then determine yp = z/3 + xp−1 − xp+1 . xp describes the buckling of the line of bonds along ap , so that when xp = 0 the line of bonds is straight. z describes the asymmetry in the sizes of the two triangles. (0, 0, 0; 0) is the undistorted √ kagome lattice, while (x, x, x; 0) is the twisted kagome lat- tice, studied in22 , with twist angle θ = tan−1 (2 3x). It is straightforward to solve for the bulk normal modes of the periodic lattice. When any of the xp are zero the gap van- ishes on the line k · ap = 0 in the BZ. This line of zero modes is a special property of this model that follows from the pres- ence of straight lines of bonds along ap . When xp = 0 the system is at a critical point separating topologically distinct bulk phases. The phase diagram features the eight octants specified by the signs of x1,2,3 . (+ + +) and (− − −) de- scribe states, topologically equivalent to the twisted kagome lattice, with RT = 0. The remaining 6 octants are states that are topologically distinct, but are related to each other by C6 3(cid:88) rotations. We find p=1 ap sgnxp/2 RT = (11) is independent of z . Fig. 2b shows a ternary plot of the phase diagram as a function of x1 , x2 , x3 for z = 0 and a fixed value of x1 + x2 + x3 . Fig. 2c,d,e show representative structures for the RT = 0 phase (Fig. 2c), the RT (cid:54)= 0 phase (Fig. 2e), and the transition between them (Fig. 2d). The insets show density plots of the lowest frequency mode, which highlight the gapless point due to the acoustic mode in Fig. 2c and the gapless line due to states of self stress in Fig. 2d. In Fig. 2e, the gap vanishes only at the origin, but the cross arises because acoustic modes vary quadratically rather than linearly with k along its axes. This behavior will be discussed in the next section. We next examine the boundary modes of the deformed kagome lattice. Fig. 3 shows a system with periodic boundary conditions in both directions that has domain walls separating (.1, .1, .1; 0) from (.1, .1, −.1; 0). Since there are no broken bonds, the local count is ν L S = 0. On the two domain walls, equation (8) predicts T ) = +1(−1), T − R2 νT = G · (R1 for the left (right) domain wall, where R1 T and R2 T charac- terize the material to the left and right of the domain wall, respectively (See fig. 1 in the supplementary material). Note that the indices of the two domain walls are opposite in sign so that the total index ν of equation (1) is zero as required. Fig. 2c shows the spectrum of H (which has both positive and negative eigenvalues) as a function of the momentum kx par- allel to the domain wall. The zero modes of H include both the floppy modes and the states of self-stress. In the vicinity of kx = 0 the zero modes on the two domain walls couple (12) FIG. 2: Deformed kagome lattice model. a shows our convention for labeling the states. b is a ternary plot of the phase diagram for fixed x1 + x2 + x3 > 0. The phases are labeled by R, which is zero in the central phase and a nearest neighbor lattice vector in the other phases. c, d and e show representative structures for R = 0 (c) and R (cid:54)= 0 (e) and the transition between them (d). The insets are density plots of the smallest mode frequency as a function of k in the BZ. In c the gap vanishes only at k = 0, while in d it vanishes on the line kx = 0. In e the gap vanishes only at k = 0, but has a quadratic dependence in some directions for small k. placed on the sites (bonds) in a manner analogous to the “peb- ble game”4 . This can be found by defining a bulk unit cell with basis vectors da that accommodates the surface with no leftover sites or bonds (see Fig. 4a below). Note that this unit cell depends on the surface termination and, in general, will T . The lo- be different from the “standard” unit cell used for ν S cal count is then the surface polarization charge given by the dipole moment per unit cell. We find νL ≡ ν S L /Ncell = G · RL /2π , (cid:88) di − (cid:88) sites i bonds m RL is similar to r0 defined above (which is assumed to be zero), but it is in general a different Bravais lattice vector. The total zero mode count on the surface then follows from equa- tions (3), (8), and (9). RL = d where (9) dm . (10) Deformed Kagome Lattice Model We now illustrate the topological boundary modes of a two- dimensional lattice with the connectivity of the kagome lat- tice, but with deformed positions. The deformed kagome lat- tice is specified by its Bravais lattice and basis vectors for the three atoms per unit cell. For simplicity, we fix the Bra- vais lattice to be hexagonal with primitive vectors ap+1 = (cos 2πp/3, sin 2πp/3). We parameterize the basis vectors as d1 = a1/2 + s2 , d2 = a2 /2 − s1 and d3 = a3 /2. Defining s3 = −s1 − s2 , sp describe the displacement of s2s3s1a2a3a1d1d2d3x1x3x2abc (.1,.1,.1;0)d (0,.1,.1;0)e (-.1,.1,.1;0)x1x2x30cdeR=Continuum Elasticity Theory 5 Unlike electronic spectra, phonon spectra have acoustic modes whose frequencies vanish as k → 0. These exci- tations along with macroscopic elastic distortions and long- wavelength surface Rayleigh waves are described by a contin- uum elastic energy quadratic in the elastic strain tensor uij . Elastic energies are restricted to small wavenumber and can- not by themselves determine topological characteristics, like those we are considering, that depend on properties across the BZ. Nevertheless the elastic energies of our model isostatic lattices fall into distinct classes depending on the topological class of the lattice. For simplicity we focus on (x1 , x2 , x2 ; 0) states, where x2 > 0 is fixed and x1 is allowed to vary. The elastic energy density f can be written [(uxx − a1uyy )2 + 2a4u2 K (13) f = xy ]. 2 We find that a1 ∝ x1 for small x1 , while a4 > 0 and K are positive and smoothly varying near x1 = 0. Thus, the RT = 0 and RT (cid:54)= 0 sectors are distinguished by the sign of a1 . f = 0 for shape distortions with uxx = a1uyy and uxy = 0. When a1 > 0, the distortion has a negative Poisson ratio47 , expanding or contracting in orthogonal directions (a feature shared by the twisted kagome lattice22 ); when a1 < 0, the distortion has the more usual positive Poisson ratio. Finally when a1 = 0, uniaxial compressions along y costs no energy. Expanding det QT for small k provides useful information about the bulk- and surface-mode structure. To order k3 , x − 3kxk2 y )] + O(k4 ), (14) x + a1k2 det QT = A[k2 y + ic(k3 where A, c > 0 for small x1 . a1 is the same as in equation (13). Long-wavelength zero modes are solutions of det QT = 0. The quadratic term, which corresponds to the elastic the- ory, equation (13), reveals an important difference between though, to order k2 , det QT = 0 for kx = ±(cid:112)a1 ky , so the bulk acoustic modes of RT = 0 and RT (cid:54)= 0. In the for- mer case, a1 > 0, det QT = 0 only at k = 0. For a1 < 0, the elastic theory predicts lines of gapless bulk modes. The degeneracy is lifted by the k3 term, leading to a k2 dispersion along those lines, which can be seen by the cross in the density map of Fig. 2e. det QT (k → 0) vanishes for complex wavenumbers asso- ciated with zero-frequency Rayleigh surface waves. Writing k = k⊥ n + k z × n for a surface whose outward normal n makes an angle θ with x, there is an ω = 0 Rayleigh wave with penetration depth Im k⊥ −1 if Im k⊥ < 0. To order k2 there are two solutions, √ sin θ ± i a1 cos θ √ cos θ ∓ i a1 sin θ √ i(3 + a1 )d a1 sin θ)3 k2 . 2(cos θ ± i (15) When a1 > 0, the linear term is always finite and nonzero, and Im k± ⊥ have opposite signs, indicating that there can be acoustic surface zero modes on all surfaces. These are the classical Rayleigh waves predicted by the elastic theory, with penetration depth O(k−1 ). When a1 < 0, the linear term in k± ⊥ = k + FIG. 3: Zero modes at domain walls. a shows a lattice with peri- odic boundary conditions and two domain walls, the left one between (.1, .1, .1, 0) and (.1, .1, −.1, 0) with zero modes and the right one between (.1, .1, −.1, 0) and (.1, .1, +.1, 0) with states of self stress. The zero mode eigenvectors at kx = π are indicated for the floppy mode (arrows) and the state of self-stress (red (+) and green (-) thick- ened bonds). b shows the vibrational mode dispersion as a function of kx . FIG. 4: Zero modes at the edge. a shows a (−.05, .05, .05, 0) lat- tice indicating three edges. b, c and d show the vibrational mode spectrum computed for a strip with one edge as shown in a and the other edge with a clamped boundary condtion. The zero mode count on each surface is compared with equations (3,8,9). and split because their penetration depth diverges as kx → 0. The eigenvectors for the zero modes at kx = π are indicated in Fig. 3a by the arrows and the thickened bonds. Fig. 4a shows a segment of a (−.05, .05, .05; 0) lattice with three different different edges. For each edge, a unit cell that accommodates the edge is shown, along with the correspond- ing RL , from which νL is determined. In the interior, a “stan- dard” unit cell, with r0 = 0 is shown. Figs. 4b, c, d show the spectrum for a strip with one edge given by the correspond- ing edge in Fig. 4a with free boundary conditions. The other edge of the strip is terminated with clamped boundary condi- tions, so that the floppy modes are due solely to the free edge. The number of zero modes per unit cell agrees with equations (8) and (9) for each surface given RL , RT . The zero modes acquire a finite frequency when the penetration length of the zero mode approaches the strip width, which leads to Gaus- sian “bumps” near k = 0, which will be discussed in the next section. In Fig. 4d, one of the three zero modes can be iden- tified as a localized “rattler”, which remains localized on the surface sites, even for k → 0. RT=0RT=0RT=00.10.20-ππ-0.1-0.2abxkxω00.10.20-ππ00.10.20-ππ00.10.20-ππr0=0RL(b)RL(d)RL(c)(a)(b)(c)(d)RTυT = -2υL = +2υT = 0υL = +1υT = +1υL = +2~~~~~~⊥ ∝ k2 . The number of long wavelength k is real and Im k± θ < θc = cot−1 (cid:112)a1 , Im k± surface zero modes depends on the angle of the surface. When ⊥ are both positive, and there are no acoustic surface zero modes. The opposite surface, θ − π < θc , has two acoustic surface modes. For θc < θ < π − θc Im k± ⊥ have opposite sign, so there is one mode. This is consistent with the mode structure in Fig. 4: The O(k−2 ) penetration depth explains the Gaussian profile of the k → 0 bumps in the zero modes, which are due to the finite strip width. In (b) a θ = 0 surface has no zero modes. (c) shows a θ = π/2 > θc surface with one long-wavelength surface (d) shows the spectrum with π − θ = π/6 < zero mode. θc with two bumps indicating two deeply penetrating long- wavelength zero modes in addition to one non-acoustic mode localized at that surface. Conclusions We have developed a general theory of topological phases of isostatic lattices, which explains the boundary zero modes and connects to the topological band theory of electronic sys- 6 tems. This points to several directions for future studies. It will be interesting to study 3D lattice models, as well as lat- tices that support point singularities in det Q(k) analogous to Dirac semimetals. Finally, it will be interesting to explore connections with theories of frustrated magnetism48 . Correspondence and requests for materials should be sent to Tom Lubensky. Acknowledgments TCL is grateful for the hospitality of the Newton Institute, where some of this work was carried out. This work was sup- ported in part by a Simons Investigator award to CLK from the Simons Foundation and by the National Science Founda- tion under DMR-1104707 (TCL) and DMR-0906175 (CLK). Author contributions: CLK and TCL contributed to the formulation of the problem, theoretical calculations, and the preparation of the manuscript. Competing financial interests statement: The authors have no competing financial interests to declare. 1 Phillips, J. C. Topology of Covalent Non-Crystalline Solids 2. Medium-Range Order in Chalcogenide Alloys and α − Si((Ge). Journal of Non-Crystalline Solids 43, 37–77 (1981). 2 Thorpe, M. F. Continuous Deformations in Random Networks. Journal of Non-Crystalline Solids 57, 355–370 (1983). 3 Feng, S. & Sen, P. N. Percolation on Elastic Networks - New Exponent and Threshold. Physical Review Letters 52, 216–219 (1984). 4 Jacobs, D. J. & Thorpe, M. F. Generic Rigidity Percolation - the Pebble Game. Physical Review Letters 75, 4051–4054 (1995). 5 Liu, A. J. & Nagel, S. R. Nonlinear dynamics - Jamming is not just cool any more. Nature 396, 21–22 (1998). 6 Liu, A. J. & Nagel, S. R. Granular and jammed materials. Soft Matter 6, 2869–2870 (2010). 7 Liu, A. J. & Nagel, S. R. The Jamming Transition and the Marginally Jammed Solid. Annual Review of Condensed Matter Physics 1, 347–369 (2010). 8 Torquato, S. & Stillinger, F. H. Jammed hard-particle packings: From Kepler to Bernal and beyond. Reviews of Modern Physics 82, 2633–2672 (2010). 9 Wyart, M., Nagel, S. R. & Witten, T. A. Geometric origin of ex- cess low-frequency vibrational modes in weakly connected amor- phous solids. Europhysics Letters 72, 486–492 (2005). 10 Wyart, M. On the rigidity of amorphous solids. Annales De Physique 30, 1–96 (2005). 11 Wilhelm, J. & Frey, E. Elasticity of stiff polymer networks. Phys- ical Review Letters 91, 108103 (2003). 12 Heussinger, C. & Frey, E. Floppy modes and nonaffine defor- mations in random fiber networks. Physical Review Letters 97, 105501 (2006). 13 Huisman, L. & Lubensky, T. C. Internal stresses, normal modes and non-affinity in three-dimensional biopolymer networks. Phys- ical Review Letters 106, 088301 (2011). 14 Broedersz, C., Mao, X., Lubensky, T. C. & MacKintosh, F. C. Criticality and isostaticity in fiber networks. Nature Physics 7, 983–988 (2011). 15 Souslov, A., Liu, A. J. & Lubensky, T. C. Elasticity and response in nearly isostatic periodic lattices. Physical Review Letters 103, 205503 (2009). 16 Mao, X. M., Xu, N. & Lubensky, T. C. Soft Modes and Elas- ticity of Nearly Isostatic Lattices: Randomness and Dissipation. Physical Review Letters 104, 085504 (2010). 17 Mao, X. M. & Lubensky, T. C. Coherent potential approximation of random nearly isostatic kagome lattice. Physical Review E 83, 011111 (2011). 18 Mao, X. M., Stenull, O. & Lubensky, T. C. Elasticity of a fila- mentous kagome lattice. Physical Review E 87, 042602 (2013). 19 Kapko, V., Treacy, M. M. J., Thorpe, M. F. & Guest, S. D. On the collapse of locally isostatic networks. Proc. Royal Soc. A 465, 3517–3530 (2009). 20 Maxwell, J. C. On the Calculaton of the Equilibirum Stiffness of Frames. Philosophical Magazine 27, 294-299 (1865). 21 Calladine, C. R. Buckminster Fuller’s “Tensegrity” Structures and Clerk Maxwell’s Rules for the Construction of Stiff Frames. In- ternational Journal of Solids and Structures 14, 161–172 (1978). 22 Sun, K., Mao, X. & Lubensky, T. C. Surface phonons, elastic response, and conformal invariance in twisted kagome lattices. PNAS 109, 12369–12374 (2012). 23 Halperin, B. I. Quantized Hall Conductance, Current-Carrying Edge States, and the Existence of Extended States in a Two- Dimensional Disordered Potential. Physical Review B 25, 2185– 2190 (1982). 24 Haldane, F. D. M. Model for a Quantum Hall Effect With- out Landau Levels - Condensed Matter Realization of the Parity Anomaly. Physical Review Letters 61, 2015–2018 (1988). 25 Kane, C. L. & Mele, E. J. Z(2) topological order and the quantum spin Hall effect. Physical Review Letters 95, 146802 (2005). 26 Bernevig, B. A., Hughes, T. L. & Zhang, S.-C. Quantum Spin Hall Effect and Topological Phase Transition in HgTe Quantum Wells. Science 314, 1757–1761 (2006). 27 Moore, J. E. & Balents, L. Topological invariants of time-reversal- invariant band structures. Physical Review B 75, 121306 (2007). 28 Fu, L., Kane, C. L. & Mele, E. J. Topological Insulators in Three Dimensions. Physical Review Letters 98, 106803 (2007). 29 Hasan, M. Z. & Kane, C. L. Colloquium: Topological insulators. Reviews of Modern Physics 82, 3045–3067 (2010). 30 Qi, X.-L. & Zhang, S.-C. Topological insulators and supercon- ductors. Reviews of Modern Physics 83, 1057–1110 (2011). 31 Su, W. P., Schrieffer, J. R. & Heeger, A. J. Solitons in Polyacetal- ene. Physical Review Letters 42, 1698 (1979). 32 Nakahara, M. Geometry, Topology and Physics (Hilger, Bristol, 1990). 33 Dirac, P. A. M. The Quantum Theory of the Electron. Royal Society of London Proceedings Series A 117, 610–624 (1928). 34 Witten, E. Dynamical breaking of supersymmetry. Nuclear Physics B 188, 513 – 554 (1981). 35 Cooper, F., Khare, A. & Sukhatme, U. Supersymmetry and quan- tum mechanics. Physics Reports 251, 267 – 385 (1995). 36 Schnyder, A. P., Ryu, S., Furusaki, A. & Ludwig, A. W. W. Clas- sification of topological insulators and superconductors in three spatial dimensions. Physical Review B 78, 195125 (2008). 37 Jackiw, R. & Rebbi, C. Solitons with Fermion Number 1/2. Phys- ical Review D 13, 3398–3409 (1976). 38 Volovik, G. E. The Universe in a Helium Droplet (Clarenden, Oxford, 2003). 39 Haldane, F. D. M. & Raghu, S. Possible Realization of Di- rectional Optical Waveguides in Photonic Crystals with Broken Time-Reversal Symmetry. Physical Review Letters 100, 013904 (2008). 40 Wang, Z., Chong, Y. D., Joannopoulos, J. D. & Solja ci ´c, M. Reflection-Free One-Way Edge Modes in a Gyromagnetic Pho- tonic Crystal. Physical Review Letters 100, 013905 (2008). 41 Prodan, E. & Prodan, C. Topological Phonon Modes and Their Role in Dynamic Instability of Microtubules. Physical Review Letters 103, 248101 (2009). 42 Berg, N., Joel, K., Koolyk, M. & Prodan, E. Topological phonon modes in filamentary structures. Physical Review E 83, 021913 (2011). 43 Callias, C. Axial Anomalies and Index Theorems on Open Spaces. Communications in Mathematical Physics 62, 213–234 (1978). 44 Bott, R. & Seeley, R. Some Remarks on Paper of Callias. Com- munications in Mathematical Physics 62, 235–245 (1978). 45 Hirayama, M. & Torii, T. Fermion Fractionalization and Index Theorem. Progress of Theoretical Physics 68, 1354–1364 (1982). 46 Niemi, A. J. & Semenoff, G. W. Fermion Number Fractional- ization in Quantum Field Theory. Physics Reports 135, 99–193 (1986). 47 Lakes, R. Foam Structures with a Negative Poisson’s Ratio. Sci- ence 235, 1038–1040 (1987). 48 Lawler, M. J. Emergent gauge dynamics of highly frustrated mag- nets. New Journal of Physics 15, 043043 (2013). SUPPLEMENTARY ONLINE MATERIALS Proof of Index Theorem In this appendix we provide details of the proof of the index theorem discussed in the text. Our starting point is equation (4), which describes the zero-mode count in a region S of a larger system. Using the fact that {H, τ z } = [ρS (r), τ z ] = 0 7 Tr τ z . (16) (cid:21) (cid:20) it is straightforward to check that equations (3-5) imply that [ρS (r), H] 1 1 ν S H + i lim T = →0 2 Since [ρS , H] = 0 for ρS = 1, and H has a finite range a, T comes only from the boundary of region S where ρS (r) ν S varies. If we assume that the boundary region is gapped and that ρ(r) varies slowly on the scale L (cid:29) a, then we safely take to zero and expand to leading order in a/L. Since [ρS (r), H]αβ = Hαβ (ρS (rα ) − ρS (rβ )) ∼ Hαβ (rα − rβ ) · ∇ρ(rβ ), we may write Tr (cid:2)τ z∇ρ(r) · H−1 [r, H](cid:3) . 1 2 We next suppose that in the boundary region the lattice is periodic, so that the trace may be evaluated in a basis of plane (cid:88) waves: k, a(cid:105) = R exp ik · (R + da )R + da (cid:105), 1√ N ν S T = (17) (18) where R is a Bravais lattice vector in a system with periodic boundary conditions and N unit cells. da are basis vectors for the dns + nb sites and bonds per unit cell. The phases are chosen such that the position operator is r ∼ i∇k . In this basis, the Bloch Hamiltonian H(k) is a dns + nb square matrix with off diagonal blocks Q(k) and Q† (k), where Qab (k) = (cid:104)k, aQk, b(cid:105). (cid:90) T then has the form ν S ν S T = ∂S where the integral is over the boundary of S with inward nor- (cid:90) mal n, and BZ ddk (2π)d Im Tr[Q−1∇kQ]. dd−1S PT · n PT = (19) (21) (20) (22) where It is useful, to write ImTr[Q−1∇kQ] = ∇k Im log det Q. It is then straightforward to show that det Q(k + G) = det Q(k) exp[−iG · r0 ], (cid:88) di − (cid:88) r0 = d nation number of site i is zi then r0 = (cid:80) sites i bonds m For a general lattice, r0 is non zero. However, if the coordi- i (d − zi /2)di + R, where R is a Bravais lattice vector. Thus, for an isostatic lat- tice with uniform coordination z = 2d, r0 is a Bravais lattice vector, and it is always possible to shift dm by lattice vectors to make r0 = 0. In the text of the paper, we assumed r0 = 0. (23) (24) dm . 8 and Acell = Vcell /dcell is the projected area of the surface unit cell, which can be expressed in terms of the volume of the bulk unit cell Vcell and the distance dcell between Bragg planes. Referring to Supplementary Fig. 1a, we use equa- tion (25) to evaluate equation (20) away from the domain wall T − R2 T ) · n, where n is the unit to give νT = (A/Vcell )(R1 vector pointing to the right. The zero mode count per unit cell can be expressed in terms of the reciprocal lattice vector G = 2π n/dcell that indexes the domain wall as T − R2 T /Ncell = G · (R1 ν S T )/2π . (27) Application to zero modes at the edge (28) We next determine the number of zero modes localized on a surface (or edge in 2d) indexed by a reciprocal lattice vec- tor G. Consider a cylinder with axis perpendicular to G and define the region S to be the points nearest to one end of the cylinder, as shown in Supplementary Fig. 1b. A similar con- struction can be used to count the zero modes on a surface in d dimensions. T is determined by evaluating equation (6) deep in the bulk ν S of the cylinder where the lattice is periodic. From equation (21) we may write T /Ncell = G · (RT − r0 )/2π . ν S L , depends on the details of the termina- The local count, ν S tion at the surface and is given by the macroscopic “surface charge” that arises when positive charges +d are placed on the sites and negative charges −1 are placed on the bonds. As discussed in the text, it can be determined by evaluating the dipole moment of a unit cell with site and bond vectors da that is defined so that the surface can be accomodated with no left over sites or bonds. This unit cell is in general different T , and its dipole moment from the unit cell used to compute ν S is in general not quantized. However, since the difference is due to a redefinition of which bond is associated with which unit cell, the dipole moment differs from r0 by a Bravais lat- di − (cid:88) (cid:88) tice vector, sites i bonds m It follows that the local count may be written L /Ncell = G · (RL + r0 )/2π . ν S The total zero mode count on the edge is then ν S /Ncell = G · (RL + RT )/2π . dm − r0 . RL = d (29) (30) (31) It can be seen that the dependence on r0 , which depends on the arbitrary unit cell used to define ν S T cancels. FIG. 1: Evaluating the zero mode count. a Cylindrical geometry for evaluating the zero mode count for a domain wall between R1 T T , indicated by the dashed line. b Cylindrical geometry for and R2 evaluating the zero mode count for a surface indexed by reciprocal lattice vector G. The region S covers half the cylinder. The boundary ∂S is deep in the interior. b also shows our notation for the surface unit cell. Here we will keep it general, and show that while r0 affects T , its effect is canceled by a compensating term in ν S L . ν S For the general case, let us write det Q(k) = q0 (k) exp[−ik · r0 ], where q0 (k) = q0 (k + G) is periodic in the BZ. Equation (21) then involves two pieces: [−r0 + RT ] . 1 PT = Vcell may be written RT = (cid:80) Here RT is a Bravias lattice vector describing the winding numbers of the phase of q0 (k) around the cycles of the BZ. It (cid:90) i niai with dk · ∇k log q0 (k) 1 ni = 2π i Ci (25) (26) where as in the text, we assume that for a given cycle Ci of the BZ the winding number is path indpendent. Application to zero modes at a domain wall To determine the zero mode count at a domain wall be- tween topological states R1 T and R2 T , we consider a cylin- der perpendicular to the domain wall (or a similar construc- tion for d dimensions). We expect the zero mode count to be proportional to the “area” A (or length in 2D) of the domain wall. We will, therefore, be interested in the zero mode count per unit cell, ν S /Ncell , where Ncell = A/Acell , ∂SAcelldcelln^GSρS(r) = 0ρS(r) = 1S∂S∂Sn^n^RTRTGρS(r) = 1ρS(r) = 0ρS(r) = 0abVcell12
1505.01882	1	1505	2015-05-07T22:29:28	Formation of Bright Solitons from Wave Packets with Repulsive Nonlinearity	[ "cond-mat.mes-hall", "nlin.PS" ]	Formation of bright envelope solitons from wave packets with a repulsive nonlinearity was observed for the first time. The experiments used surface spin-wave packets in magnetic yttrium iron garnet (YIG) thin film strips. When the wave packets are narrow and have low power, they undergo self-broadening during the propagation. When the wave packets are relatively wide or their power is relatively high, they can experience self-narrowing or even evolve into bright solitons. The experimental results were reproduced by numerical simulations based on a modified nonlinear Schr\"odinger equation model.	cond-mat.mes-hall	cond-mat	Formation of Bright Solitons from Wave Packets with Repulsive Nonlinearity Zihui Wang,1 Mikhail Cherkasskii,2 Boris A. Kalinikos,2 Lincoln D. Carr,3 and Mingzhong Wu1* 1Department of Physics, Colorado State University, Fort Collins, Colorado 80523, USA 2St.Petersburg Electrotechnical University, 197376, St.Petersburg, Russia 3Department of Physics, Colorado School of Mines, Golden, Colorado 80401, USA Formation of bright envelope solitons from wave packets with a repulsive nonlinearity was observed for the first time. The experiments used surface spin-wave packets in magnetic yttrium iron garnet (YIG) thin film strips. When the wave packets are narrow and have low power, they undergo self-broadening during the propagation. When the wave packets are relatively wide or their power is relatively high, they can experience self-narrowing or even evolve into bright solitons. The experimental results were reproduced by numerical simulations based on a modified nonlinear Schrödinger equation model. 1 Solitons are a universal phenomenon in nature, appearing in systems as diverse as water, optical fibers, electromagnetic transmission lines, deoxyribonucleic acid, and ultra-cold quantum gases.1,2,3,4,5 The formation of solitons from large-amplitude waves can be described by paradigmatic nonlinear equations, one of which is the nonlinear Schrödinger equation (NLSE). In the terms of the NLSE model, two classes of envelope solitons, bright and dark, can be excited in nonlinear media. A bright envelope soliton is a localized excitation on the envelope of a large-amplitude carrier wave. It typically takes a hyperbolic secant shape and has a constant phase across its width.6 A dark envelope soliton is a dip or null in a large-amplitude wave background. When the dip goes to zero, one has a black soliton. When the amplitude at the dip is nonzero, one has a gray soliton. A dark soliton has a jump in phase at its center. For a black soliton, such a phase jump equals to π. For a gray soliton, the phase jump is between 0 and π. The envelope of a dark soliton can be described by a unique function.3 For a black soliton, this function is typically a hyperbolic tangent function. According to the NLSE model, the formation of a bright soliton from a large-amplitude wave packet is possible in systems with an attractive (or self-focusing) nonlinearity and is prohibited in systems with a repulsive (or defocusing) nonlinearity. The underlying physics is as follows. The attractive nonlinearity produces a pulse self-narrowing effect; at a certain power level the self-narrowing can balance the dispersion-induced pulse self-broadening and give rise to the formation of a bright envelope soliton. In contrast, in systems with a repulsive nonlinearity the nonlinearity induces self-broadening of the wave packet, just as the dispersion does, and thereby disables the formation of a bright soliton. Previous experiments show good agreements with these theoretical predictions: the formation of bright solitons from wave packets has been demonstrated in different systems with an attractive nonlinearity,3,7 while the self-broadening has been observed for wave packets in systems with a repulsive nonlinearity.8 This letter reports on the first observation of the formation of bright solitons from wave packets with a repulsive nonlinearity. The experiments made use of spin waves traveling along long and narrow magnetic yttrium iron garnet (Y3Fe5O12, YIG)9 thin film strips. The YIG strips were magnetized by static magnetic fields applied in their planes and perpendicular to their length directions. This film/field configuration supports the propagation of surface spin waves with a repulsive nonlinearity.10,11 To excite a spin wave packet in the YIG strip, a microstrip line transducer was placed on one end of the YIG strip and was fed with a microwave pulse. As the spin wave packet propagates along the YIG strip, it was measured by either a secondary microstrip line or a magneto-dynamic inductive probe located above the YIG strip. When the input microwave pulse is relatively narrow and has relatively low power, one 2 observes the broadening of the spin wave packet during its propagation. At certain large input pulse widths and high power levels, however, the spin wave packet undergoes self-narrowing and evolves into a bright envelope soliton. The formation of this soliton is contradictory to the prediction of the standard NLSE model, but was reproduced by numerical simulations with a modified NLSE model that took into account damping and saturable nonlinearity. Figure 1 shows representative data on the formation of bright solitons from surface spin-wave packets. Graph (a) shows the experimental configuration. The YIG film strip was cut from a 5.6-mm-thick (111) YIG wafer grown on a gadolinium gallium garnet substrate. The strip was 30 mm long and 2 mm wide. The magnetic field was set to 910 Oe. The input and output transducers were 50-mm-wide striplines and were 6.3 mm apart. The input microwave pulses had a carrier frequency of 4.51 GHz. Note that, in Fig. 1 and other figures as well as the discussions below, Pin denotes the nominal microwave pulse power applied to the input transducer, tin denotes the half-power width of the input microwave pulse, Pout is the power of the output signal, and tout represents the half-power width of the output pulse. In Fig. 1, graphs (b), (c), (d), (f), and (g) give the power profiles of the output signals measured with different FIG. 1. Propagation of spin-wave packets in a 2.0-mm-wide YIG strip. (a) Experimental setup. (b), (c), (d), (f), and (g) Envelopes of output signals obtained at different input pulse power levels (Pin) and widths (tin). (e) Phase (q) profile for the signal shown in (d). (h) Width of output pulse (tout) as a function of Pin. (i) Width of output pulse as a function of tin. 3 Pin andtin values, as indicated. The circles in (d) shows a fit to the hyperbolic secant squared function.1,3 Graph (e) shows the corresponding phase (q) profile of the signal shown in (d). Here, the profile shows the phase relative to a reference continuous wave whose frequency equals to the carrier frequency of the input microwave.6 Graph (h) shows the change of tout with Pin for a fixed tin, as indicated, while graph (i) shows the change of tout with tin for a fixed Pin, as indicated. The data in Fig. 1 show three important results. (1) The data in Figs. 1 (b)-(e) and (h) show the change of the output signal with the input power Pin. One can see that the output pulse is broader than the input pulse when Pin=13 mW, as shown in (b), and is significantly narrower when Pin>30 mW, as shown in (c), (d), and (h). This indicates that the spin-wave packet undergoes self-broadening at low power and self-narrowing at relatively high power. (2) The data in Figs. 1 (d), (f), (g), and (i) show the change of the output signal with the input pulse width tin. It is evident that the width of the output pulse increases with tin when tin<50 ns and then saturates to about 19.5 ns when tin>50 ns. These results indicate that the spin-wave packet experiences strong self-narrowing when it is relatively broad. (3) The pulses shown in (d) and (g) are indeed bright solitons. As shown representatively in (d) and (e), they have a hyperbolic secant shape and a constant phase profile at their centers, which are the two key signatures of bright solitons.1,6 The data from Fig. 1 clearly demonstrate the formation of bright solitons from surface spin-wave packets when the energy of the initial signals (the product of Pin and tin) is beyond a certain level. This result is contradictory to the predictions of the NLSE model. One possible argument is that the width of the YIG strip might play a role in the observed formation of bright solitons. To rule out this possibility, similar measurements were carried out with an YIG strip that is an order of magnitude narrower. The main data are as follows. Figure 2 gives the data measured with a 0.2-mm-wide YIG strip. This figure is shown in the same format as in Fig. 1. In contrast to the data in Fig. 1, the data here were measured by a 50-W inductive probe,12 rather than a secondary microstrip transducer. The distance between the input transducer and the inductive probe was about 2.6 mm. The magnetic field was set to 1120 Oe. The input microwave pulse had a carrier frequency of 5.07 GHz. The data in Fig. 2 show results very similar to those shown in Fig. 1. Specifically, the low-power, narrow spin- wave packets undergo self-broadening, as shown in (b), (c), (f), and (h); as the power and width are increased to certain levels, the spin-wave packets experience self-narrowing, as shown in (h) and (i), and can also evolve into solitons, as shown in (d), (e), and (g). Therefore, the data in Fig. 2 clearly confirm the results from Fig. 1. This indicates that the formation of solitons reported here is not due to any effects associated with the YIG strip width. Note that the solitons 4 FIG. 2. Propagation of spin-wave packets in a 0.2-mm-wide YIG strip. (a) Experimental setup. (b), (c), (d), (f), and (g) Envelopes of output signals obtained at different input pulse power levels (Pin) and widths (tin). (e) Phase (q) profile for the signal shown in (d). (h) Width of output pulse (tout) as a function of Pin. (i) Width of output pulse as a function of tin. shown in Fig. 2 are narrower than those shown in Fig. 1. This difference results mainly from the fact that the spin- wave amplitudes and dispersion properties were different in the two experiments. The spin-wave dispersion differed in the two experiments because the magnetic fields were different and the wave numbers of the excited spin-wave modes were also not the same. Turn now to the spatial formation of solitons from surface spin-wave packets. Figure 3 shows representative data. Graph (a) gives the profile of an input signal. The power and carrier frequency of the input signal were 700 mW and 5.07 GHz, respectively. Graphs (b)-(f) give the corresponding output signals measured with the same experimental configuration as depicted in Fig. 2(a). The signals were measured by placing the inductive probe at different distances (x) from the input transducer, as indicated. The red curves in (b)-(f) are the corresponding phase profiles. The data in Fig. 3 show the spatial evolution of a spin-wave packet. At x=1.1 mm, the packet has a width similar to that of the input pulse. As the packet propagates to x=2.1 mm, it develops into a soliton, which is not only much 5 ) W ( r e w o P ) W m ( r e w o P ) W m ( r e w o P 0.8 0.6 0.4 0.2 0.0 800 0.04 0.03 0.02 0.01 0.00 800 0.02 0.01 0.00 800 (a) Input pulse 850 900 950 1000 (c) x=2.1 mm e s a h P º 0 8 1 850 900 950 1000 (e) x=3.6 mm º 0 8 1 e s a h P 850 900 Time (ns) 950 1000 0.06 0.04 0.02 ) W m ( r e w o P 0.00 800 0.02 0.01 0.00 800 0.02 0.01 0.00 800 ) W m ( r e w o P ) W m ( r e w o P (b) x=1.1 mm e s a h P º 0 8 1 850 900 950 1000 (d) x=2.6 mm e s a h P º 0 8 1 850 900 950 1000 (f) x=4.6 mm º 0 8 1 e s a h P 850 900 Time (ns) 950 1000 FIG. 3. Spatial formation of a spin-wave soliton in a 0.2-mm-wide YIG strip. (a) Profile of an input signal. (b)-(f) Profiles of output signals measured by an inductive probe placed at different distances (x) from the input transducer. The red curves in (b) and (f) are the corresponding phase profiles. narrower than both the initial pulse and the packet at x=1.1 mm but also has a constant phase at its center portion, as shown in (c). At x=2.6 mm, the packet has a lower amplitude due to the magnetic damping but still maintains its solitonic nature, as shown in (d). As the packet continues to propagate further, it loses its solitonic properties and undergoes self-broadening, as shown in (e) and (f), due to significant reduction in amplitude. Note that the phase profiles for all the signals in (b), (e), and (f) are not constant. These results support the above-drawn conclusion, namely, that it is possible to produce a bright soliton from a surface spin-wave packet. The data in Fig. 3 also indicate the other two important results. (1) The development of a soliton takes a certain distance, about 2 mm for the above-cited conditions, due to the fact that the nonlinearity effect needs a certain propagation distance to develop. (2) The soliton exists only in a relatively short range, about 1-2 mm for the above- cited conditions, due to the damping of carrier spin waves. To increase the "life" distance or lifetime of a spin-wave soliton, one can take advantage of parametric pumping13 or active feedback9 techniques. As mentioned above, the soliton formation presented here is contradictory to the standard NLSE model. However, it can be reproduced by numerical simulations based on the equation 6 i u ¶ t ¶ Ø OE º + v g u ¶ x ¶ u h + - ø oe ß 1 2 D u 2 2 ¶ x ¶ ( + 2 N u + S u 4 ) u = 0 (1) where u is the amplitude of a spin-wave packet, x and t are spatial and temporal coordinates, respectively, vg is the group velocity, h is the damping coefficient, D is the dispersion coefficient, and N and S are the cubic and quintic nonlinearity coefficients, respectively. The quantic nonlinearity term is included because the cubic nonlinearity is insufficient to capture the experimental observations presented above. This additional term is an expansion to the lowest order of saturable nonlinearity. The simulations used the split-step method to solve the derivative terms with respect to x and used the Runge-Kutta method to solve the equation with the rest of the terms.14,15 A high-order Gaussian profile was taken in simulations for the input pulse because it is much closer to the experimental situation than a squared pulse. The use of a square pulse as in the input pulse gave rise to numerical noise due to the discontinuity at the pulse's edges. The use of a fundamental Gaussian function did not onsiderably change the simulation results. It should be noted that both the standard and modified NLSE models are for nonlinear waves in one-dimensional (1D) systems, and previous work had demonstrated the feasibility of using the 1D NLSE models to describe nonlinear spin waves in quasi-1D YIG film strips.16,17 Figure 4 shows representative results obtained for different initial pulse amplitudes (u0), as indicated. In each panel, the left and right diagrams show the power and phase profiles, respectively. The simulations were carried out for a 20-mm-long 1D film strip and a total propagation time of 250 ns. The film strip was split into 9182 steps, and the temporal evolution step was set to 0.05 ns. The input pulse was a high-order Gaussian profile with an order number of 20 and a half-power width of 15 ns. The other parameters used are as follows: vg=3.8×106 cm/s, h=3.1×106 rad/s, D=-4.7×103 rad(cid:215)cm2/s, N=-10.1×109 rad/s, and S=1.8×1012 rad/s. Among these parameters, vg, D, h, and N were calculated according to the properties of the YIG film,9 and the S was optimized for the reproduction of the experimental responses. The profiles in Fig. 4 indicate that, at low initial power, the pulse is broader than the initial pulse and has a phase profile which is not constant at the pulse center, as shown in (a) and (b); at relatively high power, however, the pulse is not only significantly narrower than the initial pulse but also has a constant phase across its center portion, as shown in (c). These results agree with the experimental results presented above. The reproduction of the experimental responses with the modified NLSE model indicates the underlying physical 7 (a) Initial pulse amplitude u0=0.0005 180 ) e e r g e d ( e s a h P 90 0 -90 -180 160 180 Time (ns) 200 150 160 170 180 190 200 210 Time (ns) 150 160 170 180 190 200 210 150 160 170 180 190 200 210 Time (ns) Time (ns) (b) Initial pulse amplitude u0=0.005 180 ) e e r g e d ( e s a h P 90 0 -90 -180 (c) Initial pulse amplitude u0=0.071 180 ) e e r g e d ( e s a h P 90 0 -90 -180 ) . . u a ( r e w o P ) . . u a ( r e w o P ) . u . a ( r e w o P 1.0 0.8 0.6 0.4 0.2 0.0 1.0 0.8 0.6 0.4 0.2 0.0 1.0 0.8 0.6 0.4 0.2 0.0 FIG. 4. Power (left) and phase (right) profiles of spin-wave packets propagating in a YIG strip. The profiles were obtained from simulations with different initial pulse amplitudes, as indicated, for a propagation distance of 4.9 mm. 150 160 170 180 190 200 210 150 160 170 180 190 200 210 Time (ns) Time (ns) processes for the formation of bright solitons from surface spin-wave packets. In comparison with the standard NLSE, the additional terms in the modified equation are uh and 4 S u u . The term uh accounts for the damping of spin waves in YIG films, while the term 4 S u u is needed for the reproduction of the experimental responses. Since the sign of S was opposite to that of N, the term 4 S u u played a role opposite to saturation. In particular, for the configuration cited for Fig. 4(c) the term 2N u u and caused nonlinearity 2N u u , S u u overwhelmed the term 4 resulting in a repulsive-to-attractive nonlinearity transition and the formation of a bright soliton. Thus, one can see that the saturable nonlinearity played a critical role in the formation of the bright solitons from surface spin-wave packets. It should be noted that the saturable nonlinearity has been known as a critical factor for the formation of solitons in optical fibers.18 In summary, this letter reports the first observation of the formation of bright solitons from surface spin-wave packets propagating in YIG thin films. The formation of such solitons was observed in YIG film strips with significantly different widths. The spatial evolution of the solitons was measured by placing an inductive probe at different positions along the YIG strip. The experimental observation was reproduced by numerical simulations based 8 on a modified NLSE model. The agreement between the experimental and numerical results indicates that the saturable nonlinearity played important roles in the soliton formation. This work was supported in part by U. S. National Science Foundation (DMR-0906489 and ECCS-1231598) and the Russian Foundation for Basic Research. *Corresponding author. E-mail: [email protected] 9 1 M. J. Ablowitz and H. Segur, Solitons and the Inverse Scattering Transform (SIAM, Philadelphia, 1985). 2 A. Hasegawa and Y. Kodama, Solitons in Optical Communications (Oxford, New York, 1995). 3 M. Remoissenet, Waves Called Solitons: Concepts and Experiments (Springer-Verlag, Berlin, 1999). 4 Y. S. Kivshar and G. P. Agrawal, Optical Solitons: From Fibers to Photonic Crystals (Academic, New York, 2003). 5 P. G. Kevrekidis, D. J. Frantzeskakis, and R. Carretero-González, Emergent Nonlinear Phenomena in Bose-Einstein Condensates (Springer-Verlag, Berlin, 2008). 6 J. M. Nash, P. Kabos, R. Staudinger, and C. E. Patton, J. Appl. Phys. 83, 2689 (1998). 7 M. Chen, M. A. Tsankov, J. M. Nash, and C. E. Patton, Phys. Rev. B 49, 12773 (1994). 8 M. Chen, M. A. Tsankov, J. M. Nash, and C. E. Patton, Phys. Rev. Lett. 70, 1707 (1993). 9 Mingzhong Wu, “Nonlinear Spin Waves in Magnetic Film Feedback Rings,” in Solid State Physics Vol. 62, Edited by Robert Camley and Robert Stamps (Academic Press, Burlington, 2011), pp. 163-224. 10P. Kabos and V. S. Stalmachov, Magnetostatic Waves and Their Applications (Chapman and Hall, London, 1994). 11D. D. Stancil and A. Prabhakar, Spin Waves – Theory and Applications (Springer, New York, 2009). 12 M. Wu, M. A. Kraemer, M. M. Scott, C. E. Patton, and B. A. Kalinikos, Phys. Rev. B 70, 054402 (2004). 13 A. V. Bagada, G. A. Melkov, A. A. Serga, and A. N. Slavin, Phys. Rev. Lett. 79, 2137 (1997). 14 J. A. C. Weideman and B. M. Herbst, SIAM Journal on Numerical Analysis 23, 485 (1986). 15 D. Pathria and J. L. Morris, J. Comp. Phys. 87, 108 (1990). 16 H. Y. Zhang, P. Kabos, H. Xia, R. A. Staudinger, P. A. Kolodin, and C. E. Patton, J. Appl. Phys. 84, 3776 (1998). 17 Z. Wang, A. Hagerstrom, J. Q. Anderson, W. Tong, M. Wu, L. D. Carr, R. Eykholt, and B. Kalinikos, Phys. Rev. Lett. 107, 114102 (2011). 18 S. Gatz and J. Herrmann, J. Opt. Soc. Am. B 8, 2296 (1991). 10
1209.5406	1	1209	2012-09-24T20:10:30	Insulating Behavior of an Amorphous Graphene Membrane	[ "cond-mat.mes-hall" ]	We investigate the charge transport properties of planar amorphous graphene that is fully topologically disordered, in the form of sp2 three-fold coordinated networks consisting of hexagonal rings, but also including many pentagons and heptagons distributed in a random fashion. Using the Kubo transport methodology and the Lanczos method, the density of states, mean free paths and semiclassical conductivities of such amorphous graphene membranes are computed. Despite a large increase in the density of states close to the charge neutrality point, all electronic properties are dramatically degraded, evidencing an Anderson insulating state caused by topological disorder alone. These results are supported by Landauer-Buttiker conductance calculations, which show a localization length as short as 5 nanometers	cond-mat.mes-hall	cond-mat	Insulating Behavior of an Amorphous Graphene Membrane Dinh Van Tuan,1 Avishek Kumar,2 Stephan Roche,1,3 Frank Ortmann,1 M. F. Thorpe,2 Pablo Ordejon4 1CIN2 (ICN-CSIC) and Universitat Aut´onoma de Barcelona, Catalan Institute of Nanotechnology, Campus UAB, 08193 Bellaterra, Spain 2Department of Physics and Astronomy, Arizona State University, Tempe, AZ 85287, USA 3ICREA, Instituci´o Catalana de Recerca i Estudis Avanc¸ats, 08070 Barcelona, Spain 4Centre d'Investigaci´o en Nanoci`encia i Nanotecnologia - CIN2 (CSIC-ICN), Campus UAB, 08193 Bellaterra, Spain (Dated: May 14, 2018) We investigate the charge transport properties of planar amorphous graphene that is fully topologically dis- ordered, in the form of sp2 three-fold coordinated networks consisting of hexagonal rings, but also including many pentagons and heptagons distributed in a random fashion. Using the Kubo transport methodology and the Lanczos method, the density of states, mean free paths and semiclassical conductivities of such amorphous graphene membranes are computed. Despite a large increase in the density of states close to the charge neutral- ity point, all electronic properties are dramatically degraded, evidencing an Anderson insulating state caused by topological disorder alone. These results are supported by Landauer-Buttiker conductance calculations, which show a localization length as short as 5 nanometers. PACS numbers: 72.80.Vp, 73.63.-b, 73.22.Pr, 72.15.Lh, 61.48.Gh interest3. The physics of disordered graphene is at the heart of many fascinating properties such as Klein tunnel- ing, weak antilocalization or anomalous quantum Hall ef- fect (see reviews1,2). The precise understanding of in- dividual defects on electronic and transport properties of graphene is currently of great For instance, graphene samples obtained by large-scale production meth- ods display a huge quantity of structural imperfections and defects that jeopardize the robustness of the other- wise exceptionally high charge mobilities of their pristine counterparts4. Indeed, the lattice mismatch-induced strain between graphene and the underlying substrate generates polycrystalline graphene with grain boundaries that strongly impact transport properties5. However, despite the large amount of disorder, such graphene flakes (when deposited onto oxide substrates) usually maintain a finite conductiv- ity down to very low temperatures owing to electron-hole puddles-induced percolation effects that preclude localiza- tion phenomena close to the Dirac point6. The predicted An- derson localization in two-dimensional disordered graphene has been hard to measure in non intentionally damaged graphene, in contrast to chemically modified graphene7,8. Nevertheless, in a recent experiment it was possible to screen out electron-holes puddles using sandwiched graphene in be- tween two boron-nitride layers, together with an additional graphene control layer9. As a result of puddles screen- ing, a large increase of the resistivity was obtained at the Dirac point, evidencing an onset of the Anderson localiza- tion regime. Beyond individual defects and polycrystallinity, a higher level of disorder can be induced on graphene to the point of obtaining two-dimensional fully amorphous networks com- posed of sp2 hybridized carbon atoms. Such networks con- tain rings other than hexagons in a disordered arrangement. The average ring size is six according to Euler's theorem, allowing these systems to exist as flat 2D structures. Ex- perimentally, such amorphous two-dimensional lattices have been obtained in electron-beam irradiation experiments10,11, and directly visualized by high resolution electron trans- mission microscopy. Previously, indirect evidence for the formation of an amorphous network was obtained using Raman spectroscopy in samples subject to electron-beam irradiation12, ozone exposure13 and ion irradiation14. In all these cases, an evolution from polycrystalline to amor- phous structures was observed upon increase of the dam- In14, further evidence of the formation of age treatment. an amorphous network was obtained through transport mea- surements. These indicate the transition from a weak lo- calization regime in the polycrystalline samples to variable range hopping transport in the strongly localized regime for amorphous samples, as evidenced by the temperature depen- dence of the conductivity. Localization lengths were esti- mated to be of the range 0.1 to 10 nm in the amorphous samples, depending on the degree of amorphization. From the theoretical side, models of the amorphous network have been proposed using stochastic quenching methods15, and molecular dynamics16 -- 18. Electronic structure calculations show that the amorphization yields a large increase of the density of states in the close vicinity of the charge neutrality point15 -- 17. Despite the expected reduction of the conduc- tion properties due to strong localization effects, Holmstrom et al.16 suggest that disorder could enhance metallicity in amorphized samples, in contrast with the experimental evi- dence. In this Letter, we explore the transport properties of two- dimensional sp2 lattices with a massive amount of topolog- ical disorder, encoded in a geometrical mixture of hexagons with pentagon and heptagon rings. The calculations are done using two complementary approaches: a Kubo formulation in which the conductivity of bulk 2D amorphous graphene lattices is determined, and a Landauer-Buttiker formulation where the conductance of stripes of amorphous graphene contacted to semi-infinite pristine graphene electrodes is cal- culated. Both approaches lead to similar findings. Depend- ing on the ratio between odd versus even-membered rings, a transition form a graphene-like electronic structure to a to- tally amorphous and smooth electronic distribution of states is obtained. The stronger the departure from the pristine graphene, the more insulating is the corresponding lattice, which transforms into a strong Anderson insulator with elas- tic mean free paths below one nanometer and localization lengths below 10 nm close to the charge neutrality point. Those structures are therefore inefficient to carry any sizable current, and unsuitable for practical electronic applications such as touch screens displays or conducting electrodes, but interesting for scrutinizing localization phenomena in low dimensional materials. (a) (b) (c) L y W x (d) Pristine Graphene S1 S2 ] 2 - Å 1 − 0 γ [ ) Ε ( ρ 0.4 0.35 0.3 0.25 0.2 0.15 0.1 0.05 0 -3 -2 -1 0 E [γ0] 1 2 3 FIG. 1: (color online). (a), (b) Details of amorphous graphene sam- ples S1 and S2, respectively, used to compute the conductivity with the Kubo approach. (b) Scheme of setup for the Landauer-Butikker calculations. The grey area represents the amorphous graphene stripe. Periodic boundary conditions are used in the y direction. The graphene electrodes are semi-infinite in the x direction. (d) Total density of states of the two amorphous samples. The pristine crystalline graphene case (dashed lines) is also shown for compar- ison. Models of amorphous graphene.- Amorphous models of graphene are prepared using the Wooten-Winer-Weaire (WWW) method19,20, introducing Stone-Wales defects21 into the perfect honeycomb lattice. These networks can be considered as the amorphous versions of the Haeckelite structures proposed by Terrones et al.22. To generate the structures, periodic boundary conditions are imposed and the entire network is relaxed with the Keating-like potential15,23. A further relaxation of these models using forces from Den- sity Functional Theory leads to slight changes in the bonding distances and angles, but to virtually identical radial distribu- tions functions and, most importantly, electronic properties 2 TABLE I: Characteristic parameters for the two samples of amor- phous graphene, S1 and S2 number of atoms % of n-membered rings (n = 5/6/7) < n2 > − < n >2 RMS deviation of bond angles RMS deviation of bond lengths Fermi energy (in units of γ0) S1 10032 24/52/24 0.47 11.02◦ 0.044 A 0.03 S2 101640 44/12/44 0.88 18.09◦ 0.060 A 0.05 like the density of states, as shown in Ref. 17. Fig. 1(a)-(b) show pieces of the two studied sample S1 and S2, which re- spectively contain 10032 and 101640 atoms (all of them with three-fold coordination as the honeycomb lattice, but topo- logically distinct). The amorphous character of the samples is demonstrated by analyzing the radial distribution function, as shown in Ref. 15. Table I shows the parameters that char- acterize the two samples. For sample S1, 24% of the elemen- tary rings are pentagons, 52% hexagons and 24% heptagons, while sample S2 has a larger share of odd-membered rings. In both samples, the number of heptagons is the same as that of pentagons, as required by Euler's theorem, and these sys- tems can exist without an overall curvature as flat 2D struc- tures with some distortions of bond lengths and angles, al- though they may pucker under some circumstances. We will only be concerned with the planar structures here. Sample S2 is an extreme case, having very few hexagons and being furthest from the pristine honeycomb lattice. This is useful to accentuate the differences between crystalline and amor- phous samples, and to gain perspective. Nevertheless, it is likely that sample S1 is nearer to physical reality as it is less strained. For the calculation of the Landauer-Buttiker conductance, we set up models in which an amorphous stripe is contacted by two pristine graphene electrodes separated by a distance L, as shown in Fig. 1(c). We use models with increasing values of L, to explore the dependence of the conductance on the length of the amorphous contact in the transport di- rection. The models are periodic in the direction perpendic- ular to the stripe, with a periodicity of W =11.4 nm, and they have the same ring statistics as the bulk sample S1 described above. Electronic Properties.- The electronic and transport prop- erties of these disordered lattices are investigated using π-π* orthogonal tight-binding (TB) model with nearest neighbor hopping γ0 = 2.8 eV15 and zero onsite energies. No varia- tion of the hopping elements with disorder is included in the model as bond-length variation does not exceed a few per- cent (cf. Table I); all dependence on disorder stems from the ring statistics, which is expected to be the dominant effect24. Fig. 1(d) shows the density of states (DOS) of the two disor- dered samples, together with the pristine case (dashed line) for comparison. Sample 1, which keeps 52% of hexagonal rings, displays several noticeable features, similar to those 1 1 0.8 0.8 S1, E=-2γ0 S2, E=-2γ0 m m r r o o n n 0.6 0.6 D D ] m n [ ξ 100 80 60 40 20 0 -1 S1 S2 0 1 E [γ0] 2 3 0.4 0.4 0.2 0.2 0 0 S1, E=0 S2, E=0 0.2 0.2 0.4 0.4 t [ps] t [ps] 0.6 0.6 0.8 0.8 1 1 FIG. 2: (color online) Normalized time-dependent diffusion coeffi- cients for two selected energies for both samples S1 and S2. Inset: localization lengths as a function of the carrier energy found in previous studies15,16. First, the DOS at the charge neutrality point (Fermi level) is found to be increased by a large amount25. Additionally, the electron-hole symme- try of the band structure is broken by the presence of odd- membered rings which generate quasibound states at reso- nant energies26. The hole part of the spectrum is still remi- niscent of the graphene DOS, with a smoothened peak at the van Hove singularity, while for the electron part a second maximum appears close to the upper conduction band edge. By reducing further the ratio of even versus odd-membered rings (sample S2), the second maximum develops to a strong peak at about E = 2.5γ0 while spectral weight at E = 3γ0 is suppressed. The redistribution of DOS at the upper con- duction band edge is a signature of odd-membered rings and its strength with increasing number of such rings relates the statistical distribution of rings with the DOS features. Transport Methodology.- To explore quantum transport in these topologically disordered graphene samples, a real- space order-N quantum wavepacket evolution approach is employed to compute the Kubo-Greenwood conductivity27. Such method, pioneered in28, has been successfully applied to many different types of systems, and in particular it has allowed to scrutinize Anderson localization in oxygen func- tionalized graphene8. The zero-frequency conductivity for carriers at energy E is computed as σdc = e2ρ(E) lim t→∞ d dt ∆X 2(E, t) (1) where ρ(E) is the density of states and ∆X 2(E, t) is the mean quadratic displacement of the wave packet at energy E and time t: ∆X 2(E, t) = Tr(cid:2)δ(E − H) X(t) − X(0)2(cid:3) Tr[δ(E − H)] (2) A key quantity in the analysis of the transport properties is the diffusion coefficient: Dx(E, t) = d dt ∆X 2(EF , t), 3 which in the long time limit gives the conductivity through Eq. 1. Assuming an isotropic system in the x and y direc- tions, the 2D diffusion coefficient becomes D(t) = Dx(t) + Dy(t) = 2Dx(t). All information about multiple scattering effects is contained in the time-dependence of D(t). times, Numerically, whatever the initial wavepacket features, D(t) starts increasing ballistically at short then reaches a maximum value which depends on the disorder strength, and finally decays as a result of quantum inter- ferences, the strength of which will dictate either a weak or a strong Anderson localization regime. The semiclas- sical quantities (elastic mean free path ℓe(E) and semi- classical conductivity σsc) are derived from the maximum of D(t) as ℓe(E) = Dmax(E)/2v(E) and σsc(E) = 4 e2ρ(E)Dmax(E), respectively (with v(E) being the carrier velocity). 1 The conductance of the amorphous stripes contacted to pristine graphene electrodes is computed using the Landauer-Buttiker approach29: G(E) = G0T (E) = 2e2 h Tr(cid:2)t†t(cid:3) (3) where T (E) and t(E) are the transmission probability and transmission matrix, respectively, which can be computed from the Green's function G(E) in the contact region and the broadening Γ(E) of the states due to the interaction with the left and right electrodes. We calculate the conductance of the stripes, which are infinite and periodic in the direc- tion parallel to the interface with the graphene electrodes (y axis in Fig. 1(c)). Despite the very large periodicity of our models, we perform a thorough sampling of the ky-points in that direction30, to obtain the appropriate V-shaped conduc- tance of graphene in the thermodynamic limit. G is given per supercell of periodicity W =11.4 nm. Note that, with this geometry, conductivity and conductance are related though σ = L W G. Mean Free Path, Conductivity and Localization Effects.- Fig. 2 shows the time dependence of the normalized diffu- sion coefficient D(t)/Dmax for two chosen energies and for both samples S1 and S2. For energy E = −2γ0, the short- time ballistic regime is followed by the saturation of the dif- fusion coefficient typically after 0.1 ps (for both samples S1 and S2). From the saturation values, ℓe(E) and σsc are de- duced and reported in Fig. 3. A striking feature is the very low value of the mean free path ℓe, below 0.5 nm for the energy window around the Fermi level (where the DOS is considerably larger than that of pristine graphene). For neg- ative energies (holes) far from the charge neutrality point, a considerable increase of more than one order of magnitude in the mean free paths is observed. The increase occurs for smaller binding energies for sample S1 than for sample S2, in good correlation with the changes observed in the DOS (which, around the van Hove singularity, deviates from the pristine graphene one more strongly for sample S2). We observe in the inset of Fig. 3 that σsc shows a min- in agreement with the about 4e2/πh, imum value σmin sc ] m n [ e l 4 3.5 3 2.5 2 1.5 1 0.5 0 -3 ] 0 G [ c s σ 10 5 0 S1 S2 rescaled DOS -1 -2 S1 S2 2/π -2 -1 1 2 0 E [γ0] 0 E [γ 0] 1 2 3 FIG. 3: (color online) Elastic mean free path versus energy for the two samples. DOS of sample S1 is also shown for comparison, in rescaled unts. Inset: semiclassical conductivity of corresponding lattices. theoretical limit in the diffusive regime, already confirmed for other types of disorder31,32. However, in contrast to prior studies, conductivity values near the minimum are ob- tained over an energy range of several eV around the charge neutrality point. This indicates that transport is strongly degraded in the amorphous network compared to pristine graphene, for which the conductivity increases rapidly when the Fermi level is shifted away from the Dirac point. The charge mobility, µ(E) = σsc(E)/en(E), with n(E) be- ing the carrier density, is found to be about 10 cm2V−1s−1 for n = 1011 − 1012 cm−2, orders of magnitudes lower than those usually measured in graphene samples33. Such low conductivity and mobility values should be measured at room temperature, where the semiclassical approximation is expected to hold. The obtained short mean free paths and minimum (semi- classical) conductivities indicate a further marked contribu- tion of quantum interferences turning the system to a weak and strong insulating system with temperature drop. Inter- ference effects are evidenced by the time-dependent decay of the diffusion coefficient D(t)/Dmax as clearly seen in Fig. 2, although with large differences depending on the chosen energy. For E = −2γ0, such decay is weak but clearly more pronounced for the S2 sample (which is more disordered than S1). In sharp contrast, localization effects are much stronger at the charge neutrality point and develop from much shorter timescale (few nanoseconds). These dif- ferences will be reflected in the corresponding localization lengths. Based on the scaling theory of localization, an estimate of the localization length of electronic states is inferred from ξ(E) = ℓe(E) exp(πhσsc(E)/2e2)34. The results are shown in Fig. 2 (inset). The amorphous samples are con- firmed to be extremely poor conductors, with localization lengths as low as ξ ∼ 5 − 10 nm over a large energy window around the charge neutrality point. One also observes that ξ can vary by more than one order of magnitude depending on the disordered topology of the sample and rings statistics. 4 20 ] h / 2 e 2 [ G 10 ] 0 G [ > σ < 1.2 0.9 0.6 0.3 0 0 L = 1.6 nm L = 8.6 nm graphene 5 10 15 L(nm) 20 25 30 0 -0.6 -0.3 0.3 0.6 0 E[γ 0] FIG. 4: (color online) Landauer-Buttiker conductance (for W =11.4 nm) of two amorphous stripes contacted to graphene electrodes with L = 1.6 and 8.6 nm, respectively. The conductance of a pristine graphene contact with the same lateral size (11.4 nm) is shown for comparison. The inset shows the dependence of the con- ductivity on the stripe size L; symbols: calculated points; line: fit to σ(L) ∼ L W e−L/ξ. To further confirm the localization lengths estimated us- ing scaling theory, we compute explicitly the conductance of the amorphous graphene stripes contacted with pristine graphene electrodes, using the geometry shown in Fig. 1(c), as a function of the length of the amorphous contact L. Fig. 4 shows the conductance curves for two stripes with L = 1.6 and 8.6 nm, respectively, compared to that of a graphene contact with the same lateral size in the supercell (W = 11.4 nm). It is clear that the conductance of the amorphous samples is greatly reduced with respect to that of graphene, and that the reduction is more pronounced as the length of the amorphous contact increases. Also, while the conductance for the stripe with the smallest length is relatively smooth, it becomes more noisy with increasing L. This reflects the transition from a diffusive system, in which the length of the amorphous contact is longer than the mean free path but shorter than the localization length, to a strongly localized one in which the localization length is shorter than the length of the amorphous region. From the variation of the Landauer-Buttiker conductance with L, we can extract reliable values of the localization lengths, as in the Anderson regime the conductance should decay as G(L) ∼ e−L/ξ. The inset in Fig. 4 shows the value of the conductivity (obtained from the Landauer-Butikker conductance) for several stripes with L ranging from 1.6 to 15.3 nm, averaged over an energy window of 1.5γ0 around the Fermi energy. A fit of the results to σ(L) ∼ L W e−L/ξ yields ξ = 5.8 nm. This value is fully consistent with that obtained above using scaling theory for energies close to the Fermi level, and confirms that, in these amorphous struc- tures, strong localization effects should occur at low tem- peratures at distances of less than 10 nm. Our results are consistent with the experimental ones from transport mea- surements by Zhou et al.14, which show values in the range between 0.1 and 10 nm for samples amorphized by ion radi- ation. In conclusion, we have shown that topological disorder alone causes amorphous graphene to be a strong Anderson insulator. The increase of the density of states close to the charge neutrality point is associated with quantum interfer- ence which inhibits current flow at low temperatures. Very short mean free paths and localization lengths are predicted, in line with recent experimental evidence in graphene under heavy ion irradiation damage14. P.O. acknowledges support from Spanish MICINN (Grants FIS2009-12721-C04-01 and CSD2007-00050), M.F.T from the US National Science Foundation under grant DMR-0703973, and A.K. from the US Department of Edu- cation under grant GAANN P200A090123 and the ARCS Foundation. This work is supported by the European Com- munity through the Marie Curie Actions. 5 1 D.S.L. Abergel, V. Apalkov, J. Berashevich, K. Ziegler, T. Chakraborty, Adv. Phys. 59, 261 (2010). 2 S. Roche, N. Leconte, F. Ortmann, A. Lherbier, D. Soriano, J.-C. Charlier, Sol. Stat. Comm. 152, 1404 (2012). 3 A.V. Krasheninnikov, and F. Banhart, Nature Materials 6, 723 (2007); F. Banhart, J. Kotakoski and A.V. Krasheninnikov, ACS Nano 5, 26 (2011); O.V. Yazyev and S.G. Louie, Nature Materi- als 9, 806 (2010); O.V. Yazyev and S.G. Louie, Phys. Rev. B 81, 195420 (2010); E. Cockayne, G. M. Rutter, N.P. Guisinger, J.N. Crain, P.N. First, J.A. Stroscio, Phys. Rev. B 83, 195425 (2011). 4 K.S. Novoselov, Rev. Mod. Phys. 83, 837 (2011). 5 S. Bae, H. Kim, Y. Lee, X. Xu, J.-S. Park, Y. Zheng, J. Bal- akrishnan, T. Lei, H.R. Kim, Y.I. Song, Y.-J. Kim, K.S. Kim, B Ozyilmaz, J.-H. Ahn, B.H. Hong, S. Iijima, Nature Nanotech- nology 5, 574 (2010); Q. Yu, L.A. Jauregui, W. Wu, R. Colby, J. Tian, Z. Su, H. Cao, Z. Liu, D. Pandey, D. Wei, T.F. Chung, P. Peng, N.P. Guisinger, E.A. Stach, J. Bao, S.-S. Pei, Y.P. Chen, Nature Materials 10, 443 (2011). 6 S. Das Sarma, S. Adam, E.H. Hwang, E. Rossi, Rev. Mod. Phys. 83, 407 (2011). 7 A.B. Kaiser, C. G´omez-Navarro, R.S. Sundaram, M. Burghard, K. Kern, Nano Lett. 9, 1787 (2009); J. Moser, H. Tao, S. Roche, F. Alzina, C. M. Sotomayor Torres, A. Bachtold, Phys. Rev. B 81, 205445 (2010); S.H. Cheng, K. Zou, F. Okino, H.R. Gutier- rez, A. Gupta, N. Shen, P.C. Eklund, J.O. Sofo, and J. Zhu, Phys. Rev. B 81, 205435 (2010); W. Li, Y. He, L. Wang, G. Ding, Z.- Q. Zhang, R.W. Lortz, P. Sheng, N. Wang, Phys. Rev. B 84, 045431 (2011). 8 N. Leconte, J. Moser, P. Ordej´on, H. Tao, A. Lherbier, A. Bach- told, F. Alsina, C.M. Sotomayor-Torres, J.-C. Charlier, S. Roche, ACS Nano 4, 4033 (2010); A. Cresti A. Lopez-Benazilla, P. Or- dej´on, S. Roche, ACS Nano 5 (11), 9271 (2011); N. Leconte, A. Lherbier, F. Varchon, P. Ordej´on, S. Roche, J.-C. Charlier, Phys. Rev. B 84, 235420 (2011). 9 L. A. Ponomarenko, A. A. Zhukov, R. Jalil, S.V. Morozov, K.S. Novoselov, V.V. Cheianov, V.I. Fal'ko, K. Watanabe, T. Taniguchi, A.K. Geim, R.V. Gorbachev, Nature Physics 7, 958 (2011). 10 J. Kotakoski, A.V. Krasheninnikov, U. Kaiser, J.C. Meyer, Phys. Rev. Lett. 106, 105505 (2011). 11 J. C. Meyer, F. Eder, S. Kurasch, V. Skakalova, J. Kotakoski, H.J. Park, S. Roth, A. Chuvilin, S. Eyhusen, G. Benner, A.V. Krasheninnikov, U. Kaiser, Phys. Rev. Lett. 108, 196102 (2012). 12 D. Teweldebrhan and A. A. Balandin, Applied Phys. Lett. 94, 013101 (2009). 13 H. Tao, J. Moser, F. Alzina, Q. Wan, C.M. Sotomayor-Torres, J. Phys. Chem. C 115, 18257 (2011). 14 Y.-B. Zhou, Z.-M. Liao, Y.-F. Wang, G. S. Duesberg, J. Xu, Q. Fu, X.-S. Wu, D.-P. Yu, J. Chem. Phys. 133. 234703 (2010). 15 V. Kapko, D.A. Drabold, M.F. Thorpe, phys. stat. sol. B 247, 1197 (2010). 16 E. Holmstrom, J. Fransson, O. Eriksson, R. Lizarraga, B. Sanyal, M.I. Katsnelson, Phys. Rev. B 84, 205414 (2011). 17 Y. Li, F. Inam, A. Kumar, M. F. Thorpe, D. A. Drabold, phys. stat. sol. B, 248, 2082 (2011). 18 A. Kumar, M.F. Thorpe and M. Wilson (unpublished). 19 F. Wooten, K. Winer, D. Weaire, Phys. Rev. Lett. 54, 1392 (1985). 20 H. He, PhD thesis, Michigan State University (1985). 21 A.J. Stone and D.J. Wales, Chem. Phys. Lett. 128, 501 (1986). 22 H. Terrones, M. Terrones, E. Hernndez, N. Grobert, J-C. Char- lier, P.M. Ajayan, Phys. Rev. Lett. 84, 1716 (2000). 23 P. N. Keating, Phys. Rev. 145, 637 (1966). 24 Conductance calculations done using a simple 1/d2 model for the distance-depence of the hopping yield negligible differences with the results presented here. 25 An analysis of the character of the wavefunctions near the Fermi level can be found in the Supplementary Material at [URL by publisher]. 26 A. Lherbier, S M.-M. Dubois, X. Declerck, S. Roche, Y.M Ni- quet, J.C. Charlier, Phys. Rev. Lett. 106, 046803 (2011). 27 R Kubo, Rep. Prog. Phys. 29, 255 (1966). 28 S. Roche, and D. Mayou, Phys. Rev. Lett. 79, 2518 (1997); S. Roche, Phys. Rev. B 59, 2284 (1999); S. Roche and R. Saito, Phys. Rev. Lett. 87, 246803 (2001). 29 S. Datta, Electronic Transport in Mesoscopic Systems (Cam- bridge University Press, Cambridge 1995). 30 K. S. Thygesen and K. W. Jacobsen, Phys. Rev. B 72, 033401 (2005); F. D. Novaes, R. Rurali, P. Ordej´on, ACS Nano 4, 7596 (2010). 31 N.H. Shon, and T. Ando, J. Phys. Soc. Jpn. 67, 2421 (1998); N.M.R. Peres, F. Guinea, and A.H. Castro Neto, Phys. Rev. B 73, 125411 (2006); P.M. Ostrovsky, I.V. Gornyi, A.D. Mirlin, Phys. Rev. B 74, 235443 (2006). 32 K. Nomura, and A.H. MacDonald, Phys. Rev. Lett. 98, 076602 (2007); A. Lherbier, X. Blase, Y.-M. Niquet, F. Triozon, S. Roche, Phys. Rev. Lett. 101, 036808 (2008). 33 Y.-W. Tan, Y. Zhang, K. Bolotin, Y. Zhao, S. Adam, E.H. Hwang, S. Das Sarma, H.L. Stormer, P. Kim, Phys. Rev. Lett. 99, 246803 (2007). 34 P. A. Lee and T. V. Ramakrishnan, Rev. Mod. Phys. 57, 2 (1985).
1111.3566	1	1111	2011-11-15T16:13:13	Quantum Transport in Chemically-modified Two-Dimensional Graphene: From Minimal Conductivity to Anderson Localization	[ "cond-mat.mes-hall" ]	An efficient computational methodology is used to explore charge transport properties in chemically-modified (and randomly disordered) graphene-based materials. The Hamiltonians of various complex forms of graphene are constructed using tight-binding models enriched by first-principles calculations. These atomistic models are further implemented into a real-space order-N Kubo-Greenwood approach, giving access to the main transport length scales (mean free paths, localization lengths) as a function of defect density and charge carrier energy. An extensive investigation is performed for epoxide impurities with specific discussions on both the existence of a minimum semi-classical conductivity and a crossover between weak to strong localization regime. The 2D generalization of the Thouless relationship linking transport length scales is here illustrated based on a realistic disorder model.	cond-mat.mes-hall	cond-mat	Quantum Transport in Chemically-modified Two-Dimensional Graphene: From Minimal Conductivity to Anderson Localization 1 1 0 2 v o N 5 1 ] l l a h - s e m . t a m - d n o c [ 1 v 6 6 5 3 . 1 1 1 1 : v i X r a N. Leconte1, A. Lherbier1, F. Varchon1, P. Ordejon2, S. Roche3,4, J.-C. Charlier1 1Universit´e catholique de Louvain, Institut de la Mati`ere Condens´ee et des Nanosciences (IMCN), NAPS-ETSF, Chemin des Etoiles 8, B-1348 Louvain-la-Neuve, Belgium 2Centre de Investigaci´o en Nanoci`encia i Nanotecnologia, CIN2 (CSIC-ICN), Campus de la UAB, 08193 Bellaterra (Barcelona), Spain 3CIN2 (ICN-CSIC) and Universitat Autonoma de Barcelona, Catalan Institute of Nanotechnology, Campus de la UAB, 08193 Bellaterra (Barcelona), Spain 4ICREA, Instituci Catalana de Recerca i Estudis Avancats, 08010 Barcelona, Spain (Dated: September 21, 2018) An efficient computational methodology is used to explore charge transport properties in chemically-modified (and randomly disordered) graphene-based materials. The Hamiltonians of various complex forms of graphene are constructed using tight-binding models enriched by first-principles calculations. These atomistic models are further implemented into a real-space order-N Kubo-Greenwood approach, giving access to the main transport length scales (mean free paths, localization lengths) as a function of defect density and charge carrier energy. An extensive investigation is performed for epoxide impurities with specific discussions on both the existence of a minimum semi-classical conductivity and a crossover between weak to strong localization regime. The 2D generalization of the Thouless relationship linking transport length scales is here illustrated based on a realistic disorder model. PACS numbers: 73.63.-b, 72.15.Lh, 73.63.Fg, 63.22.-m I. INTRODUCTION Ever since graphene was experimentally synthesized in 20041, interest in its promising conduction properties has increased considerably2 -- 4. Owing to its two-dimensionality and reported large charge mobility, monolayer graphene has been initially envisioned as a genuine candidate to replace silicon in nano-electronics4 -- 6. But despite its realistic poten- tial in high-frequency device applications7,8, the absence of a substantial band-gap hinders its use for replacing silicon MOSFET devices in logic applications9. Various solutions have already been proposed to over- come this hurdle such as opening a wide band-gap, us- ing quantum confinement in ribbons10 -- 12 or using chemi- cal oxidation or hydrogenation to break the symmetry of the graphene lattice13 -- 22. Both of these methods have how- ever been demonstrated to be far too invasive9,23,24, generat- ing a large quantity of defects and damaging the otherwise Dirac-like properties of electronic excitations. Other more seducing proposals include the use of a laser field in the mid-infrared range which can induce tunable band gaps25, electric-field assisted gap opening in bilayers26 or chemical doping which in certain conditions allow to engineer con- trolled mobility gaps as large as 1 eV27,28. In all cases, the precise understanding of the impact of disorder on electronic and (charge, spin and phonon) trans- port properties of graphene appears of paramount impor- tance. Disorder in graphene exhibits many different fla- vors from structural defects to adsorbed impurities, recon- structed edges or long range Coulomb scatterers trapped in the graphene substrate (oxide layer). To date, the detailed re- lationship between microscopic complexity of disorder fea- tures and the onset of graphene unique transport properties remains elusive. This is particularly debated in relation with the so-called Klein tunneling mechanism29 and the weak anti-localization phenomenon which are both manifestations of pseudospin effects30 -- 34. Disorder first comes as a source of elastic scattering which limits the mean free path in a way which strongly depends on the disorder potential characteristics. The energy depen- dence of the mean free path and associated semi-classical transport quantities such as the Drude conductivity and the charge mobility can be indeed connected to the short or long range nature of the scattering potential35. Beyond the occur- rence of a diffusive regime, quantum interferences contribute significantly to the transport features at sufficiently low tem- peratures. In addition to the conventional weak localiza- tion phenomenon36,37, crossovers from weak localization to weak anti-localization have been predicted and experimen- tally observed30 -- 32,34,38 -- 44. Pseudospin-related quantum in- terferences are however maintained provided disorder does not break all underlying symmetries. This is not the case in presence of chemical defects which damage the sp2 lattice symmetry. Such stronger disturbances of graphene structure maximize localization effects, eventually turning the mate- rial to a two-dimensional insulator (Anderson localization). If Anderson localization has been highly debated and contro- versial for long-range disorder45 -- 52, its relevance for strongly damaged graphene is now well documented both theoreti- cally and experimentally53 -- 56. A recent theoretical study has however related the existence of a robust metallic state in presence of local magnetic ordering for partly hydrogenated graphene57,58 pinpointing possible subtleties between corre- lated impurity distribution and transport features. The main objective of this paper is to illustrate how an insulating regime can be tuned by intrusive functionaliza- tion of a graphene sheet caused by oxygen atoms bound in the epoxy position. Epoxide defects are for instance incor- porated on graphene after ozone treatment59. These epoxy impurities have a drastically different impact on resonant en- ergy peaks in the vicinity of the Dirac point when compared to single impurities60. To address this objective, the oxygen in epoxy position is studied by means of accurate ab initio techniques. This model allows us, on the one hand, to prove that the oxygen epoxy bonding lies in between a pure sp3-like covalent bond and an ionic bond, and on the other hand, to supply a suitable tight-binding (TB) model for further studies in very large scale systems. Using this TB model, the Kubo-Greenwood formalism is implemented in real space to obtain meaningful transport length scales and conduction properties. Several quantities such as the mean free path, the semi-classical con- ductivity and the localization length are analyzed in depth. The ongoing debate concerning long and short-range scat- tering behavior is also briefly commented in light of our re- sults. The crossover to the strongly localized regime is then investigated. Finally, conventional scaling laws are tested on our model in this localization regime. II. EPOXY DEFECTS V.V. Cheianov et al. [61] demonstrated the tendency of epoxy-bound adatoms to form spatially correlated states. The interaction between epoxy groups is mediated by the conduction electrons, similar to the Ruderman-Kittel- Kasuya-Yosida (RKKY) interaction which correlates mag- netic impurities62. These ordered states only exist for low impurity densities and disappear at a critical temperature Tc. For high concentrations of epoxy groups due to oxidation63 of graphene, J.L. Li et al. [64] argue that two epoxy groups attaching on the opposite ends of a carbon hexagon create more open rings inducing cracks along neighboring rings. Similarly, more recently, S. Fabris et al.[65] present a mech- anism giving rise to more complex crack propagation. How- ever, crack propagation has, up to our knowledge, only been reported under strong reduction and oxidation treatments. Furthermore, an aqueous environment seems to be manda- tory. H.J. Xiang et al. [66] also report on these unzipped chains caused by epoxy groups thus inducing lower energy conformations. Their study is however limited to concentra- tions above 25 % of epoxy density. The model investigated in the present paper takes ad- vantage of this literature while limiting its complexity to avoid any loss of generality for the simulations of differ- ent moderate concentrations (epoxy density ranging from 0.01 % to 5 %). Consequently, epoxy groups are assumed to be randomly distributed over the graphene sheet and 2 FIG. 1: (color online) An oxygen atom in epoxy position on a 5× 5 cell (a). 3D (b) and 2D (c) charge density difference as defined in the text. Charge accumulation in red/green and charge depletion in blue. Isovalues of 0.006 e/ A3 and −0.006 e/ A3 for the 3D charge density difference. the model prohibits the destructive presence of two oxygen atoms on the same hexagon. This simplified model could well describe the functionalization of graphene due to ozone treatment59 and comparison53 with experimental results67 backs this up. III. NUMERICAL TECHNIQUES The first part of this section presents ab initio calcula- tions performed to predict the structural properties of epoxy bound oxygen. Likewise, it handles how the TB parameters were extracted from these calculations. The second part of the section sets out the foundations of the Kubo-Greenwood formalism developed in a TB framework. A. From ab initio to tight-binding models The Density Functional Theory (DFT) calculations are conducted using the SIESTA code68 -- 70, within the local den- sity approximation (LDA) on the exchange-correlation func- tional in the Ceperley-Alder71 form parametrized by Perdew and Zunger72. Core electrons are included using Troullier- Martins73 pseudopotentials. Double ζ plus polarization or- bitals are used to define the basis set. Fig. 1(a) illustrates the oxygen atom in epoxy position af- ter ab initio geometry optimization. In order to investigate the chemical bonding of oxygen on graphene, the charge density difference is calculated. Plotting such charge den- sity difference allows for a visual understanding on how the electronic clouds associated with the orbitals are altered by chemical bonding. Fig. 1(b-c) illustrates this charge density 2callytotwocarbonatomstoformanepoxyaddendwhiletheremainingoxygenmoleculeremainschemicallyinert.Ex-perimentaldatawereproducedbythesamegrouptoconfirmtheirsimulations.V.V.Cheianovetal.[40]demonstratedthetendencyofepoxy-boundadatomstoformspatiallycor-relatedstates.Theinteractionbetweenepoxygroupsisme-diatedbytheconductionelectrons,similartotheRKKYin-teractionwhichcorrelatesmagneticimpurities41.Theseor-deredstatesonlyexistforlowimpuritydensitiesanddis-appearatacriticaltemperatureTc.Forhighconcentrationsofepoxygroupsduetooxidation42ofgraphene,J.L.Lietal.[43]assertthattwoepoxygroupsattachingontheoppo-siteendsofacarbonhexagoncreatemoreopenringsinduc-ingcracksalongneighboringrings.However,crackpropa-gationhas,uptoourknowledge,onlybeenreportedunderstrongreductionandoxidationtreatments.Furthermore,anaqueousenvironmentseemstobeaprerogative.H.J.Xi-angetal.[44]alsoreportontheseunzippedchainscausedbyepoxygroupsthusinducinglowerenergyconformations.Theirstudyislimitedhowevertoconcentrationsabove25%ofepoxydensity.Themodelinvestigatedinthepresentpapertakesadvan-tageofthisliteraturewhilelimitingitscomplexitytoavoidanylossofgeneralityforthesimulationsofdifferentmoder-ateconcentrations(epoxydensityrangingfrom0.01to5%).Consequently,epoxygroupsareassumedtoberandomlydistributedoverthegraphenesheettomimictheeffectofozonetreatmentwhileprohibitingthepresenceoftwooxy-genatomsonthesamehexagoncausinghighlydestructivechangeofthebaregraphenesheet'sstructure.Preliminaryresultscorroborateexperimentalmeasurements,thusvali-datingthissimplifiedapproach45.III.NUMERICALTECHNIQUESThefirstpartofthissectionpresentstheabinitiocalcula-tionsperformedtoobtainthestructuralpropertiesofepoxyboundoxygen.Likewise,ithandleshowtheTBparameterswereextractedfromthesecalculations.ThesecondpartofthesectionsetsoutthefoundationsoftheKubo-GreenwoodformalismdevelopedinaTBframework.A.Fromabinitiototight-bindingmodelsTheDFTcalculationsareconductedusingtheSIESTAcode46 -- 48,withinthelocaldensityapproximation(LDA)ontheexchange-correlationfunctionalintheCeperly-Alder49formparametrizedbyPerdewandZunger50.CoreelectronsareincludedusingTroullier-Martins51pseudopotentials.Figure1(a)illustratetheoxygenatominepoxypositionafterabinitiogeometryoptimization.Inordertoinvestigatethechemicalbondingofoxygenongraphene,thechargedensitydifferenceiscalculated.Plottingsuchchargeden-sitydifferenceallowsforavisualunderstandingonhowtheFIG.1:(coloronline)Typicaloxygenatominepoxypositionona5×5sheet(a).3D(b)and2D(c)chargedensitydifferenceasdefinedinthetext.Chargeaccumulationinred/greenandchargedepletioninblue.Isovaluesof0.006e/Å3and−0.006e/Å3forthe3Dinset.electroniccloudsassociatedwiththeorbitalsarealteredbytochemicalbinding.Fig.1(b-c)illustratesthischargeden-sitydifferenceρdiffdefinedasfollows:ρdiff=ρtot−ρO−ρgraph(1)whereρOandρgraphrepresentthechargedensitiesoffree-standingoxygenandgraphenerespectively.ρtotisthechargedensityofthetotalsysteminitsboundstate.InFig.1(b),thechargeaccumulationinredandthechargedepletioninbluedictateacovalentbondingbetweenthecarbonandtheoxygenatoms.Oxygenisknowntoexhibitanacceptorbe-havior(attractselectronsonthesandpyorbitals).Thepyorbitalstaysmainlyunaffectedbythebondingastheelec-troniccloudaroundtheoxygenkeepsitsconventionalp-orbitalform.InFig.1(c),theaccumulationofelectronsisrenderedinredandgreenandthedepletioninblue.Thetypicalπorbitalsofsp2graphenearebrokenbytheepoxyboundoxygen.Indeed,thebondingcauses(blue)chargedepletionclouds(π-orbitalgeometry)aboveandundertheusual(green)σbondsbetweencarbonatoms.Additionally,theelectronicchargetransfercalculatedwithaHirshfeld52,53(Voronoi53,54)integrationaddsupto−0.256e(−0.266e)ofchargetransfertowardstheoxygen,whichindicatesthebondingisalsoslightlyionic.Thesetwointegrationtech-niquesyieldbetterresultsthanthefrequentlyusedMullikenintegrationsincetheyarebasisindependant55.Inconclu-sion,theoxygenboundinepoxypositionisfoundtoperturbgraphene,butnotinasimplesp3-hybridizationwaylikeforhydrogenation.ACrystalorbitalHamiltonpopulation(COHP)study56,57isarefineddensityofstates(DOS)studywhichgivestheelectronicdensitiesassociatedwiththebondingbe-tweenspecificorbitals.Positivedensityvaluesrepresentthebondingstates,whilenegativevaluesdescribeanti-bondingstates.Figure2presentsaCOHPstudyontheorbitalspar-ticipatingtothebondingbetweentheoxygenatombound3 FIG. 2: (color online) Ab initio COHP analysis of the orbitals participating to the bonding of the oxygen atom in epoxy position with its neighboring carbon atoms. Region of interest between −1 eV and 1 eV with respect to the Fermi level. Positive and negative values indicate bonding and anti-bonding interactions, respectively. difference ρdiff defined as follows: ρdiff = ρtot − ρO − ρgraph (1) where ρ0 and ρgraph represent the charge densities of free- standing oxygen and graphene respectively. ρtot is the charge density of the total system in its bound state. In Fig. 1(b), the charge accumulation in red and the charge depletion in blue indicate a covalent bonding between the carbon and the oxygen atoms. Oxygen is known to exhibit an acceptor behavior (attracts electrons on the s and py or- bitals). Its py orbital does not participate in the bonding as the electronic cloud around the oxygen keeps its conven- tional p-orbital form. In Fig. 1(c) the accumulation of electrons is rendered in red and green and the depletion in blue. The typical π- orbitals of sp2 graphene are broken by the epoxy bonds. Indeed, this bonding causes a charge depletion (blue re- gion) in the π electron cloud located above and under the σ bond associated with the underlying carbon atoms. The latter σ bond thus encounters a slight charge accumulation (green region). Additionally, the electronic charge trans- fer calculated with a Hirshfeld74,75 (Voronoi75,76) integration adds up to −0.256 e (−0.266 e) of charge transfer to- wards the oxygen (red and green region close to the oxygen atom), which indicates that the bonding is also slightly ionic. These two integration techniques yield better results than the frequently used Mulliken integration since they are basis independent77. The oxygen bound in epoxy position is found to perturb graphene, but not in a purely sp3-hybridization way as in the hydrogenation case. A Crystal Orbital Hamilton Population (COHP) study78,79 partitions the band structure energy in terms of orbital pair contributions. Positive density values represent the bond- ing states, while negative values describe anti-bonding states when plotting the conventional -COHP. Fig. 2 presents a COHP study on the orbitals participating to the bonding between the oxygen atom in epoxy position and its neigh- boring carbon atom. In the following analysis, conclusions are drawn for the region of interest for transport properties -101Os-CsOs-CpxOs-CpyOs-Cpz-101E-EF (eV)Opx-CsOpx-CpxOpx-CpyOpx-Cpz-101Opz-CsOpz-CpxOpz-CpyOpz-Cpz-101Cs-CsCs-Cpx-COHP (arb. units)(c)(a)(b)-0.01-0.0050.050-0.050-0.1-0.054 FIG. 4: (color online) DFT STM image of empty states, ie. Local DOS for one oxygen atom in epoxy postion, integrated between 0 and 0.5 eV. Both sublattices are affected equally preserving the pseudospin symmetry in graphene. FIG. 3: (color online) Electronic band structure of a 5× 5 graphene supercell containing a single epoxy group. The DFT (red-solid) and TB (blue-dashed) band structures (a) are described along high- symmetry paths as described in the Brillouin zone (b). Nomencla- ture of extracted TB parameters is illustrated in (c). One possible orientation of the oxygen atom in epoxy position is described here- with. The two other inequivalent orientations are obtained by rotat- ◦ around the central ing the oxygen atom in epoxy position by 120 carbon atom in (d). lines in (a), (b) and (c)] interact mainly with the px, pz and s orbitals of oxygen. Finally, the py orbital of oxygen does not interact with carbon in this energy window. Consequently, a TB model with first-neighbor interactions including both px and hybridized s/pz orbitals of oxygen with the pz and s/px orbitals of carbon should be sufficient to accurately model the effect of epoxy groups on graphene. The combined contributions of the three orbitals of carbon binding with oxygen is renormalized to only one orbital in a π-like model. The matrix elements of the TB Hamiltonian are given by: [−1 eV; 1 eV] only; given bonds may have different bonding or antibonding behaviors away from the Fermi energy. One first notes that the contributions of the s (a) and pz (c) orbital of oxygen bonding with a neighboring carbon are analog. Both these orbitals have a dominant antibonding contribution with the px and s orbital of carbon at the right of the Fermi energy. The COHP study between both first neigh- boring carbon atoms [Fig. 2 (a), inset] shows that these two orbitals of carbon, also responsible for the σ bond hybridiza- tion, strongly bind at the same energies (E ∼ 0.5 eV and 1.2 eV), with analog contributions, typical for the σ-like hy- bridization between carbon atoms. These four orbitals (Opz, Os, Cpx and Cs) thus form a hybridized electronic cloud with bonding contributions between the two carbon atoms and antibonding contributions between oxygen and carbon. This result is in agreement with the charge density rearrange- ment observed in Fig. 1. The electronic charge depletion of the π orbitals observed in Fig. 1 can be rationalized with a COHP study for a larger energy window (not shown here). The π electrons are partly drawn into the σ bond between the carbon atoms. The remaining pz electrons of carbon [blue † i aj +(cid:88)i † ia i ai. (2) Hij =(cid:88)ij γija At first, a √3×√3 R 30◦ supercell with one epoxy atom was simulated using DFT to extract a band structure with limited folding of the Brillouin zone (not shown here). The bands near the Fermi energy are nicely fitted using the following TB parameters [nomenclature, see Fig. 3(c)]: x = −2.5 eV, z = −1.0 eV, 1 = 1.5 eV, γx = 1.8γ0, γz = −1.5γ0 and γ1 = 0.0 eV with γ0 = −2.6 eV. Finally, to confirm the validity of the model, these TB pa- rameters are used to generate the band structure of a 5 × 5 supercell containing a single epoxy oxygen, which is super- imposed with its DFT counterpart [see Fig. 3(a)]. The elec- tronic path chosen to plot the band structure contains all in- equivalent high-symmetry segments in the 2D Brillouin zone of a 5× 5 graphene supercell80[see Fig. 3(b)]. Note the band crossing at the Fermi energy is shifted away from the K2 point. The localized flat band (visible around the Γ point) appearing in the DFT band structure around -2.5 eV is miss- ing in the TB band structure. This originates from a strong interaction between the py orbital of oxygen with the py or- bital of carbon. Both these orbitals are missing in the TB model. These band structures were calculated for one ori- entation of the epoxy group on graphene[see Fig. 3(d)]. For (d)(c)the two other possible orientations of the epoxy oxygen the crossings occur close to K1 or K3 for symmetry reasons. In conclusion, our TB model seems to be sufficient to ac- curately model random positions and random orientations of impurities, as long as these epoxy oxygens do not inter- act with each other which is assumed to be satisfied for the range of concentrations of impurities considered here. Finally, in Fig. 4, the twofold D2h symmetry in the sim- ulated STM image obtained by integrating the local DOS (LDOS) proves the analogy with a double impurity defect, by comparison with the LDOS in Fig. 2 of Ref. [60]. The LDOS of empty states is spatially integrated between 0 and 0.5 eV. A similar pattern (not shown here) is obtained for hole carriers, by integration between and 0 and −0.5 eV. B. Kubo Formalism Transport properties for large mesoscopic-sized systems can be simulated efficiently using an order-N method based on the Kubo formalism82 -- 89. Assuming the electronic trans- port in the system is isotropic for the in plane x and y direc- tions, the 2D diffusion coefficient D(t) is obtained by D(t) = Dx(t) + Dy(t) = 2Dx(t) (3) Within this formalism, the diffusion coefficient Dx(t) in the transport direction x is calculated at each time step using where ∆X 2(E, t) = (4) (5) 2(cid:21) Dx(t) = ∆X 2(t) t Tr(cid:20)δ(E − H)(cid:12)(cid:12)(cid:12) X(t) − X(0)(cid:12)(cid:12)(cid:12) Tr(cid:104)δ(E − H)(cid:105) where X(t) is the position operator in Heisenberg represen- tation at time t: † X(t) = U (t) X(0) U (t) (6) and where U (t) = e−i Ht/¯h is the time-evolution operator. The trace, which is a sum over wavepackets initially lo- calized on each orbital of the system, is replaced by an ini- tial state with a random phase on each orbital of the sys- tem. Taking the average of ten initial random phase states already yields very satisfactory results on the smoothness of the curves. This greatly reduces computation time. U (t) can be expanded using Chebyshev polynomials to allow for the mandatory order-N method to achieve reasonable computa- tion time for systems containing millions of orbitals. Both the numerator and denominator in Eq. (5) are calculated us- ing the Lanczos recursion scheme thanks to continued frac- tions expansions. The termination term is the one usually 5 used for metals, which considers that the oscillation of the recursion coefficients is rapidly damped with the number of recursion steps. We checked that for the recursion step n = 500 the damping is sufficient, although a very small remnant oscillation caused by the small energy gap at high is observed. This is quantitatively correct at low energies and qualitatively sufficient at the border of the energy spec- trum where the energy gaps occur, by comparison with other more sophisticated termination methods. From the diffusion coefficient, the mean free path (cid:96)e(E) and the semi-classical conductivity σsc(E) can be calculated using respectively: and (cid:96)e(E) = Dmax(E) 2v(E) σsc(E) = 1 4 e2ρ(E)Dmax(E) (7) (8) where v(E) is the charge carrier velocity at energy E, Dmax the maximum value of D(t), e the electronic charge, and ρ(E) the DOS at energy E. The semi-classical Kubo- Greenwood conductivity σsc can be compared to the Drude approximation close to the Dirac point: σD(E) = 4e2 h k(E)(cid:96)e(E) 2 (9) where E = ¯hvFk with vF the Fermi velocity close to the Dirac point (dcc ≈ 1.42 A): 3γ0dcc 2¯h ≈ 1 × 106ms vF ≈ (10) −1 The limitations of the Drude approximation have recently been discussed and put into context in a Review paper [90] and the limitations of the Born approximation in the Boltz- mann theory of conductivity have been analyzed in Ref. [91]. IV. RESULTS To begin with, we discuss the effect of various oxygen concentrations on the density of states (DOS) in compari- son with pristine graphene. Then, the results obtained for the diffusion coefficient D(t) are analyzed. The other trans- port quantities are calculated within the Kubo formalism as introduced in the previous paragraph. Particular attention is given to the scaling behavior of the Kubo conductivity. All TB calculations were performed on systems containing 2560000 carbon atoms, which corresponds approximately to systems of 300 nm by 200 nm. A. DOS and energy shift The evolution of Tr(cid:104)δ(E − H)(cid:105), with increasing impurity density is re- ported on Fig. 5. Although DFT calculations on similar the DOS, calculated from 6 FIG. 5: (color online) Main frame: DOS for various impurity den- sities ranging from 0.05 to 4.42 %. The minimum of DOS shifts slightly with the addend concentration due to the changes in hop- ping parameters and on-site energies in the TB model. Side panels: δ-like peaks corresponding to impurity bands at lower (left) and higher (right) energies. FIG. 6: (color online) Main frame: Realignment of the minimum of DOS and charge neutrality point (CNP: position obtained by DOS integration) at 0 eV. Inset: linear increase of energy shift ∆n with increasing concentrations of epoxy groups (x in %). Rigid band theorem (see text) implies an impurity induced potential of ∼ −3.7 eV. concentrations exhibit a shift of the Fermi energy compared to the pristine case (not shown here), no shifts are applied yet to point out the similarity with other studies92,93 cov- ering the effect of ideal impurities (ie. the influence of the change of respectively the on-site energy and the hopping parameters) on the DOS of pristine graphene. A shift of the minimum of the DOS is observed with in- creasing impurity concentrations. Even though formally the rigid band theorem cannot be used for atoms that do not have the same valence (ie. carbon and oxygen)94, the shift is found to be linear with increasing concentrations (x) and the second order corrections O(2) can thus be neglected [see Fig. 6 (inset)]: ∆n = xU + O(2) + ... (11) In addition, where U is the local potential induced by the epoxy defect. A linear fit of ∆n versus x implies a value of U equal to ∼ −3.7 eV. the Van Hove Singularities (VHS) are smoothened out and decreased in amplitude with increasing concentrations of epoxide groups, in agreement with previ- ous observations82,92,93. Also, a small increase of the DOS appears at the minimum of the DOS with increasing impu- rity concentration. The bumps in densities of states on the left and on the right of the minimum of DOS (Fig. 5, middle panel) cor- respond to the resonant energies between the oxygen atoms in epoxy position and the graphene sheet60, while the δ-like peaks in the side panels at high energies correspond to flat impurity bands. As discussed in Section III A, the localized FIG. 7: (color online) Projected densities of states. The px (pz) orbital of oxygen mainly contributes on the left (right) side of the Fermi energy. Negligible contributions are predicted for first- nearest neighboring carbon atoms. Second carbon nearest neigh- bors do contribute to the resonant energy bump with oxygen. state close to the left VHS caused by the strong interaction between both py orbitals of oxygen and carbon is missing in this simplified TB model. Finally, oxygen in epoxy position triggers a shift of the Fermi energy which compensates the shift of the minimum of DOS discussed above (see Fig. 6). This Fermi energy is obtained by integrating the TB DOS and counting the number of electrons present in the system. In the next Sec- tions, transport calculations will implicitly include this re- alignment of minimum of DOS with the Fermi energy, thus locating the charge neutrality point (CNP) at E = 0 eV. In Fig. 7, a Projected Densities of States (PDOS) evidence -3-2-10123(eV)0.05%0.49%0.95%1.77%3.22%4.42%-9-800.20.40.60.81DoS(eV-1Ang-2)6789E-1-0.500.51E-EF(eV)00.050.10.150.2DoS(eV-1Ang-2)0.05%0.49%0.95%1.77%3.22%4.42%x(%)Δεn(eV)0-0.1-0.2024-1.5-1-0.500.511.5E-EF(eV)00.10.20.30.40.50.6DoS(eV-1Ang-2)O:px+pzO:pxO:pzC:farawayC:1stneighborC:2ndneighborO:px+pzO:pxO:pzC:farawayC:1stneighborC:2ndneighbor7 FIG. 8: (color online) Time evolution of diffusion coefficient (nor- malized) at E = 0.5 eV for different impurity concentrations. FIG. 9: (color online) Normalized diffusion coefficient versus time for x = 1.77 % at different energies. that the bump on the right side of the Fermi level originates from the pz orbital and the one on the left from the px or- bital of oxygen, in agreement with the previous COHP DFT study. Fig. 7 also indicates that the first-nearest neighboring carbon atoms do not contribute to the total DOS in contrast to the second nearest neighboring carbon atoms. The PDOS suggest the oxygen atom attracts most of the electronic den- sity and thus weakens the density on the first-neighboring carbon. Such analysis agrees with existing literature95,96. B. Diffusion coefficient and transport regimes The diffusion coefficient D(t) inherently contains all the information needed to calculate transport properties [see Eqs. (7) and (8)]. Its time evolution or dynamics also clar- ifies the dominant transport regime at the considered time scale (ie. ballistic, diffusive or localized). Fig. 8 illustrates the typical behaviors of the normalized diffusion coefficients Dnorm(t) = D(t) Dmax (12) for an energy E = 0.5 eV. As expected, the conduction in pure graphene is simply ballistic. The smallest simulated impurity concentration (ie. 0.01 %) does not reach its max- imum for the diffusion coefficient for the total elapsed time considered here (approx. 3800 fs). The slope of D(t) gives access to a v2(E) value still in close agreement with the the- oretical (analytical) value v2 F for energies E close to EF . For instance, at 0 eV and for 0.01 %, v(E) = 2.11 Aγ0/¯h, while vF = 2.13 Aγ0/¯h. For intermediate densities (0.05 and 0.1 %) Dnorm(t) evolve from the ballistic regime to the diffusive regime for the simulated times. At higher concen- trations (0.49 %) Dnorm(t) reach a diffusive regime sharply, followed by a clear decrease with time, signature of quantum interferences leading the charge carrier localization. Fig. 9 presents diffusion coefficients Dnorm(t) of 1.77 % of impurities at different energies. Localization effects fol- low an asymmetric behavior between electrons and holes. The predominant resonant peak at the right of the charge neutrality point (CNP) causes much stronger localization ef- fects than the weaker and more smeared out peak at the left of the CNP (in analogy with the PDOS in Fig. 7). Such asymmetry is of crucial importance to synthesize possible electronic devices made of functionalized graphene requir- ing an efficient switch between a conducting and an insu- lating behavior27,28. Localization effects are more signifi- cant around 0.75 eV. One finally notes the peculiar behavior of the diffusion coefficient exactly at the CNP. Numerically, this point is more problematic to simulate since the density of charge carriers may be very scarce. Nevertheless, our re- sults suggest that the saturation limit of the Diffusion coef- ficient is reached for longer simulation time and then turns into a moderate decrease characteristic of quantum effects. This discussion on the diffusion coefficient points out two different regimes which are approached separately in the remaining Sections. Firstly, the semi-classical quanti- ties, which neglect quantum effects, are discussed in Section IV C. Secondly, the localization regime is analyzed in Sec- tion IV D. C. Semi-classical regime 1. Analysis of elastic mean free paths Using Eq. (7), the mean free path (cid:96)e is calculated using the diffusion coefficient and plotted in Fig. 10 for energies between the two VHS at −2.6 eV and 2.6 eV. For the con- 6-1-0.500.51E - EF (eV)00.050.10.150.2DoS (eV-1 Ang-2)0.05 %0.49 %0.95 %1.77 %3.22 %4.42 %FIG.8:(coloronline)RealignmentofDiracpointandchargeneu-tralitypointat0eV.PositionofCNPobtainedbyintegrationofDOS.00.20.40.60.81Dnorm0.01 %0.05 %0.1 %0100020003000t (fs)00.20.40.60.81Dnorm0.49 %0.95 %1.77 %3.22 %4.42 %FIG.9:(coloronline)Timeevolutionofdiffusioncoefficient(nor-malized)atE=0.5eVfordifferentimpurityconcentrations.chargeneutralitypoint(CNP).B.DiffusiveandlocalizedregimesThediffusioncoefficientD(t)inherentlycontainsalmostalltheinformationneededtocalculatetransportproperties(seeEqs.(6)and(7)).Itstimeevolutionordynamicsalsoclarifiesthedominanttransportregimeattheconsid-eredtimescale(ie.ballistic,diffusiveorlocalized).Figure9illustratesthetypicalbehaviorsofthenormalizeddiffusioncoefficientsDnorm(t)=D(t)Dmax(11)whereDmaxisthemaximumofD(t)foranenergyE=0.5eV.Asexpected,puregrapheneissimplyballistic.The0100020003000t (fs)00.20.40.60.81Dnorm-1.5 eV-1.0 eV-0.5 eV0 eV0.5 eV0.75 eV1.0 eV1.5 eVElectron diffusionHole diffusionDiffusion at Dirac pointFIG.10:(coloronline)Normalizeddiffusioncoefficientversustimeforx=1.77%atdifferentenergies.Localizationeffectsaremoreimportantforthepositivechargecarriers.Maximumaround0.75eV.smallestsimulatedimpurityconcentration(ie.0.01%)doesnotreachitsmaximumforthediffusioncoefficientforthetotalelapsedtimeconsideredhere(≈3800fs).TheslopeofD(t)givesav2(E)valuestillincloseagreementwiththetheoretical(analytical)valuev2FforvaluesofEclosetoEF.Forinstance,at0eV,v(E)=1.57Åγ0/¯h,whilevF=2.13Åγ0/¯h.Forintermediatedensities(0.05and0.1%)Dnorm(t)evolvefromtheballisticregimetothedif-fusiveregimeforthesimulatedtimes.Athigherconcentra-tions(0.49%)Dnorm(t)reachadiffusiveregimesharply,furtherevolvingintoa1/√tbehavior,signatureofquantuminterferencescausinglocalization.Figure10presentsdiffusioncoefficientsDnorm(t)of1.77%ofimpuritiesatdifferentenergies.Localizationef-fectsfollowanasymmetricbehaviorbetweenelectronsandholes.Thepredominantresonantpeakattherightofthechargeneutralitypoint(CNP)causesmuchstrongerlocal-izationeffectsthantheweakerandmoresmearedoutpeakattheleftoftheCNP(inanalogywiththePDOSinFig.7).Suchasymmetryisofcrucialimportancetosynthesizepossibleelectronicdevicesmadeoffunctionalizedgraphenerequiringanefficientswitchbetweenaconductingandaninsulatingbehavior.Localizationeffectsarethemostseverearound0.75eV.OnefinallynotesthepeculiarbehaviorofthediffusioncoefficientexactlyattheDiracpoint.Numer-ically,thispointismoredifficulttosimulatesincetheden-sityofchargecarriersmaybeveryscarce.Nevertheless,ourresultssuggestthatthediffusiveplateauisreachedatlatertimeandthenturnsintoamoderatedecreasecharacteristicofquantumeffects.6-1-0.500.51E - EF (eV)00.050.10.150.2DoS (eV-1 Ang-2)0.05 %0.49 %0.95 %1.77 %3.22 %4.42 %FIG.8:(coloronline)RealignmentofDiracpointandchargeneu-tralitypointat0eV.PositionofCNPobtainedbyintegrationofDOS.00.20.40.60.81Dnorm0.01 %0.05 %0.1 %0100020003000t (fs)00.20.40.60.81Dnorm0.49 %0.95 %1.77 %3.22 %4.42 %FIG.9:(coloronline)Timeevolutionofdiffusioncoefficient(nor-malized)atE=0.5eVfordifferentimpurityconcentrations.chargeneutralitypoint(CNP).B.DiffusiveandlocalizedregimesThediffusioncoefficientD(t)inherentlycontainsalmostalltheinformationneededtocalculatetransportproperties(seeEqs.(6)and(7)).Itstimeevolutionordynamicsalsoclarifiesthedominanttransportregimeattheconsid-eredtimescale(ie.ballistic,diffusiveorlocalized).Figure9illustratesthetypicalbehaviorsofthenormalizeddiffusioncoefficientsDnorm(t)=D(t)Dmax(11)whereDmaxisthemaximumofD(t)foranenergyE=0.5eV.Asexpected,puregrapheneissimplyballistic.The0100020003000t (fs)00.20.40.60.81Dnorm-1.5 eV-1.0 eV-0.5 eV0 eV0.5 eV0.75 eV1.0 eV1.5 eVElectron diffusionHole diffusionDiffusion at Dirac pointFIG.10:(coloronline)Normalizeddiffusioncoefficientversustimeforx=1.77%atdifferentenergies.Localizationeffectsaremoreimportantforthepositivechargecarriers.Maximumaround0.75eV.smallestsimulatedimpurityconcentration(ie.0.01%)doesnotreachitsmaximumforthediffusioncoefficientforthetotalelapsedtimeconsideredhere(≈3800fs).TheslopeofD(t)givesav2(E)valuestillincloseagreementwiththetheoretical(analytical)valuev2FforvaluesofEclosetoEF.Forinstance,at0eV,v(E)=1.57Åγ0/¯h,whilevF=2.13Åγ0/¯h.Forintermediatedensities(0.05and0.1%)Dnorm(t)evolvefromtheballisticregimetothedif-fusiveregimeforthesimulatedtimes.Athigherconcentra-tions(0.49%)Dnorm(t)reachadiffusiveregimesharply,furtherevolvingintoa1/√tbehavior,signatureofquantuminterferencescausinglocalization.Figure10presentsdiffusioncoefficientsDnorm(t)of1.77%ofimpuritiesatdifferentenergies.Localizationef-fectsfollowanasymmetricbehaviorbetweenelectronsandholes.Thepredominantresonantpeakattherightofthechargeneutralitypoint(CNP)causesmuchstrongerlocal-izationeffectsthantheweakerandmoresmearedoutpeakattheleftoftheCNP(inanalogywiththePDOSinFig.7).Suchasymmetryisofcrucialimportancetosynthesizepossibleelectronicdevicesmadeoffunctionalizedgraphenerequiringanefficientswitchbetweenaconductingandaninsulatingbehavior.Localizationeffectsarethemostseverearound0.75eV.OnefinallynotesthepeculiarbehaviorofthediffusioncoefficientexactlyattheDiracpoint.Numer-ically,thispointismoredifficulttosimulatesincetheden-sityofchargecarriersmaybeveryscarce.Nevertheless,ourresultssuggestthatthediffusiveplateauisreachedatlatertimeandthenturnsintoamoderatedecreasecharacteristicofquantumeffects.8 (color online) Energy dependent mean free paths for FIG. 10: different impurity concentrations. Inset: mean free path versus hole (dashed/squares) and electron (solid/circles) carrier density for epoxy concentrations of 0.49 % and 0.95 %. FIG. 11: (color online) Ratio of mean free paths at two selected impurity densities. (cid:96)e(x1)/(cid:96)e(x2) (solid lines) and x2/x1 (dashed lines) with x2 = 0.49 %. sidered elapsed times, the diffusive regime is not reached at every energy for the smallest impurity concentrations (ie. ni < 0.1 %, see Fig. 8, upper panel). Therefore, the mean free path can not be estimated for these small concentra- tions. For concentrations larger than 0.5 %, the diffusive regime is reached within the entire energy window. The dip in the mean free path at the right of the CNP (and in a lesser extent, at the left of the CNP), indicating larger scattering effects, shifts away from the CNP and becomes smoother with increasing concentration of oxygen atoms in epoxy po- sition. Such dips correspond to the resonance peaks found in the DOS which are induced by the oxygen. The inset of Fig. 10 confirms the predicted asymmetry, affecting the electrons more strongly. The largest concentrations scatter more uniformly across the entire energy window. In addi- tion, the evolution of the mean free path with impurity con- centration follows a simple scaling law as expected from a Fermi golden rule (Fig. 11): (cid:96)e(x1) (cid:96)e(x2) = x2 x1 2. Mobility (13) In Fig. 12 (mainframe) the mobility of the charge carriers is estimated theoretically using: µ(E) = σsc(E) ne . (14) Scattering effects are affecting the electron mobility more strongly than the hole mobility. This asymmetry is reduced for the largest impurity concentrations [see Fig. 12 (inset)]. FIG. 12: (color online) Main frame: electron and hole mobility for usual experimental carrier densities. More severe scattering effects for negative charge carriers cause an asymmetry in mobility com- pared to the hole mobility. Inset: same mobilities with respect to the gate voltage Vg. Experimentalists usually consider the absolute value of the mobility as a key quantity to characterize samples and corresponding inherent disorder. Temperature breaks the phase coherence of electrons along the scattering path and generally reduces quantum interference effects. Accord- ingly, the use of the semi-classical conductivity in evalua- tion of µ(E) (Eq. 14) is a reasonable approximation to an- alyze the experimental data. On the same basis, computed semi-classical conductivities are expected to be more valu- able for comparison with conductivities measured experi- mentally at room temperature. One may argue that in such a non-zero temperature environment, electron-phonon cou- pling may play also a significant role. However, inelastic scattering lengths due to electron-phonon coupling are ex- tremely long in graphene and may thus be disregarded too, 7-3-2-10123E - EF (eV)1101001000le (nm)0.49 %0.95 %1.77 %3.22 %4.42 %1e+111e+121e+131e+14n (cm-2)110100le (nm)ElectronsHolesFIG.11:(coloronline)Energydependentmeanfreepathsfordif-ferentimpurityconcentrations.Diffusioncausedbythescatter-ingcentersimplyshortermeanfreepaths.Inset:meanfreepathwithcarrierdensitycorrespondingtoconcentrationsof0.49%and0.95%forholes(dashed/squares)andelectrons(solid/circles).C.MeanfreepathsUsingEq.(6),themeanfreepath\x{FFFF}eiscalculatedusingthediffusioncoefficientandplottedinFig.11forenergiesbetweenthetwoVanHovesingularitiesat−2.6and2.6eV.Fortheconsideredtimes,thediffusiveregimeisnotreachedateveryenergyforthesmallestimpurityconcentrations(ie.ni<0.1%,seeFig.9,upperpanel).Therefore,themeanfreepathcannotbeestimatedforthesetinyconcentrations.Forconcentrationslargerthan0.5%thediffusiveregimeisreachedwithintheentireenergywindow.ThedipattherightoftheCNP(andinalesserextent,attheleftoftheCNP),in-dicatinglargerscatteringeffects,shiftsawayfromtheDiracpointandbecomessmootherwithincreasingconcentrationofepoxyoxygens.Suchdipscorrespondtothedenserstatesinducedbytheoxygen.TheinsetofFig.11confirmsthepredictedasymmetry,affectingtheelectronsmorestrongly.Thelargestconcentrationsscattermoreuniformlyacrosstheentireenergywindow.Inaddition,theevolutionofthemeanfreepathwithimpurityconcentrationfollowsaFermigoldenrulewith\x{FFFF}e(x1)\x{FFFF}e(x2)=x2x1,(12)asillustratedinFig.12.Forconcentrationslargerthan2%,thecomparisonbecomeslessaccurate.Thevanish-ingofthegoldenruleattheDiracpointimpliesshort-rangescattering75.Long-rangescatteringinducesdivergenceattheDiracpoint.Thisdifferenceinrangebehavioristreatedinthefollowingparagraph.-3-2-10123E - EF (eV)0.010.11le(x )/le(0.49 %)0.49 %0.95 %1.77 %3.22 %4.42 %FIG.12:(coloronline)\x{FFFF}e(x1)/\x{FFFF}e(x2)(solidlines)andx2/x1(dashedlines)withx2=0.49%.D.Short-rangevslong-rangescatteringThescatteringtimeisdefinedbyτ(E)=\x{FFFF}E(E)v(E).(13)UsingthedispersionrelationE=¯hvFkwithk=√πn,thescatteringtimecanbeestimatedintermsofthecarrierdensityn(seeFig.(??)).AccordingtoNomuraetal.[75]whoalsousedtheKubo-Greenwoodapproachintheirpaper,followingconclusionsapply.Longrange(lr)andshortrange(sr)disorderscatteringtimesscalerespectivelyasfollows:τlr∝√nlr(14)τsr∝1√nsr.(15)Accordingtothissamepaper,longrangeconductivityscaleslinearlywithn,whileshortrangeconductivitydisplaysanon-linearbehavior.Figure??depictsthisconductivity.Withinthisframework,thenon-linearbehaviorindicatesoxygeninepoxypositionisashortrangescatterer.Nevertheless,thissubjectisopentodiscussion,asothermodels61,77 -- 79,alsogoingbeyondtheDrudeapproximation,predictadominantlineardependency(withlogarithmiccor-rections)forbothlongrangeandshortrangeconductivity:σ=4e2nhni(ln√πnR0)2(16)withnithenumberofimpuritiesandR0thescatterer'sra-dius.Additionally,ithasbeenobservedexperimentally76thataquasi-ballisticregimehasanon-linearbehaviorwithn,whileamoredisorderedsystemscaleslinearlywithn.7-3-2-10123E - EF (eV)1101001000le (nm)0.49 %0.95 %1.77 %3.22 %4.42 %1e+111e+121e+131e+14n (cm-2)110100le (nm)ElectronsHolesFIG.11:(coloronline)Energydependentmeanfreepathsfordif-ferentimpurityconcentrations.Diffusioncausedbythescatter-ingcentersimplyshortermeanfreepaths.Inset:meanfreepathwithcarrierdensitycorrespondingtoconcentrationsof0.49%and0.95%forholes(dashed/squares)andelectrons(solid/circles).C.MeanfreepathsUsingEq.(6),themeanfreepath\x{FFFF}eiscalculatedusingthediffusioncoefficientandplottedinFig.11forenergiesbetweenthetwoVanHovesingularitiesat−2.6and2.6eV.Fortheconsideredtimes,thediffusiveregimeisnotreachedateveryenergyforthesmallestimpurityconcentrations(ie.ni<0.1%,seeFig.9,upperpanel).Therefore,themeanfreepathcannotbeestimatedforthesetinyconcentrations.Forconcentrationslargerthan0.5%thediffusiveregimeisreachedwithintheentireenergywindow.ThedipattherightoftheCNP(andinalesserextent,attheleftoftheCNP),in-dicatinglargerscatteringeffects,shiftsawayfromtheDiracpointandbecomessmootherwithincreasingconcentrationofepoxyoxygens.Suchdipscorrespondtothedenserstatesinducedbytheoxygen.TheinsetofFig.11confirmsthepredictedasymmetry,affectingtheelectronsmorestrongly.Thelargestconcentrationsscattermoreuniformlyacrosstheentireenergywindow.Inaddition,theevolutionofthemeanfreepathwithimpurityconcentrationfollowsaFermigoldenrulewith\x{FFFF}e(x1)\x{FFFF}e(x2)=x2x1,(12)asillustratedinFig.12.Forconcentrationslargerthan2%,thecomparisonbecomeslessaccurate.Thevanish-ingofthegoldenruleattheDiracpointimpliesshort-rangescattering75.Long-rangescatteringinducesdivergenceattheDiracpoint.Thisdifferenceinrangebehavioristreatedinthefollowingparagraph.-3-2-10123E - EF (eV)0.010.11le(x )/le(0.49 %)0.49 %0.95 %1.77 %3.22 %4.42 %FIG.12:(coloronline)\x{FFFF}e(x1)/\x{FFFF}e(x2)(solidlines)andx2/x1(dashedlines)withx2=0.49%.D.Short-rangevslong-rangescatteringThescatteringtimeisdefinedbyτ(E)=\x{FFFF}E(E)v(E).(13)UsingthedispersionrelationE=¯hvFkwithk=√πn,thescatteringtimecanbeestimatedintermsofthecarrierdensityn(seeFig.(??)).AccordingtoNomuraetal.[75]whoalsousedtheKubo-Greenwoodapproachintheirpaper,followingconclusionsapply.Longrange(lr)andshortrange(sr)disorderscatteringtimesscalerespectivelyasfollows:τlr∝√nlr(14)τsr∝1√nsr.(15)Accordingtothissamepaper,longrangeconductivityscaleslinearlywithn,whileshortrangeconductivitydisplaysanon-linearbehavior.Figure??depictsthisconductivity.Withinthisframework,thenon-linearbehaviorindicatesoxygeninepoxypositionisashortrangescatterer.Nevertheless,thissubjectisopentodiscussion,asothermodels61,77 -- 79,alsogoingbeyondtheDrudeapproximation,predictadominantlineardependency(withlogarithmiccor-rections)forbothlongrangeandshortrangeconductivity:σ=4e2nhni(ln√πnR0)2(16)withnithenumberofimpuritiesandR0thescatterer'sra-dius.Additionally,ithasbeenobservedexperimentally76thataquasi-ballisticregimehasanon-linearbehaviorwithn,whileamoredisorderedsystemscaleslinearlywithn.n(cm-2)10100100010000µ(cm2V-1s-1)0.49%-Electron0.49%-Hole0.95%-Electron0.95%-Hole-200-1000100200V(Vg)1010000.49%0.95%1.77%4.42%10¹¹10¹²10¹³10¹⁴10⁵9 stronger classical scattering effects on the same energy win- dow (see Fig. 9). This will be emphasized in Section IV D by studying the evolution of the conductivity with time deep into the diffusive regime. 4. Short-range vs long-range scattering The nature of scattering range induced by oxygen atoms placed in epoxy position can be discussed using the scaling properties of semi-classical quantities. Both the scattering time and the conductivity are here briefly outlined98. The elastic scattering time is defined by τ (E) = (cid:96)e(E) v(E) . (15) Using the dispersion relation E = ¯hvF k, τ (E) can be es- timated in terms of the Fermi wave vector k. According to Nomura et al. [99] who also used the Kubo-Greenwood ap- proach, following conclusions apply. Long range (lr) and short range (sr) disorder scattering times scale respectively as follows: τlr ∝ k and τsr ∝ 1 k (16) Following such criterion, our data with corresponding nu- merical fits (Fig. 14) clearly indicate a short-range scatter- ing behavior of oxygen in epoxy position. Such short-range scattering time should diverge at the CNP. This does not hap- pen within the Kubo formalism since the DOS remains finite close to the CNP, in contrast with the prediction of the Drude approximation. A similar analysis based on the scaling of conductivity is not straightforward. According to Ref. [99], long range conductivity should scale linearly with n, while short range conductivity should display a non-linear behavior, approach- ing the constant Boltzmann (or Drude) conductivity σD for EF (cid:29) ¯h/τ. In Fig. 13, the behavior close to the Dirac point behaves differently depending on the impurity concen- tration. For smaller impurity concentrations, the conduc- tivity tends to decrease for energies corresponding to rea- sonable carrier concentrations (up to 1013cm−2), while it increases slightly for larger impurity concentrations. The Drude conductivity never reaches a constant plateau for the whole of the energy E or carrier n range. The short-comings of this conductivity have already been pointed out. this subject We note that is still debated, as other models41,91,100,101 predict a dominant linear dependency (with logarithmic corrections) for both long range and short range conductivity. Additionally, it has been observed experimentally102 that a quasi-ballistic regime exhibits a non-linear behavior with n, while a more disordered system scales linearly with n. Comparison with experiment becomes particularly tricky as FIG. 13: (color online) Comparison between the numerical Kubo conductivity σsc (solid lines) and the Drude approximation σD (dashed lines) for different impurity concentrations. Minimum the- oretical value 2/πG0 is plotted in horizontal dashed green line. at least as a first approximation. 3. Numerical Kubo conductivity and Drude approximation Fig. 13 compares the semi-classical value of the Kubo- Greenwood conductivity σsc (solid lines) with the Drude conductivity σD (dotted lines), extracted from Eq. (8) and (9) respectively. The Drude approximation seems to be valid only for small impurity concentrations in the energy window [−1 eV; 1 eV]. For larger densities, the conductivities are underestimated by a factor of two for energies close to the CNP. Moreover, σD is ill-defined at the CNP [see Eq. (9)]. Indeed, at the CNP, k = 0 and consequently, σD vanishes. However, theoretical work41,83,97 reports extensively on a minimum value of the semi-classical conductivity equal to 2/πG0 = 4e2/πh (when neglecting quantum interferences). To avoid the singularity at 0 eV, the mean free path could be calculated analytically, including its k dependence explic- itly. As illustrated by the calculations for larger impurity concentrations, the variations of the DOS with disorder have to be included, as it is presently the case in the Kubo formula [Eq. (8)]. These variations account for single and multiple scattering events mentioned in Ref. [103]. Up to some nu- merical discrepancy, the semi-classical minimum value of conductivity is confirmed. Additionally, an asymmetry exists between the electron and hole conductivities. This effect is weakened but not cancelled with increasing impurity concentrations. This is similar to the asymmetry already observed for both charge carriers in their respective mean free paths and mobilities. Albeit physically different, the stronger quantum localiza- tion effects on the electron side are directly linked to the 8conductivityWithDensMissingfigureE.MobilityInFig.13themobilityofthechargecarriersisestimatedtheoreticallyusing:µ=σscne.(17)Firstly,themobilityisnotconstantwithnvalidatingthepre-viousobservationthattheconductivityisnotlinearwithnandthus,thatourimpuritiesournotlongranged.Secondly,scatteringeffectsareaffectingtheelectronmo-bilitymorestronglythantheholemobility.Thisasymmetryisreducedforthelargestimpurityconcentrations(seeinset).However,themobilitiesarecalculatedforvariousim-purityconcentrationsusingthesemi-classicalconductivityonly.Localizationeffectsarewillinglyneglected,sinceex-perimentalistsareusuallyusingthemobilitytocharacterizesamplesandtheircorrespondingdisorderatnon-zerotem-perature.Temperaturebreaksthephasecoherencebetweeninteractingimpuritiesandannihilatesinthiswaythequan-tumeffects.Accordingly,thesemi-classicalconductivity,basedonsimpleelasticscatteringonly,isareasonableap-proximationtounderstandtherealconductivityinthiscon-text.Onemayarguethatinsuchanon-zerotemperatureen-vironment,electron-phononcouplingmayplayasignificantrole.However,inelasticscatteringlengthsduetoelectron-phononcouplingareextremelylongingrapheneandmaythusbedisregardedtoo.F.NumericalKubo-GreenwoodconductivityandDrudeapproximationFig.14comparestheKubo-Greenwoodconductivity[Eq.(7),solidlines]withtheDrudeconductivityσD[Eq.(8),dottedlines].Theapproximationseemsvalidonlyforsmallimpurityconcentrationsintheenergywindow[−1;1]eV.Forlargerdensities,itunderestimatestheconductivitiesbyafactoroftwoforenergiesclosetotheDiracpoint.More-over,σDisill-definedattheDiracpoint[seeEq.(8)].In-deed,attheDiracpoint,k=0andconsequently,σDvan-ishes.Theoreticalliterature80howeverreportsextensivelyonaminimumvalueofconductivityequalto2πG0.Toavoid1e+111e+121e+131e+14n (cm-2)10100100010000µ (cm2 V-1 s-1)0.49 % - Electron0.49 % - Hole0.95 % - Electron0.95 % - Hole-200-1000100200V (Vg)101001000100001e+051e+060.49 %0.95 %1.77 %4.42 %FIG.13:(coloronline)Mainframe:electronandholemobilityforusualexperimentalcarrierdensities.Moreseverescatteringeffectsfornegativechargecarrierscauseanasymmetryinmobilitycom-paredtotheholemobility.Inset:samemobilitieswithrespecttothegatevoltageVg.-1-0.500.51E-EF (eV)0.11101001000σ (G0)0.01 %0.05 %0.1 %0.49 %1.77 %4.42 %2/πFIG.14:(coloronline)Mainframe:Comparisonbetweenthenu-mericalKuboconductivity(solidlines)andtheDrudeapproxima-tion(dashedlines)forlowimpurityconcentrations.Inset:sameresultsforlargerconcentrations.MinimumtheoreticalvalueoftheDrudeconductivityisconfirmedforalldensities.thesingularityat0V,themeanfreepathcouldbecalculatedanalytically.Asillustratedbythecalculationsforlargerim-purityconcentrations,thevariationsoftheDOSwithdisor-derhavetobeincluded,aspossiblewiththeKuboformula[Eq.(7)].Uptosomenumericaldiscrepancy,theminimumvalueofconductivityisconfirmed.Additionally,anasymmetryexistsbetweenthedonorandacceptorconductivities.Thiseffectisweakenedbutnotcan-celledwithincreasingimpurityconcentration.Thisissim-ilartotheasymmetryalreadyobservedforthemeanfreepathsandthemobilities.Albeitphysicallydifferent,thestrongerclassicalscatteringeffectsonthedonorsidearedi-rectlylinkedtothestrongerquantumlocalizationeffectson10 FIG. 15: (color online) Kubo conductivities at different elapsed times of wave packet evolution for 4.42 % of impurities. When replacing Dmax by the new expression of D(t) [Eq. (4)] in Eq. (8), the Kubo conductivities are obtained at dif- ferent time scales as depicted in Fig. 15 for a 4.42 % im- purity density. The time-evolution of the conductivity is found not to be uniform over the energy spectrum, indicat- ing there might be different transport regimes depending on the charge carrier energies and impurity concentrations for a given length scale. According to the scaling theory of localization36, there are two possible behaviors for the conductivity corresponding to the weak and strong localization regimes which read as follows: e2 σ(L) = σ(cid:12)(cid:12)Dmax − σ(L) ∼ exp(cid:18)− ¯hπ2 ln(cid:18) L(t) √2(cid:96)e(cid:19) ξ (cid:19) L(t) (18) (19) where the localization length ξ gives an estimation of the dis- tance covered by a charge carrier before it is totally trapped due to this multiple scattering effects. The diffusion length is defined as L(t) = 2(cid:112)2∆X 2(t). (20) This definition of L is reasonable when saturation of the diffusion coefficient has been reached. The extra fac- tor √2 in Eq. 18 compared to the correction obtained by Lee et al. [36] comes from a different definition of D(t). Both numerical estimation of σ(L) (symbols) and analyti- 2(cid:96)e(cid:17)(cid:105) from Eq. (18) (solid lines) cal(cid:104)σ(cid:12)(cid:12)Dmax − e2/¯hπ2 ln(cid:16) L(t)√ are plotted in Fig. 16 for L > Lmax. The numerical part con- tains small jiggling caused by the very sensitive derivation in Eq. (17). FIG. 14: (color online). Scattering times in function of Fermi wave vector k for different oxygen densities. Numerical fits for each data set based on a regression algorithm with fitting parameters A and B (black) to account for the generalized inverse function. Hole and electron scattering times were fitted separately. most of previously mentioned analytical models disregard important multiple scattering effects on the computed con- ductivity. Another important remark is that most theoretical predic- tions have been derived assuming restrictions on disorder models which are partly invalidated in the present study. Indeed, the epoxy defects have been derived from accu- rate first-principle calculations, and the resulting TB model brings more realism and generality when compared to sim- plified academic models. As a matter of fact, the DOS (Fig. 6) evidences resonant energy bumps, driven by ran- domly distributed oxygen, which could cause squared loga- rithmic corrections41. Our data cannot really be accurately fitted to obtain the different corrections to the scaling in this context. This dominant short-range disorder is at the root of the quantum effects presented in the remaining Sections. D. Evolution of the Kubo conductivity with time scale (or length) The semi-classical expression of the Kubo-Greenwood conductivity [Eq. (7)] restricts the transport to the diffusive regime, ie. when suppressing quantum interferences. To fol- low the time evolution spreading of the quantum wave pack- ets, the expression of Dx(t) in Eq. (3) should be replaced as follows: Dx(t) = ∂(∆X 2(t)) ∂t (17) -1!109-5!10805!1081!109k - kF (cm-1)110(cid:111)"(fs)0.49 %0.95 %1.77 %3.22 %4.42 %1/(A+Bk)9-2-1012E-EF (eV)012σ (G0)t at Dmaxt = 250 fst = 1000 fst = 2000 fs2/πFIG.15:(coloronline)Kuboconductivityatdifferenttimesofwavepacketevolutionfor4.42%ofimpurities.thesameenergywindow(cf.Fig.10ofthediffusioncoef-ficientatdifferentenergies).Thiswillbeemphasizedinanextparagraphbystudyingtheevolutionoftheconductivitywithtimedeepintothediffusiveregime.G.EvolutionoftheconductivitywithtimeThesemi-classicalexpressionoftheKubo-Greenwoodconductivity[Eq.(7)]restrictsthetransporttothediffusiveregime,thusneglectingquantuminterferences.Moreprecisely,theexpressionofD(t)=2∆X2(t)tin[Eq.(3)]shouldbereplacedbyD(t)=d(2∆X2(t))dt(18)tofollowtheexacttimeevolutionspreadingofthequan-tumwavepackets.UsingthisnewexpressionofDx(t)in[Eq.(8)],theKuboconductivitiesareobtainedatdifferenttimesasdepictedinFig.15fora4.42%impuritydensity.Thetime-evolutionoftheconductivityisfoundnottobeuniformovertheenergyspectrum,indicatingtheremightbedifferenttransportregimesdependingontheFermilevello-cation.Accordingtothestandardscalingtheoryoflocalization84therearetwopossiblebehaviorsfortheconductivitycorre-spondingtotheweakandstronglocalizationregimesde-finedrespectivelywhichreadasfollows:σ(L)−σ\x{FFFF}\x{FFFF}Dmax=−e2¯hπ2ln\x{FFFF}L√2\x{FFFF}e\x{FFFF}(19)σ(L)∝exp\x{FFFF}−L(t)2ξ\x{FFFF}(20)-2-1012E (eV)1101001000ξ (nm)0.49 %0.95 %1.77 %3.22 %4.42 %FIG.16:(coloronline)Localizationlengthsfordifferentconcen-trationsconfirmtheimportantlocalizationduetotheimpuritiesbe-tween-1and0eV.Experimentallyobservedvaluesbetween10and100nmindicatedensitiesbetween0.5and1eV.wherethelocalizationlengthξgivesanestimationofthedis-tancecoveredbyachargecarrierbeforeitistotallytrappedduetothismultiplescatteringeffects.Thediffusionlengthisdefined(somehowarbitrarily)asL=2\x{FFFF}∆X2(t).(21)whichseemsreasonablewhensaturationofthediffusionco-efficienthasbeenreached.InvokingtheIoffeandRegelcriterion82,theappearanceofstronglocalizationbecomessignificantforimpuritycon-centrationssatisfyingkF\x{FFFF}e=1.Suchacriterionimpliesameanfreepathoftheorderoftheinteratomicdistance,whichisactuallythecasefor4.42%ofimpuritieswherethemeanfreepathisestimatedtobebelow3Åfortheenergywindowfrom0.5eVto1.0eV.Fromtheexponentialscalinglaw,thetheoreticallocaliza-tionlengthcanbededucedandcompared(seeFig.16)totheThoulesslocalizationlength84definedfor2-Dsystemsasξ(E)=\x{FFFF}eexp\x{FFFF}πσD2G0\x{FFFF}(22)ξ(E)onlydependsontheDrudeconductivityσD.Exper-imentalistsoftenapproximatethislocalizationlengthwiththelawξ(E)=Lexp\x{FFFF}2σDG0\x{FFFF}(23)wherethelessaccurateDrudeapproximationisusedaswellasthecharacteristiclengthinsteadofthemeanfreepath.Thesedefinitionsofthelocalizationlengthdonotincludeexplicitlyanyquantumeffect,rathertheysimplyassumethelocalizationisexponentialandthus,onlytheclassicalcon-ductivity(equaltotheconductanceGin2-D)issufficientto11 FIG. 17: Eq. (17) for different impurity concentrations. (color online) Localization lengths estimated using FIG. 18: Exponential fits [Eq. (19)] to estimate localization lengths ξ in the strongly localized regime at energy 0.8 eV. A first fit (dashed lines) for the whole of the available data allows us to esti- mate a length L for which wavepackets are localized. A second fit (solid lines) for values larger than L gives us new estimates for ξ. eV (dashed lines). Refitting σ(L) for the region at the right of these values (solid lines), we obtain convincing exponen- tial decays and more accurate estimates for ξ equal to 11.2 and 5.3 nm, respectively. Both these estimates and the ones obtained by Eq. (21) in Fig. 17 are of the same order of mag- nitude, thus validating our results. Experimentalists however often use the Drude approximation σD instead of the correct semi-classical conductivity σsc in Eq. (21). The inaccuracy of the Drude approximation for largest impurity concentra- tions causes the localization length to be underestimated by an order of magnitude. FIG. 16: (color online) Kubo conductivities for L > Lmax, which include weak localization corrections to the semi-classical conduc- tivity, for different impurity densities at energy 0.8 eV. The numer- ically estimated conductivity σ(L) (obtained using Eqs. 17 and 20, symbols) contains numerical jiggling caused by the very sensitive derivation in Eq. 17. The conductivity obtained using Eq. (18) is plotted in solid lines. Only one out of five points are plotted in the inset for clarity reasons. The corrections to the semi-classical conductivity in the low impurity limit (mainframe) follows the logarithmic be- havior, from which an estimation of ξ can be deduced. ξ cor- responds to the length where the cooperon corrections equal the semi-classical conductivity36. Starting from Eq. (18), the localization length is thus estimated with ξ(E) = √2(cid:96)e exp(cid:18) πσsc G0 (cid:19) (21) with the computed (cid:96)e and σsc and corresponds to the 2D gen- eralization of the Thouless relationship104,105. These local- ization lengths are plotted in Fig. 17106. For larger concentrations, the cooperon corrections to the semi-classical conductivity seem to saturate and depart from the perfect logarithmic behavior (Fig. 16, inset). The cor- rections obtained numerically become smaller than what is predicted due to a transition to the strongly localized regime following an evanescent exponential behavior. This can be rationalized invoking the Ioffe and Regel criterion107 which states that the appearance of strong localization becomes sig- nificant for impurity concentrations satisfying kF (cid:96)e = 1. Such a criterion implies a mean free path of the order of the interatomic distance, which is actually the case for 4.42 % of impurities where the mean free path is estimated to be below 3 A for the energy window from 0.5 eV to 1.0 eV. In Fig. 18, by fitting the exponential behavior of Eq. (19), values for ξ equal to 8.4 and 4.8 nm are estimated for 3.22 % and 4.42 % of impurities respectively at an energy of 0.8 0204060L (nm)23456σ (G0)0.49 %0.95 %05101500.513.22 %4.42 %-2-1012E-EF (eV)100101102103104105106(cid:106)!(nm)0.49 %0.95 %1.77 %3.22 %4.42 %051015L (nm)00.51σ (G0)3.22 %4.42 %fit gives ξ = 8.4 nm fit gives ξ = 4.8 nmfit for L > 8.4 nm gives ξ = 11.2 nmfit for L > 4.8 nm gives ξ = 5.3 nmV. CONCLUSIONS In this paper, the quantum transport properties of chem- ically damaged two-dimensional graphene based structures have been investigated. Using the Kubo-Greenwood trans- port framework, and by means of an efficient order N numer- ical implementation, mesoscopic transport features in disor- dered graphene have been explored in details, with impuri- ties (adsorbed oxygen-driven epoxide defects) described by local tight-binding parameters, deduced from first-principles calculations. In addition to the numerical calculation of the energy- dependent elastic mean free path driven by a given epox- ide density, quantum localization effects have been analyzed from the weak to the strong (Anderson) localization regimes. By applying the conventional scaling theory of localization, the 2D-localization lengths have been evaluated from the scaling behavior of the Kubo conductivity, and contrasted to the prediction deduced from the cooperon correction to the conductivity (which relates ξ to the elastic mean free path and semi-classical conductivity). A very reasonable agree- ment has been obtained, pinpointing further towards a strong energy-dependence of all transport length scales. By combining the ab-initio approach for the description 12 of the defects structure and local energetics with an efficient and exact quantum transport methodology implemented on tight-binding models, our general theoretical framework provides a solid foundation and tool to understand the origin of complex transport phenomena in strongly disordered and chemically complex graphene-based nanostructures. The extension of our study to any other kinds of defects (topo- logical, chemical, etc) and other types of two-dimensional structures is straightforward. Acknowledgments J.-C.C., N.L., A.L. and F.V. acknowledge financial sup- port from the F.R.S.-FNRS of Belgium. This work is di- rectly connected to the Belgian Program on Interuniversity Attraction Poles (PAI6) on 'Quantum Effects in Clusters and Nanowires', to the ARC on 'Graphene StressTronics' spon- sored by the Communaut´e Franc¸aise de Belgique, to the European Union through the ETSF e-I3 project (Grant N.◦ 211956), and to the NANOSIM-GRAPHENE project (Projet N.◦ ANR-09-NANO-016-01). P.O. acknowledges Spanish Grants from MICINN (FIS2009-12721-C04-01, CSD2007- 00050). Computational resources were provided by the CISM of the Universit´e catholique de Louvain: all the nu- merical simulations have been performed on the GREEN and LEMAITRE computers of the CISM. 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 K.S. Novoselov, A.K. Geim, S.V. Morozov, D. Jiang, M.I. Kat- snelson, I.V. Grigorieva, S.V. Dubonos, and A.A. Firsov, Nature (London) 438, 197 (2005). 3 A.K. Geim and K.S. Novoselov, Nat. Mater. 6, 183 (2007). 4 K. Novoselov, Rev. Mod. Phys. 83, 837849 (2011). 5 M. Lemme, Sol. Stat. Phen. 499, 156 (2010). 6 H. Yamaguchi, G. Eda, C. Mattevi, H. Kim, and M. Chhowalla, ACS Nano 4, 524-528 (2010). 7 Y. Wu, Y.-M. Lin, A.A. Bol, K.A. Jenkins, F. Xia, D.B. Farmer, Y. Zhu, Ph. Avouris, Nature 472, 74 (2011). 8 T. Palacios, Nature Nanotech. 6, 464465 (2011). 9 F. Schwierz, Nature Nanotech. 5, 487496 (2010). 10 M.Y. Han, B. Ozyilmaz, Y. Zhang, and P. Kim, Phys. Rev. Lett. 98, 206805 (2007). 1229 (2008). (2008). 11 X. Li, X. Wang, L. Zhang, S. Lee, and H. Dai, Science 319, 12 Z. Chen, Y. Lin, M. Rooks, and P. Avouris, Physica E 40, 228 13 D.C. Elias, R. Nair, T. Mohiuddin, S. Morozov, P. Blake, M. Halsall, A. Ferrari, D. Boukhvalov, M. Katsnelson, A. Geim, and K. Novoselov, Science 323, 610 (2009). 14 C. G´omez-Navarro, R.T. Weitz, A.M. Bittner, M. Scolari, A. Mews, M. Burghard, and K. Kern, Nano Lett. 7, 3499 (2007). 15 A.B. Kaiser, C. G´omez-Navarro, R.S. Sundaram, M. Burghard, and K. Kern, Nano Lett. 9, 1787 (2009). 16 J.O. Sofo, A.S. Chaudhari, and G.D. Barber, Phys. Rev. B 75, 153401 (2007). 17 S. Ryu, M. Han, J. Maultzsch, T. Heinz, P. Kim, M. Steiger- wald, and L. Brus, Nano Lett. 8, 4597 (2008). 18 S. Stankovich, D. Dikin, G. Dommett, K. Kohlhaas, Z. Zimney, E. Stach, R. Piner, S. Nguyen, and R. Ruoff, Nature 442, 282 (2006). 19 S. Gijie, S. Han, M. Wang, K. Wang, and R. Kaner, Nano Lett. 7, 3394 (2007). 20 S. Stankovich, D. Dikin, R. Piner, K. Kohlhaas, A. Klein- hammes, Y. Jia, Y. Wu, S. Nguyen, and R. Ruoff, Carbon 45, 1558 (2007). 21 R. Ruoff, Nature Nanotechnology 3, 10 (2008). 22 X. Wu, M. Sprinkle, X. Li, F. Ming, C. Berger, and W.A. de Heer, Phys. Rev. Lett. 101, 026801 (2008). 23 S. Roche, Nature Nanotech. 6, 8 (2011). 24 A. Cresti, N. Nemec, B. Biel, G. Niebler, F. Triozon, G. Cunib- erti, and S. Roche, Nano Research 1, 361 (2008). 25 H.L. Calvo, H.M. Pastawski, S. Roche, L.E.F. Foa Torres, App. Phys. Lett. 98, 23 (2011). 26 E. McCann, D.S.L Abergel, and V.I Fal'ko, Solid State Comm. 27 B. Biel, X. Blase, F. Triozon, and S. Roche, Phys. Rev. Lett. 143, 110 (2007). 102, 096803 (2009). 30 A. F. Morpurgo and F. Guinea, Phys. Rev. Lett. 97, 196804 62 D.A. Abanin, A.V. Shytov, and L.S. Levitov, Phys. Rev. Lett. 13 28 B. Biel, F. Triozon, X. Blase, and S. Roche, Nano Lett. 9, 2725 29 M.I. Katsnelson, K.S. Novoselov, A.K. Geim, Nature Physics (2009). (2006). 2, 620-625 (2006). 31 H. Suzuura and T. Ando, Phys. Rev. Lett. 89, 266603 (2002). 32 H. Suzuura and T. Ando, J. Phys. Soc. Jpn. 75, 024703 (2006). 33 V.I. Fal'ko, K. Kechedzhi, E. McCann, B.L. Altshuler, H. Suzuura, T. Ando, Sol. Stat. Comm. 143, 33 (2007). 34 F. Ortmann, A. Cresti, G. Montambaux, and S. Roche, Euro Phys. Lett. 94, 47006 (2011). 35 S. Das Sarma, S. Adam, E.H. Hwang, and E. Rossi, Rev. Mod. Phys. 83, 407 (2011). 36 P.A. Lee and T.V. Ramakrishnan, Rev. Mod. Phys. 57, 2 (1985). 37 E. Akkermans and G. Montambaux, Mesoscopic Physics of Electrons and Photons, ISBN 978-0521855129 (2007). 38 F.V. Tikhonenko, D.W. Horsell, R.V. Gorbachev, and A. K. Savchenko, Phys. Rev. Lett. 100, 056802 (2008). 39 F.V. Tikhonenko, A.A. Kozikov, A.K. Savchenko, and R.V. Gorbachev Phys. Rev. Lett. 103, 226801 (2009). 40 E. McCann, K. Kechedzhi, V.I. Fal'ko, H. Suzuura, T. Ando, and B.L. Altshuler, Phys. Rev. Lett. 97, 146805 (2006). 41 P. M. Ostrovsky, I. V. Gornyi, and A. D. Mirlin, Phys. Rev. B 42 D. V. Khveshchenko, Phys. Rev. B 75, 241406 (2007). 43 S. Adam, P.W. Brouwer, S. Das Sarma, Phys. Rev. B 79, 74, 235443 (2006). 201404(R) (2009). 44 X.S. Wu, X.B Li, Z.M. Song, C. Berger, and W.A. de Heer, Phys. Rev. Lett. 98, 136801 (2007). 45 A. Altland, Phys. Rev. Lett. 97, 236802 (2006). 46 I.L. Aleiner and K.B. Efetov, Phys. Rev. Lett. 97, 236801 47 J.H. Bardarson, J. Tworzydlo, P.W. Brouwer, and C.W.J. Beenakker, Phys. Rev. Lett. 99, 106801 (2007). 48 K. Nomura, M. Koshino, S. Ryu, Phys. Rev. Lett. 99, 146806 (2006). (2007). 49 Y.Y. Zhang, J. Hu, B.A. Bernevig, X.R. Wang, X.C. Xie, and W.M. Liu, Phys. Rev. Lett. 102, 106401 (2009). 50 J.P. Robinson, H. Schomerus, L. Oroszlany, and V.I. Falko, Phys. Rev. Lett. 101, 196803 (2008). 51 J. Bang and K.J. Chang, Phys. Rev. B 81, 193412 (2010) 52 M. Titov, P.M. Ostrovsky, I.V. Gornyi, A. Schuessler, and A.D. Mirlin, Phys. Rev. Lett. 104, 076802 (2010). 53 N. Leconte, J. Moser, P. Ordejon, H. Tao, A. Lherbier, A. Bach- told, F. Alsina, C.M. Sotomayor Torres, J.C. Charlier, and S. Roche, ACS Nano 4, 7, 4033-4038 (2010). 54 A. Bostwick, J.L. McChesney, K.V. Emtsev, T. Seyller, K. Horn, S.D. Kevan, and E. Rotenberg, Phys. Rev. Lett. 103, 056404 (2009). 55 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). 56 W. Li, Y.H. He, L. Wang, G.H. Ding, Z.-Q. Zhang, R.W. Lortz, P. Sheng, and N. Wang, Phys. Rev. B 84, 045431 (2011). 57 N. Leconte, D. Soriano, S. Roche, P. Ordejon, J.-C. Charlier, J.J. Palacios, ACS Nano 5, 3987 (2011). 58 D. Soriano, N. Leconte, P. Ordejon, J.-C. Charlier, J.J. Palacios, S. Roche, Phys. Rev. Lett. 107, 016602 (2011). 59 G. Lee, B. Lee, J. Kim, and K. Cho, J. Phys. Chem. C 113, 14225 (2009). 60 T.O. Wehling, A.V. Balatsky, M.I. Katsnelson, A.I. Lichten- stein, K. Scharnberg, and R. Wiesendanger, Phys. Rev. B 75, 125425 (2007). 61 V.V. Cheianov, O. Syljuasen, B.L. Altshuler, and V.I. Fal'ko, Phys. Rev. B 80, 233409 (2009). 105, 086802 (2010). 63 L. Staudenmaier, Ber. Dtsch. Chem. Ges. 31, 1481 (1898). 64 J.L. Li, K.N. Kudin, M.J. McAllister, R.K. Prud'homme, I.A. Aksay, and R. Car, Phys. Rev. Lett. 96, 176101 (2006). 65 T. Sun, S. Fabris (submitted). 66 H.J. Xiang, S.H. Wei, and X.G. Gong, Phys. Rev. B 82, 035416 (2010). 67 J. Moser, H. Tao, S. Roche, F. Alsina, C.M. Sotomayor Torres, and A. Bachtold, Phys. Rev. B 81, 205445 (2010). 68 J.M. Soler, E. Artacho , J.D Gale , A. Garcia , J. Junquera , P. Ordejon, and D. Sanchez-Portal, J. Phys.: Condensed Matter 14, 2745 (2002). 69 E. Artacho, E. Anglada , O. Dieguez , J.D. Gale , A. Garcia , J. Junquera , R.M. Martin , P. Ordejon , J.M. Pruneda , D. Sanchez-Portal, and J.M. Soler, J. Phys.: Condensed Matter 20, 064208 (2008). 70 E. Artacho, D. Sanchez-Portal, P. Ordejon, J.M. Soler, Physica Status Solidi (b) 215, 809 (1999). 71 D.M. Ceperley and B.J. Alder, Phys. Rev. Lett. 45, 566 (1980). 72 J.P. Perdew and A. Zunger, Phys. Rev. B 23 5048 (1981). 73 N. Troullier and J. L. Martins, Phys. Rev. B 43, 1993 (1991). 74 F.L. Hirshfeld, Theo. Chem. Acta. 44, 129 (1977). 75 C. Fonseca Guerra, J.W. Handgraaf, E.J. Baerends, and F.M. Bickelhaupt, J. Comp. Chem. 25, 189 (2003). 76 F.M. Bickelhaupt, N.J.R. van Eikema Hommes, C. Fonseca Guerra, and E.J. Baerends, Organometallics 15, 2923 (1996). 77 M.A. Zwijnenburg, S.T. Bromley, C. van Alsenoy, and T. Maschmeyer, J. Phys. Chem. A, 12376 (2002). 78 R. Dronskowski, P.E. Blochl, J. Phys. Chem. 97, 8617 (1993). 79 T. Hughbanks, R. Hoffmann, J. Am. Chem. Soc. 105, 3528 80 A. Lherbier, Simon M.-M. Dubois, X. Declerck, S. Roche, Y.- M. Niquet, and J.-C. Charlier, Phys. Rev. Lett. 106, 046803 (2011). 81 A. Castro Neto, F. Guinea, and N. Peres, Rev. of Mod. Phys. 81 (1983). (2009). 82 F. Triozon, S. Roche, A. Rubio, and D. Mayou, Phys. Rev. B 83 A. Lherbier, B. Biel, Y.-M. Niquet, and S. Roche, Phys. Rev. 69, 121410(R) (2004). Lett. 100, 036803 (2008). 84 S. Roche and D. Mayou, Phys. Rev. Lett. 79, 2518 (1997). 85 S. Roche, Phys. Rev. B. 59, 2284 (1999). 86 S. Roche, F. Triozon, A. Rubio, and D. Mayou, Phys. Rev. B 64, 121401 (2001). 87 S. Roche and R. Saito, Phys. Rev. Lett. 87, 246803 (2001). 88 S. Latil, S. Roche, D. Mayou, and J.-C. Charlier, Phys. Rev. Lett. 92, 256805 (2004). 89 H. Ishii, F. Triozon, N. Kobayashi, K. Hirose, and S. Roche, C.R. Physique 10, 283 (2009). 90 E.R. Mucciolo and C.H. Lewenkopf, J. Phys. Cond. Matt. 22, 273201 (2010). (2010). 91 J.W. Klos and I.V. Zozoulenko, Phys. Rev. B 82, 081414 92 V.M. Pereira, J.M.B. Lopes dos Santos, and A. H. Castro Neto, Phys. Rev. B 77, 115109 (2008). 93 S.D. Wu, L. Jing, Q.X. Li, Q.W. Shi, J. Chen, H.B. Su, X.P. Wang, J.L. Yang, Phys. Rev. B 77, 195411 (2008). 94 C. Kittel, Th´eorie quantique du solide, ASIN: B0014UYCPE, 95 T.O. Wehling, M.I. Katsnelson, A.I. Lichtenstein, Phys. Rev. B Dunod, Paris (1967). 80, 085428 (2009). 96 K.A. Mkhoyan, A.W. Contryman, J. Silcox, D.A. Stewart, G. Eda, C. Mattevi, S. Miller, M. Chhowalla, Nano Lett. 9, 1058 (2009). 97 N.H. Shon and T. Ando, J. Phys. Soc. Jpn 67, 2421 (1998) 98 E. Pallecchi, A.C. Betz, J. Chaste, G. Feve, B. Huard, T. Kon- tos, J.-M. Berroir, and B. Placais, Phys. Rev. B 83, 125408 (2011). 99 K. Nomura and A. H. MacDonald, Phys. Rev. Lett. 98, 076602 (2007). 100 T. Stauber, N.M.R. Peres, and F. Guinea, Phys. Rev. B 76, 14 205423 (2007). 143, 3 (2007). 101 M.I. Katsnelson and K.S. Novoselov, Solid State Commun. 102 K.I. Bolotin, K.J. Sikes, J. Hone, H.L. Stormer, and P. Kim, Phys. Rev. Lett. 101, 096802 (2008). 103 J.E. Santos, N.M.R. Peres, J.M.B. Lopes dos Santos, and A.H. Castro Neto, Phys. Rev. B 84, 085430 (2011). 104 D.J. Thouless, J. Phys. C: Solid State Phys. 6, L49-L51 (1973). 105 C.W. Beenakker, Rev. Mod. Phys. 69, 731 (1997) 106 Values of ξ obtained by crossing the solid line with the G0 line in Fig. 16 are slightly different from the ones obtained in Fig. 17. Numerical inaccuracy originating in the derivative term in Eq. (17) causes the maximum of σL to be not exactly equal to σSC. 107 A.F. Ioffe and A.R. Regel, Prog. Semicond. 4, 237 (1960).
1504.05343	2	1504	2016-02-16T19:47:18	Finite-temperature effective boundary theory of the quantized thermal Hall effect	[ "cond-mat.mes-hall", "cond-mat.str-el" ]	A finite-temperature effective free energy of the boundary of a quantized thermal Hall system is derived microscopically from the bulk two-dimensional Dirac fermion coupled with a gravitational field. In two spatial dimensions, the thermal Hall conductivity of fully gapped insulators and superconductors is quantized and given by the bulk Chern number, in analogy to the quantized electric Hall conductivity in quantum Hall systems. From the perspective of effective action functionals, two distinct types of the field theory have been proposed to describe the quantized thermal Hall effect. One of these, known as the gravitational Chern-Simons action, is a kind of topological field theory, and the other is a phenomenological theory relevant to the St\v{r}eda formula. In order to solve this problem, we derive microscopically an effective theory that accounts for the quantized thermal Hall effect. In this paper, the two-dimensional Dirac fermion under a static background gravitational field is considered in equilibrium at a finite temperature, from which an effective boundary free energy functional of the gravitational field is derived. This boundary theory is shown to explain the quantized thermal Hall conductivity and thermal Hall current in the bulk by assuming the Lorentz symmetry. The bulk effective theory is consistently determined via the boundary effective theory	cond-mat.mes-hall	cond-mat	6 Finite-temperature effective boundary theory of 1 0 2 the quantized thermal Hall effect b e F 6 1 ] l l a h - s e m . t a m - d n o c [ 2 v 3 4 3 5 0 . 4 0 5 1 : v i X r a Ryota Nakai1, Shinsei Ryu2, Kentaro Nomura3 1 WPI-Advanced Institute for Materials Research (WPI-AIMR), Tohoku University, Sendai 980-8577, Japan 2 Department of Physics, University of Illinois, 1110 West Green St, Urbana IL 61801 3 Institute for Materials Research, Tohoku University, Sendai 980-8577, Japan E-mail: [email protected] Abstract. A finite-temperature effective free energy of the boundary of a quantized thermal Hall system is derived microscopically from the bulk two-dimensional Dirac fermion coupled with a gravitational field. In two spatial dimensions, the thermal Hall conductivity of fully gapped insulators and superconductors is quantized and given by the bulk Chern number, in analogy to the quantized electric Hall conductivity in quantum Hall systems. From the perspective of effective action functionals, two distinct types of the field theory have been proposed to describe the quantized thermal Hall effect. One of these, known as the gravitational Chern-Simons action, is a kind of topological field theory, and the other is a phenomenological theory relevant to the Streda formula. In order to solve this problem, we derive microscopically an effective theory that accounts for the quantized thermal Hall effect. In this paper, the two-dimensional Dirac fermion under a static background gravitational field is considered in equilibrium at a finite temperature, from which an effective boundary free energy functional of the gravitational field is derived. This boundary theory is shown to explain the quantized thermal Hall conductivity and thermal Hall current in the bulk by assuming the Lorentz symmetry. The bulk effective theory is consistently determined via the boundary effective theory. PACS numbers: 04.62.+v,74.25.F-,74.90.+n Submitted to: New J. Phys. Finite-temperature effective boundary theory of the quantized thermal Hall effect 2 1. Introduction Topology of the energy band structure in insulators and superconductors emerges in transport phenomena, and the topological number can be detected as a transport coefficient[1, 2]. The electric Hall conductivity is quantized and given by the Chern number of the occupied wave functions in two-dimensional band insulators[3, 4]. The Chern number is nonzero when an electronic system is in time-reversal symmetry broken topological phases. Similarly, two-dimensional insulators and superconductors show the quantized thermal Hall effect in time-reversal symmetry broken topological phases[5]. This similarity between electric and thermal responses reflects a parallel between topological insulators in symmetry class A and topological superconductors in symmetry class D, both of which are characterized, in two dimensions, by the Chern number of the filled energy bands[6, 7, 8]. Topological quantum field theories are efficient descriptions and characterizations of topological phases. Effective actions for the external electromagnetic and gravitational fields are given in terms of topological terms and attributed to quantum anomalies [9, 10]. They can be used to discuss the classification of topological insulators and superconductors in arbitrary dimensions[11, 12, 13]. Quantized electromagnetic responses, including the quantum Hall effect in two dimensions, can be viewed as responses resulting from the Chern-Simons action functional. For the case of the electromagnetic field, the Chern-Simons action is given as SEM[A] = Ce2 4π Z d3x ǫµνρAµ∂νAρ in the (2+1)-dimensional space-time, where C is the Chern number (here and henceforth natural units with c = = kB = 1 are used). On the other hand, the quantized thermal Hall effect has been predicted to occur in a weak pairing phase of the chiral p-wave superconductor in Ref. [5], which is considered to be a realization of a two- dimensional time-reversal symmetry broken topological superconductor. In analogy with the electromagnetic response, it has been claimed that the quantized thermal response of topological superconductors is described by the gravitational Chern-Simons action[5] (1) (2) SG[ω] = C 96πZ d3x ǫµνρ tr(cid:18)ωµ∂νωρ + 2 3 ωµωνωρ(cid:19) , where ωµ is the spin connection. In recent years, space-time curvature is widely used to study characteristic responses of topological materials[14, 15, 16]. The gravitational Chern-Simons action is associated with the gravitational anomaly in (2+1) dimensions [9, 10, 17], and can be microscopically derived from the (2+1)-dimensional massive Dirac fermion coupled with the background gravitational field at zero temperature[18, 19, 20]. However, it has been noted that a thermal Hall current in the bulk cannot be created by a uniform gravitational field gradient as a response derived from the gravitational Chern- Simons action[21]. Therefore, while the coefficient of the gravitational Chern-Simons term ("the chiral central charge") is related to the quantized thermal Hall conductance, the quantized thermal transport and the gravitational Chern-Simons term appear to Finite-temperature effective boundary theory of the quantized thermal Hall effect 3 be more remotely related then the quantized electromagnetic response and the U(1) Chern-Simons term. An inhomogeneous temperature field driving thermal transport is effectively realized by a gravitational potential field φ through the Luttinger's phenomenological argument using the Tolman-Ehrenfest relation[22] 1 T ∂T = ∂φ. (3) Owing to the relation jT = jE − (µ/e)jC, the thermal current jT is identified as the energy current jE when the charge current jC does not contribute at zero chemical potential µ = 0, which is true for superconductors. A thermal current induced by a temperature gradient is then equivalent to an energy current induced by the space-time metric gµν. From the phenomenological point of view, an analogy between the electric and the thermal transport holds. The Wiedemann-Franz law connecting the electric and the thermal Hall conductivity for the Dirac fermion[23, 24, 25] has been proved to be κH = π2 3 T e2 σH . (4) Note that in the case of topological superconductors, the coefficient is half of that in (4) since a Majorana fermion, a quasi-particle in a topological superconductor, is half of a complex fermion. Based on the phenomenological analogy of the Hall conductivity, an effective free energy for the quantized thermal Hall effect jk T = κH ǫkl∂lT of a Lorentz invariant system has been proposed as[24] (5) (6) F [φ, Ω] ∝ κH T v2 Z d2xφΩ, where Ω is the angular velocity of the system in a rotating frame, and κH = Cπ2T /6π is the thermal Hall conductivity for insulators whose occupied energy bands have a total Chern number C, and v is the Fermi velocity of the Dirac fermion. This form of the phenomenological effective free energy (6) can be realized when we consider a metric in a rotating frame with a gravitational potential. So far, a action or a free energy of the gravitational field that accounts for the quantized thermal Hall effect has not been derived directly from fermionic models of topological insulators and topological superconductors. In this study, an effective free energy functional of the gravitational field is derived microscopically from the two-dimensional massive Dirac fermion coupled with the static gravitational field. The effective free energy is derived by the following procedure. First, we consider the gapless boundary fermion, which is a manifestation of the Dirac fermion with nontrivial bulk energy band topology. Then an effective free energy of the boundary theory is calculated by the field theoretical method. We also consider the bulk thermal Hall current and a bulk effective field theory. Finite-temperature effective boundary theory of the quantized thermal Hall effect 4 This paper is organized as follows. As a preliminary to the main contents of this study, phenomenological theory of the quantized thermal Hall effect is reviewed in Sec. 2.1, and a microscopic model studied in this paper is introduced in Sec. 2.2. In Sec. 3.1 the two-dimensional bulk fermion Hamiltonian is projected onto the one- dimensional boundary to give the boundary fermion Hamiltonian, and, in Sec. 3.2, the boundary effective free energy is calculated from the boundary fermion. In Sec. 3.3, the quantized thermal Hall conductivity is derived from the boundary theory. In Sec. 4.1, the thermal Hall current of the two-dimensional gapped bulk are deduced from the boundary effective free energy. In Sec. 4.2, the bulk effective free energy is conjectured from the expression of the bulk thermal current. We give discussion about our results in Sec. 5 and summarize our conclusions in Sec. 6. 2. Preliminaries 2.1. Phenomenology In this subsection, we briefly review the phenomenology of the thermal Hall effect[23, 24] of a Lorentz invariant system in two dimensions. Quantized thermal (energetic) Hall current in a two-dimensional fully gapped topological insulator and superconductor is phenomenologically defined in analogy with the quantized part of the electric Hall current in the quantum Hall system. The electric Hall current is defined by jC = σHz × E, (7) where z is the unit vector pointing the out-of-plane direction. Combining (7) with the Streda formula[26] for the quantized electric Hall conductivity, given by the derivative of the magnetization with respect to the chemical potential σH = −e ∂M z ∂µ , jC = ∂µ × and a phenomenological relation eE = −∂µ, we have obtained the relation ∂M ∂µ . (8) (9) The electric current (9) can be regarded as a magnetization current jC = ∂ × M , if we assume that the magnetization varies according to variation of the chemical potential. Similarly, by using the Streda formula for the quantized thermal Hall conductivity[24], given by the derivative of the energy magnetization ME with respect to temperature ∂M z E ∂T the thermal Hall current κH = − , (10) jE = −κHz × ∂T (11) can be identified with the energy version of the magnetization current[23] jE = ∂×ME, when the energy magnetization changes through variation of temperature as jE = ∂T × ∂ME ∂T . (12) Finite-temperature effective boundary theory of the quantized thermal Hall effect 5 Note that M and ME have only z-component. This fact implies that the thermal Hall current in a fully gapped system can be described in terms of the magnetization type energy current caused by the energy magnetization. Lorentz invariance of the fermionic system gives rise to a symmetry of the energy- momentum tensor with respect to permutations of the indices. The energy momentum tensor has the form of T µν =  j1 E/v j2 ǫ vπ1 Σ11 vπ2 Σ21 E/v Σ12 Σ22   , (13) where ǫ is the energy density, πj is the momentum density, and Σjk is the stress tensor. The symmetry of the tensor gives the equation jE = v2π, and then it leads to an equality between the energy magnetization M µν Ei/2 and the angular momentum Lµν = hxµπν − xνπµi as E = hxµjν E − xνjµ M µν E = v2Lµν/2, (14) E and Lz = L12. Here h· · ·i indicates evaluating an operator in thermal where M z equilibrium at temperature T . Substituting the equation (12) and (14) into the definition of the thermal Hall coefficient, we obtain E = M 12 κH = − jx E ∂yT v2 2 ∂Lz ∂T = − v2 2T ∂Lz ∂φ , = − (15) where in the last equality of (15), the relation (3) is used. The angular momentum is the conjugate field of the angular velocity Ω, and is given by variation of the free energy as Lz = −δF/δΩz. The free energy functional for the thermal Hall effect is given by F [φ, Ω] = −Z d2xLzΩ =Z d2x 2κHT v2 φΩ. (16) Corresponding electromagnetic free energy is the electromagnetic Chern-Simons term with gauge fixing F EM ∝ σHR d2xA0Bz. The free energy F EM represents an quantized electric Hall response, that is, an electric current is induced by a magnetization given by M z = −δF EM/δBz. 2.2. Two-dimensional Dirac fermion under gravitational field The Dirac fermion with the Fermi velocity v is invariant under the Lorentz transformation which preserves the line element ds2 = (vdt)2 − dx2 − dy2. (17) Introducing the three coordinates xµ = (vt, x, y), a metric representing a system under a gravitational potential φ rotating with an angular velocity Ω is given by[27] gµν =  1 + 2φ Ωy/v −Ωx/v Ωy/v −Ωx/v 0 −1 −1 0 ,   (18) Finite-temperature effective boundary theory of the quantized thermal Hall effect 6 up to linear order in φ and Ω. The effective field theory addressed in this paper is the free energy functional of the metric that results from the two-dimensional Dirac fermion coupled with the gravitational field. In the following, we proceed with a general metric form, and the metric (18) is substituted into the expression when the Tolman-Ehrenfest relation (3) is used to mimic temperature gradient. The two-dimensional Dirac fermion Hamiltonian coupled with the gravitational field is separated into the flat and the remaining parts as follows. Deviation of a curved space-time metric from the flat Minkowski space-time is defined by gµν = ηµν + hµν. (19) We use the sign convention ηµν = diag(+1,−1,−1). The metrics gµν and ηµν are related by a triad field eα µ with the equation gµν = eα µeβ νηαβ. (20) When a deviation is small enough (hµν ≪ 1), we can write the triad field in terms of h as eα µ/2. Subscripts and superscripts of hµν are raised and lowered by ηµν and its inverse ηµν, like hα µ = hνµηαν. The inverse and the determinant of the triad field are given, up to linear order in h, by µ ≃ δα µ + hα α /2, α − h µ α ≃ δµ e µ √g =pdet gµν = det(eα µ) ≃ 1 + h/2, µ. The inverse of the triad satisfies eα where h = hµ β , and β gµν = ηαβ. The gamma matrices on the curved space-time γµ satisfy the e µ α e ν anticommutation relation {γµ, γν} = 2gµν. γµ is related to the gamma matrices on the flat space-time γα by α γα, β = δα α = δν γµ = e µ µ, eα µe ν (23) µe µ where γα satisfies the relation {γα, γβ} = ηαβ. The action of the (2+1)-dimensional Dirac fermion coupled with a gravitational field is S =Z dtd2x√g ¯ψ(cid:2)ivγµ∇µ − m(cid:3) ψ. Here the covariant derivative ∇µ of the spinor field is given by ∇µψ = (∂µ−ωµ)ψ, where the connection is given by ωµ = −γαβω αβ µ /4, using γαβ = [γα, γβ]/2 with γα = ηαβγβ, and (21) (22) (24) (25) (26) ω αβ µ = eβ λ − Γρ The local Hamiltonian H is defined by νgνλ(∂µeα µλeα ρ). S =Z dtd2x√g ψ†γ0γ0 (i∂t − H) ψ. H =Z d2x ψ†√gγ0γ0Hψ Then the Hamiltonian of the two-dimensional Dirac fermion coupled with a gravitational field is written as Finite-temperature effective boundary theory of the quantized thermal Hall effect =Z d2x ψ†√g(cid:2)ivγ0γ0ω0 − ivγ0γj∇j + mγ0(cid:3) ψ ≃Z d2x ψ†[H0 + U]ψ. Latin indices in (27) run over the spatial dimensions (j = 1, 2) and Greek indices run over both temporal and spatial dimensions (α = 0, 1, 2). Matrix element of the unperturbed Hamiltonian is given by H0 = −ivγ0γj∂j + mγ0. (29) The perturbation Hamiltonian consists of two parts U = H1 + H2; one from variation of the triad and metric in (27), and the other from the spin connection. The matrix elements of the perturbation terms are, respectively, 7 (27) (28) (30) (31) H1 = (h/2)H0 + iv(h j H2 = −(iv/4)γ0γµγαγβ · ω(1) µ αβ, µ αβ is a part of ωµ αβ = ηαγηβδω γδ α /2)γ0γα∂j, where ω(1) that contains at most linear order in hα ν, and we have used ωµ αβ = −ωµ βα. Later the second perturbation term (31) will be dropped out since it does not contribute to the finite temperature free energy. Although the original Lagrangian has the Lorentz invariance, symmetries of the Hamiltonian (29) and (30) are reduced down to the SO(2) rotational invariance in the x1-x2 plane. Time- reversal symmetry and parity symmetry are broken by the mass term. µ 3. Effective theory at the boundary In the (2+1)-dimensional space-time, the Chern-Simons action is induced by the parity breaking mass term of the Dirac fermion. The Chern-Simons action of the U(1) gauge field is derived as an effective action by tracing out the Dirac fermionic degrees of freedom of the action of the (2+1)-dimensional Dirac fermion coupled with an electromagnetic field[28, 29, 30]. Similarly, the gravitational Chern-Simons action, the Chern-Simons action of the spin connection, is derived by tracing out the fermionic degrees of freedom of the action of the Dirac fermion coupled with the gravitational field[18, 19, 20]. When the theory of induced Chern-Simons terms is extended to a finite temperature system by considering the imaginary time, the induced theory gives the partition function and the free energy of the gauge fields. However, if the fermionic system is fully gapped, temperature dependent part of the free energy is exponentially suppressed by the factor e−βm, where β is the inverse temperature and m is the magnitude of the gap[31, 32]. The effective free energy of the quantized thermal Hall effect will not be given in this way, since the quantized thermal Hall conductivity is linearly dependent on temperature. Therefore in order to derive the finite-temperature effective free energy for the quantized thermal Hall effect, we, at first, consider the gapless boundary fermion, the existence of which is guaranteed by the nontrivial energy band topology of gapped fermionic systems. The effective free energy of the gravitational field for the one-dimensional boundary modes is calculated by the field theoretical method. Finite-temperature effective boundary theory of the quantized thermal Hall effect 8 This calculation procedure seems consistent with the fact that the boundary theory is legitimate to study the quantized thermal Hall effect[33]. For the purpose of deriving a finite temperature effective free energy, we consider the finite temperature path integral. Tracing out fermionic degrees of freedom of the density matrix gives an effective free energy functional of the gravitational field as exp(−βF [h]) =Z Dψ†Dψ exp(−βH[ψ†, ψ, h]) DetG−1(iωn), =Yn where G(iωn) = 1/(−iωn + H0 + U) is the temperature Green's function with the Matsubara frequency for fermions ωn = (n + 1/2)2π/β (n ∈ Z), and Det is taken for the Hilbert space on the two-dimensional space and 2 × 2 matrix degrees of freedom. Expanding the temperature Green's function with respect to U, we obtain (32) (33) lnG−1 = lnG−1 = lnG−1 0 + ln(1 + G0U) (G0U)l, 0 + 1 l ∞ Xl=1 Tr lnG−1 0 , − βF (0) =Xn l Xn − βF (l) = 1 where G0(iωn) = 1/(−iωn + H0) is the temperature Green's function in the flat space-time. Then expansion of the effective free energy with respect to hµν, that is F =P∞ l=0 F (l), is given by (34) Tr (G0U)l , (35) where the identity ln Det = Tr ln is used, and Tr is taken for the same Hilbert space as Det. F (0) is independent of the gravitational field and gives only a constant, and the second order perturbation term F (2) is the focus of this paper. 3.1. Boundary fermion In two-dimensional space, consider a boundary at x1 = 0 between a gapped bulk at x1 < 0 with mass m and that at x1 > 0 with mass −m. The boundary is extended to an entire x2 space. The masses in the two gapped semi-infinite regions are smoothly connected by introducing an x1-dependent mass term, whose sign changes at the boundary (sgn[m(x1 < 0)] = −sgn[m(x1 > 0)]). Since the Chern number at both side of the boundary differs by unity, the boundary hosts a single chiral boundary mode of the massless Dirac fermion. The unperturbed Hamiltonian (29) is decoupled into the x1- and the x2-dependent parts. The wave function of the boundary mode of the Hamiltonian (29) is a product of a plane wave of the x2-coordinate and a two-components spinor wave function of the x1-coordinate satisfying (cid:0)−ivγ0γ1∂1 + m(x1)γ0(cid:1) ψ0(x1) = 0. (36) Finite-temperature effective boundary theory of the quantized thermal Hall effect 9 Formally the solution of (36) is given by ψ0(x1) = hx1ψ0i = 1 √N exp"iγ1v−1Z x1 0 dx′1 m(x′1)#si, (37) where we have used an identity (γ1)2 = η11 = −1, and the normalization constant is defined by N = R dx1ψ† 0(x1)ψ0(x1) = hψ0ψ0i. Here, x1i is an eigenstate vector of the position operator x1 in the Hilbert space on the x1-coordinate, and ψ0i is the boundary state vector of the Hilbert space on x1-coordinate with the spinor degrees of freedom. The two component spinor si corresponding to edge bound states satisfies iγ1si = sgn(m)si, where sgn(m) indicates the sign of the mass in x1 < 0, while the spinor satisfying iγ1si = −sgn(m)si corresponds to states that cannot be normalized. By using the relation γµγν = ηµν − iǫµνργρ, the condition on si can be rewritten in a convenient form, γ0γ2si = −sgn(m)si. Next, we consider the Hamiltonian of the gapless boundary states resulting from the bulk Hamiltonian by projecting the Hilbert space on the x1-coordinate onto that of the boundary mode. The projected Hamiltonian is given by defining the projection opeartor P = ψ0ihψ0 as P †HP . In the following, the x2-dependent part H ≡ hψ0P †HPψ0i = hψ0Hψ0i of the projected Hamiltonian is referred to as the boundary Hamiltonian. The boundary Hamiltonian for the unperturbed bulk Hamiltonian (29) is H0 ≡ hψ0H0ψ0i = hs − ivγ0γ2∂2si = iv sgn(m)∂2. (38) Also we consider the projection of the perturbation term (30) onto the boundary mode. For convenience, the width of the boundary states in the x1-direction is tuned to be narrow enough so that typical length scale of the gravitational field is much longer than the width of the boundary states. This situation allows us to use an assumption that the metric depends only on x2 near the boundary. Then x1 and x2 is completely decoupled in the boundary Hamiltonian. The boundary Hamiltonian for the second term of the perturbation term (30) is hψ0iv(h j α /2)γ0γα∂jψ0i = −hψ0γ0γαγ1m(x1)ψ0ivh 1 α (x2)∂2/2. + hsγ0γαsiivh 2 The first term in (39) is zero since it contains an integral α (x2)/2 (39) −∞ Z ∞ ∝ exp"2v−1sgn(m)Z x1 dx1m(x1) exp"2v−1sgn(m)Z x1 dx′1 m(x′1)#(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) 0 0 dx′1 m(x′1)# x1=∞ x1=−∞ = 0. (40) The second term in (39) is nonzero for α = 0, 2 due to the property of the two-component spinor si. The boundary Hamiltonian for the perturbation term is written as H1 ≡ hψ0H1ψ0i = ζ(x2)(−iv∂2), (41) Finite-temperature effective boundary theory of the quantized thermal Hall effect 10 where ζ(x2) = [−sgn(m)h(x2)+sgn(m)h 2 for second perturbation term (31) 2 (x2)−h 2 0 (x2)]/2. The boundary Hamiltonian H2 ≡ hψ0H2ψ0i = −(iv sgn(m)/4)hω(1) ≡ χ(x2). 0 20 + ω(1) 1 12i − (iv/4)hω(1) 2 20i 1 01 − ω(1) (42) 3.2. Effective boundary free energy Here we calculate the effective free energy functional of the gravitational field of the boundary mode. Tracing out the boundary fermionic degrees of freedom, the second order perturbation term in (35) is given by − βF bdry = 1 2Xn Z dx2 hx2 G0 U G0 Ux2i, where G0 = 1/(−iωn + H0) is the temperature Green's function of the boundary fermion, and U = H1 + H2. Tracing over the Hilbert space on the x2-coordinate, the effective free energy becomes − βF bdry =Z dq 2π Z dx2dx′2eiq(x2−x′ 2)hΠ(2)(q)ζ(x2)ζ(x′2) +2Π(1)(q)ζ(x2)χ(x′2) + Π(0)(q)χ(x2)χ(x′2)i , where q is the momentum of the gravitational field, and Π(s)(q) is the one-loop integral of the single-component boundary fermion given by (43) (44) Π(s)(q) = vs 2 Xn Z dp × −iωn + vp 2 (cid:19)s 2π(cid:18) 2p + q −iωn + v(p + q) 1 1 . (45) The leading contribution of the integral over q in (44) is of order q0, where we expand the integral in the powers of q since the long-wave length behavior of the gravitational field is of interest. Summation over the Matsubara frequencies gives Pn(−iωn + vp)−2 = βdf (vp)/d(vp), where f (ǫ) is the Fermi-Dirac distribution function at temperature T . The coefficient (45) of the order of q0 is given by Π(s)(0) = ps. (46) β 2vZ dp 2π df (p) dp Here we have replaced vp by p for simplicity. Obviously Π(1)(0) = 0, since df (p)/dp is an even function of p. Furthermore, Π(0)(0) = −β/4πv can be neglected in this study since the free energy proportional to Π(0)(0) is independent of temperature when substituted into (44), which indicates that this term is not relevant to the quantized thermal Hall effect. The only coefficient that is concerned is Π(2)(0) = β v Z dp 2π [θ(−p) − f (p)] p, (47) Finite-temperature effective boundary theory of the quantized thermal Hall effect 11 where θ is the Heaviside step function, and is regarded as the zero-temperature Fermi-Dirac distribution function. At low temperature, the distribution function in (47) can be expanded with respect to the temperature by the Sommerfeld expansion f (p) ≃ θ(−p) − (π2T 2/6)(dδ(p)/dp) as π2T 2 dδ(p) β (48) Then the effective free energy is reduced to a simple form as p = −β πT 2 12v . 6 dp Π(2)(0) ≃ F bdry = = πT 2 2π v Z dp 12v Z ∞ 48v Z ∞ −∞ −∞ πT 2 dx2[ζ(x2)]2 dx2[−sgn(m)h + sgn(m)h2 2 − h2 0]2. (49) 3.3. Bulk thermal Hall conductivity E The energy current jbdry flowing along the boundary can be read off from the boundary effective free energy (49). Using an equivalence between the energy current and the momentum density resulting from the Lorentz invariance, the boundary energy current is jbdry E ≡ vT 02 ≃ 2v(δF bdry/δh20), which results in [−sgn(m)h + sgn(m)h2 jbdry E = πT 2 12 2 − h2 0]. (50) The Streda formula for the quantized thermal Hall effect, the form of which is given in (10), implies that the thermal Hall conductivity is given by ratio of the variation of the energy magnetization and that of temperature. Thus we consider a two-dimensional system with the boundary under a spatially homogeneous gravitational potential φ, and see how the bulk energy magnetization changes according to the change of the gravitational potential homogeneously. For this purpose, only the first term of the right-hand side of (50) is considered since this term is proportional to the gravitational potential φ. The energy magnetization is defined by jE = ∂ × ME. (51) Here, we assume the energy magnetization is uniformly distributed in the bulk as a response of the uniform bulk gravitational potential φ. This assumption is same as the one used in the phenomenological theory, that is, the energy magnetization depends only on the temperature, which is noted just above (10). Since uniform energy magnetization leads to vanishing bulk energy current due to (51), the energy magnetization in the bulk can be evaluated entirely by the boundary energy current. The bulk energy magnetization can be derived from the boundary energy current by the following argument. Consider a boundary at x1 = 0 separating the bulk region with mass m (x1 < 0) and outside region with mass −m (x1 > 0). The bulk energy magnetization at the region with mass m and that with mass −m have opposite sign since the energy magnetization changes its sign by parity transformation, which is same as the orbital magnetization in the quantum Hall system. The boundary energy current Finite-temperature effective boundary theory of the quantized thermal Hall effect 12 is defined as the energy current flowing in x2-direction in the boundary region where the energy magnetization changes from bulk value to one in the outer region. Then the energy magnetization M z in the bulk x < 0 is given in terms of the boundary energy current as jbdry dx1j2 E(x1) −∞ E ≡Z ∞ =Z ∞ = −(cid:0)M z = 2M z E, −∞ E(x1)) dx1(−∂1M z E(x1 = +∞) − M z E(x1 = −∞)(cid:1) which results in M z E = −sgn(m) πT 2 24 h. (52) (53) Substituting h = hµ µ = 2φ and using the Tolman-Ehrenfest relation (3), the thermal Hall conductivity is then given by the Streda formula for the quantized thermal Hall effect: κH = − 1 T ∂M z E ∂φ = sgn(m) πT 12 , (54) which is the quantized thermal Hall conductivity for the Chern number C = sgn(m)/2. 4. Bulk physical quantities In this section, we elaborate to make a connection between the boundary effective theory and the bulk quantities of the thermal Hall effect, that is, the thermal Hall current (5) and the bulk effective free energy (6). The expression of the bulk thermal Hall current is deduced from the boundary energy current (50), and we conjecture a possible form of the bulk effective free energy which creates a bulk quantized thermal Hall current. It should be noted that while the bulk thermal Hall conductivity (54) is a direct consequence of the boundary effective free energy (49), results in this section may be regarded as somehow indirect consequence from the boundary theory, since it is unclear whether or not, at the thermal equilibrium, the thermal Hall current flows in the bulk or only along the boundary in a fully gapped two-dimensional insulator[32, 33, 34]. 4.1. Bulk thermal Hall current The expression of the bulk energy current can be obtained by considering a two- dimensional system with the boundary under spatially inhomogeneous gravitational potential. If the boundary is considered as an isolated one-dimensional system obeying the free energy (49), the energy current (50) can break the energy conservation law when the metric varies spatially: ∂2jbdry 6= 0. Here, note that the boundary energy current is not dependent on time, since the boundary free energy (49) and responses derived from it are static. The energy conservation law is retained by including incoming energy flow E Finite-temperature effective boundary theory of the quantized thermal Hall effect 13 from the gapped bulk. The bulk effective free energy is determined to precisely describe this compensating bulk energy current. The incoming energy current from one side of the semi-infinite space of the bulk is equal to the other side, since the mass term is inverted by the time-reversal operation or the parity inversion, which causes inversion of the thermal Hall coefficient. Thus, the bulk energy current jE near the boundary satisfies the following equation: E(x1 = −0) = −j1 j1 E(x1 = +0) = ∂2jbdry (55) 1 2 E . E In the following, we show that only the first term of the right-hand side of (50) contributes to jbdry in (55) by taking into account that the bulk energy current, j1 E in (55), should not be dependent on details of the boundary, since it results from the nontrivial topology of the bulk energy bands. To see this, we examine how each term in (49) changes according to the direction of the boundary. The system is rotated by an angle θ anticlockwise to define a new boundary. New coordinate variables perpendicular and parallel to the boundary are introduced as ( x′1 = x1 cos θ + x2 sin θ x′2 = −x1 sin θ + x2 cos θ ⇔ x′j = Rj k(θ)xk, (56) where Rj k(θ) is the rotation matrix. Derivatives in the new coordinates are given by ∂′ j = R k j (θ)∂k, (57) j (θ) is the inverse of Rj where R k l holds). The Hamiltonian (29) and (30) has SO(2) rotational symmetry, that is, introducing new vectors and tensors by γ′j = Rj j(−θ), and R k j (θ) = Rk k(θ) (R k l(θ) = δk j (θ)Rj (58) k(θ)γk, and h′j h′j 0(x) = Rj k(x) = Rj l(θ)hl l(θ)R m 0(x), k (θ)hl m(x), (60) the Hamiltonian is a scalar quantity with respect to the rotation Rj k(θ): H0[γ′, h′; x′] = H0[γ, h; x] and U[γ′, h′; x′] = U[γ, h; x]. Note that γ0 and other scalar quantities contained in the Hamiltonian are unchanged during the rotation. Then the effective free energy of the new boundary perpendicular to x′1 is given by (59) F bdry′ = πT 2 12v Z ∞ −∞ dx′2[ζ ′(x′2)]2. (61) Here, ζ ′(x′2) = [−sgn(m)h(x′2) + sgn(m)h′2 energy current flowing along the x′2-coordinate is [−sgn(m)h + sgn(m)h′2 πT 2 12 jbdry E = ′ 2(x′2) − h′2 0(x′2)]/2, and thus the boundary 2 − h′2 0]. (62) The second and third terms of the right-hand side of (62) explicitly depend on the boundary angle θ through (59) and (60), while the first term does not. The bulk Finite-temperature effective boundary theory of the quantized thermal Hall effect 14 energy current, which is proportional to the Chern number of the bulk energy band, should not be dependent on the details of the boundary, such as the boundary angle. Therefore the first term of the right-hand side of (62) can be regarded as a term related to the bulk energy current by (55). The same conclusion can be given by considering the number of indices in (55). The right-hand side of the (55) must be a vector quantity with respect to rotation since the left-hand side is a vector quantity. This condition is satisfied when jbdry is a scalar quantity, which results in the first term of the right-hand side of (62). The bulk energy current corresponding to the boundary energy current jbdry E = −sgn(m)(πT 2/12)h is given by ǫkl∂lh, (63) E jk E = −sgn(m) πT 2 24 the form of which is independent of the direction of the boundary. The equation (63) exactly represents the quantized thermal Hall effect (5) by substituting h = 2φ from the metric (18), identifying the energy current with the thermal current, and using the Tolman-Ehrenfest relation (3). The thermal Hall coefficient is given by κH = − 1 T jx E ∂yφ = sgn(m) πT 12 , which is quantized by the Chern number C = sgn(m)/2. (64) 4.2. Effective bulk free energy The bulk free energy for the quantized thermal Hall effect can be deduced so that it realizes the energy current (63) as a gravitational response, and is given by F [h] = −sgn(m) πT 2 48v Z d2x ǫklhk0∂lh. (65) A similar form of the bulk gravitational free energy has also been predicted in Ref. [35]. The free energy (65) is regarded as a static and Lorentz invariant version of one in Ref. [35]. Total energy conservation can be seen by the translation symmetry of the bulk free energy together with the boundary free energy in the temporal direction. Under an infinitesimal coordinate transformation x′0 → x0 + ζ 0, (66) the metric varies as δhk0 = ∂kζ0, where we assume ζ 0 is a function of x1 and x2 since we consider a static theory in this study. The translation (66) leaves the bulk free energy (65) on a region D unchanged except the boundary term on ∂D given by δF [h] = −sgn(m) = −sgn(m) πT 2 48v ZD 48v Z∂D πT 2 d2x ǫkl(∂kζ0)(∂lh) dl · (ζ0∂h). (67) Finite-temperature effective boundary theory of the quantized thermal Hall effect 15 On the other hand, by defining the boundary parallel to x2 and the bulk in x1 < 0, the translation (66) makes a change in the boundary free energy as δ(cid:16) F bdry/2(cid:17) = δ(cid:18)−sgn(m) πT 2 48v Z dx2 hh20(cid:19) = sgn(m) πT 2 48v Z dx2 ζ0∂2h, (68) (69) which precisely cancels the boundary term in (67). Therefore the total free energy is invariant under the translation (66), and it indicates the energy of the bulk and the boundary is totally conserved. The boundary free energy F bdry considered here is a part of (49) that represents the scalar boundary energy current shown above (63), while the other terms are irrelevant to the variation (66) (h2, hh2 2)2) or to the bulk thermal Hall current ((h2 0). Furthermore the boundary free energy in the parenthesis of the left-hand side of (68) is a half of it, since the other half corresponds to the other side of the bulk (x1 > 0). 0)2 and h2 2, and (h2 2h2 In the presence of a gravitational potential and a rotation (18), h and hk0 in (65) are given by h = 2φ and hk0 = (Ωx2/v,−Ωx1/v). Therefore the integrand of (65) together with that of half of the boundary free energy is ǫklh∂khl0 = −4φΩ/v. The effective free energy under the metric (18) is given by F [φ, Ω] + F bdry[φ, Ω]/2 = −sgn(m) πT 2 48v Z d2x ǫklh∂khl0 = κH T v2 Z d2x φΩ. The free energy (69) realizes the coupling of the gravitational potential and the angular momentum, which is the same form as the phenomenologically derived free energy presented in (6). However, it turns out that, from the field theoretical method, the coefficient in (69) is half of that predicted in [24]. 5. Discussion Using the finite-temperature field theoretical method, we have derived microscopically a boundary effective free energy functional of the gravitational field, and show that this free energy accounts for the quantized thermal Hall effect of a gapped fermionic system in two-dimension. This result gives a novel insight into the effective theory of the quantized thermal Hall effect that the boundary effective theory, derived microscopically from the gapped fermionic theory, is sufficient to account for the bulk quantized thermal Hall conductivity, unlike previous studies of bulk effective theories for the quantized thermal Hall effect and the case of the quantum Hall effect. However, connecting the boundary effective theory with the expression of the bulk thermal Hall current and the bulk effective theory requires further speculation. After making this connection, our boundary effective theory is shown to be consistent with some of previous bulk effective theories[24, 35]. Finite-temperature effective boundary theory of the quantized thermal Hall effect 16 The method we have used to study the effective theory is valid for a system in thermal equilibrium. Thus the thermal current and transport coefficient derived from our effective theory are considered as those for equilibrium responses. Obviously, chiral boundary energy current in a topological insulator can flow even in thermal equilibrium as the boundary electric current does. Thus there is no doubt that the quantized thermal Hall coefficient in the bulk can be derived from our boundary theory, since it depends only on the boundary energy current. On the other hand, it is not yet clear whether the bulk thermal Hall current flows in thermal equilibrium or not. In our study, however, we have also given the expression of the bulk thermal Hall current within the equilibrium thermal response. 6. Conclusion In this study, we have considered the Hamiltonian of the two-dimensional massive Dirac fermion under a static gravitational field which contains the gravitational potential mimicking temperature gradient, and the angular velocity coupling to the momentum density. We place this model in a semi-infinite plane with boundary, and consider thermal equilibrium at a finite temperature. At first, the bulk Hamiltonian is projected onto the boundary. Then the boundary free energy is calculated from the chiral gapless boundary fermion by tracing out the fermionic degrees of freedom and perturbatively expanding the free energy with respect to the metric. We have shown that the gravitational response of the boundary free energy at the boundary determines the bulk energy magnetization, and it explains the quantized thermal Hall conductivity through the Streda formula. Furthermore, the bulk energy current is evaluated from the boundary energy current to ensure that the total energy is conserved in the whole system. The resulting bulk energy current precisely represents the thermal Hall effect. The effective free energy functional (65) for the two-dimensional massive Dirac fermion is then deduced to realize the bulk energy current. The bulk effective free energy reproduces the same form as the phenomenologically derived effective free energy (6) by substituting a metric of a gravitational potential in a rotating frame. Acknowledgments This work was supported by World Premier International Research Center Initiative (WPI) and Grant-in-Aid for Scientific Research (No. 25103703, No. 26107505 and No. 26400308) from MEXT, the NSF under Grant No. DMR-1455296, and Alfred P. Sloan foundation. [1] M. Z. Hasan and C. L. Kane. Colloquium: Topological insulators. Rev. Mod. Phys., 82:3045, 2010. [2] Xiao-Liang Qi and Shou-Cheng Zhang. Topological insulators and superconductors. Rev. Mod. Phys., 83:1057, 2011. [3] R. E. Prange and S. M. Girvin, editors. The Quantum Hall Effect. Springer, 1987. Finite-temperature effective boundary theory of the quantized thermal Hall effect 17 [4] D. J. Thouless, M. Kohmoto, M. P. Nightingale, and M. den Nijs. Quantized hall conductance in a two-dimensional periodic potential. Phys. Rev. Lett., 49:405, 1982. [5] N. Read and Dmitry Green. Paired states of fermions in two dimensions with breaking of parity and time-reversal symmetries and the fractional quantum hall effect. Phys. Rev. B, 61:10267, 2000. [6] Andreas P. Schnyder, Shinsei Ryu, Akira Furusaki, and Andreas W. W. Ludwig. Classification of topological insulators and superconductors in three spatial dimensions. Phys. Rev. B, 78:195125, 2008. [7] Alexei Kitaev. Periodic table for topological insulators and superconductors. AIP Conf. Proc., 1134:22, 2009. [8] Shinsei Ryu, Andreas P Schnyder, Akira Furusaki, and Andreas W W Ludwig. Topological tenfold way and dimensional hierarchy. New J. Phys., insulators and superconductors: 12:065010, 2010. [9] R. Bertlmann. Anomalies in Quantum Field Theory. Oxford University Press, Oxford, 1996. [10] K. Fujikawa and H. Suzuki. Path Integrals and Quantum Anomalies. Oxford University Press, Oxford, 2004. [11] Xiao-Liang Qi, Taylor L. Hughes, and Shou-Cheng Zhang. Topological field theory of time-reversal invariant insulators. Phys. Rev. B, 78:195424, 2008. [12] Zhong Wang, Xiao-Liang Qi, and Shou-Cheng Zhang. Topological field theory and thermal responses of interacting topological superconductors. Phys. Rev. B, 84:014527, 2011. [13] Shinsei Ryu, Joel E. Moore, and Andreas W. W. Ludwig. Electromagnetic and gravitational responses and anomalies in topological insulators and superconductors. Phys. Rev. B, 85:045104, 2012. [14] Taylor L. Hughes, Robert G. Leigh, and Eduardo Fradkin. Torsional response and dissipationless viscosity in topological insulators. Phys. Rev. Lett., 107:075502, 2011. [15] Andrey Gromov and Alexander G. Abanov. Thermal hall effect and geometry with torsion. Phys. Rev. Lett., 114:016802, 2015. [16] Barry Bradlyn and N. Read. Low-energy effective theory in the bulk for transport in a topological phase. Phys. Rev. B, 91:125303, 2015. [17] Luis Alvarez-Gaum´e and Edward Witten. Gravitational anomalies. Nucl. Phys. B, 234:269, 1984. [18] I. Vuorio. Parity violation and the effective gravitational action in three dimensions. Phys. Lett. B, 175:176, 1986. [19] M. A. Goni and M. A. Valle. Massless fermions and (2+1)-dimensional gravitational effective action. Phys. Rev. D, 34:648, 1986. [20] Shuichi Ojima. Derivation of gauge and gravitational induced chern-simons terms in three dimensions. Prog. Theor. Phys., 81:512, 1989. [21] Michael Stone. Gravitational anomalies and thermal hall effect in topological insulators. Phys. Rev. B, 85:184503, 2012. [22] J. M. Luttinger. Theory of thermal transport coefficients. Phys. Rev., 135:A1505, 1964. [23] Tao Qin, Qian Niu, and Junren Shi. Energy magnetization and the thermal hall effect. Phys. Rev. Lett., 107:236601, 2011. [24] Kentaro Nomura, Shinsei Ryu, Akira Furusaki, and Naoto Nagaosa. Cross-correlated responses of topological superconductors and superfluids. Phys. Rev. Lett., 108:026802, 2012. [25] Hiroaki Sumiyoshi and Satoshi Fujimoto. Quantum thermal hall effect in a time-reversal- symmetry-broken topological superconductor in two dimensions: Approach from bulk calculations. J. Phys. Soc. Jpn., 82:023602, 2013. [26] P Streda. Theory of quantised hall conductivity in two dimensions. J. Phys. C: Solid State Phys., 15:L717, 1982. [27] L. D. Landau and E. M. Lifshitz. The Classical Theory of Fields. Butterworth-Heinemann, 1975. [28] A. J. Niemi and G. W. Semenoff. Axial-anomaly-induced fermion fractionization and effective gauge-theory actions in odd-dimensional space-times. Phys. Rev. Lett., 51:2077, 1983. Finite-temperature effective boundary theory of the quantized thermal Hall effect 18 [29] A. N. Redlich. Gauge noninvariance and parity nonconservation of three-dimensional fermions. Phys. Rev. Lett., 52:18, 1984. [30] A. N. Redlich. Parity violation and gauge noninvariance of the effective gauge field action in three dimensions. Phys. Rev. D, 29:2366, 1984. [31] K. S. Babu, Ashok Das, and Prasanta Panigrahi. Derivative expansion and the induced chern- simons term at finite temperature in 2+1 dimensions. Phys. Rev. D, 36:3725, 1987. [32] Juan L. Manes and Manuel Valle. Parity odd equilibrium partition function in 2 + 1 dimensions. J. High Energy Phys., 11:178, 2013. [33] C. L. Kane and Matthew P. A. Fisher. Quantized thermal transport in the fractional quantum hall effect. Phys. Rev. B, 55:15832, 1997. [34] L Smrcka and P Streda. Transport coefficients in strong magnetic fields. J. Phys. C: Solid State Phys., 10:2153, 1977. [35] Atsuo Shitade. Heat transport as torsional responses and keldysh formalism in a curved spacetime. Prog. Theor. Exp. Phys., page 123I01, 2014.
1612.09363	1	1612	2016-12-30T01:14:09	Study of 0-$\pi$ phase transition in hybrid superconductor-InSb nanowire quantum dot devices	[ "cond-mat.mes-hall" ]	Hybrid superconductor-semiconducting nanowire devices provide an ideal platform to investigating novel intragap bound states, such as the Andreev bound states (ABSs), Yu-Shiba-Rusinov (YSR) states, and the Majorana bound states. The competition between Kondo correlations and superconductivity in Josephson quantum dot (QD) devices results in two different ground states and the occurrence of a 0-$\pi$ quantum phase transition. Here we report on transport measurements on hybrid superconductor-InSb nanowire QD devices with different device geometries. We demonstrate a realization of continuous gate-tunable ABSs with both 0-type levels and $\pi$-type levels. This allow us to manipulate the transition between 0 and $\pi$ junction and explore charge transport and spectrum in the vicinity of the quantum phase transition regime. Furthermore, we find a coexistence of 0-type ABS and $\pi$-type ABS in the same charge state. By measuring temperature and magnetic field evolution of the ABSs, the different natures of the two sets of ABSs are verified, being consistent with the scenario of phase transition between the singlet and doublet ground state. Our study provides insights into Andreev transport properties of hybrid superconductor-QD devices and sheds light on the crossover behavior of the subgap spectrum in the vicinity of 0-$\pi$ transition.	cond-mat.mes-hall	cond-mat	Study of 0-π phase transition in hybrid superconductor-InSb nanowire quantum dot devices S. Li,1 N. Kang,1, ∗ P. Caroff a),2 and H. Q. Xu1, 3, † 1Key Laboratory for the Physics and Chemistry of Nanodevices and Department of Electronics, Peking University, Beijing 100871, China 2I.E.M.N., UMR CNRS 8520, Avenue Poincar´e, BP 60069, F-59652 Villeneuve d'Ascq, France 3Division of Solid State Physics, Lund University, Box 118, S-221 00 Lund, Sweden (Dated: July 2, 2021) Abstract Hybrid superconductor-semiconducting nanowire devices provide an ideal platform to investi- gating novel intragap bound states, such as the Andreev bound states (ABSs), Yu-Shiba-Rusinov (YSR) states, and the Majorana bound states. The competition between Kondo correlations and superconductivity in Josephson quantum dot (QD) devices results in two different ground states and the occurrence of a 0-π quantum phase transition. Here we report on transport measurements on hybrid superconductor-InSb nanowire QD devices with different device geometries. We demon- strate a realization of continuous gate-tunable ABSs with both 0-type levels and π-type levels. This allow us to manipulate the transition between 0 and π junction and explore charge transport and spectrum in the vicinity of the quantum phase transition regime. Furthermore, we find a coexistence of 0-type ABS and π-type ABS in the same charge state. By measuring temperature and magnetic field evolution of the ABSs, the different natures of the two sets of ABSs are veri- fied, being consistent with the scenario of phase transition between the singlet and doublet ground state. Our study provides insights into Andreev transport properties of hybrid superconductor-QD devices and sheds light on the crossover behavior of the subgap spectrum in the vicinity of 0-π transition. a)Present address: Department of Electronic Materials Engineering, Research School of Physics and En- gineering, The Australian National University, Canberra, ACT 0200, Australia 1 I. INTRODUCTION In hybrid structures where a superconductor (S) is connected to a mesoscopic normal conductor (N), an electron in the normal region is converted to a Cooper pair into the superconductor with the reflection of a hole. This well established mechanism, known as the Andreev reflections, plays a central role in the proximity effect in which the electron pairing potential in S can penetrate into N region. During the past years, based on the idea of introducing superconducting order into low-dimensional semiconductors via the proxim- ity effect, superconducting leads coupled semiconductor quantum dots (QDs) have been extensively investigated for novel quantum phenomena and new device concepts1, such as tunable Josephson junctions2 -- 6 and Cooper-pair splitters7 -- 10, and such hybrid S-QD struc- tures provide an ideal platform to study basic physical issues including the formation of Andreev Bound States11 -- 16, the interplay between the Kondo effect and the proximity in- duced superconductivity17 -- 28, as well as the search for Majorana fermions in solid state29,30. Andreev Bound States (ABS), originating from the superposition of coherent Andreev reflec- tion processes, have been attracting growing interests owing to its fundamental importance in mesoscopic superconductor related systems, and have been observed in both S-QD-S13,14and S-QD-N12,31,32 systems in the tunneling regime. In superconducting leads coupled QDs, the spectrum of ABSs can be dramatically influenced by Coulomb interaction and spin-related many-body Kondo correlation. Depending on the ratio of the corresponding energy scales kBTK/∆ where TK is the Kondo temperature and ∆ is the superconducting gap, a 0-π Josephson junction transition has been predicted theoretically15 -- 17 and observed experimen- tally by either the sign change of the Josephson supercurrent33 -- 36 or the crossing behavior of the ABSs12,27,32. When kBTK/∆ ≪ 1, the Kondo screening is suppressed due to the lack of quasiparticle DOS around the Fermi level, hence the junction is in a π state with a crossing of the two Andreev levels. For kBTK/∆ ≫ 1, a Kondo singlet forms by breaking cooper pairs at the Fermi level, thus the junction is in a 0 state and the two Andreev levels never cross. Since these two magnetic states - a doublet for the π state and a singlet for the 0 state - are two different ground states (GSs) of the QD with different spins, a quantum phase transition is expected to be induced by a change of parameters in the QD device, such as the chemical potential and the charging energy. InSb NWs own a giant g factor39 and strong spin-orbit interaction strength39 -- 41. Hybrid 2 superconductor-InSb NW devices are expected to exhibit rich physics of subgap states due to the influence of strong spin-orbit coupling and Zeeman splitting in the presence of mag- netic fields. Recently, zero-energy states have been observed in hybrid superconductor-InSb NW devices and been interpreted as Majorana Bound States42 -- 44, though under certain cir- cumstances ABSs could behave as imitations of Majorana zero-energy states, yet having no relations with topological superconducting phases. Therefore to clarify the mechanism of these subgap bound states in such systems, a systematic study of the ABSs in hybrid superconductor-InSb NWQD devices is of fundamental importance, and is still lacking. In this work we report an extensive study on the low-temperature transport measurements of hybrid superconductor-InSb NWQD devices with different device geometries, i.e. the InSb nanowire QD(NWQD)-SQUID device and the S-NWQD-S device. In the NWQD-SQUID device we observe continuous gate-tuned ABSs with both 0-type levels and π-type levels and study the evolution of the ABSs as a function of temperature, showing a tunneling transport mechanism assisted by the thermally populated ABSs. In the S-NWQD-S device, we demonstrate that two sets of 0-type and π-type ABSs overlap at the same Kondo region, though a doublet ground state is favored given the estimated Kondo temperature. We propose a possible scenario for the simultaneous emergence of the 0-type and π-type ABSs and this scenario can be supported by the temperature and magnetic field evolution of the ABSs. II. EXPERIMENTAL DETAILS The hybrid superconductor-InSb nanowire QD devices are fabricated from individual InSb segments of InAs/InSb heterostructure NWs grown on an InP substrate by MBE method52,53, with typical diameters ranging from 60 to 120 nm. After growth, the NWs are transferred onto a degenerately doped, n-type Si substrate (used as a global backgate), covered by a 110-nm thick thermal oxide. The superconducting electrodes (5 nm/100 nm Ti/Al bilayer) are fabricated using standard e-beam lithography and e-beam evaporation procedure. Before the metal deposition, the samples are etched in a diluted (NH4)2Sx solution to remove the native oxide layer. The Ti layer is used as an adhesion layer between the superconducting Al layer and the nanowire. All measurements are performed in a 3He/4He dilution refrigerator with a base temperature of T ∼ 10 mK. In order to minimize the electronic noise, we used a series of π-filters, copper-powder filter and RC filters thermalizing at different temperature 3 stages The magnetic field is perpendicularly applied to the sample plane. III. ANDREEV BOUND STATES IN NWQD-SQUID DEVICES A. Gate-tuned different types of ABSs in NWQD-SQUID devices. Fig. 1(a) shows a scanning electron microscope (SEM) picture of a typical InSb NWQD- SQUID device and a schematic of the measurement circuit. This geometry is commonly used to improve the energy resolution for probing subgap resonances of the QDs due to the sharp change in BCS density of states at the superconducting gap edges13,14,26,31,37. In our case, as can be deduced from the Coulomb diamond (see section I of supplementary material), the central electrode is weakly coupled to the nanowire and plays the role of probing the DOS of levels in the nanowire. The top panel of Fig. 1(d) shows the differential conductance dI/dV in the superconducting state as a function of bias voltage Vsd and backgate voltage Vg, acquired at base temperature and zero magnetic field. We can clearly see two main features over the entire Vg range: a pair of main resonances at higher bias voltages with high conductance (◦) and a pair of weak resonances with low conductance (⋄) lying parallel to the main resonance. We attribute the main resonances to the alignment of the ABS levels at ±εa with the singularity in the BCS density of states of quasiparticles, i.e. at bias voltages: eVsd = ±(εa + ∆) (see Fig. 1(b)). The weak resonances can be interpreted as a replica of the main ABSs when the residual quasiparticle DOS of the probe at the Fermi level is aligned with the ABS levels: eVsd = ±εa 37. In particular, we can readily identify distinct gate voltage evolution of the ABSs res- onances in charge states of different parity determined by alternate size of the Coulomb diamond patterns. In regions with even electron occupation, the main ABSs resonances appear to overlap with the elastic quasiparticle cotunneling at eVsd = ±2∆ ≈ ±300 m eV, showing little dependence on gate voltage. This denotes that the ABSs are pushed towards the superconductor gap edge in the even states. Fig. 1(e) shows a dI/dV (Vsd) trace taken at the center of the even valley indicated by the orange dashed line in Fig. 1(d). The positions of peaks in dI/dV indicate the superconducting gap, ∆ = 1.76kBTc ≈ 150 m eV, which correspond with a critical temperature of Tc ≈ 1 K. In contrast, in the odd charge states, the shape of ABSs peaks is evidently dependent on the gate voltage and shows either crossing (blue arrow) or noncrossing (green arrow) behavior within different odd-occupied 4 Coulomb valleys. Fig. 1(f) shows a dI/dV (Vsd) trace taken at the center of the odd valley indicated by the light-blue dashed line in Fig. 1(d). A differential conductance peak appears symmetrically at finite bias voltages inside the superconducting gap, V < 2∆, and a nega- tive differential conductance (NDC) dip develops at bias voltages near ±2∆. The presence of NDC can be explained by the asymmetric coupling between the QD and two supercon- ducting leads, playing the role of probing ABSs27,45. The characteristic subgap structures have been reported in hybrid superconductor-QD devices in the odd occupation regions12,46, which can be ascribed to spin-induced subgap states forming in the QD energy spectrum. As previous works have studied17,19 -- 27, these different gate-dependent behaviors of ABSs (crossing or noncrossing) in the odd charge states can be understood as a competition between different energy scales, namely, Coulomb interaction U, QD energy level ξd, and hybridization Γ to superconducting leads. The phase diagram is displayed in Fig. 1(c). In the kBTK/∆ ≫ 1 limit, the ground state of the system is always a singlet. Thus the system stays in 0 state and the ABS levels demonstrate a noncrossing behavior. In the opposite limit of kBTK/∆ ≪ 1, the system alters its ground state from a singlet to a magnetic doublet as the ABS levels cross at zero energy. By sweeping gate voltages, the strength of the coupling between the leads and the QD can be tuned, resulting in different Kondo temperature TK. As approaching the singlet-doublet phase boundaries, the position of ABS energy εa moves towards zero, and a 0-π quantum phase transition is expected to occur. The bottom panel of Fig. 1(d) displays gate voltage dependence of the zero-bias conductance. It can be readily seen that the conductance is enhanced at gate positions where the ABS levels cross zero energy and the system undergoes a parity-changing quantum phase transition. The most surprising result is the enhanced zero-bias conductance of large amplitude in the Kondo valley at -0.59 V < Vg < -0.53 V. In Fig. 1(g) we show conductance line cut through the center of Kondo valley indicated by the green dashed line in Fig. 1(d). It clearly displays a sharp zero-bias conductance peak and two subgap peaks at Vsd ∼ ±∆/e, corresponding to εa ∼ 0. To clarify the nature of this zero-bias conductance peak, we perform current bias measurements and show the I −V curve in the inset of Fig. 1(g). A typical I −V shape of an overdamped Josephson junction comprising a supercurrent branch and a dissipative branch can be recognized, with a switching current ISW ∼ 1 nA which is consistent with typical magnitude in similar nanowire Josephson junctions47. Hence, the zero-bias conductance peak can be attributed to the manifestation of a supercurrent and this is also confirmed 5 by data from a second similar device (see Fig. S2(c) in Supplementary Materials). We argue that such supercurrent peak is further enhanced by the Kondo effect near the 0-π phase boundary48. Similar behavior of Kondo-enhanced Andreev transport at finite bias has also been reported in hybrid superconductor-QD devices24,25, which can be contributed to a logarithmic enhancement of transparency, 1/ ln2(∆/kBTK), by a poor man's analysis. We speculate that a singularity of Kondo-enhanced transparency at kBTK ∼ ∆ can be responsible for the enhanced zero-bias conductance in a crossover region from a singlet to a doublet ground state. B. Thermally excited Andreev spectra. Having characterized the gate-tunable 0-π transition of the Josephson junction in a InSb NWQD device, we now focus on the thermal effect on the two types of ABSs. Figs. 2(a) and 2(c) show the differential conductance dI/dV in two types of Kondo valleys as a function of bias voltage Vsd and backgate voltage Vg measured at base temperature and zero magnetic field. As mentioned above, depending on the respective ratio of kBTK/∆, the main ABSs resonances behave as a crossing in Fig. 2(a), indicating the charge state is in a π state in the middle of this region, while in Fig. 2(c) the noncrossing ABSs resonances denote a 0 state of that region. When the temperature is elevated gradually up to 700 mK, several prominent features of the ABSs resonances emerge in both π state and 0 state regions. First we look at the π state case (Fig. 2(b)). As the temperature is increased to 700 mK, the inner weak replicas -- stem from the the alignment of residual density of states of quasiparticles in the probe at the Fermi level with the ABS levels - almost disappear. This could be explained by the thermal smearing of the quasiparticle DOS at the Fermi level. Moreover, the π-type ABSs in the center of the Coulomb valley extend out of the π state region while the 0- type ABSs outside of the crossing points penetrate into the π state region, these extending structures form continuous resonances which are parallel to the original ABSs resonances at opposite Vsd at base temperature. For the 0 state case in Figs. 2(c) and 2(d), beside the main ABSs peaks, there are no inner structures at the base temperature. However, a new pair of weak resonances develop as the temperature is increased to 700 mK and similarly to the π state case in Fig. 2(b), the newly developed resonances are parallel to the original ABSs resonances at opposite Vsd. A more detailed study of the ABSs spectroscopy at a series of elevated temperatures is given in section III of supplementary material. We note 6 that both the main ABSs resonances and their replicas shift towards lower energies in the same way due to the shrink of ∆ as the temperature is increased, suggesting the same origin of these subgap features. The evolution of the observed ABSs spectroscopy on temperature can be understood in term of thermally excited quasiparticle transport in the subgap region37. As the temperature is elevated, quasiparticles can be thermally excited from the lower electron-like ABS to the upper hole-like ABS. As a consequence, charge transport through the thermally populated ABS level could be expected to generate additional resonant peaks in differential conductance at finite bias when the singularity of the BCS density of states of the probe is aligned with the thermally populated ABS. Hence these resonances are supposed to occur at Vsd(Vg) = ±[∆ − εa(Vg)]. Given the condition for the original ABS resonances to occur, i.e. Vsd(Vg) = ∓[∆ + εa(Vg)], the position of the thermally developed resonance and that of the original ABS resonance at finite bias corresponding to the same ABS level are spontaneously parallel with a constant spacing of 2∆. Therefore, these thermally-induced subgap features in the Andreev spectrum can also be viewed as replicas of the original ABS resonances. We also note that the width of the newly developed resonances exhibits a clear level broadening with increasing temperature, being consistent with the thermally populated ABS model. Compared with previous study, e.g. ref [37], we have observed the thermally developed ABS resonances in both π-type ABS region and 0-type ABS region, thus constituting a complete experimental picture of the thermal population of different types of ABSs. IV. ANDREEV BOUND STATES IN THE S-NWQD-S DEVICE A. Coexistence of 0 and π-type ABSs in the S-NWQD-S device. We now consider another case of a S-NWQD-S device and focus on the formation (and the underlying physical mechanism) of the ABSs with unusual appearances. We first characterize the device with the leads in the normal state. Fig. 3(a) presents differential conductance dI/dV as a function of Vsd and Vg measured at the base temperature with a small magnetic field applied (B = 50 mT). For the right region at Vg > ∼ -1.2 V, the transport is in a relatively open regime, where the average differential conductance is high and exhibits a series of broad resonances at low bias voltages. As the gate voltage becomes more negative, the differential conductance becomes lower with the emergence of a characteristic Coulomb diamond in the 7 2D plot, denoting the formation of a well defined QD. The charging energy can be estimated from the size of diamond as U ∼ 2.3 meV. Looking more closely at the diamond region, two narrow ridges of high conductance around zero bias, indicated by the black arrows, can be clearly seen in the Coulomb valley (see Fig. S4(b) in supplementary material), suggesting the emergence of a Kondo resonance. By measuring the temperature dependence and magnetic field evolution of the dI/dV in the middle of the odd region (see Fig. S4(c) in supplementary material), we further verify that this split structure is a Kondo feature split by the Zeeman energy resulting from the large g-factor in InSb. From the magnetic field dependence of the Kondo splitting, we extract an effective g-factor of this Kondo valley: g∗ ∼ 35 (see Fig. S4(d) in supplementary material). From the Lorentzian fit of the zero-bias conductance in the middle of the Kondo valley as a function of the magnetic field, the zero-bias conductance Gmax of Kondo ridge at B = 0 is estimated to be 0.19 e2/h, and the asymmetry ratio γ = ΓL/ΓR ∼ 40 is obtained from the relation Gmax = (2e2/h)4ΓLΓR/(ΓL + ΓR)2.49 This high asymmetry ratio indicates that our device can be viewed as a tunneling probe model in which the ABSs are generated mainly between the QD and the strongly coupled lead, and are probed by the other weakly coupled lead12,23,24,27. Next we turn to look at the superconducting state spectrum of the device. Differential conductance dI/dV as a function of Vsd and Vg of the Kondo region measured at B = 0 and the base temperature is displayed in Fig. 3(b). While the transport at the gate region away from the left resonant point (Vg < −1.6 V) is dominated by multiple Andreev reflections, with a quasiparticle cotunneling peak at eVsd = 2∆ and a weak first-order multiple Andreev reflection peak at eVsd = ∆, the Kondo valley exhibits rich subgap structures. Beside the weak first-order multiple Andreev reflection at eVsd = ∆, we concentrate on the main ABS resonance peaks which vary in the finite bias range from eVsd = 2∆ to eVsd = ∆ as a function of the gate voltage. The most noticeable feature is that there are two different types of ABSs peaks of opposite curvature in dispersion of Vg coexisting in this Kondo valley region, namely the π-type crossed ABSs at outer positions(green dashed lines) and the 0-type noncrossed ABSs at inner positions (black dashed lines). This is more clearly characterized by two pairs of subgap dI/dV peaks in Fig. 3(c), showing the line cut of dI/dV in the middle of the Kondo valley in Fig. 3(b). In addition, these ABSs resonances appear to extend into the open regime region next to the Kondo valley, forming a set of weak ABSs resonances in the open region. As we have mentioned, the dispersion of two 8 types of ABSs in the Kondo states is dominated by the system's ground state - singlet or doublet - depending on the ratio kBTK/∆. Taking total coupling strength Γ as determined from the Coulomb resonance peaks of N state, the Kondo temperature at the middle of this valley can be estimated from the Bethe-Ansatz model35,50 as: TK ∼ 460 mK (see section IV of supplementary material), which is much smaller than ∆/kB. From the magnetic field evolution of dI/dV (Vsd), a splitting of the Kondo resonance is observed already at a minimum field B ∼ 25 mT, corresponding to a Zeeman energy of EZ ∼ 49 m eV, providing further evidence of the small value of TK. Therefore, this odd charge state is expected in a π state, and one would expect only a pair of π-type ABSs, being inconsistent with experimental observation. To understand the observed coexistence of the two types to ABSs, we propose a simple model, in which the transport processes in our device involve two parallel conduction channels with different coupling strength to the superconducting leads, which is illustrated in the inset of Fig. 3(c). Here, one channel is weakly coupled to the electrodes and sustains the weak Kondo correlation, leading to the formation of π-type ABS. The other channel, which is strongly coupled to the electrodes, accounts for the emergence of the 0-type ABS. We now discuss the detailed mechanism about the two channel model in the following paragraph. In fact, looking more closely at the normal state conductance plot in Fig. 3(a), we can see some broad resonances with crossed diagonal shapes at low bias on the right side of the Kondo charge state. For better clarity, we take a line cut in this region (grey dashed line) and plot the dI/dV (Vsd) curve in Fig. 3(d). A set of quasi-periodic broad resonances at low bias could be clearly recognized. We argue that these broad resonances is the manifestation of the Fabry-P´erot interference in a ballistic nanowire. Fabry-P´erot interference patterns have been observed in InSb nanowire Josephson junctions in our previous work47 and a mean free path le ∼ 80 nm of our InSb nanowires has been extracted. Given here the channel length L ≃ 65 nm< le, the NWQD device is supposed to work in the ballistic regime and the occurrence of the Fabry-P´erot interference in the open regime is quite reasonable. Moreover, we can extract the average energy spacing of the Fabry-P´erot resonances from Fig. 3(d) as ∆E ∼ 2.5 meV. Using the expression of energy spacing of the Fabry-P´erot interference in a 1D subband ∆EF P = ¯h2π2/2m∗L2 c, where m∗ is the electron effective mass and Lc is the cavity length, we calculate an effective cavity length Lc ∼ 75 nm, which is quite consistent with the geometric channel length of the device and thus supports our interpretation. Therefore, beside the weakly coupled quantum dot level, an additional channel with stronger coupling constituted 9 by the Fabry-P´erot resonant states could also contribute to transport. Since the device shows a quick transition from the Coulomb blockade regime to open transport regime and the Kondo valley is just next to the transitional region (Fig. 3(e)), it is very likely that these two channels with different coupling strengths are involved in the transport simultaneously, which is also supported by the dI/dV curve in Fig. 3(d) where the Fabry-P´erot resonances at low bias and the sharp tunneling resonance at high bias are observed simultaneously. Hence, the coexistent 0-type and π-type ABS could be reasonably understood on the basis of such a two-channel model. B. Temperature evolution of the coexistent 0 and π-type ABSs. To better understand the properties of the coexistent 0 and π-type ABSs, we measure the differential conductance dI/dV (Vsd) along the cut in the middle of the Kondo valley (blue dashed line in Fig. 3(b)) over a wide range of temperature. The colored dI/dV curves shown in Fig. 4(a) correspond to different temperatures, ranging from base temperature (∼10 mK) to the critical temperature of the Al leads (∼ 1 K), and they are plotted as a 2D graph in Fig. 4(b) (a temperature evolution of the overall dI/dV (Vsd, Vg) spectra is given in Fig. S5 in supplementary material). From Figs. 4(a) and 4(b), we can see that the differential conductance dI/dV barely changes in a low temperature range, i.e. T < ∼ 300 mK. However, as the temperature is further elevated, several features become pronounced. First, the most prominent feature is the emergence of a zero bias conductance peak (ZBCP), rising up from approximately 400 mK. An ABS-assisted resonant tunneling process can lead to such ZBCP, as depicted in Fig. 4(c). Specifically, quasiparticles in the superconducting leads could be thermally excited from the continuum band below the Fermi level EF to the empty band above EF as temperature increases, occupying the density of state near the upper gap edge. Meanwhile the same process occurs between a pair of ABSs, leaving each level of the ABSs partially occupied. Note that inside the Kondo region the position of the π-type ABSs is very close to the superconducting gap edge. Thus the thermally excited quasiparticles in the superconducting contacts could tunnel through the quantum dot via both π-type ABSs near the two edges of the gap. Although the scenario presented above is most likely to account for our observation, we also note that some features of the temperature dependence of the differential conductance in Fig. 4(a) can't be fully explained by a thermally activated process. First, we find the width 10 of the ZBCP changes little with increasing temperature, as indicated by the dotted line in Fig. 4(a). If the observed ZBCP is due to thermally excited quasiparticle transport, then the width of conductance peak should be broadened with increasing temperature, inconsistent with experimental result. Second, the temperature dependence of the ZBCP does not follow an exponential behavior as expected for a thermally activated process. In Fig. 4(d) we plot G0/GN as a function of kBTK/∆, where G0 is the zero bias conductance and GN is the normal state conductance at high bias of each curve in Fig. 4(a). The ratio of G0/GN first rises rapidly at small kBTK/∆, then tends to saturate as kBTK/∆ becomes larger corresponding to the increase of temperature. The transition of these two trends occurs at roughly kBTK/∆ ∼ 0.3, which is coincident with theoretical predictions for the 0-π transition to occur11,20. In a weak coupling regime, kBTK/∆ ≪ 1, the system is in the π-state. As the temperature is raised, ∆ is reduced changing the relative strength of ∆ and kBTK, undergoing a crossover from π state to 0 state. In the 0 state, Kondo resonance develops even in the presence of the superconducting gap, and contributes an conductance peak in equilibrium regime. Therefore, we speculate that it is likely that a temperature induced 0-π transition is also responsible for the emergence of the observed ZBCP. Another interesting appearance, shown in Fig. 4(b), is that the 0-type ABSs at finite bias slightly shift to higher energy as the temperature increases above 300 mK. To understand this behavior, we look at the schematic phase diagram in the parameter space (∆/Γ, U/Γ) shown in Fig. 4(e), where ∆, Γ, U represents the superconducting gap, the coupling strength between QD level and superconducting leads, and the charging energy, respectively11,36. When temperature is increased in the range of ∼ 1 K, ∆ is suppressed while no noticeable change of Γ, U is expected, resulting in a reduced ratio of ∆/Γ, represented by the red dashed line in the phase diagram in Fig. 4(e). Theoretically the ABSs tend to approach the continuum of the BCS band in the strong coupling limit ∆/Γ ≪ 111,15, hence the shift of 0-type ABSs towards higher energy (gap edge) with increasing temperature can be reasonably understood. C. Magnetic field evolution of the coexistent 0 and π-type ABSs. Finally, we explore the effect of magnetic field on the spectroscopy of ABSs. Fig. 5(a) displays the magnetic field evolution of the differential conductance measured along the blue dashed line cut in Fig. 3(b). As the magnetic field is applied, the energy of both π-type ABSs and 0-type ABSs tends to decrease. Surprisingly the π-type ABSs seem to 11 decrease in an unusual fast way and become hardly distinguishable at B ∼ 10mT -- far below the superconductor critical field Bc ≈ 25 mT. The 0-type ABSs, on the other hand, vary much more slowly compared with the π-type ABSs and remain to be resolved as the magnetic field approaches Bc. Above the critical field, the ABSs peaks are replaced by a split Kondo resonance. In the presence of a magnetic field, two mechanisms that may have impact on the ABSs should be considered. On the one hand, the superconducting gap ∆ is suppressed by increasing magnetic field, which should result in a subsequent suppression of the intragap states and may also change the relative strength between Kondo correlation and superconductivity, making it possible for a 0-π transition to happen26. But this effect is not supposed to be obvious until the magnetic field approaches Bc since ∆ changes little at magnetic fields far below Bc. On the other hand, the Zeeman effect will alter the energy of different types of ABSs, depending on the magnetic property of corresponding ground state, i.e. singlet or doublet. For a relatively small magnetic field, few tens of mT in our case, one might think of negligible effect of the Zeeman energy. However, considering the large g factors in InSb nanowires, the role of the Zeeman effect can not be neglected. The energy of the π-type ABSs is expected to increase in a magnetic field as a result of the Zeeman splitting of its doublet GS32 (see lower right inset of Fig. 5(b)). But in our case, the π-type ABS is quite close to the superconducting gap edge. Taking the level-repulsion effect32 between the ABSs and the continuum of quasiparticle states into account, it is reasonable that we could not observe an increase of the energy of π-type ABSs. As already mentioned, the π-type ABSs are related with the QD level that is weakly coupled to the superconducting contacts. Such weakly coupled level may be localized in the NW segment just in the middle of the junction by the thick barriers, while the strongly coupled level related with the 0-type ABSs could extend in a longer geometric range. Hence, we attribute the rapid shrink of the π-type ABS to the existence of magnetic flux focusing from the Meissner effect in the case of a narrow Josephson junction geometry51. As for the 0-type ABSs, in a finite magnetic field, they will also follow the suppression of ∆. Besides, a splitting in such ABSs is also expected due to the Zeeman splitting of doublet ES (upper left inset of Fig. 5(b)). In our data, the 0-type ABSs show an overall decreasing trend as B increases. Considering the relatively low magnetic field we apply limited by the critical field of Al, in such magnetic field range, the shrink of the gap is much more significant than the Zeeman energy. Thus the level repulsion effect between the ABS and the continuum of quasiparticle states leads 12 to the overall decreasing trend of the 0-type ABSs. Nevertheless, we note that there is a sign of Zeeman splitting in the 0-type ABSs at B ∼ 14 mT indicated by the black arrows in Fig. 5(a). The corresponding dI/dV (Vsd) is shown in Fig. 5(b), in which we are able to identify a split double-peak on both 0-type ABSs peaks. The energy distance between the double peaks is measured as ∆E ∼ 34 m eV. Using g∗ ≈ 35 estimated from the Zeeman splitting of the Kondo resonance, we obtain a Zeeman energy Ez = g∗µBB ≈ 31 m eV at B = 14 mT, which is consistent with the spacing of the double peaks, further supporting our interpretation. V. CONCLUSION In summary, we demonstrate gate tunable ABSs of different types in superconductor cou- pled InSb nanowire QDs with different device structures by transport measurements. The thermal effect on the two types of ABSs is extensively explored in the NWQD-SQUID de- vices and is understood with a thermally excited quasiparticle model. Two types of ABSs are observed simultaneously in the same charge state in a S-QD-S device, and explained in the senario of two conducting levels with different coupling strength to the leads. The evolution in elevated temperatures and magnetic field confirms the different natures of these ABSs. Therefore, despite the different device geometries, the spectroscopy of two types of ABSs in superconductor coupled single InSb QDs has been unambiguously displayed. Together with the variation of these ABSs, the 0-π transition process driven by different physical parameters - gate voltage (in SQUID device) or temperature (in S-NWQD-S device) - has been revealed in such hybrid systems. Moreover, our work is the first systematic study of the tunable subgap bound states and 0-π transition in InSb nanowire based hybrid super- conducting systems which is an important candidate to host Majorana bound states. These results demonstrate the key role of Andreev Bound States in the transport in mesoscopic Josephson junctions and indicate potential prospect in exploring intragap bound states such as Majorana bound states in such systems. ACKNOWLEDGMENTS We thank D. X. Fan, G. Y. Huang and L. Lu for helpful discussions. This work was financially supported by the National Basic Research Program of the Ministry of Science and Technology of China (Nos. 2012CB932703 and 2012CB932700), and by the National 13 Natural Science Foundation of China (Nos. 11374019, 91221202, 91421303, and 61321001). H. Q. Xu acknowledges also financial support from the Swedish Research Council (VR). 14 References ∗ Corresponding author: [email protected] † Corresponding author: [email protected] 1 S. De Franceschi, L. Kouwenhoven, C. Schonenberger and W. Wernsdorfer, Hybrid Superconductor-Quantum Dot Devices, Nature Nanotech. 5, 703 (2010). 2 Y. J. Doh, J. A. van Dam, A. L. Roest, E. P. Bakkers, L. P. Kouwenhoven and S. De Franceschi, Tunable Supercurrent through Semiconductor Nanowires, Science 309, 272 (2005). 3 P. Jarillo-Herrero, J. A. van Dam and L. P. Kouwenhoven, Quantum Supercurrent Transistors in Carbon Nanotubes, Nature (London) 439, 953 (2006). 4 J. Xiang, A. Vidan, M. Tinkham, R. M. Westervelt and C. M. Lieber, Ge/Si Nanowire Meso- scopic Josephson Junctions, Nature Nanotech. 1, 208 (2006). 5 J. P. Cleuziou, W. Wernsdorfer, V. Bouchiat, T. Ondarcuhu and M. Monthioux, Carbon Nan- otube Superconducting Quantum Interference Device, Nature Nanotech. 1, 53 (2006). 6 H. A. Nilsson, P. Samuelsson, P. Caroff and H. Q. Xu, Supercurrent and Multiple Andreev Reflections in an InSb Nanowire Josephson Junction, Nano Lett. 12, 228 (2012). 7 L. Hofstetter, S. Csonka, J. Nygard and C. Schonenberger, Cooper Pair Splitter Realized in a Two-Quantum-Dot Y-Junction, Nature (London) 461, 960 (2009). 8 L. G. Herrmann, F. Portier, P. Roche, A. L. Yeyati, T. Kontos and C. Strunk, Carbon Nanotubes as Cooper-Pair Beam Splitters, Phys. Rev. Lett. 104, 026801 (2010). 9 A. Das, Y. Ronen, M. Heiblum, D. Mahalu, A. V. Kretinin and H. Shtrikman, High-Efficiency Cooper Pair Splitting Demonstrated by Two-Particle Conductance Resonance and Positive Noise Cross-Correlation, Nature Commun. 3, 1165 (2012). 10 R. S. Deacon, A. Oiwa, J. Sailer, S. Baba, Y. Kanai, K. Shibata, K. Hirakawa and S. Tarucha, Cooper Pair Splitting in Parallel Quantum Dot Josephson Junctions, Nature Commun. 6, 7446 (2015). 11 J. Bauer, A. Oguri and A. C. Hewson, Spectral Properties of Locally Correlated Electrons in a Bardeen-Cooper-Schrieffer Superconductor, J. Phys.: Condens. Matter 19, 486211 (2007). 12 R. S. Deacon, Y. Tanaka, A. Oiwa, R. Sakano, K. Yoshida, K. Shibata, K. Hirakawa and S. 15 Tarucha, Tunneling Spectroscopy of Andreev Energy Levels in a Quantum Dot Coupled to a Superconductor, Phys. Rev. Lett. 104, 076805 (2010). 13 J. D. Pillet, C. H. L. Quay, P. Morfin, C. Bena, A. L. Yeyati and P. Joyez, Andreev Bound States in Supercurrent-Carrying Carbon Nanotubes Revealed, Nature Phys. 6, 965 (2010). 14 J. D. Pillet, P. Joyez, R. Zitko and M. F. Goffman, Tunneling Spectroscopy of a Single Quantum Dot Coupled to a Superconductor: From Kondo Ridge to Andreev Bound States, Phys. Rev. B 88, 045101 (2013). 15 T. Meng, S. Florens and P. Simon, Self-Consistent Description of Andreev Bound States in Josephson Quantum Dot Devices, Phys. Rev. B 79, 224521 (2009). 16 J. Baranski and T. Domanski, In-Gap States of a Quantum Dot Coupled between a Normal and a Superconducting Lead, J. Phys.: Condens. Matter 25, 435305 (2013). 17 A. V. Rozhkov and D. P. Arovas, Josephson Coupling through a Magnetic Impurity, Phys. Rev. Lett. 82, 2788-2791 (1999). 18 E. Vecino, A. Mart´ın-Rodero and A. Yeyati, Josephson Current through a Correlated Quantum Level: Andreev States and π Junction Behavior, Phys. Rev. B 68, 035105 (2003). 19 F. Siano and R. Egger, Josephson Current through a Nanoscale Magnetic Quantum Dot, Phys. Rev. Lett. 93, 047002 (2004). 20 M.-S. Choi, M. Lee, K. Kang and W. Belzig, Kondo Effect and Josephson Current through a Quantum Dot between Two Superconductors, Phys. Rev. B 70, 020502 (2004). 21 J. S. Lim and M.-S. Choi, Andreev Bound States in the Kondo Quantum Dots Coupled to Superconducting Leads, J. Phys.: Condens. Matter 20, 415225 (2008). 22 M. R. Buitelaar, T. Nussbaumer and C. Schonenberger, Quantum Dot in the Kondo Regime Coupled to Superconductors, Phys. Rev. Lett. 89, 256801 (2002). 23 C. Buizert, A. Oiwa, K. Shibata, K. Hirakawa and S. Tarucha, Kondo Universal Scaling for a Quantum Dot Coupled to Superconducting Leads, Phys. Rev. Lett. 99, 136806 (2007). 24 A. Eichler, M. Weiss, S. Oberholzer, C. Schonenberger, A. Levy Yeyati, J. C. Cuevas and A. Mart´ın-Rodero, Even-Odd Effect in Andreev Transport through a Carbon Nanotube Quantum Dot, Phys. Rev. Lett. 99, 126602 (2007). 25 T. Sand-Jespersen, J. Paaske, B. M. Andersen, K. Grove-Rasmussen, H. I. Jørgensen, M. Aage- sen, C. B. Sørensen, P. E. Lindelof, K. Flensberg and J. Nygard, Kondo-Enhanced Andreev Tunneling in InAs Nanowire Quantum Dots, Phys. Rev. Lett. 99, 126603 (2007). 16 26 E. J. H. Lee, X. Jiang, R. Aguado, G. Katsaros, C. M. Lieber and S. De Franceschi, Zero- Bias Anomaly in a Nanowire Quantum Dot Coupled to Superconductors, Phys. Rev. Lett. 109, 186802 (2012). 27 B.-K. Kim, Y.-H. Ahn, J.-J. Kim, M.-S. Choi, M.-H. Bae, K. Kang, J. S. Lim, R. L´opez and N. Kim, Transport Measurement of Andreev Bound States in a Kondo-Correlated Quantum Dot, Phys. Rev. Lett. 110, 076803 (2013). 28 Y. Kanai, R. S. Deacon, A. Oiwa, K. Yoshida, K. Shibata, K. Hirakawa and S. Tarucha, Elec- trical Control of Kondo Effect and Superconducting Transport in a Side-Gated Inas Quantum Dot Josephson Junction, Phys. Rev. B 82, 054512 (2010). 29 R. M. Lutchyn, J. D. Sau and S. Das Sarma, Majorana Fermions and a Topological Phase Transition in Semiconductor-Superconductor Heterostructures, Phys. Rev. Lett. 105, 077001 (2010). 30 Y. Oreg, G. Refael and F. von Oppen, Helical Liquids and Majorana Bound States in Quantum Wires, Phys. Rev. Lett. 105, 177002 (2010). 31 W. Chang, V. E. Manucharyan, T. S. Jespersen, J. Nygard and C. M. Marcus, Tunneling Spectroscopy of Quasiparticle Bound States in a Spinful Josephson Junction, Phys. Rev. Lett. 110, 217005 (2013). 32 E. J. Lee, X. Jiang, M. Houzet, R. Aguado, C. M. Lieber and S. De Franceschi, Spin-Resolved Andreev Levels and Parity Crossings in Hybrid Superconductor-Semiconductor Nanostructures, Nature Nanotech. 9, 79 (2014). 33 J. A. van Dam, Y. V. Nazarov, E. P. Bakkers, S. De Franceschi and L. P. Kouwenhoven, Supercurrent Reversal in Quantum Dots, Nature (London) 442, 667 (2006). 34 H. I. Jørgensen, T. Novotny, K. Grove-Rasmussen, K. Flensberg and P. E. Lindelof, Critical Current 0-π Transition in Designed Josephson Quantum Dot Junctions, Nano Lett. 7, 2441 (2007). 35 A. Eichler, R. Deblock, M. Weiss, C. Karrasch, V. Meden, C. Schonenberger and H. Bouchiat, Tuning the Josephson Current in Carbon Nanotubes with the Kondo Effect, Phys. Rev. B 79, 161407 (2009). 36 R. Maurand, T. Meng, E. Bonet, S. Florens, L. Marty and W. Wernsdorfer, First-Order 0- π Quantum Phase Transition in the Kondo Regime of a Superconducting Carbon-Nanotube Quantum Dot, Phys. Rev. X 2, 011009 (2012). 17 37 A. Kumar, M. Gaim, D. Steininger, A. L. Yeyati, A. Mart´ın-Rodero, A. K. Huttel and C. Strunk, Temperature Dependence of Andreev Spectra in a Superconducting Carbon Nanotube Quantum Dot, Phys. Rev. B 89, 075428 (2014). 38 T. Dirks, T. L. Hughes, S. Lal, B. Uchoa, Y.-F. Chen, C. Chialvo, P. M. Goldbart and N. Mason, Transport through Andreev Bound States in a Graphene Quantum Dot, Nature Phys. 7, 386 (2011). 39 H. A. Nilsson, P. Caroff, C. Thelander, M. Larsson, J. B. Wagner, L.-E. Wernersson, L. Samuel- son and H. Q. Xu, Giant, Level-Dependent g Factors in Insb Nanowire Quantum Dots, Nano Lett. 9, 3151 (2009). 40 S. Nadj-Perge, V. S. Pribiag, J. W. van den Berg, K. Zuo, S. R. Plissard, E. P. Bakkers, S. M. Frolov and L. P. Kouwenhoven, Spectroscopy of Spin-Orbit Quantum Bits in Indium Antimonide Nanowires, Phys. Rev. Lett. 108, 166801 (2012). 41 D. Fan, S. Li, N. Kang, P. Caroff, L. B. Wang, Y. Q. Huang, M. T. Deng, C. L. Yu and H. Q. Xu, Formation of Long Single Quantum Dots in High Quality InSb Nanowires Grown by Molecular Beam Epitaxy, Nanoscale 7, 14822 (2015). 42 V. Mourik, K. Zuo, S. M. Frolov, S. R. Plissard, E. P. A. M. Bakkers and L. P. Kouwenhoven, Signatures of Majorana Fermions in Hybrid Superconductor-Semiconductor Nanowire Devices, Science 336, 1003 (2012). 43 M. T. Deng, C. L. Yu, G. Y. Huang, M. Larsson, P. Caroff and H. Q. Xu, Anomalous Zero-Bias Conductance Peak in a Nb-InSb Nanowire-Nb Hybrid Device, Nano Lett. 12, 6414 (2012). 44 M. T. Deng, C. L. Yu, G. Y. Huang, M. Larsson, P. Caroff and H. Q. Xu, Parity independence of the zero-bias conductance peak in a nanowire based topological superconductor-quantum dot hybrid device, Sci. Rep. 4, 7261-7268 (2014). 45 B. M. Andersen, K. Flensberg, V. Koerting and J. Paaske, Nonequilibrium Transport through a Spinful Quantum Dot with Superconducting Leads, Phys. Rev. Lett. 107, 256802 (2011). 46 K. Grove-Rasmussen, H. I. Jørgensen, B. M. Andersen, J. Paaske, T. S. Jespersen, J. Nygard, K. Flensberg and P. E. Lindelof, Superconductivity-Enhanced Bias Spectroscopy in Carbon Nan- otube Quantum Dots, Phys. Rev. B 79, 134518 (2009). 47 S. Li, N. Kang, D. X. Fan, L. B. Wang, Y. Q. Huang, P. Caroff and H. Q. Xu, Coherent Charge Transport in Ballistic InSb Nanowire Josephson Junctions, Sci. Rep. 6, 24822 (2016). 48 K. Grove-Rasmussen, H. I. Jørgensen and P. E. Lindelof, Kondo Resonance Enhanced Super- 18 current in Single Wall Carbon Nanotube Josephson Junctions, New J. Phys. 9, 124-124 (2007). 49 T. K. Ng and P. A. Lee, On-Site Coulomb Repulsion and Resonant Tunneling, Phys. Rev. Lett. 61, 1768-1771 (1988). 50 F. D. M. Haldane, Scaling Theory of the Asymmetric Anderson Model, Phys. Rev. Lett. 40, 416-419 (1978). 51 J. Paajaste, M. Amado, S. Roddaro, F. S. Bergeret, D. Ercolani, L. Sorba and F. Giazotto, Pb/InAs Nanowire Josephson Junction with High Critical Current and Magnetic Flux Focusing, Nano Lett. 15, 1803 (2015). 52 T. Xu, K. A. Dick, S. Plissard, T. H. Nguyen, Y. Makoudi, M. Berthe, J.-P. Nys, X. Wal- lart, B. Grandidier and P. Caroff, Faceting, Composition and Crystal Phase Evolution in III-V Antimonide Nanowire Heterostructures Revealed by Combining Microscopy Techniques, Nan- otechnology 23, 095702 (2012). 53 C. Thelander, P. Caroff, S. b. Plissard and K. A. Dick, Electrical Properties of InAs1−xSbx and InSb Nanowires Grown by Molecular Beam Epitaxy, Appl. Phys. Lett. 100, 232105 (2012). 19 (a) (b) (c) U / S Γ Singlet Doublet 0 π -0.5 ξd /U 0.5 (d) ) h / 2 e ( V d / I d 3 2 1 0 (e) 0.10 ) h / 2 e ( V d / I d 0.05 0.00 -1.0 -2' 2' (V) V g -0.8 (f) 0.4 ) h / 2 e ( V d / I d 0.2 0.0 -0.6 -0.4 (g) 3 / ) h 2 e ( V d / I d 2 1 ) V m ( d s V 0.1 0.0 -0.1 -4 0 4 Isd (nA) -0.3 0.0 0.3 -0.3 0.0 0.3 -0.3 0.0 0.3 Vsd (mV) Vsd (mV) Vsd (mV) FIG. 1: Gate tuned ABSs in the NWQD-SQUID device. (a) SEM image of an InSb NWQD-SQUID device and schematic of the measurement configuration. The diameter of the nanowire is ∼ 80 nm and the channel length of the nanowire is designed to be around 100 nm. (b) Schematic of tunneling processes in the NWQD-SQUID system. The alignment of the singularity of the hole(electron)-like quasiparticle DOS of the probe with the electron(hole) type Andreev bound states results in the main resonance peaks in the dI/dV spectra. (c) Phase diagram of superconductor coupled QD system. ξd represents the relative position of the chemical potential of the QD, U is the charging energy of the QD, and Γs is the coupling strength of the QD with superconducting leads. (d) Top panel, differential conductance dI/dV plot as a function of Vsd and Vg at base temperature and zero magnetic field. The resonance features are labeled as: ◦ for the main resonance of high conductance and ⋄ for the weak conductance resonance. The letters e and o represents the even and odd charge number states, and blue or green arrows denote the ABSs in the charge state is π-type or 0-type, respectively. Bottom panel, zero-bias conductance dI/dV over the same Vg range. (e)-(g) Traces of the dI/dV taken at Vg values indicated by the colored dashed lines in (d), respectively. Inset of (g) is a current bias measurement at the same Vg, revealing transport with a 20 supercurrent in the device. FIG. 2: Temperature evolution of different types of ABSs.(a) (b) dI/dV (Vsd, Vg) for the odd charge state at Vg ≃ −1.0 V measured at (a): base temperatures T ∼ 10 mK and (b): T ∼ 700 mK. (c) (d) dI/dV (Vsd, Vg) for the odd charge state at Vg ≃ −1.35 V measured at (c): base temperatures T ∼ 10 mK and (d): T ∼ 700 mK. The distance labeling in (b) and (d) indicates the energy spacing between the original ABS resonance and the thermally developed ABS is a constant 2∆, i.e. they are parallel. 21 FIG. 3: Characterization of the S-NWQD-S device in normal and superconducting state. (a) Normal-state differential conductance dI/dV (Vsd, Vg) measured for a two terminal S-NWQD-S device (D2) with L ∼ 65 nm in contact separation and D ∼ 90 nm in nanowire diameter at T = 10 mK and B = 50 mT. (b) Low bias differential conductance dI/dV (Vsd, Vg) measured at B = 0 of the charge state corresponding to the diamond region around Vg ∼ −1.5 V shown in (a). The black and green dashed curves are guides for the two different types of ABSs resonances. (c) dI/dV (Vsd) curves at selected Vg. The blue (orange) curve corresponds to the blue (orange) dashed line cut in panel (b). The black (green) arrows refer to the resonances denoting the π(0)-type ABSs. A shoulder structure at around Vsd ∼ 150 m eV originates from the first-order multiple Andreev reflection. Inset: schematics of the model of our S-NWQD-S system. The red bars represent the weakly coupled QD level (thin) and the strongly coupled QD level (thick) which are both involved in the transport processes. (d) Normal state dI/dV (Vsd) trace taken at the grey dashed line cut in panel (a). The broad resonances at low bias are indicated by orange arrows and the sharp 22 resonance at high bias is indicated by the black arrow. (e) Normal state dI/dV (Vg) trace for the FIG. 4: Temperature evolution of the differential conductance in the Kondo valley. (a) dI/dV (Vsd) curves measured at increasing temperatures from 10 mK to 1 K at B = 0, taken at Vg = −1.49 V denoted by the blue dashed line cut in Fig. 3(b). A zero bias conductance peak can be identified from above T = 400 mK and vanishes at the critical temperature T = 1 K. Curves are successively offset upward by 0.05e2/h for clarity. (b) 2D plot of dI/dV versus Vsd and T . (c) Schematic of the formation of the zero bias conductance peak resulting from the ABS-assisted resonant tunneling of the thermally excited quasiparticles. (d) Normalized zero-bias conductances, G0/GN , versus TK/∆ obtained from traces in (a), where G0 is the zero-bias conductance and GN is the normal state conductance at high bias. (e) Schematic phase diagram for the singlet-doublet transition as a function of ∆/Γ and U/Γ. 23 (a) dI/dV (e2/h) 0 0.2 0.4 (b) ) T m ( B 30 20 10 0 E D S 0.20 0.15 0.10 EZ B E 0.05 B=14mT E E S D ) h / 2 e ( V d / I d -0.4 -0.2 0.0 0.2 Vsd (mV) 0.4 -0.2 0.0 Vsd (mV) B 0.2 FIG. 5: Magnetic field evolution of the differential conductance in the Kondo valley. (a) 2D plot of dI/dV (Vsd, B) measured at Vg = −1.49 V denoted by the blue dashed line cut in Fig. 3(b). A weak splitting in the 0-type ABS resonances is observed at B ∼ 14 mT, indicated by a pair of black arrows. (b) The dI/dV (Vsd) trace taken at B = 14 mT. A split double-peak can be seen on top of each 0-type ABS resonance. The spacing between the double peaks is denoted by ∆E , which is expected to be equal to the Zeeman energy EZ . Upper left inset: schematic of the splitting of the 0-type ABS. For a singlet (Si) ground state and a doublet (Di) excited state, the ABS will split under a magnetic field due to the Zeeman-splitting of the doublet excited state. Lower right inset: schematic of the energy change of the π-type ABS. Due to the Zeeman-splitting of doublet ground state, the energy of the ABS will increase. 24
1112.5861	1	1112	2011-12-26T14:49:31	Non-adiabatic electron charge pumping in coupled semiconductor quantum dots	[ "cond-mat.mes-hall" ]	The possibility of non-adiabatic electron pumping in the system of three coupled quantum dots attached to the leads is discussed. We have found out that periodical changing of energy level position in the middle quantum dot results in non zero mean tunneling current appeared due to non-adiabatic non-equilibrium processes. The same principle can be used for fabrication of a new class of semiconductor electronic devices based on non-stationary non-equilibrium currents. As an example we propose a nanometer quantum emitter with non-stationary inverse level occupation achieved by electron pumping.	cond-mat.mes-hall	cond-mat	6 pages, 4 figures Non-adiabatic electron charge pumping in coupled semiconductor quantum dots P. I. Arseyev,∗ N. S. Maslova,† and V. N. Mantsevich‡ P.N. Lebedev Physical institute of RAS, Moscow, Russia, 119991 and Moscow State University, Department of Physics, 119991 Moscow, Russia (Dated: September 4, 2018) The possibility of non-adiabatic electron pumping in the system of three coupled quantum dots attached to the leads is discussed. We have found out that periodical changing of energy level position in the middle quantum dot results in non zero mean tunneling current appeared due to non-adiabatic non-equilibrium processes. The same principle can be used for fabrication of a new class of semiconductor electronic devices based on non-stationary non-equilibrium currents. As an example we propose a nanometer quantum emitter with non-stationary inverse level occupation achieved by electron pumping. PACS numbers: 73.23.-b, 73.40.Gk, 85.35.Be Keywords: D. Electronic transport in mesoscopic systems; D. Tunneling; D. Electronic transport in QDs; D. Coulomb interaction; D. Quantum well devices I. INTRODUCTION Electron pumping in nanoscale structures attracts much attention nowadays [1],[2],[3],[4], [5], [6]. A great deal of the previous research works have been devoted to adiabatic electron pumping, the idea discussed by Thou- less [7] rather long time ago. The first (to our knowledge) experiment on electron pumping in single electron device was described in [2]. Then experiments in this direction were continued in a three-junction geometry by Pothier et al. [8]. Two phase shifted rf signals were used to real- ize a single electron pump: a device with current I = ef at zero bias voltage (f -frequency of the rf signal). Adi- abatic charge pumping based on periodical variation of the potential barriers formed by the finger gates was also recently investigated in [1]. In these systems quantized current is connected with periodic adiabatic changing of the population of the quantum dot. Proposed in a number of papers photo-assisted tun- neling through coupled quantum dots [9],[10], [11], [12] is also an example of an electron pump. Pumping effect is achieved by applying an oscillating signal to the gate electrode or by irradiating the structure by monochro- matic [12] and pulsed [13] microwaves. It was understood that for practical realization of quantized electron pump the phenomenon of Coulomb blockade is very important [14]. General approach to the pumping through the interacting quantum dots in this regime is based on supposition that the charge relates to instantaneous chemical potential of a dot [15]. Using Coulomb blockade ideas another class of non-adiabatic quantized pumping of electrons in hybrid normal metal- superconductor structures was proposed in [16], [17]. ∗[email protected] †Electronic address: [email protected] ‡Electronic address: [email protected] These systems are more like a "turnstile" rather than a "pump" because quantized current directly proportional to gate frequency appears at finite (nonzero) value of ap- plied bias. But at low temperatures these systems are very promising as current standards [18]. Adiabatic charge pumping through three tunnel- coupled quantum dots attached to electron leads in the regime of strong Coulomb blockade was investigated in [19]. Slow variations of coupling strength between the dots lead to adiabatic changes of energy levels in the sys- tem. Time dependent charge redistribution caused by energy levels changes results in non-stationary adiabatic tunneling current. In the present paper we suggest a new type of electron pumping also in a system with three quantum dots, but based on non-equilibrium non-stationary tunneling cur- rents. The proposed device requires only a single ac gate signal contrary to other semiconductor devices which re- quire at least two rf signals with a definite phase shift. Mean current appears in our model due to non-adiabatic changing of electron level in a single quantum dot. II. THREE DOT MODEL OF ELECTRON PUMP We investigate non-stationary currents which flow in a three dot system shown in Fig.1. The left and right dots have energy levels ε2 and ε3 constant in time. And the level position of the middle dot ε1 is modulated by external gate voltage. Quantum dots with energy levels ε2 and ε3 are also coupled to continuous spectrum states - massive leads. Hamiltonian of the system under investigation has the also presented in the equivalent differential form (except for the point t = t′): 2 (cid:2)(i ∂ ∂t − ε3 + iγ3)((i ∂ ∂t − ε3 + iγ3) − T 2 ·(i ∂ ∂t − ε2 + iγ2)(i 13(i ∂t − ε2 + iγ2)(cid:3)GR ∂ ∂t − ε1) − T 2 12 · ) = 0 11(t, t ∂ ′ (4) FIG. 1: Schematic diagram of the three coupled quantum dots with energy level position in the middle quantum dot depending on the time. form: Consequently, retarded Green's function which deter- mine spectrum re-normalization due to the tunneling be- tween the quantum dots can be written in the following form: H = i ci +Xk εic+ 3Xi=1 2 ck) +Xp 2 c1) + T13(c+ +T12(c+ Tk(c+ 1 c2 + c+ k c2 + c+ +Xk 1 c3 + c+ Tp(c+ p c3 + c+ 3 c1) + 3 cp) εkc+ k ck + ′ GR 11(t, t ) = iΘ(t − t ′ ′ )(A1e−iE1(t−t ′ ) + A2e−iE2(t−t ′ ) + + A3e−iE3(t−t )) (5) (1) Where eigenfrequencies E1,2,3 can be found from equa- tion (see (4)): T12, T13 are tunneling transfer amplitudes between the quantum dots and amplitudes Tk and Tp correspond to the tunneling processes between the quantum dots and continuous spectrum states. c+ k(p)/ck(p)- electrons creation/annihilation operators in the quan- tum dots localized states and in the continuous spectrum states correspondingly. Energy values satisfy the follow- ing ratios: ε2 > εF and ε3 < εF . Pumping effect appears if gate voltage switches the level ε1 between energies ε2 and ε3. i /ci and c+ ′ ′ ) and GR 33(t − t In order to develop a theory for non-stationary cur- rent let us first describe in details charge relaxation pro- cesses in the system if we assume that at the initial mo- ment all charge density in the system is localized in the first (middle) quantum dot and has the value n1(0). At the first step we need to calculate exact retarded Green functions of the system. In the absence of tunneling be- tween the three quantum dots Green functions GR ), GR 22(t − t GR 11(t − t GR 22(t − t GR 33(t − t ) = −iΘ(t − t ) = −iΘ(t − t ) = −iΘ(t − t kT 2 p are tunneling re- laxation rates from leftmost and rightmost dots respec- tively to the leads. Exact retarded electron Green's func- tion GR 11 in the first quantum dot can be found from the integral equation: k and γ3 = πν0 where γ2 = πν0 ) are equal to: )e−iε1(t−t )e−iε3(t−t )e−iε2(t−t 11(t−t )−γ2(t−t )−γ3(t−t p T 2 (2) ) ) ) ′ ′ ′ ′ ′ ′ ′ ′ ′ ′ ′ ′ GR 11 = G0R 11 + G0R 11 T 2 12GR 22GR 11 + G0R 11 T 2 13GR 33GR 11 (3) Acting G0R−1 11 , GR−1 22 in , GR−1 33 turn this by integral inverse operators equation can be (E − ε1) · (E − ε2 + iγ2) · (E − ε3 + iγ3) − 12 · (E − ε3 + iγ3) − T 2 13 · (E − ε2 + iγ2) = 0 −T 2 (6) And coefficients Ai can be evaluated using integral equation for GR 11: A1 = A2 = A3 = (E1 − E2)(E1 − E3) E1(E1 + E3 − eε2 − eε3) − E1E3 + eε2eε3 E2(E2 + E3 − eε2 − eε3) − E2E3 + eε2eε3 E3(E2 + E3 − eε2 − eε3) − E2E3 + eε2eε3 (E2 − E3)(E2 − E1) (E3 − E1)(E3 − E2) (7) where eεi = εi − iγi Further on we assume for simplicity that T12 = T13 = T . If ε1 = ε2 and ε1 − ε3 ≫ T, γ coefficients Ai has the form: E3 − eε3 E1 − E3 E3 − eε3 E1 − E3 ) = A0 1(1 + ) = A0 2(1 + T 2 (ε1 − ε3)2 ) (ε1 − ε3)2 ) T 2 A1 = A0 1(1 + A2 = A0 2(1 + T 2 A3 = − (8) (ε1 − ε3)2 2 = − E2− eε2 A0 E1−E2 1 = E1− eε2 where A0 If we disconnect the third quantum dot then coeffi- cients A0 2 give an exact solution for a system of two coupled quantum dots for all energy values ε1 and ε2. 1 and A0 E1−E2 . Time evolution of electron density in the middle dot is determined by the Keldysh Green function G< [20]: ′ G< 11(t, t ) = in1(t) (9) Equations for the Green functions G< ii has the form: (G0−1 11 − T 2 = T 2 12GR 12GR 12G< 22 − T 2 22GA 11 −Xk 13GR 11 + T 2 T 2 k GR 33)G< 13G< kk)G< 11 = 33GA 11 22 = (G0−1 22 − T 2 = T 2 (G0−1 33 − T 2 = T 2 12G< 11GA 22 +Xk 13GR 11 −Xp 13G< 11GA T 2 p G< ppGA 33 33 +Xp 22(0, 0) = inF (ε2) ≃ 0, G< 11(0, 0) = n1(0) then Green function G< (10) If G< 33(0, 0) = inF (ε3) ≃ 1 and G< 11(t, t) is determined by the sum of homogeneous and inhomoge- neous solutions. Inhomogeneous solution of the equation can be written in the following way: ′ G< 11(t, t ) = iT 2 ′ dt1Z t 13Z t ×nF (ε3)e−i eε3(t1−t2)GA dt2GR 11(t − t1) × ) 11(t2 − t 0 0 ′ (11) Homogeneous solution of the differential equation has the form: ′ − iG< 11(t, t ) = f1(t ′ )e−iE1t + f2(t ′ )e−iE2t + f3(t ′ )e−iE3t (12) Function G<(t, t ′ ) satisfies the symmetry relations: ′ [G< 11(t, t ′ 11(t )]∗ = −G< ) can be written as: , t) ′ Then coefficients fi(t (13) T 2 k G< kkGA 22 Now we can find all coefficients in (14) : T 2 p GR pp)G< 33 = A = with coefficients F21, F31: 3 · F21 = − (cid:2)(E2 − eε2)(E2 − eε3)((E1 − eε2)(E3 − eε2) + + (E1 − eε3)(eε2 − E3))(cid:3) · (cid:2)(E1 − eε2)(E1 − eε3)((E3 − eε3)(E2 − eε2) + + (eε2 − E3)(E2 − eε2))(cid:3)−1 (E3 − eε2)(E2 − eε2 + (E1 − eε2)F21) (E2 − eε2)(E1 − eε2) F31 = (15) n1(0) + 2ReF21 + 2ReF31 1 +(cid:12)(cid:12)F21 + F31(cid:12)(cid:12)2 21 · A; C = F212 · A 31F21 · A; Z = F312 · A; X = F ∗ B = F ∗ D = F ∗ 31 · A (16) Finally, time dependence of the filling number in the middle quantum dot n1(t) can be written as: n1(t) = n0 1 · (Ae−i(E1−E ∗ 1 )t + Ce−i(E2−E ∗ 3 )t) + 2Re(Be−i(E1−E ∗ +Ze−i(E3−E ∗ +2Re(Xe−i(E1−E ∗ 3 )t) + 2Re(De−i(E2−E ∗ (17) 2 )t + 2 )t) + 3 )t) We see that there are six typical time scales in the considered system, which are described by the expression (17). Three of them we can identify as three relaxation modes with rates 2ImE1, 2ImE2 and 2ImE3 . Three other time scales are determined by the expressions Re(E1 − 2), Re(E1− E∗ E∗ 3). These time scales are related with charge density oscillations between quantum dots, if the following ratio between Tij and γ takes place: Tij/γ > 1/√2. 3) and Re(E2− E∗ If we neglect for a moment the tunneling from the mid- dle dot to the right one, then for the system of two cou- pled quantum dots (T13 = 0) only three time scales ap- pear and the equations (15),(16),(17) are transformed in the following way: f1(t ′ ′ ′ ) = AeiE ∗ 1 t ) = CeiE ∗ 2 t ) = X ∗eiE ∗ 1 t ′ ′ + BeiE ∗ 2 t + B∗eiE ∗ 1 t + D∗eiE ∗ 2 t ′ ′ ′ f2(t ′ f3(t ′ ′ ′ + XeiE ∗ 3 t + DeiE ∗ 3 t + ZeiE ∗ 3 t ′ f1(t ) f2(t′ ) = − ε2 − E1 − iγ ε2 − E2 − iγ (18) (14) Time dependence of the filling numbers in the first quantum dot n1(t) can be written as: Since the solution has to satisfy homogeneous integro- differential equation, we are able to determine all coef- ficients. After some calculations we obtain that the fol- lowing proportionality takes place: ′ ′ f2(t f3(t ) = F21f1(t ) = F31f1(t ′ ′ ) ) n1(t) = n0 1 ·hAe−i(E1−E ∗ 2 )ti + Ce−i(E2−E ∗ 1 )t + 2Re(Be−i(E1−E ∗ 2 )t)+ (19) where coefficients A, B and C are determined as: A = E2 − ε12 E2 − E12 C = E1 − ε12 E2 − E12 (E2 − ε1)(E∗ E2 − E12 B = − 1 − ε1) (20) and coefficients X, Z and D are equal to zero. In this situation we can distinguish two relaxation rates γres and γnonres which characterises charge relaxation through an intermediate quantum dot in resonant and nonresonant cases: γres = 2T 2 γ γnonres = γres γ2 (ε1 − ε2)2 (21) As we consider ε1− ε3 ≫ T, γ then γres ≫ γnonres and small parameter γnonres/γres exists in the theory. This allows us to write the following approximate relations for the system of three coupled quantum dots in the case ε1 ≃ ε2 valid in the first order of the small parameter γ 2 (ε1−ε3)2 T 2 (ε1 − ε3)2(cid:21) E1 − E∗ E2 − E∗ E3 − E∗ E2 − E∗ E1 − E∗ E1 − E∗ (ε1 − ε3)2(cid:21) γ2 T 2 γ2 T 2 γ2 + T 2 1 = −iγres(cid:20)1 + 2 = −2iγ(cid:20)1 − 3 = 2iγ(cid:20)1 − (ε1 − ε3)2(cid:21) 3 = ε1 − ε3 − 2iγ(cid:20)(1 + 3 = ε1 − ε3 − iγres ε1 − ε3 − i 2 = iγ + 2T 2 T 2 (22) T 2 T 2 2γ2 − γ2 2(ε1 − ε3)2(cid:21) (ε1 − ε3)2 − iγ(cid:20)1 − γ (cid:20) (ε1 − ε3)2 − (ε1 − ε3)2(cid:21) (ε1 − ε3)2(cid:21) T 2 T 2 γ2 When ε1 = ε2 and ε1 − ε3 ≫ T, γ the exact equations (15) can be transformed in the following way . F21 ≃ − T 2 γ2 (cid:20)1 + F31 ≃ γ + T 2 γ2 − (ε1 − ε3)2 + i ε1 − ε3 γ2 (cid:20) T 2 T 2 T 2 γ(ε1 − ε3)(cid:21) (ε1 − ε3)2(cid:21) T 2 (23) So, coefficients D, Z and X which are resposible for the "reverse" current to the right lead are much smaller than A, B and C, which correspond to "direct" current to the left, due to the appearance of the parameter (ε1−ε3)2 . For simplicity we omit the terms with coefficients D, Z and X in equation (17) which determine time evolution of localized charge in the middle quantum dot. For any T 2 4 concrete system the accuracy of this approximation can be easily estimated from the exact equations. Pumping of electrons takes place if energy level ε1(t) is a function of time and changes periodically (Fig. 1). We shall describe the most favorable case with T ≪ γ. For current calculation we consider the situation of pe- riodically switching the position of level ε1 by external gate: ε1(t) = ε3 in the interval 0 < t < t0 it means resonant tunneling between energy levels ε1 and ε3 ε1(t) = ε2 in the interval t0 < t < 2t0 - resonance between energy levels ε1 and ε2 Time evolution of local electron density n1(t) in the central quantum dot can be determined from equation (17) (Fig.2). When 0 < t < t0 n1(t) = n0 1(cid:20)(cid:18)1 + γres γ (cid:19) e−γrest − e−γt(cid:21) γres γ and when t0 < t < 2t0 n1(t) = n0 1(cid:2)(cid:0)1 + +(cid:2)1 − (1 + γres − γ γres γ γres γ (cid:1)e−γres(t − t0) − (e−γ(t − t0)(cid:3) + )e−γres(t − t0) + e−γ(t − t0)(cid:3) γres + γ (24) (25) Taking into account periodicity condition n1(2t0) = 1, one can find n0 n0 1: n0 1 = 1 1 +(cid:16)1 + γres γ (cid:17) e−γrest0 − γres γ e−γt0 (26) Results for n1(t) are shown in Fig.2. Situation when frequency Ω ≡ 1/2t0 of the level ε1(t) switching is higher than tunneling rates γres, γ is depicted by grey line. Black line corresponds to the case when frequency Ω is lower than γres, γ. It is clear that with the increasing of frequency the value of n1(t) always tends to the value 1/2 and is almost independent on time at high gate fre- quencies. For low frequencies if γrest0 ≫ 1 the value n0 1 is al- most equal to 1. The energy level ε1 is filled up almost completely during the pumping cycle (for considered sit- uation when energy level ε2 is well above and energy level ε3 is well below the Fermi energy). Non-stationary tunneling current through the system appears for zero applied bias: e ∂ ∂t n1(t) = I(t) (27) 5 FIG. 2: Time evolution of local electron density n1(t) in the central quantum dot in the system of three coupled quan- tum dots if energy level position in the middle quantum dot depends on time. Black line corresponds to the case when fre- quency of energy level ε1(t) switching is lower than tunneling rates and grey line corresponds to the case when frequency is higher than tunneling rates. One can find for 0 < t < t0 I(t) = en0 1γres(cid:20)(1 + )e−γrest − e−γt(cid:21) γres γ Mean tunneling current value can be found as: 0 1 2 (28) FIG. 3: Frequency dependence of the mean tunneling current for the system of three coupled quantum dots (gate frequency Ω = 1/2t0). Different frequency scales are presented. < I >= e ·(cid:20)1 − (1 + γres γ 1 2t0 t0Z 0 I(t)dt = e )e−γrest0 + γres γ 1 2t0 n0 1 · e−γt0(cid:21) (29) If Ω ≡ 1/2t0 ≪ γres tunneling current mean value can be written as < I >= eΩ since n0 1 = 1 for such frequen- cies. This is the regime, when the device operates like a current standart: the current is directly proportional to the gate frequency. This regime has exponential accu- racy which is governed by the second and third terms in the square brackets in expression (29). Note, that even for very unsuitable case if ε2 − ε3 ≃ γ ≃ T the pumping effect still remains, and the cuurrent is proportional to the frequency, though it's value is suppressed compared to the ideal relation < I >= eΩ. res/4γ. For high frequencies of the gate voltage in the region γ ≫ Ω ≡ 1/2t0 ≫ γres tunneling current average value is almost independent on the frequency and equal to: < I >= eγres/4 (γres = 2T 2 γ ). With further frequency increase (Ω ≫ γ ≫ γres) mean current value decreases to < I >= eγ2 The non-stationary mean tunneling current value has non-monotonic dependence on the gate frequency with maximum at Ω ≃ γres (Fig.3). This effect can be used for frequency stabilization in nanoelectronics. The parameters of devices based on quantum dots depend on the size of quantum dots, tunneling trans- fer rates, energy levels positions and distances between them. Estimation of tunneling parameters for achievable setup gives us characteristic frequencies and currents for such devices: γ ≃ 1 ÷ 10 meV ⇒ T ≃ 1 meV, γres = 2T 2/γ ≃ 0.1 ÷ 1 meV ≃ 1010 ÷ 1011 1/sec where parameters T and γ are determined by the widths and heights of the barriers. For such values of the tunneling rates we need to have quantum dots of tens of nanometers size, for which quantum size quanti- zation energies are not less than the tunneling width of levels. Such devices could operate at gigahertz and subgi- gahertz frequencies at nano and subnanoampere currents (1nA ≃ 6 · 109 e/sec). The difference between tunneling rates in resonant and non resonant cases can be used also for creation of inverse occupation in quantum emitter based on the system of coupled quantum dots. An example of such device is shown in Fig.4. In the system of two coupled quantum FIG. 4: Schematic diagram of two coupled quantum dots which operate as an emitter due to non stationary inverse occupation of levels. dots we switch the level ε by applying gate voltage to the second quantum dot between two levels of the first quan- tum dot directly coupled to the lead. If the third non- resonant level exists in the same quantum dot between the lowest and the highest levels, then charge pumping from the initially filled state ε3 to the highest level ε1 creates inverse occupation of the states ε2 and ε1. So some nanometer quantum generators can be fabricated on this principle. III. CONCLUSION We investigated electron pumping ability of a system of three tunnel coupled quantum dots attached to the leads. Periodical changing of energy level position in the middle quantum dot by gate voltage leads to nonzero tunneling current even if applied to the structure bias is equal to zero. Our calculations of the mean current are based on ac- curate analysis of relaxation processes in quantum dots in non-adiabatic regime. Exact equations allows to investi- gate various regimes of the device and estimate the mean 6 current value and accuracy of it's operation as a current standard. For very small dots with pronounced size ef- fect a possibility of room temperature electron pumping is opened. We should like to stress that the ideas dis- cussed in the paper can be used for fabrication of a new type of electronic devices based on non-equilibrium non- stationary tunneling currents. As an example we pro- posed in the paper nanometer quantum emitter based on two coupled quantum dots. IV. ACKNOWLEDGEMENTS The financial support by RFBR, Leading Scientific School grants and grants of Russian Minestry of Science is acknowledged. [1] M.D. Blumenthal et.al., Nature, 3, 343, (2007) [2] L.G. Geerligs et.al., Phys. Rev. Lett., 64, 2691, (1990) [3] V.E. Anderegg et.al., Physica B, 165-166, 61, (1990) [4] M.W. Keller, J.M. Martins, N.M. Zimmerman A.H. Steinbach, Appl. Phys. Lett., 69, 1804, (1996) [5] M. Switkes et. al., Science, 283, 1905, (1999) [6] B. Kaestner et. al., Phys. Rev.B, 77, 153301, (2008) [7] D.J. Thouless, Phys. Rev.B , 27, 6083, (1983) [8] H. Pothier, Europhys. Lett., 17, 249, (1992) [9] H. Stoof, Yu. Nazarov, Phys. Rev. B., 53, 1050, (1996) [13] B. Hazelzet, M. Wagewijs, H. Stoof, Yu. Nazarov, Phys.Rev. B, 63, 165313, (2001) [14] E. Cota, R. Aguadado, G. Platero, Phys.Rev Lett., 94, 107202, (2005) [15] J. Splettstoesser, M. Governale, J. Konig, R. Fazio, Phys. Rev. Lett., 95, 246803, (2005) [16] J. P. Pekola, J. J. Vartiainen, M. Mottonen, O.-P. Saira, M. Meschke, D. V. Averin, Nature Phys., 4, 120, (2008) [17] D. Averin, J. P. Pekola, Phys.Rev Lett., 101, 066801-1, (2008) [10] M. Covington, M.W. Keller, R.L. Kautz, J.M. Martins, [18] V. F. Maisi, Y. A. Pashkin, et.al, New Journal of Phys., Phys. Rev. Lett., 84, 5192, (2000) 11, 113057, (2009) [11] Ph. Brune, C. Bruder, H. Schoeller, Phys. Rev. B, 56, [19] F. Rezoni, T. Brandes, Phys. Rev. Lett., 64, 245301, 4730, (1997) (2001) [12] C.A. Stafford, N. Wingreen, Phys. Rev. Lett., 76, 1916, [20] L.V. Keldysh, Sov. Phys. JETP, 20, 1018, (1964) (1996)
1408.5244	2	1408	2015-01-07T18:44:31	Quantized electromagnetic response of three-dimensional chiral topological insulators	[ "cond-mat.mes-hall", "cond-mat.str-el" ]	Protected by the chiral symmetry, three dimensional chiral topological insulators are characterized by an integer-valued topological invariant. How this invariant could emerge in physical observables is an important question. Here we show that the magneto-electric polarization can identify the integer-valued invariant if we gap the system without coating a quantum Hall layer on the surface. The quantized response is demonstrated to be robust against weak perturbations. We also study the topological properties by adiabatically coupling two nontrivial phases, and find that gapless states appear and are localized at the boundary region. Finally, an experimental scheme is proposed to realize the Hamiltonian and measure the quantized response with ultracold atoms in optical lattices.	cond-mat.mes-hall	cond-mat	Quantized electromagnetic response of three-dimensional chiral topological insulators S.-T. Wang,1, 2 D.-L. Deng,1, 2 Joel E. Moore,3, 4 Kai Sun,1 and L.-M. Duan1, 2 1Department of Physics, University of Michigan, Ann Arbor, Michigan 48109, USA 2Center for Quantum Information, IIIS, Tsinghua University, Beijing 100084, Peoples Republic of China 3Department of Physics, University of California, Berkeley, California 94720, USA 4Materials Sciences Division, Lawrence Berkeley National Laboratory, Berkeley, California 94720, USA (Dated: November 6, 2018) Protected by the chiral symmetry, three dimensional chiral topological insulators are characterized by an integer-valued topological invariant. How this invariant could emerge in physical observables is an important question. Here we show that the magneto-electric polarization can identify the integer-valued invariant if we gap the system without coating a quantum Hall layer on the surface. The quantized response is demonstrated to be robust against weak perturbations. We also study the topological properties by adiabatically coupling two nontrivial phases, and find that gapless states appear and are localized at the boundary region. Finally, an experimental scheme is proposed to realize the Hamiltonian and measure the quantized response with ultracold atoms in optical lattices. PACS numbers: 75.85.+t, 73.43.-f, 03.65.Vf, 37.10.Jk I. INTRODUCTION The periodic table of topological insulators (TIs) and superconductors classifies topological phases of free fermions according to the system symmetry and spatial dimensions [1, 2]. Notable examples include integer quan- tum Hall insulators breaking all those classification sym- metries and the time-reversal-invariant TIs protected by the time-reversal symmetry [3 -- 8]. Mathematically, these exotic states can be characterized by various topological invariants. An interesting question is how to relate these invariants to physical observables. For integer quantum Hall insulators, the Chern number (Z invariant) corre- sponds to the quantized Hall conductance [9], while for the time-reversal-invariant TIs, the Z2 invariant is asso- ciated with a quantized magneto-electric effect in three dimensions (3D) [10, 11]. The 3D chiral TIs protected by the chiral symmetry [12, 13] are of particular interest as they are 3D TIs char- acterized by a Z (instead of Z2) invariant and may be realized in ultracold atomic gases with engineered spin- orbital coupling [14 -- 16]. An experimental scheme was recently proposed to implement a three-band chiral TI in an optical lattice [17]. For such 3D chiral TIs, it is known that the topological magneto-electric effect should also arise, but in theory it captures only the Z2 part of the Z invariant due to the gauge dependence of the po- larization in translationally invariant systems [12]. It is thus an important question to find out how the Z char- It was pro- acter could manifest itself in experiments. posed in Ref. [18] that the Z effect may become visible in certain carefully engineered heterostructures, but the im- plementation of such a heterostructure is experimentally challenging. In this paper, we study the nontrivial Z character of the chiral TI by exploring the adiabatic transition be- tween two nontrivial phases and by numerically simu- lating the magneto-electric effect in a single phase. We show that not only the Z2 response but the Z character can be observed by gapping the system without adding a quantum Hall layer on the surface, i.e., the ambigu- ity resulting from different terminations appears to be avoidable in practice. Also, the quantized polarization is demonstrated to be robust against small perturbations even in the absence of a perfect chiral symmetry. This observation is important for experimental realization, be- cause in a real system the chiral symmetry is typically an approximate instead of exact symmetry. Lastly, we pro- pose an experimental scheme to realize the Hamiltonian and probe the integrally quantized response with cold atomic systems. II. MODEL AND TOPOLOGICAL CHARACTERIZATION We first introduce a minimal lattice tight-binding (cid:80) model for chiral topological insulators with the Hamil- tonian in the momentum space given by H1 = k Ψ†kH1(k)Ψk, where Ψk = (ak↑, ak↓, bk↑, bk↓)T de- notes fermionic annihilation operators with spins ↑,↓ on sublattices or orbitals a, b. In cold atom systems, the pseudospins and orbitals can be represented by different atomic internal states. The 4 × 4 Hamiltonian is −iq0 + q3 q1 − iq2 q1 + iq2 −iq0 − q3 0 0  0 0  (1) H1(k) = iq0 + q3 q1 − iq2 q1 + iq2 iq0 − q3 0 0 0 0 where q0 = h+cos kx +cos ky +cos kz, q1 = sin kx +δ, q2 = sin ky, q3 = sin kz, with h, δ being control parameters. The lattice constant and tunneling energy are set to unity. In real space, this Hamiltonian represents on- site and nearest neighbor hoppings and spin-flip hoppings between two orbitals in a simple cubic lattice. These hoppings can be realized by two-photon Raman transi- tions in cold atoms [17, 19 -- 21]. The energy spectrum for this Hamiltonian is E±(k) = ±[(sin kx + δ)2 + sin2 ky + 2 calculated numerically by discretizing the Brillouin zone and replacing the integral by a discrete sum [22]. The results are shown in Fig. 1 for both H1(k) and H2(k). III. SURFACE STATES AND HETEROSTRUCTURE OF NONTRIVIAL TOPOLOGICAL PHASES By imposing an open boundary condition along the z direction, and keeping the x and y directions in momen- tum space, surface Dirac cones are formed for nontrivial topological phases. We find that the winding number coincides with the total number of Dirac cones counted for all inequivalent surface states (i.e. not counting the twofold degeneracy for each band), which confirms ex- plicitly the bulk-edge correspondence (See Appendix A). A distinctive difference from the time-reversal invariant TI is that any number of Dirac cones on the surface is protected by the chiral symmetry [12]. With an integer number of nontrivial phases, it is in- triguing to study the topological phase transition be- tween two different phases. A simple way to explore this is to adiabatically vary h from one end to the other end of the sample. The parameter h concerns the onsite tun- neling strength between opposite orbitals (a† b↓ ↑ terms). This hopping can be realized by a two-photon Raman process, and the strength h can be controlled by the laser intensity [17]. Numerically, we vary h in the form of h = 1 + tanh(z − Lz/2), where z denotes the zth layer and Lz the total number of slabs along the z di- rection. This form ensures that h changes adiabatically from 0 on one end to 2 on the other end of the sample, so that it effectively couples two nontrivial phases. For the Hamiltonian H1, it couples two topological phases with winding numbers Γ = −2 and Γ = 1. Similar to the b↑ and a† ↓ FIG. 2. (Color online) Coupling two topologically nontriv- ial phases by varying the parameter h adiabatically from 0 to 2 along the z direction. x and y directions are periodic. Sixty slabs are taken for the z direction. (a) The energy dis- persion for the lowest conduction band and highest valence band. (b) and (c) show the surface states near the respective Dirac points. The Γ point is displaced from the center for a better display of the Dirac cones. FIG. 1. (Color online) The winding number Γ as a function of the parameter h. The Hamiltonians are H1(k) in (a) and H2(k) in (b) respectively. δ = 0.5 for both panels. sin2 kz + (cos kx + cos ky + cos kz + h)2]1/2, with two- fold degeneracy at each k. For δ = 0, the system ac- quires time-reversal symmetry T , particle-hole symmetry C, and chiral symmetry S = T C, which can be explicitly seen as [13]: T : C : S : (σx ⊗ σy)[H1(k)]∗(σx ⊗ σy)= H1(−k) (σy ⊗ σy)[H1(k)]∗(σy ⊗ σy)= −H1(−k) (σz ⊗ I2)[H1(k)](σz ⊗ I2) = −H1(k) (2) (3) (4) with σi as Pauli matrices, and I2 as the 2×2 identity ma- trix. When δ (cid:54)= 0, time-reversal and particle-hole sym- metries are explicitly broken, but the chiral symmetry survives. The topological property of the Hamiltonian H1(k) can be characterized by the winding number Γ [1, 13]. To define the winding number, let us introduce the Q matrix, Q(k) = 1 − 2P (k), where P (k) = (cid:80) (cid:18) 0 f uf (k)(cid:105)(cid:104)uf (k) is the projector onto the filled Bloch bands with wave- vectors uf (k)(cid:105). The Q matrix can be brought into the block off-diagonal form Q(k) = with the (cid:19) b(k) b†(k) 0 chiral symmetry. With the matrix b(k), one can write [1, 13] dk µρλTr[(b−1∂µb)(b−1∂ρb)(b−1∂λb)], (5) (cid:90) 1 Γ = 24π2 BZ where µρλ is the antisymmetric Levi-Civita symbol and ∂µb ≡ ∂kµb(k). The Hamiltonian H1(k) supports topo- logical phases with Γ = 1,−2. To obtain chiral TIs with arbitrary integer topological invariants, one can use the quaternion construction proposed in Ref. [22]. By con- sidering q = q0 + q1i + q2j + q3k as a quaternion and raising to a power, all Z topological phases can be re- alized by the family of tight-binding Hamiltonians. By taking the quaternion square, for example, one obtains 1 − q2 q2 = q2 3 + 2q0q1i + 2q0q2j + 2q0q3k, and we can therefore acquire another tight-binding Hamiltonian k Ψ†kH2(k)Ψk with each component of q2 re- placing the respective components q0, q1, q2, q3 in H1(k). This second Hamiltonian H2(k) contains next-nearest- neighbor hopping terms. The winding number Γ can be H2 = (cid:80) 2 − q2 0 − q2 −4−2024−2−101hwinding number Γ(a)−4−2024−4−3−2−1012hwinding number Γ(b)−0.500.511.500.511.5−101kx/πky/πE20406000.5120406000.51Lattice site along zState density(c)(a)(b)(b)(c)interface between a topological insulator and the trivial vacuum, a surface state should appear at the interface. As shown in Fig. 2, three Dirac cones are formed inside the band gap. In addition to the surface states observed on both ends of the sample, a localized state is formed at the interface between two topologically distinct regions. These interface states are always present regardless of the detailed structure of the interface. Even for sharp boundaries, the interface states remain. The Dirac cone structure may be probed through Bragg spectroscopy in cold atom experiments [23, 24]. The above heterostructure could be used to probe the topological properties of the chiral TI, but it is experi- mentally difficult to engineer such a heterostructure, es- pecially in cold atom systems. In Ref. [18], it was shown that the Z character of the topological invariant could be seen in some carefully engineered heterostructures. In the following, we show that the Z topological invariant can be observed via the magneto-electric polarization in a single nontrivial phase with a gapped surface. We fur- ther show that the detection will be robust to realistic experimental perturbations and present a feasible exper- imental scheme to observe the quantized response. IV. MAGNETO-ELECTRIC EFFECT The magneto-electric effect is a remarkable manifesta- tion of the bulk non-trivial topology. The linear response of a TI to an electromagnetic field can be described by the magneto-electric polarizability tensor as [11] (cid:12)(cid:12)(cid:12)(cid:12)E=0 (cid:12)(cid:12)(cid:12)(cid:12)B=0 αij = ∂Pi ∂Bj = ∂Mj ∂Ei , (6) where E and B are the electric and magnetic field, P and M are the polarization and magnetization. Unique to topological insulators is a diagonal contribution to the tensor with αij = θ e2 2πh δij. This is a peculiar phe- nomenon as an electric polarization is induced when a magnetic field is applied along the same direction [10]. This effect can be described by a low-energy effective field theory in the Lagrangian as (c = 1) ∆L = θ B · E e2 2πh (7) known as the "axion electrodynamics" term [11]. For time-reversal-invariant TIs, an equivalent understanding will be a surface Hall conductivity induced by the bulk magneto-electric coupling. When the time-reversal sym- metry is broken on the surface generating an insulator, a quantized surface Hall conductance will be produced: σH = θ e2 2πh (8) where θ is quantized to be 0 or π to preserve the time- reversal invariance [10]. θ = π corresponds to the non- trivial time-reversal-invariant TI with a fractional quan- tum Hall conductivity. The electric polarization can be 3 understood with Laughlin's flux insertion argument [25]. A changing magnetic field through the insulator induces an electric field (by Faraday's law), which together with the quantized Hall conductivity will produce a transverse current and accumulate charge around the magnetic flux tube at a rate proportional to σH as Q = σH Φ [26]. A. Numerical results Theoretically, chiral TIs are also predicted to have this topological magneto-electric effect [12, 13]. The field the- ory only captures the Z2 part of the integer winding num- ber due to the 2π periodicity of θ associated with a gauge freedom in transitionally invariant systems. However, we numerically show that the Z character can actually be observed by gapping the system without adding a strong surface orbital magnetic field. Apparently, this corre- sponds to a particular gauge such that the Z character can be distilled from the polarization. More concretely, we consider the chiral TI represented by both Hamil- tonians H1(k) and H2(k). A uniform magnetic field is inserted through the chiral TI sample via the Landau gauge A = Bxy with a minimal coupling by replacing ky with ky − e Bx. We keep x and y directions in momen- tum space, and the z direction in real space with open boundaries and Lz slabs. By taking a magnetic unit cell with N sites along the x direction, the Hamiltonian can be partially Fourier transformed to be a 4Lz × N matrix for each kx and ky, with 4 taking into account of spins ↑,↓ and orbitals a, b. For a unit magnetic cell with N lattice cells, the total magnetic flux through the unit cell is quantized to be integer multiples of a full flux quantum Φ = nΦ0 = n h e due to the periodic boundary condition along the x direction, so the flux through a single lattice plaquette is quantized to be Φ/N . In the weak magnetic field limit, one needs to take a large N . Besides the bulk Hamiltonian H1(k) or H2(k), we also add a surface term to break the chiral symmetry and open a gap on the sur- face, (cid:88) (cid:88) (cid:16) (cid:17) , HS = ∆S S · z Ψ†j,kx,ky (I2 ⊗ σz) Ψj,kx,ky kx,ky j∈surf (9) where S represents the unit vector perpendicular to the surface along z direction, so S·z = 1 for the upper surface, and S· z = −1 for the lower surface. This term represents a surface magnetization with a Zeeman coupling, creating a different chemical potential for spins ↑ and ↓. It can be directly verified that this surface term breaks the chiral symmetry S in Eq. (4). As the surface becomes gapped, at half filling, an in- creasing uniform magnetic field accumulates charges on the surface via the magneto-electric coupling. In the weak magnetic field limit, the charge accumulated on the surface is proportional to σH as Q = σH Φ = θ 2π ne. (10) 4 FIG. 3. (color online). Charge Q accumulated on the surface in the z direction due to a uniform magnetic field with total flux Φ at kx = ky = 0. N = 199, Lz = 20 for all three panels. (a) and (b) consider the perturbative effect of an intra-site nearest neighbor hopping. (c) adds a random onsite potential characterized by ∆rp. The insets show the charge density at Φ = 3Φ0 without perturbations, and a closeup for the slope. The parameters in each panel are: (a) Hamiltonian H1(k) with h = 2, δ = 0.5, ∆S = 1, Γ = 1; (b) Hamiltonian H2(k) with h = 0, δ = 0, ∆S = 1, Γ = −4; (c) Hamiltonian H1(k) with h = 2, δ = 0, ∆S = 1, Γ = 1. In all panels, the accumulated charge Q is relative to the case when Φ = 0 and is summed over all particle species for the upper half of the sample at half fillings. A priori, θ needs not be quantized. However, analo- gous to the role played by time-reversal symmetry, chiral symmetry pins down θ to be mπ with an integer value m [13]. The fractional Hall conductivity, which cannot be removed by surface manipulations, emerges when m is odd [10, 12, 27]. The integer part of σH , however, depends on the details of the surface [10, 11, 28]. An in- tuitive picture is that the 2π ambiguity in θ results from the freedom to coat an integer quantum Hall layer on the surface, or equivalently to change the chemical potential and hence the Landau level occupancy of the surface in an orbital magnetic field. However, once a fixed surface Hamiltonian is defined, the adiabatic change in polariza- tion associated with the increase in magnetic flux does have a physical meaning. This ambiguity can be avoided in cold atom systems, where the precise Hamiltonian can be engineered, allowing a direct link between the winding number Γ and the charge accumulation rate θ/2π. Numerical results in Fig. 3 show that Γ = θ/π, which reveals that the magneto-electric polarization is a direct indication of the non-trivial bulk topological phase char- acterized by the integer winding number. To gain some intuition for why in our Hamiltonian the value Γ = θ/π is observed, consider how a Zeeman term and an orbital magnetic term produce different quantum Hall effects for a Dirac fermion: the latter leads to Landau levels with many intervening gaps and a chemical-potential depen- dence of the Hall effect, while the former leads to a sin- gle gap and only one value of the Hall effect. The sur- face term (9) apparently acts more like a Zeeman field in leading to a single gap and a unique value of the magneto-electric effect. We have confirmed this intuition by adding a strong orbital magnetic field in a single-layer Hamiltonian H1(k) (see Appendix B). Landau-level like bands are formed, and the charge accumulation rate is changed by an integer value by varying the chemical po- tential. B. Robustness to perturbations In real physical systems, the chiral symmetry may not be strictly observed. We therefore consider the effect of weak perturbations to the charge quantization. A natural term to add is an intra-site nearest neighbor hopping term: Hp(k) = ∆p(cos kx + cos ky + cos kz)I4. (11) This term breaks the chiral symmetry and permits nearest-neighbor hoppings within the same sublattice. Figures 3(a) and 3(b) show the charge accumulation on the surface with increasing magnetic field for various strengths of perturbations. The quantized slope is in- deed robust to small perturbations in the limit of weak magnetic field. Fig. 3(c) takes into account of random onsite potential with various perturbing strengths, and it again shows the robustness of the topological effect. This includes the effect of a weak harmonic trap typically present in cold atom systems. Note that strong perturba- tions destroy the topological phase. We also performed similar calculations by squeezing the entire uniform mag- netic field into a single flux tube with open boundaries. It shows the same linear relationship between the sur- face charge accumulation and the magnetic flux. This indicates the uniformity of magnetic field is not crucial to observe the topological magneto-electric polarization, which may be an advantage to experimental realization. V. EXPERIMENTAL IMPLEMENTATION AND DETECTION In this section, we present more details on the imple- mentation scheme with ultracold atoms. In Ref. [17], an experimental proposal for a three-band chiral TI was put 02468101214012345678Φ /Φ0Q / e Theory∆p=0∆p=0.1∆p=0.2∆p=0.3∆p=0.4Lz 50N15051520−0.0200.020.9511.05.450.5 (a)02468101214−30−26−22−18−14−10−6−20Φ /Φ0Q / e Theory∆p=0∆p=0.1∆p=0.2∆p=0.3∆p=0.4Lz 50N15051520−0.04−0.0200.020.040.99511.005−2−1.99−1.98 (b)02468101214012345678Φ /Φ0Q / e Theory∆rp=0∆rp=0.1∆rp=0.2∆rp=0.3∆rp=0.4Lz 50N15051520−0.0200.020.811.20.30.40.5 (c)5 FIG. 4. (Color online) Schematics to realize the Hamiltonian H1 with cold atoms. (a) Atomic level structure of 6Li and the four internal states used to represent the spin and orbital degrees of freedom. (b) The optical lattice is tilted with a homogeneous energy gradient along each direction. (c) Laser configurations to create the first term in Hrx. The superscript on each Rabi frequency denotes the polarization of the beam. forward. The realization scheme for the four-band Hamil- tonian studied here will be similar, with the atomic in- ternal states representing the spin and orbital degrees of freedom. In the previous sections, we studied two Hamiltonians H1(k) and H2(k). The latter involves next- nearest-neighbor hopping terms, which will be very chal- lenging for experiment to engineer. In the following, we demonstrate, however, that the implementation of H1(k) is possible with current technologies. H1(k) supports topological phases with index Γ = 0, 1,−2. It would be very exciting if experiment could simulate H1(k) and probe its nontrivial topological properties via magneto- electric polarization. The Hamiltonian H1(k) was written in momentum space in Sec. II. The real space equivalent can be ex- pressed as (for simplicity, we take h = 0, δ = 0.) H1 = r (12) Hrx + Hry + Hrz, (cid:88) (cid:16) (cid:16) Hrx = −i/2 ↑,r+x + a† a† (cid:16) ↑,r−x − a† − i/2 a† ↓,r−x (cid:16) Hry = −i/2 a† ↑,r+y + ia† − i/2 ↑,r−y − ia† a† ↓,r−y Hrz = −ia† ↑,r+zb↑,r − ia† ↓,r−zb↓,r + H.c. (cid:17) (cid:17) (b↑,r + b↓,r) (cid:17) (b↑,r − b↓,r) + H.c., (cid:17) (b↑,r − ib↓,r) (b↑,r + ib↓,r) + H.c., ↓,r+x ↓,r+y where x, y, z represents a unit vector along the x, y, z- direction of the cubic lattice, and aσ,r (bσ,r) denotes the annihilation operator of the fermionic mode at the a (b) orbital and site r with the spin state σ. Basically, all terms in the Hamiltonian are some spin superposi- tions from one orbital hopping to another orbital. In the following, we take the fermionic species 6Li, for in- stance, to illustrate the implementation scheme. Other fermionic atoms can also be used with suitable atomic levels. We make use of four internal states of the hy- perfine ground state manifold to carry two pseudospins and two orbitals as depicted in Fig. 4(a). On top of the cubic optical lattice, a linear tilt is assumed along each direction to break the left-right symmetry as does the Hamiltonian [Fig. 4(b)]. This linear tilt can be accom- plished with the natural gravitational field, the magnetic field gradient, or the gradient of a dc- or ac-Stark shift [17, 19 -- 21]. The hopping between orbitals can be acti- vated by two-photon Raman transitions. Here, we show how to get the first term in the Hamiltonian Hrx, which is −i/2(a† ↓,r+x) (b↑,r + b↓,r). Every other terms are of similar forms and can be likewise laser-induced. We may decompose it to four separate hoppings from b states at site r to a states at site r + x. As shown in Fig. 4(c), each of the hopping terms can be induced by a Raman pair: ↑,r+x + a† , Ωπ (Ωπ (Ωσ+ ↑ , Ωπ) → − i , Ωπ) → − i ↑,r+xb↑,r, (Ωσ− , Ωπ)→ − i 2 a† 2 a† ↓ , Ωπ) → − i 2 a† 2 a† ↑,r+xb↓,r, (Ωπ ↓,r+xb↑,r ↓,r+xb↓,r The superscript on each Rabi frequency denotes the po- larization of the respective beam. The relative phase and amplitude of the hoppings can be controlled by the laser beams. We have four free parameters here, ↑ , Ωσ− , Ωσ+ Ωπ ↓ , each of which can be adjusted individ- ually to yield the required configuration. These degrees of freedom ensure all other terms in the Hamiltonian can be produced in a similar way. One important aspect we need to be careful is that no spurious terms will be gen- erated with undesired laser coupling. This is guaranteed by energy matching and polarization selection rules. In the undressed atomic basis, all four internal states are at different energies (split by a magnetic field for example), so the four beams coupling b states to the excited states will not interfere with each other. In addition, the differ- ent detunings along each direction, ∆x,y,z, preempt the interference of beams that induce hoppings along differ- ent directions. There will be, however, some onsite spin- flipping terms induced by the laser beams, but those can be explicitly compensated by some r.f. fields. The above scheme is hence able to engineer the Hamil- tonian H1 and is feasible with current technologies. The 2P1/22S1/2F=1/2F=3/26Lia"b"b#a#(a)(b)b"b#a"a#2P1/2⌦⇡"⌦⇡#⌦⇡⌦+⌦(c)a"b"b#a#xa"b"b#a#actual experiment will be challenging since many laser beams are involved with careful detunings. Nonetheless, all these beams can be drawn from the same laser with small frequency shifts produced by an acoustic or elec- tric optical modulator. The uniform orbital magnetic field required to observe the topological polarization can be imprinted from the phase of the laser beams as an artificial gauge field [17, 19 -- 21]. Lastly, the gap open- ing term in Eq. (9) can be created by extra laser beams focused on the surfaces, producing an effective Zeeman splitting. Other gap opening mechanisms on the sur- face should also work since the magneto-electric polar- ization is a bulk effect. The accumulated charge will be detectable from atomic density measurements [29 -- 32]. Note that the density measurements do not need to be restricted to the surfaces, as a measurement for half of the sample produces good results, as shown in Fig. 3. One half of the sample can be removed to another state by shining a laser beam. The density on the other half of the sample can in turn be measured by time-of-flight imaging [32]. As we have demonstrated in the previous section, the charge quantization is very robust to per- turbations, so any weak perturbations introduced to the Hamiltonian, even those breaking the chiral symmetry in the bulk, should not alter detection results. VI. CONCLUSIONS In summary, we study the Z character of chiral topo- logical insulators by simulating the quantized magneto- electric effect of a nontrivial phase. We show that the Z character, not only the Z2 part, can be observed through magneto-electric polarization by properly gapping the system. An experimental scheme is also proposed for implementation and detection with cold atoms. This demonstrates explicitly how the topological invariant ap- pears in physical observables for chiral TIs and will be important for experimental characterization. ACKNOWLEDGMENTS S.T.W., D.L.D., the and L.M.D. the NBRPC (973 Program) are supported 2011CBA00300 by IARPA MUSIQC program, (2011CBA00302), the ARO and the AFOSR MURI program. J.E.M. acknowledges support from NSF DMR-1206515 and the Simons Foundation. K.S. acknowledges support from NSF under Grant No. PHY1402971 and the MCubed program at University of Michigan. Appendix A: Bulk-edge correspondence The bulk edge correspondence tells us that the bulk topological index should have a surface manifestation, typically through the number of gapless Dirac cones on 6 the surface. This is generally verified for lower topologi- cal index, such as 1 or 2. By imposing an open boundary condition along the z direction for chiral topological insu- lators of different index, we find that the winding number corresponds to the total number of Dirac cones counted for all inequivalent surface states (i.e. not counting de- generacies). Following Ref. [22] to take a quaternion power n, we can generalize the Hamiltonians in the main text from H1(k) and H2(k) to Hn(k). For the Hamiltonian H1(k) (i.e. n = 1), when h = 2, the winding number Γ = 1 guarantees the existence of 1 Dirac cone [Fig. 5(a)]. For the Hamiltonian H2(k) (i.e. n = 2), when h = 0, the winding number is Γ = 4. So there are two inequivalent surface states on each surface with two Dirac cones each [Fig. 5(b)]. In general, we have m inequivalent surface states with 1 Dirac cone each for n = m, 1 < h < 3, Γ = m, and m inequivalent surface states with 2 Dirac cones each for n = m,−1 < h < 1, Γ = 2m. These have been explicitly verified up to n = 3. Hence, the winding number Γ does correspond to the total number of Dirac cones for all inequivalent surface states. Appendix B: Surface orbital field and integer quantum Hall layers The 2π periodicity of the θ term is mathematically re- lated to the gauge freedom in the low-energy effective field theory. Physically, it is associated with the freedom to coat an integer quantum Hall layer on the surface, or equivalently to change the chemical potential and hence the Landau level occupancy of the surface in an orbital magnetic field. Here, we numerically verify this physi- cal intuition. To do that, we consider a single layer of the Hamiltonian H1(k), so the z component drops out. A strong uniform orbital field is added to the layer via Peierls substitution with the Landau gauge A = Bxy. Ba2 = 1 3 Φ0, where a is the lattice constant, and Φ0 is the flux quantum. For the Hofstadter Hamiltonian, this strong orbital field will produce three gapped Landau levels. Here, a similar structure is developed as shown in Fig. 6(a). There are six bands with the middle two bands gapless. The extra number of bands are due to the spin and orbital degrees of freedom. On top of the strong uniform magnetic field, an additional weak flux tube is inserted through the center lattice. By Laughlin's flux insertion argument, the charge accumulated around the flux tube should be Q/e = CΦ/Φ0, where C is the Chern number being an integer. Figures 6(b) and 6(c) [6(d) and 6(e)] show the charge polarization at a chemical potential µ1 [µ2]. From the slope, we infer that the first band has a Chern number C = 2 and the second band has a Chern number C = −4. So by changing the surface chemical potential, we could modify the charge accumulation rate by an integer. Alternatively, in the absence of this strong orbital magnetic field, with a surface gapping term HS in Eq. (9) of the main text, there is only one gap and no (a) (b) 7 FIG. 5. (Color online) Spectrum for the surface states showing the number of Dirac cones. The upper panels in (a) and (b) show the lowest conduction and highest valence band. The lower panels show the next two bands closest to the Fermi energy. (a) For H1(k), h = 2, δ = 0.5 with winding number Γ = 1 and 1 Dirac cone. (b) For H2(k), h = 0, δ = 0 with winding number Γ = 4 and 4 Dirac cones in total. The Γ point is displaced from the center for better display of the Dirac cones. FIG. 6. (Color online) (a) Energy spectrum for the one-layer Hamiltonian H1(k) with a strong uniform magnetic field and unit cell flux as 1 3 Φ0. An additional weak flux tube is inserted through the center lattice cell of the layer, with flux up to Φ/Φ0 = 0.1. (b) and (c) [(d) and (e)] correspond to the charge polarization with respect to the increasing flux tube at a chemical potential µ1 [µ2]. Charge is accumulated around the flux tube, and by changing the chemical potential and hence the Landau level occupancy, the charge accumulation rate can be modified by an integer. such integer quantum Hall layers. Therefore the Z char- acter of the winding number can be observed through such integrally quantized magneto-electric polarization measurements. [1] A. P. Schnyder, S. Ryu, A. Furusaki, and A. W. W. [2] A. Kitaev, in AIP Conf. Proc., Vol. 1134 (2009) p. 22. Ludwig, Phys. Rev. B 78, 195125 (2008). 00.020.040.060.080.1−0.2−0.16−0.12−0.08−0.040Φ/Φ0Q/e00.020.040.060.080.100.040.080.120.160.2Φ/Φ0Q/e051015200510152000.020.04xyQ/e0510152005101520−0.04−0.020xyQ/e050010001500−2−1012State NumberEnergyµ2µ1(a)(b)(c)(d)(e)8 [3] M. Z. Hasan and C. L. Kane, Rev. Mod. Phys. 82, 3045 (2010). [4] J. E. Moore, Nature 464, 194 (2010). [5] X.-L. Qi and S.-C. Zhang, Rev. Mod. Phys. 83, 1057 (2011). [6] C. L. Kane and E. J. Mele, Phys. Rev. Lett. 95, 146802 [20] M. Aidelsburger, M. Atala, M. Lohse, J. T. Barreiro, B. Paredes, and I. Bloch, Phys. Rev. Lett. 111, 185301 (2013). [21] H. Miyake, G. A. Siviloglou, C. J. Kennedy, W. C. Bur- and W. Ketterle, Phys. Rev. Lett. 111, 185302 ton, (2013). (2005). [22] D.-L. Deng, S.-T. Wang, and L.-M. Duan, Phys. Rev. B [7] B. A. Bernevig, T. L. Hughes, and S.-C. Zhang, Science 89, 075126 (2014). 314, 1757 (2006). [8] L. Fu, C. L. Kane, and E. J. Mele, Phys. Rev. Lett. 98, 106803 (2007). [23] D. M. Stamper-Kurn, A. P. Chikkatur, A. Gorlitz, S. In- ouye, S. Gupta, D. E. Pritchard, and W. Ketterle, Phys. Rev. Lett. 83, 2876 (1999). [9] D. J. Thouless, M. Kohmoto, M. P. Nightingale, and [24] S.-L. Zhu, B. Wang, and L.-M. Duan, Phys. Rev. Lett. M. Den Nijs, Phys. Rev. Lett. 49, 405 (1982). 98, 260402 (2007). [10] X.-L. Qi, T. L. Hughes, and S.-C. Zhang, Phys. Rev. B 78, 195424 (2008). [25] R. B. Laughlin, Phys. Rev. B 23, 5632 (1981). [26] G. Rosenberg, H.-M. Guo, and M. Franz, Phys. Rev. B [11] A. M. Essin, J. E. Moore, and D. Vanderbilt, Phys. Rev. 82, 041104 (2010). Lett. 102, 146805 (2009). [12] P. Hosur, S. Ryu, and A. Vishwanath, Phys. Rev. B 81, 045120 (2010). [13] S. Ryu, A. P. Schnyder, A. Furusaki, and A. W. W. Ludwig, New J. Phys. 12, 065010 (2010). [27] 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). [14] M. Lewenstein, A. Sanpera, V. Ahufinger, B. Damski, [28] A. M. Essin, A. M. Turner, J. E. Moore, and D. Van- A. Sen De, and U. Sen, Adv. Phys. 56, 243 (2007). [15] J. Dalibard, F. Gerbier, G. Juzeli¯unas, and P. Ohberg, Rev. Mod. Phys. 83, 1523 (2011). [16] N. Goldman, G. Juzeli¯unas, P. Ohberg, and I. B. Spiel- man, Rep. Prog. Phys. 77, 126401 (2014). [17] S.-T. Wang, D.-L. Deng, and L.-M. Duan, Phys. Rev. Lett. 113, 033002 (2014). [18] K. Shiozaki and S. Fujimoto, Phys. Rev. Lett. 110, 076804 (2013). [19] D. Jaksch and P. Zoller, New J. Phys. 5, 56 (2003). derbilt, Phys. Rev. B 81, 205104 (2010). [29] K. D. Nelson, X. Li, and D. S. Weiss, Nat. Phys. 3, 556 (2007). [30] W. S. Bakr, A. Peng, M. E. Tai, R. Ma, J. Simon, J. I. Gillen, S. Foelling, L. Pollet, and M. Greiner, Science 329, 547 (2010). [31] J. F. Sherson, C. Weitenberg, M. Endres, M. Cheneau, I. Bloch, and S. Kuhr, Nature 467, 68 (2010). [32] D.-L. Deng, S.-T. Wang, and L.-M. Duan, Phys. Rev. A 90, 041601(R) (2014).
1611.04943	1	1611	2016-11-15T17:19:29	Tunneling of Hybridized Pairs of Electrons through a One-Dimensional Channel	[ "cond-mat.mes-hall" ]	Recently, the electron transport through a quasi-one dimensional (quasi-1D) electron gas was investigated experimentally as a function of the confining potential. We present a physical model for quantum ballistic transport of electrons through a short conduction channel, and investigate the role played by the Coulomb interaction in modifying the energy levels of two-electron states at low temperatures as the width of the channel is increased. In this regime, the effect of the Coulomb interaction on the two-electron states has been shown to lead to four split energy levels, including two anti-crossings and two crossing-level states. Due to the interplay between the anti-crossing and crossing of the energy levels, the ground state for the two-electron model switches from one anti-crossing state for strong confinement to a crossing state for intermediate confinement as the channel width is first increased, and then returned to its original anti-crossing state. This switching behavior is related to the triplet spin degeneracy as well as the Coulomb repulsion and reflected in the ballistic conductance. Here, many-body effects can still affect electron occupations in the calculation of quantum ballistic conductance although it cannot vary the center-of-mass velocity.	cond-mat.mes-hall	cond-mat	a Tunneling of Hybridized Pairs of Electrons through a One-Dimensional Channel Godfrey Gumbs1, Danhong Huang2, Julie Hon1, M. Pepper3,4, and Sanjeev Kumar3,4 1Department of Physics and Astronomy, Hunter College of the City University of New York, 695 Park Avenue, New York, New York 10065, USA 2Air Force Research Laboratory, Space Vehicles Directorate, Kirtland Air Force Base, New Mexico 87117, USA 3London Centre for Nanotechnology, 17-19 Gordon Street, London, WC1H 0AH, United Kingdom 4Department of Electronic and Electrical Engineering, University College London, Torrington Place, London, WC1E 7JE, United Kingdom (Dated: August 27, 2018) Abstract Recently, the electron transport through a quasi-one dimensional (quasi-1D) electron gas was investigated experimentally as a function of the confining potential. We present a physical model for quantum ballistic transport of electrons through a short conduction channel, and investigate the role played by the Coulomb interaction in modifying the energy levels of two-electron states at low temperatures as the width of the channel is increased. In this regime, the effect of the Coulomb interaction on the two-electron states has been shown to lead to four split energy levels, including two anti-crossings and two crossing-level states. Due to the interplay between the anti-crossing and crossing of the energy levels, the ground state for the two-electron model switches from one anti-crossing state for strong confinement to a crossing state for intermediate confinement as the channel width is first increased, and then returned to its original anti-crossing state. This switching behavior is related to the triplet spin degeneracy as well as the Coulomb repulsion and reflected in the ballistic conductance. Here, many-body effects can still affect electron occupations in the calculation of quantum ballistic conductance although it cannot vary the center-of-mass velocity. 1 I. INTRODUCTION In this paper, we review the importance of many-body effects on the ballistic electron transport in a quasi-one-dimensional (1D) electron gas with varied confinement potential. This area of research has been receiving a considerable amount of attention in recent times ever since it was discovered that for a range of electron distribution and potential strength, the ground state of a 1D quantum wire splits into two rows with a Wigner lattice beginning to form. It was also demonstrated that when a perpendicular magnetic field is applied, a double-row electron formation may change completely into a single row due to an en- hanced confinement potential. Furthermore, it has been verified experimentally that weak confinement, in competition with the electron-electron interaction, causes the electron level occupation to reorder so that the ground state, conforming to the standard or common type, passes through the excited levels. The data in Ref. [1] show that the energy levels may be controlled by exploiting their separate geometric dependence on confinement and electron density. This means that in simulating the electron transport data, many-body effects must be considered. It has been well known that electrostatic potential confinement of a two-dimensional electron gas (2DEG) used to create a quasi-1D wire 2 gives rise to quantization of the con- ductance 3,4 in integer multiples of 2e2/h which is not affected by a weak electron-electron interaction 5. The long-range Coulomb interaction between electrons becomes relatively im- portant at low electron densities resulting in the formation of a 1D Wigner crystal 2,6,7. But, the role played by the Coulomb repulsion between electrons is also made greater until it overcomes the confinement potential, as the density is increased, at which point the ground state and one of the excited states are interchanged 8 which may result in hybridization and anti-crossing. In Refs. [1,9], conductance measurements were reported for weakly confined quantum wires in a 2DEG and determined by the boundaries of top split-gates. Experiment has shown that making the confinement potential less effective results in the appearance of two rows, accompanied by a sudden change in conductance G from zero to 4e2/h. This behavior may be attributed to the possibility that there was no coupling be- tween these two rows so that each row contributes independently and additively. Another way to account for this is to say that their energy eigenstates become hybridized and the resulting state causes a breakdown of the single-particle picture when the Coulomb interac- 2 tion becomes important. Recent investigations have confirmed that there exists a Coulomb interaction between the rows resulting in this anomalous jump in the conductance 10. The devices used in Refs. [1] and [9] were fabricated using electron beam lithography on 300 nm deep GaAs/AlGaAs heterostructures. Typically, the sample consisted of split gates, ∼ 0.4 µm long and 0.7 ∼ 1.0 µm wide, and a top gate of width ∼ 1.0 µm defined above the split gates, separated by a 200 nm layer of cross-linked polymethyl methacrylate. After partial illumination, the carrier density and mobility were estimated to be ∼ 1.5× 1011 cm−2 and ∼ 1.3 ∼ 3 × 106 cm2/Vs, respectively. The two-terminal conductance, G = dI/dV , was measured at 70 mK, using a 77 Hz voltage of 10 µV. Previously, the conductance through two laterally aligned but uncoupled parallel wires formed by surface gates have been shown to be the sum of the conductances of each individual wire, resulting in plateaus at multiples of 4e2/h, 11,12 indicating that hybridization of states within a wire is a many-body effect but not a single-particle one. Two side-by-side wires with very small inter-wire separation have lent support for the theory, i.e., there exists a coupling between the parallel wires 13,14. When this coupling between wires becomes strong, the electron wave functions hybridize, forming bonding and antibonding states, which manifest as anticrossings in the 1D energy subbands. Our model calculations 15 further confirm that the minimum energy gap between the states occurs at the point of anticrossing but is not given by the energy difference between the symmetric and antisymmetric states. In Fig. 1, we present a schematic illustration of a device used in the experiments carried out in Refs. [1,9], showing a pair of split gates and a top gate which adjust the confinement potential and carrier density by choosing their voltages suitably. Figure 1 also shows typical conductance features obtained with the device used in Ref. [1] as a function of the split-gate voltage, Vsg, for fixed top-gate voltage, Vtg. The traces for strong confinement are on the left-hand-side, whereas those for weak confinement are on the right. When the confinement is weak, the 2e2/h step may be lost and the 4e2/h appears as the lowest plateau 1. But, as the confinement is reduced further, the 2e2/h plateau is found to be restored. With regard to the interpretation that the carriers separate into two rows, the observed emergence of the crossing or anticrossing of energy levels needs explanation, preferably with the use of a quantum-mechanical theory. This may be verified by calculations of the kinetic, direct Coulomb and exchange energies of electrons in wires with intermediate widths as 3 well as for two extreme limits of very narrow and wide wires. In fact, we have recently demonstrated that these cases may be tracked down to the physical mechanism responsible for switching of the ground state as the wire width is varied from one value to another 15. In the presence of the Coulomb interaction, two-electron states may be employed as a basis set for constructing the anticrossing-level states 1 when two electrons travel ballistically along a quasi-1D channel. The corresponding calculations have shown that the significance of the Coulomb induced level anticrossing within a quantum wire may be adjusted by varying the confinement potential with a top gate voltage 15. There has been related work on conductance measurements of a quasi-1D wire having a quantum dot within the channel due to the presence of an impurity, as well as imperfec- tions in the device geometry. 16 These undesired features may result in results differing from integer multiples of 2e2/h for the conductance steps 17or oscillations superimposed on the conductance trace 18. Electron tunneling through the quantum dot in the channel as well as interference effects due to electron back-scattering from an impurity potential are believed to be responsible for these deviations in the values of the conductance plateaus of narrow quantum wires. 18 In the next section, we present a theoretical approach for calculating the conductance for a quasi-1D quantum wire at a low density of electrons. For this, we calculate the lowest energy eigenstates for a pair of interacting electrons within a confinement region. We explicitly determine the ground state of a dilute electron liquid and consequently the lowest conductance plateau. The complex two-electron tunneling 19,20 is not included in this review since it does not contribute to the formation of conductance-plateaus. Furthermore, we highlight below that there is a range of wire widths for which two-electron transport is mediated by anticrossing level states based on the Coulomb interaction, and, therefore, it is not possible to describe the conductance by using a single-particle formalism. II. THEORETICAL FORMULATION OF THE PROBLEM We will exploit the results for the eigenstates of a pair of interacting electrons within a harmonic confining potential 21,22. In Ref. [21], a symmetric harmonic potential was intro- duced. According to Kohn's theorem 22, for this potential the Coulomb interaction should only affect the relative motion of electrons but not that for the center-of-mass. It has been 4 pointed out that as a perpendicular magnetic field is increased, the ground state will oscil- late between a spin singlet and a spin triplet. Bryant 22 showed these electron correlation effects depend on the area of containment. By solving the Schrodinger equation exactly for two interacting electrons, it becomes clear how correlations may select the ground state and give rise to quasiparticles which participate in the transport. Coherent wavefunctions of two interacting electrons may be maintained during their transport along the channel if scattering by randomly distributed impurities and defects (negligible lattice scattering at low temperatures) is very small for high-mobility short chan- nel samples. Also, if the transmission coefficient for two injected electrons is almost perfect and the inelastic scattering between different two-electron states is nearly vanishing, we are able to use a quantum ballistic transport model for two interacting electrons then the Coulomb interaction between electrons in the channel can be fully taken into account. Bal- listic transport of two-electron clusters is assumed along the channel (y) direction. However, the finite width of a conduction channel in the transverse x direction gives rise to quanti- zation of the split cluster energy levels. Each level is assigned to have a free-electron-like kinetic energy for a ballistically moving noninteracting two-electron cluster. This leads to a hierarchy of subbands with quadratic wave-vector dependence. The cluster energy levels are y/m∗ with label p and j = 1, 2, 3, 4, where ky is the wave vector given by E(p) j,ky of electrons along the channel, m∗ is the electron effective mass and the "subband edges" E(p) are presented after Eq. (27) in terms of the dimensionless Coulomb integrals. For a j + 2k2 = E(p) j fixed linear electron density n1D, the two-electron chemical potential µp(T, n1D) within the channel may be calculated using 4(cid:88) ∞(cid:90) (cid:34) dky exp (cid:32)E(p) − µp j,ky kBT (cid:33) (cid:35)−1 + 1 , (1) n1D = 2 π j=1 0 which is expressed in terms of the temperature T of the system. Also, the chemical po- tentials for the left (L) and right (R) electrodes (with areal electron density n2D) are R (Vb, T, n2D) = µp(T, n1D) − eVb, respectively, L (Vb, T, n2D) = µp(T, n1D) + eVb and µ(p) µ(p) in the presence of a low biased voltage Vb. For quantum ballistic charge/heat transport of two interacting electrons in the channel, the charge (α = 0) and heat (α = 1) current densities are calculated from 23 5 J (α)(Vb, T, n1D) = ∞(cid:90) 4(cid:88) j=1 0 (−2e)1−α π − µp)α(cid:12)(cid:12)vj,ky (cid:12)(cid:12)(cid:104) dky (E(p) j,ky (cid:105) ) , (2) fL(E(p) j,ky ) − fR(E(p) j,ky where vj,ky = ky/m∗ is the half of the total group velocity of the two-electron state, fL(E(p) j,ky and fR(E(p) ) are the Fermi functions for noninteracting two-electron states in the left (L) j,ky and right (R) electrodes with chemical potentials µ(p) R , respectively, for noninter- L and µ(p) ) acting two-electron states. In formulating our theory, we start by considering two interacting electrons in an elon- gated quantum dot within a quantum wire. For this quantum dot, an anisotropic confine- ment is assumed with a shorter confining length ξx across the channel (x direction) with a longer confining length ξy along the channel (y direction). Here, the Hamiltonian for two interacting electrons can be written as H(r1, r2) = H0(r1) + H0(r2) + UC(r1 − r2) UC(r1 − r2) = H0(ri) = 2m∗ + Vc(ri) , p2 i e2 4π0rr1 − r2 , (3) where i = 1, 2 labels each electron, pi = −i∇ri, Vc(ri) is the confining potential for the conduction channel, H0(ri) is the single-electron Hamiltonian and UC(r1 − r2) represents the electron-electron interaction in a medium with background dielectric constant r. The single-particle eigenstates φα(r) ≡ (cid:104)rφα(cid:105) can be determined from the Schrodinger equation H0(r) φα(r) = εα φα(r), where the eigenfunctions {φα(r)} constitute a complete orthonormal set φα(cid:105) in the single particle Hilbert space. After properly anti-symmetrizing the two particle basis, including both the orbital and spin parts, we obtain Ψαm,βn(r1, s1; r2, s2) 1√ 2 [φαm(r1)χm(s1) φβn(r2)χn(s2) − φβn(r1)χn(s1) φαm(r2)χm(s2)] , = (4) where χm(si) is the spinor for the spin state of an electron. Here, the basis states in Eq. (4) are degenerate eigenstates of the noninteracting Hamiltonian H0(r1, r2) ≡ H0(r1) + H0(r2) with H0(r1, r2) Ψαm,βn(r1, s1; r2, s2) =(cid:0)εαm +εβn (cid:1) Ψαm,βn(r1, s1; r2, s2). We restrict ourselves 6 in the following to the case when only the lowest two orbitals αm = α and βn = β are populated. If we assume that two electrons stay in the same spin state with χ1 = χ2 = ↑(cid:105), the spinor part can be factored out, giving rise to Ψ(r1, s1; r2, s2) = (5) However, if we assume that two electrons remain in opposite spin state with χ1 = ↑(cid:105) and χ2 = ↓(cid:105), then the result becomes [φα(r1) φβ(r2) − φβ(r1) φα(r2)] χ1(s1)χ1(s2) . 1√ 2 Ψ(r1, s1; r2, s2) = 1√ 2 [φα(r1)χ1(s1) φβ(r2)χ2(s2) − φβ(r1)χ2(s1) φα(r2)χ1(s2)] . (6) Consequently, the subspace of the lowest states for two independent electrons can be spanned by the following basis 24 Ψ1(r1, s1; r2, s2)(cid:105) = Ψ2(r1, s1; r2, s2)(cid:105) = Ψ3(r1, s1; r2, s2)(cid:105) = Ψ4(r1, s1; r2, s2)(cid:105) = Ψ5(r1, s1; r2, s2)(cid:105) = Ψ6(r1, s1; r2, s2)(cid:105) = 1√ φα(r1)φα(r2) ( ↑(cid:105)1 ↓(cid:105)2 − ↓(cid:105)1 ↑(cid:105)2) , 2 1√ φβ(r1)φβ(r2) ( ↑(cid:105)1 ↓(cid:105)2 − ↓(cid:105)1 ↑(cid:105)2) , 2 [φα(r1)φβ(r2) + φβ(r1)φα(r2)] ( ↑(cid:105)1 ↓(cid:105)2 − ↓(cid:105)1 ↑(cid:105)2) , [φα(r1)φβ(r2) − φβ(r1)φα(r2)] ↑(cid:105)1 ↑(cid:105)2 , [φα(r1)φβ(r2) − φβ(r1)φα(r2)] ↓(cid:105)1 ↓(cid:105)2 , 1 2 1√ 2 1√ 2 [φα(r1)φβ(r2) − φβ(r1)φα(r2)] ( ↑(cid:105)1 ↓(cid:105)2 + ↓(cid:105)1 ↑(cid:105)2) . 1 2 (7) Here, the six components of these two-electron sets are orthonormal, i.e., (cid:104)ΨmΨn(cid:105) = δm,n. Moreover, using the above six states, the interacting Hamiltonian matrix can be cast into the form of 7  H = 2εα + U11 U21 U31 0 0 0 U12 2εβ + U22 U13 U23 U32 εα + εβ + U33 0 0 0 0 0 0 0 0 0 εα + εβ + U44 0 0 0 0 0 0 εα + εβ + U55 0 0 0 0 0 0 εα + εβ + U66  , (8) where Umn = (cid:104)ΨmUCΨn(cid:105) = U∗ notations for the Coulomb matrix elements as nm is the Coulomb matrix element. Explicitly, we define the Mαβ; α(cid:48)β(cid:48) ≡ (cid:104)φα, φβUCφα(cid:48), φβ(cid:48)(cid:105) , Mβα; β(cid:48)α(cid:48) = Mαβ; α(cid:48)β(cid:48) . (9) Introducing the Fourier transform to the Coulomb potential, we are able to express its matrix elements using the Coulomb and exchange integrals, i.e., 1 r1 − r2 = This gives rise to (cid:90) d2q 2π eiq·(r1−r2) q . (10) (11) (12) where we have employed the form factor, given by U11 = Mαα; αα ≡ (cid:104)φα, φαUCφα, φα(cid:105) d2q , Fαα(q)2 q (cid:90) = e2 20r (cid:90) Fαβ(q) = 1 2π d2r φ∗ α(r) eiq·r φβ(r) = F∗ βα(−q) . Similarly, we can obtain other nonzero matrix elements from 8 U22 = Mββ; ββ = U12 = Mαα; ββ = e2 20r e2 20r (cid:90) d2q (cid:90) d2q q Fββ(q)2 , q Fαβ(q)Fαβ(−q) , U13 = = U23 = = 1√ (Mαα; αβ + Mαα; βα) 2 e2 √ 20r 2 1√ (Mββ; αβ + Mββ; βα) 2 e2 √ 20r 2 (cid:90) d2q q [Fαα(q)Fαβ(−q) + Fαβ(q)Fαα(−q)] , (cid:90) d2q q [Fβα(q)Fββ(−q) + Fββ(q)Fβα(−q)] , U33 = Mαβ; αβ + Mαβ; βα = U44 = Mαβ; αβ − Mαβ; βα = e2 20r e2 20r = U55 = U66 . (cid:90) d2q (cid:90) d2q q q (cid:2)Fαα(q)Fββ(−q) + Fαβ(q)2(cid:3) , (cid:2)Fαα(q)Fββ(−q) − Fαβ(q)2(cid:3) (13) After we diagonalize the Hamiltonian matrix in Eq. (8), both the energy eigenvalues and associated eigenstates can be obtained in a straightforward way. Now, let us consider explicitly a harmonic confining potential for electrons within the xy−plane, i.e., m∗(cid:0)ω2 yy2(cid:1) , (14) with ωy (cid:28) ωx. As a result, the orbital parts of the single-particle eigenstates can be written down as xx2 + ω2 Vc(r) = 1 2 φm,n(x, y) = ψm(x) ψn(y) , εm,n = ωx (m + 1/2) + ωy (n + 1/2) . (15) We will choose two eigenstates, φα(x, y) = ψ0(x) ψ0(y) and φβ(x, y) = ψ1(x) ψ0(y), where ψn(x) is the one-dimensional oscillator wavefunction for n = 0, 1, 2, ··· . This yields (cid:19)1/2 (cid:18) (cid:18) 1 πξxξy 1 2π exp (cid:18) (cid:19) (cid:19) (cid:19) − y2 2ξ2 y , (cid:18) (cid:19) exp xξ2 −q2 x 4 − x2 2ξ2 x (cid:18) exp exp yξ2 −q2 y 4 φα(x, y) = Fαα(q) = 9 , (16) as well as φβ(x, y) = Fββ(q) = (cid:19)1/2(cid:18)2x (cid:18) 1 (cid:1) exp (cid:0)2 − q2 (cid:18) 2πξxξy 1 4π √ iqxξx 2π 2 xξ2 −q2 x 4 xξ2 x ξx (cid:19) (cid:18) (cid:19) (cid:19) (cid:18) (cid:19) (cid:18) (cid:19) (cid:18) − x2 2ξ2 x exp yξ2 −q2 y 4 exp xξ2 −q2 x 4 (cid:19) , (cid:18) − y2 2ξ2 y (cid:19) , exp yξ2 −q2 y 4 Fαβ(q) = Here, ξx =(cid:112)/m∗ωx and ξy =(cid:112)/m∗ωy are the confining lengths, and then, the system (17) exp exp . dimension may be significantly larger along the y-direction compared to that in the x- direction. We know from Eqs. (15) and (16) that eigenstates φα(r) and φβ(r) have opposite parity. Consequently, we find Fαα(q) = Fαα(−q), Fββ(q) = Fββ(−q), and Fαβ(q) + Fαβ(−q) = 0. This directly leads to u13 = u23 = 0. Moreover, for a harmonic potential the block part of the truncated Hamiltonian in Eq. (8) becomes 2εα + U11 U12 0 H3×3 =  . U12 2εβ + U22 0 0 0 εα + εβ + U33 From this, we obtain two split energy eigenvalues for the states Ψ1 and Ψ2, given by E1,2 ≡ E± = εα + εβ + (U11 + U22) ± D , 1 2 and the uncoupled energy level E3 = εα + εβ + U33 for the state Ψ3, as well as the triple- degenerate energy levels E4 = E5 = E6 = εα + εβ + U44 for the states Ψ4, Ψ5 and Ψ6. In Eq. (19), the energy-level coupling D =(cid:112)[εβ − εα + (U22 − U11)/2]2 + U122, and the level splitting is E+ − E− = 2D > 0. For evaluating u11, u22, u33, u12 and u44, we require the following Coulomb integrals: (18) (19) (cid:90) d2q (cid:90) π q Iαα; αα = 2πLy Fαα(q)2 = Ly 2π dθ dq e−α(θ)q2 = = Ly√ 2π dθ x cos2 θ + ξ2 (ξ2 0 y sin2 θ)1/2 Ly√ 2πξy 0 dθ [1 + (γ2 − 1) cos2 θ]1/2 , (20) (cid:90) 2π 0 dθ(cid:112)α(θ) √ Ly 4 π (cid:90) ∞ (cid:90) π 0 (cid:90) 2π 0 = 10 where we have used qx = q cos θ, qy = q sin θ, θ is the angle between the wave vector q and x axis, γ ≡ ξx/ξy, and α(θ) = (ξ2 y sin2 θ)/2. Additionally, we obtain x cos2 θ + ξ2 (cid:90) d2q (cid:90) π q Iββ; ββ = 2πLy Fββ(q)2 = Ly 8π = Ly 4π 0 dθ [I1(θ) + I2(θ) + I3(θ)] , (cid:90) 2π (cid:90) ∞ dθ 0 0 dq(cid:0)2 − q2ξ2 x cos2 θ(cid:1)2 e−α(θ)q2 where (cid:90) ∞ 0 I1(θ) = 4 √ 2 π(cid:112)α(θ) , dq e−α(θ)q2 = (cid:90) ∞ (cid:90) ∞ 0 0 dq q2 e−α(θ)q2 = − √ 3 dq q4 e−α(θ)q2 = x cos2 θ √ πξ2 (cid:112)α3(θ) 8(cid:112)α5(θ) x cos4 θ πξ4 . I2(θ) = −4ξ2 x cos2 θ I3(θ) = ξ4 x cos4 θ By combining the results for I1(θ), I2(θ) and I3(θ), this leads to (21) (22) , (cid:90) π 0 Ly√ 2π x cos2 θ ξ2 x cos2 θ + ξ2 ξ2 γ2 cos2 θ Iββ; ββ = (cid:34) (cid:20) × × 1 − 1 − In a similar way, we find Iαβ; αβ = 2πLy = Ly 2π where dθ x cos2 θ + ξ2 (ξ2 y sin2 θ)1/2 3ξ4 x cos4 θ x cos2 θ + ξ2 3γ4 cos4 θ + + y sin2 θ 4(ξ2 y sin2 θ)2 (cid:90) π 0 = Ly√ 2πξy (cid:35) . dθ [1 + (γ2 − 1) cos2 θ]1/2 (23) (cid:21) 1 + (γ2 − 1) cos2 θ 4[1 + (γ2 − 1) cos2 θ]2 (cid:90) d2q (cid:90) π q Fαα(q)Fββ(−q) = dθ [J1(θ) + J2(θ)] , (cid:90) 2π (cid:90) ∞ dθ 0 0 Ly 4π dq(cid:0)2 − q2ξ2 x cos2 θ(cid:1) e−α(θ)q2 (24) 0 (cid:90) ∞ 0 J1(θ) = 2 √ π(cid:112)α(θ) , dq e−α(θ)q2 = (cid:90) ∞ 0 11 J2(θ) = −ξ2 x cos2 θ dq q2 e−α(θ)q2 = − √ 4(cid:112)α3(θ) x cos2 θ πξ2 . (25) By combining these results for J1(θ) and J2(θ), we have (cid:90) π (cid:90) π (cid:90) π 0 0 (ξ2 Ly√ 2π Ly√ 2πξy √ Ly 2πξy 2 0 (cid:34) (cid:20) 1 − 1 − dθ y sin2 θ)1/2 x cos2 θ + ξ2 dθ [1 + (γ2 − 1) cos2 θ]1/2 2 + (γ2 − 1) cos2 θ [1 + (γ2 − 1) cos2 θ]3/2 . dθ Iαβ; αβ = = = (cid:35) (cid:21) 2(ξ2 ξ2 x cos2 θ x cos2 θ + ξ2 γ2 cos2 θ y sin2 θ) 2[1 + (γ2 − 1) cos2 θ] (26) (27) The last integral is calculated as Iαβ; βα = Iαα; ββ = 2πLy ξ2 (cid:90) π (cid:90) π dθ 0 = = 0 √ Ly 8 π √ Ly 2πξy 2 (cid:90) d2q q Fαβ(q)Fαβ(−q) = x cos2 θ(cid:112)α3(θ) (cid:90) π (ξ2 dθ = 0 √ 1 2π 2 γ2 cos2 θ dθ [1 + (γ2 − 1) cos2 θ]3/2 . (cid:90) 2π (cid:90) ∞ dθ 0 0 x cos2 θ ξ2 Ly 4π dq q2ξ2 x cos2 θ e−α(θ)q2 x cos2 θ + ξ2 y sin2 θ)3/2 It is important to note that we have assumed a quasi-continuum energy spectrum for a traveling quasiparticle in the longitudinal direction, in contrast with split energy levels in the transverse direction. In this case, the pair of electrons forming the quasiparticle always have the lowest transverse energy plus a free electron-like kinetic energy, resulting from the longitudinal motion. However, the Coulomb interaction between a pair of electrons in this cluster will significantly modify the "subband edges" (E(p) j ) due to quantization in the transverse direction. All quasiparticles, except the transported one, may be treated as a "background" making up the total electron density and have a quasiparticle chemical potential µp which is deter- j = E(p) j,ky=0 are calculated as E(p) + = ε0 + ε1 + (u11 + u22)/2 + ∆C, E(p) mined using Eq. (1). For the interacting two-electron states, by using the above derivations, 1 ≡ E(p)− = ε0 +ε1 +(u11 +u22)/2−∆C, their energy levels E(p) 2 ≡ E(p) E(p) (cid:113) 4 = ε0 + ε1 + u44, where that Coulomb coupling term for the two-electron anticrossing states is given by [ε1 − ε0 + (u22 − u11)/2]2 + u122. In this notation, the single-particle energy levels for the harmonic-potential model with anisotropic harmonic frequencies ωx and ωy in the transverse (x) and longitudinal (y) directions, respectively, are: ε0 = (ωx +ωy)/2 and ε1 = 3 = ε0 + ε1 + u33 and E(p) ∆C = 12 (3ωx +ωy)/2, while the introduced Coulomb interaction energies are found to be u11/Ec = 1 I11,11, u33/Ec = N0N1 (I01,01 + I01,10) and 0 I00,00, u12/Ec = N0N1 I00,11, u22/Ec = N 2 N 2 u44/Ec = N0N1 (3I01,01 − I01,10), where Ec = e2/4π0rLy in terms of the channel length Ly and the background dielectric constant r, Nn = {exp[(εn − µ0)/kBT ] + 1}−1 (n = 0, 1) is the single-particle level occupation factor, and µ0(T, n1D) is the single-electron chemical potential within the channel. At the time when a quasiparticle enters a conduction channel, it will occupy single-particle energy levels ε0 and ε1, i.e., occupying the same one or different levels. Such a selection is determined from the subband occupation by the sea of electrons within the channel. After these two noninteracting electrons are injected into the channel, they will interact with each other through either the intrasubband or intersubband Coulomb coupling. The ballistic injection of two noninteracting electrons and the existence of a sea of electrons in the conduction channel are reflected through the inclusion of these two level occupation factors. The Coulomb integral is represented by Iαβ,α(cid:48)β(cid:48)(γ) for α, β, α(cid:48), β(cid:48) = 0, 1 if we only consider interacting electron states formed from the lowest ('0') and first excited ('1') single-particle states. The channel width Wx and length Ly are directly related to the frequencies ωx and ωy of the 2D harmonic-confining potential by Wx = (cid:112)4/m∗ωx and Ly = (cid:112)4/m∗ωy, respectively. Therefore, we get the simple relations, i.e., ξx = (cid:112)/m∗ωx = Wx/2, ξy = (cid:112)/m∗ωy = Ly/2, and and γ ≡ ξx/ξy = Wx/Ly ≡ R. Furthermore, for R (cid:29) 1, we find that Iαβ,α(cid:48),β(cid:48)(R) scales as 1/R for α, β, α(cid:48), β(cid:48) = 0, 1 . In the random-phase approximation (RPA), the static dielectric function at low temper- ature for screening for an electron density n1D and r in the channel, may be expressed as 25 (cid:18) m∗e2 (cid:19) (cid:18)qWx (cid:19) 1D(q) = 1 − (28) where the wave vector q ∼ kF = πn1D/2. For the parameters chosen in our numerical calculations, we found that the effect due to static screening may be neglected. On the 2π0r2n1D ln , 2 other hand, the static dielectric function for shielding by surface gate electrodes of the electron-electron interaction may be modeled as G(q) = 1 + coth(qd), for which we may take q ∼ kF = πn1D/2 and d represents the gate insulator thickness 26. For the parameters used in our numerical calculations, we found that shielding of the interaction between two- 13 electron states may also be neglected. From the calculated J (α=0)(Vb, T, n1D) in Eq. (2), the electrical conductance G(T, n1D) for interacting two-electrons may be expressed as 23 G(T, n1D) = J (α=0)(Vb, T, n1D) Vb . (29) We now present our numerical results and their relationship to recently reported experimen- tal data in Ref. [1]. III. NUMERICAL RESULTS AND EXPERIMENTAL DATA In our numerical calculations, we use the following parameters: T = 10 mK, Vb = 0.01 mV, Ly = 400 nm, r = 12, m∗ = 0.067 m0 (m0 is the free-electron mass). The chosen R values are indicated in the figure captions. Specifically, we denote the quantum ballistic transport of two-electron states with anticrossing levels through a conduction channel as one moving through either one of two states E(p)± . For clarity, we point out that as two electrons are injected into a conduction channel, they may occupy specific single-particle subbands for their ballistic transport. The selection rule is determined by the occupation factor of the electrons already sustained within the conduction channel. During the time interval that the two injected moving electrons remain within the channel, they may interact with each other through either the intrasubband or the intersubband Coulomb coupling. We emphasize that the linear density for confined electrons within the channel may be kept constant even when the channel width is varied. However, for this to occur, the Fermi energy must automatically adjust itself to accommodate all electrons and additional subbands will be populated with reduced energy level separations. Specifically, although the Fermi energy is reduced, the number of electrons in the channel is not changed at all. Furthermore, even when the Fermi energy is reduced, the second level can still be populated due to reduced level separation at the same time so as to keep the number of electrons in the channel a constant. Clearly, enhancement of the Coulomb interaction is not solely determined by the electron density, since it also depends on how electrons are distributed. For the Coulomb effect on the two-electron states, the inclusion of a new populated two-electron state, where one electron stays in a lower-energy level while 14 the other electron populates a higher level, will introduce a new Coulomb-interaction channel for the two-electron states. A. Two-Electron Energies within the Channel j We know that as the transverse confinement becomes weaker (or the R value is increased), of a two-electron state will decrease like as 1/R2 the kinetic part of the energy levels E(p) for fixed Ly. On the other hand, the Coulomb interaction only scales as 1/R as per our discussion preceding Eq. (28). Consequently, the significance of the Coulomb interaction is expected to increase relatively by increasing R. Moreover, the level separation will be reduced by increasing R, leading to occupation of the second energy level for fixed electron density. Therefore, the additional Coulomb repulsion between two electrons on different 3 4 (as N1 > 0) in the region of single-particle energy levels must be considered. This effect can be seen from Figs. 2(b), 2(c) and 2(d) as the upward shifting of energy levels E(p)− and E(p) R > 1 as n1D ≥ 0.2 × 105 cm−1. At the same time, the E(p) level is pushed upward above level due to the enhanced Coulomb interaction for R > 1. On the other hand, the E(p) 3 for the E(p) + two-electron state, which is associated with two excited-state electrons, it is largely dominated by the kinetic energy part for the whole range of R shown in this figure. When n1D is further increased, the Coulomb repulsion effect pushes into the intermediate confinement regime (R ∼ 1) in Fig. 2(d). Due to the combined effect of these two factors, we observe the recovery of the ground state E(p)− level in Fig. 2(d) for large values of R and n1D (where the Coulomb energy is dominant) from that in Fig. 2(a) for small values of R and n1D (where the kinetic energy of electrons is dominant). It is interesting to see that the Coulomb interaction between electrons stands out to give rise to a pushing up of three energy levels and the recovery of the the ground E(p)− level at the same time in an intermediate confinement regime (R (cid:38) 1) between the strong (scaling as fast drop 1/R2 for R < 1) and weak (scaling as slow drop 1/R for R (cid:29) 1) confinement regimes. B. Ballistic Conductance within a quasi-1D Channel The recovery of the ground-state in Fig. 2 plays a significant role on both the distribution of conductance plateaus and the interplay from interaction effects, as displayed in Fig. 3. We 15 know the Coulomb coupling may be neglected for small n1D, where the 2e2/h conductance plateau is found for the interacting two-electron state as shown in 3(a) with almost all values of R. As n1D increases to 0.2× 105 cm−1 in 3(b), the 2e2/h plateau shown in 3(a) disappears except for its reappearance very close to R = 2.0. If the value of n1D gets even larger, as seen from Figs. 3(c) and 3(d), the new 4e2/h conductance plateau occurs for an interacting two- electron state, corresponding to the population of the degenerate lowest E(p) energy levels after their level crossing with another E(p)− state. However, when R further increases above one in the very-weak confinement regime, the ground-state recovery, as discussed in 3 and E(p) 4 Figs. 2(c) and 2(d), enforces the reoccurrence of the 2e2/h conductance plateau due to the Coulomb repulsion between electrons in the central region of the channel. C. Dependence of interacting electron energy on Linear Density When electrons interact with each other, their energy levels E(p) are expected to depend on the electron density n1D, as shown in Fig. 4. When the geometry ratio R = Wx/Ly is small for strong confinement in Fig. 4(a), only the ground state E(p)− is affected by varying n1D due primarily to N0 (cid:54)= 0 in this case. As R is increased to 0.6 in Fig. 4(b), both the level crossing between E(p)− of the anticrossing state with the degenerate state E(p) 3 = E(p) + states occur at lower densities. As R > 1, and the level anticrossing between E(p)− and E(p) the Coulomb interaction between electrons becomes much stronger, as presented in Figs. 4(c) and 4(d). Therefore, both E(p) levels are pushed up significantly at higher 3 densities (i.e., N1 > 0), leading to a recovery of the ground state to E(p)− . At the same time, the E(p) state in Figs.4(c) and 4(d) changes from the degenerate ground state at lower n1D to 4 and E(p) 4 j 4 the highest-energy state at higher n1D. Furthermore, under a transverse magnetic field, we expect that the E(p) 3 while the degenerate E(p) 4 state should decouple from the magnetic field due to total spin S = 0, state with total spin S = 1 will be split into three by the Zeeman effect, leading to new e2/h and 3e2/h conductance plateaus 27. D. Conductance for two interacting and noninteracting Electron Pairs Figure 5 presents a comparison of the conductance G for both a noninteracting and interacting two-electron state in the range of 0.1 ≤ R ≤ 1. For very strong confinement in 16 4(a), the Coulomb-interaction effect becomes negligible in comparison with the dominant kinetic energy of electrons and a conductance 2e2/h plateau remains with increasing n1D. On the other hand, as R goes up to 0.4 in 5(b) and 0.6 in 5(c) for cases with strong confinement, although G for a noninteracting two-electron state remains largely unchanged, for an interacting two-electron state, the conductance 2e2/h plateau in Fig. 5(a) is completely destroyed by the Coulomb interaction and replaced by a new 4e2/h plateau. This unique feature is attributed to the result of both a level-crossing and a level anticrossing observed in Fig. 4(b). However, the new 4e2/h conductance plateau is severely perturbed at higher densities by a sharp spike and a followed by a deep dip to the lower 2e2/h plateau as R = 1 for intermediate confinement in 5(d). Although the dimensionless Coulomb integrals do not depend on the linear electron den- j ∼ {uij} for a two-electron cluster is proportional to the sity n1D, the energy levels E(p) occupation factors (N0 and N1) in addition to these Coulomb integrals. Moreover, these occupation factors are determined by the chemical potential µn for noninteracting electrons through the Fermi function for fixed n1D. On the other hand, the cluster chemical potential µp, determined by Eq. (1), controls the behavior of cluster ballistic transport in the presence of a bias voltage Vb. E. Conductance for weak Confinement We present in Fig. 6 the change in the conductance plateau with increasing R in the weak confinement regime. When R ≥ 1.6, conductance plateaus for the noninteracting two- electron state are washed out in Figs. 6(b)- 6(d) due to very small single-particle energy level separation compared to the thermal energy kBT . It is also evident that the incomplete 4e2/h conductance plateau in Fig. 6(a) for the interacting two-electron state is completely destroyed in this regime. However, the recovery of the single-particle-like 2e2/h plateau, as displayed in Fig. 5(a), is found in Fig. 6. Additionally, the 2e2/h plateau further expands and extends to lower and lower electron densities as R increases to 2.0 in Fig. 6(d). This unique reoccurrence feature can be fully accounted for by the rising energy levels at higher densities due to the relatively enhanced Coulomb repulsion as shown in Figs. 4(c) and 4(d). As displayed in Fig. 4, both E(p) 3 and E(p) 4 remain degenerate for all chosen values of n1D 17 as far as R < 1 or alternatively for only small n1D values as R > 1.2. The level-crossing between E(p)− and the degenerate levels E(p) is the reason behind the upward jump of the conductance from 2e2/h to 4e2/h, as can be seen from Fig. 5. However, the degeneracy levels may be lifted by an enhanced Coulomb repulsion for R > 1 as well of the E(p) as for large values of n1D. Consequently, the subsequent downward dip in the conductance 3 and E(p) 4 3 and E(p) 4 from 4e2/h to 2e2/h is observed in Fig. 6. In order to acquire a complete picture of the quantum ballistic transport of interacting two-electron states passing through a quasi-1D conduction channel, we present the contour plots of electron conductance G as functions of R and n1D in Fig. 7 for both noninteracting and interacting two-electron states as a direct comparison. By comparing Fig. 7(a) with Fig. 7(b), we find that the effect of the Coulomb coupling becomes most dominant in the upper right-hand corner of Fig. 7(b) within a weak confinement regime and with a relatively high electron density at the same time. In this case, a gradually increasing conductance for noninteracting electrons is replaced by a 2e2/h conductance plateau. This is due to the Coulomb repulsion in interacting two-electron states. In addition, another 4e2/h conduc- tance plateau shows up in the lower right-hand corner of Fig. 7(b). This is separated by a spike in G from the upper right-hand corner. In this region, confinement is intermediate or strong but the electron density is high. In our numerical results presented above, we limit the bias voltage Vb to a very small value (0.01 mV), where G becomes essentially independent of Vb. The increase of Vb can induce a "hot-carrier" effect and reduce the ballistic conductance with increasing temperature, as presented in Fig. 8, where the conductances G for both noninteracting [in Fig. 8(a)] and interacting [in Fig. 8(b)] with Vb = 0.05 mV are compared with each other. From Figs. 7(a) and 7(b), we find G for noninteracting electrons has been changed qualitatively for different values of Vb, although G for electron clusters is only modified quantitatively. We further demonstrate such a bias dependent effect on G of electron clusters in Fig. 8(c), where three different values of Vb are chosen for n1D = 0.3 × 105 cm−1. As can be seen from Fig. 8(c), the spike in G is significantly broadened and the plateau of G on both sides of the spike is reduced simultaneously with increasing Vb. This is similar to the hot-carrier effect with increased T . We now turn our attention to the experimental aspects which are related to the preceding 18 theoretical results. Two-terminal differential conductance measurements were performed with an excitation voltage of 10 µV at 73 Hz using the Oxford Instruments cryofree dilution refrigerator, where the device is estimated to have an electron temperature of around 70 mK. In order to the test the samples, a top gated, split gate device provided additional con- finement to the quasi-1D electrons. This allowed us to vary the confinement from being very strong (zero top gate) to very weak (very negative top gate voltage). In the present study, as shown in Fig. 9, the top gate voltage, Vtg, is varied from −7.21 V (left) to −9.19 V (right) in steps of 90 mV. Figure 9 shows a plot of the differential conductance in (a) for the device as a function of the split gate voltage Vsg for various values of the top gate voltage Vtg, as well as in (b) for the transconductance (dG/dVsg) drawn from the data in (a). As can be seen from Fig. 9(a), as the confinement is reduced, the 2e2/h conductance plateau is weakened. If the confinement is further reduced, the 2e2/h plateau disappears and is replaced by a direct jump in conductance to the (rounded) 4e2/h plateau at both Vtg = −8.47 and −8.56 V (indicated by arrows). Eventually the first plateau at 2e2/h is recovered on further reducing the confinement to Vtg = −9.19 V (right-most red curve). In comparison with our calculated results presented in Figs. 5 and 6, we find the sequence from the appearance of the 2e2/h conductance plateau for small values of R. We have obtained results for strong confinement as well as the 4e2/h conductance plateau for intermediate confinement, and again the 2e2/h conductance plateau in the weak-confinement regime which is preceded by a double-kink structure. In addition, from Fig. 9(b) we know the crossing/anticrossing of the ground state and the first excited states depends on the confinement strength. Here, when Vtg is around −8.6 V, the ground state and the first excited states cross, leading to energy reversal such that previous excited state becomes the new ground state, and then, the previous ground state further moves up in the energy and anticrosses with the second excited state. This observation qualitatively agrees with the calculated results presented in Fig. 2. We would like to emphasize that the appearance/disappearance/reappearance of a con- ductance plateau has been qualitatively reproduced in our numerical calculations. This is displayed in Fig. 3, although some non-monotonic features in our reported results are not verified experimentally. We acknowledge that there is some non-monotonic behavior in the 19 results of our simulations, e.g., in Figs. 3 and 6, preceding the onset of the first conductance plateau which is verified by the experimental data. However, apart from this, we do believe that we have qualitatively reproduced a significant part of the experimentally observed re- currence of the first conductance plateau with increasing channel width. This is an aim of our review, and such an observation highlights the importance of the Coulomb interaction between electrons after appreciably suppressing the electron kinetic energy contribution as the channel confinement becomes very weak. IV. CONCLUDING REMARKS The ballistic conductance for a quasi-1D channel (quantum wire) has exhibited an in- teresting behavior as functions of the electron density as well as confinement. We demon- strated that electron-electron interaction plays a crucial role in our calculations in the weak confinement regime. Extensive calculations were carried out in regards the effects due to confinement on the conductance and its associated dependence on the interplay between level anticrossing and crossing in quantum transport of two interacting-electron clusters. As shown in our numerical results, depending on the confinement parameter, the conductance manifests the signature of single-particle or interacting two-electron state behavior. This dependence can be observed in the deviation of the conductance from 2e2/h (single-particle) to 4e2/h (interacting crossing state) and back to 2e2/h (interacting anticrossing state) as a function of the width of the quantum wire. It is interesting to observe how many-body effects enter the calculation of the quantum ballistic conductance, where the center-of-mass velocity is not affected by the electron-electron interaction but the electron distribution is affected. We conclude that the experimental observations qualitatively agree well with our theo- retical calculations. Furthermore, such experimentally observed features for switching con- ductance plateau can be physically explained by the interchange of the ground between E(p)− and back to E(p)− , which is reflected as an upward jump and the degenerate E(p) 3 from 2e2/h to 4e2/h and followed by another step jump from 4e2/h back to 2e2/h with and E(p) 4 increasing channel width. 20 Acknowledgments DH would like to thank the Air Force Office of Scientific Research (AFOSR) for the support. 21 1 S. Kumar, K. J. Thomas, L. W. Smith, M. Pepper, G. L. Creeth, I. Farrer, D. Ritchie, G. Jones and J. Griffiths, Phys. Rev. B 90, 201304(R) (2014). 2 T. J. Thornton, M. Pepper, H. Ahmed, D. Andrews, and G. J. Davies, Phys. Rev. Lett. 56, 1198 (1986). 3 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 21, L209 (1988). 4 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). 5 L. Maslov and M. Stone, Phys. Rev. B 52, R5539 (1995). 6 J. Meyer and K. Matveev, J. Phys.: Condens. Matter 21, 023203 (2009). 7 H. J. Schulz, Phys. Rev. Lett. 71, 1864 (1993). 8 G. Piacente, I. V. Schweigert, J. J. Betouras, and F. M. Peeters, Phys. Rev. B 69, 045324 (2004). 9 L. W. Smith, W. K. Hew, K. J. Thomas, M. Pepper, I. Farrer, D. Anderson, G. A. C. Jones, and D. A. Ritchie, Phys. Rev. B 80, 041306(R) (2009). 10 D. Klironomos, J. S. Meyer, T. Hikihara, and K. A. Matveev, Phys. Rev. B 76, 075302 (2007). 11 G. Smith, M. Pepper, R. Newbury, H. Ahmed, D. G. Hasko, D. C. Peacock, J. E. F. Frost, D. A. Ritchie, G. A. C. Jones, and G. Hill, J. Phys. Condens. Matter 1, 6763 (1989). 12 P. J. Simpson, D. R. Mace, C. J. B. Ford, I. Zailer, M. Pepper, D. A. Ritchie, J. E. F. Frost, M. P. Grimshaw, and G. A. C. Jones, Appl. Phys. Lett. 63, 3191 (1993). 13 K. J. Thomas, J. T. Nicholls, M. Y. Simmons, W. R. Tribe, A. G. Davies, and M. Pepper, Phys. Rev. B 59, 12252 (1999). 14 I. M. Castleton, A. G. Davies, A. R. Hamilton, J. E. F. Frost, M. Y. Simmons, D. A. Ritchie, and M. Pepper, Physica B 249-251, 157 (1998). 15 D. H. Huang, G. Gumbs, Y. Abranyos, M. Pepper, and S. Kumar, AIP Advances 5, 117227 (2015). 16 J. Faist, P. Gu´eret, and H. Rothuizen, Phys. Rev. B 42, 3217 (1990). 17 G. Gumbs, A. Balassis, D. H. Huang, S. Ahmed and R. Brennan, J. Appl. Phys. 110, 073709 (2011). 22 18 L. W. Smith, K. J. Thomas, M. Pepper, D. A. Ritchie, I. Farrer, J. P. Griffiths, and G. A. C. Jones, J. Phys.: Conf. Ser. 376, 012018 (2012). 19 J. Koch, M. E. Raikh, and F. von Oppen, Phys. Rev. Lett. 96, 056803 (2006). 20 M. Leijnse, M. R. Wegewijs, and M. H. Hettler, Phys. Rev. Lett. 103, 156803 (2009). 21 M. Wagner, U. Merkt, and A. V. Chaplik, Phys. Rev. B. 45, 1951 (1991). 22 G. W. Bryant, Phys. Rev. Lett. 59, 1140 (1987). 23 S. K. Lyo and D. H. Huang, J. Phys.: Condens. Matt. 16, 3379 (2004). 24 G. Gumbs, D. H. Huang, M. D. Blumenthal, S. J. Wright, M. Pepper, and Y. Abranyos, Semicond. Sci. Technol. 24, 115001 (2009). 25 S. K. Lyo and D. H. Huang, Phys. Rev. B 73, 205336 (2006). 26 G. R. Aizin, G. Gumbs, and M. Pepper, Phys. Rev. B 58, 10589 (1998). 27 G. Gumbs, A. Balassis, D. H. Huang, S. Ahmed, and R. Brennan, J. Appl. Phys. 110, 073709 (2011). 23 FIG. 1: (Color online) Schematic illustration of a device used in our experiments (left-upper corner), including a pair of split gates and a top gate. Additionally, typical measured conductance features of the device are also shown as a function of split-gate voltage, Vsg for various fixed top-gate voltages, Vtg. FIG. 2: (Color online) Plots of cluster energy levels E(p) j as a function of geometry ratio R with four different values of linear electron density n1D. 24 0.00.51.01.52.03Ratio RE(p)30.00.51.01.52.0Ratio RE(p)(a)(b)2p) (meV)E-(p) E+(p) E3(p) E4(p)2 (p) (meV) E-() E+(p) E3(p) E4(p)(a)(b)1n1D(105 cm-1)=0.2Level Ej(01n1D(105 cm-1)=0.05 Level Ej(3V) E-(p) E+(p)(p)3 V) E-(p) E+(p)()00(c)(d)12n1D(105 cm-1)=0.4vel Ej(p) (meVE3(p) E4(p)12n1D(105 cm-1)=0.3 vel Ej(p) (meV E3(p) E4(p)0.00.51.01.52.00LeRatioR0.00.51.01.52.00LevRatioRRatio RRatio RFIG. 3: (Color online) Plots of conductance G as a function of geometry ratio R with four different values of linear electron density n1D for both noninteracting and interacting cases. FIG. 4: (Color online) Plots of cluster energy levels E(p) as a function of linear electron density n1D with four different values of geometry ratio R = Wx/Ly. Inset in 4(b) brings us a blow-out view for the anticrossing of energy levels E(p)− and E(p) + . j 25 30.00.51.01.52.0Ratio R)Noninteracting0.00.51.01.52.03)Ratio RNoninteracting(a)(b)2(105-1)005 ce G (2e2 / h) Noninteracting Interacting2ce G (2e2 / h)Noninteracting Interacting(a)(b)01n1D(105cm1)=0.05Conductanc01n1D(105 cm-1)=0.2Conductanc032e2 / h) Noninteracting Interacting3 2e2 / h) Noninteracting Interacting0(c)(d)12uctance G (212 uctance G (20.00.51.01.52.00Condun1D(105 cm-1)=0.4RatioR0.00.51.01.52.00n1D(105 cm-1)=0.3ConduRatioRRatio RRatio R0.00.10.20.30.41.5E(p)Density n1D (105 cm-1)E(p)90.00.10.20.30.4Density n1D (105 cm-1)(b)0.91.2E3 E4(p)) (meV)E- E+(p)6R = 0.1(p) (meV)(p)(b) 0.30.6R = 0.6Level Ej(p03 Level Ej( E-(p) E+(p) E3(p) E4(p)(a) 0.00.91.2R = 2.0V) E-(p) E+(p)E(p)0.91.2R = 1.2 eV)0(c)0.30.6vel Ej(p) (meVE3(p) E4(p)0.30.6 evel Ej(p) (me E-(p) E+(p) E3(p)0.00.10.20.30.40.00.3LevDensity n1D (105cm-1)0.00.10.20.30.40.00.3LeDensity n1D (105cm-1)3 E4(p)(d)y1D()y1D()FIG. 5: (Color online) Plots of conductance G as a function of linear electron density n1D with four different values of geometry ratio R for both noninteracting and interacting cases. FIG. 6: (Color online) Plots of conductance G as a function of linear electron density n1D with another four different values of geometry ratio R for both noninteracting and interacting cases. 26 0.00.10.20.30.43)Density n1D (105 cm-1)30.00.10.20.30.4Density n1D (105 cm-1))(a)(b)2R=04e G (2e2 / h)2R=01 e G (2e2 / h)(a)(b)01R= 0.4Conductanc Noninteracting Interacting01R= 0.1Conductanc Noninteracting Interacting232e2 / h) 23 2e2 / h)00(c)(d)12R = 1.0uctance G (212R = 0.6 uctance G (20.00.10.20.30.40CondDensity n1D (105cm-1)Noninteracting Interacting0.00.10.20.30.40CondDensity n1D (105cm-1) Noninteracting Interactingy1D()y1D()0.00.10.20.30.43R=16)Density n1D (105 cm-1)30.00.10.20.30.4R=12Density n1D (105 cm-1))(a)(b)2R= 1.6ce G (2e2 / h)2R= 1.2 ce G (2e2 / h)(a)(b)01Conductanc Noninteracting Interacting01Conductanc Noninteracting Interacting023R = 2.02e2 / h) 23R = 1.8 2e2 / h)0(c)(d)12uctance G (212 uctance G (20.00.10.20.30.40CondDensity n1D (105cm-1) Noninteracting Interacting0.00.10.20.30.40CondDensity n1D (105cm-1)Noninteracting Interactingy1D()y1D()FIG. 7: (Color online) Contour plots of conductance G as functions of both linear electron density n1D and geometry ratio R for either noninteracting (left panel) or interacting (right panel) case. As labeled by the color bars in this figure, the color scales (from blue up to orange) are [0.2, 2.2] (left) and [0.3, 2.5] (right), respectively. FIG. 8: (Color online) Contour plots of G as functions of both linear electron density n1D and geometry ratio R for either noninteracting [(a)] or interacting [(b)] case at Vb = 0.05 V, as well as the plot of G as a function of R [(c)] at n1D = 0.3 × 105 cm−2 for three different values of bias voltage Vb. 27 162.0 206020822.1042.1262.1482.1702.1922.2142.2362.2582.2802.3022.3242.3462.3682.3902.4122.4342.4562.4782.500162.0 180018201.8401.8601.8801.9001.9201.9401.9601.9802.0002.0202.0402.0602.0802.1002.1202.1402.1602.1802.2002.22.50.81.21.6 Ratio R10701.0921.1141.1361.1581.1801.2021.2241.2461.2681.2901.3121.3341.3561.3781.4001.4221.4441.4661.4881.5101.5321.5541.5761.5981.6201.6421.6641.6861.7081.7301.7521.7741.7961.8181.8401.8621.8841.9061.9281.9501.9721.9942.0162.0382.0602.0820.81.21.6 Ratio R090000.92000.94000.96000.98001.0001.0201.0401.0601.0801.1001.1201.1401.1601.1801.2001.2201.2401.2601.2801.3001.3201.3401.3601.3801.4001.4201.4401.4601.4801.5001.5201.5401.5601.5801.6001.6201.6401.6601.6801.7001.7201.7401.7601.7801.8001.8200.10.20.30.40.40.8InteractingG (2e2 / h)0.30000.32200.34400.36600.38800.41000.43200.45400.47600.49800.52000.54200.56400.58600.60800.63000.65200.67400.69600.71800.74000.76200.78400.80600.82800.85000.87200.89400.91600.93800.96000.98201.0041.0261.0481.0701.0920.10.20.30.40.40.8NoninteractingG (2e2 / h)R0.20000.22000.24000.26000.28000.30000.32000.34000.36000.38000.40000.42000.44000.46000.48000.50000.52000.54000.56000.58000.60000.62000.64000.66000.68000.70000.72000.74000.76000.78000.80000.82000.84000.86000.88000.90000.20.3Density n1D (105 cm-1)Density n1D (105 cm-1)162.0 184018581.8761.8941.9121.9301.9481.9661.9842.0022.0202.0382.0562.0742.0922.1102.1282.1462.1642.1822.200162.0 198020002.0202.0402.0602.0802.1002.1202.1402.1602.1802.2002.2202.2402.2602.2802.3002.3202.3402.3602.3802.400(a)2.42.2(b)081.21.6 106610841.1021.1201.1381.1561.1741.1921.2101.2281.2461.2641.2821.3001.3181.3361.3541.3721.3901.4081.4261.4441.4621.4801.4981.5161.5341.5521.5701.5881.6061.6241.6421.6601.6781.6961.7141.7321.7501.7681.7861.8041.8221.8401.858Ratio R081.21.6 112011401.1601.1801.2001.2201.2401.2601.2801.3001.3201.3401.3601.3801.4001.4201.4401.4601.4801.5001.5201.5401.5601.5801.6001.6201.6401.6601.6801.7001.7201.7401.7601.7801.8001.8201.8401.8601.8801.9001.9201.9401.9601.9802.000Ratio R010203040.40.8InteractingG (2e2 / h)0.40000.41800.43600.45400.47200.49000.50800.52600.54400.56200.58000.59800.61600.63400.65200.67000.68800.70600.72400.74200.76000.77800.79600.81400.83200.85000.86800.88600.90400.92200.94000.95800.97600.99401.0121.0301.0481.0661.0841.102R010203040.40.8NoninteractingG (2e2 / h)0.40000.42000.44000.46000.48000.50000.52000.54000.56000.58000.60000.62000.64000.66000.68000.70000.72000.74000.76000.78000.80000.82000.84000.86000.88000.90000.92000.94000.96000.98001.0001.0201.0401.0601.0801.1001.1201.140R0.40.42.5 0.05010 h)Bias: Vb(V)0010.10.20.30.4Density n1D (105 cm-1)0.10.20.30.4Density n1D (105 cm-1)(c)1.52.00.10 ce G (2e2 / 0.01()051.0n1D(105 cm-1)=0.3Conductan0.00.51.01.52.00.5Ratio RFIG. 9: (Color online) Plot of measured differential conductance in (a) as functions of split gate voltage Vsg for various values of top gate voltage Vtg, and in (b) the transconductance (dG/dVsg) plot of the data shown in (a). The confinement in (a) is controlled by making the top gate negative so that left (right) of the plot is strong (weak) confinement, where a direct jump to 4e2/h (indicated by arrows) occurs when the confinement is weakened using a top gated, split-gate device. In addition, the first trace in (a) on the left is taken at Vtg = −7.21 V and successive traces were plotted in steps of 90 mV until Vtg = −9.19 V. 28 Vtg= -7.21 VVtg= -9.19 V(a)(b)
1208.0822	2	1208	2012-10-09T09:45:19	Scattering by linear defects in graphene: a continuum approach	[ "cond-mat.mes-hall" ]	We study the low-energy electronic transport across periodic extended defects in graphene. In the continuum low-energy limit, such defects act as infinitesimally thin stripes separating two regions where Dirac Hamiltonian governs the low-energy phenomena. The behavior of these systems is defined by the boundary condition imposed by the defect on the massless Dirac fermions. We demonstrate how this low-energy boundary condition can be computed from the tight-binding model of the defect line. For simplicity we consider defect lines oriented along the zigzag direction, which requires the consideration of only one copy of Dirac equation. Three defect lines of this kind are studied and shown to be mappable between them: the pentagon-only, the zz(558) and the zz(5757) defect lines. In addition, in this same limit, we calculate the conductance across such defect lines with size L, and find it to be proportional to k_FL at low temperatures.	cond-mat.mes-hall	cond-mat	Scattering by linear defects in graphene: a continuum approach J. N. B. Rodrigues1,3, N. M. R. Peres2 and J. M. B. Lopes dos Santos1 1CFP and Departamento de Física e Astronomia, Faculdade de Ciências Universidade do Porto, P-4169-007 Porto, Portugal 2Centro de Física e Departamento de Física, Universidade do Minho, P-4710-057, Braga, Portugal and 3Graphene Research Centre, Faculty of Science, National University of Singapore, 6 Science Drive 2, Singapore 117546 (Dated: November 10, 2018) We study the low-energy electronic transport across periodic extended defects in graphene. In the continuum low-energy limit, such defects act as infinitesimally thin stripes separating two regions where Dirac Hamiltonian governs the low-energy phenomena. The behavior of these systems is defined by the boundary condition imposed by the defect on the massless Dirac fermions. We demonstrate how this low-energy boundary condition can be computed from the tight-binding model of the defect line. For simplicity we consider defect lines oriented along the zigzag direction, which requires the consideration of only one copy of Dirac equation. Three defect lines of this kind are studied and shown to be mappable between them: the pentagon-only, the zz(558) and the zz(5757) defect lines. In addition, in this same limit, we calculate the conductance across such defect lines with size L, and find it to be proportional to kF L at low temperatures. PACS numbers: 81.05.ue, 72.80.Vp, 78.67.Wj I. INTRODUCTION Graphene growth by chemical vapor deposition (CVD) on metal surfaces1 -- 4 is a very promising scalable method for producing graphene sheets. However, the present sta- tus of the method, typically results in the synthesis of polycrystalline graphene abundant in topological defects, grain boundaries (GBs) being, by far, the most common ones.5 -- 7 Due to graphene's hexagonal structure, pairs of pen- tagons and heptagons, named Stone-Wales (SW) defects, as well as octagons, are expected to form at graphene GBs.8 Recent atomic resolution TEM studies5,6,9,10 al- lowed the observation of GBs in CVD-grown graphene. These experimental studies have shown that the GBs are generally not perfect straight lines, and that the 5-7 defects along the boundaries are not periodic. Further- more, as shown by recent TEM studies,5,6 these extended pentagon-heptagon defect-lines intercept each other at random angles, forming irregular polygons with edges showing a stochastic distribution of lengths. This renders theoretical studies of such defects difficult, in particular when using microscopic tight-binding models.11 Theoretical studies have argued that GBs strongly in- fluence the properties of graphene, namely its chemical,12 mechanical13,14 and electronic ones. Electronic mobili- ties of films produced through CVD are lower than those reported on exfoliated graphene, because15,16 electronic transport17,18 is hindered by grains and GBs.11,19 Recently, the observation of a linear extended defect acting as a one-dimensional conducting charged wire10 stimulated some theoretical studies concentrated on the scattering and transport properties of such wire.20 -- 22 Most of these studies have so far been focused on tight- binding models. The use of the continuum approximation on the scope of graphene have led to a better understanding of many Figure 1: (Color online) Scheme of a graphene sheet with a pentagon-only defect line along the zigzag direction. The primitive vectors are u1 = a(1, 0) and u2 = a(−1/2, √ 3/2). important phenomena occurring in graphene. Moreover, we believe that the use of this approach in the study of the electronic scattering across extended defects in graphene, may further extend our insight onto the physics underlying these events in graphene. As is widely known, a continuum approximation of graphene's first neighbor TB Hamiltonian for states in the vicinity of the Dirac points, describes graphene's low energy charge carriers as massless Dirac fermions. These are governed by two copies of Dirac Hamiltonian, each one of them valid around each of the Dirac points.23 In this continuum limit, the finite width defect line turns out to essentially act as a one-dimensional (infinitesi- mally thin) line, separating two distinct regions governed by Dirac Hamiltonian. The defect line is modeled by a boundary condition on the Dirac spinors, imposing a 2 1 0 2 t c O 9 ] l l a h - s e m . t a m - d n o c [ 2 v 2 2 8 0 . 8 0 2 1 : v i X r a ,AB,tu1u2txydiscontinuity across the defect. This boundary condition determines the scattering properties of the defect. For simplicity, in this text we only consider extended line defects oriented parallel to the zigzag direction. In these cases, we can ignore intervalley scattering, and thus consider only one copy of the Dirac equation. While some general properties of the boundary condition and trans- mittance can be obtained exclusively from the continuum description, the specific boundary condition must be de- rived from the TB model of the defect. Nevertheless, we feel that this approach adds considerably to the un- derstanding of the low energy limit obtained from a TB description;22 in particular, it explains, as we will show later, why different defects can show exactly the same low energy transmittance. To illustrate the main physical issues and the method of approach, we start with a simplified version of a de- fect line, composed of a double line of pentagons oriented along the zigzag direction of the graphene lattice (see Fig. 1), which we dub as pentagon-only defect line. Compared to the more realistic linear defects that we treat later in the paper, the zz(588)10 and the zz(5757) defects, it has the added simplicity of full translation symmetry along the defect direction, whereas the latter display a dou- bling of the unit cell along that same direction. The low- energy boundary conditions associated with these defect lines are also computed and compared with that of the pentagon-only defect line. Suitable choices of the micro- scopic parameters lead exactly to the same transmittance as a function of angle of incidence in all three cases. Fi- nally, we also compute the conductance across a defect of length L and find it to be proportional to kF L at low temperatures, for all three defects considered. II. ELECTRON TRANSPORT ACROSS A PENTAGON-ONLY GRAIN BOUNDARY A. The continuum description A graphene plane with an extended line defect can be viewed in the low energy limit as two half-planes of mass- less Dirac Fermions, which cannot be joined smoothly, because of the defect, a line of discontinuity. To ap- proach this problem, consider a finite strip of width W in the y direction, where there is a general local poten- tial, V (y) = Vs + Vxσx + Vyσy + Vzσz, for y < W/2, such that W × (Vs, V) → (vs, v) as W → 0. Integrat- ing the Dirac equation in the y coordinate, the resultant general boundary condition for the Dirac spinor has the form (see Appendix A) Ψ(x, 0+) = MΨ(x, 0−), (1) where the 2 × 2 matrix M reads M = e−iσy(vs+v·σ)/vF , (2) and σ = (σx, σy, σz). This boundary condition has to satisfy the conservation of current in the y direction, 2 i.e.,Ψ†(x, 0+)σyΨ(x, 0+) = Ψ†(x, 0−)σyΨ(x, 0−) for any spinor, which implies that M†σyM = σy; the form given in Eq. 2 satisfies this condition. An important feature, borne out by the derivation of Appendix A, is energy independence of the boundary condition. When we inte- grate the Dirac equation across the strip, and take the limit W → 0 , the term containing the energy of the state, which, unlike the potential, is fixed, drops out. An incoming wave from y = −∞, will be partly re- flected and partly transmitted at the defect. As a conse- quence, the real-space wave-function on each side of the defect line is given by (cid:21) (cid:21) (cid:21) q (cid:20) se−iθν (cid:20) (cid:20) se−iθν se−iθν 1 1 q q 1 Ψν qs(r) = + Ψν qs(r) = 1√ 2 ρ√ 2 τ√ 2 ei(qxx+qyy) ei(qxx−qyy), y < 0 (3a) ei(qxx+qyy), y > 0 (3b) where ν = ±1 specifies the Dirac cone, θq is the complex q, the complex phase phase of νqx + iqy, and θ of νqx − iqy (see Fig. 3). The sign of the energy is noted by s. Imposing the general boundary condition gives immediately the following general expression for the transmission probability q = −θν ν (cid:12)(cid:12)(cid:12)(cid:12)eiν2θM11 + νeiνθ(cid:0)M12 − M21 4 sin2 θ (cid:1) − M22 (cid:12)(cid:12)(cid:12)(cid:12)2 , T ν(E, θ) = (4) where we used the property detM = 1, which follows from the condition of flux conservation, M†σyM = σy. A noteworthy feature, that follows naturally from this formulation, is the energy independence of the transmis- sion across the defect. To determine the actual values do the matrix elements of M for a specific defect in a graphene lattice we must consider its microscopic description. B. The low energy limit of tight binding The first-neighbor TB Hamiltonian of graphene with a pentagon-only defect line (see Fig. 1), can be written as the sum of three terms, H = H U + H D + H L, where H U ( H L) stands for the Hamiltonian above (below) the de- fect line, while the remaining term, H D, describes the de- fect line itself. In second quantization the explicit forms of H U and H L read H U (L) = −t (cid:26)(cid:104)b†(m, n) + b†(m, n − 1) (cid:27) (cid:88) (cid:88) m n (cid:105) + b†(m − 1, n − 1) a(m, n) + h.c. , (5) where for H U (H L) n ≥ 1 (n ≤ −1). The term describing the defect, H D, is H D = −(cid:88) (cid:26)(cid:104) m + tb†(m, 0) ξt d†(m + 1) + ta†(m, 0) (cid:27) (cid:105) d(m) + h.c. , (6) 3 where t is the usual hopping amplitude of pristine graphene and ξt is the hopping amplitude between the Dm atoms of the defect line, as represented in Fig. 1. If we Fourier transform the Hamiltonian along the zigzag direction (x-direction), we reduce it to an effec- tive one-dimensional chain with two atoms per unit cell and a localized defect at its center (see Fig. 2). Figure 2: (Color online) Scheme of the one-dimensional chain obtained by Fourier transformation on the x-direction of the TB Hamiltonian of a graphene layer with a pentagon-only defect line along the zigzag direction. The complex hopping amplitude t(cid:48) has the value t(cid:48) = t(1 + eikxa). The Hamiltonian of the effective chain is defined as H(kx) = H U (kx) + H D(kx) + H L(kx) , (7) where the three terms on the right hand side of Eq. (7) read (cid:26)(cid:104) H U/L(kx) = −(cid:88) (cid:27) H D(kx) = −2ξt cos(kxa) d†(kx) d(kx) −(cid:104) t(cid:48)b†(kx, n − 1) + tb†(kx, n) a(kx, n) + h.c. (cid:105) n , + tb†(kx, 0) d(kx) + h.c. . (cid:105) (8a) ta†(kx, 0) d(kx) (8b) The one-dimensional chain has alternating hopping amplitudes between the atoms, t and t(cid:48) = t(1 + eikxa). Moreover, the electron at a Dm atom acquires an on-site energy term, (cid:101)(kx) = −2ξt cos(kxa), which depends on the value of the longitudinal momentum kx. At the bulk of the one-dimensional chain (n < −1 and n > 0), the TB equations for unit cell n involve ampli- tudes at three different positions, n − 1, n and n + 1, A(kx, n) = −tB(kx, n) − (t(cid:48))∗B(kx, n − 1), B(kx, n) = −tA(kx, n) − t(cid:48)A(kx, n + 1). (9a) (9b) Nevertheless, replacing n → n + 1 in Eq. (9a), we can solve these equations for A(kx, n+1) and B(kx, n+1) and recast them as a recurrence relation relating amplitudes at unit cell n + 1 with those at unit cell n, L(n + 1) = T(, kxa).L(n), (10) where L(n) = [A(kx, n), B(kx, n)]T . The passage matrix, T(, φ), is given by 2 cos(cid:0) φ T(, φ) = − e−i φ 2 2 (cid:1)(cid:20) 1 t 4 cos2(cid:0) φ − t 2 (cid:1) − 2 t2 (cid:21) . (11) The eigenvectors of matrix T(, φ) with eigenvalues with λ2 = 1, correspond to Bloch solutions propa- gating along the one-dimensional chain (band states). The eigenvectors with eigenvalues λ2 (cid:54)= 1 correspond to evanescent states which decrease when n → +∞ (n → −∞) when λ2 < 1 (λ2 > 1). Note that the previous formulation of the TB problem, is entirely equivalent to the usual one, where translational symmetry along the lattice vectors directions, allows the use of Bloch theorem to compute the eigenvectors and eigenvalues of the TB Hamiltonian for pristine graphene. A similar construction to that of Eq. (10) can be carried out in the rows containing the defect. The TB equations for the defect and its neighbors in the one- dimensional chain are easily read from Fig. 2 +A(kx, 1) = −(t(cid:48))∗B(kx, 0) − tB(kx, 1), (12a) B(kx, 0) = −tD(kx) − t(cid:48)A(kx, 1), (12b) D(kx) = −t(cid:0)A(kx, 0) + B(kx, 0)(cid:1) − 2ξt cos(kxa)D(kx), (12c) A(kx, 0) = −(t(cid:48))∗B(kx,−1) − tD(kx), (12d) (12e) where ± = ± e∆V /2, to account for a possible poten- tial difference between the two grains separated by the −B(kx,−1) = −t(cid:48)A(kx, 0) − tA(kx,−1), 4 where ρ and τ are, respectively, the reflection and trans- mission scattering amplitudes, and Ψ> and Ψ< stand for the right and left moving eigenstates of matrix T(, kxa), the passage matrix for pristine graphene, with corre- sponding eigenvalues noted by λ> and λ<. Imposing the boundary condition, Eq. (14), it is straightforward to obtain the coefficients ρ and τ for a given energy and a given longitudinal momentum. In particular, τ reads , det(cid:101)M(cid:101)M22 (cid:12)(cid:12)(cid:12)2 (cid:12)(cid:12)(cid:12)(cid:101)M22 (18) τ = where (cid:101)M = U−1MU is the boundary condition matrix [see Eq. (14)] in the eigenbasis of the passage matrix of pristine graphene T(, kxa). The transmission probabil- ity is given by T = τ2 = 1/ , since flux conserva- tion again requires that det M = 1. But our main concern is the low energy limit. In the following we assume ∆V = 0. Let us consider in parallel the equations that propagate the state in the bulk and in the defect: (cid:20) A(kx, n + 1) (cid:20) A(kx, 1) B(kx, n + 1) (cid:21) (cid:21) B(kx, 1) (cid:20) A(kx, n) (cid:21) (cid:20) A(kx,−1) B(kx, n) B(kx,−1) = T(, φ) = M(, φ) (cid:21) bulk; (19a) defect; (19b) As is well known, near a Dirac point Kν, the slowly vary- ing Dirac spinor Ψν(r) is defined by (ignoring irrelevant normalization constants) Ψν(mu1 + nu2) = e−iKν·(mu1+nu2) ,(20) (cid:21) (cid:20) A(m, n) (cid:21) B(m, n) (cid:20) A(kx, n) and for a plane wave along u1 ei(k−Kν )·mu1 Ψν(mu1 + nu2) = e−iKν·nu2 B(kx, n) ≡ Ψν(qx, nu2)eiq·mu1 (21) where q = k − Kν. This allows us to recast Eqs. (19a) and (19b) in terms of the Dirac fields, Ψν (qx, (n + 1)u2) = e−iKν·u2T(, φ)Ψν(qx, nu2), (22a) Ψν(qx, u2) = e−iKν·2u2M(, φ)Ψν(qx,−u2), (22b) where Kν · u2 = −ν2π/3. If we take the Fourier transform with respect to the spatial variable along u2 in Eq. (22a), q = eiν2π/3e−iq·u2T(, φ)Ψν Ψν q (23) In Appendix B we show that the matrix multiplying Ψν q on the right hand side tends to the identity matrix when q, → 0; if we expand the right hand side to linear Figure 3: (Color online) (a) Graphene FBZ with the incident vectors used in the TB and CA formalism: q = k − Kν. (b) Scheme of the electron scattering through the barrier (in the low-energy limit). pentagon-only defect line. This set of TB equations can be used to construct a matrix equation relating the TB amplitudes in opposite sides of the defect. The technique is to solve each equation for the amplitude of the right- most site in Fig. 2 and then cast them as 2 × 2 matrix equations. For instance, Eq. (12a) is equivalent to (cid:21) (cid:20) B(kx, 1) A(kx, 1) = (cid:20) − + t 1 (cid:21)(cid:20) A(kx, 1) (cid:21) B(kx, 0) −(t(cid:48))∗ t 0 With this procedure, one can derive L(1) = ML(−1), . (13) (14) where M ≡ RM1(+, φ)M2(, φ)M3(, φ)M1(, φ)M2(−, φ)RT (15) is a 2×2 matrix; R is the σx Pauli matrix, used to switch rows, φ = kxa, and M1, M2 and M3 are (1 + e−iφ) (cid:20) (cid:21) , (16a) M1(, φ) = − −1 M2(, φ) = − 1 t 0 (cid:20) (cid:20) +2tξ cos(φ) 1 + eiφ M3(, φ) = − t −1 (cid:21) 1 0 t 1 −(1 + eiφ) 0 , (16b) . (16c) (cid:21) Note that the 2 × 2 boundary condition matrix, M [see Eq. (14)], depends on the energy, , on the longitudi- nal momentum, kx, and on the potential difference ∆V , through + and −. We now have all the ingredients needed to compute the scattering coefficients of an electron wave by the pentagon-only defect line. Given an incoming wave from n = −∞, the presence of the defect (line) at n = 0, pro- duces a reflected and a transmitted component. In such a case, the wave-functions on each side of the defect (line) are given by L(n < 0) = λn L(n > 0) = τ λn >Ψ> + ρλn >Ψ>, <Ψ<, (17a) (17b) order in and q, we obtain, as we should, the Dirac- Weyl equation (see Appendix B). However, at the defect, we find e−iν2π/3M(, φ) → when φ, →0, (24) which gives rise to the following equation Ψν(qx, u2) = Ψν(qx,−u2). (25) (cid:19) (cid:18) 0 1−1 ξ (cid:19) (cid:18) 0 1−1 ξ After Fourier transforming the previous equation in qx, and as the continuum approximation yields a → 0 in u2, near the Dirac point, we end up concluding that the defect introduces a discontinuity in the Dirac fields of the form we derived from general considerations, Ψν(x, 0+) = MΨν(x, 0−), with (cid:19) (cid:18) 0 1−1 ξ M = (26) The transmission probability, given by the general ex- pression of Eq. (4), becomes here T ν(θ) = sin2 θ 1 − νξ cos θ + ξ2/4 . (27) In Fig. 4, we plot the transmission probability T , in terms of the angle of incidence on the defect line, for both the TB and the continuum approximation (CA), with ξ = 1.2. The various plots refer to different energies, but always to the same Dirac point (ν = 1). As expected, the lower the energy, the better the agreement between the TB and the CA results. For the other Dirac point, the results are mirror-symmetric relatively to the normal incidence angle θ = π/2. A special case is of some interest, namely, for low en- ergies, ∆V = 0, and ξ = 2, the transmission probabil- ity becomes T ν(θ) = (1 + ν cos θ)/2, in which case the pentagon-only defect line acts as a valley filter, for angles of incidence close to θ = 0, π. This same feature has been found in another type of defect, the zz(558), which we consider in the next section, by Gunlycke and White;21 this is no accident; we will show that these two defects share the same low energy limit. It is worth noting, that since the passage matrix in the continuum limit is obtained with φ = Kν · u1 and = 0, it is easily got in a back of envelope calculation, by writing and solving the TB equations at zero energy. This procedure is carried out in Appendix C. It is expected that defect lines and grain boundaries in graphene are reactive,12 being a likely location for ad- sorption of atoms or molecules. Such adsorbates, are expected to locally perturb the properties of the defect lines. For simplicity, we may assume that the adsorbate only modifies the local energy at the atom it adsorbs to. We can account for such a phenomenon in the pentagon- only defect line, including in its TB model, an on-site energy, 0 at the D atoms of the defect line (see Fig. 1 5 Figure 4: (Color online) Plot of the transmission probability, T , in terms of the angle of incidence on the pentagon-only defect line, θ+ q (for the low-energy limit, around Kν, with ν = +1). We have used the value ξ = 1.2 to obtain these curves. In each of the panels, we compare the TB result for > 0 (full violet curves) and for < 0 (dashed violet curves), with that obtained from the CA (in green), with ∆V = 0. Panel (a), (b), (c), (d), (e) and (f), stand, respectively, for energies /t = 0.01, /t = 0.04, /t = 0.08, /t = 0.16, /t = 0.32 and /t = 0.64. or Fig. 2). Such a modification of the TB model, will necessarily modify the TB boundary condition matrix, M [see Eq. (14)], as well as the continuum approxima- tion one, M [see Eq. (26)]. The TB boundary condition matrix, [see Eqs. (16)], will have its + 2ξt cos(kxa) term modified. This will now include the on-site energy, 0, as (cid:48) = + 2ξt cos(kxa) + 0. In the CA limit, the boundary condition matrix, M, will have (ξt − 0)/t in the M22 entry of the matrix instead of ξ. Thus, the adsorption of molecules at the defect line, in very low energies, will be equivalent to rescaling the hopping between the D atoms at the defect line. III. THE zz(558) AND THE zz(5757) DEFECT LINES We now extend this treatment to the case of a zz(558) defect line10,21,22 (see Fig. 5), and of a zz(5757) defect line (see Fig. 6). We can proceed in close analogy with the case of a pentagon-only defect line treated in the previous section. But these more realistic defects exhibit a feature that is not present in the previous case, namely, the doubling of the unit cell in the direction parallel to the defect. The corresponding folded First Brillouin Zone (FBZ) has twice as many states at the same Bloch wave vector, as in the original FBZ of graphene; the real space unit cell has new Dirac points, now located at K± = ±π/3(1,−√ two A (A1, A2) and two B (B1, B2) sites. Around the 3), 6 Defining (cid:101)L(n) ≡ [A+(kx, n), B+(kx, n), A−(kx, n), B−(kx, n)]T , we have, (cid:101)L(n + 1) = Td(cid:101)L(n), where the matrix Td, writ- (30) ten in blocks of 2 × 2 matrices, is (cid:20) T+(, φ) 0 (cid:21) Td(, φ) = 0 T−(, φ) . (31) The + and − amplitudes propagate independently; T+ and T−, the passage matrices associated with the high and the low-energy TB modes, are (cid:1)(cid:20) 1 (cid:1) − 2 t 4 cos2(cid:0) φ 2 cos(cid:0) φ T+(, φ) = − e−i φ (cid:21) (cid:1)(cid:20) 1 t 4 sin2(cid:0) φ (cid:1) − 2 2i sin(cid:0) φ e−i φ T−(, φ) = − − . (32b) ,(32a) (cid:21) t t t2 2 t2 2 2 2 2 2 The above computations are due to appear in a compan- ion paper26 devoted to the study of these same systems under the TB approach. In parallel with what we have done for the pentagon- only defect line [see Eqs. (12)- (16)], using the TB equa- tions at the zz(558) or at the zz(5757) defect lines, it is possible to write an expression relating amplitudes at the two sides of these defects, L(1) = ML(−1). The matrix M is now a 4 × 4 matrix relating the four amplitudes at each side of the defect line, and admixing, in general, high and low-energy modes of different sides of the de- fect. The high energy sector passage matrix in the bulk near = 0 and φ = Kν · u1 = νπ/3 is (cid:16) (cid:17) (cid:20) 1√ (cid:21) T+ 0, ν π 3 = −e−iπ/6 √ 0 3 3 0 ; (33) The corresponding eigenstates are evanescent, one grow- ing exponentially as e(n log 3)/2, localized on the B sub- lattice, and the other decreasing as e−(n log 3)/2, localized in the A sub-lattice. This same result was obtained by Ostaay et al. in the scope of the total reconstruction of the zigzag edge by Stone-Wales defects.25 Given this, we conclude that a low energy state, must have the following form in each one of the sides of the defect (cid:101)Φ(kx, n) ≈ (cid:101)Φ(kx, n) ≈ 0   A+(kx, n) B+(kx, n) A−(kx, n) B−(kx, n) A−(kx, n) B−(kx, n) 0   n < 0; (34a) n > 0. (34b) This form fixes the B+(kx,−1) amplitude, in terms of the low energy amplitudes A−(kx,−1) and B−(kx,−1), Figure 5: (Color online) Scheme of a zz(558) defect line. Figure 6: (Color online) Scheme of a zz(5757) defect line. there will be, in addition to two low-energy Dirac cones, two high energy bands.24 At low energies, ≈ 0, the extra states show up as evanescent solutions.25,26 In pristine graphene we know the form of the high and low energy modes since they are Bloch states of different wave vectors in the unfolded Brillouin zone. We can use this to define a change of basis that decouples, in the bulk, these two energy sectors (φ = kxa):   A+ B+ A− B− B1 A2 B2  A1  = Λ(φ)  1 0 e−iφ 0 1 1 0 −e−iφ 0 1 0 0 1√ 2  . 0 e−iφ 0 e−iφ (28) (29) with Λ(φ) := yxyxsince M22B+(kx,−1) + M23A−(kx,−1) + M24B−(kx,−1) = 0, (35) 7 and leads to an effective boundary condition relation for the low energy amplitudes only. The latter reads (cid:20) A−(kx, 1) B−(kx, 1) (cid:21) (cid:20) A−(kx,−1) B−(kx,−1) (cid:21) = Meff , (36) where the effective boundary condition matrix is ob- tained from matrix M (cid:20) M33 − M32M23/M22 M34 − M32M24/M22 M43 − M42M23/M22 M44 − M42M24/M22 Meff = (cid:21) , (37) The low energy sector, with the matrix T−(, φ), can be analyzed exactly as was done in Appendix B for the pen- tagon only boundary. We define the Dirac fields as be- fore, Ψν(qx, nu2) = e−iKν·nu2 , (38) (cid:20) A−(kx, n) B−(kx, n) (cid:21) so that, in the bulk Ψν(qx, (n + 1)u2) = e−iKν·u2T−(, φ)Ψν(qx, nu2). (39) With a procedure entirely similar to the one detailed in Appendix B, one finds, after Fourier transforming in n, that Ψν Ψν(qx, u2) = e−iKν·2u2Meff(cid:16) q satisfies the Dirac equation. At the defect, Ψν(qx,−u2). (cid:17) 0, ν (40) (41) (cid:34) 0 1 −1 2 ξ2 ξ2 1 (cid:35) , π 3 (cid:35) The calculation of Meff yields Meff(cid:16) 0, ν π 3 = eiν2π/3 (cid:17) (cid:34) and so the boundary condition for the Dirac fields is Ψν(x, 0+) = 0 1 −1 2 ξ2 ξ2 1 Ψν(x, 0−). (42) It is remarkable that this has exactly the same form as found in the pentagon -- only boundary (c.f. Eq. [26]); the low energy transmission probabilities of these two line 1 = ξ. In Fig. 7 we defects are the same provided 2ξ2/ξ2 compare the transmission probabilities calculated with a full TB calculation for different values of ξ1 and ξ2 but the 1,26 and the corresponding low energy same value of ξ2/ξ2 approximation. Figure 7: (Color online) The transmission probabilities for a zz(558) defect, at /t = .03, with ξ1 = 1, ξ2 = 0.5 (dashed red), ξ1 = 1.5, ξ2 = 1.125 (dashed-dot, blue), obtained in a full TB calculation, and the corresponding low energy approx- imation (continuous black), given by Eq. (27) with ξ = 1.0. The treatment of the zz(5757) line defect presents no further novelty. Using the quick derivation method out- lined in Appendix C, we arrive at the following passage matrix for the Dirac fields −1 b + ξ2 Meff 5757 = 2ξb(ξ2 (43) (cid:21) , (cid:20) a b−b c a/4(cid:1), b = −ξa(ξ2 a/2) (cid:0)ξ2 b − ξ2 a)/ξ2 b + ξ4 a) and c = where a = 2ξ2 c c (see Fig 6 for the notation of the a + ξ2 b ξ2 2(ξ4 hopping amplitudes). The form of the passage matrix (and the transmission probability) is not identical to the previous cases, unless ξb = −ξa/2. b − ξ2 Despite the similarities, for general values of the hop- ping parameters, the transmittances originating from each of the defect lines can be considerably different. Such a case can be seen in Fig. 8, where we compare the low-energy transmission probabilities associated with each one of the previously discussed defect lines, for a special situation with all hopping amplitudes equal to t, the bulk nearest neighbor amplitude. IV. CONDUCTANCE In this section we address the calculation of the linear conductance, across a line defect, and show that the en- ergy independence of the transmission probability at low energies found in the previous three cases gives rise to a conductance linear in kF . We make the usual assumption that the electron reser- voirs at each side of the defect line are in equilibrium and are thus described by the single particle Fermi-Dirac distribution. Then the expression for the total net cur- rent across the defect line, associated with electrons liv- 0.00.20.40.60.81.00.00.20.40.60.81.0ΘΠT8 where C(cid:48) = Lgvgse2/(4π22vF ). The transmission prob- ability, T(E, ∆V = 0) = T(0, 0), does not depend on E, as was seen above; it does depend on the values of the hopping amplitudes in the vicinity of the defect, through the passage matrix. [see Fig. 9 , for the case of the pentagon only defect]. We obtain, (cid:20) (cid:18) µ (cid:19)(cid:21) e2 T(0, 0) kBT × h kBT 4π22vF G(T ) = Lgvgs , (47) ´ where the function h(x) is a Fermi integral, h(x) := +∞ −∞ dy y + x exp(y)/ [exp(y) + 1]2, with limits h(0) = 2 ln(2), and h(x) → x for x (cid:29) 1. The conductance is linear in temperature for µ (cid:28) kBT ; in the opposite limit, µ (cid:29) kBT , it is practically temperature indepen- dent, G(T = 0) = Lgvgs T(0, 0)µ e2 4π22vF e2 4π2 = gvgs T(0, 0)kF L. (48) Figure 9: (Color online) Dependence of the total angular in- tegrated transmission probability associated with one Dirac point, T(E, ∆V ) = T+(E, ∆V ) = T−(E, ∆V ), when ∆V = 0, in terms of the hopping, ξt, between the atoms D at the de- fect line (see Fig. 1). The vertical dashed line, indicates the value of ξ used to obtain the curves of Fig. 4: ξ = 1.2. V. CONCLUSION In this text we have focused on the study of the low- energy continuum limit behavior of the electronic trans- port across periodic defect lines oriented along the zigzag direction of graphene. We have argued that in this limit, such extended defects essentially act as one-dimensional infinitesimally thin lines, that separate two regions gov- erned by the Dirac Hamiltonian. In the low-energy con- tinuum limit the defective line imposes a boundary con- dition on the Dirac spinors at each side of the defect. It is this boundary condition that defines the low-energy behavior of the electronic transport of such systems. We have demonstrated how can the boundary condition valid in the low-energy limit be computed from the TB descrip- tion of the defect line. Figure 8: (Color online) Comparison between the transmis- sion probabilities, T , across the defect for: the pentagon-only [or zz(55)] defect line (full orange line); the zz(558) defect line (dashed red line); the zz(5757) defect line (dot-dashed green line); The transmission probabilities were calculated in the low energy limit from Eq. (4), with all the hopping parameters were chosen equal to 1, ξ = ξ1 = ξ2 = ξa = ξb = ξc = 1. T is q , of q = k− K+, plotted in terms of the angle of incidence, θ+ the momentum around the Dirac point K+. The transmission probabilities valid for the vicinity of the other Dirac point, K−, are obtained from the ones plotted above, by a reflection along the vertical line θ = π/2. ing around the Dirac point Kν, is given by J ν y (∆V ) = C dE ∞ −∞ (cid:20) (cid:12)(cid:12)(cid:12)(cid:12)E − e∆V 2 (cid:12)(cid:12)(cid:12)(cid:12) Tν(E, ∆V ) × f (E, µ + e∆V 2 ) − f (E, µ − e∆V 2 ) (cid:21) (44) where C = gse/(4π22vF ), gs stands for the spin degen- eracy, ∆V is the potential difference between the each side of the defect line (see Fig. 1), and f (E, µ) the Fermi-Dirac distribution function for chemical potential µ; ∆µ = e∆V is the difference between chemical poten- tials at the two grains. In addition, Tν(E, ∆V ) stands for an angle-integrated transmission probability, Tν(E, ∆V ) = 0 π T ν(E, ∆V, θ) sin θdθ, (45) The total current is obtained summing the currents asso- ciated with the two Dirac points Jy = J + y . One can verify that T+(E, ∆V ) = T−(E, ∆V ) ≡ T(E, ∆V ), and thus, Jy = gvJ + y , where gv = 2 is the valley degeneracy. The conductance is then given by G = LJy/∆V , where L is the length of the defect line, when the current is in the linear regime, y + J− (cid:18) (cid:19) G(T ) = C(cid:48) ET(E, 0) − ∂f (E, µ) ∂E dE, (46) ∞ −∞ We have presented such a reasoning while working out the problem of the electronic transport across a pentagon- only defect line, finding its transmittance to be energy in- dependent. Furthermore, we have also studied two other kinds of more realistic periodic defect lines: the zz(558) defect line, recently observed in graphene sheets,10 and the zz(5757) defect line. We have briefly examined these latter cases, emphasizing the fact that their periodicity forces the appearance of high-energy modes at the Dirac points in addition to the typical low-energy Dirac modes. It has been shown that the influence of the former, can be encompassed in an effective boundary condition seen by the low-energy massless Dirac fermions. The trans- mittance originating from such boundary conditions was again found to be energy independent. Furthermore, we have pointed out that the effective boundary conditions arising from the zz(558) and zz(5757) defect lines, turn out to be similar to the one arising from the pentagon- only defect line. Moreover, the former can be mapped into the latter by an appropriate choice of the hopping parameters at the defects. It is important to note that by expressing the trans- mission probability in terms of the boundary conditions satisfied by the Dirac fields at the defect, these results cast some light on the low-energy limit of the full TB calculations,22 leaving us with a better understanding of the physics underlying such systems. We have in addition shown how can we compute the low-energy limit conductance expression, across this kind of defect lines. Interestingly, at low temperatures, the conductance across a defect line of size L, turned out to be linear in kF L. This feature originates from the energy independence of the low-energy transmittance of our defect lines. Finally, we must mention, that the procedures pre- sented in this text, can be used to solve more complex scattering problems in graphene. Not only linear and pe- riodic defect lines, but also periodic curvilinear extended defects oriented along graphene's zigzag direction, can be worked out using the framework presented in this text. Acknowledgments J. N. B. R. was supported by Fundação para a Ciência e a Tecnologia (FCT) through Grant No. SFRH/BD/44456/2008. N. M. R. P. was supported by Fundos FEDER through the Programa Operacional Fac- tores de Competitividade - COMPETE and by FCT un- der project no. PEst-C/FIS/UI0607/2011. J. M. B. L. S. was supported by Fundacão para a Ciência e a Tecnolo- gia (FCT) and is thankful for support and hospitality of Boston University and of National University of Singa- pore. 9 Appendix A: The general low-energy boundary condition for a defect line oriented along graphene's zigzag direction In this appendix we derive Eq. (2), determining the general form of the boundary condition matrix of a zigzag oriented defect line in the continuum limit. Suppose that our defect line is located in the region defined by y ∈ [0, W ]. Assume then, that in this region, we have a constant potential term in the Dirac equation of the form V = Vs + Vxσx + Vyσy + Vzσz. (A1) Our aim is to consider the limit where W (Vs, V) → (u0, v0) when W → 0 so that we can obtain the boundary condition of the defect line in the continuum limit. We must refer that there are some works published on the literature, considering the electronic scattering across re- gions with potentials that are a particular form of that given in Eq. A1.27,28 The Dirac equation in the region of the potential is (cid:104)(cid:0)νσx(−i∂x) + σy(−i∂y)(cid:1) +(cid:0)Vs + V · σ(cid:1)(cid:105) Ψν = vF Ψν. (A2) Since the defect line is oriented along the x-direction, we can choose Ψν(x, y) = Φν(y)eiqxx which, after substitution into Eq. (A2), results in ∂yΦν(y) = i P Φν(y), (A3) (A4) where the operator P reads P = σy vF [ − νvF qxσx − Vs − V · σ] (A5) This first order differential equation (A4) can be straight- forwardly integrated, Φν(y) = eiy P Φν(0). (A6) Taking now the limit W (Vs, V) → (u0, v0) when W → 0, we obtain the following expression for the continuum limit of the boundary condition of a zigzag oriented defect line Φν(0+) = e−iσy(u0+v0·σ)/vF Φν(0−), (A7) just as in Eqs. (1) and (2). As a final comment, we must stress the fact that the remaining terms in P , namely and vF qxσx, do not contribute to the boundary condition when we take this limit; they are fixed in value, unlike the potential terms, and cannot give rise to a discontinuity when W → 0. Appendix B: Dirac equation from the passage matrix In this appendix, we show how the Dirac-Weyl equa- tion can be obtained from the low-energy passage matrix relation in Eq. (23) q = eiν2π/3e−iq·u2T(, φ).Ψν Ψν q. (B1) Since φ = (Kν + q) · u1 = ν4π/3 + q · u1, q = eiν2π/3e−iq·u2T Ψν , ν + q · u1 4π 3 (cid:18) 10 (cid:19) Ψν q (B2) As we are working out a theory valid around the Dirac points, q and are small, and we can thus expand the exponential, keeping solely the first order terms in the momentum, (cid:18) , ν (cid:19) + q · u1 4π 3 eiν2π/3e−iq·u2T = − e−iq·(u1/2+u2) (cid:21) 3 + q·u1 √ 2 cos(cid:0)ν 2π (cid:20) 0 2 ≈ I + + t− t 0 3 2 a (cid:1)(cid:20) 1 t 4 cos2(cid:0)ν 2π (cid:20) −iqy − νqx − 0 (cid:1) − 2 (cid:21) t2 t 2 3 + q·u1 −iqy + νqx 0 (cid:21) (B3) where I is the 2 × 2 identity matrix. This term cancels the one in the right hand side of Eq. (B2) and we are left (cid:21) with(cid:20) 0 t− t 0 (cid:20) −iqy − νqx 0 √ 3 2 a Ψν q + (cid:21) 0 −iqy + νqx Ψν q = 0. (B4) Upon multiplying by iσy, we obtain Dirac's equation, Ψν q = vF σν · qΨν q, (B5) where σν = (νσx, σy), with the usual notation of ν = ±1 identifying the Dirac point. Appendix C: Quick derivation of the continuum limit boundary condition for the pentagon-only defect In this brief appendix, we will present a quick deriva- tion of the pentagon-only defect low-energy boundary condition ( = 0 and qx = 0), Eq. (26). Let us start from the TB equations at the pentagon- only defect, Eqs. (12). In these equations, we begin by setting = 0. In this way, the TB equations at the defect now read 0 = −(t(cid:48))∗B(kx, 0) − tB(kx, 1), 0 = −tD(kx) − t(cid:48)A(kx, 1), 0 = −t(cid:0)A(kx, 0) + B(kx, 0)(cid:1) −2ξt cos(kxa)D(kx), 0 = −(t(cid:48))∗B(kx,−1) − tD(kx), 0 = −t(cid:48)A(kx, 0) − tA(kx,−1), (C1a) (C1b) (C1c) (C1d) (C1e) where t(cid:48) = t(1+e−ikxa/2). From now on, we set ourselves at kxa = ν4π/3. The five equations written in Eqs. (C1), contain 7 amplitudes, and we can solve them all in terms of A(kx,−1) and B(kx,−1); we obtain for A(kx, 1) and B(kx, 1), (cid:21) (cid:20) A(kx, 1) B(kx, 1) (cid:20) 0 1−1 ξ (cid:21)(cid:20) A(kx,−1) B(kx,−1) (cid:21) ,(C2) = eiν 2π 3 which, using Eq. (22b), immediately identifies the pas- sage matrix for the Dirac fields, Eq. (26). This procedure is quite general, and can be applied to the the other line defects considered in this paper. In that case, however, we must express the TB amplitudes in terms of the amplitude of the low and high energy modes and set to zero the evanescent amplitudes of the states that grow on each side of the defect. We can then solve for the low energy amplitudes on one side of the defect, and obtain directly the 2 × 2 passage matrix for the propagating modes. 1 X. Li, W. Cai, J. An, S. Kim, J. Nah, D. Yang, R. Piner, A. Velamakanni, I. Jung, E. Tutuc, et al., Science 324, 1312 (2009). 11 Y. Zhang, S. V. Dubonos, I. V. Grigorieva, and A. A. Firsov, Science 306, 666 (2004). 16 K. S. Novoselov, D. Jiang, T. Booth, V. V. Khotkevich, S. M. Morozov, and A. K. Geim, Proc. Natl. Acad. Sci. 102, 10451 (2005). 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 N. M. R. Peres, Rev. Mod. Phys. 82, 2673 (2010). 19 O. V. Yazyev and S. G. Louie, Nature Materials 9, 806 (2010). 20 D. A. Bahamon, A. L. C. Pereira, and P. A. Schulz, Phys. Rev. B 83, 155436 (2011). 21 D. Gunlycke and C. T. White, Phys. Rev. Lett. 106, 136806 (2011). (2011). 22 L. Jiang, X. Lv, and Y. Zheng, Phys. Lett. A 376, 136 23 G. W. Semenoff, Phys. Rev. Lett. 53, 2449 (1984). 24 J. N. B. Rodrigues, P. A. D. Gonçalves, N. F. G. Rodrigues, R. M. Ribeiro, J. M. B. Lopes dos Santos, and N. M. R. Peres, Phys. Rev. B 84, 155435 (2011). 25 J. A. M. van Ostaay, A. R. Akhmerov, C. W. J. Beenakker, and M. Wimmer, Phys. Rev. B 84, 195434 (2011). 26 J. N. B. Rodrigues, N. M. R. Peres, and J. M. B. Lopes dos Santos, to be submitted (2012). 27 M. I. Katsnelson, K. S. Novoselov, and A. K. Geim, Nat. 13 O. V. Yazyev and S. G. Louie, Phys. Rev. B 81, 195420 Phys. 2, 620 (2006). 14 R. Grantab, V. B. Shenoy, and R. S. Ruoff, Science 330, Condensed Matter 20, 325221 (2008). 28 J. Viana Gomes and N. M. R. Peres, Journal of Physics: 81, 165447 (2010). (2010). 946 (2011). 2 A. Reina, X. Jia, J. Ho, D. Nezich, H. Son, V. Bulovic, M. S. Dresselhaus, and J. Kong, Nano Lett. 9, 30 (2009). 3 K. S. Kim, Y. Zhao, H. Jang, S. Y. Lee, J. M. Kim, K. S. Kim, J.-H. Ahn, P. Kim, J.-Y. Choi, and B. H. Hong, Nature 457, 706 (2009). 4 S. Bae, H. Kim, Y. Lee, X. X. andJae Sung Park, Y. Zheng, J. B. andTian Lei, H. R. Kim, Y. I. Song, Y.-J. Kim, K. S. Kim, et al., Nature Nanotechnology 5, 574 (2010). 5 P. Y. Huang, C. S. Ruiz-Vargas, A. M. van der Zande, W. S. Whitney, M. P. Levendorf, J. W. Kevek, S. Garg, J. S. Alden, C. J. Hustedt, Y. Zhu, et al., Nature 469, 389 (2011). 6 K. Kim, Z. Lee, W. Regan, C. Kisielowski, M. F. Crommie, and A. Zettl, ACS Nano 5, 2142 (2011). 7 P. Nemes-Incze, K. J. Yoo, L. Tapaszto, G. Dobrik, J. Labar, Z. E. Horvath, C. Hwang, and L. P. Biro, Appl. Phys. Lett. 99, 023104 (2011). 8 A. Stone and D. Wales, Chem. Phys. Lett. 128, 501 (1986). 9 J. Meyer, C. Kisielowski, R. Emi, M. Rossell, M. Crommie, and A. Zettl, Nano. Lett. 8, 3582 (2008). 10 J. Lahiri, Y. Lin, P. Bozkurt, I. I. Oleynik, and M. Batzill, Nature Nanotechnology 5, 326 (2010). 11 A. Ferreira, X. Xu, C. Tan, S. Bae, N. Peres, B. Hong, B. Ozyilmaz, and A. Castro Neto, EPL 94, 28003 (2011). 12 S. Malola, H. Hakkinen, and P. Koskinen, Phys. Rev. B 15 K. S. Novoselov, A. K. Geim, S. V. Morozov, D. Jiang,
1802.01638	2	1802	2019-02-02T22:20:07	Quantum many-body dynamics of the Einstein-de Haas effect	[ "cond-mat.mes-hall", "cond-mat.str-el" ]	In 1915, Einstein and de Haas and Barnett demonstrated that changing the magnetization of a magnetic material results in mechanical rotation, and vice versa. At the microscopic level, this effect governs the transfer between electron spin and orbital angular momentum, and lattice degrees of freedom, understanding which is key for molecular magnets, nano-magneto-mechanics, spintronics, and ultrafast magnetism. Until now, the timescales of electron-to-lattice angular momentum transfer remain unclear, since modeling this process on a microscopic level requires addition of an infinite amount of quantum angular momenta. We show that this problem can be solved by reformulating it in terms of the recently discovered angulon quasiparticles, which results in a rotationally invariant quantum many-body theory. In particular, we demonstrate that non-perturbative effects take place even if the electron--phonon coupling is weak and give rise to angular momentum transfer on femtosecond timescales.	cond-mat.mes-hall	cond-mat	Quantum many-body dynamics of the Einstein-de Haas effect J. H. Mentink∗ and M. I. Katsnelson Radboud University, Institute for Molecules and Materials, Heyendaalseweg 135, 6525 AJ, Nijmegen, the Netherlands Institute of Science and Technology Austria, Am Campus 1, 3400 Klosterneuburg, Austria (Dated: February 5, 2019) M. Lemeshko In 1915, Einstein and de Haas and Barnett demonstrated that changing the magnetization of a magnetic material results in mechanical rotation, and vice versa. At the microscopic level, this effect governs the transfer between electron spin and orbital angular momentum, and lattice degrees of freedom, understanding which is key for molecular magnets, nano-magneto-mechanics, spintronics, and ultrafast magnetism. Until now, the timescales of electron-to-lattice angular momentum trans- fer remain unclear, since modeling this process on a microscopic level requires addition of an infinite amount of quantum angular momenta. We show that this problem can be solved by reformulating it in terms of the recently discovered angulon quasiparticles, which results in a rotationally invari- ant quantum many-body theory. In particular, we demonstrate that non-perturbative effects take place even if the electron -- phonon coupling is weak and give rise to angular momentum transfer on femtosecond timescales. I. INTRODUCTION The concept of angular momentum is ubiquitous across physics, whether one deals with nuclear collisions, chem- ical reactions, or formation of galaxies. In the micro- scopic world, quantum rotations are described by non- commuting operators. This makes the angular momen- tum theory extremely involved, even for systems consist- ing of only a few interacting particles, such as electrons filling an atomic shell or protons and neutrons compos- ing a nucleus1. In condensed matter systems, exchange of angular momentum between electrons' spins and a crys- tal lattice governs the Einstein-de Haas2 and Barnett3 effects. These effects play a key role in magnetoelastic- ity4, in the physics of molecular and atomic magnets5 -- 8, nano-magneto-mechanical systems9 -- 13, spintronics14 -- 16, and ultrafast magnetism17 -- 20. If approached from first principles, describing angular momentum transfer in condensed-matter systems repre- sents a seemingly intractable problem, since it involves couplings between an essentially infinite number of an- gular momenta of all the electrons and nuclei in a solid. As a result, although several models of spin -- lattice cou- pling have been developed21 -- 29, they either solve the problem only partially (i.e. by ignoring the orbital dy- namics of electrons) or do not account for the over- all rotational invariance of the microscopic Hamiltonian. Moreover, while non-perturbative effects of electron -- phonon coupling have been shown to play an impor- tant role in solid-state systems, most notably in the the- ory of polarons30 and in the microscopic theory of BCS superconductivity31, none of the existing theories of an- gular momentum transfer have been applied beyond the perturbative regime. As a result, over 100 years after their discovery, a fully quantum mechanical microscopic description of the Einstein-de Haas and Barnett effects remains elusive. In particular, due to these limitations existing theories cannot describe how fast angular mo- mentum can be transferred between electronic and lattice degrees of freedom. Here, we introduce a conceptually novel approach to angular momentum transfer in solids, which relies on casting both electron and lattice degrees of freedom, and -- most importantly -- the coupling between the two, di- rectly in the angular momentum basis. This results in a fully rotationally invariant quantum many-body the- ory that treats both electron spin and orbital angular momenta as well as phonon angular momentum on an equal footing. Remarkably, despite the fact that this problem involves coupling between an infinite number of angular momenta, it can be solved in closed form in terms of the angulon quasiparticle, a concept that was recently discovered in molecular physics32. In the solid- state context, the angulon represents a many-electron atom dressed by a cloud of lattice excitations carrying an- gular momentum, see Fig. 1. This quasiparticle approach not only captures perturbative effects such as the renor- malization and broadening of well-known low-frequency FIG. 1. Angulon quasiparticle in solid-state systems. A localized magnetic impurity exchanging angular momen- tum with lattice excitations can be described as the angulon quasiparticle, characterized by total (electrons+phonons) an- gular momentum. 9 1 0 2 b e F 2 ] l l a h - s e m . t a m - d n o c [ 2 v 8 3 6 1 0 . 2 0 8 1 : v i X r a AngulonMulti-orbital atomPhonon fieldTotal angular momentumSpin+orbitalangular momentumPhonon angular momentumproperties, but also makes it straightforward to take non- perturbative effects into account. We emphasize that taking a phonon-dressed many- electron atom as building block represents a key step be- yond conventional theories of electron-phonon coupling. Such theories usually account for phonons on top of an electronic Hamiltonian involving non-local interactions (electron hopping, static crystal fields), which were re- cently argued to dominate the ultrafast angular mo- mentum dynamics in electron-only theories since they break rotational symmetry33,34. However, by construc- tion, electron-only theories fail to describe how, and how fast angular momentum is transferred from electronic to lattice degrees of freedom. Moreover, even when account- ing for such non-local interactions, rotational invariance of the system as a whole should still be conserved. In this paper, as the first application of our formalism, we focus on the local angular momentum transfer be- tween electrons and phonons, which is of key importance to reveal the shortest possible timescale of the Einstein de Haas effect18,20, and for which the angulon building block alone is sufficient. If required, electron hopping and crystal fields can be introduced on top of such a build- ing block. This, however, should not alter the qualita- tive behavior of the electron-phonon system described in this paper. Interestingly, already at this level, we predict qualitatively novel non-perturbative effects taking place even if the electron -- lattice coupling is weak. These fea- tures arise at high energies and therefore enable trans- fer between electron spins and phonons at ultra short timescales. II. THE MICROSCOPIC MODEL To illustrate our approach, we consider a microscopic Hamiltonian, H = He + Hp + Hep, where He accounts for the electronic degrees of freedom, Hp describes the phonons, and Hep captures the electron -- phonon cou- pling. For concreteness, as He we take the multi-orbital (in this case, three-orbital) Hubbard-Kanamori Hamilto- nian describing localized paramagnetic atoms35 with an additional spin-orbit coupling term. We explicitly con- sider the limit where the electronic degrees of freedom are completely localized on the atom and describe the atomic Hamiltonian as L2 + ξ L · S. He = HN − 2JH S2 − JH 2 (1) Here HN = N ( N − 1)(U − 3JH)/2 + 5JH N /2, where N is the total electron number operator, U and JH parametrize the direct and exchange Coulomb interac- tions, respectively, and ξ gives the the spin-orbit coupling strength. L and S are many-electron operators for the orbital and spin angular momentum, respectively, and we use ≡ 1 such that the parameters U, JH, and ξ have the dimension of energy. In an isolated atom, angular mo- mentum J = L + S is conserved and He is diagonal in the 2 many-electron states, Γ(cid:105) = N LSJMJ(cid:105)36,37, where MJ is the projection of J onto the laboratory-frame z-axis. For the sake of simplicity, we describe the lattice de- grees of freedom by considering an isotropic elastic solid whose excitations are acoustic phonons as described by the Hamiltonian Hp = † b kλµs ωks bkλµs, (2) (cid:88) kλµs with a linear dispersion, ωks = csk, cs being the speed of sound, s the polarization index, and k = k. In Eq. (2) we have used the angular momentum representation for † the creation and annihilation operators, b kλµs and bkλµs, where λ and µ give the phonon angular momentum and its projection onto the z-axis, respectively38. The boson operators in the {k, λ, µ} basis are connected to the op- † erators in the Cartesian representation, b ks and bks, with k ≡ {kx, ky, kz}, as follows: (cid:88) k λµ (2π)3/2 † b ks = † kλµsiλY ∗ b λµ(Ωk), (3) In this angular momentum representation, each phonon carries angular momentum λ with projection µ and Hp is diagonal in the basis kλµs(cid:105). The total angular momen- tum of phonons with a given polarization is then defined by summing all excited phonons according to their occu- pations. For the three different components of the total phonon angular momentum we get the following expres- sion: Λs = † b kλµsσλ µµ(cid:48)bkλµ(cid:48)s, (4) (cid:88) kλµµ(cid:48) where σλ is the vector of matrices fulfilling the angu- lar momentum algebra in the representation of angular momentum λ = 0, 1, 2 . . .39 Hence, the total phonon an- gular momentum defined in this way is composed of non- spherical excitations of the elastic solid (e.g., p, d, f -waves for λ = 1, 2, 3). The next step is to formulate the electron -- phonon cou- pling, Hep, in a rotationally invariant way. Here we out- line the main steps of this derivation, and provide further details in Appendix A. Our starting point is the general Hamiltonian describing density -- density interactions be- tween electrons and ions of the lattice, Hep = dx dr Ψ†(x) Φ†(r)V (x, r) Ψ(x) Φ(r), (5) (cid:90) (cid:90) where Ψ(x) and Φ(r) are field operators for electrons and nuclei, respectively. Microscopically, the two-body interaction, V (x, r), stems from the Coulomb interaction between electrons and nuclei, which is obviously rotation- ally invariant, V (x, r) ≡ V (x− r). Hence, rotational in- variance is implied and the task is to describe excitations between different angular momentum states of electrons and phonons due to such an isotropic interaction. For this purpose we first expand the interaction in spherical harmonics, Ylm(Ω): Vl(x, r)Y ∗ lm(Ωx)Ylm(Ωr), (6) lm V (x, r) = (cid:88) we expand Ψ†(x) = (cid:80) (cid:88) ψ†(x) = λµ,σ Second, considering electrons localized around the nuclei, † j (x − rj) and construct the ψ † local field operators ψ j (x) from a complete set of atomic orbitals: j ρνλ(x)Y ∗ † λµ(Ωx)χ† λµσ, σ c (7) † Here χ† σ is a Pauli spinor and c λµσ is the electron cre- ation operator. The indices ν, λ, µ, σ are the principal and orbital angular momentum quantum numbers and the projections of orbital and spin quantum numbers, re- spectively. Finally, we introduce phonons by expanding V (x, r) in small displacements and subsequent transfor- mation to the spherical phonon basis. In the resulting Hamiltonian integration over electronic and nuclear an- gles can be performed analytically. Here we present the result for the case in which (phonon-mediated) hopping between different atoms in the lattice is neglected: (cid:88) (cid:88) ×(cid:104)−Aλ1µ1 × (cid:88) µ1µ2 kλµ (cid:0)1 + (−1)λ(cid:1) i 2 bkλµ + (−1)µAλ1µ1 λ−µ,λ2µ2 (8) (cid:105) † b kλµ λµ,λ2µ2 N W LL(cid:48)S MM(cid:48)µ1µ2 X(N LM SΣ , N L(cid:48)M(cid:48)SΣ ), H loc ep,λ1 = Uλ(k) N SΣ LL(cid:48)MM(cid:48) λµ,λ2µ2 It is important to note that, first, only terms with k· es(k) (cid:54)= 0 (with es(k) the polarization vector) survive, as follows from the expansion of V (x, r) to first order in nuclear displacements. This implies that only longitudi- nal phonons contribute in the case of an isotropic elastic solid. Second, in Eq. (8) we introduced the X-operators37 (or Hubbard operators40), X(Γ, Γ(cid:48)) = Γ(cid:105)(cid:104)Γ(cid:48), that de- scribe the transitions between many-electron states due † λ1µ1σcλ1µ2σ. Here Aλ1µ1 to the terms c captures the selection rules for single-electron excitations due to phonons (see Appendix A) and W LL(cid:48)S determines the allowed transitions between many-electron terms with different orbital angular momenta, LM (cid:54)= L(cid:48)M(cid:48). In contrast, SΣ = S(cid:48)Σ(cid:48), since the electron -- phonon coupling does not depend on spin S and its projection, Σ . We em- phasize that W LL(cid:48)S is based upon the exact solution of the many-electron problem, which takes into account all allowed electronic transitions with N and N ± 1 elec- trons. The coupling strength is determined by Uλ(k) that originates from the radial integrals. Explicit formulas for Uλ(k) and W are given in Appendix A. Third, we stress that although we started from a spherically symmet- ric Coulomb interaction, the charge distribution of the atomic orbitals is not spherically symmetric. As a result, MM(cid:48)µ1µ2 MM(cid:48)µ1µ2 3 the coupling between different non-spherical electron dis- tributions induced by phonons leads to angular momen- tum transfer. Indeed, when including only s-orbitals no transfer takes place, so one needs to have asymmetric p-, d-, or f -orbitals. At the same time, rotational invariance is preserved, i.e. simultaneous rotation of both electron and phonon subsytems leaves H loc ep unchanged. The full Hamiltonian, H = He + Hp + H loc ep , is rota- tionally invariant and is therefore diagonal in the basis of a given total angular momentum, JMJ(cid:105). In addition, it exhibits striking similarities with the one used to describe molecules rotating in superfluids, which were recently found to form so-called angulon quasiparticles32,39,41. Instead of mechanical rotation of a molecule, here we deal with orbital angular momen- tum of electrons. Lattice phonons, on the other hand, play the role of superfluid excitations. The anisotropic molecule -- helium interaction, in turn, is replaced with the rotationally invariant electron -- phonon coupling, Eq. (8), derived here from microscopic principles. Inspired by this analogy, in what follows we make use of the angulon con- cept in order to understand angular momentum transfer in solid-state systems. The key advantage of casting the problem in terms of angulons is that it allows for a drastic simplification. The latter, in turn, enables studying non-perturbative ef- fects based on a transparent variational ansatz. By anal- ogy with the molecular angulon, we construct an ansatz featuring all possible single-phonon excitations allowed by angular momentum conservation in the subspace of a given number of electrons, N : LSJMJ(cid:105)0(cid:105) C JMJ (cid:88) ψJMJ(cid:105) = Z 1/2 JMJ βJMJ kλl † kλµ0(cid:105)lmSΣ(cid:105), b LM,SΣ C LM (cid:88) lm,λµ (9) + kλµ lm M Σ In Eq. (9), Z 1/2 JMJ A similar ansatz has been previously shown to provide a good approximation to the energies of polarons42 and angulons43, even far beyond the weak-coupling regime and βJMJ considered in this paper. kλl are variational parameters which are determined by min- imizing (cid:104)ψJMJH − EψJMJ(cid:105). This yields the equation, E = EJMJ − ΣJMJ (E), from which the variational ground- state energy, E, is determined self-consistently. Here is the energy of the many-electron state without EJMJ phonons and ΣJMJ (E) plays the role of a self-energy de- scribing the effect of electron -- phonon interactions: (cid:88) kλl ΣJMJ(E) = Uλ(k)2Q2 λl λl − E + ωk EJMJ , (10) where Qλl are matrix elements that determine the al- lowed transitions to electronically excited states, EJMJ , λl due to phonons with angular momentum λ, which are given in an explicit form in the Appendix B. Non- perturbative effects described below originate from the energy, E, in the denominator of ΣJMJ (E). These effects do not take place in conventional second-order perturba- tion theory, which is recovered by replacing ΣJMJ (E) → ΣJMJ(EJMJ). Moreover, the quasiparticle approach enables the study of angular momentum transfer in response to a time- dependent magnetic field, as described by the Zeeman term: HZ(t) = µBB(t) ·(cid:16) gL L + gS S , (11) (cid:17) MJ write Ψ(t)(cid:105) = e−iEt(cid:80) In this case, we search for a solution based on the time- dependent variational principle44,45. Following Ref.46, we ψJMJ (t)(cid:105). Next, for each MJ, we use the variational ansatz (9) with time-dependent pa- rameters, Z 1/2 kλl (t), which are determined by minimizing (cid:104)Ψ(t)i∂t − H − HZ(t)Ψ(t)(cid:105). Crucially, this JMJ variational approach also gives rise to non-perturbative effects in the dynamical response. That is, in addition to the perturbative effects that give rise to phonon dress- ing of states with different MJ, qualitatively new features appear. (t) and βJMJ Within the quasiparticle picture this can be under- stood as follows. Due to the static phonon dressing, an external field can trigger virtual transitions to atomic (cid:54)= J, where J is the ground-state angu- states with J(cid:48) lar momentum of the isolated atom. Without phonon coupling such excitations are obviously forbidden due to selection rules. Moreover, in the presence of electron -- phonon coupling, these electronically excited states can decay by emitting phonons. This can either lead to (i) emergence of quasi-bound states of the quasiparticle it- self, where reduced angular momentum of the electrons is balanced by increased phonon angular momentum or (ii) give rise to incoherent scattering of phonons. Both effects are captured by our theory. III. STATIC EFFECTS OF PHONON DRESSING We first illustrate the appearance of non-perturbative contributions in the static case by evaluating the effect of phonon dressing on different components of angular momentum, Iz = (cid:104)ψJMJ IzψJMJ(cid:105), where I = L, J, Λ. In the variational calculation, the electronic Hamilto- nian (1) is controlled by a single parameter, which we set to ξ/JH = 0.1. Here we focus on the case with N = 1 electron, for which the configuration of the bare impurity is given by L = 1, S = J = 1/2. Further- more, we consider the state with MJ = J, which is the ground state in the presence of a static magnetic field, B0 < 0. Fig. 2a shows different components of angular momentum as a function of the dimensionless electron -- phonon coupling strength, u = (u/EL)(cid:112)EM /EL/(2π2). Here u denotes the magnitude of the interaction Uλ(k), EM = 2/(2M a2 0), with M is the atomic mass of the nuclei (using EL = (JH + ξ)/2 as the unit of energy and the lattice spacing a0 as the unit of length, see Ap- In the absence of coupling, (cid:104)Sz(cid:105) = −1/6 pendix A 5). 4 FIG. 2. Static effects of phonon dressing. a. Quenching of orbital angular momentum by phonons. Different compo- nents of angular momentum as a function of the dimensionless electron -- phonon coupling strength, u. Due to the coupling, electronic orbital angular momentum is reduced and phonon angular momentum emerges, while the total angular momen- tum, Jz = Jz + Λz, is conserved. b. Renormalization of the electron g-factor as a function of the dimensionless electron -- phonon coupling strength. In the perturbative regime (inset) the dependence on the coupling strength is quadratic, which can be understood as phononic Lamb shift41. and (cid:104)Lz(cid:105) = 2/3, such that (cid:104)Jz(cid:105) = MJ = 1/2. While Sz remains unperturbed since Hep does not depend on spin, we find a reduction of orbital angular momentum that is quite distinct from the conventional picture of orbital angular momentum quenching. Instead of static crystal fields breaking rotational symmetry, here the dynamical crystal field induced by phonons causes the reduction of (cid:104) Lz(cid:105) as well as of (cid:104) Jz(cid:105), while conserving the total an- gular momentum, MJ = (cid:104) Jz(cid:105) + (cid:104)Λz(cid:105). The presence of phonon angular momentum also influences the response to magnetic fields. For quasi-static fields, this is reflected by the renormalization of the electron g-factor, which we determine from the well-known relation47: gJ = gL + gS 2 + 2 gL − gS (cid:104)L2 − S2(cid:105) (cid:104)J2(cid:105) (12) Evaluating the second term with the variational wave functions gives the result shown in Fig. 2b. In the pertur- 00.20.40.6angularmomentum〈Lz〉〈Jz〉〈Λz〉00.10.20.30.40.70.80.91.0ugJ-factor00.010.680.70gJ0ab5 bative regime (inset) this yields a quadratic dependence on u which can be understood as the phononic analog of the Lamb shift41, where virtual phonon excitations play the role of the photon excitations of quantum electrody- namics, thereby causing angular-momentum-dependent dressing of the electronic states. For larger coupling strengths, gJ features a linear dependence until signa- tures of saturation are observed at intermediate coupling, u ∼ 0.4, where the single-phonon ansatz of Eq. (9) be- comes less reliable. Within the quasiparticle picture, the observed enhancement of the g-factor is analogous to the enhancement of the moment of inertia due to the for- mation of molecular angulons32,41 and to the increased electron effective mass in the polaron problem. IV. DYNAMIC EFFECTS OF PHONON DRESSING Next we reveal the importance of non-perturbative ef- fects in the dynamical response,by computing the linear response to an additional time-dependent magnetic field, B(t) = (Be−iωt+εt + B∗eiωt+εt)/2, B (cid:28) B0. By de- termining the time-dependent changes of the variational parameters to linear order in B(t) and by using the gen- eral relation, j eiωt+εt(cid:105) , (cid:88) (cid:104) j δIi(t) = 1 2 ij (ω)Bje−iωt+εt + α(I) α(I) ij (−ω)B∗ where δIi(t) = (cid:104) Ii(t)(cid:105) − (cid:104) Ii(cid:105)0, i = x, y, z, we can derive closed-form expressions for the magnetic susceptibilities, ij (ω), see Appendix C. In Fig. 3 we plot −ωImαI (ω)xx α(I) as a function of ω (in units of EL), for the configuration N = 5, L = 1, S = J = 1/2, MJ = −1/2, with electronic parameters ξ/JH = 0.1, µBB0/JH = 0.02, and various coupling strength u. For the smallest value of u = 10−3, the spectrum consists of a sharp peak close to the electron spin resonance (ESR) of a free atom. Furthermore, an additional broad spectral feature appears at higher ener- gies, which is associated with incoherent phonon scatter- ing. At low (high) frequencies, the response for L,−S, J and Λ have the same (opposite) sign, where the minus sign in S comes from the fact that in the ground state S is antiparallel to both L and J. Fig. 3b shows that the ESR peak width for Λ is much narrower than that for J, and increases only slightly with increasing the electron -- phonon coupling strength. Hence, at these frequencies the phonons are damped much weaker than the electron spin and orbital angular momentum. This is consistent with the interpretation that the broadening of the ESR peak for J is due to the dressing with phonons in either ground or excited states, while for Λ the decay is only possible when the dressing of distinct MJ levels is differ- ent. The red arrow shows the position of the ESR peak of the free atom. We note that phonon dressing causes a shift of the ESR peak to lower frequencies, which corre- sponds to a reduction of the effective g-factor, in contrast Dynamical effects of phonon dressing. FIG. 3. a. Magnetic susceptibililty for various electron -- phonon cou- pling strengths, u. Besides the low-frequency electron spin resonance (ESR) peak, a broad second spectral feature due to phonon dressing is observed. By increasing u gradually, a sec- ond sharp quasiparticle peak appears due to non-perturbative effects. b. Zoom-in in the vicinity of the ESR peak, demon- strating that phonon dressing causes a slight shift of the ESR peak to lower frequencies. c. Zoom-in in the vicinity of the sharp quasiparticle peak at u = 10−2. Exactly at the peak, the phonon susceptibility dips due to angular momen- tum transfer. 0.00.20.40.60.81.00.000.050.100.150.20-ωImαxx(ω)L-Sx10JΛu=10-3u=4·10-3u=7·10-3u=10-2u=10-3u=10-2Jω0Λ0.0220.0240.02600.0040.0080.383450.383500.383550.38360-0.2-0.10.00.10.2ω-ωImαxx(ω)L-SJΛ-ωImαxx(ω)abcto what is observed for the static g-factor in Fig. 2. The static and dynamical g-factors are indeed two different quantities. While both can be derived from the mag- netic susceptibility, the static gJ ∼ αzz(ω = 0), while the ESR peak follows from the pole of αxx(ω) at ω (cid:54)= 0. Similar differences between static and dynamical electron g-factors occur in the Fermi-liquid theory48. Interestingly, In addition to a shift of the ESR peak, Fig. 3a shows that upon increasing the electron -- phonon cou- pling strength the incoherent part moves towards higher frequencies. Moreover, a second sharp peak gradually emerges in between the ESR peak and the incoherent part, which is shown in Figure 3c for u = 10−2. Both high-frequency responses have opposite sign for phonon and electron angular momentum, demonstrating that magnetic fields at these frequencies induce transfer of angular momentum from electronic to lattice degrees of freedom. The second sharp peak can be identified as an additional quasiparticle peak, i.e. a metastable excited state of the atom dressed by additional phonons carry- ing angular momentum. This is reminiscent to the effect observed in conventional polaron physics. It is known as the 'relaxed excited state' in the Frolich model49 and as the 'excited phonon-polaron bound state' in the Hol- stein model50,51, which arise at intermediate and strong coupling. in the present case this non- perturbative effect emerges already at weak coupling, u (cid:28) 1. The reason is that in our model the electron couples to low-energy acoustic phonons with a linear dis- persion ω = ck, rather than to gapped optical phonons. The presence of additional peaks is rooted in the poles of the susceptibility, which involves additional self- consistent solutions, E(cid:48), to the equation E(cid:48) = EJMJ − ΣJMJ(E(cid:48)), where E(cid:48) > E, E being the ground-state en- ergy. Due to the Kramers-Kronig relations, additional ∗ − E, sharp peaks can only occur for frequencies ω > E(cid:48) where E(cid:48) ∗ is defined by Max [Im ΣJMJ (E(cid:48))], since in this range −Re [ΣJMJ(E(cid:48))] is a decreasing function. Hence, there is a threshold value, u∗ ≈ 0.0033 for the param- eters used, below which no additional metastable states can occur. Above the threshold, u > u∗, we find an approximately linear dependence of the position of the metastable state on the coupling strength. Such a behav- ior is governed by the linear dependence of the ground- state energy on u in this regime. In Appendix C we ex- plicitly confirm this analysis numerically for various cou- pling strengths. We emphasise that the appearance of such additional peaks is rooted in the general Fano-type shape of Re [ΣJMJ (E)], and should therefore be qualita- tively independent of the particular approximation used to calculate the self-energy. 6 Furthermore, we obtained that upon changing the spin-orbit coupling strength, the width of the peaks changes but their position is hardly affected. These non- perturbative effects have important consequences for the dynamics. In particular, at optical frequencies, coupling of spins with a magnetic field is usually considered negli- gible due to the small magnitude of the magnetic compo- nent of an electromagnetic wave compared to its electric component, and the absence of magnetic dipole transi- tions. Our model, however, reveals that even for u (cid:28) JH, an additional resonance emerges at an energy scale of ω ∼ JH/5 due to non-perturbative electron -- phonon in- teractions. In the presence of such resonances, a mag- netic field can induce transfer of angular momentum be- tween spin and lattice degrees of freedom at ultrafast, femtosecond timescales. V. CONCLUSIONS The results presented here demonstrate that the prob- lem of describing the quantum dynamics of angular momentum transfer in condensed matter systems with multi-orbital atoms can be greatly simplified by casting it in terms of the angulon quasiparticles. This reformula- tion is achieved by deriving the electron -- phonon interac- tion in a rotationally invariant form and using the Hub- bard operators to keep track of the total angular momen- tum of electrons. We find that the effect of dressing of electron orbital angular momentum with phonon angular momentum leads to qualitatively new, non-perturbative high-frequency effects that should be observable in elec- tron spin resonance experiments at THz and optical fre- quencies. Promising systems for experimental confirma- tion are paramagnetic CoO and FeO systems and non- magnetic oxides containing orbitally degenerate impurity atoms, which, analogously to the model system stud- ied here, contain partially filled degenerate t2g orbitals. While here we focused on local angular momentum trans- fer, which is highly relevant to understanding the fastest possible timescale for angular momentum transfer, the angulon can be used as a building block of models tak- ing into account non-local transfer terms. Furthermore, the theory can be extended to include static crystal fields and magnetic ordering, which would pave the way to a deeper understanding of lattice dynamics during ultrafast demagnetization20,52 -- 54. This can potentially resolve the long-lasting debate as to whether the angular momentum transfer during ultrafast demagnetization is local or non- local and ultimately reveal the fastest possible timescale of the Einstein-de Haas and Barnett effects. ∗ [email protected] 1 D. A. Varshalovich, A. N. Moskalev, and V. K. Khersonski, Quantum theory of angular momentum (World Scientific, Singapore, 1988). 2 A. Einstein and W. J. de Haas, "Experimenteller nachweis der ampereschen molekularstrome," Verh. Dtsch. Phys. Ges. 17, 152 (1915). 3 S. J. Barnett, "Magnetization by rotation," Phys. Rev. 6, 239 (1915). 4 K. B. Vlasov and B. Kh. Ishmukhametov, "Equations of motion and state of magnetoelastic media," Sov. Phys. JETP 19, 142 (1964), [Zh. Eksp. Toer. Fiz. 46, 201 (1964)]. 5 C. Calero, E. M. Chudnovsky, and D. A. Garanin, "Field dependence of the electron spin relaxation in quantum dots," Phys. Rev. Lett. 95, 166603 (2005). 6 D. A. Garanin and E. M. Chudnovsky, "Quantum entan- glement of a tunneling spin with mechanical modes of a torsional resonator," Phys. Rev. X 1, 011005 (2011). 7 M. Ganzhorn, S. Klyatskaya, M. Ruben, and W. Werns- dorfer, "Quantum Einstein-de Haas effect," Nature Comm. 7, 11443 (2016). 8 F. Donati, S. Rusponi, S. Stepanow, C. Wackerlin, A. Singha, L. Persichetti, R. Baltic, K. Diller, F. Patthey, E. Fernandes, J. Dreiser, Z. Sljivancanin, K. Kummer, C. Nistor, P. Gambardella, and H. Brune, "Magnetic re- manence in single atoms," Science 352, 318 (2016). 9 T. M. Wallis, J. Moreland, and P. Kabos, "Einstein-de Haas effect in a NiFe film deposited on a microcantilever," Appl. Phys. Lett. 89, 122502 (2006). 10 A. A. Kovalev, G. E. W. Bauer, and A. Brataas, "Nanome- chanical magnetization reversal," Phys. Rev. Lett. 94, 167201 (2005). 11 R. Jaafar, E. M. Chudnovsky, and D. A. Garanin, "Dy- namics of the Einstein-de Haas effect: Application to a magnetic cantilever," Phys. Rev. B 79, 104410 (2009). 12 J. Tejada, R. D. Zysler, E. Molins, and E. M. Chudnovsky, "Evidence for quantization of mechanical rotation of mag- netic nanoparticles," Phys. Rev. Lett. 104, 027202 (2010). 13 H. Keshtgar, S. Streib, A. Kamra, Y. M. Blanter, and G. E. W. Bauer, "Magnetomechanical coupling and ferro- magnetic resonance in magnetic nanoparticles," Phys. Rev. B 95, 134447 (2017). 14 A. A. Kovalev, G. E. W. Bauer, and A. Brataas, "Current- driven ferromagnetic resonance, mechanical torques, and rotary motion in magnetic nanostructures," Phys. Rev. B 75, 014430 (2007). 15 M. Matsuo, J. Ieda, E. Saitoh, and S. Maekawa, "Effects of mechanical rotation on spin currents," Phys. Rev. Lett. 106, 076601 (2011). 16 M. Matsuo, E. Saitoh, and S. Maekawa, "Spin- mechatronics," J. Phys. Soc. Jpn. 86, 011011 (2017). 17 E. Beaurepaire, J.-C. Merle, A. Daunois, and J.-Y. Bigot, "Ultrafast spin dynamics in ferromagnetic nickel," Phys. Rev. Lett. 76, 4250 (1996). 18 B. Koopmans, J. J. M. Ruigrok, F. Dalla Longa, and W. J. M. de Jonge, "Unifying ultrafast magnetization dy- namics," Phys. Rev. Lett. 95, 267207 (2005). 19 A. Kirilyuk, A. V. Kimel, and Th. Rasing, "Ultrafast opti- cal manipulation of magnetic order," Rev. Mod. Phys. 82, 2731 (2010). 20 C. Dornes, Y. Acremann, M. Savoini, M. Kubli, M. J. Neugebauer, E. Abreu, L. Huber, G. Lantz, C. A. F. Vaz, H. Lemke, E. M. Bothschafter, M. Porer, V. Esposito, L. Rettig, M. Buzzi, A. Alberca, Y. W. Windsor, P. Beaud, U. Staub, Diling Zhu, Sanghoon Song, J. M. Glownia, and S. L. Johnson, "The ultrafast einstein -- de haas effect," Na- ture 565, 209 -- 212 (2019). 21 J. H. Van Vleck, "Paramagnetic relaxation times for tita- nium and chrome alum," Phys. Rev. 57, 426 (1940). 7 22 E. R. Callen and H. B. Callen, "Static magnetoelastic cou- pling in cubic crystals," Phys. Rev. 129, 578 -- 593 (1963). 23 R. L. Melcher, "Rotationally invariant theory of spin- phonon interactions in paramagnets," Phys. Rev. Lett. 28, 165 (1972). 24 P. A. Fedders, "Effects of self-consistency on spin-lattice relaxation," Phys. Rev. B 15, 3297 (1977). 25 A. S. Mishchenko, "Crystalline fields in systems with ex- change and magnetoelastic interaction," J. Exp. Theor. Phys. 66, 487 -- 493 (1997). 26 E. M. Chudnovsky, D. A. Garanin, and R. Schilling, "Uni- versal mechanism of spin relaxation in solids," Phys. Rev. B 72, 094426 (2005). 27 L. Zhang and Q. Niu, "Angular momentum of phonons and the Einstein-de Haas effect," Phys. Rev. Lett. 112, 085503 (2014). 28 D. A. Garanin and E. M. Chudnovsky, "Angular momen- tum in spin-phonon processes," Phys. Rev. B 92, 024421 (2015). 29 M. Fahnle, Th. Tsatsoulis, Ch. Illg, M. Haag, B. Y. Muller, and L. Zhang, "Ultrafast demagnetization after femtosec- ond laser pulses: Transfer of angular momentum from the electronic system to magnetoelastic spin-phonon modes," J. Supercond. Nov. Magn. 30, 1381 (2017). 30 J. Appel, "Polarons," in Solid State Physics, Vol. 21, edited and D Turnbull (Academic by H Ehrenreich, F Seitz, Press, New York, 1968). 31 J. Bardeen, L. N. Cooper, and J. R. Schrieffer, "Micro- scopic theory of superconductivity," Phys. Rev. 106, 162 -- 164 (1957). 32 M. Lemeshko, "Quasiparticle approach to molecules in- teracting with quantum solvents," Phys. Rev. Lett. 118, 095301 (2017). 33 W. Tows and G. M. Pastor, "Many-body theory of ul- trafast demagnetization and angular momentum transfer in ferromagnetic transition metals," Phys. Rev. Lett. 115, 217204 (2015). 34 John Kay Dewhurst, Peter Elliott, Sam Shallcross, and Sangeeta Sharma, spin transfer," Nano Let- (2018), 29424230, Eberhard K. U. Gross, "Laser-induced intersite ters https://doi.org/10.1021/acs.nanolett.7b05118. pMID: 18, 1842 -- 1848 35 A. Georges, L. de' Medici, and J. Mravlje, "Strong corre- lations from Hund's coupling," Ann. Rev. of Cond. Mat. Phys. 4, 137 (2013). 36 Z.B. Rudzikas, Theoretical Atomic Spectroscopy (Cam- bridge University Press, Cambridge, 2007). 37 V. Yu. Irkhin and Yu. P. Irkhin, "Many-electron operator approach in the solid state theory," physica status solidi (b) 183, 9 -- 58 (1994). 38 M. Lemeshko and R. Schmidt, "Molecular impurities in- teracting with a many-particle environment: from ultra- cold gases to helium nanodroplets," in "Cold Chemistry: Molecular Scattering and Reactivity Near Absolute Zero", eds. O. Dulieu and A. Osterwalder, arXiv:1703.06753 (2017). 39 R. Schmidt and M. Lemeshko, "Deformation of a quantum many-particle system by a rotating impurity," Phys. Rev. X 6, 011012 (2016). 40 J. Hubbard, "Electron correlations in narrow energy bands iv. the atomic representation," Proc. Roy. Soc. A 285, 542 (1965). 41 R. Schmidt and M. Lemeshko, "Rotation of quantum im- purities in the presence of a many-body environment," Phys. Rev. Lett. 114, 203001 (2015). 42 R. Combescot and S. Giraud, "Normal state of highly po- larized Fermi gases: Full many-body treatment," Phys. Rev. Lett. 101, 050404 (2008). 43 G. Bighin, T. V. Tscherbul, and M. Lemeshko, "Dia- grammatic monte carlo approach to angular momentum in quantum many-particle systems," Phys. Rev. Lett. 121, 165301 (2018). 44 A.D. McLachlan, "A variational solution to the time- dependent schrodinger equation," Mol. Phys. 8, 39 (1964). 45 R. Jackiw and A. Kerman, "Time-dependent variational principle and the effective action," Phys. Lett. A 71, 158 (1979). 46 M. I. Katsnelson and V. Yu. Irkhin, "Metal-insulator tran- sition and antiferromagnetism in the ground state of the Hubbard model," J. Phys. C 17, 4291 (1984). 47 N. W. Ashcroft and N. D. Mermin, Solid State Physics (Saunders College, Philadelphia, 1976). 48 P. M. Platzman and P. A. Wolff, Waves and Interactions in Solid State Plasmas, Solid State Physics: Advances in Research and Applications. Supplement No. nr. 13 (Aca- demic Press, 1973). 49 J. T. Devreese, "Lectures on Frohlich polarons from 3D to 0D -- including detailed theoretical derivations," arXiv:1012.4576v6 (2015). 50 A. A. Gogolin, "The spectrum of an intermediate polaron and its bound states with phonons at strong coupling," physica status solidi (b) 109, 95 -- 108 (1982). 51 J. Bonca, S. A. Trugman, and I. Batisti´c, "Holstein po- laron," Phys. Rev. B 60, 1633 -- 1642 (1999). 52 E. Jal, V. L´opez-Flores, N. Pontius, T. Fert´e, N. Bergeard, Ch. Boeglin, B. Vodungbo, J. Luning, and N. Jaouen, "Structural dynamics during laser-induced ultrafast de- magnetization," Phys. Rev. B 95, 184422 (2017). 53 S. Bonetti, M. C. Hoffmann, M.-J. Sher, Z. Chen, S.-H. Yang, M. G. Samant, S. S. P. Parkin, and H. A. Durr, 8 "THz-driven ultrafast spin-lattice scattering in amorphous metallic ferromagnets," Phys. Rev. Lett. 117, 087205 (2016). 54 A. H. Reid, X. Shen, P. Maldonado, T. Chase, E. Jal, P. W. Granitzka, K. Carva, R. K. Li, J. Li, L. Wu, T. Vecchione, T. Liu, Z. Chen, D. J. Higley, N. Hartmann, R. Coffee, J. Wu, G. L. Dakovski, W. F. Schlotter, H. Ohldag, Y. K. Takahashi, V. Mehta, O. Hellwig, A. Fry, Y. Zhu, J. Cao, E. E. Fullerton, J. Stohr, P. M. Oppeneer, X. J. Wang, and H. A. Durr, "Beyond a phenomenological description of magnetostriction," Nat. Commun. 9, 388 (2018). 55 C. Kittel, Quantum Theory of Solids (John Wiley and Sons, Inc, New York, 1963). 56 V. Yu. Irkhin and Yu. P. Irkhin, Electronic structure, corre- lation effects and physical properties of d- and f-metals and their compounds (Cambridge International Science Pub- lishing, 2007). VI. ACKNOWLEDGEMENTS We acknowledge discussions with E. Yakaboylu. J.H.M. acknowledges support by the Nederlandse Or- ganisatie voor Wetenschappelijk Onderzoek (NWO) by a VENI grant, and is part of the Shell-NWO/FOM- initiative 'Computational sciences for energy research' of Shell and Chemical Sciences, Earth and Life Sci- ences, Physical Sciences, FOM and STW. M.I.K. ac- knowledges support by the European Research Council (ERC) Advanced Grant No. 338957 (FEMTO/NANO). M.L. acknowledges support from the Austrian Science Fund (FWF), under project No. P29902-N27. Appendix A: Derivation of the electron-phonon coupling Hamiltonian In this Appendix we provide details on the derivation of the rotationally invariant electron-phonon coupling Hamiltonian. In particular, we discuss the derivation of the local electron-phonon Hamiltonian, the integration over electronic and nuclear posi- tions to derive the allowed terms respecting rotational invariance, introduce the Hubbard operators and obtain the dimensionless electron-phonon coupling strength. 1. Local electron-phonon coupling Starting from Eq. (5), the local electron -- phonon coupling is derived by first expanding Ψ†(x) =(cid:80) identity for the nuclear density operator, Φ†(r) Φ(r) =(cid:80) (cid:90) (cid:88) j (x − rj). Inserting an † ψ i δ(r − ri), and neglecting electron hopping between different nuclei, we have j H loc ep = † dx ψ j (x) V (x, rij) ψj(x). ij 2. Integration over electronic coordinates To exploit rotational invariance of V (x, r) = V (x − r), it is convenient to expand in spherical harmonics: V (x − r) = Vl(x, r)Y ∗ lm(Ωx)Ylm(Ωr). (cid:88) lm (A1) (A2) Inserting a complete set of atomic orbitals, (cid:88) λµ,σ † ψ j (x) = ρνλ(x)Y † † ∗ σ c λµ(Ωx)χ j,λµσ, 9 (A3) where ν is the principal quantum number, λ and µ are the quantum numbers for the orbital angular momentum and its projection, respectively, and σ is the spin projection, we obtain: H loc ep = † j,λ1µ1σ cj,λ2µ2σ c dx ρνλ1 (x)Y ∗ λ1µ1 (Ωx) V (x, rij) ρνλ2 (x)Yλ2µ2 (Ωx), (A4) (cid:88) (cid:88) (cid:88) ij,σ λ1µ1 λ2µ2 where we used that V does not depend on spin. The integral in (A4) involves a radial part and an angular integral over three (cid:20)(cid:90) spherical harmonics:(cid:88) (cid:88) lm x2dxρνλ1 (x)ρνλ2 (x)Vl(x, rij) gλ1λ2,l(rij) (−1)mAλ1µ1 l−m,λ2µ2 dΩxY ∗ λ1µ1 (Ωx) Y ∗ lm(Ωx) Yλ2µ2 (Ωx) Ylm(Ωrij ) (cid:21) = Here gλ1λ2,l(rij) ≡(cid:82) dx x2ρνλ1 (x)ρνλ2 (x)Vl(x, rij) and the integration over spherical coordinates yields1: lm Ylm(Ωrij ). (cid:115) (2l + 1)(2λ2 + 1) 4π(2λ1 + 1) C λ10 l0,λ20C λ1µ1 lm,λ2µ2 , Aλ1µ1 lm,λ2µ2 = where C l1m1 l2m2,l3m3 are Clebsch-Gordon coefficients. (cid:90) (cid:21)(cid:20)(cid:90) (cid:88) k (cid:90) (A5) (A6) (cid:105) . (A7) 3. Integration over nuclear coordinates For further derivations, we write the H loc ep in the form (cid:88) (cid:88) j,σ λ1µ1 λ2µ2 (cid:88) (cid:104) i,lm H loc ep = † j,λ1µ1σ cj,λ2µ2σ c 1 2 gλ1λ2,l(rij) Y ∗ lm(Ωij) Aλ1µ1 lm,λ1µ2 + gλ1λ2,l(rij) Ylm(Ωij) (−1)mAλ1µ1 l−m,λ1µ2 i−ri, We aim to describe phonons that account for the collective dynamics of the nuclear subsystem at small deviations, u(ri) = r(cid:48) from the equilibrium positions, ri. For convenience, we take the continuum limit for the nuclear coordinates ri and focus on the coupling to a single atom (j = 0). The dependence on the nuclear coordinates r in H loc ep is then conveniently described in reciprocal space (cid:48) lm (r F λ1λ2 ) = gλ1λ2,l(r)Ylm(Ωr) = lm (k)eik·r ≈ F λ1λ2 f λ1λ2 lm (r) + u(r) · ∇rF λ1λ2 lm (r) (A8) lm (r) is assumed to vanish, since it gives rise to static crystal field terms that are absent in an isotropic elastic The term F λ1λ2 environment. The gradient is calculated from the Fourier series: lm (k) ik eik·r = f λ1λ2 ∇rF λ1λ2 (cid:88) lm (r) = k (cid:88) k 1 V −l Gλ1λ2 l i (k) Ylm(Ωk) ik eik·r, (A9) where V is the total volume of system and f λ1λ2 spherical coordinates, from which it follows that lm (k) is evaluated using the inversion formula and expansion of plane-waves in Gλ1λ2 l (k) = 4π r2 dr gλ1λ2,l(r)jl(kr), (A10) where jl(x) is the spherical Bessel function. For an isotropic elastic solid, the displacements are written in terms of phonon creation and annihilation operators as follows55: u(r) = u † (r) = 1 n (2M ωks) −1/2es(k) , (A11) (cid:88) ks (cid:104)bkseik·r + b † kse −ik·r(cid:105) where M is the nuclear mass, n is the number of nuclei, and s = 1, 2, 3 is the polarization index. The polarization vectors, es(k), are defined by the relations e1(k) = k/k and e2,3(k) · k = 0 for longitudinal and transverse phonons, respectively. Hence, from (cid:88) (cid:90) λµ 1 Vk(2π)3 k (cid:88) kλµ † b k = (2π)3/2 k † b kλµiλY ∗ λµ(Ωk). Using (cid:88) = (cid:90) dk k2 dΩk, we can integrate over angles in k-space yielding (cid:90) dr u(r) · ∇rF λ1λ2 lm (r) = (cid:104)−bkλµ(−1)λ(−1)µδλlδµ−m + b † kλµδλlδµm (cid:105) , Uλ(k) i 1 Vr where(cid:80) k ≡(cid:82) dk and U λ1λ2 λ (k) = 1 (2π)3/2 k2(2M ωk) −1/2Gλ1λ2 λ (k). 10 (A13) (A14) (A15) (A16) (A17) (cid:90) 1 Vr evaluating the scalar product in Eq. (A8) using Eq. (A11) and Eq. (A9), we obtain that only longitudinal phonons contribute. We drop the label s = 1 below and obtain: dr u(r) · ∇rF λ1λ2 lm (r) = 1 V (2M ωk) −1/2Gλ1λ2 (k) (ik) i Here Vr = V /n is the volume of the unit cell and we used(cid:82) dreikr = (2π)3δ(k). We are still left with the dependence on angles, k l (A12) Ωk, which can be removed by transforming to spherical phonon operators using the definition38: (cid:88) −l (cid:104)−Ylm(Ω−k) b−k + Ylm(Ωk) b (cid:105) † k . Finally, we obtain: H loc ep = (cid:88) (cid:88) (cid:88) kλµ jσ λ1µ1 λ2µ2 † jλ1µ1σ cjλ2µ2σ U λ1λ2 c λ (k) bkλµ + (−1)µAλ1µ1 λ−µ,λ1µ2 † b kλµ (cid:105) . i 2 (cid:16) 1 + (−1)λ(cid:17)(cid:104)−Aλ1µ1 (cid:17)† (cid:16) H loc λµ,λ1µ2 The factor of (1− (−1)λ) originates from the assumption of an isotropic elastic solid, which ensures that only even λ contributes to the transfer of angular momentum. Hermicity, H loc , is easily proved using the symmetry relations for the Clebsch- Gordon coefficients, C λ1µ2 = (−1)µC λ1µ1 ep = ep . λµ,λ1µ1 λ−µ,λ1µ2 4. Hubbard operators Since V (x, r) does not depend on spin, phonons only change the total orbital angular momentum L of the electrons. This is made explicit by transforming from single-electron operators to many-electron X-operators37 (also known as Hub- bard operators40), X(Γ, Γ(cid:48)) = Γ(cid:105)(cid:104)Γ(cid:48). For a general operator acting on a single site i we have Oi = (cid:104)Γ OiΓ (cid:48)(cid:105) Xi(Γ, Γ (cid:48) ), (cid:48) Xi(Γ, Γ ) = iΓ(cid:105)(cid:104)iΓ (cid:48). (A18) (cid:88) Γ,Γ(cid:48) Here Γ = N LM SΣ α are the quantum numbers of many-electron states, where N denotes the total number of electrons, L, S are total orbit and spin quantum numbers with projections M, Σ , and α is the Racah seniority quantum number. For the single-electron creation operator the matrix element reads37,56: (cid:104)ΓNc iλµσΓN−1(cid:105) = N 1/2GΓN † ΓN−1 C ΓN ΓN−1,λµσ, (A19) where GΓN ΓN−1 coefficients as = GLN SN LN−1SN−1 is the coefficient of fractional parentage36, and C ΓN ΓN−1,γ is expressed through the Clebsch-Gordon C ΓN ΓN−1,γ = C LN MN LN−1MN−1,lmC SN ΣN SN−1ΣN−1,sσ, with s = 1/2. Using Eq. (A19) we obtain (cid:104)ΓNc † iλ1µ1σ1 ciλ2µ2σ2Γ N(cid:105) = N (cid:48) (cid:88) Γ(cid:48)(cid:48) N−1 GΓN Γ(cid:48)(cid:48) N−1 C ΓN Γ(cid:48)(cid:48) N−1,λ1µ1σ1 GΓ(cid:48) N Γ(cid:48)(cid:48) N−1 C Γ(cid:48) N Γ(cid:48)(cid:48) N−1,λ2µ2σ2 . (A20) (A21) In the electron-phonon coupling only the summation over single-electron operators with the same spin σ1 = σ2 enters, (cid:88) σ (cid:88) Γ(cid:48)(cid:48) N−1 † c iλ1µ1σ ciλ2µ2σ = N GΓN Γ(cid:48)(cid:48) N−1 N−1,λ1µ1σGΓ(cid:48) N Γ(cid:48)(cid:48) N−1 C ΓN Γ(cid:48)(cid:48) C Γ(cid:48) N Γ(cid:48)(cid:48) N−1,λ2µ2σ (cid:48) Xi(ΓN , Γ N ), 11 (A22) which ensures that only states S(cid:48) = S, Σ (cid:48) N and σ and by using the unitarity relation for the Clebsch-Gordan coefficients. For example, for a three-orbital atom, we have λ1 = λ2 and the seniority quantum number can be omitted. In this case we obtain we obtain the following coupling term: N M(cid:48) C L(cid:48) L(cid:48)(cid:48) N−1M(cid:48)(cid:48) N = ΣN contribute, as follows from summation over both Σ (cid:48)(cid:48) GL(cid:48) N SN L(cid:48)(cid:48) N−1S(cid:48)(cid:48) (cid:88) (cid:88) L(cid:48)(cid:48) N−1M(cid:48)(cid:48) N SN N µ1µ2 L(cid:48)(cid:48) N−1S(cid:48)(cid:48) C LN MN GLN SN N−1,λ1µ1 N−1,λ1µ2 (A23) N−1 N−1 = N , L(cid:48)(cid:48) N−1S(cid:48)(cid:48) N−1 M(cid:48)(cid:48) N−1 yielding W LN L(cid:48) MN M(cid:48) (cid:88) † c iλ1µ1σ ciλ1µ2σ = σ N SN ΣN (cid:88) (cid:88) LN L(cid:48) N MN M(cid:48) N N W LN L(cid:48) MN M(cid:48) N SN N µ1µ2 Xi(N LN MN SN ΣN , N L (cid:48) N M (cid:48) N SN ΣN ) (A24) Note that it follows from the symmetry of the Clebsch-Gordan coefficients that only M(cid:48) summation. N = MN − µ1 + µ2 remains in the 5. Electron-phonon coupling strength For numerical calculations we need to evaluate the radial integrals in Uλ(k) = U λ1λ1 (k) (see Eq. (A16)) for which we use Gaussian form factors, gλ(r) = uλ λ), where uλ parametrizes the strength of the electron-phonon coupling. Introducing dimensionless units, with EL = (JH + ξ)/2 being the unit of energy and the lattice spacing a0 being the unit of length, we can write (2π)3/2 e−r2/(2r2 λ Uλ(k) = uλ k3/2 c1/2 r2 dr e −r2/(2r2 λ)jλ(kr) dimensionless electron-phonon coupling strength we obtain u = (u/EL)(cid:112)EM /EL/(2π2), where EM = 2/2M a2 We use rλ = 1 to characterize the interaction range. The interaction strength is parametrized by u0 = u, u2 = 0.5u. For the 0. For transition metal atoms M ∼ 100 · 10−27 kg, a0 ∼ 2A, EL ∼ 0.5 eV, we have EM /EL ∼ 0.1, which ensures that u (cid:28) 1 even if (u/EL) ∼ 1. For the dimensionless speed of sound we use c = 0.05, consistent with c ∼ 3 − 6 · 103 m/s for typical solid-state systems. 0 (A25) (cid:90) ∞ Appendix B: Variational solution for the static case 1. Non-perturbative self-energy Here we discuss the derivation of the variational solution in more detail, providing explicit expressions for the matrix elements that enter the final result. For the static case, we deal with the Hamiltonian H = H C e + H LS e + Hp + H loc ep + HZ , (B1) where HZ = µBB0 . Owing to the presence of spin-orbit coupling, only Γ = LSJMJ are good quantum numbers, where J = L + S with projection MJ and for brevity we omit the label N . In addition, since Hep couples directly only to orbital momentum L, we choose the variational wavefunctions as follows: ψJMJ(cid:105) = Z 1/2 LSJMJ(cid:105)0(cid:105) + βJMJ kλl C JMJ LM,SΣ C LM lm,λµ kλµ0(cid:105)lmSΣ(cid:105) = ψ1(cid:105) + ψ2(cid:105), † b (B2) (cid:88) (cid:88) (cid:88) kλµ lm M Σ where we used that LSJMJ(cid:105) =(cid:80) JMJ minimizing E = (cid:104)ψ Hψ(cid:105)/(cid:104)ψψ(cid:105). This is equivalent to minimizing F = (cid:104)ψ H − Eψ(cid:105) and the following terms enter are variational parameters to be determined from and β(JMJ ) kλl JMJ M Σ C JMJ LM,SΣLM SΣ(cid:105). Z 1/2 (cid:88) e + HZ + Hp + HZψ1(cid:105) = EJMJZ 1/2 JMJ e + HZ + Hp + HZψ2(cid:105) = (EJMJ 2 λl + ωk)βJMJ kλl 2 (cid:104)ψ1 H C (cid:104)ψ2 H C e + H LS e + H LS (cid:16) gL Lz + gS Sz(cid:17) kλl EJMJ = EN S + ELL(L + 1) + EJ J(J + 1) + EZ M z MJ EJMJ λl = EN S + ELl(l + 1) + EJ Pλl + EZ mz λl (B3) (B4) (B5) (B6) Here EL = −JH − ξ/2, EJ = ξ/2, and EN S is the energy term depending on N and S which remains constant in the variational solution. Furthermore, we have defined M z MJ = C JMJ LM,SΣ [gLM + gSΣ ] , mz λl = C JMJ LM,SΣ C LM lm,λµ [gLm + gSΣ ] (B7) (cid:88) (cid:88) (cid:16) M Σ mµ (cid:17)2(cid:16) (cid:17)2 as well as the bare spin-orbit coupling terms in the atomic state with phonons excited, C JMJ LM,SΣ C JMJ LM(cid:48),S ¯Σ C LM lm,λ(M−m)C LM(cid:48) l ¯m,λ(M−m) C jmj lm,SΣ C jmj l ¯m,S ¯Σ j(j + 1), where ¯m = m − (M − M(cid:48)), ¯Σ = Σ + (M − M(cid:48)). In addition we have (cid:104)ψ1 H loc ep ψ2(cid:105) = −i Z 1/2∗ βJMJ kλl Uλ(k)Qλl, (cid:88) (cid:16) 1 + (−1)λ(cid:17) JMJ kλl (cid:88) (cid:16) Qλl = 1 2 N C JMJ LM,SΣ ΣM µ1µ2 (cid:17)2 (cid:32)(cid:88) Aλ1µ1 λ(µ1−µ2),λ1µ2 W LlS M ¯M µ1µ2 C LM l ¯M ,λ(µ1−µ2) , (cid:33) (k). In deriving these expressions we have used several times the symmetry properties λ with ¯M = M −µ1 +µ2 and Uλ(k) = U λ1λ1 of the Clebsch-Gordon coefficients. Minimization gives the equations E = EJMJ − ΣJMJ (E), ΣJMJ (E) = Uλ(k)2Q2 λl λl − E + ωk EJMJ (cid:88) Once E is obtained, the variational parameters can be determined from the relations (cid:88) (cid:32) kλl . ≡ RJMJ kλl (E), Z 1/2 JMJ = 1 + RJMJ kλl (E)2 βJMJ kλl Z 1/2 JMJ = −i Uλ(k) Qλl λl − ωk E − EJMJ kλl (cid:33)−1/2 . (B11) (B12) 12 (B8) (B9) (B10) (B16) (C1) (C2) (cid:88) M Σ (cid:16) (cid:88) (cid:17)2 (cid:88) Pλl = MM(cid:48)Σ m,jmj Once the variational parameters are determined, observables can be directly evaluated. For the calculation of the g-factor 2. g-factor renormalization we need to evaluate Direct substitution gives (cid:104)ψL2 − S2ψ(cid:105) (cid:104)ψJ2ψ(cid:105) = Z 1/2 JMJ For weak electron-phonon interactions we have βJMJ gJ ≈ g0 J + gL − gS 2 where gL − gS (cid:104)L2 − S2(cid:105) (cid:104)J2(cid:105) gJ = gL + gS + 2 2 2(cid:2)L(L + 1) − S(S + 1)(cid:3) +(cid:80) 2J(J + 1) +(cid:80) (cid:20) l(l + 1) − S(S + 1) JMJ kλl /Z 1/2 JMJ 2 = RJMJ Z 1/2 kλl βJMJ kλl βJMJ kλl 2(cid:2)l(l + 1) − S(S + 1)(cid:3) (cid:21) kλl 2Pλl Pλl J(J + 1) J(J + 1) kλl2 (cid:28) 1 and we can write − L(L + 1) − S(S + 1) J(J + 1) RJMJ kλl2 (B13) (B14) , (B15) (cid:88) kλl g0 J = gL + gS 2 + gL − gS L(L + 1) − S(S + 1) 2 J(J + 1) . Hence, at small coupling we expect a change of the g-factor that scales quadratically with the coupling strength. Appendix C: Variational solution for dynamical response 1. Linear response formulas To derive the equations for linear response, the variational parameters are written as Z 1/2 (t) = Z 1/2 + δZ 1/2 JMJ JMJ JMJ kλµ(t) = βJMJ βJMJ kλµ + δβJMJ kλµ(t), (t) where 13 (t) = δZ 1/2 δZ 1/2 JMJ JMJ kλl (t) = δβJMJ δβJMJ kλl (ω)e (ω)e −iωt+εt + δZ 1/2 −iωt+εt + δβJMJ (−ω)eiωt+εt, kλl (−ω)eiωt+εt. JMJ and Z 1/2 JMJ and βJMJ kλµ are given by the solution of the static case. For convenience we write the dynamical contributions as δZ 1/2 JMJ (ω) = δβJMJ kλl (ω) = B · XJMJ (ω) B · χJMJ kλl (ω), 1 2 1 2 from which δZ 1/2 JMJ time-dependent changes of angular momentum, δIi(t) = (cid:104) Ii(t)(cid:105) − (cid:104) Ii(cid:105)0, yields the susceptibilities kλl (−ω) are obtained by replacing ω → −ω and B → B∗. In this notation, evaluation of the (−ω) and δβJMJ (cid:20) (cid:88) MJM(cid:48) J (cid:2)Z 1/2∗ JMJ I MJM(cid:48) J i JMJ,j(−ω)(cid:3) + ∗ (cid:88) kλl (cid:2)βJMJ∗ kλl χJM(cid:48) I MJM(cid:48) λl,i J kλl,j(−ω)(cid:3)(cid:21) , kλl,j(ω) + βJM(cid:48) kλl χJMJ∗ J J XJM(cid:48) J,j(ω) + Z 1/2 JM(cid:48) J X α(I) ij (ω) = where J i I MJM(cid:48) I MJM(cid:48) λl,i = = (cid:104)LSJMJ IiLSJM J(cid:105), (cid:48) LM,SΣ C JM(cid:48) C JMJ J J (cid:88) LM(cid:48),SΣ(cid:48) (cid:88) MM(cid:48)ΣΣ(cid:48) mm(cid:48)µµ(cid:48) lm,λµC LM(cid:48) C LM lm(cid:48),λµ(cid:48)(cid:104)lmSΣ λµ Iilm (cid:48) (cid:48) (cid:48)(cid:105), λµ SΣ (C3) (C4) are the matrix elements of angular momentum components I = L, S, J, Λ without and with phonons excited, respectively, with i = x, y, z. For practical calculations we focus to the case of a non-degenerate ground state. For example, at B0 > 0 the variational state with M0 kλl are nonzero. For numerical evaluation of α(I) kλl,i(ω) from the expressions: J = −J has the lowest energy. Hence, in the static problem only Z 1/2 ij (ω) it is convenient to determine the contributions of XJMJ,i(ω) and χJMJ and βJM0 JM0 J J +(cid:80) XJMJ,i(ω) = Z 1/2 JM0 J J i M MJM0 λl,i K1MJ λl (ω) ω + iε + ∆EJMJ − K0MJ (ω) λl mMJM0 J , (C5) where and (cid:88) k J∗ βJM0 kλl χJMJ kλl,i(ω) =Z 1/2 JM0 J K1MJ 2 K2MJ λl (ω) mMJM0 λl,i J JM0 J λl (ω)XJMJ,i(ω) + Z 1/2 (cid:17) (cid:16) (cid:17) (cid:16) + gSSMJM(cid:48) λl,i + gSSMJM(cid:48) gLLMJM(cid:48) gLLMJM(cid:48) λl,i i i J J J J , , J J i = µB M MJM(cid:48) mMJM(cid:48) λl,i = µB (cid:88) (cid:88) (cid:88) kλl k K0MJ (ω) = K1MJ λl (ω) = K2MJ λl (ω) = k (cid:1) Uλ(k) Qλl2 (cid:0)ω + iε + ∆EJMJ (cid:0)∆E (cid:0)∆E λl − ωk λl − ωk λl − ωk Uλ(k) Qλl2 (cid:1)(cid:0)ω + iε + ∆EJMJ (cid:1)2(cid:0)ω + iε + ∆EJMJ Uλ(k) Qλl2 JM0 J JM0 J λl − ωk λl − ωk (cid:1) (cid:1) , (C6) (C7) (C8) (C9) (C10) with ∆EJMJ = E − EJMJ and ∆EJMJ , where E is the variational ground-state energy. The integrals are computed numerically with ε (cid:28) 1 until convergence is achieved. Explicit expressions for the energy of the bare impurity, EJMJ , and the energy of the bare impurity with phonons excited, EJMJ λl , as well as for Qλl, are given in Appendix B. λl = E − EJMJ λl 2. Emergence of high-frequency peaks 14 The emergence of high-frequency peaks in the susceptibilities can be understood by analyzing (i) the poles of XJMJ,i (Eq. (22) (ω) and (ii) the functions K1MJ λl (ω) (Eqs. (27) -- (28) of kλl,i(ω). Here we elaborate on the poles of XJMJ,i which give rise to of the Methods section), which involves changes of Z 1/2 JMJ the Methods section), which corresponds to changes of βJMJ additional metastable states of the quasiparticle and are determined by the equation λl (ω) and K2MJ ω + ∆EJMJ − K0MJ (ω) = 0, (C11) where ∆EJMJ = E − EJMJ . Using ω = E(cid:48) − E, with E the ground-state energy and the definition of K0MJ (ω) (see Eq. (26) of the Methods section), we find that the solution of (C11) coincides with the solutions of E(cid:48) = EJMJ − ΣJMJ (E(cid:48)) corresponding to the angulon states at energies E(cid:48) > E. Such additional solutions only occur for sufficiently high electron -- phonon coupling strength, as we illustrate in Fig. 4 by plotting E(cid:48)− EJMJ (gray solid line), −Re [ΣJMJ (E(cid:48))] (blue dashed line) and Im [ΣJMJ (E(cid:48))] (red dotted line) as a function of E(cid:48) for a few different electron -- phonon coupling strengths. Vertical dashed and dotted lines indicate the variational ground state energies of the angulon and of the free atom, respectively. We observe that additional sharp peaks only occur for E(cid:48) (cid:29) E(cid:48) ∗ is defined by Max [Im ΣJMJ (E(cid:48))], since in this range −Re ΣJMJ (E(cid:48)) is a monotonically decreasing function and Im ΣJMJ (E(cid:48)) remains small but not negligible. ∗, where E(cid:48) FIG. 4. Emergence of metastable states due to the Fano-like shape of the self-energy. Self-consistent solutions are determined by crossings of the solid grey line and the blue dashed line, which represents the real part of the self-energy, −Re ΣJMJ (E(cid:48)). a. For the smallest electron -- phonon coupling strength, u = 0.001, only one self-consistent solution is found (vertical dashed line) at a slightly lower energy than the ground-state energy of the bare atom (vertical dotted line). b. For larger coupling strengths, additional self-consistent solutions emerge. The metastable state corresponds to the solution with the largest energy. c. By further increasing the coupling strength, the metastable state shifts towards higher energies E(cid:48) where −Re ΣJMJ (E(cid:48)) monotonically decreases. For comparison, the imaginary part of the self-energy is shown (red dotted line), which determines the lifetime of the metastable state. Note the different scales of the vertical axes for different coupling strengths. In addition, in order to investigate the scaling of the high-frequency peak with electron -- phonon coupling strength u, we evaluate the dependences E(u) and E(cid:48)(u). The result is shown in Fig. 5, by solid black (E) and dashed blue (E(cid:48)) lines, from which we conclude that the change of ω(u) = E(cid:48)(u) − E(u) is approximately linear with u. The results shown in Fig. 4 and Fig. 5 are computed for the same parameters as Fig. 3 of the main text: N = 5, L = 1, S = J = 1/2, MJ = −1/2, with electronic parameters ξ/JH = 0.1, µBB0/JH = 0.02. -2.2-2.1-2.0-1.9-1.8-1.7-1.6-0.010-0.0050.0000.0050.0100.0150.020E'u=0.001-2.2-2.1-2.0-1.9-1.8-1.7-1.6-0.20.00.20.4E'u=0.005E'-EJMJ-ReΣJMJImΣJMJ-2.2-2.1-2.0-1.9-1.8-1.7-1.6-1.0-0.50.00.51.01.52.0E'u=0.01abc15 FIG. 5. Scaling of the stable and metastable states with electron -- phonon coupling strength, u. E is the ground-state energy, E(cid:48), is the energy of the additional self-consistent solution of the equation E(cid:48) = EJMJ − ΣJMJ (E(cid:48)), which is found at energies above Max [−Re ΣJMJ (E(cid:48))] (see Fig.4). An approximately linear scaling with u is found for both states, yielding a linear dependence of ω = E(cid:48) − E on u in the parameter range investigated. EE'0.0000.0020.0040.0060.0080.010-2.1-2.0-1.9-1.8u
1608.08737	1	1608	2016-08-31T06:22:09	Rashba spin-splitting of single electrons and Cooper pairs	[ "cond-mat.mes-hall" ]	Electric weak links, the term used for those parts of an electrical circuit that provide most of the resistance against the flow of an electrical current, are important elements of many nanodevices. Quantum dots, nanowires and nano-constrictions that bridge two bulk conductors (or superconductors) are examples of such weak links. Here we consider nanostructures where the electronic spin-orbit interaction is strong in the weak link but is unimportant in the bulk conductors, and explore theoretically the role of the spin-orbit active weak link (which we call a "Rashba spin splitter") as a source of new spin-based functionality in both normal and superconducting devices. Some recently predicted phenomena, including mechanically-controlled spin- and charge currents as well as the effect of spin polarization of superconducting Cooper pairs, are reviewed.	cond-mat.mes-hall	cond-mat	a Rashba spin-splitting of single electrons and Cooper pairs R. I. Shekhter,1 O. Entin-Wohlman,2, 3, ∗ M. Jonson,1, 4 and A. Aharony2, 3 1Department of Physics, University of Gothenburg, SE-412 96 Goteborg, Sweden 2Raymond and Beverly Sackler School of Physics and Astronomy, Tel Aviv University, Tel Aviv 69978, Israel 3Physics Department, Ben Gurion University, Beer Sheva 84105, Israel 4SUPA, Institute of Photonics and Quantum Sciences, Heriot-Watt University, Edinburgh, EH14 4AS, Scotland, UK (Dated: September 26, 2018) Electric weak links, the term used for those parts of an electrical circuit that provide most of the resistance against the flow of an electrical current, are important elements of many nanodevices. Quantum dots, nanowires and nano-constrictions that bridge two bulk conductors (or supercon- ductors) are examples of such weak links. Here we consider nanostructures where the electronic spin-orbit interaction is strong in the weak link but is unimportant in the bulk conductors, and explore theoretically the role of the spin-orbit active weak link (which we call a "Rashba spin split- ter") as a source of new spin-based functionality in both normal and superconducting devices. Some recently predicted phenomena, including mechanically-controlled spin- and charge currents as well as the effect of spin polarization of superconducting Cooper pairs, are reviewed. PACS numbers: 72.25.Hg,72.25.Rb I. INTRODUCTION In classical electrodynamics an electric field affects the spatial (orbital) motion of a charged particle while a mag- netic field also leads to a precession of the magnetic mo- ment of a stationary magnetic particle. Additional dy- namics occur if the magnetic particle moves in an electric field since the spatial motion of the particle generates a precession of its magnetic moment. This precession oc- curs because in the reference frame of the moving par- ticle the electric field is time-dependent and therefore, according to Maxwell's equations, generates a magnetic field. It follows that the rate of the electric-field induced precession is proportional to both the momentum p of the particle and to the strength of the electric field E, giving rise to a Larmor correction in the kinetic energy of the electron of the form1,2 ∆E = µ · (p × E) , 1 2mc (1) where µ is the magnetic moment of the electron. Since electrons carry both charge and magnetic moment (spin), they are subjected to this type of coupling between or- bital and magnetic degrees of freedom, known as the spin- orbit (SO) interaction. Remarkably, if in the classical result (1) one lets µ → (−e/mc) s, where e = e and s = (/2) σ is the electron spin operator [the compo- nents of the vector σ are the Pauli matrices σx,y,z], and if one also replaces eE by ∇V , the gradient of the crystal potential, the result coincides with the SO coupling term Hso in the Pauli equation (the low-velocity approxima- tion of the Dirac equation),3 Hso = − 4m2c2 σ · [p × ∇V (r)] . (2) Being a relativistic effect, the SO coupling is small for free electrons in an external electric field but can be quite large for electrons moving in a crystal. There, the inter- nal electric (crystal) field can be very strong, leading in turn to spin-split energy bands. This is the case for crys- tals lacking spatial inversion symmetry as discovered by Dresselhaus4 for zinc blende structures (e.g., GaAs, InSb, and CdTe), and by Rashba and Sheka5 for wurtzite struc- tures (such as GaN, CdS, and ZnO). Other examples of systems without spatial inversion symmetry, which are more relevant in the context of this review, are those with a surface or an interface. Motivated by experimental work on semiconductor het- erostructures at the time, Vas'ko6 and Bychkov and Rashba7 showed theoretically that a surface potential may induce an SO coupling of the electrons, that lifts the spin degeneracy of the energy bands. The main con- tribution to the crystal field turns out to be not due to the surface potential itself, but to its effect on the atomic orbitals near the surface, which become distorted (mixed) so that their contributions to the SO coupling are not averaged out by symmetry. Although it is pos- sible in principle to calculate an effective SO Hamilto- nian for this case ab initio,8 starting from Eq. (2), or using a semi-quantitative tight-binding approach,9 it is convenient to adopt the phenomenological SO Hamilto- nian proposed in Ref. 7. In the notation of Ref. 10, this "Rashba" Hamiltonian, which is valid for systems with a single high-symmetry axis that lack spatial inversion symmetry, reads Hso = kso m∗ σ · (p × n) . (3) Here n is a unit vector along the symmetry axis (the c- axis in a hexagonal wurtzite crystal, the growth direction in a semiconductor heterostructure, the direction of an external electric field), m∗ is the effective mass of the electron, and kso is the strength of the SO interaction in units of inverse length,11 usually taken from experiments. 2 electric field to tune the SO coupling strength.10,14 A semiclassical picture of the spin evolution for an elec- tron propagating through a nanowire-based weak elec- tric link is presented in Fig. 1(b) for the simple case of an SO interaction caused by an external electric field. Here the spin twist that accompanies the propagation of an electron through the straight one-dimensional wire is pictured as a semiclassical precession of the spin dur- ing the time it takes for the electron to pass through the SO-active wire from the source to the drain elec- trode. Quantum mechanically, the effect of such a spin rotation can be accounted for by an extra semiclassical phase,(cid:82) δp· dr/, which is acquired by the electron wave function because of the renormalization of the electronic momentum, p → p + δp, as the electrons enter the weak link. This renormalization is necessary for the total en- ergy to be conserved in the weak link, where the SO in- teraction modifies the energy. For a free electron, whose kinetic energy is p2/(2m) before entering the weak link, this extra Aharonov-Casher phase15 follows from Eq. (1) [where we let µ → −e/(2mc)σ]. To lowest order in the SO interaction it takes the form ϕAC = − e 4mc2 (σ × E) · dr . (4) (cid:90) The Aharonov-Casher phase arises from the interaction between the magnetic moment (spin) of an electron and a static electric field. It is dual to the Aharonov-Bohm phase,16 which is an extra phase induced by the inter- action between the charge of an electron and a static magnetic field. Since σ is an operator in spinor space, the Aharonov- Casher phase (4) manifests itself in a splitting of the spin state of an electron that enters the wire (in a "spin-up" state, say) into a coherent superposition of spin-up and spin-down states. We call such a splitting of electronic waves in spin space "Rashba spin splitting" and the SO- active weak links that give rise to it "Rashba (spin) split- ters". It is instructive to demonstrate the spin splitting ex- plicitly for a simple case. Consider the motion of an elec- tron along the x direction. Let E = Ez for x > 0 and E = 0 for x < 0, and dr = dx x. With the convenient notation kso = eE/(4mc2) in Eq. (4), one finds eiϕAC = e−iksox σy = cos(ksox) − iσy sin(ksox). (5) When this phase factor acts on the spin state χ< = (1, 0)T =↑(cid:105) (where the subscript < indicates x < 0), in which the spin is aligned along the positive z-axis, the resulting spin state is χ> (with > denoting the region x > 0), where (cid:21) (cid:20) cos(ksox) sin(ksox) χ> =eiϕAC χ< = = cos(ksox) ↑(cid:105) + sin(ksox) ↓(cid:105) , (6) is a coherent superposition of the spin-up and spin-down states [see also Fig. 1(b), where ksox is denoted by ϕ]. FIG. 1: (a) In a semiclassical picture, the spin s = (/2)σ [which we for convenience label by σ rather than by s] of an electron moving with momentum p along a curved trajectory precesses around the effective magnetic field Bso caused by a spin-orbit (SO) interaction, induced in this simple exam- ple by an external electric field E. Being perpendicular to both p and E, the direction of Bso and hence the direction of the precession axis changes along the trajectory leading to a complex "spin twisting" effect. (b) The spin evolution for an electron propagating through an SO-active weak link bridging two SO-inactive leads. The spin twist that accom- panies the propagation of the electron through the straight one-dimensional wire is pictured as a semiclassical precession of the spin during the time it takes for the electron to pass from the source to the drain electrode. The exploration of the "Rashba physics" that follows from Eq. (3) is today at the heart of the growing research field of "spin-orbitronics", a branch of spintronics that focuses on the manipulation of nonequilibrium material properties using the SO coupling (see, e.g., the recent reviews Refs. 12 and 13). Semiclassically, the effect of the SO interaction given by Eq. (3) on an electron can be viewed as a precession of its spin around an effective magnetic field, Bso, whose di- rection is perpendicular to both the symmetry axis n and the momentum p of the electron as it propagates along a trajectory (orbit). When the electron's trajectory is bent, as in Fig. 1(a), the orientation of the precession changes along the trajectory, which makes the picture of the spin evolution more complicated than a simple pre- cession. We will call the resulting transformation "spin twisting". Spin-control of electronic transport can be achieved in principle by incorporating a finite-length SO-active el- ement into a device, for example by using a nanowire made of a material with strong SO coupling as a weak link between two SO-inactive bulk conductors. If quan- tum spin-coherence is preserved during the transfer of electrons through the weak link, the Rashba SO inter- action makes it possible to manipulate the spin currents through such devices. This is because the strong spatial inhomogeneity of the SO coupling prevents the electronic spin from being a good quantum number and produces a twisting of the spin of the electrons that enter such a spin- active weak link. As we shall see, the net spin twisting accumulated by the electrons as they leave the SO-active weak link can be controlled by mechanically bending the nanowire and possibly also by using a strong external cossinσϕϕ=↑+↓ϕp E pBso σ⊕E (a) (b) Note that the explicit form of the initial spin state χ< matters for the result of the scattering process described by Eq. (6]. A complementary view comes from noting that χ> is an eigenfunction of sin(ksox)σx + cos(ksox)σz, which corresponds to a rotation of the spin quantization axis in the XZ-plane as the electron propagates along the x-axis.11 Before ending this part of the Introduction we men- tion that strain is another mechanism for inducing an SO coupling, the precise form of which depends on the material and type of strain involved. In a single-wall carbon nanotube, for instance, strain can be thought of as occurring when a flat graphene ribbon is rolled up to form a tube. The strain-induced SO coupling in a one- dimensional model of a such a nanotube is described by the Hamiltonian Hstrain so = vFkstrain so σ · n , (7) so where vF is the Fermi velocity, kstrain is a phenomeno- logical parameter that gives the strength of the SO in- teraction in units of inverse length, and n is a unit vec- tor pointing along the longitudinal axis of the nanotube. Equation (7) is a simplified form of the SO Hamiltonian derived for a realistic model of such a nanotube.17 A number of consequences of the Rashba spin split- ting for transport phenomena suggested recently will be reviewed here. Two groups of phenomena will be consid- ered. The first concerns incoherent electron transport, where the possible spin-coherence of the Rashba split states does not play any role. In this case the Rashba weak-link can be viewed as a spin-flip scattering center for the transferred electrons. The kinetic consequences of such a spin-flip relaxation are considered in Sec. II and Sec. III. Spin-coherent effects in non-superconducting de- vices only occur in multiply-connected geometries18 and are outside the scope of this review. Non-trivial interfer- ence effects in singly-connected geometries do occur in su- perconducting structures; these are reviewed in Sec. IV. In particular, the way by which a supercurrent19 flowing through a weak link acting as a Rashba spin splitter is affected by the SO interaction is rather unique. An SO- active superconducting weak link brings the opportunity to affect the spin-sensitive pairing of electrons in the su- perconducting condensate and can be a tool that allows a spin design of superconducting Cooper pairs. Some im- mediate consequences of such a spin-polarization of the Cooper pairs are presented in Sec. IV. Is the Rashba spin-splitting experimentally important? for this to be the case it is necessary that Clearly, ϕAC ∼ 1; the Aharonov-Casher phase accumulated dur- ing the propagation of an electron through the spin- split device under consideration must be of order one. In a free-electron model ϕAC is given by Eq. (4). Its magnitude for a straight SO-active channel of length d, placed in a perpendicular electric field E, and when the spin-polarization axis is roughly perpendicular to both the electric field and the direction of the channel, is ϕAC ∼ (eEd)/(4mc2). Since mc2 = 0.5 MeV, this gives 3 for d = 1 µm a rather small value, ϕAC ∼ 10−3, even for an electric field as strong as E = 1 V/nm. Allowing for an effective electron mass m∗ (cid:54)= m and a g-factor different from two would add a factor (gm/2m∗) which could be significant if the effective mass is small and the g-factor is large. Even so it seems challenging -- although perhaps not impossible -- to find a situation where the Rashba spin-splitting directly due to an external electric field is important. An external electric field may well have an indirect effect on the SO interaction, by influencing the mixing of the atomic orbitals particularly in nanoscale systems with poor screening and large surface to volume ratios. To estimate the scale of the SO interaction due to crystal fields associated with atomic orbitals (in a crys- tal lacking spatial inversion symmetry) one may consider the electric field at a small distance r from an atomic nucleus of charge Ze, given in SI units by the expression E = Ze/(4π0r2) where 0 = 8.9 × 10−12 F/m is the permittivity of vacuum. For Z = 10 and r = 0.05 nm (= aB, the Bohr radius) one finds E ∼ 5 × 1012 V/m, which in conjunction with Eq. (4) gives ϕAC ∼ 1 for the same 1 µm long channel as above. Rather than trying to improve on this estimate by a full band-structure calcula- tion, it is common to determine the SO coupling strength from experiments. Finally, we note that when the SO in- teraction in a nanowire bridging two bulk (SO-inactive) electrodes is induced by a crystal field, then although the direction of the crystal field cannot be independently controlled, the spin precession axis in the wire can still be varied with respect to the spin quantization axes in the bulk electrodes by bending the wire. Generally, a large mechanical deformability of nanostructures, orig- inating from their composite nature complemented by the strong Coulomb forces accompanying single-electron charge transfer, offer an additional functionality of elec- tronic nanodevices.20,21 Coherent nano-vibrations in sus- pended nanostructures, with frequency in the gigahertz range, were detected experimentally.22 II. SUSPENDED NANOWIRES AS MECHANICALLY-CONTROLLED SPIN SPLITTERS In charge transport, electronic beam-splitters (e.g., us- ing tunnel barriers) are key ingredients in interference- based devices. Tunnel-barrier scatterers may serve as coherent splitters of the electronic spin when the tun- neling electrons also undergo spin (Rashba) scattering. This allows one to map various interference-based phe- nomena in charge transport onto electronic spin trans- portation. Such spin-splitters can be made functional by adding to them a mechanical degree of freedom that con- trols their geometrical configuration in space, to which the Rashba interaction is quite sensitive. Because of this, one achieves mechanical coherent control and mechanical tuning of the spin filters.23 4 the Aharonov-Bohm magnetic flux (through an effective area).31 The model system exploited in the calculations is de- picted in Fig. 3. The tunneling amplitudes through the weak link are calculated in Appendix A. It is shown there that the linear Rashba interaction manifests itself as a matrix phase factor on the tunneling amplitude.32 In the geometry of Fig. 3, this phase is induced by an elec- tric field perpendicular to the XY plane [see Eq. (3)], with RL = {xL, yL} for the left tunnel coupling and RR = {xR,−yR} for the right one, where both radius vectors RL and RR are functions of the vibrational de- grees of freedom (as specified in Sec. A 3). The quantum vibrations of the wire which modify the bending angle, make the electronic motion effectively two dimensional. This leads to the possibility of manipulating the junction via the Aharonov-Bohm effect, by applying a magnetic field which imposes a further phase on the tunneling am- plitudes φL(R) = −(π/Φ0)(BxL(R)yL(R)), where Φ0 is the flux quantum (a factor of order one is absorbed31 in B). FIG. 3: Illustration of the geometry used to calculate the spin-orbit coupling dependence of the tunneling amplitude. Two straight segments are tunnel-coupled to left L and right R electronic electrodes, with possibly different, spin-dependent, chemical potentials µL,σ and µR,σ. The setup lies in the XY plane; a magnetic field applied along z is shown by ⊗. The setup corresponds to a configuration in which the wire is controlled only mechanically, and the STM is not shown. Reprinted figure with permission from R. I. Shekhter et al., Phys. Rev. Lett. 111, 176602 (2013) c(cid:13)2013 by the American Physical Society. The calculation of the spin-resolved current through such a junction is detailed in Appendix B, see in partic- ular Sec. B 2, Eq. (B21). The flux of electrons of spin σ emerging from the left terminal can be presented in the form (cid:88) ∞(cid:88) IL,σ = 2NLNR P (n)Tnn(cid:48),σσ(cid:48) ×(cid:16) 1 − eβ(µL,σ−µR,σ(cid:48) )(cid:17) µL,σ − µR,σ(cid:48) + (n − n(cid:48))ω0 n,n(cid:48)=0 σ(cid:48) eβ[µL,σ−µ R,σ(cid:48) +(n(cid:48)−n)ω0] − 1 , (8) where NL(R) is the density of states at the common chem- ical potential of the left (right) lead, and T is the spin- dependent transmission33 Tnn(cid:48),σσ(cid:48) = W02(cid:104)n[e−iψR × e−iψL]σ(cid:48),σn(cid:48)(cid:105)2 . (9) Here, W0 is the transmission amplitude in the absence of the SO interaction. (The configuration in which the den- sities of states are spin dependent is discussed in Sec. III.) FIG. 2: A break junction supporting a nanowire of length d attached by tunnel contacts to two biased electrodes ([L] and [R]). The small vibrations of the wire induce oscillations in the angle θ around some value θ0. The upper electrode ([G]) is an STM tip biased differently. The Rashba interaction can be controlled via the bending angle θ of the wire. The latter can be modified both mechanically, by loads (shown by the arrows) applied to the substrate and electrically, by biasing the STM. Reprinted figure with permission from R. I. Shekhter et al., Phys. Rev. Lett. 111, 176602 (2013) c(cid:13)2013 by the American Physical Society. A suspended nanowire, acting as a weak link between two electronic reservoirs, is a good candidate for such a Rashba spin-splitter.14 The amount of spin splitting brought about by the Rashba interaction on the weak link can be controlled by bending the wire. This can be mechanically tuned, by exploiting a break junction as a substrate for the wire (see Fig. 2) or by electrically inducing a Coulomb interaction between the wire and an STM tip electrode (also displayed in Fig. 2). This Rashba scatterer is localized on the nanowire, and serves as a pointlike scatterer in momentum-spin space for the electrons incident from the bulky leads. When there is a spin imbalance population in one of the leads (or both), and the Rashba spin-splitter is activated (i.e., the weak link is open for electronic propagation) spin currents are generated and are injected from the pointlike scatterer into the leads. The Rashba splitter thus redistributes the spin populations between the leads. This source of spin currents need not be accompanied by transfer of elec- tronic charges. We emphasize that although this setup is similar in the latter aspect to the Datta-Das one,27 our splitter is functional even when the leads are unbiased. Such a coherent scatterer, whose scattering matrix can be "designed" at will by tuning controllably the geome- try, can be realized in electric weak links based on clean carbon nanotubes (CNT). Carbon nanotubes have a sig- nificant Rashba SO coupling (mainly due to the strain associated with the tube curvature).24,28,29 Moreover, CNT's are known to have quite long mean-free paths (longer for suspended tubes than for straight ones), al- lowing for experimental detections of interference-based phenomena (e.g., Fabry-Perot interference patterns).30 Further tunability of the Rashba spin-splitter can be achieved by switching on an external magnetic field, cou- pled to the wire through the Aharonov-Bohm effect.16 This is accomplished by quantum-coherent displacements of the wire, which generate a temperature dependence in GRLVGVLVRΘÄLRRRRLIn Eq. (8), the free vibrations are described by the Ein- stein Hamiltonian (A20) with frequency ω0, n is the vi- brations quantum number, and the weight function P (n) is given in Eq. (B19). For the geometry of Fig. 3, In the linear-response regime the spin current loses its dependence on the bias voltage (expressions for these cur- rents beyond linear response are given in Ref. 14) and becomes ψL = φL − kso(xLσy − yLσx) , ψR = φR − kso(xRσy + yRσx) . Jspin,↑ = −U Gspin , (13) (10) where it has been used that 5 µL(R),↑ = µL(R) + UL(R) 2 , µL(R),↓ = µL(R) − UL(R) , 2 (14) such that U = (UL + UR)/2, and the spin conductance Gspin is ∞(cid:88) ∞(cid:88) Gspin = G0 sin2(ksod) P (n) n=0 (cid:96)=1 × (cid:104)ne i πBd2 4Φ0 sin(2θ) cos θn + (cid:96)(cid:105)2 2(cid:96)βω0 eβ(cid:96)ω0 − 1 . (15) Here G0 is the zero-field electrical conductance divided by e2, and β = 1/(kBT ) is the inverse temperature. The amount of spin intensity is obtained upon expanding θ in the operators of the vibrations [see Eq. (A19)]. One then finds Gspin = sin2(ksod) cos2 θ0G(B) , (16) where G(B) is precisely the magnetoresistance (divided by e2) of the wire, as analyzed in Ref. 31, and thus has the same behavior at low and high temperatures (as com- pared to the vibrations' ω0). In particular, 1 − βω0 , exp[−B2/B2 B2 B2 0 6 0 ] , βω0 (cid:28) 1 , βω0 (cid:29) 1 , , (17) (cid:40) G(B) = W02 √ 2Φ0/[πda0 cos(θ0) cos(2θ0)] (a0 is the am- where B0 = plitude of the zero-point oscillations and Φ0 is the flux quantum). The flux of particles emerging from the right lead is ob- tained upon interchanging the roles of the left and the right side of the junction in Eq. (8). One notes14 that while the phase due to the magnetic field disappears in the absence of the vibrations, this is not so for the spin- orbit-phase18 (as ψL and ψR do not commute). Combining the expressions for the incoming spin cur- rents [Eq. (8) and the corresponding one for IR,σ] yields a net spin current, which is injected from the Rashba scatterer into the leads. Therefore, the scatterer can be viewed as a source of spin current maintained when the leads have imbalanced populations. The spin current, (cid:88) (cid:88) Jspin = Jspin,σ = (IL,σ + IR,σ) , (11) σ σ tends to diminish the spin imbalance in the leads, through spin-flip transitions induced by the Rashba in- teraction. In the limit of weak tunneling, we expect the spin imbalance to be kept constant in time by injecting spin-polarized electrons into the reservoirs, so that the (spin-dependent) chemical potentials do not vary. The explicit expressions for the two spin currents yield dramatic consequences. (i) Independent of the choice of the spin-quantization axis, Jspin,σ is given solely by the term with σ(cid:48) = σ in Eq. (8) and the corresponding one for IR,σ (σ is the spin projection opposite to σ). This im- plies that only the off diagonal amplitudes (in spin space) contribute. (ii) Adopting the plausible geometry detailed in Sec. A 3 [see the discussion preceding Eq. (A19)] one finds e−iψR e−iψL = e + iσy cos θ sin(ksod) − iσz sin(2θ) sin2(ksod/2) . (1 − 2 cos2 θ sin(ksod/2) i πBd2 4Φ0 sin(2θ) (12) III. SPIN-RESOLVED TRANSPORT This result is independent of the choice of the spin po- larizations in the leads, and does not involve σx. (iii) As Eq. (12) indicates, spin flips are realized for any ori- entation of the leads' polarization. Furthermore, when the average angle θ0 [see Eq. (A19)] differs from zero, then both terms on the second line in Eq. (12) yield spin flips even for the non-vibrating wire. In this respect, the spin-orbit splitting effect is very different from that of the Aharonov-Bohm phase. As mentioned, the latter re- quires the transport electrons to cover a finite area and therefore in our setup is entirely caused by the mechani- cal vibrations. When θ0 vanishes, there are spin flips only if the polarization is in the XZ plane. To be concrete, we present below explicit results for a quantization axis along z. The more general configuration is considered in Sec. III. The formalism presented in Appendix B enables us to study the case where the current through a mechanically- deformed weak link is provided by a battery of uncom- pensated electronic spins. When the magnetic polar- izations in the electronic reservoirs forming the elec- trodes are not identical, then quite generally both charge and spin currents result from the transport of electrons through the junction. The situation at hand resembles in a way thermoelectric transport in a two-terminal junc- tion: the two currents (charge and spin), flow in response to two affinities, the voltage difference and the difference in the amount of magnetic polarization between the two reservoirs. "Non-diagonal" phenomena, analogous to the thermoelectric Seebeck and Peltier effects, can therefore be expected. For instance, it is possible to generate a spin current by injecting charges into the material, which in turn may give rise to a spatially inhomogeneous spin accumulation. However, the two opposite spins can still contribute equally to the charge transport, resulting in zero net spin propagation, much like the vanishing of the thermopower when electron-hole symmetry is maintained. In the case of combined spin and charge transport, non-diagonal spin-electric effects appear once the spin and charge transports are coupled in a way that distinguishes be- tween the two spin projections. One may achieve such a spin-dependent transport by exploiting magnetic materi- als in which the electronic energy is spin-split. When the magnetization is spatially inhomogeneous (as happens in composite magnetic structures) the spin-dependent part of the energy is inhomogeneous as well, leading to a spin- dependent force acting on the charge carriers. Another possibility, feasible even in magnetically-homogeneous materials, is to exploit the Rashba SO interaction. When this interaction varies in space, the electronic spin is twisted. The end result is the same as in the first sce- nario above: a spin-dependent force (resulting from the Rashba interaction) is exerted on the electrons, opening the way for non-diagonal spintro-electric transport. FIG. 4: A curved nanowire lying in the XY plane is coupled to two magnetically-polarized electronic reservoirs with arbitrarily-oriented magnetization axes, nL and nR. Externally-pumped spins give rise to spin-dependent electro- chemical potentials. The bending of the nanowire is specified by the angle it makes with x, with an instantaneous value θ around the equilibrium angle θ0. Reprinted figure with per- mission from R. I. Shekhter et al., Phys. Rev. B 90, 045401 (2014) c(cid:13)2014 by the American Physical Society. The setup we propose is illustrated in Fig. 4. It com- prises a nanowire bridging two leads, firmly coupled to the left and right electronic reservoirs, held at spin- dependent electrochemical potentials, µL,σ = µL + σUL , µR,σ = µL + σUR , (18) [generalizing Eq. (14)]. The two bulk metals forming the reservoirs are each polarized along its own polarization axis, denoted by the unit vectors nL and nR, respectively. 6 The wire vibrates in the XY plane, such that the angle θ it makes with the x−axis oscillates around an equilibrium value, θ0. An additional (weak) magnetic field, applied along z, gives rise to an instantaneous Aharonov-Bohm effect.31 Since the electrodes are magnetically-polarized, the density of states in each of them depends on both the in- ternal exchange interaction and the external spin pump- ing as expressed by the energy split of the electrochemi- cal potentials UL,R that determine the kinetic energy of the electrons participating in the transport. However, as- suming the spin biases UL(R) to be much smaller than the Curie temperature in the magnetic leads, the latter de- pendence is weak, and to lowest order in UL,R/µ, where µ = (µL +µR)/2 is the common chemical potential of the entire device, it may be neglected. As shown in Appendix B 2 [see in particular Eq. (B21)], the spin-resolved particle currents emerging from the left and the right electrodes are [generalizing Eq. (8) to in- clude spin-dependent densities of states in the two bulky reservoirs] (cid:88) ∞(cid:88) − IL,σ = 2πNL,σ σ(cid:48) × (1 − eβ(µL,σ−µR,σ(cid:48) )) n,n(cid:48)=0 NR,σ(cid:48) P (n)Tnn(cid:48),σσ(cid:48) µL,σ − µR,σ(cid:48) + (n(cid:48) − n)ω0 eβ[µL,σ−µR,σ(cid:48) +(n(cid:48)−n)ω0] − 1 and − IR,σ(cid:48) = 2πNR,σ(cid:48) (cid:88) σ × (eβ(µL,σ−µR,σ(cid:48) ) − 1) ∞(cid:88) n,n(cid:48)=0 P (n)Tnn(cid:48),σσ(cid:48) NL,σ µL,σ − µR,σ(cid:48) + (n(cid:48) − n)ω0 eβ[µL,σ−µR,σ(cid:48) +(n(cid:48)−n)ω0] − 1 , (19) . (20) Particle number is conserved, as can be seen by adding together Eq. (19) summed over σ and Eq. (20) summed over σ(cid:48). The spin indices of the matrix element squared forming the transmission, T , in Eqs. (19) and (20) deserve some caution: the quantization axes of the magnetization in the two electronic reservoirs are generally different (see Fig. 4), and they both may differ from the quantization axis which is used to describe the Rashba interaction on the nanowire. Specifying the quantization axis in the left (right) reservoir by the angles θL (θR) and ϕL (ϕR), then Tnn(cid:48),σσ(cid:48) = W02(cid:104)n[S† (21) where the rotation transformations SL(R) are given by Re−iψR e−iψLSL]σ(cid:48)σn(cid:48)(cid:105)2 , (cid:34) SL(R) = ϕ (cid:35) e−i ei L(R) 2 cos ϕ L(R) 2 sin θL(R) 2 θL(R) 2 e−i −ei ϕ L(R) 2 ϕ L(R) 2 sin cos θL(R) 2 θL(R) 2 . (22) ΘΜR,ΣºΜR+ΣURΜL,ΣºΜL+ΣULn`Ln`Rx`z`y`x`y`z`For instance, when the quantization axes in both elec- trodes are identical, nL = nR, then SL = SR just rotates the direction of the quantization axis of the Rashba in- teraction. The linear-response regime. In the linear-response regime, the spin-resolved particle currents, Eqs. (19) and (20) become (cid:88) NR,σ(cid:48)(µL,σ − µR,σ(cid:48))Aσσ(cid:48) , IL,σ = 2πNL,σ (cid:88) σ(cid:48) NL,σ(µR,σ(cid:48) − µL,σ)Aσσ(cid:48) , IR,σ(cid:48) = 2πNR,σ(cid:48) ∞(cid:88) ∞(cid:88) P (n)Tnn,σσ(cid:48) Aσσ(cid:48) = n=0 σ (n(cid:48) − n)βω0 e(n(cid:48)−n)βω0 − 1 . P (n)Tnn(cid:48),σσ(cid:48) + nn(cid:48)=0 n(cid:54)=n(cid:48) with the transmission (23) (24) The first term in Eq. (24) gives the contribution to the spin-resolved transport from the elastic tunneling pro- cesses. The second is due to the inelastic processes, and is active at finite temperatures. Our final expressions for the charge currents are then eIL ≡ e IL,σ = e(µL − µR)C1 − eURC3 + eULC2 , (25) σ(cid:48) IR,σ(cid:48) = −eIL. The spin currents emerging from the left and right reservoirs are σ (cid:88) with eIR ≡ e(cid:80) L ≡(cid:88) (cid:88) I spin σ I spin R = σ(cid:48) σIL,σ = (µL − µR)C2 − URC4 + ULC1 , σ(cid:48)IR,σ(cid:48) = (µR − µL)C3 + URC1 − ULC4 . (26) In Eqs. response transport coefficients (25) and (26) we have introduced the linear- σσ(cid:48) (cid:88) (cid:88) (cid:88) (cid:88) σσ(cid:48) σσ(cid:48) σσ(cid:48) C1 = 2π C2 = 2π C3 = 2π C4 = 2π NL,σAσσ(cid:48)NR,σ(cid:48) , NL,σσAσσ(cid:48)NR,σ(cid:48) , NL,σAσσ(cid:48)σ(cid:48)NR,σ(cid:48) , NL,σσAσσ(cid:48)σ(cid:48)NR,σ(cid:48) , (27) giving the various transmission probabilities of the junction.33 The Onsager relations. As mentioned, there is a cer- tain analogy between the configuration studied here and that of thermoelectric transport. In order to further pur- sue this point we consider the entropy production in our 7 device, assuming that the spin imbalance in each of the two reservoirs does not vary with time and that all parts of the setup are held at the same temperature T . Under these circumstances the entropy production, S, is T S = (cid:88) (cid:88) = IL(µL − µR) + ULI spin µL,σIL,σ + σ(cid:48) σ µR,σ(cid:48)IR,σ(cid:48) L + URI spin R , (28) where the various currents are given in Eqs. (25) and (26). Obviously, the first term on the right-hand side of Eq. (28) is the dissipation due to Joule heating. The other two terms describe the dissipation involved with the spin currents. The entropy production may be presented as a scalar product of the vector of driving forces (the "affinities"), {V ≡ (µL − µR)/e, UL, UR} and the resulting currents, {eIL, I spin In the linear-response regime these two vectors are related to one another by a (3×3) matrix M, R }. L , I spin  V UL UR I spin L I spin R  = M  eIL  e2C1 eC2 −eC3 −C4 eC2 −eC3 C1 −C4 C1   . M = (29) (30) with The matrix M contains the transport coefficients which do not depend on the driving forces. One notes that this matrix obeys the Onsager reciprocity relations: reversing the sign of the magnetic field, i.e., inverting the sign of the Aharonov-Bohm phases φL and φR [see Eqs. (10) and (21)], and simultaneously interchanging the vibra- tion states indices n with n(cid:48) and the spin indices σ with σ(cid:48) in Eqs. (21) and (24) leaves all off diagonal terms in the matrix M unchanged. The transport coefficients. The full calculation of the transmission matrix A that determines the transport co- efficients Ci [see Eqs. (24) and (27)] is quite complicated, and requires a numerical computation. When the cou- pling of the charge carriers to the vibrational modes of the wire is weak, one may obtain an approximate ex- pression by exploiting the different magnitudes that cou- pling takes in the magnetic Aharonov-Bohm phase and in the Rashba one. In order to see this, it is expedient to present the phase factors in the transmission amplitude [see Eq. (10)] in the form exp(−iψR) exp(−iψL) ≡ e−iφ(A + iV · σ) , (31) where A and V are functions of the instantaneous bend- ing angle θ, Eq. (A19), A = 1 − 2 cos2(θ) sin2(ksod/2) , V = {0, cos(θ) sin(ksod),− sin(2θ) sin2(ksod/2)} , A2 + V · V = 1 , (32) and φ = φL + φR is the instantaneous Aharonov-Bohm flux in units of the flux quantum Φ0. The components of the spin-orbit vector V are given in the coordinate axes depicted in Fig. 4. The effect of the electron-vibration interaction on the Rashba coupling is of the order of the zero-point ampli- tude of the vibrations divided by the wire length, a0/d. On the other hand, using Eq. (A19), one finds that the Aharonov-Bohm phase, φ = −[πBd2/(4Φ0)] sin(2θ) (B is the strength of the magnetic field), is φ ≈ − πBd2 4Φ0 sin(2θ0) − πa0dB 2Φ0 cos(θ0) cos(2θ0)(b + b†) . (33) (The creation and destruction operators of the vibrations are denoted b† and b.) The dynamics of the Aharonov- Bohm flux is thus determined by the flux enclosed in an area of order a0d divided by the flux quantum. The latter ratio can be significantly larger than a0/d. For instance, the length of a single-walled carbon nanotube is about d = 1 µm, while the vibrations' zero-point amplitude is estimated to be 10−5 µm. This leads to a0/d ≈ 10−5, while (Ba0d)/Φ0 is of the order of 10−2 for magnetic fields of the order of a few Teslas (at which the effect of the magnetic field on the transport through the Rashba weak link becomes visible). The disparity between the way the electron-vibration coupling affects the Rashba phase factor and the manner by which it dominates the magnetic one results in a con- venient (approximate) form for the transmission matrix A,14,33 (cid:21) (cid:20) Ad And And Ad A = G(T, B) . (34) The conductance G(T, B) (divided by e2) derived in Ref. 31 [see Eqs. (15) and (16) for the definition, and Eq. (17) for the limiting behaviors], gives the transmis- sion of the junction in the absence of the Rashba interac- tion; it depends on the temperature and on the perpen- dicular magnetic field. The spin-dependent part of the transmission is given by the matrix in Eq. (34), Ad + And = 1 , Ad − And = (A2 0 − V 2 0 )nL · nR + 2A0V0 · nL × nR + 2(V0 · nL)(V0 · nR) . (35) Here A0 and V0 are given by the values of A and V de- fined in Eqs. (32) at equilibrium, i.e., when the angle θ there is replaced by θ0. Their physical meaning is ex- plained below: And = sin2(γ), where γ is the twisting angle of the charge carriers' spins, and Ad = cos2(γ). Using the explicit expression (34) for the transmission matrix A it is straightforward to find the transport coef- ficients Ci. Retaining only terms linear in the difference 8 between the densities of states of the two spin orienta- tions, we obtain C1 + C4 ≈ 8πG(T, B)AdNLNR ≈ C2 + C3 , C1 − C4 ≈ 8πG(T, B)AndNLNR , C2 − C3 = 4πG(T, B)And(NL,↑NR,↓ − NL,↓NR,↑) , (36) where NL,R is the total density of states of each electronic reservoir (summed over the two spin directions). Glanc- ing at Eq. (25) for the charge current, and taking into account the first of Eqs. (35), shows that the conduc- tance, G, of the junction is independent of the spin-orbit interaction, and is given by G = 4πe2NLNRG(T, B) . (37) Rashba twisting. When the junction is not subjected to a perpendicular magnetic field and the charge car- riers passing through it do not collect an Aharonov- Bohm phase due to it, one may safely ignore the ef- fect of the quantum flexural nano-vibrations of the sus- pended wire.14 The scattering of the electrons' momen- tum, caused by the spatial constraint of their orbital mo- tion inside the nanowire, also induces scattering of the electronic spins. The latter results from the SO Rashba interaction located at the wire. Consequently, an elec- tronic wave having a definite spin projection on the mag- netization vector of the lead from which it emerges, is not a spin eigenstate in the other lead. Thus, a pure spin state σ(cid:105) in one lead becomes a mixed spin state in the other, σ(cid:105) ⇒ α1σ(cid:105) + α2σ(cid:105) , (38) with probability amplitude α1 to remain in the original state, and probability amplitude α2 for a spin flip (σ = −σ). During the propagation through the weak link the spins of the charge carriers are twisted, as is described by the transmission amplitude [see Eq. (31)], A0 + iV0 · σ. It follows that the probability amplitude for a spin flip, α2, is given by α2 = [S† R(A0 + iV0 · σ)SL]σσ , (39) with SL,R given in Eq. (22). The Rashba twisting angle, γ, can now be defined by α2 = sin(γ)eiδ , (40) with α22 = sin2(γ) = And , (41) yielding a clear physical meaning to the transmissions Ad and And [see Eqs. (35)]. The physical quantities depend only on the relative phase between α1 and α2. Therefore, we choose α1 = cos γ. It is then easy to check that the average of the vector σ in the state of Eq. (38) is equal to {sin(2γ) cos(δ), sin(2γ) sin(δ), cos(2γ)}. This vector is rotated by the angle 2γ relative to its direction in the absence of the spin-orbit interaction. This rotation of the electronic moments in each of the two leads is a "twist" of the spins. It is distinct from simple spin precession since the axis of this precession changes its direction during the electronic motion along the curved trajectory. In the simplest configuration of parallel magnetizations in both electrodes, i.e., nL = nR ≡ n , Eqs. (35) yield sin(γ) = [V 2 0 − (n · V0)2]1/2 . (42) (43) Interestingly enough, in this simple configuration sin(γ) is determined by the component of the Rashba vector V0 normal to the quantization axis of the magnetization in the electrodes. Mechanically manipulating the bending angle that determines the direction of the Rashba vector V0, one may control the twisting angle γ. Note also that had the vectors nL and nR been antiparallel to one another then sin(γ) = [1 − V 2 0 + (n · V0)2]1/2. An even more convenient way to monitor the twisting effect may be realized by studying the spintro-voltaic ef- fect in an open circuit, i.e., when the total charge current vanishes. One then finds that the spin-imbalanced pop- ulations in the electrodes give rise to an electric voltage, Vsv. Assuming that the spin imbalances in the two reser- voirs are identical, i.e., UL = UR ≡ U , Eq. (25) yields Vsv = U . (44) C3 − C2 C1 The ratio of the voltage created by the spin imbalance, Vsv, to the amount of spin imbalance in the electrodes (expressed by U ) can be found upon using Eqs. (36), in conjunction with Eqs. (35) and (41), Vsv = sin2(γ) NL↑NR↓ − NL↓NR↑ NLNR U . (45) The voltage generated by the Rashba interaction gives directly the twisting angle; the proportionality between Vsv/U and sin2(γ) is the magnetic mismatch parameter of the junction. The twisting angle γ determines also the various spin conductances of the junction. From Eqs. (29) and (30) we find34 (cid:16) (cid:17) R = eV (C2 − C3) + (UL + UR)(C1 − C4) . NL,↑NR,↓ − NL,↑NR,↓ UL + UR + eV L + I spin I spin e2And = 2 G 9 (cid:12)(cid:12)(cid:12)UL=UR=0 and the cross spin conductance, Gspin× , (again normalized by the charge conductance), L + I spin I spin R eV G/e2 Gspin× = (48) are determined by And, that is by the twisting angle γ, Eq. (41) (the second requires the asymmetry in the spin- resolved densities of states). , For parallel magnetizations in the leads, the twisting angle [see Eq. (43)] depends solely on the SO coupling and on the equilibrium value of the bending angle. The spin twisting disappears for any direction of the polar- izations in the leads at θ0 = π/2. This can be easily un- derstood within a classical picture for the spin rotation caused by the Rashba interaction. The spin evolution of the tunneling electron can be regarded as a rotation around an axis given by the vectorial product of the ve- locity and the electric field (directed along z in our con- figuration). At this value of θ0 the tunneling trajectory is oriented along y (because then xR = xL = 0) and so the electron "rattles" back and forth along y. This leads to a cancellation of the Rashba contribution to the tunnel- ing phase [see Eq. (10)]. The other special case is when the wire is not bended, i.e., θ0 = 0. The spin twisting for leads' magnetizations along y vanishes, while for de- vices with ferromagnetic magnetizations along the x− or z−directions it reaches its maximal value, sin(ksod). The reason for this has also to do with the orientation of the spin rotation-axis. At small values of θ0 the electronic trajectory is primarily along x. Then, when the spin of the incident electron is directed along y it is parallel to the rotation axis and no rotation is taking place. In con- trast, when the spin of the incident electron is oriented along x or z, it is perpendicular to the rotation axis, leading to a full rotation. Thus, one can have spintro-electric functionalities if one uses a vibrating suspended weak link, with both a magnetic flux and an (electric field dependent) Rashba spin-orbit interaction. The twisting of the electronic spins as they move between the (spin-polarized) elec- trodes can be manipulated by the bias voltage, the bend- ing of the weak link wire, and the polarizations in the electrodes. The twisting angle, which determines the probability amplitude of the Rashba splitting, can be measured electrically through a spintro-voltaic effect. 2NLNR , IV. SPIN POLARIZATION OF COOPER PAIRS IN SPIN-ORBIT-ACTIVE SUPERCONDUCTING (46) WEAK LINKS where G is the charge conductance, Eq. (37), and we have made use of Eqs. (36) for the C's. One now observes that both the spin conductance, Gspin (normalized by the charge conductance) Gspin = I spin L + I spin R (UL + UR)G/e2 , (47) (cid:12)(cid:12)(cid:12)V =0 A remarkable consequence of superconductivity, known as the proximity effect, allows a supercurrent to flow between two superconductors connected by a non- superconducting material of a finite width. This phe- nomenon was actively studied during several decades starting with the pioneering prediction by Josephson19 in the early 1960's that a non-dissipative current may flow through a tunnel junction formed by a layered superconductor-insulator-superconductor (S-I-S) struc- ture. A number of other so-called superconducting weak links, involving a normal metal, a quantum dot and var- ious micro-constrictions, have been studied theoretically and experimentally during the past decades in order to explore and exploit this phenomenon. A phase transition to a superconducting state is accom- panied by the formation of a new ground state, which is different from the standard metal ground state and which can be viewed as a condensate of paired electrons (Cooper pairs). The pairing is supported by an indirect attractive inter-electronic interaction usually described in terms of a "pairing potential". In inhomogeneous superconductors this potential is coordinate dependent and can, in partic- ular, be suppressed in a layer perpendicular to the direc- tion of the superconducting current. In this case electron pairs that carry the supercurrent are injected into a re- gion where pairing is no longer supported by a pairing potential. However, the coherent properties of the elec- trons established inside the superconducting injector can be preserved for a certain distance (the superconducting coherence length), which allows a non-dissipative current to flow through a non-superconducting layer of a suffi- ciently small width. Nevertheless, the spatial segregation of the Cooper pairs from the pairing potential responsible for their stability suggests a unique way for manipulat- ing the Cooper pairs during their propagation through a superconducting weak link. Consider, for example, the well-known fact that elec- trons, which form a Cooper pair in a conventional (sin- glet BCS) superconductor, are in time-reversed quantum states and therefore their spins are aligned in opposite di- rections so that the pair as a whole carries no spin. This spin ordering can be distorted inside the weak link, which allows for an intentional "spin design" to be achieved by means of a superconducting weak link. In this Sec- tion we describe a particular mechanism for this kind of spin design, viz. an SO interaction localized to the non- superconducting weak link as presented in Ref. 10 We show that the splitting of the spin state of the paired electrons that carry the Josephson current may transform the spin-singlet Cooper pairs into a coherent mixture of singlet and triplet spin states. This mixture gives rise to interference between the channel in which both electrons preserve their spins and the channel where they are flipped. The resulting interference pattern, that appears in the Josephson current but does not show up in the normal-state transmission of the junction, allows for electrical and mechanical control of the Josephson current between two spin-singlet superconductors; it cor- responds to a new type of "spin-gating"35 of supercon- ducting "weak links". To illustrate our calculation, Fig. 5 uses a semiclassi- cal analogue of the quantum evolution of the spin states of electrons which move between two bulk leads via a weak link, where they are subjected to the Rashba SO interaction. For simplicity we assume for now that the 10 FIG. 5: Schematic illustrations of the lowest-order perturba- tion expansion steps for tunneling (in the Coulomb blockade regime) through a straight nanowire weak link subjected to the Rashba spin-orbit (SO) interaction caused by an electric field along z. In a semiclassical picture, the spin of each elec- tron (denoted by an arrow) is rotated in the XZ−plane as it goes through the link. (a) Single electron tunneling from one normal metal to another, via an intermediate (rotating) state (dashed circle). When the electron enters the second normal metal, its spin has been rotated. (b) Sequential tunneling in four steps of a Cooper pair between two superconductors connected by the same weak link. Because the two electrons that form the Cooper pair are in time-reversed states, the SO interaction rotates their spins in opposite directions. (c) As they enter the second superconductor, the Cooper pairs are in a coherent mixture (dash-dotted circle) of a spin-singlet and a spin-triplet state. Inside this superconductor, this state is then projected onto the singlet state (full circle). Reprinted figure with permission from R. I. Shekhter et al., Phys. Rev. Lett. 116, 217001 (2016) c(cid:13)2016 by the American Physical Society. weak link is a straight 1D wire along the x−axis. The SO interaction in the wire is due to an electric field, which for the moment is assumed to point along z and there- fore corresponds to an effective SO-interaction-induced magnetic field directed along y. Figure 5(a) illustrates a single-electron transfer from one normal metal to an- other. Without loss of generality, we choose the z−axis in spin space to be along the direction of the polarization of the electron in the first (left) metal. Semiclassically, the spin of the injected electron rotates in the XZ−plane as it passes through the wire. As a result, the spins of the electrons that enter the second metal from the wire are rotated around the y−axis by an angle proportional to the strength of the SO interaction and the length of the wire. This rotation depends on the direction of the "initial" electron's polarization. It occurs only if the po- larization has a component in the XZ−plane. Quantum (a) (b) N N S S 1 2 3 4 (c) + x y z ! ⁄ mechanically, the electron's spinor in the left metal is an eigenfunction of the Pauli spin-matrix σz, and the spinor of the outgoing electron is in a coherent superposition of spin-up and spin-down eigenstates of σz. How can this picture be generalized to describe the transfer of the two electrons of a Cooper pair between two bulk superconductors? The simplest case to con- sider, which we focus upon below, is when the single- electron tunneling is Coulomb-blockaded throughout the wire. While the blockade can be lifted for one electron, double electron occupancy of the wire is suppressed, i.e., a Cooper pair is mainly transferred sequentially, as shown in Fig. 5(b). Each electron transfer is now accompanied by the spin rotation shown in Fig. 5(a). However, since the two transferred electrons are in time-reversed quan- tum states, the time evolution of their spins are reversed with respect to one another, and their rotation angles have opposite signs [step 4 in Fig. 5(b)]. This final state [Fig. 5(c)] can be expressed as a coherent mixture of a spin-singlet and a spin-triplet state, but only the former can enter into the second superconductor. As we show below, this projection onto the singlet causes a reduction of the Josephson current. We consider a model where a Cooper pair is transferred between superconducting source and drain leads via vir- tual states localized in a weak-link wire [see Fig. 6(a)]. The corresponding tunneling process, which supports multiple tunneling channels, was analyzed in detail in Ref. 36. For simplicity, it is assumed throughout this section that the angle θ remains fixed, that is, the wire does not vibrate. A significant simplification occurs in the Coulomb-blockade regime, defined by the inequality Ee = EC(N + 1) − EC(N ) (cid:29) ∆, where ∆ is the en- ergy gap parameter in the superconducting leads,37 and EC(N ) is the Coulomb energy of the wire when it con- tains N electrons. In this regime, tunneling channels requiring two electrons to be simultaneously localized in a virtual state in the wire can be neglected, and hence the tunneling processes are sequential. Another simpli- fication follows from our assumption that the length of the wire d is short compared to the superconducting co- herence length ξ0 ≡ vF/∆,37 so that the dependence of the matrix element for a single electron transfer on the electron energy in the virtual states can be ignored. A final simplification, facilitated by the device geome- try, concerns the conservation of the electrons' longitu- dinal momenta as they tunnel between the two leads. In Fig. 6, the wire ends are placed on top of the metal leads and are separated from them by thin but long tunneling barriers. Since the direction of tunneling is nearly per- pendicular to the direction of the current along the wire, such a geometry is conducive to longitudinal momentum conservation.38 These simple but realistic assumptions allow us to de- scribe the transfer of a Cooper pair between the two su- perconductors in terms of single-electron tunneling, as given by the Hamiltonian (B6) [see also Eqs. (B1)-(B5)], with the tunneling matrix elements derived in Appendix 11 FIG. 6: (color online) Sketch (a) and simplified model (b) of a device that would allow the effects predicted in the text to be studied. Reprinted figure with permission from R. I. Shekhter et al., Phys. Rev. Lett. 116, 217001 (2016) c(cid:13)2016 by the American Physical Society. A. We assume a weak link containing a bent wire [see Fig. 6(a)]. The actual calculations are done for the ge- ometry shown in Fig. 6(b), where the weak link comprises two straight one-dimensional wires, RL and RR, of equal length d/2, connected by a "bend". The angles between these wires and the x−axis are θ and −θ, respectively. In the absence of the SO interaction, the supercurrent scales as the transmission of the junction when in the nor- mal state.39,40 As detailed in Appendix B, the SO cou- pling modifies this transmission by the factor Tr{WW†}, where W = e−iksoσ·RL×ne−iksoσ·RR×n, (49) and the trace is carried out in spin space. When kso van- ishes, this factor is simply 2, the spin degeneracy; i.e., the SO interaction does not affect the electric conductance (unless the junction allows for geometrically-interfering processes18). The superconducting Josephson current is (cid:2)Wσσ2 − Wσ ¯σ2(cid:3) = (cid:88) σ (cid:20) 1 (cid:88) 2 σ (cid:21) − Wσ¯σ2 , J(ϕ) J0(ϕ) = 1 2 (50) where J0(ϕ) ∝ sin(ϕ) is the equilibrium Josephson cur- rent in the absence of the SO interaction,39 and ϕ is the superconducting phase difference. Hence, the SO interaction modifies significantly the amplitude of the Josephson equilibrium current, while leaving the transmission of the junction in its normal state as in the absence of this coupling. The matrix W, that determines these quantities, depends crucially on the direction n of the electric field [see Eq. (3)]. In the configuration where n is normal to the plane of the junc- tion, which is described semiclassically in Fig. 5, n (cid:107) z, and then W =(cid:2) cos2(ksod/2) − sin2(ksod/2) cos(2θ)(cid:3) + iσ ·(cid:2)y sin(ksod) cos(θ) + z sin2(ksod/2) sin(2θ)(cid:3) . (51) SuspendedNanowiresasMechanicallyControlledRashbaSpinSplittersR.I.Shekhter,1O.Entin-Wohlman,2,3,andA.Aharony2,31DepartmentofPhysics,GoteborgUniversity,SE-41296Goteborg,Sweden2RaymondandBeverlySacklerSchoolofPhysicsandAstronomy,TelAvivUniversity,TelAviv69978,Israel3PhysicsDepartment,BenGurionUniversity,BeerSheva84105,Israel(Received26June2013;published23October2013)Suspendednanowiresareshowntoprovidemechanicallycontrolledcoherentmixingorsplittingofthespinstatesoftransmittedelectrons,causedbytheRashbaspin-orbitinteraction.Thesensitivityofthelattertomechanicalbendingmakesthewireatunablenanoelectromechanicalweaklinkbetweenreservoirs.Whenthereservoirsarepopulatedwithmisbalanced''spin-upandspin-down''electrons,thewirebecomesasourceofsplitspincurrents,whicharenotassociatedwithelectricchargetransferandwhichdonotdependontemperatureordrivingvoltages.Themechanicalvibrationsofthebendedwiresallowforadditionaltunabilityofthesesplittersbyapplyingamagneticfieldandvaryingthetemperature.Cleanmetalliccarbonnanotubesofafewmicronslengtharegoodcandidatesforgeneratingspinconductanceofthesameorderasthechargeconductance(dividedbye2)whichwouldhavebeeninducedbyelectricdrivingvoltages.DOI:10.1103/PhysRevLett.111.176602PACSnumbers:72.25.Hg,72.25.RbIntroduction. -- Thelackofscreeningandthewavynatureoftheelectronstogetherwiththeensuinginterfer-enceeffectsdeterminealargevarietyofCoulomb-correlationandquantum-coherencephenomenainquantumwiresanddots.Theelectronicspin,beingweaklycoupledtootherdegreesoffreedominbulkmaterials,becomesan''activeplayer''duetotheenhancedspin-orbitinteractioninducedbytheRashbaeffect[1]intheselow-dimensionalstructures[2,3].Thisinteractioncanbealsomodifiedexperimentally[4 -- 6].Thequantum-coherencecontrolofspin-relateddevicesandthespatialtransferoftheelectronspinsareamongthemostchallengingtasksofcurrentspintronics,astheycanbringupnewfunctional-ities.Thus,e.g.,quantuminterferenceofelectronicwavesinmultiplyconnecteddeviceswaspredictedtobesensitivetotheelectronicspin,leadingtospinfilteringinelectronictransport[7].Inchargetransport,electronicbeamsplitters(e.g.,bytunnelbarriers)arekeyingredientsininterference-baseddevices.InthisLetterweproposethattunnel-barrierscat-terersmayserveascoherentsplittersoftheelectronicspinwhenthetunnelingelectronsalsoundergospin(Rashba)scattering.Thisallowsustomapvariousinterferencebasedphenomenainchargetransportontoelectronicspintransportation.Suchspinsplitterscanbereadilymadefunctionalbyaddingtothemamechanicaldegreeoffreedom,whichservestocontroltheirgeometricalconfigurationinspace,towhichtheRashbainteractionisquitesensitive.Becauseofthis,oneachievesmechanicalcoherentcontrolandmechanicaltuningofthespinfilters[8].Wesuggestthatasuspendednanowire,actingasaweaklinkbetweentwoelectronicreservoirs,isagoodcandidateforsuchaRashbaspinsplitter(seeFig.1).Theamountofspinsplitting,broughtaboutbytheRashbainteractiononthewire,isdeterminedbythespin-orbitcouplingandassuchcanbecontrolledbybendingthewire.Thiscanbemechanicallytuned,byexploitingabreakjunctionasasubstrateforthewire(seeFig.1)orbyelectricallyinduc-ingaCoulombinteractionbetweenthewireandanSTMtipelectrode(alsodisplayedinFig.1).ThisRashbascat-tererislocalizedonthenanowire,andservesasapointlikescattererinmomentum-spinspacefortheelectronsinci-dentfromthebulkyleads.Whenthereisaspinimbalancepopulationinoneoftheleads(orboth),andtheRashbaspinsplitterisactivated(i.e.,theweaklinkisopenforelectronicpropagation)spincurrentsaregeneratedandareinjectedfromthepointlikescatterertotheleads.ThustheGRLVGVLVRFIG.1(coloronline).Abreakjunctionsupportingananowireoflengthd(possiblyacarbonnanotube),attachedbytunnelcontactstotwobiasedelectrodes([L]and[R]).Thesmallvibrationsofthewireinduceoscillationsintheanglearoundsomevalue0.Theupperelectrode([G])isanSTMtipbiaseddifferently.TheRashbainteractioncanbecontrolledviathebendingangleofthewire.Thelattercanbemodifiedbothmechanically,byloads(shownbythearrows)appliedtothesubstrateandelectrically,bybiasingtheSTM.PRL111,176602(2013)PHYSICALREVIEWLETTERSweekending25OCTOBER20130031-9007=13=111(17)=176602(5)176602-1Ó2013AmericanPhysicalSocietySuspendedNanowiresasMechanicallyControlledRashbaSpinSplittersR.I.Shekhter,1O.Entin-Wohlman,2,3,andA.Aharony2,31DepartmentofPhysics,GoteborgUniversity,SE-41296Goteborg,Sweden2RaymondandBeverlySacklerSchoolofPhysicsandAstronomy,TelAvivUniversity,TelAviv69978,Israel3PhysicsDepartment,BenGurionUniversity,BeerSheva84105,Israel(Received26June2013;published23October2013)Suspendednanowiresareshowntoprovidemechanicallycontrolledcoherentmixingorsplittingofthespinstatesoftransmittedelectrons,causedbytheRashbaspin-orbitinteraction.Thesensitivityofthelattertomechanicalbendingmakesthewireatunablenanoelectromechanicalweaklinkbetweenreservoirs.Whenthereservoirsarepopulatedwithmisbalanced''spin-upandspin-down''electrons,thewirebecomesasourceofsplitspincurrents,whicharenotassociatedwithelectricchargetransferandwhichdonotdependontemperatureordrivingvoltages.Themechanicalvibrationsofthebendedwiresallowforadditionaltunabilityofthesesplittersbyapplyingamagneticfieldandvaryingthetemperature.Cleanmetalliccarbonnanotubesofafewmicronslengtharegoodcandidatesforgeneratingspinconductanceofthesameorderasthechargeconductance(dividedbye2)whichwouldhavebeeninducedbyelectricdrivingvoltages.DOI:10.1103/PhysRevLett.111.176602PACSnumbers:72.25.Hg,72.25.RbIntroduction. -- Thelackofscreeningandthewavynatureoftheelectronstogetherwiththeensuinginterfer-enceeffectsdeterminealargevarietyofCoulomb-correlationandquantum-coherencephenomenainquantumwiresanddots.Theelectronicspin,beingweaklycoupledtootherdegreesoffreedominbulkmaterials,becomesan''activeplayer''duetotheenhancedspin-orbitinteractioninducedbytheRashbaeffect[1]intheselow-dimensionalstructures[2,3].Thisinteractioncanbealsomodifiedexperimentally[4 -- 6].Thequantum-coherencecontrolofspin-relateddevicesandthespatialtransferoftheelectronspinsareamongthemostchallengingtasksofcurrentspintronics,astheycanbringupnewfunctional-ities.Thus,e.g.,quantuminterferenceofelectronicwavesinmultiplyconnecteddeviceswaspredictedtobesensitivetotheelectronicspin,leadingtospinfilteringinelectronictransport[7].Inchargetransport,electronicbeamsplitters(e.g.,bytunnelbarriers)arekeyingredientsininterference-baseddevices.InthisLetterweproposethattunnel-barrierscat-terersmayserveascoherentsplittersoftheelectronicspinwhenthetunnelingelectronsalsoundergospin(Rashba)scattering.Thisallowsustomapvariousinterferencebasedphenomenainchargetransportontoelectronicspintransportation.Suchspinsplitterscanbereadilymadefunctionalbyaddingtothemamechanicaldegreeoffreedom,whichservestocontroltheirgeometricalconfigurationinspace,towhichtheRashbainteractionisquitesensitive.Becauseofthis,oneachievesmechanicalcoherentcontrolandmechanicaltuningofthespinfilters[8].Wesuggestthatasuspendednanowire,actingasaweaklinkbetweentwoelectronicreservoirs,isagoodcandidateforsuchaRashbaspinsplitter(seeFig.1).Theamountofspinsplitting,broughtaboutbytheRashbainteractiononthewire,isdeterminedbythespin-orbitcouplingandassuchcanbecontrolledbybendingthewire.Thiscanbemechanicallytuned,byexploitingabreakjunctionasasubstrateforthewire(seeFig.1)orbyelectricallyinduc-ingaCoulombinteractionbetweenthewireandanSTMtipelectrode(alsodisplayedinFig.1).ThisRashbascat-tererislocalizedonthenanowire,andservesasapointlikescattererinmomentum-spinspacefortheelectronsinci-dentfromthebulkyleads.Whenthereisaspinimbalancepopulationinoneoftheleads(orboth),andtheRashbaspinsplitterisactivated(i.e.,theweaklinkisopenforelectronicpropagation)spincurrentsaregeneratedandareinjectedfromthepointlikescatterertotheleads.ThustheGRLVGVLVRFIG.1(coloronline).Abreakjunctionsupportingananowireoflengthd(possiblyacarbonnanotube),attachedbytunnelcontactstotwobiasedelectrodes([L]and[R]).Thesmallvibrationsofthewireinduceoscillationsintheanglearoundsomevalue0.Theupperelectrode([G])isanSTMtipbiaseddifferently.TheRashbainteractioncanbecontrolledviathebendingangleofthewire.Thelattercanbemodifiedbothmechanically,byloads(shownbythearrows)appliedtothesubstrateandelectrically,bybiasingtheSTM.PRL111,176602(2013)PHYSICALREVIEWLETTERSweekending25OCTOBER20130031-9007=13=111(17)=176602(5)176602-1Ó2013AmericanPhysicalSocietyS S rL rR x y z μ μ (a) (b) RL RR 12 link. The consequence is a spin splitting of the Cooper pairs that reach the second superconducting lead, where their spin state is projected onto the singlet state. This splitting can result in a Josephson current that is an oscil- latory function of the "action" ksod of the SO interaction (Fig. 7); the effect may be absent for special directions of the electric field. Both results can be understood in terms of a semiclassical picture, Fig. 5. As seen in Eq. (50), the Josephson current can be writ- ten as a sum of two contributions. One, Wσσ2, comes from a channel where the spin projections of the Cooper pair electrons, when leaving and entering the weak link, are identical; the other, Wσ ¯σ2, arises from another chan- nel, where the electron spins are flipped during the pas- sage. It is remarkable that the two contributions have opposite signs. This is due to a Josephson tunneling "π-shift" caused by electronic spin flips (and is similar to the effect predicted for tunneling through a Kondo impurity41). In particular, a total cancellation of the Josephson current is possible when, e.g., θ = 0 and ksod = π/4; in the limit θ = 0 and ksod = π/2 the Josephson current even changes its sign. This spin-orbit induced interference effect on the Josephson current is specific to a weak link subjected to SO interaction be- tween superconductors. There is no such effect on the current through a single weak link connecting two nor- mal metals. According to Eq. (50), none or both of the Cooper pair electrons must have flipped their spins as they leave the weak link in order to contribute to the Josephson cur- rent. This is because only spin-singlet Cooper pairs can enter the receiving s-type bulk superconductor. However, single-flip processes, where only one of the two tunneling electrons flips its spin, are also possible results of inject- ing Cooper pairs into a Rashba weak link. Those pro- cesses correspond to a triplet component of the spin state of the transferred pair, and can be viewed as evidence for spin polarization of injected Cooper pairs. The triplet component could be responsible for a spin-triplet proxim- ity effect,42 and would presumably contribute a spin su- percurrent if higher-order tunneling processes were taken into account. Thus, the supercurrent can be tuned by mechanical and electrical manipulations of the spin polarization of the Cooper pairs. In particular, the Josephson current through an electrostatically-gated device becomes an os- cillatory function of the gate voltage. We emphasize that these results follow from the interference of two trans- mission channels, one where the spins of both members of a Cooper pair are preserved and one where they are both flipped, and that this interference does not require any external magnetic field. It is important, however, that those parts of the device where the superconducting pairing potential is non-zero and where the SO coupling is finite are spatially separated. To lowest order in the tunneling this separation prevents the superconductivity in the leads to have any effect on the dynamical spin evolution in the wire. FIG. 7: (color online) The Josephson current J(ϕ) divided by its value without the SO interaction, J0(ϕ), for the SO interaction Eq. (3), as a function of ksod/(2π). The largest amplitude is for zero bending angle, θ = 0, decreasing gradu- ally for θ = π/6, π/5, π/4, π/3, and π/2.5. Relevant values of kso are estimated in Sec. V. Reprinted figure with permis- sion from R. I. Shekhter et al., Phys. Rev. Lett. 116, 217001 (2016) c(cid:13)2016 by the American Physical Society. In contrast, when the electric field is in the plane of the junction, e.g., n = y, we find W = cos[ksod cos(θ)] − iσ · z sin[ksod cos(θ)] . (52) When the SO interaction is given by Eq. (7), one finds that W of the strain-induced case has the same form as Eq. (51), except that y is replaced by x. The resulting expressions for the Josephson current and for the normal- state transmission turn out to be the same as for the SO interaction of Eq. (3), with n (cid:107) z. The matrix element Wσ¯σ depends on the quantization axis of the spins. Choosing this axis to be along z, then when n (cid:107) z Eq. (51) yields (cf. Fig. 7) 1 − 2Wσ ¯σ2 = 1 − 2 cos2(θ) sin2(ksod) . (53) In contrast, when the electric field is in the plane of the junction, n = y, the matrix W is diagonal [Eq. (52)], J(ϕ) = J0(ϕ), and the superconducting current is not affected by the spin dynamics. Similar qualitative results are found for all the directions of the spin quantization axis. For example, for spins polarized along y, one finds Wσ ¯σ2 = sin4(ksod/2) sin2(2θ) when n (cid:107) z, while when n (cid:107) y it is Wσ ¯σ2 = sin2[ksod cos(θ)]. In most cases, the splitting of the Cooper-pair spin state by the SO interac- tion reduces significantly the Josephson current through the superconducting weak link under consideration. Two features determine the magnitude of the effect for a given spin quantization axis in the leads (in addition to the strength kso of the SO interaction and the length d over which it acts). One is the extent to which the nanowire is bent (θ in Fig. 6), and the other is the orienta- tion n of the electric field responsible for the SO coupling relative to the spin quantization axis. Both break spin conservation, which results in Rabi oscillations between the singlet and triplet spin states of the (originally spin- singlet) Cooper pairs passing through the SO-active weak J(φ)J0(φ)ksod2π0.51.5-11V. CONCLUSIONS In addition to the charge of electrons, their spin de- gree of freedom can also play an important role when nanometer-sized devices are used for electronics appli- cations. The electron spin naturally comes into play if magnetic materials or external magnetic fields are used. However, even in non-magnetic materials the spin may couple strongly to an effective magnetic field induced by the spin-orbit (SO) interaction, which is a relativistic ef- fect that couples the electron's spin degree of freedom to its orbital motion. Such a coupling, first discovered in bulk materials without spatial inversion symmetry, can be significantly enhanced in nanostructures where the screening of an electrostatic field is suppressed and spatial inversion symmetry can be lifted by internal or external electric fields. The concept of an enhanced spin- orbit coupling in the vicinity of crystal surfaces,6,7 can therefore naturally be extended to nanostructures, where the surface to volume ratio can be rather high. Carbon nanotubes and semiconductor wires seem par- ticularly suitable to be used as spin-splitters. Mea- sured Rashba spin-orbit-coupling induced energy gaps in InGaAs/InAlAs (∆so = 2vFkso ≈ 5 meV)43 and InAs/AlSb (∆so ≈ 4 meV)44 quantum wells correspond to kso ≈ 4 × 106 m−1. The strain-induced SO energy ≈ 0.4 gap for a carbon nanotube is ∆strain ≈ 0.4 × 106 m−1 for meV, corresponding to kstrain vF ≈ 0.5×106 m/sec.45 For d of the order of µm, k(strain) d can therefore be of order 1 − 5. = 2vFkstrain so so so so In this article we have presented a short review of re- cent theoretical predictions, which may bring new func- tionality to nanoscale devices through the electronic spin degree of freedom. The "twisting" of the electronic spin induced by an SO interaction that is geometrically lo- calized to a weak link between bulk electrodes, can be viewed as a splitting of electronic waves in spin space -- a phenomenon we call Rashba spin-splitting. A common feature of the investigated Rashba spin-splitting devices is the possibility to tune the electronic transport through an SO-active weak link mechanically and possibly also electrostatically by "spin gating".35 We have shown that this is possible both for normal and superconducting elec- tron transport. Nevertheless, more research has to be done in order to develop a complete theory of Rashba gating of normal and superconducting weak links. Here we would like to mention a few possible directions for future work. Role of the "spin quantization axis" in the leads. The electron spin projection on an arbitrary chosen axis can take the two possible values ±/2, meaning that these are the eigenvalues of a certain operator acting on the spin wave-function. The corresponding eigenfunctions span the full Hilbert space. When the electronic spin is decoupled from other degrees of freedom and external fields, this spin operator commutes with the Hamiltonian and therefore the same eigenfunctions are also eigenfunc- tions of the Hamiltonian and and thus represent station- 13 ary spin states for any choice of the spin quantization axis. This is no longer the situation if the spin is coupled to an external magnetic field, in which case the eigen- states correspond to a spin quantization axis that is par- allel to the magnetic field. It is also not the case for an SO-active material where the spin eigenstates corre- spond to wave-vector dependent directions of the spin quantization axis. When an SO-active weak link con- nects two SO-inactive electrodes, the spin quantization axis will in general point in different directions in differ- ent parts of the device. Which spin state the electrons occupy in the source electrode is therefore important. It follows that the choice of spin quantization axis in the leads, which can be accomplished by applying a weak ex- ternal magnetic field (see the discussion in Sec. III), is another tool for spin-controlled electron transport phe- nomena that needs to be fully investigated. Spin-vibron coupling in nano-electromechanical weak links. The sensitivity of spin-controlled transport through an SO-active weak link to a mechanical defor- mation of the link, which has been demonstrated in this review, leads to the question of how transport is af- fected by the coupling between the spin and the mechan- ical vibrations. Such a coupling can be strong enough to cause spin-acoustic functionality in SO-active nano- electromechanical devices, which deserves to be investi- gated. Singlet-to-triplet spin conversion in spin-orbit active superconducting weak links. The possibility of an SO- induced "spin redesign" of Cooper pairs passing through a Rashba weak link, which was demonstrated in Sec. IV, raises the question of what kind of links can be estab- lished between two superconductors based on pairs of electrons in different spin states. The spin polarization of Cooper pairs that may result from their propagation through an SO-active Rashba spin-splitter allows for a gradual change of their spin state and hence for a trans- formation between a spin-singlet Cooper pair and a spin- triplet pair. The complete theory of the above conversion can be connected to the interesting problem of proximity- induced spin polarization of superconductors. Role of the Coulomb interaction in spin-gated devices. Charge and spin are two fundamental properties of elec- trons and we have shown that, due to the SO interaction, spin as well as charge couples to an electric field. There- fore, the question of how electron transport through a weak link is affected by the interplay between the Coulomb blockade of tunneling processes and the phe- nomenon of spin splitting is an intriguing task for future research. For example, in the study of a superconducting SO-active weak link10 (see also Sec. IV), the Coulomb blockade phenomenon was used to simplify the process of spin polarization of a Cooper pair by decomposing it into a sequence of spin twists of single electrons. What the result of lifting the Coulomb blockade will be, is an important question for future research. To conclude we emphasize that the study of spin- controlled transport through SO active weak links is only in its infancy. We believe that the early progress, some of it reviewed here, has laid a solid foundation for a wealth of future experimental and theoretical achievements. Acknowledgments This work was partially supported by the Swedish Re- search Council (VR), by the Israel Science Foundation (ISF) and by the infrastructure program of Israel Min- istry of Science and Technology under contract 3-11173. Appendix A: Tunneling elements In a simplified model, the weak links between the elec- trodes are pictured as straight segments connected at contorted bends. Considering the bent regions as scat- tering centers, this Appendix outlines the derivation of the tunneling elements representing the weak links. The probability amplitude for tunneling from one lead to the other is constructed from products of propagators along the straight segments and a transfer amplitude across the contorted parts. The propagator of the electron along a straight seg- ment in the presence of spin-orbit interactions and a Zeeman field is found in Sec. A 1. Confining the dis- cussion to a bent wire that couples two electrodes, the effective tunneling elements in-between them are derived in Sec. A 2. The effect of the mechanical degrees of free- dom on the tunneling elements is introduced in Sec. A 3: the vibrations' dynamics is incorporated into the effective tunneling.31 14 Here we extend that calculation in two directions. First, we allow for other forms of the SO interaction. For instance, SO coupling may be induced by strains, as happens in carbon nanotubes, where the spin dynamics is described by the effective interaction17,47 Hstrain so = vFkstrain so k · σ . (A3) so Here, vF is the Fermi velocity [see also Eq. (7)]; for ≈ 0.4× 106 m−1.28 vF ≈ 0.5× 106 m/sec, one finds kstrain Second, one may wish to find the propagator in the pres- ence of a magnetic field B. The orbital effect of this field on the motion along a one-dimensional wire can be ac- counted for by assigning an Aharonov-Bohm phase factor to the propagator, the phase being the magnetic flux (in units of the flux quantum) accumulated from the field upon moving along the segment. (Naturally, this phase factor depends on the choice of the coordinate origin; the physical quantities, however, include only the total Aharonov-Bohm flux through closed loops.16) The mag- netic field is coupled also to the spin, adding to the the Hamiltonian H(k) the Zeeman interaction, µBB · σ. It follows that the generic form of the (ballistic) Hamilto- nian on the straight segment is H(k) = k2 2m∗ + Q(k) · σ . For example, for the SO interaction Eq. (3), Q(k) = kso m∗ (k × n) + µBB . (A4) (A5) Note that Q(k) combines together the Zeeman magnetic field, and the effective magnetic field representing the SO interaction. The Hamiltonian (A4) is easily diagonalized: the eigen- 1. Propagation along a straight segment values are The electron's propagation along a straight segment is described by the Green's function corresponding to the Hamiltonian there. Assuming that the motion is ballistic, the spatial part of the wave function is taken as a plane wave. The propagator from point r(cid:48) to point r is then G(r − r(cid:48); E) = dkeikr−r(cid:48)[E + i0+ − H(k)]−1 , (A1) (cid:90) where E is the electron's energy, and the vector k, whose length is k, lies in the direction of the segment connecting connecting r(cid:48) with r. When the SO interaction is the one given in Eq. (3), the Hamiltonian is (adopting units in which = 1) H(k) = k2 2m∗ + kso m∗ σ · (k × n) . (A2) Inserting this expression into Eq. (A1) and carrying out the integration over the length k, one obtains the propa- gator as given in Ref. 46 (see also Ref. 18). ±(k) = k2 2m∗ ± Q(k) , (A6) and the projection operators into each of the correspond- ing subspaces are Π±(k) = 1 ± Q(k) · σ 2 , (A7) where Q(k) is a unit vector in the direction of the vector Q(k), whose length is Q(k). Using the diagonalized form in Eq. (A1) yields (cid:90) (cid:17) . (A8) G(r − r(cid:48); E) = dkeikr−r(cid:48) ×(cid:16) Π+(k) E + i0+ − +(k) + Π−(k) E + i0+ − −(k) The poles of the integrand in Eq. (A8) are given by the relation k2± = k2 0 ∓ 2m∗Q(k) + i0+ , (A9) where k2 0 = 2m∗E. Hence, G(r − r(cid:48); E) = iπ (cid:16) m∗eik+r−r(cid:48) k+ m∗eik−r−r(cid:48) + k− Π+(k+) (cid:17) Π+(k−) . (A10) (Note that the angles of the vector k are not changed along the straight segment, and therefore the integra- to obtain 15 tion is carried out over the magnitude, k.) The energy E corresponds to the Fermi energy in the leads; assum- ing that it is much larger than the energy scales of the SO interaction and the magnetic field, one may use the approximation k± ≈ k0 ∓ m∗ k0 Q(k0) , (A11) (cid:16) (cid:16) G(r − r(cid:48); E) = G0(r − r(cid:48); E) = G0(r − r(cid:48); E) −i e cos k0 m∗Q0r−r(cid:48) (cid:104) m∗Q0r − r(cid:48) i (cid:105) − i sin m∗Q0r−r(cid:48) k0 (cid:104) m∗Q0r − r(cid:48) Π−(k0) Π+(k0) + e (cid:17) . (A12) k0 k0 (cid:17) (cid:105) σ · Q0 Here, G0(r − r(cid:48); E) = iπ(m∗/k0) exp[ik0r − r(cid:48)] , (A13) the spin dynamics. The latter is embedded in the matrix W, W = exp[−iψL] × exp[−iψR] , (A17) is the propagator on the segment in the absence of the SO interaction and the magnetic field, and where Q0 = Q(k0) , (A14) where the angles of the vector k are those of the straight segment. The spin dynamic, caused by the spin-orbit in- teraction and the Zeeman field, is contained in the second factor of Eq. (A12). 2. Weak links with a bend Figure 6(b) illustrates the model system used in the calculations. The weak link between two electrodes, taken to lie in the XY plane, is replaced by two straight one-dimensional wires, RL and RR, of equal length d/2, connected by a bent. The angles between these wires and the x axis are θ and −θ, respectively. This means that the direction of the vector k [see, e.g., Eq. (A5)] of the left wire is {cos θ, sin θ, 0} and that of the right wire is {cos θ,− sin θ, 0}. These unit vectors determine the cor- responding vectors Q0, Eq. (A14). For this configuration, the tunneling amplitude, a 2×2 matrix in spin space, is W = W0W , (A15) with W0 = G0(RL; E)T G0(RR; E) , (A16) where T is the transfer matrix through the bent in the wire. This scalar amplitude comprises all the character- istics of the tunneling element that are independent of ψL(R) = m∗Q0L(R)d 2k0 σ · Q0L(R) . (A18) The unitary matrix W performs two consecutive spin ro- tations of the spins, around the axes Q0L and Q0R. For the SO interaction given in Eq. (3), and in the absence of the Zeeman field, one finds that Q0L = Q0R = k0kso/m∗, and Q0L(R) = (d/2) RL(R)× n. Equation (A15) is derived to lowest possible order in the tunneling; the explicit de- pendence of W0 on the momenta is omitted for brevity. 3. Vibrational degrees of freedom Coupling the charge carriers with the mechanical vi- brations of the suspended nanowire forming the junction adds an interesting aspect to the tunneling elements. For example, it was shown that this coupling can render the conductance through a single-channel wire to be affected by a constant magnetic field. The bending vibrations modify geometrically the spatial region where an orbital magnetic field is present, leading to a finite Aharonov- Bohm effect,31 which in turn gives rise to a magnetic-field dependence of the transmission. Likewise, the effect of the SO interaction can be modified by the effective area covered by the vibrating wire.14 Consider for instance the setup depicted in Fig. 6(b). Within this plausible geometry, yL = yR = (d/2) sin θ and xL = xR = (d/2) cos θ, where θ is the instanta- neous bending angle. (An alternative geometry, with xL = xR = d/2 and yL = −yR = (d/2)tanθ, gives similar results.) In order to mimic the bending vibrations of the wire we assume that once the wire is bent by the (equi- librium) angle θ0, then the distance along x between the two supporting leads is fixed, while the bending point vi- brates along y. As a result, tanθ = 2y/[d cos θ], implying that ∆θ = (2/[d cos θ]) cos2 θ0∆y. (Here d cos θ0 is the wire projection on the x direction.]) It follows that θ = θ0 + ∆θ = θ0 + (a0 cos θ0/d)(b + b†) , (A19) where a0 is the amplitude of the zero-point oscillations and b (b†) is the annihialtion (creation) operator of the vibrations. Their free Hamiltonian is described by the Einstein model, Hvib = ω0b†b . (A20) Details of the derivation of the current through a vibrat- ing wire are given in Sec. B 2. 16 † k(p)σ creates an electron in the left (right) The operator c electrode, with momentum k(p) and a spin index σ, which denotes the eigenvalue of the spin projection along an arbitrary axis. The construction of the matrix ele- ments is detailed in Secs. A 1 and A 2. As mentioned, the electrodes are considered as BCS superconductors, (cid:88) (cid:16) k(p) HL(R) = + ξk(p)c † k(p)σck(p)σ (cid:88) k(p) ∆L(R) † c k(p)↑c † −k(−p),↓ + H.c. , (B4) where ξk(p) = k(p) − µ is the quasi-particle energy in the left (right) bulk superconducting lead, and µ is the common chemical potential. The superconductor order parameter ∆L(R) is given by the self-consistency relation ∆L(R) = VBCS (cid:104)c−k(−p)↓ck(p)↑(cid:105) , (B5) (cid:88) (cid:17) Appendix B: Spin-resolved currents k(p) This Appendix is divided into two parts. The spin- resolved currents through the weak link discussed in Secs. A 1 and A 2 are derived in Sec. B 1; that part ignores the effect of the mechanical vibrations. For the sake of com- pleteness, we allow for the possibilities that the junction couples two superconducting electrodes, a superconduct- ing and a normal one, or two normal-state electrodes. In all these cases we assume that each electrode is de- scribed by a free electron gas, augmented (in the case of a superconducting lead) by the BCS Hamiltonian. The currents through a vibrating nanowire are considered in Sec. B 2. For simplicity the discussion there is confined to a junction connecting two normal electrodes. 1. Spin-resolved current through static weak links We consider the simplified, though realistic, model, in which two electrodes are connected via a spin-dependent tunnel Hamiltonian, † (c pσ(cid:48)[Wp,k]σ(cid:48),σckσ + H.c.) . (B1) (cid:88) (cid:88) k,p σ,σ(cid:48) HT = Here, where VBCS denotes the attractive interaction among the electrons. The total Hamiltonian of the junction is thus H = HL + HR + HT . (B6) Additional comments on the calculation of the current in-between two superconducting leads are given below.49 The spin-resolved particle current emerging from the left electrode, IL,σ, is found by calculating the time evo- lution of the number operator of electrons with spin pro- jection σ, NL,σ, − IL,σ ≡ (cid:104) NL,σ(cid:105) = (cid:68) (cid:88) (cid:88) = 2Im k,p σ(cid:48) (cid:68)(cid:88) k d dt † kσckσ c (cid:69) (cid:69) , [Wp,k]σ(cid:48),σc † pσ(cid:48)ckσ (B7) where we have used the relation (B3) and the self- consistency requirement (B5). The angular brackets in Eq. (B7) denote the quantum- thermal average, which we calculate within second-order perturbation theory in the tunneling Hamiltonian HT , Eq. (B1), (cid:88) (cid:88) dt(cid:48)(cid:68)(cid:104) k,p IL,σ = 2 Re (cid:90) t −∞ σ(cid:48) (cid:105)(cid:69) † pσ(cid:48)(t)ckσ(t),HT (t(cid:48)) [Wp,k]σ(cid:48),σc . (B8) [Wp,k]σ,σ(cid:48) = ([W−p,−k]−σ,−σ(cid:48))∗ (B2) × are elements of a matrix in spin space, which obey time- reversal symmetry.48 (In the presence of a Zeeman in- teraction the sign of the magnetic field in the matrix element on the right-hand side is reversed.) The relation (B2) adds to the one imposed by the hermiticity of the Hamiltonian, [Wp,k]σ,σ(cid:48) = ([Wk,p]σ(cid:48),σ)∗ . (B3) The time-dependence of the operators should be handled with care. When both electrodes are superconducting the difference between the phases of the two order parame- ters evolves in time according to the Josephson relation, leading to an ac current. This is not taken into account in the second-order perturbation calculation presented below, and therefore when the junction couples two su- perconducting leads, our treatment is valid only for the equilibrium situation, where no bias is applied across the junction. In that case the quasi-particle current van- ishes. However, the comparison between the amplitude of the current in the normal state of the junction, i.e., the transmission of the junction, and that of the Josephson current, is of great interest since the SO interaction mod- ifies them differently. For this reason both currents are In the standard perturbation calculation carried kept. out here, the normal-state transmission is derived from that of the quasi-particles. Accordingly, the particle cur- rent is separated into two parts, 17 IL,σ = I S L,σ + I N L,σ , (B9) with † −p−σ(cid:48)(t(cid:48))c−k−σ(t(cid:48)) − c † −p−σ(cid:48)(t(cid:48))c−k−σ(t(cid:48))c (cid:69) † pσ(cid:48)(t)ckσ(t) , † † † † kσ(t(cid:48))cpσ(cid:48)(t(cid:48)) − c kσ(t(cid:48))cpσ(cid:48)(t(cid:48))c pσ(cid:48)(t)ckσ(t) pσ(cid:48)(t)ckσ(t)c (cid:69) (B10) . (B11) c (cid:88) (cid:88) k,p σ(cid:48) I S L,σ =2 Re (cid:90) t −∞ and I N L,σ =2 Re (cid:88) (cid:88) k,p σ(cid:48) † c pσ(cid:48)(t)ckσ(t)c dt(cid:48)[Wp,k]σ(cid:48),σ2(cid:68) (cid:90) t dt(cid:48)[Wp,k]σ(cid:48),σ2(cid:68) (cid:34) (cid:104)c (cid:20) † k↑(t(cid:48))ck↑(t)(cid:105) † † −k↓(t)(cid:105) k↑(t(cid:48))c −∞ (cid:104)c Ak(ω) e−iϕL Bk(ω) The quantum-thermal average is found by introducing the Green's functions of the bulk (left) lead, G+−(k; t, t(cid:48)) = i G−+(k; t, t(cid:48)) = [ G+−(k; t, t(cid:48))]∗ , , (B12) and their Fourier transforms, G+−(k, ω) = i[1 − fL(ω)] vk = vk exp[iϕL/2], with uk2 = 1 − vk2 = (1 + ξk/Ek)/2, and Ek = (cid:112)ξ2 Analogous expressions pertain for the right lead, with k replaced by p, and L by R. The superconducting gap function of the left BCS lead is ∆L = ∆L exp[iϕL], and the coherence factors there are uk = uk exp[−iϕL/2] and k + ∆L2. The spectral functions in Eq. (B13) are G−+(k, ω) = − fL(ω) 1 − fL(ω) G+−(k, ω) . (B13) (cid:35) (cid:104)c−k↓(t(cid:48))ck↑(t)(cid:105) † (cid:104)c−k↓(t(cid:48))c −k↓(t)(cid:105) (cid:21) eiϕLBk(ω) Ak(−ω) , (B14) (B15) Ak(ω) = 2π[uk2δ(ω + Ek) + vk2δ(ω − Ek)] , Bk(ω) = −2πukvk[δ(ω + Ek) − δ(ω − Ek)] , (cid:88) (cid:90) dωdω(cid:48) f (ω) − f (ω(cid:48)) and fL(ω) is the Fermi function of the quasi particles in the left lead. Inserting the relations (B12) and (B13) into Eq. (B10) gives the equilibrium Josephson current, L,σ = sin(ϕL − ϕR) I S P k,p 2π2 ω − ω(cid:48) Bp(ω)Bk(ω(cid:48)){[Wp,k]σ,σ2 − [Wp,k]σ,σ2} , where we have used the symmetry Bk(ω) = B−k(ω) = −Bk(−ω) and fL(ω) = fR(ω) ≡ f (ω) = (exp[βω] + 1)−1 (β is the inverse temperature), since as mentioned, the supercurrent is calculated at equilibrium; P denotes the principal part, and σ is the spin direction opposite to σ. The transmission of the junction in the normal state is found by inserting the relations (B12) and (B13) into Eq. (B11) for the normal part of the spin-resolved current, (cid:90) dω (cid:88) 2π k,p I N L,σ = [fL(ω) − fR(ω)] (cid:16)[Wp,k]σ,σ2 + [Wp,k]σ,σ2(cid:17) Ak(ω)Ap(ω) . (B16) Here we have used the symmetry Ak(ω) = A−k(ω). As mentioned, the quasi-particle current I N for the unbiased junction for which fL(ω) = fR(ω). L , Eq. (B16), vanishes A comparison of the two expressions, Eq. (B15) and Eq. (B16), reveals the different ways by which the total effective magnetic field [the Zeeman field and the effective magnetic field due to the SO interaction, see Eq. (A4)] affects the Josephson current and the particle current in the normal state. One notes that the diagonal (in spin space) elements of the tunneling matrix appear in these two expressions with the same sign, as opposed to the off-diagonal ones. This implies that there is a significant difference between the effect of the component of an effective magnetic field normal to the junction plane, and an effective magnetic field in the junction's plane. 2. Spin-resolved currents through vibrating nanowires 18 When the wire connecting the two electrodes [see Fig. 6(b)] is vibrating, the tunneling amplitude is a dynamical variable (see Sec. A 3), and hence depends on time. This modifies the calculation of the current. For convenience, the discussion is confined to the case where the weak link connects two normal electrodes. In second-order perturbation theory, the starting point is still Eq. (B8), but one has to include in the calculation the time dependence of the hopping amplitude. As a result, Eq. (B8) is modified, IL,σ = 2Re = 2Re (cid:88) (cid:88) k,p (cid:88) (cid:88) σ(cid:48) (cid:90) t (cid:90) t −∞ k,p σ(cid:48) −∞ † pσ(cid:48)(t)ckσ(t),HT (t(cid:48)) [Wp,k]σ(cid:48),σ(t)c (cid:105)(cid:69) dt(cid:48)(cid:68)(cid:104) dt(cid:48)(cid:16)(cid:104)[Wp,k]σ(cid:48),σ(t)[Wk,p]σ,σ(cid:48)(t(cid:48))(cid:105)(cid:104)ckσ(t)c pσ(cid:48)(t)(cid:105)(cid:17) † † − (cid:104)[Wk,p]σ,σ(cid:48)(t(cid:48))[Wp,k]σ(cid:48),σ(t)(cid:105)(cid:104)c kσ(t(cid:48))ckσ(t)(cid:105)(cid:104)cpσ(cid:48)(t(cid:48))c † † pσ(cid:48)(t)cpσ(cid:48)(t(cid:48))(cid:105) kσ(t(cid:48))(cid:105)(cid:104)c . (B17) The quantum thermal averages over the electronic operators are obtained as in Sec. B 1, using Eqs. (B12) and (B13) (with ∆L = ∆R = 0). The thermal average over the tunneling amplitudes is carried out with the Einstein Hamiltonian, Eq. (A20). Using the notations introduced in Eqs. (A16), (A17), and (A18), we find (cid:104)[Wp,k]σ(cid:48),σ(t)[Wk,p]σ,σ(cid:48)(t(cid:48))(cid:105) = W02(cid:104)[e−iψR(t) × e−iψL(t)]σ(cid:48),σ[e−iψL(t(cid:48)) × e−iψR(t(cid:48))]σ,σ(cid:48)(cid:105) = P (n)ei(n−n(cid:48))ω0(t−t(cid:48))(cid:104)n[e−iψR × e−iψL]σ(cid:48),σn(cid:48)(cid:105)2 . (cid:88) n,n(cid:48) Here n(cid:105) indexes the eigenfunctions of the Einstein Hamiltonian and (B18) (B19) (B20) auch that P (n) = e−nβω0 Tre−βHvib = e−nβω0(1 − e−βω0) , P (n) = 1 , P (n)n = 1 eβω0 − 1 ≡ NB(ω0) . ∞(cid:88) (cid:90) dωdω(cid:48) n=0 4π ∞(cid:88) (cid:88) n=0 σ(cid:48) (cid:88) n,n(cid:48) IL,σ = 2W02 NL,σ NR,σ(cid:48) P (n) ×(cid:16) Inserting the expressions for the quantum thermal averages into Eq. (B17) yields fLσ(ω)[1 − fRσ(cid:48)(ω(cid:48))]δ[ω − ω(cid:48) + (n − n(cid:48))ω0](cid:104)n[e−iψR × e−iψL]σ(cid:48),σn(cid:48)(cid:105)2 − fRσ(cid:48)(ω(cid:48))[1 − fLσ(ω)]δ[ω(cid:48) − ω + (n − n(cid:48))ω0](cid:104)n[eiψ L × eiψ † . (B21) R ]σ,σ(cid:48)n(cid:48)(cid:105)2(cid:17) † Here, NL(R),σ are the spin-resolved densities of states at the common chemical potential of the device, µ = (µL + µR)/2 [see Eqs. (14) and (18), and the discus- sion following the latter]. The reservoirs are represented by their respective electronic distributions determined by the spin-dependent electrochemical potentials, fL,σ(k,σ) = [eβ(k,σ−µL,σ) + 1]−1 , fR,σ(cid:48)(p,σ(cid:48)) = [eβ(p,σ(cid:48)−µR,σ(cid:48) ) + 1]−1 , (B22) with β−1 = kBT . As expected, the coupling with the vibrational modes of the wire introduces inelastic processes into the tunnel- ing current, in which the charge carriers exchange energy with the mechanical degrees of freedom. Another point to notice it that the current is not spin-resolved unless the electrodes are polarized. This point is further discussed in the main text. It reflects the conclusion reached in Sec. B 1: the contributions to the normal-state particle current coming from the diagonal elements of the tun- neling amplitude and that of the off diagonal ones add together. ∗ Electronic address: [email protected] 1 S. M. Al-Jaber, X. Zhu, and W. C. Henneberger, Interac- tion of a moving magnetic dipole with a static electric field, Eur. J. Phys. 12, 268 (1991). 2 In a reference system that moves with velocity v rela- tive to a static electric field E one finds a magnetic field B = −(1/c)(v × E). The expression in Eq. (1) is half as large as the interaction energy −µ· B calculated using this Lorentz transformation result for B. The occurrence of this "Thomas half" is due to general requirements of relativis- tic invariance. A change of reference frame also leads to a time transformation and, consequently, to a change of the Larmor frequency; see §41 of Ref. 3. 3 V. B. Berestetskii, L. P. Pitaevskii, and E. M. Lifshitz, Landau and Lifshitz, Vol. 4 "Quantum Electrodynamics", 2nd ed. (Butterworth-Heinemann, Oxford, 1982), §33. 4 G. Dresselhaus, Spin-orbit coupling effects in zinc blende structures, Phys. Rev. 100, 580 (1955). 5 E. I. Rashba and V. I. Sheka, Symmetry of energy bands in crystals of wurtzite type: II. Symmetry of bands with spin-orbit interaction included, Fiz. Tverd. Tela: Collected Papers 2, 162 (1959); for an English translation see the supplementary material to Ref. 13. 6 F. T. Vas'ko, Spin splitting in the spectrum of two- dimensional electrons due to the surface potential, Pis'ma Zh. Eksp. Teor. Fiz. 30 (9), 574 (1979) [JETP Lett. 30 (9), 541 (1984)]. 7 Y. A. Bychkov and E. I. Rasbha, Properties of a 2D elec- tron gas with lifted spectral degeneracy, P. Zh. Eksp. Teor. Fiz. 39, 66 (1984) [JETP Lett. 39, 78 (1984)]. 8 See, e.g., M. A. U. Absor, H. Kotaka, F. Ishii, and M. Saito, Strain-controlled spin splitting in the conduction band of monolayer WS2, arXiv:1607.07566. 9 L. Petersen and P. Hedegard, A simple tight-binding model of spin-orbit splitting of sp-derived surface states, Surf. Sci. 459, 49 (2000). 10 R. I. Shekhter, O. Entin-Wohlman, M. Jonson, and A. Aharony, Rashba splitting of Cooper pairs, Phys. Rev. Lett. 116, 217001 (2016). 11 The Rashba Hamiltonian (3) is often written as Hso = αR σ · (k × n). Clearly, kso = m∗αR/2. The sign conven- tion in Eq. (3) is the same as the one in Ref. 7 and the one used in Ref. 10; in contrast, Refs. 8 and 12 define the Rashba interaction with the opposite sign. Also, the sign in Eq. (3) is opposite to that in Eqs. (1) and (2). Inverting the sign of αR changes the direction of the "effective magnetic field" caused by the SO interaction by π. In the absence of an external magnetic field (or if the external field is perpen- dicular to the SO-induced effective field) this sign does not matter for the size of the SO-induced spin-splitting of the energy bands, but it does affect the spin projection in the upper and lower spin-split eigenstates. For convenience, we keep the convention of Eq. (3). 12 A. Manchon, H. C. Koo, J. Nitta, S. M. Frolov, and R. A. Duine, New perspectives for Rashba spin-orbit coupling, Nature Mater. 14, 871 (2015). 13 G. Bihlmayer, O. Rader, and R. Winkler, Focus on the Rashba effect, New. J. Phys. 17, 050202 (2015). 14 R. I. Shekhter, O. Entin-Wohlman, and A. Aharony, Sus- pended nanowires as a mechanically controlled Rashba spin splitters, Phys. Rev. Lett. 111, 176602 (2013); R. I. Shekhter, O. Entin-Wohlman and A. Aharony, Me- chanically controlled spin-selective transport, Phys. Rev. B 90, 045401 (2014). 15 Y. Aharonov and A. Casher, Topological Quantum Effects for Neutral Particles, Phys. Rev. Lett. 53, 319 (1984). 19 16 Y. Aharonov and D. Bohm, Significance of Electromagnetic Potentials in the Quantum Theory, Phys. Rev. 115, 485 (1959). 17 M. S. Rudner and E. I. Rashba, Spin relaxation due to deflection coupling in nanotube quantum dots, Phys. Rev. B 81, 125426 (2010); K. Flensberg and C. M. Marcus, Bends in nanotubes allow electric spin control and coupling, Phys. Rev. B 81, 195418 (2010). 18 O. Entin-Wohlman, A. Aharony, Y. Galperin, V. Kozub, and V. Vinokur, Orbital ac Spin-Hall Effect in the Hopping Regime, Phys. Rev. Lett. 95, 086603 (2005); A. Aharony, Y. Tokura, G. Z. Cohen, O. Entin-Wohlman, and S. Kat- sumoto, Filtering and analyzing mobile qubit informa- tion via Rashba-Dresselhaus-Aharonov-Bohm interferome- ters, Phys. Rev. B 84, 035323 (2011). 19 B. D. Josephson, Possible new effects in superconductive tunnelling, Phys. Lett. 1, 251 (1962). 20 S. Hong, M. S. Grinolds, P. Maletinsky, R. L. Walsworth, M. D. Lukin, and A. Yacoby, Coherent, Mechanical Control of a Single Electronic Spin, Nano Lett. 12, 3920 (2012). 21 M. Blencowe, Quantum electromechanical systems, Phys. Rep. 395, 159 (2004); K. C. Schwab and M. L. Roukes, Putting mechanics into Quantum mechanics, Phys. Today 58, 36 (2005); R. I. Shekhter, L. Y. Gorelik, M. Jonson, Y. M. Galperin, and V. M. Vinokur, J. Comput. Theor. Nanosci. 4, 860 (2007); R. I. Shekhter, F. Santandrea, G. Sonne, L. Y. Gorelik, and M. Jonson, Nonequilibrium and quantum coherent phenomena in the electromechanics of suspended nanowires, Low Temp. Phys. 35, 662 (2009). 22 A. D. O'Connell, M. Hofheinz, M. Ansmann, R. C. Bial- czak, M. Lenander, E. Lucero, M. Neeley, D. Sank, H. Wang, M. Weides, J. Wenner, J. M. Martinis, and A. N. Cleland, Quantum ground state and single-phonon con- trol of a mechanical resonator, Nature (London) 464, 697 (2010). 23 This is similar in spirit to the device of Hong et al.,20 but differs from the ones of Refs. 24, 25, and 26. 24 S. H. Jhang, M. Marganska, Y. Skourski, D. Preusche, B. Witkamp, M. Grifoni, H. van der Zant, J. Wosnitza, and C. Strunk, Spin-orbit interaction in chiral carbon nan- otubes probed in pulsed magnetic fields, Phys. Rev. B 82, 041404(R) (2010). 25 J. R. Petta, H. Lu, and A. C. Gossard, A Coherent Beam Splitter for Electronic Spin States, Science 327, 669 (2010). 26 M. D. Shulman, O. E. Dial, S. P. Harvey, H. Bluhm, V. Umansky, and A. Yacoby, Demonstration of Entanglement of Electrostatically Coupled Singlet-Triplet Qubits, Science 336, 202 (2012). 27 S. Datta and B. Das, Electronic analog of the electro-optic modulator, Appl. Phys. Lett. 56, 665 (1990). 28 F. Kuemmeth, S. Ilani, D. C. Ralph, and P. L. McEuen, Coupling of spin and orbital motion of electrons in carbon nanotubes, Nature 452, 448 (2008). 29 D. Huertas-Hernando, F. Guinea, and A. Brataas, Spin- orbit coupling in curved graphene, fullerenes, nanotubes, and nanotube caps, Phys. Rev. B 74, 155426 (2006); D. V. Bulaev, B. Trauzettel, and D. Loss, Spin-orbit interaction and anomalous spin relaxation in carbon nanotube quan- tum dots, Phys. Rev. B 77, 235301 (2008); W. Izumida, K. Sato, and R. Saito, Spin-orbit Interaction in Single Wall Carbon Nanotubes: Symmetry Adapted Tight-Binding Cal- culation and Effective Model Analysis, J. Phys. Soc. Jpn. 78, 074707 (2009); J.-S. Jeong and H.-W. Lee, Curvature- enhanced spin-orbit coupling in a carbon nanotube, Phys. Rev. B 80, 075409 (2009). 30 M. J. Biercuk, S. Ilani, C. M. Marcus, and P. L. McEuen, Carbon Nanotubes, edited by A. Jorio, G. Dresselhaus, and M. S. Dresselhaus, Topics in Applied Physics, Vol. 111 (Springer-Verlag, Berlin, Heidelberg, 2008). 31 R. I. Shekhter, L. Y. Gorelik, L. I. Glazman, and M. Jon- son, Electronic Aharonov-Bohm Effect Induced by Quan- tum Vibrations, Phys. Rev. Lett. 87, 156801 (2006). 32 Y. Meir, Y. Gefen, and O. Entin-Wohlman, Universal ef- fects of spin-orbit scattering in mesoscopic systems, Phys. Rev. Lett. 63, 798 (1989); Y. Oreg and O. Entin-Wohlman, Transmissions through low-dimensional mesoscopic sys- tems subject to spin-orbit scattering, Phys. Rev. B 46, 2393 (1992). 33 Note the peculiar way of defining the transmission as given in Eqs. (9) and (21): it has dimensions of energy squared because it is not yet multiplied by the appropriate densities of states. 2I spin↑ = −2I spin↓ 34 Because of charge conservation one has (cid:80) = σ I spin σ . 35 R. I. Shekhter and M. Jonson, Spin gating of mesoscopic devices, Synth. Met. 216, 2(2016). 36 M. Gisselfalt, Parity effects in Josephson tunneling through a metallic island, Phys. Scr. 54, 397 (1996); see in partic- ular Eqs. (18) and (39). 37 In order to keep the estimates simple, it is assumed that both the Fermi velocities, vF, and the superconducting gaps of the two leads are the same. 38 Since electron tunneling is confined to a restricted interval ∆x at one or the other end of the wire, where the over- lap between the wave functions in the lead and the wire is finite, there is an uncertainty ∆k ∼ 1/∆x in the lon- gitudinal momentum of the tunneling electrons. We view this uncertainty as a splitting of the incoming well-defined momentum state into several channels k within a range ∆k. This enables co-tunneling processes, where the mo- mentum of the electron entering the wire from the left electrode and the momentum of the electron leaving the wire for the right electrode are different. However, for the geometry portrayed in Fig. 6, the direction of tunneling is nearly orthogonal to that of the current. This is why ∆x can be chosen large enough to suppress quantum fluctua- tions of the longitudinal electron momentum (and there- fore co-tunneling). A resonant tunneling effect exploiting this longitudinal momentum-conservation was recently ob- served by L. Britnell, R. V. Gorbachev, A. K. Geim, L. A. Ponomarenko, A. Mishchenko, M. T. Greenaway, T. M. Fromhold, K. S. Novoselov, and L. Eaves, Resonant 20 tunneling and negative differential conductance in graphene transistors, Nat. Commun. 4, 1794 (2013). 39 V. Ambegaokar and A. Baratoff, Tunneling Between Su- perconductors, Phys. Rev. Lett. 10, 486 (1963); erratum, Phys. Rev. Lett. 11, 104 (1963). 40 I. O. Kulik and I. K. Janson, Josephson effect in supercon- ductive tunneling structures (Israel Program for Scientific Translations [Keter press], Jerusalem, 1972). 41 L. I. Glazman and K. A. Matveev, Resonant Joseph- son current through Kondo impurities in a tunnel barrier, Pis'ma Zh. Eksp. Teor. Fiz. 49, 570 (1989) [JETP Lett. 49, 660 (1989); B. I. Spivak and S. A. Kivelson, Negative local superfluid densities: the difference between superconduc- tors and dirty Bose liquids, Phys. Rev. B 43, 3740 (1991). 42 Spin-triplet pairing correlations in a conductor with SO interactions in the proximity of an s-wave superconductor were found in the absence of a magnetic field by C. P. Reeg and D. L. Maslov, Proximity-induced triplet super- conductivity in Rashba materials, Phys. Rev. B 92, 134512 (2015). 43 J. Nitta, T. Akazaki, H. Takayanagi, and T. Enoki Gate Control of Spin-Orbit Interaction in an Inverted In0.53Ga0.47As In0.52Al0.48As Heterostructure, Phys. Rev. Lett. 78, 1335 (1997). 44 J. P. Heida, B. J. van Wees, J. J. Kuipers, T. M. Klap- wijk, and G. Borghs, Spin-orbit interaction in a two- dimensional electron gas in a InAs/AlSb quantum well with gate-controlled electron density, Phys. Rev. B 57, 11911 (1998). 45 F. Kuemmeth, S. Ilani, D. C. Ralph, and P. L. McEuen, Coupling of spin and orbital motion of electrons in carbon nanotubes, Nature 452, 448 (2008). 46 T. V. Shahbazyan and M. E. Raikh, Low-Field Anomaly in 2D Hopping Magnetoresistance Caused by Spin-Orbit Term in the Energy Spectrum, Phys. Rev. Lett. 73, 1408 (1994). 47 In this case the electric field is the host crystal-field, see L. Chico, M. P. L´opez-Sancho, and M. C. Munoz, Curvature-induced anisotropic spin-orbit splitting in car- bon nanotubes, Phys. Rev. B 79, 235423 (2009). 48 P. W. Anderson, in Lectures on the many-body problem, vol. 2 p. 113, Ed. E. R. Caianiello (Academic Press, New York, 1964). 49 J. C. Cuevas, A. Mart´ın-Rodero, and A. Levy Yeyati, Hamiltonian approach to the transport properties of super- conducting quantum point contacts, Phys. Rev. B 54, 7366 (1996).
1803.08549	1	1803	2018-03-22T19:10:03	Kramers-Kronig relations and causality conditions for graphene in the framework of the Dirac model	[ "cond-mat.mes-hall" ]	We analyze the concept of causality for the conductivity of graphene described by the Dirac model. It is recalled that the condition of causality leads to the analyticity of conductivity in the upper half-plane of complex frequencies and to the standard symmetry properties for its real and imaginary parts. This results in the Kramers-Kronig relations, which explicit form depends on whether the conductivity has no pole at zero frequency (as in the case of zero temperature when the band gap of graphene is larger than twice the chemical potential) or it has a pole (as in all other cases, specifically, at nonzero temperature). Through the direct analytic calculation it is shown that the real and imaginary parts of graphene conductivity, found recently on the basis of first principles of thermal quantum field theory using the polarization tensor in (2+1)-dimensional space-time, satisfy the Kramers-Kronig relations precisely. In so doing, the values of two integrals in the commonly used tables, which are also important for a wider area of dispersion relations in quantum field theory and elementary particle physics, are corrected. The obtained results are not of only fundamental theoretical character, but can be used as a guideline in testing the validity of different phenomenological approaches and for the interpretation of experimental data.	cond-mat.mes-hall	cond-mat	Kramers-Kronig relations and causality conditions for graphene in the framework of the Dirac model G. L. Klimchitskaya1, 2 and V. M. Mostepanenko1, 2, 3 1Central Astronomical Observatory at Pulkovo of the Russian Academy of Sciences, Saint Petersburg, 196140, Russia 2Institute of Physics, Nanotechnology and Telecommunications, Peter the Great Saint Petersburg Polytechnic University, Saint Petersburg, 195251, Russia 3Kazan Federal University, Kazan, 420008, Russia Abstract We analyze the concept of causality for the conductivity of graphene described by the Dirac model. It is recalled that the condition of causality leads to the analyticity of conductivity in the upper half-plane of complex frequencies and to the standard symmetry properties for its real and imaginary parts. This results in the Kramers-Kronig relations, which explicit form depends on whether the conductivity has no pole at zero frequency (as in the case of zero temperature when the band gap of graphene is larger than twice the chemical potential) or it has a pole (as in all other cases, specifically, at nonzero temperature). Through the direct analytic calculation it is shown that the real and imaginary parts of graphene conductivity, found recently on the basis of first principles of thermal quantum field theory using the polarization tensor in (2+1)-dimensional space-time, satisfy the Kramers-Kronig relations precisely. In so doing, the values of two integrals in the commonly used tables, which are also important for a wider area of dispersion relations in quantum field theory and elementary particle physics, are corrected. The obtained results are not of only fundamental theoretical character, but can be used as a guideline in testing the validity of different phenomenological approaches and for the interpretation of experimental data. 8 1 0 2 r a M 2 2 ] l l a h - s e m . t a m - d n o c [ 1 v 9 4 5 8 0 . 3 0 8 1 : v i X r a 1 I. INTRODUCTION Considerable recent attention has been focused on graphene, which is a two-dimensional sheet of carbon atoms packed in a hexagonal lattice [1, 2]. This unique material is inter- esting not only for condensed matter physics due to its unusual electrical and mechanical properties, but for quantum field theory as well. The point is that the electronic excitations in graphene are either massless or very light. At energies below a few eV they possess the linear dispersion relation and obey (2+1)-dimensional Dirac equation where the speed of light c is replaced with the Fermi velocity vF ≈ c/300 [1–3]. Thus, graphene makes possible testing many predicted effects of quantum field theory and quantum electrodynamics which are not experimentally feasible with much heavier ordinary electrons. Among other effects one could mention the Klein paradox [4], the creation of particle-antiparticle pairs from vac- uum in a static [5, 6] and time-dependent [7, 8] electric field, and the relativistic quantum Hall effect in a strong magnetic field [9]. Graphene is also unique in that its response to external electromagnetic field and quan- tum fluctuations, described by the polarization tensor in (2+1)-dimensional space-time, can be found in an explicit form on the basis of first principles of thermal quantum field the- ory. Although some special cases have been considered previously (see, e.g., Ref. [10] and literature therein), the complete expression for the polarization tensor of graphene in the one-loop approximation has been derived at zero temperature in Ref. [11] and at any nonzero temperature in Ref. [12], where the area of application was limited to the pure imaginary Matsubara frequencies. In doing so both cases of zero and nonzero width of the gap ∆ between the energy bands (i.e., of gapless and gapped graphene) and chemical potential µ were considered. The results of Refs. [11, 12] have been extensively used when investigat- ing the Casimir and Casimir-Polder forces in graphene systems [13–23] (some other, more phenomenological, approaches used for this purpose are the density-density correlation func- tions, models of the response functions of graphene by Lorenz-type oscillators, and the Kubo formalism [24–35]). A more universal representation for the polarization tensor of graphene at nonzero tem- perature was derived in Ref. [36]. Unlike Ref. [12], the polarization tensor of Ref. [36] allows an analytic continuation to the entire plane of complex frequencies including the real fre- quency axis. At the pure imaginary Matsubara frequencies both representations take the 2 same values. The novel representation was applied in investigations of the Casimir force [37–41] and, after a continuation to the real frequency axis, for better understanding of the reflectances of graphene and graphene-coated plates [42–44]. In Ref. [45] the polarization tensor of Ref. [36] was generalized for the case of doped graphene with a nonzero chemical potential. This generalization was used [46] to investigate an impact of nonzero band gap and chemical potential on the thermal effect in the Casimir force. One of the most important characteristics of graphene is its electrical conductivity. This quantity possesses many surprising properties connected with an existence of the so-called universal conductivity σ0 = e2/(4) expressed via the fundamental constants, electron charge e and Planck constant . For a pure graphene, having the zero band gap and no doping, the conductivity is equal to σ0 in the limit of zero temperature. This result might be considered as paradoxical if to take into account that with vanishing temperature the concentration of charge carriers in pure graphene goes to zero and there is no scattering and no dissipation processes. The conductivity of graphene was extensively investigated by many authors using the current-current correlation functions, the Kubo formalism, the Boltzmann transport theory, and the two-dimensional Drude model (see the review papers [47–49] and references therein). Some of the results obtained employ simple intuitive models, phenomenological approaches of a limited application area and even do not agree with each other. To overcome these troubles, the conductivity of graphene at any temperature was investigated on the basis of first principles of quantum electrodynamics using the polarization tensor of Refs. [36, 45] analytically continued to the real frequency axis. In Refs. [50, 51] the cases of pure and gapped graphene were considered, respectively, and in Ref. [52] of both gapped and doped graphene characterized by nonzero band gap ∆ and chemical potential µ. The real and imaginary parts of graphene conductivity have been found in an explicit form. It was shown that the major contribution to the conductivity of graphene calculated in the framework of Dirac model is local, whereas the nonlocal corrections are negligibly small. In this paper, we consider the problem of causality in the response of graphene to elec- tric field. The demand of causality leads to some constraints on the local conductivity of graphene. Specifically, it should be an analytic function in the upper half-plane of complex frequencies and satisfy certain symmetry conditions. These result in the Kramers-Kronig relations for the real and imaginary parts of the conductivity of graphene. Until the present 3 time the Kramers-Kronig relations for graphene were discussed only using some approxi- mate, phenomenological approaches leading to incomplete and even contradictory results (see, e.g., Refs. [53–56]). Thus, the form of Kramers-Kronig relations used in Refs. [53–55] does not take into account that the imaginary part of the conductivity of graphene has a pole at zero frequency. Furthermore, Ref. [53] arrives to the Kramers-Kronig relation ex- pressing the real part of graphene conductivity via its imaginary part, but fails in obtaining a similar relation with interchanged real and imaginary parts. An existence of the universal conductivity of graphene σ0 is not taken into account. Moreover, Ref. [56] admits that under some conditions the Kramers-Kronig relations for graphene do not hold. When it is consid- ered that these relations are not only of fundamental theoretical character, but are used for interpretation of the measurement data (see, e.g., Ref. [55]), it is of prime importance to conclusively find out their specific form for graphene and directly prove their validity. Below we establish an explicit form of the Kramers-Kronig relations for graphene and demonstrate that the real and imaginary parts of its conductivity, found independently on the basis of first principles of quantum electrodynamics, satisfy these relations precisely. Depending on temperature and a relationship between the band gap ∆ and chemical po- tential µ, an additional pole term in the Kramers-Kronig relations may arise as it holds in a familiar case of metals [57]. The obtained relations take proper account for the universal conductivity of graphene σ0. In fact, there is no reason that the Kramers-Kronig relations were not satisfied for the conductivity determined from the first principles. The obtained results, however, are physically meaningful because they establish the specific form of the Kramers-Kronig relations for so unusual material as graphene and, by performing the direct verification of these relations, confirm the expressions for its conductivity found recently in Ref. [52] using the polarization tensor. In the course of our calculations, the values of two integrals, indicated incorrectly in the most comprehensive and widely used table of in- tegrals [58], have been corrected. These integrals might be also useful in a wider context of dispersion relations for the scattering amplitudes in quantum field theory and physics of elementary particles. The paper is organized as follows. In Sec. II, the brief summary for the polarization tensor, conductivity of graphene and causality conditions is presented. Section III contains the proof of the Kramers-Kronig relations for the conductivity of graphene at zero temperature. The validity of the Kramers-Kronig relations at nonzero temperature is demonstrated in Sec. IV. 4 In Sec. V, the reader will find our conclusions and a discussion. Appendices A and B contain some details of several mathematical derivations. II. POLARIZATION TENSOR, CONDUCTIVITY OF GRAPHENE AND CAUSALITY CONDITIONS The polarization tensor of graphene in the one-loop approximation in the momentum rep- resentation is defined according to Refs. [59, 60] with the following differences. We consider the (2+1)-dimensional space-time. In the free Dirac equation the speed of light c is replaced with the Fermi velocity vF ≈ c/300 although an interaction with the electromagnetic field is governed, as usually, by the coupling constant e/c. In addition, one should take into account that we consider the polarization tensor at nonzero temperature T . Because of this, according to the Matsubara formalism, an integration over the zeroth component q0 of the wave vector qµ of a loop electronic excitation should be replaced with a summation over the pure imaginary fermionic Matsubara frequencies cq0n = 2πi(cid:18)n + 1 2(cid:19) kBT , (1) where kB is the Boltzmann constant and n = 0, ±1, ±2, . . . . Finally it is necessary to replace the zeroth component k0 of the wave vector kµ of an external photon in the argument of the polarization tensor with the pure imaginary bosonic Matsubara frequencies iξl = ck0l = 2πil kBT , where l = 0, ±1, ±2, . . . . As a result, the polarization tensor takes the form [11, 12, 36, 59, 60] Πµν(iξl, k) = −8παkBT ∞ Xn=−∞(cid:18)n + 1 2(cid:19) 1 ×Z dq (2π)2 tr 1 γµ iγµqµ − ∆/(2) iγµqµ − iγµkµ − ∆/(2) (2) (3) . Here, α = e2/(c) ≈ 1/137 is the fine structure constant, qµ = (q0n, q1, q2), kµ = (k0l, k1, k2), µ = 0, 1, 2, k = (k1, k2), γµ = η µ ν = diag(c, vF , vF ) and γν are the Dirac ν γν where η µ matrices. Note also that the numerical factor on the right-hand side of Eq. (9) takes into account four fermion species for graphene [1–3]. 5 The polarization tensor (3) was calculated over the entire axis of imaginary frequencies in Ref. [36], analytically continued to the real frequency axis and used for different purposes in Refs. [36, 42–44, 50, 51]. In Ref. [45] this tensor was generalized for the case of graphene with nonzero chemical potential µ [this is reached by the replacement q0n → q0n + µ/(c)] and analytically continued to the real frequency axis in Ref. [52]. The longitudinal (in-plane of graphene) and transverse (out-of-plane) electrical conductivities are expressed via the polarization tensor as [22, 50–52] σk(ω, k, T ) = −i σ⊥(ω, k, T ) = i ω 4πk2 Π00(ω, k, T ), c2 4πωk2 Π(ω, k, T ), (4) (5) where Π(ω, k, T ) = k2trΠµν(ω, k, T ) +(cid:18)ω2 c2 − k2(cid:19) Π00(ω, k, T ) and k = k. The conductivities of graphene are the complex quantities as well as the polarization tensor along the real frequency axis. Calculations show that the major contributions to both the real and imaginary parts of σk(⊥) are given in the local limit k = 0, whereas the nonlocal corrections are of the order of (vF /c)2 ∼ 10−5. In the local limit one has σ(ω, T ) ≡ σk(ω, 0, T ) = σ⊥(ω, 0, T ). (6) Note that the quantities Π00 and Π in Eq. (4) go to zero as k2 when k goes to zero, whereas trΠµν goes to a nonzero constant. Expanding all these quantities up to the first power in the parameter (vF k/ω)2 < (vF /c)2 and using Eqs. (28), (40), and (43) in Ref. [52], one obtains that in this perturbation order Π(ω, k, T ) = − ω2 c2 Π00(ω, k, T ). (7) Taking into account Eq. (4), it is seen that Eq. (7) is in agreement with Eq. (6). The explicit expressions for the quantity σ(ω, T ) in the most general case of graphene with nonzero ∆ and µ have been derived from Eq. (4) in Ref. [52]. It is convenient to present the local conductivity of graphene (6) as the sum of two contributions σ(ω, T ) = σ(0)(ω) + σ(1)(ω, T ). (8) 6 Imσ(0)(ω) = σ0 π (cid:20) 2∆ ω − (ω)2 + ∆2 (ω)2 (ω)2 + ∆2 (ω)2 ln(cid:12)(cid:12)(cid:12)(cid:12) ω + ∆ ω − ∆(cid:12)(cid:12)(cid:12)(cid:12) (cid:21) , The quantity σ(0) on the right-hand side of this equation is the contribution to the conduc- tivity which does not depend on T and µ. It is given by [52] Reσ(0)(ω) = σ0θ(ω − ∆) , (9) where ∆ is the width of the gap in Dirac's spectrum and θ(x) is the step function equal to unity for x ≥ 0 and zero for x < 0. Note that under the condition ∆ > 2µ the quantity σ(0)(ω) defined in Eq. (9) has the physical meaning of the total conductivity of graphene at zero temperature σ(ω, 0). This means that under the condition ∆ > 2µ it holds [52] σ(1)(ω, 0) = lim T →0 σ(1)(ω, T ) = 0. (10) Thus, if ∆ > 2µ the conductivity σ(ω, 0) does not depend on µ (even if µ is not equal to zero but is smaller than ∆/2) and Reσ(0) vanishes if ω < ∆. The conductivity of graphene at T = 0 and ∆ < 2µ is considered in Sec. III. The quantity σ(1) on the right-hand side of Eq. (8) depends on T , ∆ and µ. It can be represented in the form [52] Reσ(1)(ω, T ) = −σ0θ(ω − ∆) Imσ(1)(ω, T ) = 2σ0 π Z ∞ dt(cid:20)1 + ω ∆ (ω)2 + ∆2 (ω)2 (ω)2 + ∆2 (ω)2 F (ω, T ), (11) 1 t2 − 1(cid:21) F (ωt, T ), where the function F (x, T ) is defined as F (x, T ) = Xκ=±1(cid:20)exp(cid:18) x + 2κµ 2kBT (cid:19) + 1(cid:21)−1 . (12) It is convenient to introduce the new integration variable v = ωt/∆ in the second line of Eq. (11) which takes the form Imσ(1)(ω, T ) = 2σ0 π ∆ ω Z ∞ dv(cid:20)1 + 1 ∆2 + (ω)2 (v∆)2 − (ω)2(cid:21) F (cid:18)v∆ This expression has the pole term C(T )/ω at ω = 0, where C(T ) = 2σ0 π ∆ Z ∞ 1 dv v2 + 1 v2 F (cid:18)v∆ , T(cid:19) . 7 , T(cid:19) . (13) (14) Now we separate the pole term in the imaginary part of conductivity by adding and subtracting the quantity C(T )/ω on the right-hand side of Eq. (13). Leaving the first expression in Eq. (11) unchanged, both Reσ(1) and Imσ(1) can be rewritten as Reσ(1)(ω, T ) = −σ0θ(ω − ∆) (ω)2 + ∆2 (ω)2 F (ω, T ), Imσ(1)(ω, T ) = C(T ) ω + 2σ0 π ω∆Z ∞ 1 dv v2 + 1 v2[(v∆)2 − (ω)2] F (cid:18)v∆ , T(cid:19) . (15) Now we discuss the requirements of causality imposed on the conductivity σ(ω, T ) and its constituents σ(0)(ω) and σ(1)(ω, T ). According to the principle of causality, the electric current density j(t) must not depend on the values of electric field E(t) at times greater than t, i.e., j(t, T ) =Z ∞ 0 σ(τ, T )E(t − τ )dτ. (16) Multiplying both sides of this equation by eiωt and integrating with respect to t from −∞ to ∞, we obtain an equation for the Fourier images of the field and current density j(ω, T ) = σ(ω, T )E(ω), (17) (18) where σ(ω, T ) =Z ∞ 0 σ(τ, T )eiωτ dτ. Repeating the well known reasoning contained in Ref. [57] for the case of frequency- dependent dielectric permittivity, it is easy to find the analytic properties of σ(ω, T ) in the plane of complex frequencies and the symmetry properties of its real and imaginary parts. Specifically, from Eq. (18) it follows that in the upper half-plane (Imω > 0) σ(ω, T ) is an analytic function with no singularities. The real and imaginary parts of σ(ω, T ) are the even and odd functions of real frequency, respectively. From Eq. (18) it is seen also that for the complex ω it holds σ(−ω∗, T ) = σ∗(ω, T ). Then at the pure imaginary frequencies σ(ω, T ) takes the real values. Equation (16) is also valid for the contributions σ(0)(τ ) and σ(1)(τ, T ) to the conductivity σ(τ, T ) [with the corresponding contributions to the total current j(0)(t) and j(1)(t, T ) on the left-hand side]. From this it follows that all the above properties of σ(ω, T ) are inherent also in σ(0)(ω) and σ(1)(ω, T ). 8 Note that the explicit expressions (9) and (15) may appear in disagreement with the formulated above general properties of conductivity following from the causality condition (16). The point is that it may exist several equivalent representations for some quantity along the positive frequency axis, but only one of them allows immediate analytic continuation to the entire plane of complex frequencies. Equations (9) and (15) are written in the form which is most convenient for applications only at the real, positive frequencies and can be easily compared with the results obtained using various approximate and phenomenological approaches (see below). These equations, however, can be identically rewritten in the form where the analytic continuation from the real, positive frequency axis to the entire complex frequency plane is achieved by simply putting frequency ω complex. For example, Eq. (9) can be rewritten in the form σ(0)(ω) = i 2σ0 π (cid:20) ∆ ω − (ω)2 + ∆2 (ω)2 arctanh ω ∆(cid:21) , (19) where all the above properties are evidently satisfied. An equivalence of Eqs. (9) and (19) along the real, positive frequency axis follows from the identities [61] arctanhx = 1 2 ln arctanhx = arctanh 1 + x 1 − x 1 x , 0 ≤ x2 < 1, + i π 2 , (20) where the last identity is used for x > 1 at the upper bank of the cut which passes from unity to ∞. The analytic properties of the functions σ(ω, T ), σ(0)(ω) and σ(1)(ω, T ) result in the validity of the Kramers-Kronig relations which can be proven in exactly the same manner as it is done in Ref. [57] for the case of dielectric permittivity. The form of the Kramers-Kronig relations depends on the behavior of σ at zero frequency. As is seen in Eq. (9), both the real and imaginary parts of σ(0) are regular at ω = 0 (the first order pole in the first term of Imσ(0) is canceled by a similar pole with an opposite sign in the second term). At ω → ∞ the quantity Reσ(0) goes to σ0. Because of this, the Kramers-Kronig relation is valid for the function Reσ(0) − σ0. The result is similar to that presented in Ref. [57] for the dielectric permittivity Reσ(0)(ω) = σ0 + Imσ(0)(ω) = − 2ω π 9 ξImσ(0)(ξ) ξ2 − ω2 dξ, 0 Reσ(0)(ξ) ξ2 − ω2 dξ, 2 π −Z ∞ −Z ∞ 0 (21) where the crossed sign of integration means that the principal value of the integral is taken. We note also that −Z ∞ 0 dξ ξ2 − ω2 = 0. (22) Because of this it is not necessary to subtract σ0 in the nominator of the second equality in Eq. (21). Now we consider the second contribution to the conductivity of graphene, i.e., σ(1). As is seen in Eq. (15), the imaginary part of σ(1) has the first-order pole. Because of this, the Kramers-Kronig relations are similar to those obtained in Ref. [57] for the dielectric permittivity of conductors Reσ(1)(ω, T ) = 2 π ξImσ(1)(ξ, T ) − C(T ) ξ2 − ω2 dξ, (23) 0 −Z ∞ −Z ∞ 2ω π 0 Imσ(1)(ω, T ) = − Reσ(1)(ξ, T ) ξ2 − ω2 dξ + C(T ) ω , where C(T ) is defined in Eq. (14). We note that both the real and imaginary parts of σ(1) defined in Eq. (15) go to zero when ω → ∞. Because of this, it is not needed to subtract any constant from Reσ(1) like it was done in Eq. (21). At the same time, it is necessary to subtract C(T ) in the nominator of the first dispersion relation in Eq. (23). This subtraction does not change the value of the integral at all ω 6= 0 due to Eq. (22), but makes the Kramers-Kronig relation correct at ω = 0 (see the relevant discussions in Ref. [57] for the dielectric permittivity of metals and in Sec. III). By combining Eqs. (21) and (23), one arrives to the Kramers-Kronig relations for the total conductivity of graphene at any temperature ξImσ(ξ, T ) − C(T ) ξ2 − ω2 dξ, (24) Reσ(ω, T ) = σ0 + Imσ(ω, T ) = − 2ω π 0 2 π −Z ∞ −Z ∞ 0 Reσ(ξ, T ) ξ2 − ω2 dξ + C(T ) ω . The Kramers-Kronig relations (21), (23), and (24) follow from the discussed above general analytic properties of the local conductivity of graphene. None of the expressions for the graphene conductivity obtained in the previous literature using various approximate and phenomenological methods satisfy these relations precisely. Below we demonstrate, however, that the conductivity (8), (9), (15), derived independently on the basis of first principles of 10 quantum electrodynamics at nonzero temperature using the polarization tensor, is in full agreement with the Kramers-Kronig relations and, thus, with the demands of causality. III. KRAMERS-KRONIG RELATIONS FOR THE CONDUCTIVITY AT ZERO TEMPERATURE We begin with the case ∆ ≥ 2µ when the total conductivity of graphene at T = 0 is given by Eq. (9), i.e., σ(ω, 0) = σ(0)(ω). It is straightforward to substitute the first line of Eq. (9) in the right-hand side of the second Kramers-Kronig relation in Eq. (21) and obtain − 2ω π = − −Z ∞ 0 2σ0 π Reσ(0)(ξ) ξ2 − ω2 dξ ω(cid:20)−Z ∞ ∆ dζ ζ 2 − 2ω2 + ∆2−Z ∞ ∆ (25) dζ ζ 2(ζ 2 − 2ω2)(cid:21) , where the integration variable ζ = ξ is introduced. Integrating on the right-hand side of Eq. (25) we find − Reσ(0)(ξ)dξ ξ2 − ω2 = −Z ∞ 2ω π (ω)2 + ∆2 0 − (ω)2 σ0 ω ω + ∆ π (cid:20) 2∆ ω − ∆(cid:12)(cid:12)(cid:12)(cid:12) (cid:21) = Imσ(0)(ω) ln(cid:12)(cid:12)(cid:12)(cid:12) (26) if to take into account Eq. (9). Thus, the second Kramers-Kronig relation in Eq. (21) is really satisfied. Now we substitute the second line of Eq. (9) in the right-hand side of the first Kramers- Kronig relation in Eq. (21) and obtain σ0 + 2 π −Z ∞ 0 ξImσ(0)(ξ) ξ2 − ω2 dξ = σ0 − 2σ0 π2 −Z ∞ 0 where we have taken into account Eq. (22). ξ ξ2 − ω2(cid:20)ln(cid:12)(cid:12)(cid:12)(cid:12) + ξ + ∆ ξ − ∆(cid:12)(cid:12)(cid:12)(cid:12) ∆2 (ξ)2 ln(cid:12)(cid:12)(cid:12)(cid:12) (cid:21) dξ, (27) ξ + ∆ ξ − ∆(cid:12)(cid:12)(cid:12)(cid:12) In the Appendix A, we calculate the following important integral: I(b) ≡ −Z ∞ 0 y y2 − b2 ln(cid:12)(cid:12)(cid:12)(cid:12) y + 1 y − 1(cid:12)(cid:12)(cid:12)(cid:12) dy =  π2 2 , b < 1, 0, b > 1 (28) and indicate relevant incorrect results contained in Ref. [58]. 11 Introducing the variable y = ξ/∆ in the first integral on the right-hand side of Eq. (27) and using Eq. (28), one obtains ξ + ∆ ξ − ∆(cid:12)(cid:12)(cid:12)(cid:12) dξ = −Z ∞ 0 y y2 − b2 ln(cid:12)(cid:12)(cid:12)(cid:12) y + 1 y − 1(cid:12)(cid:12)(cid:12)(cid:12) dy (29) 0 ξ −Z ∞ ξ2 − ω2 ln(cid:12)(cid:12)(cid:12)(cid:12) =  π2 2 , ω < ∆, 0, ω > ∆, b ≡ ω ∆ . The second integral on the right-hand side of Eq. (27) can be evaluated similarly ξ(ξ2 − ω2) 0 ∆2 2 −Z ∞ b2 (cid:20)−Z ∞ = 1 0 1 ξ + ∆ ln(cid:12)(cid:12)(cid:12)(cid:12) y2 − b2 ln(cid:12)(cid:12)(cid:12)(cid:12) ξ − ∆(cid:12)(cid:12)(cid:12)(cid:12) y − 1(cid:12)(cid:12)(cid:12)(cid:12) y + 1 y 0 dξ = −Z ∞ dy − −Z ∞ 0 y(y2 − b2) 1 y + 1 y + 1 y − 1(cid:12)(cid:12)(cid:12)(cid:12) ln(cid:12)(cid:12)(cid:12)(cid:12) (cid:21) = y − 1(cid:12)(cid:12)(cid:12)(cid:12)  0, ln(cid:12)(cid:12)(cid:12)(cid:12) dy y dy (30) ω < ∆, − 1 ω(cid:1)2 , ω > ∆. 2(cid:0) π∆ In obtaining this result we have used Eq. (28) for two times, namely with b 6= 0 and b = 0. Substituting Eqs. (29) and (30) in the right-hand side of Eq. (27) and using the first line of Eq. (9), one arrives at σ0 + 2 π −Z ∞ 0 ξImσ(0)(ξ) ξ2 − ω2 dξ =  0, σ0 ω < ∆, (ω)2+∆2 (ω)2 , ω > ∆ (31) = Reσ(0)(ω). Thus, the conductivity σ(0) in Eq. (9) satisfies the first Kramers-Kronig relation in Eq. (21). Now we continue to consider the case of zero temperature, but assume that ∆ < 2µ. In this case it holds and σ(1)(ω, 0) = lim T →0 σ(1)(ω, T ) 6= 0 σ(ω, 0) = σ(0)(ω) + σ(1)(ω, 0). Calculations show that under the condition ∆ < 2µ we have [52] Reσ(ω, 0) = σ0θ(ω − 2µ) Imσ(ω, 0) = σ0 π (cid:20) 4µ ω − (ω)2 + ∆2 , (ω)2 (ω)2 + ∆2 (ω)2 ln(cid:12)(cid:12)(cid:12)(cid:12) ω + 2µ ω − 2µ(cid:12)(cid:12)(cid:12)(cid:12) (cid:21) . 12 (32) (33) (34) Similar to Eq. (9), this result is valid at the real, positive frequency axis. It is easily seen that in the limiting case ω → 0 one has where Imσ(ω, 0) = C(0) ω 2µ(cid:19) , + O(cid:18) ω C(0) = σ0 π (2µ)2 − ∆2 µ . (35) (36) The last equation is also obtainable as a particular case of Eq. (14) if one puts there T = 0. In so doing it is necessary to take into account that at T → 0 only the interval of v from unity to 2µ/∆ contributes to the integral. Taking into account that Imσ(ω, 0) has a pole at zero frequency, the Kramers-Kronig relations are given in this case by Eq. (24) where one should replace σ(ω, T ) with σ(ω, 0) and C(T ) with C(0). It is easily seen that both of them are satisfied. Really, substituting the first line of Eq. (34) in the right-hand side of the second Kramers-Kronig relation in Eq. (24) with C defined in Eq. (36) and introducing the variable ζ = ξ, one obtains − 2ω π σ0 π (2µ)2 − ∆2 µω ζ 2 − 2ω2 + ∆2−Z ∞ 2µ dζ ζ 2(ζ 2 − 2ω2)(cid:21) , −Z ∞ 0 2σ0 π σ0 π = − + Reσ(ξ, 0) ξ2 − ω2 dξ + ω(cid:20)−Z ∞ dζ 2µ (2µ)2 − ∆2 . µω Calculating the integrals in Eq. (37), we arrive at − = 0 2ω π σ0 −Z ∞ π (cid:20) 4µ ω Reσ(ξ, 0)dξ ξ2 − ω2 + (ω)2 + ∆2 − (ω)2 in accordance with Eq. (24). σ0 π (2µ)2 − ∆2 µω . ln(cid:12)(cid:12)(cid:12)(cid:12) (cid:21) = Imσ(ω, 0) ω + 2µ ω − 2µ(cid:12)(cid:12)(cid:12)(cid:12) (37) (38) Now we verify the first Kramers-Kronig relation in Eq. (24), when C(T ) is replaced with C(0) from Eq. (36). It is more illustrative to consider first the case ω 6= 0 when C(0) can be simply omitted due to Eq. (22). Substituting the second line of Eq. (34) in the right-hand side of the first Kramers-Kronig relation in Eq. (24), we find σ0 + 2 π −Z ∞ 0 ξImσ(ξ, 0) ξ2 − ω2 dξ = σ0 − 2σ0 π2 −Z ∞ 0 ξ ξ2 − ω2(cid:20)ln(cid:12)(cid:12)(cid:12)(cid:12) ξ + 2µ ξ − 2µ(cid:12)(cid:12)(cid:12)(cid:12) + ∆2 (ξ)2 ln(cid:12)(cid:12)(cid:12)(cid:12) 13 ξ + 2µ ξ − 2µ(cid:12)(cid:12)(cid:12)(cid:12) (cid:21) dξ. (39) The first integral on the right-hand side of this equation is calculated like in Eq. (29) with y = ξ/(2µ) and b = ω/(2µ) using Eq. (28). The result is given by Eq. (29) where ∆ is replaced with 2µ. The second integral is calculated like in Eq. (30). It is equal to zero when ω < 2µ and to −(π∆)2/(22ω2) when ω > 2µ. Substituting the values of both integrals in Eq. (39) and taking into account the first line in Eq. (34), one finds σ0 + 2 π −Z ∞ 0 = Reσ(ω, 0). ξImσ(ξ, 0) ξ2 − ω2 dξ =  0, σ0 ω < 2µ, (ω)2+∆2 (ω)2 , ω > 2µ. (40) Thus, with account of Eq. (22), the first Kramers-Kronig relation in Eq. (24) is proven for all ω 6= 0. At ω = 0 the validity of this Kramers-Kronig relation is achieved by the subtraction of C(0) in the first line of Eq. (39). To see this, we substitute Imσ(ω, 0) from Eq. (34) and C(0) from Eq. (36) in the right-hand side of the first Kramers-Kronig relation of Eq. (24) at ω = 0 and obtain σ0+ 2 π −Z ∞ 0 ξImσ(ξ, 0) − C(0) ξ2 dξ = σ0+ 2σ0 π2 −Z ∞ dξ(cid:20) ∆2 µξ2 − 0 1 ξ Calculation of all the three integrals on the right-hand side of this equation (see Appendix − ln(cid:12)(cid:12)(cid:12)(cid:12) ξ + 2µ ξ − 2µ(cid:12)(cid:12)(cid:12)(cid:12) ∆2 (ξ)3 ln(cid:12)(cid:12)(cid:12)(cid:12) ξ + 2µ ξ − 2µ(cid:12)(cid:12)(cid:12)(cid:12) (cid:21) . (41) B) results in σ0 + 2 π −Z ∞ 0 as it should be because in accordance to the first line of Eq. (34) ξImσ(ξ, 0) − C(0) ξ2 dξ = σ0 − σ0 = 0, Reσ(0, 0) = 0. (42) (43) This concludes the proof of the Kramers-Kroniog relations for the conductivity of graphene at zero temperature and validates the fact that expressions (9) for ∆ > 2µ and (34) for ∆ < 2µ satisfy the condition of causality. In some particular cases Eqs. (9) and (34) have been derived using various models and phenomenological approaches (see, for instance, Refs. [47, 62, 63]). Note, however, that the additional terms in the conductivity of graphene at zero temperature containing the δ-function of ω, which were obtained within some approaches (see, e.g., Refs. [47, 64]), are not obtainable in our formalism based on the 14 first principles of quantum electrodynamics. Such terms would violate the Kramers-Kronig relations and, thus, lead to contradiction with the principle of causality [65]. IV. KRAMERS-KRONIG RELATIONS AT NONZERO TEMPERATURE At first, we prove the validity of the Kramers-Kronig relations for the temperature- dependent part of the conductivity of graphene σ(1)(ω, T ) defined in Eq. (15). As usual, we start from the second Kramers-Kronig relation in Eq. (23). Substituting the first line of Eq. (15) in the right-hand side of the second Kramers-Kronig relation in Eq. (23), one obtains − 2ω π 0 −Z ∞ π2 −Z ∞ 2ωσ0 ∆ = Reσ(1)(ξ, T ) ξ2 − ω2 dξ + C(T ) ω (44) (ξ)2 + ∆2 ξ2(ξ2 − ω2) F (ξ, T )dξ + C(T ) ω . Introducing the new integration variable v = ξ/∆ and using the secons line of Eq. (15), we find − = Reσ(1)(ξ, T ) ξ2 − ω2 0 2ω π 2σ0 π −Z ∞ ω∆−Z ∞ 1 dv = Imσ(1)(ω, T ), dξ + C(T ) ω v2 + 1 v2[(v∆)2 − (ω)2] (45) F (cid:18) v∆ , T(cid:19) + C(T ) ω i.e., the second Kramers-Kronig relation in Eq. (23) is satisfied. Now we substitute the second line of Eq. (15) to the right-hand side of the first Kramers- Kronig relation in Eq. (23). Taking into account that in the second line of Eq. (15) the pole term is already separated, one can consider both cases ω 6= 0 and ω = 0 simultaneously. The result is dξ (46) J ≡ = = 0 2 π 4σ0 π2 4σ0 −Z ∞ ∆−Z ∞ π2 −Z ∞ 0 0 ξImσ(1)(ξ, T ) − C(T ) ξ2 − ω2 dv ξ2dξ ξ2 − ω2 −Z ∞ y2 − b2 −Z ∞ y2dy dv 1 1 , T(cid:1) v2[(v∆)2 − (ω)2] (v2 + 1)F (cid:0) v∆ F (cid:0) v∆ , T(cid:1) v2 − y2 , v2 + 1 v2 where the integration variable y = ξ/∆ was introduced and b = ω/∆. 15 Note that if b < 1, i.e., ω < ∆, then v 6= b holds over the entire integration region from unity to infinity. Taking into account that −Z ∞ 0 y2dy (y2 − b2)(y2 − v2) = 0 for b 6= v, (47) one immediately concludes that J = 0. It remains to consider the case b > 1, i.e., ω > ∆. To deal with this case, we present our integral (46) in the form J = − Integrating here by parts we find v + y v − y(cid:12)(cid:12)(cid:12)(cid:12) . (48) (49) 0 1 2σ0 0 1 4σ0 2σ0 y dy ∞ 1 y dy v + y J = − v2 + 1 y2 − b2 π2 −Z ∞ v + y v − y(cid:12)(cid:12)(cid:12)(cid:12) (cid:21)(cid:12)(cid:12)(cid:12)(cid:12) , T(cid:19) ln(cid:12)(cid:12)(cid:12)(cid:12) d(cid:20)v2 + 1 , T(cid:19)(cid:21)(cid:27) . v2 F (cid:18)v∆ v2 F (cid:18) v∆ , T(cid:19) d ln(cid:12)(cid:12)(cid:12)(cid:12) y2 − b2 −Z ∞ π2 −Z ∞ ×(cid:26)(cid:20)v2 + 1 v2 F (cid:18)v∆ −−Z ∞ v − y(cid:12)(cid:12)(cid:12)(cid:12) ln(cid:12)(cid:12)(cid:12)(cid:12) , T(cid:19) −Z ∞ 1 − y(cid:12)(cid:12)(cid:12)(cid:12) y2 − b2 ln(cid:12)(cid:12)(cid:12)(cid:12) π2 F (cid:18) ∆ π2 −Z ∞ y2 − b2 −Z ∞ d(cid:20)v2 + 1 ln(cid:12)(cid:12)(cid:12)(cid:12) v − y(cid:12)(cid:12)(cid:12)(cid:12) v2 F (cid:18)v∆ , T(cid:19)(cid:21) −Z ∞ v2 F (cid:18)v∆ , T(cid:19)(cid:21) −Z ∞ π2 −Z ∞ π2 −Z ∞ d(cid:20)v2 + 1 v2 F (cid:18) v∆ d(cid:20)v2 + 1 y dy y2 − b2 ln(cid:12)(cid:12)(cid:12)(cid:12) ln(cid:12)(cid:12)(cid:12)(cid:12) t2 − b2 t dt v + y 1 + y y dy 0 1 y dy 0 0 0 1 v + y v − y(cid:12)(cid:12)(cid:12)(cid:12) t − 1(cid:12)(cid:12)(cid:12)(cid:12) t + 1 , J = 2σ0 + 2σ0 J = 2σ0 J = 1 (50) , T(cid:19)(cid:21) . . (51) (52) Taking into account that in accordance to Eq. (12) F (x, T ) → 0 when x → ∞, Eq. (49) leads to The first integral on the right-hand side of this equation is equal to zero due to Eq. (28) and, changing the integration order with respect to y and v, we have Now we introduce the integration variable t = y/v in the last integral and obtain where b = b/v can be both larger and less than unity. According to Eq. (28), the last integral on the right-hand side of Eq. (52) is equal to zero if b > 1 (i.e., v < b) and to π2/2 if b < 1 (i.e., v > b). As a result, Eq. (52) is simplified to J = σ0−Z ∞ = −σ0 b b2 + 1 v2 F (cid:18)v∆ d(cid:20)v2 + 1 b2 F (cid:18)b∆ , T(cid:19)(cid:21) , T(cid:19) = −σ0 (53) (ω)2 + ∆2 (ω)2 F (ω, T ). 16 Combining together the results for b < 1 (i.e., ω < ∆) and b > 1 (i.e., ω > ∆) and using the first line of Eq. (11), we conclude from Eq. (46) that J = −σ0θ(ω − ∆) (ω)2 + ∆2 (ω)2 F (ω, T ) = Reσ(1)(ω, T ), (54) i.e., the first Kramers-Kronig relation in Eq. (23) is satisfied. The total conductivity of graphene at nonzero temperature is given by Eq. (8). Using Eqs. (9) and (15), one obtains Reσ(ω, T ) = σ0θ(ω − ∆) (ω)2 + ∆2 (ω)2 [1 − F (ω, T )], (55) Imσ(ω, T ) = σ0 π (cid:26)(cid:20)2∆ + πC(T ) σ0 v2 + 1 v2[(v∆)2 − (ω)2] +2ω∆−Z ∞ 1 dv (ω)2 + ∆2 ω − (cid:21) 1 F (cid:18)v∆ (ω)2 , T(cid:19)(cid:27) . ln(cid:12)(cid:12)(cid:12)(cid:12) ω + ∆ ω − ∆(cid:12)(cid:12)(cid:12)(cid:12) Note that Reσ(ω, T ) can be rewritten in especially simple and transparent equivalent form. For this purpose we use the definition of F in Eq. (12) and the following identity The result is 1 2 − 1 ey + 1 = 1 2 tanh y 2 . Reσ(ω, T ) = σ0θ(ω − ∆) (ω)2 + ∆2 2(ω)2 ×(cid:18)tanh ω + 2µ 4kBT + tanh ω − 2µ 4kBT (cid:19) . (56) (57) As to Imσ(ω, T ), simple asymptotic expressions for it in different regions of parameters and the results of numerical computations can be found in Refs. [50–52]. Although in the general case of gapped graphene with nonzero chemical potential Eqs. (55) and (57) were derived in Ref. [52], in different special cases similar dependences have been obtained previously using various approaches based on the Kubo formalism and two-dimensional Drude model (see, e.g., Refs. [66–71]). The Kramers-Kronig relations (24) for the total conductivity of graphene (55) are satisfied automatically, because they are obtained by the combination of already proven Kramers- Kronig relations (21) and (23) satisfied for σ(0)(ω) and σ(1)(ω, T ), respectively. 17 V. CONCLUSIONS AND DISCUSSION In the foregoing, we have investigated the problem of causality for the conductivity of graphene in the framework of the Dirac model. Until recently, only some partial results for the conductivity of graphene have been obtained using some models and phenomenological approaches. To investigate the problem of causality, we use the complete results for the spatially local conductivity found on the basis of first principles of thermal quantum field theory using the polarization tensor of graphene in (2+1)-dimensional space-time [50–52]. The spatially nonlocal corrections to these results were shown to be of the order of 10−5 of the local contributions and, thus, are of no physical significance in the framework of Dirac's model. General discussion of causality presented in the paper leads to the conclusion that both the total conductivity of graphene and contributions to it σ(0)(ω), depending on the band gap, and σ(1)(ω, T ), depending on the band gap and chemical potential, are the analytic functions in the upper half-plane of complex frequencies and possess all the standard sym- metry properties. Hence it follows that the real and imaginary parts of the conductivity of graphene derived in any specific formalism must satisfy the Kramers-Kronig relations. The form of these relations, as shown above, depends on the presence of a pole at zero frequency and takes into account an existence of the universal conductivity. There is no pole for the conductivity of graphene at zero temperature under the condition that the band gap is larger than twice the chemical potential, and there is such a pole in all remaining cases. The fulfilment of the Kramers-Kronig relations can be considered as a basic guideline in deciding which specific expression for the conductivity of graphene is correct. We have shown through the direct analytic calculations that the real and imaginary parts of the conductivity of graphene, found in Ref. [52] in the most general case of nonzero temperature, band gap and chemical potential on the basis of first principles of thermal quantum field theory, satisfy both Kramers-Kronig relations precisely. In the process, the values of two important integrals in the widely used tables have been corrected, which might be useful in the context of dispersion relations for the scattering amplitudes in quantum field theory. One can conclude that the obtained results are not of only fundamental theoretical character, but they also open fresh opportunities for the use of Kramers-Kronig relations in different fields of physics and for the interpretation of experimental data. 18 (A1) (A2) , . I+(b) = −Z ∞ I−(b) = −Z ∞ 0 0 dy dy y + b y − b ln(cid:12)(cid:12)(cid:12)(cid:12) ln(cid:12)(cid:12)(cid:12)(cid:12) y + 1 y + 1 y − 1(cid:12)(cid:12)(cid:12)(cid:12) y − 1(cid:12)(cid:12)(cid:12)(cid:12) Acknowledgments The work of V.M.M. was partially supported by the Russian Government Program of Competitive Growth of Kazan Federal University. Appendix: A Here, we calculate the integral (28) and correct relevant integrals in Ref. [58] which are important for various applications in a wide context of dispersion relations in different branches of physics. The integral in Eq. (28) can be presented in the form I(b) = 1 2 [I+(b) + I−(b)], where We consider the case b ≥ 0, b 6= 1. It is easily seen that the integrals in Eq. (A2) converge at the points y = 1, y = b. Integrating by parts in Eq. (A2), one obtains I±(b) = 2−Z ∞ 0 dy ln y ± b y2 − 1 , (A3) where the out-of-integral terms vanish and the lower indices ± correspond to plus and minus on the right-hand side, respectively. From Eq. (A3) we find the derivative of I± with respect to b dI±(b) db = ±2−Z ∞ 0 dy 1 (y ± b)(y2 − 1) . Calculating this integral, we obtain the result dI±(b) db = ±2 ln b 1 − b2 . From this it follows that dI(b) db = 1 2(cid:20) dI+(b) db + dI−(b) db (cid:21) = 0, 19 (A4) (A5) (A6) i.e., I(b) takes the constant values in the intervals [0,1) and (1,∞), where it is a continuous function. Let us consider first the interval [0,1) and find the values I+(0) = I−(0) = −Z ∞ =Z 1 0 1 + y 1 − y dy y dy y ln(cid:12)(cid:12)(cid:12)(cid:12) +Z ∞ ln 0 1 y + 1 y − 1(cid:12)(cid:12)(cid:12)(cid:12) ln dy y y + 1 y − 1 . (A7) Changing the integration variable according to y = 1/x in the second integral on the right- hand side of Eq. (A7), one obtains I±(0) = 2Z 1 0 dy y ln 1 + y 1 − y = 4 ∞ Xk=1 1 2k − 1Z 1 0 y2k−2dy. Calculating this integral and taking into account that [72] we find 1 (2k − 1)2 = π2 8 , ∞ Xk=1 I±(0) = π2 2 . (A8) (A9) (A10) Then from Eq. (A1) it follows that I(b) = π2/2 under the condition 0 ≤ b < 1 in agreement with the first line of Eq. (28). Note that it is also possible now to find the values of integrals I±(b) at any b. By integrating Eq. (A5) with respect to b for b < 1, we have I±(b) = ± ln b ln 1 + b 1 − b ∓ Li2(b) ± Li2(−b) + π2 2 , (A11) where Lin(x) is the polylogarithm function. This equation can be checked by differentiation taking into account that dLi2(±b) db = − ln(1 ∓ b) b . (A12) The value of the arbitrary integration constant in Eq. (A11), C = π2/2, is determined from Eq. (A10) taking into account that Li2(0) = 0. The result (A11) is in disagreement with the formula 2.6.14.27 of Ref. [58] where the independent on b value of the integrals I±(b) equal to π is indicated leading to an incorrect result I(b) = π. This formula is also in contradiction with the formula 2.6.14.24. The latter is in agreement with our result (A10). 20 Now we consider the case when b varies in the interval (1, ∞), where the dilogarithm function has a cut. Using Eq. (A12), one can easily check that the integration of Eq. (A5) (A13) results in I±(b) = ± ln b ln b + 1 b − 1 ± Li2(cid:18)1 b(cid:19) ∓ Li2(cid:18)− 1 b(cid:19) . The integration constant C = 0 is found from the fact that I±(b) → 0 when b → ∞. From Eqs. (A1) and (A13) we have I(b) = 0 over the entire interval (1, ∞) which concludes the proof of Eq. (28). The result (A13) contradicts to the formula 2.6.14.26 of Ref. [58], where instead of Eq. (A13) an incorrect value I±(b) = 0 is indicated. Appendix: B Here, we calculate the integrals contained in Eq. (41). Introducing the new variable y = ξ/(2µ), the right-hand side of Eq. (41) takes the form where I1 = σ0 + 2σ0 π2 (I1 − I2), ∆2 (2µ)2 −Z ∞ 0 (cid:18) 2 y2 − ln(cid:12)(cid:12)(cid:12)(cid:12) y − 1(cid:12)(cid:12)(cid:12)(cid:12) y + 1 dy y I2 = −Z ∞ 0 1 y3 ln(cid:12)(cid:12)(cid:12)(cid:12) . y + 1 y − 1(cid:12)(cid:12)(cid:12)(cid:12) (cid:19) dy, (B1) (B2) Using Eq. (28) with b = 0, which is proven in Appendix A, one obtains I2 = π2/2. Because of this below we consider only I1. It is easily seen that this integral converges at y = 0. Really, for y < 1 it holds ln(cid:12)(cid:12)(cid:12)(cid:12) y + 1 y − 1(cid:12)(cid:12)(cid:12)(cid:12) In a similar way, for y > 1 one obtains ln(cid:12)(cid:12)(cid:12)(cid:12) y + 1 y − 1(cid:12)(cid:12)(cid:12)(cid:12) = ln 1 + y 1 − y = 2 y2k−1 2k − 1 ∞ Xk=1 y2k+1 2k + 1 ∞ Xk=1 = ln 1 + 1 y 1 − 1 y 1 (2k − 1)y2k−1 . = ln y + 1 y − 1 ∞ = 2 Xk=1 21 Substituting Eqs. (B3) and (B4) in the first line of Eq. (B2), we find = 2y + 2 . (B3) (B4) I1 = = − ∆2 (2µ)2 (cid:20)Z 1 0 (cid:18) 2 y2 − (2µ)2 (Z 1 Xk=1 2∆2 ∞ 0 1 y3 ln y2k−2 2k + 1 1 + y y + 1 1 − y(cid:19) dy +Z ∞ y − 1(cid:19) dy(cid:21) 1 (cid:18) 2 y2 − (2k − 1)y2k+2# dy) . 1 " 1 dy −Z ∞ Xk=1 y2 − 1 y3 ln 1 ∞ Integrating on the right-hand side of this equation, one arrives at I1 = − 2∆2 (2µ)2" ∞ Xk=1 1 4k2 − 1 − 1 + 1 4k2 − 1# . ∞ Xk=1 (B5) (B6) (B7) Taking into account that [72] ∞ 1 4k2 − 1 = 1 2 , Xk=1 we finally obtain that I1 = 0. Substituting the values of both I1 and I2 in Eq. (B1), one finds in accordance with Eq. (42). σ0 + 2σ0 π2 (I1 − I2) = 0 (B8) [1] M. I. Katsnelson, Graphene: Carbon in Two Dimensions (Cambridge University Press, Cam- bridge, 2012). [2] Physics of Graphene, ed. H. Aoki and M. S. Dresselhaus (Springer, Cham, 2014). [3] 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, 109 (2009). [4] M. I. Katsnelson, K. S. Novoselov, and A. K. Geim, Chiral tunneling and the Klein paradox in graphene, Nature Phys. 2, 620 (2006). [5] D. Allor, T. D. Cohen, and D. A. McGady, Schwinger mechanism and graphene, Phys. Rev. D 78, 096009 (2008). [6] C. G. Beneventano, P. Giacconi, E. M. Santangelo, and R. Soldati, Planar QED at finite tem- perature and density: Hall conductivity, Berry's phases and minimal conductivity of graphene, J. Phys. A 42, 275401 (2009). [7] G. L. Klimchitskaya and V. M. Mostepanenko, Creation of quasiparticles by a time-dependent electric field, Phys. Rev. D 87, 125011 (2013). 22 [8] I. Akal, R. Egger, C. Muller, and S. Villarba-Ch´avez, Low-dimensional approach to pair production in an oscillating electric field: Application to bandgap graphene layers, Phys. Rev. D 93, 116006 (2016). [9] M. O. Goerbig, Electronic properties of graphene in a strong magnetic field, Rev. Mod. Phys. 83, 1193 (2011). [10] P. K. Pyatkovsky, Dynamical polarization, screening, and plasmons in gapped graphene, J. Phys.: Condens. Matter 21, 025506 (2009). [11] M. Bordag, I. V. Fialkovsky, D. M. Gitman, and D. V. Vassilevich, Casimir interaction between a perfect conductor and graphene described by the Dirac model, Phys. Rev. B 80, 245406 (2009). [12] I. V. Fialkovsky, V. N. Marachevsky, and D. V. Vassilevich, Finite-temperature Casimir effect for graphene, Phys. Rev. B 84, 035446 (2011). [13] M. Bordag, G. L. Klimchitskaya, and V. M. Mostepanenko, Thermal Casimir effect in the interaction of graphene with dielectrics and metals, Phys. Rev. B 86, 165429 (2012). [14] M. Chaichian, G. L. Klimchitskaya, V. M. Mostepanenko, and A. Tureanu, Thermal Casimir- Polder interaction of different atoms with graphene, Phys. Rev. A 86, 012515 (2012). [15] G. L. Klimchitskaya and V. M. Mostepanenko, Van der Waals and Casimir interactions be- tween two graphene sheets, Phys. Rev. B 87, 075439 (2013). [16] B. Arora, H. Kaur, and B. K. Sahoo, C3 coefficients for the alkali atoms interacting with a graphene and carbon nanotube, J. Phys. B 47, 155002 (2014). [17] K. Kaur, J. Kaur, B. Arora, and B. K. Sahoo, Emending thermal dispersion interaction of Li, Na, K and Rb alkali-metal atoms with graphene in the Dirac model, Phys. Rev. B 90, 245405 (2014). [18] G. L. Klimchitskaya and V. M. Mostepanenko, Classical Casimir-Polder force between polar- izable microparticles and thin films including graphene, Phys. Rev. A 89, 012516 (2014). [19] G. L. Klimchitskaya and V. M. Mostepanenko, Classical limit of the Casimir interaction for thin films with applications to graphene, Phys. Rev. B 89, 035407 (2014). [20] G. L. Klimchitskaya and V. M. Mostepanenko, Observability of thermal effects in the Casimir interaction from graphene-coated substrates, Phys. Rev. A 89, 052512 (2014). [21] G. L. Klimchitskaya and V. M. Mostepanenko, Impact of graphene coating on the atom-plate interaction, Phys. Rev. A 89, 062508 (2014). 23 [22] G. L. Klimchitskaya, V. M. Mostepanenko, and Bo E. Sernelius, Two approaches for describing the Casimir interaction with graphene: density-density correlation function versus polarization tensor, Phys. Rev. B 89, 125407 (2014). [23] G. L. Klimchitskaya, U. Mohideen, and V. M. Mostepanenko, Theory of the Casimir in- teraction for graphene-coated substrates using the polarization tensor and comparison with experiment, Phys. Rev. B 89, 115419 (2014). [24] J. F. Dobson, A. White, and A. Rubio, Asymptotics of the dispersion interaction: Analytic benchmarks for van der Waals energy functionals, Phys. Rev. Lett. 96, 073201 (2006). [25] G. G´omez-Santos, Thermal van der Waals interaction between graphene layers, Phys. Rev. B 80, 245424 (2009). [26] D. Drosdoff and L. M. Woods, Casimir forces and graphene sheets, Phys. Rev. B 82, 155459 (2010). [27] D. Drosdoff and L. M. Woods, Casimir interaction between graphene sheets and metamaterials, Phys. Rev. A 84, 062501 (2011). [28] Bo E. Sernelius, Casimir interactions in graphene systems, Europhys. Lett. 95, 57003 (2011). [29] T. E. Judd, R. G. Scott, A. M. Martin, B. Kaczmarek, and T. M. Fromhold, Quantum reflection of ultracold atoms from thin films, graphene and semiconductor heterostructures, New J. Phys. 13, 083020 (2011). [30] J. Sarabadani, A. Naji, R. Asgari, and R. Podgornik, Many-body effects in the van der Waals- Casimir interaction between graphene layers, Phys. Rev. B 84, 155407 (2011); Phys. Rev. B 87, 239905(E) (2013). [31] D. Drosdoff, A. D. Phan, L. M. Woods, I. V. Bondarev, and J. F. Dobson, Effects of spatial dispersion on the Casimir force between graphene sheets, Eur. Phys. J. B 85, 365 (2012). [32] Bo E. Sernelius, Retarded interactions in graphene systems, Phys. Rev. B 85, 195427 (2012). [33] A. D. Phan, L. M. Woods, D. Drosdoff, I. V. Bondarev, and N. A. Viet, Temperature depen- dent graphene suspension due to thermal Casimir interaction, Appl. Phys. Lett. 101, 113118 (2012). [34] A. D. Phan, N. A. Viet, N. A. Poklonski, L. M. Woods, and C. H. Le, Interaction of a graphene sheet with a ferromagnetic metal plate, Phys. Rev. B 86, 155419 (2012). [35] N. Knusnutdinov, R. Kashapov, and L. M. Woods, Casimir-Polder effect for a stack of con- ductive planes, Phys. Rev. A 94, 012513 (2016). 24 [36] M. Bordag, G. L. Klimchitskaya, V. M. Mostepanenko, and V. M. Petrov, Quantum field theoretical description for the reflectivity of graphene, Phys. Rev. D 91, 045037 (2015); 93, 089907(E) (2016). [37] G. L. Klimchitskaya and V. M. Mostepanenko, Origin of large thermal effect in the Casimir interaction between two graphene sheets, Phys. Rev. B 91, 174501 (2015). [38] G. L. Klimchitskaya, Quantum field theory of the Casimir force for graphene, Int. J. Mod. Phys. A 31, 1641026 (2016). [39] V. B. Bezerra, G. L. Klimchitskaya, V. M. Mostepanenko, and C. Romero, Nernst heat theorem for the thermal Casimir interaction between two graphene sheets, Phys. Rev. A 94, 042501 (2016). [40] G. Bimonte, G. L. Klimchitskaya, and V. M. Mostepanenko, How to observe the giant thermal effect in the Casimir force for graphene systems, Phys. Rev. A 96, 012517 (2017). [41] C. Henkel, G. L. Klimchitskaya, and V. M. Mostepanenko, Influence of chemical potential on the Casimir-Polder interaction between an atom and gapped graphene or graphene-coated substrate, Phys. Rev. A 97, 032504 (2018). [42] G. L. Klimchitskaya, C. C. Korikov, and V. M. Petrov, Theory of reflectivity properties of graphene-coated material plates, Phys. Rev. B 92, 125419 (2015); 93, 159906(E) (2016). [43] G. L. Klimchitskaya and V. M. Mostepanenko, Reflectivity properties of graphene with nonzero mass-gap parameter, Phys. Rev. A 93, 052106 (2016). [44] G. L. Klimchitskaya and V. M. Mostepanenko, Optical properties of dielectric plates coated with gapped graphene, Phys. Rev. B 95, 035425 (2017). [45] M. Bordag, I. Fialkovskiy, and D. Vassilevich, Enhanced Casimir effect for doped graphene, Phys. Rev. B 93, 075414 (2016); 95, 119905(E) (2017). [46] G. Bimonte, G. L. Klimchitskaya, and V. M. Mostepanenko, Thermal effect in the Casimir force for graphene and graphene-coated substrates: Impact of nonzero mass gap and chemical potential, Phys. Rev. B 96, 115430 (2017). [47] V. P. Gusynin, S. G. Sharapov, and J. P. Carbotte, AC conductivity of graphene: From tight- binding model to 2+1-dimensional quantum electrodynamics, Int. J. Mod. Phys. B 21, 4611 (2007). [48] N. M. R. Peres, The transport properties of graphene: An introduction, Rev. Mod. Phys. 82, 2673 (2010). 25 [49] S. Das Sarma, S. Adam, E. H. Hwang, and E. Rossi, Electronic transport in two-dimensional graphene, Rev. Mod. Phys. 83, 407 (2011). [50] G. L. Klimchitskaya and V. M. Mostepanenko, Conductivity of pure graphene: Theoretical approach using the polarization tensor, Phys. Rev. B 93, 245419 (2016). [51] G. L. Klimchitskaya and V. M. Mostepanenko, Quantum electrodynamic approach to the conductivity of gapped graphene, Phys. Rev. B 94, 195405 (2016). [52] G. L. Klimchitskaya, V. M. Mostepanenko, and V. M. Petrov, Conductivity of graphene in the framework of Dirac model: Interplay between nonzero mass gap and chemical potential, Phys. Rev. B 96, 235432 (2017). [53] D. Liu and S. Zhang, Kramers-Kronig relation of graphene conductivity, J. Phys.: Condens. Matter 20, 175222 (2008). [54] M. Jablan, H. Buljan, and M. Soljaci´c, Transverse electric plasmons in bilayer graphene, Optics Express 19, 11236 (2011). [55] J. Horng, Chi-Fan 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, Drude conductivity of Dirac fermions in graphene, Phys. Rev. B 83, 165113 (2011). [56] V. U. Nazarov, Negative static permittivity and violation of Kramers-Kronig relations in quasi-two-dimensional crystals, Phys. Rev. B 92, 161402(R) (2015). [57] L. D. Landau, E. M. Lifshitz, and L. P. Pitaevskii, Electrodynamics of Continuous Media (Pergamon, Oxford, 1984). [58] A. P. Prudnikov, Yu. A. Brychkov, and O. I. Marichev, Integrals and Series. Vol.1. Elementary Functions (Gordon and Breach, New York, 1986). [59] A. I. Akhiezer and V. B. Berestetskii, Quantum Electrodynamics (Interscience, New York, 1965). [60] S. S. Schweber, An Introduction to Relativistic Quantum Field Theory (Dover, New York, 2005). [61] Handbook of Mathematical Functions with Formulas, Graphs and Mathematical Tables, eds. M. Abramowitz and I. A. Stegun (Dover Publications, New York, 2012). [62] T. G. Pedersen, A.-P. Jauho, and K. Pedersen, Optical response and excitons in gapped graphene, Phys. Rev. B 79, 113406 (2009). [63] T. Stauber, Plasmonics in Dirac systems: from graphene to topological insulators, J. Phys.: 26 Condens. Matter 26, 123201 (2014). [64] V. P. Gusynin, S. G. Sharapov, and J. P. Carbotte, Magneto-optical conductivity in graphene, J. Phys.: Condens. Matter 19, 026222 (2007). [65] G. L. Klimchitskaya and V. M. Mostepanenko, Comment on "Lifshitz-Matsubara sum formula for the Casimir pressure between magnetic metallic mirrors", Phys. Rev. E 94, 026101 (2016). [66] S. Ryu, C. Mudry, A. Furusaki, and A. W. W. Ludwig, Landauer conductance and twisted boundary conditions for Dirac fermions in two space dimensions, Phys. Rev. B 75, 205344 (2007). [67] L. A. Falkovsky and S. S. Pershoguba, Optical far-infrared properties of a graphene monolayer and multilayer, Phys. Rev. B 76, 153410 (2007). [68] L. A. Falkovsky and A. A. Varlamov, Space-time dispersion of graphene conductivity, Eur. Phys. J. B 56, 281 (2007). [69] T. Stauber, N. M. R. Peres, and A. K. Geim, Optical conductivity of graphene in the visible region of the spectrum, Phys. Rev. B 78, 085432 (2008). [70] K. F. Mak, M. Y. Sfeir, Y. Wu, C. H. Lui, J. A. Misewich, and T. F. Heinz, Measurement of the optical conductivity of graphene, Phys. Rev. Lett. 101, 196405 (2008). [71] L. A. Falkovsky, Optical properties of graphene, J. Phys.: Conf. Series 129, 012004 (2008). [72] I. S. Gradshtein and I. M. Ryzhik, Table of Integrals, Series and Products (Academic Press, New York, 1980). 27
1208.3760	1	1208	2012-08-18T16:13:37	Geometrical-induced rectification in two-dimensional ballistic nanodevices	[ "cond-mat.mes-hall" ]	The paper demonstrates that a two-dimensional ballistic nanodevice in which the electron gas satisfies either the Schroodinger equation (as in quantum wells in common semiconductor heterostructures) or the Dirac equation (as in graphene) is able to rectify an electric signal if the device has a non-uniform cross section, for instance a taper configuration. No p-n junctions or dissimilar electrodes are necessary for rectification.	cond-mat.mes-hall	cond-mat	1 GEOMETRICAL-INDUCED RECTIFICATION IN TWO-DIMENSIONAL BALLISTIC NANODEVICES Daniela Dragoman1(a) and Mircea Dragoman2 1Univ. Bucharest, Physics Dept., P.O. Box MG-11, 077125 Bucharest, Romania 2National Research and Development Institute in Microtechno logies, Str. Erou Iancu Nicolae 126 A , 077190 Bucharest, Romania The paper demonstrates that a two-dimensional ballist ic nanodevice in which the electron gas satisfies either the Schrödinger equation (as in quantum wells in common semiconductor heterostructures) or the Dirac equat ion (as in graphene) is able to rectify an electric signal if the device has a non-uniform cross sect ion, for instance a taper configurat ion. No p-n junct ions or dissimilar electrodes are necessary for rectification. (a) Corresponding author email: [email protected] 1. Introduction 2 The rect ificat ion process transforms an alternative signal into a DC signal. Rect ificat ion is a key signal processing tool in many applications, ranging from power supplies up to high- frequency detectors, and is present nowadays in smart phones, computers, TV-sets and radios wireless communications. Rectificat ion is implemented using a semiconduct ing p-n junct ion obtained by chemical doping. However, in two-dimensional electron gas (2DEG) devices based on semiconduct ing heterostructures, carbon nanotubes or graphene, chemical doping becomes difficult and is replaced by electrostatic doping performed by mult iple metallic electrodes polarized with different DC vo ltages [1-3]. Also, in CNTs rect ificat ion can be achieved in Schottky contacts implemented with the help o f dissimilar or asymmetric electrodes made from different metals [4,5]. However, metallic gates on nanostructures complicate very much the architecture of rectifying devices, use extensive nano lithography techniques, and quite often introduce parasit ic effects in the gated rectifying nanodevices. Therefore, it is very important to be able to rectify alternative signals based on a different mechanism than chemical doping or Schottky contacts. Such a different mechanism is the geometrical asymmetry in 2DEG nanodevices with non-uniform cross-sect ions, for example, in taper configurations. Tailoring the shape of semiconductor channels was used already in self-switching devices, but was not related to ballist ic transport [6], while a unipo lar diode that acts as room-temperature THz detector based on symmetry breaking o f a semiconduct ing channel was demonstrated recent ly [7]. Geometrical asymmetries are used for metal-insulator-metal diodes working in a non-ballist ic regime to enhance the sensit ivity o f electromagnet ic wave detect ion at 900 MHz, which remains, however, quite low [8]. Recent ly, a graphene trapezoidal-shaped diode deposited on Si/SiO2 and coupled to a bowtie metallic antenna was shown to detect an electromagnetic radiat ion at 28 THz generated by a CO2 laser [9]. However, the rectification properties o f this diode are very poor and its nonlinearit ies are vey weak even at gate vo ltages o f 40 V. These poor results are caused by the high losses of the 3 doped Si substrate, which appear even at few GHz [10], and by the non-ballist ic transport o f charge carriers. At room temperature, the ballist ic transport in graphene occurs for a mean-free path of 0.4 m, reaching an intrinsic mobility o f 44 000 cm2V-1s-1 [11], mean-free carrier paths longer than 1 m and mobilit ies higher than 100 000 cm2/s being observed only in the exceptional case when graphene is deposited on a hexagonal boron nitride substrate [12] matching the graphene lattice. So, the role of this paper is to investigate the rectifying properties of a 2DEG nanodevice with a tapered geometry working in the ballist ic regime, where the transport of charge carriers is described by either the Schrödinger or the Dirac equat ion. We show that such a device can rect ify alternat ive electrical signals under certain condit ions, at room temperature, thus paving the way for generat ion o f THz signals via mult iplicat ion and for detection o f THz signals used in imaging applications. 2. Signal rectification of a tapered 2DEG satisfying the Schrödinger equation A 2DEG sat isfying the Schrödinger equat ion is in fact a quantum well formed in a common semiconductor heterostructure. A schematic representation of a tapered 2DEG connected at two metallic electrodes is displayed in Fig. 1; we assume throughout this paper that d  in d out . Rect ificat ion occurs for current flowing along the x direct ion. We assume that the boundaries of the tapered 2DEG act as regions with an infinite energy potential and model the transport o f charge carriers by replacing the taper with a number o f N regions with constant but different widths shown as rectangles with dotted boundaries in Fig. 1. In each region of width jd , j = in,1,…,N,out, the wavevector component along y has discrete values jd n / , so that the wavevector component along x is equal to k  jn , VEm (2  2 /)  j n (  / d 2 ) j , where m and E are the effect ive mass and energy o f the electron, respectively, and jV is the potential energy 4 in region j. A solut ion of the Schrödinger equation corresponding to a given n value for which jnk , is real is referred to as a mode (no charge transport takes place for an imaginary jnk , since tunneling place no role in this device); the number o f modes in the jth region is N j  Int [ d j VEm (2  /) j ] , where Int[] denotes the integer part and d j  d in  ( d in  d out ) Nj /(  )1 , for j = 1,…,N. We assume throughout this paper that a bias vo ltage V is applied across the tapered structure, such that 0inV , Vout  eV , and V j  jeV N /(  )1 , j = 1…N. The rect ificat ion principle can be understood from Fig. 2, which shows the variat ion o f the ratio of outgoing and incoming mode numbers with V for three energy values: E = 0.1 eV (green dashed line), 0.2 eV (so lid black line) and 0.3 eV (red dotted line). We consider throughout this paper ind = 50 nm, outd = 10 nm and L = 100 nm. Figure 2 shows that the number o f emerging modes, and hence the current, which is proportional to the mode number in the Landauer theory, depends on the applied voltage: the current is high for one polarity and decreases to zero for the other polarity, as the number of outgoing modes decreases. A more refined analys is o f charge transport in the tapered 2DEG must take into account that scattering occurs between different modes in quantum waveguides with a non-uniform cross-section. Therefore, the transmission probability is determined by the requirement of flux conservation and is given by T N out  n 1  k n , out A n out , 2 N in k  n 1  A inn , 2 inn , (1) where the coefficients jnA , , j = in,out represent the amplitudes o f the forward-propagating components o f the wavefunct ion in the incidence and outgoing media, respect ively, and are found by imposing the continuity condit ions for the wavefunct ion 5 A jn , exp( ik jn , x )  B jn , exp( ik  jn , x )] sin( 2 dyn /  j ) (2)  j yx ,( ) N j [  n 1  and it s derivat ive with respect to x at each interface between different regions (for details on electron propagation in waveguides with non-unifo rm cross-section, see [13,14]). The vo ltage dependence o f the transmission coefficient for E = 0.2 eV is represented in Fig. 3 for the case of a step interface between the incoming and outgoing media, i.e. for N = 0 (red dotted line), and for N = 1 (solid black line) and 2 (green dashed line). As N increases, the transmission coefficient resembles the smooth curve represented with magenta dashed-dotted line in Fig. 3. As the number of modes, the transmission through the tapered configurat ion depends asymmetrically on the applied vo ltage: it takes high values for posit ive biases and low values for negat ive biases, indicat ing no current transport for V values for which there are no propagating outgoing modes. As suggested by Fig. 2, the bias vo ltage for which T increases sharply increases as the energy decreases. The current, calculated with the Landauer formula, is represented in Fig. 4 for three values o f the Fermi energy level: FE = 0 (green dashed line), 0.1 eV (so lid black line) and 0.2 eV (red dotted line). Although current rectificat ion occurs in all cases, the shape of the I-V characterist ic depends on the posit ion of the Fermi level. A low FE corresponds to lower energy values o f the charge carrying electrons, for which T can have significant values for only one polarity (see the green dashed line), while fo r high enough Fermi levels the transmissio n probability has a finite value at V = 0 (see also Fig. 3), so that current flows also for a limited range of negat ive V values. Therefore, to have current flow only for one polarizat ion, one must choose FE appropriately. 6 3. Signal rectification of a tapered graphene sheet Graphene is a natural 2DEG, in which the charge carriers sat isfy a massless Dirac equation. As a result, an infinite graphene sheet has no bandgap, although a finite bandgap exists in narrow graphene strips, called also nanoribbons. In our tapered configuration, there is a small bandgap because the spinorial wavefunction cannot extend outside the taper, and hence such a taper could rect ify an electrical signal. However, unlike for Schrödinger-type charge carriers, rectification is possible only for electrical signals with small enough amplitudes. The mathemat ical treatment of the graphene taper parallels that of the previous Sect ion. In particular, the taper is again modeled as a succession o f N regions with widths jd , j = in,1,…,N,out, so that the wavevector component along y is again discrete and given by the same expression as in the previous sect ion, the wavenumbers along x being now k jn ,  sgn( VE  ) j ( VE  2 ) /(  j 22 v F ) n (  / d 2 ) j [15], where v F  /c 300 is the Fermi velocity. If sgn( jVE  ) is posit ive the electrical charge is carried by electrons, otherwise by ho les. In the tapered waveguide charge propagation occurs for real jnk , values, the number of modes in the jth region being now given by N  [ Int d VE  /(  v F j j j )] . The expressions o f jd and jV are the same as in the previous case. In the graphene taper, with ident ical dimensions as those considered in Sect ion 2, the vo ltage dependence of the ratio of outgoing and incoming mode numbers is represented in Fig. 5 for E = 0.1 eV (green dashed line), 0.2 eV (solid black line) and 0.3 eV (red dotted line). Although there are a finite number of outgoing modes for both vo ltage polarizations, current rectification can be achieved if we explo it the small vo ltage range in which outN , and hence T vanishes. The width of this vo ltage range is given by v / F d  out : it increases as outd decreases. Please note that this vo ltage range does not depend on the energy of the incident electrons, but shifts to lower values as the energy increases. The transmission coefficient is now defined as 7 N out  n 1  A n , out 2 N in  n 1  A inn , 2 (3) T where the coefficients where the coefficients jnA , , j = in,out are calculate by imposing the continuity condit ion at each interface between different regions for the two components of the spinorial wavefunct ion  j yx ,( )  N j [  n 1  N j [  n 1        A jn , exp( ik jn , x )  B jn , exp( ik  jn , x )] sin( 2 dyn /  A jn , exp( ik jn , x )  B jn , exp(  ik jn , x )] sin( 2 dyn /  ) ) j j       . (4) Figure 6 represent the transmission coefficient dependence on V for E = 0.2 eV and for N = 0 (red dotted line), 1 (so lid black line) and 2 (green dashed line). T tends to the magenta dashed-dotted line for large N values. Unlike for Schrödinger-type charge carriers, the transmission has significant values, except for a narrow voltage range in which there are no outgoing modes. The current through the tapered graphene, calculated also with the Landauer formula, is represented in Fig. 7 for three values of the Fermi energy level: FE = 0 (green dashed line), 0.1 eV (solid black line) and 0.2 eV (red dotted line). As fo llows from Fig. 7, the shape of the I-V characterist ics is the same in all cases, but the current amplitude depends on the Fermi leve l (on the energy of electrons that contribute to the current flow). Unlike for Schrödinger-type charge carriers, the I-V characterist ics are symmetric around a bias vo ltage that represents the center of the region in which the transmission coefficient/number o f outgoing modes vanishes. Current rectification can occur only if the Fermi level is chosen such that the sudden current 8 increase or decrease takes place at V = 0. Such a situation is represented with the red dotted line in Fig. 7. In this case current rectificat ion, in the sense that finite current values exist for one polarizat ion only, occurs only if the incident vo ltage signal has an amplitude (in this simulat ion) lower than 0.4 V. Voltage signals with higher amplitudes could be rectified if outd decreases. Conclusions We have demonstrated that rectificat ion, which is a key electronic funct ion, takes place in two- dimensional ballist ic nanodevices, in which the electron gas is described by either the Schrödinger equation or the Dirac equat ion, only when the 2DEG geometry is asymmetric wit h respect to current flow direction. While in quantum wells described by Schrödinger equat ion the rectificat ion is produced for a single, positive polarity of the electrical field (or negative polarity in two-dimensional ho le gases), in the case of graphene, which satisfy a Dirac-like equation, rect ification is ambipo lar; ambipolarity is an imprint of all graphene device. Because the cutoff frequencies o f the above devices is located in the THz region [16], our results open up the perspective o f generating THz frequencies using mult ipliers with a much simpler architecture and of imaging THz radiat ion by connecting the above rect ifiers with small antennas on a single Si wafer. References 9 [1] D.J. Reilly, and C.M. Marcus, “Fast single-charge sensing with a rf quantum po int contact”, Appl. Phys. Lett. 91, 162101 (2007). [2] J.U. Lee, P.P. Gipp, and C.M. Heller, “Carbon-nanotube p-n junct ion diodes”, Appl. Phys. Lett. 85, 145 (2004). [3] J.R. Williams, L. Dicarlo and C.M. Marcus, “Quantum-controlled p-n junctions o f graphene”, Science 317, 638 (2007). [4] M.H. Yang, K.B.K. Teo, and W.I. Milne, “Carbon nanotube Schottky diode and direct ionally depend field-effect transistor using asymmetric contacts”, Appl. Phys. Lett. 87, 253116 (2005). [5] C. Lu, L. An, Q. Fu and J. Liu, “Schottky diodes from asymmetric metal-nanotube contacts”, Appl. Phys. Lett. 88, 133501 (2006). [6] A.M. Song, M. Missous, P. Omling, A.R. Peaker, L. Samuelson, and W. Seifert, “Unidirect ional electron flow in a nanometer-scale semiconductor channel: A self- switching device”, Appl. Phys. Lett. 83, 1881 (2003). [7] C. Balocco, S.R. Kasjoo, X.F. Lu, L.Q. Zhang, Y. Alimi, S. Winnerl, and A.M. Song, “Room temperature of a unipo lar nanodiode at terahertz frequencies”, Appl. Phys. Lett. 98, 223501 (2011). [8] K. Cho i, G. Ryu, F. Yesilkoy, A. Chryssis, N. Goldsman, M. Dagenais, and M. Peckerar, “Geometry enhanced asymmetric rectifying tunnel diodes”, J. Vac. Sci B28, C6050 (2010). [9] G. Moddel, Z. Zhu, S. Grover, and S. Joshi, “Ultrahigh speed graphene diode with reversible polarity”, Solid Sate Commun., doi:10.1016/j.ssc.2012.06.013 (2012). 10 [10] G. Deligiorgis, M. Dragoman, D. Neculo iu, D. Dragoman, G. Konstant inidis, A. Cismaru, and R. Plana, “Microwave propagation in graphene”, Appl. Phys. Lett. 95, 073107 (2009). [11] R.S. Shishir and D.K. Ferry, “Intrinsic mobility in graphene”, J. Phys.: Condens. Matter. 21, 23204 (2009). [12] A.S. Mayorov, R.V. Gorbachev, S.V. Morozov, L. Britnell, R. Jail, L.A. Ponomarenko, K.S. Novoselov, K. Watanabe, T. Taniguchi, and A.K. Geim, “Micrometer-scale ballist ic transport in encapsulated graphene at room temperature”, Nano Lett. 11, 2396 (2011). [13] H. Wu, D.W.L. Sprung, J. Martorell, and S. Klarsfeld, “Quantum wire with periodic serial structure”, Phys. Rev. B 44, 6351 (1991). [14] A. Weisshaar, J. Lary, S.M. Goodnick, and V.K. Tripathi, “Analysis o f discont inuit ies in quantum waveguide structures”, Appl. Phys. Lett. 55, 2114 (1989). [15] D. Dragoman, “Evidence against Klein paradox in graphene”, Phys. Scr. 79, 015003, (2009). [16] D. Dragoman and M. Dragoman, “T ime flow in graphene and its implicat ions on cutoff frequency o f ballist ic graphene devices”, J. Appl. Phys. 110, 014302 (2011). Figure Captions 11 Fig. 1 Schemat ic representation of the ballist ic rect ifier. Fig. 2 Voltage dependence of the ratio of outgoing and incoming mode numbers in a 2DEG for E = 0.1 eV (green dashed line), 0.2 eV (solid black line) and 0.3 eV (red dotted line). Fig. 3 Voltage dependence of the transmission coefficient in a 2DEG for E = 0.2 eV and N = 0 (red dotted line), 1 (solid black line) and 2 (green dashed line). For large N, T tends to the magenta dashed-dotted line. Fig. 4 Current-voltage characterist ics in a 2DEG for FE = 0 (green dashed line), 0.1 eV (solid black line) and 0.2 eV (red dotted line). Fig. 5 Voltage dependence of the ratio of outgoing and incoming mode numbers in graphene for E = 0.1 eV (green dashed line), 0.2 eV (solid black line) and 0.3 eV (red dotted line). Fig. 6 Voltage dependence of the transmission coefficient in graphene for E = 0.2 eV and for N = 0 (red dotted line), 1 (solid black line) and 2 (green dashed line). T tends to the magenta dashed-dotted line for large N values. Fig. 7 Current-voltage characterist ics in graphene for FE = 0 (green dashed line), 0.1 eV (solid black line) and 0.2 eV (red dotted line). 12 metallic contact Fig. 1 y x dout din rectification L metallic contact 13 Fig. 2 Fig. 3 14 Fig. 4 Fig. 5 15 Fig. 6 Fig. 7
1608.01617	1	1608	2016-08-04T17:17:42	Black Phosphorus Plasmonics: Anisotropic Elliptical Propagation and Nonlocality-Induced Canalization	[ "cond-mat.mes-hall" ]	We investigate unusual surface plasmons polariton (SPP) propagation and light-matter interactions in ultrathin black phosphorus (BP) films, a 2D material that exhibits exotic electrical and physical properties due to its extremely anisotropic crystal structure. Recently, it has been speculated that the ultra-confined surface plasmons supported by BP may present various topologies of wave propagation bands, ranging from anisotropic elliptic to hyperbolic, across the mid- and near-infrared regions of the electromagnetic spectrum. By carefully analyzing the natural nonlocal anisotropic optical conductivity of BP, derived using the Kubo formalism and an effective low-energy Hamiltonian, we demonstrate here that the SPP wavenumber cutoff imposed by nonlocality prohibits that they acquire an arbitrary hyperbolic topology, forcing operation in the canalization regime. The resulting nonlocality-induced canalization presents interesting properties, as it is inherently broadband, enables large light-matter interactions in the very near field, and allows extreme device miniaturization. We also determine fundamental bounds to the confinement of BP plasmons, which are significantly weaker than for graphene, thus allowing a larger local density of states. Our results confirm the potential of BP as a promising reconfigurable plasmonic platform, with exciting applications, such as planar hyperlenses, optoelectronic components, imaging, and communication systems.	cond-mat.mes-hall	cond-mat	Black Phosphorus Plasmonics: Anisotropic Elliptical Propagation and Nonlocality-Induced Canalization D. Correas-Serrano1,2, J. S. Gomez-Diaz1, A. Alvarez Melcon2, and Andrea Alù1* 1 Department of Electrical and Computer Engineering, The University of Texas at Austin, Austin, TX 78712, USA 2Universidad Politecnica de Cartagena, 30202 Cartagena, Murcia, Spain. Keywords: Plasmonics, uniaxial, hyperbolic materials, black phosphorus, nonlocality. Abstract: We investigate unusual surface plasmons polariton (SPP) propagation and light-matter interactions in ultrathin black phosphorus (BP) films, a 2D material that exhibits exotic electrical and physical properties due to its extremely anisotropic crystal structure. Recently, it has been speculated that the ultra-confined surface plasmons supported by BP may present various topologies of wave propagation bands, ranging from anisotropic elliptic to hyperbolic, across the mid- and near-infrared regions of the electromagnetic spectrum. By carefully analyzing the natural nonlocal anisotropic optical conductivity of BP, derived using the Kubo formalism and an effective low-energy Hamiltonian, we demonstrate here that the SPP wavenumber cutoff imposed by nonlocality prohibits that they acquire an arbitrary hyperbolic topology, forcing operation in the canalization regime. The resulting nonlocality-induced canalization presents interesting properties, as it is inherently broadband, enables large light-matter interactions in the very near field, and allows extreme device miniaturization. We also determine fundamental bounds to the confinement of BP plasmons, which are significantly weaker than for graphene, thus allowing a larger local density of states. Our results confirm the potential of BP as a promising reconfigurable plasmonic platform, with exciting applications, such as planar hyperlenses, optoelectronic components, imaging, and communication systems. Introduction The rise of graphene in the past decade [1]–[4] has triggered extensive research on other two- dimensional materials with properties able to surmount some of graphene's shortcomings. Great effort has been devoted to transition metal dichalcogenides (TMDs) such as MoS2 and WS2 [5]– [8] due to their direct bandgap, an attractive feature to achieve high on-off ratios, which has proven elusive in graphene devices [9]. However, the low mobility in these materials (orders of magnitude below the one of graphene) has hindered their applicability. More recently, black phosphorus (BP) thin films, or phosphorene, has been rediscovered as a material able to simultaneously exhibit high on-off ratios and carrier mobility comparable to that of graphene [10]. Importantly, it may have yet untapped potential for mid-IR nano-optics, since an adequate selection of the film thickness results in a small bandgap and good optical transitions for realistic doping levels [11]. In this direction, perhaps the most striking feature of BP compared to other 2D materials is the naturally 1 high in-plane anisotropy of its macroscopic physical properties, which arises due to its peculiar orthorhombic lattice with two clearly distinguishable axes, as depicted in Fig. 1. BP has been successfully isolated as a monolayer and few-layer form by means of mechanical exfoliation and plasma thinning [12]. In addition, exciting possibilities to control the optical properties of BP have been demonstrated using different methods. For instance, Ref. [12] experimentally demonstrated high-resolution control of these properties by laser pruning, reducing thick BP films to few-layer forms and enabling local and straightforward engineering of the bandgap and the optical conductivity. Strong normal fields induced by dopants were used in [13] to experimentally demonstrate drastic tuning of the bandgap due to giant Stark effect, allowing to efficiently and swiftly switch the nature of BP from a moderate-bandgap semiconductor to a band- inverted semimetal. In the critical point before band inversion, BP becomes an anisotropic Dirac semimetal similar to graphene only in one direction, providing unprecedented design flexibility that may open venues for novel functionalities in planar optics. Importantly, this behavior can be obtained through an externally applied gate bias similarly to graphene devices. Mechanical strain has been predicted to enable extensive reconfiguration capabilities as well [14], e.g. by allowing to swap the higher-conductivity direction with a moderate biaxial or uniaxial strain [15]. Until now, most studies on the applied physics of BP have focused on its DC properties, aiming to realize fast and efficient electronic transistors and devices. Even though interesting optical responses, such as strong layer-dependent photoluminescence have been reported [16] – similarly to the case of MoS2 [17] –, they have been mainly applied to confirm the direct and layer-sensitive bandgap of the material, rather than to the design of practical optoelectronic devices. One exciting direction to realize such components would be to exploit the extremely-confined anisotropic surface plasmons polaritons (SPPs) that BP supports at mid-IR owing to its inductive nature [18]- [19]. It is important to note that graphene cannot support low-loss plasmons at such high frequencies, due to the strong coupling of SPPs to optical phonons [3]. Localized plasmons in patterned BP films have also been theoretically studied in the context of plane wave transmission/reflection, showing the expected polarization dependence due to the material anisotropy [20]. Unfortunately, to our knowledge the few existing works on BP plasmonics have all considered a local description of the optical conductivity, so the fundamental performance bounds and potential practical applications of BP are not yet fully understood. In this context, this paper presents a comprehensive study of BP plasmonic properties from terahertz to infrared (IR), unveiling the potential of this material as a platform for future planar and reconfigurable mid-IR devices with improved functionalities over alternative technologies. The main contributions of this paper are: (i) a clear discussion, using realistic material parameters, of the different SPPs topologies supported in BP films when taking into account both intraband and interband contributions; and (ii) the study of nonlocal effects in BP plasmons, shedding light into fundamental bounds on field confinement while also reporting unusual light-matter interactions and quantum nonlocality-related phenomena. Specifically, our study demonstrates 2 Figure 1. Optical properties of black phosphorus. (a) Lattice structure of a two-layer ultrathin film. Different colors are employed for visual clarity. (b) Energy-wavenumber dispersion diagram of electrons in BP. 𝐸𝐹, 𝐸𝑐,1, and 𝜇𝑐 are the Fermi and first-conduction band energies and three quantized valence/conduction subbands along 𝑘𝑥 and 𝑘𝑦 and associated interband transition processes. (e) BP bandgap dependence versus layer thickness. the chemical potential, respectively. (c)-(d) First that the hyperbolic response predicted by local conductivity models [19] cannot be realized in practice, and instead becomes a broadband, nonlocality-induced and deeply subwavelength canalization regime that can be directly applied to realize in-plane hyperlenses. Optical conductivity of BP thin films Throughout this paper we assume infinitesimally-thin layers of BP, which can be modeled by the space- and time-dispersive conductivity tensor , (1) where 𝜔 = 2𝜋𝑓 is the angular frequency, 𝒒 is the wavevector, and the different components can be computed through the well-known Kubo formalism. Here we use the approximate two- dimensional Hamiltonian developed in [14] to model the in-plane electron dispersion, which is based on 𝑘 ⋅ 𝑝 theory and is valid near the Γ point (low to moderate electron energies). We reproduce it here for completeness. It reads 3 ,,,,,xxxyyxyyqqqqq , (2) where 𝐸𝑐 and 𝐸𝑣 are the energies of the conduction and valence band edges, respectively, and 𝜂𝑐,𝑣 and 𝜈𝑐,𝑣 are related to the in-plane effective masses. For multilayer films, the above Hamiltonian describes electron dispersion in each layer. In this case, different valence and conduction subband energies must be considered for each layer j, with 𝐸𝑐,𝑣 being replaced by 𝐸(𝑐,𝑣),𝑗, the edge energy values of the corresponding subband pairs. By choosing the previous values to match experimental data for the effective masses and bandgap of few-layer and bulk BP, we are able to accurately model its conductivity. We use the same values as in [21], which are consistent with experimental reports. This model allows to phenomenologically take into account bandgap shrinkage and widening due to giant Stark effect [13], [22] or mechanical strain, among others. The Kubo formula can be used calculate all the components of the conductivity tensor as a function of frequency and the wavevector as (3) In this equation, , is the reduced Planck constant, is the electron charge, is the finite damping, is the Fermi-Dirac distribution function, is the velocity operator, and are the eigenvalues and eigenfunctions of , and the indices and denote the conduction/valence band and the subband indices, respectively. Due to the nonexistent overlap between different subbands in this model [14], [18], optical intra/interband transitions are only allowed between equal subbands, i.e. , as illustrated in Fig. 1c-d (otherwise the matrix elements in the Kubo integral vanish). Lastly, we define the chemical potential 𝜇𝑐 as the difference between the Fermi level and the first conduction band, i.e., 𝐸𝐹 = 𝐸𝑐,1 + 𝜇𝑐, and use a scattering rate of 3 meV. In the local case, i.e., when from the EM wavevector, yielding the commonly used local tensor , the optical conductivity tensor is diagonal and independent . (4) ) of a 10 nm thick Figs. 2a-b illustrate the imaginary and real pats of the local conductivity ( BP multilayer versus frequency for various values of chemical potential 𝜇𝑐. For this thickness, the bandgap is estimated to be 0.48 eV. Results confirm a Drude-like dispersion below the interband transition threshold and a high degree of anisotropy with ample tunability for small variations of the chemical potential. The y-component of the conductivity presents metallic behavior (inductive, under a 𝑒𝑖𝜔𝑡 time convention) for a wider range of frequencies than the x- i.e., component, which is the high-conductivity direction and shows a very distinct transition from metallic to dielectric (capacitive, i.e., ) response due to interband transitions. Interband processes become significant for frequencies larger than 2𝜋𝑓0ℏ ≈ 𝐸𝑔 + 2𝜇𝑐 , where 𝐸𝑔 is the 4 222222ccxcyxyxyvvxvyEEkkkkkkkk2''2'ΦΦΦ,(cid:160)2ΦsjsjsjsjsjsjssjjsjsjsjsjfEfEvviedEEEEikkkkkkkkkkqkk'=k+qef1kvsjEkΦsjk'1ss'jj'jjq000xxyy0qIm[]0Im[]0 Figure 2. Dispersion of BP optical conductivity. Imaginary (a) and real (b) components of BP conductivity versus frequency plotted for various values of chemical potential. Different curves correspond to growing (as indicated by the arrow) values of chemical potential: 0.005 eV, 0.05 eV, 0.1 and 0.2 eV. Blue and red lines corresponds to 𝜎𝑥𝑥 and 𝜎𝑦𝑦 conductivity components, respectively. (c), (d) Anisotropy ratio of the imaginary and real parts of the conductivity (insets show detail at low frequencies). Hyperbolic regime appears at the frequencies where the ratio Im[𝜎𝑥𝑥]/Im[𝜎𝑦𝑦] is negative (shaded region in panel c). Other parameters are similar to Fig. 1. bandgap energy. Due to Pauli blocking, these frequency points are separated by approximately Δ𝑓 ≈ Δ𝜇𝑐/𝜋ℏ and can therefore be controlled in practice by modifying BP chemical potential through an externally applied electrostatic bias [18]. The peaks observed in the conductivity above this frequency appear due to the different interband transition frequencies of the subband pairs associated to each layer (see Fig. 1c-d). 5 Figs 2c-d explicitly show a large degree of anisotropy in the real and imaginary parts of the conductivity components. In the Drude regime, we find that the ratios Im[𝜎𝑥𝑥]/(Im[𝜎𝑦𝑦] and (Re[𝜎𝑥𝑥]/Re[𝜎𝑦𝑦] are weakly dependent on 𝜇𝑐, and hover around 5. This anisotropy becomes strongly dependent on chemical potential as we approach the interband threshold for the x- component, which marks the beginning of the hyperbolic regime predicted in [19]. Before this point, the anisotropy in the imaginary part continuously decreases due to the interband processes counteracting the inductive intraband contributions, while loss monotonically increases. This ratio evidently crosses unity, allowing isotropic phase propagation with anisotropic plasmon decay, something unique to BP. The hyperbolic regime has a broad bandwidth, until 𝜎𝑦𝑦 also becomes dielectric. After this point, BP behaves as a thin dielectric layer with very large anisotropy. Overall, this study demonstrates the dispersive, tunable and extremely anisotropic electromagnetic response of BP. This response is indeed very rich and includes anisotropic elliptic quasi-transverse magnetic (TM, i.e., Im[𝜎𝑥𝑥] > 0, Im[𝜎𝑦𝑦] > 0 ) and quasi-transverse electric (TE, i.e., Im[𝜎𝑥𝑥] < 0, Im[𝜎𝑦𝑦] < 0 ) behaviors at low and very high frequencies, respectively, as well as an intrinsic hyperbolic frequency band ( Im[𝜎𝑥𝑥] < 0, Im[𝜎𝑦𝑦] > 0 ) and two clearly-defined topological transitions that implement 𝜎-near-zero topologies. To our knowledge, hyperbolic plasmons have only been previously reported at THz and far-IR frequencies using nanostructured graphene [23], [24] and at optics by patterning silver [25] and therefore BP may potentially fill this frequency gap by enabling SPPs with hyperbolic topologies at mid-IR. However, as detailed later in the text, nonlocal effects may play a dominant role in such scenarios, dramatically modifying the expected hyperbolic response of BP. Anisotropic surface plasmons in BP thin films In this section, we solve the exact dispersion relation of surface waves in BP and compute the associated fields, in order to quantitatively characterize the operation regime of the supported SPPs discussed earlier. For now, we still consider a local description of the conductivity. The dispersion of the supported plasmons is given by [24], [26], [27] (5) where , is the free-space impedance, and the possible wavenumber- dependence of the conductivity components has been omitted. This equation can be numerically solved with conventional root-finding algorithms in the complex plane [24]. Fig. 3 illustrates the three main topologies discussed in the previous section, which can be found in BP at different operation frequencies. For each topology, the figure shows the associated isofrequency contour (IFC) and the normal component of the electric field along the BP excited by a z-oriented dipole located 10 nm above the layer. Fig. 3a corresponds to the elliptical regime that is found at "low" frequencies (50 THz) where the intraband contributions dominate. The degree of anisotropy Im[𝜎𝑥𝑥]/Im[𝜎𝑦𝑦] is roughly 5 in this scenario, which results into a significant canalization along the x-axis (armchair direction). Note that in this Drude regime, i.e., when interband transitions are negligible, the ratio between the axial inductances is weakly dependent on frequency and chemical potential (see Fig. 2), so this type of topology is the most common at low frequencies. 6 0222200004220zxxyyxyyxxxyyxxxyyyxyxyyxkqkqqqq22220xyzkqqq0 Figure 3. Anisotropic surface plasmons supported by black phosphorus. Isofrequency contours are computed using Eq. (3). Color maps show the z-component of the electric field excited by a dipole located 10 nm above the surface. (a) Anisotropic elliptic plasmons, operation frequency is 𝑓 = 50 THz, chemical potential 𝜇𝑐 = 0.1 eV (b) Interband-induced quasi-isotropic plasmons, 𝑓 = 100 THz, 𝜇𝑐 = 0.1 eV (c) Hyperbolic plasmons, 𝑓 = 80 THz, 𝜇𝑐 = 0.05 eV. The middle panel corresponds to the quasi-isotropic point for 𝜇𝑐 = 0.1 eV, which occurs around 100 THz. As anticipated in the previous section, phase propagation is uniform, but loss is clearly higher along the x-axis due to interband transitions. This unusual operation regime may be useful to realize polarization-sensitive absorbers. We note that around this frequency the conductivity is very sensitive to changes in doping. Lastly, the right panel illustrates the natural hyperbolic regime for 𝜇𝑐 = 0.05 eV at 80 THz. We do stress that local, homogeneous BP is able to intrinsically support hyperbolic plasmons thanks to its inherent anisotropy, and without requiring complicated patterning process as the case of graphene [23], [24] and silver [25] A remarkable property of the latter two examples is the extremely large wavenumbers of the supported plasmons, with wavelengths hundreds of times smaller than free-space. Properly designed, such wave- confinement can potentially lead to ultra-miniaturized devices. We have found that this response is always present in BP films operating near interband frequencies, as a result of 𝜎𝑦𝑦 always being very small when 𝜎𝑥𝑥 approaches the transition from metallic to dielectric. This can be mitigated only partially by modifying the doping or film thickness. Essentially, we have found that it is impossible to obtain hyperbolic responses with wavenumbers significantly lower than the ones considered in Fig. 3. It is well known that extremely confined surface waves are associated with a drastic enhancement of the local fields and the spontaneous emission rate of nearby sources, both of which are desired features in applications such as hyperlensing, nonlinear optics, or heat transfer [28]. However, several factors limit the response of any plasmonic device, and values as large as those reported here may lead to important practical difficulties or unexpected effects. These include achieving unwanted coupling to sources and detectors, increased loss due to impurities, and spatial dispersion phenomena. In- and out-coupling of energy to BP films can be done with near-field techniques similar to those used to excite plasmons in graphene [29], and its study is out 7 Figure 4. Contour plots of the electron energies in the first conduction band, overlaid with a colormap of the electron velocity. of the scope of this paper. Loss mechanisms can be phenomenologically added to the model through the parameter 𝜂 in Eq. (3) with predictable results. On the other hand, spatial dispersion has less evident effects that depend on the electron dispersion in the lattice even in the absence of impurities, and as such they are not expected to be mitigated by improved fabrication processes. Lastly, note that due to the conductivity model used, we have neglected possible finite-thickness effects. Spatial dispersion in realistic BP thin films Results in the previous section evidenced that the most exotic operation regimes in BP thin films involve extremely confined plasmons. Thus, the common assumption of a local conductivity, i.e. 𝑞 → 0 in Eq. (3) may not be valid. In the following we explicitly consider the wavevector- dependence of the conductivity to more accurately predict the topology of BP plasmons by assuming 𝑞 ≠ 0 in the Kubo formula and finding the fully populated conductivity tensor for any electromagnetic wavevector. Although computationally cumbersome, this approach is expected to be very accurate and to yield reliable insights into the performance limits of BP plasmons. We expect simplified models to be developed in the near future, for instance by using a phenomenological approach to introduce nonlocal corrections to the local Drude model within the Thomas-Fermi approximation. Similar models have been used in the study of traditional metallic nanostructures such as spheres and dimers [30] and we recently outlined how this nonlocal description of 3D media may be adapted to ultrathin materials [31]. In this context, it is illustrative to discuss the differences between BP and graphene with respect to nonlocal response. The simplest way to qualitative understand the influence of this phenomena is in terms of the electron velocity, with the so-called Fermi velocity 𝑣𝐹 corresponding to electrons with energies near the Fermi level. In general, the electron velocity is given by 𝑣± = 1 ℏ ∇𝑘𝐸±(𝑘), (6) 8 Figure 5. Nonlocal conductivity components of black phosphorus versus real in-plane wavenumbers 𝑞𝑥/𝑘0 and 𝑞𝑦/𝑘0 , computed for 𝜇𝑐 = 0.1 eV at 50 THz. White dashed lines indicate zero-crossings where 𝐸±(𝑘) is the electron dispersion relation, given by the eigenvalues of the Hamiltonian. In graphene, 𝐸±(𝑘) = ±ℏ𝑣𝐹𝑘 [2] is the well-known linear dispersion law near the Dirac points, and leads to an energy-independent 𝑣± = 𝑣𝐹. As shown in [31], this implies that the magnitude of the wavevector of propagating plasmons is necessarily bounded by 𝑘 < 𝑐/𝑣𝐹𝑘0 ≈ 300𝑘0 for any device based on graphene [31]. In BP, however, the picture is more complicated due to anisotropy, the non-symmetric conduction and valence bands, and the energy-dependent electron velocities as a result of the quadratic k-terms in the Hamiltonian. Analytical expressions can be obtained for 𝑣± by applying Eq. (6) to the eigenvalues of the Hamiltonian, but they involve cumbersome algebra and are omitted here for the sake of clarity. The color maps in Fig. 4 show the velocity in the conduction (left) and valence (right) bands in the vicinity of the Γ point. The contours represent the electron energies. We observe velocity values range from roughly 105 m/s for the lowest energy electrons to around 106 for energies around 1 eV (recall that 𝑣𝐹 = 106 m/s = 𝑐/300 in graphene). Interestingly, this suggests that (i) plasmonic devices based on BP may support larger wavenumbers than their graphene counterparts, and (ii) the bounds may be doping dependent. Fig. 5 shows the nonlocal conductivity components for 𝑓 = 50 THz and 𝜇𝑐 = 0.1 eV for a wide range of in-plane wavenumbers 𝑞𝑥 and 𝑞𝑦 . This scenario corresponds to the elliptic regime illustrated in Fig. 3a. For small wavenumbers, these values converge to the local conductivity, with 𝜎𝑥𝑦 = 𝜎𝑦𝑥 = 0. As 𝒒 increases, so do the non-diagonal terms and the real parts of the diagonal components. This implies that large wavenumbers force a non-negligible in-plane rotation of 9 Figure 6. Nonlocal conductivity components of black phosphorus versus real in-plane wavenumbers 𝑞𝑥/𝑘0 and 𝑞𝑦/𝑘0, at operation frequency of 80 THz and 𝜇𝑐 = 0.05 eV. While dashed lines indicate zero-crossings power from the longitudinal to the transverse direction, similarly to graphene [31] and other plasmonic systems [30]. Note, however, that this is not a dissipative process. In addition, very large 𝑞 may cause the conductivity components to change from metallic to dielectric or vice versa (white dashed lines show the zero-crossings). In this example the transition from metallic to dielectric response is different for 𝜎𝑥𝑥 and 𝜎𝑦𝑦 , but in both cases occurs for 𝑞 ≈ 600𝑘0 . As expected from earlier discussions based on 𝑣±, this value is significantly larger than in graphene, where 𝒒 < 300𝑘0 [31]. Fig. 6 shows the nonlocal conductivity components corresponding to the hyperbolic regime previously considered (Fig. 3c), i.e., 𝑓 = 80 THz and 𝜇𝑐 = 0.05 eV. In this case, the local model predicts capacitive 𝜎𝑥𝑥 and metallic 𝜎𝑦𝑦, hence the hyperbolic response. However, it is clear that nonlocality causes 𝜎𝑥𝑥 to become metallic for a wide range of 𝑞𝑥 , potentially changing the plasmon dispersion, as we will see later. Regarding 𝜎𝑦𝑦, it once again remains metallic until 𝒒 ≈ 600𝑘0. Let us focus again on the surface waves supported by the BP films under consideration, now taking its nonlocal response into account. Fig. 7 shows the IFCs computed with Eq. (3) assuming the fully populated nonlocal 𝜎, with each 𝜎𝛼𝛽 being replaced by 𝜎𝛼𝛽(𝑞𝑥, 𝑞𝑦). Panels (a), (c) correspond to the case of elliptical plasmons (Fig. 5). Here, nonlocality has a moderate effect on the IFCs due to the relatively low wavenumbers involved. Interestingly, spatial dispersion increases the confinement along the y-direction, something that does not happen in graphene plasmons along 10 infinite sheets [32]. This is in agreement with Figs. 5a-5c, which show a slight reduction of Im[𝜎𝑦𝑦] and Im[𝜎𝑥𝑥] when 𝑞𝑦 increases from zero up to 𝑞𝑦 ≈ 500𝑘0. The increased inductance makes plasmons more confined. Plasmons along the x direction are barely affected, and overall loss increases for all directions. Panels (b), (d) show the IFCs of the hyperbolic case, related to Fig. 6. The effect of nonlocality is much more pronounced here, with the hyperbolic dispersion being replaced by an almost perfect canalization along y. In order to understand this phenomenon, it is helpful to once again consider the corresponding conductivity plots in Fig. 6. First, it is clear that increasing 𝑞𝑦 for low 𝑞𝑥 results in higher Im[𝜎𝑦𝑦] (lower inductance, contrary to the elliptic scenario), which causes a significant shift of the vertex of the hyperbola to lower 𝑞𝑦. Near 𝑞𝑥 = 0, the hyperbolic shape can still be noticed, but as 𝑞𝑥 increases the topology changes drastically and the IFC closes around 𝑞𝑥 ≈ 400𝑘0. This is related to the unusual nonlocality induced dielectric-metallic transition shown in Fig. 6a around 𝑞𝑥 ≈ 200𝑘0. We have found that, within this model, nonlocality always prevents plasmons from exhibiting a true hyperbolic topology for a significant range of wavenumbers, in contrast with previous reports that considered only local response and predicted hyperbolic plasmons [19]. We stress, however, that more complex structures such as arrays of strips or patches might still support hyperbolic plasmons similar to other nanostructured metasurfaces [23], [25] although they should be designed to operate with lower wavenumbers. We emphasize that this study assumes an infinitesimally thin layer, i.e., we are assuming a certain wavevector on the surface studying the material response, and therefore we do not consider possible bulk modes supported by the actual BP thin film. Within this assumption, our results aimed to probe the quantum-induced limits of BP plasmonics associated to spatial dispersion and have further demonstrated the immense potential of this material for manipulating IR light at the nanoscale. Specifically, we remark the looser physical bounds in terms of field confinement and the unusual phenomena related to nonlocality, in stark contrast with graphene sheets, where spatial dispersion reduces field confinement and increases loss. We envision that these features may be exploited in the future as additional degrees of freedom in the design of deeply subwavelength plasmonic devices. Light-matter interactions in BP thin films An effective and simple metric of the magnitude of light-matter interactions enabled by a metasurface is given by the spontaneous emission rate (SER) of nearby sources [23],[24]. Therefore, it is of evident importance to understand how this metric compares to other similar technologies, and how spatial dispersion may affect it. It can be computed as , (6) where 𝑃0 and 𝑃 are the power generated by the emitter when isolated in free-space and when located above the BP film; 𝜇𝑝⃗⃗⃗⃗ and 𝑟0⃗⃗⃗ denote the dipole orientation and position, 𝑘0 is the free- space wavenumber and 𝐺𝑆(𝑟0⃗⃗⃗ , 𝑟0⃗⃗⃗ , 𝜔) is the scattered term of the Green's function of the entire 11 000061Im,,pSppPSERGrrPkFigure 7. Influence of spatial dispersion in the isofrequency contours of SPP supported by black phosphorus films. Results are computed using Eq. (6), modelling BP optical conductivity with the Kubo formula (see Eq. (3)) with and without nonlocal effects taken into account. (a), (c) Operation frequency 𝑓 = 50 THz, chemical potential 𝜇𝑐 = 0.1 eV (same as Fig. 5). (b), (d) Operation frequency 𝑓 = 80 THz, chemical potential 𝜇𝑐 = 0.05 eV (same as Fig. 6). system. In plasmonic structures, a drastic enhancement of power radiated by a nearby source is typically observed due to efficient coupling of the evanescent spectrum to surface waves with high q. Fig. 8 shows the SER of a z-oriented dipole versus its distance to the two BP films studied in the previous section, for the local and nonlocal models. We also plot the SER for a homogeneous graphene sheet with the same 𝜇𝑐 and 𝜂. Although this comparison is somehow arbitrary (many practical factors such as achievable scattering rate and ease of doping should be taken into account), it allows to easily compare results with a more familiar platform. Fig. 8a corresponds to the elliptical regime at 50 THz, with 𝜇𝑐 = 0.1 eV. In this case the local model presents good accuracy due to the small effect that nonlocality has on plasmon dispersion. Since spatial dispersion slightly increases the confinement along the y-direction in this case (see Fig. 7a), the SER is in fact higher than predicted by the local model when the source is very close to the surface, and lower as it is moved farther. This occurs because high-q waves decay faster in free-space. The SER associated to a homogeneous graphene sheet shows a similar trend versus distance, but there is a larger discrepancy between local and nonlocal models. This is an expected effect of spatial dispersion in both materials, as discussed in previous sections. Fig. 8b corresponds to the hyperbolic regime at 80 THz for 𝜇𝑐 = 0.05 eV. For sources very close to the surface, the SER is larger than in Fig. 8a, but it decays more rapidly with distance due to the larger q of the supported 12 Figure 8. Spontaneous emission rate of a z-oriented point source versus its distance to BP thin films and graphene sheets, considering local and realistic nonlocal responses. Parameters are the same as (a) Fig. 7a ( 𝑓 = 50 THz , 𝜇𝑐 = 0.1 eV ) and (b) Fig. 7b (𝑓 = 80 THz , 𝜇𝑐 = 0.05 eV) plasmons (see Fig. 7b). The disagreement between nonlocal and local models is more pronounced in this case for the same reason. As a consequence, the energy coupled is orders of magnitude larger than predicted by the local mode when the source separation above a few nanometers. Overall, BP enables strong light-matter interactions, comparable to graphene, while providing additional freedom in tailoring plasmon dispersion and higher upper limits in the SER. Conclusion and Outlook We have investigated BP thin films as a potential IR plasmonic material with exotic optical properties, such as strong broadband anisotropy and a natural hyperbolic response. To this end, we have used a 2D description of its conductivity based on the Kubo formula and an approximate Hamiltonian. We have illustrated different plasmonic operation regimes, from elliptic to hyperbolic, depending on frequency and doping. By considering the nonlocal response of BP for the first time, we have determined fundamental bounds in the confinement of BP plasmons. Importantly, results demonstrate that these bounds are looser than in graphene devices, allowing more tightly confined fields and stronger light-matter interactions in nanostructures. As expected, the quantum effects of nonlocality are anisotropic, with transitions between inductive and capacitive responses that depend on the in-plane direction and material properties. We have also demonstrated that nonlocality prevents plasmons from exhibiting true hyperbolic topology, contrary to previous predictions, and shown unusual nonlocality-induced topological transitions that leads to a broadband canalization regime. We envision a promising future for BP as a reconfigurable plasmonic platform whose electromagnetic properties can be further enhanced by nano-structuring or by realizing hybrid systems composed of BP, graphene, metals or other 2D materials, with exciting opportunities to realize a wide variety of optoelectronic devices, sensors, lenses and communications system at IR frequencies. 13 Corresponding author *To whom correspondence should be addressed: [email protected] Acknowledgements This work was supported by the Air Force Office of Scientific Research grant No. FA9550-13-1- 0204, the Welch Foundation with grant No. F-1802, the Simons Foundation, and the National Science Foundation with grant No. ECCS-1406235. D. C. S. acknowledges financial support by the Spanish MECD under Grant FPU13-04982. References [1] K. Geim and K. S. Novoselov, "The rise of graphene," Nat. Mater., vol. 6, pp. 183–91, 2007. [2] A. H. C. Neto, N. M. R. Peres, K. S. Novoselov, and A. K. Geim, "The electronic properties of graphene," Rev. Mod. Phys., vol. 81, pp. 109–162, 2009. [3] M. Jablan, H. Buljan, and M. Soljacic, "Plasmonics in graphene at infrared frequencies," Phys. Rev. B, vol. 80, p. 245435, 2009. [4] F. J. Garcia de Abajo, "Graphene plasmonics: challenges and opportunities," Acs Photonics, vol. 1, no. 3, pp. 135–152, 2014. [5] R. Mas-Ballesté, C. Gómez-Navarro, J. Gómez-Herrero, and F. Zamora, "2D materials: to graphene and beyond," Nanoscale, vol. 3, no. 1, pp. 20–30, Jan. 2011. [6] K. F. Mak, C. Lee, J. Hone, J. Shan, and T. F. Heinz, "Atomically Thin MoS2: A New Direct- Gap Semiconductor," Phys. Rev. Lett., vol. 105, no. 13, p. 136805, Sep. 2010. [7] W. Zhao, Z. Ghorannevis, L. Chu, M. Toh, C. Kloc, P.-H. Tan, and G. Eda, "Evolution of electronic structure in atomically thin sheets of WS2 and WSe2," ACS Nano, vol. 7, no. 1, pp. 791–797, Jan. 2013. [8] H. S. S. Ramakrishna Matte, A. Gomathi, A. K. Manna, D. J. Late, R. Datta, S. K. Pati, and C. N. R. Rao, "MoS2 and WS2 Analogues of Graphene," Angew. Chem. Int. Ed., vol. 49, no. 24, pp. 4059–4062, Jun. 2010. [9] 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, "Vertical field-effect transistor based on graphene-WS2 heterostructures for flexible and transparent electronics," Nat. Nanotechnol., vol. 8, no. 2, pp. 100–103, Feb. 2013. [10] H. Liu, A. T. Neal, Z. Zhu, Z. Luo, X. Xu, D. Tománek, and P. D. Ye, "Phosphorene: An Unexplored 2D Semiconductor with a High Hole Mobility," ACS Nano, vol. 8, no. 4, pp. 4033–4041, Apr. 2014. [11] F. Xia, H. Wang, and Y. Jia, "Rediscovering black phosphorus as an anisotropic layered material for optoelectronics and electronics," Nat. Commun., vol. 5, p. 4458, Jul. 2014. [12] J. Lu, J. Wu, A. Carvalho, A. Ziletti, H. Liu, J. Tan, Y. Chen, A. H. Castro Neto, B. Özyilmaz, and C. H. Sow, "Bandgap Engineering of Phosphorene by Laser Oxidation toward Functional 2D Materials," ACS Nano, vol. 9, no. 10, pp. 10411–10421, 2015. [13] J. Kim, S. S. Baik, S. H. Ryu, Y. Sohn, S. Park, B.-G. Park, J. Denlinger, Y. Yi, H. J. Choi, and K. S. Kim, "Observation of tunable band gap and anisotropic Dirac semimetal state in black phosphorus," Science, vol. 349, no. 6249, pp. 723–726, 2015. 14 [14] A. S. Rodin, A. Carvalho, and A. H. Castro Neto, "Strain-Induced Gap Modification in Black Phosphorus," Phys. Rev. Lett., vol. 112, no. 17, p. 176801, May 2014. [15] R. Fei and L. Yang, "Strain-Engineering Anisotropic Electrical Conductance of Phosphorene and Few-Layer Black Phosphorus," Nano Lett., vol. 14, no. 5, pp. 2884–2889, 2014. [16] S. Zhang, J. Yang, R. Xu, F. Wang, W. Li, M. Ghufran, Y.-W. Zhang, Z. Yu, G. Zhang, Q. Qin, and Y. Lu, "Extraordinary Photoluminescence and Strong Temperature/Angle- Dependent Raman Responses in Few-Layer Phosphorene," ACS Nano, vol. 8, no. 9, pp. 9590–9596, Sep. 2014. [17] A. Splendiani, L. Sun, Y. Zhang, T. Li, J. Kim, C.-Y. Chim, G. Galli, and F. Wang, "Emerging Photoluminescence in Monolayer MoS 2," Nano Lett., vol. 10, no. 4, pp. 1271– 1275, Apr. 2010. [18] T. Low, A. S. Rodin, A. Carvalho, Y. Jiang, H. Wang, F. Xia, and A. H. Castro Neto, "Tunable optical properties of multilayer black phosphorus thin films," Phys. Rev. B, vol. 90, no. 7, p. 75434, Aug. 2014. [19] A. Nemilentsau, T. Low, and G. Hanson, "Anisotropic 2D Materials for Tunable Hyperbolic Plasmonics," Phys. Rev. Lett., vol. 116, no. 6, p. 66804, Feb. 2016. [20] G. C. Constantinescu and N. D. M. Hine, "Multipurpose Black-Phosphorus/hBN Heterostructures," Nano Lett., p. acs.nanolett.6b00154, 2016. [21] T. Low, R. Roldán, H. Wang, F. Xia, P. Avouris, L. M. Moreno, and F. Guinea, "Plasmons and Screening in Monolayer and Multilayer Black Phosphorus," Phys. Rev. Lett., vol. 113, no. 10, p. 106802, Sep. 2014. [22] D. A. B. Miller, D. S. Chemla, and S. Schmitt-Rink, "Relation between electroabsorption in bulk semiconductors and in quantum wells: The quantum-confined Franz-Keldysh effect," Phys. Rev. B, vol. 33, no. 10, pp. 6976–6982, May 1986. [23] J. S. Gomez-Diaz, M. Tymchenko, and A. Alù, "Hyperbolic Plasmons and Topological Transitions Over Uniaxial Metasurfaces," Phys. Rev. Lett., vol. 114, no. 23, p. 233901, 2015. [24] J. S. Gomez-Diaz, M. Tymchenko, and A. Alù, "Hyperbolic metasurfaces: surface plasmons, light-matter interactions, and physical implementation using graphene strips [Invited]," Opt. Mater. Express, vol. 5, no. 10, p. 2313, Oct. 2015. [25] A. a. High, R. C. Devlin, A. Dibos, M. Polking, D. S. Wild, J. Perczel, N. P. de Leon, M. D. Lukin, and H. Park, "Visible-frequency hyperbolic metasurface," Nature, vol. 522, no. 7555, pp. 192–196, 2015. [26] H. J. Bilow, "Guided Waves on a Planar Tensor Impedance Surface," IEEE Trans. Antennas Propag., vol. 51, no. 10 II, pp. 2788–2792, 2003. [27] R. Quarfoth and D. Sievenpiper, "Artificial tensor impedance surface waveguides," IEEE Trans. Antennas Propag., vol. 61, no. 7, pp. 3597–3606, 2013. [28] A. Poddubny, I. Iorsh, P. Belov, and Y. Kivshar, "Hyperbolic metamaterials," Nat. Photonics, vol. 7, no. 12, pp. 948–957, Nov. 2013. [29] A. Woessner, M. B. Lundeberg, Y. Gao, A. Principi, P. Alonso-González, M. Carrega, K. Watanabe, T. Taniguchi, G. Vignale, M. Polini, J. Hone, R. Hillenbrand, and F. H. L. Koppens, "Highly confined low-loss plasmons in graphene–boron nitride heterostructures," Nat. Mater., vol. 14, no. April, pp. 421–425, 2014. [30] S. Raza, S. I. Bozhevolnyi, M. Wubs, and N. A. Mortensen, "Nonlocal optical response in metallic nanostructures," J. Phys. Condens. Matter, vol. 27, no. 4, p. 183204, 2015. [31] D. Correas-Serrano, J. S. Gomez-Diaz, M. Tymchenko, and A. Alu, "Nonlocal response of hyperbolic metasurfaces," Opt. Express, vol. 23, no. 23, p. 29434, 2015. 15 [32] D. Correas-Serrano, J. S. Gómez-Díaz, J. Perruisseau-Carrier, and A. Alvarez-Melcon, "Spatially dispersive graphene single and parallel plate waveguides: analysis and circuit model," IEEE Trans. Microw. Theory Tech., vol. 61, no. 12, pp. 4333–4344, Dec. 2013. 16
1505.07592	1	1505	2015-05-28T08:28:47	Effect of the interface resistance in non-local Hanle measurements	[ "cond-mat.mes-hall" ]	We use lateral spin valves with varying interface resistance to measure non-local Hanle effect in order to extract the spin-diffusion length of the non-magnetic channel. A general expression that describes spin injection and transport, taking into account the influence of the interface resistance, is used to fit our results. Whereas the fitted spin-diffusion length value is in agreement with the one obtained from standard non-local measurements in the case of a finite interface resistance, in the case of transparent contacts a clear disagreement is observed. The use of a corrected expression, recently proposed to account for the anisotropy of the spin absorption at the ferromagnetic electrodes, still yields a deviation of the fitted spin-diffusion length which increases for shorter channel distances. This deviation shows how sensitive the non-local Hanle fittings are, evidencing the complexity of obtaining spin transport information from such type of measurements.	cond-mat.mes-hall	cond-mat	Effect of the interface resistance in non-local Hanle measurements Estitxu Villamor,1 Luis E. Hueso,1,2 and Fèlix Casanova,1,2,* 1CIC nanoGUNE, 20018 Donostia-San Sebastian, Basque Country (Spain) 2IKERBASQUE, Basque Foundation for Science, 48011 Bilbao, Basque Country (Spain) *E-mail: [email protected] We use lateral spin valves with varying interface resistance to measure non-local Hanle effect in order to extract the spin-diffusion length of the non-magnetic channel. A general expression that describes spin injection and transport, taking into account the influence of the interface resistance, is used to fit our results. Whereas the fitted spin-diffusion length value is in agreement with the one obtained from standard non-local measurements in the case of a finite interface resistance, in the case of transparent contacts a clear disagreement is observed. The use of a corrected expression, recently proposed to account for the anisotropy of the spin absorption at the ferromagnetic electrodes, still yields a deviation of the fitted spin-diffusion length which increases for shorter channel distances. This deviation shows how sensitive the non-local Hanle fittings are, evidencing the complexity of obtaining spin transport information from such type of measurements. I. INTRODUCTION Pure spin currents are a key ingredient in the field of spintronics [1], which takes advantage not only of the charge of the electron, but also of its spin as an alternative to transport information. Lateral spin valves (LSVs), consisting of two ferromagnetic (FM) electrodes bridged by a non-magnetic (NM) channel (see Fig. 1(a)), are widely used to electrically create pure spin currents due to their non-local geometry, in which a spin- polarized current is injected from one of the FM electrodes (the injector) into the NM channel, and the pure spin current at the second FM electrode (the detector) is measured [2-15]. Hanle effect is based on the precession of spins under a perpendicular magnetic field. Due to the diffusive nature of the spin transport through the NM, there is dispersion on the time that spins need to travel from the FM injector to the detector, which in turn originates an angular dispersion on the orientation of the spins arriving at the FM detector. This causes the measured spin current at the FM detector to be zero for high enough magnetic fields [3-9]. In addition to being an effective tool for spin manipulation, it presents an important advantage in the study of the spin-injection and transport mechanisms, because it permits to obtain the spin polarization of the FM (PF), of the FM/NM interface (PI) and the spin-diffusion length of the NM (N) by using a single LSV [3-8], as opposed to the conventional non-local spin valve (NLSV) method, which needs several LSVs with different distances (L) between the FM electrodes in order to obtain these parameters [10-15]. However, Hanle measurements are very sensitive to different device details, such as the interface resistance [7,8] or the finite length of the NM channel [9]. The used model has also been widely discussed in terms of the liability of the obtained information. It has been suggested that it is not possible to measure Hanle effect with transparent interfaces [3,12] or that, if doing so, the equation needs to be carefully chosen [7,8]. In the present work, we analyze the validity of the general expression for the study of spin injection and transport in LSVs with any FM/NM interface resistance, presented from Ref. 5. We do so by fitting the equation to measurements of the Hanle EXPERIMENTAL DETAILS effect in LSVs with different interface resistances and comparing the obtained parameters to those obtained from the fitting of the NLSV measurements as a function of L in the very same devices. Whereas in the presence of a contact resistance both methods are in good agreement, we observe an anomalous behaviour for the case with transparent contacts, where there is a clear mismatch between both methods. While, for L larger than N, this disagreement can be solved by taking into account the recently proposed spin absorption anisotropy at the FM electrodes [8], it is still present when L is shorter than N, evidencing that an additional effect is influencing the spin precession. Our analysis shows the complexity of an accurate fitting of non-local Hanle measurements, a widely used technique to extract relevant spin-transport parameters. II. The LSVs employed in this work were fabricated by a two-step electron-beam lithography, ultra-high-vacuum (base pressure 110-8mbar) evaporation and lift-off process. In the first step, FM electrodes were patterned in PMMA resist on top of a Si/SiO2 substrate and 35nm of permalloy (Py) or cobalt (Co) were evaporated. Different widths of the FM electrodes were chosen, wF185nm and wF2140nm, in order to obtain different switching magnetic fields. In the second step, the NM channel with a width of wN190nm was patterned and Cu was thermally evaporated with a thickness t 150nm. Ar-ion milling was performed prior to the Cu deposition in order to remove resist left- overs [14]. The reason for choosing different materials as FM electrodes is the need of different FM/NM interface resistances. Py has given us high-quality transparent interfaces with a high spin polarization [13,14], whereas Co is easily oxidized allowing the fabrication of an interface with a non-zero resistance [15]. The interface resistance (RI) was measured in all samples, where a cross-shaped junction was fabricated in addition to the regular LSVs. Several samples were fabricated and measured (all of them containing LSVs with different L). Since the obtained results are reproducible [16], only two samples will be compared in this paper. Sample #1, containing Co/Cu LSVs, has an RIAI product (AI is the contact area) of 2.810-2m2 (the RI’s have values of RI1=1.6 and RI2=1, which fall in the intermediate regime, i.e. they are not transparent interfaces but they cannot be considered to be in the fully tunneling regime [17]). The measured RI at the Py/Cu junctions of sample #2 is negative, meaning that RI is of the order or lower than the resistance of the electrodes and RIAI10-3m2 [14,18,19]. Therefore, sample #2 is in the transparent regime [14,17]. All measurements were performed in a liquid He cryostat at 10 K, applying a magnetic field B and using a “DC-reversal” technique [11]. The voltage V, normalized to the applied current I, is defined as the non-local resistance RNL=V/I (see Fig. 1(a) for a scheme of the measurement). This magnitude is positive [negative] when the magnetization of the electrodes is parallel (P) [antiparallel (AP)], depending on the value of B. Two types of measurements have been performed: (i) RNL as a function of the in- plane magnetic field along the FM electrodes (BY from Fig. 1(a)), so-called NLSV measurements, and (ii) RNL as a function of the out-of-plane magnetic field (BZ from Fig. 1(a)), so-called Hanle measurements. In the case of NLSV measurements, the absolute value of RNL does not vary, only its sign does change when the magnetizations of the FM electrodes change from P to AP. The difference between the positive and the negative values of RNL is the spin signal, ΔRNL=2RNL, which is proportional to the spin accumulation at the FM detector (see lower inset of Fig. 1(b)). In the case of Hanle measurements, the magnitude of the measured RNL gradually changes from positive to negative (or vice versa) due to the precession of the spins. In addition, a reduction in RNL with BZ is superimposed, due to the angular dispersion of the orientation of the spins [6]. The expression used for fitting the Hanle measurements, obtained by solving the Bloch-type equation with an added one-dimensional spin-diffusion term applied to the LSV geometry [2,5,10,20], is the following: , (1) where and are an effective spin-diffusion length and an effective spin resistance of the NM, respectively, and RFi=FF/wNwFi is the spin resistance of the FM injector (i=1) or detector (i=2). F is the spin-diffusion length of the FM, N and F are the electrical resistivities of the NM and FM, sf is the spin-relaxation time of the NM and ωL=2BBZ/ħ is the Larmor frequency, with B the Bohr magneton and ħ the reduced Planck constant. Cu(=1.2μΩcm) is obtained by measuring the resistance of Cu for every L, and performing a linear fit for each sample, whereas Py(=22.4μΩcm) and Co(=11.5μΩcm) are obtained in two different devices, where Py and Co were deposited under the same nominal conditions as for the LSVs. We use Py=5nm [21] and Co=36nm [21]. The dimensions wN, wFi and L are measured by Scanning Electron Microscopy (SEM) for each device. Therefore, PI, PF and Cu are the parameters to be fitted from Hanle measurements. To be more precise, one needs to take into account that the magnetization of the FM electrodes can be tilted out-of-plane due to BZ. The following equation is used to correct for such tilting [3,4,6]: , (2) where “+” and “-” signs correspond to the P and AP magnetizations of the FM is the one from Eq. (1), and (BZ) is the angle between the electrodes, magnetization of the FM electrodes and BZ; its dependence with BZ can be extracted from the anisotropic magnetoresistance (AMR) measurements of the FM electrodes as a function of BZ [6]. Hence, in order to obtain the spin polarizations and spin-diffusion length from the Hanle measurements, the data was fitted to Eq. (2) (see upper inset of Fig. 1(b)). In the case of NLSV measurements we have an in-plane magnetic field BY, and Eq. (1) reduces to the following: , (3) where RN=NN/tNwN and N are the regular spin resistance and spin-diffusion length of the NM metal, respectively. The measured RNL as a function of L can, thus, be fitted to Eq. (3) (see Fig. 1 (b)). Even though the values obtained from both methods should be identical, the validity of Hanle measurements in the case of transparent contacts has already been called into question [3,7,8,12]. Figure 1: (a) SEM image of a LSV. The non-local measurement configuration, materials, and the directions of the in-plane and out-of-plane magnetic fields (BY and BZ) are shown. (b) Spin signal, RNL, as a function of the distance between FM electrodes, L, measured at 10 K for sample #1, which contains Co/Cu LSVs with an interface resistance of 1. Red solid line is a fit to Eq. (3). Lower inset: non-local resistance, RNL, as a function of BY measured at 10 K for the same Co/Cu LSV with L=500 nm. Solid (dotted) line indicates the decreasing (increasing) sweep of BY. RNL is tagged in the image. Upper inset: RNL as a function BZ measured at 10 K both for the parallel (red solid squares) and anti-parallel (blue empty squares) configuration of the FM electrodes for a Co/Cu LSV with L=1.5m. Black solid lines are fits to Eq. (2), using the RNL expression from Eq. (1). III. RESULTS AND DISCUSSION NLSV=0.0380.004 and Cu For sample #1, with a non-zero interface resistance, PI NLSV=0.0430.003, NLSV=1159100nm were obtained from the fitting of the PCo NLSV measurements to Eq. (3). The measured data and the fitting are shown in Fig. NLSV is in good agreement with our previous results [13,14], 1(b). The value of Cu NLSV has also been reported and discussed before [10,14]. whereas the low value of PCo Note that PI and PF are coupled, as seen from Eqs. (1)-(3), since sample #1 is not fully in (i.e. for the tunnelling the tunnelling regime. Only when or or transparent regimes [17]) they will decouple. For Hanle measurements, RNL as a function of BZ was measured for both P and AP magnetization states (see inset of Fig. 1(b)), with identical results. For all the LSVs Hanle=98725nm and with different L, a spin-diffusion length ranging between Cu Hanle=0.0440.001 and 110727nm, and an interface polarization ranging between PI 0.0480.001 were obtained. Due to the coupling of PI and PF in Eq. (1), the spin Hanle values show polarization of Co was fixed to PCo=0.038. The obtained Cu no substantial deviation from the NLSV values for any of the distances L (see Fig. 2a). Hanle and PI For sample #2, with transparent interfaces, we can approximate RI=0 in Eqs. (1)- (3) in order to obtain PPy and Cu. From NLSV measurements as a function of L we NLSV=0.360.01 obtained PPy for Hanle Hanle=55726nm and measurements, spin-diffusion 124558nm were obtained. The spin polarization of Py also changed between Hanle=0.340.01 and 0.630.02. Note that in this case RNL as a function of BZ was PPy only measured for the P magnetization of the FM electrodes [22]. As shown in Fig. 2a, NLSV=112562nm. However, ranging between Cu and Cu lengths 120.10.20.30.4 RNL (m)L (m)(a)Co/Cu-101-50050RNL ()BZ (T)-0.10.00.1-0.20.00.2 RNL (m) BY (mT)RNL(b)Hanle tends to reduce. PPy NLSV and Cu NLSV and, for longer L (L>>Cu Hanle starts to deviate from Cu Hanle changes with the opposite tendency to that of Cu Hanle values present a clear deviation from the NLSV values with a the obtained Cu NLSV) the agreement between both strong dependence on L: for low values of L (L<<Cu NLSV. The methods is excellent but, as L increases, Cu NLSV), the deviation highest discrepancy occurs for L~Cu Hanle, of Cu showing a coupling between both fitting parameters (Fig. 2b). The observed deviation NLSV is clearly originated from a bad fitting of the data [16]. However, this for L~Cu deviation is very reproducible for all measured samples and, thus, intrinsic to LSVs with transparent contacts [16]. Figure 3 shows the measured RNL as a function of BZ in sample #2 for the three mentioned regimes, together with the simulated curves of Eq. (2), using the RNL expression from Eq. (1). For the simulations (blue solid lines), we used the NLSV values obtained from the fittings of the NLSV measurements. The PPy NLSV, figure shows a good agreement between the measured data and Eq. (1) for L<<Cu Hanle and Cu NLSV. the same way there is an excellent agreement between the fitted Cu NLSV regime, the curves are far from reproducing the measured However, in the L~Cu NLSV regime, the simulated curve tend to converge to the measured data. For the L>>Cu data again. This result suggests that Eq. (2) (with the RNL from Eq. (1)) is not valid and additional effects should be considered in the spin transport in Cu. Hanle) obtained from the fitting of Eq. (2) (using RNL from Eq. Figure 2: (a) Spin-diffusion length of Cu (Cu (1)) to the RNL vs. BZ data, as a function of L, for sample #2 containing Py/Cu LSVs with transparent interfaces (red solid squares) and sample #1 containing Co/Cu LSVs with an interface resistance of 1 NLSV) (blue solid circles). Both Cu obtained for each sample from the fitting of Eq. (3) to the RNL vs. L data. (b) Spin polarization of Py Hanle) obtained from the same fitting of Eq. (2) (using RNL from Eq. (1)) to the RNL vs. BZ data, as a (PPy NLSV) obtained for the function of L, for sample #2. PPy same sample from the fitting of Eq. (3) to the ΔRNL vs. L data. Hanle and L are normalized to the spin-diffusion length of Cu (Cu Hanle is normalized to the spin polarization of Py (PPy 01230.81.21.60.00.40.81.2 PHanlePy/PNLSVPy L/NLSVCu(b) RI = 0 RI = 1 HanleCu/NLSVCu(a)Whereas Maasen et al. reported an anomalous behaviour of the parameters obtained from Hanle measurements due to a bad fitting, where the backflow of spins at the FM electrodes was not taken into account [7], this is not the case in the present work, since Eq. (1) explicitly takes into account the role of the interface resistances. Very recently, Idzuchi and co-workers [8] have proposed the difference in the spin absorption mechanisms for longitudinal and transverse spin currents as the reason of the disagreement in Hanle measurements in LSVs without tunnel barriers. According to this work, in LSVs with transparent interfaces, the different spin absorption by the FM electrodes for different current polarizations alters the spatial distribution of the chemical potential. Therefore, the spin transport is also altered, more pronouncedly for NLSV in the short L [8]. This could explain the strong deviation between Cu NLSV L~Cu regime. Instead, we find the opposite trend. NLSV regime, but one would expect an even stronger deviation in the L<<Cu Hanle and Cu NLSV and Cu In order to clarify this issue, Fig. 3 also shows the simulated curves of Eq. (2), using now the RNL expression from Eq. (S13) in Ref. 8 (red dashed lines). For the NLSV values obtained from the fittings of simulations, in addition to the PPy the NLSV measurements, a value of Gr=3.91014-1m-2 was used as the real part of the NLSV regime, spin-mixing conductance of the Py/Cu interface [8,23,24]. For the L>>Cu Eq. (S13) from Ref. 8 follows quite accurately the measured data. However, in the NLSV regime, the simulated curves start to deviate from the experimental results. L~Cu NLSV regime, where the measured data is more The discrepancy is highest for the L<<Cu affected by the precession, suggesting that the diffusion time is longer, an effect already reported to alter the fitted PF in LSVs using Eq. (3) [11]. Figure 3: RNL measured as a function of BY (black squares) for sample #2. L ranges from 200nm to 3m. All measurements were done for a parallel configuration of the Py electrodes at 10 K. Blue solid (red NLSV dashed) line is a simulation of Eq. (2) using RNL from Eq. (1) (Eq. (S13) from Ref. 8). PPy obtained from NLSV measurements were used, and a real part of the spin-mixing conductance between Py and Cu of Gr=3.9×1014Ω-1m-2 was assumed [8,23,24]. NLSV and Cu In order to obtain the value of Cu by fitting Eq. (2) with RNL from Ref. 8, we Hanle, which was left as the fitting parameter. This fixed all the parameters except for Cu was done for the sake of simplicity, given the complexity of Eq. (S13) from Ref. 8. NLSV are in good agreement in the L>>Cu Hanle as a function of L using that equation. For Figure 4 shows the obtained values of Cu Hanle values obtained by using Eq. (1), already shown in Fig. 2a, are comparison, the Cu Hanle and also plotted. The tendency is the same observed in the simulations, where Cu Hanle decreases when Cu NLSV. Therefore, Eq. (S13) from Ref. 8, which considers both the spin backflow L<<Cu and the anisotropic spin absorption at the FM/NM interfaces, does not work at the NLSV regime, showing that both mentioned effects are not enough to account for L<<Cu the disagreement between the current Hanle models and the measured curves. NLSV regime, but Cu A possible source of interference is the effect of nearby FM electrodes in the LSVs, but it is discarded by performing control experiments [16,25]. Taking into account that the discrepancy occurs at short channel distances (see green triangles in Fig. 4), the origin could be attributed to the use of a one-dimensional spin-diffusion model to derive the used equations [5,8], which could no longer be a good approximation. Indeed, the region of the NM channel under the FM injector, where the spin-polarized electrons spend time diffusing, has been shown to influence the effective spin polarization of the FM in LSVs [11] and would also affect the non-local Hanle curves [26]. Hanle obtained from the fitting of Eq. (2) by using Eq. (1) (red squares) and Eq. (S13) from Figure 4: Cu Ref. 8 (green tringles) as a function of L for sample #2, which consists of Py/Cu LSVs with transparent interfaces. Both Cu IV. CONCLUSIONS Hanle and L are normalized to Cu NLSV. To summarize, we performed non-local Hanle measurements in LSVs with transparent and finite interface resistances, and we compared the spin-diffusion length of Cu, Cu, obtained from such measurements to the one obtained from NLSV measurements as a function of L. Whereas, in the case where we have a finite FM/NM interface resistance, both methods are in excellent agreement, in the case of transparent interfaces an anomalous behaviour is observed, which depends on the distance L between both FM electrodes. Although taking into account the spin backflow and the anisotropic spin absorption at the FM/NM interfaces can explain some of the observed disagreements, an additional the non-local Hanle measurements is detected when L<<Cu. Such effect is beyond the understanding of the current one-dimensional spin diffusion models, evidencing the need for a more complete model that takes into account three dimensional effects. Hence, care should be taken interference that influences 01230.00.51.0 Eq. (1) Ref. 8HanleCu/NLSVCuL/NLSVCuRI = 0when obtaining spin-transport information from such type of measurements in LSVs with transparent interfaces. ACKNOWLEDGEMENTS The authors thank Asier Ozaeta and F. Sebastián Bergeret for fruitful discussions. This work was supported by the European Union 7th Framework Programme under the Marie Curie Actions (256470-ITAMOSCINOM) and the European Research Council (257654- SPINTROS), and by the Spanish MINECO under Project No. MAT2012-37638. E. V. thanks the Basque Government for a PhD fellowship (Grant No. BFI-2010-163). REFERENCES [1] Spin Current, edited by S. Maekawa, S. O. Valenzuela, E. Saitoh and T. Kimura (Oxford University Press, Oxford, 2012). [2] M. Johnson and R. H. Silsbee, Phys. Rev. Lett. 55, 1790 (1985). [3] F. J. Jedema, H. B. Heersche, A. T. Filip, J. J. A. Baselmans, and B. J. van Wees, Nature 416, 713 (2002). [4] G. Mihajlović, J. E. Pearson, S. D. Bader, and A. Hoffmann, Phys. Rev. Lett. 104, 237202 (2010). [5] Y. Fukuma, L. Wang, H. Idzuchi, S. Takahashi, S. Maekawa, and Y. Otani, Nat. Mater. 10, 527 (2010). [6] J.-C. Rojas Sánchez, P. Laczkowski, W. F. Savero Torres, M. Cubukcu, V. D. Nguyen, L. Notin, C. Beigné, C. Vergnaud, A. Marty, M. Jamet, L. Vila, and J. P. Attané, Appl. Phys. Lett. 102, 132408 (2013). [7] T. Maasen, I. J. Vera-Marun, M. H. D. Guimarães, and B. J. van Wees, Phys. Rev. B 86, 235408 (2012). [8] H. Idzuchi, Y. Fukuma, S. Takahashi, S. Maekawa, and Y. Otani, Phys. Rev. B 89, 081308(R) (2014). [9] M. Wojtaszek, I. J. Vera-Marun, and B. J. van Wees, Phys. Rev. B 89, 245427 (2014). [10] F. J. Jedema, M. S. Nijboer, A. T. Filip, and B. J. van Wees, Phys. Rev. B 67,085319 (2003). [11] M. Erekhinsky, A. Sharoni, F. Casanova, and I. K. Schuller, Appl. Phys. Lett. 96, 022513 (2010). [12] T. Kimura, T. Sato, and Y. Otani, Phys. Rev. Lett. 100, 066602 (2008). [13] E. Villamor, M. Isasa, L. E. Hueso and F. Casanova, Phys. Rev. B 87, 094417 (2013). [14] E. Villamor, M. Isasa, L. E. Hueso and F. Casanova, Phys. Rev. B 88, 184411 (2013). [15] E. Villamor, M. Isasa, S. Vélez, A. Bedoya-Pinto, L. E. Hueso, F. S. Bergeret, and F. Casanova, Phys. Rev. B 91, 020403(R) (2015). [16] See Supplementary Online Material at [URL will be inserted by AIP] for details on the fitting of the Hanle data and on experimental results in additional samples. [17] S. Takahashi and S. Maekawa, Phys. Rev. B 67, 052409 (2003). [18] R. J. Pedersen and F. L. Vernon, Jr., Appl. Phys. Lett. 10, 29 (1976). [19] J. M. Pomeroy and H. Grube, J. Appl. Phys. 105, 094503 (2009). [20] T. Valet and A. Fert, Phys. Rev. B 48, 7099 (1993). [21] S. Dubois, L. Piraux, J. M. George, K. Ounadjela, J. L. Duvail, and A. Fert, Phys. Rev. B 60, 477 (1999). [22] Py is a soft magnetic material, therefore, when starting from an initial AP state, the magnetization of one of the electrodes was always swiped back into the P state in the presence of a high enough BZ, preventing us from measuring RNL at the AP state for the whole range of BZ. [23] Y. Tserkovnyak, A. Brataas, and G. E. W. Bauer, Phys. Rev. B 66, 224403 (2002). [24] G. E. W. Bauer, Y. Tserkovnyak, D. Huertas-Hernando, and A. Brataas, Phys. Rev. B 67, 094421 (2003). [25] The effect of the nearby electrodes is considered due to the design of our devices, which consist of several LSVs on a row. However, by systematically varying the distance of the nearby Py/Cu LSVs with transparent interfaces, the same behaviour as in Fig. 2a is observed [16], ruling out any effect coming from the adjacent electrodes. [26] M. J. Erickson, Surface and Interface Effects in Nanoscopic Metallic Spin Transport Devices. Ph.D. Thesis. University of Minnesota, U.S. (2013). Effect of the interface resistance in non-local Hanle measurements Estitxu Villamor,1 Luis E. Hueso,1,2 and Fèlix Casanova,1,2 1CIC nanoGUNE, 20018 Donostia-San Sebastian, Basque Country (Spain) 2IKERBASQUE, Basque Foundation for Science, 48011 Bilbao, Basque Country (Spain) SUPPLEMENTAL MATERIAL S1. Hanle fittings from sample #2 NLSV (Fig. S1(a)), (ii) L~Cu NLSV (Figs. S1(b) and S1(c)), (iii) L>>Cu Figure S1 shows the measured RNL as a function of BZ in sample #2, as well as the fittings to Eq. (2) (using RNL from Eq. (1)), for three different regimes: (i) NLSV (Fig. L<<Cu S1(d)). The agreement between the measured data and the fitted curve is only good for NLSV are also in good agreement. For Hanle and Cu the first regime, where the values of Cu Hanle and Cu NLSV tend to be similar again. As seen from this the L>>Cu figure, in the intermediate regime the fitted curve tends to be wider than the measured Hanle (increasing, in turn, the data, which decreases considerably the fitted value of Cu value of PPy NLSV regime, Cu Hanle). Figure S1: Non-local resistance RNL measured as a function of BZ (black squares) for sample #2, which consists of Py/Cu LSVs with transparent interfaces. L ranges from 200 nm to 3 m. Red solid lines are fits to Eq. (2) (using RNL from Eq. (1)). S2. Additional samples In the main text, only two samples are compared for the sake of clarity: sample #1 (containing Co/Cu LSVs with a non-zero interface resistance) and sample #2 (containing Py/Cu LSVs with transparent interfaces). However, as mentioned in the text, more samples were measured. The results from these extra samples are shown in this section in order to emphasize that samples with the same interface resistance have the same behavior. Figure S2 shows the spin diffusion length of copper (Cu) obtained from Hanle, as a function of the distance L between FM electrodes; Hanle measurements, Cu both quantities are normalized to the spin diffusion length of Cu obtained from NLSV NLSV, for six different samples. For details about the measurements measurements, Cu and how the spin diffusion length is obtained we refer the reader to the main text. It is observed that in samples with a non-zero interface resistance (Fig. S2(a)) both methods NLSV), whereas are in good agreement (i.e., Cu in samples with transparent interfaces (Fig. S2(b)) there is a strong deviation which depends on L, and has the same trend for all the measured samples. Hanle does not deviate significantly from Cu In addition, the effect of the nearby electrodes is considered as a possible source of interference, due to the design of our devices, which consist of several LSVs on a row. However, by systematically varying the distance of the nearby Py/Cu LSVs with transparent interfaces, the same behavior is observed, ruling out any effect coming from the adjacent electrodes. Two of the control samples where the distance between the Py electrodes was varied are shown in Fig. S2(b). Hanle as a function of L for (a) Co/Cu LSVs with RI ~1Ω and (b) Py/Cu LSVs with Figure S2: λCu transparent interfaces. Both λCu Hanle and L are normalized to the λCu NLSV value of each sample.
1005.3075	2	1005	2010-11-29T20:27:42	Metal-terminated Graphene Nanoribbons	[ "cond-mat.mes-hall" ]	We have investigated structure, electronic, and magnetic properties of metal-terminated zigzag graphene nanoribbons (M-ZGNRs) by first-principles calculations. Two families of metal terminations are studied: (1) 3d-transition metals (TMs) Fe, Co, and Ni and (2) noble metals (NMs) Cu, Ag, and Au. All systems have spin-polarized edge states with antiferromagnetic (AFM) ordering between two edges, except Co-ZGNRs and Ni-ZGNRs which exhibit negligibly small energy differences between AFM and ferromagnetic states with the given ribbon width. In the AFM state the TM terminations transform semiconducting ZGNRs into metallic ones while the band gap remains in ZGNR with NM terminations. Ferromagnetic states of M-ZGNRs with TM terminations show a high degree of spin polarization at the Fermi energy. We predict a large magnetoresistance in Fe-ZGNR junctions with a low, uniform magnetic switching field.	cond-mat.mes-hall	cond-mat	Metal-terminated Graphene Nanoribbons Department of Physics and Quantum Theory Project, University of Florida, Gainesville, Florida 32611, USA Yan Wang, Chao Cao, and Hai-Ping Cheng We have investigated structure, electronic, and magnetic properties of metal-terminated zigzag graphene nanoribbons (M-ZGNRs) by first-principles calculations. Two families of metal terminations are studied: 1) 3d-transition metals (TM) Fe, Co, and Ni; and 2) noble metals (NM) Cu, Ag, and Au. All systems have spin-polarized edge states with antiferromagnetic (AFM) ordering between two edges, except Co-ZGNRs and Ni-ZGNRs which exhibit negligibly small energy differences between AFM and ferromagnetic (FM) states with the given ribbon width. In the AFM state the TM terminations transform semiconducting ZGNRs into metallic ones, while the band gap remains in ZGNR with NM terminations. Ferromagnetic states of M-ZGNRs with TM terminations show a high degree of spin polarization at the Fermi energy. We predict a large magnetoresistance in Fe-ZGNR junctions with a low, uniform magnetic switching field. PACS numbers: 73.22.Pr, 75.47.-m, 71.15.Mb I. INTRODUCTION Giant magnetoresistance (GMR)1,2 and tunneling magne- toresistance (TMR) effects3 -- 7 in spin-valves and magnetic tun- nel junctions have been the focus of spintronics studies in the past two decades. Conventional GMR and TMR devices, based on inorganic multilayered structures whose critical ele- ment is a sandwich of two magnetic layers separated by non- magnetic metallic or insulating layers, exhibit a large change in conductance when the orientation of the magnetic layers is changed from parallel (P) to antiparallel (AP). The mag- netic layers are usually made of transition metal ferromagnets, which gives a high spin polarization at the Fermi energy. Graphene nanoribbons (GNRs) with armchair (AGNR) or zigzag (ZGNR) edges8 -- 12, a novel organic material system produced by cutting graphene along two high-symmetry crys- tallographic directions, have recently attracted attention in spintronics. ZGNR is predicted to be a semiconductor with spin-polarized electronic states localized on the two edges that are individually ferromagnetically (FM) ordered but an- tiferromagnetically (AFM) coupled to each other through the graphene backbone13. Son et al.14 show that ZGNRs become half-metallic when an external transverse electric field is ap- plied across the edges. Kim et al.15 propose a spin-valve based on a ZGNR connected by two ferromagnetic electrodes using a local magnetic field to switch the magnetization of one of the electrodes. Very recently a magnetoresistance de- vice based on the transformation from the semiconducting state to the metallic state of the ZGNR has been proposed by Munoz-Rojas et al.16; an estimated 200 T magnetic field is required to switch the AFM to the FM state at room tem- perature. Although these work demonstrate convincingly that spin-polarization and magnetoresistance can be realized us- ing ZGNRs, the large magnitude of the electric field required to close the bandgap, the need of local magnetic control or a huge magnetic field, and small spin correlation length due to the lack of magnetic anisotropy in ZGNRs17 are serious lim- iting factors for future applications. Moreover, the predicted magnetic states in ZGNRs have not been experimentally ob- served so far. Most of the theoretical studies on GNRs for spintron- ics applications so far only address ribbons with hydro- gen terminations14 -- 16. Recent experiments have shown that graphitic surface curvature can be used as constraint and guide in which metal clusters aggregate to form linear islands18,19. The presence of metal atoms adsorbed on the graphene20 -- 23 or GNR24 changes its electronic and magnetic properties and thus is certainly of great interest in physics. In this paper we show that hybrid metal-terminated ZGNRs (M-ZGNRs) can be excellent candidate for spintronic appli- cations. We studied various M-ZGNRs with terminations of 3d transition metals (TM) Fe, Co, Ni and noble metals (NM) Cu, Ag, Au using first-principles calculations. Compared with hydrogen terminated ZGNR (H-ZGNR), the TM and NM ter- minations lead to drastically different and unexpected effects in the structural, electronic and magnetic properties of the rib- bon. A very large spin polarization is predicted for M-ZGNR with TM terminations in FM state. We propose a new type of junction with a low, uniform magnetic switching field that utilizes the electronic properties of M-ZGNRs, for which a magnetoresistance value as high as 200% can be obtained. II. COMPUTATIONAL METHODS The electronic structure calculations are performed using density functional theory (DFT) implemented in the PWSCF code25. Each M-ZGNR is simulated within a supercell geom- etry containing two pristine ZGNR unit-cells with 40 C atoms and 4 metal atoms at two edges. Ultrasoft pseudo-potentials with kinetic energy cutoff of 500 eV are employed in all sim- ulations. For the exchange and correlation functional we used the Perdew-Burke-Ernzerh generalized gradient approxima- tion (GGA)26. A vacuum layer of 15 Å is used between two edges and of 15 Å between two graphene planes to prevent in- teraction between adjacent images. Brillouin-Zone sampling uses a grid of 12 Monkhorst-Pack27 k-points along the peri- odic direction of the ribbon. The optimization of atomic po- sitions proceeds until the force on each atom is less than 0.01 eV/Å. The transport calculations of the proposed Fe-ZNGR mag- netic junction are performed by using the TRANSIESTA code37, which is based on nonequilibrium Green's function 2 minations, the relaxed structures are LA type. For Cu-ZGNR the most favorable structure is LB type, which is 0.98 eV/unit- cell more stable than the LA type structure, in good agreement with the result of Wu et al.28. The Fe atoms dimerize at the edge of Fe-ZGNR due to the Peierls distortion29. The most fa- vorable structures for Ag-ZGNR and Au-ZGNR are ZZ type, with a metal-metal distance larger than the length of pristine ZGNR unit-cell. The binding energies Eb of the metal atom in M-ZGNRs, defined as Eb = (EM−ZGNR − EZGNR)/4 − Eatom M , and the formation energy of the metal chain, Echain formation = Echain M /2 − Eatom M , are shown in Table. I. The large differences between Eb and Echain formation indicate a large direct binding be- tween the metal and carbon atoms at the ribbon edge. Bader analysis based on the real-space-charge density30 for the M- ZGNRs shows charge transfer from metal atom to C atom for all cases. The amounts of charge transfer for Au-ZGNR and Ag-ZGNR are 0.08 and 0.19 e/metal-atom, respectively. Such a difference can be important for nanocatalysis. For each M-ZGNR considered, the ground state has FM spin ordering of the metal atoms at each edge, while AFM order between the two edges is favored over FM, similar to H-ZGNRs and ZGNRs without H-passivation13. AFM cou- pling of the two edges is energetically more stable due to the indirect RKKY exchange interaction as a result of the bipar- tite lattice31, since the metal terminations always stand at the opposite sublattices at the two edges. For Fe-ZGRNs, the en- ergy difference ∆E between the FM and AFM states increases with decreasing ribbon width32 due to increased interaction between enhanced spins in inner C sites inside the ribbon. Without the ribbon host the two Fe chains can only show di- rect FM exchange coupling through vacuum, and the calcu- lated ∆E of the two Fe chains becomes less than 0.1 meV/Å when the distance is larger than 8 Å. Nonetheless, as shown in Table. I, the ∆E for all M-ZGNRs are less than that of H- ZGNR, indicating a relatively small effective magnetic cou- pling between two edges, especially for TM terminations. In general, the total magnetic moment of the system comes mostly from the metal atoms and their nearest-neighbor C atoms at the M-ZGNR edges. For TM-ZGNRs, the edge C atom presents magnetization antiparallel to the nearby TM atom. For Co-ZGNR and Ni-ZGNR in particular (see Fig. 1) the magnetic moments of the ribbon become more local- ized on the metal atoms at the edges compared to the inner C atoms in the ribbon, which leads to a negligibly small ex- change energy shown in Table. I. For NM-ZGNRs, the metal atom shows magnetization parallel to that of the neighboring C atom at the edge, of order 0.01 µB/metal-atom, mainly due to the proximity effect. The large difference between magnetic moments in TM and NM terminations reflects their different physical origins. In TM-ZGNRs, the relatively large moments of TM atoms come from unfilled d-shells and the TM-C co- valent bond reduces the number of unpaired d-electrons, giv- ing a reduced moment compared to the freestanding TM atom or monoatomic chain. For Cu, Ag and Au terminations with filled d-shells, the s valence electron saturates the dangling σ bonds similar to the hydrogen terminations in H-ZGNRs. The proximity-induced local magnetic moments for Cu, Ag, Au atoms are negligible, but the total edge moments of about 0.3 FIG. 1: (Color online). Four types of metal-terminated ZGNR: (a) Ni-ZGNR as linear A type (LA); (b) Cu-ZGNR as linear B type (LB); (c) Fe-ZGNR as dimerized linear type (DL); (d) Au-ZGNR as zigzag type (ZZ). The atomic configuration appears on the left, and on the right the corresponding spin density (ρ = ρα − ρβ) isosurfaces of the M-ZGNRs. The red (grey) and the blue (dark) isosurfaces denote ±0.007 e/Å3, respectively. The dashed box shows the edge view. formalism and DFT as implemented in the SIESTA package37. In order to reproduce the correct band structure of the lead, it is necessary to use double-ζ basis with polarization for Fe atoms, while single-ζ basis with polarization is sufficient for C atoms. The equivalent plane-wave cutoff for the real-space grid of 150 Ry is used in throughout the calculations. Current- voltage characteristics within small bias are calculated from the transmission coefficient T (E) using I = R T (E)[ f (E − µL) − f (E − µR)]dE. The magnetoresistance can therefore be calculated via MR = (IP − IAP)/IAP. III. RESULTS AND DISCUSSION A. Structure, electronic, and magnetic properties To determine the lowest-energy structures of the M-ZGNR, first we carried out structural relaxations for four possible ter- mination configurations: the linear A type (LA), the linear B type (LB), the dimerized linear type (DL), and the zigzag type (ZZ), as shown in Fig. 1. Both edges of the ribbon have the same configuration for all systems. The calculated ground- state configurations as well as energetic and electronic prop- erties are summarized in Table I. We find that for all cases the metal atoms bond strongly with edge carbon atoms, though the bonding configurations are quite different from one an- other, as shown in Fig. 1. For M-ZGNRs with Co and Ni ter- TABLE I: Structural, energetic and electronic properties of M-ZGNR. Columns show: metal termination, structures, metal-metal (dM−M) and metal-carbon (dM−C) bond length (in Å), the binding energy Eb and the formation energy of the metal chain Echain formation (in eV per metal atom), energy difference ∆E = EFM − EAFM between the ferromagnetic and the antiferromagnetic state (in meV/unit-cell), charge transfer of the metal atom (in e) from Bader analysis, total magnetic moment mtot of the ribbon in the FM state (in Bohr magneton µB per edge termination), local magnetic moment of the metal atom µM and its nearest-neighbor C atom µC (in µB), numbers of conduction bands crossing the Fermi level (nAFM: the AFM state; nα F M: majority and minority-spin in the FM state), and spin polarization PFM of the ribbon at the Fermi energy in the FM state. Results for H-terminated ZGNR are also listed for comparison. FM and nβ 3 structures Fe-ZGNR Co-ZGNR Ni-ZGNR Cu-ZGNR Ag-ZGNR Au-ZGNR H-ZGNR DL LA LA LB ZZ ZZ LA aFe dimerization. dM−M 2.19(2.74)a 2.46 2.46 2.46 2.74 2.71 2.46 dM−C 1.87 1.82 1.80 2.07 2.15 2.07 1.09 Eb 4.13 4.26 4.44 3.42 2.38 3.00 4.89 Echain formation 1.86 1.79 1.74 1.37 1.20 1.65 0.14 ∆E 6.8 1.3 <0.1 10.1 7.8 11.5 15.1 Charge transfer mtot 2.19 1.17 0.18 0.26 0.28 0.26 0.26 −0.31 −0.22 −0.19 −0.41 −0.19 −0.08 0.01 µC µM 2.47 −0.27 1.33 −0.11 0.23 −0.01 0.29 0.01 0.26 0.01 0.02 0.26 0.01 −0.30 6 10 8 0 0 0 0 nAFM nα F M) F M(nβ 3(4) 3(7) 3(5) 1(1) 1(1) 1(1) 1(1) PFM 82 % 78 % 55 % ≈ 0 ≈ 0 ≈ 0 ≈ 0 B # (a) (b) Majority Minority EF 2 1 0 ) V e ( y g r e n E -1 -2 0 k ( /L) 1 B DOS 0 # k ( /L) 1 DOS (d) Majority Minority EF (c) ) V e ( y g r e n E 2 1 0 -1 -2 0 k ( /L) 1 DOS 0 k ( /L) 1 DOS FIG. 2: (Color online). Energy band structures (left) and density of states (DOS, right, in states/eV) for the M-ZGNRs. (a): Fe-ZGNR in AFM state; (b): Fe-ZGNR in the FM state; (c) Au-ZGNR in the AFM state; (b): Au-ZGNR in the FM state. The dashed (red) and solid (blue) lines show the majority and minority spin, respectively. In the AFM state, majority and minority bands (DOS) coincide. µB/metal-atom are very close to that of H-ZGNR. Although Mermin-Wagner theorem33 excludes long-range order in 1D spin lattice model at finite temperature which would limit the dimension of devices based on H-ZGNRs to several nanome- ters at room temperature17, improvements can be expected in systems of M-ZGNRs by increasing the magnetic anisotropy as well as the spin correlation length. The ground-state TM-ZGNRs are all metallic and the NM- ZGNRs are all semiconductor. We show the band structures and density of states (DOS) of Fe-ZGNR and Au-ZGNR in Fig. 2. For Fe-ZGNR, both AFM and FM states are metallic. In the FM state, more flat bands of minority spin cross the Fermi level compared to those of the majority spin, resulting in a high spin polarization at the Fermi level, as can be seen in Fig. 2 (b). A detailed analysis of the projected DOS indicates that the states near the Fermi level are mainly from the Fe d orbital. Spin polarization PFM can be calculated as PFM = Nα − Nβ/(Nα + Nβ), where Nα (Nβ) represents the DOS of majority (minority) spins at the Fermi energy. The PFM of the Fe-ZGNR at EF is about 82% which is much higher than that of bulk Fe34,35. This implies that in the ballistic transport regime, the conductance of electrons with one type of spin is larger than the opposite spin. Similar to Fe-ZGNRs, the FM states of Co and Ni terminated ribbons also show high spin polarization, 78% and 55% respectively. However, all studied NM-ZGNRs, have a bandgap compa- rable to that of the hydrogenated ribbon (0.32 eV) in the AFM ground-state. The band structure and density of states of Au- ZGNR as a typical case for NM-ZGNRs are displayed in Fig. 2 (c) and (d). In the FM metallic state the Au-ZGNR shows a large spin polarization only at the energies slightly away from the EF. We also note that the band structure of NM-ZGNR is very sensitive to the edge structure. The metastable LA type of Au-ZGNR is metallic for both FM and AFM states. B. Metal-terminated graphene nanoribbon junction Our calculations of truncated Fe-ZGNRs provide an inter- esting result: if the transverse armchair edge is terminated by Fe, the strong magnetic coupling will align all edge spins in the same direction. Such highly spin polarized FM states can be useful in spintronics; in particular, a junction consisting of two TM-ZGNR leads separated by a spacer can operate as a magnetoresistance device. Fig. 3 (a) is a prototype multi- layer junction stacking three layers of Fe-ZGNRs. The center spacer can be a nonmagnetic metal such as Cr or Cu, since (a) P AP 3 2 1 4 i i n o s s m s n a r T (Color online). FIG. 3: (a) Schematic figure of a multi- layer Fe-ZGNR/spacer/Fe-ZGNR junction prototype with antiparal- lel ground-state configuration. (b) Calculated ground-state (AFM) spin density isosurfaces of the Fe-AGNR with a width of 7.2 Å. the behavior of the AFM exchange coupling between mag- netic layers through a nonmagnetic spacer is well-known36 and a GMR junction can thus be made; one can also use a small piece of non-conducting AGNR connected to the two leads via Fe chains, as the spacer which will provide a mag- netoresistance junction. The key issue is the AP configura- tion being the ground state through the spacer: the junction can be switched to the P configuration under a uniform exter- nal magnetic field applied to both leads, without the need for a local magnetic field. A conductance difference between P and AP is in general expected. Our calculations show that the Fe-terminated two armchair edges favor either AFM or FM configurations depending on the width of the spacer. In Fig. 3 (b) we show an Fe-AGNR with a width of 7.2 Å which has an AFM ground-state that is 1.1 meV/Å lower than the FM state. AFM coupling of the two armchair edge states of the spacer can lead to an AP ground-state magnetic coupling between the two Fe-ZGNR leads. This energy difference can be tuned by increasing the thickness of the lateral Fe chain and by using multiple layers as shown in Fig. 3 (a). We also note that such a junction can be made by two Fe-AZGNR leads separated by a Fe-ZGNR spacer with similar function39. To illustrate the possible magnetoresistance, we have per- formed quantum transport calculations for a model magnetic tunnel junction based on two semi-infinite Fe-ZGNR leads. A 3.1 Å gap is set between two leads to represent the Fe- AGNR spacer. The spin-dependent current-voltage charac- teristics for small bias (linear response region) are calculated from the energy-dependent transmission coefficients in the ballistic tunneling regime. Large transmission differences are found near the Fermi energy, as shown in Fig. 4 (a). By com- paring the current-voltage characteristics between P and AP configurations in Fig. 4 (b), we find a magnetoresistance up to 200% within the small bias region, which is comparable to the TMR ratio observed in conventional Fe/MgO/Fe sandwiched junctions6,7. Negative magnetoresistance is also found in the region from 0 to 0.2 eV, as shown in Fig. 4 (c), primary due to the existence of several large peaks in the transmission of AP configuration which are very close to the Fermi energy. Finally we discuss the operational conditions of our pro- posed magnetic junction of Fe-ZGNR/Fe-AGNR/Fe-ZGNR. 0 -1.0 (b) (c) 0.4 0.3 0.2 0.1 0.0 200 100 0 ) A ( t n e r r u C ) % ( R M -0.5 0.0 E-EF (eV) 0.5 1.0 IP IAP -100 0.0 0.1 0.2 0.3 V (V) 0.4 0.5 FIG. 4: (Color online). Transport properties for a model magnetic tunnel junction with two semi-infinite Fe-ZGNR leads separated by a 3.1 Å vacuum gap. (a) Energy-dependent transmission coefficients and (b) current-voltage characteristics within small bias region. The solid (red) and dashed (blue) lines represent the results for parallel (P) and antiparallel (AP) magnetic configuration of the two leads, respectively. (c) Calculated magnetoresistance (MR) as a function of bias voltage. For a lead of infinite size any external field would switch the magnetization of the ZGNR lead. For a junction with finite- size leads, the critical switching magnetic field is defined as B = ∆E/gµBMtot, where ∆E is the energy difference between P and AP configurations corresponding to the Zeeman energy splitting, Mtot is the total magnetic moment of the junction in the P configuration, g = 2 for graphene and µB=0.058 meV/T. A junction with two Fe-ZGNR leads of length 10 nm and a Fe- AGNR spacer 2.4 nm in length and 7.2 Å in width produces a ∆E of about 26 meV, which is comparable to room tem- perature kT . The junction with ideal edge termination of Fe atoms has a Mtot of 415 µB, thus resulting in a critical switch- ing field as low as 0.54 T, which is much less than that of the junction proposed by Munoz-Rojas et al.16. In experiments the possible formation of Fe wires of finite width at the Fe- ZGNR lead edges would enhance the total magnetization and thus further reduce the effective switching field. This junction shows a big advantage over the devices based on H-ZGNRs which mainly rely on ferromagnetic electrodes as well as lo- cal magnetic control15 or a huge switching magnetic field16. Current nano-size lithographic techniques can provide the ba- sis for fabricating and patterning of such junctions. IV. CONCLUSION In summary, we have found that TM and NM terminated ZGNRs have completely different structures and magnetic properties. The TM terminations transform semiconducting ZGNRs to metallic, while NM termination does not affect the semiconducting nature of the ZGNR. ZGNRs with TM ter- minations and ferromagnetically coupled edges states show a high spin polarization at the Fermi energy. We propose a mag- netic junction with low, uniform switching field using the Fe- ZGNRs as the ferromagnetic electrodes, and quantum trans- port calculations indicate a large magnetoresistance is possi- ble in such junctions. Our results outline a roadmap to ob- serve the magnetic order in graphene nanoribbon and hope- fully stimulate experimental activities in the near future. 5 Note added in proof. In an extended study we have exam- ined the ribbon width dependence of magnetic coupling be- tween two edges of metal-terminated ZGNRs and AGNRs us- ing projector augmented wave (PAW) potentials within DFT, and the results will be published elsewhere. Acknowledgments This work was supported by US/DOE/BES/DE-FG02- 02ER45995. The authors acknowledge DOE/NERSC and the UF-HPC center for providing computational resources. C. Cao acknowledges support from NSFC (Grant No. 10904127). 1 M. N. Baibich et al., Phys. Rev. Lett. 61, 2472 (1988). 2 G. Binasch, P. Grunberg, F. Saurenbach, and W. Zinn, Phys. Rev. B 39, 4828 (1989). 3 T. Miyazaki and N. Tezuka, J. Magn. Magn. Mater. 139, L231 (1995). 4 J. S. Moodera, L. R. Kinder, T. M. Wong, and R. Meservey, Phys. Rev. Lett. 74, 3273 (1995). 5 W. H. Butler, X.-G. Zhang, T. C. Schulthess, and J. M. MacLaren, Phys. Rev. B 63, 054416 (2001). 6 S. S. P. Parkin, C. Kaiser, A. Panchula, P. M. Rice, B. Hughes, M. Samant, and S.-H. Yang, Nat. Mater. 3, 862 (2004) 7 S. Yuasa, T. Nagahama, A. Fukushima, Y. Suzuki, and K. Ando, Nat. Mater. 3, 868 (2004). 8 K. Nakada, M. Fujita, G. Dresselhaus, and M. S. Dresselhaus, Phys. Rev. B 54, 17954 (1996). 9 M. Fujita, K. Wakabayashi, K. Nakada, and K. Kusakabe, J. Phys. Soc. Jpn. 65, 1920 (1996). 10 M. Y. Han, B. Ozyilmaz, Y. Zhang, and P. Kim, Phys. Rev. Lett. 98, 206805 (2007). 11 B. Ozyilmaz, P. Jarillo-Herrero, D. Efetov, D. A. Abanin, L. S. Levitov, and P. Kim, Phys. Rev. Lett. 99, 166804 (2007). 12 X. Li, X. Wang, L. Zhang, S. Lee, and H. Dai, Science 319, 1229 (2008). 13 Y.-W. Son, M. L. Cohen, and S. G. Louie, Phys. Rev. Lett. 97, 216803 (2006). 14 Y.-W. Son, M. L. Cohen, and S. G. Louie, Nature 444, 347 (2006). 15 W. Y. Kim and K. S. Kim, Nature Nanotech. 3, 408 (2008). 16 F. Munoz-Rojas, J. Fern´andez-Rossier, and J. J. Palacios, Phys. Rev. Lett. 102, 136810 (2009). 17 O. V. Yazyev and M. I. Katsnelson, Phys. Rev. Lett. 100, 047209 (2008). 18 M. Schmidt, N. K´ebaıli, A. Lando, S. Benrezzak, L. Baraton, Ph. Cahuzac, A. Masson, and C. Br´echignac, Phys. Rev. B 77, 205420 (2008). 19 A. F. Kemper, H-P. Cheng, N. K´ebaıli, S. Benrezzak, M. Schmidt, A. Masson, and C. Br´echignac, Phys. Rev. B 79, 193403 (2009). 20 K. T. Chan, J. B. Neaton, and M. L. Cohen, Phys. Rev. B 77, 235430 (2008). 21 H. Johll, H. C. Kang and E. S. Tok, Phys. Rev. B 79, 245416 (2009). 22 A. V. Krasheninnikov, P. O. Lehtinen, A. S. Foster, P. Pyykko, and R. M. Nieminen, Phys. Rev. Lett. 102, 126807 (2009). 23 C. Cao, M. Wu, J. Jiang, and H.-P. Cheng, Phys. Rev. B 81, 205424 (2010). 24 V. A. Rigo, T. B. Martins, A. J. R. da Silva, A. Fazzio, and R. H. Miwa, Phys. Rev. B 79, 075435 (2009). 25 P. Giannozzi, S. Baroni, N. Bonini, M. Calandra, R. Car, C. Cavazzoni, D. Ceresoli, G. L. Chiarotti, M. Cococcioni, I. Dabo, A. Dal Corso, S. Fabris, G. Fratesi, S. de Gironcoli, R. Gebauer, U. Gerstmann, C. Gougoussis, A. Kokalj, M. Lazzeri, L. Martin- Samos, N. Marzari, F. Mauri, R. Mazzarello, S. Paolini, A. Pasquarello, L. Paulatto, C. Sbraccia, S. Scandolo, G. Sclauzero, A. P. Seitsonen, A. Smogunov, P. Umari, R. M. Wentzcovitch, J. Phys.: Condens. Matter, 21, 395502 (2009). 26 J. P. Perdew, K. Burke, and M. Ernzerhof, Phys. Rev. Lett. 77, 3865 (1996). 27 H. J. Monkhorst and J. D. Pack, Phys. Rev. B 13, 5188 (1976). 28 M. Wu, Y. Pei, and X. C. Zeng, J. Am. Chem. Soc. 132, 5554 (2010). 29 R. E. Peierls, Quantum theory of solids. Oxford University Press, London (1955). 30 G. Henkelman, A. Arnaldsson, and H. J´onsson, Comput. Mater. Sci. 36, 254-360 (2006). 31 S. Saremi, Phys. Rev. B 76, 184430 (2007). 32 S. V. Ong, R. Roblesa and S. N. Khanna, Chem. Phys. Lett. 492, 127 (2010). 33 N. D. Mermin and H. Wagner, Phys. Rev. Lett. 17, 1133 (1966). 34 P. M. Tedrow and R. Meservey, Phys. Rev. B 7, 318 (1973). 35 R. J. Soulen Jr., J. M. Byers, M. S. Osofsky, B. Nadgorny, T. Am- brose, S. F. Cheng, P. R. Broussard, C. T. Tanaka, J. Nowak, J. S. Moodera, A. Barry, J. M. D. Coey, Science 282, 85 (1998). 36 P. A. Grunberg, Rev. Mod. Phys. 80, 1531 (2008). 37 J. M. Soler, E. Artacho, J. D Gale, A. Garc´ıa, J. Junquera, Pablo Ordej´on and D. S´anchez-Portal, J. Phys.: Condens. Matter 14, 2745 (2002). 38 M. Brandbyge, J. L. Mozos, P. Ordej´on, J. Taylor, and K. Stokbro, Phys. Rev. B 65, 165401 (2002). 39 The Fe-AGNR in the FM state also shows a large spin polarization at the Fermi energy, as confirmed by our calculation.
1512.08153	1	1512	2015-12-26T22:51:15	On-chip integrated, silicon-graphene plasmonic Schottky photodetector, with high responsivity and avalanche photogain	[ "cond-mat.mes-hall" ]	We report an on-chip integrated metal-graphene-silicon plasmonic Schottky photodetector with 85mA/W responsivity at 1.55 um and 7% internal quantum efficiency. This is one order of magnitude higher than metal-silicon Schottky photodetectors operated in the same conditions. At a reverse bias of 3V, we achieve avalanche multiplication, with 0.37A/W responsivity and avalanche photogain~2. This paves the way to graphene integrated silicon photonics.	cond-mat.mes-hall	cond-mat	On-chip integrated, silicon-graphene plasmonic Schottky photodetector, with high responsivity and avalanche photogain Ilya Goykhman1, Ugo Sassi1, Boris Desiatov2, Noa Mazurski2, Silvia Milana1, Domenico de Fazio1, Anna Eiden1, Jacob Khurgin3, Joseph Shappir2, Uriel Levy2, and Andrea C. Ferrari1∗ 1Cambridge Graphene Centre, University of Cambridge, 9 JJ Thomson Avenue, Cambridge CB3 OFA, UK 2Department of Applied Physics, The Benin School of Engineering and Computer Science, The Hebrew University, Jerusalem, 91904, Israel and 3Department of Electrical and Computer Engineering, Johns Hopkins University, Baltimore, Maryland 21218, USA We report an on-chip integrated metal-graphene-silicon plasmonic Schottky photodetector with 85mA/W responsivity at 1.55µm and 7% internal quantum efficiency. This is one order of magnitude higher than metal-silicon Schottky photodetectors operated in the same conditions. At a reverse bias of 3V, we achieve avalanche multiplication, with 0.37A/W responsivity and avalanche photogain∼ 2. This paves the way to graphene integrated silicon photonics. Over the past decade silicon photonics[1] has pro- gressed towards miniaturization and on-chip integra- tion of optical communication systems, where data are encoded by light signals and distributed over waveg- uides, rather than conventional metal-based electronic interconnects[2, 3]. So far, a variety of passive and active photonic devices in Si have been demonstrated, including low-loss (∼0.3dB/cm) waveguides[4, 5], high- quality factor optical cavities (∼106)[6 -- 8], high-speed (tens of GHz)[9 -- 11] electro-optic modulators and Si light sources based on Raman gain[12, 13]. The wealth of devices, together with the well established complemen- tary metal-oxide-semiconductor (CMOS) fabrication pro- cesses make Si photonics a promising technology for short range (board-to-board, chip-to-chip or intra-chip)[1] op- tical communications. The photodetector (PD) is one of the basic build- ing blocks of an opto-electronic link, where it performs optical-to-electrical signal conversion. Development of Si PDs for telecom wavelengths (1.3-1.6µm) based on the mature CMOS technology is an essential step for monolithic, on-chip, opto-electronic integration[1]. While Si PDs are widely employed in the visible spectral range[14](0.4-0.7µm), they are not suitable for detect- ing near-infrared (NIR) radiation above 1.1µm, because the energy of NIR photons at telecom wavelengths (0.78- 0.95eV) is not sufficient to overcome the Si bandgap (in- direct, 1.12eV) and induce photogeneration of electron- hole (e-h) pairs, i.e no photocurrent (Iph) is generated. Over the years, the Si photonics industry has devel- oped solutions to overcome this deficiency by combin- ing Ge (bandgap 0.67eV) with Si[15 -- 17] and integrating compound (III-V) semiconductors on the Si chip[18, 19] using wafer bonding techniques[20]. While these ap- proaches provide a path towards photodetection in the telecom spectral range[1], they either require advanced and complex fabrication processes in the case of SiGe devices[21], or rely on III-V materials systems not com- patible with standard CMOS technology[14]. Motivated by the need of developing Si based PDs for telecom wave- lengths several approaches were proposed to date. These include two photon absorption (TPA)[22, 23], defect me- diated band-to-band photogeneration via mid-bandgap localized states[24 -- 26], deposition of polysilicon[27] for NIR absorption, enhancement by optical cavities[23, 25 -- 29]. However, in the cases of defect-mediated and poly- Si PDs, the overall concentration of defects in the Si lattice affects both Iph and the leakage (dark) current Idark[14, 24, 25], i.e a higher defects density increases both the sub-bandgap optical absorption and thermal generation processes[14], thus increasing both Iph and Idark[14, 24, 25]. As a result, PDs with reduced de- fects concentration are typically needed[24, 25], coupled to optical resonators to amplify the optical power and to enhance the absorption without increasing either de- vice length or defect density. On the other hand, nonlin- ear optical process, such as TPA, could potentially con- tribute to all-Si NIR-PDs[1], but this approach requires increased optical power[23] with respect to linear absorp- tion, or PD integration with high quality factor cavities to achieve enhanced photon density[23]. An alternative exploits internal photoemission (IPE) in a Schottky diode[14, 30, 31]. In this configuration, photoexcited ("hot") carriers from the metal are emit- ted to Si over a potential ΦB, called Schottky barrier (SB), that exists at the metal-Si interface[14, 32]. In Si, the injected carriers are accelerated by an electric field in the depletion region of a Schottky diode and then col- lected as a photocurrent at the external electrical con- tacts. Typically, a SB is lower (0.2-0.8eV) than the Si bandgap[14], thus allowing photodetection of NIR pho- tons with energy hν > ΦB. The advantages of Schot- tky PDs are the simple material structure, easy and inexpensive fabrication process, straightforward integra- tion with CMOS technology and broadband (0.2-0.8eV) operation[14]. The main disadvantage is the limited IPE quantum yield, i.e the number of carriers emitted to Si divided by the number of photons absorbed in the metal, typically< 1%[33, 34]. This is mainly due to the momen- tum mismatch between the electron states in the metal and Si, resulting in specular reflection of "hot" carriers upon transmission at the metal-Si interface[33, 34]. The quantum yield is often called internal quantum efficiency (IQE)[14], so that IQE=Iph/Pabs · hν/q, where Pabs is the absorbed optical power, hν is the photon energy, q is the electron charge and Iph/Pabs is the PD responsivity (Rph) in units of A/W. One way to improve the IQE, is to confine light at the metal-Si interface by coupling to plasmonic modes[35, 36]. Following this concept, sev- eral NIR Si plasmonic Schottky PDs have been demon- strated, exploiting both localized plasmons[37 -- 40] and guided surface plasmons polaritons (SPP)[41 -- 46]. Yet, in these device the Rph reported to date does not exceed few tens mA/W with maximum IQE∼1%[43]. These values are significantly below SiGe PDs (Rph ∼0.4-1A/W and IQE∼ 60 − 90%)[15 -- 17]. Consequently, Rph of Schot- tky PDs should be further improved both by developing advanced device designs and/or using novel CMOS com- patible materials. Graphene is appealing for photonics and optoelectron- ics because it offers a wide range of advantages com- pared to other materials[47 -- 52]. A variety of proto- type optoelectronic devices exploiting graphene have al- ready been demonstrated, such as transparent electrodes in displays[53], photovoltaic modules[54, 55], optical modulators[56 -- 58], plasmonic devices[59 -- 63], and ultra- fast lasers[51]. Amongst these, a significant effort has been devoted to PDs, due to a number of distinct char- acteristics of graphene[47 -- 50, 52]. Single layer graphene (SLG) is gapless. This enables charge carrier generation by light absorption over a very wide energy spectrum. In addition, SLG has an ultrafast carrier dynamics[64], wavelength-independent absorption[65, 66], tuneable op- tical properties via electrostatic doping[67, 68], high mobility[69], and the ability to confine electromagnetic energy to unprecedented small volumes[49, 50]. The high carrier mobility enables ultrafast conversion of photons or plasmons to electrical currents or voltages[70, 71]. By integration with local gates, this process is in-situ tuneable[72, 73] and allows for sub-micron detection res- olution and pixilation[74]. SLG absorbs 2.3% of the incident light[65, 66], which is remarkably high for an atomically-thin material. This is an appealing property for flexible and transparent opto-electronic devices[47]. (MGM) configuration, The most common SLG PDs exploit the metal- in which a graphene-metal is contacted between source and drain SLG channel electrodes[70 -- 72, 75]. MGM devices are easy to fabricate[70, 71], they are able to operate over a broad- band wavelengths range[70, 71] and have demonstrated ultrahigh (∼230GHz)[76] operation speed. However, for visible and NIR wavelengths free-space illuminated MGM 2 PDs have Rph ∼few mA/W[70, 71]. This is primar- ily because of the finite optical absorbtion[65, 66] and limited photoactive area (Aphoto)[77]. In the MGM configuration, the built-in electric field that separates the photoexcited e-h pairs is localized in very narrow (∼100-200nm)[77] regions next to the edges of the SLG- metal contacts, whilst the rest of the SLG channel area does not contribute to Iph. One way to increase Rph is to apply a voltage between source-drain electrodes and increase the electric field penetration into the SLG channel[70, 71]. However, this will drive a current into SLG (dark current, Idark), which could be of the same order or even larger than Iph[70, 71]. Thus, this ap- proach can significantly reduce the signal-to-noise ra- tio (SNR) and increase power consumption. Another way consists in combining MGM devices with metal nanostructures[63, 78] and enabling light coupling to lo- calized and SPP modes, thus enhancing light-graphene interaction and light absorption. MGM-PDs can be also integrated with microcavities[79, 80], where at resonance the optical absorption in graphene is amplified by multi- ple light round trips[79, 80]. High Rph can be achieved using a hybrid configuration, in which a MGM structure is combined with semiconductor quantum dots (QD) as light absorbing media[81]. This gave Rph ∼ 107A/W [81] with a photoconducitve gain (i.e. the number of de- tected charge carriers per single incident photon, Gph) up to 107. Similar performances to graphene-QD hy- brid devices were also demonstrated in graphene tun- neling PDs[82], comprising two SLGs separated by a thin (< 10nm) dielectric layer. However, in both QDs- integrated or tunneling-based PDs the typical response time is limited to ms[81, 82], not suitable for high-speed (tens of GHz) optical communications. Another important performance metric of PDs the Normalized Photo-to-Dark-current Ratio, is NPDR=Rph/Idark[83]. The larger the NPDR, the better PD noise rejection and ability to perform when a interference (noise) is present. To achieve higher NPDR, Idark must be reduced and Rph must be in- creased. However, since SLG has no gap, a trade-off between improving Rph by using source-drain bias and minimizing Idark exists in all MGM-PDs[52]. In telecom applications, where power consumption and SNR are parameters of great importance for achieving energy efficient data transmission with reduced errors rate[1], MGM-PDs should be operated near zero bias, which in turn limits Rph. Even though MGM-PDs can perform in photovoltaic mode at zero bias with zero dark current[52, 84], the conductance of graphene can lead to enhanced thermal noise as a result of reduced channel resistance[84]. A promising route to increase Rph, while minimizing Idark, is to create a Schottky junction with rectifying characteristics (i.e a diode) at the SLG-Si interface[85 -- 89]. By operating a Schottky diode in reverse bias (photoconductive mode), Idark is suppressed compared to Iph, while the entire Schottky contact area contributes to photodetection[85 -- 89]. Several PDs have been reported to date, operating at telecom wavelengths and integrating on-chip graphene with Si photonics, based on MGM structures evanes- cently coupled to Si waveguides[89 -- 93]. In these cases, the guided mode approach enables longer interaction be- tween SLG and the optical waveguide modes than free- space illumination[52]. This raises the optical absorption in PD beyond 2.3% and, by increasing the interaction length, 100% light power can be absorbed and contribute to Iph[91]. Nevertheless, because of the evanescent cou- pling, the typical length needed to achieve nearly com- plete absorbtion in MGM-PDs is∼ 40 − 100µm[89 -- 93]. However, for on-chip optoelectronic integration, where scalability, footprint and cost play an important role, the development of miniaturized, simple to fabricate, Si- based PDs for telecoms, with Rph comparable to the SiGe devices currently employed in Si photonics, is needed[1]. Here, we report a compact (5µm length), waveguide integrated, plasmonic enhanced metal/ graphene/Si (M- SLG-Si) Schottky PD with Rph ∼0.37A/W at 1.55µm. The M-SLG-Si structure supports SPP guiding and ben- efits from optical confinement at the Schottky interface. Our data show that graphene integration in M-SLG-Si PDs increases Rph by one order of magnitude compared to the standard M-Si configuration without SLG. The SLG-integrated device has Rph ∼ 85mA/W at 1V re- verse bias, with Idark ∼ 20nA. Taking advantage of the Shottky diode operation in the reverse bias, Rph can be further increased up to∼ 0.37A/W at 3V. To the best of our knowledge, this is the highest Rph reported so far for waveguide-integrated Si-PDs operating at 1.55µm, and it is comparable to state-of-the-art SiGe devices[15 -- 17]. This is a simple, easy-processed approach for high responsivity Si PDs in the telecom spectral range, and paves the way to graphene-Si optoelectronic integration. Our PD is schematically shown in Fig.1a. The device consists of a Si-waveguide coupled to a SLG/Au contact that electrically forms a Schottky diode, while support- ing the fundamental SPP waveguide mode (Fig.1b). The SPP guiding provides optical confinement at the M-SLG- Si interface (Fig.1b), where the IPE process takes place. This can increase light absorption in SLG, adjacent to the Schottky interface and, as a result, enhance Rph. The fabrication process is discussed in Methods. We prepare on the same chip two types of devices: 1) M-SLG- Si Schottky PDs (our target device) and 2) a reference M-Si PD for comparison. Fig.2 shows a scanning electron microscope (SEM) micrograph of the resulting M-SLG-Si Schottky PD integrated with locally-oxidized[5] Si waveg- uides. The PD length is∼ 5µm and the Si waveguide width is∼ 310nm. Fig.3 plots a typical current-voltage (I-V) characteris- tic of our devices, measured using a probe station and a parameter analyzer (Keithley 4200). The device shows 3 FIG. 1: a) Schematic drawing of M-SLG-Si Schottky PD. SOI: silicon-on-insulator. BOX: buried oxide; b) Finite ele- ment (COMSOL Multiphysics)[94] simulated optical intensity profile of a SPP waveguide mode supported by a M-SLG-Si structure. FIG. 2: a) SEM micrograph of Schottky PD coupled to Si photonic waveguide. False colors: brown-Si, yellow-Au; b) Layout of waveguide integrated Schottky PD. electrical rectification (e.g diode behavior). The cur- rent in forward bias is limited by series resistance[14], while at reverse bias the leakage current I0 is limited by thermionic emission from Au/SLG to Si. In reverse bias, I0 grows with increasing temperature, consistent with what expected for thermionic-emission in a Schot- tky diode[14]. In the thermionic regime, the variations of I0 are reflected in the forward bias region, where the for- ward current also increases (Fig.3). Using the I-V char- acteristics in forward bias, and following the procedure described in Ref.95, 96, we extract the M-SLG-Si devices Schottky barrier height ΦB ∼ 0.34 and a diode ideality factor n∼ 1.8 (defined as the deviation of the measured FIG. 3: I-V characteristics of our M-SLG-Si Schottky PD for various temperatures. 4 FIG. 4: I-V characteristics of a) graphene-integrated and b) reference M-Si PDs for different optical powers coupled to the Schottky region. Measured photocurrent in c) graphene-integrated and d) reference M-Si PDs as a function of optical power coupled to the Schottky region. The slope of the lines in (c,d) corresponds to Rph. I-V curve from the ideal exponential behavior)[14]. For the reference M-Si devices we get ΦB ∼ 0.32 and n ∼ 1.7, similar to M-SLG-Si. This indicates that SLG does not significantly affect the electrical properties of the Schot- tky contact. For opto-electronic characterization, we use 1.55µm transverse-magnetic (TM) polarized light from a tun- able laser source (Agilent 81680A) butt-coupled to the waveguide using a polarization-maintaining (PM) ta- pered fiber. At the output facet of our waveguide the light is collected with a similar fiber and detected by an external InGaAs power meter (Agilent 81634a). Fig.3 shows that our device has a symmetric Y-branch to split the optical signal between the active arm with integrated Schottky PD and the reference waveguide. This is con- tinuously monitored to avoid optical power fluctuations. To test the opto-electronic response, we measure the I- V curves of graphene-integrated M-SLG-Si and reference M-Si devices at different Popt inside the SPP waveguide, as shown in Fig.4a,b. The PDs operate in photoconduc- tive mode[14], when a Popt increase results in larger re- verse current, since Iph acts as an external current source added to the Schottky diode I0 . Fig.4c,d plot Iph as a function of Popt as derived from the I-V curves in Fig.3. Iph grows linearly with Popt and the slope corresponds to Rph, i.e Iph = Rph ·Popt. We esti- mate Popt inside the Schottky PDs by taking into account a coupling loss of ∼ 18.5dB (98.5%) between the external tapered fiber and the Si waveguide (as measured by mon- itoring the output signal in the reference waveguide), a propagation loss (scattering + free carriers)∼ 1.5dB/mm (29% per mm) in the waveguide,∼ 3dB (50/50) power splitting, and∼ 1.5dB (29%) power loss in the Y-branch. Consequently, based on our I-V measurements and our Popt, we calculate and plot Rph as a function of re- verse voltage VR in Fig.5a. We get Rph ∼ 85mA/W with I0 ∼ 20nA at VR = 1V . The former corresponds to IQE ∼ 7%. For the reference M-Si PDs we get 5 FIG. 5: a) Rph of M-SLG-Si and reference M-Si PDs as a function of reverse bias for different optical powers coupled to the Schottky region; b) Rph of M-SLG-Si and reference M-Si PDs for 0 < VR < 3V . Colored solid lines show a fit of the bias dependent Rph based on combined thermionic-field emission and avalanche multiplication processes. Rph ∼ 9mA/W (at VR = 1V ), similar to state of the art Si Schottky PDs at 1.55µm[41 -- 46]. We conclude that the presence of SLG at the Schottky interface improves Rph by one order of magnitude compared to our reference M- Si PD. We attribute this to light absorption in the SLG adjacent to the Schottky barrier, where the IPE process takes place. The absorption is enhanced by SPP optical confinement at the M-SLG-Si interface (Fig.1b). The sig- nificant increase of Rph in SLG-integrated devices could be due to an higher transmission probability of "hot" carriers from SLG to Si when compared to the M-to-Si photoemission process. We then measure Rph for VR > 1V . Fig.5b shows that Rph grows monotonically up to VR ∼ 2V , then abruptly increases to∼ 0.37A/W at VR = 3V . To the best of our knowledge, this is the highest Rph reported so far for waveguide-integrated Si-PDs at 1.55µm, and it is com- parable to state-of-the-art Si-Ge devices currently em- ployed in Si photonics[15 -- 17]. We attribute this to the combined effect of two processes that can enhance Iph. First: thermionic-field emission (TFE), i.e tunneling of photoexcitepd carriers from the M-SLG contact to Si at energies EF < E < ΦB. The relative contribution of TFE with respect to IPE depends on Si doping, oper- ation temperature and the electric field applied to the Schottky junction[14, 97]. TFE tends to dominate at higher (> 1018cm−3) doping levels[14, 97] and its volt- age dependence is∝pVR + ΦB/E0 · exp(qVR/ǫ′), where E0 and ǫ′ are two analytically defined constants[14, 97]. In our device, with Si doping ∼ 7 · 1017cm−3 at room temperature, we calculate using Eqs.3,4 (see Methods) E0 and ǫ′ to be∼ 1.04V and∼ 2.1eV respectively[14]. Second: avalanche multiplication of photoexcitepd car- riers inside the Si depletion region, where the electrons (or holes) can lose their energy upon scattering with the Si lattice creating other charge carriers (i.e impact ion- ization). This process can be empirically modeled by M = 1/[1 − (VR/VBD)k][14], where M is the avalanche multiplication factor, VBD is the breakdown voltage at which M goes to infinity, and k is a power coefficient that empirically acquires values between 2 < k < 6[14]. As first order approximation, we assume independent con- tribution of each process. We show in Fig.5b that our data is well fitted by Rph(V ) ∝ T F E · M with VBD and k as free parameters. From the fit we get VBD ∼ 3.75V and k ∼ 3.2, corresponding to M ∼ 2 at VR = 3V . We note that, under avalanche conditions, the dark current also increases (∼ 3µA), and operation at elevated VR (>2.5V) reveals a trade-off between improving Rph and higher dark current. In summary, we demonstrated an on-chip, compact, waveguide-integrated metal-graphene-silicon plasmonic Schottky photodetector operating at 1.55µm. The pres- ence of graphene at the Schottky interface significantly improves the PD responsivity. The device has 85mA/W responsivity at 1V reverse bias, corresponding to 7% in- ternal quantum efficiency. This is one order of magni- tude higher compared to a reference metal-Si PD un- der the same conditions. We attribute this improve- ment to the combined effect of light confinement and graphene absorption at the metal-graphene-silicon Schot- tky interface, as well as enhanced carriers injection from graphene-to-silicon as compared to the metal-silicon in- terface. Avalanche multiplication for higher (>2V) re- verse biases allows us to reach a responsivity∼ 0.37A/W , corresponding to a photogain∼ 2. Our device paves the way towards graphene integrated silicon photonics. We acknowledge funding from EU Graphene Flagship 604391), ERC Grant Hetero2D, EPSRC Grants EP/N010345/1, (no. EP/K01711X/1, EP/M507799/1, EP/L016087/1 EP/K017144/1, 6 FIG. 6: Fabrication process of Si-SLG Schottky PD integrated with a photonic waveguide. a) Planar SOI substrate; b) PECVD deposition and patterning of SiN mask; c) Local oxidation; d) Etching of SiN and SiO2. Al ohmic contact to Si; e) SLG transfer; f) Formation of Schottky contact and consequent SLG etching. METHODS Si-SLG Schottky PD Fabrication Fig.6 outlines the fabrication process of our devices. We start with a commercial silicon on insulator (SOI, from SOITEC) substrate with a 340nm p-type (7 · 1017cm−3) Si layer on top of a 2µm buried oxide (BOX). First, a 100nm SiN mask is deposited by plasma en- hanced chemical vapor deposition (PECVD, Oxford Plas- maLab100) onto the SOI substrate at 300◦C (Fig.6b). Next, a Si photonic waveguide and the PD area are de- fined by electron beam lithography (EBL, Raith eLine 150) using positive e-beam resist (ZEP 520A). The EBL pattern is subsequently transferred to SiN by RIE (Ox- ford Plasmalab 100) with a CHF3/O2 gas mixture. Then, the SOI substrate is locally oxidized (wet, 1000◦C), to grow a SiO2 layer only in localized patterns defined by EBL where Si is exposed to O2, while at the same time a SiN mask prevents O2 diffusion into the Si in protected areas (Fig.6c). After oxidation, the sacrificial SiN mask layer is etched in hot phosphoric acid (H3PO4, 180◦), followed by SiO2 removal in a buffered oxide etch (BOE) solution. The ohmic contact to Si is realized by Al evap- oration, followed by metal lift-off and thermal alloying at 460◦C in a forming gas (H2/N2, 5%/95%) environ- ment. This fabrication process is based on the technique of local-oxidation of Si (LOCOS), in which a Si waveg- uide is defined by oxide spacers[5] rather than reactive ion etching (RIE). The LOCOS process enables the re- alization of low-loss(∼0.3dB/cm)[5] Si photonic waveg- uides coupled to a Schottky PD using the same fabrica- tion step. SLG is grown on a 35µm Cu foil, following the pro- cess described in Ref.98. The substrate is annealed in hydrogen atmosphere (H2, 20sccm) up to 1000◦C for 30 minutes. Then 5sccm CH4 is added to initiate growth[98, 99]. The substrate is subsequently cooled in vacuum (1mTorr) to room temperature and removed from the chamber. After growth, the quality and uni- formity of SLG are monitored by Raman spectroscopy using a Renishaw InVia equipped with a 100X objec- tive (numerical aperture NA=0.85). The Raman spec- trum of SLG on Cu at 514nm is shown in Fig.7(b) (green curve). This has a negligible D peak, thus in- dicating negligible defects[100 -- 104]. The 2D peak is a single sharp Lorentzian with full width at half maxi- mum, FWHM(2D)∼29cm−1, a signature of SLG[100]. Different (∼ 20) point measurements show similar spec- tra, indicating uniform quality. The position of the G peak, Pos(G), is∼1589cm−1, with FWHM(G)∼13cm−1. The 2D peak position, Pos(2D) is∼2698cm−1, while the 2D to G peak intensity and area ratios, I(2D)/I(G) and A(2D)/A(G), are 2.6 and 5.8, respectively, indicating a p-doping∼300meV[105, 106], which corresponds to a car- rier concentration∼ 5 · 1012cm−2. SLG is then transferred onto the SOI with Si waveg- uides. A∼500nm thick layer of polymethyl methacrylate (PMMA) is spin coated on the SLG/Cu sample, then placed in a solution of ammonium persulfate (APS) in DI water until Cu is completely etched[98, 107]. After Cu etching, the PMMA membrane with attached SLG is transferred to DI water for cleaning APS residuals. To obtain a Schottky interface between the Si waveg- uide and SLG without the native oxide layer we perform the transfer in diluted hydrofluoric acid (HF) and DI wa- ter (1:100) solution. After cleaning from APS residuals, a SLG/PMMA membrane is placed on a plastic beaker containing 5ml/500ml HF and DI water. Next, the tar- get SOI chips are firstly dipped in BOE for 5sec to etch the Si native oxide and then immediately used to lift a floating SLG/PMMA membrane from diluted HF. As a result, during drying the presence of HF at the SLG/Si 7 3nm/100nm Cr/Au metal strip intersecting the Si waveg- uide with SLG on top (Fig.6f) (or without SLG for refer- ence devices) and forming a Schottky interface for pho- todetection. Finally, the samples are placed in a reactive O2 plasma, to remove superfluous SLG. Thermionic Field Emission The TFE current is given by[14]: JT F E = A∗∗T k sπE00 × q(cid:20)VR + ΦB cosh2(E00/kT )(cid:21) (1) ×exp( ) × exp( −qΦB E0 qVR ǫ′ ) where A∗∗ is the effective Richardson constant, k is the Boltzmann constant, T is the temperature, q is the elec- tron charge. The contribution of TFE to charge injection across the M-Si interface can be evaluated by comparing the thermal energy kT to E00, defined as[14]: E00 = q 2 r N m∗ǫs (2) where is the reduced Planck constant, N is the Si doping, m∗ is the effective mass of the charge carriers in Si and ǫs is the dielectric permittivity of Si. When kT ≈ E00 the TFE process mainly contributes to charge carriers injection across the Schottky interface[14]. The parameters E0 and ǫ′ are analytically defined as[14]: E0 = E00 × coth( E00 kT ) ǫ′ = E00 (E00/kT ) − tanh(E00/kT ) (3) (4) In our case, for Si doping 7 · 1017cm−3 using Eq.2 we get E00 = 45meV , comparable to the thermal energy at room temperature of 26meV , reflecting a significant TFE contribution to carriers injection at the Schottky interface. Hence, we calculate E0 and ǫ′ to be∼ 1.04V and ∼ 2.1eV , respectively. ∗ Electronic address: [email protected] [1] Reed, G. 2008, Wiley, UK. [2] Silicon Photonics. Focus Issue, Nat. Nanotechnol. 2010, 4, 491-578. [3] Doylend, J. K.; Knights, A. P. Laser & Photon. Rev. 2012, 6, 504-525. [4] Cardenas, J.; Poitras, C. B.; Robinson, J. T.; Preston, K. ; Chen, L.; Lipson, M. Opt. Express 2009, 17, 4752- 4757. FIG. 7: a) Raman spectra of (red curve) Si substrate and (black curve) SLG transferred on Si. b) Raman spectra of (green curve) SLG on Cu , and (blue curve) after normal- ized, point-to-point subtraction of the Si substrate spectrum (shown in (a) red curve) from the spectrum of SLG transferred on Si (shown in (a) black curve). interface prevents Si oxidation and allows formation of "oxide free" SLG/Si Schottky contacts. After drying, PMMA is removed in acetone leaving SLG to entirely cover the SOI. We also transfer SLG from the same Cu foil using the same transfer procedure onto Si. This is used to check the SLG quality after transfer by Raman spectroscopy. The Raman spectrum of SLG transferred on Si is shown in Fig.7(a)(black line). This is measured at 514.5nm and with laser power below 300µW to avoid possible heating effects or damage. The D peak re- gion overlaps the bands at∼1200-1500cm−1, attributed to third order Raman scattering from TO phonons in the Si substrate[108]. The peaks at∼1550 and∼2330cm−1 in the Raman spectrum of Si substrate (red line) arise from molecular vibrations of ambient oxygen (O2)[109] and nitrogen (N2)[110]. The Raman spectra of the transferred SLG film (black line) and reference Si sub- strate (red line) are acquired using identical exposure time and laser power. After normalizing the intensity of the third order Si peak at∼1450cm−1 in the Si refer- ence spectrum (red line) to the same peak in the spec- trum of the transferred SLG film (black line), a point-to- point subtraction is implemented [Fig. 7 (b)(blue line)]. The resulting spectrum shows D to G intensity ratio, I(D)/I(G)∼0.04, indicating negligible defects[100 -- 104]. The 2D peak retains its single-Lorentzian line-shape with FWHM(2D)∼33cm−1, validating that SLG has been suc- cessfully transferred. Pos(G)∼1584cm−1, FWHM(G) ∼17cm−1 and Pos(2D)∼2687cm−1, while I(2D)/I(G) and A(2D)/A(G) are 3.2 and 5.9, respectively, suggesting a p-doping∼4·1012 cm−2 (∼200meV)[105, 106]. After SLG transfer, we use an additional EBL step fol- lowed by O2 plasma etching to selectively remove SLG from the substrate area containing 5 waveguides and ded- icated to the reference M-Si devices. Then, a Schot- tky contact is prepared by evaporation and liftoff of an 8 [5] Desiatov, B.; Goykhman, I.; Levy, U. Opt. Express [35] Sipe, J. E.; Becher, J. J. Opt. Soc. Am. 1981, 71, 1286- 2010, 18, 18 592-18 597. 1288. [6] Velha, P.; Picard, E.; Charvolin, T.; Hadji, E. ; Rodier, J. C.; Lalanne, P.; Peyrade, D. Opt. Express 2007, 15, 16090-16096. [7] Kuramochi, E.; Taniyama, H.; Tanabe, T.; Kawasaki, K. ; Roh, Y.-G.; Notomi, M. Opt. Express 2010, 18, 15 859-15 869. [36] Endriz, J. G. Appl. Phys. Lett. 1974, 25, 261-262. [37] Fukuda, M.; Aihara, T.; Yamaguchi, K.; Ling, Y. Y.; Miyaji, K.; Tohyama, M. Appl. Phys. Lett. 2010, 96, 153 107+. [38] Knight, M. W.; Sobhani, H.; Nordlander, P.; Halas, N. Science 2011, 332, 702-704. [8] Naiman, A.; Desiatov, B.; Stern, L.; Mazurski, N.; [39] Zhu, S.; Chu, H. S.; Lo, G. Q.; Bai, P.; Kwong, D. L. Shappir, J.; Levy, U. Opt. Lett. 2015, 9, 1892-1895. Appl. Phys. Lett. 2012, 100, 061 109+. [9] Xu, Q.; Schmidt, B.; Pradhan, S.; Lipson, M. Nature [40] Lin, K.-T.; Chen, H.-L.; Lai, Y.-S.; Yu, C.-C. Nat. Com- 2005, 435, 325-327. munications 2014, 5. [10] Liu, A.; Liao, L.; Rubin, D.; Nguyen, H. ; Ciftcioglu, B.; Chetrit, Y.; Izhaky, N.; Paniccia, M. Opt. Express 2007, 15, 660-668. [41] Akbari, A.; Tait, R. N.; Berini, P. Opt. Express 2010, 18, 8505-8514. [42] Goykhman, I.; Desiatov, B.; Khurgin, J. B.; Shappir, [11] Reed, G. T.; Mashanovich, G.; Gardes, F. Y.; Thomson, J.; Levy, U. Nano Lett. 2011, 11, 2219-2224. D. J. Nat. Photonics 2010, 4, 518-526. [12] Rong, H.; Liu, A.; Jones, R.; Cohen, O. ; Hak, D.; Nicolaescu, R.; Fang, A.; Paniccia, M. Nature 2005, 433, 292-294. [13] Liang, D.; Bowers, J. E. Nat. Photonics 2010, 4, 511- 517. [14] Sze, S. M., Kwok, K. N. 2006, Wiley, New York. [15] Kang, Y. et al. Nat. Photonics 2009, 3, 59-63. [16] Michel, J.; Liu, J.; Kimerling, L. C. Nat. Photonics 2010, 4, 527-534. [43] Goykhman, I.; Desiatov, B.; Khurgin, J. B.; Shappir, J.; Levy, U. Opt. Express 2012, 20, 28594-28602. [44] Aihara, T.; Nakagawa, K.; Fukuhara, M.; Yu, Y. L.; Yamaguchi, K.; Fukuda, M. Appl. Phys. Lett. 2011, 99, 043 111+. [45] Sobhani, A.; Knight, M. W.; Wang, Y.; Zheng, B.; King, N. S.; Brown, L. V.; Fang, Z.; Nordlander, P.; Halas, N. J. Nat. Communications 2013, 4, 1643+. [46] Desiatov, B.; Goykhman, I.; Mazurski, N; Shappir, J.; Khurgin, J. B.; Levy, U. Optica 2015, 4, 335-338. [17] Assefa, S.; Xia, F.; Vlasov, Y. A. Nature 2010, 464, [47] Bonaccorso, F.; Sun, Z.; Hasan, T.; Ferrari, A. C. Nat. 80-84. [18] Hawkins, A. R.; Wu, W.; Abraham, P.; Streubel, K.; Bowers, J. E. Appl. Phys. Lett. 1997, 70, 303305. [19] Kang, Y.; Mages, P.; Clawson, A. R.; Yu, P. K. L.; Bitter, M.; Pan, Z.; Pauchard, A.; Hummel, S.; Lo, Y. H. IEEE Photon. Tech. Lett. 2002, 14, 15931595. [20] Liang, D.; Roelkens, G.; Baets, R.; Bowers, J. Materials 2010, 3, 1782-1802. [21] Wang, J.; Lee, S. Sensors 2011, 11, 696-718. [22] Liang, T. K.; Tsang, H. K.; Day, I. E.; Drake, J.; Knights, A. P.; Asghari, M. Appl. Phys. Lett. 2002, 81, 1323-1325. [23] Tanabe, T.; Sumikura, H.; Taniyama, H.; Shinya, A.; Notomi, M. Appl. Phys. Lett. 2010, 96(10), 101 103+. [24] Bradley, J. D. B.; Jessop, P. E.; Knights, A. P. Appl. Phys. Lett. 2005, 86, 241 103+. [25] Ackert, J. J.; Fiorentino, M.; Logan, D. F.; Beausoleil, R. G.; Jessop, P. E.; Knights, A. P. J. Nanophotonics 2011, 5, 059 507+. [26] Desiatov, B.; Goykhman, I.; Shappir, J.; Levy, U. Appl. Phys. Lett. 2014, 104, 091 105+. [27] Preston, K.; Lee, Y. H.; Zhang, M.; Lipson, M. Opt. Lett. 2011, 36, 52-54. Photonics 2010, 4, 611-622. [48] Ferrari, A. C. et al. Nanoscale 2015, 7, 4598. [49] Grigorenko, A. N.; Polini, M.; Novoselov, K. S. Nat. Photonics 2012, 6, 749-758. [50] Koppens, F. H. L.; Chang, D. E.; Garcia de Abajo, F. J. Nano Lett. 2011, 11, 3370-3377. [51] Sun, Z.; Hasan, T.; Torrisi, F.; Popa, D.; Privitera, G.; Wang, F.; Bonaccorso, F.; Basko, D. M.; Ferrari, A. C. ACS Nano 2010, 4, 803-810. [52] Koppens, F. H. L.; Mueller, T.; Avouris, P.; Ferrari, A. C.; Vitiello, M. S.; Polini, M. Nat. Nanotechnol. 2014, 9, 780-793. [53] Kim, K. S.; Zhao, Y.; Jang, H.; Lee, S. Y.; Kim, J. M.; Kim, K. S.; Ahn, J.-H.; Kim, P.; Choi, J.-Y.; Hong, B. H. Nature 2009, 457, 706-710. [54] Baugher, B. W. H.; Churchill, H. O. H.; Yang, Y.; Jarillo-Herrero, P.Nat. Nanotechnol. 2014, 9, 262-267. [55] Pospischil, A.; Furchi, M. M.; Mueller, T. Nat. Nan- otechnol. 2014, 9, 257-261. [56] Liu, M.; Yin, X.; , Ulin-Avila, E.; Geng, B.; Zentgraf, T.; Ju, L.; Wang, F.; Zhang, X. Nature 2011, 474, 64- 67. [57] Phare, C. T.; Daniel Lee, Y.-H.; Cardenas, J.; Lipson, [28] Chen, H.; Luo, X.; Poon, A. W. Appl. Phys. Lett. 2009, M. Nat. Photonics 2015, 9, 511-514. 95, 171 111+. [29] Casalino, M.; Coppola, G.; Iodice, M.; Rendina, I.; Sir- leto, L. Opt. Lett. 2012, 20, 12 599-12 609. [30] Casalino, M.; Sirleto, L.; Iodice, M.; Saffioti, N.; Gioffr, M.; Rendina, I.; Coppola, G. Appl. Phys. Lett. 2010, 96, 241 112+. [31] Zhu, S.; Lo, G. Q.; Kwong, D. L. IEEE Photon. Technol. Lett. 2008, 20, 1396-1398. [32] Peters, D. W. Proc. IEEE 1967, 55, 704. [33] Goykhman,I.; Desiatov, B.; Shappir, J.; Khurgin, J.B.; Levy, U. 2010, http://arxiv.org/abs/1401.2624. [34] Brongersma, M. L.; Halas, N. J.; Nordlander, P. Nat. Nano 2015, 10, 25-34. [58] Ding, Y.; Zhu, X.; Xiao, S.; Hu, H.; Frandsen, L. H.; Mortensen, N. A.; Yvind, K. Nano Lett. 2015, 15, 4393- 4400. [59] Chen, J. et al. Nature 2012, 487, 77-81. [60] Fei, Z. et al. Nature 2012, 487, 82-85. [61] Ju, L. et al. Nat. Nanotechnol. 2011, 6, 630-634. [62] Yan, H.; , Li, X.; Chandra, B.; Tulevski, G.; Wu, Y.; Freitag, M.; Zhu, W.; Avouris, P.; Xia, F. Nat. Nan- otechnol. 2012, 7, 330-334. [63] Echtermeyer, T. J.; Britnell, L.; Jasnos, P. K.; Lom- bardo, A.; Gorbachev, R. V.; Grigorenko, A. N.; Geim, A. K.; Ferrari, A. C.; Novoselov, K. S. Nat. Communi- cations 2011, 2, 458. [64] Brida, D.; Tomadin, A.; Manzoni, C.; Kim, Y. J.; Lom- bardo, A.; Milana, S.; Nair, R. R.; Novoselov, K. S.; Ferrari, A. C.; Cerullo, G.; Polini, M. Nat. Communi- cations 2013, 4, 1987. [65] Kuzmenko, A. B.; van Heumen, E.; Carbone, F.; van der Marel, D. Phys. Rev. Lett. 2008, 100, 117401. [66] Nair, R. R.; Blake, P.; Grigorenko, A. N.; Novoselov, K. S.; Booth, T. J.; Stauber, T.; Peres, N. M. R.; Geim, A. K. Science 2008, 320, 1308. [67] Li, Z. Q.; Henriksen, E. A.; Jiang, Z.; Hao, Z.; Mar- tin, M. C.; Kim, P.; Stormer, H. L.; Basov, D. N. Nat. Physics 2008, 4, 532-535. [68] Wang, F.; Zhang, Y.; Tian, C.; Girit, C.; Zettl, A.; Crommie, M.; Shen, Y. R. Science 2008, 320, 206-209. [69] Mayorov, A. S. et al. Science 2011, 333, 860-863. [70] Xia, F.; Mueller, T.; Lin, Y. M.; Valdes-Garcia, A.; Avouris, P. Nat. Nanotechnol. 2009, 4, 839-843. [71] Mueller, T.; Xia, F.; Avouris, P. Nat. Photonics 2010, 4, 97-301. [72] Lee, E. J. H.; Balasubramanian, K.; Weitz, R. T.; Burghard, M.; Kern, K. Nat. Nanotechnol. 2008, 3, 486-490. [73] Xia, F.; Mueller, T.; Golizadeh-Mojarad, R.; Freitag, M.; Lin, Y.-m.; Tsang, J.; Perebeinos, V.; Avouris, P. Nano Lett. 2009, 9, 1039-1044. [74] Lemme, M. C.; Koppens, F. H. L.; Falk, A. L.; Rudner, M. S.; Park, H.; Levitov, L. S.; Marcus, C. M. Nano Lett. 2011, 11, 4134-4137. [75] Park, J.; Ahn, Y. H.; Ruiz-Vargas, C. Nano Lett. 2009, 9, 1742-1746. [76] Urich, A.; Unterrainer, K.; Mueller, T. Nano Lett. 2011, 11, 2804-2808. [77] Mueller, T.; Xia, F.; Freitag, M.; Tsang, J.; Avouris, Ph. Phys. Rev. B 2009, 79, 245430. [78] Echtermeyer, T. J.; Milana, S.; Sassi, U.; Eiden, A.; Wu, M.; Lidorikis, E.; Ferrari, A. C. 2015, http://arxiv.org/abs/1505.06721. [79] Furchi, M.; Urich, A.; Pospischil, A.; Lilley, G.; Un- terrainer, K.; Detz, H.; Klang, P.; Andrews, A. M.; Schrenk, W.; Strasser, G.; Mueller, T. Nano Lett. 2012, 12, 2773-2777. [80] Engel, M.; Steiner, M.; Lombardo, A.; Ferrari, A. C.; Lahneysen, H.; Avouris, P.; Krupke, R. Nat. Comms 2012, 3, 906. [81] Konstantatos, G.; Badioli, M.; Gaudreau, L.; Osmond, J.; Bernechea, M.; de Arquer,; Gatti, F.; Koppens, F. H. L. Nat. Nanotechnol. 2012, 7, 363-368. [82] Liu, C.-H.; Chang, Y.-C.; Norris, T. B.; Zhong, Z. Nat. Nanotechnol. 2014, 9, 273-278. [83] Chui, C. O.; Okyay, A. K.; Saraswat, K. C. Photonics Technology Letters, IEEE 2003, 15, 1585-1587. [84] Gulbahar, B.; Akan, O. B. Nanotechnology, IEEE Transactions on 2012, 11, 601-610. [85] Chen, C.-C.; Aykol, M.; Norris, C.-C.; Zhong, Z.; Cronin, S. B. Graphene-silicon Schottky diodes, Nano Lett. 2011, 11(5), 1863-1867. [86] An, Y.; Behnam, A.; Pop, E.; Ural, A. Appl. Phys. Lett. 2013, 102, 013 110+. [87] An, X.; Liu, F.; Jung, Y. J.; Kar, S. Nano Lett. 2013, 13, 909-916. 9 [88] Amirmazlaghani, M.; Raissi, F.; Habibpour, O.; Vuku- sic, J.; Stake, J. IEEE J. Quantum Electron. 2013, 49, 589-594. [89] Wang, X.; Cheng, Z.; Xu, K.; Tsang, H. K.; Xu, J.-B. Nat. Photonics 2013, 7, 888-891. [90] Gan, X.; Shiue, R.-J.; Gao, Y.; Meric, I.; Heinz, T. F.; Shepard, K.; Hone, J.; Assefa, S.; Englund, D. Nat. Photonics 2013, 7, 883-887. [91] Pospischil, A.; Humer, M.; Furchi, M. M.; Bachmann, D.; Guider, R.; Fromherz, T.; Mueller, T. C Nat. Pho- tonics 2013, 7, 892-896. [92] Youngblood, N.; Anugrah, Y.; Ma, R.; Koester, S. J.; Li, M. Nano Lett. 2014, 14, 2741-2746. [93] Shiue, R.-J.; Gao, Y.; Wang, Y.; Peng, C.; Robertson, A. D.; Efetov, D. K.; Assefa, S.; Koppens, F. H. L.; Hone, J.; Englund, D. Nano Lett. 2015, 15, 7288-7293. [94] COMSOL Multipphysics http://www.comsol.com. [95] Cheung, S. K.; Cheung, N. W. Appl. Phys. Lett. 1986, 49, 85-87. [96] Sato, K.; Yasumura, Y. J. Appl. Phys. Lett. 1985, 58, 3655-3657. [97] Padovani F. A, Stratton R. Solid State Electron. 1966, 9, 69. [98] Bae, S., Kim, H., Lee, Y., Xu, X., Park, J. S., Zheng, Y., Balakrishnan, J., Lei, T., Kim, H. R., Song, Y. I., et al. Nat. Nanotechnol. 2010, 5, 574-8. [99] Li, X. S., Cai, W. W., An, J. H., Kim, S., Nah, J., Yang, D. X., Piner, R., Velamakanni, A., Jung, I., Tutuc, E., et al. Science 2009, 324, 1312-1314. [100] Ferrari, A. C., Meyer, J. C., Scardaci, V., Casiraghi, C., Lazzeri, M., Mauri, F., Piscanec, S., Jiang, D., Novoselov, K. S., Roth, S., et al. Phys. Rev. Lett. 2006, 97, 187401. [101] Cancado, L. G., Jorio, A., Ferreira, E. H., Stavale, F., Achete, C. A., Capaz, R. B., Moutinho, M. V., Lom- bardo, A., Kulmala, T. S., Ferrari, A. C. Nano Lett. 2011, 11, 3190-6. [102] Ferrari, A. C., Basko, D. M. Nat. Nanotechnol. 2013, 8, 235-46. [103] Ferrari, A. C., Robertson, J. Phys. Rev. B 2000, 61, 14095-14107. [104] Bruna, M., Ott, A. K., Ijas, M., Yoon, D., Sassi, U., Ferrari, A. C. Acs Nano 2014, 8, 7432-7441. [105] Das, A., Pisana, S., Chakraborty, B., Piscanec, S., Saha, S. K., Waghmare, U. V., Novoselov, K. S., Krishna- murthy, H. R., Geim, A. K., Ferrari, A. C., Sood, A. K. Nat. Nanotechnol. 2008, 3, 210-5. [106] Basko, D. M., Piscanec, S., Ferrari, A. C. Phys. Rev. B 2009, 80, 165413. [107] Bonaccorso, F., Lombardo, A., Hasan, T., Sun, Z. P., Colombo, L., Ferrari, A. C. Mater. Today 2012, 15, 564-589. [108] Temple, P. A.; Hathaway, C. E. Phys. Rev. B 1973, 7, 3685. [109] A. Weber, E. A. McGinnis, J. Mol. Spectrosc. 1960, 4, 195. [110] Lofthus, A.; Krupenie, P. H. J. Phys. Chem. Ref. Data 1977, 6, 113.
1606.08614	2	1606	2016-06-30T10:31:47	Zeeman splitting of conduction band in HgTe quantum wells near the Dirac point	[ "cond-mat.mes-hall" ]	The Zeeman splitting of the conduction band in the HgTe quantum wells both with normal and inverted spectrum has been studied experimentally in a wide electron density range. The simultaneous analysis of the SdH oscillations in low magnetic fields at different tilt angles and of the shape of the oscillations in moderate magnetic fields gives a possibility to find the ratio of the Zeeman splitting to the orbital one and anisotropy of g-factor. It is shown that the ratios of the Zeeman splitting to the orbital one are close to each other for both types of structures, with a normal and inverted spectrum and they are close enough to the values calculated within kP method. In contrast, the values of g-factor anisotropy in the structures with normal and inverted spectra is strongly different and for both cases differs significantly from the calculated ones. We believe that such disagreement with calculations is a result of the interface inversion asymmetry in the HgTe quantum well, which is not taken into account in the kP calculations.	cond-mat.mes-hall	cond-mat	a Zeeman splitting of conduction band in HgTe quantum wells near the Dirac point G. M. Minkov,1 O. E. Rut,1 A. A. Sherstobitov,1, 2 S. A. Dvoretski,3 and N. N. Mikhailov3, 4 2M. N. Miheev Institute of Metal Physics of Ural Branch of Russian Academy of Sciences, 620137 Ekaterinburg, Russia 1Institute of Natural Sciences, Ural Federal University, 620002 Ekaterinburg, Russia 3Institute of Semiconductor Physics RAS, 630090 Novosibirsk, Russia 4Novosibirsk State University, Novosibirsk 630090, Russia (Dated: September 6, 2018) The Zeeman splitting of the conduction band in the HgTe quantum wells both with normal and inverted spectrum has been studied experimentally in a wide electron density range. The simultaneous analysis of the SdH oscillations in low magnetic fields at different tilt angles and of the shape of the oscillations in moderate magnetic fields gives a possibility to find the ratio of the Zeeman splitting to the orbital one and anisotropy of g-factor. It is shown that the ratios of the Zeeman splitting to the orbital one are close to each other for both types of structures, with a normal and inverted spectrum and they are close enough to the values calculated within kP method. In contrast, the values of g-factor anisotropy in the structures with normal and inverted spectra is strongly different and for both cases differs significantly from the calculated ones. We believe that such disagreement with calculations is a result of the interface inversion asymmetry in the HgTe quantum well, which is not taken into account in the kP calculations. I. INTRODUCTION A HgTe/CdTe quantum well is a system where the Dirac fermions appear only in a single valley, at the Γ point of the Brillouin zone, unlike graphene where there are two valleys of the Dirac fermions with a strong inter- valley scattering. The energies of spatially quantized sub- bands at the quasimomentum k = 0 and energy spectrum E(k) for different widths of the quantum well (d) were calculated within kP method in numerous papers [1 -- 5]. As seen from Fig. 1, various types of energy spectrum are realized upon increasing the HgTe quantum-well width; namely, "normal", when d is less than a critical width dc ≃ 6.3 nm, Dirac-like at small quasimomenta for d = dc, inverted when d > dc, and finally, semimetallic when d > 14 − 16 nm. To interpret experimental data, these calculations of the energy spectrum are used practically always. They well describe the width dependence of the energies of both electron and hole subbands at k = 0 and the energy dependence of the electron effective mass (me). However, quite a lot of differences between the ex- perimental data and the results of these calculations on the energy spectrum of the carriers have been accu- mulated to date. First of all, they refer to the spec- trum of the valence band. The hole effective mass (mh) at d ≈ 20 nm within the wide hole density range p = (1 − 4) × 1011 cm−2 is substantially less than the calculated one: mh ≃ (0.15 − 0.3)m0 [6, 7] instead of (0.5−0.6)m0 [5]. The top of the valence band in the nom- inally symmetric structures with d ≈ dc (d = 5.5 − 7 nm) was found to be very strongly split by spin-orbit (SO) in- teraction [8]. Therewith, the SO splitting of the conduc- tion band in the same structures does not reveal itself [9]. It is surprising that such SO splitting is observed in struc- tures both with inverted and normal spectrum despite the fact that at d < dc and d > dc the conduction band is formed from different terms (see Fig. 1). At d < dc, the conduction band is formed from electron states and states of light hole, while at d > dc, it is formed from heavy-hole states. Such SO splitting was not described by Byckov-Rashba effect taken into account within kP method. It was assumed [9] that such a surprising be- havior of SO splitting is a result of the interface inversion asymmetry (IIA) in the HgTe quantum well, which was not taken into account in kP calculations in [1 -- 5]. The question arises: how other spin-dependent effects, for example, the Zeeman splitting, depend on the spec- trum type -- normal or inverted. We found only two pa- pers where the Zeeman splitting of electron spectrum was measured in the HgTe quantum wells with the width d which is more or less close to dc [10, 11]. In Ref. [10], the Zeeman splitting was determined in a structure with nor- mal spectrum, d = 6.1 nm, at very large electron density n = 1.4 × 1012 cm−2. In Ref. [11], it was determined in a structure with inverted spectrum with d = 9 nm that is noticeably larger than dc, at n = 6.59 × 1011 cm−2. So, up to now a systematic study of the Zeeman split- ting and a comparison of it with theoretical calculations are absent. In this paper, we present the results of the in- vestigation of the Shubnikov-de Haas (SdH) oscillations in tilt magnetic fields in the HgTe quantum wells with normal and inverted spectra. To find the ratio of the Zeeman splitting to the orbital one, we have used a mod- ified coincidence method which consists in measuring the angle dependence of amplitudes of the SdH oscillations in low magnetic fields. The simultaneous analysis of this dependence and the shape of oscillations of ρxx made it possible to determine both the ratio of the Zeeman split- ting to the cyclotron one and the anisotropy of g-factor (gk/g⊥) over a wide electron-density range, where gk and g⊥ are the in-plane and transverse g-factor, respectively. (a) 50 0 -50 -100 50 0 -50 (b) d=5 nm 50 0 -50 (c) d=7 nm 0 0.5 k (nm-1) 0 0.5 k (nm-1) Figure 1. (Color online) (a) -- The quantum well width de- pendence of the subband energies at k = 0. The dependences E(k) of the conduction and valence bands for d < dc (b) and d > dc (c). The marked area in (a) shows the range of quantum well widths under study. II. EXPERIMENT Our samples with the HgTe quantum wells were real- ized on the basis of HgTe/Hg1−xCdxTe (x = 0.55 − 0.65) heterostructures grown by the molecular beam epitaxy on a GaAs substrate with the (013) surface orientation [12]. The samples were mesa etched into standard Hall bars of 0.5 mm width and the distance between the po- tential probes was 0.5 mm. To change and control the carrier density in the quantum well, the field-effect tran- sistors were fabricated with parylene as an insulator and aluminium as a gate electrode. For each heterostructure, several samples were fabricated and studied. The Zee- man splitting of the conduction band has been obtained from measurements of the SdH effect in a tilted mag- netic field, i.e. we used the so-called coincidence method. This method is based on the fact that the spin splitting, gµBB, depends on the total magnetic field (B) whereas the orbital splitting of the Landau levels (LLs) in 2D sys- tems, ωc, is proportional to the component of the mag- netic field which is perpendicular to the 2D plane (B⊥): ωc = (e/me)B⊥ = (e/me)bB, where b ≡ B⊥/B. Thus, the ratio of the Zeeman splitting to the orbital one, gµBB/ωc, will change upon varying the tilt an- gle as X(b) = gµBme/(eB b). It is clear that there are particular angles bc when X(bc) = 1/2, 1, 3/2... At inte- 2 Table I. The parameters of heterostructures under study number structure d (nm) type p, n(Vg = 0) (cm−2) 1 2 3 4 5 1520 1122 1023 H725 HT71 4.6 5.6 6.5 8.3 9.5 n p p p p 1.5 × 1011 1.3 × 1011 1.0 × 1010 5.0 × 1010 6.0 × 1010 ger X(bc) values, the energies of the LLs with different numbers and opposite spin coincide with each other and the distances between the pairs of such degenerate LLs are equal to ωc. When X(bc) is half-integer, the energy distances between nearest LLs are twice as low, 0.5ωc. As a result, the oscillation periods will differ twice for these cases. Knowing values of bc, one can find the ratio X(1) ≡ X. For example, when X(bc) = 1/2, X = bc/2. In this paper, we used the modified coincidence method [13 -- 15]. We have measured the oscillations at low mag- netic field when: (i) B⊥ is significantly less than the field of the onset of the quantum Hall effect (QHE); (ii) ampli- tude of the oscillation is small so that oscillations of the Fermi energy are negligible; (iii) the SdH oscillations are spin-unsplit. In this case, the study of the angular de- pendence of the oscillation amplitude A(b) at a given B⊥ value within the entire range of angles (rather than the determination of critical angles) allows one not only to determine the ratio of the Zeeman splitting to the orbital one, but estimate the g-factor anisotropy. To find analytic expression for the tilt-angle depen- dence of the oscillation amplitude A(b), it is convenient to represent the oscillations as the sum of the contribu- tions from the two series of the Landau spin sublevels. At low magnetic field, when the SdH oscillations are unsplit, the main contribution to the oscillations of ρxx comes from the first harmonic, [16] and the well known Lifshits- Kosevitch (LK) formula for the SdH oscillations reduces to the following expression ∆ρxx(B, b) = ρxx(B, b) − ρxx(0) = a↑ cos(cid:20)2π(cid:18) EF + a↓ cos(cid:20)2π(cid:18) EF ωc ωc + + 1 2 1 2 + − X(b) 2 (cid:19)(cid:21) 2 (cid:19)(cid:21) . X(b) (1) Here, the factors a↑ and a↓ depend on the Dingle factor, temperature, and magnetic field. When a↑ = a↓ = a, Eq. (1) is ∆ρxx(B, b) = 2a cos [πX(b)] cos(cid:20)2π(cid:18) EF ωc + 1 2(cid:19)(cid:21). (2) Thus, over this magnetic field range, the values of B⊥ corresponding to the extremes of ρxx should not depend on the tilt angle, while the amplitudes of oscillations A(b) = 2a cos [πX(b)] should periodically change with ( a ) n =10 n =12 3000 ) m h O ( y x 2000 1000 ( b ) b=1.00 0.80 0 0.65 0.53 0.43 0.34 0.31 0.28 0.23 400 ) m h O ( x x 200 ) s t i n u . b r a ( B^ d / x x dr 0 0 1 B (T) 2 -200 1 B^ (T) 2 Figure 2. (Color online) (a) The magnetic field dependences of ρxx and ρxy at normal magnetic field in structure 1520 at n = 5.45 × 1011 cm−2. T = 4.2 K. (b) The dependences of dρxx/dB⊥ on B⊥ for different b. b and the angular dependence of the relative amplitude is A(b) A(1) = cos[πX(b)] cos[πX(1)] . (3) III. RESULTS AND DISCUSSION Let us begin our analysis with the results obtained in the structures with a normal spectrum (d < dc) (see Ta- ble I). As an example, consider the results for the struc- ture 1520. Before discussing the oscillations in the tilt magnetic field, it is necessary to estimate the magnetic field range where Eq. (1) is valid for this structure. To this end, let us inspect the magnetic field dependences of ρxx and ρxy in the normal field, which are presented in Fig. 2(a) for the electron density n = 5.45 × 1011 cm−2. It is seen that at B < 1.5 T, the amplitude of oscilla- tions of ρxx is 10 percent less, and the steps in ρxy(B) (with even filling factors ν) appear only at B > 1.5 T; therefore one can neglect the oscillations of the Fermi en- ergy within this range of B. The electron density found from the period of oscillations under assumption that the Landau levels are two-fold degenerate, coincides with the Hall density [17]. Thus, at B < 1.5 T, the conditions of applicability of Eq. (1) are met. Now let us inspect the SdH oscillations in the tilt mag- netic field. To remove the monotonic part we plotted in Fig. 2(b) the dρxx/dB⊥ versus B⊥ dependences, mea- sured at different tilt angles. To make it easier to trace the position of oscillations at different angles, we mark the position of one of the maxima dρxx/dB at the normal field B = 1.02 T by a dashed line. It is clearly seen that the positions of extremes of dρxx/dB⊥ do not change with tilt angle but the maxima are transformed to the minima at b ≃ 0.65, and upon further rotation they are 3 ) 1 ( A / ) b ( A 2 0 -2 -4 -6 X g^ /gII 0.39 1 0.4 1 0.405 0.9 0.415 0.8 0.39 1.05 0.8 1.0 B^ (T) 1 1.2 1.4 0.6 b 0.2 0.4 Figure 3. (Color online) The relative oscillations ampli- tude for the structure H1520 plotted against b at n = 5.45 × 1011 cm−2. Points are experimental data found at B⊥ = 1.0, 1.2 and 1.4 T. The inversion of the amplitude sign corresponds to the change of the oscillation phase by π. Solid and dash lines are the dependences Eq.(3) with X = 0.39 and 0.4, respectively. Other lines are the calculated dependences with taking into account g-factor anisotropy. transformed to the maxima again at b ≃ 0.23. Note, a noticeable difference was not observed when the parallel component of the field was along or perpendicular to the current. For the quantitative analysis, the amplitude of oscil- lations A(B⊥, b) at a given b was found by fitting of the oscillatory part of A(B⊥, b) versus B⊥ curves to the oscillating function corresponding to the electron den- sity n = 1/eRH measured in the normal field, with the amplitude A(B) = k0 exp(k1B). The relative am- plitudes of the oscillations found in this way A(b)/A(1) as a function of b are plotted for some values of B⊥ in Fig. 3. The inversion of the amplitude sign corre- sponds to the change in the oscillations phase by π. Note, that A(b)/A(1) does not depend practically on B⊥ when the magnetic field is sufficiently small so that Eq. (1) is valid. In the same figure we have shown the dependences which are given by Eq. (3) for some values of the ratio X(b)=gµBB/(eB⊥/me). One can see that Eq. (3) well describes the experimental data with X = 0.39 ± 0.01. Some deviation at b < 0.3 can result from the g-factor anisotropy. Indeed, taking this possibility into account in simplest form g(b) = qg 2 ⊥b2 + g 2 k(1 − b2) (4) with g⊥/gk = 0.9 we obtain the exact coincidence over all range of tilt angles with X = 0.405 (see dot line in Fig. 3). In Fig. 3, we have plotted also the angular dependences of A(b)/A(1) with close pairs of parameters. This makes it possible to assess how uniquely these parameters are determined. Note that the value of X < 0.5 is consistent with the fact that the onset of QHE is observed with even numbers [see Fig. 2(a)]. Such data treatment was carried out for other elec- tron densities and all the results for gµBB/ωc versus r r 2000 ( a ) ) m h O ( y x 1000 0 0.0 175 ) m h O ( x x 150 ) s t i n u . b r a ( B^ d / x x dr ( b ) b=1 0.67 0.50 0.44 0.33 0.26 0.19 0.5 1.0 1.5 0.4 B (T) 0.6 B^ (T) 0.8 Figure 4. (Color online) (a) The magnetic field dependences of ρxx and ρxy at normal magnetic field for the structure 1023 (d < dc) at n = 4.6 × 1011 cm−2. T = 4.2 K. (b) The dependences of the derivatives dρxx/dB⊥ on B⊥ for different b. the electron density are plotted in Fig. 7 together with the results obtained for another structure with d < dc, 1122 (see Table I). Before comparing these results with the theoretical de- pendences of gµBB/ωc on the electron density, let us consider the data for structures with inverted spectrum (d > dc). As an example, in Fig. 4, we have presented the data for structure 1023 at n = 4.6 × 1011 cm−2. One can see that at B < 0.8 T, the oscillations of ρxy are practically absent and the spin splitting of the oscillations of ρxx is not observed. So, at B < 0.8 T, the conditions for appli- cability of Eq. (1) are met. The derivatives dρxx/dB⊥, measured at different tilt angles as a function of B⊥ are presented in Fig. 4(b) for different b. For clarity, we have plotted the dashed line at B⊥ = 0.68 T which cor- responds to the position of one of maxima. It is clearly seen that with the tilt angle increase, the positions of the extremes of dρxx/dB in B⊥, similarly to in structures with d < dc (see. Fig. 2b), does not change but the am- plitude of the oscillations decreases significantly slower than in the structure with d < dc, and the inversion of the oscillations phase is not observed up to b = 0.19 [compare Fig. 2(b) and Fig. 4(b)]. The dependence of the amplitude of oscillations on b together with the calculated dependence, Eq. (3), with the isotropic g-factor is presented in Fig. 5. One can see that this simple dependence describes well the data over b range from 1 to (0.25 − 0.2) with X = 0.13. Note, this value is three-four times as low as that for the structures with normal spectrum (see Fig. 3). Let us check how un- ambiguously the value of X is determined for this case. To this end, we have plotted in Fig. 5 the A(B⊥, b) ver- sus b dependences which were calculated using two free parameters, namely gk/g⊥ and X. One can see that the experimental data are equally well described with very different pairs of X and gk/g⊥. To avoid such a large ambiguity, let us consider os- 4 cillations of ρxx(B) in a larger magnetic field, where the Zeeman splitting starts to be observed but lower than the onset of QHE. Such experimental dependence in the nor- mal field is presented in Fig. 6 together with the curves calculated using the LK formula with different X values. We assumed the Lorentz broadening of LLs with param- eter ∆L = 4.5 meV which was found from the magnetic field dependence of ρxx(B) at B < 1 T. The inset shows that the calculated curve with X = 0.13 radically differs from the experimental curve, it does not demonstrate the Zeeman splitting of the oscillations up to 3 T. The calcu- lated curves with X = 0.35 − 0.48 are significantly closer to the experimental dependence ρxx(B), therewith the curve with X = 0.42 practically reproduces the exper- imental dependence. Thus, the comparison of the data with the calculated curves in Fig. 5 and Fig. 6 gives the possibility to find unambiguously the values of X and gk/g⊥ as 0.42 ± 0.03 and 0.19 ± 0.02, respectively. As seen from Fig. 5, some discrepancy between the A(b)/A(1) data and calculated curves remains at b < 0.2. The reasons for this discrepancy are not clear. TThey may be: (i) effect of sufficiently large in-plane component of B because at b = 0.2, the in-plane component of B is about 3 − 4 T, which corresponds to the magnetic length L = (13 − 14) nm so that L became comparable to the width of the quantum well width (8 − 9) nm. That can change the energy spectrum noticeably. (ii) a difference in the broadening (or difference in contribution to the oscillations) of different spin sub-levels; (iii) imperfect flatness of the quantum well and so on. Nevertheless, we believe that the value of X = 0.42 ± 0.02 corresponds to the ratio of the Zeeman splitting to the cyclotron energy in the normal magnetic field and g-factor anisotropy is gk/g⊥ = 0.19 ± 0.02. The described above measurements and data treat- ment were carried out for all structures from Table I over the wide electron density range. All obtained values of X and gk/g⊥ versus the electron density are summarized in Fig. 7. One can see that the ratio of the Zeeman splitting to the orbital one is close to each other for both types of the structures, with the normal and inverted spectrum. This ratio decreases slightly from X ≈ 0.5 to ≈ 0.4, as the electron density increases from n ≈ 1 × 1011 cm−2 to 5 × 1011 cm−2. The values of the g-factor anisotropy in the structures with normal and inverted spectra differ significantly [see Fig. 7(b)]. The values of gk/g⊥ in the structures with the normal spectrum are in the range of 0.6 − 0.9, while in the structures with inverted spectrum they are in the range of 0.05−0.015. For both types of the structures the values of gk/g⊥ increase with increasing electron density. Let us compare our data with the results of previous studies. We have found only two articles [10, 11] where the Zeeman splitting was studied in the structures with d ≈ dc and we plotted them in Fig. 7. In paper [11], the Zeeman splitting was determined for the normal mag- netic field only in the structure with d = 9 nm at the r r 0.6 T 0.7 T 0.8 T 1.0 0.5 0.0 ) 1 ( A / ) b ( A X gII/g^ 0.13 1 0.42 0.19 0.30 0.38 0.2 0.63 0.2 0.4 0.6 b 0.8 1.0 Figure 5. (Color online) The relative oscillation amplitude for structure 1023 plotted against b. Points are experimental data found at B⊥ = 0.6, 0.7, and 0.8 T at the electron density n = 4.6 × 1011 cm−2. The lines are the dependences Eq. (3) with different pairs of the parameters X and gk/g⊥ shown in figure. electron density n = 6.59 × 1011 cm−2. The value of X agrees well with our data (see Fig. 7). In Ref. [10], both the Zeeman splitting and g-factor anisotropy were found for the structure with d = 6.1 nm for the very high electron density, n = 1.46 × 1012 cm−2. The value of the g-factor anisotropy is found to be close to our data for the structure with d = 5.6 nm [see Fig. 7(b)], while the ratio gµBB/ωc is significantly larger than our data: 1.26 instead of 0.4 − 0.5. Such difference is unclear. One of possible reasons is role of spin-orbit interaction, which can be large for so high electron density and was not taken into account in the analysis of the data. Now let us compare the obtained results with the theo- retical ones. To find X=gµBB/(eB⊥/me), the positions of the Landau levels have been calculated in framework of the 8-band kP model [5]. Since there are different no- tations of the Landau levels in various papers, we have numbered the levels in a row, starting from unity for the lowest LL of the conduction band. The Zeeman splitting was found as the energy distance between the levels n and n + 1 with odd n, while the orbital splitting was found as the distance between the levels n and n + 2 [18]. The cal- culated gµBB/ωc versus n dependences are plotted in Fig. 7(a) by solid lines. It is seen that the experimental values are slightly ower than the calculated ones for the structures with the normal and inverted spectrum. It is instructive to compare the results of the calculations per- formed in the framework of the 8-band kP model with those obtained within the framework of the Bernevig- Hughes-Zhang (BHZ) model [4], which is often used to analysis various effects. We have used the parameters of the BHZ model which give the dependence E(k) very close to that calculated in framework of the 8-band kP model. However, the Zeeman splitting in this case ap- os(3.140.3sqrt(x^2+0.38^2(1-x^2))/x)/cos(3.140.3) cos(3.140.42sqrt(x^2+0.19^2(1-x^2))/x)/cos(3.140.42) cos(3.140.2sqrt(x^2+0.63^2(1-x^2))/x)/cos(3.140.2) cos(3.140.13/x)/cos(3.140.13) 3000 5 X=0.13 ) m h O ( y x 2000 1 1000 X=0.46 0.42 0.38 0 -1000 0 2 B (T) 3 200 ) m h O ( x x 100 2 1 B (T) Figure 6. (Color online) The experimental dependences of ρxx and ρxy for structure 1023 in the normal magnetic field at n = 4.6 × 1011 cm−2 for magnetic field range larger than in Fig. 4 (points). The solid lines are the calculated ρxx curves with different values of X. These curves are shifted for clarity. The inset shows comparison of the data with the calculation with X = 0.13 up to B = 3 T (see text). pears to be 20 − 30 percent larger [see dashed lines in Fig. 7(a)]. To compare the data for the g-factor anisotropy with the theory, one needs to know the values of gk together with g⊥ calculated just above. The dependences of gk on electron density were calculated using the results of the paper [19] where the energy spectrum of the HgTe quantum wells in the in-plane magnetic field was stud- ied. The calculated dependences of gk/g⊥ versus electron density are shown in Fig. 7(b). It is seen that for both types of the structures, with the normal and inverted spectrum, the theoretical values of gk/g⊥ are small, they are close to each other, and increase with the electron density increase. The calculated values of gk/g⊥ signif- icantly differ from the experimental data for both types of the structures. For the structures with the inverted spectrum (d > dc), the experimental values are to 1.5 − 2 times lower. For the structures with the normal spec- trum, the difference is larger and the experimental data are to 3 − 4 times higher than the calculated ones. To discuss possible reasons for the discrepancy, let us remind the results of our previous paper [9]. We have shown (see Introduction) that for d ≈ dc in nominally symmetric structures the top of the valence band is very strongly split by SO interaction [8]. Therewith, the SO splitting of the conduction band in the same structures does not reveal itself [9]. It is surprising that such SO splitting is observed in structures with the inverted and r r )^ c hw / B B m g ( 1.4 1.2 1.0 0.8 0.6 0.4 0.2 0.0 ( a ) 1 2 3 4 5 d<6.3 d>6.3 5 nm 7 nm 1.0x1011 1.0x1012 n (cm-2) / I I 0.4 0.2 0.0 ( b ) 1.0 0.8 0.6 g g 4.6 nm d<6.3 5.6 nm d>6.3 1.0x1011 1.0x1012 n (cm-2) Figure 7. (Color online) The ratio of the Zeeman splitting to the orbital one at normal magnetic field (a) and g-factor anisotropy (b) plotted against the electron density. The solid and open symbols are the experimental data for the struc- tures with normal and inverted spectra, respectively. The structures numbers are presented in (a). The diagonal crosses are the result of [11] and straight cross are the result of [10]. The lines are the calculated dependences for d = 5 nm and d = 7 nm (see text). normal spectrum despite the fact that at d < dc and d > dc the conduction band is formed from different terms (see Fig. 1). It was assumed in [9] that such sur- prising behavior of the SO splitting is a result of the in- terface inversion asymmetry in the HgTe quantum well, which is not taken into account in kP calculations [1 -- 5]. We believe that the disagreement between the experi- mental data on g-factor anisotropy and calculations is also a result of the interface inversion asymmetry in the 6 HgTe quantum well, which is not taken into account in kP calculations. In summary, the ratio of the Zeeman splitting to the orbital one and anisotropy of the g-factor in the HgTe quantum wells both with normal and inverted spectrum have been studied experimentally within a wide electron density range. To obtain two these parameters unam- biguously, we have analyzed both the tilt angle depen- dence of the SdH oscillations in low magnetic fields and the shape of the oscillations in moderate magnetic fields. It has been shown that the ratios of the Zeeman splitting to the orbital one are close to each other in the structures with normal and inverted spectra, these ratios decrease when the electron density increases and they are quite close to the values calculated within the kP method. In contrast, the anisotropy of g-factor in the structures with the normal and inverted spectrum is strongly different and for both cases differ significantly from the calculated ones. We believe that such disagreement with the calcu- lations is a result of the interface inversion asymmetry in the HgTe quantum well, which is not taken into account in the kP calculations. ACKNOWLEDGMENTS We are grateful to S. Studenikin, V. Aleshkin, A. V. Germanenko and O. E. Raichev for useful discus- sions and M. Zholudev for calculations of LLs in frame- work of 8-band kP model. The work has been supported in part by the Russian Foundation for Basic Research (Grants No. 16-02-00516 and No. 15-02-02072) and by Act 211 Government of the Russian Federation, agree- ment No. 02.A03.21.0006. A.V.G. and O.E.R. grate- fully acknowledge financial support from the Ministry of Education and Science of the Russian Federation under Projects No. 3.571.2014/K and No. 2457. 1 L. G. Gerchikov and A. Subashiev, Phys. Stat. Sol. (b) 160, 443 (1990). 2 X. C. Zhang, A. Pfeuffer-Jeschke, K. Ortner, V. Hock, and G. Landwehr, H. Buhmann, C. R. Becker, Phys. Rev. B 63, 245305 (2001). 3 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). 4 B. A. Bernevig, T. L. Hughes, and S.-C. Zhang, Science 314, 1757 (2006). 5 M. Zholudev, Ph.D. thesis, University Montpellier 2, France (2013). 6 D. A. Kozlov, Z. D. Kvon, N. N. Mikhailov, S. A. Dvoret- skii, and J. C. Portal, Pis'ma Zh. Eksp. Teor. Fiz. 93, 186 (2011), [JETP Lett. 93, 170 (2011)]. 7 G. M. Minkov, A. V. Germanenko, O. E. Rut, A. A. and N. N. Mikhailov, Sherstobitov, S. A. Dvoretski, Phys. Rev. B 88, 155306 (2013). 8 G. M. Minkov, A. V. Germanenko, O. E. Rut, A. A. and N. N. Mikhailov, Sherstobitov, S. A. Dvoretski, Phys. Rev. B 89, 165311 (2014). 9 G. M. Minkov, A. V. Germanenko, O. E. Rut, A. A. Sher- stobitov, M. O. Nestoklon, S. A. Dvoretski, and N. N. Mikhailov, Phys. Rev. B 93, 155304 (2016). 10 M. Pakmehr, C. Bruene, H. Buhmann, L. W. and B. D. McCombe, Molenkamp, A. V. Stier, Phys. Rev. B 90, 235414 (2014). 11 X. C. Zhang, K. Ortner, A. Pfeuffer-Jeschke, C. R. Becker, and G. Landwehr, Phys. Rev. B 69, 115340 (2004). 12 N. N. Mikhailov, R. N. Smirnov, S. A. Dvoretsky, Y. G. Sidorov, V. A. Shvets, E. V. Spesivtsev, and S. V. Rykhlit- ski, Int. J. Nanotechnology 3, 120 (2006). 13 F. F. Fang and P. J. Stiles, Phys. Rev. 174, 823 (1968). 14 S. A. Studenikin, P. T. Coleridge, G. Yu, and P. Poole, Semicond. Sci. Technol. 20, 1103 (2005). 15 E. V. Kurganova, H. J. van Elferen, A. McCollam, L. A. Ponomarenko, K. S. Novoselov, A. Veligura, B. J. van Wees, J. C. Maan, and U. Zeitler, Phys. Rev. B 84, 121407 (2011). 16 An exception is the case when ∆E = 0.5ωc, and the am- plitude of the first harmonic is zero. ^ 17 The electron density for all structures investigated linearly depends on gate voltage Vg, n(Vg) = n(0) + αVg and α, within experimental error, coincides with C/(Se), where C is the capacitance measured in the same sample; S is the gate area. 18 The facts that the onset of QHE is observed with even numbers and the decrease in the amplitude of the oscilla- tions at deviation of the magnetic field from the normal to 2D plane show that the ratio of namely these values is determined from dependence A(b). 19 O. E. Raichev, Phys. Rev. B 85, 045310 (2012). 7
1104.2032	1	1104	2011-04-11T19:46:23	On the imaging of electron transport in semiconductor quantum structures by scanning-gate microscopy: successes and limitations	[ "cond-mat.mes-hall" ]	This paper presents a brief review of scanning-gate microscopy applied to the imaging of electron transport in buried semiconductor quantum structures. After an introduction to the technique and to some of its practical issues, we summarise a selection of its successful achievements found in the literature, including our own research. The latter focuses on the imaging of GaInAs-based quantum rings both in the low magnetic field Aharonov-Bohm regime and in the high-field quantum Hall regime. Based on our own experience, we then discuss in detail some of the limitations of scanning-gate microscopy. These include possible tip induced artefacts, effects of a large bias applied to the scanning tip, as well as consequences of unwanted charge traps on the conductance maps. We emphasize how special care must be paid in interpreting these scanning-gate images.	cond-mat.mes-hall	cond-mat	On the imaging of electron transport in semiconductor quantum structures by scanning-gate microscopy: successes and limitation. H. Sellier1, B. Hackens2,§, M.G. Pala3, F. Martins2, S. Baltazar3, X. Wallart4, L. Desplanque4, V. Bayot1,2, S. Huant1,§ 1 Institut Néel, CNRS and Université Joseph Fourier, BP 166, 38042 Grenoble, France 2 IMCN, Pôle NAPS, Université Catholique de Louvain, B-1348 Louvain-la-Neuve, Belgium 3 IMEP-LAHC, UMR 5130, CNRS/INPG/UJF/UdS, Minatec, Grenoble INP, Grenoble, France 4 IEMN, UMR CNRS 8520, UST Lille, BP 60069, 59652 Villeneuve d’Ascq, France § Authors to whom correspondence should be addressed E-mail : [email protected] [email protected] Short title: Imaging electron transport with scanning-gate microscopy PACS: 73.21.La, 73.23.Ad, 03.65.Yz, 85.35.Ds Keywords: scanning-gate microscopy, nanostructures, electron transport Abstract. This paper presents a brief review of scanning-gate microscopy applied to the imaging of electron transport in buried semiconductor quantum structures. After an introduction to the technique and to some of its practical issues, we summarise a selection of its successful achievements found in the literature, including our own research. The latter focuses on the imaging of GaInAs-based quantum rings both in the low magnetic field Aharonov- Bohm regime and in the high-field quantum Hall regime. Based on our own experience, we then discuss in detail some of the limitations of scanning-gate microscopy. These include possible tip induced artefacts, effects of a large bias applied to the scanning tip, as well as consequences of unwanted charge traps on the conductance maps. We emphasize how special care must be paid in interpreting these scanning-gate images. 1. Introduction Scanning-gate microscopy (SGM) uses the electrically polarized tip of a low-temperature atomic-force microscope (AFM) to scan above a semiconductor device while the conductance changes due to the tip perturbation are simultaneously mapped in real space. Since its introduction in the late nineties [1- 3], SGM has proven powerful to unravel the local details of electron behaviour inside modulation- doped nanostructures whose active electron systems are, in contrast to surface electron-systems, not accessible to scanning-tunnelling microscopy (STM) because they are buried too deep below the free surface. In this paper, we first describe the technique and some technical issues related to it (section 2). Then, we give in section 3 a review of some of its instructive achievements found in the literature, with a special emphasis on our own research that deals with the imaging of electron transport across GaInAs quantum rings both in the (low-field) Aharonov-Bohm (AB) regime and in the (high-field) quantum Hall regime. Starting from our own experience, we discuss in section 4 the main limitations of the SGM technique that may be encountered and emphasize where caution must be paid either to operate SGM or to interpret the images that it produces. Figure 1. Principle of SGM. A low-frequency (about 1 kHz) probe current Iapplied (of typically 1-10 nA) is applied to the device, here a quantum ring patterned from a 2DEG buried at a few tens of nanometres below the free surface. The measurement of the voltage drop Vmeasured across the device gives access to its electrical conductance G or resistance R. The AFM tip is biased with a voltage Vtip typically of the order of a few tenths of volts up to a few volts and scanned at a distance d of typically 20 to 50 nm above the device (or alternatively, the sample is scanned under the fixed tip). This modifies the conductance (resistance). Imaging the conductance changes ΔG(x,y) or, alternatively, resistance changes ΔR(x,y) due to the tip perturbation as function of the lateral tip position (x,y) builds what is called a conductance, respectively resistance, SGM image in this paper. 2. The technique of scanning-gate microscopy 2.1 Principle The principle of SGM as applied to the imaging of a semiconductor nanostructure (here ring-shaped) patterned from a buried two-dimensional electron gas (2DEG) is sketched in figure 1. A voltage- biased (Vtip) AFM tip is scanned at an altitude of 20 to 50 nm over the device surface to perturb locally one of its macroscopic electrical properties whose induced changes are mapped as function of the tip position. In the example of figure 1, it is the conductance G (or resistance) of the device that is imaged as function of the tip position in a current-biased device (Iapplied) configuration. Alternatively, a voltage- biased device configuration can be used in particular in high impedance devices like quantum point contacts (QPCs) (see eg. [2,3]). The whole setup is plunged into a cryostat to operate at low temperature, down to below 100 mK in a dilution refrigerator for the coldest SGM setups [4,5]. Optionally, an external magnetic field can be applied. The combined low-temperature and magnetic-field environment requires the use of cryogenic magnetic-free displacement units, such as for example titanium-made inertial step motors [6] for the in situ coarse positioning of the tip relative to the nanostructure over a few millimetres. Fine positioning over a few micrometres for the image acquisition is ensured by commercially available piezoelectric scanner elements. 2.2 Some technical issues SGM experiments have several difficult steps to deal with. The most critical one perhaps is to locate the region of interest, without damage (either electrical or mechanical) to both the tip and sample, in a top-loading microscope that has no optical access to the user. This can be achieved by using the microscope in the AFM “tapping” mode to measure the topography of the sample. The surface can be supplied with a set of patterned arrows and other “traffic signs” pointing to the nanostructure, and locating these signs by recording successive images can serve as a guide to drive the tip towards the active device for subsequent SGM imaging. Measuring the sample topography requires measuring the AFM cantilever deflection by some means. Instead of using an optical method as commonly done in AFM, it is recommended using a light-free setup such as soft piezoresistive cantilevers [1-3] or commercially available quartz tuning forks [7]. This is because semiconductor materials are photosensitive and illuminating them at low temperature can induce charge redistributions in their electrostatic background and undesired persistent changes in their carrier density [8]. Tuning forks have originally been introduced for near-field scanning optical microscopy [6,7,9] and extended later to other scanning-probe microscopy (SPM), such as for example the low- temperature combined AFM-STM [10] used to study partially conducting hybrid devices. For the purpose of SGM, either a metallic tip or a commercial conductive AFM cantilever can be electrically anchored on one metallic pad of the tuning fork. A further advantage of using a tuning fork is that both the topographic profile and conductance images can be recorded simultaneously if required. A drawback however is that the tuning fork is much stiffer than a standard AFM cantilever (static spring constant in excess of 20.000 N/m compared to a few N/m for a cantilever), and special care must be taken not to apply a too large force [7] on the sample. This can be achieved by using a phase-locked loop regulation of the tuning fork vibration. Otherwise, it could irreversibly damage the sample or rearrange residual charges in the device environment, thereby altering its electronic properties. Once the tip has located the device, it is lifted at some tens of nanometres above the surface and scanned in a plane parallel to the 2DEG (the tip in general does not follow the topography). Therefore, all subsequent SGM measurements are carried out without contact to the surface. However, it may happen that drifts occur during SGM imaging, especially if parameters like temperature or magnetic field are varied. In this case, additional topographic images are acquired to compare with the initial topography and possibly adjust the tip position. 3. Some successful achievements of scanning-gate microscopy Here, we give a summary of a few selected achievements of SGM, which, we believe, have given strong impulse to this imaging technique. We refer the reader to the cited literature for more information. 3.1 Quantum-point contacts in two-dimensional semiconductor heterostructures One spectacular achievement is the imaging of coherent electron flow from QPCs patterned in high- mobility GaAs/GaAlAs 2DEGs. In QPCs, the conductance is quantized in units of 2e2/h (e is the electron charge, h the Planck’s constant), each plateau corresponding to a conductance channel propagating through the QPC with near unity transmittance [11]. SGM has been able to image the three lowest conductance modes in real space and has shown that each mode contributes to a number of spatial electron “strands”. The latter could be indexed in correspondence with their parent’s mode index, i.e., the lowest mode contributes one strand; the second mode contributes two strands, and so on. Each mode has a well-defined nodal structure that agrees well with the corresponding squared wavefunction [2]. In addition to this nodal structure, the electron flow images are decorated with fine undulations at half the Fermi wavelength that are oriented perpendicularly to the flow and extend to micrometres away from the QPC [2,3]. These interference patterns are due to coherent backscattering of electron waves between the tip depleted region and the QPC at short distance from the QPC, or between the tip and localized scattering centres at larger distances. Their occurrence confirms that the electron flow is imaged in the coherent regime of transport (the phase coherence length is of the order of a few µm in these systems at low temperature), whereas their spacing at half the Fermi wavelength can be used to probe the spatial profiles of the electron density in the 2DEG [12]. A similar electron interference imaging experiment has been done using an additional mirror gate to backscatter part of the electron flow towards the QPC [13]. Another interesting feature of the coherent flow from a QPC is that it splits in an increasing number of smaller ramifications at increasing distances from the QPC. This branching is explained as due to focusing in the valleys of the background potential and to the presence of hard scattering centres [3]. Therefore, SGM turns out to provide additionally a visual insight into this background potential. More recent work has revealed that the larger the 2DEG mobility, the farther from the QPC does the above branching occur [14], in agreement with an increased mean free path. In the highest mobility samples at an usual operation temperature of 1.7 K, the fine decoration at half the Fermi wavelength mentioned above does not survive at large distances from the QPC as a consequence of a cleaner background potential landscape and a finite thermal smearing of the electron interferences. However, these fringes are recovered at lower temperatures (350 mK) and allow probing spatially phase coherent properties in the “clean” electron interferometer formed by the SGM tip and the QPC [15]. In a very recent work, electron-electron scattering in the 2DEG has also been studied in these interferometers [16]. The above imaging of QPCs has been extended along several directions. One is dealing with “magnetic steering”. In presence of a perpendicular magnetic field, the Lorentz force bends the electron flow in the 2DEG that forms closed cyclotron orbits. SGM has revealed these cyclotron trajectories in a sample containing two QPCs in series [17,18]. Another extension deals with the visualization of quantum interference patterns within a QPC whose area distribution could be shown by SGM to be similar to that of the magneto-conductance fluctuations of the QPC, thereby indicating a common origin for both effects [19]. The last extension we wish to mention goes beyond imaging, in that the SGM tip has been used to finely tune the conductance quantization of the QPC, in particular to investigate the nature of the anomalous 0.7 plateau that often appears below the conductance quantum [20,21]. The imaging of coherent electron flow from QPCs has given strong confirmation that SGM is a powerful tool to image at the local scale, study and control electron transport in mesoscopic structures that are not accessible to STM. It has stimulated several theoretical analyses based on Green’s function techniques or perturbation theory [22-25] and has led to an increasing number of experimental groups developing the technique and applying it to a broadening range of buried structures and devices as written in the remaining of this section. 3.2 Semiconductor quantum dots and rings Since the SGM tip behaves as a local gate, it can be used to study Coulomb blockade in quantum dots (QDs). Scanning the biased tip around a dot draws a set of concentric isopotential lines corresponding to successive charging states of the dot. This effect allows one to precisely locate the position of a dot when it is unknown, like in the case of multiple dots along nanowires [26] and nanotubes [27], or traps in the barrier of a 2DEG heterostructure [28-30]. Coulomb blockade has also been used to investigate the spatial extension of a localized wavefunction inside a single-electron QD [31] or, conversely, to map the spatial distribution of the tip induced potential [32,33]. SGM was also applied to large open QDs, called quantum billiards, to image scarred wavefunctions corresponding to classical orbits and their periodicity in magnetic field [34]. SGM has also proven to be successful in imaging quantum rings (QRs), where it has been able to give images of Aharonov-Bohm interferences [35], of the electronic local density of states [36-38], and of Coulomb islands in the Quantum Hall regime [5]. These results on QRs are the focus of two subsections below. 3.3 Carbon-based micro- and nanostructures If SGM has been primarily applied to modulation-doped structures made of conventional semiconductors, it has also been very successful with other systems such as carbon-based structures. Carbon nanotubes, for example, have been the focus of one of the very first SGM experiments that revealed the exact position of several quantum dots appearing along the nanotube and giving rise to complex Coulomb blockade conductance oscillations in transport experiments [27]. This widening interest has been confirmed by the recent publication of several papers dealing with the SGM imaging of single-layer graphene flakes [39], the study of the influence of a tip-induced scattering potential on universal conductance fluctuations [40] and on weak localization [41] in graphene devices and the imaging of graphene QDs coupled to source and drain by narrow constrictions [42]. In the latter case for instance, SGM conductance images draw three overlapping sets of concentric Coulomb rings that are centred on the dot, for one of them, and on the two constrictions, for the two others. These ring-like patterns build real-space images of the Coulomb blockade in the dot and the two constrictions, and reveal the exact location of the Coulomb-blockading species, in pretty much the same way as in the semiconductor QDs summarized in the previous subsection. In addition to this imaging ability, SGM is able to extract valuable quantitative information such as the extension of the involved localized states and the charging energy of the dot. 3.4 Quantum rings in the Aharonov-Bohm regime An open QR in the coherent regime of transport sees its conductance increasing to a maximum when electron waves interfere constructively at the output contact and decreasing to a minimum for destructive interferences. Varying either the magnetic flux captured by the QR or the electrostatic potential in one arm allows the interference to be tuned by changing the phase accumulated by electrons as they are transmitted through the QR’s arms. This gives rise to the well-known magnetic and electrostatic [43,44] AB oscillations in the ring conductance [45]. An example of AB oscillations is given in figure 4a, showing the magnetoresistance of a QR with a radius of 520 nm, and a temperature of 90 mK. The temperature dependence of the AB oscillations amplitude is directly related to that of the phase coherence length or time, and saturates at low temperature, when the phase coherence time is equal to the dwell time (i.e., the time spent by electrons inside the device, statistically), as illustrated in figure 1 of [36]. Since the dwell time depends on the QR area and on the number of modes in the QR openings, this saturation occurs at different temperatures for different quantum ring geometries [46]. SGM can image these archetypical interference phenomena in real space [35]. The electrostatic AB effect gives rise at low magnetic field to a well-developed fringe pattern in the conductance image of GaInAs-based QRs in the coherent regime of transport when the tip scans outside the QR, as illustrated in [35]. This outer pattern is mainly concentric with the ring geometry. The qualitative interpretation (see [47] for a more quantitative approach) in terms of a scanning-gate- induced electrostatic AB effect is that as the tip approaches the QR, either from the left or right, the electrical potential mainly increases on the corresponding side of the QR. This induces a phase difference between electron wavefunctions travelling through the two arms of the ring, and/or bends the electron trajectories, which produces the observed pattern. Modifying the magnetic field strength contributes another phase term through the magnetic AB effect and displaces the whole pattern with respect to the QR. This displacement is periodic in magnetic field strength with the same periodicity as the AB oscillations seen in the magneto-conductance, which gives further support to the interpretation in terms of AB effects. Figure 2. (a) Fluctuations of magnetoresistance measured at 90 mK on a large GaInAs QR (outer diameter 1.2 µm, rim width 400 nm). (b-c) Unfiltered SGM resistance maps at two different temperatures of a large GaInAs QR (Vtip = -1 V). The black lines are guide to the eye pointing to the QR topography that has been recorded prior to the acquisition of the resistance maps. At 4.2 K (left image) the resistance drifted as the tip was scanning over the quantum ring, due to tip-induced switching of traps in the vicinity of the QR. At 90 mK, the SGM image contrast can be interpreted in terms of coherent effects, as explained in the text. In this framework, one can expect that the evolution of the SGM contrast with temperature be related to the temperature dependence of the AB effect. The confirmation of this hypothesis is shown in [36] (figure 1), for the range 3-40 K. Note that in the case of this particular QR, both the SGM contrast and the AB oscillation amplitude saturate below 10 K, an effect related to the dwell time, as explained above. In QRs with different geometrical parameters, it can be advantageous to go to lower temperatures to enhance the SGM contrast, as illustrated in figure 2. Here, the QR has both a larger diameter (1.2 µm) and wider arms (400 nm) than those studied in our previous works [35,36]. In this geometry, electrons have the opportunity to probe a much larger set of semiclassical trajectories. As a consequence, the signature of electron interference in the magnetoresistance is no longer made of oscillations with a single characteristic frequency, but rather aperiodic reproducible fluctuations, the so-called “Universal Conductance Fluctuations”, shown in figure 2a. The superposition of trajectories also induces an “averaging” effect, which reduces the amplitude of the contribution of coherent effects in the magnetoresistance, compared to rings with narrower arms. As a consequence, it is necessary to reach lower temperatures to recover a good contrast in SGM images. This is shown in figures 2b and 2c, comparing SGM resistance images measured on this “large” QR at two temperatures, T= 4.2 K and 90 mK (and zero magnetic field). Note that the images of figure 2 are unfiltered original images in contrast with [35], where a slowly varying background tip effect was subtracted. The amplitude of the contrast related to coherent effects in SGM images is strongly enhanced at the lowest temperature, and a rich set of details is revealed. These data therefore clearly show that the improvement of SGM when the temperature decreases depends on the system under investigation. QR conductance images also exhibit a complex pattern when the tip scans directly over the QR region (see figure 4d and data in [35-38]). These inner fringes have been discussed in our previous papers and linked to the electron-probability density in the QR [36-38]. A detailed analysis based on quantum mechanical simulations of the electron probability density, including the perturbing tip potential, the magnetic field, and the presence of randomly distributed impurities, is able to reproduce the main experimental features and demonstrates the relationship between conductance maps and electron probability density maps at the Fermi level. An example of such a relationship is shown in figure 3 in the case of a realistic QR perturbed by negatively charged impurities [48]. Although the impurities distort the electronic local density-of-states (LDOS), this distortion is reflected back in the conductance image in such a way that the conductance map can still be seen as a mirror of the LDOS, provided that the perturbation induced by the tip is “small and narrow enough” (details concerning the range of validity of the correspondance LDOS-SGM map in QRs can be found in [36-38]; see also subsection 4.3 below). As seen in figure 3, both the LDOS and conductance images tend to develop radial fringes, which are mostly, but not entirely, anchored to the impurity locations and are reminiscent of the experimental images that are often found to be asymmetric. Note that it is hopeless to try to find a direct and exact correspondence between simulations such as those presented in figure 3 and experimental data, unless the precise geometry of the experimental device and impurity distribution are taken into account. Indeed, the LDOS pattern in mesoscopic structures is extremely sensitive to slight changes in the electron confining potential (as shown in [36]). However, qualitative and general trends, such as the tendency to develop radial fringes, can be compared. Figure 3. Quantum mechanical simulation of a SGM experiment on a QR in the presence of negatively charged impurities. The outer diameter, inner diameter and opening width are 530 nm, 280 nm, and 120 nm, respectively. The effective mass is 0.04 m0, with m0 the free electron mass. The Fermi energy is EF= 95.9 meV. (a) and (b) are the simulated images of the LDOS and conductance changes (in units of G0, the quantum of conductance), respectively, calculated for the random distribution of negatively-charged impurities shown in (c). In the simulation, the tip potential has a Lorentzian shape with a 10 nm range and repulsive amplitude of EF/50. Reprinted from [38]. The discussion above, as well as data presented in [35-38], suggest that SGM can be viewed as the analogue of STM [49] for imaging the electronic LDOS at the Fermi level in open mesoscopic systems buried under an insulating layer or the counterpart of the near-field scanning optical microscope that can image the photonic LDOS in confined nanostructures, provided that the excitation light source can be considered as point-like [50] such as in active tips based on fluorescent nano- objects [51,52]. However, imaging the LDOS of a buried electron system with SGM requires some caution to be taken, as the technique is less direct than STM applied to surface electron systems. This point is discussed further below in subsection 4.3. 3.5 Quantum rings in the quantum Hall regime Several groups have applied SGM to study 2DEGs in the quantum Hall regime, for instance to image edge states, localized states, and inter-edge-state scattering centres in this high magnetic field regime [53-55]. In this subsection, we wish to show how SGM can be useful in the study of the QHE physics by describing in some detail our study dealing with QRs in this regime. Figure 4. (a-c) Magnetoresistance of a QR patterned from an InGaAs/InAlAs heterostructure. The lithographic inner and outer radii of the SR are 215 and 520 nm, respectively, and the opening widths are 300 nm. A 2DEG is buried at 25 nm below the free surface with a carrier density of 1.4x1012cm-1. The curve in (c) is the QR magnetoresistance, whose low-field and high-field parts are zoomed in (a) and (b). (d-e) Two typical unfiltered SGM resistance maps taken at zero magnetic field (d) and in a high field of 9.65 T (e). The voltage applied to the tip is +0.5 V at zero field and +1 V at high field. The tip to surface distance is 50 nm. All data are taken at a temperature of 100 mK. In QRs, the AB related phenomena described in the previous subsection occur at low magnetic field and they all disappear at higher field. This is because at high field, the cyclotron radius shrinks below the width of the QR arms and openings, and electrons tend to propagate along the edges of the device. This completely changes the electron dynamics and the contrast in SGM images, as shown hereafter. The gradual evolution to the high-field regime is illustrated in figures 4a-c, displaying the magnetoresistance of a 1 µm-diameter QR, measured at 100 mK (same ring as in [5]). Below B = 2 T, the magnetoresistance shows periodic oscillations superimposed on a smooth background that can clearly be seen in the low-field zoom on the magnetoresistance in figure 4a. In this B-range, the SGM contrast is determined by interference effects (concentric features in the vicinity of the quantum ring), and by the details of the electron probability density in the quantum ring region, as explained above (see also [35]). Above 2T, the ring magnetoresistance first exhibits Shubnikov-de Haas oscillations, and, at higher B, signatures of the quantum Hall (QH) regime [56], with an important difference with respect to the QH effect in macroscopic samples. Indeed, in the B-ranges where the resistance drops to zero in macroscopic Hall bars (around integer filling factors), the ring magnetoresistance shows a rich pattern of fluctuations [57-59]. An example of such fluctuations is shown in the high-field zoom on the magnetoresistance, in figure 4b. To understand the origin of these fluctuations, different models have been proposed, based either on direct tunnelling between counter-propagating QH edge states, or on tunnelling through localized states (“Coulomb islands”) inside the device [60]. Such Coulomb islands can be pinned by local maxima or minima in the 2DEG potential. When the flux of magnetic field through the area of the island increases by one magnetic flux quantum, one electron must be added to each occupied Landau level in the island. This induces a change in the island charging energy, which therefore oscillates, with a period given by φ0/fc, where fc is the number of occupied Landau levels and φ0 is the flux quantum [60]. When the biased tip of the SGM approaches such a Coulomb island, the associated modification in the potential pinning the island results in a change of the island area and, hence, of the magnetic flux through this island. As this magnetic flux controls the number of trapped electrons in the island, this induces Coulomb-type oscillations in SGM images. This is illustrated on the high-field (9.65 T) low-temperature resistance map in figure 4e. Different sets of concentric resistance fringes are visible in the image, in the lower left part of the ring, and above the central hole of the ring. The centre of each set of fringes corresponds to a localized Coulomb island, as shown in [5]. In this regime, SGM therefore allows to locate individual QH Coulomb islands, and to “manipulate” them individually, through local modifications of the electron confining potential. Knowing the location of each island, and observing the dependence of the SGM fringe pattern with the magnetic field leads to a detailed understanding of the magnetoresistance fluctuations around integer filling factors in mesoscopic devices. These experiments also open the way to the spectroscopic investigation of each QH island by changing the voltage on the tip, and the bias through the QR. 3.6 Related scanning probe techniques for buried 2DEG investigations Finally, it is worth mentioning here two other cryogenic SPM techniques related to SGM, which have also proven very powerful in imaging buried electron systems, namely scanning single- electron transistor (SET) microscopy and scanning subsurface charge accumulation (SCA) microscopy. In scanning-SET microscopy [61,62], a SET is fabricated at the apex of a tapered optical fibre and used as a scanning electric probe sensitive to any potential change in the scanned device. This technique has enabled to image the distribution of charges and the way they localize in space in GaAs/GaAlAs heterostructures under various ground state conditions including the integer [62,63] and fractional [64] QHEs. It has recently been used to image electron-hole puddles in the vicinity of the neutrality point in graphene monolayers [65,66]. In scanning SCA microscopy [67-69], the sharp metallic tip of a scanning microscope is connected to a very sensitive charge detector that records the local ac-charge accumulation in the 2DEG in response to a small ac-voltage applied to the sample. When the tip scans above a conducting region the ac-current is significant, whereas it disappears above insulating, incompressible, or Coulomb blocked regions. This imaging technique has been applied on 2DEG in the quantum Hall regime, revealing a rich structure of filaments and droplets [67], random electrostatic potential fluctuations [68], and short length-scale scattering potentials [69]. 4. Some limitations of the SGM technique Based on our experience in the SGM imaging of QRs, we exemplify in this section a few limitations of the technique that can be helpful to understand its capabilities and that should be considered carefully during SGM experiments. 4.1 Tip-induced effects and artefacts in SGM Experimentally evaluating [33] or theoretically computing the tip potential seen by the electrons probed in SGM is a complex matter. But this is required for a proper understanding of the SGM images. For instance, the AB effects in QRs introduced earlier in this paper have indicated that the tip potential is strongly screened by the electron system [37,47]. Figure 5. Spatial extension as a function of the tip voltage of the tip-induced potential on the 2DEG present at the interface of a GaInAs/AlInAs heterostructure with a silicon doping layer of 1.7 x 1019 cm-3 in the barrier. Solving self-consistently the three-dimensional Schrödinger and Poisson equations that govern the properties of the electron system coupled to the tip potential can simulate this screening. This allows estimating the spatial extension of the tip-induced perturbation, which is a critical factor to ensure a correct interpretation of conductance images of QRs in terms of LDOS [36,37]. As an example, we show in figure 5 the dependence of the tip-induced spatial extension on the tip voltage for an SGM tip located 10 nm above the free surface of a GaInAs/AlInAs heterostructure. This spatial extension is defined as the length of the area occupied by a potential higher than a threshold value of 0.3 meV. The linear increase of the spatial extension at low tip-bias is followed by a power-law dependence due to the screening of electrons in the 2DEG. In figure 5, there is a distinctive non-zero spatial extension at zero tip voltage, or conversely a finite voltage must be applied to have a vanishing extension. This is due to the dielectric screening arising from fixed charges in the doping layer of the heterostructure and on the exposed surface. This screening is modelled as explained in [70]. On the experimental side, some caution must be paid to avoid tip-induced artefacts. A first artefact is related to possible distortions of the tip, which may happen after thorough scanning of the sample or inadvertent contact to the surface. It results in distortions of the SGM images, which can also be simulated (see figure 2 in [37]). For large distortions, it may even happen that the respective roles of the electron system and of the tip are reversed, as suggested by the SGM image in figure 6b. Here the tip has been damaged during scanning of the sample in such a way that the tip potential is imaged rather than the intrinsic properties of the electron system. Fortunately, a strong indication that the tip has been damaged and that the SGM data are consequently not reliable is given by the strongly distorted topographic image in figure 6a, where the ring geometry can hardly be recognized in contrast with topographic images acquired with good tips (see e.g. figure 1a in [35]). It may also happen that the tip is contaminated by particles that distort the induced potential. In this case a high-field tip treatment could be used to recover a smooth and symmetrical potential: a high voltage pulse is applied on the SGM tip in STM mode on a metallic electrode to eject the unwanted particles before SGM imaging of the nanostructure [71]. 4.2 Distortions due to random residual charges SGM images can be disturbed by the presence of parasitic electron traps coupled by tunnelling to the 2DEG of the semiconductor device under investigation. These traps may originate from impurities located in the spacer layer of the heterostructure very close to the 2DEG, closer than the doping layer whose charge transfer is blocked at low temperature. The time fluctuation of these electron traps is a frequent source of low-frequency noise in high mobility transistors. When the SGM tip scans over such an impurity, its charge state can change, and consequently, the potential landscape for the 2DEG electrons also changes, thereby affecting the conductance of the device. At low temperature, these traps are subjected to Coulomb blockade, and the number of electrons changes one by one in a discrete manner when the tip is approached. Since the charge state is the same when the tip is at a constant distance from the trap, SGM images display sets of concentric rings centred on each trap coupled to the 2DEG. The observed pattern changes with illumination (that acts as a charge reset) or after a new cooling down of the sample. This parasitic effect in SGM has been observed for instance in QPC experiments (see figure 3 in [28]) and in QD studies (see figure 2c in [29]). 4.3 On the imaging of the LDOS of a buried electron system STM in its spectroscopic mode – scanning tunnelling spectroscopy - is known to probe directly the electronic LDOS of a surface electron system, such as for example the quantum corral [49] or the inversion layer on an InSb surface in the QHE regime [72], which can be computed accurately [73]. Since SGM probes tip-induced changes in the conductance maps of a buried 2DEG, the connection of the collected information with the LDOS is indirect and some caution must be taken in interpreting the images. We have already mentioned this issue earlier in this paper (subsection 3.2). Here we wish to put more emphasis on this important point in the case of QRs. According to the analytical model presented in [37], in the single-channel transmission case a generalized Kramers–Kronig relation holds between the LDOS and the conductance variation due to the tip-induced potential. This is valid in the linear response regime where the local perturbation due to the tip is so weak that the energy levels of the system are unchanged. In the multi-channel transmission case, an exact correspondence between LDOS and conductance maps is not possible due to the deficient spatial resolution of the tip-induced potential and to the uneven effect of the tip on the different transverse modes involved in the transport. However, a noticeable correspondence is still achievable in some cases of interest for which the total wavefunction is determined only by the few states close to the Fermi energy and presents a large and uniform spatial extension. Examples are given by mesoscopic systems such as QRs where the LDOS is dominated by localized states due to randomly distributed impurities or by semi-classical periodic orbits inducing scarring effects in the total wavefunction. In addition to these fundamental limitations, special care must be taken to apply a small bias to the tip. Indeed, applying too high a voltage will obviously strongly couple the tip to the buried electron system and eventually modify its properties. This can be simulated without resorting to self-consistent simulations as done in [36] where it was found that for small tip potentials the conductance image essentially reflects the electronic LDOS in the QRs. However, applying too high a tip potential progressively introduces spurious features in the conductance image that are not present in the LDOS (see figure 4 in [36]). Experimentally, the low tip-potential regime is maintained as long as the magnitude of the imaged tip-induced features, such as the AB “inner fringes” mentioned earlier, increases linearly with the tip bias [36]. Beyond this regime, this magnitude tends to increase sub- linearly, or to saturate, and spurious features, not seen at low bias, appear. This regime must be avoided. Note however that this limitation of a low bias voltage applied to the tip holds essentially for the SGM experiments aiming at measuring the unperturbed electronic properties of the electron system such as the LDOS. It does not hold for measurements where the tip is used to scatter the electron waves as done, for instance, in the imaging of coherent electron flow through QPCs or related experiments [2,3,12-16]. 5. Conclusion The few examples summarized in this paper confirm that SGM is very powerful in imaging the electronic transport in various low-dimensional semiconductor devices and to reveal how electrons behave down there. It often gives valuable complementary view on phenomena that are usually considered within a macroscopic experimental scheme. The ability of locating precisely compressible Coulomb islands in a quantum Hall interferometer is illustrative of this claim. Although some attention must be paid to avoid possible artefacts, the broad applicability range of SGM makes it a powerful tool for the electron diagnose of nanodevices in the coherent regime of transport, or even in the quantum Hall regime. Therefore, more spectacular achievements can be expected in the future. Acknowledgments B. H. is FNRS research associate and F. M. is FNRS postdoctoral researcher. F. M. acknowledges a former postdoctoral grant with FCT - Portugal. This work has been supported by FRFC grant no. 2.4.546.08.F, and FNRS grant no 1.5.044.07.F, by the Belgian Science Policy (Interuniversity Attraction Pole Program IAP-6/42) as well as by the PNANO 2007 program of the Agence Nationale de la Recherche, France (“MICATEC” project). VB acknowledges the award of a “Chaire d’excellence” by the Nanoscience Foundation in Grenoble. References [1] Eriksson M A, Beck R G, Topinka M, Katine J A, Westervelt R M, Campman K L and Gossard A C 1996 Appl. Phys. Lett. 69 671 [2] Topinka M A, LeRoy B J, Shaw S E J, Heller E J, Westervelt R M, Maranowski K D and Gossard A C 2000 Science 289 2323 [3] Topinka M A, LeRoy B J, Westervelt R M, Shaw S E J, Fleischmann R, Heller E J, Maranowski K D and Gossard A C 2001 Nature 410 183 [4] Gildemeister A E, Ihn T, Barengo C, Studerus P, Ensslin K 2007 Rev. Sci. Instr. 78 013704 [5] Hackens B, Martins F, Faniel S, Dutu C A, Sellier H, Huant S, Pala M, Desplanque L, Wallart X and Bayot V 2010 Nat. Commun. 1 39 [6] Obermüller C, Deisenrieder A, Abstreiter G, Karrai K, Grosse S, Manus S, Feldmann J, Lipsanen H, Sopanen M and Ahopelto J 1999 Appl. Phys. Lett. 75 358 [7] Karrai K and Grober R D,Appl. Phys. Lett. 1995 66 1842 [8] Lang D V and Logan R A 1977 Phys. Rev. Lett. 39 635; Störmer H L, Dingle D, Gossard A C, Wiegmann W and Sturge M D 1979 Solid State Commu n. 29 705; Huant S, Mandray A, Martinez G and Etienne B 1997 Semicond. Sci. Technol. 12 813 [9] Brun M, Huant S, Woehl J C, Motte J F, Marsal L and Mariette H 2001 J. Microsc. 202 202 [10] Senzier J, Luo P S and Courtois H 2007 Appl. Phys. Lett. 90 043114 [11] van Wees B J, Kouwenhoven L P, van Houten H, Beenakker C W J, Mooij J E, Foxon C T and Harris J J 2008 Phys. Rev. Lett. 60 848 [12] LeRoy B J, Topinka M A, Westervelt R M, Maranowski K D and Gossard A C 2002 Appl. Phys. Lett. 80 4431 [13] LeRoy B J, Bleszynski A C, Aidala K E, Westervelt R M, Kalben A, Heller E J, Shaw S E J, Maranowski K D and Gossard A C 2005 Phys. Rev. Lett. 94 126801 [14] Jura M P, Topinka M A, Urban L, Yazdani A, Shtrikman H, Pfeiffer L N, West K W and Goldhaber-Gordon D 2007 Nature Phys. 3 841 [15] Jura M P, Topinka M A, Grobis M, Pfeiffer L N, West K W and Goldhaber-Gordon D 2009 Phys. Rev. B 80 041303(R) [16] Jura M P, Grobis M, Topinka M A, Pfeiffer L N, West K W and Goldhaber-Gordon D 2010 Phys. Rev. B 82 155328 [17] Crook R, Smith C G, Simmons M Y and Ritchie D A 2000 Phys. Rev. B 62 5174 [18] Aidala K E, Parrott R E, Kramer T, Heller E J, Westervelt R M, Hanson M P and Gossard A C 2007 Nature Phys. 3 464 [19] da Cunha C R, Aoki N, Morimoto T, Ochiai Y, Akis R and Ferry D K 2006 Appl. Phys. Lett. 89 242109 [20] Crook R, Graham A C, Smith C G, Farrer I, Beere H E and Ritchie D A 2003 Nature 424, 751 [21] Crook R, Prance J, Thomas K J, Chorley S J, Farrer I, Ritchie D A, Pepper M and Smith C G 2006 Science 312, 1359 [22] Metalidis G and Bruno P 2005 Phys. Rev. B 72 235304 [23] Cresti A 2006 J. Appl. Phys. 100 053711 [24] Freyn A, Kleftogiannis I and Pichard J L 2008 Phys. Rev. Lett. 100 226802 [25] Jalabert R A, Szewc W, Tomsovic S and Weinmann D 2010 Phys. Rev. Lett. 105 166802 [26] Bleszynski A C, Zwanenburg F A, Westervelt R M, Roest A L, Bakkers E P A M and Kouwenhoven L P 2007 Nano Lett. 7 2559 [27] Woodside M T and McEuen P L 2002 Science 296 1098 [28] Aoki N, Burke A, da Cunha C R, Akis R, Ferry D K and Ochiai Y 2006 J. Phys. Conf. Ser. 38 79 [29] Gildemeister A E, Ihn T, Schleser R, Ensslin K, Driscoll D C and Gossard A C 2007 J. Appl. Phys. 102 083703 [30] Pioda A, Kicin S, Brunner D, Ihn T, Sigrist M, Ensslin K, Reinwald M and Wegscheider W 2007 Phys. Rev. B 75 045433 [31] Fallahi P, Bleszynski A C, Westervelt R M, Huang J, Walls J D, Heller E J, Hanson M and Gossard A C 2005 Nano Lett. 5 223 [32] Pioda A, Kicin S, Ihn T, Sigrist M, Fuhrer A, Ensslin K, Weichselbaum A, Ulloa S E, Reinwald M and Wegscheider W 2004 Phys. Rev. Lett. 93 216801 [33] Gildemeister A E, Ihn T, Sigrist M, Ensslin K, Driscoll D C and Gossard A C 2007 Phys. Rev. B 75 195338 [34] Crook R, Smith C G, Graham A C, Farrer I, Beere H E and Ritchie D A 2003 Phys. Rev. Lett. 91 246803 [35] Hackens B, Martins F, Ouisse T, Sellier H, Bollaert S, Wallart X, Cappy A, Chevrier J, Bayot V and Huant S 2006 Nature Phys. 2 826 [36] Martins F, Hackens B, Pala M G, Ouisse T, Sellier H, Wallart X, Bollaert S, Cappy A, Chevrier J, Bayot V and Huant S 2007 Phys. Rev. Lett. 99 136807 [37] Pala M G, Hackens B, Martins F, Sellier H, Bayot V, Huant S and Ouisse T 2008 Phys. Rev. B 77 125310 [38] Pala M G, Baltazar S, Martins F, Hackens B, Sellier H, Ouisse T, Bayot V and Huant S 2009 Nanotechnology 20 264021 [39] Connolly M R, Chiou K L, Smith C G, Anderson D, Jones G A C, Lombardo A, Fasoli A and Ferrari A C 2010 Appl. Phys. Lett. 96 113501 [40] Berezovsky J, Borunda M F, Heller E J and Westervelt R M 2010 Nanotechnology 21 274013 [41] Berezovsky J and Westervelt R M 2010 Nanotechnology 21 274014 [42] Schnez S, Guttinger J, Huefner M, Stampfer C, Ensslin K and Ihn T 2010 Phys. Rev. B 82 165445 [43] Webb R A, Washburn S, Umbach C P and Laibowitz R B 1985 Phys. Rev. Lett. 54 2696 [44] Washburn S, Schmid H, Kern D and Webb R.A. 1987 Phys. Rev. Lett. 59 1791 [45] Note that strictly speaking, only scalar or vector potentials associated with the electric or magnetic fields are applied to the electron waves. In experiments on mesoscopic QRs, however, there is a common use of the term “Aharonov-Bohm effect” even though electric and magnetic fields are applied to electron waves. [46] Hackens B, Faniel S, Gustin C, Wallart X, Bollaert S, Cappy A and Bayot V 2005 Phys. Rev. Lett. 94, 146802 [47] Martins F, Hackens B, Sellier H, Liu P, Pala M G, Baltazar S, Desplanque L, Wallart X, Bayot V and Huant S, Acta Phys. Polon. A (to appear) [48] Najda S P, Armistead C J, Trager C and Stradling R A 1989 Semicond. Sci. Technol. 4 439; Huant S, Najda S P and Etienne B 1990 Phys. Rev. Lett. 65 1486; Huant S, Mandray A, Martinez G, Grynberg M and Etienne B 1992 Surf. Sci. 263 565 [49] Crommie M F, Lutz C P and Eigler D M 1993 Science 262 218 [50] Colas des Francs G, Girard C, Weeber J C, Chicane C, David T, Dereux A and Peyrade D 2001 Phys. Rev. Lett. 86 4950 [51] Cuche C, Drezet A, Sonnefraud Y, Faklaris O, Treussart F, Roch J F and Huant S 2009 Opt. Express 17 19969 [52] Cuche A, Mollet O, Drezet A and Huant S 2010 Nano Lett. 10 4566 [53] Aoki N, da Cunha C R, Akis R, Ferry D K and Ochiai Y 2005 Phys. Rev. B 72 155327 [54] Baumgartner A, Ihn T, Ensslin K, Maranowski K and Gossard A C 2007 Phys. Rev. B 76 085316 [55] Woodside M T, Vale C, McEuen P L, Kadow C, Maranowski K D and Gossard A C 2001 Phys. Rev. B 64 041310(R) [56] Halperin B I 1982 Phys. Rev. B 25 2185 [57] van Wees B J, Kouwenhoven L P, Harmans C J P M, Williamson J G, Timmering C E, Broekaart M E I, Foxon C T and Harris J J 1989 Phys. Rev. Lett. 62 2523 [58] Zhang Y, McClure D T, Levenson-Falk E M, Marcus C M, Pfeiffer L N and West K W 2009 Phys. Rev. B 79 241304 [59] Giesbers A J M, Zeitler U, Katsnelson M I, Reuter D, Wieck A D, Biasiol G, Sorb L and Maan J C 2010 Nature Phys. 6 173 [60] Rosenow B and Halperin B I 2007 Phys. Rev. Lett. 98 106801 [61] Yoo M J, Fulton T A, Hess H F, Willett R L, Dunkleberger L N, Chichester R J, Pfeiffer L N and West K W 1997 Science 276 579 [62] Yacoby A, Hess H F, Fulton T A, Pfeiffer L N and West K W 1999 Solid State Commun. 111 1 [63] Zhitenev N B, Fulton T A, Yacoby A, Hess H F, Pfeiffer L N and West K W 2000 Nature 404 473; Ilani S, Martin J, Teitelbaum E, Smet J H, Mahalu D, Umansky V and Yacoby A 2004 Nature 427 328 [64] Martin J, Ilani S, Verdene B, Smet J, Umansky V, Mahalu D, Schuh D, Abstreiter G and Yacoby A 2004 Science 305 980 [65] Martin J, Akerman N, Ulbricht G, Lohmann T, Smet J H, von Klitzing K and Yacoby A 2008 Nature Phys. 4 144 [66] Martin J, Akerman N, Ulbricht G, Lohmann T, von Klitzing K, Smet J H and Yacoby A 2009 Nature Phys. 5 669 [67] Tessmer S H, Glicofridis P I, Ashoori R C, Levitov L S and Melloch M R 1998 Nature 392 51 [68] Finkelstein G, Glicofridis P I, Ashoori R C and Shayegan M 2000 Science 289 90 [69] Steele G A, Ashoori R C, Pfeiffer L N and West K W 2005 Phys. Rev. Lett. 95 136804 [70] Pala M G, Iannaccone G, Kaiser S, Schliemann A, Worschech L and Forchel A 2002 Nanotechnology 13 373 [71] Gildemeister A E, Ihn T, Sigrist M, Ensslin K, Driscoll D C and Gossard A C 2007 Appl. Phys. Lett. 90 213113 [72] Hashimoto K, Sohrmann C, Wiebe J, Inaoka T, Meier F, Hirayama Y, Romer R A, Wiesendanger R and Morgenstern M 2008 Phys. Rev. Lett. 101 256802 [73] Champel T and Florens S 2009 Phys. Rev. B 80 161311(R)
1712.02751	2	1712	2018-08-08T19:02:51	Electrical Reservoirs for Bilayer Excitons	[ "cond-mat.mes-hall" ]	The ground state of two-dimensional (2D) electron systems with equal low densities of electrons and holes in nearby layers is an exciton fluid. We show that a reservoir for excitons can be established by contacting the two layers separately and maintaining the chemical potential difference at a value less than the spatially indirect band gap. Equilibration between the exciton fluid and the contacts proceeds via a process involving virtual intermediate states in which an unpaired electron or hole occupies a free carrier state in one of the 2D layers. We derive an approximate relationship between the exciton-contact equilibration rate and the electrical conductances between the contacts and individual 2D layers when the contact chemical potentials align with the free-carrier bands, and explain how electrical measurements can be used to measure thermodynamic properties of the exciton fluid.	cond-mat.mes-hall	cond-mat	Electrical Reservoirs for Bilayer Excitons Department of Physics, The University of Texas at Austin, Austin, TX 78712, USA Ming Xie and A. H. MacDonald (Dated: August 10, 2018) The ground state of two-dimensional (2D) electron systems with equal low densities of electrons and holes in nearby layers is an exciton fluid. We show that a reservoir for excitons can be established by contacting the two layers separately and maintaining the chemical potential difference at a value less than the spatially indirect band gap, thereby avoiding the presence of free carriers in either layer. Equilibration between the exciton fluid and the contacts proceeds via a process involving virtual intermediate states in which an unpaired electron or hole virtually occupies a free carrier state in one of the 2D layers. We derive an approximate relationship between the exciton-contact equilibration rate and the electrical conductances between the contacts and individual 2D layers when the contact chemical potentials align with the free-carrier bands, and explain how electrical measurements can be used to measure thermodynamic properties of the exciton fluids. PACS numbers: 71.35.-y, 73.21.-b Introduction. -- Excitons are composite bosonic parti- cles in which conduction band electrons bind with va- lence band holes. Excitons normally exist as excited states of semiconductors and insulators, and can have extremely long lifetimes when the electron and hole are separated in momentum-space, or in real-space [1], or both. Bose-Einstein condensation of long-lived excitons was predicted several decades ago [2], and is thought to have been realized relatively recently in semiconductor quantum well [3 -- 7] double-layers. Closely related polari- ton condensate states, in which longer range coherence is assisted by the small masses of 2D vertical cavity pho- tons, are regularly realized and have been studied exten- sively over the past decade [8 -- 18]. In typical exciton- condensation experiments a population of electrons and holes is generated in nearby 2D layers by optical excita- tion. Free electrons and holes can also be injected elec- trically if contacts can be established to conduction and valence bands [19 -- 22]. The electrons and holes then com- bine to form excitons and the excitonic state is revealed by photons emitted during the exciton radiative decay process [23 -- 25]. In this paper we propose and the- oretically analyze a mechanism that allows direct elec- trical control of the chemical potential of spatially indi- rect exciton fluids without populating free electron and hole states. The mechanism requires substantial exci- ton binding energies in systems with long electron-hole recombination times. Our proposal is motivated by the properties of van der Waals (vdW) heterojunction sys- tems in which single-layer semiconductors are separated by hexagonal boron nitride (hBN) barrier layers. In recent years, 2D transition metal dichalcogenide (TMD) semiconductors have been established as an ex- citing exciton physics platform [26 -- 28] in which energy scales are enhanced by strong Coulomb interactions. Sur- prising flexibility in the design of optical and electronic properties can be achieved [29 -- 31] by stacking vdW cou- pled 2D materials in a variety of different arrangements. FIG. 1. (Color online) Schematic illustration of a 2D mate- rial heterojunction capable of supporting a spatially indirect exciton condensate, and of an electrode pair that can act as a reservoir for spatially indirect excitons. vdW heterojunctions involving 2D semiconductors can host spatially indirect excitons formed from electrons and holes in two different layers with binding energies that remain large, even when the electron and hole layers are separated by hBN layers that increase exciton lifetimes by orders of magnitude from the nanosecond range [32 -- 34] that applies in the absence of spacer layers. When the chemical potentials of contact electrodes are inside the energy gaps of the 2D semiconductors, elec- trons cannot tunnel into double-layer band states. How- ever, because of the Coulomb interaction and the exciton binding energy that it produces, correlated pair tunnel- ing from electrodes connected to the two different lay- ers is possible. In this Letter, we develop a microscopic model of this two-particle tunneling process and argue that it can allow electrode pairs to act directly as exciton reservoirs with a well-defined chemical potential set by the source-to-drain bias. Direct exciton reservoirs have advantages for exciton generation and control over the commonly employed indirect optical and electrical gen- eration processes that start by generating free electrons and holes, and we expect in particular that they will en- able electrical measurements of the transport properties of exciton fluids. Correlated pair tunneling. -- We consider the vertically stacked multilayer heterostructure system illustrated in VbGraphene(S)TMD1(T)TMD2(B)Graphene(D)hBNhBNhBNdSdDLdD-+-+-+-+-+2 p, the band index λ = c, v and the valley index τ of the graphene electrode states. In Eq. (2), a is the creation † operator in the electrodes and c T (B) is the creation oper- ator for conduction band electrons in T and valence band electrons in B. For single-grain hBN tunnel barriers, the tunneling properties can have very specific momentum dependence which does not play an essential role and is not accounted for below, but is sensitive to the relative orientation of the various 2D material layers [37, 38]. We neglect interlayer tunneling between the T and B primar- ily because we are interested in a bias voltage regime in which free carriers are not present to tunnel. We also set the interlayer radiative recombination rate to zero in order to focus on double-electrode reservoir properties. In practice we anticipate that the interplay between our exciton reservoirs and interlayer radiative recombination, whose strength can be adjusted over orders of magnitude by varying the thickness and orientation of the hBN bar- rier layer, opens up a rich range of opto-electronic phe- nomena for study that are a primary motivation for this work. The band diagram of the vertical vdW heterostruc- ture system is shown schematically in Fig. 2. At zero bias (Fig. 2(a)), we assume that both graphene electrodes are neutral and that the aligned Dirac points are in the middle of the spatially indirect band gap Eg. When a bias voltage in the subgap regime (Eg > eVb > 0) is ap- plied, tunneling between the electrodes and free-carrier states in the TMD layers is prohibited by energy conser- vation. (Note that Eg increases with Vb, and that direct tunneling of electrons from source to drain is extremely strongly suppressed because it must navigate three tun- neling barriers.) Our interest here is in the bias regime ex, where µ0 Eg > Vb > µ0 ex is the energy of an isolated spatially indirect exciton. In this bias voltage regime energy conservation can be achieved by the two-particle tunneling process illustrated in Fig. 2(b). The state cre- ated when an electron from S tunnels to a virtual state in T and a hole from D subsequently tunnels to B has a finite overlap with an exciton fluid state (path 1). An alternative and equally possible path is for a hole from D to tunnel to a virtual state in B first (path 2). Each tunneling process effectively transfers one electron from S to D and creates an exciton in the DL. We concen- trate here on the case of T < Tc for which the excitons form a condensate, although the main idea of using an electrode pair as a reservoir for excitons applies equally well when the excitons form a non-condensed gas. For the low temperature case we find that because of the stimulated scattering characteristic of bosonic statistics, a major fraction of the excitons added or removed from the system are simply added or removed from the con- densate. The condensate state of spatially indirect excitons has been extensively studied in previous work [39 -- 41] using a FIG. 2. (Color online) Schematic band diagrams for the vertical vdW heterostructure systems of interest for the case of (a) zero applied bias and (b) a finite bias Vb satisfying indirect gap Eg > eVb > µ0 ex is the energy of an isolated spatially indirect exciton. In case (b), the filled (empty) black circles represents electrons (holes) and the gray (pink) circle represents a virtual state in the T(B) layer. 1 and 2 label two possible tunneling paths as discussed in detail in the text. ex where µ0 Fig. 1, which contains a TMD semiconductor double- layer (DL) with a hBN barrier layer sufficiently thick to suppress tunneling, and source (S) and drain (D) elec- trodes that contact the two layers separately. For the sake of definiteness we have assumed that the electrodes are formed from graphene sheets instead of metals since these have less influence on exciton binding energies [35], but this detail is inessential. Dynamic screening due to source-drain electron-hole pair excitations is negligible because of the suppression of source-to-drain tunneling by the tunnel barrier. We also assume that the top (T) layer and bottom (B) layer materials are chosen so that the conduction band minimum and valence band maxi- mum are respectively above and below but close to the graphene sheet Dirac point (for example, T=MoS2 and B=WSe2 [36]) as illustrated in Fig. 2(a). Once the bias voltage between S and D, µS − µD = eVb, exceeds the energy needed to create an isolated indirect exciton, an exciton fluid will form whose equilibrium chemical poten- tial equals eVb, where e > 0. We discuss the equilibration process below. The total Hamiltonian of the four-layer system is H = HS + HD + HDL + Ht. (1) where HS and HD are the linear band Hamiltonians of the graphene electrodes, and HDL is the DL Hamiltonian including Coulomb interactions. In this paper we assume that the TMD DL is in its exciton condensate ground state and ignore its spin degree of freedom. Tunneling between the electrodes and the double-layer system is accounted for by (cid:88) Ht = † tS k,T a ¯p,S + tD k ¯p c † k ¯p c k,Ba ¯p,D + h.c. (2) k, ¯p is a tunneling matrix element. ¯p ≡ (p, λ, τ ) where tS(D) is a compound index that combines the 2D momentum k ¯p Eg2Eg2STBDµSµD(a)(b)STBDµSµDµSeVb1122duction and valence bands. Here N = (cid:80) BCS-like mean field theory approach in which the ground state is found by minimizing (cid:104) HDL − µex N(cid:105) in the space of Slater determinant states with coherence between con- † k(c k,T ck,T + † ck,B c k,B)/2 is the total number of electron-hole pairs. The mean-field Hamiltonian of the exciton condensate system is HM F DL = EG + † k,1γk,1) + µex N (3) † k,0γk,0 − γ (cid:88) Ek(γ k (cid:113) k − µex/2)2 + ∆2 (T where EG is the condensate ground state energy and µex the exciton chemical potential. The quasiparticle energy k, where we have dispersion is Ek = k = −B assumed that the DL energy dispersion T k = 2k2/(2m) + Eg/2 and that the order parameter ∆k is † † † k,T − ukc real. γ k,1 = vkc k,B are creation operators for states in the empty and oc- cupied dressed quasiparticle bands, respectively, and uk and vk are coherence factors that depend on the pair potential ∆k which is determined in turn by solving a self-consistent equation that has solutions only if µex ex- ceeds µ0 † † † k,T + vkc k,0 = ukc k,B and γ ex [39]. The two-particle tunneling rate can be obtained by ap- plying Fermi's golden rule to the second order tunneling process. We find that the net rate at which excitons are added to the condensate is M ¯p ¯p(cid:48)2 (f S ¯p(cid:48) − µex) (4) ¯p − f D ¯p − D (cid:88) ¯p(cid:48) )δ(S = dnex dt 2π A ¯p, ¯p(cid:48) where f α tion. The matrix element in Eq. (4) ¯p , with α = S, D, is the Fermi distribution func- 1 ¯p(cid:48) − E1 D k (5) 1 ¯p + k − S E0 k = −E0 k = Ek + µex/2 and E1 where E0 k are the energies required to add quasiparticles of momentum k to bands 0 and 1, respectively. The energy denominators account for the finite energy cost of hopping to the intermediate virtual states, and never vanish in the bias voltage range of interest. The two terms in the matrix element account for the two tunneling paths depicted in Fig. 2(b). The evaluation of M ¯p ¯p(cid:48)2 in Eq. (5) requires knowl- edge of the momentum dependent tunneling amplitudes tα k ¯p. We simplify our calculation by assuming that in- terfacial disorder plays an important role in determining the tunneling amplitude. We employ a Gaussian random tunneling model for which (cid:104)tα(r)(cid:105)dis = 0 and the second order correlation functions satisfy (cid:104)tα(r)tα(cid:48)∗(r (cid:48))(cid:105)dis = ∆t2F(r − r (cid:48))δαα(cid:48), (6) where (cid:104)···(cid:105)dis is the disorder average and F(r − r(cid:48)) is a smoothly decaying function of the distance r − r(cid:48). For (cid:88) k M ¯p ¯p(cid:48) = ukvktS k ¯ptD∗ k ¯p(cid:48) (cid:40) (cid:41) 3 FIG. 3. Equilibrium electrode densities (a) and exciton den- sities (b) at different values of β = gXC /gH , where gH is fixed with its value set by choosing dDL = 1nm. These re- sults were calculated for dS = dD = dDL spatially indirect gap Eg(0) = 1.1eV , and exciton binding energy Eb = 0.2eV . The extended (colored) dashed line represents electrode den- sities in the case in which the TMD double-layer that hosts excitons is absent. The vertical gray dashed lines indicates the threshold voltage at which the dual electrodes establish a reservoir for excitons. Inset: Threshold voltage as a func- tion of dS/dDL. The dashed line indicates the zero electrode density limit of the isolated exciton energy µ0 ex. low exciton densities and Vb > µ0 tunneling current-voltage equation ex limit, we obtain the Iex ≈ Gex(Vb − µex/e). (7) (µex > µ0 ex at finite exciton density.) The effective ex- citon tunneling conductance Gex is given approximately by Gex = AgS N nexa2 N gD B e2/ 8 ρ0Eb , (8) N and gD where gS N are the normal tunneling conductances per unit area between S and T and between D and B, respectively, ρ0 is the density of states of quasiparticle band 0, aB is the Bohr radius, and Eb the exciton binding energy. The tunneling conductance in Eq. (8) is propor- tional to the exciton condensate density nex because of the bosonic stimulated scattering effect. Since the frac- tion of uncondensed excitons is small in the low den- sity BEC limit, we have assumed in deriving this simple result that the contributions from processes with a fi- nal state exciton outside the condensate are negligible. N ∼ 10−2e2/h · µm−2, Using the typical values gS B ∼ 0.01, and taking the quasiparticle band masses nexa2 in the TMD layers close to the bare electron mass, we estimate that Gex is in the order 10−11e2/h · µm−2. N = gD (7) Eq. states that a time-independent quasi- equilibrium is reached when eVb = µex. We say quasi- equilibrium here, rather than simply equilibrium, to em- phasize that we are assuming that excitons cannot an- nihilate by radiative recombination. As long as all pro- cesses in which electrons move between the two TMD lay- ers are absent, we effectively have an equilibrium problem 02468−2nS(1012cm−2)00.51.01.52.0Vb(V)Vthβ=0.20−0.2−0.4−0.6(a)01020304050nex(1012cm−2)00.51.01.52.0Vb(V)Vthβ=−0.6−0.4−0.200.2(b)0.81.01.2µ0exVth(V)024dS/dDL4 the same chemical potential in either environment i.e. when and eVb = µex. Fig. 3 shows equilibrium densities calculated for several typical values of the dimensionless exchange-correlation coupling strength β ≡ gXC/gH . Below we take β as an unknown parameter and show that its value can be mea- sured electrically. When estimated using self-consistent mean-field theory β changes sign from positive to neg- ative when dDL exceeds around a quarter of an exci- tonic Bohr radius, and using TMD semiconductor pa- rameters has the value β = −0.6 for dDL = 1nm in [41]. Below the threshold voltage Vth, which satisfies eVth = µ0 ex + gH nS(Vth) and depends on dS/dDL, no excitons are injected and nS(Vb) is independent of β, as shown in Fig. 3(a). When Vb > Vth, electrons and holes can enter the TMD layers by forming excitons via the two-particle tunneling process. The slope of the nS(Vb) curve is reduced and becomes negative when β changes sign from positive to negative. For β < 0, we find that nS becomes negative when Vb = −µ0 ex/β; we show later that the dynamic response is anomalous at this point. √ [42]). √ 2nS/( low = a0CDL, C ef f The two-particle tunneling rate which we have esti- mated theoretically can be measured by performing ac electrical measurements, letting Vb(t) = Vdc + Vac cos ωt, where Vac is small. The linear response of the sys- tem to Vac can be extracted by linearizing Eq. (7) and (10) (see details in Supplemental Material In Fig. 4 we plot normalized amplitudes of the differen- tial conductance dI/dV / ¯Gex, where ¯Gex(Vdc) is the dc exciton conductance Gex. In the low and high fre- quency limits, the system behaves effectively as a ca- pacitor with C ef f high = a∞CDL, where γ = (1 + 2Cgeo/CQ)dtot/dDL and a0 = {(γ − 1)2/[(1 + β)γ − 1] + 1}/γ and a∞ = 1/γ. CQ = πv) is the quantum capacitance of graphene [43]. dI/dV devi- ates from linear frequency dependence when the scaled frequency x = ωCDL/ ¯Gex ∼ x0 = [(1 + β)γ − 1]/γ. Measuring the crossover frequency ω0, gives the tunnel- ing conductance ¯Gex = ω0CDL/x0. The dc bias voltage dependence of the differential conductance is shown in Fig. 4(b). For β < 0 a peak appears at Vpeak = −E0 ex/β, the point at which nS reaches 0, as mentioned above. This suggests a way to measure the exciton exchange- correlation energy parameter gXC, provided that Eg and Eb are known. For β > 0, the differential conductance in- creases slowly with increasing Vb in the regime Vb > Vth. Optical recombination, which provides a mechanism for excitons to leak out of the system of interest, can eas- ily be added to the theory explained here, and in the dc bias voltage case converts the equilibrium exciton fluid into a steady state. When the steady state exciton fluid condenses, it will emit coherent light forming a state sim- ilar to a polariton laser but subject to precise electrical control. This work was supported by the Army Research Of- fice under Award W911NF-17-1-0312 and by the Welch FIG. 4. (Color online) Differential conductance as a function of (a) x = ωCDL/ ¯Gex and (b) bias voltage Vb. The dashed line and the dash-dotted line in (a) shows the linear relation at low and high frequency limits, respectively. in which the spatially indirect band gap is tuned electri- cally by varying Vb. We do not emphasize this distinction between equilibrium and quasi-equilibrium below. Electrical characteristics of exciton reservoirs. -- Because of repulsive interactions between excitons, their chemical potential increases with exciton density [39 -- 41]. For spatially indirect excitons µex = µ0 ex+(gH +gXC)nex, where gH = e2/CDL = /(4πdDL) and gXC are the Hartree and exchange-correlation contributions to the ef- fective exciton-exciton interaction. The Hartree term ac- counts for the capacitive coupling between T and B lay- ers, whereas gXC, which is density-dependent and cannot be evaluated exactly, accounts for exchange and corre- lation effects contributions due to both intralayer and interlayer Coulombic interactions and therefore also de- pend on the interlayer spacing dDL. When we add the potential energy associated with charged electrodes, the exciton chemical potential in our geometry has an addi- tional electrostatic contribution: µex = µ0 ex + (gH + gXC)nex + gH nS (9) where nS = nD are the equal densities of electrons and holes in the two electrodes. Eq. (9) can be obtained by minimizing the total energy, E[nS, nex], with respect to nex (see Supplemental Material [42]). By minimiz- ing E[nS, nex] with respect to nS we obtain the following expression for the bias potential, (cid:32) √ 2v e2(cid:112)nS + 2π (cid:33) nex CDL + nS 1 Cgeo = Vb e . (10) In Eq. (10), Cgeo = /(4πe2dtot) is the geometric capac- itance and dtot = dS + dDL + dD; the energy function used to derive Eq. (10) does not account for exchange and correlation in the electrodes, i.e. for interaction cor- rections to the electrode quantum capacitance, but these can easily be added when relevant. A time-independent equilibrium between the electrodes and the exciton fluid is established when nex (cid:54)= 0 and electron-hole pairs have 00.51.01.52.0dI/dV/¯Gex02468x=ωCDL/¯Gexy=a0xy=a∞xx=[(1+β)γ−1]/γ(a)00.20.40.60.81.0dI/dV/¯Gex00.51.01.52.02.5Vb(V)VthVpeakβ=−0.6β=0.2(b)Foundation under grant TBF1473. Ming Xie was sup- ported by The Center for Dynamics and Control of Ma- terials (CDCM) under NSF Award DMR-1720595. The authors acknowledge helpful discussions with Hui Deng, Jim Eisenstein, Emanuel Tutuc, and David Snoke. [1] Y. Lozovik and V. Yudson, JETP Lett. 22, 274 (1976). [2] L. V. Keldysh and A. N. Kozlov, Sov. Phys. JETP 27, 521 (1968). [3] L. V. Butov, A. L. Ivanov, A. Imamoglu, P. B. Little- wood, A. A. Shashkin, V. T. Dolgopolov, K. L. Camp- man, and A. C. Gossard, Phys. Rev. Lett. 86, 5608 (2001). [4] L. V. Butov, C. W. Lai, A. L. Ivanov, A. C. Gossard, and D. S. Chemla, Nature 417, 47 (2002). [5] A. V. Gorbunov and V. B. Timofeev, JETP Lett. 83, 146 (2006). [6] A. A. High, J. R. Leonard, A. T. Hammack, M. M. Fogler, L. V. Butov, A. V. Kavokin, K. L. Campman, and A. C. Gossard, Nature 483, 584 (2012). [7] A. A. High, J. R. Leonard, M. Remeika, L. V. Butov, M. Hanson, and A. C. Gossard, Nano Lett. 12, 2605 (2012). [8] H. Deng, G. Weihs, C. Santori, J. Bloch, and Y. Ya- mamoto, Science 298, 199 (2002). [9] J. Kasprzak, M. Richard, S. Kundermann, A. Baas, P. Jeambrun, J. M. J. Keeling, F. M. Marchetti, M. H. Szymanska, R. Andre, J. L. Staehli, V. Savona, P. B. Lit- tlewood, B. Deveaud, and L. S. Dang, Nature 443, 409 (2006). [10] R. Balili, V. Hartwell, D. Snoke, L. Pfeiffer, and K. West, Science 316, 1007 (2007). [11] J. J. Baumberg, A. V. Kavokin, S. Christopoulos, A. J. D. Grundy, R. Butt´e, G. Christmann, D. D. Solnyshkov, G. Malpuech, G. Baldassarri Hoger von Hogersthal, E. Feltin, J.-F. Carlin, and N. Grandjean, Phys. Rev. Lett. 101, 136409 (2008). [12] E. Wertz, L. Ferrier, D. D. Solnyshkov, R. Johne, D. San- vitto, A. Lemaitre, I. Sagnes, R. Grousson, A. V. Ka- vokin, P. Senellart, G. Malpuech, and J. Bloch, Nat Phys 6, 860 (2010). [13] D. Snoke, Science 298, 1368 (2002). [14] H. Deng, H. Haug, and Y. Yamamoto, Rev. Mod. Phys. 82, 1489 (2010). [15] I. Carusotto and C. Ciuti, Rev. Mod. Phys. 85, 299 (2013). [16] T. Byrnes, N. Y. Kim, and Y. Yamamoto, Nat Phys 10, 803 (2014). [17] D. Sanvitto and S. Kena-Cohen, Nat Mater 15, 1061 (2016). [18] Y. Sun, P. Wen, Y. Yoon, G. Liu, M. Steger, L. N. Pfeif- fer, K. West, D. W. Snoke, and K. A. Nelson, Phys. Rev. Lett. 118, 016602 (2017). [19] C. Schneider, A. Rahimi-Iman, N. Y. Kim, J. Fis- cher, I. G. Savenko, M. Amthor, M. Lermer, A. Wolf, 5 L. Worschech, V. D. Kulakovskii, I. A. Shelykh, M. Kamp, S. Reitzenstein, A. Forchel, Y. Yamamoto, and S. Hofling, Nature 497, 348 (2013). [20] P. Bhattacharya, B. Xiao, A. Das, S. Bhowmick, and J. Heo, Phys. Rev. Lett. 110, 206403 (2013). [21] P. Bhattacharya, T. Frost, S. Deshpande, M. Z. Baten, A. Hazari, and A. Das, Phys. Rev. Lett. 112, 236802 (2014). [22] Y. Yao, A. J. Hoffman, and C. F. Gmachl, Nat. Photon- ics 6, 432 (2012). [23] For systems in which dark excitons (excitons with mo- mentum or spin quantum numbers that don't match those of the photons) are favored however, direct opti- cal access is blocked. [24] K. Cohen, Y. Shilo, K. West, L. Pfeiffer, and R. Rapa- port, Nano Lett. 16, 3726 (2016). [25] Y. Shilo, K. Cohen, B. Laikhtman, K. West, L. Pfeiffer, and R. Rapaport, Nat. Comm. 4, 2335 EP (2013). [26] D. Y. Qiu, F. H. da Jornada, and S. G. Louie, Phys. Rev. Lett. 111, 216805 (2013). [27] K. F. Mak, K. He, C. Lee, G. H. Lee, J. Hone, T. F. Heinz, and J. Shan, Nat. Mater. 12, 207 (2013). [28] M. M. Fogler, L. V. Butov, and K. S. Novoselov, Nat. Comm. 5, 4555 EP (2014). [29] A. K. Geim and I. V. Grigorieva, Nature 499, 419 (2013). [30] K. S. Novoselov, A. Mishchenko, A. Carvalho, and A. H. Castro Neto, Science 353, aac9439 (2016). [31] K. F. Mak and J. Shan, Nat. Photonics 10, 216 (2016). [32] P. Rivera, J. R. Schaibley, A. M. Jones, J. S. Ross, S. Wu, G. Aivazian, P. Klement, K. Seyler, G. Clark, N. J. Ghimire, J. Yan, D. G. Mandrus, W. Yao, and X. Xu, Nat. Commun. 6, 6242 (2015). [33] M. Palummo, M. Bernardi, and J. C. Grossman, Nano Lett. 15, 2794 (2015). [34] E. V. Calman, M. M. Fogler, L. V. Butov, S. Hu, A. Mishchenko, and A. K. Geim, arXiv:1709.07043. [35] A. Raja, A. Chaves, J. Yu, G. Arefe, H. M. Hill, A. F. Rigosi, T. C. Berkelbach, P. Nagler, C. Schuller, T. Korn, C. Nuckolls, J. Hone, L. E. Brus, T. F. Heinz, D. R. Reichman, and A. Chernikov, Nat. Comm. 8, 15251 EP (2017). [36] C. Zhang, C. Gong, Y. Nie, K.-A. Min, C. Liang, Y. J. Oh, H. Zhang, W. Wang, S. Hong, L. Colombo, R. M. Wallace, and K. Cho, 2D Materials 4, 015026 (2017). [37] R. Bistritzer and A. H. MacDonald, Phys. Rev. B 81, 245412 (2010). [38] K. Zhou, D. Wickramaratne, S. Ge, S. Su, A. De, and R. K. Lake, Phys. Chem. Chem. Phys. 19, 10406 (2017). [39] C. Comte and P. Nozires, J. Phys. France 43, 1069 (1982). [40] X. Zhu, P. B. Littlewood, M. S. Hybertsen, and T. M. Rice, Phys. Rev. Lett. 74, 1633 (1995). [41] F.-C. Wu, F. Xue, and A. H. MacDonald, Phys. Rev. B 92, 165121 (2015). [42] See Supplemental Material for the discussion of the to- tal energy functional E[nS, nex], the existence of a global minimum and a detailed derivation of the ac response equations. [43] J. Xia, F. Chen, J. Li, and N. Tao, Nat. Nanotechnol. 4, 505 (2009).
1807.01652	2	1807	2019-01-08T04:01:44	Gate controlled large resistance switching driven by charge density wave in 1T-TaS2/2H-MoS2 heterojunction	[ "cond-mat.mes-hall", "cond-mat.mtrl-sci" ]	1T-TaS2 is a layered material that exhibits charge density wave (CDW) induced distinct electrical resistivity phases and has attracted a lot of attention for interesting device applications. However, such resistivity switching effects are often weak, and cannot be modulated by an external gate voltage - limiting their widespread usage. Using a back-gated 1T-TaS2/2H-MoS2 heterojunction, here we show that the usual resistivity switching in TaS2 due to different phase transitions is accompanied with a surprisingly strong modulation in the Schottky barrier height (SBH) at the TaS2/MoS2 interface - providing an additional knob to control the degree of the phase-transition-driven resistivity switching by an external gate voltage. In particular, the commensurate (C) to triclinic (T) phase transition results in an increase in the SBH owing to a collapse of the Mott gap in TaS2. The change in SBH allows us to estimate an electrical Mott gap opening of ~71 +/- 7 meV in the C phase of TaS2. On the other hand, the nearly-commensurate (NC) to incommensurate (IC) phase transition results in a suppression in the SBH, and the heterojunction shows a gate-controlled resistivity switching up to 17.3, which is ~14.5 times higher than that of standalone TaS2. The findings mark an important step forward showing a promising pathway to externally control as well as amplify the CDW induced resistivity switching. This will boost device applications that exploit these phase transitions, such as ultra-broadband photodetection, negative differential conductance, fast oscillator and threshold switching in neuromorphic circuits.	cond-mat.mes-hall	cond-mat	Gate controlled large resistance switching driven by charge density wave in 1T-TaS2/2H-MoS2 heterojunction Mehak Mahajan, Krishna Murali, Nikhil Kawatra, and Kausik Majumdar* Department of Electrical Communication Engineering, Indian Institute of Science, Bangalore 560012, India These authors contributed equally, Corresponding author, email: [email protected] ABSTRACT: 1T-TaS2 is a layered material that exhibits charge density wave (CDW) induced distinct electrical resistivity phases and has attracted a lot of attention for interesting device applications. However, such resistivity switching effects are often weak, and cannot be modulated by an external gate voltage -- limiting their widespread usage. Using a back-gated 1T-TaS2/2H- MoS2 heterojunction, here we show that the usual resistivity switching in TaS2 due to different phase transitions is accompanied with a surprisingly strong modulation in the Schottky barrier height (SBH) at the TaS2/MoS2 interface -- providing an additional knob to control the degree of the phase-transition-driven resistivity switching by an external gate voltage. In particular, the commensurate (C) to triclinic (T) phase transition results in an increase in the SBH owing to a collapse of the Mott gap in TaS2. The change in SBH allows us to estimate an electrical Mott gap opening of ~71 ± 7 meV in the C phase of TaS2. On the other hand, the nearly-commensurate (NC) to incommensurate (IC) phase transition results in a suppression in the SBH, and the heterojunction shows a gate-controlled resistivity switching up to 17.3, which is ~14.5 times higher than that of standalone TaS2. The findings mark an important step forward showing a promising pathway to externally control as well as amplify the CDW induced resistivity switching. This will boost device applications that exploit these phase transitions, such as ultra-broadband photodetection, negative differential conductance, fast oscillator and threshold switching in neuromorphic circuits. Keywords: Charge density wave (CDW), Transition metal dichalcogenide, 1T-TaS2, MoS2, Mott transition, Schottky barrier height. 1 Transition metal dichalcogenides are emerging out as pertinent materials for applications in high performance flexible electronics and optoelectronics1,2,3. 1T-TaS2 is a distinct layered material that exhibits multiple conductivity phases resulting from strong electron-phonon and electron-electron interactions. This material hosts a wide variety of charge density wave (CDW)4,5 states and the CDW amplitude is significant in 1T-polytype as compared to 2H-polytype of TaS2 6,7. Depending on the temperature of the 1T-TaS2 crystal, CDW exists in different phases subject to the alignment with the underlying lattice6,8,9,10. As the temperature is reduced below 550 K, the metallic crystal undergoes a CDW phase transition, however the CDW remains incommensurate (IC) with the underlying crystal lattice. On further cooling, it undergoes IC to nearly-commensurate (NC) CDW phase transition at 340 K, and an NC to commensurate (C) phase transition at 180 K. The NC to C transition is accompanied with a Mott transition resulting in strong suppression of conductivity. On heating, the crystal undergoes a new phase transition, from C phase to the triclinic (T) phase appearing at 223 K11,12,13, followed by a T to NC, and NC to IC phase transitions at 283 K and 353 K, respectively. The different CDW phase transitions can also be controlled by pressure8,14, doping15,16, thickness17,18 and photoexcitation 9,19. Among these different phase transitions, the C-T transition is of great scientific interest due to a large resistivity switching of more than an order of magnitude7,8,18 owing to a Mott gap opening associated with the phase transition. While the Mott gap opening has been extracted by several reports using various optical techniques9,10,20,21,22,23,24,25, an estimation of the electrical gap directly from transport measurement of a TaS2 device is missing. On the contrary, the NC-IC transition can be electrically driven while operating at room temperature and thus has attracted a lot of attention in device applications, including wideband photodetector26, fast oscillator27 and neuromorphic circuits28. However, there are two intrinsic bottlenecks with such resistivity 2 switching. First, the resistivity switching ratio during the NC-IC phase transition is quite weak (< 2)7,8,18. Second, the phase transition driven resistivity switching of TaS2 cannot be controlled by an external gate voltage. Hence, improving the switching ratio in the NC-IC phase transition and adding a possible gate controllability would be of great importance for the advancement of these applications. In order to address these issues, in this work, we employ a 1T-TaS2/2H-MoS2 heterojunction (lattice mismatch of ~6.33%29,30) in a back gated field effect transistor (FET) structure. We show that the TaS2/MoS2 interface exhibits a low barrier, high performance van der Waals (vdW) electrical contact31,32,33,34,35,36, which is promising for pathway towards "all-2D" flexible devices. We also demonstrate that both the C-T and NC-IC phase transitions not only result in a change in the resistivity in the TaS2 film, but also bring about a change in the Schottky barrier height (SBH) at the TaS2/MoS2 interface. This allows us to control the phase transition driven carrier transport through the heterojunction device by the application of a gate voltage. The C-T phase transition results in an increase in the SBH, which allows us to electrically estimate the Mott gap opening in 1T-TaS2 at the C phase. On the other hand, the NC-IC phase transition reduces the SBH at the TaS2/MoS2 interface, which, depending on the gate voltage applied, enhances the switching ratio by a factor as much as 14.5X compared with 1T-TaS2 control. Results and Discussions: 2H-MoS2 is a layered transition metal di-chalcogenide (TMDC) semiconductor, which is appealing as a channel material in electronic device applications owing to its appreciable bandgap, moderate carrier mobility, and channel length scalability. The heterojunction device used in this work is schematically shown in Fig. 1a, where a back gated MoS2 channel is formed with asymmetric contacts, namely Ni and TaS2 contacts on two different sides. We note that Ni makes 3 good electrical contact with MoS2 37,38,39 (see Supplemental Material S140) owing to efficient interfacial charge transfer resulting from strong hybridization of partially filled Ni-3d and S-3p orbitals39. By switching the polarity of the drain voltage, we study characteristics of the carrier injection through the TaS2/MoS2 contact interface while taking Ni/MoS2 junction of the same device as the reference interface. To fabricate the heterojunction device, we first exfoliate few- layer 2H-MoS2 flakes on 285 nm thick SiO2 coated heavily-doped Si substrate. We next transfer thin layers of 1T-TaS2 on top of the MoS2 flake under a microscope using a micromanipulator. The contact electrodes are patterned by electron beam lithography, followed by electron-beam evaporation of Ni (10nm)/Au (50 nm), and subsequent lift-off. Fig. 1b shows the optical image of the device after completion of fabrication. The top panel of Fig. 1c depicts the corresponding thickness mapping image using Atomic Force Microscope (AFM). The bottom panel of Fig. 1c shows the thickness of the MoS2 and the TaS2 flakes are 6.4 nm and 43.6 nm, respectively, as measured along the green dashed arrow. The devices reported in this work are measured multiple times over a period of several weeks and no noticeable degradation of the device characteristics is observed due to surface oxidation and other ambience induced effects. 4 Figure 1. 1T-TaS2/2H-MoS2 heterojunction. (a) Schematic of the device with terminals 1 and 2 probing the TaS2 control device (TS12) and terminals 2 and 3 probing the heterojunction device (H23). (b) Optical image of the fabricated device. Scale bar: 5 m. (c) Top panel: AFM thickness mapping image of the heterojunction device. Scale bar: 5 m. Bottom panel: Thickness of the MoS2 and TaS2 flakes along the green dashed arrow in the top panel. Raman spectroscopy is a useful tool to characterize CDW phase transitions in 1T-TaS2 41,42,43,44,45. When the crystal has not undergone any CDW phase transition, due to the high symmetry of the crystal, specific zone center phonons participate in the first order Raman scattering in order to maintain both energy and momentum conservation. However, once a CDW phase change sets in, the lattice distorts, reducing the translational symmetry of the crystal. This relaxes the condition of first order Raman scattering at the zone center and results in a large number of Raman active vibrational modes41. Fig. 2a shows the acquired Raman spectra from 1T-TaS2 in the heating cycle using a 532 nm laser excitation at 193 K (C phase) and 300 K (NC phase), which are in agreement with previous reports41,42,45. In the C phase, the distinct Raman peaks at the lower frequencies (between 90 cm-1 and 140 cm-1) result from acoustic branches and directly correlate with the signature of the commensurate nature of the C phase. The higher frequency peaks (between 200 cm-1 to 400 cm-1) originate from optical phonons42, and can be observed in both C and NC phases. Fig. 2b depicts the resistance (𝑅) - temperature (𝑇) characteristics of a representative two-probe TaS2 device (in the inset) in vacuum in the heating cycle under small electric field condition. Each layer of 1T-TaS2 crystal structure is composed of tantalum (Ta) atoms, which are surrounded by sulphur atoms in an octahedral arrangement13. The high resistance state at low temperature results from the C phase, where the David-star structure6,46, as depicted in Fig. 2c, forms a commensurate structure with the underlying lattice. This commensurate phase results from the inward 5 displacement of the twelve Ta atoms, located at the star corners, towards the thirteenth Ta atom at the center of the star. The atomic displacement results in the deformation in the structure, including a swelling at the star center30,47. The reduction in the interatomic distance strengthens the bonds inside the David-star in comparison to the bonds outside the star, resulting in the disintegration of the band structure into submanifolds. The twelve corner Ta atoms of the star contribute electrons to the two three-band submanifolds in the valence band, whereas the thirteenth atom at the center of the star contributes one electron to the submanifold in the conduction band (Fig. 2d -- left panel). It has been suggested that the spin-orbit coupling forces further reconstruction in the band structure and results in a unique narrow band at the Fermi level that is partially filled48,49. This facilitates electron-electron interaction induced Mott transition in the lattice and a Mott gap opens up (Fig. 2d -- right panel)47,49. 6 Figure 2. Electrical tuning of phase transitions in 1T-TaS2. (a) Raman shift of 1T-TaS2 in NC phase (at 300 K in red) and in C phase (at 193 K in black) during the heating cycle. (b) Temperature dependent resistance of a representative TaS2 flake under low field condition. Inset: Optical image of the device. (c) David star structure formation in the C phase of 1T-TaS2. (d) Hubbard model of 1T-TaS2 depicting Mott gap opening. (e) Current -- electric field characteristics of 1T-TaS2 two probe device (TS12 -- probing terminals 1 and 2 in Fig. 1a-b) in the temperature range 116 K to 240 K. The black arrows indicate the bias sweep direction. (f) Resistance -- temperature plot of the forward sweep in (e) indicating multiple resistance states. (g) Colour plot of resistance in the temperature-field space. (h)-(l) Joule heating induced localized phases of the device giving rise to multiple resistance states in (f). Electrically accessing different TaS2 phases: In order to access the different CDW phases of TaS2 electrically, we next apply a high field across the probes 1 and 2 of the TaS2 device TS12 in Fig. 1a. The hysteretic bi-stable switching, as observed in Fig. 2e, is indicative of external bias controlled phase change of the TaS2 flake. The sharp change in resistance is observed around 160 K (Fig. 2f), which is lower than the C-T phase transition temperature (~220 K) under low field in Fig. 2b. This suggests that Joule heating induced increase in local temperature plays a key role in the phase change. A color phase plot of the different resistance-states in the temperature-electric field space is shown in Fig. 2g. We construct a simple model for the multi-state resistance switching in the TaS2 flake, as explained in Fig. 2h-l. Initially, at low temperature, the whole flake is in C phase, with linear current-field characteristics (Fig. 2h). This situation is denoted by point A in Fig. 2f-g. As the sample is heated close to the C-T phase transition temperature (𝑇𝑐𝑇), the current induced Joule heating drives the local temperature at the central part of the flake (which is farthest from contact heat sinks) at a higher value than the rest of the flake. Note that, in the C phase, particularly, close to 𝑇𝑐𝑇, an 7 increase in the temperature results in a steep reduction in the lattice component of the thermal conductivity, suppressing the overall thermal conductivity50. This provides a positive feedback and further helps to increase the local temperature. Eventually, the temperature of the central part is driven beyond 𝑇𝑐𝑇, forcing a local C-T phase transition (point B), as schematically depicted in Fig. 2i. This corresponds to a steep jump in the overall resistance of the sample. With further increase in the drain field or heating of the sample (point C), the local temperature of the surrounding portion also increases causing a gradual increase in the size of the central T phase region (Fig. 2j), and, in turn, results in the gradual reduction in the resistance. When the temperature and field are increased further, the whole flake is eventually converted into T phase (Fig. 2k), and no further change in resistance is observed beyond this point (point D). At higher temperature (point E - beyond 283 K lattice temperature), the whole flake transforms into NC phase (Fig. 2l), however, T-NC phase transition has an almost negligible impact on further change in resistance11,13. The above-mentioned Joule heating induced phase transition mechanism is qualitatively supported by the hysteresis observed in the current-field plot in Fig. 2e. The flake undergoes a C-T phase transition due to Joule heating in the forward sweep. When the field is withdrawn, Joule heating is suppressed, but the flake does not immediately come back to high resistance C phase until the flake cools down below 180 K, resulting in hysteresis. Efficient carrier injection by 1T-TaS2 contact: We next explore the carrier injection efficiency from 1T-TaS2 to 2H-MoS2 in the heterojunction device (H23) shown in Fig. 1a, by probing the terminals 2 and 3. We take TaS2 and Ni as the source (S) and the drain (D), respectively, in the rest of the paper. Thus, owing to the asymmetric design of the device, for 𝑉𝑑(= 𝑉32) > 0, electrons are injected from the TaS2 contact, while for 𝑉𝑑 < 0, electrons are injected from Ni into the MoS2 channel. Hence, by switching the polarity of 𝑉𝑑, we can probe the carrier injection from individual 8 contacts, as schematically depicted in the insets of Fig. 3a-b. The transfer characteristics of the device for the two cases, shown in Fig. 3a-b at 𝑇 = 240 K (T-phase), indicate an on-off ratio in excess of 106, irrespective of the carriers being injected from Ni or TaS2. Fig. 3c shows the output characteristics at different back gate voltages (𝑉𝑔). We clearly observe that the magnitude of the drive current is similar for both TaS2 and Ni injection cases, suggesting excellent carrier injection efficiency of TaS2/MoS2 interface. Such highly efficient carrier injection from the TaS2/MoS2 junction is promising for vdW-vdW contact engineering. Figure 3. Current-voltage characteristics of the heterojunction device H23. Probing terminal 2 and 3 in Fig. 1a-b. (a)-(b) Transfer characteristics at 240 K (T phase) at different drain voltages, with (a) 𝑉𝑑 > 0 (Ni is under positive bias and TaS2 sourcing electrons) and (b) 𝑉𝑑 < 0 (TaS2 is under positive bias and Ni sourcing electrons). (c) Output characteristics of the same device. Modulating drive current by C-T phase transition: We next turn our attention to the control of the carrier injection as the phase of the 1T-TaS2 source undergoes a C-T phase transition. The two different situations are schematically depicted in Fig. 4a-b. Note that, in this heterojunction device, the overall current density through the TaS2 source is smaller compared to the high field case discussed in Fig. 2e (TS12) owing to the series resistance offered by the MoS2 channel, and hence 9 the role of local Joule heating in the phase transition in TaS2 source can be ruled out. This results in a more uniform phase transition in TaS2 source controlled by the external temperature. The measured device current with 𝑉𝑑 > 0 and 𝑉𝑑 < 0 are plotted as a function of temperature in Fig. 4c and 4d, respectively. The effect of the TaS2 phase change on the device current manifests as a sharp increase in the drive current for both TaS2 and Ni injection, as indicated by the black arrows. The change in drain current (𝐼𝑑) can be attributed to the change in the series resistance offered by the TaS2 portion of the device due to the C-T phase transition. The fractional enhancement of drive current during phase change is stronger at higher 𝑉𝑔 due to reduced MoS2 channel resistance. Figure 4. Drive current impact of C-T phase transition in 1T-TaS2/2H-MoS2 heterojunction device. (a)-(b) Schematic of the heterojunction with TaS2 in C-phase [in (a)] and in T-phase [in (b)]. (c)-(d) Drive current of device H23 under 𝑉𝑑 > 0 [in (c)] and 𝑉𝑑 < 0 [in (d)]. The C-T phase transition temperature of TaS2 is indicated by black arrows. (e) Temperature dependent peak 10 extrinsic mobility of H23. (f) Resistance ratio (𝜌 = 𝑅𝐶/𝑅𝑇) dependence on the gate voltage for TaS2 injection (𝑉𝑑 = 0.2 𝑉) and Ni injection (𝑉𝑑 = −0.2 𝑉). The MoS2 extrinsic electron mobility (i.e., the effect of series resistance in mobility calculation has not been de-embedded) in the heterojunction device has been extracted by using the relation: 𝜇 = 𝐿 𝑊𝐶𝑜𝑥𝑉𝑑 × 𝑑𝐼𝑑 𝑑𝑉𝑔 where 𝐿 is the channel length, 𝑊 is the channel width, 𝐶𝑜𝑥 is the back-gate oxide capacitance. The peak extrinsic mobility decreases with an increase in temperature, as shown in Fig. 4e. Note that the suppression in the TaS2 series resistance manifests itself by the sharp increase in the peak extrinsic mobility at the phase transition temperature, indicated by the black arrow. Fig. 4f shows the gate voltage dependence of the ratio of the measured resistances in the C-phase and the T-phase (𝜌 = 𝑅𝐶/𝑅𝑇). Here, we define 𝑅𝐶 and 𝑅𝑇 as the total resistance measured right before (at 220 K) and right after (at 240 K) C-T phase transition. At large negative 𝑉𝑔, the total resistance is governed by thermionic injection over the source-channel barrier and consequently the ratio exponentially increases for the Ni injection case (𝑉𝑑 = −0.2 V) due to an increase in the temperature. However, at large positive 𝑉𝑔, where the current injection is dominated by tunneling through the Schottky barrier, the ratio becomes close to unity, but remains larger than 1. In fact, at larger positive 𝑉𝑔, the ratio increases with an increase in 𝑉𝑔. This effect is slightly more prominent in the TaS2 injection (𝑉𝑑 = 0.2) case. Such an increase in the ratio is due to a gradual reduction in the MoS2 channel resistance with 𝑉𝑔, and hence the effect due to TaS2 series resistance change becomes more pronounced. Phase transition induced SBH modulation and extraction of TaS2 C-phase Mott gap: In Fig. 4f, we observe a surprisingly large difference in the ratio 𝜌 in the TaS2 and Ni injection cases when 𝑉𝑔 is below threshold voltage. 𝜌 is found to be suppressed and strongly non-monotonic in 𝑉𝑔 at 11 large negative 𝑉𝑔 for the TaS2 injection case, where the carrier injection is governed by thermionic injection over the Schottky barrier height. This suggests a change in the barrier height at the TaS2/MoS2 interface due to the C-T phase change. To explore the SBH at the TaS2/MoS2 heterojunction, in Fig. 5a, we plot the H23 drain current (in log scale) at 𝑉𝑑 = 0.1 V as a function of temperature, for different gate voltages. As indicated by the dashed box, the device current is found to be suppressed by the C-T phase transition at low gate voltages, although this effect smears out at higher gate voltage. This suppression of current is in contrary to the current enhancement effect at large 𝑉𝑔 in Fig. 4c. Note that at such a low gate voltage, the channel resistance is high, hence TaS2 series resistance or channel mobility do not play any role in such flattening of the device current. This indicates an increase in the SBH at the TaS2/MoS2 contact interface once the phase change in TaS2 sets in. At small 𝑉𝑔, the current injection is completely governed by thermionic emission over the TaS2/MoS2 barrier. An increase in the temperature increases the thermionic emission probability, but the C-T phase change abruptly increases the SBH, compensating for the temperature increase effect. Note that, no such current suppression behavior is observed when Ni injects the electrons (i.e., for 𝑉𝑑 < 0) into MoS2 channel (dashed rectangular box in Fig. 5b). This is due to lack of any phase change of the Ni source, unlike TaS2, hence SBH at the Ni source remains the same before and after 𝑇𝐶𝑇. We extract the total effective barrier (𝜑𝐵) offered by the TaS2/MoS2 junction at a given 𝑉𝑔 using Richardson equation. As recently proposed39, we use a modified Richardson equation for such a top contact geometry: 𝐼𝑑 = 𝐴∗𝑇𝛼𝑒−𝑞𝜑𝐵/𝑘𝐵𝑇 where 1 ≤ 𝛼 ≤ 1.5. This differs in the power of 𝑇 from the Richardson equation typically used for the interface between a metal and a conventional bulk semiconductor owing to a change in the dimensionality. 𝜑𝐵 is extracted from the slope of 12 log ( 𝐼𝑑 𝑇𝛼) versus 𝑞 𝑘𝐵𝑇 as depicted in Fig. 5c. One could clearly observe the abrupt increase in the slope as the phase transition happens, indicating an increase in 𝜑𝐵 in the T phase. The extracted 𝜑𝐵 is plotted in Fig. 5d as a function of 𝑉𝑔 for both C and T phases. We note here that, the extracted 𝜑𝐵 is not very reliable at large positive 𝑉𝑔 due to strong tunneling current and mobility degradation with temperature, which tend to underestimate the extracted effective barrier height. Figure 5. Phase dependent Schottky barrier height at 1T-TaS2/2H-MoS2 interface of H23 and Mott gap estimation in C-Phase of 1T-TaS2. (a) Drain current as a function of temperature under small positive drain bias (TaS2 injection). The dashed box indicates the suppression of drain current by the C-T phase transition under low 𝑉𝑔, which smears out at higher 𝑉𝑔. (b) Drain current as a function of temperature under small negative drain bias (Ni injection), showing no such suppression in the dashed box. (c) Richardson plot (with 𝛼 = 1) at different gate voltages in two different phases, indicating a C-T phase transition driven change in slope at negative 𝑉𝑔 due to 13 change in barrier height. (d) Extracted barrier height plotted as a function of gate voltage in two different phases. The dashed parallel lines indicate linear increase of barrier height for larger negative 𝑉𝑔. (e)-(f) Schematic representation of the barrier height increase mechanism from C- phase to T-phase. In Fig. 5d, extracted barrier height is found to increase linearly with decrease in 𝑉𝑔 when 𝑉𝑔 is small. The 'knee point' in the curve, where the barrier height deviates from linearity, is indicative of the true Schottky barrier height 𝜑𝐵0 (i.e. under flat band condition) of the TaS2/MoS2 interface51. This is much lower compared with the difference between the work function of 1T-TaS2 (5.2 eV)52 and the electron affinity of multi-layer MoS2 (4 eV)53,54. This suggests a strong Fermi level pinning at the interface, close to the conduction band edge of MoS2 -- supporting the excellent carrier injection efficiency through the interface. Such a strong Fermi level pinning is a unique feature of TaS2/MoS2 vdW contact, suggesting the vdW gap does not efficiently suppress the evanescent wave function of the TaS2 states, likely resulting in metal induced gap states (MIGS)55 in MoS2. Note that the increase in barrier height with negative 𝑉𝑔 in C and T phases can be fitted by two parallel lines, as shown in Fig. 5d. This observation indicates that the band bending (∆𝜑) in MoS2 is similar in both cases at large negative 𝑉𝑔. This is schematically explained in Fig. 5e-f, which allows us to write the total barrier as: 𝜑𝐵,𝑝(𝑉𝑔) = 𝜑𝐵0,𝑝 + ∆𝜑(𝑉𝑔), where 𝜑𝐵0,𝑝 is the true SBH of the TaS2/MoS2 interface (under flat-band condition) and 𝑝 ∈ {𝐶, 𝑇} represents the phase of TaS2. Assuming a symmetric Mott gap opening, ∆𝜑𝐵 provides an estimate of the Mott gap in TaS2 in the C phase: ∆𝜑𝐵 = 𝜑𝐵,𝑇 − 𝜑𝐵,𝐶 ≈ 𝜑𝐵0,𝑇 − 𝜑𝐵0,𝐶 ≈ 𝐸𝑔,𝑀𝑜𝑡𝑡 2 14 The vertical separation of the dashed fitting lines in Fig. 5d is an indicator of ∆𝜑𝐵. The extracted Mott gap of TaS2 in the C-phase is estimated to be 𝐸𝑔,𝑀𝑜𝑡𝑡 ≈ 71 ± 7 meV. This is in reasonable agreement with reported numbers in literature from different optical techniques viz. infrared reflectivity20,21, time resolved photoemission spectroscopy9,22, angle resolved photoemission spectroscopy10,23,24 and angle resolved inverse photoemission spectroscopy25 -- providing an independent verification using pure electrical transport method. 15 Figure 6. Enhanced resistance switching during NC-IC phase transition in 1T-TaS2/2H- MoS2 heterojunction device H45. (a)-(b) Drive current of device H45 under 𝑉𝑑 > 0 [in (a)] and 𝑉𝑑 < 0 [in (b)]. The different phase transition temperatures of TaS2 are indicated by black arrows. (c) Resistance ratio (𝑅𝑁𝐶/𝑅𝐼𝐶) as the function of gate voltage for TaS2 injection (in greed symbols, 𝑉𝑑 = 0.2 𝑉) and Ni injection (in orange symbols, 𝑉𝑑 = −0.2 𝑉). The dashed line indicates 1T- TaS2 control. (d)-(e) Schematic for SBH height reduction from NC [in (d)] to IC-phase [in (e)]. (f)-(g) Schematic of the heterojunction with TaS2 in NC-phase [in (f)] and in IC-phase [in (g)]. Enhancing resistance switching during NC-IC phase transition: Fig. 6a-b depict the temperature dependent drain current characteristics from another device (H45) possessing a lower threshold voltage than H23, and the device is driven deep into the inversion by increasing the overdrive voltage. An optical image of the device H45, along with the control TaS2 transport characteristics are provided in Supplemental Material S240. In this device, we could modulate the drive current by as much as ~40% through the C-T phase change under large gate overdrive condition. The strong suppression of the drive current at higher temperatures (beyond 223 K) is due to the temperature induced mobility degradation effect. The temperature dependent peak extrinsic mobility extracted from H45 is shown in Supplemental Material S240. When we drive the temperature of the device up to 360 K, which is beyond the NC-IC phase transition temperature at 353 K, we observe that the NC-IC phase transition manifests as a step jump in the drive current, both for TaS2 and Ni injection cases as in Fig. 6a and b, respectively. The corresponding resistance switching ratios ( 𝑅𝑁𝐶(𝑇=350 𝐾) 𝑅𝐼𝐶(𝑇=360 𝐾) ) at 𝑉𝑑 = ±0.2 V are plotted in Fig. 6c. For reference, we also show the ratio for the TaS2 control as a dashed line in the same plot. For TaS2 injection case (𝑉𝑑 = 0.2), the switching ratio is a strong function of 𝑉𝑔 and is remarkably large at negative 𝑉𝑔 reaching a value of 17.3 at 𝑉𝑔 = −80 V, which is 14.5 times higher than the TaS2 control device. On the other hand, 16 the ratio remains a weak function of 𝑉𝑔 for Ni injection case and remains close to the value of TaS2 control. These observations point to a suppression of the SBH at the TaS2/MoS2 interface due to the NC- IC phase transition, as schematically depicted in Fig. 6d-g. Consequently, for electron injection from TaS2 source, the current modulation is much higher, while for Ni injection, we only get small effect due to a change in the series resistance of TaS2 during the phase transition. While the origin of such a change in SBH requires further investigation, it is likely that during the NC-IC phase transition, as the hexagonal David-star clusters are broken (Fig. 6f-g) to increase conductivity, there is a more pronounced effect of the MIGS from TaS2 into the bandgap of MoS2. This causes the Fermi level to be pinned closer to the conduction band edge of MoS2, reducing the SBH. Conclusion: In conclusion, we have demonstrated a low-barrier efficient electrical contact between 1T-TaS2 source and 2H-MoS2 channel, which is promising for "all-2D" flexible electronics. Along with the usual conductivity switching of 1T-TaS2 during different phase transitions, we discovered that these transitions also bring about a change in the Schottky barrier height at the 1T-TaS2/2H-MoS2 interface. The phase transition driven resistance switching ratio of the heterojunction thus shows a large modulation that can be controlled by an external gate voltage. This enhancement and additional gate control provide an unprecedented opportunity for boosting different device applications which exploit such phase transition induced resistance switching, such as broadband photodetection, neuromorphic circuits, negative differential conductance and fast oscillator. 17 Acknowledgements: K. M. acknowledges the support a grant from Indian Space Research Organization (ISRO), grants under Ramanujan Fellowship, Early Career Award, and Nano Mission from the Department of Science and Technology (DST), Government of India, and support from MHRD, MeitY and DST Nano Mission through NNetRA. Conflicts of Interest: The authors declare no conflict of interest. References: 1. Butler, S. Z. et al. Progress, challenges, and opportunities in two-dimensional materials beyond graphene. ACS Nano 7, 2898 -- 2926 (2013). 2. Jariwala, D., Sangwan, V. K., Lauhon, L. J., Marks, T. J. & Hersam, M. C. Emerging device applications for semiconducting two-dimensional transition metal dichalcogenides. ACS Nano 8, 1102 -- 1120 (2014). 3. Akinwande, D., Petrone, N. & Hone, J. Two-dimensional flexible nanoelectronics. Nat. Commun. 5, 1 -- 12 (2014). 4. Thorne, R. E. Charge-density-wave conductors. Phys. Today 49, 42 -- 47 (1996). 5. Grüner, G. The dynamics of charge-density waves. Rev. Mod. Phys. 60, 1129 -- 1181 (1988). 6. Wilson, J. A., Di Salvo, F. J. & Mahajan, S. Charge-density waves and superlattices in the metallic layered transition metal dichalcogenides. Adv. Phys. 24, 117 -- 201 (1975). 7. Williams, P. M., Scruby, C., Clark, W. & Parry, G. Charge Density Waves in the Layered Transition Metal Dichalcogenides. J. Phys. Colloq. 37 (C4), C4 139-C4 150 (1976). 18 8. Sipos, B. et al. From Mott state to superconductivity in 1T-TaS2. Nat. Mater. 7, 960 -- 965 (2008). 9. Hellmann, S. et al. Ultrafast melting of a charge-density wave in the mott insulator 1T-TaS2. Phys. Rev. Lett. 105, 187401 (2010). 10. Rossnagel, K. On the origin of charge-density waves in select layered transition-metal dichalcogenides. J. Phys. Condens. Matter 23, 213001 (2011). 11. Manzke, R. et al. On the phase transitions in 1T-TaS2. Europhys. Lett. 8, 195 -- 200 (1989). 12. Ishiguro, T. & Sato, H. Electron microscopy of phase transformations in 1T-TaS2. Phys. Rev. B 44, 2046 -- 2060 (1991). 13. Thomson, R. E., Burk, B., Zettl, A. & Clarke, J. Scanning tunneling microscopy of the charge-density-wave structure in 1T-TaS2. Phys. Rev. B 49, 16899 -- 16916 (1994). 14. Tani, T., Osada, T. & Tanaka, S. The pressure effect on the CDW-transition temperatures in 1T-TaS2. Solid State Commun. 22, 269 -- 272 (1977). 15. Di Salvo, F. J., Wilson, J. A., Bagley, B. G. & Waszczak, J. V. Effects of doping on charge- density waves in layer compounds. Phys. Rev. B 12, 2220 -- 2235 (1975). 16. Ang, R. et al. Real-space coexistence of the melted Mott state and superconductivity in Fe- substituted 1T-TaS2. Phys. Rev. Lett. 109, 176403 (2012). 17. Yoshida, M. et al. Controlling charge-density-wave states in nano-thick crystals of 1T-TaS2. Sci. Rep. 4, 07302 (2014). 18. Yu, Y. et al. Gate-tunable phase transitions in thin flakes of 1T-TaS2. Nat. Nanotechnol. 10, 270 -- 276 (2015). 19 19. Stojchevska, L.; Vaskivvskyi, I.; Mertelj, T.; Svetin, D.; Brazovskii, S.; Mihailovic, D. Ultrafast Switching to a Stable Hidden Quantum State in an Electronic Crystal. Science (80- . ). 344, 177 -- 180 (2014). 20. Barker, A. S., Ditzenberger, J. A. & DiSalvo, F. J. Infrared study of the electronic instabilities in tantalum disulfide and tantalum diselenide. Phys. Rev. B 12, 2049 -- 2054 (1975). 21. Gasparov, L. V. et al. Phonon anomaly at the charge ordering transition in 1T-TaS2. Phys. Rev. B - Condens. Matter Mater. Phys. 66, 943011 -- 943015 (2002). 22. Perfetti, L. et al. Femtosecond dynamics of electronic states in the Mott insulator 1T-TaS2 by time resolved photoelectron spectroscopy. New J. Phys. 10, 053019 (2008). 23. Pollak, R. A., Eastman, D. E., Himpsel, F. J., Heimann, P. & Reihl, B. 1T-TaS2 charge- density-wave metal-insulator transition and Fermi-surface modification observed by photoemission. Phys. Rev. B 24, 7435 -- 7438 (1981). 24. Pillo, T. et al. Fine structure in high-resolution photoemission spectra of quasi-two- dimensional 1T-TaS2. Phys. Rev. B 64, 245105 (2001). 25. Sato, H. et al. Conduction-band electronic structure of 1 T -TaS2 revealed by angle-resolved inverse-photoemission spectroscopy. Phys. Rev. B - Condens. Matter Mater. Phys. 89, 155137 (2014). 26. Wu, Dong; Ma, Y. C.; Niu, Y. Y.; Liu, Q. M.; Dong, Tao; Zhang, S. J.; Niu, J. S.; Zhou, H. B.; Wei, Jian; Wang, Y. X.; Zhao, Z. R.; Wang, N. L. Ultrabroadband photosensitivity from visible to terahertz at room temperature. arXiv:1806.08952 20 27. Liu, G. et al. A charge-density-wave oscillator based on an integrated tantalum disulfide- boron nitride-graphene device operating at room temperature. Nat. Nanotechnol. 11, 845 -- 850 (2016). 28. Khitun, A., Liu, G. & Balandin, A. A. Two-dimensional oscillatory neural network based on room temperature charge-density-wave devices. IEEE Trans. Nanotechnol. 16, 1 -- 8 (2017). 29. Young, P. a. Lattice parameter measurements on molybdenum disulphide. J. Phys. D. Appl. Phys. 1, 936 -- 938 (1968). 30. Spijkerman, A., de Boer, J. L., Meetsma, A., Wiegers, G. A. & van Smaalen, S. X-ray crystal-structure refinement of the nearly commensurate phase of 1T-TaS2 in (3+2)- dimensional superspace. Phys. Rev. B 56, 13757 (1997). 31. Lee, Y. T. et al. Graphene versus ohmic metal as source-drain electrode for MoS2 nanosheet transistor channel. Small 10, 2356 -- 2361 (2014). 32. Du, Y. et al. MoS2 Field-Effect transistors with graphene/metal heterocontacts. IEEE Electron Device Lett. 35, 599 -- 601 (2014). 33. Leong, W. S. et al. Low resistance metal contacts to MoS2 devices with nickel-etched- graphene electrodes. ACS Nano 9, 869 -- 877 (2015). 34. Andleeb, S., Eom, J., Naz, N. R. & Singh, A. K. MoS2 field effect transistor with graphene contacts. J. Mater. Chem. C 5, 8308 -- 8314 (2017). 35. Sata, Y. et al. N- and p-type carrier injections into WSe2 with van der Waals contacts of two-dimensional materials. Jpn. J. Appl. Phys. 56, 04CK09 (2017). 21 36. Shin, H. G. et al. Vertical and In-Plane Current Devices Using NbS2/n-MoS2 van der Waals Schottky Junction and Graphene Contact. Nano Lett. (2018). doi:10.1021/acs.nanolett.7b05338 37. Yang, L. et al. Chloride molecular doping technique on 2D materials: WS2 and MoS2. Nano Lett. 14, 6275 -- 6280 (2014). 38. English, C. D., Shine, G., Dorgan, V. E., Saraswat, K. C. & Pop, E. Improved contacts to MoS2 transistors by ultra-high vacuum metal deposition. Nano Lett. 16, 3824 -- 3830 (2016). 39. Somvanshi, D. et al. Nature of carrier injection in metal/2D-semiconductor interface and its implications for the limits of contact resistance. Phys. Rev. B 96, 205423 (2017). 40. See Supplemental Material at [xxx] for output characteristics of MoS2 channel with Ni contact on both sides, characteristics of TaS2 control and heterojunction from device H45. 41. Smith, J. E., Tsang, J. C. & Shafer, M. W. Raman spectra of several layer compounds with charge density waves. Solid State Commun. 19, 283 -- 286 (1976). 42. Duffey, J. R., Kirby, R. D. & Coleman, R. V. Raman scattering from 1T-TaS2. Solid State Commun. 20, 617 -- 621 (1976). 43. Sugai, S.; Murase, K.; Uchida, S.; Tanaka, S. Comparison of the soft modes in Tantalum Dichalcogenides. Phys. B+C 105, 405 -- 409 (1981). 44. Fu, W. et al. Controlled Synthesis of Atomically Thin 1T-TaS2 for Tunable Charge Density Wave Phase Transitions. Chem. Mater. 28, 7613 -- 7618 (2016). 45. He, R. et al. Distinct surface and bulk charge density waves in ultrathin 1T−TaS2. Phys. Rev. B 94, 201108 (2016). 22 46. Tosatti, E. & Fazekas, P. On the nature of the low-temperature phase of 1T-TaS2. J. Phys. Colloq. 37, C4-165-C4-168 (1976). 47. Fazekas, P. & Tosatti, E. Electrical , structural and magnetic properties of pure and doped 1T-TaS2. Philos. Mag. B 39, 229 -- 244 (1979). 48. Smith, N. V, Kevan, S. D. & DiSalvo, F. J. Band structures of the layer compounds 1T- TaS2 and 2H-TaSe2 in the presence of commensurate charge-density waves. J. Phys. C Solid State Phys. 18, 3175 -- 3189 (1985). 49. Rossnagel, K. & Smith, N. V. Spin-orbit coupling in the band structure of reconstructed 1T−TaS2. Phys. Rev. B 73, 073106 (2006). 50. Núñez-Regueiro, M., Lopez-Castillo, J. & Ayache, C. Thermal Conductivity of 1T-TaS2 and 2H-TaSe2. Phys. Rev. Lett. 55, 1931 -- 1934 (1985). 51. Das, S., Chen, H. Y., Penumatcha, A. V. & Appenzeller, J. High performance multilayer MoS2 transistors with scandium contacts. Nano Lett. 13, 100 -- 105 (2013). 52. Shimada, T. & Ohuchi, F. S. Work Function and Photothreshold of Layered Metal Dichalcogenides. Jpn. J. Appl. Phys. 33, 2696 -- 2698 (1994). 53. Schlaf, R., Lang, O., Pettenkofer, C. & Jaegermann, W. Band lineup of layered semiconductor heterointerfaces prepared by van der Waals epitaxy: Charge transfer correction term for the electron affinity rule. J. Appl. Phys. 85, 2732 -- 2753 (1999). 54. Howell, S. L. et al. Investigation of Band-Offsets at Monolayer-Multilayer MoS2 Junctions by Scanning Photocurrent Microscopy. Nano Lett. 15, 2278 -- 2284 (2015). 55. Tersoff, J. Schottky Barrier Heights and the Continuum of Gap States. Phys. Rev. Lett. 52, 23 465 -- 468 (1984). 24 Supplemental Material: Gate controlled large resistance switching driven by charge density wave in 1T-TaS2/2H- MoS2 heterojunction Mehak Mahajan, Krishna Murali, Nikhil Kawatra, and Kausik Majumdar Department of Electrical Communication Engineering, Indian Institute of Science, Bangalore 560012, India These authors contributed equally *Corresponding author, email: [email protected] 25 Supplemental Material S1: Output characteristics of MoS2 channel with Ni contact on both sides Figure S1. Id-Vd characteristics of MoS2 channel with Ni contact on both sides for different gate voltages, at T = 290 K, with channel length of 2.8 m. 26 -1.0-0.50.00.51.0-16-80816Vg= 50 V T = 290 K Current (A/m)Drain Voltage (V)Vg= -50 V Supplemental Material S2: Characteristics of TaS2 control and heterojunction from device H45 Figure S2. (a) Optical image of the heterojunction device H45. T1 and T2 are the contacts to the TaS2 flake, while M1 and M2 are contacts to MoS2. Scale bar is 5 m. (b) Current versus electric field characteristics of TaS2 control (probed between T1 and T2) at different temperatures and low electric field condition. (c) The resistance versus temperature plot extracted from (b) clearly indicating the C-T transition around 220 K and the NC-IC phase transition around 350 K. (d) Output characteristics of the heterojunction H45 (probed between T1 and M1) at 290 K. (e) Temperature dependent peak extrinsic mobility, as extracted from H45. 27
1812.09226	1	1812	2018-12-21T16:11:42	Thermoelectric transport in two-dimensional topological insulator state based on HgTe quantum well	[ "cond-mat.mes-hall" ]	The thermoelectric response of HgTe quantum wells in the state of two-dimensional topological insulator (2D TI) has been studied experimentally. Ambipolar thermopower, typical for an electron-hole system, has been observed across the charge neutrality point, where the carrier type changes from electrons to holes according to the resistance measurements. The hole-type thermopower is much stronger than the electron-type one. The thermopower linearly increases with temperature. We present a theoretical model which accounts for both the edge and bulk contributions to the electrical conductivity and thermoelectric effect in a 2D TI, including the effects of edge to bulk leakage. The model, contrary to previous theoretical studies, demonstrates that the 2D TI is not expected to show anomalies of thermopower near the band conductivity threshold, which is consistent with our experimental results. Based on the experimental data and theoretical analysis, we conclude that the observed thermopower is mostly of the bulk origin, while the resistance is determined by both the edge and bulk transport.	cond-mat.mes-hall	cond-mat	Thermoelectric transport in two-dimensional topological insulator state based on HgTe quantum well G. M. Gusev,1 O. E. Raichev,2 E. B. Olshanetsky,3 A. D. Levin,1 Z. D. Kvon,3,4 N. N. Mikhailov,3 and S. A. Dvoretsky,3 1Instituto de F´ısica da Universidade de Sao Paulo, 135960-170, Sao Paulo, SP, Brazil 2Institute of Semiconductor Physics, NAS of Ukraine, Prospekt Nauki 41, 03028 Kyiv, Ukraine 3Institute of Semiconductor Physics, Novosibirsk 630090, Russia 4Novosibirsk State University, Novosibirsk, 630090, Russia E-mail: [email protected] Abstract. The thermoelectric response of HgTe quantum wells in the state of two-dimensional topological insulator (2D TI) has been studied experimentally. Ambipolar thermopower, typical for an electron-hole system, has been observed across the charge neutrality point, where the carrier type changes from electrons to holes according to the resistance measurements. The hole-type thermopower is much stronger than the electron-type one. The thermopower linearly increases with temperature. We present a theoretical model which accounts for both the edge and bulk contributions to the electrical conductivity and thermoelectric effect in a 2D TI, including the effects of edge to bulk leakage. The model, contrary to previous theoretical studies, demonstrates that the 2D TI is not expected to show anomalies of thermopower near the band conductivity threshold, which is consistent with our experimental results. Based on the experimental data and theoretical analysis, we conclude that the observed thermopower is mostly of the bulk origin, while the resistance is determined by both the edge and bulk transport. 1. Introduction The HgTe-based quantum well (QW) is a semiconductor system, where the two-dimensional (2D) conduction and valence subbands, divided by a narrow variable gap, can be created. The switch between the electron and the hole types of transport is achieved by the gate control. The gap energy is controlled by the well width d and goes to zero at d ≃ 6.3 nm, when the electron energy spectrum resembles a Dirac cone, like in graphene. Wider wells have an inverted energy band order, known to be in the state of 2D topological insulator (TI) [1] characterized by a pair of counterpropagating gapless edge modes. The edge states have a helical spin structure and are supposed to be robust to backscattering. The edge state transport in such HgTe QWs has been confirmed experimentally for ballistic transport in mesoscopic samples [2-4] and for diffusive transport in macroscopic samples [4,5]. Similar results have been obtained in experiments on Si-doped InAs/GaSb quantum wells [6-9] which are also believed to be 2D TI. Application of novel experimental methods for the study of the transport properties of 2D TI is of particular interest. The thermoelectric measurements can give complementary information about electron transport in metals and semiconductors and are used as a powerful tool for probing the sign of the charge carriers and the transport mechanisms. The diffusive thermopower is often described using the Mott relation, as the logarithmic derivative of the energy-dependent electrical conductivity. Apart from necessitating the validity of the Boltzmann equation and the degeneracy of the electron gas, the Mott relation has also other limitations discussed in the literature. In particular, a strong energy dependence of the relaxation time, when this time changes considerably within the kT interval around the Fermi level, causes a failure of the simple Mott relation. Indeed, the deviation from the Mott relation has been observed in thermopower experiments in graphene near the charge neutrality point (CNP) and at high temperatures [10,11]. Application of the original Mott relation [12], which is more general because it is not based on the approximate Sommerfeld expansion, allows one to overcome these limitations and describe the thermopower in a wide range of temperature and carrier density [13]. However, in the case of strongly inelastic scattering by optical phonons, considerable deviations even from the Mott relation in its general form are expected, as recently demonstrated in high mobility graphene samples [14,15]. Similar to graphene, the system based on HgTe quantum wells reveals ambipolar Hall effect accompanied by the resistance peak near the CNP. However, in contrast to the gapless graphene, in 2D TI the transition between the electron and hole types of conduction as the gate voltage is swept through the CNP, occurs when the Fermi level stays in the insulating gap and transport is determined by the edge states. The position of the Fermi level in the gap is stabilized by the bulk states which are present in the gap because of a random spatial inhomogeneity (disorder). These states are often considered as localized ones. However, electron transitions between the edge and the bulk states are possible and may influence transport properties in 2D topological insulators [4,5,16]. The edge to bulk mixing is believed to cause a strong enhancement of the thermopower in 2D TI. In particular, it is suggested [17,18] that when the Fermi level approaches the bulk band edge, the scattering rate of electrons in the edge states increases rapidly and significantly, which is expected to cause an anomalous growth of the amplitude of Seebeck signal and a change of its sign. This offers a new opportunity to improve the thermoelectric parameter, such as the figure of merit zT , which is defined as zT = GS2T /(Ke + Kph), where G is the electron conductance, S is the Seebek coefficient, Ke and Kph are the thermal conductances of electrons and phonons consequently. The interplay between the edge and the bulk conductances leads to a strong dependence of the parameter zT on the sample geometry and size. It has been predicted that the value of the figure of merit can be improved by more than ∼ 1 for a certain geometry at room temperature [17,18]. Despite the interest to the thermoelectric properties in 2D topological insulators, the experimental studies have almost all been focused on the measurements of the electrical resistance. In the present paper we report an experimental study of the thermopower in band-inverted HgTe-based quantum wells. At the CNP where the resistance reaches its maximum, the thermopower changes its sign, showing the ambipolar behavior. The nonlocal resistance in our samples is comparable with the local one. This observation clearly proves the presence of the edge state transport, which dominates within the bulk gap. Importantly, we do not observe any of the anomalies of the Seebeck effect predicted in Refs. [17,18], in particular, the sign of the effect changes like in a normal electron-hole system. This apparently suggests that the effect of the edge to bulk scattering on the transport is not as significant as it was expected. To verify this statement, we have carried out a calculation of conductivity and thermopower in the 2D TI, taking into account both the particle and energy balance in the coupled system of edge and bulk states. In brief, we demonstrated that the transport properties are determined not by the edge to bulk scattering rate alone, but by the spin current flowing between the edge and the bulk. If the spin relaxation in the bulk is slow, a bottleneck effect takes place, when the spin current is limited by the bulk conductivity rather than by the edge to bulk scattering rate. Therefore, in the region where the bulk conductivity is smaller than the conductance quantum e2/h the scattering between the edge and the bulk is expected to be insignificant, while in the regions of larger bulk conductivity the edge-state contribution to transport is no longer important. Further, from a qualitative analysis of our experimental data supported by the theoretical considerations, we conclude that the observed thermopower is mostly of the bulk transport origin. The paper is organized as follows. Section II contains the description of measurements and experimental results. Section III is devoted to the theoretical model and to discussion of the results based on this model. The concluding remarks are given in the last section. Figure 1. (Color online) Sample geometry and resistance of shortest segment (RI=1,6;V =4,5) as a function of the gate voltage for different temperatures (Sample A). Left-schematic structure of the sample. 2. Experiment To probe the carrier transport, we measured thermoelectric voltage V and thermopower S together with the resistance. The quantum well structures Cd0.65Hg0.35Te/HgTe/Cd0.65Hg0.35Te with [013] surface orientations and widths d = 8 − 8.3 nm were prepared by molecular beam epitaxy (figure 1, left panel). The sample is a long Hall bar consisting of three 3.2 µm wide consecutive segments of different length (3, 9, and 35 µm) and seven voltage probes, covered by the TiAu gate ( see figure 1, top panel). The measurements were performed in a variable temperature insert cryostat in the temperature range 1.4 − 10 K using the standard four point scheme. A detailed description of the sample structure has been given in Ref. [5]. The electrically powered heater placed symmetrically near the contact 1 (see Fig. 2, top panel) creates temperature gradient in the system, while the other end is indium soldered to a small copper slab that serves as a thermal ground. The copper slab is, in turn, connected to the copper rod of the sample holder. One calibrated thermo sensor is attached at the end of the sample near the heater while the other is attached to the heat sink. The thermo sensors were used to measure the ∆T along the sample. The voltages induced by this gradient were measured by a lock-in detector at the frequency of 2f0 = 0.8 − 2 Hz across various voltage probes. The thermal conductance of the sample is overwhelmingly dominated by phonon transport in the GaAs substrate [19,20]; diffusive heat transport by the two-dimensional gas is negligible in comparison. The thermal conductivity κ of a pure dielectric crystal is usually determined by the boundary scattering at low temperatures and depends on the temperature as κ ∼ T 3. We performed the measurements of the thermal conductivity in our samples and obtained the value 300 − 200W m−1K−1 at T=4.2K for different substrates, which agrees with the previous measurements [19,20]. We did not directly measure the temperature difference between the voltage probes, since the distance is very small. We estimate this difference between probes 3 and 2 as ∼ 20mK K for the heater power used in our experiment. For a given temperature difference between the sample extremities the temperature profile along the sample could be nonlinear, especially close to the ends of the substrates. The situation is somewhat similar to the electrical measurements, when the electric potential profile is inhomogeneous near the metallic contacts. However, we expect that in the center of the sample and along the short distances the profile is linear. We have also checked that the temperature difference varies linearly with heater power. Four different devices have been studied. Below we show the results obtained in two representative samples ( A and B). ) m h O k ( , 5 4 = V 6 1 = , ; I R 40 30 20 10 0 (a) T=4.2 K holes electrons 1.0 0.5 0.0 ) V ( , 5 4 V (b) P(arb.units) 1 0.64 0.44 0.25 T=4.2 K 0 Vg-VCNP (V) 2 -1 0 1 Vg-VCNP(V) 2 3 Figure 2. (indicated) as a function of gate voltage , T = 4.2 K (sample A). (Color online) Resistance (a) and thermovoltage (b) for different heater powers The variation of the resistance with the gate voltage and lattice (bath) temperature is shown in Figure 1. The resistance of the shortest segment reveals a broad peak whose amplitude is larger than the value h/2e2 expected in the ballistic case. We see that the resistance decreases sharply for temperatures above 15 K while saturating below 10 K. We find that the profile of the resistance temperature dependencies above T > 15K fits very well the activation law R exp(∆/2kT ), where ∆ is the activation gap. Insert in Figure 1 shows the peak maximum resistance versus temperature. The thermally activated behavior of resistance above 15 K corresponds to a gap of 10 meV between the conduction and valence bands in the HgTe well. The mobility gap can be smaller than the energy gap due to disorder. Figure 2 shows the (a) 250 200 150 (b) 4.2 K 3.5K 3.2K 2.4K 200 150 100 ) / K V ( S 50 0 -50 -2 -1 0 1 Vg-VCNP(V) ) / K V ( S 100 50 2 2 (c) 1000 800 600 400 ) 1 - K 1 - m W ( ~T3 200 3 T(K) 4 2 3 T(K) 4 (Color online) (a) Thermopower for different temperatures. (b) Temperature Figure 3. dependence of thermopower at Vg − VCN P = −1.2 V. (c) Lattice thermoconductivity of the GaAs substrate as a function of the temperature (Sample B). resistance and thermovoltage as a function of the gate voltage measured between probes 4 and 5 at T = 4.2 K. The thermovoltage increases nearly linearly with heater power, which proves that we measure the longitudinal (Seebeck) thermoelectric effect. The signal shows a behavior similar to other electron-hole systems such as graphene [10,11]. It changes sign at the charge neutrality point (CNP) and decreases with the carrier density increasing. The voltage interval between the electron-like and hole-like regimes (∆Vg ∼ 1 V) is almost two times smaller than the half-width of the resistance peak. The Figure 3a displays the traces of the thermopower versus Vg for different temperatures. Figure 3b shows the temperature dependence of the thermopower measured across a longer bridge at a selected gate voltage Vg− VCN P = −1.2 V (hole side) where the thermopower approaches its maximum. It is found that the signal grows almost linearly with temperature: S ∼ T 1.3±0.1 ( figure 3b) in the temperature interval 2.2 < T < 4.2K. This temperature interval was selected, because resistance becomes temperature dependent above T > 10K( see figure 1), and metallic approximation for thermopower is no longer valid. It is worth noting that prior to the thermoelectric measurements the thermal conductance of the sample has been determined. The thermoconductance is dominated by the phonon transport in the substrate; the contribution from the diffusive heat transfer by the electrons is negligibly small. The thermal conductivity of the GaAs substrate is usually determined by the boundary scattering at low temperatures [19, 20]. The thermoconductivity is given by: κ = 2π2 15 kΛ v2 h (cid:19)3 ph (cid:18) kT , (1) where Λ is the phonon mean-free path and vph = 3300m/s the appropriate mean acoustic phonon velocity. Figure 3c shows the thermal conductivity of the GaAs substrate as a function of the lattice temperature. The thermal conductivity follows κ ∼ T 3 law, in accordance with eq.1. The value of κ is ∼ 600W m−1K −1 at T=4.2K, which agrees well with the previously measured thermoconductivity in pure GaAs substrates [19,20]. Once the thermal conductance is found, the temperature gradient can be used to convert the measured thermoelectric voltages into the thermopower. Figure 4 shows resistance as a function of the gate voltage measured (a) RI=1,6;V=2,3 T=4.2 K 400 300 200 S3,4 S2,3 (b) T=4.2K ) K V / 100 RI=1,6;V=3,4 ( S 0 -100 h/2e2 S4,5 400 300 ) / K V ( S 200 100 (c) 3 2 ) h / 2 e ( G 1 RI=1,6;V=4,5 1 0 Vg-VCNP(V) -1 2 -200 -2 3 -1 0 1 Vg-VCNP(V) 2 3 0 5 10 15 30 35 0 40 20 25 Lgate( m) ) m h O k ( R 80 60 40 20 0 (Color online) (a) Resistance R as a function of gate voltage measured between Figure 4. various voltage probes, T = 4.2 K, I = 10−9 A. (b) Thermopower as a function of the gate voltage measured between various voltage probe, T = 4.2 K. (c) Resistance at the CNP and thermopower at Vg − VCN P = −0.7V as a function of the distance between the voltage probes L (Sample B). between different probes at T = 4.2 K. It is worth noting that the edge current flows along the gated sample edge whose length Lgate is longer than the distance between the probes L (bulk current path) and corresponds to 5-6 µm. For longer distances between the probes we see higher resistances. The large resistance can appear because of multiple transitions of electrons between counterpropagating helical edge states caused by either direct backscattering or electron transfer mediated by the bulk states in the puddles [21] emerging due to spatial potential fluctuations near the edge. The observation of a nonlocal resistance constitutes the main proof of the presence of the edge state transport in a 2D TI. A systematic study of the local and nonlocal transport in 2D TI has been preformed in the previous works in the ballistic [3,4] and in the diffusive regimes [16]. The dependence of the resistance peak on the length Lgate is shown in figure 4c. It is found to be very close to the 1/Lgate dependence. The thermopower signal has a nonmonotonic dependence on the distance Lgate (figure 4c) in contrast to the resistance. Below we consider the theory, which accounts for both the edge and bulk contributions to the electrical conductance and thermoelectric effect in 2D TI, including the effects of edge to bulk leakage. Finally a few words need to be said about the thermoelectric efficiency in the 2D topological insulator regime. As mentioned in the introduction, in conventional semiconductors the factor zT is size independent because the geometrical factor is canceled between the conductance and the thermoconductance. In 2D topological insulator regime factor zT can be optimized by choosing appropriate geometries [18]. A large enhancement of the power factor is predicted for the gapless edge states near the charge neutrality point. Figure 5a shows the conductance measured near the CNP as a function of the gate voltage. It is expected that when the edge-state contribution to the transport is important, the nonlocal resistance should be observed [3,4]. 100 ) h / 2 e ( G 10 1 edge transport regime 1E-11 1E-12 ) 2 K / 2 V h / 2 e ( 2 S G 1E-13 1E-14 0.1 ) 2 e h ( / L N R (a) 0.01 2 1E-15 1E-16 -1 (b) 1 0 Vg-VCNP(V) -1 0 1 Vg-VCNP(V) Figure 5. RN L = RI=4,8;V =5,7 as a function of the gate voltage. the gate voltage near CNP, T=4.2K (Sample B). (Color online) (a) Conductance G = 1/RI=1,6;V =4,5 and nonlocal resistance (b) Coefficient GS2 as a function of For comparison figure 5a presents an example of the nonlocal resistance when the current flows between contacts 4-8 and the voltage is measured between contacts 5-7, i.e. RI=4,8;V =5,7. The nonlocal resistance peaks are narrower and lower as compared to the local resistance peaks measured in the same device. Simple estimation from Kirchoff formula (see for details [16]), gives RI=4,8;V =5,7 = 0.6 h e2 for the mean free path l = 10µm, which in 2 times large than the experimental value RI=4,8;V =5,7 = 0.3 h e2 . It is not surprising, since when the edge current flows over a long distance (in this particular case L4,8 >> l) there is a high probability for the coupling with the bulk states, and the total current experiences considerable leakage into the bulk. An advanced theory, considered in [16], is required for a more detailed analysis. However, we may conclude here that the edge state transport dominates in the voltage interval −1V < Vg < 1V . Figure 5b shows the coefficient GS2 as a function of the gate voltage near the CNP in the edge state transport regime at T=4.2K. It is clear that an enhancement of GS2 is observed compared to other than TI cases on the hole-side of the resistance peak. Unfortunately this enhancement is observed at low temperature and is unlikely to survive at high T ( see figure 1). However, quite recently the TI state has been observed at T ∼ 100K [22], and one may hope that the thermopower characteristics presented here could be valid at higher temperatures in these new materials. 3. Theory Since the thermovoltage follows almost linear temperature dependence and monotonically decreases with increasing carrier density far away from the CNP, the thermopower is likely determined by the diffusive mechanism. The phonon drag thermopower would have different temperature dependence. In comparison to GaAs quantum wells, where the phonon drag mechanism is essential, in HgTe quantum wells this mechanism should be much less important because of relative smallness of deformation-potential and piezoelectric constants in HgTe, and because of relative smallness of the density of states of 2D carriers. In the edge state transport, the possibility of phonon drag is negligible because of topological protection of the edge states. Therefore, we consider the diffusive mechanism in the following. The transport in the edge channel k (since the channel is helical, k = 1, 2 denote both the direction of motion and spin orientation) along the axis Ox is described by the Boltzmann equations for the energy distribution functions of electron, fkε(x): sk ∂fkε ∂x = γε(fk′ε − fkε) + gε(Fkε − fkε) + J ee kε v + J ph kε v , (2) ε /v and gε = νeb where k′ 6= k, v is the edge state velocity, Fkε(x, y) is the isotropic part of the electron distribution function in the bulk, γε = νbs ε /v are the inverse mean free path lengths for elastic backscattering and edge to bulk scattering (νbs and νeb are the corresponding scattering rates). Next, J ee and J ph are the collision integrals for inelastic processes, electron-electron and electron-phonon scattering. Assuming that the electrons in the state 1 move from the left to the right (i.e., have positive velocity), we put s1 = 1 and s2 = −1. The transitions between counterpropagating edge states (either elastic or inelastic [23]) are rare because of topological protection, while the scattering between the edge and the bulk is weak because of low probability of finding the bulk-state puddles [21] in the close vicinity to the edge (it may become strong, however, above the threshold of band conductivity). On the other hand, the inelastic electron-electron scattering within a single edge channel is free from these restrictions and, therefore, is much stronger. Moreover, such kind of scattering has a very large phase space, especially if v is energy-independent so that momentum and energy conservation rules are satisfied simultaneously for any two electrons participating in the collisions [24]. As the electron-electron scattering controls electron distribution, we search the distribution function in the Fermi-like form, fkε(x) = {exp[(ε − ϕk(x))/Tke(x)] + 1}−1 [24], characterized by the coordinate-dependent elecrochemical potential ϕk and temperature Tke. Assuming that in the bulk the different spin states are weakly coupled, we use the similar form for Fkε(x, y). Indeed, in the original Bernevig-Hughes-Zhang Hamiltonian [25] describing HgTe quantum wells the spin states are uncoupled, the coupling appears when the spin-orbit terms are introduced [2]. The coupling in symmetric wells is described by the bulk inversion asymmetry parameter ∆ [2], and the probability ratio of spin-flip to spin-conserving transitions is estimated as a ratio of the energy ∆ to the gap energy. This ratio is small (about 0.05) for our quantum wells. Integrating Eq. (2) over ε, then multiplying Eq. (2) by ε and integrating again, we obtain four balance equations expressing conservation of particles and energy (for details see Supplementary information). They are solved together with the balance equations for the bulk states and boundary conditions expressing conservation of currents and energy fluxes at the edge. We consider the case when particle transfer between edge and bulk occurs only in the narrow regions near the contacts, while in the most part of the edge a dynamical equilibrium is reached, when only spin currents between edge and bulk are flowing. This corresponds to the condition g ≫ 1/L ( g is energy averaged quantity of gε, defined below) and leads to a homogeneous distributions of temperatures and potentials at the edge: the spin- averaged potentials and temperatures linearly depend on x, while the quantities describing spin polarization, δϕ = ϕ1 − ϕ2 and δTe = T1e − T2e, are constants. Only this condition is relevant, because under the opposite condition g ≤ 1/L the particle transfer between edge and bulk cannot have a sizeable influence on the transport. The current carried by a single edge is given as Ie = (e/h)R dε(f1ε− f2ε) = (e/h)δϕ, and the thermal current (energy flux minus µIe/e) along the edge is We = (1/h)R dε(ε− µ)(f1ε − f2ε) = (π2/3h)T δTe, where µ and T are the equilibrium chemical potential and temperature. The total current I = Ibulk + 2Ie and the total thermal current W = Wbulk + 2We are connected to the voltage ∆V and temperature difference ∆T between the contacts by a linear relation (cid:18) I (e/T )W (cid:19) = 2e2 h G(cid:18) ∆V ∆T /e (cid:19) , where we introduced a thermoelectric response matrix G = wσ L M + c"c + Lγ + L 2 (cid:18) w 2σ M −1 + g−1(cid:19)−1#−1 c. (3) (4) The first and the second terms of this matrix describe the bulk and the edge contributions, respectively, and the term in the round braces describes the influence of the edge to bulk leakage on the edge transport. Here and below, σ = σ/(e2/h), σ is the bulk conductivity per spin, L is the distance between the contacts, w is the sample width, and the matrices are defined as γ = (cid:18) γ γI γII (cid:19) , g = (cid:18) g γI gI gI gII (cid:19) , c = (cid:18) 1 0 π2/3 (cid:19) , 0 M = 1 σI σ (cid:18) σ σI σII (cid:19) =(cid:18) 1 eSb eSb e2(κb/σT + S2 b ) (cid:19) . (5) (6) The matrix M describes thermoelectric response in the bulk (for details see Supplementary information) and is expressed through the bulk thermopower S = Sb and electronic thermal conductivity κb. The quantities X, XI , and XII are introduced as the averages X ≡ hXεi = R dε(−∂f (0) ε /∂ε)Xε, XI ≡ hXε(ε − µ)/Ti, and XII ≡ hXε(ε − µ)2/T 2i, where f (0) ε = {exp[(ε − µ)/T ] + 1}−1 is the equilibrium distribution. Equation (3) is written under an additional assumption that spin relaxation length in the bulk exceeds the sample halfwidth w/2. We also neglected energy transfer between the edge states and the lattice, which is justified, according to our estimates, in the samples of submillimeter length. The total thermopower Stot = −(∆V /∆T )I=0 is determined by the ratio of non-diagonal to diagonal elements of the matrix (3), Stot = e−1G12/G11. The total conductance is Gtot = I/∆V = (2e2/h)G11. In the general case, Eq. (3) tells us that under conditions σ < 1, where the edge conductivity can be significant, the contribution due to g is not important (unless gw/2 < σ < 1 which necessarily implies w ≪ L because g ≫ 1/L is assumed). The whole contribution of the term describing edge to bulk leakage in G cannot exceed (L/w)σ M and can be neglected in the case of γL ≫ 1. This is a manifestation of the bottleneck effect described in the Introduction. On the other hand, when σ ≫ 1, there is no need to take the edge transport into account. In the case when Sommerfeld expansion for each of the energy-dependent parameters is valid, i.e., X ≃ cX + (π2/3)T X ′ σx, where X ′ = dXµ/dµ and σx =(cid:18) 0 1 1 0 (cid:19), the Mott relation S = (π2/3e)T (G′/G) is valid as well. Then the second term in Eq. (3) is simplified and leads to the edge state contribution to the conductance and thermopower as follows: Ge = e2 h F −1, F = 1 + γL + ( L w )σgL ( 2L w )σ + gL , π2T F ′ F 3e while the total conductance and thermopower are Se = , Gtot = 2(Ge + Gb), Stot = SeGe + SbGb Ge + Gb , (7) (8) (9) where Gb = σw/L = (e2/h)(w/L)σ is the bulk conductance per spin. Though the theory leading to Eqs. (7) and (8) initially assumes gL ≫ 1, these equations remain formally valid in the opposite limit gL ≪ 1, when the edge state contribution to the conductance is not influenced by the bulk-edge currents and Ge = e2 h (1 + γL)−1. This means that Eqs. (7) and (8) can be used as a reasonable approximation for arbitrary gL. The main results of the our theoretical model is given by equations 7 and 8. One can see that the function F contains two terms: the first depends on the scattering between edges and the second describing the edge to bulk leakage. Two cases can be considered: a) Fermi level is inside of the gap. In this case if the bulk transport is suppressed by localization the conductance and thermopower are governed by the backscattering between the edge states. b) The Fermi level enters conductance (valence ) band. The edge state conductance and thermopower are determined by equations 7 and 8. If (L/w)σ ≪ gL, the presence of the edge to bulk scattering does not strongly affect either the resistance or the thermopower, because: F = 1 + γL + L w σ (10) This also means that even when energy dependence of g is strong, the thermopower sign alteration considered in model [18] does not occur. For the long narrow sample and for high mobility bulk carriers (high conductivity), if (L/w)σ ≫ gL the edge state contribution to function F should be proportional to gL, and the conductance and thermopower should be given by : G = e2 h (1 + γL + g)−1 (11) Se = π2T 3e ! γ ′ + g′ 1 + γL + gL (12) Note that for the limiting case γ ′ ≪ g′ ≪ 0 the thermopower is large and has a positive anomalous sign for electrons ( for conventional 2D and 3D system Seebek coefficient always negative for electrons). This mechanism has been predicted in model [18]. Thus, for observations of the anomalous thermopower long samples with suppressed edge to edge scattering is required. 4. Discussion and comparison with experiment In the ballistic case, the edge state contribution to the thermopower is absent. However, in the experiment we observe thermopower signal in a quasi-ballistic case (Figs. 2 and 4). Generally, it is expected that γ is a smooth function of chemical potential, monotonically decreasing away from the CNP, because the CNP roughly corresponds to the crossing point in the edge-state spectrum, where the transferred momentum is zero, and an elastic scattering rate usually decreases with increasing transferred momentum. This property of the scattering should cause a decrease of the resistance with gate voltage Vg − VCN P and a negative slope of the thermopower near the CNP, as it is observed in experiment. If the energy is counted from the Dirac point, the transferred momentum is q = 2ε/ve. Assuming, for example, that the backscattering rate νε is proportional 0)/[1 + (ε/ε0)2]. The to a Lorentzian function γ0/[1 + (ε/ε0)2], one can find γ ′/γ = −(2µ/ε2 thermopower is a linear function of the Fermi energy near the CNP, negative in the electron part and positive in the hole part. If we assume that the transport near the CNP is dominated by the edge states, then the slope of the thermopower near the CNP is entirely determined by the energy dependence of the backscattering rate. In this region the assumed approximation g = 0 is relevant. 150 100 50 L, m: 0.5 5 20 100 ) / K V ( e S 0 -50 -100 (a) -150 -40 -20 0 20 40 E(meV) 1E-12 1E-13 1E-14 ) 2 K / 2 V h / 2 e ( 2 e S G 1E-15 1E-16 (b) 1E-17 -40 L, m: 0.5 5 20 100 -20 0 E(meV) 20 40 Figure 6. (Color online) (a) Edge states thermopower calculated from eqs.7 and 8 as a function of the Fermi energy for different distances between the probes, T=4.2K. (b) Calculated coefficient GS2 for different distances between the probes as a function of the Fermi energy. Figure 6a shows the thermopower calculated from eqs.7 and 8 taking into account the Lorentzian energy dependence of the backscattering rate for different edge channel lengths L with the following parameters : ε0 = 10meV , γ0 = 1/µm, σ = 0.1. One can see a nonlinear length dependence. A nonmonotonic length dependence has also been observed in our samples (see figure 4c). Figure 6b shows the coefficient GS2 as a function of the energy near the CNP in the edge state transport regime at T=4.2K for different edge channel lengths. The theory (eq.7 and 8) confirms the predictions of the model [18] concerning the dependence on geometry. In practice, however, geometry-related optimization would necessitate the possibility to minimize the thermoconductance of a realistic 2D TI. This is not feasible in our case, since the thermoconductance in our structures is mostly determined by the GaAs substrate. One can see that on approaching the CNP GS2 increases, goes through a maximum, and then reaches its minimum at the CNP, which has also been observed in our experiment ( figure 5b). Note, however, that the theory is simplified and does not consider the asymmetry between electron and hole side near the Dirac point. The model reproduces the key feature of the thermopower, for example, the ambipolar behaviour of the signal, the linear temperature dependence and the nonmonotonic dependence on the length. As we mentioned above, for narrow and long sample, when if (L/w)σ ≫ gL, the contribution of the γ ′ and g′ are equally important. Note , however, that observation of the nonlocal resistance ( figure 5a) clearly demonstrates that the edge state transport inside of the mobility gap (−1.5V < V g < 1.5V ). Let us assume that gL ∼ γL ∼ 1. In this case the mixing between edge states and the bulk becomes important for probes 3-2 (L/w = 10) if σ ≫ 0.1e2/h. From our model (ref.16) we estimated approximately equal value of the bulk conductivity ∼ 0.08e2/). Note, however, that alternatively, this bulk conductivity may lead to bulk mechanism of the thermopower, considered below. We can suggest that the bulk transport is important in our samples even in the vicinity of the CNP, despite the existence of a considerable nonlocal resistance signifying the presence of the edge transport. One would assume that our samples are characterized by a significant disorder level which results in the persistence of the bulk electron transport even inside the gap, so that the bulk component of the conductance exists in the whole gate voltage range. Near the center of the gap, the bulk transport may be of a hopping variety. A signature of such transport would be an exponential temperature dependence of the kind R ∝ exp[(Tc/T )1/3] corresponding to the Mott law for 2D carriers. However, we do not see such a dependence because the bulk conductivity is short-circuited by the edge states. On the other hand, the temperature behavior of the bulk thermopower is not expected to change considerably at the transition from the band transport to the hopping transport. For the band transport Sb ∝ T . For the hopping transport, the thermopower was calculated in Ref. 27. Applied to 2D electrons the result of Ref. 27 can be written as Sb = − λ e (cid:20) π − 2 π (T 2 c T )1/3 + 2π 3 T(cid:21)(cid:18) d ln ρµ dµ (cid:19) (13) where λ is a numerical constant of the order of 1, ρε is the density of states, Tc is a characteristic temperature proportional to 1/(ρµa2 0), and a0 is a localization length of the electron wavefunction. Since the Mott law is only valid for Tc considerably larger than T , the first term in Eq. (10) is important together with the second one. However, in the regime close to the onset of the band transport the localization length becomes large and Tc is of the order of T or smaller than T , so only the second term in Eq. (10) remains important and Sb is linear in T . The presence of the logarithmic derivative of the density of states is quite a general feature, because the conductivity in the hopping transport regime is proportional to the square of the density of states. Any physically reasonable model of the density of states gives a dependence of S qualitatively similar to the experimental one. We approximate the density of states by the 4 3 2 ) 1 - V e m 2 - m c 1 1 0 1 ( -1 Vg(V) 0 1000 experiment 1 (b) (a) 500 ) K V / Eg ( b S 1 holes electrons theory 0 0 -40 -20 0 20 40 -50 E(meV) 0 E(meV) 50 Figure 7. red line- with disorder. experimental thermopower at 4.2 K. (Color online) (a) The calculated density of states: black line-without disorder, (b) The calculated (hopping transport, diffusive mechanism) and function: ρe = ρ0 + ε πA2 , ε > 0 ρh = ρ0 + β ε πA2 , ε < 0 (14) (15) where the energy ε is counted from the center of the gap εg and ρ0 is the constant background density of states describing localized states. The linear function ε/πA2, where A =0.36 eV nm for a well of width d =8 nm, describes the band density of states at ε > εg/2 in the Dirac model with the electron spectrum ε = q(εg/2)2 + (Ak)2. To avoid discontinuities at ε = εg/2, the function ε/πA2 is extended to the region 0 < ε < εg/2. The eq.(12) gives a reasonably good description of the density of states in the electron part of the spectrum. In the hole part, where the dispersion relation is more complicated and may include several subbands, no simple model exists. However, it can be roughly approximated by the same form, with an extra coefficient β (β > 1) describing the electron-hole asymmetry, which leads to Eq. (13). Any physically reasonable model of the density of states results in a dependence of S qualitatively similar to the experimental one. For example, applying a Lorentz spectral function to describe the broadening of electron states, one obtains ρe = 1 πA2 (cid:20)ε arctan ε − ∆/2 δ + ε arctan ε + ∆/2 δ + 1 2 δCε(cid:21) , where Cε = ln (ε − Ec)2 + δ2 (ε − ∆/)2 + δ2! + ln (ε + Ec)2 + δ2 (ε + ∆/2)2 + δ2! , (16) (17) where δ is the broadening energy, Ec is a large cutoff energy. In the hole region (ε < 0) the extra coefficient β ≈ 6 is added, and the density of band states is larger. An example of the density of states calculated according to this expression is shown in Fig.7a with disorder parameters Ec = 150 meV and δ = 6meV . Applying Eq. (11) with a scaling constant λ ≈ π, one may plot the thermopower in the hopping transport regime as shown in Fig. 7b. This calculation is in a reasonable agreement with experiment in the region close to the CNP. The numerical constant is closer to the metallic case. Strictly speaking, neither the hopping nor metallic transport are expected to work well in this intermediate regime. Still, it is important to get the idea of how the thermopower can depend on the electron density around the CNP. Note that the better agreement of the experimental results with equation (11) requires exact knowledge of the bulk conductivity behavior in the gap and hole side regions. The temperature dependence of thermopower does not allow to separate the bulk transport from the edge transport, because Se ∝ T as well. Nevertheless, there are several reasons to argue that the observed thermopower is mostly of the bulk origin. First, the asymmetry of the thermopower signal is much larger than the asymmetry of the resistance peak. The large asymmetry is associated with the bulk transport and is the consequence of electron-hole asymmetry in HgTe quantum wells (the asymmetry increases with the width d in the 2D TI regime d > 6.3 nm). Second, the absolute value of Se, according to Eq. (10), is determined by the energy derivative of γ, which is not expected to be large (γ ′ is small if the backscattering rate is weakly sensitive to the momentum transfer). In contrast, the absolute value of Sb is not small even in the region of hopping transport, where Sb is determined by the energy derivative of the density of states according to Eqs. (14-15). In the transition region between the hopping and band transport, Sb is expected to be large because both the density of states and the bulk conductivity σ strongly depend on the energy in this region. Thus, we expect Se ≪ Sb. The total thermopower in these conditions is reduced to S ≃ Sb/(1 + Ge/Gb). To conclude this chapter we should note that until now only few 2D topological insulators has been discovered: HgTe, InAs/GaSb based quantum wells, and recently W T e2 monolayer. Unfortunately in all these systems e2/2h in the conductance quantization has been observed in micron size devices, which indicates the presence of the backscattering between the counter propagating edge states. Thus, for observation of anomalous sign of the Seebek coefficient, predicted in the paper [18] it is important to develop a technology that reduces the impurity concentration in the bulk. 5. Summary and conclusions In summary, we have studied the thermoelectric power together with the resistance behavior in HgTe quantum wells. The dependence of the thermopower on the gate-controlled carrier density, temperature, and device length has been investigated. The thermopower shows a behavior expected for electron-hole systems, changing sign at the CNP, where the resistance reaches the maximum, and decreasing with carrier density increasing. The hole thermopower is much stronger than the electron one. The temperature dependence of the thermopower is close to linear. Near the CNP, when the Fermi level lies in the gap, the resistance is comparable to or larger than h/2e2. The bulk of the sample is likely to be localized under these conditions, and the resistance is determined mostly by the edge transport. Away from the CNP, the bulk diffusive transport takes place. In contrast, the thermopower appears to be mostly of the bulk origin, regardless to the position of the Fermi level. The transition from the localized states in the gap to the band conductance does not show the anomalies such as strong enhancement and sign variation of the thermopower recently suggested in the theoretical works [17,18]. Our theory of a linear thermoelectric response in 2D TI explains the absence of these anomalies and supports the conclusion about the bulk origin of the observed thermopower. 6. Acknowledgment. The financial support of this work by the Russian Science Foundation (Grant No. 16-12-10041, MBE growth of HgTe QWs, fabrication of the field effect transistors and carrying out of the experiment and data analysis) and FAPESP (Brazil), CNPq (Brazil) is acknowledged. References [1] M. Z. Hasan and C. L. Kane, Rev. Mod. Phys. 82, 2045 (2010); X.-L. Qi, S.-C. Zhang, Rev. Mod. Phys. 83, 1057 (2011). [2] M. Konig, S. Wiedmann, C. Brune, A. Roth, H. Buhmann, L. W. Molenkamp, X.-L. Qi, and S.-C. Zhang, Science 318, 766 (2007). [3] A. Roth, C. Brune, H. Buhmann, L. W. Molenkamp, J. Maciejko, X.-L. Qi, and S.-C. Zhang, Science 325, 294 (2009). [4] E. B. Olshanetsky, Z. D. Kvon, G. M. Gusev, A. D. Levin, O. E. Raichev, N. N. Mikhailov, and S. A. Dvoretsky, Phys.Rev.Lett.114, 126802 (2015). [5] G. M. Gusev, Z. D. Kvon, O.A.Shegai, N. N. Mikhailov, S. A. Dvoretsky and J. C. Portal, Phys. Rev. B 84, 121202(R), (2011). [6] I. Knez, R.-R. Du, and G. Sullivan, Phys. Rev. Lett. 107, 136603 (2011); 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] L. Du, I. Knez, G. Sullivan, and R.-R. Du, Phys. Rev. Lett. 114, 096802 (2015). [8] F. Nichele, A. N. Pal, P. Pietsch, T. Ihn, K. Ensslin, C. Charpentier, and W. Wegscheider, Phys. Rev. Lett. 112, 036802 (2014). [9] K. Suzuki, Y. Harada, K. Onomitsu, and K. Muraki, Phys. Rev. B 91, 245309 (2015). [10] Y. M. Zuev, W. Chang, and P. Kim, Phys. Rev. Lett. 102, 096807 (2009). [11] P. Wei, W. Bao, Y. Pu, C. N. Lau, and J. Shi, Phys. Rev. Lett. 102, 166808 (2009). [12] M. Cutler and N. F. Mott, Phys. Rev. 181, 1336 (1969). [13] E. H. Hwang, E. Rossi, and S. Das Sarma, Phys. Rev. B 80, 235415 (2009). [14] F. Ghahari, H.-Y. Xie, T. Taniguchi, K. Watanabe, M. S. Foster, and P. Kim, Phys. Rev. Lett. 116, 136802 (2016). [15] H.-Y. Xie and M. S. Foster, Phys. Rev. B 93, 195103 (2016). [16] A. Rahim, A. D. Levin, G. M. Gusev, Z. D. Kvon, E. B. Olshanetsky, N. N. Mikhailov, S. A. Dvoretsky, 2D Materials, 2, 044015 (2015). [17] R. Takahashi and S. Murakami, Phys. Rev. B 81, 161302(R) (2010); Semicond. Sci. Technol. 27, 124005 (2012). [18] Y. Xu, Z. Gan, and S.-C. Zhang, Phys. Rev. Lett. 112, 226801 (2014). [19] R. Fletcher, J. C. Maan, K. Ploog, G. Weimann,Phys. Rev. B 33, 7182 (1986). [20] W. E. Chickering, J. P. Eisenstein, L. N. Pfeiffer, and K. W. West, Phys. Rev. B 81, 245319 (2010). [21] J. I. Vayrynen, M. Goldstein, and L. I. Glazman, Phys. Rev. Lett. 110, 216402 (2013). [22] Sanfeng Wu,Valla Fatemi, Quinn D. Gibson, Kenji Watanabe, Takashi Taniguchi, Robert J. Cava, Pablo Jarillo-Herrero, Science 359, 7679 (2018) [23] T. L. Schmidt, S. Rachel, F. von Oppen, and L. I. Glazman, Phys. Rev. Lett. 108, 156402 (2012). [24] M. Kuroda and J. P. Leburton, Phys. Rev. Lett. 101, 256805 (2008). [25] B. A. Bernevig, T. L. Hughes, and S. C. Zhang, Science 314, 1757 (2006). [26] J. Wang, Y. Meir, and Y. Gefen, Phys. Rev. Lett. 118, 046801 (2017). [27] O. E. Parfenov and F. A. Shklyaruk, Semiconductors (St. Petersburg) 41, 1021 (2007). Supplemental material: Thermoelectric transport in two-dimensional topological insulator state based on HgTe quantum well G. M. Gusev,1 O. E. Raichev,2 E. B. Olshanetsky,3 A. D. Levin,1 Z. D. Kvon,3,4 N. N. Mikhailov,3 and S. A. Dvoretsky,3 1Instituto de F´ısica da Universidade de Sao Paulo, 135960-170, Sao Paulo, SP, Brazil 2Institute of Semiconductor Physics, NAS of Ukraine, Prospekt Nauki 41, 03028 Kyiv, Ukraine 3Institute of Semiconductor Physics, Novosibirsk 630090, Russia 4Novosibirsk State University, Novosibirsk, 630090, Russia E-mail: [email protected] Abstract. in HgTe quantum wells in topological insulator regime. In this supplementary, we provide the linear theory of resistance and thermopower 7. Linear theory of resistance and thermopower in HgTe quantum wells in topological insulator regime 7.1. Glossary Let us introduce a bit of terminology. k = 1, 2 - spin numbers. fk - distribution function of the edge states. Fk - isotropic part of the distribution function of the bulk states. These distribution functions are assumed locally Fermi-like ones, because electron-electron scattering controls the distribution within each subsystem. For edge states this statement is proved rigorously in the Fermi liquid theory, in helical states even better (the Luttinger liquid effects are ignored). The coordinate dependence of fk and Fk enters through the parametric variables, electrochemical potentials and temperatures listed below: ϕ = (ϕ1 + ϕ2) /2, δϕ = ϕ1 − ϕ2 (below these definitions of spin-average and spin-polarized quantities apply to all variables). Φk(x, y) - electrochemical potentials of the bulk states. Φ = (Φ1 + Φ2) /2, δΦ = Φ1 − Φ2 ΦkB(x) - electrochemical potentials of the bulk states near the edge. ΦB = (Φ1B + Φ2B) /2, δΦB = Φ1B − Φ2B Tke(x) - effective temperatures of the edge states. Te = (T1e + T2e) /2, δTe = T1e − T2e Tk(x, y) - effective temperatures of the bulk states. T = (T1 + T2) /2, δT = T1 − T2 TkB(x) - effective temperatures of the bulk states near the edge. TB = (T1B + T2B) /2, δTB = T1B − T2B Energy-dependent quantities: σε = e2Dερε- bulk energy-dependent conductivity (per spin); Dε is the diffusion coefficient, ρε- is the density of states (per spin). In zero B, bulk states are degenerate in spin, so σε, Dε , ρε and are the same for both states. Total bulk conductivity is 2 σε: χε = νs ε ρε- spin relaxation rate in the bulk multiplied by the density of states. γε = νe−e /ve - inverse length of free path for edge to edge elastic scattering. ε gε = νe−b E /ve- inverse length of free path for edge to bulk elastic scattering. υe - velocity of edge states (constant). In the linear theory, there appear the following averages for any energy-dependent quantity XE: ε where f (0) ∂ε (cid:19) Xε X ≡ hXεi = R dε(cid:18)− ∂f (0) XI ≡ hXε(ε − µ)/T0i , XII ≡ (cid:10)Xε(ε − µ)2/T 2 0(cid:11) ε = 1/ [exp [(ε − µ)/T0] + 1] is the equilibrium distribution, µ and T0 are the chemical potential and the temperature in equilibrium. For example, when substituting the energy-dependent conductivity for Xε, one gets the averaged conductivity σ = hσεi. Further, σI = eβ defines the thermoelectric coefficientβ and bulk thermopower S = Sb = β/σ. Also, σII defines the thermal conductivity κ according to κ = (cid:0)σII − σ2 I /σ(cid:1) T0/e2 Other parameters are obtained in a similar way. of T (this means, for example, σ ≫ σ′T0 ≫ σ′′T 2 are possible: XI ≃ (cid:0)π2/3(cid:1) X ′T0, XII ≃ (cid:0)π2/3(cid:1) X, where X ′ denotes the energy derivative of Xε at ε = µ. This leads to Mott relation S ≃ (cid:0)π2/3e(cid:1) (σ′/σ) T0 and Widemann-Frantz law ε ≃ σII T0/e2 ≃ (cid:0)π2/3e2(cid:1) σT0 ≡ κ0. If σ, X, γ, and g are weakly changing in the energy interval 0 and so on), the Sommerfeld expansions 7.2. Currents In this subsection we describe the basic equations for current, which we used in the theory. Bulk current density per spin: jk = −e−1σ (∇Φk + eS∇Tk) Bulk energy density flow per spin: Hk = (Φ + eST )jk/e − κ∇Tk (18) (19) (22) (23) Since the theory is linear, Φ and T in this expression can be replaced by µ and T0. Edge current (for a single edge). It is assumed that the state 1 moves in positive direction and the state 2 in negative direction: Edge energy density flow: Ie = e h Z dε (f1ε − f2ε) = e h δϕ He = 1 h Z dεε (f1ε − f2ε) = 1 h "ϕδϕ + π2 3 TeδTe# (20) (21) In the linear theory ϕ and Te can be replaced by µ and T0. Total electric current along x in the Hall bar of width w: J = Z w 0 dyj + 2Ie Total energy flow: H = Z w 0 dyH + 2He where j = j1 + j2, H = H1 + H2. J is independent of x. H is independent of x under condition of no dissipation of energy by phonons. However, both Ie and He can depend on coordinate x in the presence of particle exchange between the bulk and the edge. 7.3. Equations In this subsection we describe the basic equations derived from kinetic equations, which we used in the theory. 1. For edge states (propagating along x): ∂ϕ ∂x = −(γ + g/2)δϕ − (γI + gI /2) δTe + (g/2)δΦB + (gI /2) δTB, ∂δϕ ∂x = −2g (ϕ − ΦB) − 2gI (Te − TB) , π2 3 ∂Te ∂x = − (γI + gI /2) δϕ − (γII + gII /2) δTe + (gI /2) δΦB + (gII /2) δTB, π2 3 ∂δTe ∂x = −2gI (ϕ − ΦB) − 2gII (Te − TB) . (24) (25) (26) (27) Four 1-st order equations. Require 4 boundary confitions. These equations are derived from kinetic equations. The first pair satisfies the current (particle flow) conservation, the second one for energy flow conservation. In the second pair, the dissipation of edge state energy by the lattice (through electron-phonon interaction) is neglected. Estimates was made to justify this neglect in 10µm size samples. In centimeter size samples one needs to take the dissipation into account. 2. In the bulk: Particle flow and energy flow conservation impose equations for bulk temperatures and electrochemical potentials. The total currents and energy flows satisfy obvious relations ∇ · j = 0, ∇ · H + P ph = 0, (28) where P ph is the power lost to phonons. More detailed analysis is needed for partial flows jk and Gk. There must be taken into account particle exchange between states 1 and 2 due to spin-flip scattering and the energy exchange between these states due to spin- flip scattering, electron-electron interaction, and electron-phonon interaction [corresponding isotropic collision integrals are J s k (ε). Writing the collision integrals for spin-flip as J s 1,2(ε) = ∓Ree (T1 − T2) , and P ph ε (F1ε − F2ε) and assuming R dερεεJ ee 2 νs k (ε) = Rph (Tk − TL) (TL is lattice temperature), we obtain: k (ε), J ee k (ε) , and J ph 1,2(ε) = ∓ 1 k = −R dερεεJ ph ∇2Φ + eS∇2T = 0, κ∇2T = Rph (T − TL) , ∇2δΦ + eS∇2δT = Γ2[δΦ + ηδT ], S∇2δΦ + ξ∇2δT = Γ2[ηδΦ + ζδT ]. (29) (30) (31) (32) (33) (34) where ξ = e2(cid:0)κ + σS2T0(cid:1) / (σT0) = σII/σ is a dimensionless quantity which is equal to π2/3 if Widemann-Frantz law works. Next, Γ = √e2χ/σ is the inverse length of spin relaxation, η = χI /χ, and ζ = χII /χ + (2Ree + Rph) /χI0 are dimensionless quantities related to spin relaxation and energy exchange. Four 2-nd order equations. Require 8 boundary conditions. 7.4. Boundary conditions In this subsection we describe the boundary conditions, which we used in the theory. 1. Outside the contacts. The total currents and energy flow normal to the boundary should be zero: n · jkboundary = e h n · (Hk − µjk/e)boundary = [g (ϕk − ΦkB) + gI (Tke − TkB)] 1 h T0 [gI (ϕk − ΦkB) + gII (Tke − TkB)] where n is a unit vector perpendicular to the boundary and directed inside the sample. Since k takes two values, there are 4 boundary conditions per each boundary. Assuming the lower boundary at y = 0 (because we already assumed that the edge states are propagating along x), we rewrite the boundary conditions as: + eS ∂Φ ∂y + ξ ∂y (cid:20) ∂Φ (cid:20)eS (cid:20) ∂δΦ (cid:20)eS ∂δΦ ∂y ∂y ∂T ∂T ∂y(cid:21)y=0 ∂y(cid:21)y=0 ∂y (cid:21)y=0 ∂y (cid:21)y=0 ∂δT + eS ∂δT + ξ = −σ−1 [g (ϕ − ΦB) + gI (Te − TB)] , = −σ−1 [gI (ϕ − ΦB) + gII (Te − TB)] , = −σ−1 [g (δϕ − δΦB) + gI (δTe − δTB)] , = −σ−1 [gI (δϕ − δΦB) + gII (δTe − δTB)] , (35) (36) (37) (38) where σ = σ/G0 defines conductivity in the units of fundamental conductance quantum G0 = e2/h. 2. At the contacts. If contact a is at x = xa, y = ya, contact b is at x = xb, y = yb and the electrons in the state 1 move from a to b (in the state 2 move from b to a): Edge: ϕ1 (xa) = eVa, ϕ2 (xb) = eVb T1e (xa) = Ta, T2e (xb) = Tb, (39) (40) Total 4 boundary conditions. Bulk (full spin mixing and thermalization at each contact): Φ1 (xa, ya) = Φ2 (xa, ya) = eVa, T1 (xa, ya) = T2 (xa, ya) = Ta, Φ1 (xb, yb) = Φ2 (xb, yb) = eVb, T1 (xb, yb) = T2 (xb, yb) = Tb, (41) (42) where Va, Ta, Vb, Tb are voltages and temperatures at the contacts. 4 boundary conditions per contact. 7.5. Solutions In this subsection we show the solutions of the differential equations with boundary conditions described above. Note, that we show the final solutions and discussion in the main text. In general, numerical solution is required. If edge to bulk scattering is ignored (g = 0), analytical solution is possible because edge and bulk subsystems are not coupled. Assume two contacts at x = 0 and x = L, with voltage and temperature differences between them ∆V and ∆T , and boundaries at y = 0 and y = w. We can assume a constant temperature gradient (the same for electrons and lattice), so the bulk solution is straightforward (δΦ = 0, δT = 0, Φ = Φ(0) − e∆V x/L, T = T (0) − ∆T x/L) and the current is J = 2σw(∆V + S∆T )/L. For edge states the solution is δϕ = const, δTe = const, and the current is where Ge and Se are edge conductance and thermopower: Ie = Ge (∆V + Se∆T ) , Ge = 1 Re , Re = h e2 "1 + γL − (γI L)2 π2/3 + γII L# (43) (44) Se = 1 e γI L 1 + (3/π2) γII L (45) Parameter γε describes edge backscattering. It is very unlikely that this parameter strongly changes with energy at the temperature scale. Therefore, one can apply the Sommerfeld expansion, which leads to: Ge = e2 h 1 1 + γL Se = γ ′L 1 + γL , π2T 3e (46) (47) e/Ge). Se and Ge satisfy the Mott relation Se = −(cid:0)π2T /3e(cid:1) (G′ If edge to bulk scattering is persistent (finite g) the problem is much more complicated. For local transport (current passes along x) it can be solved analytically in the case gL ≫ 1, because edge to bulk particle transfer in this case occurs only in the short regions near the contacts, while in the most part of the sample edge a dynamical equilibrium is reached, when only spin currents between edge and bulk are flowing. This assumes a homogeneous distributions of temperatures and potentials at the edge: the spin-average quantities linearly depend on x, ϕ(x) = ΦB(x) = φ0 − φx/L, Te(x) = TB(x) = t0 − tx/L, while the spin-polarized quantities δϕ, δΦB, δTe and δTB are constants. The constants φ0 and t are nor essential, while φ = e∆V −δϕ and t = ∆T − δTe according to the boundary conditions (22) and (23). The behavior of ΦB and TB is consistent with homogeneous behavior of the spin-average bulk potentials, which follows from the first pair of bulk eqations [Eqs. (12), (13)] if local thermal equilibrium with the lattice takes place, T (x, y) = TL(x, y). Thus, in the bulk we need to solve only the second pair of equations, where δΦ and δT depend only on the transverse coordinate y. Moreover, this depenence is antisymmetric with respect to y = w/2 (symmetry axis of the sample), so it is enough to apply boundary conditions for one edge (y = 0) only. The bulk solution is searched in the form: δΦ(y) = C1 sinh λ1(y − w/2) + C2 sinh λ2(y − w/2), δT (y) = C1b1 sinh λ1(y − w/2) + C2b2 sinh λ2(y − w/2), where λ1 and λ2 are positive solutions of the biquadratic equation and λ4(cid:16)ξ − e2S2(cid:17) − λ2Γ2(ζ + ξ − 2eSη) + Γ4(cid:16)ζ − η2(cid:17) = 0, bj = − j − Γ2 λ2 eSλ2 j − Γ2η From Eqs. (31) - (34), one can find the matrix relation: (cid:18) ∇yδΦ ∇yδT (cid:19)y=0 = − A(cid:18) δΦB δTB (cid:19) with A = 1 b2 − b1 (cid:18) u1b2 − u2b1 b1b2 (u1 − u2) u2b2 − u2b1 (cid:19) , u2 − u1 (48) (49) (50) (51) (52) (53) Thus, one has a set of 4 linear equations given below in the matrix form: uj = λj coth (λjw/2) (cid:16)1 + g−1 M A(cid:17)(cid:18) δΦB δTB (cid:19) = (cid:18) δϕ δTe (cid:19) , and [c + Lγ + (L/2)g](cid:18) δϕ δTe (cid:19) − (L/2)g(cid:18) δΦB δTB (cid:19) = (cid:18) where we defined the following 2x2 matrices: e∆V (cid:0)π2/3(cid:1) ∆T (cid:19) , M = σ(cid:18) 1 eSb eSb ξ (cid:19) , g = (cid:18) g gI gI gII (cid:19) , γ = (cid:18) γ γI γII (cid:19) , c = (cid:18) 1 0 π2/3 (cid:19) , γI 0 (54) (55) (56) (57) and g−1 means inverted matrix. A more detailed analysis is done below in the limit of narrow samples, when λjw ≪ 1. This means that neither direct spin-flip scattering nor energy exchange between bulk states 1 and 2 can influence the distribution of potential and temperature of bulk states. In this approximation, A = (2/w)1. The expressions for potential and temperature differences in the edge states is where K = c + Lγ + (cid:18) δϕ δTe (cid:19) = K −1(cid:18) e∆V (cid:0)π2/3(cid:1) ∆T (cid:19) , 2 h(w/2) M −1 + g−1i−1 L (58) (59) . The problem is reduced to finding the matrix inversion of K. Then, expressing the current Eq. (3) through δϕ and δTe found from Eqs. (41) and (42), one can find the conductance and thermopower for edge states. Expressing the energy flow Eq. (4), one finds also the thermal conductivity. The matrix inversion leads to large and complicated expressions not given here. Instead, we present the results obtained in approximation of slow energy dependence of σ g, and γ, when Sommerfeld expansions are valid for all quantities, σ g, and γ. In this case, Ge ≃ e2 h F −1, F = 1 + γL + (L/w)σgL (2L/w)σ + gL , In the main text we compare the theoretical results with experiment. Se ≃ π2T 3e . F ′ F (60) (61)
1704.02879	2	1704	2017-04-12T09:19:33	Hole weak anti-localization in a strained-Ge surface quantum well	[ "cond-mat.mes-hall" ]	We report a magneto-transport study of a two-dimensional hole gas confined to a strained Ge quantum well grown on a relaxed Si0.2Ge0.8 virtual substrate. The conductivity of the hole gas measured as a function of a perpendicular magnetic field exhibits a zero-field peak resulting from weak anti-localization. The peak develops and becomes stronger upon increasing the hole density by means of a top gate electrode. This behavior is consistent with a Rashba-type spin-orbit coupling whose strength is proportional to the perpendicular electric field, and hence to the carrier density. By fitting the weak anti-localization peak to a model including a dominant cubic spin-orbit coupling, we extract the characteristic transport time scales and a spin splitting energy of ~1 meV. Finally, we observe a weak anti-localization peak also for magnetic fields parallel to the quantum well and attribute this finding to a combined effect of surface roughness, Zeeman splitting, and virtual occupation of higher-energy hole subbands.	cond-mat.mes-hall	cond-mat	Hole weak anti-localization in a strained-Ge surface quantum well R. Mizokuchi,1, 2 P. Torresani,1, 2 R. Maurand,1, 2 M. Myronov,3 and S. De Franceschi1, 2 1CEA, INAC-PHELIQS, F-38000 Grenoble, France 2University Grenoble Alpes, F-38000 Grenoble, France 3Department of Physics, University of Warwick, Coventry CV4 7AL, United Kingdom We report a magneto-transport study of a two-dimensional hole gas confined to a strained Ge quantum well grown on a relaxed Si0.2Ge0.8 virtual substrate. The conductivity of the hole gas measured as a function of a perpendicular magnetic field exhibits a zero-field peak resulting from weak anti-localization. The peak develops and becomes stronger upon increasing the hole density by means of a top gate electrode. This behavior is consistent with a Rashba-type spin-orbit cou- pling whose strength is proportional to the perpendicular electric field, and hence to the carrier density. By fitting the weak anti-localization peak to a model including a dominant cubic spin-orbit coupling, we extract the characteristic transport time scales and a spin splitting energy of ∼1 meV. Finally, we observe a weak anti-localization peak also for magnetic fields parallel to the quantum well and attribute this finding to a combined effect of surface roughness, Zeeman splitting, and virtual occupation of higher-energy hole subbands. Hole spins in p-type SiGe-based heterostructures are promising candidates for quantum spintronic applica- tions [1, 2]. They are expected to display a relatively small in-plane effective mass [3, 4], favoring lateral con- finement, as well as long spin coherence times [5], stem- ming from a reduced hyperfine coupling (natural Ge is predominantly constituted of isotopes with zero nuclear spin and holes are less coupled to nuclear spins due to the p-wave symmetry of their Bloch states [6]). In addition, low-dimensional, SiGe-based structures benefit from a strong and electrically tunable spin orbit coupling [4, 7 -- 11]. This property could be exploited for purely electrical spin control [5, 12 -- 14]. Finally, hybrid superconductor- semiconductor devices based on strained Ge quantum wells confining holes should provide a favorable platform for the development of topologically protected qubits based on Majorana fermions of parafermions [15, 16]. Here we consider a SiGe-based heterostructure with a compressively strained Ge quantum quantum well (QW) at its surface. This heterostructure presents two main ad- vantages: 1) in a metal-oxide-semiconductor field-effect transistor (MOSFET) device layout, it allows for an ef- ficient gating of the accumulated two-dimensional hole gas (2DHG); 2) the surface position of the QW en- ables an easier fabrication of low-resistive contacts to the 2DHG. The strained SiGe heterostructure was grown on a 200 mm Si(001) substrate by means of reduced pres- sure chemical vapor deposition (RP-CVD). Growth was realized using an industrial-type, mass-production sys- tem (ASM Epsilon 2000 RP-CVD), which is a horizontal, cold-wall, single wafer, load-lock reactor with a lamp- heated graphite susceptor in a quartz tube. RP-CVD offers the major advantage of unprecedented wafer scal- ability and is nowadays routinely used by leading com- panies in the semiconductor industry to grow epitaxial layers on Si wafers of up to 300 mm diameter. The het- erostructures, shown schematically in Fig. 1.a, consists of a 3 µm thick reverse linearly graded, fully relaxed Si0.2Ge0.8/Ge/Si(001) virtual substrate with a 32-nm- thick strained Ge QW surface layer. This is a typical design for surface channel structures employed in mod- ern MOSFET devices. The full structure was grown in a single process without any external treatment. The sur- face of the Si wafers was cleaned by an in situ thermal bake in H2 ambient at high temperature, above 1000◦C. The Ge epilayer was grown from a commercially avail- able and widely used germane (GeH4) gas precursor at a relatively low substrate temperature (<450◦C), as it is known that the growth temperature of the compressively strained Ge epilayers has to be sufficiently low to sup- press surface roughening and retain compressive strain in the epilayers. Further details of materials growth and characterization are described elsewhere [17]. The same epitaxial growth technology resulted in the creation of strained Ge QW heterostructures with superior low- and room-temperature electronic properties [18 -- 20] enabling the observation of various quantum phenomena includ- ing the fractional quantum Hall effect [21], mesoscopic effects due to spin-orbit interaction[2, 22 -- 25] and tera- hertz quantum Hall effect [26]. The studied devices have a Hall-bar geometry defined by a top-gate electrode operated in accumulation mode (Fig. 1.b). Due to the absence of intentional doping, the Ge QW contains no carrier at low temperature. A suffi- ciently negative voltage applied to the top gate induces the accumulation of a 2DHG in the QW region under- neath. Device fabrication involves the following steps : a relatively large (tens of microns wide), 55-nm-thick mesa structure is initially defined by optical lithography and dry etching in Cl2 plasma ; ohmic contacts are succes- sively fabricated using optical lithography, followed by Ar etching (to remove the residual oxide) and Pt depo- sition in an e-beam evaporator system; 30 nm of Al2O3 are deposited everywhere using atomic layer deposition at 250◦C ; finally, the Hall-bar-shaped top gate accu- mulation electrode is defined by e-beam lithography and deposition of 60 nm of Ti/Au. Magneto-transport measurements were performed in a 3He cryostat with a base temperature of 300 mK. Lon- gitudinal (ρXX) and Hall (ρXY ) resistivities were mea- 7 1 0 2 r p A 2 1 ] l l a h - s e m . t a m - d n o c [ 2 v 9 7 8 2 0 . 4 0 7 1 : v i X r a 2 FIG. 1. a) Schematic of the heterostructure. The 2DHG (blue) lays on top of the 32 nm Ge layer. b) Optical image of the Hall bar devices. The blue line highlights the mesa and the white dotted lines the Pt contacts. We measure the transverse Hall voltage (VH) and the longitudinal channel voltage (Vch) from which we extract Hall resistivity and channel resistivity respectively. The directions of the applied fields B⊥ and B// are also indicated. c) Channel resistivity ρXX (red) and Hall resistivity ρXY (blue) as a function of out of plane magnetic field. Channel resistivity shows a dip at low field which is a signature of weak anti-localization. d) Mobility µ (red) and carrier density nhole (blue) as a function of accumulation gate voltage Vtg. sured as a function of magnetic field, B⊥, perpendicu- lar to the 2DHG, and top gate voltage (Vtg), controlling the carrier density. Examples of such traces are given in Fig. 1.c. From Hall resistivity we extracted the hole mobility (µ) and carrier density (nhole) ranging from 800 to 2400 cm2/Vs and from 1.3 to 1.7 ×1011 cm−2 respec- tively, for top gate voltages verying between -4.1 and - 4.4V (the threshold Vtg for the onset of conduction is at -3.8V, as shown in Fig. 1.d). The mobility is much lower than the one reported in other strained Ge heterostruc- tures [27]. This difference is likely due to the presence of charge traps at the Ge/Al2O3 interface. Following basic Hall-effect characterization we now turn to a more in-depth investigation of the magneto- transport properties. In Fig. 1.c, the longitudinal re- sistivity (red trace) exhibits a pronounced dip at zero magnetic field. Such a dip is a characteristic signature of weak anti-localization, a mesoscopic phenomenon as- sociated with spin-orbit coupling [28]. At zero magnetic field the latter leads to a reduced backscattering result- ing in a resistivity minimum. This quantum interference a) Traces of the weak anti-localization contribu- FIG. 2. tion to the channel conductivity ∆σW AL as a function of B⊥ for different accumulation gate voltages and carrier densities from 1.3 × 1011 cm−2 (top trace) to 1.7 × 1011 cm−2 (bottom trace, traces are offset for better visibility). The weak anti- localization peaks emerges as carrier density is increased. b) Evolution of scattering time τtr (red crosses), phase relaxation time τϕ (blue circles) and spin relaxation time τSO (blue tri- angles) as a function of carrier density. c) Evolution of the spin splitting energy ∆so as a function of carrier density. effect is suppressed by a magnetic field perpendicular to the 2DHG, accounting for the observed resistivity dip at B⊥ = 0. This phenomenon is further investigated in Fig. 2.a, where the longitudinal conductivity is now plotted as a function of B⊥ and for a range of Vtg values, after having removed the feature-less back-ground contribution from classical Drüde conductivity. As a matter of fact, ∆σW AL represents the quantum correction resulting from weak anti-localization. Interestingly, this data set shows that the weak anti-localization peak develops upon increas- ing the gate voltage and, correspondingly, the hole den- sity nhole and the perpendicular electric field in the QW. This trend suggests that the spin-orbit coupling respon- sible for the weak anti-localization effect is of a Rashba type. In our strained QW system, where holes occupy the first heavy-hole subband, we expect the dominant Rashba component to be cubic in the in-plane momen- tum. [22, 23, 29, 30]. The weak anti-localization peak can be fitted to the formula [31]: 13.011.0-60B⊥[T]2001.81.2-4400-410021Vtg[mV]vchvHb)c)a)d)B⊥B//420-2-1.00.01.0B [T]a)642τφ,τSO [ps]∆SO [meV]1.601.40nhole [1011 cm-2]∆σWAL [µSsq]b)1.401.60nhole [1011 cm-2]c)nhole=1.3 x 1011 cm-2nhole=1.7 x 1011 cm-22π2{Ψ(1 ∆σW AL(B⊥) = e2 ) − 1 2Ψ(1 + Bϕ + 2 BSO B⊥ B⊥ − 1 2 ln ( Bϕ B⊥ 2 + Bϕ B⊥ 2 + Bϕ B⊥ + 2BSO B⊥ + BSO B⊥ ) − ln ( Bϕ B⊥ ) + 1 2 ln ( Bϕ B⊥ 2Ψ(1 ) + 1 2 + BSO ) B⊥ )} (1) where Ψ(X) is the digamma function, Bϕ is the phase coherence field and BSO is the characteristic field associ- ated with the Rashba spin orbit coupling. From the fitting parameters Bϕ and BSO we can ex- tract the phase coherence time τϕ and the spin relaxation time τso with τi = m∗/4πµnholeBi, i being either ϕ or SO and m∗ being the effective in-plane hole mass (from earlier studies in similar heterostructures [22, 23, 32, 33], we assumed m∗ = 0.1 m0 where m0 is the bare elec- tron mass). We note that the large width of the ob- served weak anti-localization peak is consistent with the relatively small values obtained for the scattering time (τtr = m∗µ/e). These values, as well as those for τϕ, τSO are displayed as a function of carrier density in Fig. 2.b. The evo- lution of these characteristic time scales with respect to nhole provides a hint on the underlying mechanism for spin relaxation. If spin relaxation were due to impurity scattering (Elliot-Yafet mechanism [34, 35]), then τSO should increase with τtr and decrease with the carrier density ( τso ∝ τtr/n2 hole). This does not correspond to the observed trend. On the other hand, if spin relaxation occurred in between scattering events, due to spin-orbit- induced rotation (Dyakonov-Perel mechanism [36]), the spin relaxation time should decrease with τtr and with the spin splitting energy ∆SO (τso ∝ 1/(τtr × ∆2 SO), where ∆SO ≈ α3Ezk3 F where α3 is the cubic Rashba coupling, Ez the vertical electric field and kF the Fermi wave number). Our experimental finding is consistent with this second scenario, which allows us to deduce the the spin splitting energy and its dependence on the car- rier density (see Fig. 2.c). The obtained values of the spin splitting energy are around 1 meV, i.e. comparable to those obtained in similar heterostructures and using different experimen- tal methods [22, 23, 29]. To further investigate the nature of the zero-field conductivity enhancement, magneto-transport measure- ments were performed also with the magnetic field ap- plied in the plane of the 2DHG, as indicated in Fig. 3. To first order, an in-plane magnetic field is not expected to break the effect of weak anti-localization, because it pro- duces no flux through the time-reversed back-scattering trajectories. Contrary to this expectation, the longitudi- nal conductivity measured as a function of the in-plane magnetic field, B//, does exhibit a clear zero-field peak. The characteristic half width at half maximum of the peak is ∼ 0.7 T, i.e. several times larger than in the case of perpendicular field. 3 We can rule out the possibility of a misalignment of the magnetic field with respect to the plane of the 2DHG. In fact, from a simultaneous measurement of the Hall resis- tivity, also shown in Fig. 3, we estimate a misalignment of only 2◦. Therefore, the out-of-plane component of the applied field is far too small to explain the observed con- ductivity peak. FIG. 3. Black trace: Hall resistivity ρXY as a function of in plane magnetic field. The small dependence on field results from a small perpendicular field component. We estimate an angle of only 2◦ between B// and the chip plane. Black circles: quantum correction to channel conductivity ∆σW AL revealing a weak anti-localization peak. The blue dashed line and the red solid line are fits to the model from Minkov et al. [37] without and with the addition of a B6 // term respectively. Instead, the effect can be ascribed to the finite thick- ness of the 2DHG. This type of problem was theoret- ically investigated by Minkov et al.[37]. According to this work, the weak anti-localization correction to the magneto-conductivity can be expressed as: ∆σW AL(B//) = e2 4π2 2 ln (cid:20) (cid:19) (cid:18) Bϕ + 2BSO + ∆r (cid:18) Bϕ + ∆r (cid:18) Bϕ + BSO + ∆r (cid:19) (cid:18) Bϕ + ∆r + ∆s (cid:19)(cid:21) (cid:18) Bϕ (cid:19) Bϕ + BSO Bϕ + 2BSO +S − ln − S + ln Bϕ (cid:19) (2) BSO BSO where ∆r and ∆s are B//-dependent corrections to Bϕ taking into account the effect of surface roughness and Zeeman splitting, respectively. Following Ref. [37], we assume ∆r = rB2 //. The S(x) function in Eq. 2 can be explicitly written as: // and ∆s = sB2 S(x) = 8√ 7 + 16x arctan 7 + 16x 1 − 2x − πΘ(1 − 2x) (3) where Θ is the Heaviside step function. For the ef- fective fields BSO and Bϕ we take the values extracted (cid:20) (cid:18)√ (cid:19) (cid:21) B// [T]ρXY[Ω]∆σWAL[µSsq]40-430-34000from the previously discussed magneto-transport mea- surements in perpendicular magnetic field, for the same carrier density, i.e. BSO = 170 mT and Bϕ= 19 mT. // + q × B6 //, i.e. ∆r = r × B2 The dotted blue line in Fig. 3 is a fit to equation 2 using the proportionality factors r and s as fitting pa- rameters. The fit shows only moderate agreement with the data. A significantly improved fit can be obtained by introducing in the expression of ∆r a second orbital term proportional to B6 //, with the additional fitting parameter q. This second term de- scribes the time reversal symmetry breaking due to mag- netic field via the virtual occupation of unoccupied higher energy subbands[38 -- 40]. The new fit, shown by a solid red line in Fig. 3, is in remarkably good agreement with the experimental data set over the entire magnetic-field range. Following Ref. [39], the value of the fit parameter q can be related to the effective thickness d of the 2DHG, . We find d ∼ 14 nm. Taking into account the thickness of the Ge QW (32 nm) and the gate-induced band bending, this is a perfectly realistic estimate and a validity check for the adopted model. i.e. d ∼(cid:16) q×Φ5 (cid:17)1/14 4π2n2 hole 0 In conclusion, we have reported magneto-transport measurements of a 2DHG confined to a compressively strained Ge QW on the surface of a relaxed Si0.2Ge0.8 virtual substrate. The 2DHG is formed by gate-induced hole accumulation up to carrier densities of the order of 1011 cm−2. 4 We find that the hole mobility is highly reduced as compared to similar heterostructures where the QW is buried well below the surface. This can be explained by a high density of traps at the Ge/Al2O3 interface, which is in line with the fact that germanium native oxide is known to be of poor quality. In order to enhance hole mobility, a higher interface quality would be required. There exist possible solutions to tackle this problem [41], which could be explored in forthcoming studies. Despite the relatively low mobility, weak anti- localization is observed, exhibiting a strong dependence on the carrier density, which is directly related to the perpendicular electric field. This points to a spin-orbit coupling of the Rashba type, with an expected dominant cubic component. The estimated characteristic times τtr, τso and τϕ, as well as the spin-splitting energy ∆SO are consistent with values measured in buried Ge QWs [22, 23, 29]. Finally, we find that weak anti-localization can as well be suppressed by an in-plane magnetic field, reflecting the finite thickness of the 2DHG and a contri- bution from Zeeman effect, surface roughness, and virtual inter-subband scattering processes. The authors would like to thank M. Houzet, J. Meyer, Y.-M. Niquet, S. Oda, M. Sanquer and Z. Zeng for the fruitful discussions. We acknowledge financial support from the Agence Nationale de la Recherche, through the TOPONANO project, from the EU through the ERC grant No. 280043, and from the Fondation Nanosciences, Grenoble. [1] F. A. Zwanenburg, A. S. Dzurak, A. Morello, M. Y. Sim- mons, L. C. L. Hollenberg, G. Klimeck, S. Rogge, S. N. Coppersmith, and M. A. Eriksson, Rev. Mod. Phys. 85, 961 (2013). [2] C. Morrison and M. Myronov, physica status solidi (a) 213, 2809 (2016). [3] K. Sawano, K. Toyama, R. Masutomi, T. Okamoto, N. Usami, K. Arimoto, K. Nakagawa, and Y. Shiraki, Applied Physics Letters 95, 2109 (2009). [4] C. Morrison, C. Casteleiro, D. R. Leadley, and M. My- ronov, Applied Physics Letters 109, 102103 (2016), http://dx.doi.org/10.1063/1.4962432. [5] A. P. Higginbotham, T. W. Larsen, J. Yao, H. Yan, C. M. Lieber, C. M. Marcus, and F. Kuemmeth, Nano Letters 14, 3582 (2014). [6] C. Testelin, F. Bernardot, B. Eble, and M. Chamarro, Physical Review B 79, 195440 (2009). [7] H. Watzinger, C. Kloeffel, L. Vukušić, M. D. Rossell, V. Sessi, J. Kukučka, R. Kirchschlager, E. Lausecker, A. Truhlar, M. Glaser, A. Rastelli, and G. Katsaros, Nano A. Fuhrer, D. Loss, Letters 27656760, http://dx.doi.org/10.1021/acs.nanolett.6b02715. [8] N. Ares, V. N. Golovach, G. Katsaros, M. Stoffel, F. Four- nel, L. I. Glazman, O. G. Schmidt, and S. De Franceschi, Phys. Rev. Lett. 110, 046602 (2013). (2016), pMID: 6879 16, [9] C. Kloeffel, M. Trif, and D. Loss, Phys. Rev. B 84, 195314 (2011). [10] C. Morrison, P. Wiśniewski, J. Foronda, D. R. Leadley, 105, Applied http://dx.doi.org/10.1063/1.4901107. Physics Letters S. D. Rhead, and M. Myronov, (2014), 182401 [11] A. Zarassi, Z. Su, J. Danon, J. Schwenderling, M. Ho- cevar, B.-M. Nguyen, J. Yoo, S. A. Dayeh, and S. M. Frolov, arXiv preprint arXiv:1610.04596 (2016). [12] N. Ares, G. Katsaros, V. N. Golovach, J. J. Zhang, A. Prager, L. I. Glazman, O. G. Schmidt, and S. D. Franceschi, Applied Physics Letters 103, 263113 (2013). [13] M. Brauns, J. Ridderbos, A. Li, W. G. van der Wiel, E. P. A. M. Bakkers, and F. A. Zwanenburg, Applied Physics Letters 109, 143113 (2016). [14] R. Maurand, X. Jehl, D. Kotekar-Patil, A. Corna, H. Bo- huslavskyi, R. Laviéville, L. Hutin, S. Barraud, M. Vinet, M. Sanquer, et al., Nature Communications 7, 13575 (2016). [15] F. Maier, J. Klinovaja, and D. Loss, Phys. Rev. B 90, 195421 (2014). [16] Z. Su, A. Zarassi, B.-M. Nguyen, J. Yoo, S. A. Dayeh, and S. M. Frolov, arXiv preprint arXiv:1610.03010 (2016). [17] M. Myronov, A. Dobbie, V. A. Shah, X.-C. Liu, V. H. Nguyen, and D. R. Leadley, Electrochemical and Solid- State Letters 13, H388 (2010). [18] A. Dobbie, M. Myronov, R. J. H. Morris, A. H. A. Has- san, M. J. Prest, V. A. Shah, E. H. C. Parker, T. E. Whall, and D. R. Leadley, Applied Physics Letters 101, 5 172108 (2012), http://dx.doi.org/10.1063/1.4763476. [19] M. Myronov, C. Morrison, J. Halpin, S. Rhead, C. Casteleiro, J. Foronda, V. A. Shah, and D. Leadley, Japanese Journal of Applied Physics 53, 04EH02 (2014). [20] M. Myronov, C. Morrison, J. Halpin, S. Rhead, J. Foronda, in 2014 7th Interna- tional Silicon-Germanium Technology and Device Meet- ing (ISTDM) (2014) pp. 11 -- 12. and D. Leadley, [21] Q. Shi, M. A. Zudov, C. Morrison, and M. Myronov, Phys. Rev. B 91, 241303 (2015). [22] C. Morrison, P. Wiśniewski, S. Rhead, J. Foronda, D. R. Leadley, and M. Myronov, Applied Physics Letters 105, 182401 (2014). [23] M. Failla, M. Myronov, C. Morrison, D. Leadley, and J. Lloyd-Hughes, Physical Review B 92, 045303 (2015). [24] C. Morrison, J. Foronda, P. Wiśniewski, S. Rhead, D. Leadley, and M. Myronov, Thin Solid Films 602, 84 (2016), the 9th International Conference on Silicon Epitaxy and Heterostructures. [25] J. Foronda, C. Morrison, J. E. Halpin, S. D. Rhead, and M. Myronov, Journal of Physics: Condensed Matter 27, 022201 (2015). [26] M. Failla, J. Keller, G. Scalari, C. Maissen, J. Faist, C. Reichl, W. Wegscheider, O. J. Newell, D. R. Leadley, M. Myronov, and J. Lloyd-Hughes, New Journal of Physics 18, 113036 (2016). [27] A. Dobbie, M. Myronov, R. J. Morris, A. Hassan, M. J. Prest, V. Shah, E. H. Parker, T. E. Whall, and D. R. Leadley, Applied Physics Letters 101, 172108 (2012). [28] W. Knap, C. Skierbiszewski, A. Zduniak, E. Litwin- Staszewska, D. Bertho, F. Kobbi, J. L. Robert, G. E. Pikus, F. G. Pikus, S. V. Iordanskii, V. Mosser, K. Zekentes, and Y. B. Lyanda-Geller, Phys. Rev. B 53, 3912 (1996). [29] R. Moriya, K. Sawano, Y. Hoshi, S. Masubuchi, Y. Shi- raki, A. Wild, C. Neumann, G. Abstreiter, D. Bougeard, T. Koga, and T. Machida, Physical review letters 113, 086601 (2014). [30] R. Winkler, D. Culcer, S. J. Papadakis, B. Habib, and M. Shayegan, Semiconductor Science and Technology 23, 114017 (2008). [31] S. Iordanskii, Y. B. Lyanda-Geller, and G. Pikus, ZhETF Pisma Redaktsiiu 60, 199 (1994). [32] M. A. Zudov, O. A. Mironov, Q. A. Ebner, P. D. Martin, Q. Shi, and D. R. Leadley, Phys. Rev. B 89, 125401 (2014). [33] D. Laroche, S.-H. Huang, Y. Chuang, J.-Y. Li, C. W. Liu, and T. M. Lu, Applied Physics Letters 108, 233504 (2016), http://dx.doi.org/10.1063/1.4953399. [34] R. J. Elliott, Physical Review 96, 266 (1954). [35] Y. Yafet, Physics Letters A 98, 287 (1983). [36] M. Dyakonov and V. Perel, Soviet Physics Solid State, Ussr 13, 3023 (1972). [37] G. Minkov, A. Germanenko, O. Rut, A. Sherstobitov, L. Golub, B. Zvonkov, and M. Willander, Physical Re- view B 70, 155323 (2004). [38] V. I. Fal'ko, Journal of Physics: Condensed Matter 2, 3797 (1990). [39] J. S. Meyer, A. Altland, and B. L. Altshuler, Phys. Rev. Lett. 89, 206601 (2002). [40] D. M. Zumbühl, J. B. Miller, C. M. Marcus, K. Camp- man, and A. C. Gossard, Phys. Rev. Lett. 89, 276803 (2002). [41] R. Zhang, T. Iwasaki, N. Taoka, M. Takenaka, and S. Takagi, IEEE Transactions on Electron Devices 59, 335 (2012).
1203.3848	1	1203	2012-03-17T09:20:25	Topological Superfluid Transition Induced by Periodically Driven Optical Lattice	[ "cond-mat.mes-hall", "cond-mat.other" ]	We propose a scenario to create topological superfluid in a periodically driven two-dimensional square optical lattice. We study the phase diagram of a spin-orbit coupled s-wave pairing superfluid in a periodically driven two-dimensional square optical lattice. We find that a phase transition from a trivial superfluid to a topological superfluid occurs when the potentials of the optical lattices are periodically changed. The topological phase is called Floquet topological superfluid and can host Majorana fermions.	cond-mat.mes-hall	cond-mat	Topological Superfluid Transition Induced by Periodically Driven Optical Lattice Guocai Liu,1 Ningning Hao,2, 3 Shi-Liang Zhu,4, 5 and W. M. Liu2 1School of Science, Hebei University of Science and Technology, Shijiazhuang 050018, China 2Beijing National Laboratory for Condensed Matter Physics, Institute of Physics, Chinese Academy of Science, Beijing 100190, China 3Department of Physics, Purdue University, West Lafayette, Indiana 47907, USA 4Laboratory of Quantum Information Technology, School of Physics and Telecommunication Engineering, South China Normal University, Guangzhou, China 5Center for Quantum Information, IIIS, Tsinghua University, Beijing, China We propose a scenario to create topological superfluid in a periodically driven two-dimensional square optical lattice. We study the phase diagram of a spin-orbit coupled s-wave pairing superfluid in a periodically driven two-dimensional square optical lattice. We find that a phase transition from a trivial superfluid to a topological superfluid occurs when the potentials of the optical lattices are periodically changed. The topological phase is called Floquet topological superfluid and can host Majorana fermions. PACS numbers: 03.75.Lm, 05.30.Pr, 71.70.Ej I. INTRODUCTION Optical lattice system has gradually become a promis- ing platform to study many-body quantum systems be- cause of lots of significant advances in cold-atom exper- iments [1]. In particular, recent theoretical and experi- mental progress in laser-induced-gauge-field [2 -- 7] makes it a hot spot to study topological quantum states in cold atoms system [8 -- 11]. Recently, topological quan- tum states have attracted considerable interest in con- densed physics; however, subject to the compounds' nat- ural properties [12], we have to rely on serendipity in looking for topological materials in solid-state structures [13 -- 16]. In contrast, one can engineer the Hamiltonian of an optical lattice system to realize variant quantum phase states [17 -- 19]. In this paper, we show that periodically driven per- turbations may give rise to a phase transition from a trivial superfluid to a topological one, which carries the hallmark with topological protected gapless edges on the boundaries of the system. Time-periodic depen- dent Hamiltonian can be described by Floquet's theo- rem, which is used to explain quantized adiabatic pump- ing phenomena [20 -- 23]. Recently, it demonstrated that the phase transition from a superfluid to a Mott insula- tor in one-dimensional Bose-Hubbard model can be in- duced by a periodically driven optical lattice [24]. We extend this phase transition mechanism to explore the topological phase transition in a two-dimensional opti- cal lattice. We study the phase diagram of a spin-orbit coupled s-wave pairing superfluid in a periodically driven two-dimensional (2D) square optical lattice. We find that a topological phase transition from a trivial superfluid to a topological superfluid can be induced in periodi- cally modulated optical lattices. The topological phase is called Floquet topological superfluid [25 -- 29] and can host Floquet Majorana fermions. It was proposed that a topological phase can be realized in a BCS s-wave super- fluid of ultracold fermionic atoms in the presence of both a Rashba spin-orbit (SO) interaction and a large perpen- dicular Zeeman field [30 -- 33]; however, the Rashba spin- orbit coupling and a large perpendicular Zeeman field are hard to be simultaneously realized for cold fermionic atoms [32, 33]. We will prove that if one replaces the Zee- man field by a periodically driven optical lattice [24], a spin-orbit coupled BCS s-wave superfluid will still allow a realization of topological superfluid through modifying the oscillating amplitude (or modulation strength) of op- tical lattice. Therefore, we provide an alternative method to create an important topological superfluid which can host Majorana fermions. The paper is organized as follows: In Sec. II, we in- troduce the s-wave superfluid model in a square optical lattice in the presence of both a Rashba SO coupling and a periodically modulated optical lattice potential. A Zeeman-magnetic-field-like term will be derived un- der the first order approximation; In Sec. III we present a two-band approximation and explain the topological phase transition at the Γ point in the first Brillouin zone (BZ). At last, we give a brief summary in Sec. IV. II. MODEL The tight-binding Hamiltonian, which describes an s- wave superfluid of neutral fermionic atoms in a 2D optical square lattice, is given by H (t) = H0 + Hd (t) , (1) where H0 = −tXhiji + µXi c† i cj − iλXhiji i ci + UXi c† c† i↑c† i↓ci↓ci↑, c† i (cid:16)σ × dij(cid:17)z cj (2) and tions with the scalar product given by Hd (t) = K (t) ·Xi ric† i ci. (3) hh··ii = 1 T Z T 0 dth··i , 2 (6) i↑, c† i =(cid:16)c† i↓(cid:17) with c† Here t is the hopping amplitude between the nearest neighbor link hi, ji, c† iα (ciα) denot- ing the creation (annihilation) operator of a fermionic atom with pseudospin α (up or down) on lattice site i. The second term in Eq. (2) represents a Rashba SO cou- pling interaction which can be obtain by laser-induced- gauge-field method, λ is the coupling coefficient, σ are the Pauli matrices and dij is a unit vector along the bond that connects site j to i. µ is the chemical po- tential and U < 0 denotes an on-site attractive inter- action which is easy to obtain via an s-wave Feshbach resonance in cold atom system. The oscillating Hamil- tonian Hd, with K (t) = K (cos (ωt) , sin (ωt)), mimics a monochromatic electric dipole potential with frequency ω and amplitude K. This term can be realized experimen- tally by periodically shifting the position of a mirror em- ployed to generate the standing laser waves along x- and y-directions, and transforming to the comoving frame of reference [24]. We choose t=1 as the energy unit and the distance a between the nearest sites as the length unit throughout this paper. It was demonstrated in Ref. [30] that the Hamiltonian H0 in a mean field approxima- tion and combination with a perpendicular Zeeman field can support a topological superfluid. On the other hand, replaced the Hamiltonian H0 with an one-dimensional Bose-Hubbard Hamiltonian, it was shown in Ref. [24] that a phase transition from a superfluid to a Mott insu- lator can be induced by Hd in its one-dimensional form. When a Hamiltonian of quantum system has a peri- odic dependence on time, i.e., H (t)=H (t + T ) with pe- riod T =2π/ω, the Hamiltonian satisfies the discrete time translational symmetry, t → t + T , which can been de- scribed by Floquet's theorem [24, 34, 35]. Floquet's the- orem tell us that the Schrodinger equation with time- periodic dependent Hamiltonian has a complete set of solutions with the form ψn (t)i=un (t)i exp (−iεnt/). Here, the periodic function un (t)i = un (t + T )i, an analog of Bloch states known from spatially periodic crys- tals, satisfies the eigenvalue equation [H (t) − i∂t]un (t)i = εn un (t)i . (4) We call H (t)=H (t) − i∂t as the Floquet Hamiltonian and the eigenvalues εn as quasienergies which are defined modulo the frequency ω=2π/T . The Floquet basis is (7) (8) (9) i↑c† the Bessel i} , m′ c† Dh{n′ i} , m′ c† i.e., by the usual scalar product h··i combined with time- averaging. Hence, the quasienergies are obtained by computing the matrix elements of the Floquet operator H (t) − i∂t in the basis (5) with respect to the scalar product (6), and diagonalizing. By a straightforward calculation, we can obtain the matrix elements of some operators in Floquet Hamiltonian H (t): Dh{n′ iαcjα′ {ni} , miE = e−i(m′−m)θij Jm′−m (zij) , i} , m′ c† iαciα {ni} , miE = niαδm,m′, i↓ci↓ci↑ {ni} , miE = ni↑ni↓δm,m′ , Dh{n′ function of where Jm′−m (zij) the (m′ − m)th order and zij= K ij. Here, xij = (ri)x − (rj)x, yij = (ri)y − (rj)y and tan θij = xij /yij. In the above matrix, the diagonal block of the Floquet Hamiltonian, H(mm), i.e., the subspace with n photons and the non-diagonal blocks H(m′m) with m′ 6= m correspond to the interaction be- tween different subspaces [27]. For sufficiently high fre- quencies, we can argue, from Eq. (7-9), that the driven system (1) behaves similar to the undriven system (2), but with the tunneling matrix element t and the SO cou- pling λ of the latter being replaced by the effective ma- trix element t ∼ tJ0 (zij) and λ ∼ λJ0 (zij), respectively. Now, suppose that we enhance the modulation strength K, then we have to consider the coupling of other pho- ton sectors. For simplify, we only consider coefficient of subspace with n = 1 photon on the subspace with n = 0 photon. When K is strong enough but still satisfy zij << 1, the system has the effective Hamiltonian [29] is the n-photon sector, ωqx2 ij + y2 Hef f = H(00) + 1 ω (cid:2)H−1,H+1(cid:3) . (10) Here, H−1 (H+1) denotes the non-diagonal block with m′−m = −1 (+1) around the 0-photon sector. According to Eqs. (7-9), we can obtain J0 (zij) c† i cj − iλXhiji J0 (zij) c† i(cid:16)σ × dij(cid:17)z cj i ω Z t {ni} , mi = {ni}i exp"− (5) where {ni}i indicates a Fock state with ni particles on the ith site, and m accounts for the zone structure [24], consist of an extended Hilbert space of T -periodic func- dt′K (t′)·Xi −∞ rini+imωt# , c† H(00) = −tXhiji + µXi H−1 = −tXhiji − iλXhiji i ci + ψsXi (cid:16)c† i↑c† i↓ + H.c.(cid:17) , eiθij J−1 (zij) c† i cj eiθij J−1 (zij) c† i(cid:16)σ × dij(cid:17)z (11) cj, (12) e−iθij J+1 (zij) c† i cj the Hamiltonian (10) in square lattice system can be rewritten in the momentum space as 3 H+1 = −tXhiji − iλXhiji e−iθij J+1 (zij) c† cj. (13) i(cid:16)σ × dij(cid:17)z In the derivation of Eq. (11), we have made a mean field approximation and ψs is the gap function. By using the Fourier transform of atomic operators c† iσ, i.e., c† iσ = 1 √N Xk c† kσe−ik·Ri, (14) Hef f =Xk ψ+ k (Hef f (k)ψk, (15) where we have defined the four-component basis operator ψk=(ck↑, ck↓, c† −k↓)T. The effective Hamiltonian in momentum space is given by −k↑, c† Hef f (k) =  0 ψs εk − Γ (k, z) 2λJ0 (z) α (k) 2λJ0 (z) α∗ (k) εk + Γ (k, z) 0 −ψs ψs 0 −ε−k + Γ (k, z) 2λJ0 (z) α∗ (k) 2λJ0 (z) α (k) −ε−k − Γ (k, z) ,   (16) −ψs 0 = where α (k) = 16λ2J+1(z)J−1(z) sin ky + i sin kx, z cos kx cos ky, Ka ω , Γ (k, z) εk = −2tJ0 (z) (cos kx + cos ky) − µ. = ω and Following the method outlined in Ref. Hamiltonian [30], one can obtain a "dual" HD (k) =  ψs − Γ (k, z) 2λJ0 (z) α (k) 2λJ0 (z) α∗ (k) −ψs + Γ (k, z) 0 −εk εk 0 0 εk −εk 0 ψs + Γ (k, z) −2λJ0 (z) α∗ (k) −2λJ0 (z) α (k) −ψs − Γ (k, z) ,   (17) where the unitary transformation HD (k) = DHef f D† with D = 1 √2  1 0 0 1 0 1 −1 0 0 1 1 0 −1 0 0 1 .   It is easy to obtain the eigenvalues of Eq. (17) with E1 = −pf1 + f 2; E2 = −pf1 − f2, E3 = +pf1 − f2; E4 = +pf1 + f2, where we have defined each other. Next, we discuss the levels E1 and E2 (or E3 and E4). These two levels can touch each other if only the following two relations Γ2 (k, z) = 0 and J 2 0 (z) ε2 k(cid:0)sin2 kx + sin2 ky(cid:1) = 0. are simultaneously satisfied. From Γ2 (k, z) = 0, we have kx = ± π 2 in the first BZ. So, when J0 (z) 6= 0, we have 2 or ky = ± π −µ = 2tJ0 (z) (cos kx + cos ky) . k + ψ2 s(cid:1) + 4J 2 0 (z) λ2 α (k)2 , f1 = Γ2 (k, z) +(cid:0)ε2 f2 = 2qΓ2 (k, z) (ε2 It is obvious that if only ψs 6= 0, i.e., the system lies in the superfluid phase, the energy levels E1 and E4, denoting the lowest and the highest band, will not touch k α (k)2. s ) + 4λ2J 2 0 (z) ε2 k + ψ2 the and E2 touch each other When −µ > 2tJ0 (z), (or E3 and E4) will not ever. E1 levels E1 and E2 for Else, when − 2tJ0 (z) ≤ −µ ≤ 2tJ0 (z), touch each 2tJ0(z)(cid:17)(cid:17) and 2(cid:17). In the following, we 2 , ky = arccos(cid:16) −µ other at points (cid:16)kx = ± π (cid:16)kx = arccos(cid:16) −µ 2tJ0(z)(cid:17) , ky = ± π and E4) will (or E3 only consider the case, −µ > 2tJ0 (z), which means bands E1 and E4 will be off away from the other two levels E2 and E3, and then the topological properties of bands E1 and E4 will not change if we vary some pa- rameters. Therefore, we will only consider the topologi- cal properties of band E2 and E3, which have chances to contact each other at the high symmetry points, Ki=1,...,4 = (0, 0) ; (0, π) ; (π, 0) ; (π, π) in the first BZ, when satisfying the condition Γ2 (z) = ε2 Ki + ψ2 s , (18) because band-gap closing is an essential condition for the topological characteristic changes. We denote Γ2 = Γ2 (z). Consider- Ki ing z << 1 and −µ > 2tJ0 (z), the two bands can only (Ki, z) = (cid:0)16λ2J 2 +1 (z) /ω(cid:1)2 touch at point K1 = (0, 0) when varying the parameter z. 4 III. TOPOLOGICAL PHASE TRANSITION We now study the topological properties of these two bands by two-band approximation at point K1 = (0, 0) in the first BZ. We will not consider the other points since the gaps at other high symmetry points will not shut down, leaving no influence to the topological changes. To have a basic idea of the topological features of the system, we explore it by using a two-band approximation at point K1 in the first BZ. We expand the Hamiltonian (17) at point K1 and obtain K1 (q) = HD  2λJ0 (z) q+ ψs + Γ (z) 2λJ0 (z) q− − (ψs + Γ (z)) − (4tJ0 (z) + µ) 4tJ0 (z) + µ − (4tJ0 (z) + µ) 0 0 0 4tJ0 (z) + µ 0 ψs − Γ (z) −2λJ0 (z) q− −2λJ0 (z) q+ − (ψs − Γ (z))   (19) with q±=qy ± iqx. Because ψs < 0 and Γ (z) > 0, when taking 4tJ0 (z) + µ ≪ 1, we can see that the major contribution to bands E2 and E3 comes from the up- diagonal sector in above matrix and thus we may treat the others as a perturbation. Under this condition, we obtain an effective two-band Hamiltonian given by HK1 ef f (q) = 2λJ0 (z) qyσx − 2λJ0 (z) qxσy + M (z) σz, (20) where the corresponding mass term M (z)=ψs + Γ (z) + (4tJ0 (z) + µ)2 / (ψs − Γ (z)) and σν=x,y,z the Pauli ma- Let M (z)=0, we obtain the gapless condi- trices. (18) again at K1 point. Eq.(20) can be tion Eq. written as HK1 ef f (q) = σ · d, where the vector d = {2λJ0 (z) qy,−2λJ0 (z) qx, M (z)}. The topological fea- tures of the system can be characterized by the winding number ( first Chern number) of the Berry phase gauge field C = 1 ) in the first Bril- louin zone, where d = d/d. When M (z) 6= 0, it is straightforward to obtain the winding number for the ef- fective system described by Eq. (20), i.e., 4π R dkxR dky ∂kx × ∂ d d · ( ∂ d ∂ky C = 1 2 sign (M (z)) . (21) This non-integral winding number appears since the devi- ations from this two-band approximation model at large momenta are not included in the above calculation of the winding number. So it can not be directly related to the topological features of the system; however, the change in the winding numbers is independent of the large-momentum contribution [36]. Let us discuss the change of the topological properties of superfluid system when we adjust the oscillating amplitude K of optical lat- tice. It is obvious that the initial non-driven system is in a trivial state which corresponds to z = 0 and M (z) < 0. Now, we apply the driven field to the system and make M (z) < 0 to M (z) > 0, the change in Chern number is: 1 ∆C = 2(cid:2)sign(cid:0)M (z)>0(cid:1) − sign(cid:0)M (z)<0(cid:1)(cid:3) = +1. (22) So, we get a topological superfluid state with C = +1 for M (z) > 0 . It is notable that gapless chiral edge states are usu- ally the hallmark of a topological system. Therefore, to further prove the above argument, we show the phase diagram and the band structures of the effective Hamil- tonian (10) in a striped geometry in Fig. 1. In Fig. 1(a), we plot the mass M (z) as a function of z, which can be adjusted by changing the modulation strength K. The point B denotes M (z) = 0 with z = 0.1320, where the bands E2 and E3 contact each other at Γ point. The band structures in a striped geometry with 60 sites in x direction are shown in (b), (c) and (d) corresponding to the points A (z = 0), B (z = 0.132), and C (z = 0.2) in Fig.(a), respectively. In the numerical calculation, we take the typical parameters t = 1, λ = 0.6, ψs = −0.5, ω = 0.05, and µ = −4. It is clear that there is no edge state in the region I where the mass M (z) < 0 with 0 ≤ z < 0.1320, so it is a trivial superfluid. In contrast, there are a pair of edge states in the region II where the mass M (z) > 0 with z > 0.1320, so it is a topological superfluid. There is a phase transition from a trivial superfluid to a topological superfluid occurs at 5 ψs → ψseimθ, which is similar with the case in the topo- logical superconductor [38]. Both cases give the similar Majorana fermion obviously confirmed from the Eq. (16) and Eq. (17) connected by the unitary transformation D. The zero mode solutions of the Majorana fermion can be obtained from the Bogoliubov-de Gennes (BdG) equa- tion, which has the similar form compared with that in Ref. [30]. Moreover, such Majorana fermion excitations can be detected by the standard Raman spectroscopy [33, 39]. IV. CONCLUSION In summary, we have discussed the topological super- fluid phase transition in a periodically driven square op- tical lattice. By using Floquet's theorem, we find that a Floquet topological superfluid will be created when the two-dimensional square optical lattice potentials are pe- riodically driven. This topological phase is interesting in hosting a Majorana fermion excitation which can be detected by Raman spectroscopy in cold atom system. Therefore we propose a novel scenario to create Majo- rana fermions which may play a key role in topological quantum computation. FIG. 1: The phase diagram and the band structures of the system. (a) The mass M (z) as a function of z. The region I with (z) < 0 is a trivial superfluid, while the region II with M (z) > 0 is a topological superfluid. The band structures of the effective Hamiltonian (10) in a striped geometry with 60 sites in x direction are shown in (b), (c) and (d) corresponding to the points A (z = 0), B (z = 0.132), and C (z = 0.2) in (a), respectively. Other parameters t = 1, λ = 0.6, ψs = −0.5, ω = 0.05, and µ = −4. the point B where the gap is closed. Generally, there exists Majorana Fermionic excitation bounded with the vortex structure in the nontrivial topo- logical superfluid phase. Hence, we can obtain the 2-D Foquet Majorana fermions [25] if there have the vortex structures in our system. The vortex structure can be produced from two different routes, one of which can be realized through the phase twist of the SO-produced lasers: λ → λeimθ with m the vorticity [30]. Another route is that the vortex structure can come from ini- tial rotation of the atomic cloud [37]. Then the vortex structure is coupled with the superfluid order parameter: Acknowledgments G. Liu was supported by NSF of China (No.11147171), S. L. Zhu was supported in part by the NBRPC (No.2011CBA00302), the SKPBRC (No.2011CB922104), and NSF of China (No.11125417). This work was also supported by the NKBRSFC under grants 2011CB921502, 2012CB821305, 2009CB930701, Nos. 10934010, 2010CB922904, NSFC under grants Nos. 60978019, grants Nos. 11061160490 and 1386-N-HKU748/10. and NSFC-RGC under [1] M. Lewenstein, A. Sanpera, V. Ahufinger, B. Damski, A. Porto, and I. B. Spielman, Nature 462, 628 (2009). Sen(De), U. Sen, Adv. Phys. 56, 243 (2007). [2] J. Ruseckas, G. Juzeli¯unas, P. Ohberg, and M. Fleis- chhauer, Phys. Rev. Lett. 95, 010404 (2005). [3] K. Osterloh, M. Baig, L. Santos, P. Zoller, and M. Lewen- stein, Phys. Rev. Lett. 95, 010403 (2005). [4] G. Juzeli¯unas, J. Ruseckas, P. Ohberg, and M. Fleis- chhauer, Phys. Rev. A 73, 025602 (2006). [5] S. L. Zhu, H. Fu, C. J. Wu, S. C. Zhang, and L. M. Duan, Phys. Rev. Lett. 97, 240401 (2006); S. L. Zhu, D. W. Zhang, and Z. D. Wang, Phys. Rev. Lett. 102, 210403 (2009). [6] Y. J. Lin, R. L. Compton, A. R. Perry, W. D. Phillips, J.V. Porto, and I. B. Spielman, Phys. Rev. Lett. 102, 130401 (2009). [7] Y. J. Lin, R. L. Compton, K. Jim´enez-Garc´ıa, J. V. [8] T. D. Stanescu, V. Galitski, J. Y. Vaishnav, C. W. Clark, and S. Das Sarma, Phys. Rev. A 79, 053639 (2009). [9] A. Bermudez, N. Goldman, A. Kubasiak, M. Lewenstein and M. A. Martin-Delgado, New J. Phys. 12, 033041 (2010). [10] I. I. Satija, D. C. Dakin, J. Y. Vaishnav, and C. W. Clark, Phys. Rev. A 77, 043410 (2008). [11] X. J. Liu, X. Liu, C. Wu, and J. Sinova, Phys. Rev. A 81, 033622 (2010). [12] T. D. Stanescu, V. Galitski, and S. D. Sarma, arXiv:0912.3559v1. [13] S. Murakami, Phys. Rev. Lett. 97, 236805 (2006). [14] M. Konig, S. Wiedmann, C. Breune, A. Roth, H. Buh- mann, L. W. Molenkamp, X. L. Qi, S. C. Zhang, Science 318, 766 (2007). [15] D. Hsieh, D. Qian, L. Wray, Y. Xia, Y. S. Hor, R. J. Cava 6 and M. Z. Hasan, Nature 452, 970 (2008). [27] J. I. Inoue and A. Tanaka, Phys. Rev. Lett. 105, 017401 [16] H. Zhang, C. X. Liu, X. L. Qi, X. Dai, Z. Fang, and S. C. Zhang, Nature physics 5, 438 (2009); X. L. Qi, R. Li, J. Zang, S. C. Zhang, Science 323, 1184 (2009). [17] N. Goldman, I. Satija, P. Nikolic, A. Bermudez, M. A. Martin-Delgado, M. Lewenstein, and I. B. Spielman, arXiv: 1002.0219v2; A. Bermudez, L. Mazza, M. Rizzi, N. Goldman, M. Lewenstein, and M. A. Martin-Delgado, arXiv: 1004.5101v1. (2010). [28] N. H. Lindner, G. Refael and V. Galitski, Nature Physics 7, 490 (2011). [29] T. Kitagawa, L. Fu, E. Demler, T. Oka and A. Brataas, arXiv: 1104.4636v1. [30] M. Sato, Y. Takahashi, and S. Fujimoto, Phys. Rev. Lett. 103, 020401 (2009). [31] C. Zhang, S. Tewari, R. M. Lutchyn, and S. Das Sarma, [18] L. B. Shao, S. L. Zhu, L. Sheng, D.Y. Xing, and Z. D. Phys. Rev. Lett. 101, 160401 (2008) Wang, Phys. Rev. Lett. 101, 246810 (2008). [19] G. Liu, S. L. Zhu, S. Jiang, F. Sun, and W. M. Liu, Phys. [32] C. Zhang, Phys. Rev. A 82, 021607(R) (2010). [33] S. L. Zhu, L. B. Shao, Z. D. Wang, and L. M. Duan, Rev. A 82, 053605 (2010). [20] L. Fu and C. L. Kane, Phys. Rev. B 74, 195312 (2006). [21] J. I. Inoue, Phys. Rev. B 81, 125412 (2010). [22] D. J. Thouless, Phys. Rev. B 27, 6083 (1983). [23] Q. Niu and D. J. Thouless, J. Phys. A 17, 2453 (1984); S. L. Zhu and Z. D. Wang, Phys. Rev. B 65, 155313 (2002) [24] A. Eckardt, C. Weiss, and M. Holthaus, Phys. Rev. Lett. 95, 260404 (2005). [25] L. Jiang, T. Kitagawa, J. Alicea, A. R. Akhmerov, D. Pekker, G. Refael, J. I. Cirac, E. Demler, M. D. Lukin, and P. Zoller, Phys. Rev. Lett. 106, 220402 (2011). Phys. Rev. Lett. 106, 100404 (2011). [34] J. H. Shirley, Phys. Rev. 138, B979 (1965). [35] M. Grifoni, P. Hanggi, Physics Reports 304, 229 (1998). [36] X. L. Qi and S. C. Zhang, Rev. Mod. Phys. 83, 1057 (2011). [37] M. W. Zwierlein, J. R. Abo-Shaeer, A. Schirotzek, C. H. Schunck, W. Ketterle, Nature (London) 435, 1047 (2005). [38] J. D. Sau, R. M. Lutchyn, S. Tewari, and S. Das Sarma, Phys. Rev. Lett. 104, 040502 (2010). [39] S. Tewari, S. Das Sarma, C. Nayak, C. Zhang, and P. [26] T. Kitagawa, E. Berg, M. Rudner, and E. Demler, Phys. Zoller, Phys. Rev. Lett. 98, 010506 (2007). Rev. B 82, 235114 (2010).
1904.04666	2	1904	2019-05-02T01:13:44	Quantum Anomalous Valley Hall Effect for Bosons	[ "cond-mat.mes-hall" ]	We predict the emergence and quantization of the valley Hall current of indirect excitons residing in noncentrosymmetric two-dimensional materials in the absence of an external magnetic field. Thus we report on the quantum anomalous valley Hall effect (QAVHE) for bosons. The crucial ingredients of its existence are the finite anomalous group velocity of exciton motion and the presence of the Bose-Einstein condensate, from which the particles are pushed out by external illumination to contribute to the stationary nonequilibrium current, which exhibits a quantized behavior as a function of the frequency of light.	cond-mat.mes-hall	cond-mat	Quantum Anomalous Valley Hall Effect for Bosons V. M. Kovalev and I. G. Savenko A.V. Rzhanov Institute of Semiconductor Physics, Siberian Branch of Russian Academy of Sciences, Novosibirsk 630090, Russia Center for Theoretical Physics of Complex Systems, Institute for Basic Science (IBS), Daejeon 34126, Korea (Dated: May 3, 2019) We predict the emergence and quantization of the valley Hall current of indirect excitons residing in noncentrosymmetric two-dimensional materials in the absence of an external magnetic field. Thus we report on the quantum anomalous valley Hall effect (QAVHE) for bosons. The crucial ingredients of its existence are the finite anomalous group velocity of exciton motion and the presence of the Bose-Einstein condensate, from which the particles are pushed out by external illumination to contribute to the stationary nonequilibrium current, which exhibits a quantized behavior as a function of the frequency of light. The essence of the ordinary Hall effect, which discov- ery is tracing back to 1879, is in the emergence of particle transport in the transverse direction under an external magnetic field [1]. If the magnetic field becomes suffi- ciently strong, the Hall conductance takes on the quan- tized values [2, 3]; such is the nature of the quantum Hall effect (QHE). In general, all the Hall effects are based on the time reversal symmetry breaking employing either an external field, or internal magnetization, or spin-orbital coupling. For instance, there takes place the anomalous Hall effect (AHE) in certain materials due to the pres- ence of magnetic polarization and spin-orbit splitting [4]. The AHE also has its quantized counterpart [5]. Anomalous transport of carriers of charge in low- dimensional structures is a captivating fundamental and a simultaneously application-oriented subject of research, which is recently attracting growing attention [6]. One of the conceptual problems here is the study of the AHE resulting from the topological properties of the carriers of charge [7] due to the emergence of the Berry phase asso- ciated with closed trajectories in the parameter space [8]. It turns out that the Berry curvature plays a central role in various transport phenomena in condensed matter physics [9 -- 11], one of the most intriguing of which is the valley Hall effect (VHE) [12 -- 15]. In intuitively apprehensible quasiclassical language, the origin of the anomalous Hall transport lies in a spe- cific term in the expression of the particle group veloc- ity, r = ∂pε(p) − p × Ωp, where ε(p) is the dispersion relation and Ωp is the Berry curvature, expressed via the periodic amplitude of the Bloch wave function u(cid:105): Ωp = ∂p × (cid:104)ui∂pu(cid:105). For a charged particle subject to an electric field E(t), the quasiclassical equation of motion reads p = eE(t), giving the anomalous velocity term −eE(t) × Ωp acting as a pseudomagnetic field in momentum space. This very term is responsible for the anomalous Hall currents. We pose a question: Is there any influence of the Berry curvature on dynamics of overall neutral com- pound particles like bosonic excitations in semiconductor crystals, called excitons (electron-hole pairs)? For the first glance, it seems to vanish due to the insensitivity of their center-of-mass motion to external uniform electro- magnetic fields. In the meantime, transport of neutral particles is of great interest, and also it would be highly beneficial to map the topological properties of fermions on bosons. Excitons not only play an important role in optical phenomena occurring in low-dimensional struc- tures [16, 17] but also they can be widely used in appli- cations [18]. One of the advantages of excitonic devices is the possibility of exciton-photon coupling, which is es- sential for the manipulation of optical signals, including the opportunity to form coherent quasi-condensate states with the vanishing sensitivity to a disorder present in any nanostructure. In this Letter, we address this question and show that in the presence of the two-dimensional Bose-Einstein quasi-condensate (later condensate) and finite Berry phase, the exciton current density becomes quantized in the absence of an external magnetic field; we call this phenomenon the Quantum Anomalous Valley Hall Ef- fect (QAVHE) for bosons. It manifests itself in exciton current steplike behavior as a function of frequency of external polarized light, which breaks the time reversal symmetry. Excitons can form from electrons and holes sharing a common quantum well (direct excitons) or separated from each other, thus residing in a double quantum well (indirect excitons). The main feature of indirect exci- tons is that they possess a built-in dipole moment. The most well-studied system of materials for double quan- tum wells and thus for the study of optical and trans- port properties of excitons is GaAs and its alloys with aluminum and indium. The discovery of novel purely 2D materials, such as monomolecular layers of transition metal dichalcogenides (TMDs), benchmarked new chap- ter in both the direct [19 -- 23] and indirect [24 -- 27] exci- ton physics. These monolayers have a complex Brillouin zone [13], containing two inequivalent valleys located at the opposite edges. Due to sizeable intervalley distance 2 govern the transport of an individual exciton, where R and P are the exciton center-of-mass coordinate and momentum. Each exciton in a TMD consists of two Coulomb-interacting Dirac particles. The two-valley structure of MoS2 Brillouin zone per- mits the existence of two types of indirect excitons. If an electron from the top layer and a hole of the bottom layer inhabit the same valley, such an exciton is consid- ered to be direct in momentum space (we will refer to it as DMSE). The exciton is indirect in momentum space (we will use the abbreviation IMSE) if the particles reside in different valleys (Fig. 2). Below we consider the general case and later discuss how these two types of excitons participate in the QAVHE. The Hamiltonian describing an exciton in the TMD layers in the continuous model reads [29, 31] H = H0(pe) + H0(ph) + u(rh − re) + U (re, rh). (2) In the vicinity of the conduction and valence band edges, both the H0 terms can be approximated by the parabolic term H0(p) = p2/2m. In the framework of the two-band Dirac model, which describes well the low-energy spec- trum of a TMD monolayer, the electron and hole have equivalent effective masses. The account of the higher en- ergy bands results in a difference of the particle masses. However, for our purpose, which is the proof of principle, the two-band model is sufficient. The second term in Hamiltonian (2) describes the Coulomb interaction between the electron and hole, and U (re, rh) term corresponds to the exciton energy shift due to the voltage applied across the layers. The Berry- phase effect is introduced via the gauge transformation rα → rα + Ωα(pα) × pα, 1 2 (3) where α = e, h. Substituting (3) in (2) and then in (1), one finds the equation of (quasiclassical) exciton motion (see the Supplemental Material [32] for the details of cal- FIG. 1. System schematic. A two-dimensional exciton gas localized in two MoS2 layers and exposed to an electrostatic field Ez(R) by means of the top an bottom gates. The field is inhomogeneous in the in-plane directions, which is shown by black arrows becoming more dense from left to the right. in momentum space, the quantum number indicating the valley is robust against external perturbations, thus cre- ating an additional degree of freedom similar to spin or charge of the particles. This paradigm forms the basis of bosonic valleytronics. Another property of indirect excitons, which distin- guishes them from their direct counterparts, is a long lifetime, typically ranging from nanoseconds to microsec- onds. The lifetime is long due to the weak overlap of spatially separated electron and hole wave functions. Fast energy relaxation allows for cooling indirect excitons down to ∼ 0.1 K within their lifetime, that is well below the temperature of quantum degeneracy of exciton gas, thus opening an experimental way to form a coherent exciton state -- the excitonic quasi-condensate. Exper- iments shows typical temperatures of exciton condensa- tion 1−5 K in GaAs nanostructures. Recently there have been reported reasonable arguments to expect a much higher condensation temperature for indirect excitons in TMD monolayers such as MoS2 [16], which makes these materials very promising for the observation of quantum coherent phenomena at higher temperatures. If we look at the quasiclassical picture, the dipole mo- ment of an exciton p = ed (e > 0), where d is the in- terlayer separation, directs across the layers (see Fig. 1). If a voltage is applied across the layers, it generates a static electric field Ez, resulting in the exciton energy shift by U = −pEz and thus giving an opportunity to create a force F(R) = −∇RU (R) which influences the center-of-mass motion due to the change of the electric field along the layers Ez(R) [18]. The presence of the Berry phase results in an anomalous contribution to the exciton center of mass velocity [28 -- 30], proportional to −[∇RU×Ωex], where Ωex is the exciton Berry curvature. This contribution is responsible for the exciton QAVHE. The Heisenberg equations of motion, P = −∂RH, R = ∂PH, FIG. 2. Two types of indirect excitons in momentum space: the definition of the direct momentum space exciton (DMSE) and indirect momentum space exciton (IMSE). (1) Top gate Bottom gate MoS2 Ez(R) X MoS2 K K' K K' Top monolayer Bottom monolayer DMSE IMSE C V 3 (8) in the excited state, R = (cid:104)n, P WBEC(cid:105)2 2π S ×δ(EBEC(n, P) − EBEC − ω) is the rate of exciton transitions to the final state n, P(cid:105) from the equilibrium Bose-condensed state BEC(cid:105) (per second per unit area). Here ω is the frequency of exter- nal electromagnetic field, EBEC is the energy of exciton system in BEC regime and EBEC(n, P) is the energy of the excited state consisting of the condensate and a pho- toexcited exciton. We use a constant τ assuming the scattering of excited excitons via the short-range impu- rities. The Hamiltonian describing the exciton interac- tion with the external uniform electromagnetic field E is taken in the dipole approximation, W (r) = −e(rE), where r = rh − re is the relative in-plane electron-hole coordinate. The exciton energy is characterized by the momentum P and discrete energy levels n of internal exciton mo- tion, En(P) = P2/2M + n. We assume that the con- densate occurs at the energy state with zero center-of- mass momentum P = 0 and at the lowest energy level of the internal exciton motion n=0 = 0. According to the Bogoliubov model of a weakly-interacting Bose gas, the low-energy properties of a Bose condensate can be characterized by the excitations having the dispersion ωk = sk(cid:112)1 + (kξ)2, where s = (cid:112)gnc/M is a phase velocity, ξ = 1/2M s is a healing length, and g is the strength of exciton-exciton interaction. Following the Bogoliubov theory, the Bose-condensed state reads ΨBEC = a0√ S ψ0(r) (cid:88) P(cid:54)=0 (cid:16) † uPaP + vPa −P (cid:17) exp(iPR) √ S (9) . +ψ0(r) √ ncS, the first term describes Here a0 = the macroscopically-occupied BEC state, whereas the sec- ond term corresponds to the fluctuations with the dis- persion ωP, and uP, vP are the Bogoliubov transforma- tion coefficients. The function ψn(r) is the eigen func- tion of internal exciton motion, while the excited states √ with energies En(P) have the eigen functions Ψn(R, r) = ψn(r) exp(iPR)/ S. Due to the presence of two terms in Eq. (9), the matrix element of external field W (r) also consists of two terms describing two types of processes, schematically shown in Fig. 3(a). The matrix element of the first-type processes reads [32] Mn(P) = a0Wn,0 δ(P), where (10) (2π)2 (cid:90) S drψ† (cid:88) FIG. 3. (a) Two principal types of photoinduced transitions in the system. Blue arrows show the direct transitions from the quasi-condensate to excited states, whereas indirect tran- sitions accompanied by the emissions of Bogoliubov quasi- particles with frequencies ω−P are depicted by red arrows. These processes result in the emergence of two components of the electric current density presented in Fig. 4 (blue and red curves). (b) Photoinduced intra-exciton transitions in the system under the circular-polarized light. culation), R = P/M − 1 8 [F0 × Ωex], (4) where F0 = F(0) is the in-plane static force acting on the exciton center-of-mass and Ωex = 2Ωe is the exciton Berry curvature, taken at the band edges. The corre- sponding exciton current reads j = Rf (P, R, t). (5) P In equilibrium, the distribution function is an even func- tion of the center-of-mass momentum f0(−P) = f0(P). Therefore only the second term in (4) if substituted in (5) gives a non-vanishing contribution. If condensed, the ex- citons occupy the lowest energy state with zero momen- tum, f0(P) = ncδP,0, where nc is a condensate density. The resulting equilibrium valley Hall current reads jeq = − nc 8 [F0 × Ωex]. (6) Under the action of an external electromagnetic field the exciton distribution acquires the nonequilibrium cor- rection, f (P, R, t) = f0(P) + δf (P, R, t), describing the population of noncondensed excitonic states n, P(cid:105) with n being the quantum number of the internal ex- citon quantized energy levels and P (cid:54)= 0. In the sta- tionary regime, the current of photoexcited excitons is determined by the time-averaged distribution function δfn(P) = δf (P, R, t), yielding j = − 1 8 [F0 × Ωex] (cid:88) n,P δfn(P), (7) Wn,0 = n(r)W (r)ψ0(r). where δfn(P) = τ R, with τ the exciton relaxation time It corresponds to direct exciton transitions from the con- densate to a non-condensed state with the excitation of Exciton vacuum K K' (0,0) (0,0) (n, +1) (n, -- 1) (n, -- 1) (n, +1) P -P Energy Momentum (b) (a) ω-P En(P) an internal degree of freedom of individual exciton. The corresponding photoexcited Hall current reads Wn,02 (cid:18) ncτ 2 (cid:19)(cid:88) j(1) = [Ωex × F0] 1 + (ω − n)2τ 2/2 . 42 n(cid:54)=0 (11) It has a resonant structure and is proportional to the condensate density. The processes of the second type are described by the second term in Eq. (9), reflecting the density fluctuations of the BEC (the Bogoliubov quanta). At zero (small) temperature this excitation branch is (nearly) empty thus † −p con- the only term containing the creation operator a tributes to the optical transitions. This corresponds to the exciton excitation by the photon absorption accom- panied by the emission of a Bogoliubov quasiparticle, as indicated in Fig. 3(a). Contrary to the processes of the first type, here the exciton transfers into the non- condensate state with a nonzero momentum P (cid:54)= 0 from the condensate state with P = 0. Such indirect transi- tions may occur only if the momentum P is carried away by a "third body". The Bose condensate here plays this very role, carrying away the momentum by means of the Bogoliubov excitations with the energy ω−P. The matrix element describing these processes reads Mn(P(cid:48), P) = Wn0 (2π)2 S vPδ(P(cid:48) − P). (12) In the most interesting limit of linear dispersion of Bo- goliubov excitations (ωP ≈ sP (cid:29) P 2/2M ), we can carry out analytical calculations [32] and find the correspond- ing contribution to the exciton Hall current: j(2) = [Ωex × F0] ×(cid:88) n(cid:54)=0 Wn,02 + arctan 2 (cid:19) (cid:18) M τ (cid:18) π 16π3 (cid:20) (ω − n)τ (cid:21)(cid:19) (13) . Evidently, this term vanishes if we disregard the Berry phase effect (the term j(1) is also zero in this case). Note, that it is the arctangent function which ultimately de- termines the behavior of this component of the electric current. Another important property of (13) is that it does not explicitly depend on the condensate density, in contrast with (11), even though j(2) also vanishes in the absence of the condensate. The system response (13) is a unique characteristic of bosonic systems, making the QAVHE here conceptually different from the conven- tional fermionic VHE. Thus, we conclude that condensa- tion is the necessary condition for the current quantiza- tion in the absence of an external magnetic field. In an experiment, one creates the exciton gas when the sample is illuminated by light with the frequency ex- ceeding the material band gap. Due to the fast energy re- laxation, the photo-excited electrons and holes cool down 4 FIG. 4. Qualitative behaviour of the photoexcited part of ex- citon valley Hall current as a function of frequency of light (the red curve is slightly shifted down for better distinguisha- bility). The calculations are performed within an analytically- tractable model of internal exciton spectrum [32]. Red and blue curves correspond to the processes described by red and blue arrows in Fig. 3(a). ex = −ΩK(cid:48) and form excitons in the monolayers. We can expect that both the valleys will be populated with DMSE and IMSE nearly equally. The Berry curvature of IMSE Ωex = 0 since the signs of the Berry curvatures of electrons and holes located in different valleys are opposite. Thus, only DMSE will contribute to the effect. Due to the nearly equivalent population of the valleys, the equilibrium val- ley Hall current vanishes since ΩK ex . However, the photoinduced part of the valley Hall current might be nonzero. Due to the axial symmetry of the Coulomb electron-hole interaction potential u(r), the eigen states of the internal exciton motion are characterized by the principal (radial) quantum number and the projection of the angular momentum: n = (nr, m). The axial symme- try dictates the degeneracy of quantum states (nr,±m). As it is shown in Ref. 29, the Berry curvature influences these states resulting in their energy splitting of the or- der of tens of meV. Moreover, the splitting has opposite sign in different valleys, see Fig. 3(b). If the exciton gas is exposed to a circularly polarized electromagnetic field, the intra-excitonic transitions oc- cur in one valley due to the valley-dependent optical se- lection rules, simultaneously exciting the state m = +1 or m = −1 in the same valley. Thus, the photoexcited valley Hall current occurs in the system resulting in the exciton accumulation at the sample edge across F0, see Fig. 4. The latter can be detected at the sample bound- ary by measuring the polarization of luminescence. Due to the step-like behaviour of the exciton Hall current, the intensity of the luminescence should inherit analogous step-like behaviour since its intensity is proportional to the density of excitons at the sample edge, determined by the value of the Hall current. ��0.5 0.4 0.3 0.2 0.1 0.0 3.6 3.7 3.8 3.9 4.0 3.5 Light frequency [arb.units] Current density [arb.units] j(2) j(1) j(1)+j(2) The last issue we want to address is the applicability of the Bogoliubov model. Even though this model implies a weakly-interacting system, in the long-wavelength limit it works well even in the case of a strongly interacting Bose gas [33]. This is due to the low-energy excitations under strong interaction between the Bose particles also acquire linear (sound-like) dispersion (with some effective renormalized parameters). Conclusions. We have reported on the QAVHE of bosons. We considered indirect exciton transport in two parallel layers of noncentrosymmetric two-dimensional materials and shown that in the presence of the Bose- Einstein condensate, there emerges a quantized valley Hall current of excitons due to the anomalous group ve- locity of exciton center-of-mass motion under the action of an external electromagnetic field. This current experi- ences steplike behavior as a function of electromagnetic field frequency, mimicking the quantization of the inter- nal exciton motion without external magnetic field. Acknowledgments. We acknowledge the financial sup- port by the Russian Science Foundation (Project No. 17- 12-01039) and the Institute for Basic Science in Korea (Project No. IBS-R024-D1). [1] E. H. Hall, Am. J. Math. 2(3), 287 (1879). [2] T. Ando, Y. Matsumoto, Y. Uemura, J. Phys. Soc. Jpn. 39(2), 279 (1975). [3] K. v. Klitzing, G. Dorda, M. Pepper, Phys. Rev. Lett. 45(6), 494 (1980). [4] R. Karplus and J. M. Luttinger, Phys. Rev. 95(5), 1154(1954). [5] C.-Z. Chang et al., Science 340(6129), 167 (2013). Phys. Rev. Lett. 49, 405408 (1982). [6] N. Nagaosa, J. Sinova, S. Onoda, A. H. MacDonald, and N.-P.-Ong, Rev. Mod. Phys. 82, 1539 (2010). [7] M. Z. Hasan, C. L. Kane, Rev. Mod. Phys. 82, 3045 (2010). [8] M. V. Berry, Proc. R. Soc. Lond. A Math. Phys. Sci. 392, 45 (1984). [9] M.-C. Chang and Q. Niu, Phys. Rev. B 53, 7010 (1996). [10] G. Sundaram and Q. Niu, Phys. Rev. B 59, 14 915 (1999). [11] Di Xiao, Ming-Che Chang and Qian Niu, Rev. Mod. Phys. 82, 1959 (2010). [12] D. J. Thouless, M. Kohmoto, M. P. Nightingale, M. den Nijs, Phys. Rev. Lett. 49, 405 (1982). 5 [13] Di Xiao, Gui-Bin Liu, Wanxiang Feng, Xiaodong Xu and Wang Yao, Phys. Rev. Lett. 108, 196802 (2012). [14] Di Xiao, Wang Yao, and Qian Niu, Phys. Rev. Lett. 99, 236809 (2007). [15] Wang Yao, Di Xiao, and Qian Niu, Phys. Rev. B 77, 235406 (2008). [16] M. M. Fogler, L. V. Butov and K. Novoselov, Nature Commun. 5, 4555 (2016). [17] L. Butov, Solid State Communications 127, 89 (2003). [18] L. Butov, Superlattices and Microstructures 108, 2 (2017). [19] Gang Wang, Alexey Chernikov, Mikhail M. Glazov, Tony F. Heinz, Xavier Marie, Thierry Amand, and Bernhard Urbaszek Rev. Mod. Phys. 90, 021001 (2018). [20] J.-S. You, S. Fang, S. Y. Xu, E. Kaxiras, and T. Low, Phys. Rev. B 98, 121109(R) (2018). [21] B. Radisavljevic, A. Radenovic, J. Brivio, V. Giacometti, A. Kis, Single-layer MoS2 transistors, Nature Nanotech. 6(3), 147?50 (2011). [22] R. S. Sundaram, M. Engel, A. Lombardo, R. Krupke, A. C. Ferrari, Ph. Avouris, M. Steiner, Electrolumines- cence in Single Layer MoS2, Nano Lett. 13(4), 1416 (2013). [23] X. Xu, W. Yao, D. Xiao and T. F. Heinz, Spin and pseu- dospins in layered transition metal dichalcogenides, Na- ture Phys. 10, 343 (2014). [24] A. T. Hanbicki, H.-J. Chuang, M. R. Rosenberger, C. S. Hellberg, S. V. Sivaram, K. M. McCreary, I. I. Mazin, and B. T. Jonker, ACS Nano 12, 4719 (2018). [25] E. V. Calman, M. M. Fogler, L. V. Butov, S. Hu, A. Mishchenko, A. K. Geim, Indirect excitons in van der Waals heterostructures at room temperature, Nature Commun. 9, 1895 (2018). [26] E. V. Calman, C. J. Dorow, M. M. Fogler, L. V. Butov, S. Hu, A. Mishchenko, A. K. Geim, Control of excitons in multi-layer van der Waals heterostructures, Appl. Phys. Lett. 108, 101901 (2016). [27] M. M. Fogler, L. V. Butov, K. S. Novoselov, High- temperature superfluidity with indirect excitons in van der Waals heterostructures, Nature Commun. 5, 4555 (2014). [28] W. Yao and Q. Niu, Phys. Rev. Lett. 101, 106401 (2008). [29] J. Zhou, W.-Yu Shan, W. Yao, and D. Xiao, Phys. Rev. Lett. 115, 166803 (2015). [30] A. Srivastava and A. Imamoglu, Phys. Rev. Lett. 115, 166802 (2015). [31] A. Hichri, S. Jaziri, M. O. Goerbig, arXiv:1807.10838v1 (2018). [32] see Supplemental Material for the details of derivations. [33] S. T. Beliaev, Energy-spectrum of a non-ideal Bose gas, JETP 7(2), 299 (1958).
1601.02072	1	1601	2016-01-09T04:18:34	Geometry, stability and thermal transport of hydrogenated graphene nanoquilts	[ "cond-mat.mes-hall", "cond-mat.mtrl-sci" ]	Geometry, stability, and thermal transport of graphene nanoquilts folded by hydrogenation are studied using molecular dynamics simulations. The hydrogenated graphene nanoquilts show increased thermodynamic stability and better transport properties than folded graphene structures without hydrogenation. For the two-fold graphene nanoquilt, both geometry and thermal conductivity are very sensitive to the adsorbed hydrogen chains, which is interpreted by the red-shift of PDOS. For the multi-fold nanoquilts, their thermal conductivities can be tuned from 100% to 15% of pristine graphene, by varying the periodic number or length. Our results demonstrated that the hydrogenated graphene nanoquilts are quite suitable to be thermal management devices.	cond-mat.mes-hall	cond-mat	Geometry, stability and thermal transport of hydrogenated graphene nanoquilts Zhongwei Zhang1, Yuee Xie1†, Qing Peng2, and Yuanping Chen1* 1 Department of Physics, Xiangtan University, Xiangtan 411105, Hunan, P.R. China 2 Department of Mechanical, Aerospace and Nuclear Engineering, Rensselaer Polytechnic Institute, Troy, NY 12180, U.S.A Abstract: Geometry, stability, and thermal transport of graphene nanoquilts folded by hydrogenation are studied using molecular dynamics simulations. The hydrogenated graphene nanoquilts show increased thermodynamic stability and better transport properties than folded graphene structures without hydrogenation. For the two-fold graphene nanoquilt, both geometry and thermal conductivity are very sensitive to the adsorbed hydrogen chains, which is interpreted by the red-shift of PDOS. For the multi-fold nanoquilts, their thermal conductivities can be tuned from 100% to 15% of pristine graphene, by varying the periodic number or length. Our results demonstrated that the hydrogenated graphene nanoquilts are quite suitable to be thermal management devices. Keywords: graphene nanoquilts, thermal transport, stability, hydrogenation 1 Ⅰ. Introduction Graphene is a superior material for electronics and phononics [1-4] because of super high electron mobility [5] and thermal conductivity [6]. Various functionalizations [7-10] are used to tune physical properties of graphene in order to achieve desired devices. Folding [10-12] is considered as an effective way of functionalization. Although graphene is very hard in the planar direction, it can be easily warped in the out-of-plane direction. Folding can drastically change the physical properties of graphene [10-15]. For example, the Fermi velocity in the graphene is reduced by folding [13]; an armchair nanoribbon changes from semiconducting to metallic after folding [14]; the thermal conductivities of graphene decrease as folding occurs [15]. Based on these folded graphene structures, some devices have been proposed, such as resistors [16], diverters [16], and conductance modulators [17], etc. Experimentally, folded graphene can be obtained by random ultrasonic stimulation [18] or patterned etched trenches [19] i.e., by external mechanical forces. However, the geometries of these folded structures are not able to be easily controlled [18,19] and they are not stable at high temperatures, especially for multi-folded graphene [15]. The former studies indicated that the physical properties of folded graphene are very sensitive to their geometries [10-15]. Therefore, another more effective way should be found to precisely fold graphene. In fact, advances in microscopy technologies have provided an exciting opportunity to manipulate graphene morphologies by chemical functionalization [20,21]. A recent study shows that graphene can be folded by hydrogenation with different folding angles [21,22]. These hydrogen folded structures are more suited for thermal devices, however, to our 2 best knowledge thermal transport in these folded graphene, especially in the multi- folded graphene, has been reported scarcely. In this paper, we study thermal transport properties of two-fold and multi-fold graphene by applying classical molecular dynamics methods. Because graphene can be folded like a quilt by hydrogenation, as shown in Figures 1(b) and 1(d), we call these folded graphene nanoquilts. The geometric evolution from pristine graphene to a two-fold nanoquilt is discussed at first. The thermal conductivities decrease rapidly with hydrogenation, ultimately to about 30% of pristine graphene. The tuning range of thermal conductivities is much larger than the case of folded graphene without hydrogenation. On the basis of the two-fold nanoquilt, thermal transports in multi-fold nanoquilts are studied. The multi-fold nanoquilts exhibit excellent thermal stability at high temperatures, and their thermal conductivities can be well modulated by periodic number and periodic length. We find that, after pristine graphene is folded into a multi-fold structure, its thermal conductivity can at most be reduced to 15%. These findings are very useful for the applications of graphene in thermal devices. Ⅱ. Sim ulation m ethod and m odel Figure 1(a) shows a graphene nanoribbon with length L1 and width W, where n hydrogen chains are adsorbed on the middle of top surface. The optimized structure can be seen in Figure 1(b), showing the graphene nanoribbon being folded by hydrogenation. θ is used to represent the folding degree defined as the angle between two-side straight ribbons. The folding process likes a quilt being folded into two layers, so we call it a two-fold nanoquilt. Based on Figure 1(a), a periodic two-surface hydrogenated graphene nanoribbon is shown in Figure 1(c), in which L0 represents the periodic length and L2 is the total length. Figure 1(d) is the optimized structure, i.e., it 3 forms a multi-fold graphene nanoribbon with a periodic number m, which is called multi-fold nanoquilts. In our simulations, the LAMMPS package [23] with the Adaptive Intermolecular Reactive Empirical Bond Order (AIREBO) potential [24] is adopted. Periodic boundary conditions are applied only to the direction perpendicular to the functionalized rows and other two directions are set free. Then atomic positions are relaxed to obtain minimum energy configuration using the conjugate gradient algorithm. After relaxation, we extend the relaxed structure to the desired hydrogenated graphene nanoribbons. That is, we prepared the folded graphene nanoribbons by C-H chemical bonding effects. In the non-equilibrium molecular dynamics simulation (NEMD), the AIREBO potential is still adopted. On each hydrogenated graphene nanoribbon, fixed boundary conditions are implemented with the atoms at the left and right ends are fixed at their equilibrium positions. Next to the boundaries, the adjacent two cells of atoms are coupled to Nosé–Hoover thermostats [25,26] with temperatures 320 K and 280 K, respectively. That is, the average temperature is 300 K with a temperature difference 20 K. From Fourier’s law, the thermal conductivity K is defined as K = - J∇T∙S , (1) where T∇ is the temperature gradient in structure, J is the heat flux from the heat bath to the system, which can be obtained via calculating the heat baths power. S W H × = is the cross-sectional area, W is the width and H is the thickness of graphene nanoribbons, which we have chosen as H = 0.144 nm. In order to focus on the hydrogenation and deformation effect on phonons transport in nanoribbons, we 4 keep the cross section area as a constant. With a time step of 0.5 fs, the graphene ribbon is first relaxed to the equilibrium states at 300 K for up to 500 ps under the canonical ensemble (NVT). Next, a temperature gradient is achieved by thermostat bath. After reaching the non-equilibrium steady state, 100 ps been used to calculate the thermal conductivity. In addition, the adopted AIREBO potential included the LJ term [24], which can be used to describe the van der Waals forces between ribbon parts after deformation, as shown in Figures 1(b) and 1(d). To understand the underlying mechanisms of phonon transport in hydrogenated graphene, the phonon density of states (PDOS) has been calculated. The PDOS is calculated from the Fourier transform of the velocity autocorrelation function [26]: PDOS(ω)= 1√2π� e-iωt〈� vj(t)vj(0) j=1 N 〉dt (2) where vj(0) is the average velocity vector of a particle j at initial time, vj(t) is its velocity at time t, and ω is the vibration wavenumber. Ⅲ. Results and discussion In Figure 2(a), the geometric evolution of two-folded graphene nanoquilts by hydrogenation is shown. The size of the graphene nanoribbon we considered is L1 = 6.3 nm and W = 2.21 nm with armchair edge. The nanoribbon will be bended gradually as hydrogen chains are adsorbed on the middle. This is because the hydrogenation of graphene leads to a transformation of sp2 to sp3 and then causes a local distortion. The more chains, the more the bending deformation accumulates. 5 Figure 2(a) displays the relation between the bending angle θ and the number n of adsorbed hydrogen chains. As the number n increases from 1 to 9, the bending angle drops fast from 1400 to 250, indicating the graphene nanoribbon is folded more and more. On one hand, the two-side straight nanoribbons come together gradually, on the other hand the middle region of ribbon forms a half cylinder. In the range of 10 < n < 18, the bending angle approaches θ = 0°, i.e., the two-side straight nanoribbons are parallel. As n > 18, the two straight nanoribbons are folded into a half cylinder. Generally speaking, the two straight nanoribbons are considered as two thermal leads in the application. The two leads are hidden in the case of n > 18, therefore we do not show them in Figure 2(a) anymore and we do not consider this case in the following study of thermal transport. Seen in Figure 2(a), one can find that the geometries of two-fold graphene nanoquilts can be tuned by hydrogen coverage. However, our results are somewhat different with those in Ref. [22] in which the angle of folded graphene varies linearly with hydrogenation ratio. The corresponding thermal transport properties of the two-fold nanoquilts are shown in Figure 2(b). Somewhat similar to the geometric variation, thermal conductivities of the nanoquilts decrease quickly as the chains number n of hydrogenation increases from 1 to 7, and then approach a convergent value which is about 1/3 the thermal conductivity of a pristine nanoribbon. Apparently, the combined effect of deformation and hydrogenation result in a remarkable reduction of thermal conductivity. As the adsorbed chains are above 10, the thermal conductivities of the nanoquilts are insensitive to the hydrogenation coverage because of their steady geometries. The thermal conductivity of folded graphene without hydrogenation is also shown in Figure 2(b) for comparison (see the dashed line). These folded structures are obtained by relaxation of hydrogenated nanoquilts after the hydrogen 6 atoms are removed. Some folded structures are shown in the inset of Figure 2(b). One can find that the thermal conductivities of these folded nanoribbons (without-H) are less sensitive to the geometric deformation than those of hydrogenated nanoribbons. In other words, thermal transport can be more effectively modified by hydrogenated nanoquilts. As demonstrated by Ouyang [17], folded graphene is an excellent thermal conductance modulator. Obviously, the hydrogenated folded graphenes are more effective to construct thermal modulators and resistors. It is well known that the thermal transport properties are determined by phonon activities. To understand the phonon transport in the two-fold nanoquilts, in Figure 3(a), the PDOS of these nanoquilts are calculated. Compared with the PDOS of the pristine nanoribbon, the PDOS of hydrogenated nanoquilts have an additional phonon peak at frequencies about 103 THz, which corresponds to C−H bond vibrations. It has no contribution to the thermal conductivity because its frequency is too high. In the lower frequency region, one can find that the G-band peak (the higher frequency peak) of the pristine nanoribbon at frequencies about 65 THz shifts to a lower frequency as the number of hydrogen chains n increases. The detailed shift of the G-band is shown in Figure 3(b). The frequency of the G-band is very sensitive to the hydrogenation at small covered chains. The red shift of the G-band induces the acoustic phonons to shift to low frequencies and also leads to a reduction of the phonon group velocity ). Meanwhile, the phonon mean free path l is decreased by hydrogenation [28] while the heat capacity C is relatively insensitive to the shift of the (𝑣𝑣=𝑑𝑑𝑑𝑑 𝑑𝑑𝑑𝑑⁄ decrease because 𝐾𝐾=∑𝐶𝐶𝑣𝑣𝐶𝐶, from classical lattice thermal transport theory. In the G-band at room temperature [29]. Therefore, thermal conductivities K of nanoquilts bottom panel of Figure 3(a), the PDOS of folded nanoribbons without hydrogenation 7 indicates that the physical folding has little effect on the G-band and thus the change of thermal conductivity is also small. The multi-folded graphene nanoquilts can be constructed based on the two-fold nanoquilt, as shown in Figure 1. Here, the thermal stability of multi-fold nanoquilts is discussed. We consider a nanoquilts consisting of 8 periodic two-fold nanoquilt with hydrogen chains n = 5 and L0 = 6.3 nm. Figure 4(a) shows atomic structure of the nanoquilts at 0 K, while Figure 4(b) is the structure at 300 K. One can find that an increase in temperature does not destroy the form of the quilt. Our further calculations demonstrate that the folded nanoquilt maintains periodic structure up to temperature 1200 K (not shown). The main reason is the adsorbed hydrogen atoms protect the folding edges of the quilt. For comparison, Figures 4(c) and 4(d) display the folded graphene without hydrogenation at 0 K and 300 K, respectively. The periodic folded structure is destroyed obviously by the thermodynamic process. This indicates that the nanoquilts after hydrogenation are more stable than those obtained by external mechanical forces. Other nanoquilts with different periodic number m in Figures 4(e) and 4(f) further show the stability of the hydrogenated nanoquilts. In addition, our calculations demonstrate that the multi-fold hydrogenated nanoquilts can maintain their geometric form until 1200 K. So, the hydrogenated folded graphene are more suitable for thermal devices. Thermal transport in nanostructures is tightly associated with the size of the structures. In Figure 5, thermal conductivities of multi-fold nanoquilts as a function of periodic number m are calculated. As the periodic number m is small (m < 10), the thermal conductivities of all nanoquilts increase with their length, which is different with the results in Ref. [22]. As m becomes larger, however, different kinds of multi- 8 fold nanoquilts show different dependent of thermal conductivities on the length. For the multi-fold nanoquilts with n = 1 and 3, the thermal conductivities continue to increase. For the multi-fold nanoquilts with n = 5 and 7, their thermal conductivities approach converged values. The difference originates from the variation of transport mechanisms in the different kinds of nanoquilts. As the periodic number is small, the length of all nanoquilts is in the range of the phonon mean free path, i.e., phonon transport is ballistic. Because the thermal conductivity is proportional to the length of the ballistic mechanism, thermal conductivities of all nanoquilts increase in proportion to their length (𝐾𝐾~𝐿𝐿𝜆𝜆, where λ is the divergence power) [30]. The phonon mean free paths of the nanoquilts with more hydrogen chains (n = 5 and 7) decrease sharply because of larger deformations and scattering of hydrogen atoms. Therefore, phonon transport in these nanoquilts changes from ballistic to diffusive transport. In this case the thermal conductivities are not sensitive to the length anymore, in larger m, and thus converge to a steady value. For the longer multi-fold nanoquilts with less hydrogen chains (n = 1 and 3), phonon transports in these structures are between ballistic and diffusive transport, because the effect of adsorbed hydrogen atoms is relatively week. As a result, their thermal conductivities continue increasing but the rate of the increasing speed becomes slower. It is noted that the thermal conductivity of graphene nanoribbons is significantly reduced by multiple folds. For example, the thermal conductivity of a nanoquilt with n = 7 and m = 18 is 41.35 W/mK, which only about 15% of the pristine nanoribbon (287.2 W/mK). In addition, the VDWs interaction can also affects the thermal transport of nanoquilts. For example, the thermal conductivity of a nanoquilt with m = 9 and n = 3 decreases 13.5% as the VDWs force is omitted. In Figure 6, the effect of the periodic length L0 of multi-fold nanoquilts on thermal conductivity is shown. The nanoquilts we studied here are fixed at L2 = 88.2 nm. The 9 decrease of L0 leads to the increase of periodic number m. As L0 changes from 88.2 to 0 nm, it represents the structural change from a pristine nanoribbon to multi-fold nanoquilts and then to a graphane nanoribbon. It is seen from Figure 6 that, thermal conductivities of all kinds of the nanoquilts (n = 1, 3, 5 and 7) decrease with L0, which is attributed to more and more periodic folds strengthening the phonon scattering. At L0 = 0 nm, i.e., the structure is an unfolded graphane nanoribbon, the thermal conductivities undergo a small increase, 89.8 W/mK, which is about 46.5% of the pristine nanoribbon. Finally, it should be noted that we also calculated the thermal transport in two-fold and multi-fold nanoquilts with zigzag edges. The results indicate that thermal transports in the zigzag-edge nanoquilts have the same properties as those in the armchair-edge nanoquilts, even though they have some differences in the detailed quantitative values because of anisotropy. Ⅳ. C onclusion In summary, we have investigated thermal transport properties of two-fold and multi-fold graphene nanoquilts. The studies of the two-folded hydrogenated nanoquilts show that their bending angle and thermal conductivity depend on the adsorbed hydrogenated chains. The tuning range of thermal conductivity is much larger than the case of folded graphene without hydrogenation. The variation of thermal conductivity is interpreted by the red-shift of PDOS. The multi-fold hydrogenated nanoquilts are more stable than multi-folded graphene without hydrogenation at high temperatures. Moreover, the thermal conductivity of multi- folded graphene nanoquilts can be well modulated by varying periodic number m and periodic length L0. We have found that a thermal reduction of multi-folded graphene 10 nanoquilts have been acquired as high as about 85% by varying periodic parameters. Our work demonstrates that hydrogenation is an effective method to precisely control geometry and also thermal transport of folded graphene. These findings may be useful for the applications of graphene in nanoscale devices, such as modulators, resistors and graphene electronic circuits. Acknowledgments This work was supported by the National Natural Science Foundation of China (Nos. 511 76161, 51376005 and, 11474243). Corresponding author: † [email protected]; * [email protected] 11 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 306 (2004) 666. [2] A. K. Geim and K. S. Novoselov, Nature Mater. 6 (2007) 183. [3] A. H. C. Neto, F.Guinea, N. M. R. Peres, K. S. Novoselov and A. K. Geim, Rev. Mod. Phys. 81 (2009) 109. [4] N. Yang, G. Zhang and B. Li, Appl. Phys. Lett. 95 (2009) 033107. [5] C. R. Dean, A. F. Young, I . Meric, C. Lee, L. Wang, S. Sorgenfrei, K. Watanabe, T. Taniguchi, P. Kim, K. L. Shepard and J. Hone, Nature Nanotech. 5 (2010) 722. [6] A. A. Balandin, S. Ghosh, W. Bao, I. Calizo, T. Teweldebrhan, F. Miao and C. N. Lau, Nano Lett. 8 (2008) 902. [7] T. Ramanathan, A. A. Abdala, S. Stankovich,D. A. Dikin, M. H. Alonso, R. D. Piner, D. H. Adamson, H. C. Schniepp, X. Chen, R. S. Ruoff, S. T. Nguyen, I. A. Aksay, H. R. A. Prud and L. C. Brinson, Nature Nanotech. 3 (2008) 327. [8] T. M. G. Mohiuddin, A. Lombardo, R. R. Nair, A. Bonetti, G. Savini, R. Jalil, N. Bonini, D. M. Basko, C. Galiotis, N. Marzari, K. S. Novoselov, A. K. Geim and A. C. Ferrari, Phys. Rev. B 79 (2009) 205433. [9] D. Pan, J. Zhang, Z. Li and M. Wu, Adv. Mater. 22 (2010) 734. [10] N. Patra, B. Wang and P. Král, Nano Lett. 9 (2009) 3766. 12 [11] S. T. Yang, Y. Chang, H. Wang, G. Liu, S. Chen Y. Wang, Y. Liu and A. Cao, J. Colloid Interface Sci. 351 (2010) 122. [12] F. Liu, S. Song, D. Xue and H. Zhang, Adv. Mater. 24 (2012) 1089. [13] Z. Ni, Y. Wang, T. Yu, Y. You and Z. Shen, Phys. Rev. B 77 (2008) 235403. [14] Y. Xie, Y. Chen and J. Zhong, J. Appl. Phys. 106 (2009) 103714. [15] N. Yang, X. Ni, J. W. Jiangand B. Li, Appl. Phys. Lett. 100 (2012) 093107. [16] Y. Xie, Y. Chen, X. Wei and J. Zhong, Phys. Rev. B 86 (2012) 195426. [17] T. Ouyang, Y. Chen, Y. Xie, G. M. Stocks and J. X. Zhong, Appl. Phys. Lett. 99 (2011) 233101. [18] J. Zhang, J. Xiao, X. Meng, C. Monroe,Y. Huang, and J. M. Zuo, Phys. Rev. Lett. 104 (2010) 166805. [19] K. Kim, Z. Lee, B. D. Malone, K.T. Chan, B. Alemán, W. Regan, W. Gannett, M. F. Crommie, M. L. Cohen, and A. Zettl, Phys. Rev. B 83 (2011) 245433. [20] S. Akcöltekin, H. Bukowska, T. Peters, O. Osmani, I. Monnet, I. Alzaher, B. B. Etat, H. Lebius and M. Schleberger, Appl. Phys. Lett. 98 (2011) 103103. [21] C. D. Reddy and Y. W. Zhang, Carbon 69 (2014) 86. [22] C. Li, G. Li and H. Zhao, Carbon 72 (2014) 185. [23] S. Plimpton, J. Comp. Phys. 117 (1995) 1. [24] S. Stuart, A. Tutein and J. Harrison, J. Chem. Phys. 112 (2000) 6472. 13 [25] S. Nosé, J. Chem. Phys. 81 (1984) 511. [26] W. G. Hoover , Phys. Rev. A 31 (1985) 1695. [27] J. M. Dickey and A. Paskin, Phys. Re. 188 (1969) 1407. [28] C. Li, G. Li and H. Zhao,Carbon 72 (2014) 185. [29] M. G. Xia and S. L. Zhang, Eur. Phys. JB 84 (2011) 385. [30] X. Mu, X. Wu, T. Zhang, D. B. Go and T. Luo, Sci. Rep. 4 (2014) 3909. 14 Figure captions Fig. 1. (color online) (a) Schematic view of an armchair-edge graphene nanoribbon adsorbing n hydrogen chains. L1 and W represent the length and width of nanoribbon, respectively. Top panel is the top view while the bottom panel is the side view. (b) Optimized structure of (a), a two-fold nanoquilt with a bending angle θ. (c) Schematic view of a periodic hydrogenated graphene nanoribbon in units of (a). L2 is the length of the nanoribbon and L0 is the periodic length. (d) Optimized structure of (c), a multi- fold nanoquilts of m period. Fig. 2. (color online) (a) Bending angle θ of two-fold nanoquilts as a function of hydrogenated chains n. Several corresponding structures are shown in the inset. (b) Thermal conductivities of two-fold nanoquilts as a function of hydrogenated chains n. Thermal conductivities of folded graphene nanoribbons without hydrogenation are also shown for compare. The other parameters of the two-fold nanoquilt are L1 = 6.3 nm and W = 2.21 nm. Fig. 3. (color online) (a) Phonon PDOS for pristine graphene, two-fold hydrogenated nanoquilts (n = 3, 5, 7 and 11), and folded graphene without hydrogenation (corresponding to the two-fold hydrogenated nanoquilt with n = 11). The high frequency peaks represent the G-peak of phonon. (b) The G-peak for the two-fold nanoquilt as a function of hydrogenated chains n. The other parameters of the two- fold nanoquilt are L1 = 6.3 nm and W = 2.21 nm. Fig. 4. (color online) (a-b) Multi-fold nanoquilts with n = 5 and m = 8 under temperatures 0 K and 300 K, respectively. (c-d) Multi-fold nanoribbon without hydrogenation (m = 8) under temperatures 0 K and 300 K, respectively. (e-f) Multi- folded nanoquilts (n = 5) with periodic numbers m = 4 and m = 12 under 300 K, 15 respectively. The other parameters of the multi-fold nanoquilts are L0 = 6.3 nm and W = 2.21 nm. Fig. 5. (color online) Thermal conductivities of multi-fold nanoquilts with different hydrogenated chains n are as functions of periodic number m. The periodic length L0 = 6.3 nm. Fig. 6. (color online) Thermal conductivities of multi-fold nanoquilts with different hydrogenated chains n are as functions of periodic length L0. The total length of all nanoquilts are fixed at L2 = 88.2 nm. 16 Fig. 1 17 Fig. 2 18 Fig. 3 19 Fig. 4 20 Fig. 5 21 Fig. 6 22
1805.06686	1	1805	2018-05-17T10:32:25	Intrinsic valley Hall transport in atomically thin MoS2	[ "cond-mat.mes-hall" ]	Electrons hopping in two-dimensional honeycomb lattices possess a valley degree of freedom in addition to charge and spin. In the absence of inversion symmetry, these systems were predicted to exhibit opposite Hall effects for electrons from different valleys. Such valley Hall effects have been achieved only by extrinsic means, such as substrate coupling, dual gating, and light illuminating. Here, we report the first observation of intrinsic valley Hall transport without any extrinsic symmetry breaking in the non-centrosymmetric monolayer and trilayer MoS2, evidenced by considerable nonlocal resistance that scales cubically with local resistance. Such a hallmark survives even at room temperature with a valley diffusion length at micron scale. By contrast, no valley Hall signal is observed in the centrosymmetric bilayer MoS2. Our work elucidates the topological quantum origin of valley Hall effects and marks a significant step towards the purely electrical control of valley degree of freedom in topological valleytronics.	cond-mat.mes-hall	cond-mat	Intrinsic valley Hall transport in atomically thin MoS2 Zefei Wu1†, Benjamin T. Zhou1†, Gui-Bin Liu3, Jiangxiazi Lin1, Tianyi Han1, Liheng An1, Yuanwei Wang1, Shuigang Xu1, Gen Long1, Chun Cheng4, Kam Tuen Law1, Fan Zhang2, Ning Wang1* 1Department of Physics and the Center for Quantum Materials, the Hong Kong University of Science and Technology, Hong Kong, China. 2Department of Physics, University of Texas at Dallas, Richardson, TX 75080, USA. 3School of Physics, Beijing Institute of Technology, Beijing 100081, China. 4Department of Materials Science and, Southern University of Science and Technology, Shenzhen 518055, China. *Correspondence to: [email protected] (N.W.); [email protected] (F.Z.); [email protected] (Z.W.) †These authors contributed to this work equally. 1 Electrons hopping in two-dimensional honeycomb lattices possess a valley degree of freedom in addition to charge and spin. In the absence of inversion symmetry, these systems were predicted to exhibit opposite Hall effects for electrons from different valleys. Such valley Hall effects have been achieved only by extrinsic means, such as substrate coupling, dual gating, and light illuminating. Here, we report the first observation of intrinsic valley Hall transport without any extrinsic symmetry breaking in the non-centrosymmetric monolayer and trilayer MoS2, evidenced by considerable nonlocal resistance that scales cubically with local resistance. Such a hallmark survives even at room temperature with a valley diffusion length at micron scale. By contrast, no valley Hall signal is observed in the centrosymmetric bilayer MoS2. Our work elucidates the topological quantum origin of valley Hall effects and marks a significant step towards the purely electrical control of valley degree of freedom in topological valleytronics. Electron valley degree of freedom emerges as local extrema in the electronic band structures. Inequivalent valleys, well separated in the Brillouin zone, can be energetically degenerate due to symmetry and serve as novel information carriers controllable via external fields1-4. A feasible means to manipulate such a valley degree of freedom is through a valley Hall effect (VHE)3-7. Analogous to an ordinary Hall effect, in which a transverse charge current is driven by a uniform magnetic field in real space, a transverse valley current in the VHE is produced by valley-contrasting Berry curvatures in momentum space. Upon the application of an external electric field, the curvatures drive carriers from 2 different valleys to traverse in opposite directions. Therefore, the VHE has been a major theme in the study of valleytronics, particularly in those 2D materials featuring K and K' valleys in their hexagonal Brillouin zones8-17. As Berry curvature is even under spatial inversion (P) and odd under time reversal (T), the VHE cannot survive when both P and T symmetries are present. To achieve VHEs in monolayer and bilayer graphene, an elaborately aligned h-BN substrate8 and a strong dual gating field9,10 were respectively utilized to break the P symmetry. To excite VHEs in specific valleys15,16, circularly polarized lights18-20 were used for breaking the T symmetry in atomically thin transition-metal dichalcogenides (TMDC). Monolayer TMDCs have direct band gaps of optical frequencies at two inequivalent K-valleys21,22, due to the intrinsic P asymmetry in their unit cells depicted in Fig. 1a. Thus, Berry curvatures with opposite signs naturally emerge at the two K-valleys. Moreover, a mirror symmetry locks the spin and valley indices of the sub-bands split by the spin-orbit couplings, both of which are flipped under T; the spin conservation suppresses the inter-valley scattering. Therefore, monolayer TMDCs have been deemed an ideal platform for realizing intrinsic VHE without extrinsic symmetry breaking13,14. However, the quantum transport in atomically thin TMDCs has been a long-standing challenge due to the low carrier mobility and the large contact resistance in their field-effect devices prepared by the exfoliation method. Recent breakthroughs in the fabrication of low- temperature ohmic contacts for high-mobility 2D TMDC devices23-26 have already facilitated the observation of transport hallmarks of Q-valley electrons26,27, K-valley holes28-30, and Γ-valley holes31. These discoveries have revealed the rich and unique valley 3 physics in the platform of atomically thin TMDCs. In this work, we systematically examine intrinsic VHEs through measuring nonlocal resistances and their layer dependence in n-type 2H-MoS2. For the first time, we observe nonlocal resistances that exhibit cubic power-law scaling with the local resistances in the monolayers and trilayers, evidencing intrinsic VHEs. Because of the large intrinsic bandgaps of TMDCs, such VHEs can even be observed at room temperature in our monolayer devices. Beyond critical carrier densities (5×1010 cm−2 for monolayers and 1.1×1011 cm−2 for trilayers), the cubic scaling turns into linear scaling. Notably, only linear scaling is observed in bilayer MoS2, where the P symmetry is restored. Intriguingly, although the monolayer and trilayer feature respectively K- and Q-valleys near their conduction-band edges, they display comparable valence-band Berry curvatures, valley Hall signatures, and micron-sized valley diffusion lengths. Our results not only offer the first experimental evidence for the intrinsic VHE but also help elucidate its fundamental origin in odd-layer TMDCs and pave the way for realizing room-temperature low- dissipation valleytronics by purely electronic means. Figure 1b is an optical image of a monolayer MoS2 field-effect transistor. The structure of this transistors is sketched in Fig. 1c. The device fabrication process includes a dry transfer step followed by a selective etching step25,26,31 (Please see Methods and Supplementary Fig.1 for details). A low contact barrier formed on the n-type MoS2 is evidenced by the I-V curves and the field-effect mobilities 𝜇 varied from 1,000 cm2V−1s−1 to 25,000 cm2V−1s−1 at T = 2 K (Supplementary Figs. 2 and 3). 4 As for the electronic measurement, an inverse VHE is exploited to detect a valley current, as sketched in Fig. 1d. An applied current I12 through probes 1 and 2 induces charge imbalance in a remote region, as measured by the voltage drop V34 between probes 3 and 4 (Supplementary Fig. 4). The nonlocal resistance RNL = V34 / I12 mediated by the valley Hall current was predicted32 to present cubic power-law dependence on the local resistance RL = V24 / I13. Nonlocal transport in monolayer MoS2. Nonlocal resistance RNL in an n-type monolayer MoS2 (sample B of length 𝐿 = 6 μm and width 𝑊 = 1.5 μm illustrated in Fig. 1d), measured as a function of gate voltage Vg at varied temperatures, is shown in Fig. 1e. A giant RNL is observed in the range of Vg ~ −15 to −25 V that amounts to the electron density n ~ 1010 to 1011 cm−2. In particular, the observed RNL ~ 106 Ω exceeds the classical ohmic contribution 𝑅CL = 𝑅L W πL e−π𝐿/𝑊 ~ 104 Ω by two orders of magnitude in the range of Vg ~ −15 to −18 V at 2 K and Vg ~ −22 to −25 V at 300 K. Another unexpected feature of RNL is its Vg dependence. In sharp contrast to the classical contribution 𝑅CL , which decreases gradually with increasing Vg, the observed RNL drops by at least one order of magnitude within an increase of several volts in Vg. Both the pronounced nonlocal signal and its unusual sensitivity to Vg suggest that the observed RNL has a physical origin different from the classical ohmic contribution 𝑅CL. The temperature dependence of RL and RNL uncovers the mesoscopic mechanism of both the local and nonlocal transport. The conduction can be separated into three regimes: 5 the thermal activation (TA) at 250 K<T<130 K, the nearest-neighbor hopping (NNH) at 130 K<T<60 K, and the variable-range hopping (VRH) below 60 K (sample A of 𝐿 = 3.6 μm and 𝑊 = 1.5 μm, see Fig. 2f, Supplementary Figs. 5a and 5b). Interestingly, the characteristic temperatures of both NNH and VRH for RNL are much larger than those for RL in the range of Vg ~ −60 to −58 V (Supplementary Figs. 5d and 5e). This indicates a higher energy barrier in the nonlocal transport and implies an anomalous origin of the nonlocal signal. To determine the origin of the observed RNL, we investigate the scaling relation between RNL and RL as functions of Vg at different temperatures for both sample A (Fig. 2) and sample B (Supplementary Fig. 6). For a fixed Vg, both RL and RNL increase when the temperature is lowered. In sample A, two regimes with distinct scaling behaviors become clearly visible in Fig. 2c, the logarithmic plot of RL and RNL at different temperatures. Above 160 K, the slopes of the lnRNL versus lnRL curves are 1, indicating that the nonlocal signal is proportional to the classical contribution. Below 160 K, the slopes turn to 3 in the low electron density regime (RL ≈ 108–109 Ω), which amounts to 𝑅NL ∝ 𝑅L 3. Such a cubic scaling is reminiscent of the nonlocal charge transport mediated by spin diffusion in the spin Hall effect32. As introduced above and calculated later (Fig. 4), the massive Dirac band structure produces large valley Hall conductivity 𝜎xy V (ref. 13,14) for monolayer MoS2. Thus, it is natural to attribute the observed nonlocal signal to the VHE, and the VHE mediated 𝑅NL follows32 𝑅NL = 2 ) 1 2 V 𝜎xy ( 𝜎xx 𝑊 𝜎xx𝑙V − e 6 𝐿 𝑙V ∝ (𝜎xy V )2𝑅L 3. (1) Here 𝑙V is the valley diffusion length (or inter-valley scattering length), and 𝜎xx and 𝑅L have the simple relation of 𝜎xx = 𝐿 𝑅L𝑊 . The RNL and RL data measured at different temperatures for the case of Vg = −60 V are plotted in Fig. 2d. The cubic law is not applicable above 160 K, probably due to the suppression of valley Hall conductivity at high temperatures. This is consistent with previous experiments in bilayer graphene9,11. Below 160 K, Eq. (1) can be employed to estimate 𝑙V . For the case of intermediate inter-valley scattering and edge roughness, 𝑙V~0.36 μm if we assume 𝜎xy V ~ 1 e2/h. In the limit of strong inter-valley scattering and edge roughness, 𝑙V ~ 0.43 μm if we assume 𝜎xy V ~ 0.1 e2/h. These values of 𝑙V are comparable to those obtained in graphene systems8-12,15,16. We further investigated the length dependence of the nonlocal valley transport. Apart from sample A (𝐿 = 3.6 μm) and sample B (𝐿 = 6 μm), two more samples (𝐿 = 11 μm and 16 μm ) are investigated (Supplementary Fig. 6). The semilog plot of RNL at n = 4×1010 cm-2 (extracted from the Hall measurement, see Supplementary Fig. 8) versus the sample length yields an estimation of 𝑙V ~ 1 μm (Fig. 2g). This value is very close to W and much larger than the electron mean free path lm ~ 20 nm estimated from the sample mobility 𝜇 for the range of n where the cubic scaling appears. Nevertheless, these estimations based on the observed nonlocal signals are suggestive of 𝑙V in the order of micron. In sample B, the cubic scaling remains even at room temperature, attributed to the dominant valence-band contribution to 𝜎xy V and particularly the large intrinsic bandgap that is impossible for graphene systems. 7 Nonlocal transport in bilayer and trilayer MoS2. For bilayer MoS2, the measured RL and RNL as functions of Vg at different temperatures are plotted in Figs. 3a–3b. As the carrier density increases, RL and RNL decrease in a similar fashion in the temperature range of 5–50 K. This yields a linear scaling behavior between RL and RNL, as analyzed in Figs. 3c–3d, and no cubic scaling is detected. We note that extrinsic P symmetry breaking can be introduced into atomically thin bilayers via external gating, as achieved in bilayer graphene9,10, and that detecting a nonlocal signal in gated bilayer graphene requires a threshold gating strength9,10. In our devices, however, Vg is too low to reach the threshold estimated by an recent optical experiment16; the estimated potential difference between the top and bottom layers is ~ 9.2 meV at Vg = −60 V16. This weak symmetry breaking produces little change in the total Berry curvature as compared with the pristine case (Supplementary Fig. 7), given the facts that the induced potential is much smaller than the bandgap and that the valence-band contribution to 𝜎xy V is dominant. In light of this analysis, the gating-induced P symmetry breaking is negligible in our bilayer MoS2. Therefore, we conclude that the absence of cubic scaling in bilayer MoS2 indicates the crucial role of strong P symmetry breaking in generating VHE. This key conclusion can be immediately tested in thicker MoS2 samples. Given that P symmetry is broken (respected) in pristine odd-layer (even-layer) MoS2, one might wonder whether the intrinsic VHE and its cubic scaling could be detected in trilayer MoS2. Figs. 3e and 3f display our RL and RNL data measured in trilayer MoS2 as functions of Vg at different temperatures. Evidently, the measured RNL rapidly decreases as Vg increases in the narrow range of −20 V ＜ Vg＜ −18.4 V, which is reminiscent of the behavior of RNL in our 8 monolayer devices in the low density regime. Similar to the monolayer case, the logarithmic plots of RL and RNL in Fig. 3g exhibit clear changes in slop from 1 to 3 near Vg = −18.4 V, further confirming the observation of the nonlocal signal of VHE in trilayer MoS2. To illustrate the temperature dependence, Fig. 3h plots the scaling relation between RL and RNL at different temperatures for the case of Vg = −20 V. Again, there is a clear change in slop from 1 to 3 near 30 K. Moreover, the valley diffusion length can be extracted based on Fig. 3h and Eq. (1). We obtain 𝑙V~ 0.5 μm and ~1 μm, respectively, for the aforementioned two limits 𝜎xy V ~1 e2/h and ~0.1 e2/h. Both the observed amplitude of nonlocal signal and the estimated valley diffusion length in the trilayer MoS2 devices are comparable to those in the monolayer case. In addition to the crucial role of P symmetry breaking, these observations are suggestive of a universal physical origin of VHEs in odd- layer TMDCs. Discussion To better understand the thickness dependent observations, we calculate13,14 the electronic band structures and Berry curvatures for monolayer, bilayer, and trilayer MoS2. The band structures in Figs. 4a-4c are indeed thickness dependent. In particular, the conduction-band minima lie at the K-valleys for the monolayer, whereas they shift to the Q-valleys for the bilayer and trilayer. Given the low electron densities in our samples (~3×1010 cm-2 in monolayers and ~1×1011 cm-2 in bilayers and trilayers), the Fermi levels only cross the lowest sub-bands, as indicated by the green lines in Figs. 4a-4c. As bilayer MoS2 has a restored P symmetry that is intrinsically broken in odd-layer MoS2, the sub- bands are spin degenerate in the bilayer yet spin split in the monolayer and trilayer. With 9 the band structures, we further derive the Berry curvatures that can drive the VHEs. Berry curvature vanishes if both P and T are present. As plotted in Figs. 4d-4f, our calculations reveal that the curvatures are indeed trivial in the bilayer yet substantial in the monolayer and trilayer. This explains the reason why no cubic scaling is observed in bilayer MoS2 and highlights the role of P symmetry breaking in producing VHEs. It is puzzling to understand and compare the nonlocal signals of VHEs in monolayer and trilayer MoS2. Similar cubic scaling behaviors and their transitions to linear ones above the critical densities or temperatures are observed in both cases. However, the conduction- band Berry curvatures (the difference between the blue and orange curves in Fig. 4d and Fig. 4f) are considerably large in the monolayer K-valleys yet negligibly small in the trilayer Q-valleys. This implies that the geometric VHE requiring finite doping3 might not be the origin9, which is further evidenced by the fact that the cubic scaling behaviors weaken rapidly with increasing the electron densities. On the other hand, these facts appear to be in harmony with the topological VHE that arises from the valence-band contributions4-7. Particularly, the monolayer and trilayer share almost identical valence-band Berry curvatures (the orange curves in Fig. 4d and Fig. 4f) and valley Hall conductivities, due to the extremely weak interlayer couplings. Recently, nearly quantized edge transports have been observed along the designed or selected domain walls in graphene systems11,12 and even in artificial crystals17. In our case, the roughness of natural edges can cause edge inter-valley scattering6 and passivate the edge states33. The electrons in the passivated edge states are likely responsible for the nonlocal signals of VHE; these edge states are partially gapped and well spread into the interiors, leading to 10 bulk instead of edge valley transport. The pronounced nonlocal signals are observed in our MoS2 samples with length up to 16 μm and at temperature up to 300 K. The valley diffusion lengths are also estimated to be in the order of micron. The low carrier concentration ensures the low possibility of bulk inter-valley scattering and maintains a long valley diffusion length. In addition, the mirror and T symmetries lock the spin and valley indices of the lowest sub-bands, preventing bulk inter-valley scattering via spin conservation. Our observed intrinsic VHEs and their long valley diffusion lengths are promising for realizing room-temperature low-dissipation valleytronics. To better elucidate the outstanding problems of both geometric3 and topological4-7 VHEs, our observations and analyses call for future efforts, particularly complementary experiments in p-type TMDCs. 11 Methods (a) Van der Waals structures MoS2 bulk crystals are bought from 2Dsemiconductors (website: http://www.2dsemiconductors.com/), and the h-BN sources (grade A1) were bought from HQgraphene (website: http://www.hqgraphene.com/). To fabricate van der Waals heterostructures, a selected MoS2 sample is picked up from the SiO2/Si substrate by a thin h-BN flake (5-15 nm thick) on PMMA (950 A7, 500 nm) via van der Waals interactions. The h-BN/MoS2 flake is then transferred onto a fresh thick h-BN flake lying on another SiO2/Si substrate, to form a BN-MoS2-BN heterostructure (step 1 in Supplementary Fig. 1). (b) Selective etching process A hard mask is patterned on the heterostructure by the standard e-beam lithography technique using PMMA (step 2 in Supplementary Fig. 1). The exposed top BN layer and MoS2 are then etched via reactive ion etching (RIE), forming a Hall bar geometry (steps 3 & 4 in Supplementary Fig. 1). Then a second-round e-beam lithography and RIE is carried out to expose the MoS2 layer (steps 5 & 6 in Supplementary Fig. 1). The electrodes are then patterned by a third-round e-beam lithography followed by a standard e-beam evaporation (steps 7 & 8 in Supplementary Fig. 1). To access the conduction band edges of MoS2, we choose Titanium as the contact metal, as the work function of Titanium (~4.3 eV) matches the band-edge energy of MoS2 (~4.0 to 4.4 eV depending on the layer numbers). (c) Electronic measurement 12 The I–V curves are measured by Keithley 6430. Other transport measurements are carried out by using: (i) low-frequency lock-in technique (SR 830 with SR550 as the preamplifier and DS 360 as the function generator, or (ii) Keithley 6430 source meter (> 1016 Ω input resistance on voltage measurements). The cryogenic system provides stable temperatures ranging from 1.8 to 300 K. A detailed discussion of the nonlocal measurement is presented in Supplementary Fig. 4. References 1 Rycerz, A., Tworzydlo, J. & Beenakker, C. W. J. Valley filter and valley valve in 2 3 graphene. Nat. Phys. 3, 172-175 (2007). Zhang, F. Brought to light. Nature Physics 14, 111 (2017). Xiao, D., Yao, W. & Niu, Q. Valley-Contrasting Physics in Graphene: Magnetic Moment and Topological Transport. Phys. Rev. Lett. 99, 236809 (2007). 4 Zhang, F., Jung, J., Fiete, G. A., Niu, Q. & MacDonald, A. H. Spontaneous quantum Hall states in chirally stacked few-layer graphene systems. Phys. Rev. Lett. 106, 156801 (2011). 5 Martin, I., Blanter, Y. M. & Morpurgo, A. F. Topological Confinement in Bilayer Graphene. Phys. Rev. Lett. 100, 036804 (2008). 6 Zhang, F., MacDonald, A. H. & Mele, E. J. Valley Chern numbers and boundary modes in gapped bilayer graphene. Proc. Natl. Acad. Sci. 110, 10546-10551 (2013). 7 Vaezi, A., Liang, Y., Ngai, D. H., Yang, L. & Kim, E.-A. Topological Edge States at 13 a Tilt Boundary in Gated Multilayer Graphene. Phys. Rev. X 3, 021018 (2013). 8 Gorbachev, R. V. et al. Detecting topological currents in graphene superlattices. Science 346, 448-451 (2014). 9 Sui, M. et al. Gate-tunable topological valley transport in bilayer graphene. Nat. Phys. 11, 1027-1031 (2015). 10 Shimazaki, Y. et al. Generation and detection of pure valley current by electrically induced Berry curvature in bilayer graphene. Nat. Phys. 11, 1032-1036 (2015). 11 Ju, L. et al. Topological valley transport at bilayer graphene domain walls. Nature 520, 650-655 (2015). 12 Li, J. et al. Gate-controlled topological conducting channels in bilayer graphene. Nat. Nanotechnol. 11, 1060-1065 (2016). 13 Xiao, D., Liu, G., Feng, W., Xu, X. & Yao, W. Coupled spin and valley physics in monolayers of MoS2 and other group-VI dichalcogenides. Phys. Rev. Lett. 108, 196802 (2012). 14 Li, X., Zhang, F. & Niu, Q. Unconventional Quantum Hall Effect and Tunable Spin Hall Effect in Dirac Materials: Application to an Isolated MoS2 Trilayer. Phys. Rev. Lett. 110, 066803 (2013). 15 Mak, K. F., McGill, K. L., Park, J. & McEuen, P. L. The valley Hall effect in MoS2 transistors. Science 344, 1489-1492 (2014). 16 Lee, J., Mak, K. F. & Shan, J. Electrical control of the valley Hall effect in bilayer MoS2 transistors. Nat. Nanotechnol. 11, 421-425 (2016). 17 Lu, J. et al. Observation of topological valley transport of sound in sonic crystals. 14 Nat. Phys. 13, 369-374 (2017). 18 Mak, K. F., He, K., Shan, J. & Heinz, T. F. Control of valley polarization in monolayer MoS2 by optical helicity. Nat. Nanotechnol. 7, 494-498 (2012). 19 Zeng, H., Dai, J., Yao, W., Xiao, D. & Cui, X. Valley polarization in MoS2 monolayers by optical pumping. Nat. Nanotechnol. 7, 490-493 (2012). 20 Cao, T. et al. Valley-selective circular dichroism of monolayer molybdenum disulphide. Nat. Commun. 3, 887 (2012). 21 Mak, K. F., Lee, C., Hone, J., Shan, J. & Heinz, T. F. Atomically thin MoS2: A new direct-gap semiconductor. Phys. Rev. Lett. 105, 136805 (2010). 22 Splendiani, A. et al. Emerging photoluminescence in monolayer MoS2. Nano Lett. 10, 1271-1275 (2010). 23 Cui, X. et al. Multi-terminal transport measurements of MoS2 using a van der Waals heterostructure device platform. Nat. Nanotechnol. 10, 534-540 (2015). 24 Movva, H. C. P. et al. High-Mobility Holes in Dual-Gated WSe2 Field-Effect Transistors. ACS Nano 9, 10402-10410 (2015). 25 Shuigang, X. et al. Universal low-temperature Ohmic contacts for quantum transport in transition metal dichalcogenides. 2D Mater. 3, 021007 (2016). 26 Wu, Z. et al. Even-odd layer-dependent magnetotransport of high-mobility Q-valley electrons in transition metal disulfides. Nat. Commun. 7, 12955 (2016). 27 Pisoni, R. et al. Gate-Defined One-Dimensional Channel and Broken Symmetry States in MoS2 van der Waals Heterostructures. Nano Lett. 17, 5008-5011 (2017). 28 Fallahazad, B. et al. Shubnikov-de Haas Oscillations of High-Mobility Holes in 15 Monolayer and Bilayer WSe2: Landau Level Degeneracy, Effective Mass, and Negative Compressibility. Phys. Rev. Lett. 116, 086601 (2016). 29 Movva, H. C. P. et al. Density-Dependent Quantum Hall States and Zeeman Splitting in Monolayer and Bilayer WSe2. Phys. Rev. Lett. 118, 247701 (2017). 30 Gustafsson, M. V. et al. Ambipolar Landau levels and strong band-selective carrier interactions in monolayer WSe2. Nat. Mater. advance online publication, doi:10.1038/s41563-018-0036-2 (2018). 31 Xu, S. et al. Odd-Integer Quantum Hall States and Giant Spin Susceptibility in p- Type Few-Layer WSe2. Phys. Rev. Lett. 118, 067702 (2017). 32 Abanin, D. A., Shytov, A. V., Levitov, L. S. & Halperin, B. I. Nonlocal charge transport mediated by spin diffusion in the spin Hall effect regime. Phys. Rev. B 79, 035304 (2009). 33 Wu, D. et al. Uncovering edge states and electrical inhomogeneity in MoS2 field- effect transistors. Proc. Natl. Acad. Sci. 113, 201605982 (2016). 16 Acknowledgments We thank Wing Ki Wong for the assistance in the sample preparation process. We acknowledge the financial support from the Research Grants Council of Hong Kong (Project Nos. 16302215, HKU9/CRF/13G, 604112 and N_HKUST613/12) and the UT- Dallas Research Enhancement Funds (toward F.Z.). We are grateful for the technical support of the Raith–HKUST Nanotechnology Laboratory at MCPF. F.Z. is grateful to Di Xiao and Allan MacDonald for valuable discussions. 17 Figure 1 Valley Hall transport induced nonlocal resistance in a monolayer MoS2 field- effect transistor. a, Crystal structure of an odd/even-layer 2H-MoS2 is inversion asymmetric/symmetric. b, Optical image of a typical monolayer MoS2 device. A MoS2 Hall bar is sandwiched between the top and bottom h-BN flakes. Scale bar: 3 μm. c, Schematic of the h-BN encapsulated MoS2 field-effect transistor. d, Schematic of the nonlocal resistance measurement and the VHE-mediated nonlocal transport. The applied charge current in the left circuit generates a pure valley current in the transverse direction via a VHE. This valley current induces opposite chemical potential gradients for the two valleys over the inter-valley scattering length, which, in turn, generates a voltage drop measured by probes 3 and 4 in the right circuit via an inverse VHE. e, Nonlocal resistance RNL (upper panel) and the classical ohmic contribution RCL (lower panel) as functions of gate voltage Vg at varied temperatures. 18 Figure 2 Local and nonlocal resistances of monolayer MoS2. a, b, Semilog plots of RL and RNL as a function of Vg measured at varied temperatures. Inset of b: optical micrograph of our typical h-BN/MoS2/h-BN device with multi-terminal Hall Bar configurations. c, Scaling relation between ln RL and ln RNL at temperatures ranging from 300 K to 2 K. When the electron density is relatively high, i.e., RL and RNL are small, RNL is linearly proportional to RL. Below 200 K, RNL and RL follow a cubic relationship at low electron densities, indicating that the contribution from valley Hall transport dominates. The crossover from linear to cubic scaling is observed around the critical density of nc = 5×1010 cm−2, with gate voltage Vg = −57 V. d, Crossover phenomenon by considering classical diffusion (RNL∝ RL) and valley Hall transport (RNL∝RL 3). The experimental data (black circles, Vg = −60 V) clearly show two different regimes which are fitted by two linear curves (orange dashed line with slope 1 and blue dashed line with slope 3). e, RNL plotted as a function of Vg at 19 low temperatures. The ohmic contribution, calculated according to RL and device geometry, is deducted from the measured RNL at different temperatures. f, 1/RNL (orange circles) and 1/RL (blue circles) in log scale plotted as functions of 1/T at Vg = −60 V. Three distinct transport regimes were observed: the thermal activation (TA) transport, nearest neighbor hopping (NNH) transport, and the variable range hopping (VRH) transport. g, Semilog plot of RNL as a function of L at n = 4×1010 cm−2 (orange squares). Nonlocal signal decays exponentially with increasing L. The dashed line yields a valley diffusion length of ~ 1 μm. 20 Figure 3 Local and nonlocal resistances of bilayer and trilayer MoS2. a,b,e,f, Gate- dependence of RL and RNL at different temperatures in bilayer (a,b) and trilayer (e,f) samples. c,g, Scaling relation between ln RL and ln RNL is obtained at different temperatures in bilayer (c) and trilayer (g) samples. For the trilayer case, RNL scales linearly with RL in the high electron density regime, whereas the cubic scaling law RNL∝RL 3 is observed in the low electron density regime (nc = 1.1×1012 cm−2 or Vg = −18.4V). d, lnRL v.s. lnRNL for bilayer MoS2. In the full range of gate voltages, RNL scales linearly with RL, and the experimental data (black dots, Vg = −60 V) is fitted by a linear curve (red solid line). h, lnRL v.s. lnRNL for trilayer MoS2. The experimental data (black dots, Vg = −20 V) clearly show two different regimes which are fitted by two linear curves (red solid line with slope 1 and blue solid line with slope 3). Evidently, a crossover exists from linear (RNL∝RL) to cubic scaling behaviors (RNL∝RL 3). 21 Figure 4 Calculated band structures and Berry curvatures of monolayer, bilayer, and trilayer MoS2. a-c, Band structure of (a) monolayer, (b) bilayer, and (c) trilayer MoS2. The conduction band edges lie at the K-valleys in the monolayer but at the Q-valleys in the bilayer and trilayer. Insets of a-c: The Fermi levels only cross the lowest sub-bands, which are spin degenerate in (b) but spin split in a and c. d-f, Berry curvatures of (d) monolayer, (e) bilayer, and (f) trilayer MoS2. The blue curves are the total curvatures of all occupied states below the Fermi levels, whereas the orange curves are the total curvatures of all valence-band states. The red arrow in f points out a tiny bump at a Q valley. Insets of d-f, 2D mapping of Berry curvatures in the 2D Brillouin zone (white dashed lines). 22
1408.4740	3	1408	2015-10-30T20:51:28	Capacitively coupled singlet-triplet qubits in the double charge resonant regime	[ "cond-mat.mes-hall" ]	We investigate a method for entangling two singlet-triplet qubits in adjacent double quantum dots via capacitive interactions. In contrast to prior work, here we focus on a regime with strong interactions between the qubits. The interplay of the interaction energy and simultaneous large detunings for both double dots gives rise to the double charge resonant regime, in which the unpolarized (1111) and fully polarized (0202) four-electron states in the absence of interqubit tunneling are near degeneracy, while being energetically well-separated from the partially polarized (0211 and 1102) states. A rapid controlled-phase gate may be realized by combining time evolution in this regime in the presence of intraqubit tunneling and the interqubit Coulomb interaction with refocusing ${\pi}$ pulses that swap the singly occupied singlet and triplet states of the two qubits via, e.g., magnetic gradients. We calculate the fidelity of this entangling gate, incorporating models for two types of noise -- charge fluctuations in the single-qubit detunings and charge relaxation within the low-energy subspace via electron-phonon interaction -- and identify parameter regimes that optimize the fidelity. The rates of phonon-induced decay for pairs of GaAs or Si double quantum dots vary with the sizes of the dipolar and quadrupolar contributions and are several orders of magnitude smaller for Si, leading to high theoretical gate fidelities for coupled singlet-triplet qubits in Si dots. We also consider the dependence of the capacitive coupling on the relative orientation of the double dots and find that a linear geometry provides the fastest potential gate.	cond-mat.mes-hall	cond-mat	Capacitively coupled singlet-triplet qubits in the double charge resonant regime V. Srinivasa1, 2, 3, ∗ and J. M. Taylor1, 2, 4 1Joint Quantum Institute, University of Maryland, College Park, Maryland 20742, USA 2National Institute of Standards and Technology, Gaithersburg, Maryland 20899, USA 3Laboratory for Physical Sciences, College Park, Maryland 20740, USA 4Joint Center for Quantum Information and Computer Science, University of Maryland, College Park, Maryland 20742, USA 5 1 0 2 t c O 0 3 ] l l a h - s e m . t a m - d n o c [ 3 v 0 4 7 4 . 8 0 4 1 : v i X r a We investigate a method for entangling two singlet-triplet qubits in adjacent double quantum dots via capacitive interactions. In contrast to prior work, here we focus on a regime with strong interactions between the qubits. The interplay of the interaction energy and simultaneous large detunings for both double dots gives rise to the "double charge resonant" regime, in which the unpo- larized (1111) and fully polarized (0202) four-electron states in the absence of interqubit tunneling are near degeneracy, while being energetically well-separated from the partially polarized (0211 and 1102) states. A rapid controlled-phase gate may be realized by combining time evolution in this regime in the presence of intraqubit tunneling and the interqubit Coulomb interaction with refo- cusing π pulses that swap the singly occupied singlet and triplet states of the two qubits via, e.g., magnetic gradients. We calculate the fidelity of this entangling gate, incorporating models for two types of noise – charge fluctuations in the single-qubit detunings and charge relaxation within the low-energy subspace via electron-phonon interaction – and identify parameter regimes that optimize the fidelity. The rates of phonon-induced decay for pairs of GaAs or Si double quantum dots vary with the sizes of the dipolar and quadrupolar contributions and are several orders of magnitude smaller for Si, leading to high theoretical gate fidelities for coupled singlet-triplet qubits in Si dots. We also consider the dependence of the capacitive coupling on the relative orientation of the double dots and find that a linear geometry provides the fastest potential gate. I. INTRODUCTION (qubits) Electrons spins confined within semiconductor quan- tum dots form the basis of a highly controllable and potentially scalable approach to solid-state quan- tum information processing.1–5 The encoding of spin quantum bits in two-electron singlet and triplet states of a double quantum dot2,6–8 enables rapid, universal manipulation via tuning of the singlet- triplet (exchange) splitting through electrical control over the double-dot potential9 combined with static magnetic field gradients,10,11 without requiring time- dependent magnetic fields and while simultaneously pro- viding protection against errors induced by hyperfine interaction.8,9,12–18 Coherent control of singlet-triplet qubits has been experimentally demonstrated in the con- text of both single-qubit manipulation9,11,17,19–21 and two-qubit entanglement.22,23 For a pair of singlet-triplet qubits coupled via tunnel- ing, the effective exchange interaction can be used to carry out two-qubit gates;7,24–28 however, this approach typically requires an accompanying mechanism for sup- pressing errors due to leakage out of the qubit subspace during gate operation. Alternatively, two singlet-triplet qubits in adjacent double dots may be entangled via capacitive coupling.2,22,23,29–36 In this case, interqubit tunneling is absent and the entanglement instead origi- nates from the Coulomb interaction of the multipole mo- ments associated with the different charge distributions of the singlet and triplet states.15 The spin-dependent charge dipole moments of spatially separated singlet- triplet qubits can also be coupled to microwaves, enabling long-range, high-frequency gating.37,38 Nevertheless, re- alizing robust entangling gates in the presence of the charge-based decoherence mechanisms typically existing in the solid state, including both dephasing8,15,39–42 and relaxation via, e.g., coupling to phonons,8,12,43–48 re- mains challenging. Here, we consider a pair of capacitively coupled singlet- triplet qubits in the absence of interqubit tunneling. In contrast to the repulsive interqubit dipole-dipole interac- tion originally considered in Ref. 2, we focus specifically on the case of an attractive dipole-dipole interaction, im- plemented by adjusting via external gate voltages the energy detunings between the singly and doubly occu- pied two-electron charge configurations such that they are large for both double dots. The interplay of these large detunings and the Coulomb interaction energy gives rise to the "double charge resonant" regime, as we de- scribe below. Combining time evolution in this regime with single- qubit π pulses that swap the singly occupied singlet and triplet states of both qubits using, e.g., static magnetic gradients23 leads to a controlled π-phase (or controlled- Z) entangling gate. As a consequence of the attractive dipole-dipole interaction, increasing the speed of this gate simultaneously decreases the gate error due to charge noise. We calculate the gate fidelity in the presence of charge fluctuations in the double-dot detunings and iden- tify gate voltages and coupling strengths at which the fidelity is optimized. We then investigate charge relax- ation due to electron-phonon coupling for both GaAs and Si double quantum dots in linear and purely quadrupo- lar dot configurations and determine the effects of both this relaxation and fast charge noise on the gate fidelity. Finally, we consider the geometry dependence of the in- terqubit capacitive coupling and identify the linear geom- etry as a configuration that maximizes the gate speed. II. MODEL AND DOUBLE CHARGE RESONANT REGIME We consider two singlet-triplet qubits, realized within a pair of adjacent two-electron double quantum dots [Fig. 1(a)] with only the lowest orbital level of each dot taken into account. Each two-electron double dot encodes one qubit. As in Ref. 2, we initially assume a linear geometry in which the tunnel barriers are adjusted via gates such that tunneling occurs only between the dots within each qubit, while adjacent dots belonging to different qubits are coupled purely capacitively. We can write a Hubbard Hamiltonian for the system49 as Hhub = Ha + Hb + Hint, where (cid:21) Uα 2 nαi(nαi − 1) Hα = Hαn + Hαt, (cid:20) (cid:88) (cid:88) (cid:88) i(cid:54)=j σ αinαi + i=1,2 + Vαnα1nα2, † tαc αiσcαjσ, Hαn = Hαt = (1) (2) (3) is the Hamiltonian for double dot α = a, b, and Hint is the capacitive interaction between the double dots. For sim- plicity, we initially include only the dominant interaction term for the linear geometry we consider, Hint = Uabna2nb1. σ nαiσ = (cid:80) tron number operators nαi = (cid:80) (4) Equations (2) and (4) are expressed in terms of the elec- † σ c αiσcαiσ, † where c αiσ creates an electron in dot i of qubit α with spin σ =↑,↓ and orbital energy αi. These terms deter- mine the energy of each four-electron charge configura- tion na1 na2 nb1 nb2(cid:105) in the absence of interdot tunnel- ing. The quantities Uα and Vα are the Coulomb repulsion energies for two electrons in the same dot and in differ- ent dots within qubit α, respectively. Hαt couples the double-dot charge configurations (nα1, nα2) via tunnel- ing, and tα denotes the tunneling amplitude for double dot α. As discussed in Refs. 2 and 8, each double dot can be described in the two-electron regime as an effective three-level system with a state space spanned by (cid:16) (cid:16) (cid:17) (cid:17) † † † † 2↑ 1↓c 2↓ + c 1↑c c † † † † 1↑c 2↓ − c 1↓c 2↑ c 0(cid:105) , 0(cid:105) , (5) (6) T11(cid:105) ≡ (1, 1) T0(cid:105) = 1 √2 1 √2 S11(cid:105) ≡ (1, 1) S(cid:105) = † 2↑c S02(cid:105) ≡ (0, 2) S(cid:105) = c (7) where the qubit index α has been suppressed for clarity. † 2↓ 0(cid:105) , 2 Figure 1. (a) Schematic diagram of capacitively coupled dou- ble quantum dots in the charge states 1111(cid:105) and 0202(cid:105) . The interdot spacing within each double dot is d, and the separa- tion between the centers of the double dots is R. (b) Energy level diagram for the main four-electron charge configurations considered in the present work, illustrating the double charge resonant regime. 0202(cid:105) , In our analysis of the capacitively coupled double- dot pair system, we focus on the four-electron charge subspaces 1111(cid:105) , 1102(cid:105) , and 0211(cid:105) . Noting that Hhub conserves both the total spin and the to- tal z component of spin and that Hαt couples only the two-electron singlet states S11(cid:105) and S02(cid:105) within double dot α, we may consider the subspace spanned by product states of the form Sa, Sb(cid:105) ≡ Sa(cid:105) ⊗ Sb(cid:105) , where Sα(cid:105) ∈ {S11(cid:105) ,S02(cid:105)} for α = a, b. In the basis {S11, S11(cid:105) ,S02, S02(cid:105) ,S11, S02(cid:105) ,S02, S11(cid:105)} , the Hamil- tonian has the representation  0 0 √2tb √2ta δ √2ta √2tb 0 ∆a 0 √2tb √2ta ∆b √2ta √2tb 0  , Hhub = (8) where ∆a ≡ −a +Ua−Va +Uab and ∆b ≡ −b +Ub−Vb− Uab are the effective energy detunings between S11(cid:105) and S02(cid:105) for double dots a and b, respectively (accounting for coupling to the other double dot), α ≡ α1 − α2, and δ ≡ ∆a + ∆b − Uab is the energy difference be- tween S11, S11(cid:105) and S02, S02(cid:105) . The detunings ∆α are controlled via tuning of the on-site energies via gate volt- ages, which set a and b. The controlled-phase gate for two singlet-triplet qubits III involves tunneling from S11(cid:105) to discussed in Sec. S02(cid:105) for α = a, b, which simultaneously induces dipole moments in both double dots. Given that the interqubit Coulomb interaction strength Uab > 0, the regime of interest for the operation of this gate is that in which 111102021111021111020202(cid:303)(cid:168)a(cid:168)b(cid:168)a(cid:14)(cid:3)(cid:168)bUabRdd(a)(b)EtaUabtba1a2b1b2∆α (cid:29) δ > 0 for α = a, b, so that S11, S11(cid:105) is lower in energy than S02, S02(cid:105), while S11, S11(cid:105) and S02, S02(cid:105) are energetically well-separated from S11, S02(cid:105) and S02, S11(cid:105) [Fig. 1(b)]. We refer to this regime as the "double charge resonant regime," as the attractive interaction be- tween the dipole moments of the two double dots in the state S02, S02(cid:105) effectively brings it into near-resonance with S11, S11(cid:105) . Note that this regime is not accessible in the scenario originally studied in Ref. 2, where the state S02, S20(cid:105) is considered instead of S02, S02(cid:105) and the interqubit Coulomb interaction between the dipole mo- ments is repulsive. We show in Sec. III that, in contrast to the nonreso- nant regime of Ref. 2, the double charge resonant regime enables a controlled-phase gate to be generated by dy- namics within an effective low-energy subspace derived from S11, S11(cid:105) and S02, S02(cid:105) . In order to compare two- qubit phase gates in the nonresonant and double charge resonant regimes, we now estimate the scaling of the phase gate errors in the presence of detuning noise. Us- ing Eq. (8), we calculate the fourth-order energy shift for the state S11, S11(cid:105) due to the interqubit capacitive coupling (i.e., the additional energy shift for Uab (cid:54)= 0), which gives the rate of the phase gate. For the non- resonant regime, the Hamiltonian has the same form as Eq. (8) with the replacements S02, S02(cid:105) → S02, S20(cid:105) , S11, S02(cid:105) → S11, S20(cid:105) , and δ = ∆a + ∆b + Uab (here, ∆b ≡ b + Ub − Vb + Uab). Assuming ∆a = ∆b ≡ ∆ and ta = tb ≡ t for simplicity, we find a phase gate rate ε1111 = 8t4 (δ − 2∆) /∆3δ. Setting ∆(cid:48) = ∆+ξ, where ξ represents classical, static, Gaussian-distributed noise in the single-qubit detunings due to gate voltage fluctuations,42 we can write the non- trivial phase factor acquired by the state derived from , with φ(cid:48) = φ0 + φξ. Here, S11, S11(cid:105) for Uab (cid:54)= 0 as eiφ φ0 ≡ ε1111τgate is the phase acquired during the gate time τgate in the absence of noise and φξ ≈ (cid:48)(cid:69) (cid:16) represents the phase fluctuations, approximated using the second-order energy shift. Averaging over the noise −Γ2 gives eiφ0 with = e 2,ξ (T ∗ Γξ ∼ t2/∆2T ∗ 2,ξ denotes the dephasing time as- sociated with the charge fluctuations ξ), so that the phase gate error can be approximated as err ≈ Γ2 (cid:16) (cid:17)2 ξτ 2 gate ∼ δ2∆2/t4 (δ − 2∆)2 T ∗2 2,ξ. For the nonresonant regime, δ = 2∆ + Uab and err ∼ , while for the double charge resonant regime, δ = 2∆−Uab tT ∗ . Thus, the phase gate er- ror due to gate voltage fluctuations has the scaling ∼ ∆2/t2 for the nonresonant regime, which is unfavorable for suppressing errors arising from dipole transitions to S11, S20(cid:105) and S02, S11(cid:105) by keeping ∆/t large. On the other hand, the error scales as ∼ δ2/t2 in the double charge resonant regime, so that faster gates (correspond- ing to stronger coupling Uab and therefore smaller δ for fixed ∆) are also associated with smaller error. (cid:0)∆2/t2(cid:1) (1 + 2∆/Uab)2 / (cid:17)2 (cid:0)4t2/∆2(cid:1) ξτgate ξτ 2(cid:17) and err ∼ δ2/t2(cid:16) eiφ0 ≈ 1 − Γ2 tT ∗ (cid:68) eiφ ξτ 2 2,ξ 2,ξ (cid:48) 3 We now proceed with a more detailed analy- sis of the double charge resonant regime and con- sider the low-energy effective Hamiltonian in the sub- space {S11, S11(cid:105) ,S02, S02(cid:105)} . Applying a Schrieffer-Wolff transformation of the form H = eλAHhube−λA with λ ∝ tα (assuming ta ∼ tb) to the Hamiltonian in Eq. (8), we choose A such that the coupling to the higher- energy states S11, S02(cid:105) and S02, S11(cid:105) is eliminated up to O this transformation are given in Appendix A. (cid:0)λ2(cid:1) . Expressions for the basis states resulting from (cid:12)(cid:12)(cid:12) − (cid:12)(cid:12)(cid:12) , the effective Hamiltonian within (cid:12)(cid:12)(cid:12) (cid:94)S02, S02 (cid:69)(cid:68) (cid:69)(cid:68) (cid:94)S11, S11 (cid:94)S02, S02 Defining σz = the transformed subspace is Heff = − Ja + Jb − jd 2 σz − jxσx, (9) where, in terms of the charge admixture parameters ηα ≡ tα/∆α, δ, and the difference of the detunings ∆d ≡ ∆a − ∆b, (cid:12)(cid:12)(cid:12) (cid:94)S11, S11 (cid:19) jd 2 1 − (cid:18) Ja ≡ η2 a (Uab + δ + ∆d) , b (Uab + δ − ∆d) , Jb ≡ η2 (cid:18) δ + ∆d Uab − δ − ∆d δ − ∆d Uab − δ + ∆d (cid:19) , jd ≡ δ − 2 Ja +Jb (cid:20) (cid:18) jx ≡ 2ηaηbUab 1 + δ 1 (10) (11) (12) (cid:19)(cid:21) . (13) (cid:12)(cid:12)(cid:12) (cid:94)S11, S11 (cid:69) Diagonalization of Eq. g(cid:105) = cos θ sin θ (cid:12)(cid:12)(cid:12) (cid:94)S11, S11 (cid:69) (cid:12)(cid:12)(cid:12) (cid:94)S02, S02 + cos θ − sin θ (cid:69) Uab − δ − ∆d + 1 Uab − δ + ∆d (cid:12)(cid:12)(cid:12) (cid:94)S02, S02 (cid:69) (9) yields the eigenstates and e(cid:105) = , where (14) tan θ = (cid:112) j2 d + 4j2 jd − Ω 2jx and Ω ≡ Ee− Eg = x is the energy gap between g(cid:105) and e(cid:105) . The spectrum of Heff is shown in Fig. 2 as a function of δ for Uab = 200 µeV, ∆d = 0, and ηa = (cid:69) ηb ≡ η0 = 0.1. An avoided crossing occurs at δ = 0, i.e., when S11, S11(cid:105) and S02, S02(cid:105) are resonant. For δ (cid:29) 0, g(cid:105) ≈ (cid:12)(cid:12)(cid:12) (cid:94)S11, S11 (cid:12)(cid:12)(cid:12) (cid:94)S02, S02 and e(cid:105) ≈ (cid:69) . (cid:110) a rotation around the z axis of the Bloch sphere for the singlet-triplet qubit.2,8,9 Combining HαJ for α = a, b and Heff [Eq. (9)] yields, in the two-qubit basis T11, T11(cid:105) , (cid:12)(cid:12)(cid:12) (cid:94)S11, S11 (cid:12)(cid:12)(cid:12) S11, T11 (cid:12)(cid:12)(cid:12)T11, S11 (cid:69)(cid:111) (cid:69) (cid:69) (cid:69) , , , ,  0 −Ja −Jb HJ = −Ja − Jb −jx −jx −Ja − Jb + jd (17) The dynamics generated by HJ are described by the op- erator 4 (cid:12)(cid:12)(cid:12) (cid:94)S02, S02  .  , (cid:19)(cid:21) (18) (19) Figure 2. Spectrum of Heff [Eq. (9)] as a function of δ for Uab = 200 µeV, ∆d = 0, and ηa = ηb ≡ η0 = 0.1. III. CONTROLLED-PHASE GATE UJ (τ ) ≡ e −iHJ τ = eiJaτ eiJbτ , [Eq. (cid:69) (cid:69)(cid:111) (cid:12)(cid:12)(cid:12) (cid:94)S02, S02 (cid:110)(cid:12)(cid:12)(cid:12) (cid:94)S11, S11 The time evolution generated by the Hamiltonian Heff (9)] within the two-singlet subspace spanned by leads to a controlled-phase gate between the two singlet-triplet qubits that is based on ca- pacitive coupling in the double charge resonant regime. In order to obtain this two-qubit entangling gate, we now incorporate the triplet states of the double dots [Eq. (5)] into the analysis. Since the full sequence for the controlled-phase gate also involves single-qubit rotations around two orthogonal axes, we consider the Hamiltonian Hhub + HaZ + HbZ, where HαZ = gµB 2 c † αiσ (Bαi · σα) cαiσ(cid:48) (15) (cid:88) (cid:88) i=1,2 σ,σ(cid:48) represents Zeeman coupling to magnetic fields Bαi for double dot α (here, g is the effective electron g fac- tor and µB denotes the Bohr magneton). Combining this Zeeman coupling with the spin-independent Hub- bard term Hα [Eqs. (1)-(3)] that leads to the exchange Jα enables universal one-qubit control of singlet-triplet qubits.2,7–9,11,17,19–21 In the basis {T11(cid:105) ,S11(cid:105) ,S02(cid:105)} [Eqs. (5)-(7)], H(cid:48) α ≡ Hα + HαZ takes the form2,8  0 (cid:48) α = H ωαZ 0 0 √2tα ωαZ 0 √2tα ∆α  , (16) where we have defined the energy associated with a static magnetic field gradient of magnitude dBα ≡ (Bα1 − Bα2) /2 along the single-spin quantization axis as ωαZ ≡ gµBdBα. We initially consider the double charge resonant regime (∆α (cid:29) δ) in the limit ωαZ → 0 and keep only the term Hα in the Hamiltonian for double dot α. Elimination of the doubly occupied singlet state S02(cid:105) gives the effective Hamiltonian HαJ ≡ −Jα where Z ≡ T11(cid:105)(cid:104)T11 − (cid:12)(cid:12)(cid:12) = Jα (Zα − 1) /2, (cid:12)(cid:12)(cid:12) , which generates (cid:12)(cid:12)(cid:12) S11 (cid:69)(cid:68) (cid:12)(cid:12)(cid:12) S11 (cid:69)(cid:68) S11 S11  1 (cid:18) Ωτ (cid:69) 2 where e −iHeff τ = ei(Ja+Jb−jd/2)τ cos (cid:20) (cid:19)(cid:18) jd Ω e−iHeff τ (cid:18) Ωτ (cid:19) 1 2 2jx Ω σz + σx . (cid:12)(cid:12)(cid:12) (cid:94)S11, S11 (cid:69) +i sin (cid:12)(cid:12)(cid:12) (cid:94)S02, S02 and The gate UJ thus describes an oscillation between with frequency Ω, together with z-axis rotations of the individual qubits. This evolution occurs in the double charge resonant regime illustrated in Figs. 1 and 2. To obtain a controlled-phase gate using UJ that incor- porates robustness to single-qubit exchange errors, we construct a gate sequence that includes spin-echo (re- focusing) pulses.50 For a singlet-triplet qubit, phase er- rors accumulated due to exchange fluctuations can be canceled via a π rotation about the x axis of the Bloch sphere,2,51 and simultaneous π pulses can be applied to both qubits.23 Since single-qubit x-axis rotations are gen- erated by the terms HaZ and HbZ [see Eqs. (15) and (16)], the refocusing pulses are applied in the regime ωαZ (cid:29) Jα for α = a, b. This regime can be reached by adjusting the double-dot detunings such that the 0211(cid:105) and 1102(cid:105) states are energetically closer than 0202(cid:105) to the 1111(cid:105) state, with ∆α (cid:46)(cid:12)(cid:12)(cid:12)δ (cid:12)(cid:12)(cid:12)−tα for α = a, b (here, we use a new symbol δ in order to indicate that the range (cid:12)(cid:12)(cid:12) (cid:94)S02, S02 (cid:12)(cid:12)(cid:12) , (cid:69)(cid:68) of values of ∆a + ∆b − Uab is different from that of δ in the double charge resonant regime). The effective Hamil- (cid:12)(cid:12)(cid:12) + (cid:12)(cid:12)(cid:12) S11 (cid:68) (cid:69) (cid:94)S02, S02 tonian is HZ ≡ ωZ (Xa + Xb) + δ where X ≡ T11(cid:105) (cid:104)T11 and we choose (cid:12)(cid:12)(cid:12) . We note that (cid:12)(cid:12)(cid:12) (cid:94)S02, S02 (cid:69)(cid:68) dBa = dBb for simplicity. The associated evolution is e−iHZ τ , which for τ = π/2ωZ is equal to Rπ ≡ −XaXb + e−iπδ/2ωZ applying Rπ results in the accumulation of a relative phase between the 0202(cid:105) and 1111(cid:105) charge subspaces. (cid:94)S02, S02 S11 !S02,S02!!!S11,S11!!S11,S11!!60!40!200204060!60!40!200204060∆ΜeVEΜeV!!S02,S02!!2jxΓΩ∆d=0The full sequence for the controlled-phase gate in terms of the exchange gate in the double charge reso- nant regime, UJ , and the refocusing pulse gate, Rπ, is given by PSe−iφe−iHaJ τa e−iHbJ τb Rπ UJ (τn) Rπ UJ (τn) PS =  1 (cid:12)(cid:12)(cid:12) S11, T11 (cid:69) 1 1 e2iφ  . (cid:12)(cid:12)(cid:12) (cid:94)S11, S11 (cid:69) , Here, τn = 2πn/Ω and φ = (1 − jd/Ω) nπ, where (cid:110) n is an integer, τα = φ/Jα, and PS is the projec- tor onto the four-dimensional 1111(cid:105) subspace spanned . In the by next section, we consider the controlled π-phase gate, which corresponds to φ = π/2. (cid:12)(cid:12)(cid:12)T11, S11 T11, T11(cid:105) , (cid:69)(cid:111) , IV. CHARGE NOISE AND GATE FIDELITY In practice, the performance of the controlled-phase gate in Eq. (20) is affected by charge noise.8,15,40,42,52,53 We now investigate the effects of classical, Gaussian- distributed noise in δ and ∆d due to gate voltage fluc- tuations and set δ(cid:48) = δ + ξs, ∆(cid:48) d = ∆d + ξd, where ξs and ξd are assumed to be uncorrelated and have the dis- β /√2πσβ with charge noise tributions ρβ (ξβ) = e standard deviations σβ for β = s, d. In what follows, we assume that Rπ and the single-qubit rotations in Eq. (20) are ideal in order to focus on effects due to errors in UJ , which is the gate derived from the capacitive interaction of the double dots in the double charge resonant regime. Errors due to residual magnetic gradient terms are dis- cussed briefly at the end of this section. β /2σ2 −ξ2 We therefore consider the simpler gate sequence  eiφ eiφ 1 1  , (21) Uφ ≡ UJ (τn) Rπ UJ (τn) = eiζ (cid:111) (cid:110) 2 (Ja + Jb − jd) n/Ω + 1 − δ/2ωZ π and we where ζ = have neglected a trivial global phase factor. Equation (21) represents the ideal gate sequence. We determine the gate sequence U(cid:48) φ in the presence of charge noise by expanding the terms in the Hamiltonian HJ [Eq. (17)], which are defined in Eqs. (10)-(13), up to second order in the fluctuations ξβ. For h = Ja, Jb, jd, jx, (cid:12)(cid:12)(cid:12)(cid:12)0 (cid:12)(cid:12)(cid:12)(cid:12)0 ξd (cid:48) h (cid:48) , ∆ ≡ h (δ (cid:48) d) ≈ h (δ, ∆d) + (cid:12)(cid:12)(cid:12)(cid:12)0 ∂h ∂δ(cid:48) 1 2 ξ2 s + ∂h ∂∆(cid:48) d ξ2 d (cid:12)(cid:12)(cid:12)(cid:12)0 ξs + ∂2h ∂∆(cid:48)2 d + 1 2 ∂2h ∂δ(cid:48)2 5 (cid:12)(cid:12)(cid:12)(cid:12)0 ξsξd, + ∂2h ∂δ(cid:48)∂∆(cid:48) d where we use the notation 0 ≡ δ(cid:48)=δ,∆(cid:48) of these expressions into Eq. (18) then yields U(cid:48) φ. For an initial state ψin(cid:105) , we define the minimum fi- delity as . Substitution d=∆d (20) Fmin = ρ(0) out ρout (cid:68) ∞ Tr (cid:104) ∞ = Tr −∞ −∞ (cid:105)(cid:69) (cid:104) (cid:105) ξs,ξd ρ(0) out ρout (cid:104) (cid:112) ×ρs (ξs) ρd (ξd) dξsdξd, (22) We (cid:69) (cid:69) the of T11, T11(cid:105) + corresponding charge noise. (cid:12)(cid:12)(cid:12)T11, S11 (cid:12)(cid:12)(cid:12) S11, T11 state the final sequence gate state the final pres- the ψin(cid:105) = † φ is out ≡ Uφ ψin(cid:105)(cid:104)ψin U ideal under evolution the (cid:48)† is φ ψin(cid:105)(cid:104)ψin U φ evolution where ρ(0) after and ρout ≡ U(cid:48) (cid:16) after ence 1 2 in order to maximize the error (see Appendix B) and assume that this state can be prepared without errors. Fmin is then independent of ζ, as there is initially zero probability that the system is in the state [see Eq. (21)]. We find (cid:12)(cid:12)(cid:12) (cid:94)S11, S11 (cid:69)(cid:17) (cid:12)(cid:12)(cid:12) (cid:94)S02, S02 (cid:69) (cid:19)(cid:21) in choose + + (cid:20) nπ (cid:18) j(cid:48) (cid:105) Tr ρ(0) out ρout = cos2 , (23) d Ω(cid:48) − jd Ω 2 where Ω(cid:48) = j(cid:48)2 d + 4j(cid:48)2 x . We now calculate Fmin for the controlled π-phase gate. The associated constraint φ = π/2 [see Eq. (20)] leads −1 . Since n must be an integer, to n = [2 (1 − jd/Ω)] this relation restricts the possible values of δ and ∆d for fixed values of the charge admixture parameters ηa, ηb and the capacitive coupling strength Uab. For the pa- rameter regime we consider in the present work, we find that n varies more strongly with δ than with ∆d and set ∆d = 0 for simplicity in the remainder of the anal- ysis. Choosing ηa = ηb ≡ η0 = 0.1 and σs = 2 µeV,54 we solve the constraint for the values of δ correspond- ing to n = 1, 2, . . . , 15 and calculate Fmin via numeri- cal integration using Eq. (22). By repeating this cal- culation for a range of coupling strengths Uab, we ob- tain the variation of Fmin with δ and Uab shown in Fig. 3. We see that Fmin increases with increasing coupling strength Uab as expected. For a given value of Uab, Fmin also increases as δ increases, i.e., as the energy sepa- ration between S11, S11(cid:105) and S02, S02(cid:105) becomes larger and the contribution of S02, S02(cid:105) to the ground state de- creases. For δ = 41 µeV (corresponding to n = 14) and Uab = 150 µeV, Fmin > 0.999 and the time for the gate UJ is τn = 2πn/Ω ≈ 1 ns. While we assume ideal echo pulses Rπ in the present analysis, switching times for the magnetic gradients used 6 quantum dots and calculate Γ for the higher-energy state, assuming kBT (cid:46) jx. The electron-phonon interaction for GaAs is described by the Hamiltonian55 HGaAs = 2ρ0V0cµk (kΞlδµ,l − iβ) (cid:17) † µ,−k aµ,k + a Mk, (24) while the Hamiltonian for Si has the form56 (cid:115) (cid:88) µ,k (cid:16) (cid:115) (cid:88) × HSi = i µ,k 2ρ0V0cµk (cid:48) + kz(cid:48) z · µ,k Ξu) (k · µ,k Ξd (cid:16) † µ,−k aµ,k + a (cid:17) Mk. (25) † In Eqs. (24) and (25), a µ,k creates an acoustic phonon with wave vector k, polarization µ [the sum is taken over one longitudinal mode (µ = l) and two transverse modes (µ = p)], phonon speed cµ, energy εph = cµk, and unit polarization vector µ,k, ρ0 is the mass density of the material, V0 is the crystal volume, Ξl is the defor- mation potential and β is the piezoelectric constant for GaAs, Ξd (Ξu) is the dilation (uniaxial) deformation po- tential for Si, z(cid:48) denotes the direction of uniaxial strain, and δµ,l is the Kronecker delta function. The different phonon terms appearing in Eqs. (24) and (25) reflect the fact that the crystal structure of GaAs lacks a center of symmetry, whereas unstrained Si has a centrosymmetric crystal structure: while both deformation potential and piezoelectric phonons contribute to the electron-phonon coupling in GaAs, there is no contribution from piezoelec- tric phonons for Si.55 Thus, the strength of the electron- phonon coupling and the associated relaxation rate are expected to be much smaller for Si quantum dots.5 The factor in HGaAs and HSi encompassing the cou- pling to electron charge degrees of freedom is Mk = M (a) k + M (b) k , where (cid:104)α, i eik·r α, j(cid:105) c † αiσcαjσ (26) (cid:88) (cid:88) i,j=1,2 σ M (α) k ≡ and r is the electron position operator. Note that in our calculation, we take each double dot to be cou- pled independently to the same phonon bath.39,43 Thus, we implicitly assume that the phonon mean free path is greater than the size of the system, so that scat- tering of phonons between interactions with the elec- tron pairs in the two double dots can be neglected. The matrix elements in Eq. (26) depend on the spatial configuration of the four quantum dots and are evaluated using the two-dimensional Gaussian wave functions Ψαi (r) ≡ (cid:104)rα, i(cid:105) = ψ (x − xαi) ψ (y − yαi) i = 1, 2 and α = a, b, where ψ (q) = for e−q2/4σ2 and M (b) in the basis {S11, S11(cid:105) ,S02, S02(cid:105) ,S11, S02(cid:105) ,S02, S11(cid:105)} and using the same Schrieffer-Wolff transformation used II to write Mk = eλAMke−λA ≈ for Hhub in Sec. /(cid:0)2πσ2(cid:1)1/4 . Re-expressing M (a) k k Figure 3. Minimum fidelity Fmin [Eq. (22)] of the controlled- phase gate sequence [Eq. (21)] for φ = π/2, ∆d = 0, and η0 = 0.1 as a function of the energy difference δ and the capacitive coupling strength Uab [see Fig. 1(b)]. The values of δ in the plot are calculated for each value of Uab by solving the constraint φ = π/2 with the chosen parameter values for n = 1, 2, . . . , 15 (see the main text). Note that this implies that the range of δ varies with Uab for fixed η0. The values of δ are therefore given in units of the value of Uab with which they are associated. to generate Rπ which are longer than the timescale of the qubit dynamics will lead to residual magnetic gradi- ents that remain during the action of the exchange gate UJ. The residual Zeeman energy ωZ,res will modify the Hamiltonian in Eq. (17) and thus lead to errors. Setting Ja = Jb ≡ J0 and assuming ωZ,res (cid:28) J0, we can regard the residual magnetic gradient terms as a perturbation to HJ and obtain an upper bound for the allowable residual gradient dBres using the condition τnω2 Z,res/J0 (cid:28) 1. For η0 = 0.1, δ = 41 µeV, and Uab = 150 µeV, this condition yields ωZ,res (cid:28) 1 µeV, corresponding to dBres (cid:28) 40 mT for GaAs dots (with g = −0.44) and dBres (cid:28) 10 mT for Si dots (with g = 2). V. CHARGE RELAXATION VIA PHONONS A. Relaxation rate In addition to charge noise arising from gate voltage fluctuations, charge relaxation due to electron-phonon coupling also affects the coherence of the capacitively coupled double-dot system we consider. Here, we de- termine the rate Γ of relaxation via phonons that oc- curs between the eigenstates g(cid:105) and e(cid:105) of Heff [Eq. (9)], as illustrated in Fig. 2. From Fermi's golden rule, Γ ∼ (cid:104)g Hep e(cid:105)2 ρ (Ω) , where Hep is the electron-phonon interaction Hamiltonian and ρ (Ω) is the phonon density of states at the gap energy Ω. In the following analy- sis, we consider acoustic phonons in both GaAs and Si 0.65000.80000.90000.95000.98000.99000.99600.99800.99900.99960.9998Fmin0.050.100.150.200.250.3050100150200250300∆UabUabΜeV7 Figure 4. Rate of relaxation via electron-phonon coupling for capacitively coupled GaAs double quantum dots [Eq. (24)]. The rate is calculated as a function of δ and the total interqubit capacitive coupling strength [see Fig. 1(b)], which is equal to Uab for the linear dot geometry (a) and 2Uab for the purely quadrupolar dot geometry (b) considered in the present work (see the main text). The parameter values used are ∆d = 0, η0 = 0.1, d = 140 nm, R = 2d, the GaAs effective mass m∗ = 0.067me (where me is the free-electron mass), dot size σ = 20 nm, and phonon parameter values ρ0 = 5.3 × 103 kg/m3, cl = 5.3 × 103 m/s, ct = 2.5 × 103 m/s, Ξl = 7 eV, and β = 1.4 × 109 eV/m.57 (c) Dipolar (Γdip) and quadrupolar (Γquad) contributions to the full relaxation rate (Γfull) for the linear geometry of Fig. 1(a) with Uab = 200 µeV. (d) Comparison of Γquad and Γfull for the purely quadrupolar geometry, corresponding to θ = π/2 and ϕ = 0 in Fig. 7(a), with 2Uab = 200 µeV. Mk + λ [A, Mk] + λ2 sition matrix element 2 [A, [A, Mk]] , we determine the tran- (cid:104)g Mk e(cid:105) = cos θ sin θ (cid:104)S11, S11 Mk S11, S11(cid:105) (cid:16) (cid:17) −(cid:104)S02, S02 Mk S02, S02(cid:105) + cos2 θ (cid:104)S11, S11 Mk S02, S02(cid:105) − sin2 θ (cid:104)S02, S02 Mk S11, S11(cid:105) . The relaxation rates for GaAs and Si are given by ΓGaAs = glI (Ω/cl) + gpI (Ω/cp) and ΓSi = slKl (Ω/cl) + spKp (Ω/cp) , with the momentum-space angular integrals (cid:0)1 + γ cos2 χ(cid:1)2 (cid:104)g Mk e(cid:105)2 dΩang, I (k) ≡ (cid:104)g Mk e(cid:105)2 dΩang, (28) (29) γ2 cos2 χ sin2 χ(cid:104)g Mk e(cid:105)2 dΩang Kl (k) ≡ Kp (k) ≡ (27) and the factors gl = (cid:18) Ω2 Ω 8π22ρ0c3 l 2c2 l (cid:19) , Ξ2 l + β2 gp = sµ = 2Ω 8π22ρ0c3 p β2, Ω3 8π24ρ0c5 µ Ξ2 d, µ = l, p. (30) (31) (32) In writing Eqs. (28) and (29), we have chosen one of the two transverse (µ = p) phonon polarization axes to lie orthogonal to z(cid:48) [see Eq. (25)] and defined χ as the angle between k and z(cid:48). We also define γ ≡ Ξu/Ξd, and Ωang denotes the momentum-space solid angle. We calculate the relaxation rates via numerical inte- gration for the linear geometry depicted in Fig. 1(a), which has both dipolar and quadrupolar moments, as well as for a purely quadrupolar geometry, which cor- responds to a rectangular arrangement of the dots ob- tained from the general configuration illustrated in Fig. 0.050.100.150.200.250.3050100150200250300∆2Uab2UabΜeV0.050.100.150.200.250.3050100150200250300∆UabUabΜeV(a)(b)(c))(dfulldipquad0102030405060700.010.1110100∆ΜeVGHzfullquad0102030405060700.010.1110100∆ΜeVGHz0.050.10.20.512510GHz0.00010.0010.010.050.10.20.512510GHz8 Figure 5. Rate of relaxation via electron-phonon coupling for capacitively coupled Si double quantum dots [Eq. (25)]. The rate is calculated as a function of δ and the total interqubit capacitive coupling strength [see Fig. 1(b)], which is equal to Uab for the linear dot geometry (a) and 2Uab for the purely quadrupolar dot geometry (b) considered in the present work (see the main text). The parameter values used are ∆d = 0, η0 = 0.1, d = 140 nm, R = 2d, the Si effective mass m∗ = 0.19me, dot size σ = 22 nm, and phonon parameter values ρ0 = 2.33 × 103 kg/m3, cl = 9.33 × 103 m/s, ct = 5.42 × 103 m/s, Ξd = 5 eV, and Ξu = 8.77 eV.56,58 (c) Dipolar (Γdip) and quadrupolar (Γquad) contributions to the full relaxation rate (Γfull) for the linear geometry of Fig. 1(a) with Uab = 200 µeV. (d) Comparison of Γquad and Γfull for the purely quadrupolar geometry, corresponding to θ = π/2 and ϕ = 0 in Fig. 7(a), with 2Uab = 200 µeV. 7(a) by setting θ = π/2 and ϕ = 0. For the linear case (corresponding to θ = 0, ϕ = π), we set xa1 = − (R + d) /2, xa2 = − (R − d) /2, xb1 = (R − d) /2, xb2 = (R + d) /2, and yαi = 0 for all α and i. The coordinates of the dot centers for the pure quadrupole are (xa1, ya1) = (−d/2, R/2) , (xa2, ya2) = (d/2, R/2) , (xb1, yb1) = (d/2,−R/2) , (xb2, yb2) = (−d/2,−R/2) , and we take as the interqubit Coulomb interaction term for the quadrupolar geometry Hint = Uab (na2nb1 + na1nb2) . (33) Equation (33) leads to δ = ∆a + ∆b − 2Uab and corre- sponding modifications to Eqs. (10)-(13) for the case of the purely quadrupolar system. Note that the total cou- pling strength between the qubits for the quadrupolar geometry is effectively twice that for the linear geome- try. We therefore vary 2Uab for the quadrupolar system over the same range of values of Uab considered for the linear configuration, in order to focus on the geometry- dependent variation in the relaxation rate. The calculated relaxation rates are shown for GaAs in Fig. 4 and for Si in Fig. 5 as a function of δ and the interqubit capacitive coupling strength, where we choose ∆d = 0, η0 = 0.1, d = 140 nm, and R = 2d. Comparing Figs. 4(a) for the linear geometry and 4(b) for the purely quadrupolar geometry, we see that both relaxation rates increase with increasing Uab but exhibit a nonmonotonic dependence on δ. While the largest rates shown for both geometries are ∼ 10 GHz, the rate for the quadrupolar geometry reduces to (cid:46) 100 kHz for the smallest values of δ and Uab considered. On the other hand, we see from Fig. 4(a) that the rate reduces only to ∼ 10 MHz for the linear geometry. Figures 5(a) and 5(b) reveal that the relaxation rates for Si dots are several orders of magni- tude smaller than those for GaAs dots, as expected due to the absence of piezoelectric phonons in Si.5 The rates increase as both Uab and δ are increased, with a maxi- mum rate ∼ 1 MHz for the parameter ranges considered. At the smallest values of δ and Uab shown, relaxation for the linear geometry has a rate ∼ 1 Hz [Fig. 5(a)], while the rate for the purely quadrupolar geometry [Fig. 5(b)] is two orders of magnitude smaller. We now consider separately the contributions of the 0.050.100.150.200.250.3050100150200250300∆2Uab2UabΜeV0.050.100.150.200.250.3050100150200250300∆UabUabΜeV(a)(b)(c))(dfulldipquad0102030405060701040.011100104∆ΜeVkHzfullquad0102030405060701040.011100104∆ΜeVkHz0.0010.010.11101001000kHz0.000010.00010.0010.010.11101001000kHzdipolar and quadrupolar terms in Mk to the total re- laxation rates for coupled GaAs and Si double dots in both the linear and the purely quadrupolar geome- tries. For phonon wavelengths long compared to the size of the quantum dot system, we can write eik·r ≈ 1 + ik · r − (k · r)2 /2. The dipolar (Γdip) and quadrupo- lar (Γquad) contributions to the rate are then obtained by calculating the relaxation rates with the transition matrix elements (cid:104)g ik· re(cid:105) and (cid:104)g (k · r)2 /2e(cid:105) , respec- tively, substituted for the full matrix element (cid:104)g Mk e(cid:105) in Eqs. (27)-(29). We see in Figs. 4(c) and 5(c) that, for both GaAs and Si, the full relaxation rate Γfull for the linear geometry contains a large dipolar contribution and a much smaller quadrupolar contribution. The large dipolar term can be understood from the fact that a net dipole moment exists for the four-electron system in the linear configuration. In contrast, the purely quadrupo- lar geometry [Figs. 4(d) and 5(d)] lacks a net dipole moment, so that Γdip = 0 in this case. While a large discrepancy exists between the quadrupolar contribution Γquad and Γfull for GaAs, Γfull for Si is well described by the quadrupolar term. This can be understood from the fact that, over the range of δ (and thus Ω) we con- sider, the ratio of the system size (∼ R) to the phonon wavelength is less than 1 for Si. On the other hand, the corresponding ratio for GaAs becomes larger than 1 at sufficiently large values of δ. B. Modification of controlled-Z gate fidelity , (cid:69) (cid:69)(cid:111) (cid:12)(cid:12)(cid:12) (cid:94)S02, S02 (cid:110)(cid:12)(cid:12)(cid:12) (cid:94)S11, S11 Having calculated the rate of phonon-induced charge relaxation within the two-singlet subspace spanned by , we now determine the effect of this decay on the gate fidelity calculated in Sec. IV for the linear quantum dot geometry. In order to incorpo- rate the relaxation into the dynamics, we consider the Lindblad master equation for the density matrix within the two-singlet subspace ρs, which can be written in the form (cid:69)(cid:68) ρs = −i (cid:12)(cid:12)(cid:12) (cid:94)S11, S11 (cid:94)S02, S02 with a ≡ H ≡ Heff − i † a Γ 2 a = Heff − i (cid:104) H, ρs † + Γaρsa (cid:105) (cid:12)(cid:12)(cid:12) and (cid:12)(cid:12)(cid:12) (cid:94)S02, S02 Γ 2 (34) (cid:69)(cid:68) (cid:94)S02, S02 (cid:12)(cid:12)(cid:12) . (35) We assume Γ, jx (cid:28) jd and neglect the final (quantum- jump) term in Eq. (34). Within this approximation, we can regard the dynamics in the two-singlet subspace [Eq. (19)] as being generated by the non-Hermitian "Hamilto- nian" in Eq. (35) instead of Heff . We can then estimate the effect of the phonon decay 9 on the dynamics by making the replacements ei[(Ja+Jb)τn+φ](cid:12)(cid:12)(cid:12) (cid:94)S11, S11 (cid:12)(cid:12)(cid:12) −Γeff τn ei[(Ja+Jb)τn+φ](cid:12)(cid:12)(cid:12) (cid:94)S11, S11 (cid:69)(cid:68) ei[(Ja+Jb)τn+φ](cid:12)(cid:12)(cid:12) (cid:94)S02, S02 (cid:12)(cid:12)(cid:12) −Γ2τn ei[(Ja+Jb)τn+φ](cid:12)(cid:12)(cid:12) (cid:94)S02, S02 (cid:69)(cid:68) (cid:69)(cid:68) (cid:69)(cid:68) (cid:94)S02, S02 (cid:94)S11, S11 → e → e (cid:12)(cid:12)(cid:12) , (cid:12)(cid:12)(cid:12) (cid:94)S11, S11 (cid:94)S02, S02 x/2j2 d and the in UJ (τn) [see Eq. (18)]. Here, Γeff ≡ Γj2 explicit form of Γ2 does not enter into the calculation for our choice of initial state ψin(cid:105) , which is defined en- tirely within the effective 1111(cid:105) subspace. Incorporating these modifications into the gate sequence in Eq. (21), we determine the resulting modified minimum gate fidelity F (cid:48) min using Eq. (22). The results are shown in Fig. 6(a) for GaAs dots and in Fig. 6(b) for Si dots. Comparing these plots with Fig. 3, we see that the phonon-induced decay results in a large reduction of the gate fidelity for GaAs, as ex- pected from the fact that ΓGaAs ∼ 100 MHz − 10 GHz is comparable to 1/τn. In contrast, essentially no mod- ification to the fidelity occurs for the case of Si, since ΓSi (cid:46) 1 MHz (cid:28) 1/τn. Thus, we find that an implemen- tation of the controlled-Z gate based on Si quantum dots provides robustness to phonon-induced decay. While the analysis so far has assumed quasistatic charge noise, fast charge noise also affects the coherence of singlet-triplet spin qubits in practice.42 Accordingly, we also calculate the modified minimum fidelity F (cid:48)(cid:48) min of the controlled-phase gate in the presence of decay with an effective rate Γ(cid:48) eff ≡ Γeff + Γchg, where Γchg is the fast charge noise frequency. Note that, for realistic devices, Γchg may vary with the particular operating point via the dependence of the detuning noise spectral density on this point.42 Here, we assume that Γchg is independent of the operating point for simplicity. The results for GaAs (Si) dots are shown for Γchg = 1 MHz (representing the high-frequency limit of the noise spectra analyzed in Ref. 42) in Fig. 6(c) [Fig. 6(d)] and for Γchg = 1 GHz in Fig. 6(e) [Fig. 6(f)]. We find that, while the gate fidelity for GaAs dots is further degraded in the presence of fast charge noise, fidelities of up to F (cid:48)(cid:48) min ∼ 0.999 are in prin- ciple possible for Si dots even in the presence of 1 MHz charge noise. Charge noise of frequency Γchg = 1 GHz results in a significant reduction of the gate fidelity for Si dots, which becomes similar to that for GaAs dots with Γchg = 1 GHz [Fig. 6(e)] as expected from the fact that ΓSi (cid:28) 1 GHz [see Fig. 5(a)]. VI. DEPENDENCE OF CAPACITIVE COUPLING ON DOT GEOMETRY Finally, we consider how the capacitive coupling strength varies with the relative orientation of the double dots.53 Specifically, we consider the geometry shown in 10 Figure 6. (a),(b) Modified minimum fidelity F (cid:48) min of the controlled-phase gate sequence [Eq. (21)] for φ = π/2, in the presence of decay due to electron-phonon coupling in (a) GaAs and (b) Si quantum dots arranged in the linear geometry of Fig. 1(a). The parameters used in the calculation are identical to those given in the captions of Figs. 3,4, and 5. (c)-(f) Modified minimum fidelity F (cid:48)(cid:48) min in the presence of both phonon-induced decay and charge noise of frequency Γchg, calculated for (c) GaAs dots with Γchg = 1 MHz, (d) Si dots with Γchg = 1 MHz, (e) GaAs dots with Γchg = 1 GHz, and (f) Si dots with Γchg = 1 GHz. Fig. 7(a) and the general form of the capacitive interac- tion term in the Hamiltonian, given by center positions Ri = (xi, yi) and Rj = (xj, yj) , respec- tively, is55 (cid:88) i(cid:54)=j HC = 1 2 Uijninj. (36) In Eq. (36), we have for notational convenience re- defined the dot indices a1, a2, b1, and b2 as 1, 2, 3, and 4, respectively. The matrix element of the Coulomb in- teraction between the electrons in dot i and dot j with Uij ≡ (cid:104)ij ≡ 1 r ij(cid:105) Ψi (r)2 Ψj (r(cid:48))2 r − r(cid:48) drdr(cid:48) , (37) where Ψi (r) ≡ (cid:104)ri(cid:105) = ψ (x − xi) ψ (y − yi) and ψ is the one-dimensional Gaussian function defined in Sec. V. We assume R (cid:29) d [see Fig. 7(a)] and estimate Uij by 0.6500.8000.9000.9500.9800.9900.9960.9980.999Fmin0.3000.3500.4000.4500.5000.5500.600Fmin0.6000.7000.8000.8500.9000.9500.9700.980Fmin0.6000.7000.8000.8500.9000.9500.9700.9800.985Fmin0.050.100.150.200.250.3050100150200250300∆UabUabΜeV0.050.100.150.200.250.3050100150200250300∆UabUabΜeV(a)(b)0.65000.80000.90000.95000.98000.99000.99600.99800.99900.99960.9998Fmin(c))(d0.050.100.150.200.250.3050100150200250300∆UabUabΜeV0.050.100.150.200.250.3050100150200250300∆UabUabΜeV(e))(f0.050.100.150.200.250.3050100150200250300∆UabUabΜeV0.050.100.150.200.250.3050100150200250300∆UabUabΜeV0.3000.3500.4000.4500.5000.5500.6000.6500.700Fminthe leading order term in the multipole expansion of the Coulomb interaction as Uij ∼ 1 Ri − Rj . (38) Defining E (n1 n2 n3 n4) as the energy of the charge state n1 n2 n3 n4(cid:105) , the Coulomb energy that sets the speed of the controlled-Z gate in the double charge reso- nant regime is given by 11 U0202 ≡ E (0202) − E (1111) − [E (0211) − E (1111)] − [E (1102) − E (1111)] = E (0202) + E (1111) − E (0211) − E (1102) = U13 + U24 − U14 − U23. (39) We note that Eq. (39) includes the term U23 = Uab, which is the dominant term in U0202 for the parameter regime we consider in the present work. U0202 depends on the parameters R, d, θ, and ϕ through Eq. (38). The dependence of U0202 on the relative orientation of the two double dots, determined by θ and ϕ [see Fig. 7(a)], is shown in Fig. 7(b) for fixed R/d. From this dependence, we see that the linear geometry (θ = 0, ϕ = π) is associated with a minimum energy, corresponding to an attractive dipole-dipole interaction of maximum strength, and therefore provides the fastest gate. On the other hand, the case θ = 0, ϕ = 0 corresponds to a maximum repulsive interaction strength. Note that Eq. 38 is approximately independent of the dot size σ. Thus, to leading order, the interqubit capaci- tive coupling strength for R (cid:29) d is largely insensitive to variations in the sizes of the dots and depends primarily on the dot center positions. While the sensitivity of the intraqubit tunneling amplitudes to dot size differences may modify the charge admixture ηα and thus lead to changes in the gate speed, this sensitivity will not qualita- tively affect the approach discussed in the present work. In addition, knowledge of the dot size variation should in principle enable tuning of the gate voltages control- ling the double dot potentials in order to compensate for changes in the intraqubit charge admixture and thereby optimize the fidelity. VII. CONCLUSIONS In the present work, we have investigated capacitively coupled singlet-triplet qubits in a pair of adjacent dou- ble quantum dots in the double charge resonant regime, where the interqubit Coulomb interaction leads to near- degeneracy between the 1111(cid:105) and 0202(cid:105) charge states. This regime is different from that considered in Ref. 2 and subsequent work, where the two-qubit coupling relies on a repulsive dipole-dipole interaction. Using the dynamics generated within the two-singlet subspace by the capacitive coupling, we derived a sequence for a controlled-phase gate that includes spin echo pulses Figure 7. (a) Illustration of a pair of double dots having interdot distance d, separated by distance R, and with relative in-plane orientation determined by the angles θ and ϕ. (b) Variation of the capacitive coupling U0202 [Eq. (39)] with θ and ϕ for R/d = 3. to correct for single-qubit dephasing. For this gate se- quence, we showed that rapid gates with fidelities greater than 0.9998 in the presence of classical, static charge noise are in principle achievable by adjusting the indi- vidual qubit detunings to appropriate values. We also studied the relaxation of coupled singlet-triplet qubits via electron-phonon interaction for quantum dots in both GaAs and Si. The full relaxation rates, as well as their dipolar and quadrupolar contributions, were calculated for both linear and purely quadrupolar dot geometries. For the linear dot geometry, we showed that the presence of phonon-induced decay results in a large decrease in the gate fidelity for GaAs dots but does not significantly af- fect the fidelity in the case of Si dots due to much slower charge relaxation. In addition, we found that fidelities greater than 0.999 are in principle possible for Si dots even in the presence of 1 MHz charge noise. Finally, we showed that the linear geometry gives rise to the fastest two-qubit gate. These results demonstrate that the intraqubit detun- ings, interqubit interaction strengths, and geometry of a capacitively coupled pair of double dots can be chosen in order to optimize the controlled-Z gate fidelity. Im- plementations of this gate in the double charge resonant 0.00.20.40.60.81.00.00.51.01.52.0ΘΠΠ0.080.060.040.020.000.020.040.060.08U0202a.u.(a)(b)Rϕddθregime using Si dots arranged in a linear geometry should lead to high fidelities in the presence of both quasistatic and fast charge noise as well as relaxation via phonons. Improvements to the results of the present work might be found by considering the double charge resonant regime for, e.g., multi-electron singlet-triplet qubits,59–62 which are expected to have enhanced robustness to charge noise due to screening of the Coulomb interaction by the addi- tional electrons in the dots. Finally, we note that mea- sured relations for the exchange coupling as a function of detuning in double dots20,42 deviate from the detun- ing dependence in Eqs. (10) and (11) derived from the Hubbard model and thus may lead to different optimal operating points for the controlled-phase gate. Potential future directions therefore also include exploring exten- sions to the Hubbard model as well as more sophisticated charge noise models34,36 in order to obtain a more accu- rate description of capacitively coupled double dots in the double charge resonant regime. (cid:12)(cid:12)(cid:12) (cid:94)S11, S02 (cid:69) (cid:34) ≈ 1 − η2 a (cid:34) (cid:12)(cid:12)(cid:12) (cid:94)S02, S11 (cid:69) ≈ 1 − η2 12 (cid:35) S11, S02(cid:105) (A3) S02, S11(cid:105) −2ηaηb √2ηb S11, S11(cid:105) − √2ηa − b (Uab + δ + ∆d)2 (cid:1) (cid:0)U 2 (Uab − δ − ∆d)2 − η2 ab + δ2 − ∆2 (Uab − δ)2 − ∆2 d d S02, S11(cid:105) (Uab + δ + ∆d) (Uab − δ − ∆d) S02, S02(cid:105) , (cid:35) (cid:0)U 2 (Uab + δ − ∆d)2 (Uab − δ + ∆d)2 (cid:1) ab + δ2 − ∆2 (Uab − δ)2 − ∆2 S11, S02(cid:105) a − η2 d d b −2ηaηb √2ηa S11, S11(cid:105) − √2ηb − (Uab + δ − ∆d) (Uab − δ + ∆d) S02, S02(cid:105) . (A4) ACKNOWLEDGMENTS Appendix B: Minimum fidelity We acknowledge useful discussions with A. Yacoby, S. Das Sarma, B. Halperin, A. Pal, and S. Yang. We also thank M. Maghrebi and G. Solomon for helpful com- ments. This work was supported by DARPA MTO and the NSF-funded Physics Frontier Center at the JQI. Appendix A: Basis states obtained via Schrieffer-Wolff transformation Here, we give expressions for the corrected states re- sulting from the Schrieffer-Wolff transformation used to obtain the effective Hamiltonian Heff [Eq. (9)]. Up to second order in the charge admixture parameters ηa and ηb defined in Sec. II, we find a − η2 (cid:12)(cid:12)(cid:12) (cid:94)S11, S11 S11, S11(cid:105) (cid:69) (cid:1) ≈ b (cid:0)1 − η2 (cid:0)U 2 (cid:1) ab − δ2 + ∆2 (Uab − δ)2 − ∆2 d d (cid:12)(cid:12)(cid:12) (cid:94)S02, S02 (cid:69) ≈ 1 − η2 a S02, S02(cid:105) − 2ηaηb + √2ηb S11, S02(cid:105) + √2ηa S02, S11(cid:105) ,(A1) (cid:34) (cid:0)U 2 S02, S02(cid:105) (Uab + δ + ∆d)2 (cid:35) (Uab − δ − ∆d)2 (Uab + δ − ∆d)2 (cid:1) (Uab − δ + ∆d)2 ab − δ2 + ∆2 (Uab − δ)2 − ∆2 (Uab + δ + ∆d) (Uab − δ − ∆d) S11, S02(cid:105) (Uab + δ − ∆d) (Uab − δ + ∆d) S02, S11(cid:105) , (A2) S11, S11(cid:105) d d −η2 b −2ηaηb +√2ηa +√2ηb The expression in Eq. (23) for the minimum fidelity of the controlled π-phase (or controlled-Z) gate is obtained using the particular state ψin(cid:105) chosen for the analysis in the present work. Here, we show that this initial state represents only one possible element of a more general class of states that minimize the gate fidelity (and thus maximize the error) for a given charge noise distribution. (cid:48)† We write ρ(0) φ ψ(cid:105)(cid:104)ψ U φ for an arbitrary initial state ψ(cid:105) ≡ cT T T11, T11(cid:105) + , where cST cT T , cST , cT S, and cSS are complex coefficients. The gate fidelity then becomes [see Eq. (22)] † (cid:69) φ and ρout ≡ U(cid:48) (cid:12)(cid:12)(cid:12) (cid:94)S11, S11 out ≡ Uφ ψ(cid:105)(cid:104)ψ U + cT S + cSS (cid:69) (cid:69) (cid:12)(cid:12)(cid:12) S11, T11 (cid:104) (cid:12)(cid:12)(cid:12)(cid:104)ψ U f ≡ Tr = (cid:12)(cid:12)(cid:12)T11, S11 (cid:105) (cid:12)(cid:12)(cid:12)2 ρ(0) out ρout † φU (cid:48) φ ψ(cid:105) (cid:20) nπ (cid:18) j(cid:48) d Ω(cid:48) − (cid:19)(cid:21) jd Ω , (B1) = A2 + B2 + 2AB cos with A ≡ cT T2 +cSS2 and B ≡ cT S2 +cST2 . Noting that the minimum value of the cosine function is -1, we then find f min = A2 + B2 − 2AB = (A − B)2 . This has a minimum value of zero for A = B. Together with the normalization condition A + B = 1 for ψ(cid:105) , this yields A = B = 1/2, so that (cid:19)(cid:21) (cid:18) j(cid:48) (cid:20) (cid:18) j(cid:48) nπ d Ω(cid:48) − jd Ω d (cid:19)(cid:21) Ω(cid:48) − jd Ω f min = 1 2 + = cos2 cos 1 2 (cid:20) nπ 2 , (B2) which minimum fidelity agrees with the in Eq. expression (23) for the determined (cid:16) specific (cid:12)(cid:12)(cid:12) S11, T11 (cid:69) (cid:12)(cid:12)(cid:12)T11, S11 (cid:69) the T11, T11(cid:105) + using 1 2 Note that for this state, cT T = cSS = cT S = cST = 1/2, which satisfies A = B = 1/2. Thus, ψin(cid:105) represents a input + state + ψin(cid:105) = . (cid:12)(cid:12)(cid:12) (cid:94)S11, S11 (cid:69)(cid:17) particular initial state that minimizes the gate fidelity. All such states lead to the same expression for fmin [Eq. (B2)]. 13 ∗ [email protected] 1 D. Loss and D. P. DiVincenzo, Phys. Rev. A 57, 120 (1998). 2 J. M. Taylor, H. A. Engel, W. Dur, A. Yacoby, C. M. Marcus, P. Zoller, and M. D. Lukin, Nature Phys. 1, 177 (2005). 3 R. Hanson, L. P. Kouwenhoven, J. R. Petta, S. Tarucha, and L. M. K. Vandersypen, Rev. Mod. Phys. 79, 1217 (2007). 4 C. Kloeffel and D. Loss, Annual Review of Condensed Mat- ter Physics 4, 51 (2013). 5 F. A. Zwanenburg, A. S. Dzurak, A. Morello, M. Y. Sim- mons, L. C. L. Hollenberg, G. Klimeck, S. Rogge, S. N. Coppersmith, and M. A. Eriksson, Rev. Mod. Phys. 85, 961 (2013). 6 G. Burkard, D. Loss, and D. P. DiVincenzo, Phys. Rev. B 59, 2070 (1999). 7 J. Levy, Phys. Rev. Lett. 89, 147902 (2002). 8 J. M. Taylor, J. R. Petta, A. C. Johnson, A. Yacoby, C. M. Marcus, and M. D. Lukin, Phys. Rev. B 76, 035315 (2007). 9 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). 10 M. Pioro-Ladriere, T. Obata, Y. Tokura, Y. S. Shin, T. Kubo, K. Yoshida, T. Taniyama, and S. Tarucha, Na- ture Phys. 4, 776 (2008). 11 S. Foletti, H. Bluhm, D. Mahalu, V. Umansky, and A. Ya- coby, Nature Phys. 5, 903 (2009). 12 A. C. Johnson, J. R. Petta, J. M. Taylor, A. Yacoby, M. D. Lukin, C. M. Marcus, M. P. Hanson, and A. C. Gossard, Nature 435, 925 (2005). 13 J. M. Taylor, W. Dür, P. Zoller, A. Yacoby, C. M. Marcus, and M. D. Lukin, Phys. Rev. Lett. 94, 236803 (2005). 14 F. H. L. Koppens, J. A. Folk, J. M. Elzerman, R. Han- son, L. H. W. van Beveren, I. T. Vink, H. P. Tranitz, W. Wegscheider, L. P. Kouwenhoven, and L. M. K. Van- dersypen, Science 309, 1346 (2005). 15 W. A. Coish and D. Loss, Phys. Rev. B 72, 125337 (2005). 16 E. A. Laird, J. R. Petta, A. C. Johnson, C. M. Marcus, A. Yacoby, M. P. Hanson, and A. C. Gossard, Phys. Rev. Lett. 97, 056801 (2006). 17 H. Bluhm, S. Foletti, I. Neder, M. Rudner, D. Mahalu, V. Umansky, and A. Yacoby, Nature Phys. 7, 109 (2011). 18 X. Wang, L. S. Bishop, J. P. Kestner, E. Barnes, K. Sun, and S. Das Sarma, Nat. Commun. 3, 997 (2012). 19 C. Barthel, J. Medford, C. M. Marcus, M. P. Hanson, and A. C. Gossard, Phys. Rev. Lett. 105, 266808 (2010). 20 B. M. Maune, M. G. Borselli, B. Huang, T. D. Ladd, P. W. Deelman, K. S. Holabird, A. A. Kiselev, I. Alvarado- Rodriguez, R. S. Ross, A. E. Schmitz, M. Sokolich, C. A. Watson, M. F. Gyure, and A. T. Hunter, Nature 481, 344 (2012). 21 X. Wu, D. R. Ward, J. R. Prance, D. Kim, J. K. Gam- ble, R. T. Mohr, Z. Shi, D. E. Savage, M. G. Lagally, M. Friesen, S. N. Coppersmith, and M. A. Eriksson, Proc. (2012). (2014). (2007). (2007). (2014). (2006). (2006). (2006). (2002). Nat. Acad. Sci. USA 111, 11938 (2014). 22 I. van Weperen, B. D. Armstrong, E. A. Laird, J. Medford, C. M. Marcus, M. P. Hanson, and A. C. Gossard, Phys. Rev. Lett. 107, 030506 (2011). 23 M. D. Shulman, O. E. Dial, S. P. Harvey, H. Bluhm, V. Umansky, and A. Yacoby, Science 336, 202 (2012). 24 J. Klinovaja, D. Stepanenko, B. I. Halperin, and D. Loss, Phys. Rev. B 86, 085423 (2012). 25 R. Li, X. Hu, and J. Q. You, Phys. Rev. B 86, 205306 26 J. P. Kestner, X. Wang, L. S. Bishop, E. Barnes, and S. Das Sarma, Phys. Rev. Lett. 110, 140502 (2013). 27 M. P. Wardrop and A. C. Doherty, Phys. Rev. B 90, 045418 28 S. Mehl, H. Bluhm, and D. P. DiVincenzo, Phys. Rev. B 90, 045404 (2014). 29 W. G. van der Wiel, S. De Franceschi, J. M. Elzerman, T. Fujisawa, S. Tarucha, and L. P. Kouwenhoven, Rev. Mod. Phys. 75, 1 (2002). 30 R. Hanson and G. Burkard, Phys. Rev. Lett. 98, 050502 31 D. Stepanenko and G. Burkard, Phys. Rev. B 75, 085324 32 G. Ramon, Phys. Rev. B 84, 155329 (2011). 33 L. Trifunovic, O. Dial, M. Trif, J. R. Wootton, R. Abebe, A. Yacoby, and D. Loss, Phys. Rev. X 2, 011006 (2012). 34 E. Nielsen, R. P. Muller, and M. S. Carroll, Phys. Rev. B 85, 035319 (2012). 35 X. Wang, E. Barnes, and S. Das Sarma, arXiv:1412.7756v2 36 F. A. Calderon-Vargas and J. P. Kestner, Phys. Rev. B 91, 035301 (2015). 37 G. Burkard and A. Imamoglu, Phys. Rev. B 74, 041307 38 J. M. Taylor and M. D. Lukin, arXiv:cond-mat/0605144 39 S. D. Barrett and C. H. W. Barnes, Phys. Rev. B 66, 125318 (2002). 40 X. Hu and S. Das Sarma, Phys. Rev. Lett. 96, 100501 41 D. Culcer, X. Hu, and S. Das Sarma, Applied Physics Letters 95, 073102 (2009). 42 O. E. Dial, M. D. Shulman, S. P. Harvey, H. Bluhm, V. Umansky, and A. Yacoby, Phys. Rev. Lett. 110, 146804 (2013). 43 T. Brandes and T. Vorrath, Phys. Rev. B 66, 075341 44 T. Meunier, I. T. Vink, L. H. Willems van Beveren, K.- J. Tielrooij, R. Hanson, F. H. L. Koppens, H. P. Tranitz, W. Wegscheider, L. P. Kouwenhoven, and L. M. K. Van- dersypen, Phys. Rev. Lett. 98, 126601 (2007). 45 C. Barthel, J. Medford, H. Bluhm, A. Yacoby, C. M. Mar- cus, M. P. Hanson, and A. C. Gossard, Phys. Rev. B 85, 035306 (2012). 46 M. Raith, P. Stano, and J. Fabian, Phys. Rev. B 86, 54 K. D. Petersson, J. R. Petta, H. Lu, and A. C. Gossard, 205321 (2012). 47 J. Danon, Phys. Rev. B 88, 075306 (2013). 48 F. R. Braakman, J. Danon, L. R. Schreiber, W. Wegschei- der, and L. M. K. Vandersypen, Phys. Rev. B 89, 075417 (2014). 49 S. Yang, X. Wang, and S. Das Sarma, Phys. Rev. B 83, 50 L. M. K. Vandersypen and I. L. Chuang, Rev. Mod. Phys. 51 L.-A. Wu and D. A. Lidar, Phys. Rev. Lett. 88, 207902 161301 (2011). 76, 1037 (2005). (2002). 52 T. Meunier, V. E. Calado, and L. M. K. Vandersypen, Phys. Rev. B 83, 121403 (2011). 53 S. Yang and S. Das Sarma, Phys. Rev. B 84, 121306 (2011). 14 Phys. Rev. Lett. 105, 246804 (2010). 55 G. Mahan, Many-Particle Physics (Plenum, 1990). 56 P. Yu and M. Cardona, Fundamentals of Semiconduc- tors: Physics and Materials Properties, Graduate Texts in Physics (Springer, 2010). 57 P. Stano and J. Fabian, Phys. Rev. Lett. 96, 186602 (2006). 58 C. Tahan and R. Joynt, Phys. Rev. B 89, 075302 (2014). 59 S. Vorojtsov, E. R. Mucciolo, and H. U. Baranger, Phys. Rev. B 69, 115329 (2004). 60 E. Barnes, J. P. Kestner, N. T. T. Nguyen, S. Das Sarma, Phys. Rev. B 84, 235309 (2011). 61 A. P. Higginbotham, F. Kuemmeth, M. P. Hanson, A. C. Gossard, and C. M. Marcus, Phys. Rev. Lett. 112, 026801 (2014). 62 S. Mehl and D. P. DiVincenzo, Phys. Rev. B 88, 161408 and (2013).
1805.05894	1	1805	2018-05-15T16:33:50	Comments on the Fractal Energy Spectrum of Honeycomb Lattice with Defects	[ "cond-mat.mes-hall" ]	We address the energy spectrum of honeycomb lattice with various defects or impurities under a perpendicular magnetic field. We use a tight-binding Hamiltonian including interactions with the nearest neighbors and investigate its energy structure for two different choices of point defects or impurities. In the first case, we fix a unit cell consisting of 8 lattice points and survey the energy eigenvalues in the presence of up to 2 point defects. Then it turns out that the existence of the fractal energy structure, called Hofstadter's butterfly, depends on the choice of defect pairs. In the second case, we extend the size of a unit cell which contains a single point defect and up to 32 lattice points. The fractal structures indeed appear for those cases and there exist a robust gapless point in the $E=0$ eV line without depending on both the size of unit cells and the shape of lattices. Therefore we keep an immortal butterfly since such a robust point corresponds to the center of a butterfly. Consequently we predict that the presence of a butterfly in a graph is equivalent to that of fractality.	cond-mat.mes-hall	cond-mat	Comments on the Fractal Energy Spectrum of Honeycomb Lattice with Defects Yoshiyuki Matsuki∗ and Kazuki Ikeda† Department of Physics, Osaka University, Toyonaka, Osaka 560-0043, Japan (Dated: May 16, 2018) We address the energy spectrum of honeycomb lattice with various defects or impurities under a perpendicular magnetic field. We use a tight-binding Hamiltonian including interactions with the nearest neighbors and investigate its energy structure for two different choices of point defects or impurities. In the first case, we fix a unit cell consisting of 8 lattice points and survey the energy eigenvalues in the presence of up to 2 point defects. Then it turns out that the existence of the fractal energy structure, called Hofstadter's butterfly, depends on the choice of defect pairs. In the second case, we extend the size of a unit cell which contains a single point defect and up to 32 lattice points. The fractal structures indeed appear for those cases and there exist a robust gapless point in the E = 0 eV line without depending on both the size of unit cells and the shape of lattices. Therefore we keep an immortal butterfly since such a robust point corresponds to the center of a butterfly. Consequently we predict that the presence of a butterfly in a graph is equivalent to that of fractality. I. INTRODUCTION It has been widely known that the graph of the spectrum over a wide range of rational mag- netic fields passing through a two-dimensional lattice cell shows fractal, called the Hofstadter butterfly [1], which is confirmed by various ex- perimental studies [2–4]. This intriguing be- havior of Bloch electrons have long attracted the major interest of researchers from vari- ous perspectives. In a two-dimensional system, the Hofstadter problem is addressed for vari- ous choice of lattices: general Bravais lattice [5], buckled graphene like materials [6] and square lattice with next-nearest-neighbor hopping [7]. More recently Hofstadter's butterfly appears in relation to the quantum geometry [8, 9] and it is also surveyed from a perspective of mathe- matical physics [10, 11]. This problem is ex- tended to higher dimensional cases [12–14] and it is known that analog of Hofstadters butterfly exists in more general systems. All of the above theoretical studies are based on the perfect lattice structures, therefore a sim- ple question "Does the fractal energy structure ∗ [email protected] † [email protected] appear even in systems with defects?" comes to our mind. As well as investigating the origin of the fractal nature, we aim at giving a posi- tive answers to this question. Apparently it is a formidable task to find such a fractal energy spectrum in a system with defects since they usually make energy bands gapped. Indeed, as we report in this article the energy spectrum highly depends on a choice of defect pairs and analogue of the Hofstadter butterfly exists only under appropriate conditions. It is briefly re- ported that the honeycomb lattice with a single point defect in a unit cell accommodates the fractal spectrum with large band gaps nearby the E = 0 eV line (Fig.3) and point defects are responsible for modifying the Hofstadter but- terfly structure [15]. As a succeeding study, we investigate more on roles of defects or im- purities in order to enhance our knowledge on its fractal natures. Though it is obvious that the fractal structure cannot be obtained if too many defects are introduced into a unit cell, as a non-trivial result, we find that the large band gaps reported in the previous paper [15] are closed and the graph of the energy spec- trum looks like symmetric rather than fractal when another atom is vacant so that the honey- comb lattice accommodates line defects (Fig.4). In addition, even for several enlarged unit cells, 8 1 0 2 y a M 5 1 ] l l a h - s e m . t a m - d n o c [ 1 v 4 9 8 5 0 . 5 0 8 1 : v i X r a we also observe such a robust gapless point in the E = 0 eV level that exists without depend- ing on a choice of the unit cell size. Moreover, it can be found at the center of the butterfly shape (φ = 0.5 in Fig.1) and it is crucial to the fractal nature of the graph. This article is orchestrated as follows. In Sec- tion II we explain our problem setting. We use the tight-binding Hamiltonian including the nearest neighbor interactions of electrons in en- larged unit cells with point defects or impuri- ties. In Section III, we present a number of graphs obtained by numerical calculation for various choice of defects and for several enlarged unit cells. We also give several theoretical rea- sons for the obtained family of energy structures and consider the origin of the fractal nature from a purely linear algebraic perspective. This article is concluded in Section IV with some pro- posed future works. 2 Ri(cid:105)(cid:104)Rj , (1) substitution[16] Ri(cid:105)(cid:104)Rj → eiΘmag ji (cid:82) Ri ji = − e where Θmag gauge (0, Bx, 0), it corresponds to Rj A · dl In the Landau ji = −πφ(Rj + Ri) · x ((Ri − Rj) · y) ,(2) Θmag where Ri is a position vector specifying the atom, Ri = mαa1 + nαa2 with labels α of the atoms in the enlarged unit cell, φ = BS/φ0, φ0 = h/e and S is the area of the enlarged unit cell. Under this setup, our system can be written by means of a new matrix equation which can be called the generalized Harper equation[17] II. MODEL AND FORMULATION EΨm = UmΨm + VmΨm+1 + WmΨm−1.(3) We introduce the magnetic field into the lattice system by the Peierls 2 dimensional Using matrix representation, we rewrite it as 0 U1 V1 W2 U2 V2 0 . . . 0 0 W3 U3 V3 ... . . . . . . V ∗ 0 . . . 0 W ∗ 1 0 0 . . . 0 0 . . . ... . . . 0 Wq Uq . . .    , Ψ1 Ψ2 ... Ψq−1 Ψq E Ψ1 Ψ2 ... Ψq−1 Ψq    = q = Vqe−ikxa1xq (cid:0) − π q ,−π ≤ ky ≤ π(cid:1). Note that the index m where Um, Vm, Wm are certain matrixes and q ≤ W ∗ kx ≤ π is periodic in q with a period q therefore wave functions are written 1 = W1e−ikxa1xq, V ∗ ... 0 q ψm+q = eikxa1xqψm (5) by Bloch's theorem. For example, if the unit cell (a) in Fig.2 is preferred, then Ψm has the (4) (6)  form  Ψm = ψA m ψB m ψC m ψD m ψE m ψF m ψG m ψH m and Um, Vm, Wm are 8× 8 matrices. Then solv- ing the characteristic equation det(E − H) = 0 with respect to each φ, where H is defined by the left hand side of the equation (4), we obtain the butterfly shape (Fig.1). FIG. 1. Energy spectrum plotted over a wide range of φ on a pure honeycomb lattice. We investigate energy spectra of Bloch elec- trons in the honeycomb lattice whose unit cell is chosen in three different ways as displayed in Fig.2, and consider its dependence on both position- and density-dependence of impurities or defects. In the first study, we treat the atom at E as a defect and introduce additional im- purities at C, F and H, respectively. (Those three pairs are enough for our purpose since the honeycomb lattice has the π/6 rotation symme- try as well as the transition symmetry.) (E, F ) is a nearest neighbor pair of defects, (E, C) is a next-nearest neighbor pair of defects, and (E, H) is a third nearest neighbor pair of de- fects. In the second study, we extend the size of unit cells (from (a) to (c) in Fig.2) and con- sider the size dependence of the energy spec- trum. Each unit cell contains a single defect (E in (a), L in (b) and Q in (c)) and purity of the honeycomb lattice would get close to a real- istic situation (0.125 (a), 0.056 (b), 0.031 (c)). Throughout this article, all interactions among Bloch electrons are treated as the nearest neigh- bor interactions. 3 III. RESULTS AND DISCUSSIONS A. 2 Defects in a Unit Cell We first begin with the case where a single defect exists at E in a unit cell (Fig. 3). Com- pared with studies on Hofstadter's butterfly on a perfect lattice, a similar fractal nature can be found. Note that there are gapless points at φ = k + 1/2 (k ∈ N≥0) in the E = 0 eV line and large band gaps having triangular shapes whose boundaries are formed by cos curves are formed elsewhere. We next leave the atom at E vacant and introduce additional impurities or defects. Graphs of the spectrum over a wide range of ra- tional magnetic fields φ are displayed in Fig.4. We denote by H F ppπ the interaction of the π or- bital of the atom at F with the neighboring π or- bital. We first consider the case where an impu- rity is inserted at F . According to the figures in Fig.4, there are three different patterns based on positions of impurities. As interactions with the ppπ → 0), energy levels atom E get weaken (H F split and form gaps around E ∼ 1.8 eV. Con- sequently, energy levels within ±1.8 eV ranges are modified and the blank area having a rugby ball shaped energy gap within E < 1.0 eV shrinks. If atoms at E and F are missed, then the graph acquires a new transition symme- try whose period is one-quarter of the origi- nal. When only a single impurity exists in a unit cell, those central band gaps are robust: it does not disappear for arbitrary non-zero value of hopping parameter H E ppπ [15]. However this is not true if additional impurity is considered. In contrast to this case, flow of spectra formed ppπ → 0 show completely in the process of H C different behavior. As seen in the figures of H C ppπ, the arcwise sets of spectra in Fig.3 be- comes boundaries which divide the chunks of spectra into three parts. As a result, all spectra within the range E < 0.8 eV in the figure of H C ppπ = 1/3Hppπ approach to E = 0 eV, namely the corresponding states get degenerated, and the graph of H C ppπ = 0 does not accommodate any fractal nature. Note that when atoms at E and C are missed, then the honeycomb lattice 4 (a) (b) (c) FIG. 2. Enlarged unit cells of the honeycomb lattice. Each of them consists of 8 (a), 18 (b) and 32 (c) atoms respectively. For example, the lattice vector are a1 = 3ax + √ 3ay in the case of (a). √ 3ay, a2 = 2 band gaps around E = 1.8 eV. The last case ppπ → 0 is also different from the above two H H cases. Energy bands localize around six inde- pendent stripes (or cos curves) which remain at H H ppπ = 0. They would correspond to energy bands of the six atoms living in the unit cell. It should be noted that there are band cross- ing points in the E = 0 eV line throughout the process, and the fractal nature of graphs also observed for all non-zero H H ppπ. According to the figures, it is only H H ppπ = 0 that the frac- tality becomes invisible. We may state things more formally as follows. Note that any square matrix M over C is similar to a Jordan matrix (uniquely determined by M ) J1  , J2 . . . Jk FIG. 3. Spectrum for a honeycomb lattice whose unit cell has a single defect. J = (7) is no more arcwise connected and the bound- aries are zigzag. One may naively think that this is consistent with the fact that Hofstadter's butterfly never appears on any one-dimensional lattice system, but this is not correct here. To confirm this, we also tried a case with armchair boundaries and the obtained graph somehow looks fractal (Fig.5). Though there is a wide band gap in E ≤ 1 eV, the other regions are gapless. This is in contrast to the case H F ppπ = 0, which has where Ji = J(λi, di) is labeled by its eigenvalue λi and dimensions di, which are equal to the number of the corresponding degenerated quan- tum states. Thereby the smaller the number of Jordan cells, the more complicated the frac- tal structure becomes. The figure of H H ppπ = 0 would suggest that the minimum requirement for a graph to acquire fractality is k ≥ the number of atoms in a unit cell. (8) 5 H F ppπ = 2/3Hppπ H F ppπ = 1/3Hppπ H F ppπ = 0 H C ppπ = 2/3Hppπ H C ppπ = 1/3Hppπ H C ppπ = 0 H H ppπ = 2/3Hppπ H H ppπ = 1/3Hppπ H H ppπ = 0 FIG. 4. Graphs of energy spectrum over φ = 4P/Q, where P and Q are positive integers. Defects or impurities are introduced at points (E, F ), (E, C) and (E, H). The hopping parameter Hppπ stands for the interaction of the π orbital with the neighboring π orbital. H E ppπ is set to 0 for all cases, i.e. E is vacant. 6 butterflies and their basic shapes are protected (compare also with Fig.4) unless the fractality is lost. Therefore it would be natural to guess that such a gapless point exist without depend- ing on lattice forms. This fact is very particular since the other gapless points which exist in the perfect lattice case disappear when there is a defect in a unit cell. IV. CONCLUSION AND FUTURE WORK In this note, we considered the Hofstadter problem on the honeycomb lattice with defects using the tight-binding Hamiltonian including the nearest-neighbor interactions. We examined the contributions from atoms to the formulation of Hofstadter's butterfly. Naively, increasing the number of defects in a unit cell make graphs less fractal. However we found a case where it recovers its fractal nature and acquires a new translation symmetry. In addition, we also sur- veyed the dependence on the size of unit cells with a single defect and unexpectedly found ro- bust gapless points in the E = 0 eV line. Those points exist without depending on the size of a unit cell and they exist at the center of but- terflies. Therefore we may conclude that the butterfly at φ = 0 point is immortal as long as graphs have fractal structures. For future work, it will be interesting to consider Hamiltonian with the next-nearest- interactions. We found that when a unit cell has more than two defects, choices of missing points are crucial to energy structures. But we also admit that interactions among Bloch wave functions are also crucial. It would be also in- triguing to find a topological method to describe butterflies with or without defects. Especially we are curious how the robustness of the gap- less points can be described in a more theoreti- cal way. When a standard topological insulator has a gapless point, then its topological number usually change there. We expect there would be a certain topological number whose parameter is φ. Moreover, extension to higher dimensional case remains open question. Does such a robust FIG. 5. Butterflies on the honeycomb lattice with armchair boundaries. We use the lattice (b) in Fig.2 cut along A, E, F, M, N, R. In addition, it turns out by numerical calcula- tion that the H C ppπ = 0 case is correspond to the case rankJ = 1 2 dimJ. Therefore, rankJ > 1 2 dimJ will be also necessary to generate a frac- tal graph. B. Dependence on the Size of Unit Cells We next consider the unit cell size depen- dence of the energy spectrum. To address this problem, we extend the previous unit cell which contains 8 atoms as shown in Fig. 2. The graphs of spectrum on unit lattice (b) in Fig.2 with a single defect correspond to (b-1) and (b- 2) in Fig.6, and similarly defined for spectra on unit cell (c). Fractal structures can be found in either case. One significant aspect is that there are robust gapless points (φ = 0.5 for exam- ple) in the E = 0 eV line even if a single de- fect is inserted in a unit cell, without depend- ing on its size (see (b-2) and (c-2) for detail). Those points exist in the Hofstadter problem on the perfect honeycomb lattice (Fig.1) and in the previous case with a defect at E (Fig.3). In Hofstadter's original case on a square lattice, those points also correspond to the center of 7 (b-1) (b-2) (c-1) (c-2) FIG. 6. Spectra on the different enlarged unit cells. (b-1) corresponds to a two-period of the energy spectra considered on the unit cell (b) in Fig.2 and the area around φ = 0.5 is picked up and displayed in (b-2). (c-1) and (c-2) are labeled in a same manner. gapless point generally exist in higher dimen- sional lattice spaces? If so, does it depend on space's dimension? How can it be generically described in a uniformed manner? ACKNOWLEDGMENTS We wish to thank Mikito Koshino for useful discussions and advices. 8 [1] D. R. Hofstadter, Phys. Rev. B 14, 2239 (1976). [2] C. R. Dean, L. Wang, P. Maher, C. Forsythe, F. Ghahari, Y. Gao, J. Katoch, M. Ishigami, P. Moon, M. Koshino, T. Taniguchi, K. Watan- abe, K. L. Shepard, J. Hone, and P. Kim, Na- ture 497, 598 EP (2013). [3] 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, Nature 497, 594 EP (2013). [4] B. Hunt, J. D. Sanchez-Yamagishi, A. F. Young, M. Yankowitz, B. J. LeRoy, K. Watan- abe, T. Taniguchi, P. Moon, M. Koshino, P. Jarillo-Herrero, and R. C. Ashoori, Science 340, 1427 (2013). [5] F. Yılmaz and M. O. Oktel, Phys. Rev. A 95, 063628 (2017), arXiv:1703.08810 [cond- mat.quant-gas]. [6] V. M. Apalkov and T. Chakraborty, Phys. Rev. B 91, 235447 (2015). [7] Y. Hatsugai and M. Kohmoto, Phys. Rev. B 42, 8282 (1990). [8] Y. Hatsuda, H. Katsura, and Y. Tachikawa, (2016), 103023 J. New 18, arXiv:1606.01894 [hep-th]. Phys. [9] Y. Hatsuda, Y. Sugimoto, and Z. Xu, Phys. Rev. D95, 086004 (2017), arXiv:1701.01561 [hep-th]. [10] K. Ikeda, (2017), arXiv:1708.00436 [cond- mat.mes-hall]. [11] K. Ikeda, (2017), arXiv:1708.00419 [cond- mat.mes-hall]. [12] M. Koshino, S. Kagoshima, Lett. 86, 1062 (2001). H. Aoki, K. Kuroki, and T. Osada, Phys. Rev. [13] J. Bruning, V. V. Demidov, and V. A. Geyler, Phys. Rev. B 69, 033202 (2004). [14] T. Kimura, Progress of Theoretical and Exper- imental Physics 2014, 103B05 (2014). [15] S. Islamoglu, M. O. Oktel, and O. b. u. Gulseren, Phys. Rev. B 85, 235414 (2012). [16] R. Peierls, Zeitschrift fur Physik 80, 763 (1933). [17] P. G. Harper, Proceedings of the Physical So- ciety. Section A 68, 879 (1955).
1101.4673	1	1101	2011-01-24T21:56:58	Determination of substrate pinning in epitaxial and supported graphene layers via Raman scattering	[ "cond-mat.mes-hall" ]	The temperature-induced shift of the Raman G line in epitaxial graphene on SiC and Ni surfaces, as well as in graphene supported on SiO2, is investigated with Raman spectroscopy. The thermal shift rate of epitaxial graphene on 6H-SiC(0001) is found to be about three times that of freestanding graphene. This result is explained quantitatively as a consequence of pinning by the substrate. In contrast, graphene grown on polycrystalline Ni films is shown to be unpinned, i.e., to behave elastically as freestanding, despite the relatively strong interaction with the metal substrate. Moreover, it is shown that the transfer of exfoliated graphene layers onto a supporting substrate can result in pinned or unpinned layers, depending on the transfer protocol.	cond-mat.mes-hall	cond-mat	Determination of substrate pinning in epitaxial and supported graphene layers via Raman scattering Nicola Ferralis,∗ Roya Maboudian, and Carlo Carraro Department of Chemical and Biomolecular Engineering, University of California, Berkeley, California 94720† (Dated: August 13, 2018) The temperature-induced shift of the Raman G line in epitaxial graphene on SiC and Ni sur- faces, as well as in graphene supported on SiO2, is investigated with Raman spectroscopy. The thermal shift rate of epitaxial graphene on 6H-SiC(0001) is found to be about three times that of freestanding graphene. This result is explained quantitatively as a consequence of pinning by the substrate. In contrast, graphene grown on polycrystalline Ni films is shown to be unpinned, i.e., to behave elastically as freestanding, despite the relatively strong interaction with the metal substrate. Moreover, it is shown that the transfer of exfoliated graphene layers onto a supporting substrate can result in pinned or unpinned layers, depending on the transfer protocol. PACS numbers: 65.80.Ck, 68.65.Pq, 63.22.Rc Graphene films on insulating substrates have great po- tential for realization of high speed electronic devices [1], as demonstrated recently by graphene transistor opera- tion near terahertz frequencies [2]. Compared to research into graphene's remarkable electronic properties, much less is known about its mechanical properties. Yet, the response of graphene films to mechanical stimuli is an im- portant fundamental topic, with potentially far reaching applications. Noteworthy among these is strain engineer- ing, which offers the tantalizing prospect of manipulating graphene's electromagnetic properties [3, 4]. Moreover, mechanical energy transfer processes dominate the re- sponse (and hence, dictate the design) of graphene-based nanomechanical transducers and actuators [5]. Lastly, the success of many promising graphene-on-insulator fab- rication processes hinges on suitable mechanical manip- ulation of single layers of material [6 -- 8]. Modeling of a graphene film as an elastic continuum can offer much valuable insight into its mechanical re- sponse [9, 10]. The constitutive ingredients of such model, namely Young's modulus and bending rigidity, are known experimentally [11]. However boundary con- ditions must also be specified in order to solve the model. These are determined by the complex interactions be- tween graphene film and substrate, which ultimately re- sult into pinned or unpinned films. A pinned layer im- plies continuity of tangential displacements, while an un- pinned layer implies zero tangential stress. In this Letter, we show that information about film pinning is obtained unambiguously from measurements of the temperature dependence of the Raman G line po- sition, which allow one to separate out "spurious" effects, like charge or strain, that dominate the absolute line po- sition at any given temperature. The significance of the result is illustrated in three different systems. Growth by sublimation on SiC(0001) is shown to yield pinned films. Growth by chemical vapor deposition on Ni films yields unpinned films. Finally, transfer of an exfoliated film 0.1% strained monolayer 0.5% strained monolayer 1595 1590 1585 1580 1575 ] 1 - m c [ n o i t i s o p k a e p G n a m a R 300 350 400 Temperature [K] 450 500 550 FIG. 1. Raman thermal line shift of the G vibrational mode for epitaxial graphene films, prepared with different amounts of strain on 6H-SiC(0001). onto an oxidized Si substrate results in either pinned or unpinned films depending on the transfer protocol. These results may seem surprising at first, given that graphene on SiC is known to grow on top of a "buffer" layer, which decouples it from the substrate [12 -- 15], while on metals, strong interactions involving hybridization of electronic orbitals are believed to exist [16]. The explanation of this apparent paradox resides likely in the nature of the potential energy landscape on semiconductor vs metal surfaces, the former having a highly corrugated poten- tial, the latter a much smoother one. It is the strength of the lateral, rather than the vertical, interactions between film and substrate that is responsible for pinning. Figure 1 shows the values of the frequency of the zone- center optical phonon of graphene (the G line) [17, 18] recorded on two epigraphene monolayer samples grown in ultrahigh vacuum by sublimation of Si from the (0001) face of a 6H-SiC crystal under Si flux [19]. Note that epigraphene films on SiC can be deliberately produced with different amounts of internal stress at room tem- perature, and hence, the G line frequency at room tem- perature could take on any value between 1580 and 1605 cm−1 in a given sample, depending on sample prepara- tion [20]. The samples in Fig. 1 had strain values of 0.1% and 0.5% at room temperature. Previous measurements on freestanding graphene monolayers produced by exfo- liation from HOPG crystals [21, 22] observed an average redshift rate of -0.016 cm−1/K in the temperature range 100 K < T < 300 K, in agreement with theoretical calcu- lations on 2D graphene [23]. The redshift rates observed in epigraphene are almost three times as large, regardless of the amount of internal stress in the films observed at room temperature. Thus, the slope of the curves is in- dependent of the initial amount of stress present in the film, while the absolute position of the G line can vary for several reasons (such as engineered strain [24, 25] or doping [26, 27]), which are decoupled from the thermal redshift. This implies that a single measurement at con- stant temperature is not sufficient to determine whether a film is pinned to the substrate or not. Pinning is in- stead ascertained by performing measurements at differ- ent temperatures. To understand the behavior of the thermal line shift, recall that at constant pressure, the rate of change of phonon frequency with temperature is given by the sum of two terms, dT (cid:19)P (cid:18) dω = χT (T ) + χa(T ) (1) where: χT (T ) = (cid:18) ∂ω ∂a(cid:19)T (cid:18) ∂a ∂T (cid:19)P ; χa(T ) = (cid:18) ∂ω ∂T (cid:19)a . (2) Phonon frequencies depend on temperature in two ways, explicitly and implicitly through the lattice con- stant a. The explicit dependence is contained in the con- stant specific area term χa, which is due to many-phonon interactions: it reflects the anharmonicity of the C-C potential in the graphene lattice. The implicit depen- dence, χT , is given by the product of the rate of change of frequency with lattice constant a, times the rate of lattice thermal expansion. It is this latter term that is affected by pinning of the layer to the substrate, i.e., by boundary conditions at the substrate-film interface. For the pinned layer, we assume continuity of tangential dis- placements, and calculate the frequency shift of the G line in epigraphene by using the coefficient of linear ther- mal expansion (CTE) of SiC, αSiC [28], but keeping for all other terms the ab initio values appropriate for free- standing graphene (denoted by the the subscript "free" and taken from ref. [23]): ω(T ) = ω(T0) +Z T T0 dT ′χa,free(T ′) +Z T T0 dT ′χT,free(T ′) αSiC(T ′) αfree(T ′) . (3) ] 1 - m c [ t f i h s y c n e u q e r F 5 0 -5 -10 -15 -20 -25 "Freestanding" graphene - ref. [21] "Freestanding" graphene - this work "Freestanding" graphene - calculation, ref. [23] Epitaxial graphene on SiC - this work Epitaxial graphene on SiC - calculation, this work 100 200 300 400 Temperature [K] 2 500 FIG. 2. Comparison between theoretical and experimental thermal G-line shifts for freestanding and epitaxial graphene films (re-plotted from Fig. 1). The thermal line shifts in both the theoretical curves and the experimental data are referred to the position of the G line at absolute zero. Note that the integration constant, ω(T0) (where T0 is an arbitrary reference temperature), need not coincide with the line position of a freestanding film in equilib- rium: its value may be shifted due to engineered stress or charge effects, for example. The results of the calcu- lations are shown in Fig. 2, along with the data. For comparison, we also plot the calculated thermal G-line shift of free graphene [23], along with experimental mea- surement from ref. [21], which we have augmented here by performing measurements on free films at higher tem- peratures. The agreement between theory and experiment con- firms that on the SiC(0001) surface, substrate effects are entirely accounted for by using the boundary conditions appropriate for a pinned monolayer. However, as shown below, this is not to be regarded as a forgone conclu- sion: we have observed strikingly different behavior on different substrates. Graphene monolayer growth by chemical vapor depo- sition (CVD) has been demonstrated on several metal surfaces, including Ni substrates [6], where strong film- substrate interactions have been observed [29, 30]. In some cases, interactions lead to the suppression of the Raman signal, possibly owing to the hybridization be- tween the metal d bands and the carbon pz bands [31]. On polycrystalline Ni films, perfect epitaxy is not ex- pected, and in fact, a Raman signal is always observed [6]. We have measured the thermal line shifts for graphene monolayers grown by CVD on thin Ni films (300 nm) on SiO2/Si substrates [32]. The results are shown in Fig. 3, along with calculated values for freestanding graphene [23], for graphene on Si, and for graphene on bulk Ni. We emphasize that the mechanical behavior of a well ] 1 - m c [ t f i h s y c n e u q e r F 5 0 -5 -10 -15 -20 -25 250 300 350 400 Graphene on Ni-film/SiO2/Si "Freestanding" graphene - calculation ref. [23] Graphene on Si - calculation, this work Graphene on Ni - calculation, this work Sample dω/dT [cm−1/K] T range Ref. Theory Ref. [K] 3 Freestanding Pressed on SiO2/Si -0.052±0.004 300-400 -0.009±0.002 150-250 [21] -0.015±0.003 300-400 [21] ∗∗ -0.011 [23] -0.017 [23] ∗∗ -0.046∗ -0.040±0.002 400-500 [38] -0.052∗ -0.043±0.013 300-400 -0.048 ∗ Calculation treats the substrate as pure silicon. On Au/SiNx/Si Epi-G on SiC ∗∗ ∗∗ ∗∗ ∗∗ This work. 550 600 650 700 TABLE I. Comparison between values of the thermal line- shift rates dω/dT measured over different temperature ranges for graphene films prepared by different methods and calcu- lated values from Eq. 3. 450 500 Temperature [K] FIG. 3. The experimental thermal line shifts of the Raman G line for a monolayer graphene grown via CVD on a Ni/SiO2/Si substrate is compared to the calculated values of the thermal shifts for freestanding graphene, graphene pinned to Si, and graphene pinned to Ni (from Eq. 3, using the linear thermal expansion coefficient of Si [33] and Ni [34], respectively.) adhered thin film (or thin film stack) on a thick sub- strate is dictated by the properties of the bulk substrate. When comparing to our experiment, the relevant CTE to be used in Eq. 3 is that of Si [33], since the Si wafer thickness is 500 µm vs. 300 nm Ni film thickness. The good agreement between the experimental data and the theoretical curve for freestanding graphene, rather than for graphene on Si substrate, indicates that on polycrys- talline Ni films, graphene is completely unpinned from the substrate and it behaves essentially like a freestand- ing layer. This free-like behavior can be rationalized by considerations of surface potential corrugation, which is usually much smaller on metal surfaces compared to in- sulators (or semiconductors). Since film pinning involves tangential displacements, it is plausible that graphene monolayers grown on metal surfaces, are free to slide lat- erally, the rather strong electronic interactions with the substrate notwithstanding. Other fabrication methods involve transfer of graphene layers onto an inert substrate, usually an insulator [6 -- 8]. The question then arises whether these supported layers are free or pinned. Interestingly, two different transfer methods lead consistently to contrasting answers. In the first method, we produced graphene films by mechanical exfoliation of HOPG crystals [35]; the substrate die (a 300 nm thick thermal oxide film on Si) was dipped in a toluene solution where graphene flakes were dispersed. Graphene layers thus collected are referred to as "free- standing." In the second method, graphite platelets were exfoliated from HOPG samples using ScotchT M tape, and immediately pressed against the SiO2/Si substrate die. After sonication of the die to remove the larger flakes, ∼1 µm single and double layers of graphene (char- acterized by Raman spectroscopy [36] and optical mi- croscopy [37]) remained attached to the surface. Figure 4 "Freestanding" graphene - calculation ref. [23] "Freestanding" graphene - this work Graphene pressed on SiO2/Si - this work Graphene on Au/SiNx/Si - ref. [38] Strongly interacting graphene on Si - calculation, this work 10 5 0 -5 -10 -15 -20 ] 1 - m c [ t f i h s y c n e u q e r F 250 300 350 400 450 500 550 600 650 700 Temperature [K] FIG. 4. Experimental thermal line shifts of the Raman G-line for freestanding graphene and for graphene films pressed onto a SiO2 and Au/SiNx [38] substrates. shows the thermal line shifts measured for these two sam- ples (solid and empty squares, respectively). For compar- ison, data corresponding to graphene layers pressed onto Au/SiNx/Si substrates are plotted from ref. [38]. A vi- sual inspection shows very similar thermal line shifts for graphene pressed on SiO2/Si and Au/SiNx/Si substrates (Figure 4). The experimental data is compared with cal- culations for freestanding graphene and graphene pinned to a Si substrate, (solid and dashed lines in Fig. 4, re- spectively). Agreement between theory and experiments points out that only the films dispersed from solution behave as freestanding, whereas pressing or stamping re- sults in pinned films. Consistent with our discussion of tangential vs vertical interactions, we believe pinning of pressed films is brought about by their better conforma- tion to nanoscale surface roughness, as well as by the squeezing out of liquid-like layers at the film-substrate interface (e.g., physisorbed water [39]). The values of dω/dT and their corresponding temperature ranges are reported in Table I. In summary, we have shown that the position of the Raman G line of graphene monolayers at a given tem- perature can be influenced by the graphene interaction with the substrate, resulting in thermal line shifts that can be calculated based on simple thermodynamic ar- guments, the only inputs being thermomechanical prop- erties of free graphene (known, e.g., from density func- tional calculations) and the thermal expansion coefficient of the substrate. Conversely, experimental determination of the temperature shift of the G peak in the Raman spectrum allows one to determine whether a graphene film, grown or transferred onto a substrate, is pinned to it, or elastically decoupled from it. These results shed light on the diverse effects of substrate interactions in some of the most common graphene production meth- ods. Furthermore, they imply that pinning effects must be considered properly when using Raman spectroscopy for thermal measurements, such as in the determination of thermal conductivity [38, 40 -- 42]. Support from the National Science Foundation (Grants EEC-0832819 and CMMI-0825531), and from DARPA- MTO is gratefully acknowledged. ∗ [email protected] † Current address: Department of Materials Science and Engineering, MIT, MA 02139 [1] F. Schwierz, Nat. Nano. 5, 487 (2010). [2] Y.-M. Lin, C. Dimitrakopoulos, K. Jenkins, D. Farmer, H.-Y. Chiu, A. Grill, and P. Avouris, Science 327, 662 (2010). [3] F. Guinea, M. Katsnelson, and A. Geim, Nature Phys. 6, 30 (2009). [4] N. Levy, S. Burke, K. Meaker, M. Panlasigui, A. Zettl, F. Guinea, A. Castro Neto, and M. Crommie, Science 329, 544 (2010). [5] J. Bunch, A. van der Zande, S. Verbridge, I. Frank, D. Tanenbaum, J. Parpia, H. Craighead, and P. McEuen, Science 315, 490 (2007). [6] K. Kim, Y. Zhao, H. Jang, S. Lee, J. Kim, K. Kim, J.-H. Ahn, P. Kim, J.-Y. Choi, and B. Hong, Nature 457, 706 (2009). [7] A. Reina, X. Jia, J. Ho, D. Nezich, H. Son, V. Bulovic, M. Dresselhaus, and J. Kong, Nano Lett. 9, 30 (2009). [8] X. Li, W. Cai, J. An, S. Kim, J. Nah, D. Yang, R. Piner, A. Velamakanni, I. Jung, E. Tutuc, S. Banerjee, L. Colombo, and R. Ruoff, Science 324, 1312 (2009). [9] E.-A. Kim and A. Castro Neto, Eur. Phys. Lett. 84, 57007 (2008). [10] J.H. Los, M.I. Katsnelson, O.V. Yazyev, K.V. Za- kharchenko, and A. Fasolino, Phys. Rev. B 80, 121405 (2009). [11] I. Frank, D. Tanenbaum, A. Van der Zande, and 4 [15] K.V. Emtsev, F. Speck, T. Seyller, L. Ley, and J.D. Riley, Phys. Rev. B 77, 155303 (2008). [16] Y. Wu, T. Yu, and Z. Shen, J. Appl. Phys. 108, 071301 (2010). [17] A. Ferrari, Sol. St. Comm. 143, 47 (2007). [18] M. Malard, M. Pimenta, G. Dresselhaus, and M. Dres- selhaus, Phys. Reports 473, 51 (2009). [19] Raman spectra were acquired ex situ in a JYHoriba LabRAM spectrometer in backscattering configuration (excitation line: 633 nm). Substrate temperature was controlled in air using a heating stage (Linkham THMS350V), with a measurement error of about 1 K. [20] N. Ferralis, R. Maboudian, and C. Carraro, Phys. Rev. Lett. 101, 156801 (2008). [21] I. Calizo, A. Balandin, W. Bao, F. Miao, and C. Lau, Nano Lett. 7, 2645 (2007). [22] I. Calizo, F. Miao, W. Bao, C. Lau, and A. Balandin, Appl. Phys. Lett. 91, 071913 (2007). [23] N. Bonini, M. Lazzeri, N. Marzari, and F. Mauri, Phys. Rev. Lett. 99, 176802 (2007). [24] M. Huang, H. Yan, C. Chen, D. Song, T. Heinz, and J. Hone, Proc. Nat. Ac. Sci. 106, 7304 (2009). [25] T.M.G. Mohiuddin, A. Lombardo, R.R. Nair, A. Bonetti, G. Savini, R. Jalil, N. Bonini, D.M. Basko, C. Galiotis, N. Marzari, K.S. Novoselov, A.K. Geim, and A.C. Fer- rari, Phys. Rev. B 79, 205433 (2009). [26] S. Pisana, M. Lazzeri, C. Casiraghi, K. Novoselov, A. Geim, A. Ferrari, and F. Mauri, Nat. Mat. 6, 198 (2007). [27] A. Das, S. Pisana, B. Chakraborty, S. Piscanec, S. Saha, U. Waghmare, K. Novoselov, H. Krishnamurthy, A. Geim, A. Ferrari, and A. Sood, Nat. Nano. 3, 210 (2008). [28] G. Slack and S. Bartram, J. Appl. Phys. 46, 89 (1975). [29] Y.S. Dedkov, M. Fonin, U. Rudiger, and C. Laubschat, Phys. Rev. Lett. 100, 107602 (2008). [30] A. Gruneis and D.V. Vyalikh, Phys. Rev. B 77, 193401 (2008). [31] A. Allard and L. Wirtz, Nano Lett.(2010), doi: 10.1021/nl101657v. [32] Graphene growth on Ni substrates is carried out in a LP-CVD reactor for 5 minutes at 1000◦C, in a flow of 50 sccm CH4, 200 sccm Ar and 65 sccm H2. [33] H. Watanabe, N. Yamada, and M. Okaji, Inter. J. Ther- mophys. 25, 221 (2004). [34] F. Nix and D. MacNair, Phys. Rev. 60, 597 (1941). [35] K. Novoselov, A. Geim, S. Morozov, D. Jiang, Y. Zhang, S. Dubonos, I. Grigorieva, and A. Firsov, Science 306, 666 (2004). [36] P. Blake, E. W. Hill, A. Castro Neto, K. Novoselov, D. Jiang, R. Yang, T. J. Booth, and A. Geim, Appl. Phys. Lett. 91, 063124 (2010). [37] Z. Ni, H. Wang, J. Kasim, H. Fan, T. Yu, Y. Wu, Y. Feng, and Z. Shen, Nano Lett. 7, 2758 (2007). [38] W. Cai, A. Moore, Y. Zhu, X. Li, S. Chen, L. Shi, and R. Ruoff, Nano Lett. 10, 1645 (2010). P. McEuen, J. Vac. Sci. Technol. B 25, 2558 (2007). [39] C.-C. Chen, W. Bao, J. Theiss, C. Dames, C. Lau, and [12] W. Chen, H. Xu, L. Liu, X. Gao, D. Qi, G. Peng, S. Tan, Y. Feng, K. Loh, and A. Wee, Surf. Sci. 596, 176 (2005). [13] A. Mattausch and O. Pankratov, Phys. Rev. Lett. 99, 076802 (2007). [14] F. Varchon, R. Feng, J. Hass, X. Li, B.N. Nguyen, C. Naud, P. Mallet, J.-Y. Veuillen, C. Berger, E.H. Con- rad, and L. Magaud, Phys. Rev. Lett. 99, 126805 (2007). S. Cronin, Nano Lett. 9, 4172 (2009). [40] A. Balandin, S. Ghosh, W. Bao, I. Calizo, D. Tewelde- brhan, F. Miao, and C. Lau, Nano Lett. 8, 902 (2008). [41] C. Faugeras, B. Faugeras, M. Orlita, M. Potemski, R. Nair, and A. Geim, ACS Nano 4, 1889 (2010). [42] S. Ghosh, W. Bao, D. Nika, S. Subrina, E. Pokatilov, C. Lau, and A. Balandin, Nat. Materials 9, 555 (2010).
1905.11565	1	1905	2019-05-28T01:44:26	Spin-Orbit Torque in a Single Ferromagnetic Layer with Large Spin-Orbit Coupling	[ "cond-mat.mes-hall", "cond-mat.mtrl-sci" ]	Spin-orbit torque in heavy metal/ferromagnet heterostructures with broken spatial inversion symmetry provides an efficient mechanism for manipulating magnetization using a charge current. Here, we report the presence of a spin torque in a single ferromagnetic layer in both asymmetric MgO/Fe0.8Mn0.2 and symmetric MgO/Fe0.8Mn0.2/MgO structures, which manifests itself in the form of an effective field transverse to the charge current. The current to effective field conversion efficiency, which is characterized using both the nonlinear magnetoresistance and second-order planar Hall effect methods, is comparable to the efficiency in typical heavy metal/ferromagnet bilayers. We argue that the torque is caused by spin rotation in the vicinity of the surface via impurity scattering in the presence of a strong spin-orbit coupling. Instead of cancelling off with each other, the torques from the top and bottom surfaces simply add up, leading to a fairly large net torque, which is readily observed experimentally.	cond-mat.mes-hall	cond-mat	Spin-Orbit Torque in a Single Ferromagnetic Layer with Large Spin-Orbit Coupling Ziyan Luo1, Qi Zhang1,2, Yanjun Xu1, Yumeng Yang1, Xinhai Zhang2, and Yihong Wu1,a) 1 Department of Electrical and Computer Engineering, National University of Singapore, 4 Engineering Drive 3, Singapore 117583, Singapore 2 Department of Electrical & Electronic Engineering, Southern University of Science and Technology, Xueyuan Rd 1088, Shenzhen 518055, China Spin-orbit torque in heavy metal/ferromagnet heterostructures with broken spatial inversion symmetry provides an efficient mechanism for manipulating magnetization using a charge current. Here, we report the presence of a spin torque in a single ferromagnetic layer in both asymmetric MgO/Fe0.8Mn0.2 and symmetric MgO/Fe0.8Mn0.2/MgO structures, which manifests itself in the form of an effective field transverse to the charge current. The current to effective field conversion efficiency, which is characterized using both the nonlinear magnetoresistance and second-order planar Hall effect methods, is comparable to the efficiency in typical heavy metal/ferromagnet bilayers. We argue that the torque is caused by spin rotation in the vicinity of the surface via impurity scattering in the presence of a strong spin-orbit coupling. Instead of cancelling off with each other, the torques from the top and bottom surfaces simply add up, leading to a fairly large net torque, which is readily observed experimentally. a) Electronic mail: [email protected] 1 I. INTRODUCTION When a charge current passes through a ferromagnet (FM), due to the imbalance in electron density of states at the Fermi level and scattering asymmetry between spin-up and spin-down electrons, it becomes polarized, thereby generating a net spin current flowing in the charge current direction. In addition to longitudinal charge and spin currents, transverse charge and spin currents are also generated by the anomalous Hall effect (AHE), leading to charge and spin accumulation at the side surfaces or edges at steady-state [1]. So far, the study of AHE has been mainly focused on charge accumulation because it can be detected directly as a voltage signal, and very little attention was devoted to the spin accumulation. Recently, several groups have attempted to exploit the AHE-induced spin accumulation and related spin torque for magnetization switching applications in FM/nonmagnet (NM)/FM trilayers [2-5] which, compared to spin-orbit torque (SOT) generated by the spin Hall effect (SHE), offers the possibility of controlling the spin polarization direction by manipulating the magnetization direction of one of the FM layers. In addition, experiments have been carried out to detect the AHE-induced spin current through spin injection and nonlocal electrical detection [6-9]. We have recently demonstrated that, for a FM with large spin-orbit coupling (SOC), the spin accumulation would also result in an AHE-related magnetoresistance (MR), we termed it as anomalous Hall magnetoresistance (AHMR) [10]. The AHMR exhibits the same magnetization angle-dependence as that of spin Hall magnetoresistance (SMR) in heavy metal (HM)/FM [11-16], both of which show a linear dependence on current. In all these studies, the magnetization is typically in a saturation state, and therefore, nonlinear effect induced by spin torque, if any, is often suppressed by the large field and can hardly be observable. The AHE occurs in solids either from extrinsic mechanisms like skew scattering and side-jump or intrinsic mechanism related to the Berry phase of electronic band structure [1, 17], with the extrinsic mechanism dominated in highly conductive ferromagnet. Regardless of its origin, it is commonly believed that spin polarization of the deflected electrons will follow the local magnetization direction (parallel for 2 down-spin and antiparallel for up-spin) due to strong exchange interaction, so does the polarization of spins accumulated at the surfaces. Such kind of self-alignment of spin direction with the local magnetization facilitates the control of polarization direction of AHE-generated non-equilibrium spins using an external field, which is not possible for spin current generated by SHE. In the meantime, this also means that the accumulated spins near the surfaces are unable to exert any torque on the magnetization itself, limiting its practical applications. To circumvent this problem, several groups have proposed to employ a FM/NM/FM trilayer structure, in which the first FM serves as the spin generator whereas the second FM functions as a spin detector whose magnetization direction can be manipulated by the spin- transfer torque from the spin current generated by the first FM layer [2-5]. However, the spin current will decay while traveling across the NM layer, which works as a separator to decouple the two FM layers, resulting in a low efficiency. The same difficulty also faces FM/NM/FM trilayers in which spin current generated at the bottom FM layer and spacer layer interface exerts a torque on the top FM layer [18, 19]. Here, we demonstrate that a damping-like (DL) SOT exists in a single thin FM layer as long as it exhibits a sizable AHE arising from a large SOC. Specifically, for a thin Fe0.8Mn0.2 layer with in-plane anisotropy, we observed a DL SOT which manifests itself in the form of an effective in-plane field transverse to the charge current. The strength of the DL effective field is characterized by measuring both the nonlinear magnetoresistance and second-order planar Hall effect (PHE) induced by the SOT. For a symmetric structure of MgO(2)/Fe0.8Mn0.2(5)/MgO(2), an effective field to current density ratio of 0.242 Oe/(1010A/m2) is obtained, which is comparable to that of FL effective field in Pt/NiFe bilayers [20, 21]. We argue that part of the AHE-generated electron spins near the surfaces is misaligned with the magnetization direction due to precession upon scattering by the SOC scattering center, and the backflow of these misaligned spins exerts a torque on the magnetization. The effects from top and bottom surfaces simply add up rather than cancelling out with each other due to the scattering asymmetry of up-spin and down-spin electrons at the two surfaces, which eventually leads to an observable net spin torque. 3 II. EXPERIMENTAL DETAILS AND RESULTS A. Nonlinear magnetoresistance measurement Samples with the structure MgO(d1)/Fe0.8Mn0.2(d2)/sub and MgO(d1)/Fe0.8Mn0.2(d2)/MgO(d1)/sub are deposited on quartz substrate. The numbers inside the parentheses indicate thickness in nm. All the materials are prepared by magnetron sputtering with a base pressure of 2 × 10-8 Torr and working pressure of 3 × 10-3 Torr, respectively. Devices are patterned and contacted by Pt(10)/Cu(200)/Ta(5) (Ta is deposited first) electrodes using combined techniques of photolithography and lift-off. Both nonlinear MR and second-order PHE were measured to characterize the SOT. The MR measurements were performed using a Quantum Design Versalab Physical Property Measurement System. During the measurements, the sample is rotated relative to an in-plane field with fixed direction and then measure the nonlinear MR by applying an AC current along x direction. Instead of using the standard lock-in technique, we employed a Wheatstone full bridge to measure the nonlinear magnetoresistance [20, 22]. Compared to the lock-in technique, the latter is not affected by thermal drift and low-noise signals can be readily obtained without resorting to any post-measurement data processing. On the other hand, Hall bars are used for the second- order PHE measurements; more details will be presented when we discuss the PHE data. FIG. 1. (a) Schematic of Wheatstone bridge used for nonlinear MR measurement. (b) nlR vs m at different external fields. (c) Experimental (symbol) and fitting (solid-line) results for the nlR vs m curve obtained at Hex = 200 Oe. Dash-dotted and dashed- lines refer to the decomposed nlR  and , m nlR  components, respectively. ,3 m 4 Figure 1(a) shows the schematic of a Wheatstone bridge which consists of four rectangular shaped elements with a dimension of 100 μm (width) × 500 μm (length). When an AC current, acI t 0 sin I , is applied to two terminals of the bridge along x-axis, half of the current ( 1 2 I 0 sin t ) will flow in each element. Any nonlinear MR ( nlR ) can be readily detected from the bridge output voltage which is given by: V out  1 2 I R 0 nl  1 2 I R 0 nl cos2 t . (1) Here nlR is the change in resistance originated from both the SOT ( nlR ) and thermoelectric effect ( SOT T nlR ) [23, 24]. The effect of Oersted field can be excluded because it is almost cancelled out in a single metallic layer. As shown in Eq. (1), one can directly measure the DC component as the nonlinear signal because it is the same as the amplitude of the second harmonic signal, therefore, there is no need to use the lock-in technique [20, 22]. We first performed the angular dependent harmonic measurement at room temperature by applying a rotational field (Hex) in the xy-plane with the strength varying from 100 Oe to 30 kOe. Figure 1(b) shows the typical result of nlR as a function of field angle ( m ) of MgO(2)/Fe0.8Mn0.2(5). The shape of angle- dependence curve depends strongly on the strength of the external field, but in general can be fitted very well using the following equation: R nl  R nl ,  m sin( )   0  m  R nl ,3  m sin3( )   0  m  R 0 , (2) where 0R and 0 are the resistance and angle offsets due to experimental setup, and nlR  and , m nlR  are ,3 m the coefficient of sin m and sin 3 m components, respectively. As an example, Fig. 1(c) compares the experimental and fitting results at exH = 200 Oe. The square symbols are the measured results and black solid-line is the overall fitting based on Eq. (2). The dash-dotted and dashed lines correspond to the 1st and 5 2nd terms of Eq. (2), respectively. The best fitting to Eq. (2) yields nlR  = 6.83 mΩ, , m nlR  = 4.91 mΩ, ,3 m 0R = 3.14 mΩ, 0 = 3.34o. Based on previous studies of HM/FM bilayers, the symmetry of the signal shown in Fig. 1(b) suggests that there is an SOT effective field present in the single FM layer and its direction is along y-axis [23]. Without losing generality, the longitudinal resistance of each element can be considered as consisting of two parts: one is the linear resistance ( lR ) which is independent of the current and the other is the nonlinear resistance ( nlR ) which is proportional to the current. The former can be rewritten as: R R l z   ( R R x z  )sin 2   m cos m 2  ( R y  R z )sin 2   m sin m 2 , (3) where m , m are the polar and azimuthal angles of m, respectively, and iR is the longitudinal resistance measured when m is saturated along the direction i  , , x y z . The second term corresponds to the anisotropic magnetoresistance (AMR). As for the third term, although it exhibits a symmetry similar to that of spin Hall magnetoresistance (SMR) in HM/FM bilayer, as we demonstrated in the previous work, it has an AHE origin and is dubbed as anomalous Hall magnetoresistance (AHMR) [10]. On the other hand, the nonlinear part caused by the small changes in m and m by m and m , due to current-induced effective field  , is approximately given by IH R SOT nl ( R   xz 2 cos  m R   yz 2 sin  m )sin 2   m R    m 2 sin yx    m sin 2  , (4) m m and where R   ij ( , R R i j i  j  , , ) x y z . For a film with in-plane anisotropy, m m can be estimated is the () component of using the first order approximation [23-27], i.e.,   m H  I / ( H ex  H d ) and H H  / m I ; here, IH ( IH ) ex  IH , dH is the out-of-plane demagnetizing field. Note that we have ignored the in-plane anisotropic field as it is much smaller than the external field. The contribution of m induced change to nlR SOT , if any, should be very small as m  / 2 and H H  . Therefore, we only need to I d 6 consider the change in nlR SOT caused by m . When the SOT effective field is in negative y-direction (based on our measurement geometry), its projection in the azimuthal direction is given by H  I   H I cos  m . Since the effective field is proportional to the current density, it can be written as H I j 0_ eff rms ; here, eff is the strength of effective field per unit current density in the FM layer, which represents the SOT generation efficiency, and j 0_ rms  I 0 / (2 2 wd ) is the root-mean-square (rms) amplitude of the applied AC current density, with w the width of individual elements and d the thickness of FM layer. Substitute all these parameters into Eq. (4), one obtains R SOT nl   R xy 2 2  m (   m sin eff sin H ex ) j 0 _ rms   R xy 2 2  m (  eff sin 3  m ) j 0_ rms sin H ex . (5) By adding to the thermoelectric contribution, R T   nl  Il 0 2 w  T z sin   m sin m [28], where  is the effective coefficient that accounts for anomalous Nernst effect (ANE) and spin Seebeck effect (SSE), l is the length of the individual elements, and zT is the thermal gradient along z direction, the total change of resistance induced by the current may be written as 2  m (   m sin eff sin H ex ) j 0_ rms  R  xy 2 R nl  R  xy 2 2  m (  eff sin3  m ) j 0_ rms  sin H ex Il 0  2 w T  z sin   m sin m . (6) As mentioned above, when an AC current, acI t 0 sin I , is applied to two terminals of the bridge along x- axis, half of the current ( 1 2 I 0 sin t ) will flow in each element. The Wheatstone bridge consists of 4 identical elements (by design), if we consider one of them first (e.g., element #1), the voltage across its two terminal is given by V 1  [ R z  ( R x  R z )sin 2 ( R   yz R   xz )sin 2 2 m cos   m  eff  j 0_   m sin 2 m ( R rms 2 2 sin )sin cos t R  z y m sin   m H   m I 0 2 l  w  ex sin  t T  z sin   m sin m . ] 1 2 I o sin t  7 Similarly, for element #2, the voltage across its two terminals is given by (note the current is in opposite phase of element #1): (7) V 2 [   R z  ( R x  R z )sin 2 ( R   yz R   xz )sin 2   m sin 2 m 2 m   m j 0_ cos  eff  ( rms 2 2 m )sin R R  z y sin cos t   m H   m I 0 2 sin l  w  ex sin The algebraic sum of V1 and V2 gives the bridge output voltage  t T  z sin   m sin m . I o sin t  ] 1 2 (8) V out  V V 1 2  [(   R xz   R )sin 2  m sin 2  m yz  eff [(   R xz   R )sin 2  m sin 2  m yz  eff j 0_ rms 2 H cos  m ex j 0_ rms H ex l w   cos  m   l w I 0 2  T z sin   m sin m I 0 4  T z sin   m sin m ] I o (1 cos 2  t  ) ] I o 2 sin t  . (9) From the time-average or DC output of the bridge voltage, we can obtain the nonlinear resistance: R nl  V 1 2 out I 0  R SOT , nl  m sin  m  R SOT ,3 nl  m sin3  m  R  nl T ,3  m sin  m . (10) Here R SOT , nl  m  R SOT ,3 nl  m ( R   xy 2 H ex )  eff j 0_ rms , and R T l   w , nl  m I 0 2 T  z . nlR  T  , m is independent of the external field as long as the external field is sufficiently large to saturate the magnetization in the field direction [23, 24, 28] (note: m  / 2 ). Eq. (10) reproduces the experimental results well except for the angle and resistance offsets which are caused by the sample fabrication and measurement processes rather than the intrinsic properties of the sample. 8 nlR  and SOT , m FIG. 2. (a) nlR  as a function of Hex. Symbols are extracted from the decomposed fitting results. Solid-lines are SOT ,3 m the fitting results. (b) AC current density dependence of nlR  at Hex = 100, 200, and 500 Oe, respectively. (c) Geometry of SOT ,3 m angle-dependent magnetoresistance (ADMR) measurement in xy plane. (d) Results of xyR as a function of m in xy plane measured at Hex = 30 kOe. To examine the field dependence of nlR  and SOT , m nlR  , we perform the angle-dependence fitting of all SOT ,3 m data measured at different external fields, and the extracted values for nlR  are plotted in Fig. SOT ,3 m 2(a) as a function of Hex. As shown in the figure, nlR  decrease sharply as Hex increases in the SOT ,3 m nlR  and SOT , m nlR  and SOT , m low field range (< 5 kOe ), and at high filed, nlR  and SOT , m nlR  gradually diminishes. The solid-lines are the SOT ,3 m fitting results using R SOT , nl  m  R SOT ,3 nl  m  ( C H SOT ex ) with CSOT = 970 mΩꞏOe (at j0_rms = 0.71×1010 A/m2). The excellent agreement with Eq. (10) suggests that, indeed, there exists a SOT effective field in y-direction. To further ascertain its current related origin, we plot in Fig. 2(b) nlR  as a function of current density at SOT ,3 m Hex = 100, 200, and 500 Oe, respectively. As expected, an almost linear dependence is found between nlR  and the current density at all field values. The slight deviation from the linear trend at high current SOT ,3 m density is presumably caused by the thermal effect. To estimate the SOT efficiency, we measured xyR of an individual element by aligning m with a rotating external field of 30 kOe in xy plane as indicated in xyR as a function of m is plotted in Fig. 2(d). For a MgO(2)/Fe0.8Mn0.2(5) sample, xyR Fig. 2(c). The turned out to be 7.38 Ω, from which eff is calculated as 0.372 Oe/(1010A/m2); this value is comparable to the efficiency reported for Pt/NiFe bilayer at similar thickness range [20, 21]. 9 MgO Fe0.8Mn0.2 Quartz MgO Fe0.8Mn0.2 MgO Quartz MgO NiFe Quartz MgO NiFe MgO Quartz 0.4 0.3 0.2 0.1 0.0 (i) (ii) (iii) (iv) ) 2 m A / 0 1 0 1 ( / e O ( f f e FIG. 3. Values of eff obtained for different structures: (i) MgO/Fe0.8Mn0.2, (ii) MgO/Fe0.8Mn0.2/MgO, (iii) MgO/NiFe, (iv) MgO/NiFe/MgO. The error bars indicate the fitting accuracy. To shed light on the respective role of the FM and its capping layer and underlayer, we compare the eff values for the following four samples: i) MgO(2)/Fe0.8Mn0.2(5), ii) MgO(2)/Fe0.8Mn0.2(5)/MgO(2), iii) MgO(2)/NiFe(5), iv) MgO(2)/NiFe(5)/MgO(2). The rms amplitude and frequency of the applied AC current density were fixed at 0.71×1010A/m2 and 5000 Hz for all the samples. As shown in Fig. 3, the eff value for MgO(2)/Fe0.8Mn0.2(5)/MgO(2) is 0.242 Oe/(1010A/m2); although it is smaller than that of MgO(2)/Fe0.8Mn0.2(5), the difference is not that large. Considering the fact that the crystalline structure and interface roughness for these two samples might be different, further structural analysis are required in order to reveal the true cause for this difference. In a sharp contrast, the eff values for MgO(2)/NiFe(5) is about one order of magnitude smaller than that of MgO(2)/Fe0.8Mn0.2(5). Again, there is no noticeable difference between the eff values for MgO(2)/NiFe(5) and MgO(2)/NiFe(5)/MgO(2) either. In fact, the eff value for MgO(2)/NiFe(5)/MgO(2) is even slightly larger than that of MgO(2)/NiFe(5). If we compare (i) with (iii) and (ii) with (iv), it is rather clear that, by simply changing the FM layer, the eff varies in a large magnitude. Therefore, eff depends more on the FM layer than the asymmetry of the interfaces. 10 These results suggest that Rashba-Edelstein effect can be excluded as the origin for the observed SOT [29- 31], which if exists, would have been largely cancelled out in symmetric structures due to different sign of SOT at top/bottom surfaces. B. Second-order planar Hall effect measurement The current-induced SOT indeed can induce nonlinear MR, but one cannot completely rule out other possibilities. To further test the relevance of SOT, we measure the current-induced effective field directly using the second-order planar Hall effect (PHE) technique [32-36], from which we extract eff and compare it with the values obtained from the nonlinear MR measurements. As illustrated in Fig. 4(a), a bias current I is injected into a 200-µm-wide, 2000-µm-long Hall bar consisting of MgO(2)/Fe0.8Mn0.2(5)/MgO(2). A small transverse bias magnetic field Hy generated by a pair of Helmholtz coils is superimposed with the effective field. The planar Hall voltage was measured at different Hy with a sweeping external magnetic field Hx in x-direction, and at a positive and a negative bias current, respectively. The second-order planar Hall voltage is calculated from the measured Hall voltages as V H  xy ( y )  V xy (  H I H  , , y )  ex V H I H xy   ( , , y ) ex . (11) Under the small perturbation limit, i.e., both the current induced field (Heff) and applied transverse bias field (Hy) are much smaller than the external field (Hex), the change in in-plane magnetization direction is proportional to ( H H  y ) / H ex eff . The linear dependence of second-order PHE voltage on the algebraic sum of Hy and Heff allows one to determine the effective field by varying Hy as both fields play an equivalent role in determining the magnetization direction. After some algebra, one can derive the relation . Therefore, by linearly fitting V H  xy ( y  0) /[ V H  xy  ( ) y   V H xy ( )]  H y eff 2 H y V H ( xy  0) against y y y ) ) ( 2 H  V H xy V H  xy (  , the ratio H can be determined from the slope of the curve, which in turn can be used to extract Heff. Figure 4(b) shows one set of second-order PHE voltage curves for MgO(2)/Fe0.8Mn0.2(5)/MgO(2) with bias current of 5mA at different Hy (0, (cid:3397)0.5 and (cid:3398)0.5 Oe). The eff y 11 different magnitude of the signal is attributed to the change of the total field along y-direction. The ratio of Heff to 2Hy can be extracted from the slope of V H ( xy  0) y versus diffV plot, where V  diff   V H xy ( y   0.5 ) Oe   V H xy ( y   0.5 ) Oe , as shown in Fig. 4(c). The data for Hex < 30 Oe are not included in the fitting in order to satisfy the requirement of small angle perturbation. Fig. 4(d) shows the dependence of Heff on current density. As expected, Heff changes linearly with the current density, and from the slope, the SOT efficiency, eff , is extracted as 0.187 Oe/(1010A/m2); this is comparable to the value of 0.242 Oe/(1010A/m2) obtained from the nonlinear MR measurement. FIG. 4. (a) Schematic of the second-order PHE measurement. (b) Second-order planar Hall voltage at 5 mA bias current with different transverse fields (c) Linear fitting of V H ( xy  0) y against yH (0, (cid:3397) 0.5 and (cid:3398) 0.5 Oe). V  diff   V H ( xy   0.5 ) Oe y   V H ( xy   0.5 ) Oe . (d) Linear dependence of effH as a function of current density. Error bars y indicate the accuracy of linear fitting. III. DISCUSSION After the confirmation of presence of an effective field along y-axis by both techniques, we investigate how it depends on the Fe0.8Mn0.2 thickness by varying d from 4 nm to 10 nm. As shown in Fig. 12 5(a), the results extracted from the two different techniques are fairly close to each other, with both scaling almost inversely with d, which implies that the SOT is of surface or interfacial nature. We have fabricated samples with even smaller thicknesses, but the sharp increase of resistance at small thickness does not allow to obtain reproducible results; therefore, we only focus on samples with a thickness larger than 4 nm. In our previous study on anomalous Hall effect induced magnetoresistance, i.e., AHMR, we found that the absolute value of AHMR scales as  length of FM, respectively,  is the polarization for longitudinal conductivity, and AH is the anomalous , where d and ls are the thickness and spin diffusion  2 2 l s d tanh  AH   d 2 l s  Hall angle [10]. If the SOT observed in this study is indeed from the backflow of reflected spins whose polarization is misaligned with the FM magnetization, the average SOT effective field should scale as tanh 1 2 d  d 2 l s  . The additional 1/d pre-factor comes from the fact that the torque generated is proportional to the total angular momentum transferred divided by the moment per unit area of the FM film. In Fig. 5(a), we plot the fitting results by using ls = 2-5 nm (shadowed region). The lower and upper boundary denoted the minimum and maximum values obtained by ls = 2 nm and ls = 5 nm. As can be seen, ls = 3.5 nm (green dotted line) fits the experimental results very well. This value is in good agreement with the value we obtained previously from AHMR studies. Now the question is what could be the underlying mechanism for the observed SOT and related effective field? According to Manchon and Zhang [37], when a charge current flows through a single FM layer, the non-equilibrium transverse spin induced by the SOC exerts a torque on the local magnetization, but this is expected to be a small higher-order effect for inversion-symmetry-preserved SOC such as the impurity SOC and Luttinger spin-orbit band, which is presumably the present case. Recently, it has been theoretically predicted by Pauyac et al. [38] that transversely polarized spin current can exist in ferromagnet due to spin-orbit interaction, which generates spin current to compete with spin dephasing. Similarly, based on the first principles calculations, Amin et al. proposed that the intrinsic spin 13 current is not subject to dephasing even though its spin polarization is misaligned with the magnetization [39], therefore, in addition to the AHE-induced spin current whose spin direction is along the magnetization direction, there is also an intrinsic spin current with the spin direction transverse to the magnetization in a single layer ferromagnet. Very recently, Wang et al. [40] experimentally observed the anomalous spin- orbit torques (ASOT) along y-direction in magnetic single-layer films (Py, Co, Ni and Fe), and attributed it to bulk SOI-generated transversely polarized spin current. Due to the exchange coupling in magnetic materials, the ASOT-induced magnetization tilting is observable when the FM is much thicker than the exchange length, whereas the ASOTs at top and bottom surfaces nearly cancel out in FM with thickness smaller than the exchange length. Although we don't completely rule out the relevance of bulk spin current and related torque, the thickness-dependence shown in Fig. 5(a) strongly suggests that that the SOT observed in this study is originated from the spin scattering and precession near the surface instead of bulk spin current. As discussed below, spin-dependent scattering and related spin rotation at the sample surface offers a more feasible explanation for the observed torque. FIG. 5. (a) Fe0.8Mn0.2 thickness dependence of SOT efficiency eff . Circles and squares are the values extracted from second- order PHE and nonlinear MR measurement, respectively. The shadowed area is the calculated eff range using ls: 2-5 nm, and dotted-line is the fitting results using ls = 3.5 nm. (b) Schematic of effective fields generated at the top and bottom surface of Fe0.8Mn0.2 due to spin rotation. 14 As mentioned in the introduction, the polarization of AHE-generated spins inside the FM is presumably along the magnetization direction due to strong exchange coupling. However, in the vicinity of the interface, polarization direction of scattered electrons may deviate from the magnetization direction, depending on whether the electrons are scattered towards the surface or the interior of the FM layer; the latter will have its polarization aligned with the magnetization after it travels by a distance comparable to the exchange length, whereas the former will be reflected from the surface with its polarization determined by the scattering geometry and strength of the SOC of the scattering center, as illustrated in Fig. 5(b). When the current is in x-direction and considering the geometry of the sample, we may only need to focus on scattering in the xz plane with its normal along   direction which is parallel to y-axis. Here, k ( n k k ' 'k ) is electron moving direction before (after) scattering. For an electron with initial polarization direction , upon scattering from an SOC scattering center, the polarization of scattered electrons may have the form '   Sn Tn ( ) n U n     (  )  , where S, T and U are constants determined by the scattering angle and  strength of SOC [41-44]. The spins polarized in the ' direction will accumulate at the surface and then diffuse towards the film side, thereby exerting spin torques on the magnetization of the FM layer. Similar to the transfer of spin angular momentum from a nonlocal source to a local magnetic moment in HM/FM heterostructures, the spin torque here has the form of ( ST m m    ) '   [45-48]. To facilitate discussion, we write  as   sign m ( )  , where sign    for down (up) spins. Then, the torque can be written ) 1( 1) ( as  ST [ S sign   ( )  ( ) Tm n m m n    ]  sign Um n ( )   . In the present case, since m is in the plane, the first terms is equivalent to an effective field along z-direction with its magnitude depending on the azimuthal angle of m . This term can be ignored since the contributions from top and bottom surfaces have an opposite sign and thus are mostly cancelled out. On the other hand, the second term corresponds to a magnetization- independent effective field in y-direction with its direction depending on the direction of both  and n . We now consider the case that an electron moving in x-direction is scattered by an SOC scattering center near the top surface of the FM layer. As shown in the schematic of Fig. 5(b), we assume that up-spin 15 electrons are scattered towards the surface side and down-spin electrons are scattered towards the film side, then n is in -- y direction for up-spin electrons and in +y direction for down-spin electrons. This will lead to an effective field scattering near both surfaces, then the direction of n will be the same for both up-spin and down-spin effH  in +y direction. If we assume that the scattering asymmetry remains the same for electrons at the bottom surface. Since at the bottom surface, down-spin electrons will be scattered towards the surface side, the resultant SOT effective field effH  is also in +y direction. Although the torque from down-spin electrons may not be as large as that of up-spin electrons, when they add up it will lead to a sizable net torque. It is worth pointing out that based on this model, the field is of damping-like nature, but the effect on magnetization is equivalent to a field-like effective field along y direction in HM/FM bilayers. And since the aforementioned process is similar to AHE in a toy model, the torque should be weak in samples with small AHE. In a previous work, we have demonstrated that the anomalous Hall angle of NiFe is around one order of magnitude smaller than that of Fe0.71Mn0.29 [10], which also corresponds to the size of anomalous Hall resistivity ( AH xy ) in transition metals, i.e., Fe >> Ni [49, 50]. This explains why the SOT is weak in NiFe control samples. The spin rotation is expected to be of surface nature and is present only within the spin dephasing length. More detailed theoretical studies are required to quantify the spin accumulation caused by the spin rotation as well as the resultant torque. IV. CONCLUSIONS In summary, spin torque with a strength comparable to the field-like SOT in Pt/NiFe bilayers has been observed in both asymmetric MgO/Fe0.8Mn0.2 and symmetric MgO/Fe0.8Mn0.2/MgO structures. The thickness dependence of current-to-torque conversion efficiency suggests that the torque is originated from surface or interfaces with the oxides. Although we don't rule out other possibilities, we argue that the torque is caused by spin rotation in the vicinity of the surface via impurity scattering in the presence of a strong spin-orbit coupling. Instead of cancelling off with each other, the torques from the top and bottom surfaces 16 simply add up, leading to the observed torque. We believe our work will stimulate more studies on spin current and spin torque in single layer ferromagnet with large spin orbit coupling. ACKNOWLEDGMENTS The authors would like to acknowledge support by Ministry of Education, Singapore under its Tier 2 Grants (grant no. MOE2017-T2-2-011 and MOE2018-T2-1-076), and stimulating discussions with Shufeng Zhang of The University of Arizona. 17 References: [1] N. Nagaosa, J. Sinova, S. Onoda, A. H. MacDonald, and N. P. Ong, Anomalous Hall effect, Rev. Mod. Phys. 82, 1539 (2010). [2] T. Taniguchi, J. Grollier, and M. D. Stiles, Spin-Transfer Torques Generated by the Anomalous Hall Effect and Anisotropic Magnetoresistance, Phys. Rev. Appl. 3, 044001 (2015). [3] A. Bose, D. D. Lam, S. Bhuktare, S. Dutta, H. Singh, Y. Jibiki, M. Goto, S. Miwa, and A. A. Tulapurkar, Observation of Anomalous Spin Torque Generated by a Ferromagnet, Phys. Rev. Appl. 9, 064026 (2018). [4] J. D. Gibbons, D. MacNeill, R. A. Buhrman, and D. C. Ralph, Reorientable Spin Direction for Spin Current Produced by the Anomalous Hall Effect, Phys. Rev. Appl. 9, 064033 (2018). [5] S. Iihama, T. Taniguchi, K. Yakushiji, A. Fukushima, Y. Shiota, S. Tsunegi, R. Hiramatsu, S. Yuasa, Y. Suzuki, and H. Kubota, Spin-transfer torque induced by the spin anomalous Hall effect, Nat. Electron. 1, 120 (2018). [6] K. S. Das, W. Y. Schoemaker, B. J. van Wees, and I. J. Vera-Marun, Spin injection and detection via the anomalous spin Hall effect of a ferromagnetic metal, Phys. Rev. B 96, 220408(R) (2017). [7] K. S. Das, J. Liu, B. J. van Wees, and I. J. Vera-Marun, Efficient injection and detection of out-of- plane spins via the anomalous spin Hall effect in permalloy nanowires, Nano Lett. 18, 5633 (2018). [8] C. Qin, S. Chen, Y. Cai, F. Kandaz, and Y. Ji, Nonlocal electrical detection of spin accumulation generated by anomalous Hall effect in mesoscopic Ni81Fe19 films, Phys. Rev. B 96, 134418 (2017). [9] R. Hao, M. Xu, T. Zhou, S. Kang, G. Liu, L. Bai, G. Han, S. Yu, and S. Yan, Detection of interfacial spin accumulation induced by anomalous Hall effect in ferromagnets, J. Magn. Magn. Mater. 477, 209 (2019). [10] Y. Yang, Z. Luo, H. Wu, Y. Xu, R. W. Li, S. J. Pennycook, S. Zhang, and Y. Wu, Anomalous Hall magnetoresistance in a ferromagnet, Nat. Commun. 9, 2255 (2018). [11] H. Nakayama, M. Althammer, Y.-T. Chen, K. Uchida, Y. Kajiwara, D. Kikuchi, T. Ohtani, S. Geprägs, M. Opel, S. Takahashi, R. Gross, G. E. W. Bauer, S. T. B. Goennenwein, and E. Saitoh, Spin Hall Magnetoresistance Induced by a Nonequilibrium Proximity Effect, Phys. Rev. Lett. 110, 206601 (2013). [12] Y.-T. Chen, S. Takahashi, H. Nakayama, M. Althammer, S. T. B. Goennenwein, E. Saitoh, and G. E. W. Bauer, Theory of spin Hall magnetoresistance, Phys. Rev. B 87, 144411 (2013). [13] M. Althammer, S. Meyer, H. Nakayama, M. Schreier, S. Altmannshofer, M. Weiler, H. Huebl, S. Geprägs, M. Opel, R. Gross, D. Meier, C. Klewe, T. Kuschel, J.-M. Schmalhorst, G. Reiss, L. Shen, A. Gupta, Y.-T. Chen, G. E. W. Bauer, E. Saitoh, and S. T. B. Goennenwein, Quantitative study of the spin Hall magnetoresistance in ferromagnetic insulator/normal metal hybrids, Phys. Rev. B 87, 224401 (2013). [14] C. Hahn, G. de Loubens, O. Klein, M. Viret, V. V. Naletov, and J. Ben Youssef, Comparative measurements of inverse spin Hall effects and magnetoresistance in YIG/Pt and YIG/Ta, Phys. Rev. B 87, 174417 (2013). [15] N. Vlietstra, J. Shan, V. Castel, B. J. van Wees, and J. Ben Youssef, Spin-Hall magnetoresistance in platinum on yttrium iron garnet: Dependence on platinum thickness and in-plane/out-of-plane magnetization, Phys. Rev. B 87, 184421 (2013). [16] S. R. Marmion, M. Ali, M. McLaren, D. A. Williams, and B. J. Hickey, Temperature dependence of spin Hall magnetoresistance in thin YIG/Pt films, Phys. Rev. B 89, 220404(R) (2014). [17] Y. Omori, E. Sagasta, Y. Niimi, M. Gradhand, L-E. Hueso, F. Casanova and Y. Otani, Relation between spin Hall effect and anomalous Hall effect in 3d ferromagnetic metals, Phys. Rev. B 99, 014403 (2019). 18 [18] A. M. Humphries, T. Wang, E. R. J. Edwards, S. R. Allen, J. M. Shaw, H. T. Nembach, J. Q. Xiao, T. J. Silva, and X. Fan, Observation of spin-orbit effects with spin rotation symmetry, Nat. Commun. 8, 911 (2017). [19] S-h. C. Baek, V. P. Amin, Y.W. Oh, G. Go, S. J. Lee, G. H. Lee, K. J. Kim, M. D. Stiles , B. G. Park, and K. J. Lee , Spin currents and spin -- orbit torques in ferromagnetic trilayers, Nat. Mater. 17, 509 (2018). [20] Y. Xu, Y. Yang, Z. Luo, B. Xu, and Y. Wu, Macro-spin modeling and experimental study of spin- orbit torque biased magnetic sensors, J. Appl. Phys. 122 ,193904 (2017). [21] T. Nan, S. Emori, C. T. Boone, X. Wang, T. M. Oxholm, J. G. Jones, B. M. Howe, G. J. Brown, and N. X. Sun, Comparison of spin-orbit torques and spin pumping across NiFe/Pt and NiFe/Cu/Pt interfaces, Phys. Rev. B 91, 214416 (2015). [22] Y. Xu, Y. Yang, M. Zhang, Z. Luo, and Y. Wu, Ultrathin All-in-One Spin Hall Magnetic Sensor with Built-In AC Excitation Enabled by Spin Current, Adv. Mater. Technol. 3, 1800073 (2018). [23] C. O. Avci, K. Garello, M. Gabureac, A. Ghosh, A. Fuhrer, S. F. Alvarado, and P. Gambardella, Interplay of spin-orbit torque and thermoelectric effects in ferromagnet/normal-metal bilayers, Phys. Rev. B 90, 224427 (2014). [24] C. O. Avci, K. Garello, A. Ghosh, M. Gabureac, S. F. Alvarado, and P. Gambardella, Unidirectional spin Hall magnetoresistance in ferromagnet/normal metal bilayers, Nat. Phys. 11, 570 (2015). [25] A. Ghosh, K. Garello, C. O. Avci, M. Gabureac, and P. Gambardella, Interface-Enhanced Spin-Orbit Torques and Current-Induced Magnetization Switching of Pd/Co/AlOx Layers, Phys. Rev. Appl. 7, 014004 (2017). [26] Z. Luo, Y. Xu, Y. Yang, and Y. Wu, Magnetic angular position sensor enabled by spin-orbit torque, Appl. Phys. Lett. 112, 262405 (2018). [27] M. Hayashi, J. Kim, M. Yamanouchi, and H. Ohno, Quantitative characterization of the spin-orbit torque using harmonic Hall voltage measurements, Phys. Rev. B 89, 144425 (2014). [28] C. O. Avci, J. Mendil, G. S. D. Beach, and P. Gambardella, Origins of the unidirectional spin Hall magnetoresistance in metallic bilayers, Phys. Rev. Lett. 121, 087207 (2018). [29] V. M. Edelstein, Spin polarization of conduction electrons induced by electric current in two- dimensional asymmetric electron system, Solid State Commun. 73, 233 (1990). [30] I. M. Miron, G. Gaudin, S. Auffret, B. Rodmacq, A. Schuhl, S. Pizzini, J. Vogel, and P. Gambardella, Current-driven spin torque induced by the Rashba effect in a ferromagnetic metal layer, Nat. Mater. 9, 230 (2010). [31] A. Manchon, H. C. Koo, J. Nitta, S. M. Frolov and R. A. Duine, New perspectives for Rashba spin- [32] Y. Xu, Y. Yang, K. Yao, B. Xu, and Y. Wu, Self-current induced spin-orbit torque in FeMn/Pt orbit coupling, Nat. Mater. 14, 871 (2015). multilayers, Sci. Rep. 6, 26180 (2016). [33] X. Fan, J. Wu, Y. Chen, M. J. Jerry, H. Zhang, and J. Q. Xiao, Observation of the nonlocal spin- orbital effective field, Nat. Commun. 4, 1799 (2013). [34] X. Fan, H. Celik, J. Wu, C. Ni, K. J. Lee, V. O. Lorenz, and J. Q. Xiao, Quantifying interface and bulk contributions to spin -- orbit torque in magnetic bilayers, Nat. Commun. 5, 3042 (2014). [35]Y. Yang, Y. Xu, X. Zhang, Y. Wang, S. Zhang, R.-W. Li, M. S. Mireshekarloo, K. Yao, and Y. Wu, Phys. Rev. B 93, 094402 (2016). [36] S. Emori, T. Nan, A. M. Belkessam, X. Wang, A. D. Matyushov, C. J. Babroski, Y. Gao, H. Lin, and N. X. Sun, Interfacial spin-orbit torque without bulk spin-orbit coupling, Phys. Rev. B 93, 180402(R) (2016). [37] A. Manchon and S. Zhang, Theory of nonequilibrium intrinsic spin torque in a single nanomagnet, Phys. Rev. B 78, 212405 (2008). [38] C. O. Pauyac, M. Chshiev, A. Manchon, and S. A. Nikolaev, Spin Hall and spin swapping torques in diffusive ferromagnets, Phys. Rev. Lett. 120, 176802 (2018). 19 [39] V. P. Amin, J. Li, M. D. Stiles, P. M. Haney, Intrinsic spin currents in ferromagnets, arXiv: 1901.04022v2. [40] W. Wang, T. Wang, V. P. Amin, Y. Wang, A. Radhakrishnan, A. Davidson, S. R. Allen, T. J. Silva, H. Ohldag, D. Balzar, B. L. Zink, P. M. Haney, J. Q. Xiao, D. G. Cahill, V. O. Lorenz, X. Fan, Anomalous spin-orbit torques in magnetic single-layer films, arXiv:1902.05490v1. [41] V. P. Amin, J. Zemen, and M. D. Stiles, Interface-Generated spin currents, Phys. Rev. Lett. 121, [42] V. P. Amin and M. D. Stiles, Spin transport at interfaces with spin-orbit coupling: Formalism, Phys. 136805 (2018). Rev. B 94, 104419 (2016). Phys. Rev. B 94, 104420 (2016). [43] V. P. Amin and M. D. Stiles, Spin transport at interfaces with spin-orbit coupling: Phenomenology, [44] J. Kessler, Polarized electrons (Springer Science & Business Media, 2013), Vol. 1, p. 57. [45] J. C. Slonczewski, Conductance and exchange coupling of two ferromagnets separated by a tunneling barrier, Phys. Rev. B 39, 6995 (1989). [46] J. Slonczewski, Current-driven excitation of magnetic multilayers, J. Magn. Magn. Mater. 159, L1 (1996). 9353 (1996). [47] L. Berger, Emission of spin waves by a magnetic multilayer traversed by a current, Phys. Rev. B 54, [48] M. D. Stiles and A. Zangwill, Anatomy of spin-transfer torque, Phys. Rev. B 66, 014407 (2002). [49] Y. Tian, L. Ye and X. Jin, Proper scaling of the anomalous Hall effect. Phys. Rev. Lett. 103, 087206 (2009). [50] L. Ye, Y. Tian, X. Jin, and D. Xiao, Temperature dependence of the intrinsic anomalous Hall effect in nickel. Phys. Rev. B 85, 220403(R) (2012). 20
1202.0393	2	1202	2012-05-10T07:49:24	Influence of two-level fluctuators on adiabatic passage techniques	[ "cond-mat.mes-hall", "quant-ph" ]	We study the process of Stimulated Raman Adiabatic Passage (STIRAP) under the influence of a non-trivial solid-state environment, particularly the effect of two-level fluctuators (TLFs) as they are frequently present in solid-state devices. When the amplitudes of the driving-pulses used in STIRAP are in resonance with the level spacing of the fluctuators the quality of the protocol, i.e., the transferred population decreases sharply. In general the effect can not be reduced by speeding up the STIRAP process. We also discuss the effect of a structured noise environment on the process of Coherent Tunneling by Adiabatic Passage (CTAP). The effect of a weakly structured environment or TLFs with short coherence times on STIRAP and CTAP can be described by the Bloch-Redfield theory. For a strongly structured environment a higher-dimensional approach must be used, where the TLFs are treated as part of the system.	cond-mat.mes-hall	cond-mat	Influence of two-level fluctuators on adiabatic passage techniques Nicolas Vogt,1, 2, 3 Jared H. Cole,4, 2 Michael Marthaler,2, 3 and Gerd Schon2, 3 1Institut fur Theorie der Kondensierten Materie, Karlsruhe Institute of Technology, D-76128 Karlsruhe, Germany 2Institut fur Theoretische Festkorperphysik, Karlsruhe Institute of Technology, D-76128 Karlsruhe, Germany 3DFG-Center for Functional Nanostructures (CFN), Karlsruhe Institute of Technology, D-76128 Karlsruhe, Germany 4Chemical and Quantum Physics, School of Applied Sciences, RMIT University, Melbourne, Australia (Dated: February 2, 2012) We study the process of Stimulated Raman Adiabatic Passage (STIRAP) under the influence of a non-trivial solid-state environment, particularly the effect of two-level fluctuators (TLFs) as they are frequently present in solid-state devices. When the amplitudes of the driving-pulses used in STIRAP are in resonance with the level spacing of the fluctuators the quality of the protocol, i.e., the transferred population decreases sharply. In general the effect can not be reduced by speeding up the STIRAP process. We also discuss the effect of a structured noise environment on the process of Coherent Tunneling by Adiabatic Passage (CTAP). The effect of a weakly structured environment or TLFs with short coherence times on STIRAP and CTAP can be described by the Bloch-Redfield theory. For a strongly structured environment a higher-dimensional approach must be used, where the TLFs are treated as part of the system. PACS numbers: 74.50.+r, 03.65.Yz, 32.80.Qk, 03.67.Lx Keywords: superconducting three-state system, STIRAP, decoherence, Josephson junctions, two-level fluc- tuators I. INTRODUCTION Much of the recent excitement in low-temperature device research, for example the growing field of cir- cuit quantum electrodynamics 1 -- 5, centers around map- ping quantum optics concepts to solid-state systems. An example to be discussed in the present work is Stimulated Raman Adiabatic Passage (STIRAP), a quantum control scheme to transfer population between two levels in an adiabatic rotation via a third level. It has been widely studied in the field of quantum optics, its advantage ly- ing in the robustness against external perturbations and imperfections in the control pulses6. STIRAP, therefore, has also been proposed for achieving single-qubit rota- tions in quantum information experiments7. Specific pro- tocols were discussed for superconducting qubits, charge qubits8, phase qubits9 and the quantronium10,11. Ex- periments on electromagnetically induced transparency and the formation of a dark state, both effects closely related to STIRAP, have also been performed 12 -- 14. A related scheme called Coherent Tunnelling by Adiabatic Passage (CTAP) exists for three neighbouring quantum dots15 where the tunnel matrix elements themselves are modulated to achieve an adiabatic transition. The description of decoherence, both in quantum op- tics and circuit QED systems, is frequently based on a master equation approach as described by the Lindblad form. This formulation is sufficient for the structure- less environment which the vacuum provides in quan- tum optics, however, the solid-state environment can be much more complicated. In addition to the effects of the electric circuit, typically dominated by an Ohmic re- sponse, experiments displayed strong signatures of low- frequency or 1/f noise16. This background noise may be described by a bath of independent two-level-fluctuators (TLFs)17,18, however, recent work on qubits demon- strated the coupling to individual TLFs19. Qubits have be used to detect and manipulate them20,21, yet their true microscopic nature is still unknown21. The pres- ence of TLFs in a solid-state environment requires more complex models than are possible with the conventional Lindblad treatment. In this paper we study how decoherence effects from a structured environment influence the efficiency, i.e., the transferred population, in the STIRAP process. When analysing these problems we compare different theoret- ical approaches. We start with a short review what is known about the effect of noise on the STIRAP protocol as long as it is treated by a master equation with Lind- blad damping terms. Next we introduce the model with TLFs and show how a weakly structured noise environ- ment can be treated by the Bloch-Redfield equation for the three-level system. For a strongly structured noise environment, e.g., due to TLFs with long life times, the analysis of a higher-dimensional system formed by the three-level system and the TLFs is needed. We arrive at the following conclusions: As long as the dissipation arises due to wide-band noise its effect can be reduced by using strong driving pulses. The situation changes for a structured bath, e.g., due to the presence of TLFs. If TLFs couple to the three-level system the trans- ferred population decreases sharply when the amplitude of the driving-pulses used in STIRAP coincides with the level spacing of the fluctuator. In general this effect can not be reduced by speeding up the STIRAP process. For an efficient STIRAP protocol the driving pulse amplitude must be chosen sufficiently low or, as long as the rotat- ing wave approximation remains valid, high compared to the TLF frequency. In a final section we study how the transferred population in a CTAP process is affected by a TLF coupling to the barrier-tunnelling amplitudes. II. STIRAP WITH LINDBLAD DAMPING In this section we discuss the effects of noise on the STIRAP process in the frame of a master equation with Lindblad damping terms. As long as this description is valid, it turns out to be advantageous to use strong driving pulses for the STIRAP protocol. In subsequent sections we will generalize the analysis to the case where the environmental bath is structured and the Lindblad theory is no longer sufficient. Usually STIRAP is performed in a Λ-system consist- ing of two low-energy levels, 0i and 1i, and one at high- energy, 2i. The energy of 0i is set to zero and those of 1i and 2i are ∆ and ǫ, respectively. Ideally the energy gap ∆ is small compared to ǫ. In the STIRAP process two coupling-pulses, modulated with amplitudes Ω0(t) and Ω1(t), are applied in resonance with the energy difference between level 2i and the two low-lying levels 0i and 1i, respectively. In order to transfer population from the state 0i to 1i, the two pulses are applied in counter-intuitive order, namely first the pulse Ω1(t) and then Ω0(t). A sketch of the three-level-system with the coupling-pulses is shown in Figure 1. is known to be The STIRAP protocol robust against detuning of the coupling-pulses from the 0i-2i- transition and the 1i-2i-transition, as long as there is two-photon-resonance and both coupling-pulses are de- tuned by the same frequency ∆tp. Here for simplicity we only consider full resonance since the main effect we are interested in, the effect of a two-level-fluctuator is not qualitatively altered in the case of two-photon-resonance as will be explain below. Below we assume the coupling-pulse amplitudes Ω0(t) and Ω1(t) to have Gaussian profiles. Both pulses are applied on the time scale T . The coupling-pulse Ω0(t) is applied after the pulse Ω1(t) with a delay of τ . The whole STIRAP process is assumed to take place in the time span 0 ≤ t ≤ Tmax. Specifically, we will assume for the numerical simulations22 Ω0/1(t) = Ωmax exp(−(cid:2)(cid:0)t − Tmax 4(cid:3)2 2 (cid:1) ∓ T T 2 ) (1) with cut-offs for t < 0 and t > Tmax, and we choose √2 10 Tmax. The pulse sequence is shown in Figure 1. T = In the interaction picture, after the rotating wave ap- (a) E (b) E (c) ] 0 ω [ E 2 2 2 + + 1 1 TLF ensemble + Bath coupling to the TLFs + Bath coupling to the system Spectral function + Bath coupling to the system 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 t [T ] max 0 0 1 0.8 0.6 0.4 0.2 0 FIG. 1. (a) and (b): The STIRAP process involves steering the population within a three-state system (states 0, 1 and 2) using applied laser (or microwave) driving fields with a time dependent Rabi frequency Ω0(t) and Ω1(t). In a solid state environment, the three-level system can couple to non-trivial environmental modes stemming from two-level solid-state de- fects. This effect can be modelled in two alternative ways (illustrated conceptually here, via explicitly treating the co- herent defects in the environment (a) or via an effective spec- tral model using Bloch-Redfield theory (b). While the former is 'exact', it is computationally expensive. The later scales better but is only applicable in certain parameter regimes. (c): In order to realize the STIRAP protocol, the effective Rabi frequency of the driving pulses must be varied as a func- tion of time, plotted here for the counter-intuitive pulse se- quence. proximation (RWA) the Hamiltonian reduces to HS(t) =  0 0 0 0 Ω0(t) Ω1(t) Ω0(t) Ω1(t) 0   . (2) Its eigenstates are traditionally called bright states (+i, −i) and dark state dsi. The latter is a superposition of the states 0i and 1i, dsi = cos Θ 0i + sin Θ 1i, (3) with angle Θ = arctan (Ω1(t)/Ω0(t)), which reduces to Θ = 0 for Ω0 = 0 but Ω1 6= 0 and to Θ = π/2 for Ω1 = 0 but Ω0 6= 0, i.e., for the counter-intuitive order used in STIRAP the angle Θ changes from 0 to π/2 and, as long as the adiabaticity condition23 is met, all population originally prepared in state 0i is transferred to state 1i. The effect of dissipation on the STIRAP process has been modelled in the past by a Lindblad form for the master equation8,9, ρ = −i[H, ρ] + L(ρ) L(ρ) =Xα Γα[LαρL†α − 1 2 {L†αLα, ρ}] , (4) (5) where {Lα} is an orthonormal set of operators and the Γα denote the relaxation and dephasing rates24. Exten- sions to account for the effects of driving on these rates were discussed in Ref. 25. In addition, solid state systems frequently are coupled to a structured noise environment due to the presence of strongly coupling two-level fluc- tuators. In general their effect can not be modelled by the Lindblad equation acting on the three-level-system alone. In the STIRAP process the adiabatic rotation depends on the phase coherence between the states 0i and 1i. Therefore even pure dephasing causes population to de- cay from the dark state to the bright states, and as a result it is not transferred to the final state. Although a background bath can also lead to direct relaxation pro- cesses we consider in this paper a background that only causes dephasing. As long as the destructive effect of background dephasing can be described by a constant rate, it can be reduced by performing the STIRAP pro- cess faster. It becomes negligible if Tmax ≪ 1/Γα for all dephasing rates in the Lindblad equation. The adia- baticity condition imposes a further constraint between the strength of the coupling-pulses Ωmax and the dura- tion of the process22, Ωmax · Tmax ≫ 1. Hence the quality of the STIRAP in the presence of background dephas- ing effects can be improved by reducing the process time Tmax and simultaneously increasing Ωmax. The connection between the parameters of Gaussian coupling-pulses and the success rate of STIRAP in the case of pure dephasing has been studied by Ivanov et al.22 and Yatsenko et al.26. They showed that, as long as the adiabaticity condition is satisfied and the RWA is valid, the population lost from STIRAP depends exponentially on the length T of the pulses and the timescale is set by the dephasing rates, i.e., the transferred population is maximised by decreasing T . They further found for each given coupling-pulse amplitude a lower bound for T 22. Below this bound the adiabaticity condition is no longer fulfilled and the population transfer breaks down. The lower bound decreases with increasing strength of the driving pulses. In conclusion, as long as the noise can be described by the Lindblad equation, due to the flexibility of the driving pulses in STIRAP, it is possible to achieve a high-quality STIRAP process even in the presence of dephasing. For this reason one should choose the laser pulses as strong as possible while staying in the regime of the RWA. We will see that this is no longer the case for a structured environment. 3 III. STRUCTURED BATH A. Model with TLF For solid-sate quantum systems the noise environment frequently is structured due to the presence of two-level fluctuators (TLFs)20,21,27,28. In these cases, the Lindblad form described above is not sufficient. To proceed we can treat the TLF explicitly as a part of the larger quan- tum system to be studied. Alternatively, we can derive from the appropriate system-bath model the correspond- ing Bloch-Redfield equation29. This approach has been used to study pure dephasing by Shi et al.30. In what follows we will compare the two approaches. Environmental TLFs are regularly observed in exper- iments on low-temperature electrical circuits, although their precise microscopic origin is not yet clear. In the context of superconducting qubits, several models for TLFS have been proposed which all fit the experimen- tal data to a greater or lesser extent. Various models are summarised in Ref.21 for the case of a phase-qubit, including bistable lattice defects, charge traps and spin impurities. A possible source of TLFs which could be present in all superconducting qubits are two-level tun- neling systems19, bistable lattice defects in the Joseph- son junctions of the qubit. In this work we use a gen- eral model of TLFs as a quantum two level systems that can couple transversally and longitudinally to the three- level-system of interest. This model has been compared directly to experimental observations21,31 and therefore we need not consider the exact microscopic nature of the TLF. We consider a quantum three-level system coupled to a two-level fluctuator, and both coupled to independent baths B1 and B2, respectively. We do not assume a specific form for the baths B1 and B2 but assume that they are generic background baths that lead to constant relaxation and dephasing rates on the three-level-system and the TLF as is described in more detail below. The Hamiltonian of the three-level-system, the TLF and their coupling in the presence of generic background baths is given by H(t) = HS(t) + Hint(t) + HT LF + HB1 + HB2 HT LF = − ω0 2 τz . (6) The Hamiltonian of the driven three-level system HS(t) has been introduced above, eqn.(2). The TLF is de- scribed by HT LF and it is assumed to couple transver- sally to the 0i ↔ 1i transition, i.e., in the interaction picture, where the driving terms acquire phase factors e±i∆t, within the rotating wave approximation the cou- pling term is given by Hint = g  0 τ +ei∆t 0 τ−e−i∆t 0 0 0 0 0  . (7) To simplify the notation we define new coupling- operators time-consuming we use another form for the equation of motion, τ±(t) = τ±e±i∆t . (8) ρ = −i [HS(t), ρ] + Lback(ρ) + LTLF(ρ) , (12) 4 The background baths and couplings to the three-level system and the TLF, described by HB1 and HB2, cause dissipation which we assume is unstructured. Accord- ingly they can be characterized by Lindblad terms with relaxation and dephasing rates. In the following, for sim- plicity we ignore pure dephasing and assume low temper- atures, i.e., we take only the relaxation from the fluctu- ator state ↑i to ↓i with rate Γ↓ into account. Fur- thermore we assume that the direct coupling between the background-bath HB1 and the three-level-system is so weak that it can be ignored. Even if this was not the case, the coupling to HB1 would only decrease the effectiveness of STIRAP and not alter the qualitative be- haviour that arises from the coupling to the TLF. The equation of motion of the density matrix ρ thus reduces to a 6-dimensional system, which can be solved numerically. Results will be presented below and com- pared to other approaches. The analysis can be extended to account for more than one TLF, however, with grow- ing number the numerical analysis quickly becomes in- tractable. Other approaches, such as the Bloch-Redfield approach, provide an alternative. B. Bloch-Redfield Approach In the Bloch-Redfield approach the degrees of free- dom of the baths, in the present case including those of the TLF, are traced out of the equation of motion for the density matrix of the system29. The characteris- tics of this bath are contained in the spectral function, i.e., the Fourier-transformed correlation functions of the coupling-operators, Cτ − τ +(ω) =Z ∞ −∞ dt eiωthτ−(t)τ +(0)i . (9) The relevant frequencies are the differences between the time-dependent eigenvalues of HS(t), ωij(t) = Ej(t) − Ei(t) . (10) As usual the master equation for the time evolution of the matrix elements of the density matrix can be written in the eigenbasis of the system Hamiltonian, ρmn − i(Em − En)ρmn = Xn′m′ Rnmn′m′ ρn′m′ , (11) with relaxation and dephasing described by the ten- sor Rnmn′m′. However, for the driven system consid- ered here, the eigenbasis, i.e., the energies Em(t) and Rnmn′m′(t) also depend on time. Since the construction of the tensor Rnmn′m′ (t) at ev- ery timestep of a numerical simulation would be very with the Lindblad term Lback(ρ) responsible for the ef- fects of the background bath given in equation eq.(5) where in the case of pure relaxation considered here the prefactor of only one Lα is nonzero, Γ↓ with L↓ = τ− . The effect of the TLF is contained in 1 LTLF(ρ) = − 10(t) ρ 1ih0 10(t) ρ − α+ 4(cid:2)1ih0 α+ +ρ α−01(t) 0ih1 − 0ih1 ρ α−01(t)(cid:3) α±kl(t) = V (t)(cid:2)ξkl(t) ◦ Cτ − τ + (±ω(t))(cid:3) V −1(t) ξkl(t) = V −1(t)lihkV (t) . (13) where ◦ denotes element-wise matrix multiplication ([A ◦ B]ij = Aij · Bij), Cτ − τ + (ω(t)) is a matrix defined by, [Cτ − τ + (ω(t))]ij = Cτ − τ + (ωij(t)) , (14) and V (t) diagonalizes the Hamiltonian, V −1(t)HS(t)V (t) =qΩ2 0(t) + Ω2 . 1(t)  −1 0 0 0 0 0 0 0 1  (15) 2 ∆2 tp + Ω2 0(t) + Ω2 shifted by ∆tp need to make the substitution pΩ2 q 1 If we were to allow two-photon-resonance, all eigenen- ergies would be and we would 0(t) + Ω2 1(t) → 1(t). Similarly, the form of the cou- pling operators changes in the eigenbasis of the Hamilto- nian. This shows that two-photon-resonance would only change the frequency range of the spectral function which is scanned by the system during the STIRAP process. If TLFs are present in both frequency ranges the results are only changed quantitatively and not qualitatively. The spectral functions of the coupling-operators of a single TLF can be obtained from the quantum regres- sion theorem29,32, and in the interaction picture takes the Lorentzian shape Cτ − τ +(ω) = 8g2Γ↓ (Γ↓)2 + 4(ω − (ω0 − ∆))2 . (16) The width of the peak in the spectral function is given by the rate Γ↓. If several TLFs are present that do not couple to each other, the spectral function of the ensem- ble of TLFs is given by the sum of the spectral functions of the individual TLFs. Therefore we start by consider- ing only one TLF coupling to the three-level system and discuss the case of several TLFs later. As the time scale set by the rate Γ↓ for a certain TLF becomes equal to the timescale of the coherent time evolution of the system, the Markov approximation can no longer be applied and the TLF must be treated as part of the system. Such a non-Markovian behaviour has been studied by Kamleit- ner et al.33 in CTAP systems. We now compare the time evolution of the system during the STIRAP process obtained from the Bloch- Redfield approach with that obtained from the exact re- sult of a numerical simulation of the six-dimensional sys- tem, formed by the three level system and the two-level fluctuator. The equation of motion of the density matrix of the six-dimensional problem is given by, ρ6 = −i [H6, ρ6] + Γ↓[L↓,6 ρ6 L† ↓,6 − 1 2 {L† ↓,6 L↓,6, ρ6}] . (17) The Hamiltonian H6 contains the three-level system, the TLF and their coupling, 0 g − ω0 0 2 0 + ω0 2 g −∆ − ω0 0 2 0 0 0 0 0 0 0 0 0 H6 =   0 0 0 0 0 −∆ + ω0 2 0 0 0 0 0 ǫ − ω0 2 0 0 0 0 0  (18) ǫ + ω0 2 The six-dimensional operator in the Lindblad part of the equation of motion is L↓,6 = I3 ⊗ τ− = 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0     . (19) IV. RESULTS AND COMPARISON During the STIRAP process the Hamiltonian HS(t) changes adiabatically, and so do the system eigenvalues. Consequently, also the strength of decoherence changes, since it is determined by the spectral function evaluated at the differences of the eigenvalues, ω12(t) =qΩ2 0(t) + Ω2 1(t) . (20) The decoherence strength thus depends on the frequency interval covered by the applied pulses. It becomes strong, when the energy difference eq.(20) is brought into res- onance with the TLF's frequency ω0. The scenario is illustrated in Figure 2 for varying strength of Ωmax. If the amplitudes are smaller than ω0, the TLF remains off-resonance during the whole STIRAP process, and the decoherence it introduces is weak. When the maximal difference between the eigenvalues of HS(t) is resonant with the TLF, the decoherence reaches its highest value. If the coupling-pulse amplitude is increased further, the ] 0 ω − [ ) t ( E − ) t ( s d E 2 1.8 1.6 1.4 1.2 1 0.8 0.6 0.4 0.2 0 Ω Ω Ω = 0.7 [ω ] 0 = 1 [ω ] 0 = 1.3 [ω ] 0 max max max 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 t [T ] max 5 C(ω) FIG. 2. The time-dependent difference of the eigenvalues of HS(t) during STIRAP compared to the spectral function of a TLF with resonance frequency ω0, shown in the right-hand panel. With increasing maximum pulse strength a resonance is reached. 1 0.9 0.8 0.7 0.6 n o i t l a u p o p e t a t s k r a d 0.5 0 1 0.5 t [Tmax] Ω max = 0.7 ω Ω max = 1 ω Ω max = 1.3 ω 0 0 0 FIG. 3. The evolution of the population in the dark state during a STIRAP process, as obtained from the Bloch- Redfield approach, is plotted for three different pulse ampli- tudes. For weak pulses the decoherence is weak during the whole time, and most of the populations stays in the dark state. For Ωmax = ω0 the system gets in resonance with the TLF during the STIRAP sequence, and the population in the dark state decreases sharply. If the coupling-pulse amplitude is increased further the sharp loss of population starts earlier but lasts for a shorter time, so that a higher fraction of the population remains in the dark state. resonance no longer occurs at the maximal difference of the eigenvalues, the system spends less time in resonance with the noise source, and the overall decoherence of the whole process is reduced. The resulting time evolutions of the population of the dark state for three regimes of pulse amplitudes are illustrated in Figure 3. In the case shown in Figure 4, the coupling between 1 0.8 l n o i t a u p o p 0.6 0.4 0.2 0 0 0.5 t [Tmax] 0 Ω max = 0.7 ω Ω max = ω Ω max= 1.3 ω 0 0 1 FIG. 4. The populations of the time-dependent eigenstates, obtained from the six-dimensional simulation, are plotted for weak damping of the TLF (Γ↓ = 0.0007ω0). At the beginning of the STIRAP sequence when the TLF and the system are off resonance we observe only small oscillations in the population of the states. Once system and TLF are in resonance the population oscillates coherently between the dark and lower lying bright state. The oscillations stop when the system and TLF are again off resonance. BR, slow TLF−relaxation 6D, slow TLF−relaxation semi−classical, slow TLF−relaxation BR, fast TLF−relaxation 6D, fast TLF−relaxation 1 0.8 0.6 0.4 0.2 n o i t l a u p o p d e r r e f s n a r t 0 0.7 1.3 pulse strength Ω 1 max [ω 0] 6 the two states when the system is in resonance with the TLF. As the dephasing rate Γ↓ is small compared to the inverse timescale of the coherent evolution of the system the TLF has to be treated as part of the system. The ratio between the population in the two states at the end of the sequence depends predominantly on the coupling strength between system and TLF and the time while system and TLF are in resonance. Figure 5 shows how the transferred population de- creases as one approaches the resonance, and how it re- covers as Ωmax is further increased. In this figure we also compare results obtained from an exact treatment of the TLF and those from the Bloch-Redfield approach. They deviate for weak damping, as the effects of the TLF on the system can no longer be described by a Marko- vian part in the equation of motion containing only the spectral function of the TLF. They coincide for stronger damping when the timescale of relaxation of the TLF is smaller than the timescale of the coherent evolution of the TLF, and the Markov approximation in the deriva- tion of the Bloch-Redfield equation is justified. In the strong damping limit, the effect on the trans- ferred population during STIRAP can be described with a semi-classical approximation. We denote the proba- bility of the system to be in the dark state with pds(t). The transferred population is given by pds(Tmax). We as- sume that the interaction with the TLF will only lead to relaxation from states with high energy to lower energy states, since Γ↑ = 0. In the adiabatic basis, we obtain a time-dependent relaxation rate Γds(t) from the dark state to the bright state −i which yields the differential equation pds(t) = Γds(t)pds(t). The rate Γds(t) can be approximated by Γds(t) = g2Cτ − τ + [E1(t) − E2(t)] . (21) The time evolution of pds(t) is given by pds(t) = e−A(t) A(t) = g2Z t 0 Γ2 ↓ Γ↓ 0(t′) + Ω2 + 4(cid:16)pΩ2 1(t′) − ω0(cid:17)2 dt′ . (22) FIG. 5. The transferred population during a STIRAP pro- cess is plotted as a function of Ωmax. The results of the Bloch- Redfield approach for small Γ↓ (Γ↓ = 0.067ω0) and large Γ↓ (Γ↓ = 0.33ω0) are plotted with a dashed line and a solid line, the result of an exact treatment of the TLF for small Γ↓ (Γ↓ = 0.067ω0) is drawn with crosses, and the one for large Γ↓ (Γ↓ = 0.33ω0) with circles. The result of the semi- classical approach for small Γ↓ (Γ↓ = 0.067ω0) is drawn with a diamond shapes. TLF and background bath is so weak that the total sys- tem, formed by the three-level-system and the TLF, have a high degree of coherence. In this case the population does not decay from the dark state to the lower lying bright state. Instead we see coherent oscillations between This simplified model reproduces the result of the Bloch- Redfield approach qualitatively as seen in Figure 5. The predicted dips in the transferred population could be detected by measuring the transferred population in a series of STIRAP experiments with different pulse ampli- tudes, provided that the dips corresponding to individual TLFs are distinguishable. However, for the parameters of TLFs and qubits as observed in experiments16,20,21,34 one obtains broad dips. In addition, we only obtain sharp dips in a parameter regime where the Bloch-Redfield ap- proach no longer quantitatively reproduces the result of an exact treatment of the two-level fluctuator. If the system couples to several TLFs with different frequen- cies the dips in the transferred population merge and the transferred population just decreases with increasing 1 0.8 0.6 0.4 ] g [ ) ω C ( 0.2 l n o i t a u p o p d e r r e f s n a r t spectral functions 2 0 0 0.7 1 frequency ω [ω 0] 1 pulse strength [ω 0] Ensemble 1 TLF 1.3 FIG. 6. Transferred population as a function of the coupling- pulse strength for two different spectral functions. One Lorentzian spectral function corresponding to a single TLF, the second spectral function models the effect of several TLFs. The transferred population recovers from a dip with growing pulse amplitudes if the spectral function decays to zero for higher frequencies, as it is the case for a Lorentzian. In the case of the spectral function that is a superposition of the Lorentzians of several TLFs, little population is transferred once the coupling strength has reached the threshold of the lowest TLF resonance frequency. coupling-pulse strength. Figure 6 shows the transferred population for such a dense distribution of TLFs. This limits the possibility of improving the STIRAP process in the presence of background dephasing by reducing the process time and increasing Ωmax. For an efficient implementation of STIRAP in a sys- tem with a small number of TLFs one has to avoid using pulse amplitudes in the vicinity of their frequencies. The intervals to avoid depend on the exact form of the spec- tral function and the coupling-pulses. However, as a rule of thumb one should avoid a frequency region of two to three times the width Γ of the Lorentzian peak in the spectral function. V. CTAP IN THE PRESENCE OF TLFS In a system consisting of three quantum dots a transfer protocol similar to STIRAP, called coherent tunnelling by adiabatic passage (CTAP), has been proposed15,33. The system consists of three potential wells arranged in a line and separated by controllable potential barriers. Varying the barrier tunneling rates in an analogous way to STIRAP allows an electron to be adiabatically steered from the first to last well without populating the central well. The position of the electron in the three quantum dots corresponds to the three basis states 0i, 2i, and 1i (arranged in this order for similarity with the STI- RAP notation). An illustration of the CTAP system is shown in Figure 7. The tunneling-matrix-elements be- 7 0 TLF 2 1 FIG. 7. The three states used in the coherent tunnelling by adiabatic passage (CTAP) correspond to an electron located in one of three neighbouring quantum dots. The TLF is as- sumed to couple to the time dependent tunneling amplitude which corresponds to the coupling-pulse in STIRAP. tween the quantum dots can be tuned by varying the height of the energy barrier between the quantum dots by applying a gate voltage. We use Gaussian modula- tions of the tunnelling-matrix-elements which correspond to the microwave coupling-pulses. CTAP has also been studied in Bose-Einstein condensates35 and has recently been experimentally demonstrated using multi-core opti- cal fibres36. In the STIRAP process the microwave coupling-pulses are assumed to be fully coherent. In CTAP, however, a noisy environment, e.g., TLFs could modify the time- dependent tunnelling-rates between the quantum dots. Hence the pulse amplitudes Ω0/1(t) controlling the tran- sitions between the dot 0 and 2, or 2 and 1, acquire extra fluctuating terms, δΩ0/1(t). While this picture fixes the coupling terms in the basis of the three dot levels, it could be either longitudinal (diagonal) of transverse (off- diagonal) in the eigenbasis of the TLF. We will see that the two types of coupling cause qualitatively different be- haviour and allow for both, writing δΩi(t) = gx,z τx,z. If for simplicity we assume that all three states of the quan- tum dots are energetically degenerate, the noise on the coupling-pulses takes the form i HSB =  0 0 0 0 gx,z 1 τx,z gx,z 2 τx,z gx,z 1 τx,z gx,z 2 τx,z 0   . (23) For the present example we assume that the relaxation rates Γ↓ and Γ↑ of the TLF are equal, which corresponds to a high-temperature environment. First we consider the case of longitudinal coupling. With a pure τz coupling we obtain from the quantum regression theorem the spectral function Cτzτz (ω) = 8(gz 1)2Γ↓ · Γ↑ (Γ↓ + Γ↑)3 + (Γ↓ + Γ↑) · ω2 . (24) It is peaked around zero frequency, and its resonance condition does not depend on the eigenfrequency of the fluctuator. The TLF is in resonance with the difference between the eigenvalues of HS(t) at the beginning and at the end of the pulse sequence seen in Figure 2. Therefore, in this case, the strength of decoherence is not increased as the difference between the eigenvalues of HS(t) be- comes resonant with the TLF. Only the decreasing of the φ = 0 φ = π/8 φ = π/4 φ = 3π/8 φ = π/2 1 0.9 0.8 0.7 0.6 0.5 0.4 l n o i t a u p o p e t a t s k r a d 0.7 0.8 0.9 Ω 1 max [ω 0] 1.1 1.2 1.3 FIG. 8. The transferred population is plotted over the coupling-pulse amplitude for a CTAP process where the TLF couples to the coupling-pulse amplitude between states 0i and 2i. Different cases of mixed coupling with the σx-τz coupling amplitude g · cos(φ) and σx-τx coupling amplitude g · sin(φ) are plotted. Γ↓ and Γ↑ are assumed to be equal. process time Tmax has an effect on the transferred pop- ulation as it shortens the time during which the system suffers from strong decoherence. The transferred population increases with decreasing process time. One should also note that the efficiency of CTAP is poor for this configuration since the strong decoherence at zero frequency is always present. This is equivalent to the ubiquitous low frequency noise seen in many solid-state systems. A qualitatively different behaviour emerges in the pres- ence of transverse coupling. With the assumption that there is no relaxation rate from ↓i to ↑i in the TLF ( Γ↑ = 0), the spectral function of the τx operators is the same as that of τ−τ +, Cτx,τx(ω) = Cτ −τ +(ω) . (25) As the spectral function is the same, the behaviour is similar to a TLF that couples transversally to the bare energy levels of the system. We obtain a result that is only quantitatively but not qualitatively different from the behaviour seen in Figure 5. Results of a simulation of a CTAP process with generalised system-TLF coupling, HSB = g [sin(φ)τx + cos(φ)τz ] (0ih2 + 2ih0) , (26) are shown in Figure 8. Here we see the smooth transition from the behaviour which is pure dephasing to that which displays TLF resonance effects. 8 VI. CONCLUSION We analysed the effect of damping on the STIRAP pro- tocol in situations where the environment is dominated by TLFs. In this case the environment is structured and an approach based on a master equation with Lindblad damping terms acting on the three-level-system alone is not sufficient. We compared two approaches to account for the structured environment: (i) the Bloch-Redfield theory for a three-dimensional system where we account for the effects of the structured environment beyond simple relaxation and dephasing rates by a frequency- dependent spectral function, and (ii) a higher dimen- sional approach where the TLFs of the environment are treated as part of the system. Generally speaking we can say that the Bloch-Redfield approach works well in the case of a weakly structured environment where the peaks around the TLF frequencies in the spectral function are broad, whereas one has to use the higher dimensional approach for more coherent TLFs. Due to the time-dependence of the Hamiltonian in STI- RAP the system is sensitive to the spectral function of the environment over a wide frequency interval. As a con- sequence, for a dense distribution of TLFs, the effects of decoherence on the STIRAP process can not be reduced by accelerating the STIRAP process. In the presence of only few TLFs one has to avoid using pulse amplitudes in the vicinity of the TLF resonance frequencies. This problem is more severe for the CTAP protocol, as envi- ronmental fluctuators may also couple to the tunnelling- matrix-elements between the quantum dots. ACKNOWLEDGMENTS We want to thank J. Jeske, C. Muller, A. Shnir- man, and A. Greentree for useful discussions on this work. The work has also been supported in the frame of the EU Information Societies Technologies Programme "Solid State Systems for Quantum Information Process- ing (SOLID)". 1 O. Astafiev, K. Inomata, A. O. Niskanen, T. Yamamoto, Y. A. Pashkin, Y. Nakamura, and J. S. Tsai, Nature 449, 588 (10 2007) 2 J. M. Fink, M. Goppl, M. Baur, R. Bianchetti, P. J. Leek, A. Blais, and A. Wallraff, Nature 454, 315 (07 2008) 3 M. Hofheinz, H. Wang, M. Ansmann, R. C. Bialczak, E. Lucero, M. Neeley, A. D. O'Connell, D. Sank, J. Wen- ner, J. M. Martinis, and A. N. Cleland, Nature 459, 546 (05 2009) 4 T. Niemczyk, F. Deppe, H. Huebl, E. P. Menzel, F. Hocke, M. J. Schwarz, J. J. Garcia-Ripoll, D. Zueco, T. Hummer, E. Solano, A. Marx, and R. Gross, Nat Phys 6, 772 (10 2010) 5 M. Marthaler, J. Leppakangas, and J. H. Cole, Phys. Rev. B 83, 180505 (May 2011) 6 K. Bergmann, H. Theuer, and B. W. Shore, Rev. Mod. Phys. 70, 1003 (Jul 1998) 7 Z. Kis and F. Renzoni, Phys. Rev. A 65, 032318 (Feb 2002) 8 G. Mangano, Falci, Siewert, and G. J. Eur. Phys. J. Special Topics 160, 259 (2008) J. Paspalakis, and N. E. 9 K. 9 21 J. H. Cole, C. Muller, P. Bushev, G. J. Grabovskij, J. Lisenfeld, A. Lukashenko, A. V. Ustinov, and A. Shnir- man, Applied Physics Letters 97, 252501 (2010)J. Lisen- feld, C. Muller, J. H. Cole, P. Bushev, A. Lukashenko, A. Shnirman, and A. V. Ustinov, Phys. Rev. Lett. 105, 230504 (Dec 2010) 22 P. A. Ivanov, N. V. Vitanov, and K. Bergmann, Phys. Rev. A 70, 063409 (Dec 2004) Kylstra, 23 A. Messiah, Quantum Mechanics Volume II (Elsevier Sci- Journal of Modern Optics 51, 1679 (2004) 10 J. Siewert, T. Brandes, and G. Falci, Phys. Rev. B 79, 024504 (Jan 2009) 11 A. La Cognata, P. Caldara, D. Valenti, D. Spag- and G. Falci, nolo, A. D'Arrigo, E. Paladino, International Journal of Quantum Information (IJQI) 9, 1 (Jan 2009) 12 M. A. Sillanpaa, J. Li, K. Cicak, F. Altomare, J. I. Park, R. W. Simmonds, G. S. Paraoanu, and P. J. Hakonen, Phys. Rev. Lett. 103, 193601 (Nov 2009) 13 O. Astafiev, A. M. Zagoskin, A. A. Abdumalikov, Y. A. Pashkin, T. Yamamoto, K. Inomata, Y. Nakamura, and J. S. Tsai, Science 327, 840 (2010) 14 I.-C. Hoi, C. M. Wilson, G. Johansson, T. Palomaki, B. Peropadre, and P. Delsing, Phys. Rev. Lett. 107, 073601 (Aug 2011) 15 A. D. Greentree, J. H. Cole, A. R. Hamilton, and L. C. L. Hollenberg, Phys. Rev. B 70, 235317 (Dec 2004) 16 J. Clarke and F. K. Wilhelm, Nature 453, 1031 (06 2008) 17 A. Shnirman, G. Schon, I. Martin, and Y. Makhlin, Phys. Rev. Lett. 94, 127002 (Apr 2005) ence B.V., 1961) 24 G. Lindblad, Commun. Math. Phys. 48, 119 (1976) 25 G. Ithier, E. Collin, P. Joyez, P. J. Meeson, D. Vion, D. Es- teve, F. Chiarello, A. Shnirman, Y. Makhlin, J. Schriefl, and G. Schon, Phys. Rev. B 72, 134519 (Oct 2005) 26 L. P. Yatsenko, V. I. Romanenko, B. W. Shore, and K. Bergmann, Phys. Rev. A 65, 043409 (Apr 2002) 27 M. Neeley, M. Ansmann, R. C. Bialczak, M. Hofheinz, N. Katz, E. Lucero, A. O/'Connell, H. Wang, A. N. Cle- land, and J. M. Martinis, Nat Phys 4, 523 (07 2008) 28 R. de Sousa, K. B. Whaley, T. Hecht, J. von Delft, and F. K. Wilhelm, Phys. Rev. B 80, 094515 (Sep 2009) 29 J. Carmichael, Howard, Statistical Methods in Quantum Optics 1 (Springer, 1999) 30 Q. Shi and E. Geva, J. Chem. Phys. 119, 11773 (2003), ISSN 00219606 31 P. Bushev, C. Muller, J. H. Cole, A. Lukashenko, A. Shnirman, and A. V. Ustinov, Phys. Rev. B 82, 134530 (Oct 2010) J. Lisenfeld, 32 M. Lax, Phys. Rev. 129, 2342 (Mar 1963) 33 I. Kamleitner, J. Cresser, and J. Twamley, Phys. Rev. A 18 J. Schriefl, Y. Makhlin, A. Shnirman, and G. Schon, New 77, 032331 (Mar 2008) Journal of Physics 8, 1 (2006) 19 J. M. Martinis, K. B. Cooper, R. McDermott, M. Steffen, M. Ansmann, K. D. Osborn, K. Cicak, S. Oh, D. P. Pap- pas, R. W. Simmonds, and C. C. Yu, Phys. Rev. Lett. 95, 210503 (Nov 2005) 20 A. Lupascu, P. Bertet, E. F. C. Driessen, C. J. P. M. Harmans, and J. E. Mooij, Phys. Rev. B 80, 172506 (Nov 2009) 34 K. Kakuyanagi, T. Meno, S. Saito, H. Nakano, K. Semba, H. Takayanagi, F. Deppe, and A. Shnirman, Phys. Rev. Lett. 98, 047004 (Jan 2007) 35 M. Rab, J. H. Cole, N. G. Parker, A. D. Greentree, L. C. L. Hollenberg, and A. M. Martin, Phys. Rev. A 77, 061602 (Jun 2008) 36 S. Longhi, G. Della Valle, M. Ornigotti, and P. Laporta, Phys. Rev. B 76, 201101 (Nov 2007)
1905.11227	3	1905	2019-09-19T16:25:08	Cavity QED of Strongly Correlated Electron Systems: A No-go Theorem for Photon Condensation	[ "cond-mat.mes-hall", "cond-mat.str-el", "quant-ph" ]	In spite of decades of work it has remained unclear whether or not superradiant quantum phases, referred to here as photon condensates, can occur in equilibrium. In this Letter, we first show that when a non-relativistic quantum many-body system is coupled to a cavity field, gauge invariance forbids photon condensation. We then present a microscopic theory of the cavity quantum electrodynamics of an extended Falicov-Kimball model, showing that, in agreement with the general theorem, its insulating ferroelectric and exciton condensate phases are not altered by the cavity and do not support photon condensation.	cond-mat.mes-hall	cond-mat	a Cavity QED of Strongly Correlated Electron Systems: A No-go Theorem for Photon Condensation G.M. Andolina,1, 2 F.M.D. Pellegrino,3, 4 V. Giovannetti,5 A.H. MacDonald,6 and M. Polini2 2Istituto Italiano di Tecnologia, Graphene Labs, Via Morego 30, I-16163 Genova, Italy 1NEST, Scuola Normale Superiore, I-56126 Pisa, Italy 3Dipartimento di Fisica e Astronomia "Ettore Majorana", Universit`a di Catania, Via S. Sofia 64, I-95123 Catania, Italy 4INFN, Sez. Catania, I-95123 Catania, Italy 5NEST, Scuola Normale Superiore and Istituto Nanoscienze-CNR, I-56126 Pisa, Italy 6Department of Physics, The University of Texas at Austin, Austin, TX 78712, USA In spite of decades of work it has remained unclear whether or not superradiant quantum phases, referred to here as photon condensates, can occur in equilibrium. In this Letter, we first show that when a non-relativistic quantum many-body system is coupled to a cavity field, gauge invariance forbids photon condensation. We then present a microscopic theory of the cavity quantum elec- trodynamics of an extended Falicov-Kimball model, showing that, in agreement with the general theorem, its insulating ferroelectric and exciton condensate phases are not altered by the cavity and do not support photon condensation. Introduction. -- Superradiance [1 -- 5] refers to the co- herent spontaneous radiation process that occurs in a dense gas when a radiation field mode mediates long- range inter-molecule interactions. Superradiance was ob- served first more than 40 years ago in optically pumped gases [2, 3] and has recently been identified in optically pumped electron systems in a semiconductor quantum well placed in a perpendicular magnetic field [6]. In 1973 Hepp and Lieb [7] and Wang and Hioe [8] independently pointed out that for sufficiently strong light-matter cou- pling the Dicke model, often used to describe superradi- ance in optical cavities, has a finite temperature second- order equilibrium phase transition between normal and superradiant states. To the best of our knowledge, this phase transition has never been observed [9]. In the su- perradiant phase the ground state contains a macroscop- ically large number of coherent photons, i.e. (cid:104)a(cid:105) (cid:54)= 0, where a (a†) destroys (creates) a cavity photon. To avoid confusion with the phenomenon discussed in the original work by Dicke [1], we refer to the equilibrium superradi- ant phase as a photon condensate. Theoretical work on photon condensation has an in- teresting and tortured history. Early on it was shown that photon condensation is robust against the addition of counter-rotating terms [10] neglected in Refs. [7, 8]. Soon after, however, Rzazewski et al. [11] pointed out that addition of a neglected term related to the Thomas- Reiche-Kuhn (TRK) sum rule [12, 13] and proportional to (a + a†)2 destroys the photon condensate. These quadratic terms are naturally generated by applying min- imal coupling p → p−qA/c to the electron kinetic energy p2/(2m). More recent research has focused on ground state properties. The quantum chaotic and entangle- ment properties of the Dicke model photon condensate were studied in Refs. [14, 15]. The authors of Ref. [16] criticized these studies however, pointing again to the im- portance of the quadratic term. The no-go theorem for photon condensation was revisited in Ref. [17], where it was claimed that it can be bypassed in a circuit quantum electrodynamics (QED) system with Cooper pair boxes capacitively coupled to a resonator. Soon after, however, Ref. [18] showed that the no-go theorem for cavity QED applies to circuit QED as well. The claims of Ref. [18] were then criticized in Ref. [19]. (See also subsequent discussions [20, 21] on light-matter interactions in cir- cuit QED.) Later it was argued [22] that the linear band dispersion of graphene provides a route to bypass the no- go theorem, and that photon condensation could occur in graphene in the integer quantum Hall regime. This claim was later countered in Refs. [23, 24], where it was shown that a dynamically generated quadratic term again for- bids photon condensation. Recent experimental progress has created opportuni- ties to study light matter interactions in new regimes in which direct electron-electron interactions play a promi- nent role. For example [25] two-dimensional (2D) elec- tron systems can be embedded in cavities or exposed to the radiation field of metamaterials, making it possible to study strong light-matter interactions in the quantum Hall regime [26 -- 31]. Other emerging possibilities include cavity QED with quasi-2D electron systems that exhibit exciton condensation, superconductivity, magnetism, or Mott insulating states. This Letter is motivated by in- terest in strong light-matter interactions in these new regimes and by fundamental confusion on when, if ever, photon condensation is allowed. We present a no-go the- orem for photon condensation that is valid for generic non-relativistic interacting electrons at T = 0. This re- sult generalizes to interacting systems existing no-go the- orems for photon condensation in two-level [11, 17] and multi-level [18] Dicke models. We then present a theory of cavity QED of an extended Falikov-Kimball model [32], which, in the absence of the cavity, has insulating ferro- electric and exciton condensate phases. We show through explicit microscopic calculations how the theorem is satis- fied in this particular strongly correlated electron model. Gauge invariance excludes photon condensation. -- We consider a system of N electrons of mass mi described by a non-relativistic many-body Hamiltonian of the form (cid:20) p2 i 2mi N(cid:88) i=1 H = (cid:21) (cid:88) i(cid:54)=j 1 2 + V (ri) + v(ri − rj) . (1) Here, V (r) is a generic function of position and v(r) is a generic (non-retarded) two-body interaction, which need not even be spherically symmetric. In a solid V (r) is the one-body the crystal potential. Below we first exclude the possibility of a continuous transition to a condensed state, and then use this insight to exclude first-order tran- sitions. For future reference, we denote by ψm(cid:105) and Em the exact eigenstates and eigenvalues of H [33, 34], with ψ0(cid:105) and E0 denoting the ground state and ground-state energy, respectively. We treat the cavity e.m. field in a quantum fashion, via a uniform quantum field A corresponding to only one (cid:112) mode [5, 7, 8, 10, 11, 13, 14, 17 -- 24, 35, 51, 52], i.e. A = A0u(a + a†), where u is the polarization vector, A0 = 2πc2/(V ωcr), V is the volume of the cavity, r is its relative dielectric constant, and the photon Hamiltonian Hph = ωca†a, where ωc is the cavity frequency. The full Hamiltonian, including light-matter interactions in the Coulomb gauge [35 -- 37] is: N(cid:88) a + † HA0 = H + ωca e2A2 † 0 2mic2 (a + a N(cid:88) + i=1 i=1 e mic pi · A0(a + a † ) )2 , (2) i=1 e2A2 0/(2mic2). (cid:80)N where A0 ≡ A0u and −e < 0 is the electron charge. The third and fourth terms in Eq. (2) are often re- ferred to respectively as the paramagnetic and diamag- netic contributions to the light-matter coupling Hamil- tonian. Our aim is to make general statements about the ground state Ψ(cid:105) of HA0. For future reference we define i) the paramagnetic (number) current opera- i=1 pi/mi tor [33, 34], jp ≡ (c/e)δ HA0/δA0A0=0 = (cid:80)N and ii) ∆ ≡ The term proportional to ∆ in Eq. (2) can be re- moved by performing the transformation b = cosh(x)a + sinh(x)a†, where cosh(x) = (λ + 1)/(2√λ) and sinh(x) = (λ − 1)/(2√λ) with λ = 1 + 4∆/(ωc). The Hamil- tonian (2) becomes: HA0 = H + (e/c)jp · A0λ−1/2(b + b†) + ωcλb†b. It can be shown (see Sec. I of the Supple- mental Material (SM) [38]) that in the thermodynamic limit (N → ∞, V → ∞ limit at fixed N/V ), the ground state Ψ(cid:105) of HA0 does not contain light-matter entangle- ment, i.e. we can take Ψ(cid:105) = ψ(cid:105)Φ(cid:105), where ψ(cid:105) and Φ(cid:105) are matter and light wave functions. Using this property we see that in the thermodynamic limit the ground state (cid:112) Φ(cid:105) of the effective photon Hamiltonian (cid:104)ψ HA0 ψ(cid:105) is a coherent state [39, 40] β(cid:105) satisfying bβ(cid:105) = β β(cid:105). The ground-state energy is therefore given by 2 Eψ(β) = (cid:104)ψ Hψ(cid:105) + e c (cid:104)ψjpψ(cid:105) · A0 2Re[β] √λ + ωcλβ2 . (3) We need to minimize Eψ(β) with respect to β and ψ(cid:105). The minimization with respect to β can be done analyt- ically. We find that the optimal value ¯β is a real number given by: ¯β = − 1 ωcλ3/2 e c (cid:104)ψjpψ(cid:105) · A0 . (4) We are therefore left with a constrained minimum prob- lem for the matter degrees of freedom. Its solution must be sought among the normalized anti-symmetric states ψ(cid:105) which yield (4). This is the typical scenario that can be handled with the stiffness theorem [34]. For photon condensation to occur we need Eψ( ¯β) < Eψ0(0) or, equivalently, ωcλ ¯β2 > (cid:104)ψ Hψ(cid:105) − (cid:104)ψ0 Hψ0(cid:105) , (5) where, because of (4), ψ(cid:105) depends on ¯β. The dependence of (cid:104)ψ Hψ(cid:105) − (cid:104)ψ0 Hψ0(cid:105) on ¯β can be calculated exactly up to order ¯β2 by using the stiffness theorem [34]. We find (cid:104)ψ Hψ(cid:105) − (cid:104)ψ0 Hψ0(cid:105) = α ¯β2/2 + O( ¯β3), where α = −1/χ(0) > 0 and 2 2ω2 c λ3 (cid:104)ψnjp · A0ψ0(cid:105)2 χ(0) ≡ − (cid:88) e2 c2 < 0 (6) n(cid:54)=0 En − E0 (cid:88) is proportional to the static paramagnetic current- current response function in the Lehmann representa- tion [33, 34]. We have used that (e/c)(cid:104)ψ0jpψ0(cid:105)·A0 = 0, as proven in Sec. II of the SM [38]. It follows that photon condensation occurs if and only if (cid:104)ψnjp · A0ψ0(cid:105)2 (cid:88) However, as shown in Sec. III of the SM [38], > ωc + 4∆ . En − E0 e2 c2 n(cid:54)=0 (7) 4 (cid:104)ψnjp · A0ψ0(cid:105)2 = ∆ . (8) e2 c2 n(cid:54)=0 En − E0 Eq. (8) is the TRK sum rule [12] which expresses the fact that the paramagnetic and diamagnetic contributions to the physical current-current response function cancel in the uniform static limit [33, 34], as discussed more fully in Sec. III of the SM [38], i.e. it expresses gauge invari- ance. Using Eq. (8) we can finally rewrite Eq. (7) as c24∆ > c2(ωc + 4∆) which cannot be satisfied. We con- clude that photon condensation cannot occur and that, upon minimization with respect to ψ(cid:105), the ground state is ψ0(cid:105) and ¯β = 0. From this analysis it is clear that first- order transitions to states with finite photon density are also excluded, because interactions with a coherent equi- librium photon field do not lower the matter energy [41]. Gauge invariance excludes photon condensation for any Hamiltonian of the form (2). This is the first important result of this Letter. Cavity QED of an extended Falikov-Kimball model. -- We now illustrate how this general conclusion applies to a specific properly gauge invariant model of strongly corre- lated electrons in a cavity. We consider spinless electrons in a one-dimensional (1D) inversion-symmetric crystal with N sites, each with one atom with two atomic or- bitals of opposite parity (s and p). When this lattice model is augmented by the addition of on-site repulsive electron-electron interactions, it is often referred to as an extended Falikov-Kimball (EFK) model [32]. The EFK model has been used to discuss exciton condensation [42] and electronic ferroelectricity [43, 44]. The coupling of cavity photons to the matter degrees of freedom of a 1D EFK model can be described [45 -- 48] by employing a Peierls substitution in the site representation with a uniform linearly-polarized vector potential of amplitude A0, as detailed in Sec. IV of the SM [38]. We obtain HA0 = H0 + Hee + ωca )2 , † a + a g0√N † jp(a + a ) † T (a + a † k,α,β c k,αHαβ(k)ck,β is the band Hamilto- (9) g2 0 2N − where H0 =(cid:80) nian, Hαβ(k) = 2it sin(ka) −2it sin(ka) Ep + 2tp cos(ka) and the Hubbard interaction term (cid:18)Es − 2ts cos(ka) N(cid:88) Hee = U (cid:19) , (10) (11) † † j,scj,sc c j,pcj,p . j=1 In Eq. (9), jp = (cid:80) erator, and T = (cid:80) † k,αjαβ(k)ck,β with jαβ(k) ≡ k,α,β c −1∂Hαβ(k)/∂k is the paramagnetic number current op- † k,αTαβ(k)ck,β with Tαβ(k) ≡ k,α,β c −a−2∂2Hαβ(k)/∂k2 is the diamagnetic operator. In Eq. (10), Es and Ep are on-site energies for the s and p orbitals, ts ∈ R and tp ∈ R are hopping parameters, and t ∈ R is the inter-band hopping parameter. At the single- particle level (i.e. for U = 0), t is the only term responsi- ble for inter-band transitions due to light. All sums over with the usual rule N−1(cid:80) the wave number k are carried out in the 1D Brillouin zone and become integrals in the thermodynamic limit −π/a dk/(2π), where a is the lattice constant. In these equations the Greek labels take values α, β = s, p. The momentum-space and site representations for field operators are linked by k → a(cid:82) +π/a per site. j,α = N−1/2(cid:80) 3 † † k,αe−ikja. The the usual relationship c k c (cid:112) dimensionless light-matter coupling constant in Eq. (9) is defined by g ≡ eaA0/(c) = g0/√N , where g0 ≡ 2πe2/(v0ωcr) and v0 = V /N is the cavity volume We emphasize that the operators jp and T describ- ing light-matter interactions are completely determined by the matrix elements Hαβ(k) of the band Hamilto- nian. This property is crucial to have a properly gauge- invariant model [49] and must be a general feature of any strongly correlated lattice model coupled to cavity photons. † k,ασ(z) k,α,β c a term proportional to(cid:80) In the limit g0 → 0, the model reduces to a 1D EFK model [32, 43, 44]. In the limit ka → 0 and U = 0, Eq. (9) reduces to the Dicke model, augmented by the addition of αβ ck,β(a + a†)2 [50 -- 52], where σ(z) αβ are the matrix elements of the correspond- ing 2 × 2 Pauli matrix. For non-interacting systems, the diamagnetic term prevents photon condensation from oc- curring in the thermodynamic limit [11, 17]. We now show that interactions do not help. HA0 does not sup- port photon condensation. To make progress in analyzing the interacting problem we treat the Hubbard term using an unrestricted Hartree- Fock (HF) approximation [34, 53]. As detailed in Sec. V of the SM [38] we arrive at † † k,pck,p − c (c k,sck,s) † c k,pck,s) + U ∗ 2 ee = −U M (cid:88) H(HF) k † (cid:18) (Ic k,sck,p + I M2 − n2 − U + U N k 0 (cid:19) + I2 (cid:88) k,α n0 2 nk,α . (12) In Eq. (12) we have introduced the following self- consistent fields: i) the electronic polarization (13) (14) M ≡ 1 N † † k,pck,p(cid:105) − (cid:104)c ((cid:104)c k,sck,s(cid:105)) , ii) the complex excitonic order parameter (cid:88) k I ≡ 1 N † (cid:104)c k,pck,s(cid:105) , N−1(cid:80) and iii) the number of electrons per site n0 ≡ † k,α(cid:104)nk,α(cid:105), where nk,α ≡ c k,αck,α. The term pro- portional to n0/2 in Eq. (12) acts as a renormalization of the chemical potential in the grand-canonical Hamil- tonian and can be discarded in this study since we study the phase diagram only at half filling and n0 = 1 in all phases. In order to reduce the number of free parameters in the problem, from now on we enforce particle-hole symmetry in the bare band Hamiltonian H0 by setting Es ≡ −Ep = −Eg/2 and ts ≡ tp = t (with t > Eg/4, (cid:88) 4 (cid:88) k mind the reader that the photon condensate order param- see Fig. S1). In order to find the ground state of the Hamiltonian (9) with Hubbard interactions treated as in Eq. (12), we follow the same steps outlined in the proof of the no-go theorem above. We seek a ground state of the unentangled form Ψ(cid:105) = ψ(cid:105)Φ(cid:105). After removing the term proportional to (a+a†)2, one finds that Φ(cid:105) must be a co- herent state ¯β(cid:105) with ¯β = −g0J √N /(λ3/2ωc). (We re- ¯β(cid:105)/√N = eter is (cid:104) ¯βa ¯β(cid:105) /√N = (cid:104) ¯β cosh(x)b− sinh(x)b† ¯β/√N λ. See Sec. VI of the SM [38].) Here, J ≡ (cid:104)ψ jp ψ(cid:105) /(aN ), λ has the same expression as in the proof of the no-go theorem with ∆ = −g2 0T /2, and T ≡ (cid:104)ψ T ψ(cid:105). Note that both J and T have units of energy and are finite in the N → ∞ limit. The resulting effective Hamiltonian for the matter de- grees of freedom, i.e. (cid:104) ¯β HA0 ¯β(cid:105), can be diagonalized exactly since, after the HF decoupling, it is quadratic † k,α. To this end, it is in the fermionic operators ck,α, c † k,− = sufficient to introduce the Bogoliubov operators γ † † † ukc k,s+vkc k,p and γ kc k,p, where the quan- tities uk and vk depend on the parameters of the bare Hamiltonian H0, on the Hubbard parameter U , on the light-matter coupling constant g0, and on the quantities † k,− vac(cid:105) † † k,+ = v∗ k,s−u∗ kc can be written in a BCS-like fashion, k γ (cid:2)uk + vkc I, M, J , and T . The ground state ψ(cid:105) =(cid:81) where ∅(cid:105) = (cid:81) in terms of uk, vk: M = N−1(cid:80) kuk, J = 2N−1(cid:80) N−1(cid:80) kvk)], and T = 2N−1(cid:80) uk2(cid:1) † k,s vac(cid:105) and vac(cid:105) is the state with k c no electrons. The final ingredients which are needed are the expressions for the quantities M, I, J , and T k(vk2 − uk2), I = k[−t sin(ka)(vk2 − uk2) − vk2− kvk)]. The technical details of this k v∗ − 2t sin(ka)Im(u∗ k[t cos(ka)(cid:0) 2t cos(ka)Im(u∗ calculation are summarized in Sec. VI of the SM [38]. † k,pck,s ∅(cid:105) , (cid:89) k ψ(cid:105) = (cid:3) (15) The quantities I, M, J , and T can be determined by solving this nonlinear system of equations. A typical so- lution is shown in Fig. 1. We have found that all observ- ables are independent of g0. In other words, in the ther- modynamic limit the ground state is given by Eq. (15) with uk and vk evaluated at g0 = 0, in agreement with the general theorem proven above. The self-consistent solutions always have J = 0 (i.e. ¯β = 0), as clearly seen in Fig. 1(c), and therefore display no photon condensa- tion but may have finite polarization and exciton order parameters. This is the second important result of this Letter. At t = 0 the HF ground state has a single tran- sition at U = UXC. For 0 < U < UXC the ground state is an exciton condensate with spontaneous coherence be- tween s and p bands [43, 44] which are not hybridized when U = 0. The ordered state appears on the small U side of the transition because interactions favor or- bital polarization over coherence. The value of UXC can be determined analytically as detailed in Sec. VIII of the (a) (b) 4 (c) (d) FIG. 1. (Color online) Panel (a) The excitonic order pa- rameter I is plotted as a function of U (in units of Eg). Numerical results have been obtained by setting t = 0.5 Eg and ωc = Eg. Different curves correspond to different val- −4 Eg. Black dotted line: ues of t. Red solid line: t = 10 t = 0.05 Eg. Blue dashed line: t = 0.1 Eg. Green dash- dotted line: t = 0.15 Eg. Note that for t (cid:54)= 0, I (cid:54)= 0 for Uc1 < U < Uc2. Panel (b) Same as in panel (a) but for the electronic polarization M. Panel (c) Same as in other panels but for J . Note that J = 0 for all values of t and U/Eg. This implies ¯β = 0 and therefore no photon condensation. Panel (d) Same as in other panels but for T (in units of Eg). SM [38]. We find, in agreement with earlier work [54, 55], that UXC = 8t2/Eg − Eg/2. In the limit t = 0, HA0 separately conserves the num- ber of electrons with band indices α = s, p, and has a global U (1) symmetry associated with the arbitrariness of the relative phase between s and p electrons [32]. The HF ground state breaks this symmetry. For t (cid:54)= 0 the U (1) symmetry is reduced to a discrete Z2 symmetry re- flecting the invariance of the Hamiltonian under spatial inversion. This symmetry is broken for Uc1(t) < U < Uc2(t). Note that limt→0 Uc2(t) = UXC. Corrections to Uc2(0) can be found perturbatively for t/t (cid:28) 1 and are of O(t2) (see Sec. VIII of the SM [38]). For 0 < U < Uc1(t) inversion symmetry is unbroken and I = 0. For U > Uc1(t) the ground state is an insulating ferroelectric that breaks the Z2 symmetry (see Sec. IX of the SM [38]). The dependence of Uc1 on t in non-analytical and can be extracted asymptotically for t/t (cid:28) 1. We find that g /4)1/2/ ln(t/t) (see Sec. VIII of the Uc1(t) → π(4t2 − E2 SM [38]). In summary, we have presented a no-go theorem for photon condensation that applies to all quantum many- body Hamiltonians of the form (1), greatly extend- ing previous no-go theorems for Dicke-type Hamiltoni- ans [17, 18]. Since the proof is non-perturbative in the strength of electron-electron interactions, our arguments 00.511.52U[Eg]00.10.20.3I00.511.52U[Eg]−1−1/20M00.511.52U[Eg]−0.100.1J[Eg]00.511.52U[Eg]−1−0.50T[Eg]against photon condensation apply to all lattice models of strongly correlated electron systems that can be de- rived from Eq. (1). We have then explained how the the- orem manifests in practice, presenting a theory of cavity QED of a 1D model that supports insulating ferroelet- ric and exciton condensate phases. We have shown that these electronic orders are never entwined with photon condensation [56]. In the future, it will be interesting to study the role of spatially-varying multimode cavity fields and their interplay with retarded interactions [57, 58], or strong magnetic fields [59]. Our work emphasizes that theoretical models of interacting light-matter sys- tems must retain precise gauge invariance, which is often lost when the matter system is projected onto a low- energy model. Acknowledgements. -- A.H.M. was supported by Army Research Office (ARO) Grant # W911NF-17-1-0312 (MURI), and by Welch foundation Grant TBF1473. It is a great pleasure to thank M.I. Katsnelson and F.H.L. Koppens for useful discussions. [1] R.H. Dicke, Phys. Rev. 93, 99 (1954). [2] M. Gross and S. Haroche, Phys. Rep. 93, 301 (1982). [3] K. Cong, Q. Zhang, Y. Wang, G.T. Noe II, A. Belyanin, and J. Kono, J. Opt. Soc. Am. B 33, C80 (2016). [4] A.F. Kockum, A. Miranowicz, S. De Liberato, S. Savasta, and F. Nori, Nat. Rev. Phys. 1, 19 (2019). [5] P. Kirton, M.M. Roses, J. Keeling, and E.G. Dalla Torre, Adv. Quantum Technol. 2, 1970013 (2019). [6] G.T. Noe II, J.-H. Kim, J. Lee, Y. Wang, A.K. W´ojcik, S.A. McGill, D.H. Reitze, A.A. Belyanin, and J. Kono, Nat. Phys. 8, 219 (2012). [7] K. Hepp and E.H. Lieb, Ann. Phys. 76, 360 (1973) and Phys. Rev. A 8, 2517 (1973). [8] Y.K. Wang and F.T. Hioe, Phys. Rev. A 7, 831 (1973). [9] See, however, experimental work on Dicke superradiance in open systems formed by Bose-Einstein condensates cou- pled to optical cavities: K. Baumann, C. Guerlin, F. Bren- necke, and T. Esslinger, Nature 464, 1301 (2010); K. Bau- mann, R. Mottl, F. Brennecke, and T. Esslinger, Phys. Rev. Lett. 107, 140402 (2011); H. Ritsch, P. Domokos, F. Brennecke, and T. Esslinger, Rev. Mod. Phys. 85, 553 (2013). [10] H.J. Carmichael, C.W. Gardiner, and D.F. Walls, Phys. Lett. 46A, 47 (1973). [11] K. Rzazewski, K. W´odkiewicz, and W. Zakowicz, Phys. Rev. Lett. 35, 432 (1975); I. Bialynicki-Birula and K. Rz¸azewski, Phys. Rev. A 19, 301 (1979); K. Gaw¸edzki and K. Rz¸azewski, Phys. Rev. A 23, 2134 (1981). [12] J.J. Sakurai, Modern Quantum Mechanics (Addison- 5 [15] V. Buzek, M. Orszag, and M. Rosko, Phys. Rev. Lett. 94, 163601 (2005). [16] K. Rz¸azewski and K. W´odkiewicz, Phys. Rev. Lett. 96, 089301 (2006). [17] P. Nataf and C. Ciuti, Nature Commun. 1, 72 (2010). [18] O. Viehmann, J. von Delft, and F. Marquardt, Phys. Rev. Lett. 107, 113602 (2011). [19] C. Ciuti and P. Nataf, Phys. Rev. Lett. 109, 179301 (2012). [20] T. Jaako, Z.-L. Xiang, J.J. Garcia-Ripoll, and P. Rabl, Phys. Rev. A 94, 033850 (2016). [21] M. Bamba, K. Inomata, and Y. Nakamura, Phys. Rev. Lett. 117, 173601 (2016). [22] D. Hagenmuller and C. Ciuti, Phys. Rev. Lett. 109, 267403 (2012). [23] L. Chirolli, M. Polini, V. Giovannetti, and A.H. MacDon- ald, Phys. Rev. Lett. 109, 267404 (2012). [24] F.M.D. Pellegrino, L. Chirolli, V. Giovannetti, R. Fazio, and M. Polini, Phys. Rev. B 89, 165406 (2014). [25] F.M.D. Pellegrino, V. Giovannetti, A.H. MacDonald, and M. Polini, Nature Commun. 7, 13355 (2016). [26] G. Scalari, C. Maissen, D. Turcinkov´a, D. Hagenmuller, S. De Liberato, C. Ciuti, C. Reichl, D. Schuh, W. Wegscheider, M. Beck, and J. Faist, Science 335, 1323 (2012). [27] V.M. Muravev, P.A. Gusikhin, I.V. Andreev, and I.V. Kukushkin, Phys. Rev. B 87, 045307 (2013). [28] S. Smolka, W. Wuester, F. Haupt, S. Faelt, W. Wegschei- der, and A. Imamoglu, Science 346, 332 (2014). [29] S. Ravets, P. Knuppel, S. Faelt, M. Kroner, W. Wegschei- der, and A. Imamoglu, Phys. Rev. Lett. 120, 057401 (2018). [30] G.L. Paravicini-Bagliani, F. Appugliese, E. Richter, F. Valmorra, J. Keller, M. Beck, N. Bartolo, C. Rossler, T. Ihn, K. Ensslin, C. Ciuti, G. Scalari, and J. Faist, Nat. Phys. 15, 186 (2019). [31] P. Knuppel, S. Ravets, M. Kroner, S. Falt, W. Wegschei- der, and A. Imamoglu, arXiv:1903.09256. [32] For a recent review see e.g. J. Kunes, J. Phys.: Condens. Matter 27, 333201 (2015). [33] D. Pines and P. Nozi`eres, The Theory of Quantum Liq- uids (W.A. Benjamin, Inc., New York, 1966). [34] G.F. Giuliani and G. Vignale, Quantum Theory of the Electron Liquid (Cambridge University Press, Cambridge, 2005). [35] J. Keeling, J. Phys.: Condens. Matter 19, 295213 (2007). [36] A. Vukics and P. Domokos, Phys. Rev. A 86, 053807 (2012). [37] A. Stokes and A. Nazir, arXiv:1905.10697. [38] See the Supplemental Material file for technical details on the ground-state factorization, on the stiffness theo- rem, on the TRK sum rule, on the coupling of light to the EFK model degrees of freedom, on the HF treatment of electron-electron interactions and the resulting Bogoli- ubov transformation, on the f -sum rule, and on the phase diagram of the EFK model. [39] D.F. Walls and G.J. Milburn, Quantum Optics (Springer, Wesley, Reading-MA, 1994). Berlin, 2007). [13] T. Tufarelli, K.R. McEnery, S.A. Maier, and M.S. Kim, Phys. Rev. A 91, 063840 (2015). [14] C. Emary and T. Brandes, Phys. Rev. Lett. 90, 044101 (2003) and Phys. Rev. E 67, 066203 (2003); N. Lambert, C. Emary, and T. Brandes, Phys. Rev. Lett. 92, 073602 (2004). [40] A. Serafini, Quantum Continuous Variables: A Primer of Theoretical Methods (CRC Press, Boca Raton, FL, 2017). [41] In the thermodynamic limit, the ground-state factoriza- tion implies a photon coherent state. The matter state for a given β is therefore equivalent to a matter state with a time-independent spatially constant vector potential. In 6 turn, this implies a matter state energy independent of β. According to Eq. (3), introducing photons only costs en- ergy, forbidding a first-order transition to a photon con- densate. [42] See, for example, D. J´erome, T.M. Rice, and W. Kohn, Phys. Rev. 158, 462 (1967); B.I. Halperin and T.M. Rice, Solid State Phys. 21, 115 (1968); B.I. Halperin and T.M. Rice, Rev. Mod. Phys. 40, 755 (1968). [43] T. Portengen, Th. Ostreich, and L.J. Sham, Phys. Rev. Lett. 76, 3384 (1996) and Phys. Rev. B 54, 17452 (1996). [44] C.D. Batista, Phys. Rev. Lett. 89, 166403 (2002). [45] W. Kohn, Phys. Rev. 133, A171 (1964). [46] B.S. Shastry and B. Sutherland, Phys. Rev. Lett. 65, 243 (1990). [47] A.J. Millis and S.N. Coppersmith, Phys. Rev. B 42, 10807(R) (1990). [48] R.M. Fye, M.J. Martins, D.J. Scalapino, J. Wagner, and W. Hanke, Phys. Rev. B 44, 6909 (1991). [49] Gauge invariance (of linear-response functions) and the f -sum rule follow from particle (charge) conservation [33, 34]. Consider Eq. (9) at g0 = 0 and arbitrary values of U . In this case, the conservation law reads as following, ∂t n(cid:96)+ (cid:80) (jp,(cid:96)+1 − jp,(cid:96))/a = 0, which is the lattice version of 1D continuity equation. Here, we have defined the local den- † sity operator in the site representation, n(cid:96) ≡ α c (cid:96),αc(cid:96),α and the local paramagnetic (number) current operator † † † (cid:96)−1,sc(cid:96),s)/ − itpa(c jp,(cid:96) ≡ itsa(c (cid:96),pc(cid:96)−1,p − (cid:96),sc(cid:96)−1,s − c † † † † (cid:96)−1,pc(cid:96),p)/+ita(c (cid:96)−1,pc(cid:96),s)/+ita(c (cid:96),sc(cid:96)−1,p−c (cid:96)−1,sc(cid:96),p− c † (cid:96),pc(cid:96)−1,s)/. Coupling to the uniform vector potential of c the cavity must be done via the paramagnetic current op- jp,(cid:96) (while, at the same time, including the diamagnetic term). This is manifestly displayed by our Hamiltonian (9) at g0 (cid:54)= 0. Further details on the f -sum rule can be found in Sec. VII of the SM [38]. [50] O. Di Stefano, A. Settineri, V. Macr`ı, L. Garziano, R. erator jp =(cid:80)N (cid:96)=1 Stassi, S. Savasta, and F. Nori, Nat. Phys. (2019). [51] D. De Bernardis, T. Jaako, and P. Rabl, Phys. Rev. A 97, 043820 (2018). [52] D. De Bernardis, P. Pilar, T. Jaako, S. De Liberato, and P. Rabl, Phys. Rev. A 98, 053819 (2018). [53] J.A. Verg´es, E. Louis, P.S. Lomdahl, F. Guinea, and A.R. Bishop, Phys. Rev. B 43, 6099 (1991); J.A. Verg´es, F. Guinea, and E. Louis, Phys. Rev. B 46, 3562 (1992). [54] A.N. Kocharian and J.H. Sebold, Phys. Rev. B 53, 12804 (1996). [55] S. Ejima, T. Kaneko, Y. Ohta, and H. Fehske, Phys. Rev. Lett. 112, 026401 (2014). [56] The influence of strong interactions combined with a uni- form single-mode cavity field on the phase diagram of small-gap semiconductors has also been considered in an interesting recent paper by G. Mazza and A. Georges, Phys. Rev. Lett. 122, 017401 (2019). Some of the con- clusions of this paper are incorrect because the model is not precisely gauge invariant. [57] F. Schlawin, A. Cavalleri, and D. Jaksch, Phys. Rev. Lett. 122, 133602 (2019). [58] J.B. Curtis, Z.M. Raines, A.A. Allocca, M. Hafezi, and V.M. Galitski, Phys. Rev. Lett. 122, 167002 (2019). [59] S.M. Girvin, A.H. MacDonald, and P.M. Platzman, Phys. Rev. Lett. 54, 581 (1985) and Phys. Rev. B 33, 2481 (1986); S.M. Girvin and A.H. MacDonald, Phys. Rev. Lett. 58, 1252 (1987); F.D.M. Haldane, Phys. Rev. Lett. 107, 116801 (2011). 1 Supplemental Material for "Cavity QED of Strongly Correlated Electron Systems: A No-go Theorem for Photon Condensation" G.M. Andolina,1, 2 F.M.D. Pellegrino,3 ,4 V. Giovannetti,5 A.H. MacDonald,6 and M. Polini1 1Istituto Italiano di Tecnologia, Graphene Labs, Via Morego 30, I-16163 Genova, Italy 2NEST, Scuola Normale Superiore, I-56126 Pisa, Italy 3Dipartimento di Fisica e Astronomia "Ettore Majorana", Universit`a di Catania, Via S. Sofia 64, I-95123 Catania, Italy 4INFN, Sez. Catania, I-95123 Catania, Italy 5NEST, Scuola Normale Superiore and Istituto Nanoscienze-CNR, I-56126 Pisa, Italy 6Department of Physics, The University of Texas at Austin, Austin, TX 78712, USA In this Supplemental Material we provide additional information on the ground-state factorization (Sec. I), on the stiffness theorem (Sec. II), on the TRK sum rule (Sec. III), on the coupling of light to the EFK model degrees of freedom (Sec. IV), on the Hartree-Fock treatment of electron-electron interactions (Sec. V) and the resulting Bogoliubov transformation (Sec. VI), on the f -sum rule (Sec. VII), and on the phase diagram of the EFK model (Sects. VIII and IX). SECTION I: DISENTANGLING LIGHT AND MATTER In this Section we show that, in the thermodynamic N → ∞ limit, it is permissible to assume a factorized ground state of the form We begin by defining the electron-photon Hamiltonian Ψ(cid:105) = ψ(cid:105)Φ(cid:105) . N(cid:88) i=1 Hel−ph = e mic † pi · A0(a + a † ) + ∆(a + a )2 , where ∆ has been defined in the main text. The electron Hamiltonian H and the photon Hamiltonian Hph have been defined in the main text. Each of the Hamiltonians H, Hph, and Hel−ph scales extensively in N . While this is obvious for H, we note that also Hph and Hel−ph scale with N since A0 ∝ 1/√N and a, a† ∝ √N . Below, we therefore work with the operators H/N , Hph/N , and Hel−ph/N which are well defined in the N → ∞ limit. For the sake of simplicity, we now assume that all electrons have the same mass, i.e. mi = m ,∀i = 1 . . . N . In order to prove Eq. (S1) we will prove that, in the limit N → ∞ (S1) (S2) (S3) (S4) (S5) and Hph N [ Explicitly, the left-hand side of Eq. (S3) reads as following: N(cid:88) i=1 H N [ , Hel−ph N ] = [ V (ri) + 1 2 v(ri − rj), i(cid:54)=j N(cid:88) j=1 e mc pj · A0] (a + a†) N 2 . H N [ , Hel−ph N ] → 0 ] → 0 . , N Hel−ph (cid:88) Using that [f (ri), pj] = δi,ji∇ ri f (ri) and introducing the external force F ext F C i,j = −∇ ri v(ri − rj)/2 we get: i = −∇ riV (ri) and the Coulomb force H N [ , Hel−ph N ] = − ie(a + a†)A0 mcN 2 · N(cid:88) i=1 F ext i , (S6) Exploiting the commutator [a, a†] = 1, we can rewrite the left-hand side of Eq. (S4) as: where F C i,j dropped out of the commutator since (cid:80) which does not scale with N , and that(cid:80)N (cid:40) N(cid:88) 1/N , and therefore vanishes in the thermodynamic limit. i=1 Hph N Hel−ph [ , Again, this quantity scales like 1/N , since(cid:80)N ] = i=1 N e mc † pi · A0(a ωc N 2 − a) + ∆(cid:2)(a i=1 pi ∼ N and ∆ ∼ 1. 2 i,j = 0. Noticing that (a + a†)A0 is an intensive quantity, F C F ext i ∼ N , we obtain that the commutator [ H/N, Hel−ph/N ] scales like i,j (cid:41) † † + a)(a † − a) + (a † − a)(a + a)] . (S7) SECTION II: ON THE STIFFNESS THEOREM In this Section we prove that (cid:104)ψ0jpψ0(cid:105)·A0 = 0. We used this property to evaluate the quantity (cid:104)ψ Hψ(cid:105)−(cid:104)ψ0 Hψ0(cid:105) i ri and note that, because of the fundamental commutator We introduce the total dipole operator d = −e(cid:80) up to order ¯β2, via the stiffness theorem. [r(cid:96)α, pkβ] = iδ(cid:96),kδα,β, we have and − iejp = [ d, H] [jpα, dβ] = i(cid:88) e m(cid:96) (cid:96) δα,β . Using Eq. (S8) we immediately find for a large but finite system (cid:104)ψ0jpψ0(cid:105) · A0 = i e (cid:104)ψ0[ d, H]ψ0(cid:105) · A0 = i e (E0 − E0)(cid:104)ψ0 dψ0(cid:105) · A0 = 0 . SECTION III: ON THE TRK SUM RULE, I.E. EQ. (8) IN THE MAIN TEXT In this Section we prove the TRK sum rule, i.e. Eq. (8) in the main text. Eq. (S8) implies that (cid:88) n(cid:54)=0 e2 c2 (cid:104)ψnjp · A0ψ0(cid:105)2 En − E0 = 1 2c2 (En − E0)(cid:104)ψn d · A0ψ0(cid:105)2 . (cid:88) n(cid:54)=0 (S8) (S9) (S10) (S11) (S12) (S13) (S14) Eq. (S9) implies that we can rewrite ∆ as: N(cid:88) ∆ = e2A2 0 2m(cid:96)c2 = We then manipulate the right-hand side of Eq. (S12) by inserting exact identities [S1, S2] 11 =(cid:80) (cid:88) n (cid:104)ψ0jp · A0ψn(cid:105)(cid:104)ψn d · A0ψ0(cid:105) (cid:104)ψ0[ d · A0, jp · A0]ψ0(cid:105) = ie 2c2(cid:104)ψ0[ d · A0, jp · A0]ψ0(cid:105) . appropriate positions, (cid:96)=1 (cid:88) (cid:88) n (cid:104)ψ0 d · A0ψn(cid:105)(cid:104)ψnjp · A0ψ0(cid:105) − i (cid:88) n (cid:104)ψ0 d · A0ψn(cid:105)(cid:104)ψn[ d · A0, H]ψ0(cid:105) − e 2i 2i (En − E0)(cid:104)ψn d · A0ψ0(cid:105)2 = − e e = = − n i e (cid:88) (cid:88) n (cid:104)ψ0[ d · A0, H]ψn(cid:105)(cid:104)ψn d · A0ψ0(cid:105) (En − E0)(cid:104)ψn d · A0ψ0(cid:105)2 . n(cid:54)=0 n ψn(cid:105)(cid:104)ψn in the Using the previous result inside Eq. (S12), we find (cid:88) n(cid:54)=0 ∆ = 1 2c2 (En − E0)(cid:104)ψn d · A0ψ0(cid:105)2 . Comparing Eq. (S14) with Eq. (S11) we reach the desired result, i.e. Eq. (8) of the main text. 3 We now present a more physical, alternative proof. We first remind the reader that the physical current operator corresponding to the Hamiltonian HA0 , Eq. (2) in the main text, is N(cid:88) Jphys = = jp + c e δ HA0 δA0 e mic A0 . (S15) i=1 We now observe that the electron system cannot respond to A0, since the latter is uniform and time-independent. (A current cannot flow along u in response to A0.) This property, i.e. gauge invariance, implies that the physical current-current response function in response to A0 must vanish [S1, S2], i.e. (cid:88) n(cid:54)=0 N(cid:88) i=1 0 = − 2 Ld (cid:104)ψnjp · uψ0(cid:105)2 En − E0 + 1 Ld 1 mi , (S16) where the first (second) term on the right-hand side is the paramagnetic (diamagnetic) contribution and Ld is the electron system volume. Eq. (S16) can be written as (cid:88) n(cid:54)=0 2 (cid:104)ψnjp · uψ0(cid:105)2 En − E0 = N(cid:88) i=1 1 mi , (S17) which is easily seen to be equivalent to Eq. (8). In other words, Eq. (8) simply expresses the fact that paramagnetic and diamagnetic contributions to the physical current-current response function cancel out in the uniform and static limit [S1, S2]. SECTION IV: COUPLING THE EFK MODEL TO CAVITY PHOTONS Consider spinless electrons hopping in a one-dimensional inversion-symmetric crystal with N sites, one atom per site, and two atomic orbitals of opposite parity (s and p), in a tight-binding scheme. The second-quantized single-particle Hamiltonian in the site representation reads as following: H0 = (cid:88) N(cid:88) N(cid:88) j=1 α=s,p − t j=1 † j,αcj,α − ts Eαc † † j+1,scj,s + c (c j,scj+1,s) + tp † † j+1,pcj,p + c (c j,pcj+1,p) † † j,pcj+1,s) + t j+1,scj,p + c (c † † j+1,pcj,s) ≡ j,scj+1,p + c (c † j,αcj,α + T , Eαc (S18) N(cid:88) j=1 N(cid:88) (cid:80)N j=1 N(cid:88) j=1 N(cid:88) (cid:88) j,pcj+1,p) − t(cid:80)N α=s,p j=1 where ts, tp, t ∈ R are for the moment completely arbitrary, we assumed periodic boundary conditions (cN +1,α = t(cid:80)N † † c1,α), and defined the kinetic operator T = tp j+1,scj,p + c j=1(c j,pcj+1,s) + † † j,scj+1,p + c j=1(c j+1,pcj,s). We now add repulsive on-site electron-electron interactions in the site representa- † † j+1,pcj,p + c j=1(c tion: Hee = U N(cid:88) j=1 nj,s nj,p , † where U > 0 and nj,α ≡ c j,αcj,α is the orbitally-resolved local density operator. The full Hamiltonian of our 1D EFK model in the absence of cavity photons is H = H0 + Hee . (S19) (S20) We now couple the matter Hamiltonian H in Eq. (S20) to light by employing a uniform linearly-polarized vector potential A(t) = A(t) u where u = ± x in the ring geometry above with periodic boundary conditions. This is accomplished, as usual [S3 -- S6], by means of the Peierls factor: HA(t) = † j,αcj,α − ts Eαc † † −ieaA(t)/(c)c j+1,scj,s + e+ieaA(t)/(c)c j,scj+1,s e N(cid:88) (cid:16) j=1 4 (cid:17) α=s,p j=1 N(cid:88) (cid:88) (cid:16) N(cid:88) (cid:16) N(cid:88) (cid:16) N(cid:88) j=1 j=1 e + tp − t + t † † −ieaA(t)/(c)c j+1,pcj,p + e+ieaA(t)/(c)c j,pcj+1,p e † † −ieaA(t)/(c)c j+1,scj,p + e+ieaA(t)/(c)c j,pcj+1,s † e+ieaA(t)/(c)c j,scj+1,p + e † −ieaA(t)/(c)c j+1,pcj,s j=1 + Hee , where a is the lattice constant. We expand HA(t) in powers of A(t) for small A(t), retaining terms of O(A2(t)). We find: HA(t) = H0 + Hee + e c A(t)jp − 1 2 e2a2A2(t) 2c2 where c e δ HA(t) δA(t) jp ≡ (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)A(t)=0 N(cid:88) N(cid:88) j=1 j=1 = itsa ita + † † j+1,scj,s − c j,scj+1,s) − (c itpa † † j+1,pcj,p − c (c j,pcj+1,p) † † j+1,scj,p − c (c j,pcj+1,s) + ita † † j,scj+1,p − c (c j+1,pcj,s) is the paramagnetic (number) current operator and T is the kinetic operator in Eq. (S18). The physical (number) current operator is therefore Jphys ≡ δ HA(t) δA(t) c e = jp − e c A(t) a2 2 T , (S24) with contains paramagnetic and diamagnetic terms. We finally quantize the e.m. field by writing A(t) → A0(a + a†), where A0 has been defined in the main text, and we give dynamics to the field by means of the photon Hamiltonian Hph = ωca†a. The full Hamiltonian, which includes light-matter interactions, is therefore given by: HA0 = H0 + Hee + Hph + † A0jp(a + a e c 1 2 e2a2 2c2 A2 0 † T (a + a ) − )2 . (S25) The fourth and fifth terms in the right-hand side of Eq. (S25) are called paramagnetic and diamagnet contributions. Eq. (S25) is written in the site representation. In the main text, however, the quantities H0, jp, and T have been given in momentum space. The link between momentum-space and site representations is offered by † c j,α = −ikja , † c k,αe (cid:88) (cid:90) +π/a k∈BZ 1 √N (cid:88) 1 N → a −π/a k∈BZ dk 2π . where the sum is carried over the 1D Brillouin zone (BZ). In the thermodynamic N → ∞ limit we can replace We find H0 = (cid:88) (cid:16) k∈BZ † † k,s c c k,p (cid:17)(cid:18)Es − 2ts cos(ka) 2it sin(ka) −2it sin(ka) Ep + 2tp cos(ka) (cid:19)(cid:18)ck,s ck,p (cid:19) ≡ (cid:88) (cid:88) k∈BZ α,β=s,p † c k,αHαβ(k)ck,β , (S28) (cid:17) (cid:17) (cid:17) T , N(cid:88) N(cid:88) j=1 j=1 (S21) (S22) (S23) (S26) (S27) (cid:16) (cid:88) (cid:16) (cid:88) k∈BZ k∈BZ (cid:17)(cid:18) ts sin(ka) (cid:17)(cid:18) (cid:19)(cid:18)ck,s (cid:19) (cid:19)(cid:18)ck,s (cid:19) ck,p ck,p ≡ ≡ (cid:88) (cid:88) k∈BZ (cid:88) (cid:88) α,β=s,p k∈BZ α,β=s,p † c k,αjαβ(k)ck,β , † k,αTαβ(k)ck,β . c jp = 2a † † k,s c c k,p it cos(ka) −it cos(ka) −tp sin(ka) and T = 2 † † k,s c c k,p it sin(ka) −ts cos(ka) −it sin(ka) +tp cos(ka) It is easy to check that and jαβ(k) = Tαβ(k) = − 1 1 a2 ∂Hαβ(k) ∂k ∂2Hαβ(k) ∂k2 . Equations (S31)-(S32) heavily constrain the paramagnetic and diamagnetic terms of the full Hamiltonian HA0 , which rule light-matter interactions. In other words, one cannot simply couple light to matter with arbitrary operators jp and T . Instead the form of these operators is specified by H0 and must be constructed with perfect consistency. SECTION V: HARTREE-FOCK TREATMENT OF ELECTRON-ELECTRON INTERACTIONS We treat the electron-electron interaction term in Eq. (11) of the main text -- or, equivalently, Eq. (S19) -- within Hartree-Fock (HF) mean-field theory (see, e.g., Chapter 2 of Ref. [S1]). We replace Hee with 5 (S29) (S30) (S31) (S32) (S33) (S34) N(cid:88) (cid:88) j=1 i H(HF) ee ≡ U − U † † † † † † † † j,scj,s(cid:104)c j,pcj,p(cid:105) + c j,pcj,p(cid:104)c j,scj,s(cid:105) − c j,scj,p(cid:104)c j,pcj,s(cid:105) − c j,pcj,s(cid:104)c j,scj,p(cid:105)] [c † † † † j,scj,p(cid:105)(cid:104)c j,pcj,p(cid:105) − (cid:104)c j,scj,s(cid:105)(cid:104)c [(cid:104)c j,pcj,s(cid:105)] . Each of the mean fields above can be written as † (cid:104)c j,αcj,β(cid:105) = 1 N e −i(k−k(cid:48))ja(cid:104)c † k,αck(cid:48),β(cid:105) . (cid:88) k,k(cid:48)∈BZ † † k,αck(cid:48),β(cid:105) = j,αcj,β(cid:105) is independent of the site index j (translational invariance), i.e. we take (cid:104)c We assume that (cid:104)c † δk,k(cid:48)(cid:104)c k,αck,β(cid:105). We are therefore naturally led to introduce the following quantities: † † M ≡ (cid:104)c j,pcj,p(cid:105) − (cid:104)c j,scj,s(cid:105) = 1 N † I ≡ (cid:104)c j,pcj,s(cid:105) = 1 N and † † ((cid:104)c k,pck,p(cid:105) − (cid:104)c k,sck,s(cid:105)) , † (cid:104)c k,pck,s(cid:105) , † † j,pcj,p(cid:105) + (cid:104)c n0 ≡ (cid:104)c j,scj,s(cid:105) = 1 N † † k,pck,p(cid:105) + (cid:104)c ((cid:104)c k,sck,s(cid:105)) . Under the assumption of homogeneity, we can rewrite the HF interaction term (S33) as k∈BZ (cid:88) (cid:88) (cid:88) k∈BZ k∈BZ (cid:20) (cid:88) (cid:88) k∈BZ n0 2 k∈BZ H(HF) ee = −U + U M 2 † † † k,sck,s) + Ic k,pck,p − c k,sck,p + I (c M2 − n2 † † k,pck,p + c (c k,sck,s) + U N (cid:18) 0 4 ∗ † c k,pck,s (cid:19) + I2 (S35) (S36) (S37) (S38) (cid:21) . K = HA0 − µ N , where N =(cid:80) (cid:80) The term proportional to n0/2 acts as a renormalization of the chemical potential µ in the grand-canonical Hamiltonian † α=s,p c k,αck,α is the total electron number operator. In this work we study only the phase diagram at half filling. We therefore have n0 = 1 ∀j = 1 . . . N in all phases and can discard such term. The last term in Eq. (12) instead must be retained (after discarding n0) since it takes different values in different phases. (It is a trivial constant: it therefore only matters when one compares total energies of different phases.) k∈BZ 6 The HF mean-field Hamiltonian (S38) can be written in a 2 × 2 fashion: (cid:88) (cid:16) k∈BZ (cid:17)(cid:18) H(HF) ee = U † † k,s c c k,p ∗ M/2 −I −I −M/2 (cid:19)(cid:18)ck,s (cid:19) ck,p (cid:18) (cid:19) + U N M2 4 + I2 . (S39) SECTION VI: DETAILS ON THE BOGOLIUBOV TRANSFORMATION In this Section we give all technical details relevant to the diagonalization of the problem posed by Eq. (9) in the main text, with Hee replaced by its HF mean-field expression (S39): H(HF) A0 ≡ H0 + H(HF) ee + Hph + g0√N † jp(a + a g2 0 2N † T (a + a ) − )2 . (S40) a An effective mean-field Hamiltonian for matter degrees of freedom can be obtained by taking the expectation value We seek ground-state wave functions of the unentangled form Ψ(cid:105) = ψ(cid:105)Φ(cid:105), where ψ(cid:105) is the wave function for the matter degrees of freedom and Φ(cid:105) is the analog for the e.m. field. In order to reduce the number of free parameters in the problem, we enforce particle-hole symmetry by setting Es = −Eg/2 = −Ep and ts = tp = t. of H(HF) over the light state Φ(cid:105), i.e. Heff−matter ≡ (cid:104)Φ H(HF) (cid:19) H(k) can be conveniently written in terms of ordinary 2 × 2 Pauli matrices {σi, i = 1, 2, 3}, i.e. H(k) =(cid:80) † + ωc (cid:104)Φa (cid:18)ck,s aΦ(cid:105) + U N † † k,s c c k,p (cid:88) + I2 . (S41) i hi(k)σi M2 4 A0 Φ(cid:105) = (cid:19) H(k) ck,p (cid:18) k∈BZ (cid:16) (cid:17) A0 with h1(k) = −U Re(I) , (cid:18) (cid:19) h2(k) = −2t sin(ka) 1 − g2 0 2 A2 and h3(k) = − Eg 2 − 2t cos(ka) In Eqs. (S42)-(S44) we have introduced (cid:18) − 2tg0 cos(ka)A1 + U Im(I) , (cid:19) + 2tg0 sin(ka)A1 + U M 2 g2 0 2 A2 1 − (S42) (S43) . (S44) and A1 ≡ † 1 √N (cid:104)Φa + a (cid:0)a + a †(cid:1)2 Φ(cid:105) Φ(cid:105) . A2 ≡ 1 N (cid:104)Φ The Hamiltonian (S41) can be diagonalized by introducing the following Bogoliubov transformation: † † † k,− = ukc γ k,s + vkc k,p , † † † ∗ ∗ k,s − u γ k,+ = v kc kc k,p , (S45) (S46) (S47) (S48) where uk = cos(θk/2) and vk = sin(θk/2)eiφk with cos(θk) = − sin(θk) = − , h3(k) (k) 1(k) + h2 h2 (k) 2(k) , eiφk = (k) = h1(k) + ih2(k) 2(k) h2 1(k) + h2 , h2 1(k) + h2 2(k) + h2 3(k) . (cid:112) (cid:112) (cid:113) 7 (S49) (S50) (S51) (S52) Note that uk and vk are functions of I, M, A1, and A2, i.e. uk = uk(I,M,A1,A2) and vk = vk(I,M,A1,A2). We find ξ(k)γ † k,ξ γk,ξ + C , (S53) (cid:88) k∈BZ (cid:88) (cid:19) ξ=± Heff−matter = (cid:18) M2 4 where theory, we find that The ground state of (S53) is ψ(cid:105) =(cid:81) where ∅(cid:105) = (cid:81) † k,pck,sψ(cid:105) = v∗ (cid:104)ψc + I2 C ≡ U N † k,− vac(cid:105), where vac(cid:105) is the state with no electrons. Mimicking the BCS aΦ(cid:105) . k∈BZ γ (S54) † + ωc (cid:104)Φa (cid:3) † k,pck,s ∅(cid:105) , (cid:89) (cid:2)uk + vkc ψ(cid:105) = k∈BZ † † † k,sck,sψ(cid:105) = uk2, (cid:104)ψc k,pck,pψ(cid:105) = vk2, and k∈BZ c k,s vac(cid:105). The following quantities (cid:104)ψc (cid:88) kuk are useful to write the order parameters in terms of uk and vk. We find (cid:88) I = 1 N k∈BZ † (cid:104)ψc k,pck,sψ(cid:105) = 1 N ∗ kuk v k∈BZ and M = 1 N (cid:88) k∈BZ † † k,pck,pψ(cid:105) − (cid:104)ψc ((cid:104)ψc k,sck,sψ(cid:105)) = (cid:88) (cid:0) k∈BZ 1 N We also write the expectation values of jp and T over the HF state ψ(cid:105) in terms of uk and vk: vk2 − uk2(cid:1) . kvk)(cid:3) kvk)(cid:3) . k∈BZ (cid:88) (cid:2) ∗ −t sin(ka)(vk2 − uk2) − 2t cos(ka)Im(u (cid:88) (cid:2)t cos(ka)(cid:0) vk2 − uk2(cid:1) ∗ − 2t sin(ka)Im(u J ≡ and aN (cid:104)ψjpψ(cid:105) = 2 N T ≡ (cid:104)ψ T ψ(cid:105) = 2 N k∈BZ Note that both J and T have units of energy and are finite in the thermodynamic N → ∞ limit. Following exactly the same steps described in the proof of the no-go theorem in the main text and defining and ∆ = − (cid:114) g2 0 2 T λ = 1 + 4∆ ωc , (S55) (S56) (S57) (S58) (S59) (S60) (S61) we find that Φ(cid:105) must be a coherent state ¯β(cid:105), i.e. b ¯β(cid:105) = ¯β ¯β(cid:105), with √N , g0 ¯β = − λ3/2 Jωc 8 (S62) to be compared with Eq. (4) in the main text. As in the case of the proof of the no-go theorem, b = cosh(x)a+sinh(x)a†, with cosh(x) = (λ + 1)/(2√λ) and sinh(x) = (λ − 1)/(2√λ). The inverse transformation reads as following: a = cosh(x)b − sinh(x)b†. Note that J depends on ¯β and therefore the previous equation defines ¯β only implicitly. Since we have found the ground state Φ(cid:105) (i.e. a coherent state ¯β(cid:105) of the b operator), we can evaluate Eqs. (S45)- (S46): 2 ¯β √N λ 1 and A1 = ¯β(cid:105) = † √N λ(cid:104) ¯βb + b (cid:0)b + b †(cid:1)2 To derive Eq. (S64) we have used that (cid:0)b + b†(cid:1)2 = b2 + b†2 + 2b†b + 1. Using Eqs. (S53)-(S54) and using that a†a = (λ2 + 1)b†b/(2λ)− (λ2 − 1)(b2 +b†2)/(4λ) + (λ− 1)2/(4λ), we can also write the ground-state energy per particle 1 N λ (cid:104) ¯β λ2 Jωc ¯β(cid:105) = = A2 4 ¯β2 + 1 A2 = 1 λN = − (S63) (S64) 1 + N λ 2g0 . as . (S65) GS = EGS N 1 N = − (k) + (cid:88) k∈BZ (cid:90) +π/a In the thermodynamic N → ∞ limit we find A2 = A2 dk 2π N→∞ GS = −a lim −π/a 4λN ωc[4 ¯β2 + (λ − 1)2] (cid:18) ωc 4 A2 1 + U (k) + (cid:19) + I2 (cid:19) (cid:18) + U M2 4 M2 4 + I2 1 (i.e. the vacuum contribution can be neglected) and . (S66) Since (k) depends only on ka, it is useful to change integration variable in Eq. (S66) from k to k(cid:48) = ka ∈ (π, +π). Eqs. (S56), (S57), (S58), (S59), (S63), and (S64) fully determine all the relevant quantities in the problem, i.e. I, M, J , and T . In Fig. S1 we present a summary of our main results for the bands ±(k) both in the simple non-interacting U = 0 case -- panel a) -- and in the interacting U (cid:54)= 0 case -- panel b). SECTION VII: OPTICAL CONDUCTIVITY, DRUDE WEIGHT, AND THE f -SUM RULE In this Section we discuss the optical conductivity σ(ω) and the f -sum rule for the EFK model. The longitudinal conductivity σ(ω) is defined as the response of the physical charge current operator to the electric field E(t) = −c−1∂A(t)/∂t. Assuming A(t) = Aωe−iωteηt+c.c. with η = 0+ (as usual, for the applicability of linear response theory the applied field must vanish in the far past [S1, S2]), we have E(t) = ic−1(ω +iη)Aωe−iωteηt +c.c. = Eωe−iωteηt +c.c.. We therefore find that the response of the physical current is given by − eδJphys(ω) ≡ σ(ω)Eω = i c σ(ω)(ω + iη)Aω . (S67) We conclude that the pre-factor in front of Aω in the right-hand side Eq. (S67) can be calculated from the current- current response function [S2, S34], with its paramagnetic and diamagnetic contributions. Here, we focus on the EFK model, i.e. Eq. (9) for g0 = 0. Using Eq. (S67) and linear response theory [S1, S2], we immediately find that the optical conductivity of the EFK model is given by i σ(ω) = (Pm − Pn)(cid:104)ψnjpψm(cid:105)2 ω − ωnm + iη denotes a thermal average, and Pn = exp(−βEn)/Z, with β = (kBT )−1 and Z = (cid:80) where ψn(cid:105) are the exact eigenstates of the Hamiltonian H0 + Hee with eigenvalues En, (cid:104). . .(cid:105) ≡ e2a2 2L (cid:104)− T (cid:105) + e2 L ω + iη ω + iη n,m i , (cid:80) n Pn(cid:104)ψn . . .ψn(cid:105) n exp(−βEn) is the canonical (S68) (cid:88) (a) (b) 9 FIG. S1. (Color online) Panel (a) The non-interacting U = 0 spectrum ξ(k) = ξ(k) = ±(k) (in units of Eg) as a function of ka in the first BZ, ka ∈ (−π/π). The red dashed line is the spectrum in an insulating non-interacting case t < Eg/4, t = 0.1 Eg, and t = 0. The black dotted line shows the metallic phase t > Eg/4, t = 0.5 Eg, and t = 0. The blue solid line -- obtained by setting t = 0.5 Eg and t = 0.1 Eg -- shows that a finite value of t opens a single-particle hybridization gap. Panel (b) A comparison between the non-interacting and the interacting spectrum. The black dotted line is the non-interacting spectrum (i.e. obtained by setting U = 0), for t = 0.5 Eg and t = 0.1 Eg. The red solid line is the HF mean-field spectrum obtained for the same values of t and t, at U = Eg (i.e. U/t = 2). partition function. In deriving the exact eigenstate representation (S68) we have used that j† ∗. (cid:104)ψnjpψm(cid:105) = (cid:104)ψmjpψn(cid:105) P0 = 1), we finally find: Separating the real and imaginary parts of σ(ω) and taking the zero-temperature limit (Pn = 0 for n (cid:54)= 0 and p = jp and therefore (cid:88) n(cid:54)=0 (cid:104)ψnjpψ0(cid:105)2 En − E0 (cid:88) 2πe2 L n(cid:54)=0 (S69) (S70) Re[σ(ω)] = Dδ(ω) + πe2 L [δ(ω − ωn0) + δ(ω + ωn0)] where D is the so-called Drude weight [S3 -- S6] D = πe2a2 2L (cid:104)ψ0 − T ψ0(cid:105) − (cid:104)ψnjpψ0(cid:105)2 En − E0 ≡ Dd + Dp . Here, Dd (Dp) defines the diamagnetic (paramagnetic) contribution to D. We immediately notice the f -sum rule [S1, S2, S4 -- S6] (cid:90) +∞ (cid:90) +∞ (cid:88) −∞ dωRe[σ(ω)] = 2 dωRe[σ(ω)] = D + 0 2πe2 L n(cid:54)=0 (cid:104)ψnjpψ0(cid:105)2 En − E0 = Dd . (S71) We now show that the f -sum rule is satisfied in our HF treatment of the EFK model. In the absence of light, the complete HF Hamiltonian including electron-electron interactions and neglecting an irrelevant constant is (see Eq. (S53)): ξ(k)γ † k,ξ γk,ξ . (S72) † ξ,k vac(cid:105) and ξ(k) = ξ(k). We remind the reader that (k) has been The eigenstates and eigenvalues are ξ, k(cid:105) = γ defined in Eq. (S52) and, in this Section, needs to be evaluated at g0 = 0. The ground state, as noticed above, is ψ0(cid:105) =(cid:81) k∈BZ γ † −,k vac(cid:105). We have Dp = − 2πe2 N a (cid:104)+, kjp−, k(cid:105)2 2(k) (S73) (cid:88) (cid:88) k∈BZ ξ=± HHF = (cid:88) k∈BZ −π−π/20π/2πka−1−0.500.51ξ(k)[Eg]−π−π/20π/2πka−1−0.500.51ξ(k)[Eg](a) (b) 10 (c) (d) FIG. S2. (Color online) (a) The quantity 2(k) (in units of Eg) as a function ka in the first BZ. Different curves refer to different values of the Hubbard U parameter: U = Eg/10 (black), U = Eg (red), and U = 2Eg (blue). (b) The Drude weight D (black solid line) and the contributions Dp (blue dash-dotted line) and Dd (red dashed line) are plotted as functions of U/Eg. The three quantities D, Dp, and Dd are in units of e2aEg/2. (c) The smooth contribution σ (ω) -- see Eq. (S78) -- to the real part of the optical conductivity (in units of e2aEg/) is plotted as a function ω/Eg. Different curves refer to different values of the Hubbard U parameter: U = Eg/10 (black solid line), U = Eg (red solid line), and U = 2Eg (blue solid line). Black, red, and blue vertical dashed lines mark the energy EVHS = mink∈BZ[2(k)] at which a logarithmic divergence of σ (ω) occurs. Clearly, EVHS shifts with U . (d) The paramagnetic contribution to the Drude weight Dp (black solid line, in units of e2aEg/2) is (cid:48) compared with the quantity S p (red circles) defined in Eq. (S79). All numerical results in this figure have been obtained by setting t = Eg/2 and t = Eg/10. (cid:48) (cid:48) and (cid:88) k∈BZ Dd = πe2a 2N (cid:104)−, k − T −, k(cid:105) . (S74) The quantity 2(k) is the energy necessary to promote an electron with wave number k vertically from the lower band ξ = − to the upper band ξ = +. Fig. S2(a) shows the quantity 2(k) (in units of Eg) as a function of ka. The extrema of 2(k) give rise to logarithmic divergences in the optical conductivity. In order to calculate both contributions to the Drude weight, we need the following matrix elements: (cid:104)+, kjp−, k(cid:105) = 2a eiφk [t sin(ka) sin(θk) + t cos(ka) cos(θk) sin(φk) + it cos(ka) cos(φk)] and (cid:104)−, k − T −, k(cid:105) = [t cos(ka) cos(θk) + t sin(ka) sin(θk) sin(φk)] , (S75) (S76) where θk and φk are the Bogoliubov angles defined in Sec. VI. From Eq. (S75), we notice that, for t = 0 and U = 0, one has (cid:104)+, kjp−, k(cid:105) = 0. This is expected since, in the absence of many-body effects and for t = 0, the eigenstates −1.0−0.50.00.51.0k[π/a]01234562(k)[Eg]0.00.51.01.52.0U[Eg]−1.5−1.0−0.50.00.51.0D[e2aEg/¯h2]012345¯hω[Eg]0.00.51.01.52.0σ0(ω)[e2a/¯h]00.511.52U[Eg]−2−1.5−1−0.50Dp[e2aEg/¯h2]have no orbital mixing (i.e. sin(θk) = 0). Switching on t or U , however, yields (cid:104)+, kjp−, k(cid:105) (cid:54)= 0. In particular, for t = 0, repulsive interactions allow (cid:104)+, kjp−, k(cid:105) (cid:54)= 0 when sin(θk) (cid:54)= 0, i.e. for 0 < U < UXC = Uc2(0), since, from Eqs. (S42)-(S44), one has sin2(θk) = [UI/(k)]2. Using Eqs. (S75) and (S76), it is possible to calculate Dp and Dd, and therefore D. Fig. S2(b) shows these quantities as functions of U/Eg. We now calculate the smooth part σ(cid:48)(ω) of Re[σ(ω)], (cid:48) σ (ω) ≡ πe2 N a (cid:104)+, kjp−, k(cid:105)2 2(k) [δ(ω − 2(k)/) + δ(ω + 2(k)/)] . (S77) Note that σ(cid:48)(ω) = Re[σ(ω (cid:54)= 0)]. Because of δ(ω ∓ 2(k)/) in the integrand of Eq. (S77), the integral over k (in the thermodynamic limit) can be carried out analytically. We find (cid:88) k∈BZ (cid:88) i (cid:80)3 (cid:104)+, kjp−, k(cid:105)2 j=1 hj(k)∂khj(k) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)k=ki(ω) (cid:48) σ (ω) = e2 8 , (S78) where ki(ω) are the solutions of (ki(ω)) = ω/2 and the quantities h1(k), h2(k), and h3(k) have been defined in Eqs. (S42)-(S44). Fig. S2(c) shows σ(cid:48)(ω) as a function ω/Eg. Each vertical dashed line marks the energy EVHS = mink∈BZ[2(k)] at which the quantity 2(k) is minimal. At this energy a logarithmic enhancement of σ(cid:48)(ω) occurs. (Similarly, another singularity occurs at E(cid:48) VHS = maxk∈BZ[2(k)], but that is weaker in our numerical calculations.) Using Eq. (S78) we can finally calculate numerically the quantity (cid:90) +∞ (cid:48) p = −2 S 0 (cid:48) (ω) . dω σ 11 (S79) (S80) (S81) (S82) (S83) We have verified numerically that This is seen in Fig. S2(d). It follows that which is exactly the f -sum rule (S71). (cid:48) p = Dp . S (cid:90) +∞ 2 0 dωRe[σ(ω)] = Dd , SECTION VIII: ON THE PHASE DIAGRAM OF THE EFK MODEL In this Section we demonstrate the existence of two critical values of U , Uc1 and Uc2, at which I = 0, where I has been introduced in the self-consistent field equation (Eq. (14)) of the main text and Eq. (S56). The latter yields the following equations: and (cid:34) 1 N 1 − (cid:34) (cid:88) (cid:35) k∈BZ 1 − (cid:88) 1 N U k∈BZ 2(k) (cid:35) U 2(k) Re(I) = 0 (cid:88) sin(ka) k∈BZ (k) . 1 Im(I) = −t N In the absence of hybridization, i.e. for t = 0, the previous expressions become identical. This implies a degeneracy with to respect the phase of I. For t (cid:54)= 0, these equations can be satisfied by solutions of the HF equations which yield Im(I) = 0 and (−k) = (k). The latter condition implies that the left-hand side of Eq. (S83) vanishes. All the solutions we find are of this type. The condition Re(I) (cid:54)= 0 implies that the following equation must be satisfied: (cid:88) 1 N 1 − U = 0 . k∈BZ 2(k) 12 (S84) Before proceeding to prove the existence of Uc1 and Uc2 we write Eq. (S84) in a more appealing form. We define ψk = Re(I)/[2(k)], and we rewrite Eq. (S84) as U ψk(cid:48) . (S85) (cid:88) k(cid:48)∈BZ 2(k)ψk = 1 N This establishes an immediate link between Eq. (S84) and the equation for the exciton binding energy. We first demonstrate the existence of upper critical value of U , i.e. Uc2(t). Let us first set t = 0. For U/t (cid:29) 1, the system is in a trivial insulating phase in which all electrons occupy the s band and M = M0 ≡ −1. Upon decreasing U down to UXC = Uc2(0), the system develops an infinitesimal excitonic order parameter. The value UXC at which this occurs can be found by solving Eq. (S84) for an infinitesimal I, i.e. (cid:88) 1 N UXC = 1 , k∈BZ 4t cos(ka) + Eg + UXC or, in the thermodynamic limit, (cid:90) π −π dx 2π 1 cos(x) + [Eg + UXC]/(4t) = 4t UXC . Carrying out the integral analytically we find UXC = 8t2 Eg − Eg 2 . (S86) (S87) (S88) Corrections to UXC and M0 due to t (cid:54)= 0 can be found perturbatively in the limit t/t (cid:28) 1. They start at second order in the small parameter t/t: δUc2(t) = (t/t)2u and δM0(t) = (t/t)2m0. The latter is the change in the electronic polarization from the value M0 = −1 in the limit t = 0. We find m0 = Eg/(2UXC) and u = E3 g + 16t2)]. For example, for t = Eg/2 we find m0 = 1/3 and u = 2/7. In conclusion, we have g /[t(3E2 Uc2(t) = UXC + E3 t(3E2 g (t/t)2 g + 16t2) + O(t3) . (S89) We now demonstrate the existence of a lower critical value of U , i.e. Uc1(t). Following similar steps to the ones above, one can demonstrate that there is also a lower-threshold for the existence of the exciton insulating phase. Up to leading order in an asymptotic expansion for small t/t (and under the single-particle condition t > Eg/4 discussed in Fig. S1) we find . (S90) We clearly see that limt→0 Uc1(t) = 0. But, for finite t/t, Uc1(t) (cid:54)= 0. We will come back to Eq. (S90) below. of validity. We have checked that the analytical results (S89) and (S90) match very well our numerical results, in their regime SECTION IX: PSEUDOSPIN ANALYSIS In this Section we present a few more remarks on the ground state of the EFK model in the HF approximation. We view the mean-field problem as a variational problem and use a trial ground-state wave function of the form (cid:89) ψ(cid:105) = k∈BZ † k,−vac(cid:105) . γ (S91) (cid:113) 4t2 − E2 ln(t/t) g /4 π Uc1(t) → We then express the full Hamiltonian of the 1D EFK model defined by Eq. (S20) in terms of the Bogoliubov operators † k,± and γk,±: γ 13 † k,s = uk γ c † ∗ c k,p = v k γ † k,− − vk γ † ∗ k,− + u k γ † k,+ , † k,+ , (S92) (S93) where uk = cos(θk/2) and vk = sin(θk/2)eiφk . By writing the Hamiltonian in its normal ordered form, exploiting the following property of the variational wave function † k1,λ1 . . . γ † kn,λn γkn+1,λn+1 . . . γk2n,λ2nψ(cid:105) = (cid:104)ψγ 2n(cid:89) j=1 δλj ,−(cid:104)ψγ † k1,− . . . γ † kn,−γkn+1,− . . . γk2n,−ψ(cid:105) , and enforcing particle-hole symmetry, we find the following ground-state energy: (cid:88) (cid:8)[Eg/2 + 2t cos(ka)] cos(θk) + 2t sin(ka) sin(θk) cos(φk)(cid:9) sin(θk(cid:48)−q/2) cos(θk+q/2) cos(θk/2) sin(θk(cid:48)/2)ei(φk(cid:48)−φk(cid:48)−q) (cid:104)ψγ k∈BZ (cid:104)ψ Hψ(cid:105) = (cid:88) U N k,k(cid:48),q∈BZ + † k,−γk,−ψ(cid:105) (cid:104)ψγ † k(cid:48)−q,−γ † k+q,−γk,−γk(cid:48),−ψ(cid:105) . Using the ansatz (S91) in the previous equation and the properties † k,−γ † k,−ψ(cid:105) = 1 , (cid:104)ψγ † k(cid:48)−q,−γ † k+q,−γk,−γk(cid:48),−ψ(cid:105) = δq,0 − δq,k(cid:48)−k , (cid:104)ψγ we finally find (cid:104)ψ Hψ(cid:105) = − (cid:88) k∈BZ U N (cid:88) k,k(cid:48)∈BZ [Eg/2 + 2t cos(ka)] cos(θk) + 2t sin(ka) sin(θk) cos(φk) + cos2(θk/2) sin2(θk(cid:48)/2) sin(θk/2) cos(θk/2) sin(θk(cid:48)/2) cos(θk(cid:48)/2) cos(φk(cid:48) − φk) . (S94) (S95) (S96) (S97) (S98) (cid:88) U N k,k(cid:48)∈BZ (cid:88) k,k(cid:48)∈BZ We therefore note that the ground-state energy can be written in a form that resembles the energy of a chain of classical interacting spins in an external magnetic field, i.e. E[τk] ≡ (cid:104)ψ Hψ(cid:105) − U N 4 = − [By(k)τ y k + Bz(k)τ z k ] − U 4N τk · τk(cid:48) , (S99) (cid:88) k∈BZ where B(k) = [0, 2t sin(ka), Eg/2 + 2t cos(ka)]T and τk is a unit vector with Cartesian components τ x k ≡ † † † † τ y s,kcp,kψ(cid:105) = sin(θk) sin(φk), and τ z (cid:104)ψc s,kcp,kψ(cid:105) = sin(θk) cos(φk), k ≡ i(cid:104)ψc p,kcs,k − c p,kcs,k + c k = † † (cid:104)ψc p,kcp,kψ(cid:105) = cos(θk). Notice that the "lattice" of spins is in momentum rather than real space. s,kcs,k − c In this pseudospin description, the repulsive Hubbard-U interaction becomes a ferromagnetic rotationally-invariant spin- spin interaction term. The spin configuration which minimizes the energy satisfies the self-consistent field equations τ x k = − τ y k = − τ z k = − (cid:35)2 (cid:34) (cid:88) U τ x k , 2µkN k∈BZ By(k) µk − Bz(k) µk − U 2µkN U 2µkN (cid:88) k∈BZ U 2N τ x k + By(k) − k∈BZ (cid:88) (cid:88) (cid:35)2 k∈BZ τ y k , τ z k , (cid:34) τ y k + Bz(k) − (S100) (S101) (S102) (S103) (cid:35)2 τ z k . (cid:88) k∈BZ U 2N where (cid:118)(cid:117)(cid:117)(cid:116)(cid:34) (cid:88) k∈BZ U 2N µk = − The quantities I and M can also be expressed in terms of pseudospins: and M = − 1 N (cid:88) k∈BZ I = 1 2N k − iτ y (τ x k ) (cid:88) τ z k . k∈BZ 14 (S104) (S105) Within this description, we consider two limiting cases. Firstly, we consider the limit U (cid:29) Eg, t, t. Neglecting Eg, t, t with respect to U , it follows that the following configurations and τ (n) k = − z , τ (fx) k = x , τ (fy) k = y , (S106) (S107) (S108) are degenerate (i.e. E[τ (n) The configuration corresponding to τ (n) and τ (fy) of the form ] = −U N/2). For U (cid:29) Eg, t, t the system is invariant under rotations. describes the normal phase (I = 0), while the ones corresponding to τ (fx) correspond to HF states with Re(I) (cid:54)= 0 and Im(I) (cid:54)= 0, respectively. This implies that all configurations k ] = E[τ (fx) ] = E[τ (fy) k k k k k τ (θ,φ) k = cos(θ)τ (n) k + sin(θ) cos(φ)τ (fx) k + sin(θ) sin(φ)τ (fy) k (S109) are degenerate. By turning on Eg, t and t, and treating them as weak perturbations, we find that the energy associated to the configurations (S109) is given by E[τ (θ,φ) ] = −[U + Eg cos(θ)]N/2. This means that the gap energy Eg makes the normal phase expressed in (S106) energetically preferred. This simple example shows why at large values of the Hubbard-U parameter, the HF phases with I (cid:54)= 0 do not occur. Before concluding, we discuss a second limiting case. We set Eg = 0 (which is compatible with the condition t > Eg/4 described in Fig. S1) and we assume 0 < t < t. Under these conditions, the external magnetic field B(k) k∈BZ B(k) = 0, but B(k) (cid:54)= 0 ∀k if t, t (cid:54)= 0. The spin configuration which minimizes the energy is lays on the y- z plane and and its average value is zero, i.e. N−1(cid:80) k τ x k = − τ y k = − τ z k = − µk = − (cid:88) k∈BZ (cid:113) U 2N (cid:112) t U = 2π U Re(I) µk , 2t sin(ka) , µk , µk 2t cos(ka) (cid:113) [U Re(I)]2 +(cid:2)2t sin(ka)(cid:3)2 [U Re(I)]2 +(cid:2)2t sin(ka)(cid:3)2 (cid:115) K 1 1 1 + [U Re(I)]2/(4t2) (S110) (S111) (S112) (S113) + [2t cos(ka)]2 , + [2t cos(ka)]2 = 1 . (S114)  , (S115) 1 − (t/t)2 1 + [U Re(I)]2/(4t2) with Re(I) (cid:54)= 0 only if the following implicit equation is satisfied: In the thermodynamic limit the latter can be rewritten as where K(x) is the complete elliptic integral of the first kind. Inspecting the right-hand side of the previous equation, one finds that, for fixed values of t, t, and U , (cid:18)(cid:113) (cid:19) t U ≤ 0 ≤ t Uc1(t) ≡ 1 2π K 1 − (t/t)2 , (S116) where Uc1 is the minimum value of the Hubbard U parameter which gives Re(I) (cid:54)= 0. For small values of t we find 15 (cid:113) Uc1(t) → 2πt ln(t/t) . (S117) Note the logarithmic divergence, as we has seen previously in Eq. (S90). The only role of Eg (cid:54)= 0 is to replace 2t → g /4 in Eq. (S117). 4t2 − E2 If 0 < U < Uc1, the configuration which minimizes the energy is τ x k = 0 , τ y k = − τ z k = − µk = − , , 2t sin(ka) µk 2t cos(ka) (cid:113)(cid:2)2t sin(ka)(cid:3)2 µk (S118) (S119) (S120) (S121) + [2t cos(ka)]2 , which means that Re(I) = 0. This second liming case well describes what occurs in the EFK model in the HF approximation for U ∼ Uc1. [S1] G.F. Giuliani and G. Vignale, Quantum Theory of the Electron Liquid (Cambridge University Press, Cambridge, 2005). [S2] D. Pines and P. Nozi`eres, The Theory of Quantum Liquids (W.A. Benjamin, Inc., New York, 1966). [S3] W. Kohn, Phys. Rev. 133, A171 (1964). [S4] B.S. Shastry and B. Sutherland, Phys. Rev. Lett. 65, 243 (1990). [S5] A.J. Millis and S.N. Coppersmith, Phys. Rev. B 42, 10807(R) (1990). [S6] R.M. Fye, M.J. Martins, D.J. Scalapino, J. Wagner, and W. Hanke, Phys. Rev. B 44, 6909 (1991).
1008.1979	2	1008	2010-11-15T16:55:52	Local Charge of the nu=5/2 Fractional Quantum Hall State	[ "cond-mat.mes-hall", "cond-mat.str-el" ]	Electrons in two dimensions and strong magnetic fields effectively lose their kinetic energy and display exotic behavior dominated by Coulomb forces. When the ratio of electrons to magnetic flux quanta in the system is near 5/2, the unique correlated phase that emerges is predicted to be gapped with fractionally charged quasiparticles and a ground state degeneracy that grows exponentially as these quasiparticles are introduced. Interestingly, the only way to transform between the many ground states would be to braid the fractional excitations around each other, a property with applications in quantum information processing. Here we present the first observation of localized quasiparticles at nu=5/2, confined to puddles by disorder. Using a local electrometer to compare how quasiparticles at nu=5/2 and nu=7/3 charge these puddles, we are able to extract the ratio of local charges for these states. Averaged over several disorder configurations and samples, we find the ratio to be 4/3, suggesting that the local charges are e/3 at seven thirds and e/4 at five halves, in agreement with theoretical predictions. This confirmation of localized e/4 quasiparticles is necessary for proposed interferometry experiments to test statistics and computational ability of the state at nu=5/2.	cond-mat.mes-hall	cond-mat	Local Charge of the ν = 5/2 Fractional Quantum Hall State Vivek Venkatachalam, Amir Yacoby, Loren Pfeiffer†, Ken West† * Department of Physics, Harvard University, Cambridge, MA, USA † Department of Electrical Engineering, Princeton University, Princeton, NJ, USA Abstract Electrons in two dimensions and strong magnetic fields effectively lose their kinetic energy and display exotic behavior dominated by Coulomb forces. When the ratio of electrons to magnetic flux quanta in the system is near 5/2, the unique correlated phase that emerges is predicted to be gapped with fractionally charged quasiparticles and a ground state degeneracy that grows exponentially as these quasiparticles are introduced. Interestingly, the only way to transform between the many ground states would be to braid the fractional excitations around each other, a property with applications in quantum information processing. Here we present the first observation of localized quasiparticles at ν = 5/2, confined to puddles by disorder. Using a local electrometer to compare how quasiparticles at ν = 5/2 and ν = 7/3 charge these puddles, we are able to extract the ratio of local charges for these states. Averaged over several disorder configurations and samples, we find the ratio to be 4/3, suggesting that the local charges are 7/3 = e/3 and e∗ e∗ 5/2 = e/4, in agreement with theoretical predictions. This confirmation of localized e/4 quasiparticles is necessary for proposed interferometry experiments to test statistics and computational ability of the state at ν = 5/2. When a two-dimensional electron system (2DES) is subject to a strong perpendicular magnetic field, the physics that emerges is controlled by interelectron Coulomb interactions. If the 2DES is tuned such that the ratio of electrons to magnetic flux quanta in the system (ν) is near certain rational values, the electrons condense into so-called fractional quantum Hall (FQH) phases [1]. These strongly-correlated states are gapped and incompressible in the bulk of the sample, but metallic and compressible along the sample boundary, allowing current to flow around the perimeter in such a way that the transverse conductance is precisely quantized to Gxy = ν(e2/h). Additionally, the electronic correlations encoded in FQH states give rise to local excitations with a fraction of an electron charge and braiding statistics that fall outside the conventional classification of bosonic or fermionic. The state at ν = 5 2, unlike its conventional odd- denominator relatives, is predicted to have the additional property that particle interchange can evolve the system adiabatically between orthogonal ground states [2]. This property, dubbed non-abelian braiding statistics, has been proposed as the basis for a topological quantum computer that would be insensitive to environmental decoherence [3, 4]. One necessary (but insufficient) condition for exotic braiding statistics at ν = 5 2 is for the ground state to support local excitations with a charge of e∗ 5/2 = e/4, where e is the charge of an electron [2]. Though a charge of e/4 had previously been measured using shot noise techniques [5], more recent data from the same group [6] suggest that the value of the measured charge changes continuously as the point contact conductance and temperature are varied, reaching an inferred charge of unity in the weak and strong tunneling limits. Unexpected charges have also been reported for the more conventional fractions at 1/3, 2/3, and 7/3 [7, 6]. Moreover, DC conductance measurements in the weak tunneling regime [8] suggest a quasiparticle charge of e∗ 5/2 = 0.17e, in stark contrast to the shot noise results. Clearly, a better understanding of the tunneling processes that take place between quantum Hall edges in the quantum point contact is needed in order to interpret the shot noise results. Alternatively, one can employ a thermodynamic approach [9] that probes the quasiparticle charge in the bulk of the sample in order to infer quasiparticle charge. Here we use a single electron transistor as a sensitive electrometer to measure the 1 equilibrium charge distribution in the bulk and its dependence on the average density and magnetic field. Our results provide clear evidence for localized charge e/4 quasiparticles at ν = 5/2. Our measurement employs a fixed single electron transistor (SET) as a gated device capable of sensitively measuring the local incompressibility (κ−1 = ∂µ ∂n) of a high-mobility 2DES [10]. The 2DES has a 200 nm deep, 30 nm wide MBE-grown GaAs/AlGaAs quantum well, with symmetric Si δ-doping layers 100 nm on either side. A metallic backgate grown 2 µm below the 2DES allows us to tune the global density, n, in the well over a typical range of 2.3− 2.5× 1015m−2, with some variation between samples. The SET is fabricated on top of the sample using standard electron beam lithography and shadow-evaporation techniques (Figure 1), creating an island with dimensions 500 nm × 80 nm. All measurements were carried out in a dilution refrigerator with an electron temperature of 20 mK, verified using standard Coulomb blockade techniques. As we adjust the density and magnetic field we expect to see regions of incompressibility when a gap is present, which will only happen precisely when the system is in a QH state. The slope of these incompressible regions in the nB-plane corresponds to the filling factor of the state [11]. Figure 2 shows incompressibility versus density and magnetic field between ν = 2 and ν = 3, with the two highlighted regions corresponding to FQH states at ν = 5/2 and ν = 7/3. Additionally, due to the rough disorder potential created by remote donors, we can expect different points in space to develop gaps at different values of the global density. Because of this, we expect a well-developed QH state to have a percolating incompressible region punctured by small compressible puddles which behave as either dots or anti-dots [11]. As the global density is varied, a given compressible puddle will occasionally be populated by quasiparticles or quasiholes of the surrounding incompressible state. This creates a jump ∂n. The magnitude and in the local chemical potential, µ(n), and a spike in the local incompressibility ∂µ spacing of these spikes is determined by the charging spectrum of the puddle, which in turn is dictated by the quasiparticle charge in the surrounding incompressible region. Namely, if the quasiparticle charge were reduced by a factor of three for a fixed disorder potential, we should see three times as many compressible spikes as a function of global electron density (Figure 1 b,c). This difference in spike frequencies has previously been used to measure the local charge at ν = 1/3 and ν = 2/3 [9]. Unlike shot noise measurements [7], these local compressibility measurements find a quasiparticle charge of e/3 at both filling factors. Additionally, because of the spatial resolution afforded by the scanning technique in that measurement, it was possible to establish that the disorder potential landscape does not change as the electron system is tuned between Hall states with comparable gaps. Transport measurements confirm that the gap inferred from activation of Rxx minima is comparable for the states at 5/2 and 7/3 [12, 13], so we can expect similar potential landscapes for the two states. Our procedure begins with obtaining charging spectra (incompressibility versus density) at ν = 5/2 and ν = 7/3. Because the gap for these states is comparable, and the disorder potential is not altered as we change the magnetic field or density, we expect the spacing between charging features to reflect the quasiparticle charge in each state. In the limit of an isolated compressible puddle surrounded by an incompressible fluid, this relationship is particularly simple - if the ratio of local charges between the two spectra is β, the spectra should be identical after one of the density axes is rescaled by a factor of β, and shifted by some amount (Figure 3a). To proceed, we choose a value of β and stretch one of the spectra by this factor. We then calculate the correlation between the two spectra as a function of density offset and record the highest value. Finally, we repeat this for many scaling factors to obtain quality-of-fit versus β, as depicted in Figure 3b. This procedure was repeated for 20 different disorder configurations, obtained by changing samples, mea- suring with different SETs, or thermal cycling to change the disorder. A summary of the data is shown in Figure 4a, with an average over the measured ensemble in Figure 4b. The peak observed at β = 1.31 suggests a charge ratio of 4:3 between the two states, and a qualitative inspection of spectra overlap (as in Figure 3a) corroborates this. To determine the significance of the peak value, we repeated our analysis with pairs of spectra from different disorder configurations, which should be less correlated. For each scale, we characterized the distribution of best correlations with a mean and standard deviation. These, in turn, can be simply converted to the expected mean and standard error for our data (if it were uncorrelated). (cid:18) (cid:104)C1(x) C2(x)(cid:105) (cid:19) √ (cid:104)C1(x)2(cid:105)(cid:104)C2(x)2(cid:105) 2 The 1σ region around the uncorrelated mean is depicted in red in Figure 3b. Our averaged correlation at β = 1.31 lies 3.8 standard errors above the uncorrelated mean, corresponding to a one-tailed P-value of 7 × 10−5. Assuming a charge of e∗ 5/2 = (e/3)/(1.31) = 0.254e, in agreement with the Moore-Read prediction of e∗ To better understand why some configurations seem to provide weaker (and sometimes different) measure- ments of β, it helps to abandon the assumption that we are charging and monitoring single puddles, as well as the assumption that quasiparticles in different puddles do not interact. A free energy for our system that takes these into account is given by 7/3 = e/3, this measured value of β suggests e∗ 5/2 = e/4 [2]. (cid:88) (cid:88) VijQiQj −(cid:88) ∆ (cid:22) Qi (cid:23) . 2 (cid:88) i F = (i − VBG)Qi + 1 2 UiQi(Qi − 1) + i i<j i Here, Ui and i are the onsite interaction (self-capacitance) and bare disorder potential for puddle i respec- tively. Vij is a pairwise interaction, or cross-capacitance, between puddles i and j, and ∆ is the energy gained by forming a bound pair of quasiparticles. For now, we will let ∆ = 0. We assume that some subset of the puddles is capacitively coupled to and measured by the SET. To compute charging spectra from this model, we first choose values of U, V , and for each puddle from Gaussian distributions. We then discretize Qi into units of e/3 or e/4 and determine how many units of charge to put in each puddle to minimize the above free energy. This is done for each value of VBG and converted into a charging spectrum. Finally, we can take the resulting spectra and repeat the processing performed on data to obtain summary statistics for comparison. The result, with = 0 ± .3U and Vij = 0.3U ± 0.2U, is shown in Figure 4c. Results for other parameter choices in a large range are qualitatively similar, with smaller values of σ and Vij corresponding to sharper peaks and less spread. As expected, these simulations tell us that both and Vij can distort spectra in such a way that the maximum cross-covariance will shift slightly or even dramatically away from 4/3. Still, we should always expect some weight at 4/3, and this can be extracted by averaging over disorder configurations (Figure 4d). Recently, there has been some suggestion that e/2 quasiparticles at the ν = 5/2 edge may be present and relevant to interference measurements [14]. In the context of our model, we can consider the effect weak binding of quasiparticles would have on measured spectra. This binding is parameterized by ∆ above, and we only consider the case where pairing affects the e/4 quasiparticles. As the strength of pairing is increased relative to the onsite interaction (Figure 4d), we expect weight to shift from the peak at 4/3 to a peak at 2/3 (corresponding to e/2 quasiparticles), with considerable weight at 2/3 even when ∆ = 0.1 U. Our data show no appreciable evidence for a peak at 2/3, suggesting that the only quasiparticles participating in localization are have charge e/4. These measurements constitute the first direct measurement of incompressibility and localized states at ν = 5/2, and provide an equilibrium probe of the local charge that is insensitive to complications that arise from measurements of transport through nanostructures. The measured value, e∗ 5/2 = e/4, indicates that the FQH state at ν = 5/2 demonstrates pairing, in agreement with proposed non-Abelian variational wavefunctions and different from other observed FQH states. Finally, the localization of e/4 quasiparticles is essential to the development of interferometers capable of detecting and exploiting these exotic braiding properties [15, 16], and our measurements suggest that e/4 localization does indeed occur in a well-behaved way. Acknowledgements: We would like to acknowledge Basile Verdene, Jonah Waissman, and Johannes Nübler for technical assistance, and we are grateful to Bertrand Halperin for theoretical discussions. This research has been funded by Microsoft Corporation Project Q. 3 References [1] Girvin, S. & Prange, R. The Quantum Hall Effect (Springer, 1987). [2] Moore, G. & Read, N. Nonabelions in the fractional quantum hall effect. Nuclear Physics B 360, 362 -- 396 (1991). [3] Nayak, C., Simon, S. H., Stern, A., Freedman, M. & Sarma, S. D. Non-abelian anyons and topological quantum computation. Reviews of Modern Physics 80, 1083 (2008). [4] Das Sarma, S., Freedman, M. & Nayak, C. Topologically protected qubits from a possible non-abelian fractional quantum hall state. Phys. Rev. Lett. 94, 166802 (2005). [5] Dolev, M., Heiblum, M., Umansky, V., Stern, A. & Mahalu, D. Observation of a quarter of an electron charge at the ν = 5/2 quantum hall state. Nature 452, 829 -- 834 (2008). [6] Dolev, M. et al. Dependence of the tunneling quasiparticle charge determined via shot noise measure- ments on the tunneling barrier and energetics. Phys. Rev. B 81, 161303 (2010). [7] Bid, A., Ofek, N., Heiblum, M., Umansky, V. & Mahalu, D. Shot noise and charge at the 2/3 composite fractional quantum hall state. Phys. Rev. Lett. 103, 236802 (2009). [8] Radu, I. et al. Quasi-particle properties from tunneling in the v= 5/2 fractional quantum hall state. Science 320, 899 (2008). [9] Martin, J. et al. Localization of fractionally charged quasi-particles. Science 305, 980 (2004). [10] Ilani, S., Yacoby, A., Mahalu, D. & Shtrikman, H. Microscopic structure of the metal-insulator transition in two dimensions. Science 292, 1354 -- 1357 (2001). [11] Ilani, S. et al. The microscopic nature of localization in the quantum hall effect. Nature 427, 328 (2004). [12] Dean, C. R. et al. Intrinsic gap of the ν = 5/2 fractional quantum hall state. Phys. Rev. Lett. 100, 146803 (2008). [13] Choi, H. C., Kang, W., Das Sarma, S., Pfeiffer, L. N. & West, K. W. Activation gaps of fractional quantum hall effect in the second landau level. Phys. Rev. B 77, 081301 (2008). [14] Bishara, W., Bonderson, P., Nayak, C., Shtengel, K. & Slingerland, J. K. Interferometric signature of non-abelian anyons. Phys. Rev. B 80, 155303 (2009). [15] Stern, A. & Halperin, B. I. Proposed experiments to probe the non-abelian ν = 5/2 quantum hall state. Phys. Rev. Lett. 96, 016802 (2006). [16] Bonderson, P., Kitaev, A. & Shtengel, K. Detecting non-abelian statistics in the ν = 5/2 fractional quantum hall state. Phys. Rev. Lett. 96, 016803 (2006). 4 Figure 1: Filling puddles with fractional charge. a, The sample well width is 30 nm, with symmetric Si δ-doping layers 100 nm on either side indicated by orange bands. Donors in these layers create a disorder potential in the 2DES, which produce puddles of localized states when the bulk is tuned to an incompressible, percolating Hall state. These puddles have some charging energy associated with adding electrons (Ui), and possibly some interaction with surrounding puddles (Vij). Incompressibility (κ−1 = ∂µ ∂n) is measured using an SET fabricated on the surface. b, While the global chemical potential should increase smoothly with density (black dashed line), the local chemical potential will increase in jumps (red line), with charge being added when the global chemical potential aligns with a localized state. c, Repeating the charging of an identical puddle with charge e/3 objects instead of charge e objects results in three times as many charging events in the same range of global density. Scaling the density axis of the charge e spectrum by 1/3 and shifting by some amount (green curve) should result in good overlap of the incompressibility spectra. Figure 2: Incompressibility and localized states at 5/2. a, By varying the magnetic field and the backgate voltage (density), we can identify incompressible phases of the 2DES. Our samples show clear incompressible FQH states at 5/2 and 7/3, with the expected slopes in the nB-plane. b, Zooming in shows repeatable charging events associated with quasiparticles localizing in puddles under the SET, stable on a timescale of days. c, A linecut showing the charging spectrum of any puddles coupled to the SET. Downwards spikes correspond to quasiparticles entering puddles beneath the SET. 5 eeeUiVijµµn∂∂µnne= ee*= e/3abc2D ELECTRON SYSTEMBACKGATEDensity (x 10 cm )11-2Magnetic Field (T)2.322.423.84.35/22.382.392.42.41-2-10Incompressibility (A.U.)Density (x 10 cm )11-27/3abcFigure 3: Comparison of spectra at 5/2 and 7/3. a, To determine the charge, we first choose a relative scale between the two density axes (β), and determine the offset between the two spectra that maximizes the cross-covariance. Here the density for the spectrum at 5/2 is scaled up by a factor of 1.29 and shifted to match up with the spectrum at 7/3. The guide lines show the density change required to add 1 electron to an area of 100 nm x 500 nm, approximately the size of our SET. We would therefore expect, very roughly, 3 e/3 charging events in a window this size. b, Repeating this for many values of β suggests that a relative scale of 1.29 best describes this data set. Figure 4: Summary of Data and Model. a, Repeating the measurement over many disorder configurations and samples shows that the peak at 4/3 is usually present. b, Averaging over all measurements yields a clear peak at β = 1.31, 3.8σ above the uncorrelated background for that scale (P = 7 × 10−5), suggesting a local charge ratio of 4/3. c, d, Running our model with parameters = 0±.3, V = 0.3±0.2, and ∆5/2 = 0.01, 0.1, and 1.0 (all in units of U, the on-site charging energy). We simulated charging of four puddles, of which two were capacitively coupled to the SET. 6 0.40.81.622.40.20.30.40.50123456789-8-6-4-202Relative Scale (β)∆n at 7/3 (x 10 m )13-25/27/3Incompressibility (A.U.)Best Correlationab1.290.41.11.82.50.320.431.310.41.11.82.50.260.440.41.11.82.50.41.11.82.5adcbMeasurementModelBest CorrelationRelative Scale (β)Relative Scale (β)∆=0∆=U4/32/3
1810.02628	1	1810	2018-10-05T11:54:54	Heat transfer between two metals through subnanometric vacuum gaps	[ "cond-mat.mes-hall" ]	We theoretically investigate the heat transfer between two metals across a vacuum gap in extreme near-field regime by quantifying the relative contribution of electrons, phonons and photons. We show that electrons play a dominant role in the heat transfer between two metals at subnanometric distance subject to a temperature gradient. Moreover, we demonstrate that this effect is dramatically amplified in the presence of an applied bias voltage. These results could pave the way to novel strategies for thermal management and energy conversion in extreme near-field regime.	cond-mat.mes-hall	cond-mat	Heat transfer between two metals through subnanometric vacuum gaps Riccardo Messina,1 Svend-Age Biehs,2 Till Ziehm,2 Achim Kittel,2 and Philippe Ben-Abdallah1, ∗ 1Laboratoire Charles Fabry, UMR 8501, Institut d'Optique, CNRS, Université Paris-Saclay, 2 Avenue Augustin Fresnel, 91127 Palaiseau Cedex, France 2Institut für Physik, Carl von Ossietzky Universität, D-26111 Oldenburg, Germany (Dated: October 8, 2018) 8 1 0 2 t c O 5 ] l l a h - s e m . t a m - d n o c [ 1 v 8 2 6 2 0 . 0 1 8 1 : v i X r a We theoretically investigate the heat transfer between two metals across a vacuum gap in extreme near-field regime by quantifying the relative contribution of electrons, phonons and photons. We show that electrons play a dominant role in the heat transfer between two metals at subnanometric distance subject to a temperature gradient. Moreover, we demonstrate that this effect is dramatically amplified in the presence of an applied bias voltage. These results could pave the way to novel strategies for thermal management and energy conversion in extreme near-field regime. The transition from radiation to conduction regimes between two bodies at different temperatures when their separation distance is reduced to subnanometric gaps is an emerging problem in physics [1, 2]. Recent experimen- tal works carried out by two different teams [3, 4] have explored this problem by making a direct measurement of heat power exchanged between a metallic tip and a metallic plate separated by vacuum gaps of angstrom to nanometer width. Nevertheless, these experiments come to radically different conclusions. On one hand, Kittel et al. have reported a strong deviation between their exper- imental results and the predictions coming from Rytov's fluctuational electrodynamic theory (FED) [5, 6]. On the other hand, Reddy et al. found a relatively good agree- ment with this theory down to few-angstrom separation distances. Unfortunately, due to the lack of a complete theory to describe the multichannel energy exchange at this scale, this problem still remains open. The two ex- perimental setups are similar. They are based on custom- fabricated Scanning Thermal Microscope (SThM) gold probes above a substrate in an ultra-high vacuum envi- ronment. In Kittel's experiment the extreme end of the probe can be modeled as a sphere with a radius of cur- vature of about 30 nm, while in Reddy's experiment the tip has a radius of 150 nm. Moreover, in the first ex- perimental setup heat transfer takes place at cryogenic temperatures (Tprobe = 280 K and Tsample = 120 K), while in the second setup around ambient temperature (Tprobe = 303 K and Tsample = 343 K). In this Letter we explore fundamentally heat transfer in extreme near-field regime by considering the contri- bution of all possible channels: electrons, phonons and photons. We first describe the transfer between two bulk metallic samples in plane -- plane geometry when the separation distance is reduced from nanometer down to angstrom gaps. In particular, we discuss quantitatively and qualitatively the role played by each carrier on the heat transfer in this range of distances, showing that elec- trons play a major role when approaching the contact. We also show how heat transfer is dramatically affected by the presence of an applied bias voltage. Next, we develop a simple model to evaluate the heat transfer be- tween a SThM tip and a sample and compare our theo- retical predictions with recent experimental results. Figure 1: Two solids out of thermal equilibrium separated by a vacuum gap of thickness d exchange heat. When the separa- tion distance d is of the order of the thermal wavelength λth, heat transfer is mainly due to evanescent photons. When d is much smaller than λth (typically at subnanometric distances for reservoir temperatures around the ambient temperature) and becomes even smaller than a nanometer, electrons and acoustic phonons contribute to the heat transfer by tunneling effect through the vacuum gap. We start by considering a system (see Fig. 1) made of two reservoirs at fixed temperature and chemical poten- tial TL,R and µL,R separated by a vacuum gap of thick- ness d. Neglecting, in first approximation, the coupling between electrons, phonons and photons, the heat flux transferred from the left (L) to the right reservoir (R) can be decomposed into three contributions Jh = J (el) h + J (ph) h + J (rad) h (1) respectively due to electrons (el), acoustic phonons (ph) and photons (rad). We stress that the contribution of optical phonyons is taken into account in the photonic contribution J (rad) through the dielectric permittivity de- scribing the optical response of each material. We detail eElectrontunnelingd << lthPhonontunnelingTLµLTRµRPropagativephotonsSurface waveFrustratedphotons2 Ex and Ex + dEx across a unit area per unit time. Elec- trons above (below) the chemical potential of each reser- voir contribute positively (negatively) to the flux. Simi- larly the electric current density reads Je = e dEx[NL(Ex, TL) − NR(Ex, TR)]T (Ex). (4) (cid:90) ∞ 0 The heat flux carried by conduction is associated with acoustic-phonon tunneling. Recent theoretical works have quantified the contribution of this channel for ex- changes between piezoelectric materials [11], polar ma- terials [2], semiconductors [12] and metals [13] and com- pared it with near-field heat exchanges. However, to date its relative contribution with respect to the electronic channel has not been addressed. In the long-wavelength approximation this energy transfer can be calculated us- ing the continuous elastic solid theory. In this case, the net heat flux exchanged between two reservoirs reads (cid:90) +∞ (cid:88) l=L,T 0 (cid:90) kc 0 (ph) J h = ω∆n dω 2π κT (ph) l dκ 2π (ω, κ), (5) κ being the parallel component of phonons wavector, κc = π/a its cutoff (a being the lattice constant) and ∆ = [n(ω, TL) − n(ω, TR)] the difference of two distri- bution functions n(ω, Ti) = [exp(ω/kBTi) − 1]−1 at Ti (i = L, R). The transmission coefficients for the energy of longitudinal (L) and transversal (T ) phonons read [13] T (ph) L = T (ph) T = TLL2c2 LkLx + TT L2c2 TkT x LkLx c2 TLT2c2 LkLx + TT T2c2 TkT x TkT x c2 , , (6) (7) where klx (l = L, T ) denotes the normal component of wave vector (with klx + κ2 = k2 l , kl = ωl/cl being the modulus of wavector at the frequency ωl and cl the ve- locity of l−phonons), while Tlq (l, q = L, T ) are the trans- mission coefficients of phonons across the gap (defined in Ref. [13]) which are directly related to the elastic prop- erties of the solid and to its lattice constants. As for the flux carried by photons, it can be evaluated from FED using the Polder and van Hove formalism [6] by using a nonlocal effects [14]. The net exchanged heat flux reads (cid:88) (cid:88)(cid:90) ∞ p=T E,T M 0 (rad) J h = (cid:90) ∞ 0 ω∆n dω 2π κT (rad) p dκ 2π (ω, κ), where T (rad) p (8) (ω, κ) denotes the energy transmission coef- Figure 2: Heat transfer at subnanometric scale between two gold samples in plane -- plane geometry. The temperatures are TL = 280 K and TR = 120 K and no bias voltage is applied. (2) below the calculation of each of these contributions, start- ing with the electronic flux by tunneling effect. This flux can be calculated once the effective potential barrier de- scribing the vacuum gap is estimated. In the case of a simple rectangular barrier, the transmission probability T (Ex) of electrons of normal energy Ex through this bar- rier reads [7] 4Ex(Ex − V ) + V 2 sin2(cid:0)k2x(Ex, V )d(cid:1) , where k2x(Ex, V ) =(cid:112)2me(Ex − V )/ denotes the nor- 4Ex(Ex − V ) T (Ex) = mal components of wavectors inside the gap, me being the electron mass and the reduced Plank constant. We emphasize that the transmission probability T (Ex) also implicitly depends on the distance d and barrier height V . The latter is written here as V (d) = VeV(d) + EF , i.e. the sum of the Fermi energy EF (EF = 5.53 eV for gold) and the distance-dependent potential VeV(d), for which the data taken from [8] have been fitted from DFT calcu- lations with the log-scale law VeV(d) = V0 ln(1 + d/1Å), where V0 = 1.25 eV. This result can be easily generalized to the case of a spatially dependent potential barrier [9] It follows that the heat flux carried by the electrons through tunnel effect can be calculated by summing over all energies Ex = 1 x in the x-direction normal to the surface. Following Simmons [10], the net electronic heat flux reads 2 mev2 0 (el) J h = dExEx[NL(Ex, TL) − NR(Ex, TR)]T (Ex), (3) where Ni(Ex, Ti)dEx (i = L, R), with Ni(Ex, Ti) = )], denotes the number of mekB Ti 2π23 electrons in the reservoir i with a normal energy between ln[1 + exp(− Ex−EF −µi kB Ti (cid:90) ∞ 0.10.20.512510-411041081012Fh = ∆T /T 2 by the Onsager relations [15] (cid:19) (cid:18) Je Jh (cid:18) Lee Leh Lhe Lhh (cid:19)(cid:18) FeFh = 3 , (10) (cid:19) where La,b (a, b = e, h) are the Onsager coefficients which are related to the familiar transport coefficients. These fluxes can be calculated using the general expressions we have introduced. 1 − T 4 Fig. 2 shows the net heat flux exchanged between two gold parallel samples with respect to their separa- tion distance in the range from 1 Å to 5 nm in a dou- ble logarithmic plot. The temperatures TL = 280 K and TL = 120 K correspond to the parameters of the experiment [3]. Above the nanometer we observe that the heat exchange is mainly driven by evanescent pho- tons (for the temperatures chosen here the blackbody 2 ) is around 300 W·m−2). Below limit φBB = σ(T 4 one nanometer the contribution of photons is slightly en- hanced by nonlocal effects (Landau damping). A satura- tion sets in for a distance on the order of the Thomas- Fermi length. Around d = 1 nm the electronic contri- bution becomes comparable with the photonic part and finally dominates heat exchanges at subnanometric dis- tances. We also notice that the contribution of phonons, which scales as d−10, becomes also more important than the one of photons at a few angstrom separation dis- tance. This confirms the conclusions obtained by previ- ous works [11, 13]. However, the magnitude of the flux carried by elastic waves remains several orders of magni- tude smaller than the flux associated to electronic tun- neling. We also stress that, while the contribution of the three carriers discussed above quantitatively depend on the choice of the two temperatures TL and TR, the main features discussed here, namely the leading role of electronic flux at subnanometric distances and the tran- sition around 1 nm between mainly photonic and mainly electronic flux, are basically unaffected for temperatures close to ambient temperature. It is now important to address the issue of the impact of the presence of an applied bias voltage between the two bodies. This situation corresponds for example to Kit- tel's experimental setup where a bias voltage of 600 mV is applied between the tip and the sample. The pres- ence of a non-vanishing voltage modifies the electronic chemical potential µ, thus their distribution function. In Fig. 3 we show the value of the thermal power and cur- rent exchanged between two gold planar samples in the temperature conditions Kittel's experiment with respect to the separation distance for different values of chemical potential. We observe that below one nanometer both the total power and the current are increased by sev- eral orders of magnitude because of this external voltage. This enhancement is of the order of three orders of mag- nitude at 1 Å separation distance, with an applied bias voltage of 600 mV, i.e. the one of Kittel's experiment. This suggests that the discrepancy observed between the Figure 3: Influence of a bias voltage on the heat flux between two gold samples in plane -- plane geometry. The temperatures are TL = 280 K and TR = 120 K. The main part (inset) of the plot shows the heat flux (current density) for three different values of the chemical potential µR, assuming µL = 0. ficient T (rad) p (ω, κ) =  (1 − ρL,p2)(1 − ρR,p2) 1 − ρL,p ρR,p e2ikxd2 4 Im (ρL,p) Im(ρR,p)e−2Im(kx)d 1 − ρL,p ρR,p e−2Im(kx)d2 , κ < ω c , , κ > ω c , (9) for the two polarizations p = T E, T M, taking into ac- count the contributions of propagating (κ < ω/c, κ be- ing the component of the wave vector parallel to the slabs) and evanescent waves (κ > ω/c). ρL,p and ρR,p are the reflection coefficients of the two reservoirs and kx =(cid:112)ω2/c2 − κ2 is the normal component of the wave vector inside the vacuum gap. When nonlocal effects are neglected and the reservoirs are made of bulk gold, the reflection coefficients reduce to the Fresnel coeffi- cients defined with the dielectric permittivity (ω) = 1 − ω2 p/ω(ω + iγ) (ωp = 13.71 × 1015 rad s−1 being the plasma frequency and γ = 4.05 × 1013 s−1 the damping coefficient). If the nonlocal scenario, the Fresnel coeffi- cients must be replaced by reflections coefficients defined in terms of surface impedances [14]. In the following, our results for the photonic contribution to heat flux always include nonlocal effects. We emphasize that the expressions given above are valid for any choice of temperatures and voltage bias. In the linear-response approximation (for small ∆T = TL − TR and ∆µ = µR − µL) the electric and heat currents Je and Jh are linearly related to the thermo- dynamic forces Fe = ∆V /T (where ∆V = ∆µ/e) and 0.10.20.51251051071091011101310150.10.20.512100105108101110144 treme end of the tip (probe) is cut in planes of finite area parallel to the sample and the heat flux exchanged between the tip and the sample is simply obtained by summing the heat flux between the sample and all these elementary surfaces. For a spherical shape of radius R, the net power exchanged between the tip and the sample reads (d + R −(cid:112) R2 − r2). (11) (cid:90) R 0 (ph) h = 2π P (ph) dr r J h Concerning the electronic contribution, the PA is gener- ally not valid anymore. Based on the theory of scanning tunneling microscopy [16 -- 19] we proceed differently. We model the electronic flux and current between tip and sample by taking the plane -- plane result and multiply- ing it by πr2 Au, where rAu = 1.35 Å represents the radius of a gold atom. This simple approach is validated by the verification that when approaching contact it gives results in agreement with the quantum of electrical con- ductance G0 = 2e2/. Another very relevant issue is the choice of the barrier height V (d). This is typically determined in any SThM experiment by looking at the rate of exponential decay of the observed current with respect to the distance [16]. In both experiments men- tioned above [3, 4], the authors obtain values for this barrier between 1 eV and 2.5 eV, far below the theoret- ical value (4.7 eV) expected for ideally clean and bulk samples. As a consequence, since the theoretical re- sults depend (as expected) strongly on the value of the potential barrier [20], for these simulations we replace our model V (d) with values directly taken from the ex- perimental data under scrutiny. Unfortunately, the re- sults presented in Ref. [3] are associated with different barrier heights, making the theory-experiment compar- ison more challenging. As a consequence, we compare our theoretical results with more recent experimental re- sults [21], obtained in a different temperature configu- ration (Tprobe = 295 K and Tsample = 195 K) and corre- sponding to a given barrier height. We have fitted the data for the current [show in the inset of Fig. 4(a)] with either a exp[−2κd] or a exp[−2κ(d − d0)] dependence in order to take into account an error on the determination of the distance. We find a value of the barrier height close to 1 eV and an error bar on the distance of the order of 1 Å. In the inset of Fig. 4(a), we present our theoretical predictions for the current (the shaded region is the one between our theoretical results for d and d + 1 Å), along with experimental data, showing a good agreement. We remind that in this experiment a bias voltage of 600 mV is applied. The main part of Fig. 4(a) shows instead the conductance for the same configuration. Albeit the absence of a good agreement between theoretical and ex- perimental results, we confirm indeed the existence of a strong deviation (amplification) with respect to FED re- sults, due to the electronic contribution. We now focus on the experimental results of Reddy and collaborators, Figure 4: Thermal conductance at subnanometric distance in a SThM geometric configuration made of a Au probe above a Au sample in (a) Kittel's experimental conditions (Tprobe = 295 K, Tsample = 195 K and R = 150 nm) and (b) Reddy's experimental condition (Tprobe = 343 K, Tsample = 303 K and R = 30 nm). Inset: tunneling current as a function of distance d. The bias voltage taken in theoretical results is 600 mV in (a), 8 mV in (b). The points correspond to experimental data. power measured in this experiment and the predictions of conventional FED could be indeed attributed to the contribution of electrons. Based on this last result, we develop now a simple model to describe the values for the current and heat flux observed in recent experiments [3, 4]. Based on the results shown in Fig. 2, we can neglect the role of phonons. As far as the photonic flux is concerned, we exploit the so-called proximity approximation (PA), al- ready employed in the theoretical simulations described in Refs. [3, 4]. According to this approximation, the ex- and refer in particular to Fig. 2(c) of Ref. [4], correspond- ing to the cleanest sample and smallest temperature dif- ference (Tprobe = 343 K and Tsample = 303 K). For the barrier height we take the value of 1.7 eV given by the authors and we keep our theoretical error bar of 1 Å. Concerning the current, we are able to reproduce the values measured by the authors by considering an ap- plied bias voltage of 8 mV, much smaller than the one of the experiment discussed above. Concerning the current, our model gives a good agreement with experimental re- sults, as shown in the inset of Fig. 4(b). Concerning the flux, we observe a strong reduction with respect to the other experiment. This is due to a much lower value of the electronic flux, which is in turn due to a two-order- of-magnitude lower applied voltage. Moreover, at the smallest value of the distance shown in Ref. [4] (approxi- mately 2 Å), our theoretical estimate of the conductance is between 0.5 nW·K−1 and 2.6 nW·K−1, in good agree- ment with the upper boundary of 2.5 nW·K−1 claimed by the authors. In conclusion, we have introduced the first theoretical framework to investigate individually the contribution of all channels to the heat transfer between two solids near the physical contact and analyzed the role played by the different energy carriers on the heat transfer between two metals at subnanometric distance. We have predicted a giant heat transfer before contact and demonstrated its electronic origin. We have highlighted the strong tun- ability of heat flux in extreme near-field regime by ap- plying an external bias voltage. These effects could be used to develop novel strategies of thermal management at this scale. They also could be exploited in the field of heat-assisted magnetic recording as well as in energy- conversion technology. Further theoretical developments should lead to a re- finement of comparison of predictions with the experi- mental results. Among them are a more precise modeling of electron tunneling, a self-consistent calculation of the potential barrier for a given geometric configuration, as well as a more precise description of optical, electronic and mechanical properties. R. M. and P. B.-A. acknowledge discussions with A. W. Rodriguez. The authors acknowledge financial support by the DAAD and Partenariat Hubert Curien Procope Program (project 57388963). 5 ∗ Electronic address: [email protected] [1] S. Xiong, K. Yang, Y. A. Kosevich, Y. Chalopin, R. D'Agosta, P. Cortona, and S. Volz, Phys. Rev. Lett. 112, 114301 (2014). [2] V. Chiloyan, J. Garg, K. Esfarjani, and G. Chen, Nat. Commun. 6, 6755 (2015). [3] K. Kloppstech, N. Könne, S.-A. Biehs, A. W. Rodriguez, L. Worbes, D. Hellmann, and A. Kittel, Nat. Commun. 8, 14475 (2017). [4] L. Cui, W. Jeong, V. Fernandez-Hurtado, J. Feist, F. J. Garcia-Vidal, J. C. Cuevas, E. Meyhofer, and P. Reddy, Nat. Commun. 8, 14479 (2017). [5] S. M. Rytov, Y. U. A. Kravtsov, and V. I. Tatarskii, Prin- ciples of statistical radiophysics (Springer, Berlin, 1988). [6] D. Polder and M. van Hove, Phys. Rev. B 4, 3303 (1971). [7] C. Cohen-Tannoudji, J. Dupont-Roc, G. Grynberg, Atom-photon interactions (Wiley, New York, 1992). [8] A. Kiejna, Ultramicroscopy 42, 231 (1992). [9] P. Zhang, Sci. Rep. 5, 9826 (2015). [10] J. G. Simmons, J. Appl. Phys., 34, 6, 1793 (1963). [11] M. Prunnila and J. Meltaus, Phys. Rev. Lett. 105, 125501 (2010). [12] D. P. Sellan, E. S. Landry, K. Sasihithlu, A. Narayanaswamy, A. J. H. McGaughey, and C. H. Amon, Phys. Rev. B 85, 024118 (2012). [13] J. B. Pendry, K. Sasihithlu, and R. V. Craster, Phys. Rev. B 94, 075414 (2016). [14] G. W. Ford, W. H. Weber, Phys. Rep. 113, 195 (1984). [15] S. R. de Groot and P. Mazur, Non-equilibrium thermo- dynamics (Dover, NewYork, 1984). [16] C. J. Chen, Introduction to Scanning Tunneling Mi- croscopy (Oxford University Press, New York, 2008). [17] G. Binnig, H. Rohrer, C. Gerber, and E. Weibel, Appl. [18] G. Binnig, H. Rohrer, C. Gerber, and E. Weibel, Phys. Phys. Lett. 40, 178 (1982). Rev. Lett. 49, 57 (1982). Rev. Lett. 50, 120 (1983). [19] G. Binnig, H. Rohrer, C. Gerber, and E. Weibel, Phys. [20] See EPAPS Document No. [number will be inserted by publisher] for a discussion of the dependence of the elec- tronic heat flux and current on the barrier height V . [21] T. Ziehm et al., in preparation.
1704.09024	1	1704	2017-04-28T17:44:41	Revealing topological Dirac fermions at the surface of strained HgTe thin films via Quantum Hall transport spectroscopy	[ "cond-mat.mes-hall" ]	We demonstrate evidences of electronic transport via topological Dirac surface states in a thin film of strained HgTe. At high perpendicular magnetic fields, we show that the electron transport reaches the quantum Hall regime with vanishing resistance. Furthermore, quantum Hall transport spectroscopy reveals energy splittings of relativistic Landau levels specific to coupled Dirac surface states. This study provides new insights in the quantum Hall effect of topological insulator (TI) slabs, in the cross-over regime between two- and three-dimensional TIs, and in the relevance of thin TI films to explore novel circuit functionalities in spintronics and quantum nanoelectronics.	cond-mat.mes-hall	cond-mat	Revealing topological Dirac fermions at the surface of strained HgTe thin films via Quantum Hall transport spectroscopy. C. Thomas,1 O. Crauste,2, 3 B. Haas,4 P.-H. Jouneau,4 C. Bauerle,2, 3 L.P. L´evy,2, 3 E. Orignac,5 D. Carpentier,5 P. Ballet,1 and T. Meunier2, 3 1Univ. Grenoble Alpes, CEA, LETI, MINATEC campus, F38054 Grenoble, France. 2CNRS, Inst NEEL, F-38042 Grenoble, France. 3Univ. Grenoble Alpes, Inst NEEL, F-38042 Grenoble, France. 4Univ. Grenoble Alpes, CEA, INAC, F38054 Grenoble, France. 5Univ Lyon, ENS de Lyon, Univ Claude Bernard, CNRS, Laboratoire de Physique, F-69342 Lyon, France. (Dated: July 15, 2021) We demonstrate evidences of electronic transport via topological Dirac surface states in a thin film of strained HgTe. At high perpendicular magnetic fields, we show that the electron transport reaches the quantum Hall regime with vanishing resistance. Furthermore, quantum Hall transport spectroscopy reveals energy splittings of relativistic Landau levels specific to coupled Dirac surface states. This study provides new insights in the quantum Hall effect of topological insulator (TI) slabs, in the cross-over regime between two- and three-dimensional TIs, and in the relevance of thin TI films to explore novel circuit functionalities in spintronics and quantum nanoelectronics. PACS numbers: Similar to the case of graphene, the charge carriers at the surface of topological insulators are expected to be massless Dirac fermions but with a real spin locked to the momentum [1 -- 3]. This has strong implications when the electrons experience a large perpendicular magnetic field B and enter the quantum Hall regime. Indeed, each surface is then characterized by non-degenerate Landau levels (LL) and the associated Hall conductance is ex- pressed as σxy = (N + 1 h , where N is the LL in- In topological insulator slabs, however, two sur- dex. face states of extension w have to be considered and are separated by a thickness t. When the wave functions of the two surfaces do not overlap (t (cid:29) w), they are only connected at the boundaries of the sample [4] and the transport properties are obtained by summing the distinct contributions of each surface. The Hall conduc- tance σxy = (Ntop + Nbottom + 1) e2 h is then ex- pected where Ntop and Nbottom are the LL index for the top and bottom surfaces, respectively. In this regime, integer filling factors ν have been observed in relatively thick strained HgTe [5 -- 7] and Bi-based [8, 9] topological insulators. By decreasing t down to w, the two surface states start to overlap giving rise to a non-negligible hy- bridization energy ∆ [10 -- 12] and the transport then oc- curs through states delocalized between the two surfaces. Degenerate Dirac LLs are then expected to emerge at N B. Moreover, addi- high B with energies scaling as tional dispersive couplings between the two surfaces are expected to lift the LL degeneracy with an energy split- ting linear in B [12, 13]. While of orbital nature, this splitting reveals the microscopic coupling of spin and orbital degrees of freedom in the Dirac Landau levels. Therefore analysing the energy gaps of both odd and even filling factors and their B-dependences provides a h = ν e2 2 ) e2 √ powerful tool to reveal the Dirac surface states of a thin topological insulator slab. Decreasing further t would re- sult in the opening of a large gap in the surface states and the emergence of the quantum spin Hall phase [11 -- 14]. In this letter, the magneto-transport properties of strained HgTe thin films at high magnetic fields are inves- tigated. We study films with a thickness of about 15 nm, characterized by two tunnel-coupled surfaces, where elec- tron transport is solely mediated by surface charge carri- ers. The quality of the material allows reaching the quan- tum Hall regime with vanishing resistance for magnetic field larger than 1.5 T. At higher magnetic fields, non- degenerate Landau levels are observed. By analysing the temperature dependence of the magneto-conductance, a clear difference of the energy gaps corresponding to odd and even filling factors is noticed and is consistent with a Landau level energy spectrum characteristic of two cou- pled Dirac surfaces. We investigate top-gated Hall bars (see Fig.1 (a)) fab- ricated from a 15 nm-thick HgTe layer surrounded by two 30 nm-thick Hg0.3Cd0.7Te barriers and grown on a (100) CdTe substrate. Particular attention to lower the de- fects present in the HgTe layer and to obtain sharp HgTe / Hg0.3Cd0.7Te interfaces was paid during the structure growth [15] (see Fig.1 (b)). Two different structures with similar thickness were grown and gave very similar re- sults. The Hall bar is 40 µm long and 10 µm wide. A top gate covering the Hall bar enables to change the car- riers from holes to electrons as illustrated by the density sign inversion in Fig.1 (c). In HgTe layers, the light hole band Γ8,LH band is lying 0.3 eV above the Γ6. Such an inverted band structure at the Γ point results in topological surface states, robust to the presence of the heavy hole band Γ8,HH [16]. At zero 2 Vgate through n = αVgate = ν eB h where α is the depleting factor of the gate (see Fig. 1 (c)). Derived from the ex- tracted electron density at low B, the expected positions of the Rxx minima are represented by the black dashed lines in Fig. 2 (a). They are reproducing properly the minima of the fan diagram considering only odd filling factors on the hole-side (Vgate ≤ VDP ) and both even and odd integers on the electron-side (Vgate ≥ VDP ), in agreement with the corresponding quantized plateaus on the Rxy mapping on Fig. 2 (b). At B = ± 3 T, the quantum Hall regime is achieved on the electron-side with vanishing resistance and Hall con- ductance plateaus corresponding to integer filling factors (see Fig. 3). Such observations are strong signatures that there is no extra bulk contribution to the transport in the electron-regime in contrary to what is observed in thicker samples [5 -- 7]. On the hole-side, σxy plateaus cor- responding to odd filling factors are observed with σxx no longer completely vanishing. Moreover, it is worth notic- ing that σxx is characterized by broader peaks than on the electron-side. The observed differences between holes and electrons are explained by the coupling of the surface states with the heavy hole Γ8,HH bulk band. From the band struc- ture of strained HgTe [16], the Γ8,HH band is expected to efficiently couple to the hole part of the surface states. This coupling opens up scattering channels resulting in the broadening of σxx peaks. As a consequence, a larger magnetic field is needed on the hole-side to resolve the observed spin splitting on the electron-side. Whereas the ν=2 plateau is obtained for B larger than ∼ 1.5 T with a clear separation of the Rxx maxima into two distinct branches (see Fig. 2 (a)), additional measurements allow detecting the ν = -2 plateau on the hole-side only when B equals 5.5 T [17]. We can thus conclude that only one set of fan diagrams, associated to the same surface state on both electron- and hole-sides, is observed in the quantum Hall regime. Moreover, the degeneracy of the Landau levels is lifted for a 15 nm-thick HgTe topological film at high perpendicular magnetic fields. To gain knowledge on the nature of the Landau levels, we analysed the temperature dependence of the magneto- conductance at high fields [18 -- 20]. In such a procedure, the energy difference between LLs is estimated via ther- mal activation. More precisely, the temperature depen- dence of the longitudinal resistance Rxx minima is fit- ted using the Arrhenius law (see Fig. 4(a) for ν = 1): xx ∝ exp ( −∆E Rmin 2kB T ) where ∆E is the activation energy gap, kB the Boltzmann constant and T the temperature. This analysis has only been performed on the electron- side since the influence of the Γ8,HH bulk band on the hole-side changes drastically with temperature [17]. We first focus on odd filling factors where two successive LLs have different orbitals (see Fig. 4(b) and (c)). We clearly observe non-regular energy separations between the suc- cessive LLs. Indeed, they decreases with N and increases FIG. 1: (a) Optical image of the Hall bar sample realized after nanofabrication of the structure. (b) Scanning Trans- mission Electron Microscopy (STEM) image in a High Angle Annular Dark Field (HAADF) mode of a 15 nm-thick HgTe layer embedded between two Hg0.3Cd0.7Te barriers. Inset: High resolution STEM HAADF image zooming on the HgTe layer. Low defect density and quality of the interfaces are evidenced. (c) Evolution of the slope RH = dRxy dB at low per- pendicular magnetic fields with the voltage Vgate applied on In a one-carrier model, the density n is equal to the gate. where e is the charge of an electron. Vgate adjusts n in the range of order of 1011 cm−2. The red dashed line corre- sponds to a fit, which allows to extract the depleting factor of the gate α equal to ∼ 5.1011 cm−2.V−1. Inset: Schematics of the strained HgTe topological insulator structure used in the experiment. (d) Longitudinal resistance Rxx as a func- tion of Vgate at zero magnetic field showing the Dirac point at Vgate = VDP ∼ −0.5V. eRH 1 magnetic field and close to the charge neutrality point, the longitudinal resistance Rxx presents a peak at gate voltage VDP (see Fig. 1 (d)), whose amplitude depends on the size of the sample and can be as low as 1 kΩ for Hall bars of 1 micrometer [17]. We conclude that the structure has a metallic behavior as expected for electron transport through surface states. To probe the nature of the surface state carriers, we analyse the Hall bar magneto-conductance at high per- pendicular magnetic fields and at a temperature of 100 mK. Shubnikov-de Haas (SdH) oscillations on Rxx and quantized plateaus on the Hall resistance Rxy are ob- served (see Fig. 2(a) and (b)) and point at the emergence of LLs in the structure. Both B and the top gate volt- age Vgate allow controlling the filling of individual LLs. Indeed, the carrier density n is directly related to B and 3 FIG. 3: The Hall σxy = ρxy ρ2 xy +ρ2 and the longitudinal σxx = conductances as a function of Vgate for B= -3 T with W L , L and W respectively the length ρxx ρ2 xy +ρ2 ρxy = Rxy and ρxx = Rxx and the width of the Hall bar. xx xx about 2.07 ± 0.65 meV/T from the red solid line fit of Fig. 4 (d), which is similar to the Landau level splitting observed in HgTe non-topological quantum wells [21, 22]. To quantitatively model the energy gaps for odd and even filling factors, we consider an effective low-energy model of a thin 3D topological insulator: two Dirac sur- face states of opposite chirality vf τz(σx.ky−σy.kx) cou- pled by a k-dependent hopping amplitude of ( ∆ 2M )τx where vf is the surface state band velocity equal to 5.105 m.s−1 in our structures [16]. This coupling is purely of orbital nature, it does not depend on spins and originates from the overlap of the two surface states that opens a constant gap ∆ and introduces a quadratic k2 2M hopping In these conditions, we obtain the fol- amplitude [12]. lowing LL energy spectrum with β = e 2M explained in further details in ref [17]. 2 − k2 FIG. 2: (a) Longitudinal resistance Rxx as a function of Vgate and the perpendicular magnetic field B at 100 mK. The dashed lines represent the expected position of the Rxx min- ima extracted from the density evolution with Vgate (see Fig.1 (c)). (b) Hall resistance Rxy (in units of h2 e ) as a function of Vgate and B. (cid:114) EN,± = ± 2N ev2 f B + ( − N βB)2 ± βB,N ≥ 1 (1) ∆ 2 non-linearly with B (mostly noticed for the ν = 1 gap). Both observations are in qualitative agreement with a Dirac-like LL energy spectrum scaling as N B. √ The resulting energy gaps for even filling factors do not show the same behavior (see Fig. 4 (d)). Indeed, the activation energy gaps for all these LLs are similar and linear with B. The slope 2β has been estimated to be E0,± = ±βB − ∆ 2 (2) While of orbital nature, the term βB acts as an effective Zeeman energy and is fixed by the analysis of the even filling factor energy gaps performed in the previous paragraph. This spectrum evidences a splitting of each orbital LL due to the presence of the quadratic 100102104Rxx(Ω)Vgate(V)(a)-101Rxy(h/e2)Vgate(V)(b)ν=1ν=2ν=3ν=4ν=5ν=-1ν=-3ν=-50.40.20.0-0.2-0.4-0.6-0.80.40.20.0-0.2-0.4-0.6-0.8-3-2-10123B(T)-3-2-10123B(T)-5σxy(e2/h)σxx(e2/h)-131-2-3-4-6-7B = -3 TVgate(V)43210-1.0-0.8-0.6-0.4-0.20.00.20.44 SDH oscillations on the electron-side [17]. Nevertheless, for ν = 1, the model requires a three times larger Γ (solid red line in Fig. 4 (c)) that can be explained by the proximity in energy of the Γ8,HH band [17]. These analysis point at the Dirac nature of the surface-state carriers of strained HgTe thin films. In conclusion, we have shown that the quantum Hall regime with vanishing longitudinal resistance is achieved in 15 nm-thick strained HgTe layers. Dirac carriers are revealed via quantum Hall spectroscopy and propagate solely through states delocalized between the top and bottom topological surfaces of the HgTe slab. The overlap between the surface states results in the mixing between the two Dirac species thus in the splitting of the Landau levels. Dirac-surface-restricted transport opens the route towards the realization of topological circuits based on strained HgTe thin films, with potential applica- tions in quantum nanoelectronics[23] and spintronics[24]. We acknowledge N. Mollard for TEM sample prepa- ration and the technical support from the technologi- cal centres of the Institut N´eel, CEA-LETI and PTA. T.M. acknowledges financial support from Fondation Nanosciences and ANR Semitopo. [1] C. Kane and E. Mele, Phys. Rev. Lett. 98, 106803 (2007). [2] J. Moore and L. Balents, Phys. Rev. B 75, 121306(R) (2007). [3] R. Roy, Phys. Rev. B 79, 195322 (2009). [4] D. Lee, Phys. Rev. Lett. 103, 196804 (2009). [5] 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). [6] D. A. Kozlov, Z. D. Kvon, E. B. Olshanetsky, N. N. Mikhailov, S. A. Dvoretsky, and D. Weiss, Phys. Rev. Lett. 112, 196801 (2014). [7] D. A. Kozlov, D. Bauer, J. Ziegler, R. Fischer, M. L. Savchenko, Z. D. Kvon, N. N. Mikhailov, S. A. Dvoretsky, and D. Weiss, Phys. Rev. Lett. 116, 166802 (2016). [8] B. Sac´ep´e, J. Oostinga, J. Li, A. Ubaldini, N. Couto, E. Giannini, and A. Morpurgo, Nature Communications 2, 575 (2011). [9] Y. Xu, I. Miotkowski, C. Liu, J. Tian, H. Nam, N. Ali- doust, J. Hu, C.-K. Shih, M. Hasan, and Y. Chen, Nature Physics 10, 956 (2014). [10] J. Linder, T. Yokoyama, and A. Sudbo, Phys. Rev. B 80, 205401 (2009). [11] C.-X. Liu, H. Zhang, B. Yan, X.-L. Qi, T. Frauenheim, X. Dai, Z. Fang, and S.-C. Zhang, Phys. Rev. B 81, 041307 (2010). [12] H.-Z. Lu, W.-Y. Shan, W. Yao, Q. Niu, and S.-Q. Shen, Phys. Rev. B 81, 115407 (2010). [13] S.-B. Zhang, H.-Z. Lu, and S.-Q. Shen, Scientific Reports 5, 13277 (2015). [14] M. Konig, S. Wiedmann, C. Brune, A. Roth, H. Buh- FIG. 4: Minimum of Rxx as a function of temperature T for the ν=1 filling factor (a). This analysis allows to extract the activation energy gap ∆E for all positive filling factors at different B. ∆E evolution with B for odd filling factors ν = 1, 3, 5 and 7: (b) from the model of eq. 1 and 2: ∆Eν=2N +1 = EN +1,− − EN,+, (c) from the experimental data. (d) ∆E In (c) and (d), as a function of B for even filling factors. the red solid lines represent the result of the model with a constant offset Γ (see text and ref. [17]). In (c), ∆Eν=2N +1 = EN +1,− − EN,+ + Γ with Γ = 11 meV. In (d), ∆Eν=2N = EN,+ − EN,− + Γ with Γ = 3 meV. term. It is worth noting that for a vanishing overlap between the two surfaces both the gap and the quadratic term go to zero and one should observe only odd filling factors [12]. The results of this model for the estimated vf are presented in Fig. 4(b), (c) and (d) through the solid red lines. To obtain a good agreement between the data and the model for all energy gaps, we need to take into account the width of the Landau level Γ as a constant offset in energy, estimated to 3 meV from the B(T)ΔE(meV)ΔE(meV)0.040.101/T(K-1)Log(Rxx)(c)(d)(a)(b)0.162345B=3 Tν=10246810ν=3ν=1ν=7ν=50.00.51.01.52.02.53.0543210ν=4ν=2ν=8ν=6B=2.5 TB=2 TB=1.5 TB=1 TΔE(meV)05101520B(T)1235 mann, L. Molenkamp, X.-L. Qi, and S.-C. Zhang, Sci- ence 318, 766 (2007). [15] P. Ballet, C. Thomas, X. Baudry, C. Bouvier, O. Crauste, T. Meunier, G. Badano, M. Veillerot, J. P. Barnes, P. Jouneau, and L. L´evy, J. Electron. Mater. 43, 2955 (2014). [16] O. Crauste, Y. Ohtsubo, P. Ballet, P. Delplace, D. Car- pentier, C. Bouvier, T. Meunier, A. Taleb-Ibrahimi, and L. P. L´evy, arXiv:1307.2008 . [17] see supplementary materials, [18] Z. Jiang, Y. Zhang, H. L. Stormer, and P. Kim, Phys. . Rev. Lett. 99, 106802 (2007). [19] A. J. M. Giesbers, U. Zeitler, M. I. Katsnelson, L. A. Ponomarenko, T. M. Mohiuddin, and J. C. Maan, Phys. Rev. Lett. 99, 206803 (2007). [20] K. Takase, H. Hibino, and K. Muraki, Phys. Rev. B. 92, 125407 (2015). [21] Y. S. Gui, C. R. Becker, N. Dai, J. Liu, Z. J. Qiu, E. G. Novik, M. Schafer, X. Z. Shu, J. H. Chu, H. Buhmann, and L. W. Molenkamp, Phys. Rev. B 70, 115328 (2004). [22] B. Buttner, C. X. Liu, G. Tkachov, E. G. Novik, C. Brune, H. Buhmann, E. M. Hankiewicz, P. Recher, B. Trauzettel, S. C. Zhang, and L. W. Molenkamp, Na- ture Physics 7, 418 (2011). [23] L. Jiang, C. Kane, and J. Preskill, Phys. Rev. Lett. 106, 130504 (2011). [24] J.-C. Rojas-S´anchez, S. Oyarz´un, Y. Fu, A. Marty, C. Vergnaud, S. Gambarelli, L. Vila, M. Jamet, Y. Oht- subo, A. Taleb-Ibrahimi, P. Le F`evre, F. Bertran, N. Reyren, J.-M. George, and A. Fert, Phys. Rev. Lett. 116, 096602 (2016). 6 Supplementary materials: Revealing topological Dirac fermions at the surface of strained HgTe thin films via Quantum Hall transport spectroscopy. In the main text of this letter, we have presented the results obtained on a first sample (called sample 1) fabricated from a 15 nm-thick HgTe layer surrounded by two Hg0.3Cd0.7Te barriers on a (100) CdTe substrate. These supple- mentary materials aim to complete the analysis by, first of all, adding information about the metallic behavior of the longitudinal resistance at zero magnetic field, the Landau level splitting as well as the Landau level broadening. To do so, experimental results from a second sample (called sample 2), fabricated from another 15 nm-thick HgTe layer, are presented and discussed. It is worth noting that the two samples have been made using the same growth and nanofabrication recipes. Finally, a model describing the hybridization origin of the Landau level splitting is also devel- oped thus providing the justification of the observed integer quantization. With such contents, these supplementary materials aim to strengthen our experimental investigation of thin films of strained HgTe 3D topological insulators. LONGITUDINAL RESISTANCE Rxx AS A FUNCTION OF THE HALL BAR LENGTH The gate dependence of Rxx was measured at zero magnetic field for several Hall bars with lengths L ranging from 1 to 40 µm keeping the aspect ratio constant. All the Hall bars considered in this section were fabricated from sample 2. Similar observations were performed on sample 1. Figure 5 displays Rxx as a function of the gate voltage for B=0 T and for T lower than 100 mK. For all the bars, Rxx evolves the same way as a function of the gate voltage Vgate and exhibits a maxima at the Dirac point (DP) for Vgate = VDP . Resistance maxima are to first approximation increasing with L from 1 kΩ to 25 kΩ. This witnesses a non-ballistic transport suggesting a metallic behaviour of the surface states for which the resistance value is governed by scattering disorder potential. These observations are inconsistent with quantum spin Hall effect expected for thin HgTe topological quantum wells [1]. It is worth noting some divergences in the resistance of Hall bars of same dimensions. FIG. 5: Longitudinal resistance Rxx as a function of Vgate − VDP at zero magnetic field for sample 2. VDP is defined for each curve as the voltage Vgate corresponding to the maximum Rxx and slightly varies from one Hall bar to another. The Hall bar length varies from 1 µm to 40 µm. -0.3-0.2-0.10.00.30.4Rxx(kΩ)0B = 0 T 0.20.1510152025L = 8 μmL = 40 μmVgate - VDP(V)L = 10 μmL = 8 μmL = 10 μmL = 8 μmL = 1 μm7 EXPERIMENTAL EVIDENCE OF THE LANDAU LEVEL SPLITTING IN THE HOLE-SIDE In the main text, Hall bar magnetoconductance measurements were presented for magnetic fields up to 3 T where quantization between electron- and hole-side differs. All integer filling factors were observed in the electron-side whereas only odd ones for holes. Additional measurements have been performed for magnetic field values B up to 5.5 T on sample 2. In this section, the Hall bar of interest is characterized by a length L = 8 µm and a width W = 2 µm. Supplementary Fig. 6 displays the longitudinal resistance Rxx as well as the Hall resistance Rxy as a function of B and Vgate for T = 100 mK. Quantum Hall effect is also well-defined for this sample with vanishing minima of Rxx directly associated with plateaus of Rxy. Using the same analysis than in the main text of the letter, the black dashed lines represent the expected filling factors from the low-magnetic field Hall measurement. These two mappings unambiguously evidence the appearance of the ν = −2 plateau for B ≥ 5 T. This is emphasized on Supplementary Fig. 7 where two traces of σxy for both B=3 T and B=5.3 T are displayed. A very clear and well-defined ν = −2 plateau exists for B=5.3 T whereas for B=3 T the quantization in the hole side jumps from -1 to -3. The observed difference in the required magnetic field to resolve the Landau level splitting in the hole- and electron- side is explained by the presence of the Γ8,HH bulk band. It results in the enlargement of the Landau level (LL) width in the hole-side and, in comparison with the electron-side, it then requires larger magnetic fields to resolve the even filling factors. FIG. 6: (a) Rxx as a function of Vgate and the perpendicular magnetic field B. The black dashed lines represent the expected position of the minima of Rxx from the density evolution with Vgate extracted at low magnetic field (see Fig. 1). (b) Hall resistance Rxy (in unit of h e2 ) as a function of Vgate and B. These data are extracted from sample 2. Vgate(V)B(T)34521-0.4-0.3-0.2-0.10.00.1Vgate(V)B(T)4521-0.4-0.3-0.2-0.10.00.1-3-2-11233100102104Rxx(Ω)a-101Rxy(h/e2)b8 FIG. 7: Appearance of even filling factor on the hole-side for high magnetic fields. Hall conductance σxy as a function of Vgate at B = 3 T (red) and B = 5.3 T (blue) for sample 2. Rxx TEMPERATURE DEPENDANCE Quantum Hall effect measurements have been performed for temperatures going from the mK range to several Kelvins. In the main text, such analysis is used to extract the activation gap energy ∆E for odd and even filling factors presented in Fig. 4. Supplementary Fig. 8 displays typical Rxx of sample 1 as a function of the gate voltage for the different temperatures. The energy gap ∆E associated to each filling factors are evidenced as Rxx minima. As expected, these resistance minima are increasing with temperature. One can notice that they are still visible at 20 K in the electron-side while they disappear around T= 4 K in the hole-side. Such difference is interpreted as the impact of the Γ8,HH bulk band on the hole-side. Such thermally activated behavior in the electron-side is fitted using Arrhenius law and evidences the Dirac nature of the charge carriers. σxy(e2/h)B = 3 TB = 5.3 T43210-1-2-3-0.4-0.3-0.2-0.10.00.1Vgate(V)9 FIG. 8: Rxx as a function of Vgate for temperatures T ranging from 0.5 K to 20 K for sample 1. The inset is a zoom on the electron-side. The thermally activated behavior of Rxx minima is evidenced. Rxx(kΩ)Vgate(V)-1.0-0.8-0.6-0.4-0.20.00.20.4020406080-0.4-0.20.00.20.4Rxx(kΩ)Vgate(V)T=20 KT=0.5 Kν=1ν=2ν=30246LANDAU LEVEL BROADENING 10 In the main text, Landau level broadening Γ estimation is used to analyse the conditions of appearance of the splitting for both electron- and hole-sides. As explained in [2] and [3], the analysis of the Shubnikov-de Haas (SdH) oscillation amplitude as a function of the magnetic field allows to obtain values of LL broadening through the deter- mination of the quantum scattering lifetime τq: ∆Rxx 4R0 = γth exp(− π wcτq ) (3) where ∆Rxx is the amplitude of the SdH oscillations, R0 is the resistance background of the oscillations, γth describes the temperature effect on the LL broadening, and wc = eB m∗ is the cyclotron frequency. The cyclotron effective mass m∗ value is prior determined from the analysis of the SdH oscillation amplitude as a function of the temperature [4] as explained in the following subsection. Estimation of the carrier cyclotron effective mass The cyclotron effective mass can be extracted from the temperature dependence of the SdH oscillation analysis using the Lifshits-Kosevich (LK) formula: ∆R(B, T ) ∆R(B, Tref ) T sinh (2π2kBTref m∗/eB) Tref sinh (2π2kBT m∗/eB) = (4) with kB the Boltzmann constant. Figure 9(a) displays Rxx as a function of B for Vgate = 0.4 V. The temperature T is increased from 0.4 K to 20 K with 0.1 K steps below 1 K, 0.8 K steps between 1 and 5 K and finally 5 K steps between 5 and 20 K. The oscillation amplitude is decreasing with T and the oscillations become hardly distinguishable at 20 K. The determination of the carrier cyclotron effective mass m∗ is thus possible using the data from Fig. 9(a) and equation 4 with Tref = 0.4 K. Note that amplitude variations of the oscillations have been studied for B close to 1 T to avoid spin splitting and only consider the surface state carrier intrinsic properties. The curve of ∆Rxx(T )/∆Rxx(T = 0.4K) (see Fig. 9(b)) is fitted using the red dashed line demonstrating m∗ value of about 0.027 ± 0.004 m0. However, such fit is not so accurate. This is mainly due to the lack of data at larger temperatures but as can be seen in Fig. 9(a), oscillations for T larger than 10 K are difficult to resolve for small B. Due to this issue, determination of m∗ is difficult in our structures. Similar study, but in the hole-side, is presented in Fig.9(c) and (d). In this regime, the SdH oscillations disappear very quickly with temperature and are no more distinguishable above 3.4 K. This behavior is attributed to the influence of the Γ8,HH bulk band. To properly estimate m∗ in this regime, we consider only T from 0.4 to 1 K, range of T for which the oscillations are the best defined. Amplitude dependence of SdH oscillations is reported in Fig.9(d) and considerably differs from what has just been reported for electrons. No saturation is visible at the lower temperatures, but on the contrary a striking increase is noticeable. Such a behavior can not be fitted by a standard metallic LK model. The coupling with bulk Γ8,HH states in the hole regime is expected to be at the origin of this particular feature. Nevertheless, it is possible to have an idea of the carrier cyclotron effective mass in our HgTe structures by using analogy with graphene. Due to a linear energy spectrum, one expects a zero effective mass for Dirac fermions. However, as already reported for graphene [5], Dirac fermions rest mass is zero but not the cyclotron mass. Note that SdH oscillation amplitude dependence with temperature gives directly access to this cyclotron mass [4]. However, standard definition for m∗ relies on energy second derivative and is only adapted for materials defined with parabolic dispersion. An alternative definition has been developed for materials with linear dispersion law [6]: m∗ = k vF = = p δE δp √ πn vF (5) This relation directly links the cyclotron effective mass m∗ and the carrier density n. Fig. 10 displays the evolution of m∗, expected from the density variation (see Fig. 1(c) in main text), as a function of Vgate. Note the red square which represents the effective mass estimated just before. These two methods appear to give consistent results. Therefore, we will use relation 5 to determine m∗ thereafter. 11 FIG. 9: SdH oscillations of the longitudinal resistance Rxx of sample 1 for Vgate = 0.4 V (a) and for Vgate = −1 V (c) for temperatures ranging from 0.4 K to 20 K. (b) The corresponding Lifshits-Kosevich (LK) fit for the extraction of the cyclotron effective mass m∗ on the electron-side. (d) Amplitude variations of the SdH oscillations on the hole-side. T(K)Rxx(Ω)051015200.00.20.40.60.81.0B(T)0.00.51.01.52.02.53.0ΔRxx(T)/ΔRxx(T=0.4K)0100200300400500abT=0.47KT=207KVgate=70.4VVgate=70.4VRxx(kΩ)B(T)0.00.51.01.52.02.53.00246810T(K)051015200.00.20.40.60.81.0ΔRxx(T)/ΔRxx(T=0.4K)dcT=0.47KT=207KVgate=7-1VVgate=7-1VLK7fitm=0.0277±70.0047m012 FIG. 10: Carrier cyclotron effective mass for particles with linear energy dispersion. Evolution of m∗ as a function of Vgate for a Dirac cone (blue dotted line). Estimation of m∗ using SdH oscillations has been performed for Vgate = 0.4 V and is represented by a red square. Error bar corresponds to the LK fit error. Estimation of the LL broadening Due to a non negligible bulk contribution in the hole-side, the LL broadening estimation method of equation 3 has only been performed in the electron-side where amplitude of SdH oscillations only depends on the surface state properties. τq is then directly related to Γ through Γ = . 2τq Figure 11(a) presents the evolution of Rxx as a function of the inverse of the magnetic field 1/B for Vgate = 0 V at T = 0.4 K. The red dots represent both the maxima and the minima which define an envelope function. ∆Rxx is extracted from the difference between successive maxima and minima and R0 results from the mean value of these extrema. Figure 11(b) displays the evolution of ∆Rxx/R0 as a function of 1/B in a logarithmic scale. The red dashed line represents the linear fit to the data using equation 3. The extracted value of τq provides direct access to the broadening value here equal to 3.91 ± 0.04 meV. The same analysis has been performed for several values of Vgate. Figure 12(a) displays the extracted values of electron-side broadening Γe. Note that the broadening is decreasing with Vgate and thus with density n. This behavior is due to the variations of m∗ with n and is consistent with Rxx mapping of Fig. 2(a) (see main text) Indeed, we can see that the closer are the LLs from the Dirac point, the more important is the magnetic field value at which the splitting appears. This is especially visible with the ν = 2 plateau appearing for B ≥ 1.5 T while ν = 4 shows up at larger gate voltage value for B ≥ 1.0 T. To have an estimation of the broadening in the hole-side, we have fitted N = 1 and N = −1 LL peaks with a Gaussian distribution to extract the full width at half maximum (see inset of Fig. 12(b)). Figure 12(b) demonstrates the difference through the ∆Vgate values between hole and electron peaks. With a ratio of about 3.7, the hole broadening Γh has been estimated to be larger than 11 meV. More precise determination of Γh is complicated as it also depends on the gate voltage. m (m0)Vgate (V)-0.4-0.20.00.20.40.0000.0050.0100.0150.0200.0250.030-0.3-0.10.10.313 FIG. 11: Estimation of the Landau level broadening for Vgate = 0 V. (a) Longitudinal resistance Rxx as a function of the inverse of the magnetic field for Vgate = 0 V and T = 0.4 K for sample 1. The red dots represent maxima and minima defining the envelope function. The difference between maxima and minima gives ∆Rxx. (b) ∆Rxx normalized by the background R0 as a function of the inverse of the magnetic field. Blue dots represent the data and the red dashed line is a fit using relation 3. FIG. 12: Estimation of the Landau level broadening on the electron- and the hole-sides. (a) Determination of the LL energy broadening Γe as a function of Vgate in the electron-side for sample 1. (b) Estimation of the energy broadening Γh on the hole-side using comparison between the full width at half maximum ∆Vgate of the Rxx peaks for the N = 1 (blue) and N = −1 (red) peaks (shown in the inset) as a function of B. Log(ΔRxx/4R0)1/B(T-1)-0.9-0.8-0.7-0.6-0.5-0.41/B(T-1)-0.9-0.7-0.5-1.1Rxx/4 (Ω)0100200300400500-1001.00.0-1.0-2.0-3.0abΓe(meV)3.13.23.32.72.82.93.0-0.3-0.2-0.10.00.10.20.32.52.62.72.82.93.0B(T)ΔVgate(V)0.120.100.080.060.020.04-1.0-0.8-0.6-0.4-0.20.0Rxx(104Ω)3210ab3.43.5Vgate(V)Vgate(V)B=2.9TLANDAU LEVEL SPLITTING APPEARANCE: DISCUSSION Supplementary Fig. 13 summarizes on the same diagram all the relevant energy scales discussed in the letter. All the energy gaps ∆E of both odd and even filling factors for different magnetic fields discussed in Fig. 4 (see main text) are displayed as squares. The color bands correspond to the Landau level broadening Γe and Γh extracted in the previous section. If the energy gaps are smaller than Landau level broadening, they can not be resolved by the quantum Hall transport spectroscopy. This diagram suggests the appearance of even filing factors in the electron side for Be ≥ 2.4 T and in the hole side for Bh ≥ 6.3 T. Such values are consistent in term of ratio with experimental data for which Be and Bh were estimated around 1.5 T and 4.5 T respectively. 14 FIG. 13: Final fan diagram with the odd ν=1, 3, 5, 7 (black squares), and even ν=2 (red squares) filling factors and the "effective Zeeman splitting" energy (red dotted line). Γh (respectively Γe) range is represented by a red (blue) color band. Note that the energy ranges considered for broadening values take into account variations with gate voltage. MODEL AND ORIGIN OF THE LANDAU LEVEL SPLITTING Model We consider a thin layer of topological insulator material, with two parallel surfaces indexed by τ z = +1 for the top (T) and τ z = −1 for the bottom (B) surface. The effective Hamiltonian for the surface states reads H = vf τz ⊗ (z × (cid:126)σ).(−i(cid:126)∇) + τx ⊗ I + V τz ⊗ I, (6) (cid:18) ∆ 2 2∇2 2M + (cid:19) where vf is the Fermi velocity of the surface states and σx, σy describes Pauli matrices acting on the spin Hilbert space while τα acts on the T/B space. The first term describes the Dirac dispersion relation for the top and bottom surface states with opposite chirality (the spin winds in opposite ways around the Fermi surface), while ∆/2 − (k)2/2M is a momentum dependent tunneling between the two surfaces, and V the chemical potential difference between the surfaces. The band structure corresponding to this model is symmetric with respect to E = 0, and reads (cid:18) ∆ 2 − 2k2 2M (cid:19)2 E2±(k) = + (V ± vfk)2 . (7) 6420B(T)ΔE(meV)73102681012ΓhΓe2βBν=1ν=3ν=5ν=7ν=2Landau spectrum 15 We consider a magnetic field B applied along z, perpendicular to the two surfaces. We obtain a spectrum with a chiral symmetry with respect to E = 0. The Zeeman coupling is neglected in the Hamiltonian (6), so the effect of the magnetic field is purely orbital, and enters into the Hamiltonian by the minimal substitution −i(cid:126)∇ → −i(cid:126)∇ − e (cid:126)A where (cid:126)A is the electromagnetic vector potential. In the gauge (Ax = 0, Ay = Bx), we introduce the ladder operators a = q = 1√ 2 (q − ip), x − kyl2 (q + ip) ; a† = 1√ 2 X lB B = /eB. The Hamiltonian in the presence of a magnetic field can be ; p = −ilB∂X , (9) (8) l2 B = B where we introduced the magnetic length l2 rewritten as H = −η τz ⊗(cid:0)aσ− + a†σ+ (cid:1) + (cid:18) ∆ 2 (cid:19) − ω 2 (2a†a + 1) τx ⊗ I + V τz ⊗ I where we introduced 2σ± = σx ± iσy and η2 = 2 2v2 l2 B f = 2v2 F eB ; ω = 2 M l2 B eB M . = We consider the basis N(cid:105) of quanta of the operators a, a†. In the basis T,↑, N(cid:105),T,↓, N − 1(cid:105),B,↑, N(cid:105),B,↓, N − 1(cid:105), the reduced Hamiltonian reads for N ≥ 1: √ V,−η √ ∆ −η 2 − ω 2 − ω 0, ∆ Hn = √ 2 (2N + 1), 0 2 (2N − 1) √ N N ,−V 2 − ω N , ∆ 2 − ω N , V, 0, ∆ 2 (2N + 1), 0,−V, η 2 (2N − 1), η (cid:19)2 (cid:18)ω (cid:18) ∆ 2 + 2 − Nω (cid:21) (cid:19) − Nω (cid:20)(cid:18) ∆  = τx ⊗ (cid:32) (cid:19)2 ± N η2(cid:0)4V 2 + (ω)2(cid:1) + (ω)2 ω 2 I + σz 2 E2 N,± = V 2 + N η2 + Its spectrum is symmetric with respect to 0 (chiral symmetric) and the energies satisfy √ N τz ⊗ σx + V τz ⊗ I. (12) − η (cid:19)2(cid:33) 1 2 for N (cid:54)= 0, (13) − Nω (cid:18) ∆ 2 which identifies exactly with the spectrum of [7] with the notation B = 2/2M . and by a†σ+, and the effective Hamiltonian is then: In the case N = 0, we have to consider the basis T,↑, 0(cid:105) and B,↑, 0(cid:105). These basis states are annihilated by aσ− with eigenvalues ±(cid:112)V 2 + (ω − ∆)2/4. H0 = V τz ⊗ 11 + (cid:18)ω − ∆ (cid:19) 2 τx ⊗ 11, (10) (11) (14) (15a) (15b) Landau spectrum with V = 0 In the special case V = 0, we can rewrite Eq. (13) as E2 N,± = N η2 + + − Nω (cid:18)ω (cid:19)2 (cid:18)ω − ∆ (cid:19)2 2 (cid:18) ∆ 2 E2 0 = 2 (cid:19)2 ± ω (cid:32) N η2 + (cid:19)2(cid:33) 1 2 (cid:18) = − Nω (cid:18) ∆ 2 N ± ω 2 (cid:19)2 where (cid:32) N = N η2 + (cid:33) 1 2 (cid:19)2 . − Nω (cid:18) ∆ 2 16 (16) Hence the degeneracy of the Landau levels N of the two relativistic surfaces corrected by the non-relativistic couplings is lifted by a splitting ±ω/2 (±β in the notation of the main text). The N = 0 Landau level was initially not degenerate. Note that while this splitting is not of magnetic origin, it is linear in magnetic field, and can be viewed as an "effective Zeeman splitting" originating from the k2 dependence of hybridization term between the Top and Bottom surface [7]. Such Landau energy spectrum is represented in Supplementary Fig. ??. Nature of the "Landau Level Splitting" When V = 0 and in the limit ∆ → 0, ω = 0 the reduced Hamiltonian (12) reads √ N = −η H (0) N τz ⊗ σx. The corresponding Laudau eigenstates are (in the gauge we have chosen): ψ+ N,+ = ψ− N,+ = 1√ 2 1√ 2 (T,↑, N(cid:105) − T,↓, N − 1(cid:105)) , (B,↑, N(cid:105) − B,↓, N − 1(cid:105)) , ψ+ N,− = ψ− N,− = 1√ 2 1√ 2 (B,↑, N(cid:105) + B,↓, N − 1(cid:105)) for N = N η, √ (T,↑, N(cid:105) + T,↓, N − 1(cid:105)) for N = − N η. √ They are entirely localized in either the top (T) or bottom (B) surface, and carry no net magnetization: (cid:104)ψ+ N,±Sxψ+ N,±(cid:105) = (cid:104)ψ+ N,±Syψ+ N,±(cid:105) = (cid:104)ψ+ N,±Szψ+ N,±(cid:105) = 0. (17) (18a) (18b) (19) Let us now discuss the effect of various terms of the Hamiltonian (12), treated as perturbations of the relativistic Landau Hamiltonian (17): • Chemical Potential Asymmetry V . The corresponding term is +V τz ⊗ I which merely shifts the energies of the Landau levels of the top surface with respect to those of the bottom surface without modifying their nature. • Top / Bottom Hybridization: the corresponding term is(cid:0) ∆ 2 − Nω(cid:1) τx ⊗ I. This operator will shift the energies of the Landau levels and delocalize the eigenstates on both top and bottom surfaces but without any splitting between the two degenerate eigenstates. This can be inferred from the matrix elements (20) • Splitting. The most striking consequence of the coupling between the two surface states arises from the k2 term τx ⊗ Iψ± N,s(cid:105) = ψ∓ N,s(cid:105) ⇒ (cid:104)ψ+ N,+τx ⊗ Iψ+ N,−(cid:105) = 0. in (6) and leads in the reduced Hamiltonian (12) to a term HSplitting = ω 2 τx ⊗ σz, which satisfies (21) HSplittingψ+ N,+(cid:105) = ψ+ N,−(cid:105). (22) Indeed, while the two eigenstates ψ+ N,±(cid:105) carry no magnetization, the spin and orbital degree of freedom (eigen- states N ) are tightly bound together in their structure. This is illustrated by the absence of magnetization along x. It is this relation between spin and orbital degree of freedom which is broken by the perturbation HSplitting, whose eigenstates are symmetric (S) or antisymmetric (A) between the two surfaces and read respectively for the positive and negative eigenvalues: ↑, S(cid:105) = ↑, A(cid:105) = 1√ 2 1√ 2 (T,↑, N(cid:105) + B,↑, N(cid:105)) , (T,↑, N(cid:105) − B,↑, N(cid:105)) , ↓, A(cid:105) = ↑, S(cid:105) = 1√ 2 1√ 2 (T,↓, N − 1(cid:105) − B,↓, N − 1(cid:105)) , (T,↓, N − 1(cid:105) + B,↓, N − 1(cid:105)) . (23a) (23b) Note that these eigenstates carry a magnetization along z: the splitting of the Landau level eigenstates will be associated with the appearance of finite magnetization along z of the eigenstates, linear in magnetic field (in ω). 17 FIG. 14: Energy spectrum as a function of magnetic field for LL index N ranging from 0 to 4 on both electron- and hole-sides. LL splitting as well as hybridization gap ∆ are evidenced. ∆ has been set to 2 meV, 2β to 2.07 meV/T and vf to ≈ 4.105 m.s−1. Bibliography: [1] Konig, M. et al. Quantum spin Hall insulator state in HgTe quantum wells. Science, 318, 766-770 (2007). [2] Hong, X., Zou, K. & Zhu, J. Quantum scattering time and its implications on scattering sources in graphene. Phys. Rev. B, 80, 241415(R) (2009). [3] Young, A.F. et al. Spin and valley quantum Hall ferromagnetism in graphene. Nature Physics, 8, 550-556 (2012). [4] Fuchs, J.-N. Dirac fermions in graphene and analogues: magnetic field and topological properties. Habilitation diriger des recherches (2013). [5] Novoselov, K. et al. Two-dimensional gas of massless Dirac fermions in graphene. Nature, 438, 197-200 (2005). [6] Ariel, V. & Natan, A. Electron Effective Mass in Graphene. arXiv, 1206.6100 (2012). [7] Zhang, S., Lu, H. & Shen, S. Edge states and integer quantum Hall effect in topological insulator thin films. Scientific Reports, 5, 13277 (2015). B(T)0.00.51.01.52.02.53.0N=0+N=0-N=1+N=1-ħω = 2βBN=2+N=2-ν=1ν=3ν=5N=3+N=3-ν=-1E(meV)6040200-20-40-60ΔN=4+N=4-ν=7
1006.3748	2	1006	2010-09-13T15:48:16	Dirac-Kronig-Penney model for strain-engineered graphene	[ "cond-mat.mes-hall" ]	Motivated by recent proposals on strain-engineering of graphene electronic circuits we calculate conductivity, shot-noise and the density of states in periodically deformed graphene. We provide the solution to the Dirac-Kronig-Penney model, which describes the phase-coherent transport in clean monolayer samples with an one-dimensional modulation of the strain and the electrostatic potentials. We compare the exact results to a qualitative band-structure analysis. We find that periodic strains induce large pseudo-gaps and suppress charge transport in the direction of strain modulation. The strain-induced minima in the gate-voltage dependence of the conductivity characterize the quality of graphene superstructures. The effect is especially strong if the variation of inter-atomic distance exceeds the value a^2/l, where a is the lattice spacing of free graphene and l is the period of the superlattice. A similar effect induced by a periodic electrostatic potential is weakened due to Klein tunnelling.	cond-mat.mes-hall	cond-mat	Dirac-Kronig-Penney model for strain-engineered graphene S. Gattenlohner,1 W. Belzig,2 and M. Titov1, 3 1School of Engineering & Physical Sciences, Heriot-Watt University, Edinburgh EH14 4AS, UK 2Department of Physics, Konstanz University, D -- 78457 Konstanz, Germany 3Institut fur Nanotechnologie, Karlsruhe Institute of Technology, 76021 Karlsruhe, Germany (Dated: November 8, 2018) Motivated by recent proposals on strain-engineering of graphene electronic circuits we calculate conductivity, shot-noise and the density of states in periodically deformed graphene. We provide the solution to the Dirac-Kronig-Penney model, which describes the phase-coherent transport in clean monolayer samples with an one-dimensional modulation of the strain and the electrostatic potentials. We compare the exact results to a qualitative band-structure analysis. We find that periodic strains induce large pseudo-gaps and suppress charge transport in the direction of strain modulation. The strain-induced minima in the gate-voltage dependence of the conductivity characterize the quality of graphene superstructures. The effect is especially strong if the variation of inter-atomic distance exceeds the value a2/ℓ, where a is the lattice spacing of free graphene and ℓ is the period of the superlattice. A similar effect induced by a periodic electrostatic potential is weakened due to Klein tunnelling. PACS numbers: 73.23.-b, 73.22.Pr, 73.21.Cd I. INTRODUCTION is responsible in graphene Graphene is recognized as the only two-dimensional crystal that withstands large deformations and high temperatures and is readily integrated into electronic circuits.1 -- 3 The high electron mobility and unique spec- tral characteristics single out graphene as a promising key element for electronic and optoelectronic applications.4 -- 9 The well-known Dirac dispersion of electronic ex- citations for a num- ber of interesting analogies with relativistic quantum electrodynamics,10 but prevents the use of graphene in field-effect transistors due to the absence of energy band- gaps.11,12 The strain-engineering of graphene circuits has been recently suggested as a means to bypass this difficulty.13,14 In the Dirac-fermion picture of graphene, the strain gradient is equivalent to the presence of a pseudo-magnetic field, which can be manipulated to in- duce a zero-field quantum Hall effect and a topological isolator state.15 The variation of strain may also induce substantial variation in the on-site electron energies, how- ever, this effect is reduced by screening.16 FIG. 1: (Color online) Schematic illustration of the graphene setup with metal leads for x < 0 and x > L. The angle θ specifies the orientation of the honeycomb lattice with respect to the transport direction x. The experimental realizations of strained graphene confirm its surprisingly high elasticity and pave the way for the development of strain-engineered graphene electronics.8,17 Different techniques have been proposed to produce graphene samples with controlled periodical variations of strain.18 -- 20 First transport measurements of graphene superstructures have been already reported.21 Below we analyze the phase-coherent charge transport in periodically strained graphene samples and propose a way to characterize the quality of graphene superstruc- tures on the basis of their transport properties. More specifically we use the exact solution of the Dirac- Kronig-Penney model to calculate density of states, con- ductance, and shot noise in transport through finite size graphene samples with periodic potentials. The scatter- ing off the metal leads is taken into account in all quan- tities. The position, the width, and the shape of the con- ductance minimum as well as the shot noise maximum associated with the periodic superstructure are analyzed in detail. This paper is organized as follows. In Section II the effective Dirac Hamiltonian for deformed graphene is de- rived. The scattering approach to transport is briefly de- scribed in Section III. Section IV is devoted to the exact solution of the Dirac-Kronig-Penney model for graphene with one-dimensional modulations of strain in transport direction. The generalization of this model is discussed at the end of the Section IV. The results for conduc- tance, shot-noise, and the density of states are quali- tatively understood on the basis of the band structure analysis, which is presented in Appendix A. In Section IV D we discuss transport in the direction perpendicular to the strain modulation. We summarize our results in the concluding Section V. II. EFFECTIVE HAMILTONIAN The electromechanical coupling in deformed graphene membranes has been investigated theoretically by many authors22 -- 25 following earlier publications on carbon nanotubes.26 -- 29 The deformations affect the hopping in- tegrals in the tight-binding description of graphene in two distinct ways: by changing the distance between carbon atoms and by tilting the electronic pz-orbitals, which are responsible for conduction. However, for most cases of interest, the in-plane strains play the major role in de- termining the electronic properties of deformed graphene, while the tilting can be neglected. The effects of in-plane strain are well captured by the tight-binding hamiltonian 2 approximation, which is formulated in a generic reference frame shown in Fig. 1. In this frame we find δα = a(cid:18) sin (θ + θα) − cos (θ + θα) (cid:19) , θα = 2πα/3, (4) while the positions of the two non-equivalent Dirac points in the reciprocal space are given by K1,2 = ± Using the Fourier ansatz 4π 3a√3 (cid:18) cos θ sin θ (cid:19) . (5) aR = 2 Xs=1 eiK sRas(R), bR = 2 Xs=1 eiK sRbs(R), (6) we obtain, to the leading order in spatial gradients, the effective model H = Z d2R Ψ† R [−i¯hvσ∇ + τzσAθ(R)] ΨR, (7) (1) where ¯hv = 3ta/2, τz is the Pauli matrix in the valley space and the operators are arranged into the four-spinor 3 H = −XR Xα=1 [t + δtα(R)](cid:16)a† R+δα bR + h.c.(cid:17) , where the summation runs over the atomic positions, R, of a honeycomb lattice, the annihilation operators aR and bR refer to the occupation of pz-orbitals at the two non-equivalent positions, A and B, of the unit cell, and the three vectors δα shown in Fig. 1 are directed from a B-atom to its three nearest neighbors. If the strain varies smoothly on atomic distances, the deviation of the hopping integral from its unperturbed value t ≈ 2.7 eV in a perfect crystal can be parameterized as δtα(R) = (βt/a2)δT α u(R)δα, (2) where β = −∂ ln t/∂ ln a ≈ 2 and u(R) is the strain ten- sor of the graphene membrane. The dimensionless strain tensor describes a change in a metric that can be expressed as dℓ2 = dℓ2 0 + 2uijdsidsj, (3) where the summation over the spatial indices i, j = x, y is assumed. The length elements dℓ2 0 = δijdsidsj and dℓ2 = gijdsidsj correspond to the metric δij in the flat space and to the local metric gij in the membrane, re- spectively. In subsequent formulas we let conventionally sx = x and sy = y. The graphene crystal withstands very large internal strains so that the values of the strain tensor elements may reach 20%.30 Even though the strain tensor makes no reference to the crystal structure of graphene, the lattice symmetry is entering the hopping integral in Eq. (2) due to the vectors δα, α = 1, 2, 3. The slow variation of u(R) on the scale of the lattice spacing justifies the effective mass ΨR = (cid:0) a1 eiθb1 −e−iθb2 a2 (cid:1). (8) The vector field Aθ = (Aθx, Aθy) is real and its compo- nents are related to the strain tensor in a simple way, Aθx − iAθy = (¯hvβ/a)e3iθ(uxx + 2iuxy − uyy). (9) In the derivation of the effective Hamiltonian (7) we ne- glected the velocity renormalization, which would appear as a correction of the order δtα to the prefactor of the spatial gradients.31 Unlike the usual vector potential, the vector field Aθ preserves the time reversal symmetry of the Hamiltonian (7). It also exhibits the discrete rotational invariance of the honeycomb lattice. Indeed, the Hamiltonian (7) remains invariant under the rotation through the angle θ = 2π/3, while the rotation through the angle θ = π/3 is equivalent to an interchange of valleys. The effect of a constant uniaxial strain has been stud- ied both theoretically32,33 and experimentally8,34 and falls beyond the scope of our consideration. If the con- stant uniaxial strain (in x direction) exceeds a certain critical value, the Dirac points merge and a band gap opens. It is predicted32 that, for a crystal expanded uni- formly in the zigzag direction (θ = 0), the critical expan- sion takes on its minimal value (≈ 23%), while uniaxial strain in armchair direction, θ = π/2, never generates a gap. These predictions await experimental verification. The strain also induces an electrostatic potential due to a change of the on-site energies of the tight-binding model. This effect leads to the appearance of a scalar electrostatic potential, which has to be added to the ef- fective Hamiltonian (7). III. SCATTERING APPROACH In this Section we formulate the scattering approach to transport through a deformed graphene sample with metallic leads. We take advantage of the effective single- particle Hamiltonian H = −iσ∇ + τzσAθ + V, (10) where the fictitious vector potential Aθ is related to the strain tensor, u(R), by means of the relation (9), and the scalar field V (R) describes strain-induced and exter- nal electrostatic potentials. In most of the intermediate expressions we let ¯hv = 1 for simplicity. The metal leads are modeled by letting V = −Ulead for x < 0 and x > L,35 in Eq. (10). The vector potential is assumed to be zero in the leads. The width of the sample in y direction is denoted as W . The scattering off the metal leads are fully taken into account in the subsequent analysis. The rectangular sample geometry makes it convenient to employ the Fourier transform in y, ΨR = Xq eiqyΨq(x), (11) where q is the quasiparticle momentum component par- allel to the graphene-metal interface. The momentum takes on the quantized values, q = qn, which depend on the boundary conditions in y direction, for example, qn = 2πn/W with integer n for periodic boundary con- ditions. In the limit W ≫ L, which we assume below, the particular type of the boundary conditions is not im- portant. We also restrict our consideration to small energies, ε ≪ Ulead, where the energy ε is measured with respect to the Dirac point. In this approximation we derive the equation on the transfer matrix in the form36 ∂M ∂x = hσx(q + τz Aθy) + iσz( V −ε) + iτz AθxiM, (12) where we introduced matrix notation in Fourier (channel) space, e.g. Vnm(x) = 1 W Z W 0 dy ei(qn−qm)y V (x, y), (13) and qnm = δnmqn. The transfer matrix has the following structure in σ-space M = (cid:18) 1/t′† −t−1r′ 1/t (cid:19) , rt−1 (14) where r(r′) for x = L is the matrix of reflection am- plitudes for quasiparticles entering the sample from the left(right) lead. The matrices t and t′ contain the corre- sponding transmission amplitudes. Then, the Landauer formula for the conductance can be cast in the following form 3 where the trace is referred to the valley and channel space and the symbol M11 stands for the 11 block of the trans- fer matrix in σ-space. The Hamiltonian (10) with vanishing electrostatic po- tential, V = 0, obeys the chiral symmetry σzHσz = −H, which is responsible for a non-Abelian Aharonov-Casher gauge invariance at zero energy,37 Ψ′ 0 = eiφ+τzσzχΨ0. (16) The spatially dependent phases φ and χ can be cho- sen in such a way that the zero-energy spectral equation HΨ0 = 0 is reduced to the Dirac equation, −iσ∇Ψ′ 0 = 0, with zero vector potential. This gauge transformation can be applied in the scattering approach in order to demonstrate that the presence of arbitrary vector po- tential has no effect on charge transport at the Dirac point as far as the contribution from edge states can be disregarded.38,39 IV. 1D DIRAC-KRONIG-PENNEY MODEL A. Transport One-dimensional modulations of strain were realized experimentally in suspended graphene films using the remarkably large and negative thermal expansion of graphene.18 Motivated by these experiments we calculate the transport properties of the one-dimensional Dirac- Kronig-Penney model, which has been introduced earlier by several authors.40 -- 42 Below we consider a general form of the model, where the variation of both electrostatic as well as vector potentials is included. Using the scattering approach formulated in the Section III we provide simple analytical solutions for transport and density of states in this model, which complement previous theoretical stud- ies of Dirac Fermions in periodic potentials.43 -- 52 A graphene sample with one-dimensional modulation of strain is characterized only by the component uxx(x) of the strain tensor, which depends solely on the x- coordinate. From Eq. (9), the components of the pseudo- vector potential take the form Aθx = βuxx a cos 3θ, Aθy = − βuxx a sin 3θ . (17) In addition, the strain induces a spatial variation of elec- trostatic potential, V = V (x). Its relation to the strain tensor is, however, complicated by screening effects. Since the potentials depend only on x, the transverse momentum q is conserved. In this case the matrices V , Aθx and Aθy in Eq. (12) are diagonal in channel space. Since the considered potentials also do not couple the valleys, the scattering problem is reduced to the solution of 2 × 2 matrix equation, G = 2e2 h Tr tt† = 2e2 h Tr h M11 M† 11i−1 , (15) ∂M ∂x = [σx(q + Aθy) − iσz(ε − V ) + iAθx]M. (18) gauge transformation M → M exp(iR x The x-component of the vector potential, Aθx, enters the equation in a trivial way and can be excluded by the 0 Aθxdx′), which does not affect any observable. The first and the most trivial consequence of this transformation is that the one- dimensional strain modulations in x direction have no effect on transport in the zigzag direction (θ = 2πn/3 with integer n) provided the effect of V is ignored. For other angles the one-dimensional strain modula- tions lead to the appearance of new minima in the gate voltage dependence of the conductivity of the graphene sample. The numerical solution of Eq. (18) suggests that a periodic vector potential induces much more pro- nounced minima than a periodic scalar potential of equiv- alent amplitude. The weaker effect of V can be associated with Klein tunnelling, which leads to the suppression of pseudo-gaps in the spectrum of the superstructure. In order to describe the effect of periodic potentials analytically we introduce the one-dimensional Dirac- Kronig-Penney model, which is characterized by the vec- tor potential A given by Eq. (17) with uxx(x) = −u0/2 + u0Xj Θ(x− 2ℓj) Θ((2j+1)ℓ−x), (19) where the scale ℓ stands for half the period of the super- lattice, Θ(x) is the Heaviside step function, and the di- mensionless parameter u0 specifies the amplitude of the strain modulation. The vector potential introduced by Eqs. (17,19) corresponds to a strain field which is smooth on atomic scale but changes abruptly on distances smaller than the Dirac quasiparticle wave length, ¯hv/ε. The pe- riodic electrostatic potential is introduced in a similar manner, V (x) = −V0/2 + V0Xj Θ(x − 2ℓj) Θ((2j+1)ℓ−x), (20) where V0 is the amplitude of the modulation. For the sake of simplicity we consider a sample of length L = 2ℓN with N being an integer. In this case the solution to Eq. (18) satisfies M(L) = MN 0 , where M0 is the transfer matrix corresponding to the size of the supercell, 2ℓ. Disregarding the x-component of the vector potential, Aθx, which has been argued above to have no effect on transport, we write M0 = M+M−, M± = e(q±σx−iε±σz )ℓ, (21) where we take advantage of the definitions q± = q ± (βu0/2a) sin 3θ, ε± = ε ± V0/2. (22) The eigenvalues of the transfer matrix, M0, are conve- niently parameterized as e±2ik0ℓ = λ ± ip1 − λ2, where we introduced the real function (23) λ = cos k+ℓ cos k−ℓ+ q+q−−ε+ε− k+k− sin k+ℓ sin k−ℓ, (24) 4 4 10π 4 50π i h ℓ 2 / W 2 e h G 10 8 6 4 2 0 0.5 0.4 0.3 F 0.2 0.1 0 0 0 0.1 0.2 N = 10 N = 50 N (cid:1) ∞ 1 3 0 0 0.1 0.2 π 2π 2ℓE/¯hv 3π FIG. 2: (Color online) Conductance and Fano factor calcu- lated from Eq. (28) for finite systems comprising N = 10 and N = 50 supercells are compared with the averaged ex- pressions to be derived in Eqs. (33,34), which correspond to the limit N → ∞ and ignore contributions from evanescent modes. The plots are for graphene with strain modulations, rθx = 1. In a small vicinity of the Dirac point, ε ≪ ¯hv/L, (shown at the insets) the transport is insensitive to strain due to the extended gauge invariance (16) and is dominated by the evanescent modes. ±−q2 with k± = (ε2 ±)1/2 the x-component of the momenta. The wave number k0 becomes imaginary for some values of q and ε indicating the appearance of pseudo-gaps in the superstructure. In the considered Dirac-Kronig-Penney model the value of k0 is nothing but the x-component of the quasi-momentum in the reciprocal space associated with the superstructure. In order to calculate the element M11 of the full transfer-matrix M(L) we use the Chebyshev identity to calculate the N -th power of an unimodular matrix53 0 = M0 sin k0L − sin k0(L−2ℓ) . (25) MN sin 2k0ℓ The matrix element (M0)11 = λ − iη is readily deter- mined from Eqs. (21,25) with η = ε− k− sin k−ℓ cos k+ℓ + ε+ k+ sin k+ℓ cos k−ℓ. (26) Using the fact that both λ and η are real functions of the energy ε, and the conserved momentum component q, we obtain from Eqs. (14,25) the exact transmission probabilities for each value of q, Tq = (cid:20)cos2 k0L + η2 1 − λ2 sin2 k0L(cid:21)−1 . (27) 5 v ¯h / ℓ ε 2 v ¯h / ℓ ε 2 3π 2π π 0 3π 2π π 0 r0 = 0.0 rθx = 0.1 r0 = 0.1 rθx = 0.0 rθx = 2.0 r0 = 0.0 rθx = 0.0 r0 = 2.0 −3π −2π −π 0 π 2π 3π −3π −2π −π 0 π 2π 3π 2qℓ 2qℓ 1 0.8 0.6 0.4 0.2 0 rθx 0.1 2 0 0 r0 0 0 0.1 2 i h ℓ 2 / W 2 e h ¯G 10 8 6 4 2 0 0.5 0.4 0.3 ¯F 0.2 0.1 0 0 π 2π 2ℓE/¯hv 3π FIG. 3: The density plot for the averaged transmission spec- tra, T q(ε), calculated from Eq. (32) for the Dirac-Kronig- Penney model with strain modulation rθx = 0.1 (top left), rθx = 2.0 (bottom left) and electrostatic field modulation r0 = 0.1 (top right), r0 = 2.0 (bottom right). This expression is reduced to the well-known result for a purely ballistic system,35 i.e. for u0 = V0 = 0, by the substitution η = (ε/k) sin 2kℓ and λ = cos 2kℓ. The transmission probabilities (27) determine the energy- dependent transport quantities G = 4e2 h Xq Tq, F = Pq Tq(1 − Tq) Pq Tq , (28) where G is the Landauer conductance and F is the Fano factor for the shot noise. In the limit W ≫ L the sum- mation over the scattering channels can be replaced by the integration, Pq → (W/2π)R dq. More generally one can define the cumulant generating function for the transport, F (χ) = Xq ln(1 − Tq + eχTq). (29) The dimensionless cumulants cn = limχ→0 ∂nF /∂χn de- termine the so-called full counting statistics of the charge transport. The conductance and the Fano factor are given by G = (4e2/h)c1 and F = c2/c1, respectively. The strength of strain-induced and electrostatic poten- tials in the Dirac-Kronig-Penney model (19,20) is char- acterized by the dimensionless parameters rθx = (2u0ℓ/a) sin 3θ, r0 = ℓV0/¯hv, (30) respectively. In Fig. 2 the conductance and the Fano factor in graphene with periodic modulations of strain calculated from Eq. (28) are plotted for systems with finite lengths L = 20ℓ and L = 100ℓ for rθx = 1. FIG. 4: (Color online) The averaged conductance and Fano factor calculated from Eqs. (33,34) for the one-dimensional Dirac-Kronig-Penney model with different field strengths (30). The effect of the scalar potential is suppressed due to Klein tunnelling through potential barriers. The energy dependence of conductance and Fano- factor (28) reveal fast Fabry-P´erot oscillations on the scale ¯hv/L. From a physics point of view, these oscil- lations originate from multiple reflections of propagating modes (real k0) at the metal-graphene interfaces. The channels with imaginary values of k0 (evanes- cent modes or metal-induced states) also contribute to transport. Even though the individual contribution of each evanescent mode is exponentially small, Tq ∼ exp(−2LIm k0), their combined effect becomes essential for energies in a vicinity of band-edges. The role of evanescent modes is especially important at the Dirac point due to the absence of propagating modes. Indeed, the conductance and the Fano-factor determined by (28) take on the values G = 4e2W/πhL and F = 1/3 for ε ≪ ¯hv/L irrespective of the vector potential. This uni- versality is due to the extended gauge invariance (16). For N = L/2ℓ ≫ 1, the amplitude of the Fabry-P´erot oscillations in G and F decreases (provided W ≫ L) and the relative contribution of evanescent modes becomes less important. Taking the limit N → ∞ is equivalent to ignoring the imaginary values of k0 and averaging over the rapid phase, k0L. This approximation has been used e.g. in Ref. 39 to obtain the full counting statistics of few-layer graphene. The averaged generating function (29) takes the form F (χ) = 2Xq ln(cid:18)eχ/2 +qT −2 q + eχ − 1(cid:19) , (31) 3π 2π ℓ q 2 π 0 8 6 4 2 0 ] h ℓ 2 / W 2 e [ G 1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 0 π 2π 3π 2ℓε/¯hv FIG. 5: (Color online) The averaged conductance of Eq. (33) (solid) is compared with an estimate (35) based on the mode- counting argument (dashed) for the Dirac-Kronig-Penney model with rθx = 0.3, r0 = 0. The upper panel shows the corresponding transmission spectrum. where we introduce the mean transmission probability T q = η−1Rep1 − λ2. (32) The averaged transport quantities do not depend on the system size L and reveal no Fabry-P´erot oscillations. For a ballistic sample, u0 = V0 = 0, one finds T q = k/ε, where k = pε2 − q2 is the x-component of the momen- tum. ball From the generating function (31) we readily find the averaged conductance and noise. The conductance is given by the Landauer formula, G = 4e2 h Xq T q, (33) while the Fano-factor is related to the averaged transmis- sion probabilities in a less evident way F = Pq T q(cid:16)1 − T 2Pq T q 2 q(cid:17) . (34) The exact and averaged transmission probabilities (27,32) together with the corresponding expressions for the full counting statistics (29,34) provide the complete analytical description of the transport properties in the 1D-Dirac-Kronig-Penney model with scalar and vector potentials of arbitrary strength. 6 rθx 0.3 1 2 0 r0 0 0 0 2 ℓ 4 / ρ 2 π v ¯h 20 15 10 5 0 0 1 2 3 4 5 6 2ℓε/¯hv FIG. 6: (Color online) The averaged density of states obtained from Eq. (43) for the Dirac-Kronig-Penney model with differ- ent strengths of the periodic potentials. The dependence of the transmission coefficient on the transversal momentum component, q, is called the trans- mission spectrum. In Fig. 3 we plot the transmission spectra obtained from Eq. (32) for different strengths of the scalar and vector potentials. The transmission spectra shown in Fig. 3 are qualita- tively different for weak, rθx ≪ 1, and strong, rθx ≫ 1, strain modulations. It has to be stressed that the latter regime is well within the experimentally accessible range of parameters. Indeed, for an achievable strain modula- tion of 0.2% (u0 = 0.002) and the period ℓ = 70 nm one finds rθx ≈ 2. Neither the modulated strain nor the electrostatic field does open up a full band gap for any values of the pa- rameters rθx or r0. Nevertheless, very large pseudo-gaps are generated by strain modulations with rθx ≫ 1. One can see from the lower left panel in Fig. 3 that the charge transport in this regime is suppressed for almost all di- rections of the momenta (different q values) in a wide energy range. In contrast, the pseudo-gaps do not open at normal incidence (q = 0) in the Dirac-Kronig-Penney model with periodic electrostatic potential due to the Klein-tunnelling of quasiparticles through electrostatic barriers.54 The effects of strain and electrostatic field modulations on the conductance and noise are compared in Fig. 4. The most prominent feature of these plots are the dips in conductivity centered around the energies εn = nπ¯hv/2ℓ with n = 1, 2, 3, . . . . The physical mechanism, which is responsible for the appearance of the pseudo-gaps, is equivalent to that of the parametric resonance.55 For rθx ≫ 1, the conductance is suppressed in wider energy intervals around ε = εn. For weak strain or po- tential amplitude, rθx ≪ 1 or r0 ≪ 1, significant pseudo- gaps are only located around the energy values εn with odd values of n, since only odd Fourier components of the potentials (19,20) exist. For a weak harmonic potential, a significant pseudo-gap would only arise around ε = ε1. The areas of vanishing transmission probability in Fig. 3 (top left) can be understood on the basis of pertur- bation theory as presented in Appendix A for small am- plitudes of periodic potentials. The pseudo-gap emerging due to a weak periodic strain is described by the func- tions εg−(q) and εg+(q) given in the first row of Tab. I. These functions are shown with dashed lines in the upper panel of Fig. 5. ball Inside the pseudo-gap, i.e. q = pε2 − q2/ε, for εg−(q) < ε < εg+(q), the transmission coefficient is exponentially small, while it is close to the ballistic value, T outside the pseudo-gap (one can disregard the change in the resistance of the metal-graphene interface due to the weak periodic potentials). Note further, that the pseudo- gap is located near q = 0 for energies near the dip such that T q ≈ 1 in this region. We can, therefore, estimate the conductance in the vicinity of the lowest dip (ε = ε1) by subtracting the contribution of ballistically propagat- ing modes inside the pseudo-gap from the ballistic result as ball Gmc = G0 − 4e2 h Xq Θ(ε−εg−(q)) Θ(εg+(q)−ε), (35) where Θ(x) is the Heaviside step function, G0 = e2W ε/h, and the limit Wε ≫ 1 is assumed. The generalisation of Eq. (35) around the higher resonant energies, ε = εn, n = 3, 5, . . . , is straightforward. It is shown in Fig. 5 for rθx = 0.3 that the re- sult of Eq. (35) agrees with the averaged conductance calculated from Eq. (33). The band-structure analy- sis (35,A5) predicts characteristic dips in the conduc- tance at ε = nπ¯hv/2ℓ, n = 1, 3, 5 . . . of the depth δG = (8e2W/hπ)pAnnπ/2ℓ, where An stands for n- th Fourier-component of the vector potential. For the Dirac-Kronig-Penney model one finds An ∝ 1/n hence the value of the conductance dip does not depend on n. We note that the validity of Eq. (35) is restricted to weak potentials. For rθx, r0 > the perturbation theory of Appendix A is no longer ap- plicable. ∼ 1, the pseudo-gaps overlap and B. Density of states Let us now extend the solution of the Dirac-Kronig- Penney model to the density of states. For purely ballis- tic system with metal leads, both the local and the inte- grated density has been found in Ref. 56 using a Green's function approach. Below we take advantage of an alter- native route and relate the partial density of states in the channel q to the corresponding transmission amplitude, tq, by the well-known formula57 νq = − 1 π ∂ε (cid:21) . Im(cid:20) ∂ ln tq (36) 7 Then, the integrated density of states per unit volume is given by ρ(ε) = 4 LW Xq νq(ε), (37) where the factor 4 takes into account the spin and val- ley degeneracy. The transmission amplitude is readily obtained from Eqs. (14,25) as tq = 1 cos k0L + iζq sin k0L , ζq = η sin 2k0ℓ , (38) where the wave-number k0 and the real quantity η are determined from the expressions (23) and (26), respec- tively. Therefore, the partial density of states can be calculated exactly as Lζq(dk0/dε) + sin k0L cos k0L (dζq/dε) . (39) νq = π(cid:0)cos2 k0L + ζ2 q sin2 k0L(cid:1) Unlike the conductance or the shot noise (28), the den- sity of states depends on the value of Ulead, because the spectrum in a vicinity of the band-edges is dominated by the metal-induced states (evanescent modes). The metal proximity effect56 can be seen already for purely ballistic system near the Dirac point. Indeed, for u0 = V0 = 0, one finds νball q (ε) = Lε2 − (q2/k) cos kL sin kL π(ε2 − q2 cos2 kL) , (40) q where k = pε2 − q2. At zero energy the ballistic result (40) is reduced to νball (0) = tanh (qL)/πq. Therefore, the density of states at ε = 0 in Eq. (37) acquires a loga- rithmic divergency. This divergency is regularized by the largest available transversal momentum qmax, which is simply equal to the Fermi-momentum in the metal lead, qmax = Ulead/¯hv. Thus, the density of states in a close vicinity of the Dirac point, ε ≪ ¯hv/L, is given by ρball(0) = 4 π2¯hvL ln UleadL ¯hv , (41) in agreement with Ref. 56. Similar logarithmic depen- dence of the density of states on Ulead takes place in the Dirac-Kronig-Penney model for energies inside the pseudo-gaps. In full analogy with the averaged generating function (31) we can introduce the averaged density of states, ρ, which corresponds to the limit L → ∞. In this limit we disregard the contribution of the metal-induced states by projecting on the real values of the momentum k0 and average the result of Eq. (40) over the rapid phase k0L. This procedure leads to the simple result νq = dRe k0 dε L π (cid:12)(cid:12)(cid:12)(cid:12) , (cid:12)(cid:12)(cid:12)(cid:12) (42) q ℓ 2 q ℓ 2 4 2 0 -2 -4 4 2 0 -2 -4 q ℓ 2 4 2 0 -2 -4 -4 -2 0 2 4 -4 -2 0 2 4 2ℓk0 2ℓk0 q ℓ 2 4 2 0 -2 -4 -4 -2 0 2 4 -4 -2 0 2 4 2ℓk0 2ℓk0 FIG. 7: (Color online) The contour plot of the dispersion re- lation for the Dirac-Kronig-Penney model (19,20) with strain modulation rθx = 0.3 (top left), rθx = 2.0 (bottom left) and electrostatic field modulation r0 = 0.3 (top right), r0 = 2.0 (bottom right) at energy ε = 3.3¯hv/2ℓ. hence, in the limit W ≫ L, the mean density is given by 2 ρ = π2 Z ∞ dq (cid:12)(cid:12)(cid:12)(cid:12) −∞ . (43) dRe k0 dε (cid:12)(cid:12)(cid:12)(cid:12) In the ballistic limit, u0 = V0 = 0, Eq. (43) yields the den- sity of states of the clean graphene ρball = 2ε/π¯h2v2. In Fig. 6 we plot the averaged density of states calculated from Eq. (43) for different strengths of the periodic po- tentials. C. Beyond the Dirac-Kronig-Penney model The results of the previous subsections are readily gen- eralized to a model with arbitrary periodic variation of strain and electrostatic potential in x direction. The analysis of the model is reduced to the calculation of the transfer matrix, M0, which corresponds to the wave propagation over the distance 2ℓ, the period of the po- tential. In this generalized model, both the exact and the averaged full counting statistics as well as the density of states are still given by the expressions (29,31,39,42) with λ = 1 2 TrM0, η = i 2 Tr σzM0, (44) and the x-component, k0, of the quasi-momentum is related to λ by Eq. (23). Thus, for a periodic potential of a general type, the full solution of the problem is re- duced to the straightforward numerical evaluation of the 8 Aθyn = 1 Aθxn = 1 ε 12 10 8 6 4 2 -10 -5 0 kx Vn = 1 5 10 0 ky Vzn = 1 5 10 -10 -5 ε 12 10 8 6 4 2 -10 -5 0 ky 5 10 -10 -5 5 10 0 ky 12 10 8 6 4 2 12 10 8 6 4 2 ε ε FIG. 8: (Color online) Typical band gap boundaries. In all plots n, b was set to 1. functions λ and η. Note, that the exact analytical expres- sions (24,26) are restricted to the Dirac-Kronig-Penney model and do not apply generally. We have used Eqs. (44) to calculate the energy- dependent conductance for different amplitudes of the periodic strain in the harmonic potential, uxx = u0 sin(πx/ℓ). In this case, the higher pseudo-gaps are found to be suppressed for weak potentials rθx, r0 ≪ 1 as compared to the Dirac-Kronig-Penney model. The low- est conductance minima around ε = ε0 is essentially the same in both models. The models become even more sim- ilar with increasing potential strength. For rθx, r0 > ∼ 1 the transmission spectra of the single-harmonic model, becomes almost equivalent to those of the Dirac-Kronig- Penney model for all energies. D. Transport in y direction We have demonstrated that periodic and x-dependent scalar and vector potentials modify transport in x- direction due to the appearance of pseudo-gaps. Let us now argue that such potentials hardly affect transport in y direction, i.e. in the direction parallel to the equi- potential lines. In Fig. 7 we plot the Fermi surface slightly above the energy value ε = ε0 using the exact dispersion relation, ε = ε(k0, q), obtained from Eqs. (23,24) for the Dirac- Kronig-Penney model. We have seen that the transport in x direction is dominated by the modes with the mo- menta q ∈ (−ε, ε). For some values of q in this interval one cannot find any propagating states, which correspond to the real values of k0. This can be seen as the forma- tion of pseudo-gaps in the transmission spectra shown in Fig. 3, which is especially strong for rθx, r0 > ∼ 1. In con- trast, for each real value of k0 one always finds real values of q. Therefore, the transport in y-direction is not af- fected by the formation of the pseudo-gaps, and depends only on the details of scattering at the metal-graphene interfaces. The latter effect is, however, weak and falls beyond the scope of our consideration. We therefore con- clude that one-dimensional superlattices have a major impact only on the transport properties along the direc- tion of their periodicity. V. CONCLUSION The Dirac-Kronig-Penney model has been applied in order to study phase-coherent charge transport in peri- odically strained graphene samples. Using the exact re- lations for the band-structure in the considered superlat- tice, we calculate the density of states and the full count- ing statistics for charge transport. The exact quantities are simplified in the thermodynamic limit by neglecting the contribution of evanescent modes and averaging over the Fabry-P´erot oscillations. The conductance is found to be suppressed at energies corresponding to the posi- tions of the pseudo-gaps in the modified spectra of the superlattice. The periodic strain and electrostatic po- tentials are argued to have the largest effect on transport in the direction perpendicular to the equipotential lines. The vector potentials are shown to play a greater role in confining the Dirac quasi-particles due to the suppression of Klein tunnelling. Acknowledgments We thank W.-R. Hannes for discussions. S.G. acknowl- edges support by the EPSRC Scottish Doctoral Training Centre in Condensed Matter Physics. W. B. acknowl- edges support by the DFG through SFB 767 Controlled Nanosystems , by the Research Initiative UltraQuantum, and in part by the Project of Knowledge Innovation Pro- gram (PKIP) of Chinese Academy of Sciences, Grant No. KJCX2.YW.W10. Appendix A: Band structure for Dirac particles in a superlattice In this appendix we perturbatively calculate the band structure of a graphene sample described by the single- valley Dirac equation [−iσ · ∇ + V(r)] Ψ(r) = εΨ(r), (A1) where V(r) = (cid:18) V (r) + Vz(r) Aθx(r) − iAθy(r) V (r) − Vz(r) (cid:19) (A2) Aθx(r) + iAθy(r) is a general 1d-periodic potential that is smooth on the scale of the graphene lattice constant a. To do so, we gen- eralize the standard methods of elementary band struc- 9 ture theory to the Dirac case with its non-trivial ma- trix structure. We find that for purely scalar potentials, this additional structure prevents the creation of a gap at ky = 0 thus making Aθ and Vz more relevant for the transport properties; this fact can be related to the chirality of the quasi-particles in graphene and the phe- nomenon of Klein tunnelling. Let b = (bx, by)T span the 1d-super-lattice (so that V(r + nb) = V(r)) and g be the corresponding reciprocal vector fixed by the conditions g · b = 2π and g k b. Making the Bloch ansatz Ψ(r) = Xn ψ(k − ng) ei(k−ng)r (A3) for the wave function and writing the periodic potential down the central equation k∗ − ng∗ −ε (cid:19) ψ(k−ng)+Xm as a Fourier series, V(r) = Pn Vn eingr, we can write (cid:18) −ε k − ng Vm−nψ(k−mg) = 0, (A4) where k = kx + iky and g = gx + igy. The central equa- tion (A4) represents an infinite set of linear equations for the coefficients ψ, which can in general only be solved numerically. The solution, however, simplifies for suffi- ciently weak periodic potentials. For k close to a degeneracy, k = k− ng ≈ ε, we can neglect all but the two coefficients ψ(k) and ψ(k − ng) in Eq. (A4). The resulting linear system has the simple form −ε k k∗ −ε z A∗ V ∗ + V ∗ A∗ θx + iA∗ θy θx − iA∗ V ∗ − V ∗ z θy (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) V + Vz Aθx + iAθy −ε kn Aθx − iAθy V − Vz k∗ n −ε = 0. (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) (A5) Here, we introduced the abbreviation kn = k − ng and omitted the Fourier index n of the potentials for brevity. The solution of Eq. (A5) is particularly compact when only one of the coefficients Aθxn, Aθyn, Vn or Vzn is dif- ferent from zero. In this case we obtain ε2 = k2 + kn2 2 + Un2 ± 1 2p(k2 − kn2)2 + 4Un22p − ng2, (A6) where the pair of quantities (Un, p) take on one of the fol- lowing values (Axn, kx), (Ayn,−iky), (Vn, k), or (Vzn, 0). We compare the energy values of the forbidden zone's upper/lower boundary, εg±, as well as the width of the gap in forward direction for the different types of periodic potentials in Tab. I. To calculate εg± we have chosen k in Eq. (A6) such, that k· b/b = ng/2. Typical band gap boundaries are shown in Fig. 8. εg± ∆(ky = 0) 10 Aθxn 6= 0 Aθyn 6= 0 Vn 6= 0 Vzn 6= 0 ng ng 2 ± Aθxn 2 (cid:12)(cid:12) 2 ± Aθyn 2 (cid:12)(cid:12) 2 + Vn2 ± kyVn qk2 x + (cid:12)(cid:12) qk2 y + (cid:12)(cid:12) qk2 2 (cid:12)(cid:12) y + (cid:12)(cid:12) qk2 2 + Vzn2 ± 2 ng 2 (cid:12)(cid:12) y + (cid:12)(cid:12) ng ng 0 2Aθyn 0 2Vzn 2 Vzn TABLE I: Energy value of the forbidden zone's upper/lower boundary, as well as the width of the gap in forward direction for the different types of periodic potentials. In the first row, b was chosen parallel to ey, for the remaining rows b k ex. 1 A. Geim, Science 324, 1530 (2009). 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 A. K. Geim, K. S. Novoselov, Nature Materials 6, 183 (2007). 4 J. C. Meyer, A. K. Geim, M. I. Katsnelson, K. S. Novoselov, T. J. Booth, and S. Roth, Nature 446, 60 (2007). 5 K. I. Bolotin, K. J. Sikes, J. Hone, H. L. Stormer, and P. Kim, Phys. Rev. Lett. 101, 096802 (2008). 6 L. A. Ponomarenko, F. Schedin, M. I. Katsnelson, R. Yang, E. W. Hill, K. S. Novoselov, and A. K. Geim, Science 320, 356 (2008). 7 Y.-M. Lin, C. Dimitrakopoulos, K. A. Jenkins, D. B. Farmer, H.-Y. Chiu, A. Grill, and Ph. Avouris, Science 327, 662 (2010). 8 K. S. Kim, Y. Z. H. Jang, S. Y. Lee, J. M. Kim, Kw. S. Kim, J.-H. Ahn, P. Kim, J.-Y. Choi, and B. H. Hong, Nature 457, 706 (2009). 9 X. Wang, L. Zhi, and K. Mullen, Nano Lett. 8, 323 (2008). 10 M. I. Katsnelson and K. S. Novoselov, Solid State Com- mun. 143, 3 (2007). 11 K. S. Novoselov, A. K. Geim, S. V. Morozov, D. Jiang, and I. V. Grigorieva, Y. Zhang, S. V. Dubonos, A. A. Firsov, Science 306, 666 (2004). 12 K. S. Novoselov, D. Jiang, F. Schedin, T. Booth, V. V. Khotkevich, S. V. Morozov, and A. K. Geim, PNAS 102, 10451 (2005). 13 V. M. Pereira and A. H. Castro Neto, Phys. Rev. Lett. 103, 046801 (2009). 14 F. Guinea, A. K. Geim, M. I. Katsnelson, and K. S. Novoselov, Phys. Rev. B 81, 035408 (2010). 15 F. Guinea, M. I. Katsnelson, and A. K. Geim, Na- ture Physics 6, 30 (2010). 16 F. von Oppen, F. Guinea, and E. Mariani, Phys. Rev. B 80, 075420 (2009). 17 J. Bunch, S. S. Verbridge, J. S. Alden, A. M. van der Zande, J. M. Parpia, H. G. Craighead, and P. L. McEuen, Nano Lett. 8, 2458 (2008). 18 W. Bao, F. Miao, Z. Chen, H. Zhang, W. Jang, C. Dames, and C. Lau, Nature Nanotechnology 4, 562 (2009). 19 I. Pletikosic, M. Kralj, P. Pervan, R. Brako, J. Coraux, A. T. N'Diaye, C. Busse, and T. Michely, Phys. Rev. Lett. 102, 056808 (2009). 20 A. L. Vazquez de Parga, F. Calleja, B. Borca, M. C. G. Passeggi, J. J. Hinarejos, F. Guinea, and R. Mi- randa Phys. Rev. Lett. 100, 056807 (2008) . 21 M. L. Teague, A. P. Lai, J. Velasco, C. R. Hughes, A. D. Beyer, M. W. Bockrath, C. N. Lau, and N.-C. Yeh, Nano Letters 9, 2542 (2009). 22 S. V. Morozov, K. S. Novoselov, M. I. Katsnelson, F. Schedin, L. A. Ponomarenko, D. Jiang and A. K. Geim, Phys. Rev. Lett.97, 016801 (2006). 23 F. Guinea, B. Horovitz and P. Le Doussal, Phys. Rev. B 77, 205421 (2008). 24 E. Kim, A. H. Castro Neto, Europhys. Lett. 84, 57007 (2008). 25 A. Isacsson, L. M Jonsson, J. M. Kinaret and M. Jonson, Phys. Rev. B 77, 035423 (2008). 26 C. L. Kane and E. J. Mele, Phys. Rev. Lett. 78, 1932 (1997). 27 A. Kleiner and S. Eggert, Phys. Rev. B 64, 113402 (2001). 28 K. Sasaki, Y. Kawazoe and R. Saito, Prog. Theor. Phys. 113, 463 (2005). 29 D. Huertas-Hernando, F. Guinea and A. Brataas, Phys. Rev. B 74, 155426 (2006). 30 C. Lee, X. Wei, J. W. Kysar, J. Hone, Science 321, 385 (2008). 31 F. de Juan, A. Cortijo, M. A. H. Vozmediano Phys. Rev. B 76, 165409 (2007). 32 V. M. Pereira, A. H. C. Neto, and N. M. R. Peres, Phys. Rev. B 80, 045401 (2009). 33 M. M. Fogler, F. Guinea, and M. I. Katsnelson, Phys. Rev. Lett. 101, 226804 (2008). 34 T. M. G. Mohiuddin, A. Lombardo, R. R. Nair, A. Bonetti, G. Savini, R. Jalil, N. Bonini, D. M. Basko, C. Galiotis, N. Marzari, K. S. Novoselov, A. K. Geim, and A. C. Fer- rari, Phys. Rev. B, 79, 205433 (2009). 35 J. Tworzydlo, B. Trauzettel, M. Titov, A. Rycerz, and C. W. J. Beenakker, Phys. Rev. Lett. 96, 246802 (2006). 36 M. Titov, Europhys. Lett. 79, 17004 (2007). 37 Y. Aharonov and A. Casher, Phys. Rev. A 19, 2461 (1979). 38 A. Schuessler, P. M. Ostrovsky, I. V. Gornyi, and A. D. Mirlin, Phys. Rev.B 79, 075405 (2009). 39 W.-R. Hannes and M. Titov, Europhys. Lett. 89, 47007 (2010). 40 M. R. Masir, P. Vasilopoulos, F. M. Peeters, New J. Phys. 11, 095009 (2009). 41 M. Barbier, P. Vasilopoulos, and F. M. Peeters, 11 Phys. Rev. B 81, 075438 (2010); Phys. Rev. B 80, 205415 (2009). 50 L. Z. Tan, C.-H. Park, S. G. Louie, Phys. Rev. B 81, 195426 (2010). 42 D. P. Arovas, L. Brey, H. A. Fertig, Eun-Ah Kim, 51 C.-H. Park, L. Z. Tan, S. G. Louie, Physica E (2010), K. Ziegler, arXiv:1002.3655 (2010). doi:10.1016/j.jphyse.2010.07.022 43 C.-H. Park et al., Phys. Rev. Lett. 103, 046808 (2009); 52 D. S. Novikov, Phys. Rev. B 72, 235428 (2005); Nano Lett. 8, 2920 (2008). 44 C.-H. Park et al., Nature Physics 4, 213 (2008). 45 L. Dell'Anna and A. De Martino, Phys. Rev. B 79, 045420 (2009). 46 I. Snyman, Phys. Rev. B 80, 054303 (2009). 47 L. Brey and H. A. Fertig, Phys. Rev. Lett. 103, 046809 (2009). 48 M. Esmailpour, A. Esmailpour, R. Asgari, E. Elahi, and M. R. Rahimi Tabar, Solid State Commun. 150, 655 (2010). 49 L.-G. Wang, and S.-Y. Zhu, Phys. Rev. B 81, 205444 (2010). Phys. Rev. Lett. 95, 066401 (2005). 53 M. Born and E. Wolf, Principles of Optics, Cambridge Uni- versity Press, 1999. 54 M. I. Katsnelson, K. S. Novoselov, and A. K. Geim, Na- ture Phys. 2, 620 (2006). 55 V. Arnold, Mathematical methods of classical mechanics, Springer Verlag, 1989. 56 M. Titov, P. M. Ostrovsky, I. V. Gornyi Semi- cond. Sci. Technol. 25, 034007 (2010). 57 M. Buttiker, J. Phys.: Condensed Matter 5, 9361 (1993).
1204.5210	1	1204	2012-04-23T21:14:16	Transport Properties of Graphene Nanoroads in Boron-Nitride Sheets	[ "cond-mat.mes-hall" ]	We demonstrate that the one-dimensional (1D) transport channels that appear in the gap when graphene nanoroads are embedded in boron-nitride (BN) sheets are more robust when they are inserted at AB/BA grain boundaries. Our conclusions are based on ab-initio electronic structure calculations for a variety of different crystal orientations and bonding arrangements at the BN/C interfaces. This property is related to the valley-Hall conductivity present in the BN band structure and to the topologically protected kink states that appear in continuum Dirac models with position dependent masses.	cond-mat.mes-hall	cond-mat	Transport Properties of Graphene Nanoroads in Boron-Nitride Sheets Jeil Jung,† Zhenhua Qiao,∗,† Qian Niu,†,‡ and Allan H. MacDonald† Department of Physics, The University of Texas at Austin, Austin, Texas 78712, USA, and International Center for Quantum Materials, Peking University, Beijing 100871, China E-mail: [email protected] KEYWORDS: h-Boron Nitride, Graphene ribbon, 1D conducting state, Kink State, Ballistic Transport Abstract We demonstrate that the one-dimensional (1D) transport channels that appear in the gap when graphene nanoroads are embedded in boron-nitride (BN) sheets are more robust when they are inserted at AB/BA grain boundaries. Our conclusions are based on ab-initio electronic structure calculations for a variety of different crystal orientations and bonding arrangements at the BN/C interfaces. This property is related to the valley-Hall conductivity present in the BN band structure and to the topologically protected kink states that appear in continuum Dirac models with position dependent masses. Metallic transport channels appear at the edges, surfaces, and interfaces of two and three di- mensional bulk insulators when a bulk topological index changes value as the interface region is crossed.1 -- 10 This property can provide transport channels in otherwise insulating materials. The ∗To whom correspondence should be addressed †Department of Physics, The University of Texas at Austin, Austin, Texas 78712, USA ‡International Center for Quantum Materials, Peking University, Beijing 100871, China 1 metallic states possess an internal structure related to their sense of propagation which leads to special transport properties including zero bend resistance at sharp turns in the current propagation trajectory, pseudospin memory, and suppressed backscattering.7 Same topology Zigzag B N Opposite topology Zigzag Armchair Armchair H - (BN)n -- Cm -- (BN)n - H H - (BN)n -- Cm -- (NB)n - H Figure 1: (Color online) Schematic plot of two types of zigzag and armchair BNC nanoroads. The integer n is the number of atoms in each BN partition whereas m is the number of C atoms across the nanoroad. The outermost BN atoms are passivated with hydrogen. Left Panel: The same topology arrangement in which the sign of the valley Hall effect is the same in both BN sheets. In the absence of C atoms the BN fragments can be joined seamlessly. Right Panel: The opposite topology arrangement in which the sign of the valley Hall effect is opposite in the two BN sheets. In the absence of C atoms it is necessary to form a nearest-neighbor bond between atoms of the same species- either N−N and/or B−B. In this Letter we show that two-dimensional hybrid structures consisting of graphene nanoroads13 (See 1) embedded in hexagonal boron nitride (BN) sheets can be an attractive host for topologi- cally assisted one-dimensional (1D) transport channels. Our study is motivated by the observa- tion that the p -bands of BN are similar to those of a graphene sheet, except that the honeycomb sublattices have different p -electron site energies. (The electrostatic potential is more attractive on the higher-Z N atom sites). Because valence band states near the K and K′ Brillouin-zone corners are strongly localized on N sites whereas those elsewhere in the Brillouin-zone are di- vided more evenly, the bands of BN contribute with valley Hall conductivities and associated Berry curvatures of opposite signs near K and K′ whose continuum model expression is given by W (q) = t z(cid:0)3a2D t2(cid:1) /(cid:16)2(cid:0)D 2 + 3q2a2t2(cid:1)3/2(cid:17) where q is the momentum measured from a Dirac point, t z = ±1 is the valley, a is the lattice constant of the honeycomb lattice and D gap.14 If the B-site vs. N-site potential difference was small, these Berry curvatures would be is the band strongly concentrated near the BZ corners [see 2]. Indeed, in the graphene case it is known that 2 2 U = 0.1 t 2 U = 2 t K K ′ K K ′ 0.5 0.4 0.3 0.2 0.1 ) K y e l l a V ( r e b m u N n r e h C 0.0 0.0 0.5 1.0 1.5 2.0 2U (Band Gap) 2U / t (Band gap) Figure 2: Berry curvatures obtained from a tight-binding model of the honeycomb lattice with different band gap sizes D = 2U where U is the staggering site potential. The Berry curvatures are more sharply peaked near the Brillouin zone corners K and K′ when the band gap is small and they spread out more when the gap is larger. The rightmost panel shows the deviation of the valley resolved Chern number from the ideal 1/2 value of the continuum model obtained integrating the Berry curvature in one equilateral triangle corresponding to one half of the primitive cell associated to valley K given by CK = (2p )−1RK d2kW (k). The dotted vertical line at 2U /t = 2 indicates the valley resolved Chern number obtained for the tight-binding hamiltonian we used to approximate the BN bands. electronic properties in systems with sublattice-staggered potentials that are weak compared to the p -band width can be described using a two-dimensional massive Dirac equation. Under these circumstances Dirac-equation continuum models are valid and predict one-dimensional (1D) local- ized states along lines where the sublattice-staggered potential (the Dirac-equation mass) changes sign.3,4 In the closely related bilayer5 -- 7 and multilayer6 graphene cases, the sublattice staggered potentials are readily generated experimentally by applying an electric field across the layers and varied spatially by appropriately arranging external gate voltages. In the single-layer case, how- ever, it has not been obvious how the sublattice-staggered potentials could be realized, although gaps may be present with a lattice matched BN substrate11 or when gate potential profiles are carefully correlated with strains.12 This work is motivated by the idea that the difference between B-atom and N-atom potentials in BN provides the desired staggered potential. Of course the stag- gered potential is not weak in the case of a BN crystal, so that expectations based on massive Dirac continuum models must be checked by ab initio electronic structure calculations. The topological index that is inherent in BN bands can be viewed as a valley Hall effect since each valley separately supports half integer quantum Hall effects of opposite sign, or at least does in 3 the limit that the potential difference is small compared to the p -band width. The properties that we discuss below are therefore closely related to the simple momentum space Berry curvature pattern in the bands of BN sheets. Hybrid BN/graphene systems of the type we study occur naturally in patched sheets containing a mixture of atomically thin graphene and BN,15 and have recently become a subject of great interest,13,16 -- 29,29 -- 33 as the basis of a possible strategy for controlling graphene system band gaps. In this work we show that the properties of the transport channels formed by graphene nanoroads in BN strongly depend on the valley Hall effect of the surrounding material. The simplest graphene nanoroad geometries have C/BN interfaces with crystallographic ori- entations along the zigzag and armchair directions. The unit cells considered in our ab initio calculation are illustrated in 1. In each case a graphene nanoribbon is flanked on the left and right by BN sheets with B and N atoms either on the same or opposite sublattices. When the two dif- ferent BN fragments can be joined seamlessly preserving the crystalline B and N atom sequences upon carbon atom removal, as in the cases illustrated in the left panels of 1, the BN sheets on left and right have valley Hall conductivities of the same sign and we will say that they have the same topology. The right and center panels show junctions in which the BN sheets on opposite sides of the graphene nanoroad have opposite valley Hall conductivities. We will refer to these configurations as having opposite topologies. When the carbon atoms are removed, joining the BN fragments would in this case require B−B or N−N bond. As shown in 3, a difference in the topology of the BN sheet arrangement on the nanoroad shoulders invariably leads to a qualitative change in nanoroad quasi-1D band structure. The indices n and m in 3 represent respectively the number of BN and carbon atoms in the 1D unit cell. The width of the BN fragment, specified by the index n, plays a minor role in mid-gap state properties. We begin by discussing results for zero, one, and three hexagon width zigzag nanoroads presented in the upper panels of 3. The zero width cases illustrates the electronic structure of seamless BN nanoribbons with 2n = 80 atoms of BN, and therefore a width of 86 nm. The large band gap of 4 eV is comparable to that of bulk BN. There is a notable lack of 4 particle-hole symmetry in the band structure, which is partly due to the difference between B and N site potentials mentioned previously but cannot be captured by a nearest neighbor tight- binding model. We are mainly interested in the mid-gap states which emerge when a graphene nanoroad is embedded in the BN sheet. As expected on the basis of the properties of pure graphene nanoribbons, we find that zigzag graphene nanoroads do have strongly reduced gaps because of the presence of zigzag edge states.36 -- 38 There is still however a topology sensitive feature; in the opposite topology case a mid-gap state appears that is peaked near the center of the nanoroad, rather than at its edges, with some penetration into the BN region. This state is indicated by blue shading in Figs. 3(d) to 3(i). We view this state as the lattice remnant of the mass-reversal (kink) state which would appear in the opposite topology case if the staggered potential was weak and the continuum model applied. Even in the absence of carbon atoms its appearance reduces the energy gap by approximately 2 eV. If the continuum model was accurate there would be opposite velocity kink states associated with the K and K′ valleys and the gap would disappear entirely. Although gaps are absent for wide nanoroads in both like cases, the additional kink state present for opposite topologies closes the gap already at one hexagon width in the opposite topology case. For same topology configurations, the gaps are closed because the kx ∼ 0 states that are most strongly localized at the edge are shifted in opposite directions by coupling to the BN shoulders.38 For inverted topology, since the same type of atoms N or B are attached at both edges of the graphene nanoroad, we see a shift of onsite energies at the edge sites in the same direction, upwards for N atom bonds and downwards for B atom bonds. In this case the states closing the band gap have kink-state character and are spread widely over the C atoms of the nanoroad, and with some penetration into the BN region.7 We now turn to armchair nanoroads. In ribbons this orientation does not support metallic edge states,40 and instead yields gaps which scale as the inverse ribbon width. Continuum model considerations suggest that reversed topology would lead to vanishing gaps for all widths.6,7 The band structures in the lower panel of 3 demonstrate that there are gaps in the reversed topology case, but that the gaps are much smaller than in the same topology case. The magnitude of the small 5 avoided crossing gap between opposite velocity kink states at kx ∼ 0 shows the same oscillatory decline as a function of ribbon width that has been heavily studied in ribbons.39,40 The gap reaches a value as large as 0.75 eV for armchair single-hexagon nanoroads. It is noteworthy that this avoided crossing is always smaller than the finite-size gap in either nanoribbons of the same width or same topology nanoroads, a fact that allows the kink channels to become the main current conducting path when the Femi level of the system is adequately shifted. The same topology and and d , are plotted as a function of nanoroad width in the bottom right opposite topology gaps, D panel of 3. Zigzag graphene nanoroads have bearded interfaces terminations when they have an odd num- ber of carbon atoms or non modulo 4 even integer number of C atoms in the unit cell.36,37 These interfaces are not expected to be common in experimental systems because of stability consid- erations. A combination of zigzag and armchair fragments35 are more abundant than bearded configurations with repeated dangling atoms. Bearded edges lead to more complicated band struc- tures that we discuss further in the supplementary information, but generally preserve the tendency toward greater conduction in the opposite topology case. To address the influence of topology on nanoroad transport, we consider a two-terminal geom- etry with a central scattering region containing nanoroads that are either straight, or curved with both armchair and zigzag segments [See 4]. In our model, the source and drain electrodes are semi- infinite zigzag graphene nanoroads. The central scattering region contains 96×60 lattice sites and the nanoroad width is fixed at 32 carbon atoms. The BNC nanoroad system is modeled using a nearest neighbor tight-binding Hamiltonian of the form H = −t (cid:229) hi ji i c j + (cid:229) c† i∈A UAc† i ci − (cid:229) i∈B UBc† i ci, (1) i (ci) is a p -orbital creation (annihilation) operator for an electron at the site i, UA and UB where c† are the BN p -orbital site energies on the A and B subllices, and t = 2.6 eV is the nearest neighbor hopping energy for both graphene and h-BN. We introduce a staggered AB sublattice potential 6 to describe the site energy differences between B and N sublattices of h-BN, i.e. UB = −UN = l U0. Here, l ='±' defines the crystal topology of the h-BN regions in the graphene nanoroad. U0 measures the strength of the staggered potential, and for the purposes of this illustrative calculation we have assumed it to be U0 = t. The temperature T = 0 conductance from the left lead to the right lead can be calculated using the multi-probe Landauer-Büttiker formula:41 GRL = e2 h Tr[G RGrG LGa], (2) where Gr,a are the retarded and advanced Green functions of the central scattering region. The quantity G R/L is a linewidth function describing the coupling between left/right-lead and the central region and is obtained by calculating the self-energy S r R/L of left/right-lead using a variant transfer matrix method.42 -- 44 5 plots conductance GRL as a function of Fermi energy e for the straight (zigzag-edged) and curved BNC nanoroads. In panel (a), the crystal topologies are the same in upper and lower h- BN regions. We observe that the conductance of the straight BNC nanoroad is quantized to be G = 2n e2/h (n = 0, 1, 2...). The conductance plotted here is per spin, so that the even conductance values can be attributed to the presence of ballistic states that are localized at both edges. The nanoribbon finite size effect for this nanoroad width produces a small band gap with vanishing conductance near the charge neutrality point. When the graphene nanoroad becomes curved in the central regime, disorder produces backscattering which supresses the conductance below the quantized value. The case of opposite BN shoulder topologies is illustrated in 5(b). In this case the conductance rule for straight BNC nanoroads is changed to G = (2n − 1) e2/h (n = 1, 2...) as shown in 5(b). This change reflects the addition of a topologically supported kink state. For the curved reversed topology nanoribbon, the conductance does not decrease substantially in the region near the charge neutrality point where the only remaining channel is the kink state. Near the neutrality point the 7 conductances show a surprising plateau feature with minor fluctuations that may be suppressed further by increasing the width of the graphene nanoroad. This finding strongly reflects the robust- ness of the crystal topology induced 1D topological conducting state against BNC nanoroad bends and is consistent with the finding of the zero bend resistance for 1D topological confinement states in gated bilayer graphene.7 To visualize the 1D topological conducting channel, we can evaluate the position dependence of its local density of states (LDOS) of the current originating from Pth lead using r p(r,e ) = 1 2p [GrG pGa]rr. (3) Here e is the Fermi energy. In 6, we plot the LDOS distribution of the topologically assisted conducting channel from the left lead into a snake-shaped BNC nanoroad. Our calculation shows that it is localized across the graphene atomic sites with only a small portion on the interface between h-BN and graphene. The transport calculations for the straight and curved nanoroad geometries confirms that the kink states picture remains essentially valid even when both zigzag and armchair edges are simultaneously present. The almost ballistic transport is preserved in spite of irregular bonds at the graphene/BN interfaces. In summary, we have shown that graphene nanoroads embedded in BN sheets host kink-like transport channels when surrounded by opposite topology BN fragments. Transmission by the kink-states is topologically assisted and nearly perfect even when the graphene nanoroad is curved and has irregular edges consisting of combinations of both for zigzag and armchair edges. Calculation Details The self-consistent ab-initio calculations were carried out using the PAW-LDA parametrization of the exchange correlation functional in a plane wave basis code, as implemented in Quantum Espresso.45 We used the 1.443 Angstrom for the separation between neighboring B, N and C atoms and 1.09 for B−H and N−H separations. These values are consistent with optimized lattice 8 constants obtained from total energy minimization. Relaxation to obtain a final crystal geometry with atomic forces below 0.01 eV /Angstrom had a negligible effect in modifying the interatomic distances and band structures. For our nanoroad calculations in ribbon geometries we used 30, 1 and 1 k-points in the kx, ky and kz directions, and a plane wave energy cutoff of 60 Ry. The vacuum separation distances between supercell repeated nano ribbons was set to 20 in both y and z directions. Acknowledgement This work was supported by NSF (DMR 0906025), NRI-SWAN, Welch Foundation (TBF1474, F- 1255), DOE (DE-FG03-02ER45958, Division of Materials Science and Engineering), and Texas Advanced Research Program. We gratefully acknowledge the computation resources and assis- tance provided by the Texas Advanced Computing Center. Supporting Information Available We have also attached a supporting information material for a more detailed understanding of our work. This material is available free of charge via the Internet at http://pubs.acs.org/. References (1) Heeger A. J.; Kivelson S.; Schrieffer J. R.; Su W.-P. Rev. Mod. Phys. 1988, 60, 781. (2) Volovik G. E., 'The Universe in a Helium Droplet', Oxford University Press (2003). (3) Semenoff G. W.; Semenoff V.; Zhou F. Phys. Rev. Lett. 2008, 101, 087204. (4) Yao W.; Yang S. A.; Niu Q. Phys. Rev. Lett. 2009, 102, 096801. (5) Martin I.; Blanter Ya. M.; Morpurgo A. F. Phys. Rev. Lett. 2008, 100, 036804. (6) Jung J.; Zhang F.; Qiao Z. H.; MacDonald A. H. Phys. Rev. B 2011, 84, 075418. 9 (7) Qiao Z. H.; Jung J.; Niu Q.; MacDonald A. H. Nano Lett. 2011, 11 (8), 3453. (8) Killi M.; Wei T.-C.; Affleck I.; Paramekanti A. Phys. Rev. Lett. 2010, 104, 216406. (9) Killi M.; Wu S.; Paramekanti A. Phys. Rev. Lett. 2011, 107, 086801. (10) Wu S.; Killi M.; Paramekanti A. E-print arXiv:1202.1441 (2012). (11) Giovannetti G.; Khomyakov P. A.; Brocks G.; Kelly P. J.; van den Brink J.; Physl Rev. B 2007, 76, 073103. (12) Low T., Guinea F., and Katsnelson M. I., Phys. Rev. B 2011, 83, 195436. (13) Bhowmick S.; Singh A. K.; Yakobson B. I. J. Phys. Chem. C 2011, 115, 9889. (14) Xiao D.; Yao W.; and Niu Q., Phys. Rev. Lett. 2007, 99, 236809. (15) Ci L.; Song L.; Jin C.; Jariwala D.; Wu D.; Li Y.; Srivastava A.; Wang Z. F.; Storr K.; Balicas L.; Liu F.; Ajayan P. M. Nat. Mat. 2010, 9, 430. (16) Rubio A. Nat. Mat. 2010, 9, 379. (17) Seol G.; Guo J. App. Phys. Lett. 2011, 98, 143107. (18) Shinde P. P.; Kumar V. Phys. Rev. B 2011, 84, 125401. (19) Pruneda J. M. Phys. Rev. B 2010, 81, 161409(R). (20) Li J.; Shenoy V. B. Appl. Phys. Lett. 2011, 98, 013105. (21) Liu Y.; Bhowmick S.; Yakobson B. I. Nano Lett. 2011, 11, 3113. (22) Ding Y.; Wang Y.; Ni J. Appl. Phys. Lett. 2009, 95, 123105. (23) Liu Y.; Wu X.; Zhao Y.; Zeng X. C.; Yang J. J. Phys. Chem. C 2011, 115, 9442. (24) Fan Y.; Zhao M.; Zhang X.; Wang Z.; He T.; Xia H.; Liu X. J. Appl. Phys. 2011, 110, 034314. 10 (25) Lam K.-T.; Lu Y.; Feng Y. P.; Liang G. Appl. Phys. Lett. 2011, 98, 022101. (26) Berseneva N.; Krasheninnikov A. V.; Nieminen R. M. Phys. Rev. Lett. 2011, 107, 035501. (27) Ramasubramaniam A.; Naveh D. Phys. Rev. B 2011, 84, 075405. (28) Obodo K. O.; Andrew R. C.; Chetty N. Phys. Rev. B 2011, 84, 155308. (29) Manna A. K.; Pati S. K. J. Phys. Chem. C 2011, 115, 10842. (30) Song L. et al., E-print arXiv:1105.1876 (2011). (31) Modarresi M.; Roknabadi M. R.; Shahtahmasbi N. Physica E 2011, 43 (9), 1751. (32) Qiu M.; Liew K. M. J. Appl. Phys. 2011, 110, 064319. (33) Cao T.; Feng J.; Wang E. G. Phys. Rev. B 2011, 84, 205447. (34) Wakabayashi K. Phys. Rev. B 1999, 59, 8271. (35) Jia X. et al. Science 2009, 323, 1701. (36) Fujita M.; Wakabayashi K.; Nakada K.; Kusakabe K. J. Phys. Soc. Jpn. 1996, 65, 1920. (37) Nakada K.; Fujita M.; Dresselhaus G.; Dresselhaus M. S. Phys. Rev. B 1996, 54, 17954. (38) Jung J.; Pereg-Barnea T.; MacDonald A. H. Phys. Rev. Lett. 2009, 102, 227205. (39) Son Y.-W.; Cohen M. L.; Louie S. G. Phys. Rev. Lett. 2006, 97, 216803. (40) Brey L.; Fertig H. A. Phys. Rev. B 2006, 73, 235411. (41) Datta S., 'Electronic Transport in Mesoscopic Systems', Cambridge University Press (1995). (42) López Sancho M. P.; López Sancho J. M.; Rubio J. J. Phys. F: Met. Phys. 1984, 14, 1205. (43) Qiao Z. H.; Wang J. Nanotech. 2007, 18, 435402. (44) Wang J.; Guo H. Phys. Rev. B 2009, 79, 045119. 11 (45) Giannozzi P. et al., http://www.quantum-espresso.org. 12 4 3 2 1 (a) 0 (b) 3 2 1 0 (c) -1 (b) -2 -1 -2 -3 -1 (c) -2 -1 -2 -3 -4 X -3 X -4 -4 Γ X -3 X -4 AC, same topology Γ 5 4 3 2 (b) 1 0 (b) -1 -1 -2 -2 -3 -3 5 4 3 2 1 0 -1 -2 -3 (c) (cid:127)∆ (c) -1 -2 -3 6 5 4 3 2 0 -1 ) V e ( y g r e n E ) V e ( y g r e n E (a) 1 X Γ -1 -2 -2 6 5 ) ) V 4 e ( 3 y g r e 2 n E 1 V e ( y g r e n E (a) (a) 0 -1 -1 -2 -2 X 4 3 2 1 (d) 0 (d) -1 -1 -2 -2 -3 -1 -2 -3 3 2 1 0 -1 (e) -2 -3 -4 3 2 1 0 (f) (e) -1 (f) -2 -3 -4 X X Γ -1 -2 X -1 -2 -3 -4 -3 X X Γ -5 X -4 X -5 Γ AC, opposite topology 5 4 3 2 1 0 (d) (d) -1 -2 -3 -1 -2 -3 X (e) -1 (e) -2 -3 -4 4 3 2 1 0 -1 -2 -3 -4 (f) Δ (f) δ -1 -2 -3 -4 4 3 2 1 0 -1 -2 -3 -4 5 4 3 2 1 0 (g) -1 -2 5 4 3 2 (g) 1 (h) 0 (h) -1 -1 -2 (i) -1 -2 Γ 5 4 3 2 1 0 -1 -2 (i) X X Γ X (BNCNB) (BNCNB) (BNCBN) (BNCNB) (BNCNB) (BNCBN) ! X Γ -2 X X #% ) V e ( p a G y g r e n E 6 ) 5 - 4 3 2 1 0 / . - , + V e ( p a G y g r e n E !" !& !! ! 0 0 8 16 24 32 40 48 56 64 72 80 " #$ %& '% &! &" ($ $& )% "! 8 16 24 32 40 48 56 64 72 80 Carbon Atom Number m ZZ, same topology ZZ, opposite topology A ZZ, opposite topology B Γ X X Γ X X Γ X Γ X X Γ X X Γ X 72.89, :;9< =><8-. < Carbon Atom Number m (BN) (NB) where nitrogen bind with carbon and (NB) , also showing small fluctuations with a period of three hexagons. (jeil) for all cases. We consider the case of same topology (BN) (NB) where nitrogen bind with carbon and (NB) (jeil) for all cases. We consider the case of same topology (BN) , also showing small fluctuations with a period of three hexagons. FIG. 2: (Color online) Upper Panel: Band structures of zigzag graphene nanoroads embedded in hexagonal boron nitride (Color online) Upper Panel: Quasi 1D band structures of zigzag (ZZ) graphene Figure 3: FIG. 2: (Color online) Upper Panel: Band structures of zigzag graphene nanoroads embedded in hexagonal boron nitride obtained for di erent nanoroad widths corresponding to zero, one and three complete hexagons or equivalently = 0, 4 and obtained for di erent nanoroad widths corresponding to zero, one and three complete hexagons or equivalently = 0, 4 and nanoroads of different widths embedded in hexagonal boron nitride. The three band structures are 12 carbon atoms in the unit cell. The number 12 carbon atoms in the unit cell. The number (BN) and reversed topology configurations (BN) for widths corresponding to zero, one and three complete hexagons or equivalently m = 0, 4 and (BN) and reversed topology configurations (BN) (BN) where the boron bind with carbon. The kink bands are represented with a thick blue line. Lower Panel: Band structures of where the boron bind with carbon. The kink bands are represented with a thick blue line. Lower Panel: Band structures of 12 carbon atoms in a unit cell. The BN width number n = 40 for all cases. We consider the same armchair graphene nanoroads embedded in hexagonal boron nitride obtained for di erent nanoroad widths corresponding to armchair graphene nanoroads embedded in hexagonal boron nitride obtained for di erent nanoroad widths corresponding to = 0, 4 and 12 carbon atoms in the unit cell. Same and reversed topology configurations are considered. The kink bands that topology configuration, (BN)n− Cm− (BN)n, the reversed topology configuration A, (BN)n− Cm− = 0, 4 and 12 carbon atoms in the unit cell. Same and reversed topology configurations are considered. The kink bands that appear for opposite topology configurations are plotted with thick blue lines. A a trivial avoided gap oscillates with a period appear for opposite topology configurations are plotted with thick blue lines. A a trivial avoided gap oscillates with a period (NB)n, in which nitrogens bind with carbon and reversed topology configuration B, (NB)n− Cm− of three multiples of 4 C atoms in the unit cell whereas the band gap due to the finite confinement of an armchair nanoribbon of three multiples of 4 C atoms in the unit cell whereas the band gap due to the finite confinement of an armchair nanoribbon (BN)n, in which borons bind with carbon. States that are peaked in the middle of the nanoroads decreases proportionally to decreases proportionally to are represented by thick blue lines. Lower Panel: Band structures of armchair (AC) graphene nanoroads of different widths embedded in hexagonal boron nitride. The two band structures are II. QUANTUM TRANSPORT IN NANOROADS ' defines the crys- i.e., II. QUANTUM TRANSPORT IN NANOROADS for m = 0, 4 and 12 carbon atoms in a unit cell, corresponding to nanoroads that are 0, 1 and 3 tal topology of the upper and lower regions outside the graphene nanoroad in Fig. 3. measures the strength of hexagons wide. Same and reversed topology configurations are considered. The states that are So far we have discussed the electronic structure of the staggered potential in the BN fragment for which we So far we have discussed the electronic structure of localized in the middle of the nanoroads are again indicated by thick blue lines. As in the case have used X eV . The conductance from p-lead to graphene nanoroads in straight edges. In order to fur- graphene nanoroads in straight edges. In order to fur- of graphene nanoribbons the gaps have an oscillatory behavior with period 12 (corresponding to 3 q-lead can be evaluated using the multi-probe Landauer- ther understand what happens when we have a mixture ther understand what happens when we have a mixture Buttiker formula37 hexagons), embedded in a smooth envelope that follows a m−1 law. of zigzag and armchair crystal fragments we consider a of zigzag and armchair crystal fragments we consider a graphene nanoroad of curved shape as illustrated in Fig. graphene nanoroad of curved shape as illustrated in Fig. 3. For this model we used a finite central region consist- 3. For this model we used a finite central region consist- ing of nhor X nver (jeil) lattices. The electronic structure ing of nhor X nver (jeil) lattices. The electronic structure was modeled through a nearest neighbor tight-binding was modeled through a nearest neighbor tight-binding hamiltonian of the form hamiltonian of the form i.e., ' defines the crys- tal topology of the upper and lower regions outside the graphene nanoroad in Fig. 3. measures the strength of the staggered potential in the BN fragment for which we have used X eV . The conductance from p-lead to q-lead can be evaluated using the multi-probe Landauer- Buttiker formula37 WITH GENERAL EDGES WITH GENERAL EDGES . Here, =' . Here, =' Tr[ (2) Tr[ (2) pq pq ij ij i,A i,A i,B i,B (1) (1) ) is a -orbital creation (annihilation) oper- where -orbital site ator for an electron at the site energy, and = 2 eV is the nearest neighbor hopping energy both for graphene and BN. We introduce a stag- gered AB sublattice potential to describe the BN treat- ing the site energies at sublattice to be opposite, ) is a -orbital creation (annihilation) oper- where 13 -orbital site ator for an electron at the site energy, and = 2 eV is the nearest neighbor hopping energy both for graphene and BN. We introduce a stag- gered AB sublattice potential to describe the BN treat- is a is a where Gr,a are the retarded and advanced Green func- tions of the central scattering region. The quantity is a linewidth function describing the coupling between p-lead and the central region and can be obtained calcu- lating the self-energy of p-lead using a transfer ma- trices method.38 To visualize the 1D kink state transport channel we represent the local density of states (LDOS) at each site where Gr,a are the retarded and advanced Green func- tions of the central scattering region. The quantity is a linewidth function describing the coupling between p-lead and the central region and can be obtained calcu- lating the self-energy of p-lead using a transfer ma- trices method.38 To visualize the 1D kink state transport channel we represent the local density of states (LDOS) at each site evaluated through evaluated through ) = [G ) = rr [G (3) (3) (a) (b) h-BN graphene h-BN h-BN graphene h-BN Left lead Central scattering regime Right lead Figure 4: Schematic plot of the two-terminal BNC nanoroad transport measurement that we model. (a) Straight BNC nanoroads; (b) Curved BNC nanoroads: the two leads are exactly the same as those in panel (a). The nanoroad will support kink-like states when the two BN regions have opposite topology. ) h / 2 e ( L R G 5 4 3 2 1 0 (a) straight curved BN-C-BN (b) straight curved BN-C-NB 5 4 3 2 1 0 -0.4 -0.2 0.0 0.2 0.4 / t -0.4 -0.2 0.0 0.2 0.4 / t Figure 5: (Color online) Conductance calculations for straight zigzag-edged and curved graphene nanoroads embedded in the h-BN as a function of the Fermi energy e . Panel (a): conductance for graphene nanoroads embedded within h-BN sheets with the same topology. An insulating gap clearly arises near the charge neutrality point. Panel (b): graphene nanoroads embedded within h-BN sheets with different topologies. The conductance plateau near the charge neutrality region is due to a topologically supported edge state. Irregular and curved edges introduce a extremely small backscattering probability. 14 4 + 0.000 0.007 0.2180 0.218 + ) m n ( y 0 -4 -6 - - e /t = 0.116 G = 0.998 e2/h + - 0 x (nm) 6 + - Figure 6: (Color online) Local density-of-states of a kink state in a curved carbon nanoroad em- bedded between opposite topology BN sheets, suggested here by the '+' and '−' signs, calculated through a nearest neighbor tight-binding model as discussed in the main text. The edges consist of a mixture of zigzag and armchair fragments but the disorder does not degrade transmission. 15
1902.00708	1	1902	2019-02-02T12:29:15	Hybrid plasmon-magnon polaritons in graphene-antiferromagnet heterostructures	[ "cond-mat.mes-hall" ]	We consider a hybrid structure formed by graphene and an insulating antiferromagnet, separated by a dielectric of thickness up to $d\simeq 500 \,nm$. When uncoupled, both graphene and the antiferromagnetic surface host their own polariton modes coupling the electromagnetic field with plasmons in the case of graphene, and with magnons in the case of the antiferromagnet. We show that the hybrid structure can host two new types of hybrid polariton modes. First, a surface magnon-plasmon polariton whose dispersion is radically changed by the carrier density of the graphene layer, including a change of sign in the group velocity. Second, a surface plasmon-magnon polariton formed as a linear superposition of graphene surface plasmon and the antiferromagnetic bare magnon. This polariton has a dispersion with two branches, formed by the anticrossing between the dispersive surface plasmon and the magnon. We discuss the potential these new modes have for combining photons, magnons, and plasmons to reach new functionalities.	cond-mat.mes-hall	cond-mat	Hybrid plasmon-magnon polaritons in graphene-antiferromagnet heterostructures Y. V. Bludov1, J. N. Gomes1, G. A. Farias3, J. Fernández-Rossier2,4, M. I. Vasilevskiy1,2, N. M. R. Peres1,2 1Department of Physics, Center of Physics, and QuantaLab, University of Minho, Campus of Gualtar, 4710-057, Braga, Portugal 2QuantaLab,International Iberian Nanotechnology Laboratory (INL), Av. Mestre José Veiga, 4715-330 Braga, Portugal 3Departamento de Física, Universidade Federal do Ceará, Caixa Postal 6030, Campus do Pici, 60455-900 Fortaleza, Ceará, Brazil and 4Departamento de Física Aplicada, Universidad de Alicante, Carretera de San Vicente del Raspeig 03690 San Vicente del Raspeig, Alicante, España 9 1 0 2 b e F 2 ] l l a h - s e m . t a m - d n o c [ 1 v 8 0 7 0 0 . 2 0 9 1 : v i X r a We consider a hybrid structure formed by graphene and an insulating antiferromagnet, separated by a dielectric of thickness up to d (cid:39) 500 nm. When uncoupled, both graphene and the antiferro- magneic surface host their own polariton modes coupling the electromagnetic field with plasmons in the case of graphene, and with magnons in the case of the antiferromagnet. We show that the hy- brid structure can host two new types of hybrid polariton modes. First, a surface magnon-plasmon polariton whose dispersion is radically changed by the carrier density of the graphene layer, includ- ing a change of sign in the group velocity. Second, a surface plasmon-magnon polariton formed as a linear superposition of graphene surface plasmon and the antiferromagnetic bare magnon. This polariton has a dispersion with two branches, formed by the anticrossing between the dispersive surface plasmon and the magnon. We discuss the potential these new modes have for combining photons, magnons, and plasmons to reach new functionalities. I. INTRODUCTION Plasmons, excitons, phonons, and magnons are typi- cal examples of collective excitations in condensed mat- ter systems. They all imply the presence of poles with frequencies Ωp in the spectrum of the response function that describes the interaction of the system with elec- tromagnetic waves. As a result, the propagation of elec- tromagnetic waves with frequency ω in a material that hosts these collective modes is strongly modified, or even suppressed altogether, for ω (cid:39) Ωp. This general physical phenomenon is rationalized in terms of the formation of new collective modes known as polaritons. Quantum mechanically, polaritons are described as hy- brid collective excitations that are linear superpositions of a matter collective excitation and a photon. Semi- classically, they are described using Maxwell equations and constitutive relations that include the frequency de- pendent response functions. In both instances, the un- derlying physical phenomenon is the emergence of a new type of wave or excitation, with properties different from those of the constituent collective mode and electromag- netic wave. Excitons, phonons, and plasmons, couple predomi- nantly to the electric component of the electromagnetic wave. In contrast, for magnon-polaritons, it is the mag- netic field that couples to the spins. Most of the work so far has focused on polaritons that couple electromag- netic waves with just one type of collective mode (exci- tons, phonons, plasmons, spin waves). Interestingly, the same electromagnetic field would couple both to the spin and charge sector in a system that hosts both spin and charge collective modes. The electromagnetic field of po- laritons would thus provide a coupling channel between excitations that are normally un-coupled. The fabrication of nanostructures offers a new arena to explore hybrid systems with collective modes in the spin and charge sectors, that could result in a new type of polariton, mixing spin and charge collective modes. Here we explore this possibility in a system that seems easy to fabricate with state of the art techniques. We consider the coupling of surface magnon polaritons of an uniax- ial antiferromagnet (AF) to surface plasmon polaritons (SPPs) in graphene. The antiferromagnetic resonance (AFMR) frequency in insulating unixaxial antiferromagnets, that ultimately determines the magnon-polariton frequency, occurs in the THz range, well above the typical GHz range for ferro- magnetic resonance and, importantly, within the spectral range of graphene SPPs. The difference between AFMR and ferromagnetic resonance arises from the fact that the former is determined by the interplay of exchange and anisotropy [1], whereas the latter is only given by mag- netic anisotropy, which is much smaller in most cases. It has been shown [2, 3] that uniaxial AFs, such as FeF2, host both bulk and surface magnon polaritons (SMPs). These surface polaritons decay exponentially as we move away from the antiferromagnet-dielectric interface. The formation of hybrid modes occurs when the un- coupled modes are degenerate. Therefore, the existence of an experimental knob to tune the frequency of the modes is very convenient. In the case of graphene, gat- ing controls the carrier density, leading to a change of the dispersion curve of SPPs. Therefore, here we consider a graphene sheet at a distance d from the surface of an in- sulating AF, as shown in Fig. 1. Since both graphene and the insulating AF host their own polariton modes, here we explore whether this hybrid artificial material system hosts a new type of hybrid polariton that couples graphene surface plasmons and AF magnons at the same romagnet. Thus, we consider the semi-infinite AF, oc- cupying the half-space z < 0. The other half-space z > 0 is supposed to be occupied by the vacuum with the graphene monolayer, arranged at plane z = d paral- lel to the AF surface (see Fig. 1). The semi-infinite uniaxial antiferromagnet, such as FeF2 or MnF2[2, 3], is described by the permeablilty ten- sor µ(ω) = 2 (1)  µa 0 −iµb 0 0 iµb 0 µa 1  . The off-diagonal component µb is finite in the presence of an external magnetic field H0, that permits to tune the antiferromagnetic resonance frequency. In addition, application of the latter provides tunability of the reso- nance. In this work we only consider the case H0 = 0, so that µb = 0 and [1]: √ (2) µa(ω) = 1 + Here, Ω0 = γµ0 (cid:112)2HaHe + H 2 2Ω2 s 0 − ω2 , Ω2 a is the antiferromag- netic resonance frequency, that, unlike the case of uniax- ial ferromagnets, depends not only on the anisotropy field Ha, but as well on the exchange field He, which makes Ω0 much larger than the usual ferromagnetic resonance frequencies. The gyromagnetic ratio is given by γ = e/(2m) where e and m are the charge and mass of free electron, corre- spondingly. The so called saturation frequency is given by Ωs = γµ0 2HaMs, where Ms denotes the saturation magnetization of each sublattice. A calculation of the permeability tensor for this sys- tem was performed long ago [1, 5 -- 8]. Equation (2) can be obtained from a microscopic model Hamiltonian for spins, using both the spin wave approximation and Kubo formula for linear response to a transverse ac field of fre- quency ω and zero wavevector. Expression (2) is real, ignoring thereby losses. These could be included by re- placing ω by ω + iΓ, where Γ describes a scattering rate. The spectral range for which µa(ω) < 0 plays a very special role, as it becomes evident below. The condition µa(ω) < 0 is met for Ω0 < ω <(cid:112)Ω2 s. 0 + 2Ω2 B. Maxwell equations and boundary conditions The electromagnetic waves in such a layered structure are governed by macroscopic Maxwell equations, + J (2D)δ (z − d) ∂D rotH = ∂t rotE = − ∂B divD = ρ(2D)δ (z − d) divB = 0, ∂t (3) (4) (5) (6) Figure 1. Schematic drawing of the system considered in this work: a graphene sheet is located at a distance d from the surface of an antiferromagnet, characterized by a magnetic permeability tensor µ(ω). time. In this work we show that indeed the tunability of the electromagnetic properties of an antiferromagnetic insu- lator can be achieved by gating a graphene sheet (see Fig. 1). In particular, we find a smooth transition from the conventional regime where the system has the en- ergy propagation oriented along the same direction of the SMP's wave vector to a regime where the energy flux is opposite to the wavevector, i.e. the group velocity of the hybrid excitation is negative. If the dielectric layer be- tween graphene and the antiferromagnet has a negative electric permittivity, as it happens in a polar crystal near optical phonon resonances, a metamaterial [4] with both negative and µ can be achieved, thus exhibiting nega- tive refraction. Such a tunable system allows to control the direction of energy flow at the surface of the anti- ferromagnet, thus providing a mechanism for directional propagation of the electromagnetic energy, without the need of an external magnetic field. II. PROBLEM STATEMENT AND MAIN EQUATIONS A. Antiferromagnetic permeability The main objective of this paper is to investigate how the presence of graphene in the vicinity of an antiferro- magnet influences the spectrum of SMPs, and vice versa, how the SPPs in graphene are affected by the antifer- where delta-functions in Eqs.(3) and (5) describe the two- dimensional nature of charges ρ(2D) and current J (2D) in the graphene monolayer. Maxwell equations (3) -- (6) can be solved separately in three spatial domains z < 0, 0 < z < d and z > d, which further in the paper will be denoted by j = 1, 2 and 3, correspondingly. In the framework of this formalism, the Maxwell equations have the form: , ∂D(j) rotH(j) = ∂t rotE (j) = − ∂B(j) divD(j) = 0, divB(j) = 0, ∂t , (7) (8) (9) (10) and media indices are added as superscripts to the elec- tric and magnetic fields E (j), H(j) D(j), B(j). It is no- table that charges and currents induced in graphene do not enter explicitly Eqs. (7). These quantities are present in boundary conditions, which couples the electromag- netic fields in media j = 2 and j = 3. The boundary conditions at graphene plane take the explicit form: uz ×(cid:16)E (3) − E (2)(cid:17)(cid:12)(cid:12)(cid:12)z=d uz ×(cid:16)H(3) − H(2)(cid:17)(cid:12)(cid:12)(cid:12)z=d (cid:16)D(3) − D(2)(cid:17)(cid:12)(cid:12)(cid:12)z=d (cid:16)B(3) − B(2)(cid:17)(cid:12)(cid:12)(cid:12)z=d = 0, = J (2D), · uz = ρ(2D), · uz = 0. Here uz is a unit vector in the direction z, "×" and "·" mean vector and scalar products, respectively. The antiferromagnet is insulating, and therefore has no free charges and currents. In addition, we are assuming there is no surface magnetization. As a result, the boundary conditions between media j = 1 and j = 2, at the surface of the antiferromagnet, can be written as: = 0, uz ×(cid:16)E (2) − E (1)(cid:17)(cid:12)(cid:12)(cid:12)z=0 uz ×(cid:16)H(2) − H(1)(cid:17)(cid:12)(cid:12)(cid:12)z=0 (cid:16)D(2) − D(1)(cid:17)(cid:12)(cid:12)(cid:12)z=0 (cid:16)B(2) − B(1)(cid:17)(cid:12)(cid:12)(cid:12)z=0 = 0, · uz = 0, · uz = 0. In the following we look for the equations describing electromagnetic waves with the propagation vector k ly- ing in-plane. There are two cases, k parallel and perpen- dicular to the AF's staggered magnetization Ms, that we take along y (see Fig. 1). 3 C. In plane propagation perpendicular to the staggered magnetization We consider first the case where the electromagnetic wave propagates in the direction x, perpendicular to the direction of magnetization. This means that the problem under consideration is uniform in the direction y (i.e. ∂/∂y ≡ 0), and Maxwell equations (7)-(10) can be de- composed into two independent subsystems, which cor- respond to TE and TM polarizations. The TE-polarized wave includes the y-component of the electric field E (j) as well as x- and z-components of the magnetic field H(j), i.e. E (j)(x, z, t) = uyE (j) H(j)(x, z, t) = uxH(j) y (x, z, t), x (x, z, t) + uzH(j) z (x, z, t). (19) (20) Here ux and uy are unit vectors in directions x and y, respectively. The second subsystem, describing the TM- polarized wave, possesses x- and z-components of the electric field and y-component of the magnetic field, H(j)(x, z, t) = uyH(j) E (j)(x, z, t) = uxE (j) y (x, z, t), x (x, z, t) + uzE (j) z (x, z, t). (21) (22) Moreover, one can assume the temporal and spatial de- pendencies of the electromagnetic fields as those of a plane wave with frequency ω, travelling along the x-axis with wave-number k, that is, we can write E (j)(x, z, t) = E(j)(z) exp(ikx − iωt), H(j)(x, z, t) = H(j)(z) exp(ikx − iωt). In this formalism the wave amplitudes E(j) and H(j) depend upon z-coordinate only. We now take into account the constitutive relations: D(j) = ε0E(j) exp(ikx − iωt), B(1) = µ0 µH(1) exp(ikx − iωt), B(j(cid:54)=1) = µ0H(j(cid:54)=1) exp(ikx − iωt). (23) (24) (25) Such form of the constitutive relations describes the fact that the dielectric permittivities of all three media are equal to vacuum permittivity ε0, and the magnetic per- meability tensor of antiferromagnetic medium (j = 1) is equal to µ0 µ. Under all these assumptions, Maxwell equations (7) for the TE-polarization take the form dH (j) dz − ikH (j) z = −iωε0E(j) y , x y dE(j) dz = −iωµ0 [µa (ω) δj,1 + (1 − δj,1)] H (j) x , y = iωµ0 [µa (ω) δj,1 + (1 − δj,1)] H (j) ikE(j) z (26) (27) (28) where δj,1 is the Kronecker delta. Correspondingly, the (11) (12) (13) (14) (15) (16) (17) (18) Maxwell equations for the TM-polarization read: dE(j) x dz − ikE(j) z = iωµ0H (j) y , dH (j) y = iωε0E(j) x , dz y = −iωε0E(j) z , ikH (j) (29) (30) (31) It is crucial that Eqs. (26) -- (28) for the TE-polarization involve only the µxx = µzz components of the magnetic permeability tensor µ(ω) [see Eq. (1)]. As a consequence, the magnetic medium is effectively isotropic with respect to the TE-polarized wave, when electromagnetic wave propagates along x-direction (perpendicular to the stag- gered magnetization). At the same time, only yy compo- nent of the magnetic permeability tensor µ(ω) is present in the Maxwell equations for the TM-polarized wave (29), which is equal to unity [see Eq. (1)]. Therefore, we would expect that the AF medium in the structure depicted in Fig. 1 would not exert any influence on the spectrum of the TM-polarized wave. As we will see this is not exactly the case near the resonance frequency Ω0. D. In plane propagation parallel to the staggered magnetization We now consider the propagation along the y direction, parallel to the staggered magnetization. In this case the homogeneity of the system under consideration in the direction x (∂/∂x ≡ 0) also implies the separation of Maxwell equations (7)-(10) into the TE subsystem: ikH (j) z − dH (j) y dz = −iωε0E(j) x , dE(j) x dz − ikE(j) and the TM subsystem: = iωµ0H (j) y , x = iωµ0 [µa (ω) δj,1 + (1 − δj,1)] H (j) z (32) (33) (34) (35) (36) (37) (38) z − dE(j) y dz = ikE(j) = iωµ0 [µa (ω) δj,1 + (1 − δj,1)] H (j) x , dH (j) x dz − ikH (j) obtaining = −iωε0E(j) y , x = −iωε0E(j) these . While z equation, we used the plane-wave spatio-temporal dependence of the field E (j)(y, z, t) = E(j)(z) exp(iky − iωt), H(j)(y, z, t) = H(j)(z) exp(iky − iωt), as well as the constitutive rela- tions, similar to Eqs. (23) (except the dependence upon y-coordinate instead of x). As a result, the antiferro- magnet, whose response involves components yy and zz 4 Magnon Plasmon Surface polariton type System Pol. Wavevector TE k · Ms = 0 TM isotropic AF+G TE k · Ms = 0 AF+G TM k × Ms = 0 Magnon-plasmon Plasmon-magnon AF G Table I. Summary of the different surface polariton excitations discussed in this work. The type of the polariton indicates the elementary excitation coupled to the EM field, except for the third line where plasmons are not directly involved in the hybrid wave. of the magnetic permeability tensor (1), is effectively anisotropic with respect to the TE-polarized waves [see Eqs. (32)]. Furthermore, the AF medium influences the properties of the TM-polarized waves [see Eq. (35)]. III. UNCOUPLED MODES: SURFACE PLASMON-POLARITONS AND SURFACE MAGNON-POLARITONS In this section we briefly revisit the properties of the SPPs in graphene, on one side, and the SMPs in the AF in the other, ignoring their mutual coupling. This provides a background to understand the nature of the new hybrid collective modes that arise in the combined graphene/AF structure. We keep the discussion at a qualitative level. The quantitative theory presented in the next sections includes, as limiting cases, a theoretical description of these excitations. A. Magnon -- polaritons The case of bulk magnon-polaritons for a uniaxial an- tiferromagnet was studied by [2]. It was found that only TE modes exist, with a dispersion relation that we derive in the appendix A and is depicted in Fig.2 by dashed red lines. The magnon -- polariton dispersion is mathemati- cally identical to the case of Hopfield exciton -- polaritons in a semiconductor. Magnon -- polaritons come in two branches [acoustical ωa(k) and optical ωo(k)], both twice degenerate on account of the dimension of the symme- try plane perpendicular to the easy axis. At frequen- cies far from the AFMR resonance, ω ≶ Ω0 these two branches are close to the photon dispersion curve ω = ck, while in the frequency range ω (cid:46) Ω0 the lowest, acoustic branch asymptotically approaches the AFRM frequency as k → ∞, i.e. ωa(∞) = Ω0. In the vicinity of the AFMR frequency two modes are separated by the frequency gap, whose value is roughly given by ωo(0)− ωa(∞) ≈ Ω2 s/Ω0. At the surface of AF, the collective excitations of spins, i.e. magnons can be coupled to an electomag- netic wave, forming surface magnon-polaritons (SMP). The key property of SMPs [see the first line of Table I] is that they are TE-polarized waves and it was first consid- 5 since they are characterized by both a longer lifetime and a higher degree of field confinement [10, 11]. If graphene layer is deposited on a polar substrate, the electromag- netic field of graphene SPPs can interact with optical phonons in the substrate, thus forming hybrid modes called surface plasmon-phonon-polaritons [12 -- 16]. Here, we also expect hybrid polaritons invloving two physically distinct elementary excitations in the materials involved. This mode will be called surface plasmon-magnon po- lariton (SPMP) and its properties are summarized in the forth line of Table I. The study of this mode will be considered in detail in Sec. V. However, the system considered in the present work is different from sur- face plasmon-phonon-polaritons in one important aspect. The two materials combined in our system, if taken sep- arately, support surface waves whose polarizations are orthogonal to each other. From Table I, it is apparent that graphene SPPs are TM modes whereas AF's surface magnon -- polaritons are TE modes. Therefore, in order to study the polaritons of the hybrid system, we need to consider both TM and TE modes. IV. SURFACE MAGNON-PLASMON POLARITONS A. General equations for TE modes In this Section we shall demonstrate that interaction between magnons and free charges in graphene via elec- tromagnetic radiation modifies the spectrum of SMP. Boundary condition on graphene (12) couples the in- plane components of the electric and magnetic fields; in the case opf TE modes it involves transverse plas- mons in graphene. For such plasmons, the current is perpendicular to the wavevector, the charge density is kept constant[17] and they do not interact with the elec- tromagnetic radiation directly. However, in the hybrid structure considered here they can interact indirectly, through the AF whose magnons do couple to the TE- polarized radiation. Such hybrid evanescent waves will be called surface magnon -- plasmon -- polaritons (SMPP). They propagate along the direction perpendicular to the staggered magnetization in the antiferromagnet. It should be noticed that the frequency range of the an- tiferromagnetic resonance lies in the THz spectral region, where interband transitions in graphene play no role. Therefore, we consider the optical response of graphene described by a Drude formula without losses:[18] σ(ω) = i 2e2 h EF ω , (39) with EF being the Fermi energy of doped graphene. This equation is valid as long as the Fermi energy is much larger than kBT (kB is the Boltzmann constant and T the temperature). Figure 2. Schematic dispersion relation of surface (blue solid line) and bulk (red dashed lines) magnon-polaritons in the system without graphene, EF = 0, and with the AF param- eters Ωs = 0.5Ω0. The black dashed line corresponds to the s, 0 + Ω2 s are depicted by horizontal dash-and-dotted vacuum light line ω = kc, while frequencies Ω0, (cid:112)Ω2 and (cid:112)Ω2 0 + 2Ω2 black lines (from bottom to top, respectively). ered by Camley e Mills [2] (see also Ref. [9]). One of the first reported [3] observations of SMPs was in the anti- ferromagnet FeF2 using the technique of attenuated total reflection. The same method first used to observe SPPs in metallic-dielectric interfaces. These surface magnon -- polaritons [depicted by blue solid line in Fig. 2] only exist for ck > Ω0, as we show below [see subsection IV B]. In Sec. IV we study how the interaction between elec- tromagnetic field of SMP at vacuum/AF interface and forced charge-carrier oscillations in graphene modify the SMP spectrum. The resulting hybrid mode will be re- ferred to as surface magnon-plasmon polariton and its fundamental properties are summarized in the third line of Table I. B. Graphene surface plasmon-polaritons Graphene SPPs can be understood as solutions of the Maxwell equations that describe an electromagnetic wave propagating along a conductive graphene sheet. The elec- tromagnetic field is strongly confined in the neighbour- hood of the graphene, with evanescent off-plane tails. Graphene SPPs are TM-polarized and their dispersion curve, ω(k) < ck, can be tuned by controlling the carrier density. Fundamental properties of SPPs in graphene are briefly summarized in the second line of Table I. Compared to the SPPs at the surface of noble met- als, the graphene polaritons have significant advantages The propagation of TE-polarized waves is governed by the Maxwell equations in the form (26) -- (28). Substitu- tion of Eqs. (27) and (28) into Eq. (26) results into the Helmholtz equation, − d2E(j) y y dz2 + k2E(j) ω2 c2 [µa (ω) δj,1 + (1 − δj,1)] E(j) y , = (40) whose solution in semi-infinite media j = 1 and j = 3 can be expressed as E(1) y (z) = E(1) y (0) exp(β(1)z), E(3) y (z) = E(3) y (d) exp (cid:104)−β(2) (z − d) (cid:105) (41) (42) , y (0) and E(3) with E(1) (d) being the values of electric field at the surface of the antiferromagnet and graphene, re- spectively, and y β(1) =(cid:112)k2 − ω2µa (ω) /c2, β(2) =(cid:112)k2 − ω2/c2. (43) (44) In the considered framework we assume, for simplicity, that both the antiferromagnet and graphene are lossless media, which means that both the in-plane wavevector k and frequency ω have real and positive values. Moreover, since we are interested in studying the surface wave, β(1) and β(2) are also real and characterize the inverse pen- etration length of the evanescent fields. The respective signs of the exponents in Eqs. (41) and (42) were chosen to satisfy the boundary conditions at z = ±∞ [namely, (∞) = 0], which describe the absence E(1) of modes, growing exponentially towards z → ∞. (−∞) = E(3) The respective magnetic fields in media j = 1 and j = 3 can be obtained by substituting Eqs. (41) and (42) into Eqs. (27) and (28), and expressing the fields as y y x (z) = − H (1) β(1) iωµ0µa (ω) E(1) y (0) ez1, H (1) z (z) = H (3) x (z) = H (3) z (z) = k ωµ0µa (ω) β(2) iωµ0 E(3) y E(1) y (0) ez1, (d) e−(z2−d2), k ωµ0 E(3) y (d) e−(z2−d2), (45) (46) (47) (48) where we have adopted the notation zj ≡ β(j)z, dj ≡ β(j)d with j = 1, 2. For the spacer layer j = 2 that separates graphene and the AF's surface, 0 < z < d, there are no restrictions on the presence of exponentially growing or decaying modes. As a result, the solution is composed as the superposition of these two modes. In terms of hyperbolic functions and amplitudes of electric field at boundaries z = 0 and z = d, this solution can be represented as y (d)FS(z2), E(2) y (z) = E(2) (cid:104) y (0)FS(d2 − z2) + E(2) (cid:105) y (0)FC(d2 − z2) E(2) (cid:104) y (d)FC(z2) y (0)FS(d2 − z2) + E(2) E(2) H (2) x (z) = ηx −E(2) H (2) z (z) = ηz , (cid:105) y (d)FS(z2) 6 (49) (50) (51) , (52) where we have defined FS(z2) ≡ sinh(z2) sinh(d2) , FC(z2) ≡ cosh(z2) sinh(d2) , ηx ≡ β(2) iωµ0 , ηz = k ωµ0 . (53) (54) For the particular case of TE-polarized wave (and plane wave temporal and spatial dependencies, men- tioned above), first and second relations in the boundary conditions (11) and (16) are expressed as (d) , y x (d) = σ(ω)E(2) y y E(3) (d) = E(2) x (d) − H (2) H (3) E(2) (0) = E(1) x (0) − H (1) H (2) y (0) , y x (0) = 0. (d) , (55) (56) (57) (58) It should be pointed out that boundary condition (56) was obtained by using the two-dimensional current in graphene J (2D) expressed via the graphene conductivity and electric field as J (2D) = uyσ (ω) E(2) y (d) exp (ikx − iωt) . (59) Substitution of Eqs. (45) and (51) into boundary condi- tions (56) results into the homogeneous linear system of equations:(cid:32) M11 (cid:33)(cid:32) −β(2) sinh(d2) sinh(d2) M22 −β(2) E(2) E(2) (d) y y (0) (cid:33) (cid:32) (cid:33) 0 0 = , (60) where M11≡ β(2) + β(2) tanh(d2) − iωµ0σ (ω) , M22≡ β(2) tanh(d2) + β(1) µa (ω) . (61) (62) Non trivial solutions of Eq. (60) occur when the deter- minant of the matrix vanishes: (cid:18) β(2) (cid:19)2 sinh(d2) M11M22 − = 0 (63) For a given value of k, this equation can be satisfied by several values of ων(k) that define the different polariton modes in the system, which are distinguished by the in- dex ν. In the following we discuss these polariton modes in two cases. First, we take the d → ∞ limit, where there is no coupling between graphene and the AF's sur- face. This permits to recover an expression for the surface magnon-polariton [2]. Later we shall also take the limit where β(2)d is small, for which the presence of graphene modifies the SMP properties. B. No dressing: the d → ∞ limit In the case of infinite distance d → ∞ between the an- tiferromagnet surface and graphene monolayer, the dis- persion relation (63) transforms into (cid:105)(cid:20) (cid:21) (cid:104) 2β(2) − iωµ0σ (ω) β(2) + β(1) µa (ω) = 0. (64) The term in the first braces of Eq. (64) is always positive owing to the positiveness of the imaginary part of Drude conductivity (39)[19], while setting equal to zero the sec- ond term in braces in (64) yields the dispersion relation of the SMP existing at single interface between vacuum and antiferromagnet[7]. (cid:112)Ω2 Since both β(1) and β(2), defined in Eqs.(43), are positive, Eq.(64) only has solutions when µa (ω) < 0, in the aforementioned frequency range Ω0 ≤ ω ≤ i.e. s. The simultaneous positiveness of the ar- 0 + 2Ω2 guments of β(1) and β(2) in that range takes place when k ≥ ω/c, i.e., at the right of the "light line" ω = ck (which is depicted in Fig. 2 by black dashed line). SMP's disper- sion relation (64) after some algebra can be expressed in the explicit form (cid:20) (cid:115)(cid:18) ω = Ω2 s + Ω2 0 2 + c2k2 (cid:19)2 − c2k2 − Ω2 0 2 + Ω2 s (Ω2 0 + Ω2 s) 1/2 , (65) which is depicted in Fig. 2 by solid blue line. The SMP's spectrum starts on the light line at frequency ω = Ω0 and wavevector k = Ω0/c. In the vicinity of this point, the dispersion relation is described approximately by the relation ω ≈ Ω0 + 2Ω2 s 0 + 2Ω2 s Ω2 (ck − Ω0) . asymptotically approaches the frequency ω =(cid:112)Ω2 At large wavevectors, k → ∞, the SMP's spectrum 0 + Ω2 s (66) as ω ≈(cid:113) (cid:20) (cid:21) Ω2 s + Ω2 0 1 − Ω2 4c2k2 s . (67) 7 Expectedly, the SMP's spectrum appears in the gap between two branches of the TM-polarized bulk magnon- polariton dispersion relation, β(1) = 0, (68) which is depicted in Fig. 2 by red dashed lines (see Ap- pendix A). C. Dressing: the finite β(2)d < 1 case We now discuss the influence of graphene on the prop- erties of the SMP. This happens when the distance be- tween antiferromagnet and graphene is such that β(2)d is not very large and the Fermi energy in graphene is not at the Dirac point, EF (cid:54)= 0. In this case the spectrum of SMPs is strongly modified owing to the influence of free charges in graphene on the electromagnetic field of the SMP, supported by the surface of the AF. The SMP spectrum (63) for relatively small distance d = 500 nm is depicted in Fig. 3(a) for different values of the Fermi en- ergy. Thus, for finite doping of the graphene, the dressed SMP spectrum has a starting-point with the frequency ω > Ω0, lying on the light line. An increase of the Fermi energy EF results into the shift of the starting-point of the spectrum towards higher frequencies. An expression for the starting-point frequency (cid:1) s + Ω2 0 (cid:34) Ω2 0a − 2b(cid:0)2Ω2 (cid:35)1/2 2 (a − b) ωi = (cid:112)Ω4 + 0a2 + 8abΩ2 2 (a − b) s (2Ω2 s + Ω2 0) > Ω0, (69) (cid:112)Ω2 (cid:112)Ω2 can be obtained explicitly from the dispersion relation (63) by putting the condition β(2) = 0. In Eq. (69) a = [1 + 4αEF d/(c)]2, b = 8 [αEF /(Ωs)]2, and α = e2/(4πε0c) is the fine-structure constant. It is apparent that results are independent of the sign of EF , i.e., for graphene with extra electrons or holes. In the k → ∞ limit, the spectrum tends to ω = s. Therefore, as EF is ramped up and the 0 + Ω2 spectrum is pushed up in frequency at the smallest al- lowed values of k, so that the starting SMPP's fre- quency becomes larger than that limiting frequency (ωi > s), their group velocity vg = dω/dk has to be negative. This happens at experimentally attain- able dopings of the graphene. For EF = 0.01 eV and EF = 0.03 eV, orange and green lines in Fig. 3(c), respec- tively, vg < 0 in a range of high values of k. For higher values of EF [ EF = 0.4 eV, red line in Fig. 3(c)] vg < 0 for all values of k. 0 + Ω2 This result is distinct from the zero Fermi energy case, where SMPP's group velocity [slope of the dispersion curve, ω(k), depicted by solid blue line in Fig. 3(a)] is positive in all range of frequencies and wavevectors. It is possible to see, that the group velocity is much smaller 8 V. SURFACE PLASMON-MAGNON POLARITONS A. Equations for TM modes We now consider the case of TM-polarization, for which the graphene layer is able to sustain SPPs, os- cillations of charge-carrier density in graphene coupled to the electromagnetic radiation[21, 22]. We anticipate our main finding: in the AF-graphene coupled structure, the graphene SPP is hybridized with the AF magnon, resulting in a polariton spectrum with 2 branches, that reflects the emergence of a hybrid collective mode that combines graphene plasmons with AF magnons. We shall call these hybrid excitations surface plasmon -- magnon -- polaritons (SPMPs). We address the case where the electromagnetic field is TM polarized and wave propagates along the y-axis, i.e. parallel to the staggerred magnetization. In this situa- tion the electromagnetic field is defined by the Maxwell equations in the form of Eqs. (35). Their solutions in the AF region can be expressed as (0) exp(z1), y E(1) ωε0 β(1) x (z) = −i H (1) (z) = E(1) E(1) (0) exp(z1), (z) = −i k E(1) E(1) y y z β(1) y (0) exp(z1), The solutions in the spacer region j = 2 can be written up as: (cid:104) (70) (71) (72) (cid:105) H (2) x (z) = iωε0 β(2) y (0)FC(d2 − z2) − E(2) E(2) y (d)FC(z2) (cid:105) y (d)FC(z2) , (73) (74) E(2) E(2) y (d)FS(z2), (cid:104) y (0)FS(d2 − z2) + E(2) y (z) = E(2) ik β(2) y (0)FC(d2 − z2) − E(2) E(2) . (75) Finally, the solutions in the j = 3 region, above z (z) = graphene, read: H (3) x (z) = i ωε0 β(2) (z) = E(3) y k (z) = i β(2) x (d)e−(z2−d2), E(3) (d) e−(z2−d2), (d) e−(z2−d2). E(3) y E(3) y E(3) z (76) (77) (78) Here β(1) and β(2) are defined in Eqs. (43). Using the same boundary condition equations as those in Eqs. (55)-(58) adapted for the TM polarization we ob- tain the linear homogeneous system of equations in the form(cid:32) β(2) sinh(d2) M12 −1 M21 1 β(2) sinh(d2) (cid:33)(cid:32) (cid:33) (cid:32) (cid:33) 0 0 E(2) E(2) y (d) y (0) = , (79) Figure 3. (a) Surface magnon -- plasmon -- polariton (SMPP) spectrum in the AF/graphene structure for Fermi energy val- ues EF = 0 (solid blue line), 0.01 eV (solid orange line), 0.03 eV (solid green line), and 0.4 eV (solid red line). Black dashed line stands for the light line in vacuum, ω = ck; (b) Spatial distributions of the SMPP electric field for the modes with ck/Ω0 = 1.46 and 0.01 eV (orange line A), 0.03 eV (green line B), and 0.4 eV (red line C). These modes are indicated in panel (a) by the respective letters A, B and C. The re- gion occupied by the antiferromagnet is shadowed in panel (b) and the position of graphene is shown by vertical bold black solid line; (c) Group velocity, vg = dω/dk (in dimen- sionless units vg/c) of the SMPP modes with EF = 0.01 eV (orange line), 0.03 eV (green line), and 0.4 eV (red line). In all panels the fields and magnetization of the antiferromagnet are µ0Ha = 0.787 T, µ0He = 55.3 T, and µ0Ms = 0.756 T, for the antiferromagnet MnF2 [20]. The spacer between the antiferromagnet and graphene has thickness d = 500 nm. The magnitude of the fields was chosen arbitrarily for convenient visualization of their profiles. than the speed of light in vacuum, c. Even more, in short-wavelength limit ck/Ω0 (cid:38) 30 the group velocity is less than 10−5c, i.e. SMPPs are slow waves. Examples of spatial profiles of SMPP modes are shown in Fig. 3(b). As can be seen from the figure, in the case of TE-polarized wave Ey(0) > Ey(d), so the field is mainly concentrated nearby of the antiferromagnet surface. From the comparison of lines A, B, and C in Fig. 3(b) it is possible to conclude that higher graphene doping level leads to a stronger localization of the elec- tromagnetic field in the vicinity of the antiferromagnet surface. where M12 = M21 = 1 β(2) tanh(d2) 1 β(1) , + 1 1 β(2) + β(2) tanh(d2) − σ (ω) iωε0 . 9 (80) (81) sented as Consequently, the dispersion relation can be repre- M21M12 − (cid:0)β(2)(cid:1)2 1 sinh2(d2) = 0 . (82) B. Undressed SPP: the d → ∞ limit For infinite separation d → ∞ between the graphene monolayer and the antiferromagnet the dispersion rela- tion becomes:(cid:20) 2 (cid:21)(cid:20) 1 (cid:21) − σ (ω) iωε0 β(2) + 1 β(1) β(2) = 0. (83) Setting equal to zero the first term in brackets in Eq. (83) yields the dispersion relation of the SPPs in a free -- standing graphene monolayer, (cid:114) β(2) = k2 − ω2 c2 = 2iω0 σ(ω) . (84) This dispersion curve is shown in Fig.4(a) for EF = 0.03 eV by solid pink line (this small Fermi energy is chosen for clarity of the figure). At low frequencies the dispersion relation (84) can be expressed as ω = ck − 2 (ck)3 . (85) 8α2E2 F As a consequence, the dispersion curve (85) appears slightly below the light line. The second term in brackets in Eq. (83) is always pos- itive, so that it does not provide additional modes. This situation, however changes when d is finite. C. SPP dressed by the antiferromagnet Figure 4. (a,b) Dispersion relations of TM-polarized SPMP in the graphene-antiferromagnet structure with EF = 0.03 eV [solid blue lines in panel (a)], EF = 0.1 eV [solid orange lines in panel (b)], and EF = 0.3 eV [solid green lines in panel (b)]. For comparison, the SPP dispersion relation in bare graphene layer (with EF = 0.03 eV) is shown in panel (a) by solid pink line and the dashed black line corresponds to the bare photon, ω = ck, while the horizontal dash-and-dotted line corresponds to the AFMR frequency, ω = Ω0. (c) Spatial profiles of the electric fields, corresponding to SPMP modes with EF = 0.03 eV and ck/Ω0 = 1.18 (green line A and orange line C) and ck/Ω0 = 2.0 (blue line B and black line D). AF region is shadowed and graphene is at z = 0. The field pro- files are normalized to have the same magnitude on graphene. (d) Group velocity vg = dω/dk (in dimensionless units vg/c) of the SPMP low-frequency [ω−(k), solid lines] and high- frequency [ω+(k), dashed lines] modes with EF = 0.03 eV (blue lines), EF = 0.1 eV (orange lines), and EF = 0.4 eV (green lines). Other parameters of the structure are the same as those in Fig. 3. We now consider the effect of a finite value of d in Eq. (82). We plot the spectrum of this hybrid modes -- SPMPs [solid blue lines in Fig. 4]. We find that it consists of two branches ω−(k) and ω+(k) with an anti-crossing between them [see inset in Fig. 4], which takes place in the vicinity of the AFMR frequency, ω = Ω0. Away from the AFMR frequency ω ≶ Ω0, both modes follow the dispersion of the graphene SPP [like points A and D in Fig 4(a)]. In the vicinity of the antiferromagnetic resonance the lower mode approaches asymptotically the AFMR frequency Ω0 as k → ∞ [alike point B in Fig 4(a), i.e., ω−(∞) = Ω0], and the other one has the starting point of the spectrum at AFMR frequency, i.e. ω+(k+) = Ω0 [point C in Fig 4(a)]. The strong enhancement of the density of states of the SPP at the bare (non-polaritonic) AFMR frequency is a distinctive feature of the SPMPs, which are hybrid mode formed by the SPPs and the bare magnons. The magni- tude of the SPMP mode is quantified by: 1. Absence of the energy gap between two branches, since ω = Ω0 is both the maximum energy of the lower branch, and the minimum point of the upper branch; 2. Presence of the infinite gap in momentum space at AFMR frequency Ω0 = ω+(k+) = ω−(∞). This can be inferred by group velocities of upper and lower branches, shown in Fig. 4(d); 3. The asymmetry of the decay of the Ey component at two sides of the graphene sheet. The electromagnetic field is, for all modes, predomi- nantly concentrated nearby the graphene layer. The dis- tribution of the electric field depends strongly on the mo- mentum k and the SPMP branch. In Fig. 4(c) we show Ey(z), in units of Ey(d), for 4 different SPMP modes, la- belled with A,B,C and D, shown in Fig. 4(a). Modes A and D lie away from the anti-crossing of the branches and have a marked surface-plasmon character: their decay is the same at both sides (z < d and z > d) of graphene, and their profile almost does not change at the AF surface (z = 0). In contrast, modes B and C with frequencies nearby the AFMR frequency are asymmetric and change radically at the AF surface. The evolution of the group velocity of the two branches as a function of k, for three values of EF , shown in Fig. 4(d), shows very clearly that the hybrid SPMP modes are combining the dispersive graphene SPP with a non- dispersive mode with ω = Ω0. As the carrier density in graphene is varied, and thereby EF , SPMP dispersion is changed, and as a result, so is the value of k at which the anticrossing takes place. VI. DISCUSSION AND CONCLUSIONS In this work we have investigated the electromagnetic properties of an antiferromagnetic insulator in the prox- imity of a graphene sheet. We have found two new types of hybrid polaritons that combine the electromagnetic field with the magnetization in the magnetic material and the free carrier response of Dirac electrons in graphene: 1. A TE-polarized surface magnon-plasmon polariton (SMPP), propagating perpendicular to the direc- tion of staggerred magnetization in the AF. The group velocity of this mode becomes negative as EF is ramped up, resulting in a collective mode in the AF surface whose propagation direction can be steered upon gating the graphene layer, located at a distance of a few hundred nanometers away. 2. A TM-polarized surface plasmon-magnon polari- ton (SPMP), propagating along the staggerred magnetization direction, which hybdridizes the surface-plasmon polariton in graphene and the bare magnons at the AF. In both instances, a quantized theory of this new po- laritons implies a new type of hybrid collective modes that combine of surface plasmons in graphene, magnons in the antiferromagnets and the photon field. These new collective excitations have very exotic properties: 10 1. They are a mixture of spin excitations (magnon), charged excitations (plasmon) and electromagnetic field (photon). The first term in the names that we attributed to these hybrid polaritons indicate the condensed matter excitation that primarily inter- cats with the field; it also determines its polariza- tion (TE or TM). 2. They are extremely non-local, as they reside simul- taneously at the graphene and the AF surface. As a rule of thumb, the electromagnetic coupling be- tween these two layers survives as long as their sep- aration d is smaller than the wavelength of the EM field at the relevant frequencies. Therefore, it sur- vives to distances way above above 500 nanometers. 3. Their properties can be tuned by changing the car- rier density in graphene. The recently discovered two-dimensional magnetic ma- terials [23 -- 27] and the fabrication of Van der Waals heterostructures integrating 2D magnetic crystals with graphene and other non-magnetic 2D crystals [28 -- 34] opens up the possibility of observing the same effects dis- cussed in this work but in the context of van der Waals heterostructures. If the antiferromagnet is also metal- lic, then for frequencies below the plasma frequency, we would have a system exhibiting both negative permit- tivity and permeability functions. This material would present intrinsic negative refraction. In addition, the proximity of a graphene layer could be used for tuning the electromagnetic properties of the material. The prospects opened by 2D materials allow to envi- sion many different arrangement of these systems leading to a new class of metamaterials with tunable electromag- netic properties promoted by the existence of magnetic order. ACKNOWLEDGMENTS Y. B., M. V. and N. M. R. P. acknowledge sup- port from the European Commission through the project "Graphene- Driven Revolutions in ICT and Beyond" (Ref. No. 785219), and the Portuguese Foundation for Science and Technology (FCT) in the framework of the Strategic Financing UID/FIS/04650/2013. Ad- ditionally, N. M. R. P. acknowledges COMPETE2020, PORTUGAL2020, FEDER and the Portuguese Founda- tion for Science and Technology (FCT) through project PTDC/FIS-NAN/3668/2013 and FEDER and the por- tuguese Foundation for Science and Technology (FCT) through project POCI-01-0145-FEDER-028114. G. A. Farias acknowledge support from the Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq) of Brazil. J. F.-R. acknowledge financial support from FCT for the P2020-PTDC/FIS-NAN/4662/2014, the P2020-PTDC/FIS-NAN/3668/2014 and the UTAP- EXPL/NTec/0046/2017 projects, as well as Generali- 11 (cid:115) µ0µa (ω) (A9) ε0 tat Valenciana funding Prometeo2017/139 and MINECO Spain (Grant No. MAT2016-78625-C2). owing the Eq.(A3). The components of the respective electric field can be defined from Eq. (A1) Appendix A: Transverse bulk waves E = H Let us imagine that antiferromagnetic medium charac- terized by the magnetic permeability tensor (1) occupies all the space −∞ < z < ∞. In this case wave propa- gation is governed by Maxwell equations (7), which solu- tions we will seek in the form of travelling waves E(r, t) = E exp(ikr−iωt), H(r, t) = H exp(ikr−iωt), propagating in arbitrary direction k with amplitudes of electromag- netic field E, H. Under this assumption, jointly with the constitutive relations D(r, t) = ε0E exp(ikr − iωt), B(r, t) = µ0 µ (ω) H exp(ikr−iωt) Maxwell equations (7) will be rewritten as × [−ux cos ϕ cos θ + uy sin θ − uz sin ϕ cos θ] . Notice that in this representation the electric field is per- pendicular to magnetic field, E ⊥ H, what follows from the scalar product E · H = 0. Eq. (68) have two solution for ω -- acoustic ωa and op- tical ωo modes (cid:114) (cid:114) f (k) −(cid:113) (cid:113) f (k) + ωa = ωo = f (k)2 − c2k2Ω2 0, f (k)2 − c2k2Ω2 0, k × H = −ωε0E, k × E = ωµ0 µ (ω) H, ik · E = 0, ik · µ (ω) H = 0. with (A1) (A2) (A3) (A4) Notice that in this Appendix we omit index j for brevity. If we apply operator k× to Eq. (A1) and use Eq. (A2), we have k × (k × H) = k (k · H) − k2H = −(cid:16) ω (cid:17)2 c µ (ω) H. (A5) For the transverse waves the wavevector k should be or- thogonal to the magnetic field H, i.e. (k · H) = 0. Fur- ther this wave will be referred to as TM-polarized bulk polariton. Simultaneously with Eq. (A4) this condition can be satisfied only if Hy ≡ 0. In this case the Helmholtz equation (A5) for components of the magnetic field Hx and Hz will be rewritten as (cid:20) k2 −(cid:16) ω (cid:20) k2 −(cid:16) ω c (cid:17)2 (cid:17)2 c (cid:21) (cid:21) µ (ω) Hx = 0 = β1Hx, µ (ω) Hz = 0 = β1Hz. (A6) (A7) For nonzero amplitudes this system of equation will have solution only when condition β1 = 0 is met, thus Eq. (68) determines the dispersion relation of bulk waves. If the wavevector is represented in spherical coordinates as (cid:112)µ (ω) (ux cos ϕ sin θ + uy cos θ k = ω c +uz sin ϕ sin θ) , (A8) (cid:115) f (k) = c2k2 + Ω2 0 2 + Ω2 s. (A10) This is dissimilar to the case of a metal described by a Drude optical response, where only one transverse bulk mode exists. Spectra of these two bulk TM-polarized magnon-polariton modes are depicted in Fig. 2 by dashed red lines. The spectrum of the acoustic mode starts at zero frequency and in at long-wavelength limit k → 0 is described by the approximate expression as ωa ≈ kc Ω2 0 Ω2 0 + 2Ω2 s . (A11) In short-wavelength limit k → ∞ the dispersion curve of acoustic mode asymptotically approaches the frequency Ω0 as ωa ≈ Ω0 − Ω0Ω2 (ck)2 . s (A12) It should be underlined that the spectrum of the acoustic mode is located at the right of the light line ω = ck (depicted by dashed black line in Fig. 3). This fact means that phase velocity of the acoustic mode, ωa/k is smaller than the velocity of light in vacuum c for all values of the wavevector k. The optical mode spectrum starts at the frequency s, and in the limit k → 0 its approximate dis- 0 + 2Ω2 (cid:112)Ω2 persion relation can be represented as ωo ≈(cid:113) Ω2 0 + 2Ω2 s + c2k2Ω2 s 0 + 2Ω2 s)3/2 (Ω2 , (A13) the respective components of the magnetic field will be H = H (−ux sin ϕ + uz cos ϕ). In these equations θ is the polar angle between the y-axis and wavevector, and ϕ is the azimuthal angle in plane xz. The electric field is also perpendicular to the direction of the propagation while in the limit k → ∞ the optical mode's approximate dispersion relation is ωo ≈ kc + Ω2 s ck . (A14) Thus, at large values of wavevector k, the optical mode spectrum asymptotically approaches light line ω = ck. Contrary to the acoustic mode, the optical one is charac- terized by phase velocity larger than the velocity of light in vacuum c, and its spectrum is located at the left of the 12 light line. exists in the frequency range Ω0 < ω < (cid:112)Ω2 It is interesting that no bulk magnon polariton mode s -- between the highest frequency of the acoustic mode and lowest frequency of the optical mode. Notice, that this gap is characterized by the negative values of µa(ω) < 0. 0 + 2Ω2 [1] F. Keffer and C. Kittel, Physical Review 85, 329 (1952). [2] R. E. Camley and D. L. Mills, Phys. Rev. B 26, 1280 (1982). [3] M. R. F. Jensen, T. J. Parker, K. Abraha, and D. R. Tilley, Phys. Rev. Lett. 75, 3756 (1995). [4] W. J. Padilla, D. N. Basov, and D. R. Smith, Materials Today 9, 28 (2006). [5] M. Kaganov and V. Tsukernik, Journal of Experimental and Theoretical Physics 7, 361 (1958). [6] M. Kaganov and V. Tsukernik, Journal of Experimental and Theoretical Physics 14, 192 (1962). [7] D. L. Mills and E. Burstein, Rep. Prog. Phys. 37, 817 (1974). [8] N. S. Almeida, G. A. Farias, N. T. Oliveira, and E. F. Vasconcelos, Physical Review B 48, 9839 (1993). [9] R. H. Tarkhanyan and D. G. Niarchos, physica status solidi (b) 246, 1939 (2009). [11] A. Y. Nikitin, F. Guinea, F. J. García-Vidal, [10] F. H. L. Koppens, D. E. Chang, and F. J. García de Abajo, Nano Letters 11, 3370 (2011), arXiv:1104.2068. and L. Martín-Moreno, Physical Review B 84, 195446 (2011). [12] V. W. Brar, M. S. Jang, M. Sherrott, S. Kim, J. J. Lopez, L. B. Kim, M. Choi, and H. Atwater, Nano Letters 14, 3876 (2014). [13] Y. Liu and R. F. Willis, Physical Review B 81, 081406 (2010). [14] I. J. Luxmoore, C. H. Gan, P. Q. Liu, F. Valmorra, P. Li, J. Faist, and G. R. Nash, ACS Photonics 1, 1151 (2014). [15] 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 letters , 4701 (2011). [16] A. Kumar, T. Low, K. H. Fung, P. Avouris, and N. X. Fang, Nano letters 15, 3172 (2015). [17] A. Gutiérrez-Rubio, T. Stauber, and F. Guinea, J. Op- tics 15, 114005 (2013). [18] The inclusion of losses is straightforward but complicates the analysis and is, therefore, ignored. [19] If the interband transitions are taken into account in the expression of graphene's conductivity, its imaginary part can be negative. In this case graphene is able to sustain s-polarized surface plasmon-polaritons[? ], which disper- sion relation is described by the term in first braces of Eq. (64). Nevertheless, their typical frequencies lies in the vicinity of double graphene's Fermi energy ω ∼ 2EF , i.e. significantly higher than the antiferromagnet resonance frequency Ω0. [20] R. Macêdo, "Chapter two - tunable hyperbolic media: (Aca- Magnon-polaritons in canted antiferromagnets," demic Press, 2017) pp. 91 -- 155. [21] M. Jablan, H. Buljan, and M. Soljačić, Physical Review B 80 (2009), 10.1103/PhysRevB.80.245435. [22] Y. V. Bludov, A. Ferreira, N. M. R. Peres, and M. I. Vasilevskiy, International Journal of Modern Physics B 27, 1341001 (2013). [23] C. Gong, L. Li, Z. Li, H. Ji, A. Stern, Y. Xia, T. Cao, W. Bao, C. Wang, Y. Wang, Z. Q. Qiu, R. J. Cava, S. G. Louie, J. Xia, and X. Zhang, Nature 546, 265 (2017). [24] B. Huang, G. Clark, E. Navarro-Moratalla, D. R. Klein, R. Cheng, K. L. Seyler, D. Zhong, E. Schmidgall, M. A. McGuire, D. H. Cobden, W. Yao, D. Xiao, P. Jarillo- Herrero, and X. Xu, Nature 546, 270 (2017). [25] J. L. Miller, Physics Today 70, 16 (2017). [26] L. Hao, D. Meyers, H. Suwa, J. Yang, C. Frederick, T. R. Dasa, G. Fabbris, L. Horak, D. Kriegner, Y. Choi, J.- W. Kim, D. Haskel, P. J. Ryan, H. Xu, C. D. Batista, and J. Liu, Nature Physics 14, 806 M. P. M. Dean, (2018), arXiv:1804.08780. [27] Z. Fei, B. Huang, P. Malinowski, W. Wang, T. Song, J. Sanchez, W. Yao, D. Xiao, X. Zhu, A. May, W. Wu, D. Cobden, J.-H. Chu, and X. Xu, Nature materials 17, 778 (2018). [28] Z. Wang, I. Gutiérrez-Lezama, N. Ubrig, M. Kroner, M. Gibertini, T. Taniguchi, K. Watanabe, A. Imamoğlu, E. Giannini, and A. F. Morpurgo, Nature Communica- tions 9, 2516 (2018). [29] B. Huang, G. Clark, D. R. Klein, D. MacNeill, E. Navarro-Moratalla, K. L. Seyler, N. Wilson, M. A. McGuire, D. H. Cobden, D. Xiao, W. Yao, P. Jarillo- Herrero, and X. Xu, Nature Nanotechnology 13, 544 (2018). [30] D. R. Klein, D. MacNeill, J. L. Lado, D. Soriano, E. Navarro-Moratalla, K. Watanabe, T. Taniguchi, S. Manni, P. Canfield, J. Fernández-Rossier, and P. Jarillo-Herrero, Science 360, 1218 (2018). [31] T. Song, X. Cai, M. W.-Y. Tu, X. Zhang, B. Huang, N. P. Wilson, K. L. Seyler, L. Zhu, T. Taniguchi, K. Watanabe, M. A. McGuire, D. H. Cobden, D. Xiao, W. Yao, and X. Xu, Science 360, 1214 (2018). [32] S. Jiang, J. Shan, and K. F. Mak, Nature materials 17, 406 (2018). [33] S. Jiang, L. Li, Z. Wang, K. F. Mak, and J. Shan, Nature Nanotechnology 13, 549 (2018). [34] K. L. Seyler, D. Zhong, D. R. Klein, S. Gao, X. Zhang, B. Huang, E. Navarro-Moratalla, L. Yang, D. H. Cobden, M. A. McGuire, W. Yao, D. Xiao, P. Jarillo-Herrero, and X. Xu, Nature Physics 14, 277 (2018).
1507.05158	1	1507	2015-07-18T08:27:47	Observation of topological edge modes in bianisotropic metamaterials	[ "cond-mat.mes-hall", "cond-mat.mtrl-sci" ]	Existence of robust edge modes at interfaces of topologically dissimilar systems is one of the most fascinating manifestations of a novel nontrivial state of matter, topological insulators. Such electronic states were originally predicted and discovered in condensed matter physics, but they find their counterparts in other fields of physics, including the physics of classical waves and electromagnetics. Here, we present the first experimental realization of a topological insulator for electromagnetic waves based on engineered bianisotropic metamaterials. By employing the near-field scanning technique, we demonstrate experimentally the topologically robust propagation of electromagnetic waves around sharp corners without back reflection.	cond-mat.mes-hall	cond-mat	Observation of topological edge modes in bianisotropic metamaterials Alexey P. Slobozhanyuk1,4†, Alexander B. Khanikaev2,3†, Dmitry S. Filonov4, Daria A. Smirnova1, Andrey E. Miroshnichenko1, and Yuri S. Kivshar1,4* 1Nonlinear Physics Center, Australian National University, Canberra ACT 0200, Australia. 2Department of Physics, Queens College of City University of New York, NY 11367, USA. 3Department of Physics, Graduate Center of City University of New York, NY 10016, USA. 4ITMO University, St. Petersburg 197101, Russia *Correspondence to: [email protected] †These authors contributed equally to this work Abstract Existence of robust edge modes at interfaces of topologically dissimilar systems is one of the most fascinating manifestations of a novel nontrivial state of matter, topological insulators. Such electronic states were originally predicted and discovered in condensed matter physics, but they find their counterparts in other fields of physics, including the physics of classical waves and electromagnetics. Here, we present the first experimental realization of a topological insulator for electromagnetic waves based on engineered bianisotropic metamaterials. By employing the near-field scanning technique, we demonstrate experimentally the topologically robust propagation of electromagnetic waves around sharp corners without back reflection. Topologically nontrivial states of light1 represent a “Holy Grail” for optical applications usually suffering from undesirable backscattering and interference effects, thus dramatically limiting the bandwidth and performance of many photonic devices. Recently there were numerous theoretical predictions of topologically nontrivial states in different photonic systems ranging from microwave to optical frequencies2-10, but experimental demonstrations are very limited. In particular, the topological protection was shown experimentally for gyromagnetic photonic crystals12,13, arrays of coupled ring resonators14, chiral optical fibers15, and microwave waveguides16. In all such systems, different approaches were suggested to emulate an effective magnetic field, in a direct analogy with the familiar Quantum Hall Effect (QHE). However, another class of electromagnetic systems for which topological nontrivial states emulating Quantum Spin Hall Effect (QSHE)17-22 and robust helical edge modes have been predicted represents a bianisotropic metamaterial9,10 - a special class of synthetic periodic optical media with engineered magneto-electric response23-26. Here we suggest and demonstrate the first experimental realization of topologically nontrivial metamaterials which emulate directly the spin-orbit coupling in solids and exhibit the desirable property of topologically robust edge transport enabling guiding waves through sharp corners. The topological electromagnetic states realized in our system are based on two key ingredients: (i) “spin-degeneracy” enabling emulation of the electron spin, and (ii) bianisotropic gauge fields mimicking the spin-orbital coupling for electrons in solids. We consider a metacrystal with a square lattice of metamolecules with the shape of a ring and two symmetric slits and a wire placed in the middle. The schematic of the single metamolecule and the periodic arrangement of the metacrystal are shown in Fig. 1. The metamolecule of this geometry possesses two low-frequency magnetic and electric dipolar resonances with their field profiles shown in Fig. 1b. The magnetic resonance is formed by the antisymmetric charge distribution in the wires of the split-ring and with the inactive central wire (an upper panel in Fig. 1b). The magnetic dipole moment of this mode is oriented in the direction orthogonal to the plane of the metamolecule and originates from the currents counter-propagating in the wires. It is also easy to see that the net electric dipole moment of this mode vanishes due to cancellation of dipole moments of the individual antennas of the split-ring. In contrast, the electric dipolar resonance originates from the dipole moment of the central wire dressed up by the interaction with the wires of the split-ring. Figure 1 Structure and eigenmodes of the bianisotropic metamolecules and a metacrystal. a, Geometry of the metamolecules with bianisotropic response controllable by the asymmetry of and gaps. The radius (r), length of the wire (l), and periods (Px, Py) are equal to 2.85 mm, 0.5 mm, 7.5 mm, 13 mm and 12 mm respectively. Metal width for the central wire and split ring wires is 1 mm and 0.5 mm, respectively. b, Dipolar magnetic (top subplot) and dipolar electric (bottom panel) eigenmodes of the symmetric metamolecule ( ). c, Perspective view of the two-dimensional metacrystal formed by periodic stacking of the metamolecules. d, Photonic band structure of the metacrystal with ( ) and without ( ) bianisotropy shown by red and green markers, respectively. The right subplot shows enlarged region near the topological band gap. The yellow shaded area illustrates the spectral width of the gap. The substrate with the permittivity equal to is taken into account. Light line position is indicated by thin black dashed lines. This resonance appears to be shifted significantly towards the subwavelength range thanks to the hybridization with the split-ring by inducing symmetric currents distribution in the wires. The latter feature allows us to satisfy the spin-degeneracy condition as the geometrical parameters of the metamolecule can be tuned to ensure that the electric and magnetic dipolar resonances occur at the same frequency. As the next step, we form a metacrystal (see Fig. 1c) by arranging the metamolecules into a square two-dimensional array with only one layer in vertical direction. The open geometry considered here represents a special interest due to possible leakage of topological modes caused by the out-of-plane scattering into the radiative continuum. The square lattice of the metacrystal with in-plane and z-inversion symmetries possesses quadratic degeneracies at high symmetry and points in the Brillouin zone4,12,27, which is confirmed for both electric and magnetic dipolar modes of the metacrystal by first principle finite element method simulations (see Methods for details), and are shown in Fig. 1d (green dotted line). Here we are interested in guided waves exponentially confined to the structure in vertical (z-) direction and whose bands are located below the light cone (dashed straight lines in Fig. 1d), and, therefore, in what follows we focus on the quadratic degeneracies taking place at M points. Note, that to ensure the condition of spin-degeneracy at M-point, e.g. the two doubly-degenerate quadratic bands stemming from magnetic and electric resonances appear at the same frequency, we fine-tuned the geometry of the metamolecules to compensate for the effects of slightly different interaction between magnetic and electric dipoles in the array. The resultant band structure (shown in Fig. 1d) confirms the presence of overlaid electric-like and magnetic-like collective modes with quadratic degeneracy at the M point. The presence of the quadratic degeneracies is crucial for topological protection in the metamaterial under study; to illustrate this, here we first apply the analytic effective Hamiltonian description3,9,27. The band structure of the metacrystal in the proximity of the M-points can be described by the effective quadratic Hamiltonian , acting on a four component wavefunction , with its components being in-plane electric and magnetic dipole moments, and subscripts and indicating the electric and magnetic blocks, respectively, which are described by the expression27: . (1) Here are the Pauli matrices, is the deviation of the Bloch wavevector from the M-point, and and are the effective parameters of the model, which in general differ for electric and magnetic modes. However, in our design the dispersion of the electric and magnetic modes has been tuned to match near the M-point so that these parameters are, to a good approximation, are equal for both types of modes. This degeneracy is crucially important as it allows choosing a new set eigenmodes as any linear combination of electric and magnetic components. One of the key ingredients for engineering the topological state in the metacrystal represents the ability to emulate the spin degree of freedom of electrons in QSHE, which must be odd under time reversal (TR) operation. By noticing that electric and magnetic moments transform differently under TR and , it was shown that an appropriate choice of basis is given by the wavefunction , where and are the desirable pseudo-spin components which are time-reversal partners of each other9. Next, to endow the metacrystal with topological properties, we introduce an effective gauge field emulating the spin-orbital coupling, which is achieved by introducing bianisotropy mixing magnetic and electric degrees of freedom of the system9. The desirable form of bianisotropy should couple x-(y-) magnetic dipole moment with y-(x-) electric dipole moment of the metamolecule, and is realized by breaking its z-inversion symmetry23. Figure 1d (red dotted line) shows the band structure calculated by the FEM for the case of distorted metamolecules, where the central wire was displaced vertically between the split ring wires (so that gaps and are no longer equal), and reveals a gap that is complete below the light cone. In the context of the effective Hamiltonian description the effect of the bianisotropy induced by this symmetry reduction can be described by an spin-orbital-like potential 9, which results in opening of the photonic band gap and transition from topologically trivial to topological nontrivial state (see Supplementary Information for details). Figure 2 Experimental system and the observation of topological edge modes. a, Photograph of the fabricated metacrystal with the location of the double-bend domain wall indicated by the white dashed line. b, Transmission spectra of the metacrystal away from the domain wall (blue line) and along the domain wall (red line). c, Two- dimensional map of the magnetic field intensity indicating reflection-less propagation along the domain wall. d, Normalized magnetic field intensity along the domain wall measured for different frequencies (8.9 GHz - blue line; 8.95 GHz - green line; 8.987 GHz - red line) within the topological band gap. Appearance of topologically robust helical edge modes insensitive to local structural imperfections and avoiding backscattering represents a hallmark of topological photonic order emulating QSHE8,9,28. In the case of bianisotropic metacrystals9, the edge states have been predicted to occur at the “domain walls” representing boundaries between regions of the metacrystal with the reversed bianisotropy. The presence of such modes was confirmed by calculating the photonic band structure in the effective Hamiltonian model, as described in the Supplementary Information. From the bulk-boundary correspondence, for such a domain wall, one expects two edge modes for every spin, in accordance with the change in the spin-Chern number between the domains (ref. 9 and Supplementary Information). It is worth mentioning that one could also expect to observe the edge modes on the external boundary of the metacrystal with the air, however, in contrast to electronic systems, the external domain does not possesses a bandgap in the spectrum and is filled with the electromagnetic continuum. This makes such modes either leaky or unstable to any perturbation which couples them to the radiative continuum of the free space, thus breaking the topological protection. To verify these theoretical predictions and confirm topological robustness of the helical edge states, a metacrystal was fabricated by printing an array of metamolecules on the a 1 mm- thick dielectric board (Arlon 25N) with the dielectric permittivity . The board was cut into linear segments which were stacked to form a square lattice, as shown in Fig. 2a. To test the topological protection, which leads to the ability of the edge modes to propagate around sharp bends without back scattering9, the zigzag-shaped domain wall was created by deliberate distribution of metamolecules with the central wires shifted vertically up and down across the crystal, as indicated by the dashed white line in Fig.2a. First, the presence of the topological band gap was verified by exciting bulk modes with the dipole source placed away from the domain wall. Figure 2b clearly reveals the gap spanning frequency range from 8.6 GHz to 9.3GHz. Next, the presence of the topological edge mode was tested by placing the dipole at the domain wall and the transmission spectrum was measured. As expected, the enhanced transmission within the band gap region occurred due to the excitation of the edge mode, as shown in Fig. 2b. Note that the transmission was especially high near the gap center and gradually dropped towards the edges of the bands. To confirm that the transmission indeed occurs due to the edge mode localized to the domain wall, the near-field map of magnetic field intensity was measured across the entire sample with the use of the probe (magnetic loop antenna) mounted on the two-dimensional near- field scanning stage. The map shown in Fig 2c clearly shows that the mode excited at the frequency of GHz is indeed guided by the domain wall. In agreement with the theoretical predictions, the mode appears to be strongly localized to the wall. In addition, quite surprisingly we have not noticed any apparent effects of leakage neither to the modes above the light cone nor to the radiative continuum. Moreover, we observed that regardless of the open character of the system, the edge modes do not scatter to these modes even when they encounter sharp bends of the domain wall, and the wave flawlessly transfers between orthogonal segments of the wall. To ensure the robustness across the entire frequency range of the topological bandgap, the same measurements have been conducted for multiple frequencies. As expected for the frequencies close to the spectral edges of the bulk modes, the edge states become poorly localized and are hardly distinguishable from the bulk modes. However, as can be seen from Fig. 2d, the edge states with frequencies sufficiently apart from the bulk spectrum remains well localized and exhibits similar robustness against sharp bends of the domain wall at all frequencies. It is also important to mention that the drop in the transmission observed in Fig. 2b away from the center of the topological band gap lacks back-reflection along the domain wall (observed in Fig. 2d), which indicates that the low transmission near the edges is associated with the insertion loss. This can be explained as the consequence of increasingly poor overlap of the field produced by the dipole antenna with the field profile of the edge mode which spread over the bulk as we approach the band edges. Also note that in Fig. 2d some variations in the field intensities occur due to the local interference effects, which, however, do not lead to backscattering, but rather assist total transmission through the zigzag, as is evidenced by the rebound of the intensity away from the 90o-bends of the wall. Finally, it’s important to mention that according to our measurements the finite lateral size of the sample did not play any detrimental role and no any noticeable backscattering of the edge modes occurred at the metacrystal boundaries, where the edge modes were efficiently radiated into the propagating continuum of the free space. Topological robustness of electromagnetic helical edge-states in an open metamaterial system demonstrated here for the first time can be of significant interest both from fundamental and applied perspectives. First, it confirms the possibility to emulate exotic quantum states of solids in open metamaterial structures thus allowing direct near-field mapping of the local amplitudes and phases of topological waves. This enables the study of their topological characteristic and peculiarities in the real coordinate space as well as in the reciprocal space by performing Furrier transformation of the measured field maps. Second, the possibility to guide electromagnetic energy around arbitrarily shaped pathways avoiding undesirable back-reflection from sharp bends and without leakage into the radiative continuum brings the versatility of control of the electromagnetic energy flows in engineered photonic systems to an unprecedented level. It is only a matter of time when the full potential of topological photonic states will be exploited to study novel fundamental electromagnetic phenomena and in applied systems and devices with their physics and functionality solely relaying on the presence of the photonic topological order. Methods Simulations. All numerical results for photonic band structures were obtained by performing finite-element-method calculations in COMSOL Multiphysics (Radio Frequency module). We have assumed perfectly conducting elements (split ring and wire) on the dielectric substrate. The periodic boundary conditions were imposed in x and y directions to form an infinite square lattice (shown in Fig. 1c). The perfect matched layers were added to the domain in order to prevent back reflections from z direction. The mesh was optimized in order to reach the convergence. Experiments. The metamolecules were fabricated on the dielectric board (Arlon 25N) using chemical etching technique. Subsequently, the board was cut into linear segments which were staked to form a square lattice. Special mask made from styrofoam material with the dielectric permittivity of 1 was used to control the periodicity of the segments separation. All the measurements were performed in the anechoic chamber. We utilized subwavelength dipole as a source, which was connected to the transmitting port of a vector network analyser (Agilent E8362C). For measurement of the transmission spectra shown in Fig. 2b a similar dipole antenna was used as a receiver. To perform the near field measurements we used an automatic mechanical near-field scanning setup and a magnetic field probe connected to the receiving port of the analyser (Fig.2c). The probe was oriented normally with respect to the interface of the structure. The near field was scanned at the 1 mm distance from the back interface of the metacrystal to avoid a direct contact between the probe and the sample. Acknowledgements This work was supported by the Australian Research Council, USA Air Force Research Laboratory, the Government of the Russian Federation (Grant 074-U01), and the Dynasty Foundation (Russia). APS acknowledges a support of the SPIE scholarship. The authors are grateful to Dr. Alexander Poddubny and Dr. Ilya Shadrivov for useful discussions and suggestions. References 1. Lu, L., Joannopoulos, J. D. & Soljačić, M. Topological photonics. Nat. Photon. 8, 821-829 (2014). 2. Raghu, S. & Haldane, F. D. M. Analogs of quantum-Hall-effect edge states in photonic crystals. Phys. Rev. A 78, 033834 (2008). 3. Haldane, F. & Raghu, S. Possible realization of directional optical waveguides in photonic crystals with broken time-reversal symmetry. Phys. Rev. Lett. 100, 013904 (2008). 4. Wang, Z., Chong, Y., Joannopoulos, J. & Soljačić, M. Reflection-free one-way edge modes in a gyromagnetic photonic crystal. Phys. Rev. Lett. 100, 13905 (2008). 5. Fang, K., Yu, Z. & Fan, S. Microscopic theory of photonic one-way edge mode. Phys. Rev. B 84, 075477 (2011). 6. Fang, K., Yu, Z. & Fan, S. Realizing effective magnetic field for photons by controlling the phase of dynamic modulation. Nat. Photon. 6, 782–787 (2012). 7. Umucalılar, R. O. & Carusotto, I. Artificial gauge field for photons in coupled cavity arrays. Phys. Rev. A 84, 043804 (2011). 8. Hafezi, M., Demler, E. A., Lukin, M. D. & Taylor, J. M. Robust optical delay lines with topological protection. Nat. Phys. 7, 907–912 (2011). 9. Khanikaev, A. B. et al., Photonic topological insulators. Nat. Mater. 12, 233–239 (2013). 10. Ma, T., Khanikaev, A. B., Mousavi, S. H. & Shvets, G. Phys. Rev. Lett. 114, 127401 (2015). 11. Lu, L., Fu, L., Joannopoulos, J. D. & Soljačić, M. Weyl points and line nodes in gyroid photonic crystals. Nat. Photon. 7, 294-299 (2013). 12. Wang, Z., Chong, Y., Joannopoulos, J. D. & Soljačić, M. Observation of unidirectional backscattering-immune topological electromagnetic states. Nature 461, 772-775 (2009). 13. Poo, Y., Wu, R., Lin, Z., Yang, Y. & Chan, C. T. Experimental Realization of Self-Guiding Unidirectional Electromagnetic Edge States. Phys. Rev. Lett. 106, 093903 (2011). 14. Hafezi, M., Mittal, S., Fan, J., Migdall, A. & Taylor, J. M. Imaging topological edge states in silicon photonics. Nat. Photon. 7, 1001–1005 (2013). 15. Rechtsman, M. C. et al., Photonic Floquet topological insulators. Nature 496, 196–200 (2013). 16. Chen, W.-J. et al., Experimental realization of photonic topological insulator in a uniaxial metacrystal waveguide. Nat. Commum. 5, 6782:1-7 (2014). 17. Kane, C. L. & Mele, E. J. Quantum Spin Hall Effect in Graphene. Phys. Rev. Lett. 95, 226801 (2005). 18. Kane, C. L. & Mele, E. J. Z2 Topological Order and the Quantum Spin Hall Effect. Phys. Rev. Lett. 95, 146802 (2005). 19. Bernevig, B. A., Hughes, T. L., Zhang, S. C. Quantum spin Hall effect and topological phase transition in HgTe quantum wells. Science 314, 1757-1761 (2006). 20. König, M. et al., Quantum spin Hall insulator state in HgTe quantum wells, Science 318, 766 (2007). 21. Hasan, M. Z. & Kane, C. L. Colloquium: Topological insulators. Rev. Mod. Phys. 82, 3045- 3067 (2010). 22. Qi, X.-L. & Zhang, S. C. Topological insulators and superconductors. Rev. Mod. Phys. 83, 1057-1110 (2011). 23. Serdyukov, A. N., Semchenko, I. V., Tretyakov, S. A. & Sihvola, A. Electromagnetics of bi- anisotropic materials: Theory and applications (Gordon and Breach Science Publishers, Amsterdam, 2001). 24. Linden, S. et al., Magnetic Response of Metamaterials at 100 Terahertz. Science 306 (5700), 1351-1353 (2004). 25. Markos, O. & Soukoulis, C. M. Wave Propagation: From Electrons to Photonic Crystals and Left-Handed Materials (Princeton University Press, Princeton, 2008). 26. Cai, W. & Shalaev, V. Optical Metamaterials: Fundamentals and Applications (Springer, New York, 2009). 27. Chong, Y. D., Wen, X.-G. & Soljačić, M. Effective theory of quadratic degeneracies. Phys. Rev. B 77, 235125 (2008). 28. Mittal, S. et al., Topologically Robust Transport of Photons in a Synthetic Gauge Field. Phys. Rev. Lett. 113, 087403 (2014).
1511.04525	1	1511	2015-11-14T07:47:31	Anomalous Josephson Hall effect in magnet/triplet superconductor junctions	[ "cond-mat.mes-hall", "cond-mat.supr-con" ]	We investigate anomalous Hall effect in a magnet coupled to a triplet superconductor under phase gradient. It is found that the anomalous Hall supercurrent arises from non-trivial structure of the magnetization. The magnetic structure manifested in the Hall supercurrent is characterized by even order terms of the exchange coupling, essentially different from that discussed in the context of anomalous Hall effect, reflecting the disspationless nature of supercurrent. We also discuss a possible candidate for magnetic structure to verify our prediction.	cond-mat.mes-hall	cond-mat	a Anomalous Josephson Hall effect in magnet/triplet superconductor junctions Department of Physics, Tokyo Institute of Technology, Tokyo 152-8551, Japan (Dated: August 25, 2018) Takehito Yokoyama We investigate anomalous Hall effect in a magnet coupled to a triplet superconductor under phase gradient. It is found that the anomalous Hall supercurrent arises from non-trivial structure of the magnetization. The magnetic structure manifested in the Hall supercurrent is characterized by even order terms of the exchange coupling, essentially different from that discussed in the context of anomalous Hall effect, reflecting the disspationless nature of supercurrent. We also discuss a possible candidate for magnetic structure to verify our prediction. PACS numbers: 73.43.Nq, 72.25.Dc, 85.75.-d Recently, the coexistence of superconductivity and ferromagnetism has received much attention.1 -- 4 In fer- romagnet/superconductor junctions, equal-spin triplet pairings can be generated due to spin flip scat- tering in ferromagnetic multilayer or inhomogeneous ferromagnet5, or in uniform ferromagnet with spin- orbit coupling6. Spin-polarized supercurrent carried by equal-spin triplet pairings is a novel ingredient for spintronics applications. Recent experiments have suc- cessfully demonstrated the generation of spin-triplet pairing by observing Josephson current through strong ferromagnets.7 -- 9 Up to now, in ferromagnetic Joseph- son junctions, longitudinal Josephson current has been investigated.10,11 Ferromagnet/triplet superconductor junctions offer richer physics than singlet ones since triplet pairings have spin degree of freedom. The interaction between mag- netism and triplet superconductivity leads to interest- ing phenomena such as 0-π transitions.12 -- 17 Josephson junctions composed of triplet superconductors have been fabricated to identify the pairing symmetry of triplet superconductors.18 -- 20 Also, a highly conducting interface between a ferromagnet SrRuO3 and a triplet supercon- ductor Sr2RuO4 has been realized recently.21 The Hall effect in ferromagnets has been discussed in- tensively in the context of anomalous Hall effect.22 The anomalous Hall effect arises from non-trivial spin texture, which is associated with the spin Berry phase effect.23 -- 27 It has been shown that the Hall conductivity stems from non-trivial spin configurations such as vector spin chiral- 28 and scalar spin chirality Si · (Sj × Sk)26, ity Si × Sj where Si is a localized spin with position i. Motivated by these studies, in this paper, we investigate Hall su- percurrent driven by non-trivial magnetic structure cou- pled to triplet superconductivity under phase gradient. Since phase is odd in time-reversal, the magnetic struc- ture manifested in Hall supercurrent becomes essentially different from that in the anomalous Hall effect. In this paper, we study the anomalous Hall effect in a magnet coupled to a triplet superconductor under phase gradient. It is shown that the anomalous Hall supercur- rent arises from non-trivial structure of the magnetiza- tion. The magnetic structure manifested in the Hall su- percurrent is essentially different from that discussed in the context of anomalous Hall effect, reflecting the diss- pationless nature of supercurrent. We also discuss the condition of the generation of the anomalous Josephson Hall current based on symmetry and a possible candidate for magnetic structure to verify our prediction. We consider a magnet/triplet superconductor junction (See Fig. 1). The Hamiltonian of the superconduc- tor and the magnet are given by HS = H0 + H∆ and HM = H0 + Hex + Hϕ, respectively. The H0, H∆ and Hex represent the kinetic energy, the superconducting or- der parameter, and the exchange interaction between the conducting electrons and the local spins, respectively: φ† kξσ0⊗τ3φk, φ† k(d · σ) ⊗ τ1, H0 =Xk H∆ =Xk Hex = −JXk,q (φ† k−q σ ⊗ τ0φk) · Sq (1) (2) (3) k = (c† k↑, c† k↓, ic−k↓, −ic−k↑) and ξ = ¯h2k2 with φ† 2m − εF . Here, σ and τ denote the Pauli matrices in spin and Nambu spaces, respectively. Also, εF , d, J, and S are the Fermi energy, d-vector of the triplet pairing, the ex- change coupling, and the localized spins, respectively. The localized spins can have spatial dependence, but we consider only slowly varying case. Note that we adopt the basis in Ref.29 such that singlet pairing is proportional to the unit matrix in spin space. We consider supercurrent induced by phase gradient. The phase gradient along j direction, ∇jϕ, enters the Hamiltonian as φ† k ¯h2 m Hϕ =Xk kj∇jϕφk (4) where ∇jϕ is assumed to be spatially constant. We will treat Hex and Hϕ perturbatively. With the above Hamiltonians, the charge current op- erator (jc) in i-direction is given by jc,i = − e¯h m kiσ0 ⊗ τ0 − δij e¯h m ∇j ϕσ0 ⊗ τ3 (5) where −e is the electron charge. j DĂŐŶĞƚ dƌŝƉůĞƚƐƵƉĞƌĐŽŶĚƵĐƚŽƌ 2 0,k,ω and ga where ga 3,k,ω are normal Green's functions while f a k,ω is a 3D vector characterizing anomalous Green's function. The anomalous Green's function in the magnet arises due to the proximity effect. We take into account Hex up to second order and Hϕ as a first order perturbation. Diagrammatic representations of the Hall supercurrent are shown in Fig. 2. FIG. 1: net/triplet superconductor junction. (Color online) Schematic illustration of mag- Before proceeding to the explicit calculation, let us dis- cuss Josephson Hall current qualitatively based on sym- metry argument.30 Consider the London equation of the form: jc = − e2 m ρ · A. (6) where jc, ρ, and A are, respectively, the charge cur- rent, the superfluid density tensor, and the vector po- tential. Since charge current and vector potential are time-reversal odd, ρ describes the reversible and dissi- pationless flow of the supercurrent. Thus, the Hall cur- rent can flow without breaking time-reversal symmetry. Namely, the Hall current is allowed in even-order pertur- bation with respect to time-reversal breaking term Hex. This contrasts with the anomalous Hall effect26 where Hall current is driven by an electric field which is even under time-reversal. Thus, one can expect an essentially different magnetic structure manifested in the Hall su- percurrent. Also, since charge current and vector poten- tial are odd under spatial inversion, the Hall coefficient should be even with respect to spatial gradient. There- fore, in the lowest order, the possible form of the Hall coefficient should be ∇iS × ∇j S with i 6= j. However, this is a vector quantity while the Hall coefficient is a scalar. Here, triplet superconductivity plays an essential role: We have a vector formed by triplet pairing (such as d-vector). This way, we expect that the Hall coefficient is proportional to ∇iS × ∇j S projected on some vector due to triplet superconductivity. Below, we will show that this consideration is in fact correct by the explicit calculation of the Hall supercurrent. Now, we calculate Hall supercurrent and give their an- alytical expressions. We consider the unperturbed ad- vanced Green's functions in the magnet of the form k,ω = ga ga 0,k,ωσ0 ⊗ τ0 + ga 3,k,ωσ0 ⊗ τ3 + (f a k,ω · σ) ⊗ τ1 (7) ( )a cj ( )c cj ( )b cj (d) FIG. 2: Diagrammatic representations of the current den- sity. The wavy lines denote the interaction with the local spin S and dotted lines represent the phase gradient ∇ϕ. (a- c) The contributions from the first term in Eq.(5). (d) The contribution from the second term in Eq.(5). The Hall supercurrent can be represented as31 jc,i = +δij i¯h2e mV i¯h2e mV Xk,q ∇jϕXk,q e−iq·xTrkiG< k−q/2,k+q/2(t, t) e−iq·xTrτ3G< k−q/2,k+q/2(t, t) (8) where V is the total volume and Tr is taken over spin and Nambu spaces. G< k−q/2,k+q/2(t, t) is the lesser Green's function of the total Hamiltonian. Performing perturbations with respect to Hex and Hϕ, we expand the lesser component using the advanced Green's func- tions by the Langreth theorem.31 Noting that g< k,ω = k,ω − (ga fωhga k,ω)†i with the lesser Green's function g< k,ω and the Fermi distribution function fω, and δij = ∂ki , ∂kj we can compute the Hall supercurrent. Then, the Hall supercurrent (i 6= j) in the second orders in J and spatial derivative is given by (cid:209) j (cid:209) j 3 ×Im  2kikj(cid:16) ∂ +kin(ga 0,k,ω)2 + (ga ∂ki n(ga 0,k,ω)2 + (ga 3,k,ω)2o(cid:16)f a k,ω × ∂ ∂ki 3,k,ω)2o(cid:16)f a k,ω × ∂ ∂kj f a k,ω(cid:17)α f a k,ω(cid:17)α − ∂ ∂kj n(ga jc,i = 16¯h4eJ 2 m2V ∇jϕ Xk,ω,α 3,k,ω)2o(cid:16)f a 0,k,ω)2 + (ga k,ω × ∂ ∂ki (∇j S(x) × ∇iS(x))αfω f a k,ω(cid:17)α(cid:17)   . (9) Here, we have assumed that the proximity effect is weak: (cid:12)(cid:12)(cid:12) , and dropped the forth order terms in the anomalous Green's functions. We have also dropped the symmetric term ∇j S(x) · ∇iS(x) since we are interested in the Hall effect. If the anomalous Green's function f a k,ω becomes zero, then the supercurrent becomes zero as expected. Assuming that the proximity effect is weak, we now consider the Green's functions of the form: ga 0,k,ω + ga k,ωdk. Here, γ is the scattering rate by impurities. 0,k,ω(cid:12)(cid:12)(cid:12) 3,k,ω(cid:12)(cid:12)(cid:12) ω−iγ−ξ ≡ ga ω−iγ+ξ ≡ ¯ga ≪ (cid:12)(cid:12)(cid:12) k,ω, and f a k,ω(cid:12)(cid:12)(cid:12) 0,k,ω − ga k,ω = ga k,ω, ga 3,k,ω = 3,k,ω = ,(cid:12)(cid:12)(cid:12) k,ω¯ga ga ga f a 1 1 Then, we have jc,i = 8¯h4eJ 2 m2V (∇j S(x) × ∇iS(x))αfω ∇jϕ Xk,ω,α ×Im  4¯h2 m k2 − 4¯h2 i kjn(ga j n(ga k,ω)5(¯ga k,ω)5(¯ga k,ω)2 + (ga k,ω)2 − (ga k,ω)2 − (ga k,ω)2(¯ga k,ω)4(¯ga m kik2 +kin(ga ∂kj k,ω)2(¯ga k,ω)2(¯ga k,ω)5o(cid:16)dk × ∂ k,ω)5o(cid:16)dk × ∂ dk(cid:17)α k,ω)4o(cid:16)dk × ∂ ∂ki ∂ki dk(cid:17)α dk(cid:17)α .   (10) As an example of the d-vector, consider the form of the Rashba type with the hexagonal warping:38 dk = ∆0(cid:0) ky/kF , −kx/kF , λ(k3 x − 3kxk2 y)/k3 where kF is the Fermi wavelength. Then, under phase gradient in x-direction, we obtain the Hall supercurrent in y-direction driven by magnetic structure: F (cid:1) (11) Now, we discuss a possible candidate of the magnetic structure that may be used to verify our prediction. First, the magnetization vector S(x) should have both x and y dependences. To observe finite Hall supercur- rent, ∇xS(x) and ∇yS(x) should not be parallel to each other. One possible candidate is a magnetic skyrmion in chiral magnets33 -- 37, such as MnSi, characterized by jc,y = − 38eJ 2∆2 105¯h4k4 0λν F ε2 F (∇xS(x) × ∇yS(x))x∇xϕ (12) S(x) = S x2 + y2 + Λ2 (cid:0) −2Λx, 2Λy, x2 + y2 − Λ2 (cid:1)(14) in the limit of γ → 0. Here, ν is the density of states at the Fermi level. This is in contrast to the normal Hall current in ferro- magnet: In the normal state, the Hall current is driven by scalar spin chirality under electric field26 jc,y ∝ (∇xS(x) × ∇yS(x)) · S(x). (13) where Λ determines the size of the skyrmion. For this spin texture, we have (∇xS(x) × ∇yS(x))x = 8S2Λ3y (x2 + y2 + Λ2)3 . (15) By comparing Eq.(12) and Eq.(13), we find an essentially different magnetic structure of Hall supercurrent, which reflects that supercurrent flows in response to phase gra- dient, i.e. reflects the disspationless nature of supercur- rent. Note that the normal Hall current is driven in an electric field which is even under time reversal. Hence, the Hall coefficient in the normal state is composed of odd order terms with respect to S. Since this is odd in y, the total Hall supercurrent vanishes for infinite systems or when the skyrmion is located in symmetric points with respect to y-coordinate in finite systems. Note, however, that the skyrmion can be moved by injecting ultralow current.35,37 For example, consider the skyrmion located in the region −∞ < x < ∞, −a ≤ y ≤ b. Then, the total Hall supercurrent along the edges is proportional to Z ∞ −∞ dx (∇xS(x) × ∇yS(x))xy=b + (∇xS(x) × ∇yS(x))xy=−a = 32S2Λ3 3 " b (b2 + Λ2)5/2 − a (a2 + Λ2)5/2# (16) 4 b4 − 1 (cid:0) 1 a4(cid:1) for Λ ≪ a, b. which becomes 32S 2Λ3 3 We also discuss possible forms of the d-vector and magnetic structures from the symmetry of the Hamil- tonian. Consider quasi two dimensional systems in xy- plane, focusing on the magnetic region with proximity induced triplet superconductivity, and the current along the y-direction under the phase gradient along the x-axis. Since the current density is transformed as jc,y(x, y) → −jc,y(−x, y) under the operation of C2yT , where T de- notes time-reversal, if the Hamiltonian is invariant under this operation, jc,y is an odd function of x. Thus, if the Hamiltonian has this symmetry, the net Hall supercur- rent becomes zero. This symmetry requires the condition dx(kx, ky) = d∗ y(kx, −ky), and dz(kx, ky) = d∗ z(kx, −ky) where dk = (dx, dy, dz). Therefore, this symmetry is broken in the d-vector of Eq.(11), which is consistent with the above results of the Hall supercurrent. The skyrmion magnetic structure in Eq.(14) respects this symmetry. Thus, if we consider an s-wave singlet pairing instead of triplet one, since the singlet pairing respects the symmetry, the Hall supper- current vanishes. x(kx, −ky), dy(kx, ky) = −d∗ Let us estimate the Hall supercurrent using Eq. (12). For εF ∼ 10 eV, J ∼ 100 meV, ν ∼ 0.1 /eV/unit cell, ∆0 ∼ 0.01 meV, λ ∼ 0.1, kF ∼ 1010m−1, ∇xS(x) × ∇yS(x) ∼ 1016m−2, ∇xϕ ∼ (100 nm)−1, and the lattice constant ∼ 5 A, we estimate the magnitude of the current as jc,y ∼ 5 × 105 A/cm2 which is of measur- able magnitude39. In this paper, we have considered magnet/triplet su- perconductor junctions. Since ferromagnets with intrin- sic spin-orbit coupling can generate triplet pairings,6 one can instead consider magnet/singlet superconductor junctions with intrinsic spin-orbit coupling in the mag- net. Also, since inhomogeneous magnetism can convert singlet pairing into triplet one,5 magnet/magnet/singlet superconductor junctions such as MnSi/Ho/Nb junctions (Ho is a conical magnet) can be used. One may also consider MnSi/Bi2Te3 film/Nb junctions. Here, due to the proximity effect, triplet pairing like Eq.(11) can be induced in the Bi2Te3 film. This induced pairing can further penetrate into MnSi. Spin Hall effect due to Rashba type spin-orbit coupling in superconductor40 or Josephson junctions41 has been discussed. In this paper, we have predicted Hall super- current driven by non-trivial magnetic texture, and hence our results do not rely on spin-orbit coupling. In Ref.41, spin Hall effect is obtained by applying electric bias to the Josephson junction in order to make the Josephson cur- rent time dependent. In sharp contrast, we have consid- ered stationary supercurrent under non-trivial magnetic structure when phase gradient is applied. In summary, we have investigated anomalous Hall ef- fect in a magnet coupled to a triplet superconductor un- der phase gradient. It is found that the anomalous Hall supercurrent arises from non-trivial structure of the mag- netization. The magnetic structure manifested in the Hall supercurrent is essentially different from that dis- cussed in the context of anomalous Hall effect, reflecting the disspationless nature of supercurrent. We have also discussed the condition of the generation of the anoma- lous Josephson Hall current based on symmetry and a possible candidate of magnetic structure to verify our prediction. 1 A. I. Buzdin, Rev. Mod. Phys. 77, 935 (2005). 2 F. S. Bergeret, A. F. Volkov, and K. B. Efetov, Rev. Mod. Phys. 77, 1321 (2005). 3 M. Eschrig, Physics Today 64, 43 (2011). 4 J. Linder and J. W. A. Robinson, Nat. Phys. 11, 307 (2015). 5 F. S. Bergeret, A. F. Volkov, and K. B. Efetov, Phys. Rev. Lett. 86, 4096 (2001); Phys. Rev. B 64, 134506 (2001). 6 F. S. Bergeret and I. V. Tokatly, Phys. Rev. Lett. 110, 117003 (2013); F. S. Bergeret, and I. V. Tokatly, Phys. Rev. B 89, 134517 (2014). 7 R. S. Keizer, S. T. B. Goennenwein, T. M. Klapwijk, G. Miao, G. Xiao, and A. Gupta, Nature (London) 439, 825 (2006). 8 T. S. Khaire, M. A. Khasawneh, W. P. Pratt, Jr., and N. 9 J. W. A. Robinson, J. D. S. Witt, M. G. Blamire, Science 329, 59 (2010) 10 F. S. Bergeret, A. F. Volkov, and K. B. Efetov, Phys. Rev. B. 68, 064513 (2003); A. F. Volkov, F. S. Bergeret, and K. B. Efetov, Phys. Rev. Lett. 90, 117006 (2003). 11 A. A. Golubov, M. Yu. Kupriyanov, and E. llichev, Rev. Mod. Phys. 76, 411 (2004). 12 B. Kastening, D. K. Morr, D. Manske, and K. Bennemann, Phys. Rev. Lett. 96, 047009 (2006). 13 P. M. R. Brydon, B. Kastening, D. K. Morr, and D. Manske, Phys. Rev. B. 77, 104504 (2008). 14 P. M. R. Brydon and D. Manske, Phys. Rev. Lett. 103, 147001 (2009). 15 P. M. R. Brydon, Y. Asano, and C. Timm, Phys. Rev. B 83, 180504(R) (2011). O. Birge, Phys. Rev. Lett. 104, 137002 (2010). 16 T. Yokoyama, Phys. Rev. B 84, 132504 (2011). 5 17 P. Gentile, M. Cuoco, A. Romano, C. Noce, D. Manske, and P. M. R. Brydon, Phys. Rev. Lett. 111, 097003 (2013). 18 K. D. Nelson, Z. Q. Mao, Y. Maeno, and Y. Liu, Science 306, 1151 (2004). 19 J. D. Strand, D. J. Van Harlingen, J. B. Kycia, and W. P. Halperin, Phys. Rev. Lett. 103, 197002 (2009). 20 J. D. Strand, D. J. Bahr, D. J. Van Harlingen, J. P. Davis, W. J. Gannon, W. P. Halperin, Science 328, 1368 (2010). 21 M. S. Anwar, Y. J. Shin, S. R. Lee, S. J. Kang, Y. Sugi- moto, S. Yonezawa, T. W. Noh and Y. Maeno, Appl. Phys. Express 8, 015502 (2015). 22 N. Nagaosa, J. Sinova, S. Onoda, A. H. MacDonald, and P. Ong, Rev. Mod. Phys. 82, 1539 (2010). 23 J. Ye, Y. B. Kim, A. J. Millis, B. I. Shraiman, P. Majum- dar, and Z. Tesanovic, Phys. Rev. Lett. 83, 3737 (1999). 24 Y. Taguchi, Y. Oohara, H. Yoshizawa, N. Nagaosa, and Y. 31 H. Haug and A.-P. Jauho, Quantum Kinetics in Transport and Optics of Semiconductors (Springer, New York, 1997). 32 A. Takeuchi and G. Tatara, J. Phys. Soc. Jpn. 77, 074701 (2008); A. Takeuchi, K. Hosono, and G. Tatara, Phys. Rev. B 81, 144405 (2010). 33 U. K. Rossler, A. N. Bogdanov, and C. Pfleiderer, Nature (London) 442, 797 (2006). 34 S. Muhlbauer, B. Binz, F. Jonietz, C. Pfleiderer, A. Rosch, A. Neubauer, R. Georgii, and P. Boni, Science 323, 915 (2009). 35 F. Jonietz, S. Muhlbauer, C. Pfleiderer, A. Neubauer, W. Munzer, A. Bauer, T. Adams, R. Georgii, P. Boni, R. A. Duine, K. Everschor, M. Garst, and A. Rosch, Science 330, 1648 (2010). 36 N. Nagaosa and Y. Tokura, Nat. Nanotechnol. 8, 899 (2013). Tokura, Science 291, 2573 (2001). 37 A. Fert, V. Cros, and J. Sampaio, Nat. Nanotechnol. 8, 25 K. Ohgushi, S. Murakami, and N. Nagaosa, Phys. Rev. B 152 (2013). 62, R6065 (2000). 26 G. Tatara and H. Kawamura, J. Phys. Soc. Jpn. 71, 2613 (2002). 27 M. Onoda, G. Tatara, and N. Nagaosa, J. Phys. Soc. Jpn. 73, 2624 (2004). 28 K. Taguchi and G. Tatara, Phys. Rev. B 79, 054423 (2009). 29 D. A. Ivanov and Ya. V. Fominov, Phys. Rev. B 73, 214524 (2006). 30 S. Murakami, N. Nagaosa and Shou-Cheng Zhang, Phys. Rev. B 69, 235206 (2004). 38 L. Fu, Phys. Rev. Lett. 103, 266801 (2009). 39 V. A. Oboznov, V. V. Bol'ginov, A. K. Feofanov, V. V. Ryazanov, and A. I. Buzdin, Phys. Rev. Lett. 96, 197003 (2006). 40 H. Kontani, J. Goryo, and D. S. Hirashima, Phys. Rev. Lett. 102, 086602 (2009). 41 A. G. Mal'shukov and C. S. Chu, Phys. Rev. B 78, 104503 (2008); Phys. Rev. B 84, 054520 (2011).
1112.2204	2	1112	2012-02-15T22:18:10	Femtosecond Population Inversion and Stimulated Emission of Dense Dirac Fermions in Graphene	[ "cond-mat.mes-hall", "cond-mat.mtrl-sci" ]	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 optical conductivity for the first 200fs, where stimulated emission completely compensates absorption loss in the graphene layer. Our experiment-theory comparison with two distinct electron and hole chemical potentials reproduce absorption saturation and gain at 40fs, revealing, particularly, the evolution of the transient state from a hot classical gas to a dense quantum fluid with increasing the photoexcitation.	cond-mat.mes-hall	cond-mat	Femtosecond Population Inversion and Stimulated Emission of Dense Dirac Fermions in Graphene T. Li,1 L. Luo,1 M. Hupalo,1 J. Zhang,1, 2 M. C. Tringides,1 J. Schmalian,1, 3 and J. Wang1 1Ames Laboratory and Department of Physics and Astronomy, Iowa State University, Ames, Iowa 50011, U.S.A. 2Department of Physics, College of William and Mary, Williamsburg, Virginia 23187 3Institute for Theory of Condensed Matter and Center for Functional Nanostructures, Karlsruhe Institute of Technology, Karlsruhe 76128, Germany (Dated: June 21, 2018) 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 mani- fests 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. Our experiment-theory com- parison with two distinct electron and hole chemical potentials reproduce absorption saturation and gain at 40fs, revealing, particularly, the evolution of the transient state from a hot classical gas to a dense quantum fluid with increasing the photoexcitation. PACS numbers: 78.67.Wj, 73.22.Pr, 78.47.J-,78.45.+h Graphene is gradually emerging as a prominent plat- form for ultrafast photonics and optoelectronics [1–3]. Growing evidence was demonstrated in, e.g., broad- band transparency and universal absorption from the near-infrared to visible[4], carrier dynamics [5], sat- urable absorption [6], pulsed photoluminescence [7–9], and coherently-driven photo-currents [10]. For graphene to play a significant role in ultrafast laser technology or telecommunications that exceed semiconductor nanos- tructure performance, it is vital to investigate femtosec- ond nonlinearities of strongly photoexcited states. Prior time-resolved studies in graphene have been mostly con- cerned with the weak excitation regime where the pho- toexcited carrier density is much smaller than the initial background carrier density. In this case a linear power dependence of transient signals was observed [5, 11–13]. Ultrafast photoexcitations strongly alter the thermo- dynamic equilibrium of electronic states and lead to a series of temporally overlapping rapid processes in graphene, as illustrated in Fig. 1(a). First, during or immediately following the pulse duration τp, photoexci- tations are coherent. Then, electron-electron collisions lead to decoherence and eventually to a quasi-thermal transient distribution after a time τth. Finally, for longer times via coupling to phonons, the system relaxes back to equilibrium via cooling of the hot carriers (τc) and recombination of electron-hole pairs (τr). In most semi- conductors and their nanostructures, where τth ≫ τp for ∼10 fs laser pulses [14], ultrafast nonlinear photoexcita- tions lead to a largely non-thermal, peaked distribution close to the pump photon energy and state filling domi- nates on the 10s of fs time scale [first panel, Fig. 1(b)]. In this letter, we demonstrate that graphene is in the opposite limit τth < τp for 35 fs pulses, where a broad- band quasi-equilibrium, yet population inverted dense Fermi system emerges during the pulse propagation. The Typeset by REVTEX FIG. 1. (color online) (a) Schematics of ultrafast optical interband excitations. (b) Dispersion of our electron-doped graphene monolayers (µ = 0.4eV ) illustrating state filling (left) and band filling (right) that leads to stimulated emis- sion from a broadband, inverted population (red arrow). Also shown together is the pump pulse spectrum. (c) STM images of tomography of the sample used. (d) The static differential reflectivity spectra (red dots). The threshold ∆R/Rc for zero conductivity can be directly determined ∼-1.46% (black line, see text for details). photoexcited phase space is quickly depleted via carrier- carrier collisions, leading to band filling [second panel of Fig. 1(b)]. Most intriguingly, we demonstrate that broadband gain emerges below the excitation photon en- ergy via stimulated emission. The broadband optical gain directly manifests itself via a remarkable negative conductivity at the near-infrared probe energy detuning even ∼400 meV below the excitation level within hun- dreds of fs. These results can be quantitatively described in terms of distinct electron and hole chemical potentials, which evolves from a hot classical gas to a dense quan- tum fluid with increasing the excitation from the linear to the highly nonlinear regime. Such femtosecond build-up of high-density and broadband population inversion has implications in advancing gaphene-based above-terahertz speed modulators, saturable absorbers and gain medi- ums. 2 A Ti:Sapphire amplifier with central wavelength 800nm (¯hω = 1.55eV) and τp = 35fs is used to pump an optical parametric amplifier (900-2400 nm) to pro- duce near-infrared probes below the 1.55 eV excitation. Ultrafast degenerate and non-degenerate differential re- flectivity changes ∆R/R with ∼ 40 fs time resolution are recorded with tunable pump fluences, which exhibits no pump polarization dependence. Our epitaxial graphene monolayer sample was grown by thermal annealing in ul- trahigh vacuum on a Si-terminated 6H-SiC(0001). The Fermi energy is ∼ 0.4 eV in the sample (Fig. 1b) reflect- ing the substrate-induced electron doping as reported [15, 16]. Fig.1(c) shows the STM characterizations of to- mography of our samples used, which show homogenous carbon monolayer in atomic length scales (top) and µm scales with smooth overgrowth on the steps of SiC surface (bottom). The strong dependence of the 6√3 × 6√3R30 reconstruction modulation with the bias voltage con- firm the single layer graphene, with the homogeneity of the sample better than 90% across the entire probe re- gion. The monolayer thickness is further confirmed by the optical deferential reflection spectra [Fig. 1(d)], de- termined by the measurements with (Rg+s) or without (Rs) graphene on SiC substrate. To understand optical response in graphene, we ex- pand the solution of the Fresnel equation with respect to σ/c (of the order of the fine structure constant of quan- tum electrodynamics αQED = e2/ (¯hc) ≈ 1/137 ≪ 1) and obtain (also see supplementary) Rs+g − Rs Rs = 4 n2 s − 1 4π c σ′ (ω) + O(cid:0)α2 QED(cid:1) . (1) σ′ (ω) =Re σ(ω) is the real part of the complex opti- cal conductivity. Eq.(1) leads to two key aspects: (i) the reflection coefficient provides a direct measurement σ′ (ω) (or absorption) without reference to any model as- sumption,e.g., ∆R/R = 16π/(n2 s − 1)c ∗ (σ′ (τ ) − σ′ (0)); (ii) using the established universal value σ0 = e2 4¯h for graphene monolayer without pulse [2, 3], there exists a threshold value for the photoinduced differential re- flectivity that corresponds to the transition to a neg- ative optical conductivity and thus to gain behavior ∆R/Rc = − 4παQED −1 . With ns = 2.7 for SiC substrate follows ∆R/Rc = −1.4582%., which can be determined experimentally by measuring differential reflectivity the optical spectra [Fig. 1(d)]. Here the reflection from n2 s FIG. 2. (color online) (a) The degenerate differential reflectiv- ity at 1.55 eV for the graphene monolayer (solid dots and line) and SiC substract (empty dots) under the pump fluence of 1842 µJ/cm2. Inset: ∆R/R in logarithmic scale (black dots) together with the pump-probe autocorrelation (red line). (b), ∆R/R at 1.55 eV for different pump fluences. The dashed straight line is guide for the eyes. The different curves were taken in the order as marked. The perfect overlap of the pair of curves indicates the signals are not caused by laser-induced permanent changes. (c) The peak △R R peak as function of the pump fluence (black squares) and the corresponding conduc- tivity change (red solid dots). Blue arrow marks the thresh- old for zero conductivity ∆R/Rc = −1.4582% (see text). Dashed line: linear dependence (guide to the eyes). the zero conductivity state in the pumped garphene/SiC sample exactly corresponds to the case of bare SiC sub- strate. Critical to note also the negative σ′ (ω) repre- sents the hallmark for the existence of gain in the ex- cited graphene layer, which should not be confused with the refection loss in the substrate. Next we present ultrafast degenerate reflectivity spec- troscopy to reveal femtosecond nonlinear saturation. A typical temporal profile of ∆R/R at 1.55 eV in the graphene monolayer (black dots) is shown in Fig. 2(a), clearly showing a negative transient signal. Note the photoinduced change is negligible in the controlled ex- periment using the SiC substrate without graphene (red circles). Several temporal regimes can be identified in the logarithmic scale plot in the inset: a pulse width limited rise ∼ 40 fs (red line), followed by a bi-exponential decay of 70 fs and 2.5 ps (blue lines). The power dependence of photoinduced ∆R/R at the maximum of ∆R in Fig. 2(b) reveals a clear nonlinear behavior. Fig. 2(c) summa- rizes ∆R/R at the signal peak for different pump fluences Ip, showing a nonlinear pump fluence dependence above ≃ 1850µJ/cm2, at least one order of magnitude higher than what is reported for semiconducting single-walled carbon nanotubes [17]. Following Eq.(1), the measured ∆R/R allows us to obtain the corresponding peak con- ductivity in photoexcited graphene, as shown by the red dot in Fig. 2(c) (normalized by σ0). At the highest pump fluence used ≃ 3868µJ/cm2, ∆R/R peak appoaches 90% of the critical value ∆R/Rc and the peak conductivity drops to only 10% of σ0. The most striking response is obtained after ultrafast non-degenerated differential reflectivity. Fig. 3(a) shows dynamics using 1.55 eV pump and low energy probes at 1.16 eV and 1.33eV. It is clearly visible that, at high pump fluence, the critical value ∆R/Rc, the threshold for negative conductivity (blue lines), indeed occurs for both cases. With increasing pump fluence, Fig. 3(b) indicates that, above Ipump,c ≃ 2000µJ/cm2, the stimu- lated infrared emission surpasses absorption loss in the photoexcited graphene for the 1.16 eV probe. The tem- poral profiles for pump fluence above the gain thresh- old indicate that the negative conductivity can persist for hundreds of fs, e.g., the 3960 µJ/cm2 at 1.16 eV. We emphasize three key aspects of this conclusion: (i) the critical value has not been reached by the degener- ate pump/probe [Fig. 2(c)] and appears exclusively for non-degenerate condition when probing below 1.55 eV, (ii) the ∆R/Rc is a model independent value that corre- sponds to σ = 0, which directly indicates the transition from loss to gain behavior, (iii) the femtosecond emer- gence of stimulated emission even at the ∼400 meV below the excitation level indicates very rapid establishment of broadband population inversion and broadband gain in the strongly photoexcited graphene states. We can extract the number of photoexcited electrons in graphene immediately after the laser pulse from nex(Ip) = Z ∞ −∞ dt τp nex(t, Ip) = 1 ¯hω Z ∞ −∞ dt τp I(t, Ip)A (t) , (2) where I(t, Ip) is the Gaussian pulse envelop I(t, Ip) = dt τp −∞ π exph −4 ln 2 τ 2 p Ipq 4 ln 2 pulse fluence is Ip = R ∞ t2i, normalized such that the total I(t, Ip). For τth ≪ τp, the absorption coefficient A (t) is determined by the adiabatic dependence of the absorption on the pump fluence and can be derived from the measured saturation curve at 40 fs (Fig. 2c) by A(t) = A0 + ∆A(t) = A0(1 + ∆A(t) ) A0 (see supplementary). Without pump, the absorption of a πe2 ¯hc . Using graphene monolayer on SiC is A0 = the actual absorption A (t), instead of A0, is crucial as the linear relation ¯hωnex ≃ A0Ip substantially overestimates nex during the pulse propagation, as shown in Fig. 3(c). From Eq.(3), we extract from the data extremely dense max ≃ 0.5 × 1014 cm−2 for our photo-excited fermions nexp electron doped sample, surpassing the saturation in a 10 nm GaAs quantum well by more than two orders of magnitude under the similar excitation condition [18]. (1+nSiC )2 4 Next we analyze the transient state at τth < t < τc. Immediately after the pulse at ∆t = 40fs, energy of the electronic system is conserved because no relaxation into the phonon system has taken place yet. In the case of highly excited graphene, the phase space constraint of the Dirac spectrum leads to an approximate conservation of 3 FIG. 3. (color online) (a) Ultrafast ∆R/R at 1.55 eV pump, 1.16 eV probe, at 1116 and 3960 µJ/cm2, and 1.33 eV probe, at 4390 µJ/cm2, respectively. Blue arrow marks ∆R/Rc = −1.4582% for zero conductivity. Shown together are the pump and probe spectra. (b), The peak transient re- flectivity as function of the pump fluence. (c), The extracted transient fermion density at 40 fs (blue dots), as explained in the text, which is significantly lower than A0Ip/¯hω obtained from the universal conductivity (open circles), as illustrated in shadow area. (d), Theory (lines) vs. experimental values (rectangles) for non-degenerate (red) and degenerate (blue) transient conductivity at 40 fs. Shown together (lines) are two model simulations with the single (green) or distinct (black) chemical potentials. numbers of photoexcited holes and electrons, valid to the second order in the electron-electron Coulomb interaction [19, 20]. A recent explicit analysis of the short time dy- namics by Winzer et al. indicates that the conservation of hole and electron numbers is a good approximation for the high excitation regime [21]. Consequently, this gives rise to a slow imbalance relaxation and thus to a popula- tion inverted transient state with quasi-conserved occu- pations of the two branches of the Dirac cone in our ex- perimental condition. These, together with the assump- tion that a decohered, quasi-thermal state leads to the non-equilibrium transient distribution function (ε = vp): fe(h) (ε) = 1 exp(cid:16) ε−µe(h) kB Te (cid:17) + 1 , (3) characterized by the electron temperature, Te and two distinct chemical potentials µe and µh, for electrons in the upper and holes in the lower branch of the spectrum, respectively. Note that a scenario based on a single chem- ical potential does not explain the demonstrated popu- In thermodynamic equilibrium holds lation inversion. In general, Te and µe(h) µe = −µh = µ and Te = T . are functions of given photon energy, ¯hω, total density, 4 a crossover from a hot and dilute Maxwell-Boltzmann gas to a degenerate population inverted quantum sys- tem, with µe + µh measuring the degree of population inversion [24]. Most notably, the transient conductivity, σ (ω), of Eq.(4) changes sign if µe + µh crosses ¯hω, with broadband optical gain due to population inversion for the entire region ω < µe + µh (σ < 0), as illustrated between two intersection planes in Fig. 4(c). The sep- aration between the planes is shown to approach to the pump energy 1.55eV at high excitation density, consis- tent with our experiment. We have showed the existence of pronounced fem- tosecond population inversion and near-infrared gain in strongly photoexcited graphene monolayers. These re- sults clearly reveal the transient electron and hole poten- tials are separated on the time scale of 100s of fs. Our experimental-theory comparison explains well of the ab- sorption saturation and gain from the dense fermions, and reveals a crossover from a hot Maxwell-Boltzmann gas to a degenerate dense Fermi-Dirac system. This work was supported by by the U.S. Department of Energy-Basic Energy Sciences under Contract No. DE- AC02-07CH11358. J.S. acknowledges support by the Deutsche Forschungsgemeinschaft through the Center for Functional Nanostructures within subproject B4.5. [1] F. Bonaccorso et. al., Nature Photonics 4, 611-622 (2010) [2] A. H. Castro Neto et al., Rev. Mod. Phys. 81, 109 (2009) [3] Kin Fai Mak, et al., Phys. Rev. Lett. 101,196405 (2008). [4] R.-R. Nair et al., Science , 6, 1308 (2008). [5] J.-M. Dawlaty et al., Appl. Phys. Lett. 92, 042116 (2008); P.-A. George et al., Nano Lett. 8(12), 4248 (2008); H. Choi et al., Appl. Phys. Lett. 94, 172102 (2009). [6] Z. Sun et al., ACS Nano. 4(2), 803 (2010); Q. Bao et al., Advanced Functional Materials 19, 3077 (2009). [7] R.-J. Stohr et al., Phys. Rev. B 82, 121408 (2010). [8] C.-H. Liu et al., Phys. Rev. Lett. 105, 127404 (2010). [9] W. Lui et al., Phys. Rev. B 82, 081408 (2010). [10] D. Sun et al., Nano Lett. 10(4), 1293 (2010). [11] M. Breusing et al., Phys. Rev. Lett. 102, 086809 (2009). [12] R.-W. Newson et al., Optics Express 17, 2326 (2009). [13] D. Sun et al., Phys. Rev. Lett. 101, 157402 (2008). [14] J.-L.Oudar et al., Phys. Rev. Lett. 55, 2074 (1985) [15] I. Gierz et al., Nano Lett. 8, 4603 (2010); [16] M. Hupalo et al., Phys. Rev. B 80, 041401(R) (2009). [17] J. Wang et al, Phys. Rev. Lett. 104, 177401 (2010). [18] D. S. Chemla et al., IEEE J. Quant Electron. 20, 265 (1984). [19] L. Fritz et al., Phys. Rev. B 78, 085416 (2008). [20] M. S. Foster et al., Phys. Rev. B 79, 085415 (2009). [21] Torben Winzer et al., Nano. Lett. 10, 4839 (2010) [22] L. V. Keldysh, Sov. Phys. JETP 20, 1018 (1965). [23] This extremely good agreement between theory and ex- periment clearly corroborates τth < 40 fs (τth < τp < τc). The 70 fs and 2.5 ps relaxation components in Fig. 2a can then be naturally assigned to the cooling of hot carriers τc and recombination of electron-hole pairs τr. FIG. 4. (color online) (a) The calculated transient electron (µe), hole (µh) chemical potentials and transient temperature (inset) as function of photoexcited carrier density. (b) Tran- sient electron and hole distribution function at 40 fs are plot- ted for different pump fluence. The two intersection planes represent the occupation probability of an electron-hole pair at the given excitation photon energy is equal to unity, i.e., fe(ω/2) + fh(ω/2) = 1. and density of photoexcited carriers, nex (see supplemen- tary section). The non-equilibrium optical conductivity is calculated in Keldysh formalism [22] and follows as σ (ω) = e2 4¯h (1 − fe (¯hω/2) − fh (¯hω/2)) . (4) Fig. 3(d) compares the peak transient conductivity with the calculated conductivity σ (ω) of Eq.4 as func- tion of nex for two probe photon energies 1.55 eV and 1.16 eV. Excellent agreement between experiment and theory is found which demonstrates the faithful repre- sentation of the transient state at 40 fs by the distri- bution function, Eq.3. For the degenerate scheme, our theory (black dashed line) yields σ → 0 and thus per- max = 0.48 × 1014 cm−2. Once fect transparency for nth the systems is driven into this regime, a balance be- tween stimulated emission and absorption will lead to a transprancy. For non-degenerate scheme by probing at 1.16 eV, our theory (black solid line) predicts a crit- c = 0.32 × 1014 cm−2 for the transition ical density nth from loss to gain. All of these results agree quantita- max = 0.5 × 1014 tively with the experimental values nexp cm−2 and nexp c = 0.34 × 1014 cm−2, respectively (black In addition, the experiment-theory com- arrows) [23]. parison of the conductivity σ(ω) is shown in Fig. 3d σ0 for the distinct- (black lines) and the equal-chemical- potential model (green lines) at the probe photon energy ¯hω = 1.55eV and 1.16eV, which clearly identifies the va- lidity of the distinct-µ model calculation. The detailed information of the transient state at 40 fs as a function of nex is shown in Fig. 4(a) for the transient chemical potentials and carrier temperature (inset). For weak excitation nex ≪ ω2/v2 (but larger than the num- ber of thermally excited carriers before the pulse) we find kBTe ∼ ¯hω while µh < 0, corresponding to a hot, dilute gas of classical holes. Increasing nex changes the sign of the hole chemical potential (µh > 0) and eventually de- creases Te, i.e. as a function of pump fluence we enforce [24] Our analysis for the low photoexcited carrier density and longer, ps relaxation are consistent with the behaviors found in [25] [25] R. Kim et al. Phys. Rev. B 84, 075449 (2011) 5
1105.1125	1	1105	2011-05-05T17:36:42	Wave packet dynamics and valley filter in strained graphene	[ "cond-mat.mes-hall" ]	The time evolution of a wavepacket in strained graphene is studied within the tight-binding model and continuum model. The effect of an external magnetic field, as well as a strain-induced pseudo-magnetic field, on the wave packet trajectories and zitterbewegung are analyzed. Combining the effects of strain with those of an external magnetic field produces an effective magnetic field which is large in one of the Dirac cones, but can be practically zero in the other. We construct an efficient valley filter, where for a propagating incoming wave packet consisting of momenta around the K and K' Dirac points, the outgoing wave packet exhibits momenta in only one of these Dirac points, while the components of the packet that belong to the other Dirac point are reflected due to the Lorentz force. We also found that the zitterbewegung is permanent in time in the presence of either external or strain-induced magnetic fields, but when both the external and strain-induced magnetic fields are present, the zitterbewegung is transient in one of the Dirac cones, whereas in the other cone the wave packet exhibits permanent spatial oscillations.	cond-mat.mes-hall	cond-mat	Wave packet dynamics and valley filter in strained graphene Andrey Chaves,1, 2 , ∗ L. Covaci,1 Kh. Yu. Rakhimov,1, 3 G. A. Farias,2 , † and F. M. Peeters1, 2 , ‡ 1Department of Physics, University of Antwerp, Groenenborgerlaan 171, B-2020 Antwerp, Belgium 2Departamento de F´ısica, Universidade Federal do Cear´a, Caixa Postal 6030, Campus do Pici, 60455-900 Fortaleza, Cear´a, Brazil 3Heat Physics Department of the Uzbek Academy of Sciences, 100135 Tashkent, Uzbekistan (Dated: January 20, 2013) The time evolution of a wavepacket in strained graphene is studied within the tight-binding model and continuum model. The effect of an external magnetic field, as well as a strain-induced pseudo- magnetic field, on the wave packet tra jectories and zitterbewegung are analyzed. Combining the effects of strain with those of an external magnetic field produces an effective magnetic field which is large in one of the Dirac cones, but can be practically zero in the other. We construct an efficient valley filter, where for a propagating incoming wave packet consisting of momenta around the K and K ′ Dirac points, the outgoing wave packet exhibits momenta in only one of these Dirac points, while the components of the packet that belong to the other Dirac point are reflected due to the Lorentz force. We also found that the zitterbewegung is permanent in time in the presence of either external or strain-induced magnetic fields, but when both the external and strain-induced magnetic fields are present, the zitterbewegung is transient in one of the Dirac cones, whereas in the other cone the wave packet exhibits permanent spatial oscillations. PACS numbers: 72.80.Vp,73.23.-b, 85.75.Hh I. INTRODUCTION Since its first synthesis in 2004,1 graphene has been at- tracting much interest due to its unique electronic prop- erties arising from its singular energy spectrum, where in the vicinity of the points labelled as K and K ′ in recip- rocal space, the charge carriers behave as massless quasi- particles and exhibit an almost linear dispersion.2 These quasi-particles obey the Dirac-Weyl equations and there- fore should be sub ject to quasi-relativistic effects, such as zitterbewegung, i.e., a trembling motion caused by inter- ference between positive and negative energy states.3–5 The phenomenon of zitterbewegung was predicted in 1930 by Schrodinger6 and has been sub ject of renewed interest over the past years. Previous theoretical works have suggested few ways of experimentally observing in narrow-gap semiconductors7 , zitterbewegung, e. g. in III-V zinc-blende semiconductor quantum wells with spin-orbit coupling8 and, more recently, in monolayer9 and bilayer10 graphene. An experimental simulation of the zitterbewegung of free relativistic electrons in vacuum was performed by Gerritsma et al.11 by using trapped ions. Strain engineering in graphene has recently become a widely studied topic.12–17 The elastic properties of graphene nanoribbons were theoretically investigated by Cadelano et al.18 , which studied the in-plane stretch- ing and out of plane bending deformations by combining continuum elasticity theory and tight-binding atomistic simulations. Later, Cocco et al.19 and Lu and Guo20 showed that a combination of shear and uniaxial strain at moderate absolute deformations can be used to open a gap in the graphene energy spectrum. It has been shown recently that specific forms of strain produce a pseudo-magnetic field in graphene, which does not break the time reversal symmetry and which points in opposite directions for electrons moving around the K and K ′ points.21 The strain-induced magnetic field is expected to produce Landau levels and, consequently, the quan- tum Hall effect, even in the absence of an external mag- netic field.22,23 Guinea et al.,24 showed theoretically that an in-plane bending of the graphene sheet leads to an almost uniform field. Landau levels as a consequence of strain-induced pseudo-magnetic fields greater than 300 Tesla were recently observed with scanning tunnelling microscopy (STM) in nanometer size nanobubbles.25 Although previous works have studied wave packet propagation for the Dirac-Weyl Hamiltonian of graphene in the absence of external fields and potentials,26 or in the presence of uniformly applied external magnetic fields9,27–29 , there is still a lack of theoretical works on the wave packet propagation through potential and (pseudo) magnetic field step barriers. Moreover, the time evolu- tion of a wave packet in graphene within the tight-binding (TB) model, where the intra-valley scattering to higher energy states and inter-valley scattering due to defects appear naturally, is still hardly found in the literature. It is also interesting to see whether the results from Dirac and TB approaches for graphene differ or are similar. In this paper, we investigate the time-evolution of wavepackets in graphene within the TB model, based on the split-operator technique for the expansion of the time-evolution operator. We trace a parallel between the results from the TB model and those obtained from the Dirac approximation for particles with momentum close to one of the Dirac cones of the Brillouin zone of graphene. The proposed method is then applied to the study of the dynamics of Gaussian wave packets in graphene under external magnetic fields. The effects of the pseudo-magnetic field induced by bending the graphene sheet into an arc of circle are analyzed as well. Our results show that for an appropriate choice of strain and external magnetic field strength, the system exhibits a strong effective magnetic field for particles in one of the Dirac cones, whereas in the other cone the external and pseudo-magnetic fields cancel each other and the effective magnetic field is practically zero. We show that this effect is manifested as a transient (permanent) zitterbewegung for electrons in the cone where the effective magnetic field is zero (non-zero), which can be verified experimentally by detecting the electric field radiation emitted by the trembling wave packet.27 Moreover, our results show that with such a choice of external and strain-induced mag- netic fields, one can construct an efficient valley-filter, which can be useful for future valley-tronic devices.30 II. TIME EVOLUTION OPERATOR By solving the time-dependent Schrodinger equation, one obtains that the propagated wavefunction after a time step ∆t can be calculated by applying the time- evolution operator on the wave packet at any instant t y x 2 unstrained graphene and aij is the distance between the sites after the stress. The change in the inter-sites dis- tance affects the hopping energy between the sites, which becomes21 τij → τij (1 + 2∆aij /a0 ). A similar expres- sion can be obtained by expanding Eq. (13) of Ref. 37 in Taylor series and neglecting higher order terms in ∆aij , i.e., considering small lattice deformations. The strain-induced change in the hopping energies leads to an effective pseudo-magnetic field, which points to op- posite directions in the valley K and K ′ .22 Notice that the pseudo-magnetic field in our model is not introduced artificially by considering an additional vector potential in the Peierls phase, but it rather appears naturally after the changes in the inter-site distances due to the strain. a) {1,1} A B {1,2} {2,1} {2,2} {3,1} {3,2} b) b1 3 4 2 K 5 K’ 1 6 b2 Ψ(~r , t + ∆t) = e− i H∆tΨ(~r , t), (1) where we assumed that the Hamiltonian H is time- independent. Different techniques have been suggested for the expansion of the exponential in Eq. (1), e.g. the Chebyschev polynomials method31 and the second order differencing scheme32,33 . The numerical method that we use for the application of the time evolution operator in this work, namely, the split-operator method,34 is the sub ject of this section, where we will show how this tech- nique can be adapted for the study of the wave packet dynamics in graphene within the tight-binding and Dirac approximations. A. Tight-Binding model (2) Graphene consists of a layer of carbon atoms form- ing a honeycomb lattice, which can be described by the Hamiltonian j i , i ci + X<i,j> hτij c† HT B = Xi ij ci c† ǫi c† i cj + τ ∗ where ci (c† i ) annihilates (creates) an electron at the site i, with on-site energy ǫi , and the sum is taken only between nearest neighbor sites i and j , with hopping energy τij . The effect of an external magnetic field can be calculated by including a phase in the hopping parameters accord- ing to the Peierls substitution τij → τij exp hi e ~A · d~li, R i j where ~A is the vector potential describing the mag- netic field.35,36 In a strained graphene sheet, the dis- tance between two adjacent sites i and j is changed by ∆aij = aij − a0 , where a0 is the lattice parameter of FIG. 1: (Color online) (a) Sketch of the honeycomb lattice of graphene, made out of two superimposed triangular lattices A and B . The atoms are labelled as {n, m} according to their line and column numbers n and m, respectively. (b) Reciprocal lattice of graphene, with K (black) and K ′ (gray) Dirac points, where the area defined by the reciprocal vectors ~b1 = (−2π/3a0 , √3) and ~b2 = (4π/3a0 , 0) represents the first Brillouin zone. The numbers close to each Dirac point are explained in the text. Let us label the sites i of the graphene lattice according to their line and column numbers n and m, respectively, as shown in Fig. 1(a). The basis vector state n, mi represents an electron confined on the site of line n and column m. In a lattice consisting only of non-interacting sites, each n, mi is an eigenstate of HT B with energy ǫn,m , i.e. HT B n, mi = ǫn,m n, mi. Limiting ourselves to nearest neighbors interactions, we find HT B n, mi = ǫnm n, mi + Tm+1 n, m + 1i +Tm−1 n, m − 1i + Tn+1 n + 1, mi + Tn−1 n − 1, mi, (3) where Tn±1 and Tm±1 are equivalent to the hopping en- ergies τij between the site i = {n, m} and the adjacent sites j = {n ± 1, m} and j = {n, m ± 1}, respectively. Eq. (3) can be rewritten as HT B n, mi = Hn n, mi + Hm n, mi, where the operators Hn and Hm are defined as ǫnm 2 Hn n, mi = Tm+1 n, m+ 1i+ Tm−1 n, m− 1i+ and (4) (5a) Hm n, mi = Tn+1 n+ 1, mi+ Tn−1 n− 1, mi+ ǫnm 2 . (5b) The wavefunction at any instant t is then written as a linear combination of the basis vector states Ψt n,m = Pn,m bt n,m n, mi. The advantage of following the pro- cedure described by Eqs. (3-5) lies in the fact that the operators Hn and Hm in Eq. (5) can be represented by tri-diagonal matrices, which are easier to handle than the matrix representing the full Hamiltonian, Eq. (3). The split-operator technique can now be applied to the Hamiltonian Eq. (4), so that the time evolution operator is approximated by e− i Hn∆t e− i 2 Hm∆t e− i HT B ∆t = e− i 2 Hm∆t + O(∆t3 ), (6) where the error comes from the non-commutativity be- tween the operators Hn and Hm . We drop the O(∆t3 ) terms by considering a small time step ∆t = 0.1 fs. The propagated wavefunction is then obtained from Eq. (1), which in this case reads n,m = e− i Hn∆t e− i 2 Hm∆t e− i Ψt+∆t 2 Hm∆tΨt n,m . (7) This equation is solved in three steps: ηn,m = e− i 2 Hm∆tΨt n,m , ξn,m = e− i Hn∆tηn,m , n,m = e− i Ψt+∆t 2 Hm∆t ξn,m . (8a) (8b) (8c) Using the Cayley form for the exponentials,38 we can rewrite Eq. (8a) as ηn,m = e− i 2 Hm∆tΨt n,m = 1 − i∆t 4 Hm 1 + i∆t 4 Hm n,m + O(∆t2 ),(9) Ψt i∆t 4 (10) i∆t 4 which leads to Hm(cid:19) Ψt Hm(cid:19) ηn,m ≈ (cid:18)1 − (cid:18)1 + n,m . As the wavefunction Ψt n,m is already known, the matrix equation in Eq. (10) can be straightforwardly solved to obtain ηn,m . We repeat this procedure for the other two exponentials in Eqs. (8b) and (8c), and eventually obtain Ψt+∆t n,m . In fact, the form in Eq. (10) can also be applied to the full Hamiltonian HT B , i.e. without splitting the Hm and Hn terms. However, this would lead to matrix equations involving penta-diagonal matrices, which are harder to handle than the tri-diagonal matrices in Eq. (10). As the error produced by the splitting in Eq. (7) is smaller than the error produced by the (necessary) expansion of the exponential given by Eq. (10), it is worthy to split these terms in order to simplify the numerical calculations. 3 B. Dirac-Weyl equation (12) From the TB Hamiltonian Eq. (2), considering an in- finite graphene sheet in the absence of external potential and magnetic fields, one obtains the energy bandstruc- ture of graphene as E (cid:16)~k(cid:17) = ±τ r3 + f (cid:16)~k(cid:17) f (cid:16)~k(cid:17) = 2 cos (cid:16)√3ky a0(cid:17)+4 cos √3 ky a0! cos (cid:18) 3 kx a0(cid:19) , 2 2 (11) which is gapless in six points of the reciprocal space where E = 0, from which only two are inequivalent, labelled as K and K ′ , as shown in Fig. 1(b).2,39 In the vicinity of each of these points, the dependence of the energy spectrum on the wave vector ~k is almost linear and the electron can be described as a massless fermion by the Dirac Hamiltonian HD = hvF ~σ · (~p + e ~A) + V (x, y)Ii e−iφ , where vF = 3τ a0/2 is the Fermi velocity, ~A is the elec- tromagnetic vector potential, V (x, y ) is an external po- tential, I is the identity matrix, ~σ is the Pauli vector and the wavefunctions are pseudo-spinors Ψ = [ΨA , ΨB ]T , with ΨA(B ) as the probability of finding the electron in the sub-lattice A(B ) that composes the honeycomb lat- tice of graphene.2 The angle φ is different for electrons around the K and K ′ Dirac cones. In the vicinity of the k -th Dirac point (see labels for each Dirac point in Fig. 1(b)), one obtains φ = −π/6 + kπ/3, with k = 1 − 6. From here onwards, we will refer to the coordinates in the Dirac (TB) model as x (x) and y (y ). The exponential term in Eq. (12) is usually dropped, because it can be considered as a phase in the state vec- tors in the Dirac model. However, this term has an im- portant meaning when comparing with the TB model: this exponential can be identified as a rotation operator with angle φ. Notice that an infinite graphene hexago- nal lattice has C6v symmetry, i.e. it is symmetric only for rotation angles kπ/3 (k - integer). As a consequence, the Dirac Hamiltonian Eq. (12) without the exponen- tial term is not symmetric in the momenta px and py , as already pointed out previously.9 So, what would be the meaning of the direction-dependent observables in the Dirac description of graphene, when they are not sym- metric under rotation, exhibiting a privileged direction? Defining y (x) as the zigzag (armchair) direction, as in Fig. 1(a), the results obtained by the Dirac approxima- tion for the x and y components of any observable are compared to the armchair and zigzag directions, respec- tively, after performing a rotation φ in the coordinates of the Dirac model. From the possible values of φ, one deduces straightforwardly that at any Dirac cone, the co- ordinate x (y) of the Dirac model is always related to one of the zigzag (armchair) directions of the real graphene lattice. On the other hand, for finite rectangular sam- ples the different angles φ represent two distinguishable situations, since the rectangle does not share the C6v symmetry of the infinite graphene lattice: the x direction in the Dirac model for the odd (even) cones in Fig. 1(b) represents a diagonal (vertical) direction in the rectangle. The comparison between the TB model for a rectangu- lar graphene flake and the Dirac approximation will be discussed in details further, in the following section. In a recent work, Maksimova et al.26 presented an an- alytical study of the time evolution of a Gaussian wave packet in graphene in the absence of external potentials and/or magnetic fields within the continuum model, i. e. using the Dirac-Weyl Hamiltonian for electrons in the vicinity of the Dirac point K . In this paper, we will use an alternative and more general way of calculating the dynamics of a wavepacket in such a system,39 based on the split-operator technique, which can be applied for systems under arbitrary external potentials and magnetic fields. The Dirac-Weyl Hamiltonian HD in Eq. (12) can be separated as HD = Hk + Hr , where Hk = vF ~σ · ~k keeps only the terms which depend on the wave vector ~k, whereas Hr = vF e~σ · ~A + V I depends only on the real space coordinates x and y. Following the split-operator method, the time evolution operator for the Hamiltonian HD can be approximated as (Hk + Hr )(cid:21) ≈ exp (cid:20)− i∆t Hr (cid:21) Hk (cid:21) exp (cid:20)− Hr (cid:21) exp (cid:20)− exp (cid:20)− i∆t with an error of the order of O(∆t3 ), due to the non- commutativity of the operators involved. We invoke a well-known property of the Pauli vectors sin(S ) exp (cid:16)−i ~S · ~σ(cid:17) = cos(S )I − i (cid:16) ~S · ~σ(cid:17) S for any vector ~S , where S = ~S , and rewrite the ex- ponentials in real and reciprocal space, respectively, in matrix form Mr = (cid:20)cos (A) I − i (cid:19)(cid:21) e− i∆t 2 V (15a) κ (cid:18) (cid:19) , (15b) sin(κ) 0 κx − iκy Mk = cos(κ)I − i κx + iκy 0 where A = ~A = ∆tvF e ~A(cid:14)2, ~κ = ∆tvF ~k and κ = ~κ = ∆tvF qk2 x + k2 y , so that the time evolution of a wave packet ΨD (x, y ) = [φA , φB ]T Ψ(x, y ) can be calculated as a series of matrix multiplications Ψ(~r, t + ∆t) = Mr · Mk · MrΨ(~r, t) + O(∆t3 ). A (cid:18) sin (A) 0 Ax + iAy Ax − iAy 0 (16) i∆t 2 i∆t 2 (13) (14) 4 The matrix multiplication by Mk is made in reciprocal space by taking the Fourier transform of the functions. In the absence of magnetic fields and external potentials, one has Mr = I and Ψ(~r, t + ∆t) = MkΨ(~r, t), (17) where the matrix multiplication in reciprocal space gives the exact result for the time evolution of the wave packet, since there is no error induced by non-commutativity of operators or matrices in this case. This shows that the split-operator method provides a way to study the dy- namics of wavepackets in graphene within the continuum model where, in the presence of magnetic fields and/or external potentials, one can control the accuracy of the results by making ∆t smaller, while in their absence, the problem is solved exactly by a simple matrix multiplica- tion for any value of ∆t. III. RESULTS AND DISCUSSION We shall now discuss the results obtained for a graphene lattice with 2000×3601 atoms, with armchair (zigzag) edges in the x(y )-direction. The nearest neigh- bors hopping parameter and the lattice constant of graphene are taken as τ = −2.7 eV and a0 = 1.42 A , respectively. As initial wave packet, we consider a Gaussian centered x , q0 at ~r0 = (x0 , y0) in real space and ~q = (q0 y ) in reciprocal space: (x − x0 )2 + (y − y0 )2 2d2 c2 (cid:19) exp (cid:20)− Ψq (~r) = N (cid:18) c1 + i~q · ~r(cid:21) , (18) where N is the normalization factor. Notice that we have included a pseudo-spinor [c1 , c2 ]T in the initial wave packet, where c1(2) is the probability of finding the elec- tron in the triangular sub-lattice A(B ) that composes the graphene hexagonal lattice. One can also rewrite the pseudo-spinor as [1, eiθ ]T , where the pseudo-spin polar- ization angle θ is shown explicitly. The pseudo-spin is a concept normally attributed to the Dirac description of graphene. Indeed, the pseudo-spin of a wavefunction in the Dirac model is related to the expectation values of the Pauli matrices hσi i, which can involve integrals of the product between wavefunctions for sub-lattices A and B . Such a definition fails for the TB wavefunctions, since in this case they are defined in different points of the lat- tice, so that any integral that mixes functions of both sub-lattices gives zero. Even so, the study of the pseudo- spin related to the initial discrete wave packet helps to understand the wave packet dynamics obtained from the TB model, as we will see further in this section. A. Initial pseudo-spin polarization and zitterbewegung revisited In this subsection, we will use the TB model to review some of the main properties of the wave packet dynam- ics in graphene. Within the TB model, we consider the initial wave packet as a discrete form of the Gaussian dis- tribution in Eq. (18) for the graphene hexagonal lattice, where we multiply the Gaussian function by c1(2) in the sites belonging to the triangular sub-lattice A(B ). From Eq. (11), it is clear that in momentum space, the region of interest is the vicinity of the Dirac points K and K ′ , since the energy corresponding to wave vectors out of this region is very high. (a) ) Å ( y k0 yd = 4 k0 yd = 2 400 200 0 -200 -400 -600 200 0 -200 -400 -600 200 k0 yd = 1 0 -200 -400 -600 -500 -250 (b) 400 ) Å ( > y < 200 0 -200 (c) ) F v ( > y v < 1.0 0.9 0.8 0.7 0.6 0 x (Å) 250 500 0 20 40 60 t (fs) 80 100 FIG. 2: (Color online)(a) Contour plots of the squared mod- ulus of the wavefunction after a time evolution of t = 40 fs, for three different values of the dimensionless parameter k0 y d. (b) Expectation value of the position and (c) velocity in y - direction as a function of time. The results obtained from the TB (Dirac) model are presented as curves (symbols), for k0 y d = 1 (solid, circles), 2 (dashed, triangles) and 4 (dotted, squares) In the TB model for two-dimensional crystals, one usually considers the same Gaussian distribution for all the sites of the lattice.40 This is equivalent as choosing c1 = c2 = 1 in Eq. (18). Figure 2(a) shows the contour plots of the squared modulus of the propagated wave- functions at t = 40 fs for these values of ci , considering an initial wave vector ~q = (0, k0 y ) + K , i.e. in the vicinity of the K point labelled as 2 in Fig. 1(b). As shown in Ref. 26, the wave packet dynamics near the Dirac cones in graphene does not depend separately on the momen- tum k0 y or on the width d, but rather on the dimensionless 5 quantity k0 y d. This result was obtained from the Dirac model for graphene, i.e. considering that even high en- ergies states exhibit linear dispersion. Within the TB model we expect that wave packets with the same k0 y d behave alike only if k0 y is not too far from the Dirac cone and if d is not too small, so that the packet is well lo- calized in energy space. Within these conditions, Fig. 2 shows the time evolution for different values of this di- y d = 1 and 2, with d = 100 A , mensionless quantity: k0 y d = 4, with d = 200 A . We observe that the dis- and k0 persion of the wave packet is stronger for smaller values of k0 y d, where it becomes distorted into an arc-like shape. For larger k0 y d, on the other hand, the wave packet keeps its circularly symmetric shape for longer times. As explained in the previous subsection, in order to obtain the Dirac Hamiltonian Eq. (12), one must shift the origin of the wave vectors ~k to one of the six Dirac points shown in Fig. 1(b). Besides, one must also rotate the x and y axis by an angle φ which depends on the K or K ′ point that is considered as the origin in momentum space. For the K = (0, 4π(cid:14)3√3a0 ) point, labelled as 2 in Fig. 1(b), the Dirac Hamiltonian Eq. (12) is obtained by 1.0 0.9 0.8 0.7 0.6 0.5 (a) 0.4 0.00 0.01 0.02 0.03 0.04 0.05 y (Å-1) k0 d = 50 Å d = 100 Å d = 200 Å x 0.97 x 0.985 y = 0.01 Å-1 k0 y = 0.02 Å-1 k0 k0 y = 0.04 Å-1 80 120 160 200 d (Å) (b) 0 40 ) F v ( f y v ) F v ( f y v 1.0 0.9 0.8 0.7 0.6 0.5 0.4 FIG. 3: (Color online) Final velocities for the Gaussian wave packet in Eq. (18), with pseudo-spin c1 = c2 = 1 and mo- mentum ~q = (0, k0 y ) + K as a function of (a) the momentum y , for widths d = 50 A , 100 A and 200 A , and (b) the width k0 y = 0.01 A −1 , 0.02 A −1 and 0.04 A −1 . The d, for momenta k0 symbols (curves) are obtained from the TB (Dirac) model. y = 0.02 A −1 In (b), the results from the Dirac model for k0 (dashed) and 0.04 A −1 (solid) are multiplied by 0.985 and 0.97, respectively. rotating the axis by 90◦ , with other words, by a trans- formation of the coordinates as x → −y and y → x. The pseudo-spinor c1 = c2 = 1 represents a wave packet po- larized in x-direction, i.e. hσx i > 0 and hσz i = hσy i = 0. From the Heisenberg picture, we obtain the velocity in the x-direction for the proposed wave packet as dx dt = 1 i [x, HD ] = vF σx . (19) Performing the appropriate coordinate transformations, the velocity obtained from the TB model for the y - direction must be consistent with the prediction from the Dirac approximation, namely, vy = dx/dt = vF σx . This suggests that such a wave packet propagates towards the positive y -direction, but with non-constant velocity, since σx does not commute with HD . The expectation value of the y position of the packet hy i is shown as a function of time by the curves in Fig. 2(b), for k0 y d = 1 (solid), 2 (dashed) and 4 (dotted), where the results obtained by the Dirac equation are shown as symbols for comparison. A different linear behavior is already observed for each wave packet at large time, implying that they have dif- ferent velocities, which is kind of counter-intuitive, since low-energy electrons in graphene are expected to propa- gate always with the same Fermi velocity vF . Figure 2(c) shows the velocity vy , calculated by taking the derivative of the TB results for hy i with respect to time, which ex- hibits clear oscillations that are damped as time evolves, converging to a final value vf y < vF that depends on the initial wave packet’s width d and wave vector k0 y . The velocities obtained by the Dirac model are shown by symbols, where the same qualitative behavior is observed as obtained from the TB model, though for higher wave packet momentum and width, a small quantitative differ- ence is observed, which is a consequence of the different energy-momentum dispersion. The oscillatory behavior of the velocity is a manifesta- tion of the zitterbewegung, i.e. a trembling motion of the wave packet due to the interference between positive and negative energy states that makes up the initial Gaussian wave packet.3,9 This effect is well-known for relativistic particles, which are described by the Dirac equation, and is also observed for electrons in graphene in the vicinity of the K and K ′ points, since they can be described as massless quasi-particles by the Dirac equation as well. The velocity wiggles with shorter period and smaller am- plitudes for larger values of k0 y d. The convergence of the velocities demonstrates that the zitterbewegung is not a permanent, but a transient effect.9 Figure 3 shows the converged velocity vf y as a func- tion of (a) the momentum k0 y and (b) the width d of the Gaussian wave packet. The TB results (symbols) are compared to those calculated from Eq. (31) in Ref. 26 (curves), which was obtained analytically from the Dirac approximation in the t → ∞ limit and is repeated here just for completeness: vf y vF = 1 − 1 − e−(k0 y d)2 2(k0 y d)2 . 6 (20) Within the Dirac model, one can observe that increasing d or k0 y in Eq. (20), the final velocity increases mono- tonically and approaches vF , which is reasonable, since a wider packet in real space leads to a narrower distribu- tion in k -space, whereas a higher value of the wave vector makes the center of the packet lay far from E = 0. In both cases, the interference with negative energy states is reduced and, consequently, zitterbewegung effects are less significant. However, since the analytical formula Eq. (20) does not take into account any effect such as the curvature of the energy bands for higher energy states or trigonal warping effects, this formula is not ex- pected to give accurate results for larger k0 y . Indeed, Fig. 3(a) shows that a very good agreement between TB and Dirac results can be observed only for small values of k0 y , whereas for large k0 y , the final velocities obtained from the TB model are lower than those obtained from the (a) 250 0 -250 -500 -750 -1000 ) Å ( y 0 -250 -500 -750 -1000 (1) (2) 0 -250 -500 -750 (3) -1000 -600 -300 0 x (Å) (b) 50 0 -50 -100 -150 -200 ) Å ( > x < -250 ) Å ( > y < -392 -394 -396 -398 -400 -300 -200 -100 < x > (Å) 0 300 600 0 30 60 90 t (fs) 120 150 FIG. 4: (Color online)(a) Contour plots of the squared modu- lus of the wavefunction after a time evolution of t = 40 fs, for three initial configurations of pseudo-spin [c1 , c2 ]T and mo- mentum ~q0 : 1) [1, 0]T , k0 x = 0 and k0 y d = 4; 2) [1, i]T , k0 x = 0 and k0 x d = 4, k0 y d = 4; and 3) [1, i]T , k0 y = 0. (b) Expec- tation value of x obtained by the TB model for the initial wavepackets 1 (solid), 2 (dashed) and 3 (dotted) as a func- tion of time. The results obtained by the Dirac model, after the appropriate coordinates rotation (see text), are shown as circles, triangles and squares, respectively. The inset shows the tra jectory of the wavepacket obtained from the TB model for the initial wavepacket 3. ) F v ( > v < 1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.0 < vx >, K' < vy >, K' 0 20 40 60 t (fs) 80 100 120 FIG. 5: (Color online)(a) Expectation value of the velocity as a function of time, for wavepackets with k0 y = k0 x = 0 and pseudo-spinor [1, 1]T (solid) and [1, i]T (circles) at the Dirac point K = (0, 4π(cid:14)3√3a0 ) (point 2 in Fig. 1(b)), and [1, 1]T (triangles) at K ′ = (2π(cid:14)3a0 , 2π(cid:14)3√3a0 ) (point 1 in Fig. 1(b)). The x and y components of the velocity in the latter case are shown as dashed and dotted curves, respec- tively. Dirac model and do not increase monotonically, but de- creases slowly for very large k0 y , as a consequence of the curvature of the energy bands. On the other hand, in Fig. 3(b) we observe that varying the wave packet width for a fixed momentum, good qualitative agreement with the Dirac model is obtained for almost any value of d. The curves for larger values of k0 y (solid and dashed) are just quantitatively different from those obtained by the TB model, and they are comparable to the TB results af- ter multiplication by a factor 0.985 (0.970) for k0 y = 0.02 A −1 (0.04 A −1 ). Worse qualitative agreement between TB and Dirac results in this case is observed only for very small d, where the Gaussian width in energy space, given by ∆E = vF d−1 , incorporates higher energy val- ues, leading to deviations in vf y obtained from the TB model as compared to those from the Dirac model. In Fig. 4(a) we show contour plots of the squared modulus of the wave function at t = 40 fs for three dif- ferent choices of wave vectors ~q = (k0 x , k0 y ) + K and initial pseudo-spinors: 1) [1, 0]T , with k0 x = 0 and k0 y d = 4, 2) [1, i]T , with k0 x = 0 and k0 y d = 4, and 3) [1, i]T , with k0 x d = 4 and k0 y = 0. The curves (symbols) in Fig. 4(b) show the expectation value hxi for each case ob- tained by the TB (Dirac) model. In case 1 (solid, cir- cles), the pseudo-spinor points in the z -direction, so that hσx i = hσy i and, consequently, the velocity for both in- plane directions must be zero. Indeed, the wave packet splits into two equal parts that propagate in opposite y directions, leading to vy = 0. In the x-direction, although 7 there is still a small zitterbewegung, hxi rapidly converges to a constant, leading to vx = 0. In case 2 (dashed, tri- angles), the pseudo-spinor points in the y-direction, but the momentum of the wave packet in this direction is zero, so that the packet splits in the y -direction, since vy = vF σx = 0, but drifts slowly in the −x direction (or, equivalently, y). In case 3 (dotted, squares), both the pseudo-spin and the momentum are in the y-direction, so that the wave packet propagates in this direction without any splitting. This situation is comparable to the one in Fig. 2(a), since in both cases the pseudo-spin and mo- mentum are in the same direction and, as a consequence, the wave packet propagates in this direction practically preserving its circular symmetry. However, in the case 3, the packet still presents a very weak oscillation in the y -direction, and also drifts very slowly in this direction, as one can see from the tra jectory of the packet in the x − y plane for this case, shown in the inset of Fig. 4(b). This oscillation and drift are related to the contributions of higher energy states in the wave packet: a wave packet centered around k (0) x = 0 and k (0) 6= 0, as in Fig. 2(a), y has a symmetric distribution of momenta in x-direction even for higher energies and, consequently, there is no ad- ditional oscillation in this direction. On the other hand, 6= 0 and k (0) a packet centered around k (0) y = 0, as in Fig. x 4(b), does not have a symmetric distribution of momenta in the y -direction due to the trigonal warping for higher energies and, consequently, some oscillations are observed in this direction. As the standard Dirac Hamiltonian HD for graphene does not take trigonal warping into account, this effect is not observed in the Dirac model. In the numerical work of Thaller,3 as well as in the an- alytical work of Maksimova,26 it is demonstrated within the Dirac model that even when k0 y = k0 x = 0, wave packet motion is still observed due to zitterbewegung ef- fects. The velocities of the wave packet obtained from our TB model of graphene for wave packet momenta ex- actly at the K ′ and K , i.e. points 1 and 2 in Fig. 1(a), respectively, are shown in Fig. 5. The velocities exhibit a damped oscillation with the same time-dependent mod- ulus for any pseudo-spin and Dirac point, though they point in different directions: for [1, 1]T (solid) and [1, i]T (circles) in K , the velocity is in the y and −x direc- tions of the lattice, respectively, which are exactly the directions of polarization of these pseudo-spinors after the φ = π/2 rotation required by the K cone 2. In the K ′ cone 1, the rotation angle is φ = π/6 and, accord- ingly, the velocity points in this direction, as one can see by the decomposition of the velocity in the components hvx i (dashed) and hvy i (dotted), which obey exactly the relations hvx i = (√3/2)hvi and hvy i = (1/2)hvi. Notice that the velocities converge exactly to vF /2, a value that can also be obtained analytically by making k0 y d → 0 in Eq. (20). Henceforth, we will consider initial wave vectors ~q0 (a) W (b) y R L 0 x FIG. 6: (Color online) (a) Sketch of the strained graphene sheet: we consider a rectangular sample of width W and height L, bent into an arc of circle with inner radius R. The unstrained graphene sheet is shown as open circles, for com- parison. (b) Strain-induced magnetic field barrier step, ob- tained by bending the graphene lattice only in the y ≥ 0 region. The number of atoms was reduced in both figures, in comparison to the lattices studied in this work, in order to improve the visualization. around the Dirac points 2 and 5 of Fig. 1(b), namely 3√3a0 (cid:19) , 3√3a0 (cid:19) and K ′ = (cid:18)0, − K = (cid:18)0, 4π 4π respectively. This choice is very convenient, since the ro- tation angles for these points are φ = π/2 and 3π/2, re- spectively, so that the pseudo-spinor [1, 1]T points to the y (-y ) direction in the former (latter) case. Hence, with this pseudo-spinor, wave packets in K (K ′ ) will propa- gate with positive (negative) velocity in the vertical zig- zag direction. (21) B. External magnetic fields and strain Recently,24 it was shown theoretically that bending a graphene sheet into an arc of a circle produces a strong and almost uniform pseudo-magnetic field profile. Fig- ure 6(a) illustrates such a strained system, where the rectangular graphene sample of width W and height L is bent into an arc of a circle with inner radius R. As the (pseudo) magnetic field points in the same direction (op- posite directions) at each K and K ′ points,21 the combi- nation of both external and strain-induced magnetic field effects provides a valley-dependent magnetic field. If one applies the appropriate external magnetic field for some configuration of the strained graphene, one can obtain an almost perfect suppression of the effective magnetic field at one of the Dirac cones, while the effective field in the other cone is enhanced. This leads to a compli- cated system to be studied within the Dirac approxima- tion, since one has two completely different systems for 8 the K and K ′ valleys. Namely, Landau levels would be present only around one of the cones (though one can- not expect a perfect Landau level spectrum, since the strain-induced magnetic field is not perfectly uniform in space), whereas in the other cone, the usual continuum spectrum would be observed. This motivated us to an- alyze the tra jectories of a wave packet in such a system within the TB model, where we do not need to include the pseudo-magnetic fields artificially in the Dirac cones, since they appear naturally when we consider the effect of the strain-induced changes of the inter-site distances on the hopping energies, as explained in the previous sec- tion. In this subsection, we investigate the dynamics of a wave packet with width d = 200 A and initial wave vec- y = 0.02 A −1 around the Dirac points K tor k0 x = 0 and k0 and K ′ of Eq. (21) in the presence of external and strain- induced magnetic field barrier steps. As in the K ′ valley the pseudo-spinor [1, 1]T is polarized in the negative y - direction of the graphene lattice, we choose [1, −1]T for this case, so that a wave packet in this valley will also propagate in the positive y -direction. In order to obtain a pseudo-magnetic field barrier step, we consider that the graphene layer is strained only in the y ≥ 0 region, as sketched in Fig. 6(b). We also consider an external mag- netic field ~B = BΘ(y ) z , where Θ(y ) is the Heaviside step function, which leads to a magnetic barrier step for y ≥ 0, described by the vector potential ~A = (−B yΘ(y ), 0, 0). In order to avoid effects due to zitterbewegung in the (pseudo) magnetic field region, the wave packet starts at the position x0 = 0, y0 = −420 A , so that it can travel for some time in the magnetic field-free region y < 0 until its velocity converge to a time independent value. The influence of the external and strain-induced mag- netic field barriers on the tra jectories of the wave packet are analyzed separately in Fig. 7, which shows the tra jec- tory of the centroid of the wave packets in K (symbols) and K ′ (curves) points, calculated as hri = (hxi, hy i), (a) in a non-strained graphene sheet with magnetic field bar- riers B = 5 T (solid, circles), 7 T (dashed, triangles) and 10 T (dotted, squares) and (b) in a strained graphene sheet with radius R = 1 µm (solid, circles), 0.8 µm (dashed, triangles) and 0.6 µm (dotted, squares). All the tra jectories form semi-circles in the y ≥ 0 region, which is due to the Lorentz force produced by the (pseudo) magnetic field. As the external magnetic field (radius of the strained region) increases (decreases), the radii of these semi-circular tra jectories are reduced, since a higher (pseudo) magnetic field produces a stronger mod- ulus of the Lorentz force. Notice that the radii of tra- jectories in the external and pseudo-magnetic fields cases are comparable, which means that for radii R = 1 µm - 0.6 µm of the strained graphene, the generated pseudo- magnetic field is also within ≈ 5 T and 10 T. Indeed, the strain induced pseudo-magnetic field distribution for the (a) 400 200 0 ) Å ( y -200 400 (b) 200 0 -200 -600 -400 -200 200 400 600 0 x (Å) FIG. 7: (Color online) Tra jectories of the wave packet in the x − y plane, obtained by the TB method for such a system, for y = 0.02 A −1 around K (symbols) and K ′ initial momentum k0 (curves) points, for (a) non-strained graphene with magnetic barrier height B = 5 T (solid, circles), 7 T (dashed, triangles) and 10 T (dotted, squares), and for (b) a graphene sheet bent into an arc of circle with radius R = 1 µm (solid, circles), 0.8 µm (dashed, triangles) and 0.6 µm (dotted, squares), consid- ering B = 0 T. In (b), symbols and curves coincide for each value of R. βΦ0 aL BS (x, y ) = −4c bend graphene ribbon is given by24 2R (cid:19) cos (cid:20) 2x 2R (cid:19)(cid:21) arcsin (cid:18) L arcsin (cid:18) L L × (cid:20)1 − 2R (cid:19)(cid:21) , arcsin (cid:18) L R + y (22) L where β ≈ 2 and c is a dimensionless constant which depends on the details of the atomic displacements.22 Considering L/R → 0 in Eq. (22) the pseudo-magnetic field can be approximated as BS ≈ −cβΦ0(cid:14)aR = ω/R. Using the value ω ≈ 4.5 × 104 TA estimated numerically in Ref. 23, one obtains pseudo-magnetic fields within BS ≈ 4.5 T - 7.5 T for R = 1 µm - 0.6 µm, which are of the same order of magnitude as the external magnetic fields that we considered. For the external magnetic field barrier, the tra jectories of wave packets in K and K ′ points form circles in opposite directions, as shown in Fig. 7(a), which is reasonable, since these packets have opposite momentum, which causes a sign change in the Lorentz force. Conversely, considering the strain-induced magnetic barrier illustrated in Fig. 6(b), the tra jectories of wave packets in K and K ′ curve in the same direction, since, although their momenta have opposite signs, the pseudo-magnetic fields also point in opposite directions at each Dirac cone K and K ′ . 9 C. Strain induced valley filter Let us consider the strained sample in Fig. 6(b) with R = 1 µm. By comparing the radius of the semi-circular tra jectory of the wave packet in such a system with those obtained for different intensities of the external magnetic field barrier, one obtains the strain-induced magnetic field for this value of R as ≈ 4.9 T. Figure 8(a) shows the tra jectories in the x − y plane of the centroid of the wave packets in a system where we combine a R = 1 µm strain for y ≥ 0 with an external magnetic field barrier B = 0 T (solid, open) and 4.9 T (dashed, full), for wave packets in the K (symbols) and K ′ (curves) Dirac points. In the absence of the external magnetic field, both the K and K ′ packets exhibit the same semi-circular tra jec- tory, as discussed earlier. However, when we combine the effect of the strain-induced and external magnetic field barriers, the wave packet in K ′ undergoes a stronger Lorentz force and is readily reflected, whereas the one in the K point performs a practically straight tra jectory, as if this packet is not influenced by any Lorentz force. This is a consequence of the fact that combining the ef- fects of a pseudo-magnetic field produced by a R = 1µm strain and a B = 4.9 T external magnetic field produces a stronger magnetic field in the K ′ point, while in the K point these fields equilibrate, producing a practically magnetic field-free region for particles in this cone. In this situation, the system works as a valley filter, where only wave packets in the K Dirac cone are allowed to pass through the strained region, whereas the wave packets in K ′ are reflected. The results for the wave packet in K for two other values of the external magnetic field are shown as thin solid lines, showing that within a range of ∆B = ±0.2 T around B = 4.9 T, which is a reasonable range for magnetic field intensities in experiments, only a weak Lorentz force is observed and the valley filter works fine. The results of Fig. 8 are obtained for both external and pseudo-magnetic field barriers starting at the same position y = 0. It is straightforward to verify that if there is a mismatch between the starting points of the strained and external field regions, some deviations will occur in the tra jectories of the wave packets but, pro- vided the length of the mismatch is much smaller than the magnetic length, the filtering effect is still stable. As an example, a 30 A mismatch between the external and pseudo-magnetic field barriers in the system analyzed in Fig. 8 would produce a ≈ 5◦ deviation in the otherwise vertical tra jectory of the wave packet in K , whereas the wave packet in K ′ is still readily reflected by the combi- nation of magnetic fields in the filter region. The probability P > of finding the particle in the strained y ≥ 0 region, calculated as P > (t) = Xn> Xm n,m 2 , Ψt where n> represents the lines of atomic sites with y ≥ 0, is shown as a function of time in Fig. 8(b). In the B = 0 (23) ) Å ( y 800 700 600 500 400 300 200 100 0 -100 -200 (a) 4.7 T 5.2 T (b) 1.0 0.8 0.6 > P 0.4 0.2 0.0 0 50 100 150 200 250 t (fs) -600 -400 -200 x (Å) 0 200 FIG. 8: (Color online)(a) Tra jectories on the x − y plane for y = 0.02 A −1 around wave packets with initial momentum k0 K (symbols) and K ′ (curves) points, considering a graphene sheet bent into an arc of circle with radius R = 1 µm and an external magnetic field B = 0 T (open, solid) and 4.9 T (full, dashed). The thin solid curves show the results for two other magnetic field intensities for the K packet. (b) Probability of finding the particle in y ≥ 0 as a function of time, for wave packets with the same configurations as in (a). T case, both wave packets in K (open circles) and K ′ (solid) stay in the strained y ≥ 0 region until t ≈ 300 fs, when they turn back into the y < 0 region, reflected by the Lorentz force induced by the strain. However, for B = 4.9 T, P > already approaches zero at t ≈ 175 fs for the packet in K ′ (dashed), whereas for K (full circles), it remains close to 1 even for large t. The efficiency of the proposed valley filter is confirmed by Fig. 9, where we present P > as a function of time for initial wave packets given by a combination of Gaussians 1.0 0.8 0.6 0.4 0.2 > P 0.0 0 50 200 250 = 0.7 = 0.5 = 0.3 100 150 t (fs) FIG. 9: (Color online) Probability of finding the electron in the filter region y ≥ 0, for an initial wave packet given by a combination of Gaussian distributions around both K and K ′ Dirac points, for three different values of the K -component of the wave packet β . 10 (24) around the K and K ′ points in Eq. (21): Ψ = √αΨK ′ + pβΨK , where ΨK (K ′ ) is the Gaussian wave packet in Eq. (18) with momentum ~q around the K (K ′) Dirac point. The results are presented for three different values of β , where one can easily see that the probability of finding the packet in the strained region exhibits a peak at t ≈ 80 fs but, as the K ′ part of the packet is reflected by the mag- netic barrier, this probability decays, reaching exactly P > = β for large t. Such a system proves to be a perfect valley filter, which is able to reflect all the components of the incoming packet that are in the K ′ point and trans- mit a wave packet that is composed only of particles with momentum in the vicinity of K . We point out that when a wave packet reaches the edges of a graphene nanoribbon, it can be scattered to a different Dirac cone.41 Consequently, the efficiency of the valley filter could be compromised if one considers a thin nanoribbon, so that the filtered wave packet could still reach its boundaries and scatter back to the other valley. In order to avoid such an effect, we have considered a wide nanoribbon, so that for the time intervals we study in this work, boundary effects are not significant. D. External and pseudo magnetic field effects on the zitterbewegung In a previous paper, Rusin and Zawadzki9 used the Dirac Hamiltonian for graphene to show that the zitter- bewegung, which is transient for B = 0, as discussed ear- lier, is permanent for B 6= 0. Furthermore, the authors showed that for a Gaussian wave packet, the time evolu- tion of the average value of the current is different in x and y directions, which they explain as due to the fact that the Dirac Hamiltonian is not symmetric in the mo- menta px and py . Although the same authors say in their subsequent paper27 that this effect is unphysical, because it violates the rotational symmetry of the x − y graphene plane, we believe this result is still physical: one must re- member that the Dirac model of graphene comes from the tight-binding approach for this system, which describes a honeycomb lattice of atoms that is not symmetric in the x − y plane by definition, exhibiting C6v symmetry, as mentioned in previous section. For each K and K ′ point, the coordinates x and y in the Dirac Hamiltonian represent different directions in the real honeycomb lat- tice of graphene, where for an infinite sample the x (y) coordinate in the Dirac equation is related to one of the zigzag (armchair) directions of the real sample. In this subsection, we use our TB model of graphene to extend the previous study of Rusin and Zawadzki9 to different situations. We now study the dynamics of a wave packet with width d = 200 A , pseudo-spinor c1 = 1 and c2 = 1 and x = k0 initial wave vector k0 y = 0, i.e. exactly at one of the Dirac points K and K ′ in Eq. (21), in the presence ) c ( 0.2 0.0 -0.2 1.0 0.5 0.0 -0.5 -1.0 1 0 -1 1 0 -1 -2 0 (a) (b) (c) (d) 20 40 60 80 100 120 140 160 t (fs) FIG. 10: (Color online) Electromagnetic dipole radiation as a function of time for wave packets with initial pseudo-spinor [1, 1]T at K (thin curves) and K ′ (thick curves), considering a graphene sheet (a) in the absence of strain and magnetic fields, (b) under an uniformly applied magnetic field B = 4.9 T, (c) bent into an arc of circle with radius R = 1µm (see Fig. 6) and (d) with both the uniform magnetic field B = 4.9 T and the R = 1µm bending. Solid (dashed) curves are the results for the εy (εx ) component. of an uniform applied external magnetic field ~B = B z , instead of the magnetic field barrier step considered in the previous subsection. We also consider a pseudo- magnetic field obtained by bending the whole rectangu- lar graphene sample into an arc of a circle with radius R, as illustrated in Fig. 6(a). The radius of the circle is considered as R = 1 µm, which produces a ≈ 4.9 T pseudo-magnetic field, as demonstrated in the previous subsection. Accordingly, we consider the external uni- form magnetic field as B = 4.9 T. A few experimental techniques have been suggested in the literature for the observation of zitterbewegung.8,11 An interesting one27 is based on the fact that the wave packet Ψ(~r , t) exhibits an electric dipole moment ~D(t) = hΨ(~r, t)~r Ψ(~r , t)i and, consequently, the zitter- bewegung yields an oscillation of this dipole moment, which is a source of electromagnetic radiation, described classically42 by the equation ~ε(t) = d2 ~D(t) dt2 sin Φ 4πǫ0 s , (25) where s and Φ are the distance and angle of observation, respectively, and ǫ0 is the vacuum permittivity. Figure 10 shows the effects of external and strain- induced magnetic fields on the electric field radiation 11 produced by the zitterbewegung, written in units of εc = sin Φ(cid:14)4πǫ0s. Only weak oscillations are expected in the armchair (x) direction, since the pseudo-spin [1, 1]T points in the x-direction in the Dirac model which, as mentioned earlier, is related to the zigzag (y ) direction of the honeycomb lattice. Indeed, in Fig. 10(a-d), the x- component of the electric field (dashed) is always close to zero. In Fig. 10(a), we present the results in the absence of strain and magnetic fields, for comparison. In this case, the oscillations on the electric field are suppressed for larger time, which is due to the transient character of the zitterbewegung. Besides, the results for εy (solid) in the K (thin curves) and K ′ (thick curves) points have op- posite signs, since for these points the axis of the Dirac cones are rotated by angles which differ by π/2 differ- ence. In Fig. 10(b), the uniformly applied magnetic field B = 4.9 T in an unstrained sample leads to persistent oscillations in εy , which is related to the discrete Lan- dau level spectrum created by this field. Each Landau level that is populated by the Gaussian distribution con- tributes with a different frequency.9 Figure 10(c) shows that such a persistent behavior is also obtained by bend- ing the graphene sheet into an arc of circle with radius R = 1µm, which produces a pseudo-magnetic field of the same order of magnitude. Notice that the ampli- tude of oscillations in this case is four times smaller than those found in Fig. 10(b) for the unstrained sample under an external magnetic field. In fact, these two cases are not expected to produce the same zitterbewegung, be- cause, although both samples exhibit approximately the same magnitude of magnetic field, the strained sample has not only the pseudo-magnetic field, but also a dif- ferent distribution of atomic sites. Thus, in the strained case, there is an additional change in the direction of the pseudo-spin polarization as the wave packet drifts, due to the lattice distortion itself. As we have demonstrated in Sec. III A, the zitterbewegung strongly depends on the pseudo-spin polarization and hence, the different in- terplay between the strained atomic sites and the initial pseudo-spin polarization produces a different zitterbewe- gung for the strained case, as compared to the one of the unstrained sample under an external field. In Fig. 10(d), we combine the effects of the R = 1 µm strain and B = 4.9 T external field to produce a sys- tem where the magnetic field is practically zero in the K point, but is non-zero in the K ′ point, so that only the packet in the K ′ point exhibits persistent oscillations. For the K point, the external field compensates only the effect of the pseudo-magnetic field, namely, the persis- tent zitterbewegung, but it does not remove the effect of the lattice distortion. As a result, comparing the results for K (thin curves) in Figs. 10(a) and (d), one observes that the oscillations are transient in both cases, since there is no effective magnetic field, but they still exhibit a different oscillation profile at small t, due to the lat- tice distortion in the latter case, which is absent in the former. IV. CONCLUSION We presented a study of the dynamics of Gaussian wave packets in graphene under external and strain- induced magnetic fields, where the latter is obtained by bending the graphene sheet into an arc of a circle. The dependence of the zitterbewegung on the initial pseudo- spin of the wave packet is investigated, and the results ob- tained by means of the tight-binding model and the Dirac equation are compared. We demonstrate that the com- bination of the pseudo-magnetic field, induced by bend- ing the graphene sheet, along with an external magnetic field with appropriate strength can be used as an effi- cient valley filter. An incoming wave packet composed of momenta around the K and K ′ Dirac points is scattered such that all its components in one of the Dirac cones un- dergoes a strong Lorentz force and are readily reflected, while the components in the other cone are allowed to pass through the device with only small distortions in their tra jectory, due to the very weak residual Lorentz force. Our results also show that in the absence of exter- nal or strain-induced magnetic fields, the zitterbewegung is a transient effect, whereas in the presence of any of 12 these fields, the oscillations persist in time. In a strained sample under an external magnetic field with the appro- priate strength, the effective magnetic field in one of the Dirac cones is enhanced, whereas in the other cone it is practically cancelled. In this situation, a permanent zitterbewegung is observed only for wave packets in one of the Dirac cones. The wave packet oscillations produce electric field radiation, which can be detected experimen- tally. Finally, we believe the present work contributes to a better understanding of the relation between the results obtained from TB and Dirac approaches for graphene and those to be observed in future experiments on strained graphene-based structures. Acknowledgments This work was financially supported by CNPq, under contract NanoBioEstruturas 555183/2005-0, the Bilateral PRONEX/CNPq/FUNCAP, CAPES, programme between Flanders and Brazil, the Bel- gian Science Policy (IAP) and the Flemish Science Foundation (FWO-Vl). ∗ Electronic address: [email protected] † Electronic address: [email protected] ‡ Electronic address: [email protected] 1 K. S. Novoselov, A. K. Geim, S. V. Morozov, D. Jiang, Y. Zhang, S. V. Dubonos, I. V. Grigorieva, and A. A. Firsov, Science 306, 666 (2004). 2 A. H. Castro Neto, F. Guinea, N. M. R. Peres, K. S. Novoselov, and A. K. Geim, Rev. Mod. Phys. 81, 109 (2009). 3 B. Thaller, arXiv:quant-ph/0409079v1 (2004). 4 W. Zawadzki, T. M. Rusin, Phys. Lett. A 374, 3533 (2010). 5 G. D`avid and J. Cserti, Phys. Rev. B 81, 121417(R) (2010). 6 E. Schrodinger, Sitzungsber. Preuss. Akad. Wiss. Phys. Math. Kl. 24, 418 (1930). 7 W. Zawadzki, Phys. Rev. B 72, 085217 (2005). 8 J. Schliemann, D. Loss, and R. M. Westervelt, Phys. Rev. Lett. 94, 206801 (2005). 9 T. M. Rusin and W. Zawadzki, Phys. Rev. B 80, 045416 (2009). 10 Y.-X. Wang, Z. Yang, and S.-J. Xiong, Europhys. Lett. 89, 17007 (2010). 11 R. Gerritsma, G. Kirchmair, F. Zahringer, E. Solano, R. Blatt, and C. F. Roos, Nature (London) 463, 68 (2010). 12 V. M. Pereira and A. H. Castro Neto, Phys. Rev. Lett. 103, 046801 (2009). 13 Seon-Myeong Choi, Seung-Hoon Jhi, and Young-Woo Son, Phys. Rev. B 81, 081407(R) (2010). 14 F. M. D. Pellegrino, G. G. N. Angilella, and R. Pucci, Phys. Rev. B 81, 035411 (2010). 15 O. Bahat-Treidel, O. Peleg, M. Grobman, N. Shapira, M. Segev, and T. Pereg-Barnea, Phys. Rev. Lett. 104, 063901 (2010). 16 T. Fujita, M. B. A. Jalil, and S. G. Tan, Appl. Phys. Lett. 97, 043508 (2010). 17 Z. Wu, F. Zhai, F. M. Peeters, H. Q. Xu and K. Chang, arXiv:1008.4858v3 (2010). 18 E. Cadelano, S. Giordano, and L. Colombo, Phys. Rev. B 81, 144105 (2010). 19 G. Cocco, E. Cadelano, and L. Colombo, Phys. Rev. B 81, 241412 (2010). 20 Y. Lu and J. Guo, Nano Res. 3, 189 (2010). 21 M. A. H. Vozmediano, M. I. Katsnelson, and F. Guinea, Phys. Rep. 496, 109 (2010). 22 F. Guinea, M. I. Katsnelson, and A. K. Geim, Nat. Phys. 6, 30 (2010). 23 T. Low and F. Guinea, Nano Lett. 10, 3551 (2010). 24 F. Guinea, A. K. Geim, M. I. Katsnelson, and K. S. Novoselov, Phys. Rev. B 81, 035408 (2010). 25 N. Levy, S. A. Burke, K. L. Meaker, M. Panlasigui, A. Zettl, F. Guinea, A. H. Castro Neto, M. F. Crommie, Sci- ence 329, 544 (2010). 26 G. M. Maksimova, V. Ya. Demikhovskii, and E. V. Frolova, Phys. Rev. B 78, 235321 (2008). 27 T. M. Rusin and W. Zawadzki, Phys. Rev. B 78, 125419 (2008). 28 E. Romera and F. de los Santos, Phys. Rev. B 80, 165416 (2009). 29 V. Krueckl and T. Kramer, New J. Phys. 11, 093010 (2009). 30 A. Rycerz, J. Tworzyd lo, and C. W. J. Beenakker, Nat. Phys. 3, 172 (2007). 31 A. Weisse, G. Wellein, A. Alvermann, and H. Fehske, Rev. Mod. Phys. 78, 275 (2006). 32 S. Mahapatra and N. Sathyamurthy, J. Chem. Soc. 93, 773 (1997). 33 R. Kosloff, J. Phys. Chem. 92, 2087 (1988). 34 A. Chaves, G. A. Farias, F. M. Peeters, and B. Szafran, Phys. Rev. B 80, 125331 (2009). 35 D. A. Bahamon, A. L. C. Pereira, and P. A. Schulz, Phys. Rev. B 79, 125414 (2009). 36 M. Governale and C. Ungarelli, Phys. Rev. B 58, 7816 (1998). 37 V. M. Pereira, A. H. Castro Neto, and N. M. R. Peres, Phys. Rev. B 80, 045401 (2009). 38 N. Watanabe and M. Tsukada, Phys. Rev. E 62, 2914 (2000). 39 J. M. Pereira, A. Chaves, G. A. Farias, and F. M. Peeters, Sem. Sci. Tech. 25, 033002 (2010). 40 H. N. Nazareno, P. E. de Brito, and E. S. Rodrigues, Phys. Rev. B 76, 125405 (2007). 41 J.-H. Chen, W. G. Cullen, C. Jang, M. S. Fuhrer,and E. D. Williams, Phys. Rev. Lett. 102, 236805 (2009). 42 J. D. Jackson, Classical Electrodynamics (Wiley, New York, 1975). 13

Subsets and Splits